Perspective

Heterogeneous Combustion: Recent Developments and New Opportunities for Chemical Engineers
Arvind Varma, Victor Diakov, and Evgeny Sharovich
School of Chemical Engineering, Purdue University, West Lafayette, IN 47907 DOI 10.1002/aic.10697 Published online September 27, 2005 in Wiley InterScience (www.interscience.wiley.com).

Introduction
ombustion is arguably the rst chemical process used by humans. The reason for this is simple: Combustion is generally a self-sustained process and, once initiated, does not require external energy input; on the contrary, it releases heat that can be put to use in various ways. Since Prometheus times, the applications of combustion have become prevalent throughout our life: car engines, power generation, rearms, cigars and cigarettes, rocket propulsion, etc. While, in some cases, combustion occurs in premixed homogeneous gaseous mixtures, most of the earlier examples involve heterogeneous processes. From a chemical engineers viewpoint, the self-sustainability and relative simplicity are attractive features to use combustion in addressing important problems facing society. Novel experimental techniques are constantly bringing new insight about processes occurring during combustion of condensed media. Based on these, fundamental concepts related to heterogeneous combustion are being developed which open new possibilities for applications. The scope of this article is to describe recent ndings in the eld and to discuss how these can contribute to the solution of contemporary chemical engineering problems. In this context, we focus attention on four specic topics of signicant current and emerging interest: combustion synthesis of advanced materials, hydrogen production for fuel cells, carbon sequestration, and in-situ resource utilization in extraterrestrial environments. It is widely agreed that energy is a critical problem facing humanity today,1 so the demand for industrial processes with low energy consumption will continue to increase. An example of such technology is combustion synthesis (CS), which has been applied successfully for production of advanced materials, such as nanoscale oxide powders, implants and functionally graded materials.2,3 The development of CS processes depends on progress in understanding combustion mechanisms
Correspondence concerning this article should be addressed to A. Varma at avarma@purdue.edu.

in heterogeneous media. Recent achievements in this eld offer promise for new exciting applications. Another energy-related problem is the lack of robust power sources, which inhibits further growth opportunities for portable electronic devices. Heterogeneous combustion is likely to play an important role in this eld, leading particularly to new methods for hydrogen storage. The rapid worldwide growth in energy consumption is increasing greenhouse gas emissions. To avoid further global warming, carbon sequestration technologies are required. In this direction, along with other CO2 capture methods, novel processes, such as chemical looping combustion offer great potential. Finally, in future space exploration, the need for propellants and materials will expand chemical engineering approaches beyond the bounds of Earth. In this context, heterogeneous combustion is expected to play an important role.

Combustion Synthesis of Advanced Materials

We discuss two wide classes of CS processes. These are the original solid ame CS or self-propagating high-temperature synthesis (SHS)2,4 and the process of combusting a liquid (typically aqueous) solution of reactants.3 The CS of initially solid samples can be used to obtain such compounds as borides, carbides, hydrides, intermetallics, nitrides, oxides and silicides in the form of powders, poreless, porous or functionally graded materials, while solution combustion synthesis (SCS) is an attractive method for producing nanomaterials, particularly oxides. When using these methods, in each case some specic properties of the resulting material are targeted. For example, biocompatibility of CS products is of critical importance in fabricating biomaterials. In addition, by ne-tuning the operating pressure, it was possible to synthesize poreless orthopedic implants.5 The CS technique also proved useful in producing porous functionally graded biomaterials.6

Mechanisms of combustion wave propagation in heterogeneous media

The general idea about movement of combustion waves, suggested more than 100 years ago,7 is that layer-by-layer
Vol. 51, No. 11 AIChE Journal

2005 American Institute of Chemical Engineers

2876

November 2005

propagation occurs because of initiation of the reaction between premixed components by heat conducting from a neighbor layer in which the exothermic reaction is proceeding. Thus, the propagation is determined by two main factors: kinetics of the reaction heat evolution and rate of heat transfer from the reaction zone to the unreacted mixture. In heterogeneous systems, reaction kinetics is the controlling step for dense compact media containing ne particles (quasi-homogeneous mechanism), while particle-to-particle heat conduction is limiting for loosely packed coarse mixtures (relay-race mechanism).8 Developments in experimental techniques have led to observations of new features and to microheterogeneous modeling approaches. These are increasing fundamental understanding of the involved processes, and enhance progress toward the synthesis of advanced materials with tailored microstructures and properties. A novel digital high-speed video recording technique was developed, which allows in situ observation of rapid processes occurring at the microscopic level.9 Using this technique, signicant information about the microstructure of gasless combustion waves was obtained and a basis was created for understanding the mechanisms of fast chemical reactions occurring in heterogeneous media. It was found that the combustion wave may exhibit unexpectedly complex dynamics. Although on macroscopic length and timescales the combustion reaction front may appear to move in a steady manner, on the microscopic level it has a complex character that is related to the reaction mechanism. Over a wide range of experimental conditions, macroscopically steady reaction fronts in heterogeneous mixtures may exhibit random microscopic uctuations in shape and instantaneous velocity, which are directly related to the microstructure of the reaction mixture (i.e. the initial heterogeneity of the reaction medium).10 The reaction wave can propagate in two qualitatively different modes. The rst mode, quasi-homogeneous reaction wave (QRW), involves a continuous front which moves uniformly and there is practically no variation of behavior along the surface of this front (Figure 1a). Another type of combustion wave propagation (Figure 1b) exhibits a lower average temperature, but relatively bright spots appear randomly, indicating local regions of high temperature along the front. Thus, the front moves forward only as a consequence of appearance of the hot spots, and the overall progress of the front occurs only locally in the vicinity of the spot. Based on these features, this mode of propagation is called scintillating reaction wave (SRW). The classical consideration of combustion in condensed systems11 is based on mass and energy balances at the macroscopic scale and using average values for reactant concentrations and temperature. It appears that in heterogeneous media, the characteristic length of heat conduction may often approach the scale of particle size,12 leading to the SRW regime. The hot spots appear when the local region reaches the ignition temperature. Interestingly, in mixtures where the reaction is initiated by the melting of a component (note that the SRW mode was observed only in those systems where at least one reactant melts), the melting point plays the role of ignition temperature due to melt spreading and reaction rate acceleration induced by dramatic increase in the contact area and diffusion coefcients. To capture the features of the SRW propagation mode, a micro-heterogeneous cell model was developed, involving regAIChE Journal November 2005

Figure 1. Two modes of combustion wave propagation: (a) quasihomogeneous; (b) scintillating (from Varma et al.9).

ular lattice with random voids.13,14 The model was applied to simulate combustion wave in heterogeneous media by considering a discrete matrix of reaction cells, of particle size dimension, with physical and transport properties varying due to random organization of the phases in the reaction mixture. The model captures well the observed peculiarities of heterogeneous combustion wave propagation. Specically, model predictions agree with experimentally determined threshold between quasi-homogeneous and relay-race combustion modes (see Figure 2). While the variety of dynamic behavior of homogeneous combustion processes can itself be overwhelming,15 the features of the observed SRW9 are certainly related to the heterogeneity of the reacting media. Even a quasi-homogeneous steady-state consideration for combustion of imperfectly mixed solid-solid systems demonstrates that signicant deviations in reaction conversion and combustion temperature may arise.16 Another approach, based on localized reaction sites in a regular lattice for one17- and two18- dimensional cases, allows one to compute the instantaneous combustion velocity distributions for both relay-race and quasi-homogeneous mechanisms. It is evident from the above that signicant progress has been made, both experimentally and theoretically, in understanding how combustion waves travel in heterogeneous media leading to the CS of advanced materials. The prediction and precise control of product properties based on reactant composition and microstructure, however, still remains a challenge for the future.

Synthesis of nanoscale oxides by solution combustion

Solution combustion synthesis (SCS) is a exible technique where oxidizing and reducing precursors are mixed on the molecular level and, under unique conditions of rapid hightemperature reactions, nanoscale powders of desired compositions can be synthesized in one step. The oxidizer (typically nitrates or oxalates) and fuel (e.g., hydrazine, glycine, urea) are
Vol. 51, No. 11 2877

Figure 2. Combustion wave propagation with varying particle size at constant density; Ti-air system (from Varma et al.14).

be synthesized directly by SCS in air atmosphere, without reducing the product. Further, using the SCS of iron oxide nanopowders as an example, it was shown recently that varying the mixture compositions, both for fuels and oxidizers, is an attractive methodology to control the product composition and properties.22 Studies directed toward understanding the mechanisms of SCS are important to tailor desired product properties. In this context, experimental data concerning mixture reactivity are important and TGA/DTA measurements coupled with massspectrometry are useful.19 The data become yet more informative when related to in situ SCS measurements. Dynamic in situ temperature measurements during SCS of various iron oxides were related to TGA/DTA results.22 In this manner, the mechanisms of solution ignition and combustion were addressed in detail. It was shown that specic characteristic phase transformations, which vary with the individual fuel-oxidizer system, are responsible for the observed rapid chemical reactions. Similar experimental studies of iron(III) nitrate interactions with model fuels identied the relative reactivities of different functional groups and led to the formulation of reaction mechanisms.23 Interestingly, the variation of fuel-to-oxidizer ratio also affects the combustion mode of the mixture, as shown for SCS of lanthanum-strontium chromite perovskites.24 Depending on this ratio, the maximum reaction temperature varied, and either smoldering, volume combustion or SHS occurred, leading to different product properties. The rate-controlling step during SCS was also established. Based on limited mechanistic studies of SCS conducted to-date, it is clear that reactant composition and reaction conditions signicantly inuence product properties. The relationship between these, however, is not straightforward, and its further development is a promising direction for the future. Detailed mechanistic studies are expected to play a key role in this development.

mixed in an aqueous solution, and heated until self-ignition, yielding product gas and nanoscale oxide powders. Solution combustion is a relatively new topic in CS, and is being widely used by researchers to obtain materials for a vast variety of applications, such as oxygen storage capacitors, support and catalyst for catalytic converters, ceramic pigments, solid oxide fuel cell electrode and electrolyte materials, magnetic ferrites, dielectric and piezoelectric components.3 The exploration of SCS continues by searching for more suitable and environmentally friendly fuels. For example, Prakash et al.19 developed an SCS technique leading to the formation of complex metal oxides using hexamethylenetetramine, which unlike other fuels (e.g., oxalyldihydrazide, carbohydrazide), can be synthesized by condensation of ammonia with formaldehyde, avoiding potentially harmful hydrazine. A way to optimize the properties of SCS nanocomposites is by applying fuel mixtures and varying their ratio. Thus, depending on the fuel mixture consisting of ammonium acetate, urea and glycine, nanoscale zirconia toughened alumina composite was synthesized in various particle sizes.20 The fuel-tooxidizer ratio variation has also been shown to inuence product chemical composition in the microwave initiated SCS of Ni and NiO powders.21 It was demonstrated that Ni particles can
2878 November 2005

Hydrogen for Fuel Cells

Hydrogen storage for fuel cells is an important multidisciplinary area of contemporary research, especially attractive for chemical engineers.25 There are several possible methods for hydrogen storage, such as compressed and liqueed hydrogen, hydrogen adsorption on carbon materials of large surface area, and metal hydrides,26 but all have limitations which prevent their immediate use in practice. In this context, some promising heterogeneous combustion based techniques have recently been proposed and are described in this section.

Power sources for portable electronic devices

The number and performance characteristics of portable electronic devices in consumer, industrial, medical and military markets are increasing continuously. They include mobile phones, notebook computers, PDAs, digital cameras, DVD players, RFID and barcode scanners, autonomous robots, handheld diagnostics, infusion pumps, debrillators, etc. The total number of such units in use is difcult to evaluate, but even conservative estimates suggest more than a billion. Simultaneously with increasing performance, the power demand of portable electronic devices is also increasing. Rechargeable
Vol. 51, No. 11 AIChE Journal

batteries are approaching their limits of specic energy (0.2 Wh/g for the most advanced lithium-ion batteries). This situation is well illustrated by the fact that batteries in notebook computers typically require recharging after 2 h of operation. Fuel cells provide much higher specic energy and, for this reason, are expected to be widely used in the near future. In contrast to batteries, a fuel cell power system includes separate conversion device (fuel cell itself) and fuel storage/feeding system, analogous to internal combustion engine and gasoline tank/pump in an automobile. Thus, the future power source for portable electronics will include a permanent fuel cell unit inside the device and replaceable fuel cartridges which can be relled, recycled or discarded. Methanol and hydrogen are the most promising candidate fuels to feed portable fuel cells. They have specic energies 6.3 and 39.3 Wh/g, respectively. The assumption that direct methanol fuel cells (DMFC) can reach conversion efciency about 25% leads to the conclusion that specic energy of methanol cartridge would be 1.5 Wh/g, which is signicantly higher than for the best batteries (0.2 Wh/g). Unfortunately, DMFCs have many drawbacks, such as low power density, methanol crossover, electrode poisoning, methanol toxicity, and catalyst cost, which are serious obstacles to their use. Hydrogen fuel cells provide much higher power density, do not have any methanol-related problems, and operate with conversion efciency about 50%. The key problem in development and use of hydrogen fuel cell power sources is hydrogen storage. High pressure gas containers are too heavy (thus, leading to much lower specic energy than DMFC) and liquid hydrogen (temperature 20 K) is not possible for use in portable applications. It is generally agreed that chemical methods for hydrogen storage provide the maximum H2 yield and are the most promising for portable fuel cell applications. A simple estimate shows that 8 wt. % H2 yield results in specic energy of hydrogen fuel cartridge 1.5 Wh/g, the same as for DMFCs. The hydrogen fuel cell with the same specic energy as DMFC would be advantageous because of much higher power density and elimination of the methanol-related problems. Thus, the natural question arises: What chemicals and technology should be used to reach the maximum H2 yield?

Heterogeneous combustion methods for hydrogen generation

Borohydrides of light metals (Li, Be, Na, Mg, Al) are known to be excellent sources of hydrogen, which can be obtained by reactions of these compounds with water or oxidizing agents. Sodium borohydride (NaBH4), discovered during World War II by Nobel laureate Herbert C. Brown of Purdue, is produced for various applications at the rate of about 10,000 tons per year. Hydrolysis of sodium borohydride is a well-known process,27 extensively studied for hydrogen generation.28,29 Aqueous alkaline NaBH4 solutions are stable and their contact with selected catalysts (e.g., Ru) leads to the reaction NaBH4 2 H2O 3 NaBO2 4 H2 . (1)

to generate new sodium borohydride. Unfortunately, the practical strength of aqueous NaBH4 solution is limited to 30 wt. % borohydride, thus decreasing maximum H2 yield to 6.3 wt. %.30 Furthermore, reaction initiation requires introducing the catalyst to the mixture, which is difcult particularly for portable applications. This problem is overcome in combustion-based approaches for hydrogen generation, which require only ignition and no catalyst. It should be noted that combustion of solid mixtures is widely used for gas generation to create thrust in rocket engines, inate air bags in case of car collisions, provide emergency oxygen to airplane passengers, etc. The amounts of gases stored in chemical compounds, for example, oxygen in sodium chlorate (NaClO3) and hydrogen in sodium borohydride (NaBH4), are comparable with storage abilities of liquid (cryogenic) O2 and H2, respectively, and much larger than those of pressurized gas tanks. The gas generating compositions can be easily and safely stored for years, and ignited when the product gas is required. Recently, new hydrogen-generating pyrotechnic compositions were proposed to feed fuel cells for portable electronics by researchers from CEA (French Commission on Atomic Energy) and SNPE (French company for development of energetic materials).31-33 In these mixtures, hydrogen is generated by combustion reactions of metal borohydride with oxidizer salt, such as NH4ClO4, NaClO4 and Sr(NO3)2. These reactions are highly exothermic, easily initiated, and do not require any catalyst, which makes the method cost-effective and attractive for portable applications. Stoichiometric mixtures of the proposed reactants, however, exhibit low hydrogen yield, while those with high metal borohydride content do not burn. As a result, the maximum experimental H2 yield obtained by this method was 5.9 wt. %.31 An alternative approach for hydrogen generation, proposed in Russia, uses combustion of nanoscale aluminum with gelled water.34 Here, water acts as an oxidizer for Al and simultaneously as the sole source of hydrogen. The adiabatic combustion temperature of Al/H2O stoichiometric mixture is close to 3,000 K (at 1 atm pressure). Despite this high temperature, for self-sustained combustion, the use of nanoscale Al powder and gelling of water (for example, by adding polyacrylamide) are necessary. The use of nanopowder decreases the ignition temperature of Al, while gelling inhibits water evaporation during combustion. Hence, the mixture ignites easily and burns in inert atmosphere, producing hydrogen Al 3/2 H2O 3 1/2 Al2O3 3/2 H2 . (2)

High theoretical hydrogen yield (10.8 wt. % for stoichiometric mixture) makes this process attractive for applications. The other reaction product, sodium metaborate, is water-soluble, environmentally benign and can either be discarded or recycled
AIChE Journal November 2005

Unfortunately, low hydrogen yield of the Al/H2O system (theoretical limit 5.6 wt % for the stoichiometric mixture) is a drawback of this method. A new combustion-based method for hydrogen generation was recently proposed by us.35 To simultaneously reach high hydrogen yield and combustion efciency, we use triple sodium borohydride/aluminum/water mixtures, in which water acts as an oxidizer for both sodium borohydride and aluminum and also as a source of hydrogen (parallel reactions (1) and (2) in one mixture). Sodium borohydride is an additional hydrogen source, while aluminum increases combustion temperature, eliminating the need for catalyst. Nanoscale aluminum powVol. 51, No. 11 2879

Carbon Sequestration
Methods based on heterogeneous combustion could also help to solve another important global problem engaging attention of chemical engineers: carbon sequestration. According to the Presidents Global Climate Change Initiative, by the year 2012 the greenhouse gas intensity of the U.S. economy should be decreased by 18%. Carbon sequestration, along with making energy systems more efcient and increasing the use of low carbon fuels, is the key to accomplishing this goal. Carbon sequestration is the capture and storage of CO2 and other greenhouse gases that would otherwise be emitted to the atmosphere. A clear priority for near-term deployments, among various sequestration possibilities, is to capture CO2 from large-scale emission sources, such as power plants, and store it in underground formations.38 There are three main CO2 capture technologies, classied as post-combustion, precombustion, and oxyfuels. Post-combustion refers to capturing CO2 from a ue gas after a fuel has been burnt in air. The CO2 in ue gas is dilute (315 vol %), at low-pressure (0.1 0.2 MPa), and often contaminated with traces of sulfur and particulate matter. Thus, a very large volume of ue gas has to be treated, so that equipment is large and capital costs are high. Precombustion refers to a process where a hydrocarbon fuel is reacted with oxygen, air or steam to give mainly synthesis gas (CO and H2). The carbon monoxide then reacts with steam in a water-gas shift (WGS) catalytic reactor to give CO2 and more hydrogen. Finally, the CO2 is removed and the hydrogen is used as a fuel in a turbine cycle. This technology signicantly facilitates CO2 removal because the product gas contains CO2 in high concentration (30 50 vol %) and at high pressure (1.5 3 MPa). There are, however, few gasication-based power systems currently in operation. Oxyfuel is an approach where a hydrocarbon fuel is burnt in pure or nearly pure oxygen rather than air. The combustion product is then a mixture of CO2 and water, from which CO2 can be easily separated. The basic infrastructure for oxygen is available in coal gasication plants, but the added oxygen capacity increases capital and operating expenses. In this article, we focus on an alternative approach for oxygen supply, the so-called chemical looping combustion (CLC). The chemical-looping reaction technology is based on the cyclic use of solid medium which reacts with gaseous reactants. This technology has been used in many processes such as catalytic cracking and sulfur removal in coal gasication. The idea to use chemical looping in combustion-based power plants was originally proposed by Richter and Knoche39 in 1983. In recent years, CLC has attracted much attention because of easy separation of CO2 in this method. Ishida in Japan has conducted extensive research of CLC since 1994.40-42 In Europe, CLC has been investigated vigorously for the last ve years.43-46 To our knowledge, few research efforts have been devoted to CLC in the U.S. so far.47 In CLC technology, a metal oxide (MeO) is used as an oxygen carrier, which transfers oxygen from air to the fuel. The oxygen carrier circulates between two interconnected uidized bed reactors (see Figure 4), similar to the VPO-catalyzed butane to maleic anhydride process commercialized by DuPont.48 In one reactor, the MeO is reduced by a gaseous fuel, such as natural gas
Vol. 51, No. 11 AIChE Journal

Figure 3. Combustion front propagation in borohydride/ aluminum/water mixture; mass ratio NaBH4:Al: H2O 1:2:3 (from Sharovich et al.35).

ders are used to ensure high combustion efciency. Along with the three reactants, the mixtures include small quantities of a gelling agent (e.g., polyacrylamide) and a stabilizer (e.g., NaOH) to prevent hydrolysis of borohydride at room temperature. It should be noted that the reaction products include, besides hydrogen, sodium metaborate and alumina. Both compounds are environmentally benign materials. They can be either discarded or, if economically justied, recycled. Experiments with NaBH4/nanoAl/gelled H2O mixtures show that35 they exhibit stable self-sustained combustion (see Figure 3); hydrogen yield, measured by gas chromatography, approaches 7 wt %; hydrogen release efciency is 74 77%. The proposed novel technology of hydrogen storage, combustible borohydride/metal/water mixtures, makes it possible to develop hydrogen fuel cell power systems with high specic energy, high power density, no catalyst and safe reaction products. Power systems based on this technology can be used as chargers for various electronic devices, and have the potential to dramatically increase their portability. Larger-scale hydrogen generators can be used for emergency power supplies and in some specic applications, such as power for spacecraft and undersea vehicles. Note that the Apollo 13 accident was caused by explosion of a liquid oxygen tank, which was used to feed a fuel cell power supply of the spacecraft. The use of chemical oxygen (also using heterogeneous combustion)36 and hydrogen generators instead of cryogenic liquids in fuel cell power systems would eliminate such accidents, while providing comparable oxygen and hydrogen storage densities. In addition, the combustion heat release can be converted to electricity (for example, using turbine cycles or thermoelectric effect). This is of particular interest for apparatus with limited energy sources, such as spacecraft in deep space or autonomous weather stations and buoys in polar regions.37
2880 November 2005

to be done, however, to advance this concept from laboratory experiments and estimates to industrial technology. Experimental and modeling methods developed in studies on heterogeneous combustion of metals, coal, propellants and other energetic materials will be useful in this regard. It should be noted that combustion of light metals (such as Al and Mg) in gaseous oxidizers (including air and water vapor) was studied extensively for propulsion applications,49 see also next section. Similarly, combustion of porous carbon particles in air and oxygen was investigated widely for power generation.50 This experience could be effectively applied for oxidation and reduction reactions (35) involving oxygen carrier particles. Thus, heterogeneous combustion approaches can play an important role in solving carbon sequestration problems.
Figure 4. Chemical looping combustion.

MeO CH4 3 Me CO2 H2O.

(3)

Extraterrestrial Production of Materials and Propellants

Space exploration opens new opportunities for chemical engineers. In the near future, missions to the Moon and Mars will require propellants and structural materials to support extraterrestrial activities there and to return crews and/or samples to Earth. In situ resource utilization (ISRU) would be economically attractive as compared to transportation of propellants and materials from earth. For example, a human mission to Mars, designed by NASA in response to the Space Exploration Initiative (1989), and relying completely on earth resources, required $450 billion, while NASAs mission design using ISRU (1994) decreased the price tag to $55 billion.51 The new U.S. space policy, Vision for Space Exploration, includes ISRU as a core component:In situ resource utilization will enable the affordable establishment of extraterrestrial exploration and operations by minimizing the materials carried from earth, and by developing advanced, autonomous devices to optimize the benets of available in situ resources.52 Heterogeneous combustion can be used in space for both production of materials and propulsion. First, CS has the potential to produce materials and net-shape articles in space. Indeed, an important advantage of this method is low external energy consumption, which is required only to initiate the self-sustained combustion process. This makes CS especially attractive for use in space missions, where energy availability is limited. One potential application of CS in space is fabrication of replacement parts in long-duration exploration missions. It would be desirable to make spares of malfunctioned items aboard the spacecraft, when required, rather than carry a reserve of spares for all parts. The CS method makes it possible to fabricate replacement parts and custom tools on-site with either in situ materials or reactant powders carried from earth. They can be made to near-net shape, with tight tolerances, reliably, quickly, and safely. Furthermore, CS can be used to join different types of materials by compositional and functional grading.53 Studies in drop towers, research aircraft and space stations have demonstrated that a majority of CS technologies developed in normal gravity can be successfully adopted under microgravity conditions. In fact, many materials produced in microgravity possess superior properties as compared to those synthesized on earth.54 Besides materials synthesis, another promising application of heterogeneous combustion in space is related to propulsion.
Vol. 51, No. 11 2881

An almost pure CO2 stream can be obtained after cooling the exit gas and condensing H2O. The particles of reduced carrier (Me) are transferred to the second reactor and oxidized with air, regenerating the metal oxide Me O2 3 MeO. (4)

This reactor gives a ue gas containing only N2 and unreacted O2. The oxidized carrier (MeO) is then returned to the rst reactor for a new cycle. CLC can also be used in power plants based on coal gasication.42 In this case, synthesis gas produced from the gasier reacts with metal oxide MeO CO H2 3 Me CO2 H2O, (5)

while the other steps are the same as described earlier. The total amount of heat evolved in CLC is the same as for conventional combustion, where the oxygen is in direct contact with the fuel (e.g., CH4). The advantage of CLC is that CO2 is not diluted with N2 but obtained in a separate stream and can be easily recovered by simple steam condensation. The necessity to use the same oxygen carrier in many cycles is a critical problem for CLC technologies. Iron, nickel, cobalt, copper, manganese and cadmium (and their corresponding oxides) have been considered as possible candidates for CLC process, and iron and nickel have been shown as most promising. The nickel-based oxygen carriers withstand a higher operating temperature, while iron has lower cost and its oxides are environmentally friendly materials. A suitable carrier for CLC exhibits high reaction rate and conversion, coking resistance, cycle durability and high mechanical strength. For this reason, the metal oxides are combined with an inert (such as Al2O3, TiO2, ZrO2, SiO2, MgO), which acts as a porous support providing high surface area for reaction, and as a binder for increased mechanical strength and attrition resistance. Methods for preparation of oxygen carriers have been developed and the obtained composite particles tested using TGA and experimental reactors, which identied suitable compositions and conrmed the potential of CLC. Much work remains
AIChE Journal November 2005

Specically, metal fuel could be burnt in propulsion systems with CO2 of Mars and Venus instead of propellants brought from Earth. To demonstrate the importance of this idea for Mars exploration, let us rst briey review the history of Mars ISRU concept, which involves much chemical engineering. Most Mars ISRU suggestions are based on use of martian atmosphere which contains 95% CO2. One of the rst ideas was to produce methane and oxygen (which constitute an excellent rocket bipropellant) from martian CO2 and water in a small chemical processing plant.55 To obtain water on Mars, however, is much more difcult than CO2. For this reason, in a later modication of this approach, a small feedstock of liquid H2 is transported along with the chemical plant from Earth to Mars for use instead of water (Sabatier reaction)56 CO2 4 H2 3 CH4 2 H2O. (6)

The attractive features of this concept are the relatively simple technology and the high performance of the produced CH4/O2 bipropellant. The long-term storage and deep space transport of liquid H2 is, however, a signicant problem. This drawback is eliminated in another methodology, which postulates production of CO and O2 directly by electrolysis of the Martian CO2 in a zirconia cell.57,58 The resulting CO and O2 are then used as a liquid bipropellant. The zirconia electrolyzers, however, are fragile and require high operating temperature (1,000 K). The common problem in all Mars ISRU scenarios mentioned previously is the signicant power required to produce, liquefy, and store cryogenic propellants. An alternative, combustion-based approach in the Mars ISRU suggests burning CO2 directly as an oxidizer with metals or metal hydrides as fuel in a rocket or jet engine.59,60 For rocket applications, CO2 should be collected and liqueed, without any further chemical processing, while the fuel could be either delivered from earth or produced on mars. Carbon dioxide is not a typical oxidizing agent in industrial practice, but there exist fuels (e.g., magnesium and aluminum) which can burn with CO2. Analysis of thermodynamic performance characteristics, combustion parameters, and other properties for various candidate fuels (Li, Be, B, Mg, Al, Si, Ca, Ti, Zr, their hydrides and mixtures with hydrogen compounds) has led to the conclusion that Mg is the most promising for rocket engines using CO2 as an oxidizer.60,61 Magnesium particles ignite in CO2 environment at relatively low temperature (1,000 K) and exhibit vigorous vapor-phase combustion62-64 (see Figure 5). Because of use of in situ oxidizer, Mg-CO2 rocket engines provide signicant advantages in missions with ballistic ights on Mars, even with Earth-imported Mg fuel.65 To improve metal-CO2 engine performance, aluminum can be used instead of magnesium if a method is developed to decrease its ignition temperature in CO2 (2,300 K). One such possibility is to coat Al particles with a thin layer of nickel. Recent studies on combustion of Ni-clad Al particles have shown that they ignite at much lower temperature (1,000 K), which is associated with the intermetallic reactions between Ni and Al on the particle surface.66,67 This brief review clearly demonstrates that novel combustion-based processes have great potential for applications under extraterrestrial conditions (microgravity, atmosphere of Mars, etc.). They will facilitate the production of materials and power in future space exploration missions.
2882 November 2005

Figure 5. Spherical ame around Mg particle (50 63 m) burning in CO2 (from Legrand et al.63).

Concluding Remarks
In this article, we focused attention on four specic topics of signicant contemporary interest to society: combustion synthesis of advanced materials, hydrogen production for fuel cells, carbon sequestration, and in-situ resource utilization in extraterrestrial environments. On the basis of our expertise and because of space limitations, we did not consider other important areas, such as catalytic combustion, soot formation, droplets and spray combustion, gaseous ame synthesis, coal and explosives. All of the earlier are seemingly unrelated topics, yet each includes heterogeneous combustion as the central theme. We hope that this article will attract more chemical engineers to this fascinating area, which is rich both in fundamental concepts and important applications. Furthermore, progress in this eld will arise from a deeper understanding of the combustion mechanisms, and will be aided by development of novel experimental techniques capable of detecting phenomena at shorter time and length scales, along with detailed theoretical and simulation approaches that quantitatively describe the experimental observations. The prospects of new opportunities for chemical engineers in heterogeneous combustion are indeed glowingly bright.

Acknowledgments
We thank NSF, NASA and ACS-PRF for long-term support of our research in heterogeneous combustion (current grants, NSF: CTS-0446529; NASA: NNCO4AA36A; ACS-PRF: 36711-AC9).

Literature Cited
1. Mohanty KK. The near-term energy challenge. AIChE J. 2003;49:2454-2460.
Vol. 51, No. 11 AIChE Journal

2. Varma A, Rogachev AS, Mukasyan AS, Hwang S. Combustion synthesis of advanced materials. Advances Chem Eng. 1998;24:79-224. 3. Patil KC, Aruna ST, Mimani T. Combustion synthesis: an update. Curr Opin Solid State Mater Sci. 2002;6:507-512. 4. Merzhanov AG, Borovinskaya IP. Self-propagating hightemperature synthesis of refractory inorganic compounds. Doklady Chemistry. 1972;204:429-432. 5. Varma A, Li B, Mukasyan A. Novel synthesis of orthopaedic implant materials. Advanced Eng Mater. 2002;4: 482-486. 6. Castillo M, Ayers R A, Zhang X, Schowengerdt F, Moore JJ. Combustion synthesis of porous glasses and ceramics for bone repair. Biomed Sci Instrum. 2001;37:469-474. 7. Mallard E, Le Chatelier A. Combustion des me langes gazeux explosives. Annales de Mines. 1883;4:274-381. 8. Varma A, Rogachev AS, Mukasyan AS, Hwang S. Mechanisms of combustion wave propagation in heterogeneous reaction systems. Combust Flame.1998;115:354-363. 9. Varma A, Rogachev AS, Mukasyan AS, Hwang S. Complex behavior of self-propagating reaction waves in heterogeneous media. Proc Nat Acad Sci USA. 1998;95:1105311058. 10. Mukasyan AS, Rogachev AS, Mercedes M, Varma A. Microstructural correlations between reaction medium and combustion wave propagation in heterogeneous systems. Chem Eng Sci. 2004;59:5099-5105. 11. Aldushin AP, Khaikin BI. Combustion of mixtures forming condensed reaction products. Combust Explos Shock Waves. 1974;10:273-283. 12. Mukasyan AS, Rogachev AS, Varma A. Mechanisms of reaction wave propagation during combustion synthesis of advanced materials. Chem Eng Sci. 1999;54:3357-3367. 13. Hwang S, Mukasyan AS, Rogachev AS, Varma A. Combustion wave microstructure in gas-solid reaction systems: experiments and theory. Combust Sci Tech. 1997;123:165184. 14. Varma A, Mukasyan AS, Hwang S. Dynamics of selfpropagating reactions in heterogeneous media: experiments and model. Chem Eng Sci. 2001;56:1459-1466. 15. Bayliss A, Matkowsky BJ, Aldushin AP. Dynamics of hot-spots in solid fuel combustion. Physica D. 2002;166: 104-130. 16. Grinchuk PS, Rabinovich OS. Effect of random internal structure on combustion of binary powder mixtures. Phys Rev. 2005;E71:026116:1-7. 17. Beck JM, Volpert VA. Nonlinear dynamics in a simple model of solid ame microstructure. Physica D. 2003;182: 86-102. 18. Beck JM, Volpert VA. A simple model of two-dimensional solid ame microstructure. Combust Theory Modeling. 2003;7:795-812. 19. Prakash AS, Khadar AMA, Patil KC, Hegde MS. Hexamethylenetetramine: a new fuel for solution combustion synthesis of complex metal oxides. J Mater Synthesis Processing. 2002;10:135-141. 20. Aruna ST, Rajam KS. Mixture of fuels approach for the solution combustion synthesis of Al2O3-ZrO2 nanocomposite. Mater Res Bull. 2004;39:157-167. 21. Jung C-H, Jalota S, Bhadiri SB. Quantitative effects of fuel on the synthesis of Ni/NiO particles using a microwaveAIChE Journal November 2005

22.

23.

24.

25.

26. 27.

28.

29.

30. 31. 32. 33.

34.

35.

36.

37.

38.

39.

induced solution combustion synthesis in air atmosphere. Mater Lett. 2005;59:2426-2432. Deshpande K, Mukasyan A, Varma A. Direct synthesis of iron oxide nanopowders by the combustion approach: reaction mechanism and properties. Chem Mater. 2004;16: 4896-4904. Erri P, Pranda P, Varma A. Oxidizer-fuel interactions in aqueous combustion synthesis. 1. Iron(III) nitrate-model fuels. Ind Eng Chem Res. 2004;43:3092-3096. Mukasyan AS, Costello C, Sherlock KP, Lafarga D, Varma A. Perovskite membranes by aqueous combustion synthesis: synthesis and properties. Sep Purif Technol. 2001;25:117-126. Agrawal R, Offutt M, Ramage MP. Hydrogen economy an opportunity for chemical engineers? AIChE J. 2005;51: 1582-1589. Schlapbach L, Zuttel A. Hydrogen-storage materials for mobile applications. Nature. 2001;414:353-358. Schlesinger HI, Brown HC, Finholt AE, Gilbreath, JR, Hoekstra HR, Hyde EK. Sodium borohydride, its hydrolysis and its use as a reducing agent and in the generation of hydrogen. J Am Chem Soc. 1953;75:215-219. Amendola SC, Sharp-Goldman SL, Janjua MS, Spencer NC, Kelly MT, Petillo PJ, Binder M. A safe, portable, hydrogen gas generator using aqueous borohydride solution and Ru catalyst. Int J Hydrogen Energy. 2000;25:969975. Kim J-H, Lee H, Han S-C, Kim, H-S, Song M-S, Lee J-Y. Production of hydrogen from sodium borohydride in alkaline solution: development of catalyst with high performance. Int J Hydrogen Energy. 2004;29:263-267. Larminie J, Dicks A. Fuel Cell Systems Explained. West Sussex: Chichester, UK: Wiley; 2003. Desgardin N, Perut C, Renouard J, U.S. Patent Application 20040065395, European Patent 1405823; 2004. Desgardin N, Perut C, Renouard J. U.S. Patent Application 20040065865, European Patent 1405824; 2004. Delapierre G, Laurent J-Y, Priem T, Bloch D, Marsacq D, Perut C, Gauthier C. U.S. Patent Application 2004067396, European Patent 1344266; 2004. Ivanov VG, Gavrilyuk OV, Glazkov OV, Safronov MN. Specic features of the reaction between ultrane aluminum and water in a combustion regime. Combust Explos Shock Waves. 2000;36:213-219. Sharovich E, Diakov V, Varma A. Combustion of novel chemical mixtures for hydrogen generation. Combust Flame, in press. Sharovich E, Mukasyan AS, Varma A, Kshirsagar G, Zhang Y, Cannon JC. Mechanism of combustion in lowexothermic mixtures of sodium chlorate and metal fuel. Combust Flame. 2002;128:133-144. Weinberg F. Optimizing heat recirculating combustion systems for thermoelectric converters. Combust Flame. 2004;138:401-403. Carbon Sequestration. Technology Roadmap and Program Plan 2005. US DOE, Ofce of Fossil Energy, National Energy Technology Laboratory. May 2005. Available at: http://www.fe.doe.gov/programs/sequestration/publications/ programplans/2005/sequestration_roadmap_2005.pdf. Accessed August 12, 2005. Richter HJ, Knoche K. Reversibility of combustion pro2883

Vol. 51, No. 11

40.

41.

42. 43.

44.

45.

46.

47.

48. 49.

50.

51. 52.

53.

cesses. In: RA Gaggioli, ed. Efciency and Costing. Second Law Analysis of Processes. ACS Symposium Series 235. Washington DC: ACS, 1983:71-85. Ishida M, Jin H. A new advanced power-generation system using chemical-looping combustion. Energy. 1994;19:415422. Ishida M, Jin H. A novel chemical-looping combustor without NOx formation. Ind Eng Chem Res. 1996;35:24692472. Jin H, Ishida M. A new type of coal gas fueled chemicallooping combustion. Fuel. 2004;83:2411-2417. Lyngfelt A, Leckner B, Mattisson T. A uidized-bed combustion process with inherent CO2 separation; application of chemical-looping combustion. Chem Eng Science. 2001;56:3101-3113. Mattisson T, Johansson M, Lyngfelt A. Multicycle reduction and oxidation of different types of iron oxide particles - Application for chemical-looping combustion. Energy Fuels. 2004;18:628-637. Wolf J, Anheden M, Yan J. Comparison of nickel- and iron-based oxygen carriers in chemical looping combustion for CO2 capture in power generation. Fuel. 2005;84: 993-1006. Garc a-Labiano F, de Diego LF, Ada nez J, Abad A, Gaya n P. Temperature variations in the oxygen carrier particles during their reduction and oxidation in a chemical-looping combustion system. Chem Eng Sci. 2005;60:851-862. Yu J, Corripio AB, Harrison DP, Copeland RJ. Analysis of the sorbent energy transfer system (SETS) for power generation and CO2 capture. Advances in Environ Res. 2003; 7:335-345. Contractor RM. Duponts CFB technology for maleic anhydride. Chem Eng Sci. 1999;54:5627-5632. Yetter RA, Dryer FL. Metal particle combustion and classication. In: HD Ross, ed. Microgravity Combustion : Fire in Free Fall. San Diego: Academic Press, 2001:419478. Williams A, Pourkashanian M, Jones JM, Skorupska N. Combustion and Gasication of Coal. New York: Taylor & Francis; 2000. Frankie BM, Zubrin R. Chemical engineering in extraterrestrial environments. Chem Eng Progress. 1999;2:45-54. Vision for Space Exploration. NASA, 2004. Available at: http://exploration.nasa.gov/documents/documents.html#vision. Accessed August 12, 2005. Miyamoto Y, Kaysser WA, Rabin BH, Kawasaki A, Ford

54.

55.

56. 57.

58.

59. 60.

61.

62.

63.

64.

65. 66.

67.

RG. Functionally Graded Materials: Design, Processing and Applications. Boston: Kluwer Academic Pub; 1999. Mukasyan A, Lau C, Varma A. Inuence of gravity on combustion synthesis of advanced materials. AIAA J. 2005;43:225-245. Ash RL, Dowler WL, Varsi G. Feasibility of rocket propellant production on mars. Acta Astronautica. 1978;5: 705-724. Zubrin RM, Baker D. Humans to Mars in 1999. Aerospace America. 1990;28(8):30-32. Ramohalli K, Lawton E, Ash R. Recent concepts in missions to Mars - extraterrestrial processing. J Propulsion Power. 1989;5:181-187. Sridhar KR. Mars sample return mission with in-situ resource utilization. J Propulsion Power. 1995;11:13561362. Yuasa S, Izoda H. Carbon dioxide breathing propulsion for a Mars airplane. AIAA 89-2863, 1989. Sharovich EYa, Shiryaev AA, Goldshleger UI. Magnesium and carbon dioxide: a rocket propellant for Mars missions. J Propulsion Power. 1993;9:197-203. Sharovich EYa, Goldshleger UI. Comparison of potential fuels for Martian rockets using CO2. J Propulsion Power. 1997;13:395-397. Valov AE, Gusachenko EI, Shevtsov VI. The effect of CO2 pressure and concentration on the ignition of single Mg particles in CO2-Ar mixtures. Combust Explos Shock Waves. 1992;28:7-10. Legrand B, Sharovich E, Marion M, Chauveau C, Go kalp I. Ignition and combustion of levitated magnesium particles in carbon dioxide. Proc Combustion Institute. 1998; 27:2413-2419. Abbud-Madrid A, Modak A, Branch MC, Daily JW. Combustion of magnesium with carbon dioxide and carbon monoxide at low gravity. J Propulsion Power. 2001;17: 852-859 Sharovich E, Salomon M, Go kalp I. Mars hopper versus Mars rover. Acta Astronautica, in press. Mukasyan AS, Lau C, Varma A. Gasless combustion of aluminum particles clad by nickel. Combust Sci Technol. 2001;170:67-85. Sharovich E, Mukasyan A, Thiers L, Varma A, Legrand B, Chauveau C, Go kalp I. Ignition and combustion of Al particles clad by Ni. Combust Sci Technol. 2002;174(3): 125-140.

2884

November 2005

Vol. 51, No. 11

AIChE Journal

FLUID MECHANICS AND TRANSPORT PHENOMENA

Effect of Taylor Vortices on Mass Transfer from a Rotating Cylinder

R. Srinivasan, S. Jayanti and A. Kannan
Dept. of Chemical Engineering, IIT-Madras, Chennai 600 036, India DOI 10.1002/aic.10553 Published online August 8, 2005 in Wiley InterScience (www.interscience.wiley.com).

Mass transfer from solids, which has important applications in a number of chemical and pharmaceutical industries, has been studied experimentally and semiempirically under turbulent ow conditions, and correlations are available in the literature to calculate the mass-transfer coefcients from pellets, rotating cylinders and disks etc. However, mass transfer under laminar ow has not been sufciently addressed. One of the difculties here is the strong Reynolds number dependence of the ow pattern, for example, due to the onset of Taylor vortices for the case of a rotating cylinder. This problem is circumvented by using a computational uid dynamics (CFD)-based solution of the governing equations for the case of a cylinder rotating inside a stationary cylindrical outer vessel lled with liquid. The parameters cover a range of Reynolds number (based on the cylinder diameter, and the tangential speed of the cylinder), Schmidt number and the ratio of the outer to inner cylinder diameters. The results conrm that the circumferential velocity prole is a strong function of the Reynolds number and varies from a nearly Couette-type ow at very low Reynolds numbers to a boundary layer-like prole at high Reynolds numbers. The onset of Taylor vortices has a strong effect on the ow eld and the mass-transfer mode. The calculations show that the Sherwood number has a linear dependence on the Reynolds number in the Couette-ow regime, and roughly square-root dependence after the onset of Taylor vortices. Correlations have been proposed to calculate the Sherwood number taking account of these effects. 2005 American Institute of Chemical Engineers AIChE J, 51: 28852898, 2005 Keywords: CFD, solid dissolution, mass transfer, Taylor vortices, rotating cylinder, laminar ow

Introduction
Dissolution of solid materials in liquids nds wide-spread applications in the chemical process industries. Knowledge of the mass-transfer rates between a solid and an agitated liquid in a vessel is of considerable importance in the design of units, such as crystallizers, stirred-tank reactors for diffusion controlled soliduid reactions, and units involving physical dissolution. Some industrial examples of solid dissolution process are:

Correspondence concerning this article should be addressed to S. Jayanti at sjayanti@iitm.ac.in.

2005 American Institute of Chemical Engineers

Dissolution of metals, such as tin, lead and zinc in heat exchange media, such as mercury and sodium at high-temperatures, and their deposition at lower temperatures.1 Limestone dissolution in an aqueous solution which is a phenomenon of interest in liming of acid lakes, ue gas desulfurization. The sustained drug delivery process in pharmaceutical applications which is determined by the rate of solid dissolution in the liquid medium. Mass transfer from a rotating cylinder which is of practical importance in electrochemical processes, such as electrodeposition, coating and metal recovery.2 Recent application of mass transfer from rotating cylinVol. 51, No. 11 2885

AIChE Journal

November 2005

ders in ltration and reverse osmosis processes where Taylor vortices (see later) are induced to enhance the mass-transfer rates. Dissolution of solids has been extensively studied from both theoretical and experimental fronts. Three idealized congurations of the solid, namely, suspended pellets, rotating disk and rotating cylinder, have been considered in the literature. Eisenberg et al.3 reported on one of the rst systematic studies of solid dissolution from a rotating cylinder over a wide range of Reynolds and Schmidt numbers. Johnson and Huang4 studied the rates of dissolution of organic solid from a at surface into a turbulent liquid in an agitated vessel. Bennett and Lewis1 made a comparison between the dissolution of lead and tin in mercury with the dissolution of benzoic acid in water, and found that the metal dissolution rates were transport controlled. Sherwood and Ryan5 studied experimentally the mass transfer with and without chemical reaction from rotating benzoic acid cylinders. Marangozis and Johnson6 reviewed solid dissolution data of suspended pellets, at disks and rotating cylinders, and recommended cube-root dependency of the Sherwood number on the Schmidt number. Studies on solid-liquid mass transfer have been reviewed extensively by Pangarkar et al.7 There have been a number of more recent studies of mass transfer from a rotating cylinder with and without axial ow.8,9,10,11,12,13,14,15 The congurations studied include cases with inner cylinder alone rotating and outer cylinder alone rotating, and are spread over large ranges of Reynolds and Taylor and Schmidt numbers. Most of these have been experimental studies and do not include detailed measurement of the ow eld. An important aspect of mass transfer from a rotating cylinder is the possibility of occurrence of Taylor vortices.16,17 The rotational motion of the inner cylinder (Taylor vortices do not occur when the outer cylinder alone rotates) induces a primarily circumferential ow in the annular region between the inner and the outer cylinders. If the rotational speed is sufciently high (but not too high), then large-scale secondary ows, typically of the size of the gap width as shown in Figure 1a, are formed. It is to be expected that the presence of these Taylor vortices has a strong effect on the mass-transfer rate from the cylinder. While this ow situation has been widely investigated for very small gap widths in connection with the stability of the ow, there have been relatively few studies of the mass transfer, and also of the ow eld for large gap widths typically used in mass transfer applications. In a recent study, Baier et al.15 used computational uid dynamics (CFD) simulations to obtain the velocity eld for the case of the inner cylinder rotating, and used a boundary layer approximation to study the mass transfer aspects for small values of Taylor numbers. They found that the mass-transfer coefcient in the presence of the Taylor vortices varied as Sh Re 0.46 , which agreed with the experimental results of Coeuret and Legrand,11 Holeschovsky and Cooney14 who studied for the case of no axial ow and inner cylinder rotating, and also with the results of Kataoka et al.9 who investigated the case of the outer cylinder rotating. Earlier, Kawase and Ulbrecht13 used an analogy with the secondary ow in helical coils to arrive at the same result. Experiments at high Taylor numbers3,12 show a stronger dependence of the Sherwood number on the Reynolds number with the exponent being in the range of 0.6 to 0.72. Thus, there is considerable literature on the mass transfer from rotating cylinders. However, a detailed description of the
2886 November 2005

ow eld within the annular region is absent in these studies, presumably due to the nontrivial dependence of the velocity eld on the Reynolds number and the radii of the inner and outer cylinders, among other parameters. Baier et al.15 who included some of these effects in their computation of the velocity eld, used a boundary layer analogy for mass transfer. Thus, their study is not applicable for very low Taylor numbers. In this article, we take advantage of the accuracyfor laminar ow simulations offered by computational uid dynamics (CFD) techniques to simulate the complete problem, that is, calculate both the velocity eld and the scalar concentration eld, for a range of ow conditions including very low Taylor numbers where Taylor vortices are not present. The simulations cover a range of Reynolds numbers, diameter ratios and height to diameter ratios of the cylinders. Results from these simulations show clearly the marked change in the masstransfer coefcient upon the onset of Taylor vortices. Details of the calculation methodology, and the results obtained are discussed later.

Problem Formulation and Calculation Methodology Governing equations for uid ow

The equations solved for a CFD solution are the continuity and the momentum equation, which can be written as follows using vector notation: Continuity equation: for an incompressible Newtonian uid u0 Momentum equation 1 u uu p 2 u t (1)

(2)

Here u is the velocity; p is the static pressure, is the density, and is the kinematic viscosity. Since the uid under consideration is a liquid and the velocities are small, a constantproperty uid assumption is made. The mass concentration of the dissolving substance is calculated by solving the following generalized scalar transport equation 1 Y u YA AB YA t A

(3)

where Y A is the mass fraction of the scalar A , that is, the mass of the scalar per total mass in the control volume. AB is the molecular diffusivity (in kg/ms) of the scalar in the liquid B , dened as AB D AB , where D AB is the kinematic molecular diffusivity in m2/s. This approach has previously been used by one of the authors,18 to calculate the mixing time in jet-mixed vessels. The scalar is assumed to be a neutrally buoyant scalar so that the density and the viscosity of the uid are unaffected by the scalar concentration. Since benzoic acid, the solid under consideration, is only sparingly soluble in water, the uid under consideration, this is a reasonable assumption.
Vol. 51, No. 11 AIChE Journal

Figure 1. (a) Taylor vortices in the annular gap width of inner cylinder rotating and outer stationary, (b) geometry of the computational domain, and (c) boundary conditions.

Flow domain and boundary conditions

The ow domain consists of an inner rotating cylinder and an outer stationary cylindrical vessel lled with a liquid up to a given height. The inner solid cylinder is concentric with the axis of the vessel and rotates at a constant angular velocity. There are no internal parts, such as bafes in the vessel, and the problem is treated as being axisymmetric for the low Taylor numbers studied in this work. The ow domain consists of the liquid in the annular region between the inner rotating cylinder and the outer stationary cylinder, and is bounded at the bottom by a stationary wall and at the top by a planar shear-free interface, as shown in Figures 1b and 1c. The problem is formulated in cylindrical coordinates, the axial, radial and tangential velocity components being u , , and w , respectively. No-slip boundary conditions are imposed on all walls. Thus, a constant tangential velocity is specied at the inner cylinder,
AIChE Journal November 2005

and a zero velocity is specied on the outer wall and the bottom wall. A symmetry boundary condition is specied on the top wall to make it a zero-shear surface. Since the major ow is in the circumferential direction, all the three velocity components have to be resolved. For this reason, the ow in a 45 sector of the cylinder is simulated, and periodicity condition is imposed on the two bounding circumferential planes. For the mass fraction equation, a constant wall mass fraction, which is based on the solubility limit of benzoic acid in water, estimated to be 4.067 103 at a temperature of 30C, is applied on the surface of the inner cylinder and a Neumann boundary condition (zero normal gradient) is applied on the other boundaries. The solution of the coupled mass, momentum and masstransfer equations has been done in two stages. In the rst stage, the mass and the momentum equations are solved together under steady conditions to obtain the steady-state veVol. 51, No. 11 2887

locity and pressure elds. In the second stage, the time-dependent mass fraction equation, which is coupled in a one-way mode to the mass and momentum equations, is solved with the velocity eld obtained from the steady-state calculations as the initial condition for the velocity eld. The mass fraction in the interior of the domain at the beginning of the calculation is assumed to be zero throughout.

Formulation of the mass transfer problem

The problem under consideration is the determination of the mass-transfer coefcient from a rotating cylinder made of the solid to be dissolved. The specic case being studied is that of the dissolution of benzoic acid in water in which the former is sparingly soluble. Experimentally this is done by measuring the concentration of the benzoic acid in the solution as a function of time. It can be shown4 that the cup mean concentration with time is given by the following expression ln 1

Details of Numerical Solution Computer code

All the calculations reported here have been carried out using the commercially available CFX 4.4 computer code developed by AEA Technology, UK. CFX is a general purpose computer program, which uses a nite volume method-based discretization of the governing partial-differential equations on a nonstaggered, structured, body-tted grid. The chequer-board type of oscillations of pressure and velocity that are associated with the use of a nonstaggered grid are eliminated using the Rhie-Chow interpolation scheme19 to estimate the face velocities. The pressure-velocity coupling for incompressible ows is effected using the SIMPLE family of schemes20 adapted for a nonstaggered body-tted grid.

CA kc S/Vt CAW

(4)

Discretization schemes
The computer program has different discretization schemes for the user to choose from for a particular problem. In the present case, the convection terms in the governing ow equations have been discretized using the third-order accurate QUICK scheme,21 while the diffusion terms have been discretized, using the second-order accurate central scheme. For the solution of the mass-fraction equation, it was found that using the QUICK scheme gave negative values of the mass fraction; hence, the SUPERBEE scheme,22 which incorporates the total variation diminishing (TVD) property to eliminate oscillations associated with high order differencing schemes, has been used. The overall accuracy of the discretization can, therefore, be said to be of second-order.

Grids
The structured, orthogonal grid for the two-dimensional (2-D) ow domain consisted typically of 9,600 cells. For example, in the case with a cylinder height of 0.12 m, and a gap width of 0.035 m, a 100 96 nonuniform grid was used to discretize the 2-D ow domain in the x - r plane, x being axial direction, and r being radial direction. Since the precise location of the Taylor vortices was not known a priori, uniform grid spacing was used in the axial direction. In order to resolve the strong velocity and mass fraction gradients in the radial direction, an expanding grid was used. Preliminary calculations were made with different grids to check for grid independence of the results. The computed velocity prole near the inner cylinder was compared for three radial grids in which the grid spacing near the wall was successively reduced by a factor of two. Similarly, the grid independence in the axial direction was also veried by repeating the calculations with a grid spacing which is halved. The temporal discretization was also veried by calculating the mass-fraction equation with a series of decreasing time steps.
2888 November 2005

where k c is the mass-transfer coefcient, S is the surface area of the inner cylinder, V is the volume of the liquid, C A is the bulk mean concentration at time t , C AW is the solubility concentration of the solid in liquid. The mass-transfer coefcient is obtained from the slope by plotting LHS vs. time. In the present calculations, this is simulated by the numerical solution of the time-dependent transport equation of a passive, neutrally buoyant scalar (Eq. 3) with the appropriate diffusivity value and subject to initial and boundary conditions. (It is assumed that the dissolution rate is small enough that one does not have to consider the radial movement of the cylinder surface due to dissolution.) The mass-transfer coefcient can now be obtained in one of two ways: (1) by plotting the temporal variation of the cup mean concentration in the vessel, which gives the timeaveraged mass-transfer coefcient, and (2) from the slope at the wall of the radial prole of the instantaneous mass fraction. It was found that the two gave nearly the same results; however, the computed temporal variation was not exactly linear, especially in the initial stages of the calculation. Hence, the second method was followed to determine the mass-transfer coefcient. Here, two additional complications arise in the evaluation of the mass-transfer coefcient. First, the masstransfer rate is a function of the local velocity eld and, therefore, shows axial variation when Taylor vortices are present. An axial average of the computed wall mass ux was, therefore, taken to smooth out these variations. Second, in many cases, especially at low rotational speeds, the massfraction prole did not have a clearly dened bulk concentration. Therefore a cup mean mass fraction, dened as follows

Yc
0 Ri

L R0

w r , x Y r , x 2 rdrdx (5) w r , x 2 rdrdx

0 Ri

L R0

was used as the bulk mass fraction. The mass-transfer coefcient was then calculated as kc
Vol. 51, No. 11

jA Y * Y c

(6)
AIChE Journal

where j A is the axially-averaged mass ux at the inner cylinder at a given time wall which is directly obtained from the computer program, Y * is the wall mass fraction obtained from the solubility concentration of the solid in liquid and Y c is the computed cup mean mass fraction in the vessel at that time. This is then converted into a Sherwood number based on hydraulic diameter as follows Sh Dh where d is the gap width. k c 2 d D AB

(7)

Validation
It can be seen from the description of the earlier calculation methodology that the exact equations governing the ow are being solved. The major assumptions are the ow is laminar, axisymmetric, Newtonian with the uid having constant physical properties, which are reasonable for the problem under consideration. Thus, unlike in turbulent ow calculations, there are no modeling assumptions as far as the equations are concerned. The principal sources of error, therefore, are the discretization of the equations and the boundary conditions. The former can be reduced by employing a ne grid, which has been veried by the grid independence tests. The no-slip boundary conditions on the walls require no validation; the imposition of a shear-free interface condition at the top surface limits the application to low-rotational speeds so that the interface remains at. Finally, the assumption of axisymmetry is also a limiting condition, because it is expected that as the rotational speed increases, the axisymmetric Taylor vortices may become unstable giving rise to a wavy type of Taylor vortices.23 To demonstrate the validity of the simulations, results obtained from the present study are compared with theoretical results for the onset of Taylor vortices. This case is considered by a number of researchers. Among these, DiPrima et al24 considered the effect of the ratio of the inner and outer cylinder radii on the stability of Couette ow to axisymmetric and nonaxisymmetric perturbations for the case where the outer cylinder is at rest. They found that the ratio of the inner to outer radii , had a signicant inuence on the critical Reynolds and Taylor numbers, dened respectively as Re crit R id

Figure 2. Streamfunction contours plotted in the annular region for different Reynolds number (ReD) 25, 60, 86, 127, 203, 456, 760 for the case of radius ratio () of 0.462.

(8)

Ta crit

2 1 Re crit2 1

contours are plotted in the annular region for different Reynolds numbers, based on inner cylinder diameter Re D for the case of a radius ratio of 0.462. It can be seen that until the onset of the instability, no secondary ow is observed in the bulk of the annular region. (Since the bottom surface is a wall, this gives rise to a small amount of recirculation which cannot be associated with the Taylor vortex phenomenon; this recirculation is expected to disappear if the bottom surface too is made a shear-free surface.) Above a certain Reynolds number (or rotational speed for a given geometry), Taylor vortices appear all along the height of the annular region (see discussion on the effect of the height of the cylinder). This pattern is sustained at higher Reynolds numbers. Thus, for each radius ratio, a fairly demarcated Reynolds (or Taylor) number exists below which a Couette type of ow without signicant secondary ow and above which Taylor cells are found over the entire domain. It should be noted that these ow patterns have been obtained by solving the steady state governing equations and, hence, they do not actually mimic the transition. The transitional Taylor numbers obtained from the present CFD simulations are compared in Figure 3, with the theoretical results obtained by DiPrima et al24 for the three radius ratios. It can be seen that excellent agreement is obtained between the two.

(9)

Results from Numerical Simulations

Calculations of the solid dissolution process in a rotating cylinder have been made for several cases to investigate the effect of Reynolds number, Schmidt number and radius ratio. The diameter of the inner cylinder ( D i ) is kept constant at 0.03 m in all cases, and three different outer diameters ( D o ) 0.0475, 0.065 and 0.100 m were used to obtain radius ratios () of 0.632, 0.462 and 0.300, respectively. The height of the cylinder ( L ) was kept at 0.12 m in a majority of the cases; but some cases were repeated varied with twice the height to check
Vol. 51, No. 11 2889

for the Couette ow to Taylor vortex ow transition to occur. In this study, this theoretical dependence on the radius ratio is taken as a benchmark for the validation of the present calculation methodology. Accordingly, simulations were conducted for three radius ratios, namely, 0.632, 0.462 and 0.300, over a range of Reynolds numbers (here, the outer radius is changed while keeping the inner cylinder radius constant). Typically, the results are as shown in Figure 2 where the streamfunction
AIChE Journal November 2005

Figure 3. Transitional Taylor numbers obtained from the present CFD simulations compared with the theoretical results obtained by DiPrima et al24 for the three radius ratios ().

that the results were not dependent on the height of the domain. For the mass transfer cases, the base case corresponded to the estimated Schmidt number, Sc , of 877 for a benzoic acid-water system. Additional calculations were conducted by varying Sc in the range of 100 to 50,000 to determine the Schmidt number effect. The Reynolds number, based on the inner cylinder diameter and its surface speed, varied between 1 and 1,000. The results from these calculations are discussed later.

Typical velocity and mass fraction proles

Owing to rotation of the inner cylinder, two distinct patterns of ow arise in the annular region in the range of parameters studied here. The changeover between these can be attributed to the onset of Taylor vortices. On the basis of the nature of the velocity eld, these can be called as Couette ow (shortened as Couette ow) pattern and Taylor vortex (shortened as Taylor ow) ow pattern. Typical circumferential velocity

proles at midheight under these two conditions are shown in Figure 4a, for a radius ratio of 0.462, and a cylinder height to inner diameter of 4. The rotational speed of the inner cylinder is such that the Reynolds number, based on the inner cylinder diameter is 86 for the Couette ow case and 453 for the Taylor ow case. The velocity proles are plotted in dimensionless form in which the circumferential velocity is divided by the surface speed ( R i , where R i is the radius of inner cylinder, and is the angular velocity), and the radial distance from the wall of the inner cylinder is divided by the gap width. It can be seen that under Couette ow conditions, the velocity prole variation is nearly linear while under Taylor ow conditions, a more boundary layer-like velocity prole is obtained near the cylinder wall. The mass fractions corresponding to these two cases, for a Schmidt number of 877, are shown in Figure 4b (at a time corresponding to the same number of revolutions of the inner cylinder) in dimensionless form, in which the mass fraction is divided by the imposed mass fraction at the wall. Once again, a distinct, boundary layer type of variation is found under Taylor ow conditions while a nearly linear variation prevails under Couette ow conditions. Taylor vortices introduce a convective element into the diffusion of the species from the dissolving wall into the bulk of the uid. However, this convection is highly structured and helps only to some extent in the overall diffusion and dissolution of the substance. The nal process of mixing and dissolution is still governed by molecular diffusion to the dissolving front and is a long process. This is summarized in Figure 5a, wherein the contours of mass fraction of the solute are shown in a single Taylor cell as a function of time for a radius ratio of 0.3, Reynolds number of 760 and Schmidt number of 877. In the initial period, there is rapid transport of the solute from the inner surface to the outer surface and back towards the inner surface, in a manner consistent with the streamline pattern. However, there is very little solute concentration in the eye of the vortex and further mixing of the solute in the core uid

Figure 4. (a) Typical circumferential velocity proles at midheight for a radius ratio of 0.462. Reynolds number (ReD) is 86 for the Couette ow case, and 453 for the Taylor ow case. (b) Typical mass fraction proles at midheight for a radius ratio of 0.462. Reynolds number (ReD) is 86 for the Couette ow case, and 453 for the Taylor ow case. 2890 November 2005 Vol. 51, No. 11 AIChE Journal

Figure 5. (a) Mass fraction contours within a single Taylor vortex cell, and (b) mass fraction proles at midheight of the Taylor vortex cell at different times (s) for a radius ratio of 0.3, ReD of 760 and Sc of 877.

is governed by the slow molecular diffusion process. This is illustrated in the gradual shrinking with time of the zero concentration region at the center of the vortex. The radial variation of the mass fraction along the centerline of the Taylor cell is shown in Figure 5b at different times for this case. The convective transport of the solute is effective in increasing the solute mass fraction near the inner and the outer walls. However, the concentration in the central part of the channel remains close to zero. Large mass fraction gradients exists between the central core and the well-mixed layer outside in the path of the convection currents. From this, it is clear that there is a relatively thin layer of well-mixed zone, and a large area of the unmixed core associated with each cell. Transport of the solute between these two zones is essentially by molecular diffusion and is, therefore, very slow.

Parametric Effects of Re, , L/D and Sc Effect of Reynolds number on Taylor vortices
The effect of Reynolds number variation on the overall ow pattern in the annular region can be seen in Figure 2, where the streamfunction contours are shown at different Reynolds numbers for a gap width of 0.0175 m. As mentioned earlier, Taylor vortices are formed only after a critical Taylor number. Further increase in the Taylor number leads to some distortion of the Taylor vortices; however, their number remains constant. A comparison of the radial proles of the tangential velocity component at midheight is shown in Figure 6a in dimensionless form. Until the onset of Taylor vortices, all dimensionless proles collapse onto a single curve (Figure 6a shows the velocity proles for Re of 25 and 86 falling on the same, nearly Couette-type prole). As Re increases further, there is a gradual deviation from this, and a boundary layer type of variation
AIChE Journal November 2005

is established at a Reynolds number of about 200. Further increase in Re makes the boundary layer-type variation more pronounced. It should be noted that the plots are all at midheight, and that there is some asymmetry in the Taylor vortex pattern, presumably due to the asymmetric boundary conditions at the top and the bottom of the annular region. Thus, the midheight location does not correspond to the same location relative to a Taylor vortex, and this adds a bit of uncertainty to the comparison of the velocity proles at different Re D . (This uncertainty is not in locating the midheight of the cylinder, but in being able to impose the conditions that the midheight of the cylinder always coincides with the midheight of the Taylor cell for all Reynolds numbers.) However, the transition to a boundary layer type velocity prole at high Re is unmistakable. The mass fraction proles for different Reynolds numbers at a Schmidt number of 877 are shown in Figure 6b. These also show the Couette-type and boundary layer-type variations at low and high Reynolds numbers, respectively.

Effect of radius ratio

The effect of radius ratio on the dimensionless velocity proles is summarized in Figure 7a and Figure 7b. Here the dimensionless tangential velocity is plotted in terms of dimensionless radial distance for three radius ratios, namely, 0.632, 0.462, 0.300 at a Reynolds number of 76 (Figure 7a), which corresponds to the Couette ow condition, and at a Reynolds number of 760 (Figure 7b), which corresponds to the Taylor ow condition. From Figure 7a, we note that, for small gap widths or high values of , the radial prole of the circumferential velocity is nearly linear. As decreases, the velocity prole becomes nonlinear, though monotonically decreasing. In the Taylor ow condition also (Figure 7b) the radius ratio
Vol. 51, No. 11 2891

Figure 6. (a) Circumferential velocity proles at midheight for a radius ratio of 0.462 for different Reynolds number (ReD). (b) Mass fraction proles at midheight for a radius ratio of 0.462 for different Reynolds number (ReD).

has a signicant effect on the tangential velocity prole. It is, therefore, to be expected that the radius ratio will have some parametric inuence on the mass-transfer coefcient. Many existing correlations1,6,25 for mass-transfer coefcient do not consider this factor.

Effect of the Schmidt number

The effect of Schmidt number on the mass fraction proles is investigated by carrying out calculations in the Sc number range of 100 to 50,000. The Sc for the reference case of benzoic acid-water system is 877. It varies between 600 and 1,500 for other typical systems such as salicylic acid-water and 2-naphtol-water systems.25 For metals, such as tin, lead and zinc dissolving in liquid metals, such as mercury, the Schmidt number is in the range1 of 50 to 100, while for systems such as benzoic acid-sucrose solution, and benzoic acid-aqueous glycerol solution, it can vary in the range8 of 3,000 to 40,000. The effect of this large variation of Sc on the mass fraction prole is summarized in Figure 8, where the concentration proles are

shown for different Sc at a Reynolds number of 76 (Figure 8a), and 760 (Figure 8b). The effect of increasing Sc is to decrease the rate of diffusion of the wall-imposed mass fraction. This effect predominates at low Re , and the Schmidt number is seen to have a large effect on the mass fraction prole. At high Re , the large convective diffusion caused by the Taylor vortices restricts the mass fraction variation to close to the walls (Figure 8b (1)). However, on a magnied scale (the proles are redrawn in Figure 8b (2) over a dimensionless radial distance of 0.01), the effect of Sc on the mass fraction proles is nearly the same as at low Re . Thus, the effect of Sc on the mass-transfer coefcient scales independently of Reynolds number, as will be shown later.

Effect of L/D
It is obvious that the height of the cylinder has little effect on the overall mass-transfer coefcient under Couette ow conditions as there is nearly no axial variation of the mass ux. However, the situation is different under Taylor ow condi-

Figure 7. (a) Tangential velocity proles for three different radius ratios () at ReD 76 (Couette ow condition). (b) Tangential velocity proles for three radius ratios at ReD 760 (Taylor ow condition). 2892 November 2005 Vol. 51, No. 11 AIChE Journal

Figure 8. (a) Effect of Sc on the mass fraction prole at a ReD 76 (Couette ow condition). (b) Effect of Sc on the mass fraction prole at a ReD 760 (Taylor ow condition).

tions. Typically, a number of Taylor cells are set up if the cylinder height is long enough. This can be seen in Figure 9a, where the streamfunction contours for a radius ratio of 0.30 are shown at a Reynolds number of 760. It can be seen that a number of Taylor cells, four in this case, are established and that the ow pattern is fairly regular. The circulation within one Taylor cell is shown in Figure 9b in the form of vector plot. The corresponding wall mass ux variation along the height is shown in Figure 9c, for a Schmidt number of 877 for two different heights. (Such strong periodic variation in the local mass-transfer coefcient has also been recorded experimentally by Kataoka et al9). This shows that except close to the bottom wall, a periodic mass ux variation, consistent with the Taylor cell pattern, is obtained on the inner cylinder surface. Thus, for a long enough height, a repeating pattern is obtained. However, if the cylinder height is less, as in the base case of a height of 0.12 m, and an inner cylinder dia. of 0.03 m (both these dimensions correspond to the experimental conditions of Sherwood and Ryan5), it is possible that the L / D ratio has an effect on the Taylor cell pattern. In order to investigate this, some calculations have been for twice the base case cylinder height of 0.12 m. The resulting axial mass ux variation is compared in Figure 8c for a Reynolds number of 760, and a
AIChE Journal November 2005

Schmidt number of 877. It can be seen that the mass ux variation for the shorter cylinder case overlaps that for the longer cylinder case. Calculations with a wider gap show, however, that the predicted pattern may be affected if the gap width is increased by a factor of two while keeping the same height. These calculations show that the minimum height of the cylinder should be so as to allow a couple of Taylor cells to be resolved. Since the cell height is typically about the same as the gap width (see Figure 2), it can be concluded that the effect of liquid height would not be signicant if L / d 2.

Correlation for Mass-Transfer Coefcient

The calculation of the mass concentration prole enables the determination of the mass ux from the surface of the inner cylinder as dYA dr

j A AB

(10)
rRi

The mass-transfer coefcient k c , can then be evaluated as

Vol. 51, No. 11 2893

Figure 9. (a) Streamfunction contours for 0.30 at ReD 760. (b) Circulation within one Taylor cell in the form of vector plot. (c) Axial wall mass ux variation for two different heights at Sc 877 and ReD 760.

kc

jA Y * Y c

(11)

where Y * is the imposed mass fraction at the inner cylinder, and Y c is the mixed cup mean mass fraction dened in Eq. 5. The introduction of Y c , the cup mean value of the mass fraction, is done for two reasons. First, because of the nite size of the annular region, the concentration at large r is not zero and increases with time. Second, Y c is the desired parameter from a practical point of view as one can use it to determine how long the rotating cylinder has to be kept immersed in the solution for a specied amount of dissolution. The presence of Taylor vortices makes the mass ux vary

axially, typically as shown in Figure 9c. An axially averaged value of the mass ux is, therefore, taken to compute the mass-transfer coefcient. This is then converted into a Sherwood number, based on the hydraulic diameter as in Eq. 7. It should be noted that the gap width d here is one half of the hydraulic diameter for the same conguration. Thus, other denitions of Sherwood number, for example, based on the inner cylinder diameter or gap width are possible. The present denition is justied by the fact the Taylor number and the Reynolds number describing the ow are better with hydraulic diameter than on the diameter. Typical variation of the computed Sherwood number with the Reynolds number is plotted in Figure 10 for the three

Figure 10. Sherwood number (ShDh) variation (a) with Reynolds number (ReDh) for three radius ratios (), and (b) with Schmidt number at ReD 76 (Couette ow region) and 760 (Taylor ow region). 2894 November 2005 Vol. 51, No. 11 AIChE Journal

Figure 11. (a) Computed variation of ShDh with ReDh for very low rotational speeds for 0.3. (b) Comparison of numerically obtained Sherwood number limiting values with that of CFD.

different radius ratios under consideration. The three curves more or less merge in the Taylor ow region, but differ significantly in the Couette ow region. Closer examination of the results shows that the variation of Sh with Re is typically linear in the Couette ow region, and that it exhibits a square-root variation in the Taylor ow region, that is, Sh Re 0.5 . While the variation in the Couette ow region has not been reported hitherto, a number of authors have reported a near-square root variation with Reynolds number, as mentioned earlier. The reason for this different behavior lies in the ow eld. In the Couette ow region, the Couette-ow type of ow eld gives rise to a linear variation. In the Taylor ow region, the dominant feature of the ow is the boundary layer type variation near the wall. Since the boundary layer thickness in laminar ow typically varies as Re 0.5 , the Sherwood number also varies as Re 1/ 2 . The effect of Schmidt number on the Sherwood number is shown in Figure 10c at a Reynolds number of 76 and 760, corresponding, respectively, to the Couette ow and Taylor ow regions. It is seen that over the entire of Sc of between 100 and 50,000, Sh varies typically as Sc 1/3 in both the regions. Finally, the limiting value of the Sherwood number at very low Reynolds numbers is of interest. For mass transfer from a stationary sphere, it can be shown that the Sherwood number under stagnant conditions26 is 2. A similar analysis can be carried out for a stationary cylinder. The mass ux through a cylindrical plane at a radial distance of r under steady conditions, can be written as 1 rNAR 0 r r

Applying the boundary conditions that at r R i , C A C AW , and that at r R o , C A 0, the two constants can be evaluated as C1 C AW , lnRi /Ro C2 CAW ln Ro lnRi /Ro (14)

The mass ux at the cylinder surface can now be evaluated as N AR rRi DAB CA r

(15)
rRi

The mass-transfer coefcient and the Sherwood number, based on inner cylinder diameter can now be calculated as Sh D 2 lnRo /Ri (16)

It can be shown for the same boundary conditions that the Sherwood number based on the hydraulic diameter Sh Dh k c 2 d D AB (17)

is for the limiting case is given by Sh Dh 2 2d lnRo /Ri Di

(12)

(18)

where N AR is the mass ux per unit area. Integrating Eq. 12, we get the radial variation of concentration as C A C 1ln r C2
AIChE Journal

(13)
November 2005

The computed variation of Sh Dh with Re Dh for very low rotational speeds is illustrated in Figure 11a for the base case. Here, the speed is varied down to a Reynolds number based on the inner cylinder dia. of 0.5. It is indeed seen that as the Reynolds number is reduced, a limiting value of Sh Dh is reached. The numerically obtained limiting values are comVol. 51, No. 11 2895

Figure 12. Comparison between the Sherwood number correlation and CFD data for (a) Couette ow region, and (b) Taylor ow region.

pared in Figure 11b, with that obtained from Eq. 18 earlier for different radius ratios. As expected, the Sh (based on the hydraulic diameter) decreases as the radius ratio increases. However, there is a factor of three difference between the predictions and that given by the theoretical expression. Part of this may be attributed to the discrepancy in the denition of the Sherwood number. In the theory, it is based on the concentration difference between the wall and the bulk value, which is taken to be zero. In the CFD simulations, the bulk concentration is based on the cup mean value and takes account of velocity and concentration proles. The corresponding limiting Sh values are shown in Figure 11b with lled squares. The limiting Sh values obtained by taking the mass fraction value at the outer radius as the bulk concentration are shown here as open squares. It can be seen that these show a much closer agreement with the theoretical value for a stationary cylinder. The discrepancy that lies between the two, even with a consistent denition of the Sherwood number, may be attributed to the fact that the limiting case considered here is that of a rotating cylinder instead of a translating cylinder for which Eq. 18 is applicable. From a practical point of view, a Sherwood number denition based on the cup mean value is more useful, and the limiting value is taken as Sh lim 12 d lnRo /Ri Di

Using the theoretical analysis of DiPrima et al.,24 the following correlation is developed to determine the critical Taylor number at which the transition to Taylor vortex regime occurs Ta crit 1555 1.05

(22)

If the Taylor number for a given situation is less than Ta crit , then the Couette ow correlation, that is, Eq. 20, is used to determine Sherwood number. Otherwise, the Taylor ow correlation, that is, Eq. 21, is used. The range of parameters for these correlations to be valid is 0.3 0.7 1 Re D 2000 1 Sc 50000 A comparison between the Sherwood number correlation (Eq. 20 and Eq. 21), and the CFD data on which it is based is given in Figure 12a (Couette ow region) and Figure 12b (Taylor ow region) which shows, not surprisingly, very good agreement (within ten percent error) between the two over the range of parameters. The predictions of the correlation with the experimental data of Holman and Ashar8 is shown in Figures 13a and 13b for benzoic acid-water and benzoic acid-glycerol systems, respectively. It should be noted that Holman and Ashar8 use a rather high value of Sc of 10,000 for the rst system; this has been adjusted to a more representative value of 877 (which agrees with the estimates of Bennett and Lewis,1 Marangosis and Johnson6 and Tripathi et al25 among others) in the computation of the experimental Sh in Figure 13a. The reported Sc of 43,200 for the second system is in agreement with data reported for aqueous glycerol solutions of different concentrations, and the same value of Sc is used in the comparison in Figure 13b. Calculations using Eq. 22 show that
Vol. 51, No. 11 AIChE Journal

(19)

Using the numerical data obtained from the CFD simulations, the following correlations for the Sherwood number have been developed for the Couette ow and the Taylor ow regimes Sh Dh 12 Di d 0.0048ReDh Sc1/3 lnRo /Ri Di Do 12 Di d 1/ 2 0.46ReDh Sc1/3 lnRo /Ri Di Do

1.25

(20)

Sh Dh
2896

0.1

(21)

November 2005

Figure 13. Predictions of the correlation with the experimental data of Holman and Ashar8 for (a) benzoic acidwater, and (b) benzoic acid-glycerol systems.
Reynolds number based on hydraulic dia., R i (2 d )/ surface area of the solid Schmidt number, / D AB Sherwood number based on inner cylinder diameter, k c D i / D AB Sherwood number based on hydraulic diameter, k c (2 d )/ D AB Sherwood number limiting, 12( d / D i )/[ln( R o / R i )] time critical Taylor number, 2(1 ) Re 2 /(1 ) velocity components volume of the liquid surface speed of the inner cylinder, R i mass fraction cup mean mass fraction imposed wall mass fraction

these data lie in the Taylor ow regime. Although there is some scatter, there is good qualitative and quantitative agreement between the correlation and the data for both the systems.

Conclusions
CFD simulations are done for the case of a cylinder rotating inside a stationary cylindrical outer vessel lled with liquid. The parameters covered a range of Re , Sc , L / D ratio and radius ratio. The results conrm that the circumferential velocity prole is a strong function of Re and varies from a nearly Couette-type ow at very low Reynolds number to a boundary layer like prole at high Reynolds number. The onset of Taylor vortices has a strong effect on the ow eld and the mass transfer mode. Results show that Sherwood number has a linear dependence on the Reynolds number in the Couette-ow regime, and has linear dependence after the onset of Taylor vortices. A correlation based on two regions in the ow eld for the Sherwood number has been developed in the parameter range 0.3 0.7, 1 Re D 2000 and 1 Sc 50,000. Good agreement was found between the experimental and proposed correlation.

Re Dh S Sc Sh D Sh Dh Sh lim t Ta crit u, v, w V W tip YA Yc Y*

Greek letters
angular velocity (in radians per second) density kinematic viscosity AB molecular diffusivity radius ratio, R i / R o

Literature Cited
1. Bennett JAR, Lewis JB. Dissolution rates of solids in mercury and aqueous liquids: the development of a new type of rotating dissolution cell. AIChEJ. 1958;4(4):418 422. 2. Gabe DR, Robinson DJ. Mass transfer in a rotating cylinder cellI: laminar ow. Electrochimica Acta. 1972;17:11211127. 3. Eisenberg M, Tobias CW, Wilke CR. Mass transfer at rotating cylinders. Chem Eng Prog Symp Ser. 1955;51(16):1. 4. Johnson AI, Huang CJ. Mass transfer studies in an agitated vessel. AIChEJ. 1956;2(3):412 419. 5. Sherwood TK, Ryan JM. Mass transfer to a turbulent uid with and without chemical reaction. Chem Eng Sci. 1959;11(2):8191. 6. Marangozis J, Johnson AI. A correlation of mass transfer data of solid liquid systems in agitated vessels. Can J Chem Eng. 1962;231237. 7. Pangarkar VG, Yawalkar AA, Sharma MM, Beenackers AACM. Particle-liquid mass transfer coefcient in two/three phase stirred tank reactors. Ind Eng Chem Res. 2002;41:4141 4167. 8. Holman KL, Ashar ST. Mass transfer in concentric rotating cylinders with surface chemical reaction in the presence of taylor vortexes. Chem Eng Sci. 1971;26:18171831. 9. Kataoka K, Doi H, Komai T. Heat/mass transfer in taylor vortex ow with constant axial ow rates. Int J Heaet Mass Transfer. 1977;20: 57 63. 10. Legrand J, Dumargue P, Coeuret F. Overall mass transfer to the

Acknowledgment
The calculations reported here have been done using the computational facilities of the CFD Centre, IIT-Madras, India.

Notation
CA C AW d D D AB i, o jA L kc N AR r R Re crit Re D bulk mean concentration at time t solubility concentration of the solid in liquid gap width cylinder dia. kinematic diffusivity inner and outer cylinder average-mass ux cylinder height mass-transfer coefcient mass ux radial distance from the wall of the inner cylinder cylinder radius, m critical Reynolds number, R i d / Reynolds number based on inner cylinder dia., R i D i /

AIChE Journal

November 2005

Vol. 51, No. 11

2897

11. 12. 13. 14. 15. 16. 17. 18.

rotating inner electrode of a concentric cylindrical reactor with axial ow. Electrochimica Acta. 1980;25:669 673. Coeuret F, Legand J. Mass transfer at the electrodes of concentric cylindrical reactors combining axial ow and rotation of the inner cylinder. Electrochimica Acta. 1981;26(7):865 872. Grifoll J, Farriol X, Giralt F. Mass transfer at smooth and rough surfaces in a circular couette ow. Int J Heat Mass Transfer. 1986; 29(12):19111918. Kawase Y, Ulbrecht JJ. Laminar mass transfer between concentric rotating cylinders in the presence of taylor vortices. Electrochimica Acta. 1988;33(2):199 203. Holeschovsky UB, Cooney CL. Quantitative description of ultraltration in a rotation ltration device. AIChE J. 1991;37:1219 1226. Baier G, Grateful TM, Graham MD, Lightfoot EN. Prediction of mass transfer rates in spatially periodic ows. Chem Eng Sci. 1999;54:343 355. Taylor GI. Stability of a viscous liquid contained between two rotating cylinders. Philos Trans Roy Soc London. 1923;A223:289. Schlichting H. Boundary-layer theory (6th edition). New York: McGraw-Hill; 1968. Jayanti S. Hydrodynamics of jet mixing in vessels. Chem Eng Sci. 2001;56:193210.

19. Rhie CM, Chow WL. Numerical study of the turbulent ow past an airfoil with trailing edge separation. A I A A J. 1983;21:15271532. 20. Patankar SV. Numerical heat transfer and uid ow. New York: Hemisphere Publishing; 1980. 21. Leonard BP. A stable and accurate modeling procedure based on quadratic upstream interpolation. Comp Methods Appl Mech Eng. 1979;19:59 98. 22. Hirsch C. Numerical computation of internal and external ows. England: John Wiley & Sons, Inc.; 1990. 23. Jones CA. The transition to wavy taylor vortices. J Fluid Mech. 1984;157:135139. 24. DiPrima RC, Eagles PM, Ng BS. The effect of radius ratio on the stability of couette ow and taylor vortex ow. Phys Fluids. 1984; 27(10):24032411. 25. Tripathi G, Singh SK, Upadhay SN. Mass transfer at rotating cylinders. Indian J Tech. 1971;9:237241. 26. Bird RB, Stewart WE, Lightfoot EN. Transport phenomena. New York: Wiley, 1960.

Manuscript received Oct. 26, 2004, and revision received Mar. 9, 2005.

2898

November 2005

Vol. 51, No. 11

AIChE Journal

Study on Shrinkage Characteristics of Heated Falling Liquid Films

Feng Zhang, Zhibing Zhang, and Jiao Geng
Dept. of Chemical Engineering, Nanjing University, Nanjing, 210093, China DOI 10.1002/aic.10610 Published online August 9, 2005 in Wiley InterScience (www.interscience.wiley.com).

A liquid lm owing down along a heated solid surface may be substantially inuenced by the Marangoni effect, which arises from the existence of a surface tension gradient that is induced by the variations of surface temperature in the transverse direction of the lm, causing appreciable contraction of the lm. In this work, the governing equations for laminar falling liquid lms over a heated plate with constant surface temperature were solved and the surface temperature prole of the lm was obtained. By considering the surface tension gradient and shrinkage characteristics in the brim of the heated lm, a shrinkage model that gives the lm interfacial area was derived based on the experimental infrared images. A comparison of the model with experimental data shows that the model can satisfactorily describe the ow characteristics of falling heated lms. The model is expected to be useful for the design and operation of falling lm equipments. 2005
American Institute of Chemical Engineers AIChE J, 51: 2899 2907, 2005

Keywords: heat transfer, Marangoni effect, falling lms, shrinkage model

Introduction
Falling lms are usually used as the heat- and mass-transfer media in industrial equipment such as vertical condensers, lm evaporators, and absorption towers.1 In the processes of heated falling lms, heat transfer occurs and interacts with lm ow, leading to the variations of the liquid properties, especially viscosity and surface tension. Several researchers2,3 indicated that the variations of liquid viscosity (thermoviscosity effects) would affect the lm ow, whereas the nonuniformity of the surface tension at the liquid gas interface, usually referred to as the Marangoni effect, could strongly inuence the ow dynamics and thus the heat transfer.4 In recent years, interfacial instabilities of heated lms such as the formation of surface waves, breaking of a stream into rivulets, and evaporation/termination of the liquid layer at a contact line, have been observed and theoretically analyzed.5-10 Joo et al. studied the interfacial instability mechanism of heated falling lms and derived a long-wave evolution equation to
Correspondence concerning this article should be addressed to Z. Zhang at segz@nju.edu.cn.

2005 American Institute of Chemical Engineers

describe both the surface wave and the thermocapillary instability that were caused by either the gravity or the surface tension gradient. They concluded that the coupled temporal instabilities could create surface deformation and lead to an array of rivulets aligned with ow at moderate ow and heattransfer rates.10 In 1997, by completely solving the governing equations for the heated falling lm, they further performed the three-dimensional simulation of the instabilities and rivulet formation. It was found that the spontaneous rupture and rivulet formation occurred even in a thin lm under the combined inuences of thermocapillary and surface-wave instabilities.5 The Kabov research group4,11-13 successfully performed experiments of heated falling lms and developed a theoretical model to describe the inuence of the Marangoni effect that was induced by the temperature and concentration gradients along the liquidvapor interface. Very useful correlations for heat-transfer coefcient and lm breakdown (including surface tension effect) were obtained, and the instabilities of a thin falling lm, resulting from the Marangoni effect, were indicated to have a pattern of regular horseshoe-like structures. This pattern of instable ow caused a decrease of the heattransfer coefcient with an increase of the Reynolds number.4 Skotheim et al.6 also worked on the instability of the heated
Vol. 51, No. 11 2899

AIChE Journal

November 2005

Figure 1. Shrinkage of the heated lm.

falling lms and showed that the surface tension gradients induced by the temperature difference, curvature pressure, gravity, and heat conduction could all inuence the shape and stability of the steady ow prole. A model based on the long-wave evolution theory was proposed to describe the stationary curvature of the lm and its stability, which were in good agreement with experimental data. There are several studies on the dry patch forming in liquid lms owing over a heated surface. Some researchers14-16 reported that the Marangoni effect induced by the surface tension gradient could substantially affect the dry patch formation. Zuber and Staub14 observed the effects of the thermocapillary, the vapor thrust forces, as well as the forces arising from ow and surface wetting, on the dry patch. They thus presented a reasonable analysis for the prediction of the conditions that would permit a dry patch to form and remain stationary in a liquid lm owing over a heated surface. Shi and Zhang16 analyzed the breakage of subcooled liquid lm owing down onto a heated plate with a simplied transverse deformation model and a dry patch model. The length of deformation area and critical lm thickness were expressed as functions of surface tension coefcient and heat ux in their models, validating their analysis. As a liquid lm ows down over a heated surface, the variation of surface temperature occurs in both the streamwise and the transverse directions. The streamwise case arises from heat transfer and surface wave, whereas the transverse case arises from different ow rates in the brim of the lm. Such temperature variations generate surface tension gradient and thermocapillary phenomena that are effective on the lm ow. However, most of the work mentioned above took into account variations of surface temperature only in the streamwise direction. The variation in the transverse direction was usually ignored. In fact, the temperature gradient in the transverse direction is usually much higher than that in the streamwise direction (Figure 1) and might be more effective to heat trans2900 November 2005

fer and lm ow because the gravity-driven ow dominating in the streamwise direction is absent in the lateral direction.5 This neglect may cause the correlative theory to stray largely from the experimental observation. In 2001, using water as the working uid, our research group17 performed experiments for the heat transfer of falling lms (under no-evaporation condition) on a heated/cooled smooth stainless steel plate. It was found that the heated lm was obviously contracted to form an inverse trapezoid, whereas the cooled lm was expanded to form a right trapezoid. The width of the falling lm decreased with increasing heating temperature, and vice versa. Actually, the shrinkage of heated lm or expansion of cooled lm was concluded to arise from the surface tension gradient in the transverse direction. For a heated lm, temperature in the brim was higher than that in the center because of the slower ow and the sufcient heat transfer in the brim. This resulted in much smaller surface tension in the lm margin than that in the center. The contraction of the lm occurred under the effect of such surface tension gradient from brim to center of the lm. In the case of the cooled lm, the higher surface tension in the margin than that in the center leads to expansion of the lm, which is completely opposite to the results obtained for the heated lm. An infrared image is shown in Figure 1 to illustrate the surface temperature eld of the heated liquid lm that has initial uniform temperature. Different colors in the thermography represent different temperature magnitudes: the darker the color, the lower the temperature. The horizontal black solid line indicates the upper position of heater edge. It can be clearly seen that the heated lm below the black solid line could be divided into two sections: the uniform central part, with an almost constant temperature, and the rim parts between dashed lines and boundaries of the lm. The dashed lines are identied as temperature sidelines for the heated lm. The transverse temperature curve of the heated lm, located at about 0.05 m down from the upper edge of the heater, was recorded and shown in Figure 2 to have much higher temperature at the edges (the two peaks in the gure). Therefore, the evident surface tension gradient (shown by the arrow in Figure 1) attributed to the temperature gradient in the rim part apparently

Figure 2. Transverse surface temperature of the heated lm. Vol. 51, No. 11 AIChE Journal

small waves. Compared with the average lm thickness , the heights of these waves are negligibly small, so that the lm surface can be regarded as smooth, as shown in Figure 4. Saouli19 theoretically investigated heat transfer of a laminar falling lm along an inclined plated with constant heat ux to successfully obtain the velocity and temperature proles. In this work, a two-dimensional model for a fully developed laminar falling lm over a vertical plate of constant wall temperature would be presented to obtain the lm temperature prole in the central part. Because the inertia term in the momentum equation is negligibly small compared to the body force term, the momentum equation in the direction x of gravity is simplied as

Figure 3. Heated falling lm.

2u x g 0 y2

(1)

contracted the lm, resulting in a substantial reduction of the interfacial transfer area. Because of the absence of fundamental information concerning the hydrodynamics of the liquid lm, mass- and heattransfer analyses for industrial falling lm equipment are still based on empirical methods in general. In a wetting wall column or a packed column the partial pressure driving force is determined by the conditions in the gas phase and the equilibrium concentrations of the solutes in the solvent. The remaining factors affecting the mass-transfer rate are the mass-transfer coefcient and the interfacial transfer area. The information that would permit separation of those two factors was not available, leading to the use of a combined coefcient, K, where K is the mass-transfer coefcient and is the masstransfer area per unit volume. Because K and are both independent functions of the liquid conditions, use of a combined coefcient makes a rational analysis difcult.18 As for heat-transfer equipment, the heat-transfer rate is determined by the interfacial area and the temperature difference between working uid and heater. The interfacial area, usually regarded as the mechanical area of the heater, is overestimated because the heated liquid lm could not completely cover the solid surface. The breakage of the liquid lm and the dry patches may be frequently encountered in various common industrial heat- and mass-transfer equipment. Therefore, for both heat- and mass-transfer processes, an understanding of the liquid lm characteristics, especially the interfacial transfer area, is quite important. Based on the experimental infrared images done in this study, a governing equation for a heated lm owing down a vertical plate of constant wall temperature was presented to calculate the temperature prole of the falling lm. Then, a shrinkage model will be further developed by taking into account the effect of the transverse temperature gradient on the lm ow and Marangoni ow and be validated by designing and carrying out the experiments of heated falling lms.

where ux is the velocity component in the x-direction; , , and g represent the dynamic viscosity, liquid density, and gravitational acceleration, respectively. Because the temperature differences in both the x- and z-directions are very small compared to the temperature difference in the y-direction of the lm, and ow rate in the z-direction is also negligible compared with ux, heat transfers by convection in the z-direction and by conduction in the x- and z-directions are all neglected. Then, the governing energy equation can be expressed using the following equation ux T x, y 2T a 2 x y (2)

where a /cp represents for the thermal diffusivity. The boundary conditions of no slip at the wall and no stress in the lm surface are used for the momentum equation y0 y ux 0 ux 0 y (3a) (3b)

The energy boundary conditions are as follows:

Theoretical Model of heated liquid lm ow

As shown in Figure 3, the surface of a thin liquid lm owing down a vertical plate under the gravity emerges with
AIChE Journal November 2005 Figure 4. Physical model and coordinate system. Vol. 51, No. 11 2901

Initial Temperature of the Film T 0, y T 0 Constant Temperature of the Plate Surface T x, 0 Tw Adiabatic Film Surface T x, 0 y (4c)
Figure 5. Shrinkage of a heated lm.

(4a)

(4b)

Integrating Eq. 1 leads to the expression of the velocity prole of the lm ux um 1 1 where um 1 g 2 2

y
2

Equation 11 is the working expression for the temperature prole of the liquid lm.

(5)

Model for the shrinkage of heated lm

Consideration of the surface tension gradient in the rim part and its effect on the heated falling lm, suggests a model, as illustrated in Figure 5, for the expression of the lm shrinkage in the initial section. The lm is postulated to be symmetric to the central line (x-axis) of the lm so that it is sufcient to analyze only one side of the lm. In Figure 5, the z-direction is normal to the xy plane as shown in Figure 4. For distinguishing the two parts of the lm ow, a dashed line is drawn between the uniform central part and the rim part. The lm in the uniform central part is assumed to be of constant thickness and constant ow rate Q, whereas the liquid within the rim part is regarded as owing along the channel regulated by the dashed line and the contact line AB. Moreover, D stands for the width of the rim part and (x wcos )/D is the surface tension gradient in the rim. x and w represent the surface tension at the temperature of T(x, ) and Tw, respectively. T(x, ) represents the surface temperature of the liquid lm in the uniform central part, and Tw the temperature of the plate surface. By neglecting the effect of gravity and inertia force that are negligibly small in the z-direction, the momentum equation in the z-direction of the rim becomes

(6)

The liquid mass ow rate per unit perimeter is

u xd y

(7)

Substituting Eq. 5 in Eq. 7 generates the expression of the liquid lm thickness

3Q 2g
2

1/3

(8)

Thus, by combining Eqs. 2 and 5, the energy equation can be rewritten as follows um 1 1

y ax u m 2

T x, y 2T a 2 x y

(9)

2u z 0 y2

(12)

The following dimensionless variables were dened Y y X UY u y um T x, y T0 (10) Tw T0

The boundary conditions are as follows: At the Wall uz 0 At the Film Surface (12a)

X , Y

By substituting Eq. 10 in Eqs. 9 and 4, neglecting the entrance effects, separating the variables, and integrating Eq. 9, we have the following expression X , Y
2902

uz cos y z

(12b)

6 1 12 X 1 Y2 1 Y4 5 5 5

(11)

where uz is the velocity component of the liquid ow in the z-direction and is the angle of the tangent to the contact line with the x-direction. By integrating Eq. 12 across the lm using the above boundVol. 51, No. 11 AIChE Journal

November 2005

ary conditions and incorporating the Marangoni ow rate in the liquid layer as given by Kabov,12 the velocity in the z-direction is derived to be u z x wcos cos 4D (13)

T x , T 0 12 ax T T0 5 u m 2

(21)

The temperature of the surface without stress can thus be given as T x, T0 12 ax T 5 u m 2 (22)

where is the contact angle along the contact line. According to the illustration in Figure 5 we have uz/um tg 2 2 and cos um/u m uz . Then Eq. 13 becomes uz Dening C

x wcos um 2 4D um uz2

where T T T0. Substituting Eq. 22 in Eq. 19 gives C

(14)

12 ax w 1 cos T 1 4D 5 um 2

A0 Kx (23)

wcos 4D x

(15)

where A0

and rearranging Eq. 14, lead to the following expression uz

wcos 4D 0

(24)

2
2

um

4C2 2 um

1/ 2

1/ 2

(16)

3 Ta T 12 a 4 D 5 u m 2 5 Du m

(25)

As the liquid lm owing down x distance on the heated plate, the shrinkage width Z on one side of the lm is Z

By combining Eq. 23 with Eq. 18 and then integrating, the shrinkage width of one side of the lm can be nally expressed as Z

x/um

u zdt

(17)

2 u m
6 K

2 3/ 2 1/ 2 1/ 2 S 3/ x S 0 3 S x S 0

where t represents the time and x/um is the time duration for the lm owing over x distance. Substituting Eq. 16 into Eq. 17 produces the following expression

x 0, x e where Sx

(26)

2
2

x 0

4C 2 um

1/ 2

1/ 2

dx

(18)

2 A 0 2 Kx um 2 A0 um
2

It is assumed that the surface tension decreases linearly with the temperature: 0 (T T0); then Eq. 15 becomes C

S0

A0 0 w 1 cos um um 4D 0 3 1 Ma0 K u m 5 D Pe2 x xe when

2 1/ 2

1/ 2

wcos 1 cos 4D x 4D w
T Tw (19)

where T T(x, ). When y , Y 1, Eq. 11 is reduced to 12 X , 1 X 5 (20)

T x, Tw

By substituting Eq. 10 in Eq. 19, the following equation is obtained

AIChE Journal November 2005

where xe represents the length of the contracted lm; um/0 is denoted as a dimensionless parameter called the Capillary number; Ma0 T/a stands for Marangoni number, which is a nondimensional measure of the surface tension gradient; and Pe um/ is denoted as the Pe clet number. Equation 26 provides a new way to numerically calculate the shrinkage
Vol. 51, No. 11 2903

Figure 6. Experimental system.

Legend: 1: test section; 2: liquid collection vessel; 3: liquid storage tank; 4: pump; 5: upper tank; 6: rotameter; 7: uid distributor; 8: liquid exit gap; 9: infrared camera; 10: computer.

and then owed down to the test section at a scheduled ow rate. The ow rate ranging from 0.03 to 0.36 kg m1 s1 was measured with a rotameter. The liquid from the upper tank entered a uid distributor installed in the upper position of the test section. The liquid exit gap could be adjusted manually to create liquid lms of different thicknesses according to the needs of the experiments. The liquid owing down the plate was collected by a vessel and recycled to the storage tank. Five Pt-100 thermal resistances were also installed to measure the temperature values of the lm inlet and outlet, storage tank, upper tank, and the air near the lm, respectively. The uncertainty of these temperature values was within 0.1C. The surface temperature and the shape of the lm were determined by a highly sensitive infrared camera, Therma CAM SC3000 (FLIR Systems, Inc., Portland, OR). Because the lm edge was very thin, temperature at the edge of the lm was regarded as the surface temperature of the wall, Tw. The initial temperature of the falling water lm was set and maintained at several constant values during the experiments (T0 20 35C). The temperature of plate surface Tw was controlled as the temperature of hot water ranging from 30 to 70C. The physical characteristics of working uid and basic experiment parameters are given in Table 1.

Characteristics of heated falling liquid lms

width of a liquid lm owing down a heated vertical plate. It implies that the shrinkage width is affected by the ow rate, the physical properties of liquid, the temperature difference of (Tw T0), and the initial width of the liquid lm (contained in the three nondimensional parameters um/0, Ma0, and Pe). D, the width of the rim part, is also a very important parameter. The transverse temperature gradient and surface tension gradient are enhanced as D decreases, resulting in dramatic lm contraction. The experiment shows that the ow of the heated falling liquid lm is mainly affected by the temperature difference (Th T0) and the lm ow rate. As (Th T0) increases, the liquid lm apparently contracts and becomes very narrow. The reason is that the surface tension gradient in the transverse direction of the lm is enhanced. The effect of the ow rate on the shrinkage of the heated lm ow is quite complex. On one hand, a fast lm ow rate is benecial to the lateral expansion of the lm. On the other hand, both lm temperature and the rim width D decrease with an increase of ow rate, and the transverse surface temperature gradient correspondingly increases in the rim of the lm. Therefore, the surface tension gradient is also increased, thus apparently causing the lm to contract. For example, as shown in Figure 7, the heated lm at a ow rate 0.03 kg m1 s1 is quite narrow, but then increases appreciably with liquid ow rate, increasing from 0.03 to 0.12 kg m1 s1. However, when the ow rate is 0.12 kg m1 s1, the lm width starts to increase slowly. Interestingly, the quick liquid contraction is found only near the entrance, but the width tends to keep constant to the end after the liquid ows beyond a certain distance. This phenomenon may imply that the effect of viscous force and the lateral expansion effect of the liquid ow are nally balanced with Marangoni effect in the transverse direction of the lm.

Experimental Experimental setup

To determine the unknown variable D and to obtain the shrinkage width of the heated lm, an experimental setup as shown in Figure 6 was designed. The test section [900 300 6 mm (length width thickness)] of the stainless steel plate was carefully polished. The experiment was expected to allow the liquid lm to ow down at a scheduled ow rate over the vertical plate, being indirectly heated by hot water at a preset temperature. The surface temperature of the plate could be maintained by the hot water at a given temperature. The cool water source, used as the experiment media, was also maintained by a storage tank at a constant temperature. Distilled water in the storage tank was pumped to an upper tank

Table 1. Physical Characteristics of Working Fluid and Operation Parameters

Physical Characteristics Working uid: distilled water Initial temperature, T 0 : 2030C Density, : 995.3998.5 kg/m3 Viscosity, : 7.169.90 104 Pa s Surface tension, : 70.672.9 dyn/cm Heat conductivity, : 0.59970.6241 W m1 k1 Operation Parameters Flow rate, Q : 0.030.36 kg m1 s1 Temperature of hot water, T h : 3070C Reynolds number, Re 4 Q / : 1372024 Prandtl number, Pr C P / : 4.796.91

2904

November 2005

Vol. 51, No. 11

AIChE Journal

Figure 9. Variations of D with and .

be semicircular, as illustrated in Figure 8, where in the gure stands for the thickness of the uniform central lm. p represents the peak of bulge and is the contact angle of the liquid lm on the plate. By using the geometrical analysis, the width of the rim part can be expressed as a function of , , and p D where sin 2 1 12 1 cos 1 cos

(27)

(28)

Figure 7. Shapes of heated lms at different ow rates.

T0 20C, Tw 70C, x* x. Units are in kg m1 s1: (1) 0.03; (2) 0.06; (3) 0.09; (4) 0.12; (5) 0.15; (6) 0.18; (7) 0.24; (8) 0.30; (9) 0.36.

Results and Discussion Calculation of the rim width

As mentioned above, D represents the width of the rim part and (x wcos )/D stands for the surface tension gradient in the rim. With a decrease of D, (x wcos )/D consequently increases and the Marangoni effect in the transverse direction of lm becomes intensive. According to Figure 2, the rim part of the heated lm is not at but protuberant in the cross section. This bulgy structure is considered to result from the combination of the central lm ow and the shrinkage of the lm. Generally, the cross section of the rim part can be assumed to

where , , and the factor are, respectively, functions of physical properties of the liquid, the temperature difference between Tw and T(x, ), and lm ow rate. It is obvious that D increases with increasing lm thickness . is given by Eq. 8 when the temperature is set to the average value of the inlet and outlet of the lm. As an example for computing thickness of heated lms at the temperature of T0 20C and Th 70C, variations of D with and at a ow rate of 0.24 kg m1 s1 are illustrated in Figure 9. As can be seen, D decreases noticeably with an increase of . In fact, the wettability of the liquid on the solid is highly related to the contact angle : the larger the contact angle, the less the wettability of the solid, leading to the smaller value of D. In addition, an increase of D with the factor can be found in Figure 9, which is clearly understand-

Figure 8. Cross section view of the rim part.

B, boundary of the lm; P, peak of the rim; T, transitional point from the rim to central part of the lm.

Figure 10. Variations of Z vs. x. Vol. 51, No. 11 2905

AIChE Journal

November 2005

(Figure 1). Therefore, in the actual calculation of the shrinkage model, the average D over the contracted lm was used instead of the D at a certain distance of x.

Results of calculation
The shrinkage width Z of the heated lm (T0 20C, Th 70C) was calculated from Eq. 26. As shown in Figure 10, Z increases with the increase of the distance x of lm for all the liquid ow rates. Z also increases with decreasing liquid ow rate at the same x. Such phenomena are in accordance with the variation of the lm boundary vs. ow rate, as shown in Figure 7. Additionally, the value of Z at two ow rates of 0.06 and 0.09 kg m1 s1 differs substantially from values at other ow rates, indicating the insufcient coverage of the lm for the plate. If Z0 is assumed to be the initial width of the lm, the width Z at distance x can be calculated as Z Z0 2Z (29)

Figure 11. Variations of Z with x and liquid ow rate.

able because of the semicircular structure of the cross section of the rim in Figure 8. Moreover, D varies slightly when 50 (Figure 9). 50 can be regarded as the upper limit of the contact angle of the liquid on the plate, which is consistent with the fact that the receding contact angle of the distilled water on stainless steel at plate ranges from 12 to 46.20 In addition, it should be noted that the width of the rim D increases along the downstream and nally reaches a maximum

As shown in Figure 11, Z decreases with decreasing ow rate and is dramatically reduced as the ow rate is lower than a certain value.

Figure 12. Comparisons of experiments with calculations.

T0 20C, Th 70C.

2906

November 2005

Vol. 51, No. 11

AIChE Journal

Comparison of the model with the experiments

Comparison of the calculated results by the shrinkage model with the experimental data is made and shown in Figure 12. The symbols represent the measured lm boundaries, whereas the single lines are results calculated from Eq. 26. The calculations were in good agreement with the experiments at higher ow rates of 0.24 and 0.18 kg m1 s1, with deviations of 3.8 and 3.3%, respectively. With decreasing liquid ow rate, the deviations increase to 5.6 and 10.9% at ow rates of 0.09 and 0.06 kg m1 s1, which are still acceptable from an industrial perspective. This implies that the model can accurately describe the shrinkage of heated falling lms and thus accurately present the interfacial area. In addition, the model can be extended to a more realistic level by including the effects of evaporation at the lm edge, the effect of temperature on viscosity variation, and other dynamic parameters. This will be the focus of future work.

Literature Cited
1. Yih SM, Seagrave RC. Hydrodynamic stability of thin liquid lms owing down an inclined plane with accompanying heat transfer and interfacial shear. AIChE J. 1978;24:803-810. 2. Wilson SK, Duffy BR. On the gravity-driven draining of a rivulet of uid with temperature-dependent viscosity down a uniformly heated or cooled substrate. J Eng Math. 2002;42:359-372. 3. Goussis DA, Kelly RE. Effects of viscosity variation on the stability of a liquid lm ow down heated or cooled inclined surfaces: Finite wavelength analysis. Phys Fluids. 1987;30:974-982. 4. Kabov OA, Scheid B, Sharina IA, Legros JC. Heat transfer and rivulet structures formation in a falling thin liquid lm locally heated. Int J Therm Sci. 2002;427:664-672. 5. Ramaswamy B, Krishnamoorthy S, Joo SW. Three-dimensional simulation of instabilities and rivulet formation in heated falling lms. J Comput Phys. 1997;131:70-88. 6. Skotheim JM, Thiele U, Scheid B. On the instability of a falling lm due to localized heating. J Fluid Mech. 2003;475:1-19. 7. Schatz FM, Neitzel PG. Experiments on thermocapillary instabilities. Annu Rev Fluid Mech. 2001;33:93-127. 8. Wang KH, Ludviksson V, Lightfoot EN. Hydrodynamic stability of Marangoni lms. AIChE J. 1971;17:1402-1408. 9. Ludviksson V, Lightfoot EN. Hydrodynamic stability of Marangoni lms. AIChE J. 1968;14:620-626. 10. Joo SW, Davis SH, Bankoff SG. A mechanism for rivulet formation in heated falling lms. J Fluid Mech. 1996;321:279-298. 11. Kabov OA. Breakdown of a liquid lm owing over the surface with a local heat source. Thermophys Aeromech. 2000;7:513-520. 12. Kabov OA, Chinnov EA. Heat transfer from a local heat source to subcooled liquid lm. High Temp. 2001;39:703-713. 13. Zaitsev DV, Kabov OA, Evseev AR. Measurement of locally heated liquid lm thickness by a double-ber optical probe. Exp Fluids. 2003;34:748-754. 14. Zuber N, Staub FW. Stability of dry patches forming in liquid lms owing over heated surfaces. Int J Heat Mass Transfer. 1966;9:897903. 15. Wang BX, Zhang JT, Peng XF. Experimental study on the dryout heat ux of falling liquid lm. Int J Heat Mass Transfer. 2000;43:18971903. 16. Shi JS, Zhang QZ. Analysis on the breaking of subcooled falling liquid lms. Chin J Appl Mech. 2002;19:78-80. 17. Geng J. Inuence of Marangoni Effect on Distillation in Packing Tower. PhD Dissertation. Nanjing, China: Nanjing University; 2002. 18. Dukler AE, Bergelin OP. Characteristics of ow in falling liquid lms. Chem Eng Process. 1952;48:557-563. 19. Saouli S, Saouli SA. Second law analysis of laminar falling lm along an inclined heated plate. Int Commun Heat Mass Transfer. 2004;31: 879-886. 20. Wang XD, Peng XF, Lu JF, Wang BX. Measuring technique of contact angle and contact angle hysteresis on rough solid surface. II: Contact angle hysteresis on rough stainless surface. J Basic Sci Eng (China). 2003;11:296-303.

Conclusions
The characteristics of heated falling lms over a vertical stainless steel plate have been investigated both experimentally and theoretically, and the following conclusions were made: (1) From infrared thermographic experiments of heated lm owing down a vertical stainless steel plate, it is shown that the liquid lm is heated and contracted from the beginning stage of the entrance to form the shape of an inverse trapezoid. The lm shrinkage and the noticeable transfer area reduction are attributed to the surface tension gradient that results from the variations of temperature in the transverse direction of the lm. This indicates that the Marangoni effect arising from the existence of surface tension gradient seriously affects the falling lm process. (2) The governing equations for a laminar falling liquid lm over a vertical heated plate of constant surface temperature were solved and the temperature distribution was obtained for the exact expression of Marangoni effect in the heated lm. (3) A model describing the characteristics for the shrinkage of heated falling lm was established with consideration of the lm ow over a vertical plate and the surface tension gradient effect in the transverse direction of the lm. Comparison with the experiments demonstrated that the model could accurately describe the shrinkage characteristics of the process.

Acknowledgments
The authors are grateful for the nancial support from National 985 Project of China and the National Natural Science Foundation of China (No. 20576050). We also thank Professor Youting Wu for valuable help.

Manuscript received Nov. 21, 2004, revision received Mar. 31, 2005, and nal revision received Jun. 5, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2907

Visualizations of Boger Fluid Flows in a 4:1 SquareSquare Contraction

M. A. Alves
Depto. de Engenharia Qu mica, CEFT, Faculdade de Engenharia da Universidade do Porto, 4200-465 Porto, Portugal

F. T. Pinho
Centro de Estudos de Feno menos de Transporte, Depto. de Engenharia Meca nica, Universidade do Minho, Campus de Azure m, 4800-058 Guimara es, Portugal

P. J. Oliveira
Depto. de Engenharia Electromeca nica, Unidade Materiais Te xteis e Papeleiros, Universidade da Beira Interior, 6201-001 Covilha , Portugal DOI 10.1002/aic.10555 Published online August 12, 2005 in Wiley InterScience (www.interscience.wiley.com).

Visualizations of the 3-D ow in a 4:1 squaresquare sudden contraction for two viscoelastic Boger uids and two Newtonian uids were carried out at low Reynolds numbers. In these creeping ow conditions, the vortex length remained unchanged for Newtonian uids, whereas a nonmonotonic variation with ow rate was observed for the Boger uids. Initially, the corner vortex slightly increased with ow rate to a local peak at a Deborah number of De2 6, before decreasing signicantly to a minimum at De2 15 (De2 is based on downstream characteristics). Finally, for Deborah numbers 20 there was intense vortex enhancement until a periodic ow was established at higher ow rates (De2 4552). The strong elastic vortex enhancement was preceded by the appearance of diverging streamlines on the approach ow and, for the Boger uid with higher polymer concentration, vortex enhancement took place through a lip vortex mechanism. 2005 American Institute of Chemical Engineers AIChE J, 51: 2908 2922, 2005 Keywords: viscoelastic ow, Boger uid, diverging ow, visualization, 3D contraction ow

Introduction
Flow visualizations have always been important in uid mechanics and motivated a wealth of research in the eld, as illustrated in Van Dykes classic work An Album of Fluid Motion.1 They provide a clear insight of many phenomena and reinforce elaborate mathematics. Non-Newtonian uids are usually more viscous than common Newtonian uids and their ows are frequently investigated by ow visualization because
Correspondence concerning this article should be addressed to M. A. Alves at mmalves@fe.up.pt.

2005 American Institute of Chemical Engineers

they tend to take place in the laminar regime. A compilation of ow visualization studies specically for non-Newtonian uids, was written by Boger and Walters in 1993,2 and constitutes a representative sample of works published in this important subeld of uid mechanics. Sudden contraction ows are classical benchmark problems used in computational rheology,3 with the explicit assumption of two-dimensional (2-D) ow to simplify the simulations, and a large number of visualization studies in planar and axisymmetric contractions have been published in the literature. The ow behavior of non-Newtonian uids in these simple congurations can be very surprising; there can be substantially different ow patterns for uids with apparently similar
Vol. 51, No. 11 AIChE Journal

2908

November 2005

rheological characteristics, and distinct patterns also arise in comparable congurations as a result of geometric differences (such as planar vs. axisymmetric). The following paragraphs provide a review of relevant works in sudden contraction ows of axisymmetric, planar, and squaresquare type, and a summary of the main conclusions is presented in Table 1.

Axisymmetric contractions
For axisymmetric contractions, pioneering visualization studies with viscoelastic uids were carried out by Cable and Boger4-6 and Nguyen and Boger,7 who also reviewed previous work in the eld. At low ow rates they observed that the salient corner vortex was controlled by viscous forces and a normalized length of 0.18 was measured, with the normalization based on the upstream pipe diameter. At higher ow rates a dramatic vortex growth was reported for all geometries (cf. Table 1) and for both shear-thinning and constant viscosity elastic uids. At even higher Deborah numbers the ow became asymmetric and time dependent. At intermediate conditions Cable and Boger observed the onset of diverging ow upstream of the contraction plane and the simultaneous reduction of the salient vortex length, a phenomenon partially attributed to inertia. Fluid elasticity was considered the cause for the ow instabilities because the Reynolds numbers were well below those associated with transitions arising from inertial effects. From the mid-1980s the 4:1 contraction became the focus of a variety of experimental and numerical work. The experimental work concentrated on the investigation of low Reynolds number ow transitions and elastic instabilities and relied on several techniques, as in McKinley et al.8 At that time the numerical methods were usually unable to predict accurately the subcritical steady ows and the emphasis was focused on improving the numerical techniques. In 1986 Boger et al.9 investigated the behavior of two different Boger uids having similar shear properties and yet found quite different vortex dynamics; for a polyisobutylene in polybutene uid (PIB/PB) vortex enhancement was preceded by the growth of a lip vortex, whereas for polyacrylamide dissolved in corner syrup (PAA/CS) the lip vortex was absent. It was concluded that an additional measure of extensional properties had to be taken into account and in his 1987 review paper10 Boger suggested that parameter to be the extensional viscosity, and described in more detail the ow dynamics in the sudden contraction with increasing Deborah number. For some uids only a convex corner vortex exists, growing in size and becoming concave as elasticity increases. For other uids the corner vortex extends to the reentrant corner, and a lip vortex is formed. For high contraction ratios, the two vortices are initially separated, but similar distinct vortices were also seen by McKinley et al.8 in small contraction-ratio experiments. As the elasticity increases, the strength of the lip vortex grows at the expense of the corner vortex, whereas the length of the latter remains unchanged. Eventually, the lip vortex occupies the whole contraction plane region and a further increase in the Deborah number leads to a larger concave-shaped vortex. At even higher Deborah numbers a small pulsating lip vortex now appears and leads to unsteady behavior. All these features have
AIChE Journal November 2005

been observed in axisymmetric contractions under conditions of negligible inertia. Recent experimental investigations by Rothstein and McKinley11,12 clearly demonstrate the important role of extensional viscosity on the dynamics of vortex growth and on the associated enhanced pressure drop in contraction ows. They conjectured that the extra pressure drop results from an extra dissipative contribution to the elastic stress arising from a stress-conformation hysteresis in the nonhomogeneous extensional ow generated near the contraction plane. Such conjecture has implications on constitutive-equation modeling and led to the recent numerical simulations of Phillips et al.13 with a closed form of the adaptive length scale (ALS) model of Ghosh et al.,14 which accounts for hysteresis of the conformation tensor. These authors were able to qualitatively predict large pressure drop enhancements, although there were still discrepancies in comparison with experiments.

Planar contractions
Investigations in planar contraction ows began soon after those in axisymmetric contractions and, although they conrmed some of the ndings observed for the circular geometries, some notable discrepancies also emerged. Walters and Webster15 found no signicant vortex activity for Boger uids in the 4:1 case, in marked contrast to observations in 4.4:1 circular contractions. However, for shear-thinning uids vortex growth was observed in both planar and axisymmetric geometries. To help clarify these differences, Evans and Walters16 studied the ows of shear-thinning and constant viscosity elastic uids (aqueous solutions of polyacrylamide) through planar contractions (contraction ratios of 4:1, 16:1, and 80:1) and always found strong vortex enhancement for the shear-thinning uids, even for the smaller contraction ratio, and difculties in observing vortex activity for Boger uids. They also reported that both the contraction ratio and uid elasticity contributed to vortex enhancement with shear-thinning uids. However, for the larger contraction ratios a lip vortex was seen and a growth mechanism similar to that previously found for circular dies was observed, although in the 4:1 planar contraction there was no sign of such a lip vortex. In a subsequent paper Evans and Walters17 looked at the behavior of shear-thinning uids in the smaller contraction ratio of 4:1 to investigate whether a lip vortex mechanism of vortex growth was still at work. They found that for the less viscous/concentrated polymer solutions such lip vortex could be generated and that inertia played a critical role in separating the corner and lip vortices. Independent lip and corner vortices were also found in the numerical simulations of creeping ow by Alves et al.18,19 for a 4:1 contraction with Boger uids represented by either the upper convected Maxwell (UCM) or the Oldroyd-B models. The relevance of extensional viscosity in planar contraction ows was emphasized in the experimental investigations of White and Baird,20,21 who used two polymer melts: polystyrene (PS) and low-density polyethylene (LDPE). Whereas a vortex was observed for the LDPE, it was absent from PS ows and the difference was attributed to the distinct extensional viscosities, given their similar weak shear-thinning behavior. This was corroborated when the same authors later used a constitutive equation that represented accurately the measured extensional viscosity of both uids,22 and were able to predict
Vol. 51, No. 11 2909

Table 1. Relevant Investigations of Viscoelastic Fluid Flow in Axisymmetric, Planar, and Square Contractions*
Ref. 46 Authors (year) Cable and Boger (1978, 1979) Nguyen and Boger (1979) Walters and Webster (1982) Flow Conditions and Test Fluids Circular, 2, 4 Shear-thinning uids Circular, 4 14.6 Boger and shear-thinning uids Planar and circular, 4 Boger and shear-thinning uids Planar, 13.3 SQ/SQ, 13.3 Boger uids Main Conclusions Vortex ow regime with vortex growth at low ow rates. Divergent ow regime at intermediate ow rates and inertial effects. Unstable ow regime at large ow rates. Sequence of ow regimes referring to lip vortex behavior: (1) vortex growth; (2) asymmetric ow; (3) rotating ow; (4) helical ow (pulsating). Boger uids: substantial vortex activity in circular contraction and virtually none in planar case. Shear-thinning uid: substantial vortex activity in circular and planar contractions. Rounding the corner dramatically changes ow. No noticeable vortex activity for the 13.3:1 planar contraction. Signicant vortices observed for the 13.3:1 3-D SQ/SQ contraction. Vortex enhancement with increasing ow rate. Slight asymmetry at highest ow rates. Simultaneous lip and corner vortices, with lip vortex dominating as De increases: De 3 no inuence of on lip vortex; De 3 inuence of on lip vortex. Planar: Boger uids: difculty in observing vortex activity. Shear-thinning uids: substantial vortex activity in all cases. Large : vortex growth associated with lip vortex. Small : vortex growth is not associated with lip vortex. SQ/SQ: Strong vortex enhancement for Boger uids. Vortex growth for LDPE due to early onset of shearthinning and strain hardening extensional viscosity. No vortex growth for PS which has no strain hardening extensional viscosity. Relevance of extensional viscosity. Conrmation of the observations of White and Baird20 for lower and higher contraction ratios. Simulations conrm experiments of White and Baird21 in terms of streamlines and birefringence. Size and intensity of vortex determined by parameter , which controls extensional viscosity in the PTT model. For some uids there are lip and corner vortices, with the former responsible for the vortex enhancement mechanisms, as for large . For mobile uids inertia separates corner and lip vortices. Sequence of ow regimes near the lip vortex: (1) stable; (2) time periodic; (3) quasi-periodic; (4) aperiodic. Time-dependent 3-D ow for 2 5 at high ow rates. Rounding corner delays transitions. Each of the two uids studied react differently to the rounded corner: dramatic changes are observed for the polyacrylamide based Boger uid, in deep contrast to less dramatic changes observed for the polyisobutylene based uid. Predictions qualitatively match the experimental results of Evans and Walters.16,17 Lip vortex not associated with inertia. 2-D: no prediction of lip vortex; decrease of corner vortex size with Deborah number. 3-D: Corner vortex in center plane always smaller than in 2D case.

15

24

Walters and Rawlinson (1982)

Boger et al. (1986)

Circular, 4.4 16.5 Boger uids Planar, 4, 16, 80 Boger and shear-thinning uids SQ/SQ, 16 Boger uids

16

Evans and Walters (1986)

20

White and Baird (1986)

Planar, 5.9 Weakly shear-thinning melts Planar, 4, 8 Weakly shear thinning polymer melts (LDPE, PS) Planar, 4, 8 Numerical, Full PTT, 0 Planar, 4 Shear-thinning uids Circular, 2 8 Boger uid

21

White and Baird (1988) White and Baird (1988)

22

17

Evans and Walters (1989) McKinley et al. (1991)

46

Boger and Binington (1994)

Circular, 4 Sharp and rounded reentrant corner Boger uids Numerical simulations with a FENE-P model Planar, 4, 16, 80 Shear-thinning uids Numerical, Oldroyd-B model 2-D, 4 3-D planar 4

47

Purnode and Crochet (1996) Mompean and Deville (1997)

26

2910

November 2005

Vol. 51, No. 11

AIChE Journal

Table 1. Relevant Investigations of Viscoelastic Fluid Flow in Axisymmetric, Planar, and Square Contractions* (Continued)
Ref. 27 Authors (year) Xue et al. (1998) Flow Conditions and Test Fluids Numerical 2-D: UCM 3-D planar: Boger and shear-thinning uids (UCM, PTT with 0) Main Conclusions 2-D: Lip and corner vortex growth mechanisms depending on elasticity and Deborah numbers. Appearance of lip vortex depends on polymer viscosity and is promoted by inertia. 3-D planar UCM: conrms presence of lip vortex for some ow conditions. PTT: qualitative agreement with experiments of Quinzani et al.48 Relevance of transient extensional behavior along centerline. In 3-D vortex activity correlates well with steady extensional properties, but not in 2-D. Measurement of enhanced pressure drop. It is suggested that this is attributable to an additional dissipative contribution of polymer stress manifested in stress conformation hysteresis. Lip vortex growth mechanism for UCM uid. No lip vortex for PTT uid. Streamline divergence for high Deborah ows, enhanced by inertial effects (UCM). Importance of mesh renement and discretization schemes for numerical accuracy. 2: steady elastic lip vortex 4 8: no lip vortex observed, only corner vortex. Rounding the reentrant corner leads to shifts in the onset of ow transitions to larger Deborah numbers. The role of contraction ratio on vortex growth dynamics is rationalized by using a dimensionless ratio of the elastic normal stress difference in steady shear ow to that in transient uniaxial extension. Planar case: no sign of steady vortex enhancement Circular: lip mechanism responsible for vortex growth and enhanced pressure drop. Lip vortex enhancement for Oldroyd-B uid accompanied by decrease in corner vortex. Intense corner vortex growth for linear PTT and no lip vortex. High Deborah number simulations for PTT models.

25

Xue et al. (1998)

11

Rothstein and McKinley (1999) Oliveira and Pinho (1999); Alves et al. (2000)

Numerical 2D and 3D SQ/SQ, 4 Boger and shear-thinning uids: UCM, PTT ( 0) Circular, 4 Boger uid Planar, numerical 4 UCM and PTT uids

18 42 43

12

Rothstein and McKinley (2001)

Circular, 2 8 Various curvatures of corner Boger uid

23 19

Nigen and Walters (2002) Alves et al. (2003)

Planar and circular 2 32 Boger uids Planar, numerical; 4 Highly rened mesh Oldroyd-B and PTT (linear and exponential) models

* represents the contraction ratio. Works are presented in chronological order.

numerically, at least qualitatively, the different vortex patterns observed. The surprisingly different behavior of Boger uids in circular and planar contractions has been further studied by Nigen and Walters.23 Experiments in both types of contraction were conducted with Newtonian and Boger uids having identical shear viscosities and it was demonstrated that, although there were higher extensional strain rates in the planar geometry than in the circular die, virtually no elastic vortex enhancement was found in the former geometry in contrast to the latter. Given that the vortex dynamics is a complex function of strain rates, strains, and molecular conformation histories,12 these differences are not so surprising.

Squaresquare contractions
Experiments with ows through squaresquare contractions are scarce, partly because they are more complex but also because of their inadequacy to be used for comparison with 2-D simulations. In their 1982 experiments, Walters and Rawlinson24 reported similarities between the ows through circular and squaresquare contractions. In particular, the differAIChE Journal November 2005

ences found between ows of Boger uids in planar and circular contractions were also seen for planar and 13.3:1 squaresquare contractions and at intermediate Deborah numbers a diverging ow was seen upstream of the contraction as originally observed by Cable and Boger4 for circular geometries. In the square duct contraction, and in contrast to the 2-D planar case, isopressure lines can close in the cross-stream directions, as in the axisymmetric geometry, but there are now normal stress imbalances leading to secondary ows. The dynamics of the vortices in combination with these new secondary ow structures and uid elasticity are still to be well characterized and understood. In view of the above, it comes as no surprise that ow predictions of squaresquare contraction ows are virtually inexistent except for the work of Xue et al.,25 to be discussed below. Most of the existing three-dimensional (3-D) viscoelastic ow simulations in contractions basically considered the 2-D planar contraction extending along the spanwise direction and so the 3-D effects were attributed only to the presence of end walls. Some computations of such planar 3-D contraction
Vol. 51, No. 11 2911

ows (in fact, quasi-2-D) were presented by Mompean and Deville26 and Xue et al.27; their intention was mainly to illustrate the capabilities of their nite-volume methods in 3-D simulations, so they did not comment on the results of their calculations in relation to 3-D effects and in fact the detailed 3-D ow structure was not well resolved. The only 3-D effect mentioned by Xue et al.27 was that the average velocity in a 3-D channel differs from the average velocity in a 2-D channel, and therefore the Deborah number should be modied accordingly. In a second article, Xue et al.25 tackled the actual 3-D squaresquare contraction using the Phan-ThienTanner model (PTT) (with a nonzero second normal stress difference coefcient) and UCM constitutive equations. They found strong vortex enhancement for both uid models, in contrast to the results for the planar contraction where only the PTT uid (shear-thinning uid) exhibited vortex enhancement. These contrasting behaviors mirror those observed experimentally for axisymmetric and planar contraction ows, and are explained by the 3-D nature of circular and squaresquare contractions. The numerical investigations of Xue et al.25 related the calculated ow dynamics with the transient extensional viscosity of the uids, but nothing was reported regarding unexpected secondary ows in the contraction region or the onset of ow instabilities; they did mention, however, the existence of the well-known secondary ow in the fully developed straight upstream and downstream noncircular ducts. In conclusion, the vast majority of experimental work on contraction ows of viscoelastic liquids has been on axisymmetric and (approximately) planar congurations, with the expectation that 2-D numerical simulations would allow for adequate predictions. A squaresquare contraction arrangement is a good compromise between geometric simplicity and complex 3-D ow structure, and appears as a good candidate for a prototype 3-D test case that is necessary for validating numerical 3-D codes. However, before embarking on the heavy task of performing full 3-D simulations of such viscoelastic ows, both under steady and transient conditions, a thorough experimental investigation is required. This is the motivation for the present experimental work, which concentrates on characterizing the ow patterns in squaresquare contractions of elastic uids having a constant shear viscosity.

Figure 1. Flow rig.

PR, pressure regulator; V1 to V6, ball valves; R, reservoir; CL, cylindrical lens; T, long coiled pipe; S, free surface; C, contraction plane.

Experimental Setup
The experimental apparatus is depicted in Figure 1. The rig consisted of two consecutive square ducts (length: 1000 and 300 mm) and a vessel. The sides of the square ducts were 2H1 24.0 mm and 2H2 6.0 mm, respectively, thus dening the 4:1 contraction ratio. The ow rate was set by an adequate control of applied pressure on the upper duct and frictional losses in the long coiled pipe (T in Figure 1) located between the smaller duct and the vessel, at the bottom of the rig. To achieve low ow rates the coiled pipe was 8 m long and had a diameter of 4 mm, whereas for higher ow rates the diameter of the coiled pipe was 6 mm. Applied pressure was kept between 0.5 and 4 bar, and the dashed lines in Figure 1 represent the pressurized air lines. This ow arrangement adequately controlled the ow rate without signicant constrictions, such as valves, that would eventually degrade the poly2912 November 2005

mer molecules. The ow rate was measured by a stopwatch and the passage of the liquid-free surface at two marks in the upper duct. In all tests the uid temperature was measured and the uid properties were taken from the rheometric master curves shown in the next section. To improve the quality of the visualizations the rig was placed inside a dark room. A 10 mW HeNe laser light source was used to visualize the ow patterns. The laser beam passed through a cylindrical lens to generate a light sheet illuminating highly reective tracer particles suspended in the uid. These were 10 m PVC particles (at a concentration of about 15 mg/kg uid) added during the preparation of the uid. The path lines formed were recorded using long time exposure photography with a conventional camera (Canon EOS300 with a macro EF100 mm f/2.8 lens), as sketched in Figure 2, which includes a schematic representation of the test section. The terminal velocity of the PVC particles was negligible: assuming Stokes ow conditions it was estimated to be 0.15 m/s in the less viscous Newtonian uid used (N85), which is three orders of magnitude smaller than the minimum ow bulk velocity. The shear viscosity () and the rst normal stress difference coefcient (1) in steady shear ow, and the storage and loss moduli (G, G) in dynamic shear ow were used to characterize the rheology of the Boger uids. These properties were measured with an AR2000 rheometer from TA Instruments, using a coneplate setup with 40 mm diameter and 2 angle. A falling ball viscometer from Gilmont Instruments (model GV2200) was also used for some viscosity measurements with the Newtonian uid N85.
Vol. 51, No. 11 AIChE Journal

Figure 2. Flow visualization technique and details of the test section.

Rheological Characterization of the Fluids

Four uids based on mixtures of glycerin and water were investigated: two viscous Newtonian uids (N85 and N91) and two viscoelastic Boger uids (PAA100 and PAA300). Their compositions, densities, and zero-shear rate viscosities are listed in Table 2. The Boger uids were prepared by dissolving a small amount of polyacrylamide (PAA; Separan AP30 produced by SNF Floerger) in the Newtonian solvent N91. To minimize the intensity of shear thinning a small amount of NaCl was added, as described in detail in Stokes.28 To reduce bacteriological degradation of the solutions the biocide Kathon LXE, produced by Rohm and Haas, was also added. The uid densities were measured at 21.2C using a picnometer.

Figure 3. Material parameters in steady and dynamic shear ow for the PAA100 uid.
(a) Open symbols for shear viscosity and rst normal stress difference coefcient; solid symbols for G and ; (b) comparison between G and G data (symbols) and tting by three-mode Oldroyd-B model (solid lines).

Newtonian uids
For the N85 uid the measured shear viscosity was 0.125 Pas at 18C, the temperature at which the visualizations took place. For the N91 uid the rheometer was used to measure the shear viscosity at temperatures ranging from 15.9 to 25.0C. This second Newtonian uid, which served as solvent for the PAA solutions, was also used to evaluate the accuracy of the rheometer and to estimate the base noise level in dynamic tests (represented as the dashed baseline of 2G in Figures 3b and 4b). For N1, the measurements at 100 s1 indicated a zero reading within experimental uncertainty, as they should. The uncertainty in measuring N1 was of the order of 10 Pa, in agreement with the specications of the manu 100 s1 facturer for the cone-plate geometry used. For

inertial effects became important, and negative values of N1 were measured. This inertial effect for Newtonian uids was well predicted by the following equation29: N 1,inertia 0.152 R2 (1)

where and R represent the angular velocity and the radius of the cone, respectively. The effect of temperature on the shear viscosity for the N91 uid is well described by an Arrhenius equation, dening a shift factor aT of the form lnaT ln 1 T H 1 T0 R T T0

0 (Pa s)
0.125 (18C) 0.367 (20C) 0.487 (20C) 0.735 (20C)

(2)

Table 2. Composition and Properties of the Fluids (Mass Concentrations)

Designation N85 N91 PAA100 PAA300 PAA (ppm) 100 300 Glycerin (%) 84.99 90.99 90.99 90.97 Water (%) 15.01 7.51 7.50 7.50 NaCl (%) 1.50 1.50 1.50 Kathon (ppm) 25 25 25 25

(kg/m3)
1221 1250 1249 1247

AIChE Journal

November 2005

Vol. 51, No. 11

2913

H/R obtained for the solvent was found to be adequate for the two Boger uids. The reduced rheological quantities (subscript r) thus become as dened by Eqs. 4a 4d32

r T 0 a T T
G r G T 0 G T

r T 0 a T T
G r G T 0 G T

(4a) (4b)

r T 0 T / a T r T 0 T / a T

(4c) r T 0 T / a T
2 1r 1 T 0 1 T / a T

(4d)

Figure 4. Material parameters in steady and dynamic shear ow for the PAA300 uid.
(a) Open symbols for shear viscosity and rst normal stress difference coefcient; solid symbols for G and ; (b) comparison between G and G data (symbols) and tting by three-mode Oldroyd-B model (solid lines).

where (T0) represents the viscosity at the reference absolute temperature T0, here taken as T0 293.15 K. Fitting this equation to the experimental data gave H/R 6860 K and (T0) 0.367 Pas (more details can be found in Alves30). Generally, the shift factor aT is dened as31 aT

0 T T 0 0 0 T 0 T

(3)

where 0(T0) designates the zero shear-rate viscosity at the reference temperature T0, and and 0 are the uid densities at temperatures T and T0, respectively. When the range of temperatures is limited the denition of the shift factor can be simplied by dropping the effects of density and temperature ratios31 and thus Eq. 2 is the form adopted in this work, with (T) 0(T) and (T0) 0(T0).

The reduced steady shear viscosity and rst normal-stress coefcient data (r, 1r vs. r ) are plotted in Figures 3a and 4a for the PAA100 and PAA300 uids, respectively. Both gures also include dynamic shear data in appropriate form ( r , 2G r/ 2 r vs. r) to compare the corresponding limiting behaviors at vanishing deformations, but note their different ordinate scales. The dynamic shear data are plotted separately in Figures 3b and 4b, which include predictions of G and G (solid lines), using a three-mode Maxwell model plus a Newtonian solvent contribution. The parameters of these multimode models are listed in Table 3, at reference temperature T0. The data in Figures 3a and 4a clearly show the limiting behavior of the measured properties in steady and dynamic shear ows. The reduced shear viscosity is approximately constant at reduced shear rates in the range 0.3 to 50 s1 for the PAA100, whereas for the PAA300 it decreases approximately r 54 and 28 s1 10% per decade of reduced shear rate. At (for the PAA100 and PAA300, respectively) there is an abrupt growth in r and 1,r and this is accompanied by a slight reduction in reduced shear rate (with the rheometer operating in controlled stress mode). This phenomenon results from an elastic instability leading to 3D ow, which is frequently observed with Boger uids in coneplate and plateplate geometries, as investigated previously by Phan-Thien33 and McKinley et al.34 More details of this instability for these uids can be found in Alves.30 Figures 3b and 4b show that the three-mode Oldroyd-B model is accurate enough to predict G and G within the measured range, whereas a single-mode model was found to be unable to give accurate predictions of G and G over the whole range of frequencies. Data at low and high frequencies, leading to values of G close to the base line (noise level), were excluded. This baseline, determined as a function of angular speed for deformations of 0.10 and 0.50 using the Newtonian uids N85 and N91 and deionized water, represents the sensitivity of the rheometer and can be used to estimate the experimental uncertainty in G for the viscoelastic uids. The relaxation spectra will be useful later to quantify the
Table 3. Linear Viscoelastic Spectra for Boger Fluids at Reference Temperature (T0 293.15 K)
k (Pa s)
Mode k 1 2 3 Solvent

Boger uids
For the Boger uids the steady shear properties were measured at temperatures ranging from 15.3 to 30.0C. The time temperature superposition principle was found to be valid for both uids and was used to build master curves. The value of
2914 November 2005

k (s)
3.0 0.3 0.03

PAA100 0.075 0.027 0.018 0.367

PAA300 0.23 0.090 0.048 0.367

Vol. 51, No. 11

AIChE Journal

Re2

2H2 U2

(5)

Figure 5. Experimental (left column) and numerical (right column) streaklines for the ow of Newtonian uids N85 and N91 in the middle plane of a squaresquare sudden contraction at T 18.0 0.2C.

where H2 and U2 designate the downstream duct half-width and bulk velocity, respectively. As expected, inertia leads to a reduction of the corner vortex, especially noticeable for Reynolds numbers 1. At low Reynolds numbers, inertial effects are negligible and xR/2H1 asymptotes to 0.163. It is interesting to note that, under creeping ow conditions, the vortex size for a 4:1 circular contraction asymptotes to exactly the same value, xR/2H1 3 0.163, whereas for a 4:1 2-D planar contraction the size is somewhat different, xR/2H1 3 0.1875 (values based on numerical simulations; see Alves et al.30,35). For the circular contraction, Boger10 quotes a value of xR/(2H1) 0.17 0.01 based on both experiments and numerical predictions, in close agreement with our simulations. For Newtonian uids, all the experimental ow features are well captured by numerical simulations, shown on the right column of Figure 5; in addition, the predicted variation of the vortex length with the Reynolds number, shown by the curve in Figure 6, quantitatively matches the measured data. These numerical simulations were obtained with a nite-volume methodology developed by the authors.36 Figure 5 also highlights the point that, even though at rst sight the ow inside the vortex looks 2-D, in reality it is 3-D. In contrast with the planar and axisymmetric sudden contraction ows, none of the recirculations is ever closed in the square contraction, as revealed by a careful inspection of the streaklines. This was also conrmed by visual inspection and especially by the numerical calculations from which the ow description sketched in Figure 7 was outlined. Figure 7 shows streaklines from uid particles starting at two different planes in the upstream duct: the middle plane perpendicular to the wall (ABCD), always seen in the photos, and the second plane (EFGH) at 45 to the wall passing through opposite corners of the square cross section (hereafter referred to as corner plane).

Deborah number and for future numerical simulations of this ow. The Deborah number is used here to quantify elastic effects and is dened identically to the Weissenberg number reported in other works in the literature, based on downstream quantities.

Flow Visualization Results

Newtonian uids Flow visualizations were carried out rst with the Newtonian uids to assess the effect of inertia on the ow structure and to serve as a reference for comparison against the results for the elastic Boger uids. For the 2-D planar and axisymmetric geometries a fair amount of knowledge has been established on the behavior of Newtonian uids,10 although such information is missing for squaresquare contractions. Figure 5 shows photographs of stream traces of the ow of the Newtonian uids in the middle plane of the 3-D sudden contraction and Figure 6 plots the variation of the normalized vortex length (xR/2H1) with Reynolds number, here dened as
AIChE Journal November 2005

Figure 6. Inuence of the Reynolds number on the vortex length at the middle plane of a square square sudden contraction for Newtonian uids N85 and N91.
Comparison between experiments and numerical predictions.

Vol. 51, No. 11

2915

where p is an equivalent relaxation time based on the Oldroyd-B model, and calculated from the linear viscoelastic spectra using the following equations

k solvent

(7)

1 p

k solvent

k k

(8)

Figure 7. 3-D particle trajectories for a Newtonian uid under conditions of negligible inertia.

Particles owing near the walls along the corner plane enter the corner-plane vortex, rotate toward its center, and then drift toward the middle plane along the eye of the recirculating region, in this way entering the vortex located at the middle plane. Once here the particles now rotate toward the periphery of the middle-plane vortex and exit the vortex at the reentrant corner going into the downstream duct. Open 3-D recirculations were also predicted by other authors who simulated Newtonian ows in 3-D sudden expansions37,38 (quasi-2-D as the expansion took place in one dimension only) and in 3-D (quasi-2-D) backward-facing steps.39,40 The inhomogeneous ow conditions experienced by uid particles coming either along the upstream square duct walls or along its corners, into the contraction plane region, result in different velocity gradients, thus leading to stress gradients and complex secondary ows.

In this way, it is guaranteed that at low deformation rates (and low angular velocities), the viscoelastic behavior of the equivalent single mode Oldroyd-B model is identical to that of the multimode model (say, 1,0 lim 30 1 k 2kk 2pp). This denition of relaxation time considers only the role of the polymer additive in the absence of solvent to establish the elasticity of the uid, but it should be noted that some authors prefer to dene a Maxwell relaxation time as 0 1,0/20. Thus, from the data in Table 3 the parameters of the equivalent single-mode Oldroyd-B model are the following: for the PAA100, 0 s p 0.487 Pas, resulting in a solvent viscosity ratio s/0 0.754, and a relaxation time p 1.947 s; for the PAA300, 0 s p 0.735 Pas, s/0 0.499, and p 1.952 s. Clearly, with the relaxation time dened as in Eq. 8 we have almost the same relaxation time for both Boger uids. Because these two polymer solutions are dilute, the congurations/behavior of the individual polymer molecules in the same solvent should not differ signicantly, and so the relaxation times of the molecules should be similar. However, a higher polymer concentration leads to higher elasticity because the uid becomes more viscous and the polymer molecules take longer to relax. This can be quantied in two alternative ways, either by using the total viscosity of the uid to calculate the Maxwell relaxation time, as mentioned above (leading to 0 0.480 and 0.977 s for PAA100 and PAA300, respectively), or by using the so-called elasticity number (El2), dened by the ratio of the Deborah and Reynolds numbers El 2 De2 p 0 2 Re2 2H2 (9)

Boger uids
To quantify the strength of elastic effects with Boger uids it is convenient to use a single relaxation time in the denition of the Deborah number, which can be either based on upstream (De1) or downstream ow conditions (De2). Most studies of contraction ows dene the Deborah number with a deformation rate characteristic of downstream ow conditions (leading to De2). Alves et al.41 showed that lip vortex activity (when present) scales with De2, independently of the contraction ratio (), whereas dimensionless corner vortex size (xR/2H1) and intensity depend on De2/. Thus, hereafter the results are presented with Deborah number based on downstream quantities, De2. Nevertheless, conversion between upstream and downstream Reynolds and Deborah numbers is straightforward: Re2 4Re1; De2 64De1. The downstream Deborah number is dened as De2
2916

p TU2 aT p T0 U2 H2 H2

(6)
November 2005

which is independent of ow kinematics and takes into account the viscosity of the solution: El2 42.2 and 63.9 for the PAA100 and PAA300, respectively. This results in the PAA300 uid being 1.52 times more elastic than PAA100. We note, however, that if the elasticity number is dened as El 2 pp/2H 2 2 (1,0/4H 2 ), to be consistent with the denition of p, then the PAA300 uid is approximately three times more elastic than PAA100 (in agreement with the ratio between the zero shear rate values of the rst normal stress coefcient measured for both uids), a gure more in line with the increase in polymer concentration. In any case, whatever the denition adopted to measure elasticity, the solution of PAA300 is more elastic than that of PAA100 and this effect explains the different ow behaviors reported hereafter. Streak photographies at the middle plane of the contraction are presented in Figure 8 for the ow of the PAA100 uid for increasing values of the ow rate. The ow characteristics are
Vol. 51, No. 11 AIChE Journal

Figure 8. Inuence of elasticity on the streakline ow patterns at the middle plane of a 4:1 square square sudden contraction for PAA100 at T 18.1 0.2C.

complex, but are caused only by elastic effects because inertia is not signicant (Re2 0.95). At relatively low values of the Deborah number (De2 2.65), viscous effects predominate and the ow pattern is similar to that seen in Figure 5 for Newtonian uids, with a concave separation streakline and the 3-D nature of the recirculating ow in evidence. With increasing Deborah number changes progressively occur: rst, there is a very slight increase in vortex size while the separation streakline straightens; then, the corner vortex progressively decreases in size to about a quarter at De2 15, an effect that cannot be attributed to inertia because the Reynolds number remains well below Re2 1. McKinley et al.8 reported results in the 4:1 axisymmetric contraction bearing some resemblance with these: negligible elastic effects for De2( ) 1 (using a shearrate-dependent Maxwell relaxation time) and a decrease in the corner vortex size for Deborah numbers up to De2( ) 3. However, in contrast to McKinley et al. for the axisymmetric contraction, the formation of a strong lip vortex while the corner vortex decreases in size is not seen in the present experiments, although the higher curvature of the streaklines at the reentrant corner suggests the possibility that a weak lip
AIChE Journal November 2005

vortex does exist (see photo for De2 13.4). In fact, the decrease of xR at low De2 is compatible with the existence of a lip vortex,19 although it has been observed that xR usually grows monotonically when such a lip vortex is absent. In addition, the uids used in those references were clearly more elastic than here: in McKinley et al.8 the elasticity number is evaluated as El2 610 and in Rothstein and McKinley12 as El2 2100. As the Deborah number further increases the corner vortex starts to grow, and simultaneously there are changes in the characteristics of the streamlines approaching the contraction plane. For De2 24.1, the approach ow streaklines in the middle plane and close to the contraction are seen to progressively diverge with ow elasticity, moving away from the centerline. This anomalous effect had already been predicted by Oliveira and Pinho42,43 and Alves et al.18 for the ow of Boger uids in a 4:1 plane sudden contraction and was also observed in the experiments of McKinley et al.8 for the ow of a Boger uid in circular contractions. Attempts to explain the causes for divergence of streamlines upstream of the contraction for Boger uids have pointed to a local intense increase in extensional viscosity leading to an increased ow resistance in the centerline just upstream of the contraction plane as the extension rates grow in a region of predominantly extensional ow characteristics. Such extensional thickening is a characteristic of Boger uids, but there are not enough data to correlate rheological behavior with diverging ow. McKinley et al.8 discuss this issue and mention different degrees of ow divergence for uids with similar extensional viscosity behavior, arguing for the relevance of the total Hencky strain to this
Vol. 51, No. 11 2917

ow feature. However, the total Hencky strain is here unchanged because it is a function of the contraction ratio. Also of note here, the diverging streamlines with Boger uids were observed in the absence of noticeable lip vortex activity. This contrasts with the visualizations of McKinley et al.8 in the axisymmetric geometry, but agrees with their conclusions that diverging streamlines and lip vortex formation are probably unrelated. As discussed more recently by Rothstein and McKinley,12 extensional viscosity alone does not explain the variety of ow features in contraction ows; hysteresis effects on the stressconformation, normal stress effects in shear and extension, as well as contraction ratio and transient extensional viscosity behavior are relevant to the existence or not of a lip vortex and to the enhanced pressure drop, so the issue of diverging streamlines should be at least as complex. In fact, the early experiments of Cable and Boger4-6 had already shown strong diverging ow in the presence of combined shear-thinning and inertial effects, suggesting that inhomogeneous shear conditions near the wall are equally important. Quite interestingly, in the experiments of Walters and Rawlinson24 in a 13.3:1 squaresquare contraction, there is an abrupt expansion of the central jet as the uid is just about to enter the smaller duct (cf. their Figure 5 at W 0.002), but there is already an enhanced vortex and the diverging streamlines may be related in part to the uid particles being dragged into the recirculation, something that we did not observe in our experiments. The growth of the large vortex continues with elasticity, and the ow remains steady up to De2 52. At higher Deborah numbers, the 3-D nature of the recirculating ow seems to have changed, as can be assessed by careful comparison of streaklines in the recirculation at De2 50.4 with those taken at low Deborah numbers. As the Deborah number further increases the ow becomes periodic, possibly because of an elastic instability, and this is observed by the crossings in some streaklines at De2 54.2. The amplitude of the oscillations increases with De2, as seen in the three photos of Figure 9 taken at three moments in time within a cycle for two different supercritical ow conditions. At even higher ow rates the ow loses its periodicity. For the more concentrated Boger uid, PAA300, the streakline ow visualizations are shown in Figures 10a and 10b for conditions corresponding to two different temperatures, T 21.0 and 17.5C. The second set was taken 3 months after the rst, with the same uid sample (it was stored in a refrigerator), and with the purpose of checking the repeatability of the ow and absence of uid degradation over time under the tested conditions. In general terms, the inuence of Deborah number is the same as for PAA100, but a couple of important differences are worth mentioning. First, the increase of the corner vortex located on the axial midplane at low Deborah numbers is also observed, but it is now more intense than that with the PAA100 uid. This is clear in Figure 11, which compares the variation for both uids of the vortex size in the midplane against the Deborah number (De2): for PAA300 xR/2H1 peaks by more than 25% relative to the Newtonian value at De2 6, followed by a decrease to about 0.07 in the range De2 6 to 15. A second more important difference in relation to PAA100 is that a lip vortex is now clearly seen in the midplane for a certain range of Deborah numbers (see photos at De2 8.70 and 9.62 in Figure 10). In particular, at De2 9.62 when
2918 November 2005

Figure 9. Streaklines for the ow of PAA100 in the middle plane of a squaresquare contraction at three different moments within two different oscillating supercritical ow conditions.

inertia is still negligible, the lip vortex has similarities to that found for Boger uids in the 4:1 plane contraction calculations of Alves et al.18 (cf. their Figure 7 at De2 4 for UCM uids). It is known from experimental and numerical work in the 4:1 plane sudden contraction ow with Boger uids,23,44 that a lip vortex is formed at low to moderate Deborah numbers. This lip vortex grows with elasticity and eventually dominates the corner vortex, the characteristic ow feature of Newtonian ows and low-Deborah number viscoelastic ows. In contrast, for the axisymmetric sudden contraction ow of Boger uids the corner vortex is normally found to grow with elasticity without the presence of any lip vortex35,45; there are exceptions to this Boger uid behavior, as in the experiments of Boger et al.9 and McKinley et al.8 with PIB/PB solutions. Boger and Binnington46 also observed a lip vortex mechanism for a PAAbased uid in a 4:1 circular contraction with rounded corners, in deep contrast with the (usual) strong corner vortex enhancement observed in the same 4:1 circular contraction, but with a sharp reentrant corner. Rothstein and McKinley12 argued about the competing roles of extensional stresses and shear-induced normal stresses on vortex growth mechanisms, and concluded that lip vortices are associated with a domination of shearVol. 51, No. 11 AIChE Journal

Figure 10. Inuence of elasticity on the streakline ow patterns at the middle plane of a 4:1 square square sudden contraction for PAA300.
(a) Visualizations at 21.0C; (b) visualizations at 17.5C.

induced elasticity, whereas corner vortex growth is dominated by extensional stresses, thus predominating at large contraction ratios and/or large Deborah numbers. Given the similarities between the squaresquare contraction and the axisymmetric geometry in terms of both extensional strain rates developed along the centerline and the possibility of partial balance of cross-stream/azimuthal pressure and stress gradients, it comes as no surprise that the visualizations of Evans and Walters16 for a squaresquare 16:1 contraction do not show any lip vortex and instead illustrate a strong enhancement of the middle plane corner vortex (see their Figure 11). The presence of cross-stream secondary ows in the square contraction increases shear-induced normal stresses relative to those found in the corresponding circular contraction ow and, according to the above mechanism of Rothstein and McKinley, this will widen the range of contraction ratios where the lip
AIChE Journal November 2005 Vol. 51, No. 11 2919

Figure 11. Variation of normalized vortex length with ow elasticity for Boger uids PAA100 and PAA300.

vortex is observed. Such an effect is similar to rounding the reentrant corner in axisymmetric contractions: Rothstein and McKinley12 report for this case the existence of a lip vortex for the 4:1 contraction and a delay in corner vortex development, conrming the observations of Boger and Binnington.46 At higher ow rates/elasticity (De2 20) there is vortex enhancement through a lip vortex mechanism. A diverging streakline pattern upstream of the contraction plane is also observed, but now in the presence of a lip vortex. The appearance of diverging streaklines in a 3-D squaresquare contraction ow of Boger uids, with and without lip vortex, is documented here for the rst time and conrms the suggestion of McKinley et al.8 that the two phenomena are probably unrelated. Simulations by Oliveira and Pinho42 for creeping ow of UCM uid in a 4:1 planar contraction corroborate this point, and the numerical results presented in Oliveira and Pinho43 for the same ow under conditions where inertia is not negligible (Re2 1) illustrate that inertial effects enhance streamline divergence for high Deborah number ows. At even higher Deborah numbers, say for De2 46.7 (De1 0.726), the ow of PAA300 becomes periodic, showing a behavior similar to that reported above for the PAA100 uid. This periodic ow results from an elastic instability and occurs earlier than for the PAA100 solution, at De2 45 (De1 0.7). Figure 12 shows three photos with the PAA300 solution pertaining to a cycle of events for two different supercritical ow

Figure 12. Streaklines for the ow of PAA300 in the middle plane of a squaresquare contraction at three different moments within oscillating supercritical ow conditions.
(a) De2 55.1 and Re2 0.577; (b) De2 61.5 and Re2 0.643.

2920

November 2005

Vol. 51, No. 11

AIChE Journal

conditions, conrming that the time-average size of the timedependent vortex continues to grow with Deborah number.

Conclusions
Flow visualizations were carried out in the middle plane of a 4:1 squaresquare sudden contraction for Newtonian and Boger uids under conditions of negligible inertia, using streakline photography. The Newtonian ow patterns were in good agreement with our own numerical results and show inertia to be negligible for Reynolds numbers below Re2 1. The ow eld was clearly three-dimensional, exhibiting open vortices with both Newtonian and viscoelastic uids, and inertia tended to push the corner vortex toward the contraction plane. For the two Boger uids the upstream vortices were seen to increase slightly at low Deborah numbers, before an intense decrease took place leading to a minimum vortex size at De2 15. As the Deborah number further increased, the streaklines on the central region of the approaching ow started to diverge, whereas the vortices grew strongly with elasticity until the ow became periodic and eventually chaotic at higher ow rates. For the more concentrated Boger uid (PAA300) these effects were more intense because of the corresponding higher elasticity number, and a major difference in ow features was clearly seen: after the vortex attained its minimum size at intermediate Deborah numbers, 10 De2 15, a lip vortex appeared and grew with elasticity. For the PAA100 solution, and although only corner vortex enhancement was observed, the variation of vortex size with Deborah number was compatible with the existence of a weak lip vortex (not captured with our visualization technique). We speculate that the secondary ow in the cross section of the rectangular channel tends to increase the role of shearinduced normal stresses, thus leading to the appearance of a lip vortex at this contraction ratio and consequently delaying the vortex growth and instability to higher Deborah numbers, compared to the corresponding situation of a circular contraction. Such ow features, and the existence of a diverging ow well upstream of the contraction, are reported here for the rst time in relation to the squaresquare contraction ow of Boger uids. However, similar ow features have been previously reported for circular contractions.

Acknowledgments
The authors thank Prof. M. P. Gonc alves and D. Torres for help in characterizing the rheology of the uids used in this work. M. A. Alves thanks Fundac a o Calouste Gulbenkian for nancial support. Sponsorship of FEDER by the FCT program POCTI/EME/37711/2001 is gratefully acknowledged. Finally we are grateful for the relevant comments of one of the reviewers whose helpful suggestions signicantly improved this article.

Literature Cited
1. Van Dyke M. An Album of Fluid Motion. Stanford, CA: The Parabolic Press; 1982. 2. Boger DV, Walters K. Rheological Phenomena in Focus. Amsterdam: Elsevier; 1993. 3. Hassager O. Working group on numerical techniques. Fifth International Workshop on Numerical Methods in Non-Newtonian Flows, Lake Arrowhead, USA. J Non-Newt Fluid Mech. 1988;29:2-5. 4. Cable PJ, Boger DV. A comprehensive experimental investigation of tubular entry ow of viscoelastic uids: Part I. Vortex characteristics in stable ow. AIChE J. 1978;24:868-879.

5. Cable PJ, Boger DV. A comprehensive experimental investigation of tubular entry ow of viscoelastic uids: Part II. The velocity elds in stable ow. AIChE J. 1978;24:992-999. 6. Cable PJ, Boger DV. A comprehensive experimental investigation of tubular entry ow of viscoelastic uids: Part III. Unstable ow. AIChE J. 1979;25:152-159. 7. Nguyen H, Boger DV. The kinematics and stability of die entry ows. J Non-Newt Fluid Mech. 1979;5:353-368. 8. McKinley GH, Raiford WP, Brown RA, Armstrong RC. Non linear dynamics of viscoelastic ow in axisymmetric abrupt contractions. J Fluid Mech. 1991;223:411-456. 9. Boger DV, Hur DU, Binnington RJ. Further observations of elastic effects in tubular entry ows. J Non-Newt Fluid Mech. 1986;20:31-49. 10. Boger DV. Viscoelastic ows through contractions. Ann Rev Fluid Mech. 1987;19:157-182. 11. Rothstein JP, McKinley GH. Extensional ow of a polystyrene Boger uid through a 4:1:4 axisymmetric contraction/expansion. J Non-Newt Fluid Mech. 1999;86:61-88. 12. Rothstein JP, McKinley GH. The axisymmetric contraction-expansion: The role of extensional rheology on vortex growth dynamics and the enhanced pressure drop. J Non-Newt Fluid Mech. 2001;98:33-63. 13. Phillips SD, Joo YL, Brown RA, Armstrong RC. Modeling and simulation of the ow through an axisymmetric contraction-expansion using a multi-mode Adaptive Length Scale model, Paper NF25, Proc of the XIV International Congress on Rheology (in CD-ROM), Seoul, Korea; 2004. 14. Ghosh I, Joo YL, McKinley GH, Brown RA, Armstrong RC. A new model for dilute polymer solutions in ows with strong extensional components. J Rheol. 2002;46:1057-1089. 15. Walters K, Webster MF. On dominating elastico-viscous response in some complex ows. Philos Trans R Soc Lond A Phys Sci. 1982;308: 199-218. 16. Evans RE, Walters K. Flow characteristics associated with abrupt changes in geometry in the case of highly elastic liquids. J Non-Newt Fluid Mech. 1986;20:11-29. 17. Evans RE, Walters K. Further remarks on the lip-vortex mechanism of vortex enhancement in planar contraction ows. J Non-Newt Fluid Mech. 1989;32:95-105. 18. Alves MA, Pinho FT, Oliveira PJ. Effect of a high-resolution differencing scheme on nite-volume predictions of viscoelastic ows. J Non-Newt Fluid Mech. 2000;93:287-314. 19. Alves MA, Oliveira PJ, Pinho FT. Benchmark solutions for the ow of Oldroyd-B and PTT uids in planar contractions. J Non-Newt Fluid Mech. 2003;110:45-75. 20. White SA, Baird DG. The importance of extensional ow properties on planar entry ow patterns of polymer melts. J Non-Newt Fluid Mech. 1986;20:93-101. 21. White SA, Baird DG. Flow visualization and birefringence studies on planar entry ow behaviour of polymer melts. J Non-Newt Fluid Mech. 1988;29:245-267. 22. White SA, Baird DG. Numerical simulation studies of the planar entry ow of polymer melts. J Non-Newt Fluid Mech. 1988;30:47-71. 23. Nigen S, Walters K. Viscoelastic contraction ows: Comparison of axisymmetric and planar congurations. J Non-Newt Fluid Mech. 2002;102:343-359. 24. Walters K, Rawlinson DM. On some contraction ows for Boger uids. Rheol Acta. 1982;21:547-552. 25. Xue SC, Phan-Thien N, Tanner RI. Numerical investigations of Lagrangian unsteady extensional ows of viscoelastic uids in 3-D rectangular ducts with sudden contractions. Rheol Acta. 1998;37:158169. 26. Mompean G, Deville M. Unsteady nite volume simulation of Oldroyd-B uid through a three-dimensional planar contraction. J NonNewt Fluid Mech. 1997;72:253-279. 27. Xue SC, Phan-Thien N, Tanner RI. Three dimensional numerical simulations of viscoelastic ows through planar contractions. J NonNewt Fluid Mech. 1998;74:195-245. 28. Stokes JR. Swirling Flow of Viscoelastic Fluids. PhD Dissertation. Melbourne, Australia: Dept. of Chemical Engineering, University of Melbourne; 1998. 29. Barnes HA. A Handbook of Elementary Rheology. Aberystwyth, Wales: Institute of Non-Newtonian Fluid Mechanics, University of Wales; 2000. 30. Alves MA. Laminar Flow of Viscoelastic Fluids: Numerical, Theoret-

AIChE Journal

November 2005

Vol. 51, No. 11

2921

31. 32. 33. 34.

35. 36. 37. 38. 39.

ical and Experimental Analysis (in Portuguese). PhD Dissertation. Porto, Portugal: Dept. of Chemical Engineering, University of Porto; 2004. Bird RB, Armstrong RC, Hassager O. Dynamics of Polymeric Liquids. Volume 1: Fluid Dynamics. New York, NY: Wiley; 1987. Quinzani LM, McKinley GH, Brown RA, Armstrong RC. Modeling the rheology of polyisobutylene solutions. J Rheol. 1990;34:705-748. Phan-Thien N. Cone-and-plate ow of the Oldroyd-B uid is unstable. J Non-Newt Fluid Mech. 1985;17:37-44. McKinley GH, Byars JA, Brown RA, Armstrong RC. Observations on the elastic instability in cone-and-plate and parallel-plate ows of a polyisobutylene Boger uid. J Non-Newt Fluid Mech. 1991;40:201229. Alves MA, Oliveira PJ, Pinho FT. Numerical simulation of viscoelastic contraction ows. In: Bathe KJ, ed. Computational Fluid and Solid Mechanics. Amsterdam: Elsevier; 2003:826-829. Alves MA, Oliveira PJ, Pinho FT. A convergent and universally bounded interpolation scheme for the treatment of advection. Int J Numer Methods Fluids. 2003;41:47-75. Chiang TP, Sheu TWH, Wang SK. Side wall effects on the structure of laminar ow over a plane-symmetric sudden expansion. Comput Fluids. 2000;29:467-492. Schreck E, Scha fer M. Numerical study of bifurcation in three-dimensional sudden channel expansions. Comput Fluids. 2000;29:583-593. Williams PT, Baker AJ. Numerical simulations of laminar ow over a 3D backward-facing step. Int J Numer Methods Fluids. 1997;24:11591183.

40. Biswas G, Breuer M, Durst F. Backward-facing step ows for various expansion ratios at low and moderate Reynolds numbers. ASME J Fluids Eng. 2004;126:362-374. 41. Alves MA, Oliveira PJ, Pinho FT. On the effect of contraction ratio in viscoelastic ow through abrupt contractions. J Non-Newt Fluid Mech. 2004;122:117-130. 42. Oliveira PJ, Pinho FT. Plane contraction ows of Upper Convected Maxwell and Phan-ThienTanner uids as predicted by a nite-volume method. J Non-Newt Fluid Mech. 1999;88:63-88. 43. Oliveira PJ, Pinho FT. Numerical procedure for the computation of uid ow with arbitrary stressstrain relationships. Num Heat Transfer B. 1999;35:295-315. 44. Walters K, Webster MF. The distinctive CFD challenges of computational rheology. Int J Numer Methods Fluids. 2003;43:577-596. 45. Sasmal GPA. Finite-volume approach for calculation of viscoelastic ow through an abrupt axisymmetrical contraction. J Non-Newt Fluid Mech. 1995;56:15-47. 46. Boger DV, Binnington RJ. Experimental removal of the re-entrant corner singularity in tubular entry ows. J Rheol. 1994;38:333-349. 47. Purnode B, Crochet MJ. Flows of polymer solutions through contractions. Part 1: Flows of polyacrylamide solutions through planar contractions. J Non-Newt Fluid Mech. 1996;65:269-289. 48. Quinzani LM, Armstrong RC, Brown RA. Birefringence and laserDoppler velocimetry (LDV) studies of viscoelastic ow through a planar contraction. J Non-Newt Fluid Mech. 1994;52:1-36.
Manuscript received Oct. 8, 2004, and revision received Mar. 12, 2005.

2922

November 2005

Vol. 51, No. 11

AIChE Journal

Experimental Approaches for Understanding Mixing Performance of a Minireactor

G. G. Chen, G. S. Luo, S. W. Li, J. H. Xu, and J. D. Wang
The State Key Laboratory of Chemical Engineering, Dept. of Chemical Engineering, Tsinghua University, Beijing, 100084, China DOI 10.1002/aic.10539 Published online August 24, 2005 in Wiley InterScience (www.interscience.wiley.com).

Three experimental approaches are presented to determine the mixing characteristics of a new kind of high-efciency membrane dispersion minireactor in which microltration membranes are applied as the dispersion media. Residence time distribution (RTD) curves were measured for determining the macromixing characteristics. A typical Dushman reaction (iodideiodate) coupled with a neutralization and precipitation reaction of BaSO4 were introduced to characterize the micromixing performance of a single-phase mixing process. A dye extraction method was also applied to study the micromixing performance of a two-liquid phase-mixing system. The RTD result showed that the ow performance in the minireactor was almost in the plug ow condition. The micromixing performance was expressed with a segregation index, which could be 0.002. The single-phase micromixing performance reached the desired level. The result of precipitation of BaSO4 showed that the mixing performance had a substantial inuence on the particle size and size distribution. It was found that in the single-phase mixing process the mixing performance was mainly inuenced by the phase ux and the membrane pore size. With decreasing dispersed uid ux or the membrane pore size, or with increasing continuous uid ux, the micromixing performance was enhanced. The dye extraction method can correctly determine the mixing performance of a two-liquid phase-mixing system. Unlike the single-phase mixing process, in the two-liquid phase-mixing process there was a minimum value for the mixing efciency with the change of the continuousor dispersed-phase ux, when the membrane pore size was 0.9 m. 2005 American
Institute of Chemical Engineers AIChE J, 51: 29232929, 2005

Keywords: membrane dispersion, micromixing, macromixing

Introduction
Mixing, a widely used unit operation process in the chemical industry, has a decisive impact on the overall performance of the processing engineering. Mixing can be classied as micromixing (mixing at the molecular scale), mesomixing, and macromixing. Many methods have been developed over the last few years to describe the mixing performance in different devices. The most popular of broadly considered methods of
Correspondence concerning this article should be addressed to G. S. Luo at gsluo@tsinghua.edu.cn.

2005 American Institute of Chemical Engineers

measuring macromixing performance is to elucidate the residence time distribution (RTD). Residence times and residence time distributions are important characteristics for all chemical reactors because they provide information about macromixing and the ow behavior in the respective reactors.1,2 Some authors3-6 have reviewed the characteristics of micromixing and its available measurement methods. Almost all the developed methods are based on mixing-sensitive chemical conversions, and can be classied into three main schemes: single reaction (A B 3 R), consecutive competing reactions (A B 3 R; R B 3 S), and parallel competing reactions (A B 3 R; C B 3 S). In the latter two schemes, yield and/or selectivity, achieved for specic reaction products, constitute the parameVol. 51, No. 11 2923

AIChE Journal

November 2005

understand the mixing performance of the membrane dispersion minidevice, several experimental approaches have been introduced. The macromixing performance was characterized with RTD curves. The micromixing performance of the single phase was detected by a parallel competing reaction system and the liquidliquid phase-mixing process was denoted by a dyeextraction method. Moreover, the mixing performance was further tested by the preparation of BaSO4 nanoparticles.

Figure 1. Minireactor.

Experimental Equipment and Approaches Minireactor

An illustration of the membrane dispersion minireactor used in this work is shown in Figure 1. It can be seen that there are three main sheets in the minireactor: two stainless steel sheets and a membrane sheet. The dispersed solution inlet and dispersed solution chamber are in one sheet. The continuous solution inlet, mixing chamber, and slurry outlet are in the other sheet. The three sheets are assembled with bolts. In this minireactor, the dispersed solution (phase) is pressed through a microltration membrane into the mixing chamber to mix with another miscible solution or immiscible phase. The geometric size of the mixing chamber was 12 4 1 mm. The Ni microltration membrane with pore sizes of 0.2 and 0.9 m, and stainless steel membranes with a pore size of 5 m were applied, respectively.

ters for quantifying the mixing performance, of which the Villermaux/Dushman method has gained wide acceptance. The published results show that the micromixing performance can be determined very well when the volume ratio between the two miscible solutions is very large. On the other hand, most of the methods are designed to describe the single-phase mixing process in which one miscible solution is mixed into another solution. For the multiphase-mixing processes only few results can be referenced at the present time. In 2004 a dye extraction method7 was developed to explain a liquidliquid mixing process, which is based on the phase transfer of the solvatochromatic dye Nile red from a water/methanol phase to a heptane phase. Over the past decades, microstructured devices have triggered an explosion of scientic and industrial interest. The rapid development of microchemical systems has led to a considerable variety of microstructured devices, especially for microreaction and micromixing processes. Many researchers focused on the micromixing performances in different microstructured micromixers.7-11 The micromixing characteristic is considered to be the basis for development of new microchemical systems. Ehrfeld and coworkers 8 studied the micromixing performance in several micromixers with a competing reaction system, and compared it with that in the traditional macroscopic mixing units. Panic et al.7 provided an improved experimental protocol for the Dushman reaction by controlling the nal pH value in the reactors, which supplied more reliable experimental data than those in Ehrfeld et al.8 They also described the mixing performance of ve micromixers whose mixing principle and internal structure geometry were different. Two experimental methods of competing parallel reaction and dye extraction were introduced. In our laboratory a new kind of microstructured device, designated membrane dispersion minireactor, was developed, in which microltration membranes are applied as the dispersion media.12-16 In the reactor, single-phase mixing and multiphase-mixing processes can be successfully performed. When a liquidliquid mixing process is carried out, the dispersion phase is dispersed as small droplets into the continuous phase within a very short time. Previous works demonstrated that this type of minireactor, like most of the microstructured devices, has the characteristics of high mass transfer efciency, large capacity, low energy cost, and controllability.14 The membrane dispersion minireactor has been used in liquidliquid extraction processes, such as oil deacidication,15 extraction of citric acid and succinic acid,14,16 and in the preparation process of nanoparticles such as BaSO412 and TiO2.13 However, the mixing performance of the new device has not been tested. To better
2924 November 2005

Measurement of RTD curves

Residence time distribution curves can provide necessary information for better understanding of the macromixing performance in chemical reactors.1,2 In this work, a stimulus response method was applied to determine the residence time, and the experimental setup17,18 is shown in Figure 2. A solution of methylic orange was selected as the tracer, and the dilute aqueous solution of methylic orange was injected into the reactor with a six-way valve (Waters automated switching valve). Two online ultraviolet (UV) spectrometers (8823A, Beijing Institute of New Technology Application, China) were set at the inlet and outlet to detect the concentration.

Parallel competing reaction for single-phase mixing

Micromixing was studied by means of a parallel competing reaction: the Dushman reaction between iodide and iodate coupled with a neutralization reaction.3,4 The reaction formulas are shown as follows
H2BO3 H 3 H3BO3

quasi-instantaneous fast

(1) (2)

6H 3 3I2 3H2O 5I IO3

Figure 2. Experimental setup for measurement of RTD curve. Vol. 51, No. 11 AIChE Journal

ticles of barium sulfate were synthesized. The particles were separated by centrifugation. After being washed ve times with distilled water and once more with absolute ethyl alcohol, and then dried in an oven at 80C for 12 h, the barium sulfate nanoparticles were obtained. The morphology of the nanoparticles was characterized by SEM (JEM-6301F, JEOL, Tokyo, Japan) and TEM (JEM200CX, JEOL) images.

Dye-extraction experiment for liquidliquid mixing

The micromixing performance of a liquidliquid mixing process in the new minireactor was measured with a dyeextraction method.7 The solvatochromatic dye Nile red was selected as the indicator, water/methanol as the aqueous phase, and n-heptane as the oil phase. Nile red was extracted from the aqueous phase to the oil phase. In the membrane dispersion minireactor, the dispersed phase was dispersed as microdroplets into the continuous phase through the microltration membrane, which provided a substantial interface area for mass transport in the two-liquid mixing process. The two phases were pumped into the minireactor from the continuous-phase inlet and dispersed-phase inlet, respectively. The mixed oil-inwater or water-in-oil suspension owed out from the outlet. After quick phase separation, the Nile red concentration in the organic phase was measured with a spectrophotometer (UV vis recording spectrophotometer, UV-8345, HP) at 489 nm and the amount of the extracted dye was calculated.

Figure 3. Residence time distribution curves (0.9 m membrane).

I2 I 7 I3

(3)

Many researchers have studied the kinetics of these reactions.19 The fractional yield of I2 and I 3 were measured by spectrophotometry (UVvis recording spectrophotometer, UV-8345, HewlettPackard) at 353 and 286 nm. A segregation index, XS, is dened to explicitly quantify the micromixing quality,4 which is dened as XS
nI2 nI3 Y YST nH,0

6nIO3 ,0 nH2BO3 ,0 3nIO3 ,0

) V 6CIO3 (CI2CI3 ,0 CH2BO3 ,0

nH,0

3CIO3 ,0

Results and Discussion Macromixing performance

(4) Figures 3 and 4 show the stimulusresponse curves obtained under various dispersion-phase uxes. The response curves at the outlet are very similar to those stimulus curves at the inlet, indicating that the uid passed through the minireactor in an ideal plug-ow mode. The results mean that the residence time is the function of just two phase ow rates. The axial mixing coefcients are very small, and uneven ow could almost be ignored.

In Eq. 4, Y is the ratio of acid mole number consumed by reaction 2 divided by the total acid mole number injected; YST is the value of Y in total segregation case when the micromixing process is innitely slow. So the value of XS lies between 0 and 1; for perfect micromixing, XS 0, and in a totally segregated medium, XS 1. The experiment setup is shown in Figure 1. In the singlephase mixing process, two reactant solutions, diluted sulfuric acid solution and a solution containing the other reactants, were prepared. The acid solution was pumped into the dispersedphase chamber, and the other solution was pumped directly into the mixing chamber. The ow rates were controlled by two metering pumps. The triiodide complex formed in the minireactor was detected by sampling at the outlet.

Micromixing performance of the single-phase mixing process

Equation 4 indicates that Xs is a function of transfer rate of H. Xs will increase if the transfer rate of H decreases, which

Nanoparticle preparation
The precipitation of barium sulfate has been investigated by experimental method to determine the inuence of the mixing performance in the reactors.20-23 Precipitation experiments with barium sulfate show some inuence of micromixing intensity on the particle size and of macromixing on particle morphology.23 Two reactant solutions of sodium sulfate and barium chloride were pumped into the minireactor with two metering pumps, and the two miscible solutions were micromixed in the reactor with sodium sulfate solution as the continuous uid. Because the reaction system was at supersaturation, nanoparAIChE Journal November 2005

Figure 4. Residence time distribution curves (5 m membrane). Vol. 51, No. 11 2925

means more segregated medium to be reached. The effect of the dispersed uid ux on the segregation index Xs is shown in Figure 5. Xs decreased with an increase of the dispersed uid ux. With the increase of the dispersed uid ux, the turbulence intensity and the mass transfer efciency of H were improved. Then, the segregation index Xs decreases, and the mixing performance becomes enhanced. The mixing performance in the minireactor is determined not only by micromixing, but also by meso- or macromixing. When the ux ratio is small, the mixing performance is mainly determined by micromixing, although with increasing ux ratio the role of meso- or macromixing to the reactors performance becomes increasingly more important. Usually the ux ratio is 0.1 for determining micromixing performance in normal stirring reactors. However, the ux ratio in this work ranged from 0.1 to 10.0, which is always applied in real industry processes. Therefore the obtained data in this work are much more reliable and comparable for designing and scaling-up the equipment. Figure 6 shows the effect of the continuous uid ux on the segregation index. As the continuous uid ux increases, the segregation index decreases, which indicates that the mixing performance becomes enhanced. The segregation index could be 0.002, which means an almost perfect mixing state. Increasing continuous uid ux results in a decreasing residence time, which does not confer any benet to the mixing, but

Figure 6. Effect of continuous uid ux on segregation index.

Figure 5. Effect of dispersed uid ux on segregation index. 2926 November 2005 Vol. 51, No. 11 AIChE Journal

Figure 7. Effect of membrane pore size on micromixing.

rather results in decreasing ux ratio and increasing turbulence intensity, two phenomena that are of substantial benet to micromixing. Meanwhile an increase in continuous uid ux could reduce the mixing size, and the mixing process would be predominantly governed by the micromixing characteristic. If the micromixing characteristic is robust, the mixing performance of the minireactor will be closer to the perfect mixing state. On the other hand, it can be seen that the segregation index is in the range of 0.010 0.032, shown in Figures 5 and 6. Thus, it is possible to conclude that the micromixing performance of the minireactor is much better than that of normal reactors, and the micromixing performance can be improved through increasing the continuous uid ux and the dispersed uid ux. Figure 7 shows the effect of the membrane pore size on the mixing performance, which was substantially inuenced by the membrane pore size. When the pore size was 0.9 or 5 m, the mixing performance under the same experimental conditions was similar. When the pore size was 0.2 m, the mixing performance was enhanced to an even greater degree. Obviously the membrane pore size will inuence the mixing size. With a decrease of the pore size the mixing size decreases, and thus a better mixing state could be attained. The curves in Figure 7 also show that the micromixing performance can be further improved by increasing the continuous uid ux.

Figure 9. Inuence of membrane pore size on average size of BaSO4 nanoparticles.

Precipitation of nanoparticles
Figure 8 is a SEM microphotograph of BaSO4 nanoparticles prepared by mixing 100 mol/m3 BaCl2 solution with 400 mol/m3 Na2SO4 solution. When the continuous uid ux and the dispersed uid ux were 0.40 and 0.37 cm3/s, respectively, the particles prepared in the membrane dispersion reactor were in the sub200-nm range, which is much smaller than the size of those (range: 110 m) precipitated in the normal Taylor Couette reactor or other classic stirred tank reactors.20-23 The size of the particles is affected by the micromixing performance.23 The better performance of micromixing results in smaller particles. Thus compared against other reactors, the new minireactor has a potentially higher micromixing performance. In our previous work,12 the relationship between the operation conditions and the average particle size was investigated. The particle size decreases with increasing continuous uid ux. An increase of the continuous uid ux results in both increasing supersaturation and improvement of the micromixing, both of which favor precipitation of nanosized particles. The inuence of the membrane pore size on the average

Figure 8. SEM images of BaSO4 particles.

BaCl2: 100 mol/m3, 0.37 cm3/s dispersed uid; Na2SO4: 400 mol/m3, 0.40 cm3/s continuous uid; 0.2 m membrane.

Figure 10. Inuence of membrane pore size on size-distribution of BaSO4 nanoparticles.

Dispersed phase: BaCl2, 100 mol/m3, 0.37 cm3/s; continuous phase: Na2SO4, 400 mol/m3, 0.40 cm3/s.

AIChE Journal

November 2005

Vol. 51, No. 11

2927

Figure 11. Absorbance at 489 nm with organic phase as the dispersed phase.

Figure 13. Absorbance at 489 nm with aqueous phase as the dispersed phase.

sizes and particle size distribution of barium sulfate is shown in Figure 912 and Figure 10. It can be seen that the average particle sizes decrease and the particle size distributions become narrower with decreasing membrane pore size. As the membrane pore size decreases, the micromixing performance is vastly improved, resulting in improved morphology and size distribution of the nanoparticles.

Micromixing performance of liquidliquid mixing process

The processes of multiphase mixing are always encountered in unit operation and chemical reaction engineering. To understand the mixing performance of multiphase-mixing processes, a liquidliquid mixing process in the minireactor was carried out. After phase separation, the absorbance of the organic phase was measured. The results with the oil phase used as the dispersed phase are plotted in Figures 11 and 12. The inuences of the membrane pore size are also shown in these gures. When the membrane pore size was 0.2 m, the mixing efciency of the two-liquid phase-mixing process was similar to that of the single-phase mixing process, and was increased with increasing ow uxes. However, when the membrane pore size was 0.9 or 5 m, unlike the single-phase mixing process, the mixing efciency was initially decreased with an increase in the continuous-phase ux. After a minimum value was reached, it was increased. The results are similar to the

mass-transfer performances of a membrane dispersion extractor 24 and the results in Panic et al.7 and Benz et al.25 In the two-liquid phase-mixing process the mixing efciency is a function of the mixing condition, mixing size (droplet size), and overall mass-transfer coefcients. When the membrane pore size is smaller, the mixing size decreases and the masstransfer coefcients increase with the continuous-phase ux. The mixing becomes more uniform, and thus the mixing efciency is higher. However, when the membrane pore size is larger, the mixing size becomes larger. Therefore, the mixing conditions deteriorate and mixing efciency decreases. As the continuous-phase ux increases, both the mixing size and the residence time will decrease, although the mass-transfer coefcient will increase. At a lower ux value the mixing condition becomes worse, and the mixing efciency is predominantly controlled by the residence time. When the ux was 0.6 cm3/s, the mixing condition became more uniform and the overall mass-transfer coefcient was higher. Thus, the mixing efciency was increased with the continuous-phase ux. Figures 13 and 14 show the experimental results of the two-liquid phase-mixing process when the aqueous phase was used as the dispersed phase. The results in Figures 13 and 14 are similar to those in Figures 11 and 12. The mixing efciency for using the aqueous phase as the dispersed phase was lower than that for using the organic phase as the dispersed phase. The mixing size (droplet size) is affected not only by the membrane pore size, but also by the systems physical prop-

Figure 12. Extraction efciency with organic phase as the dispersed phase. 2928 November 2005

Figure 14. Extraction efciency with aqueous phase as the dispersed phase. Vol. 51, No. 11 AIChE Journal

erties and the wetting ability of the membrane to the two phases. When the dispersed phase was changed from the oil phase to the aqueous phase, the droplet size became larger as a result of the stronger wetting ability of the membranes to the aqueous phase. This is the possible reason for the deterioration of the mixing performance.

4.

5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25.

Conclusions
The mixing performances of a novel kind of membrane dispersion minireactor have been investigated with various experimental approaches. The RTD result showed that the ow performance in the minireactor was almost in the plug-ow condition. The singlephase micromixing performance measured with the Dushman reaction (iodideiodate) in the minireactor attained the desired condition, and the segregation index could be 0.002. The micromixing performance improved with the increasing of the continuous uid ux and dispersed uid ux. Moreover, the micromixing performance was apparently enhanced by decreasing the membrane pore size. The result of precipitation of BaSO4 showed that the mixing performance had a substantial inuence on the particle size. The membrane pore size is a major contributing factor to micromixing performance. With increasing membrane pore size, the micromixing performance became better, and the BaSO4 particle size apparently decreased. It was found that the dye-extraction method can be applied to correctly determine the mixing performance of a two-liquid phase-mixing system. Like the single-phase mixing process, the mixing performance of the two-liquid phase system was inuenced by the membrane pore size and the two-phase ow rate. When the pore size was 0.9 m, however, there was a minimum value for the mixing efciency with the changes of the continuous- or dispersed-phase ux.

Acknowledgments
We gratefully acknowledge the nancial support of the National Nature Science Foundation of China (20476050; 20490200) and SRFDP (20040003032).

Notation
d32 Fc Fd Ft mean diameter of particles, d32 d 3/ d 2, nm ux of continuous solution (phase), cm3/s ux of dispersed solution (phase), cm3/s total ux of dispersed solution (phase) and continuous solution (phase), cm3/s R ux ratio of dispersed phase and continuous phase XS segregation index u electric voltage information of absorbance of methyl orange, volt

Literature Cited
1. Amador C, Gavriilidis A, Angeli P. Flow distribution in different microreactor scale-out geometries and the effect of manufacturing tolerances and channel blockage. Chem Eng J. 2004;101:379-390. 2. Monnier H, Wilhelm AM, Delmas H. Effects of ultrasound on micromixing in ow cell. Chem Eng Sci. 2000;55:4009-4020. 3. Fournier MC, Falk L, Villermaux J. A new parallel competing reaction

system for assessing micromixing efciency-experimental approach. Chem Eng Sci. 1996;51:5053-5064. Fournier MC, Falk L, Villermaux J. A new parallel competing reaction system for assessing micromixing efciency-determination of micromixing time by a simple mixing model. Chem Eng Sci. 1996;51:51875192. Baldyga J, Bourne JR. The effect of micromixing on parallel reactions. Chem Eng Sci. 1990;44:907-916. Guichardon P, Falk L, Villermaux J. Me thode chimique de de termination des temps de microme lange. Re cents Prog Gen Proc. 1995;9: 267-272. Panic S, Loebbecke S, Tuercke T, Antes J, Bos kovic D. Experimental approaches to a better understanding of mixing performance of microuidic devices. Chem Eng J. 2004;101:409-419. Ehrfeld W, Golbig K, Hessel V, Lo we H, Richter T. Characterization of mixing in micromixers by a test reaction: Single mixing units and mixer arrays. Ind Eng Chem Res. 1999;38:1075-1082S. Monnier H, Wilhelm AM, Delmas H. Effects of ultrasound on micromixing in ow cell. Chem Eng Sci. 2000;55:4009-4020. Fang JZ, Lee DJ. Micromixing efciency in static mixer. Chem Eng Sci. 2001;56:3797-3802. Monnier H, Wilhelm AM, Delmas H. Effects of ultrasound on micromixing in ow cell. Chem Eng Sci. 2000;55:4009-4020. Chen GG, Luo GS, Xu JH, Wang JD. Membrane dispersion precipitation method to prepare nano-particles. Powder Technol. 2004;139: 180-185. Chen GG, Luo GS, Yang XY, Sun YW, Wang JD. AnataseTiO2 nano-particle preparation with a micro-mixing technique and its photocatalytic performance. Mater Sci Eng A. 2004;380:320-325. Xu JH, Luo GS, Chen GG, Tan B. Mass transfer performance and two-phase ow characteristic in membrane dispersion mini-extractor. J Membr Sci. 2005;249:75-81. Xu JH, Luo GS, Chen GG. Study on oil deacidication with a novel micro-mixer. Petroleum Process. Petrochem. 2004;35:47-49 (in Chinese). Xu JH, Luo GS, Sun Y, Chen GG. Extraction performance of a new mixer-settler with membrane dispersion technique. J Chem Eng Chin Univ. 2003;17:361-364 (in Chinese). De Bellefon C, Tanchoux N, Caravieilhes S, Grenouillet P, Hessel V. Microreactors for dynamic, high throughput screening of uid/liquid molecular catalysis. Angew Chem Int Ed Engl. 2000;39:3442-3445. Pennemann H, Hessel V, Kost HJ, Lo we H, de Bellefon C. Investigations on pulse broadening for catalyst screening in gas/liquid systems. AIChE J. 2004;50:1814-1823. Guichardon P, Falk L, Villermaux J. Characterisation of micromixing effciency by the iodideiodate reaction system. Part II: Kinetic study. Chem Eng Sci. 2000;55:4245-4253. Aoun M, Plasari E, David R, Villermaux J. Are barium sulphate kinetics sufciently known for testing precipitation reactor models? Chem Eng Sci. 1996;51:2449-2458. Barresi AA, Marchisio D, Baldi G. On the role of micro- and mesomixing in a continuous Couette-type precipitator. Chem Eng Sci. 1999;54:2339-2349. Baldyga J, Wojciech O. Barium sulphate precipitation in a pipeAn experimental study and CFD modeling. Chem Eng Sci. 2001;56:24352444. Judat B, Racina A, Kind M. Macro- and micromixing in a Taylor Couette reactor with axial ow and their inuence on the precipitation of barium sulfate. Chem Eng Technol. 2004;27:287-292. Chen GG, Luo GS, Sun Y, Xu JH, Wang JD. A ceramic microltration tube membrane dispersion extractor. AIChE J. 2004;50:382-387. Benz K, Ja ckel KP, Regenauer KJ, Schiewe J, Drese K, Ehrfeld WV, Hessel V, Lo we H. Utilization of micromixers for extraction processes. Chem Eng Technol. 2001;24:11-17.

Manuscript received Nov. 27, 2004, and revision received Mar. 1, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2929

Modeling of an IGC Experiment to Analyze Ternary PolymerSolvent Systems

Peter K. Davis, J. Larry Duda, and Ronald P. Danner
Center for the Study of Polymer Solvent Systems, Dept. of Chemical Engineering, The Pennsylvania State University, University Park, PA 16802

DOI 10.1002/aic.10551 Published online September 13, 2005 in Wiley InterScience (www.interscience.wiley.com).

A variation of the capillary column inverse gas chromatography (IGC) experiment is proposed to measure the thermodynamics and diffusion coefcients in ternary polymer solventsolvent systems, and a theoretical model has been developed for this proposed experiment. The model has been derived to include both main and mutual cross-diffusion coefcients. Ternary vaporliquid equilibria was represented by a coefcient matrix of the isotherm tangents. The governing equations of the model were uncoupled and solved using both an analytical Laplace transform technique and a numerical nite-difference technique. Solution of the model shows that in the presence of strong cross-diffusion and thermodynamic coefcients, unusual chromatographic behavior can be observed, suggesting that IGC can be a useful technique for measuring thermodynamic and masstransport properties of ternary polymersolventsolvent systems. 2005 American Institute
of Chemical Engineers AIChE J, 51: 2930 2941, 2005

Keywords: inverse gas chromatography (IGC), multicomponent diffusion and thermodynamics, polymersolvent systems, modeling

Introduction
Capillary column inverse gas chromatography (CCIGC) has proven to be a useful experimental technique to measure thermodynamic and mass-transport properties in polymersolvent systems.1-5 In CCIGC, a thin lm of polymer is coated on the inside of a capillary wall. An inert carrier gas transports a small pulse of solvent through the column and a detector measures the elution of solvent from the column. Because the solvent absorbs and diffuses through the thin polymer lm, the solvent elution can be related to the partition coefcient and diffusion coefcient by comparison to a theoretical model. Capillary columns have a major advantage over packed columns because the polymer coating thickness is much more uniform, thus
Correspondence concerning this article should be addressed to J. L. Duda at jld@psu.edu.

allowing accurate measurement of diffusion coefcients. Inverse gas chromatography has been used in the past to measure diffusion coefcients for binary and pseudobinary polymer solvent systems. Danner et al.5 used nite-concentration CCIGC to measure effective diffusion coefcients of toluene and methanol in the PVAcmethanoltoluene system. A true ternary polymersolventsolvent system has a matrix of four mutual diffusion coefcients that govern mass transport D p11 D p12 D p D D p22 p21

(1)

2005 American Institute of Chemical Engineers

The subscripts indicate solvent 1 and solvent 2. The Dp11 and Dp22 are referred to as the main terms of the diffusion coefcient matrix and Dp12 and Dp21 are the cross terms. The diffusive ux of each solvent, j i , in such a system can be dened in terms of these diffusion coefcients relative to the systems volume average velocity
Vol. 51, No. 11 AIChE Journal

2930

November 2005

j 1 Dp 11 1 Dp 12 2 j 2 Dp 21 1 Dp 22 2

(2) (3)

Here, i is the gradient in mass concentration of solvent i. Finite-concentration IGC has been used by Joffrion and Glover6 and Tsotsis et al.7 to measure ternary-phase equilibria and effective binary diffusivities, but not the cross-diffusivities. Diffusion coefcients in polymersolvent systems are usually measured by gravimetric-type experiments. Binary gravimetric experiments are fairly simple. The single binary diffusion coefcient is determined from the time needed to equilibrate a polymer sample with the vapor of solvent. Gravimetric experiments proceed by measuring the weight gain of the polymer vs. time. There are several sorption experiments that rely on this type of measurement, although these experiments fail for ternary experiments because the weight gain of the sample arises from sorption of both solvents. Thus, it is not possible to determine how much of each solvent is in the polymer vs. time. Furthermore, the vapor is no longer pure, and the composition of this phase must also be known as a function of time. A modied quartz spring gravimetric experiment has been used to measure solubility (not diffusion) of two solvents in a polymer.8,9 Although gravimetric experiments are quite common for measuring diffusion coefcients, they are not the only available techniques. Infrared spectroscopy has been used to measure the cross-diffusion coefcients in the PIBMEKtoluene system.10 Drying experiments are another possibility to measure ternary diffusion coefcients.11 In such an experiment, a polymer solventsolvent solution is devolatilized in a convective oven. Measurement of the efuent solvent concentration in the vapor phase can be used to determine the polymer solution weight loss arising from both solvents vs. time. Such an experiment relies on extensive modeling because mass and heat transfer occur in both the vapor and polymer phases. In addition, the experiment is conducted over extremely wide ranges of solvent concentration in the polymer phase. Thus, the experiment relies heavily on a model for the concentration dependency of the diffusion coefcients. Such models do exist, but have not been tested because of the absence of any ternary diffusion data.12,13 Inverse gas chromatography has a major advantage over drying because experiments are conducted at a single concentration (usually at innite dilution). However, experiments can be conducted at nite concentrations as shown by several investigators.5,6,14-16 In this communication, a new CCIGC experiment is proposed to measure the four member diffusion coefcient matrix and vaporliquid equilibria at nite concentrations in ternary polymersolventsolvent solutions. Analysis of such an experiment requires a model that describes the complex diffusion process in the column. A new CCIGC model has been developed capable of analyzing ternary CCIGC data.

Figure 1. Proposed experimental setup for a ternary CCIGC experiment.

polymer of interest and a ternary mixture of an inert carrier gas and vapor of the two solvents is continuously passed through the column. The vapor composition is controlled by the temperature of the two saturators containing each of the pure solvents. The efuent streams from the saturators are then mixed and sent through the column giving a ternary vapor mixture of constant concentration. When equilibrium is attained, a small amount of one of the solvents is injected into the column. This perturbation of solvent travels through the column and interacts with the polymersolventsolvent mixture. A suitable detector measures the vapor concentration of each solvent at the exit of the column. The detector must be capable of distinguishing between each solvent while they are eluding simultaneously. The common thermal conductivity detector (TCD) or ame ionization detector (FID) will not work because these will give an average reading for both solvents. Because the injection perturbs the vapor composition of both solvents, the experiment will produce elution proles for both solvents. Analysis of both elution proles gives information about the thermodynamics and diffusion coefcients of the ternary system.

Model for Ternary CCIGC Experiments

The original CCIGC model was developed by Macris1 for binary polymersolvent systems at innite dilution. The vapor phase was pure carrier gas containing a small solvent concentration from the injection. Here, this model is extended to ternary polymersolventsolvent systems at nite solvent gas concentrations. A diagram of the process to be modeled is show in Figure 2. In the derivation of the model, the following assumptions are made: (1) The entire system is isothermal. (2) The pressure drop in the column is insignicant. (3) The polymer coating thickness is much smaller than the column radius. (4) Gas-phase diffusion is fast enough to keep the solvent radially well mixed, making Taylor dispersion insignicant compared to conventional axial dispersion. (5) The polymer-phase diffusion coefcients are constant
Vol. 51, No. 11 2931

Proposed Experimental Procedure

The proposed experiment is an extension of that used by Danner et al.5 to measure properties of pseudobinary systems. A diagram of this new experimental setup is shown in Figure 1. A quartz capillary column is coated with a thin lm of the
AIChE Journal November 2005

 C1 2C 1 C 1V D g11 t z z2

2 1 D p11 R r

D p12
rR

2 r

rR

(6)

2C 2 C2 C 2V D g22 t z z2
Figure 2. Ternary CCIGC experiment.

2 1 D p21 R r

D p22
rR

2 r

C3 2 C1 2 C2 C 3V Dg 11 D g 22 t z z2 z2

rR

(7)

(8)

within the small concentration range of the solvent perturbations. (6) The vapor-phase diffusion coefcients are constant and the vapor-phase cross-diffusion coefcients are negligible. (7) The thermodynamics of bulk absorption can be described by a coefcient matrix of the isotherm tangent slopes. These coefcients are constant over the small concentration range of the perturbation. (8) The polymer lm has a uniform thickness through the entire column. (9) The only signicant diffusion in the polymer phase occurs in the radial direction. (10) The carrier gas is insoluble in the polymer and does not adsorb on any surfaces. (11) Surface adsorption by the solvents is negligible. (12) No chemical reactions occur. (13) The partial molar volume of the solvents in the polymer is constant. (14) Swelling of the polymer is insignicant over the concentration range of the perturbation. (15) The solvent injection can be modeled by a Dirac delta function. Based on assumption (2), the pressure in the column is constant. Because the total pressure in the column is never much greater than atmospheric, it will be assumed that the ideal gas law applies. An ideal gas at constant temperature and total pressure has a constant molar density Cmolar C1 C2 C3 C molar M1 M2 M3 constant (4)

Here, i represents the mass concentrations of the solvents in the polymer phase. In the polymer phase, the subscript 1 denotes solvent 1 and 2 represents solvent 2. In this formulation, the cross-diffusion coefcients in the gas phase are assumed to be negligible. The appropriate boundary conditions are

t0

z0

C1 C b1 C 2 C b2 C3 C b3 C1 C b1 C 2 C b2 C3 C b3

V V0

(9)

C1 t C 01 C b1 C 2 t C 02 C b2 C3 t C 03 C b3

(10)

V V0

(11)

Here, C0i is the strength of the pulse of species i and Cbi is the background (plateau) concentration of species i. Assuming that the polymer thickness is much smaller than the column radius, the two solvent species continuity equations in matrix notation for the polymer phase are
2 1 D p11 D p12 1 D p21 D p22 r 2 2 t 2

(12)
Cb2

Here, Ci is the mass concentration of each species in the gas phase and Mi is the molecular weight of each species. The subscripts dene the following species in the vapor: 1 solvent 1, 2 solvent 2, and 3 inert carrier gas. Furthermore, because the solvent perturbation is small compared to the overall concentration, the overall mass density (C) of the gas is approximately constant C C 1 C 2 C 3 constant (5)

t0

1 b1 2 b2

Cb1

K p11dC 1
Cb1

K p21dC 1

K p12dC 2
Cb2

K p22dC 2

(13)

Under this approximation, the species continuity equations in the gas phase can be expressed relative to the mass average velocity (V)
2932 November 2005

rR

1 2

C1

K p11dC 1
C1

K p21dC 1

C2

K p12dC 2
C2

K p22dC 2

(14)

Vol. 51, No. 11

AIChE Journal

rr

1 0 0 r 2

(15)

Here Kpij is the slope of the isotherm tangent and bi is the background (plateau) concentration of species i. The diffusive uxes are dened with respect to the volume average velocity in the polymer phase. Equations 6, 7, 8, and 12 constitute ve coupled partial differential equations that must be solved to model the elution of the two solvents from the column. These equations were uncoupled using a coordinate transformation17 and were solved using the technique of Laplace transforms. The details of the solution are in Appendix A. In the solution, it is shown that the solvent concentration in the gas phase at the detector can be expressed in the Laplace domain 1 i H iexp Y 2i

Figure 3. Simulation of a ternary CCIGC experiment.

Solid line is ternary model prediction for solvent 1 and dotted line is that for solvent 2. Symbols indicate ternary model simulation for insignicant cross terms (Dpij 0): solvent 1 (), solvent 2 (f).

1 i s 4i2

(16)

i is a pseudo-Laplace variable for the gas concentration Here, Y of solvent i at the detector. Hi, i, and i(s) are vectors of dimensionless parameters. Analytical inversion of Eq. 16 into the time domain is not possible, so numerical inversion was used to solve the model by a Fast Fourier Transform (FFT) algorithm.

Model Predictions for Ternary CCIGC Experiments

The purpose of deriving this model is to permit analysis of experimental ternary elution proles with the intent of determining the matrix of diffusion and thermodynamic coefcients. The Dpii and Kpii are often referred to as the main terms, whereas Dpij and Kpij are referred to as the cross terms in the coefcient matrices. The aim here is to measure all four diffusion and thermodynamic coefcients. It is questionable that all eight unique parameters can be obtained from the elution proles of a single CCIGC experiment. For this reason, experiments should be conducted with two different columns. The rst column should be designed such that there is negligible diffusion resistance in the polymer phase. This is accomplished by coating a very long column with a sufciently thin polymer coating. Because there will be negligible mass transfer resistance, experiments with this column will produce symmetric elution proles that are independent of the diffusion coefcients. These elution proles can be uniquely related to the thermodynamic coefcients. Once the thermodynamic coefcients are known, a second, shorter column should be made with a thicker coating to give signicant diffusion resistance. Experiments on this column will give asymmetric elution proles that can be used to obtain a unique set of four mutual diffusion coefcients.

pseudobinary diffusion processes because the two solvents will diffuse independent of each other. Although the proposed experiment is capable of measuring such pseudobinary diffusion coefcients, it is also possible that both main and cross-diffusivities are needed to accurately describe the mass-transport process. In such a case, the elution proles of the two solvents must provide information about all four diffusion coefcients. To demonstrate the effect of the cross-diffusion coefcients on the elution prole, the model described above was solved over varying ranges of the cross-diffusion coefcients. Nominal values of the other parameters were used in the simulations. The cross thermodynamic coefcients were assumed to be zero. The results of the simulations are shown in Figures 3, 4, and 5. The dimensionless parameters for the simulations are given in Table 1. In each of the gures, a different value of the ratio of the cross-diffusion coefcient to the main diffusion coefcient for both of the two solvents was used. In Figure 3, the cross-diffusion coefcients are 100 times smaller than the main diffusion coefcients. Under such conditions, the cross-diffusion coefcients make essentially no contribution to the shape of the elution proles. There was also only a negligible difference when the cross terms were 75% smaller than the main terms (not shown). In these cases, mass transport in the ternary system can be accurately represented as two pseudobinary systems because diffusion of the two solvents can be described by just the main diffusion coefcients.
Table 1. Values of the Dimensionless Groups for the Simulations in Figures 3, 4, and 5*
Parameter Figure 3 0.53 0.13 1.0 10.0 10.0 1.0 1.0 104 1.0 Figure 4 0.53 0.13 1.0 1.1 1.1 1.0 1.0 104 1.0 Figure 5 0.53 0.13 1.0 1.01 1.01 1.0 1.0 104 1.0

Diffusion coefcient sensitivity

Experiments on the short, thick column will produce elution proles that depend on the diffusion coefcients. It is possible that mass transport in the ternary system can be described using only the main diffusion coefcients (cross-diffusion coefcients equal to zero). This effectively corresponds to two
AIChE Journal November 2005

11 22 12 21 11 12 21 22 1 2

*Dimensionless group denitions are given in Appendix A.

Vol. 51, No. 11

2933

Figure 4. Simulation of a ternary CCIGC experiment.

Solid line is ternary model prediction for solvent 1 and dotted line is that for solvent 2. Symbols indicate ternary model simulation for insignicant cross terms (Dpij 0): solvent 1 (), solvent 2 (f).

Figure 6. Numerical simulations to determine the effect of axial dispersion on the elution prole.
Solid line is analytical model prediction for solvent 1 and dotted line is that for solvent 2 when Dg11 Dg22. Symbols: solvent 1 (), and solvent 2 (f), indicate numerical model solution when Dg22 10Dg11. All other parameters were the same as those used in Figure 4.

However, the diffusion cross terms used in Figures 4 and 5 denitely give elution proles different from those obtained without cross terms. In these cases, it is not possible to represent the ternary diffusion processes as two pseudobinary systems. Interestingly, the elution proles give very unusual behavior in the presence of signicant diffusion cross terms. Multiple peaks and solvent concentrations dipping below their baseline are observed under such conditions. This unique behavior can be quite useful when trying to determine the signicance of the cross-diffusion coefcients in polymersolvent systems. If the cross-diffusion coefcients are negligible, single peaks and positive dimensionless concentrations will be observed in a CCIGC experiment. If multiple peaks and negative dimensionless concentrations are observed from the IGC, then the cross-diffusion coefcients are signicant and can be measured. Performing experiments at different solvent concentrations can give the concentration dependency of both main and cross-diffusion coefcients in ternary polymersolvent systems. The diffusion coefcients in the matrix are not independent

of each other. They are constrained by material balances, the Onsager reciprocal relationships, and thermodynamic stability. The Onsager relationships and thermodynamic stability specify that the diffusion coefcient matrix has real-positive eigenvalues.18 For a ternary system, the necessary constraint to ensure this is D p11 D p22 0 D p11D p22 D p12D p21 0 D p11 D p22 2 D p11D p22 D p12D p21 (17)

In all simulations, these expressions were obeyed and no material balances were violated. Material balance violation can be detected by negative dimensional concentrations.19 Negative concentrations can be obtained by solving the continuity equations with unrealistic diffusion coefcients. Although the continuity equations can be mathematically satised with negative concentrations, a negative mass cannot physically exist. Thus, any negative concentrations violate material balance constraints. In these simulations, no negative concentrations were obtained.

Numerical solution
To this point, model simulations were obtained from the analytical Laplace domain solution. A numerical solution was also developed for the model based on an implicit nitedifference method. The numerical model is quite useful because it can be used to verify some of the assumptions needed for the analytical solution. For example, in the analytical solution (see Appendix A), it was assumed that the gas-phase diffusion coefcient matrix was symmetrical (Dg11 Dg22). It was argued that this assumption had no bearing on the elution proles of the solvents. To test this argument, the numerical solution was generated for Dg11 Dg22 and for Dg22 10 Dg11. The results of the simulation are shown in Figure 6. Even
Vol. 51, No. 11 AIChE Journal

Figure 5. Simulation of a ternary CCIGC experiment.

Solid line is ternary model prediction for solvent 1 and dotted line is that for solvent 2. Symbols indicate ternary model simulation for insignicant cross terms (Dpij 0): solvent 1 (), solvent 2 (f).

2934

November 2005

when the gas-phase diffusion coefcients are different by a factor of 10, the elution proles are still the same. This veries the assumption of a symmetric Dg matrix. More details for the numerical solution can be found in Appendix B.

Parameter Estimation
In the proposed experiment, the diffusion coefcient and thermodynamic coefcient matrices will be obtained from the elution proles of the two solvents. Eight unknown parameters are needed to completely describe these elution proles. As mentioned, experiments should be carried out using two different columns. The rst column should be designed to give an insignicant amount of diffusion resistance in the polymer. Experiments on this column can be used to obtain the thermodynamic coefcients Kij. As described, these are the tangents to the isotherm planes and, although constant during any experiment, they will vary with bulk composition changes. After a number of Kij with sufcient proximity in concentration have been measured, the actual bulk solubilities can be obtained by numerical integration according to Eq. 13 subject to the initial conditions (Ci 0, i 0). Experiments on a second, shorter, thicker column should then be used to obtain the diffusion coefcients. Inverse gas chromatography parameter estimation is usually carried out by moment analysis.1-4 In a binary experiment, the rst and second moments can be related to the partition and diffusion coefcient, respectively. This procedure cannot be used directly for ternary analysis because there are more parameters than moments. In addition, the complex Laplace domain expression for Yi makes the analytical expressions for the moments difcult to obtain. Thus, it is proposed that parameter estimation proceed by Fourier or time domain tting. In this procedure, the parameters are adjusted such that the experimental elution proles match the model elution proles for each solvent. This procedure was rst introduced by Pawlisch2 and used by other investigators for binary systems.4,5,20,21 In binary experiments, the partition and diffusion coefcient are adjustable parameters used to match the experimental elution to that of the binary model. Caution must be exercised when measuring the diffusion coefcients at nite concentrations arising from variation in the coating thickness . As the solvents swell the polymer coating, the lm thickness will increase. In general, this will lead to an underestimation of the diffusion coefcients. During any particular experiment, the small solvent perturbation does not signicantly change the coating thickness, but as the bulk concentration changes from experiment to experiment, swelling will become important. However, because the experiments on the column with insignicant diffusion resistance give the thermodynamics of the polymersolventsolvent system, this swelling can be predicted and accurate diffusion coefcients can be obtained.

over wide ranges of solvent concentration. The uniqueness and power of this technique is that an experiment is conducted by measuring a small solvent perturbation around a xed composition. Thus, experiments are performed at a given solvent concentration such that the diffusion and thermodynamic coefcients can be considered constant. A model has been developed that describes the complex diffusion process in the column. Both analytical and numerical solutions have been obtained. Simulation of the proposed experiment indicates that CCIGC is quite sensitive to the cross-diffusion coefcients when they are of comparable order to the main terms. This sensitivity makes CCIGC a promising technique to measure the values of all four diffusion coefcients and the true vapor liquid equilibrium in ternary polymersolventsolvent systems.

Notation
i mass concentration of species i in the polymer phase, kg/m3 bi background mass concentration of species i in the polymer phase, kg/m3 C mass density of the vapor phase, kg/m3 Ci mass concentration of species i in the vapor phase, kg/m3 Cbi background mass concentration of solvent i in the vapor phase, kg/m3 C0i strength of the solvent pulse (kg-s/m3) for species i Dpij mutual binary diffusion coefcient matrix for the polymer phase, m2/s Dgij mutual binary diffusion coefcient matrix for the gas phase, m2/s Kpij dimensionless coefcient matrix (isotherm tangent slope) describing the thermodynamics between the gas and polymer phases L column length, m Mi molecular weight of species i, kg/mol r radial coordinate, m R radius of the gaspolymer interface, m t time, s V carrier gas velocity, m/s V0 characteristic carrier gas velocity, m/s z axial coordinate, m polymer coating thickness, m ij R/Kpij thermodynamic dimensionless group 2 ij V02/LDpij polymer-phase diffusion dimensionless group ij Dgij/LV0 axial dispersion dimensionless group C01/C02 ratio of the inlet pulse strengths

Literature Cited
1. Macris A. Measurement of Diffusion and Thermodynamic Interactions in PolymerSolvent Systems Using Capillary Column Inverse Gas Chromatography. PhD Thesis. Amherst, MA: Dept. of Chemical Engineering, University of Massachusetts; 1979. 2. Pawlisch CA, Laurence RL, Macris A. Solute diffusion in polymers. 1. The use of capillary column inverse gas chromatography. Macromolecules. 1988;20:1564. 3. Arnould DD. Capillary Column Inverse Gas Chromatography (CCIGC) for the Study of Diffusion in Polymer Solvent Systems. PhD Thesis. Amherst, MA: Dept. of Chemical Engineering, University of Massachusetts; 1989. 4. Hadj Romdhane I. PolymerSolvent Diffusion and Equilibrium Parameters by Inverse GasLiquid Chromatography. PhD Thesis. State College, PA: Dept. of Chemical Engineering, The Pennsylvania State University; 1994. 5. Danner RP, Tihminlioglu F, Surana RK, Duda JL. Inverse gas chromatography applications in polymersolvent systems. Fluid Phase Equilibria. 1998;148:171. 6. Joffrion LL, Glover CJ. Vaporliquid equilibrium of the ternary n-

Summary and Conclusions

A new capillary column inverse gas chromatography experiment has been proposed to measure the four member diffusion coefcient matrix and the vaporliquid equilibrium in ternary polymersolventsolvent systems. The proposed experiment is capable of measuring the diffusion and partition coefcients
AIChE Journal November 2005

Vol. 51, No. 11

2935

7.

8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

19. 20. 21.

22.

heptane/isopropyl alcohol/atactic polypropylene mixture from perturbation gas chromatography. Macromolecules. 1986;19:1710. Tsotsis TK, Turchi C, Glover CJ. Comparison of multi-component gas chromatography measurements of vaporliquid equilibrium with static measurements using a polymer/two solvent ternary system. Macromolecules. 1987;20:2445. Tanbonliong JO, Prausnitz JM. Vapourliquid equilibria for some binary and ternary polymer solutions. Polymer. 1997;38:5775. Liu H, Huang Y, Wang K, Hu Y. Vaporliquid equilibria for mixed solventspolymer systems: Measurement and correlation. Fluid Phase Equilib. 2002;194:1067. Hong SU, Barbari TA, Sloan JM. Multicomponent diffusion of methyl ethyl ketone and toluene in polyisobutylene from vapor sorption FTIRATR spectroscopy. J Polym Sci Part B: Polym Phys. 1998;36:337. Price PE, Wang S, Romdhane IH. Extracting effective diffusion parameters from drying experiments. J Eng Appl Sci. 1997;43:1925. Alsoy S, Duda JL. Modeling of multicomponent drying of polymer lms. AIChE J. 1999;45:896. Zielinski JM, Hanley BF. Practical friction-based approach to modeling multicomponent diffusion. AIChE J. 1999;45:1. Brockmeier NF, McCoy RW, Meyer JA. Gas chromatographic determination of thermodynamic properties of polymer solutions. I. Amorphous polymer systems. Macromolecules. 1972;5:464. Brockmeier NF, McCoy RW, Meyer JA. Gas chromatographic determination of thermodynamic properties of polymer solutions. II. Semicrystalline polymer systems. Macromolecules. 1972;6:176. Brockmeier NF, McCoy RW, Meyer JA. Gas chromatographic determination of thermodynamic properties of polymer solutions. Macromolecules. 1972;5:130. Toor HL. Solution of the linearized equations of multicomponent mass transfer: I. AIChE J. 1964;10:448. Kirkaldy JS, Weichert D, Zia-Ul-Haq. Diffusion in multicomponent metallic systems. VI. Some thermodynamic properties of the D matrix and the corresponding solutions of the diffusion equations. Can J Phys. 1963;41:2166. Price PE, Romdhane IH. Multicomponent diffusion theory and its applications to polymersolvent systems. AIChE J. 2003;49:309. Surana RK, Danner RP, Tihminlioglu F, Duda JL. Evaluation of inverse gas chromatography for prediction and measurement of diffusion coefcients. J Polym Sci Part B: Polym Phys. 1997;35:1233. Surana RK. Advances in Diffusion and Partition Measurements in PolymerSolvent Systems Using Inverse Gas Chromatography. PhD Thesis. State College, PA: Dept. of Chemical Engineering, The Pennsylvania State University; 1997. Vrentas JS, Vrentas CM, Hadj Romdhane I. Analysis of inverse gas chromatography experiments. Macromolecules. 1993;26:6670.

tV L

(A3)

rR z L

(A4)

(A5)

Here, V0 is the characteristic carrier gas velocity. Substituting these variables into polymer species continuity equations (Eqs. 1215) gives

1 t 2

1 11 2 2 11 22 12 2 1 22 1 r2 2 2 2 11 21 22 0 0
1 2

(A6)

t 0

(A7)

r 0

1 2

22 21

11 12
1

C C
1 2

(A8)

r 1

0 1 0 r 2

2 r

(A9)

Summation of the three species continuity equations (Eqs. 6 8) for the gas phase gives the total continuity equation V 2 1 D p11 z R r

Appendix A: Model Solution

To aid in the model solution, the following dimensionless variables are introduced

D p12
r R

1 1b V0 K C L p11 01 1 2 2b 2 V0 K C L p22 02
C 1 C b1 C 01V 0 L C C b2 2 C 1 C C V 2 02 0 C 3 L C 3 C b3 C 03V 0 L

2 1 D p21 R r

rR

D p22
rR

2 r

r R

(A10)

(A1)

Substituting the dimensionless variables into this equation gives 2 C 01V 0 V 1 2 z 11 11 CL r

2936

2 C 01V 0 1 2 CL r 11 21

r0

2 C 02V 0 2 2 22 12 CL r

r0

2 C 02V 0 2 2 CL r 22 22

r0

r0

(A11) (A2) For a typical CCIGC experiment, (C0iV0/CL)is on the order of 105, whereas 2/12 21 is usually on the order of 1 or 10. Thus, it can be safely assumed that the mass average velocity in the column is constant
November 2005 Vol. 51, No. 11 AIChE Journal

V 0 z

and

V V0

(A12)

Under this approximation, the solvent species continuity equations simplify to C1 C1 2C 1 V0 D g11 t z z2

analytical solution possible. Macris1 showed that the equations for the binary system can be solved using Laplace transforms. This approach cannot be directly applied because the equations are coupled. To uncouple the equations a new matrix, [A], is dened. This matrix is constructed by setting its columns to the eigenvectors of the matrix in Eq. A6

2 1 D p11 R r

D p12
rR

2 r

rR

(A13) A

C2 2C 2 C2 V0 D g22 t z z2

2 1 D p21 R r

D p22
rR

2 r

rR

(A14)

2 1 1 2 22 22 2 11 21 1 1 1 2 11

(A19)

11 2 22 12

Because the velocity is assumed constant, it is no longer necessary to consider the carrier gas continuity equation. The solvent species continuity equations can be expressed in dimensionless form C 2 C C 1 1 1 t C z C z2 C 2 2 2 D where 1 1 2 2 11 11 22 12 D 2 1 2 2 11 21 22 22 and

1 and 2 are the eigenvalues of the matrix in Eq. A6. Here, They can be found from the quadratic rule to be 1 1 tr tr2 4 det 2 1 tr tr2 4 det 2

(A20)

(A21)

1 r 2

(A15)
r0

Here tr is the trace of the matrix

A 1

1 11 2 2 11 22 12 22 1 2 2 11 21 22

(A22)

and det is the determinant of the matrix. The reason for creating the [A] is that is has a unique mathematical property

D g11 V 0L 0 0 D g22 V 0L 0 C C 0
1 2

1 11 2 2 1 0 11 22 12 A 2 22 1 0 2 2 11 21 22

In dimensionless form, the necessary boundary conditions are t 0 z 0 z (A16)

Multiplying both sides of Eq. A6 by [A]1 gives

t C C t
1 2

1 A 1 A 1 t 2

(A17)

1 11 2 2 11 22 12 A 22 1 2 2 11 21 22 A 1

0 C C 0
1 2

(A18)

2 1 r2 2

(A23)

Because dimensional analysis showed the gas velocity to be constant, the governing equations are linear. This makes an
AIChE Journal November 2005

The identity matrix [I] [A][A]1 has been inserted between the matrix and the second derivative. Substituting Eq. A22 into this result gives
Vol. 51, No. 11 2937

1 0 2 1 1 2 r2 0 t 2 2

1 2

1 2

Using the initial condition from Eq. A28 (A24) 2Q 1 Q 1 s 0 r2 1 2Q 2 Q 2 s 0 r2 2 (A35)

In this equation, pseudoconcentrations have been introduced A

1

(A25)

(A36)

Equation A24 can be written as two uncoupled partial differential equations in terms of these pseudocompositions 2 1 1 1 t r2 2 2 2 2 t r2

The boundary conditions are also transformed into the Laplace domain r 0 Y Q Q B Y
1 1 2 2

(A26)

(A37)

(A27)

r 1

Q1 0 0 r Q2

(A38)

The boundary conditions are also multiplied by [A]1 to give t 0 0 0

1 2

Equations A35 and A36 are second-order linear homogeneous equations with constant coefcients. Thus, the general solutions have the form Q 1 c 1e s/1r c2 e s / 2r

(A28)

(A39) (A40)

r 0 where

C 1 1 B C 2 2

(A29)

Q 2 c 3e s/2r c 4e s/2r

B A 1 22 21 r 1

11 12
1

The ci are constants that are determined from the boundary conditions given in Eqs. A37 and A38. With these constants, the solution reduces to A Q 1 B 11Y 1 B 12Y 2 (A30)

0 1 0 r 2

1
2 1 e s/1

e s/1r

e 2 1 e s/1

e s/1
2

1r s/

(A41)

Q 2 B 21Y 1 B 22Y 2

The polymer-phase species continuity equations are now two uncoupled differential equations. These types of equations can be solved using Laplace transforms. The following Laplace domain variables are introduced s , r t , r Q Q s , r t , r
1 1 2 2

1
2 1 e s/ 2

e s/ 2r

e 2 1 e s/ 2

e s/ 2
2

2r s/

(A42)

(A31)

The gas-phase equation requires the gradient of Qi at the gaspolymer interface. This is found by differentiating Eqs. A41 and A42 with respect to r

Y 1 s , z C 1 t , z Y 2 s , z C 2 t , z

(A32) where

r
r0

Q1 Y1 Q2 C Y2

(A43)

Using these variables, Eqs. A26 and A27 can be transformed into the Laplace domain Q1 s 1 t 0 1 Q1 r2
2

C (A33)

s tanh 1 0

s 1

s tanh 2

s 2

Q2 s 2 t 0 2
2938

2Q 2 r2

(A34)
November 2005

As with the polymer phase, both sides of the gas-phase equations can be multiplied by [A]1
Vol. 51, No. 11 AIChE Journal

 A 1

C C 1 1 1 A t C z C 2 2 A A A
1

2 C 1 z2 C 2

A 1 D A A 1

r
1 2

F E

s tanh 1 0

s 1

0 s tanh 2

s 2

(A44)
r0

Next, the equations must be uncoupled by multiplying both sides of the equation by another matrix, [G]1 Y1 Y1 s G 1 Y G 1 G 1 G z Y2 2 G 1

It will be assumed that Dg11 Dg22. Although this is probably not a good approximation, the value of typically has an insignicant effect on the shape of the elution prole.22 Thus, using this approximation should not affect the end result. With this assumption, [] is a symmetric matrix, and Eq. A44 and the boundary conditions can be written as

2 Y1 Y1 G 1 F G G 1 Y z2 Y2 2

(A53)

C 2 C C 1 1 1 2 t C z C z C 2 2 2 E

Similar to [A] for the polymer phase, [G] is constructed from the eigenvectors of [F]

1 r 2

(A45)
r0

t 0

C 0 1 0 C 2

1 1 F 11 F F 12

2 F 22 F F 21 1

(A54)

(A46) 1 and F 2 are the eigenvalues of the [F] matrix. They where F can be found from the quadratic rule to be 1 F (A48) 2 F 1 tr tr2 4 det 2 1 tr tr2 4 det 2

z 0

C t 1 A 1 t C 2

(A47)

C 0 1 0 C 2

(A55)

(A56)

In Eq. A45, [E] [A] [D][A]. Using the variables dened in Eqs. A31 and A32, these equations can be converted into the Laplace domain Y1 2 Y1 Q1 Y1 E s Y z Y2 z2 Y2 r Q2 2

Here tr is the trace of the F matrix and det is the determinant of the G matrix. The reason for creating the G matrix is that it has a unique mathematical property G 1 F G

r0

(A49) z 0 1 Y Y A 1
1 1 2

1 0 F 2 0 F

(A57)

(A50)

Here, new pseudo-Laplace variables are dened:

0 Y Y 0
1 2

(A51)

1 Y Y1 1 2 G Y2 Y

(A58)

Substituting for the gradient given by Eq. A43 Y1 2 Y1 Y1 Y1 F Y s Y z Y2 z2 Y2 2 2 In this equation

AIChE Journal November 2005

Substitution of these pseudovariables gives (A52) s 1 1 1 1 0 Y 2 Y Y F 2 2 2 0 F 2 Y z Y z2 Y

1 Y 2 Y (A59)
2939

Vol. 51, No. 11

z 0

1 Y H1 2 H2 Y 1 Y 0 2 0 Y

(A60)

i H iexp Y

1 2i

1 i s 4i2

(A67)

z In Eq. A60

(A61)

Appendix B. Details of Numerical Solution

The numerical model is based on an implicit nite-difference technique. First, the species continuity equations for the gas and polymer phases are discretized using a Taylor series expansion. For example, the polymer-phase solvent 1 continuity equation given in Eq. A6 is expressed as

H H G
1 2 2

1 A 1 1

(A62)

The uncoupled equations can be written as 1 1 Y 1 Y 1 0 1 s Y 2 z 1 z 2 2 1 Y 2Y 2 0 2 s Y 2 z 2 z where

1 1,i 1,i1 1, j1 2 1, j 1, j1 2 t r2 11
(A63)

11 2, j1 2 2, j 2, j1 2 r2 22 12

(B1)

i s

i sF i

Here, the i and j subscripts represent discretized time and radial positions, t and r are the sizes of the time and radial step, respectively. A similar equation can be derived for solvent 2. The discretized gas-phase continuity equation for solvent 1 is

Like the polymer-phase equations, these are second-order linear homogeneous differential equations with constant coefcients. Thus, the general solutions have the form

C C C 1,i C 1,i1 1,k C 1,k1 1,k1 2 C 1,k C 1,k1 1 t z z2 D 11

1 c 1exp z Y

1 21

1 2 1 s 41

1 c2 exp z 21

1, j2 1, j1 2, j2 2, j1 D 12 r r

(B2)

1 2 1 s 41

(A64)

Here, k represents the discrete axial positions in the column and z is the axial step size. An expression similar to Eq. B2 can be derived for solvent 2. The discretized polymer-phase boundary conditions are

2 c 3exp z Y

1 22

1 2 2 s 42

t 0 (A65)

1 c4 exp z 22

1 2 2 s 42

0 0
1, j 2, j

(B3)

Here, ci are constants that can be determined from the boundary conditions. Using Eqs. A60 and A61, the general solution is found to be i H iexp z Y

r 0

1, j1,k 2, j1,k

22 21

11 12
1

C C
1,k 2,k

(B4)

1 2i

1 i s 4i2

(A66)

r 1

1, jN 1, jN1 2, jN 2, jN1

(B5)

Because the gas concentration at the detector is needed, this expression is evaluated at the end of the column (z 1).
2940 November 2005

Here, N is the total number of discrete points in the radial direction. Similarly, the gas-phase boundary conditions are
Vol. 51, No. 11 AIChE Journal

t 0

0 C C 0
1,k 2,k

(B6)

z 0

C C
1,k1 2,k1

1 t 1 t

(B7)

Here, M is the total number of discrete points in the axial direction. The condition at innite z was specied at a large value of z (k 10M). The gas- and polymer-phase species continuity equations were solved with a simple tridiagonal solver at each time step. A constant step size time integrator was used to solve all four equations at each time step in a marching fashion.

C 0 C 0
1,k10M 2,k10M

(B8)

Manuscript received Aug. 18, 2003, and revision received Mar. 7, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2941

PARTICLE TECHNOLOGY AND FLUIDIZATION

Generic Crystallizer Model: I. A Model Framework for a Well-Mixed Compartment

M. J. Hounslow
Dept. of Chemical and Process Engineering, University of Shefeld, Shefeld S10 2TP, U.K.

A. E. Lewis
Chemical Engineering Department, University of Cape Town, Rondebosch 7701, South Africa

S. J. Sanders
Sanders Simulation, Denville, NJ 07834

R. Bondy
OLI Systems Inc., Morris Plains, NJ 07950 DOI 10.1002/aic.10535 Published online August 3, 2005 in Wiley InterScience (www.interscience.wiley.com).

A model framework is described for crystallization of a single solid species in a well-mixed compartment at steady state. The model framework applies to both Type I (that is, nonhigh yield) crystallization and Type II (high yield) crystallization. The framework consists of population balances incorporating nucleation, growth, aggregation, breakage, classication, and dissolution, coupled with mass and energy balances. The model allows any number of product streams, any number of feed streams, one vapor product stream, nonrepresentative sampling, but only one solid species. The numerical strategy used to solve the resulting set of nonlinear integro-differential equations transforms them into a matrix of algebraic equations. Two algorithms for the solution for Type I crystallization are proposed, both of which consist of solving the material and energy balances sequentially with the population balance and iterating around only one variable. Both algorithms use an existing material and energy balance solution package, which is linked to the population balance equations. The rst solution algorithm solves the population balance equations using a NewtonRaphson solver with nite-difference approximations for the derivatives, converging around a variable related to the crystal mass and the number density for each interval. The second algorithm solves the population balance equations using a successive substitution technique with root bracketing and iterates around the suspension density. The choice of algorithm depends on the nature of the system to be modeled. A similar framework is suggested for the solution for Type II crystallization, except that the iteration variable is the growth rate at a xed supersaturation ratio. 2005
American Institute of Chemical Engineers AIChE J, 51: 29422955, 2005

Introduction
This article addresses the specication of a generic model to describe crystallization of a single solid species in a continuous
Correspondence concerning this article should be addressed to M. J. Hounslow at m.j.hounslow@shefeld.ac.uk.

2005 American Institute of Chemical Engineers

stirred tank as shown in Figure 1. The model uses a structure consisting of input stream data (temperature, pressure, compositions, ow rates, and enthalpies) combined with a function describing the particle size distribution (PSD). The modeling of a particulate system necessitates the use of the population balance equation (PBE), a statement of continuity for particulate systems, rst introduced by Hulbert and Katz1 and Randolph and Larson.2,3 The requirements necessary
Vol. 51, No. 11 AIChE Journal

2942

November 2005

temperature, pressure, and composition) is represented by a vector of state variables, s. The PBE at steady state is

k1

Q n l Q
Kin K out in in k k k1

out out k k

n l Vrn, l, s

(1)

Figure 1. CST.

where r[n, l, s] is the density functional describing the net rate of destruction of particles. For most purposes the heart of the PBE is this term, and, provided that n out k (l) can be written in terms of n(l), the denition of r in terms of n(l) converts Eq. 1 into an equation to be solved for n(l).

to completely determine the formation and dynamics of multidimensional particle distributions are covered in Randolph and Larson,4 Ramkrishna,5,6 and in extensive detail in Hounslow and Wynn.7 There are various techniques available for solving PBEs (see Ramkrishna5,6 for a general review). In this case, a discretization method is used, which considers the PSD to consist of groups of particles of different sizes, described by a vector of values whose elements correspond to discrete particle sizes. With a given set of input conditions, an appropriate set of operating conditions, and information about crystallization mechanisms, the model can describe the output stream. The real strength of a model such as this is that the solution to the PBE can be coupled to a thermodynamic package such as OLI Systems Inc.8 or Aspen Plus9 that rigorously calculates the supersaturation using the comprehensive solution chemistry embodied in such a package. For this model, there can be Kin streams owing into the vessel, each containing solid, liquid, or solid and liquid. Similarly, there are Kout streams owing out. There is also a stream of vapor leaving and an indirect heat ow. The model framework allows for nucleation, growth, aggregation, breakage, classication, and dissolution. Destruction of nes can be treated as two coupled crystallizers with classication and dissolution of nes in one of them. It is important to note that this is not an MSMPR (mixed suspension, mixed product removal) model because it is possible to have nonmixed product removal.

Mechanisms
The identication of appropriate kinetic equations for the various mechanisms included in the population balance equation is core to its predictive capacity. Thus, an overview is presented here of some of the models available to describe the various kinetic processes relevant to crystallization. Although the various crystallization mechanisms and kinetics of nucleation and growth, for example, are fundamental to the framework, the choice of function relating the kinetics to the supersaturation and other process variables is of much less importance. Nucleation. Nucleation is classied as being primary or secondary, depending on the mechanisms through which it occurs.11 Randolph and Larson4 consider the phenomenon of nucleation and discuss a number of nucleation models. These range from the fundamental Arrhenius-type expression for the rate of homogeneous nucleation12 to one accounting for heterogeneous effects.13 They conclude that an expression based on the Miers nucleation model14 has been most successful in matching experimental work. The formulation proposed by Randolph and Larson for the nucleation rate B0, the number of nuclei formed per volume per second, is
ii c c*I B 0 k nuc

(2)

Population Balance
Consider the population balance equation (PBE) for a continuous stirred tank at steady state. There are Kin streams owing in, at volumetric rates Q in k , bearing solids with number densities n in k (l). Similarly, there are Kout streams owing out, at volumetric rates Q in k , bearing solids with number densities n out k (l). The tank has a volume V and within the tank the number density is everywhere constant at n(l). This is illustrated in Figure 2. It must be noted that this representation is incomplete if more than one solid species is present. For processes excluding aggregation, a simple extension is to have a separate number density for each solid; if aggregation is present, a different approach is needed. This latter approach has been developed by Hounslow et al.10 for two solid systems and can probably be extended to multiple solids if desired. In the work discussed here, it is assumed that only one solid is present. Within the tank, the state of the solution (such variables as
AIChE Journal November 2005

where B0 is the nucleation rate (m3 s1), k ii nucis the nucleation rate constant, c is the solute concentration (kg m3), and c* is the equilibrium solute concentration (kg m3). It is presumed that k ii nuc can be a function of temperature but that I is not. Garside15 proposes an idealization of primary nucleation based on a kinetic law formally identical to the thermodynamic law for homogeneous nucleation (but where the parameters A and B do not have the same physical meaning). B 0 Ae B/ln S
2

(3)

Figure 2. Balance region for PBE. Vol. 51, No. 11 2943

where S is the supersaturation ratio. Sohnel and Garside16 conrm a similar form for primary homogeneous nucleation (applicable over a limited range of supersaturation)
iii I iv I B 0 k nuc S knuc c

where G is the linear growth rate (m s1), l is the characteristic dimension (m), t is time (s), and the mass growth rate is dened as kv 1 dm 3 sG A dt ka (8)

(4)

where B0 is the nucleation rate (m3 s1), I is the kinetic order of nucleation, S is the supersaturation ratio, c is the iv solute concentration (kg m3), and k iIi nuc and k nuc are nucleation 3 1 1 1 rate constants (m s and kg s , respectively). iv Although the nucleation rate constants k iIi nuc and k nuc have no physical signicance, Eq. 4 is very often used in practice. Secondary nucleation, besides having an obvious dependency on the level of supersaturation, is generally considered to be related to the suspension density
v J B 0 k nuc S 1I MT

(5)

where MT is the suspension density (kg m3) and k v nuc is the nucleation rate constant (kg1 s1). The nucleation rate constant k v nuc is assumed to be related to the stirring power and to exhibit a temperature dependency according to an Arrhenius-type relationship.17 The exponent I, which does not depend on temperature, lies between 0.5 and 2.5. The exponent J is generally assumed to be of the order of 1 and to depend slightly on temperature.18 Gahn and Mersmann19 proposed a new model for secondary nucleation, in which a direct relationship was formulated between the crystallization kinetics and the frequency and energy of particleimpeller collisions depending on a number of material properties of the crystals. This model has been widely accepted and is expressed as follows
J k Nimp B 0 nucS 1I MT

where A is area of the crystal (m2), m is mass of the crystal (kg), kv is the volume shape factor, ka is the area shape factor, and s is solid density (kg m3). Because crystal growth involves transport of the growth units from the bulk of the solution to the surface of the crystal and the incorporation of those units into the crystal lattice, growth can be controlled by either diffusion or surface integration. Experimental determination of the growth kinetics supported by microscopic examination of crystals often allows determination of the prevailing growth mechanism.25 For diffusion-controlled growth (crystallization of compounds with a high solubility, static crystallization, and crystallization from viscous solutions25) the growth rate becomes 1 dm kc 3 c c* A dt 1w (9)

(6)

where nuc is the secondary nucleation rate constant (m3 kg2) and Nimp is stirrer speed (s1). To assess the predictions of the GahnMersmann19 model, Bermingham et al.20 evaluated three additional kinetic models that had also been developed for crystallization processes dominated by secondary nucleation and growth. In these models, inter alia, the attrition rate is not simply a function of the overall crystal concentration, but also of the crystal size distribution. Bermingham concluded that these models,21-23 although largely empirical, do provide signicantly more physical insight into the dominant crystallization phenomena than the traditional power-law models. Growth. Crystal growth is typically denoted by G, the linear crystal growth rate, and usually has a supersaturation dependency ranging from rst to second order depending on the rate-controlling step.11 McCabe24 proposed that the linear rate of crystal growth is independent of size, an observation that is conventionally called the L law. If l is dened as a characteristic dimension, then the linear growth rate can be dened as G
2944

where kc is the mass-transfer coefcient (mol m2 s1), w is the mass fraction of solute, and c c* is the concentration driving force (mol m3). The mass-transfer coefcient, kc DAB/, can be derived from a number of correlations for the Sherwood number, Sh kcL/D, such as Sh 2 0.6Re1/2Sc1/3,25 where DAB is the binary diffusion coefcient (mol m1 s1), is the diffusion layer thickness (m), Re is the Reynolds number, and Sc is the Schmidt number. For surface integration controlled growth, Mullin26 identied three possible mechanisms: spiral growth, growth by twodimensional nucleation, and rough growth, of which spiral growth occurs most frequently. The general empirical equation for surface integration controlled growth is26 G k growS 1I (10)

dl dt

(7)
November 2005

where G is the linear particle growth rate (m s1) and kgrow is the growth rate constant (m s1). In the spiral growth mechanism, which occurs at low supersaturation levels, the growth units are incorporated only at kinks on the crystal surface. A defect must rst be generated on the surface, whereafter growth proceeds layerwise on this defect. This results in a spiral dislocation, which will continue to propagate as long as growth continues. For spiral growth at very low supersaturations, the exponent in Eq. 10 is I 2 and, for high supersaturations, I 1.25 At relatively higher supersaturations, two-dimensional nuclei occur on the crystal surface and the necessary kinks for further growth are generated. This is the two-dimensional nucleation growth mechanism, for which the exponent in Eq. 10 becomes I 2. At higher supersaturations, growth units can attach themselves anywhere on the crystal surface and the surface becomes rough. In this case, the exponent in Eq. 10 is I 1.
Vol. 51, No. 11 AIChE Journal

Table 1. Aggregation Kernels after Bramley et al.35

Mechanism Size independent Brownian motion (Smoluchowski, 1917) Gravitational (Berry, 1967) Shear [Smoluchowski (1917) and Low (1975)] Particle Inertia (Drake, 1972) Thompson kernel, empirical (Thompson, 1968) Kernel ( l , )

bl, x, s

6x2 l3

(13)

0 0 ( l )( l 1 1 ) 0( l ) 2( l ) 0( l ) 3 0( l ) 2( l 2 2) 0 ( l 3 3 )/( l 3 3 )

Similarly, Attarakih et al.,39 in their model of droplet breakage, use the following functional form for the breakage selection rate function S l k breaklm (14)

Besides the power-law formulation discussed above, other growth rate formulations are either a size-dependent linear growth rate27-29 or growth rate dispersion.30-32 Aggregation. The rst contribution to a model for particle aggregation was from Smoluchowski, who developed an aggregation frequency for particles aggregating by Brownian motion.33 Subsequently, the practice has been to describe aggregation in terms of an aggregation kernel, a measure of the frequency with which a particle of size l aggregates with one of size . Implicit in this description is the assumption that all particle collisions are binary in nature; in other words, the particle concentration is relatively low. Sastry34 rst proposed that the aggregation kernel be viewed as the product of two factors

where kbreak and m are positive parameters. For the daughter droplet distribution, b(l, x) can have two functional forms: (1) Uniform daughter droplet distribution, where it is assumed that there is an equal chance to form a daughter droplet of any smaller size when a mother droplet breaks up and thus the distribution is independent of daughter droplet volume b x, l 2 l

(15)

(2) Parabolic daughter droplet distribution, where it is assumed that there is a greater or lesser chance to form two daughter droplets of different sizes upon breakage of the mother droplet b x, l 24 x 2 xl 6 l 2 l3

L , 0f L ,

(11) (16)

The rst factor, 0, is the size-independent portion of the rate constant and depends on operating conditions such as the local uid velocity eld and chemical environment, whereas the second factor, f (L, ), is some function of particle size and often reects the mechanism of aggregation. Table 135 summarizes a range of aggregation kernels formulated in this way. Breakage. Birth and death rates of particles arising from breakage have been dened by Prasher36 and are reformulated in Nicmanis and Hounslow37 as number-based quantities. In the rate expressions, the breakage selection rate constant [S(l, s)] is the rate at which a fragment of a particle of size l will be selected to break. In the case study by the same workers,37 a binary breakage function and size-dependent breakage rate are used. Ramkrishna5 formulates an ad hoc model for breakage in which breakup of particles occurs independently of each other. The breakage frequency S(l, r, s, t) is thus sufcient to characterize the rate of destruction of particles of state (l, r) at time t, where r is the location in physical space and t is time dependency. This formulation is consistent with that suggested by Randolph and Larson4 in a general description of empirical birth and death processes. For simplicity of formulation, the rate at which particles are selected for breakage can be considered to be independent of spatial position and, at steady state, the time dependency falls away. Hounslow et al.38 use a breakage function for which there is an analytical solution. The selection rate is taken to increase simply with size cubed and the binary breakage function is formulated as one that gives uniform probability of all fragment sizes on a volume scale. That is S l , s kbreaklm (12)

Classication. A simple approach to classication is to specify the cut size of nes.4 A more comprehensive approach is to consider segregation as a function of energy dissipation and particle size11 or to include the effect of supercial ow velocity, ow direction, viscosity, and density difference between the liquid and the solid phases.11 Another approach is one where the selection function is a function of size and the state of the vessel contents. The form of the function, applied to stream k, is y k l , s Hl l0 (17)

where yk(l, s) is the selection function for stream k as a function of size and solution conditions and H(l l0) is the Heaviside step function. In this case, stream k is a representative sample of the bulk for sizes greater than some value l0, but containing no particles smaller than that size. Thus
out Hl l0 nl nk

(18)

4 where n out ) and n(l) k is the number density of stream k out (m 4 is the number density of vessel contents (m ). The hydrocyclone literature also has some useful size selectivity equations. For example, Plitt and Kawatra40 propose the following equation for the cut size

where S is the breakage selection rate (s1), kbreak is the breakage rate constant (s1 m3), l is the particle size (m), and
AIChE Journal November 2005 Vol. 51, No. 11

d 50c

0.46 0.6 1.21 0.063V Di Do e 14.8 D c 0.71 0.38 0.45 D u h Q S L0.5

(19)
2945

where d50(c) is the corrected d50 (m); Dc, Di, Do, and Du represent the inside diameters of hydrocyclone, inlet, overow, and underow, respectively (cm); V is the volumetric percentage of solids in the feed; h is the distance from the bottom of the vortex nder to the top of the underow orice (cm); Q is the ow rate of feed slurry (m3 h1); and S and L represent the density of solids and density of liquid, respectively (g cm1). The corrected d50 is taken from the corrected classication curve. The corrected d50 is calculated by assuming that, in all classiers, solids of all sizes are entrained in the coarse product liquid by short-circuiting in direct proportion to the fraction of feed water reporting to the underow.41 Lynch and Rao42 used weight percent solids, instead of using volume percent solids as in Eq. 19. Their model for calculating cut size is expressed as log10 d50 c 0.0173w 0.0695Du 0.0130Do 0.0048Q K1 (20)

where w is the weight percent solids in the feed slurry and K1 is a constant. Dissolution. Kramer et al.43 used the method proposed by Jager44 where the kinetic dissolution rate is calculated as a function of a mass-transfer coefcient and relative supersaturation. In another work, the same authors11 observe that dissolution is practically always limited by the mass-transfer step. Both the rate of dissolution and the rate of mass transfer limited growth will thus be limited by the local energy dissipation. Bermingham et al.20 used a kinetic expression that has a rstorder dependency on the supersaturation to describe the dissolution rate in their population balance equation. The rate constant for dissolution is calculated as follows k c l where
3 5 Dimp NeNimp V

D AB l4 2 0.8 3 l L

1/5

L DAB

1/3

(21)

where kc is the mass-transfer coefcient (ms1), c c* is the concentration driving force (mol m3), and m is the molar density of particles (mol m3). Despite this wealth of information relating to crystallization mechanisms, reporting of modeling studies in the open literature relating to the solution of PBEs tends to fall into one of two categories. In the rst category, the studies restrict themselves to a specic system and its measured kinetics and focus on the validation of models with experimental data. In the second category, studies tend to focus on the numerical aspects of solving the PBE, are usually restricted to the very simplest of generic mechanisms, and neglect to use the formulations that are considered to most accurately model the phenomena being considered. In the rst category, numerous studies are presented in the literature. Even the very comprehensive work undertaken by Bermingham et al.20 is limited to the ammonium sulfate system. In the second category, Kumar and Ramkrishna,45 in demonstrating a new discretization method for solving PBEs, use their technique to simulate various combinations of initial conditions, nucleation, growth, and aggregation. The functions for all the mechanisms considered are various combinations of exponential and gamma initial distributions; constant, linear, and size-dependent nucleation rates; constant, linear, and sizedependent growth rates; and constant and sum aggregation kernels. Kiparissides and Alexopoulos46 consider only constant, block pulse, and size-dependent and/or surface area dependent nucleation rates; constant and size-dependent growth rates; and a Brownian aggregation rate kernel. This work, however, is proposed to be truly generic, in that practically any chemical system for which there is thermodynamic data may be considered. In addition, the framework incorporates a comprehensive selection of kinetic processes relevant to crystallization and allows the selection of kinetic models for the processes that are believed to most accurately model the various phenomena.

The overall PBE

Once the exit ow term has been modied for classication, and the terms for net rate of destruction attributed to nucleation, growth, dissolution, aggregation, and breakage have been incorporated, the maximal form of the PBE is47

(22)

and kc is the mass-transfer coefcient (ms1), DAB is the binary diffusion coefcient (m2 s1), l is crystal length (m), is the specic power input (W kg1), L is the kinematic viscosity of the liquid (m2 s1), Ne is the power number (also known as the Newton number), Nimp is impeller frequency (s1), Dimp is impeller diameter (m), and V is tank volume (m3). Dissolution can also be incorporated as a size-independent shrinkage rate constant Dl, s D (23)

k1

Q n l nl Q
Kin K out in in k k k1

out k k

y l, s V B0 s fnucl, s

1 2

d d Gl, snl D0 l Dl, snl dl dl

l x , x, s
3 3 3

where D is the shrinkage rate constant (m s1), or as a mass transfer limited process, as follows Dl, s
2946

nl

l
33 l x3

n3 l3 x3 n xdx

l, x, sn xdx Sl, snl

2 k c c c * m

(24)
November 2005 Vol. 51, No. 11

S x, sbl, x, sn xdx

AIChE Journal

(25)

subject possibly to the boundary conditions n() n(0) 0, where y k l , s (26)

is a function of [B0(s)f nuc(, s), G(, s), D(, s), n()] and 3 l y(l, s)dl. y 3 (s) 0 Replacing the third moment with the suspension density M T SkVm3 SkV

is the selection function or grade efciency and is a function of size and the state of the vessel contents B0 s fnucl, s (27) gives

l3 nl dl

(35)

is the rate of destruction of particles attributed to nucleation d Gl, snl dl (28)

k1

Q M Q
in out
in k in Tk k1

out k

out MTk VSkVr3

(36)

is the rate of destruction of particles stemming from growth D0 l d Dl, snl dl (29)

This last equation has a crucial role in coupling the material and population balances.

Numerical form of the PBE

Analytical solutions for Eq. 25 are scarcely ever available, so it follows that some numerical strategy will be required. All such numerical strategies known to the authors represent the size distribution at a vector of sizes l, at neq discrete points l l1 , l2 , . . . , ln eq (37)

is the rate of destruction of particles attributed to dissolution and D0(l), the rate of disappearance, is a bookkeeping term to prevent the occurrence of particles of negative size 1 2

3 l3 x3 , x, s

l
3 3

l x3

n3 l3 x3 n xdx (30)

by neq discrete values N, related reasonably directly to the number density N N1 , N2 , . . . , Nn eq (38)

is the rate of destruction of particles stemming from aggregation and

Sl, snl

In this report, it is assumed that each of the Ni is the number of particles, per unit volume of suspension, in the size range (li, li1), that is Ni

S x, sbl, x, sn xdx

(31)

is the rate of destruction of particles stemming from breakage. The moment transformation so often used in the application of PBEs to process models is not possible for arbitrary size dependency. It is, however, possible to make a partial transformation for the third moment

li1

n l dl

(39)

li

m3

l 3n l dl

(32)

In the most sophisticated discretization strategies, the values for l are chosen and adjusted adaptively by the algorithm as it solves for values for N. This approach offers high accuracy and short execution times at the cost of very complex programming, and is not pursued here. A more tractable family of discretization strategies involves xing the vector l for the entire problem. In this case virtually all strategies adopt a geometric discretization of the size axis so that the ratio r l i1 2 1/3q li q J

Transformation of Eq. 25 gives

k1

Q m Q
Kin K out in k in 3k k1

(40)

out k

m s Vr3
out 3k

(33) is a constant.47 Satisfactory accuracy for CST problems is usually achieved with r 21/3 so that, for a 1000-fold range of particle sizes, neq 30 is required. (For growth-dominated PFR problems a value of r 21/12 may be needed in which case, for the same size range neq 120 is needed.) The discretized form of Eq. 33, the discretized population balance equation (DPBE), is
Vol. 51, No. 11 2947

where r 3 n , s
AIChE Journal

l3 rn, l, sdl

(34)

November 2005

k1

Q N l Q
Kin K out in k in k,i k1

out k

N l VRi N, s
out k,i

i 1 neq (41) and

M T SkVm3 SkV

i1

l N
n eq 3 i

(44)

where Nk,i is the number of crystals per unit volume in the kth stream and the ith size interval and Ri is the rate of destruction of particles in the ith size interval. The classication functions relate the exit stream size distributions to the size distribution in the bulk l i N i N k,i y k where the average size in the ith size interval is 1r li li 2 The discrete forms of Eqs. 35 and 36 are (42)

k1

Q M Q
Kin K out in k in Tk k1

out k

out Tk

VSkV

i1

l R
n eq 3 i

(45)

The task now is to determine expressions for Ri.

Nucleation, growth, and dissolution

(43) To resolve the issue of how to apply the boundary condition at zero size, it is necessary to consider nucleation, growth, and dissolution (ngd) together. Thus, adapting Hounslow et al.47 to incorporate dissolution

2 r2 r 1 2 D1 G1 N1 2 D G2 N2 1 rl1 r 1 r 1 2 2 r r D Gi 1 Ni 1 Di Gi Ni 2 D Gi 1 Ni 1 R ingd B0 finuc 1 rli r2 1 i 1 r 1 i1 2 r nuc D B0 fn Gn eq 1 Nn eq 1 Dn eq Gn eq Nn eq eq 1 rln eq r2 1 n eq 1

nuc D0 B0 f1

i1 1 i neq i neq (46)

The D0 term, the rate of disappearance, in Eq. 46 is not a constitutive equation, unlike the nucleation rate constant B0, but just a bookkeeping term to prevent the occurrence of particles of negative size. Further evidence is that the magnitude of D0 depends on the solution to the PBE, not on conditions in the vessel, such as supersaturation. As shown in Eq. 49, the value of D0 depends on n(0) and cannot be determined without knowledge of that value. By contrast the nucleation rate depends on the state of the vessel.

For the continuous system of equations, the boundary condition to be applied, n(0) 0, results in D0 D0 , s G0 , sn0 B D0 , s G0 , sn0
0

fnuc l fnuc continuous

(47)

If the rate of growth at zero size is greater than the rate of dissolution, the rate of disappearance will be zero, in which case we have

0 D 0 B D0 , s G0 , sn0 D0 , s G0 , sn0
0

D 0 , s G0 , s 0 fnuc l fnuc continuous

(48)

It is probably most sensible to apply this equation on the rst size interval and to represent n by a rst-order difference, in which case

Aggregation
The full form for arbitrary q, given by Litster et al.48 and rened by Wynn, is49
i 1

N1 B 0 D 1 G 1 l2 l1 D0 N1 D1 G1 l2 l1
2948

D1 G1 0 f nuc l f
nuc

R iagg (49)

j1

i 1, j

Ni 1 Nj

2 j i 1 / q 21/ q 1 2 ji1/q 1 2 p1/q 2 1/q 1

AIChE Journal

continuous
November 2005

p2 jip1

ip

ip, jN ipN j

Vol. 51, No. 11


q1 i1p1

1 N2 2 iq,iq iq 2
ji/q

Embedded within this is the discrete equivalent of Eq. 42 N i,k y k,iN i 2 2 21/ q 1
1/ q p/q

p1 ji1p il1

(55)

ip, jN ipN j

The suspension density is calculated from M T SkV

j1

2 ji/q i, jN iN j 1/q 2 1

ji12

i, jN iN j (50)

i1

l N
n eq 3 i

(56)

where p Int 1

q ln1 2 p / q ln 2

(51)

and Int[x] subtracts the fractional part (if any) from x. Note that care must be taken here when calculating the value of p because of rounding errors when the integer part of the calculated expression is extracted.

Breakage
The discrete form developed by Hounslow et al.38 is R ibreak Si Ni

ji

b SN
n eq i, j j

(52)

There are two possible solution strategies for the discretized population balance equations. The most obvious solution strategy is to solve the (neq 1) equations using a simple nonlinear equation solver with numerical estimation of the Jacobian (where the extra equation is Eq. 56). This has been found to be a robust and simple-to-implement technique50 and this is one of the strategies that has been adopted here. A NewtonRaphson technique has been implemented, with nite-difference approximations for the derivatives. The system of equations is dense because all equations involve all unknowns. Nonetheless, robust convergence is obtained even with poor initial estimates of the solution and a numerical Jacobian. However, the expense of evaluating the Jacobian is considerable, and so other approaches are desirable. The second solution strategy uses the observation that, in the absence of aggregation, Eq. 53 is in fact linear in N. With that observation, it is useful to write Eq. 53, with the results of Eqs. 46 and 52 but without Eq. 50 shown explicitly. For example, with 1 i neq, this becomes

Although this seems a very simple recasting of the continuous form of the equation, it must be understood that the Si and bi,j cannot be deduced by evaluating the continuous functions at li and l j . Instead, a more complicated process is required (see Eqs. 31 and 34 in Hounslow et al.38).

k1

Q N
Kin in k

K out

in k,i

Ni

k1

out k,i k

y V B0 finuc

Solving the DPBE

The maximal form of the DPBE is obtained by combining Eq. 41 with Eqs. 42, 46, 50, and 52

2 r D Gi 1 Ni 1 Di Gi Ni 1 rli r2 1 i 1

k1

Q N
Kin in k

K out

in k,i

Ni

k1

r 2 D Gi 1 Ni 1 Si Ni r 1 i1

out k,i k

y VRi N, s VRingd Riagg R

break i

j1

b SN R
n eq i, j j j

agg i

i 1 neq

(53)

This equation, and the results for i 1 and i neq, may be written as a set of equations A BN 0 (57)

For a problem where the feed size distributions are given, Eq. 53 provides neq equations in neq unknowns: that is, the vector of Ni values, N. Solving the DPBE is then a matter of solving these neq equations for these neq unknowns. Irrespective of the sophistication of the method or neness of the discretization used, Eq. 25 is transformed into neq simultaneous nonlinear algebraic equations a i B i N NT Ci N 0 i 1, 2, . . . , neq (54)

where A is a column vector of length neq and B is an neq neq matrix that can be decomposed into two matrices, E a tridiagonal matrix and F an upper-right triangular matrix BEF (58)

where ai is a (different) constant for each equation, Bi is a (different) vector of constants (of length neq) for each equation, and Ci is a (different) matrix of size neq neq for each equation.
AIChE Journal November 2005 Vol. 51, No. 11

A1 A2 An eq

(59)

2949

b1 c1 0 a2 b2 c2 0 a3 b3 c3 E an eq 1 bn eq a n eq 0

c n eq 1 b n eq

Table 2. Stream Denitions

Stream Type Variable Temperature Pressure Total ow Total solid ow Total liquid ow Liquid species ows Solid species ows Vapor species ows Volumetric ow Specic enthalpy Size distribution Unit C Pa mol s1 mol s1 mol s1 mol s1 mol s1 mol s1 m3 s1 J mol1 m3 T P n T n ST n LT n L (n L1 , n L2 , . . . , n LJ) n S (n S1 , n S2, . . . , n SJ) n V (n V1, n V2, . . . , n VJ ) Q H N ( N 1 , N 2 , . . . , N n eq) Symbol SL V

(60)

F1, n eq F1,1 F1,2 F1,3 F2, n eq 0 F2,2 F2,3 0 0 F3,3 F 0 Fn eq 1, n eq 1 Fn eq 1, n eq Fn eq, n eq 0 0

(61) (62)

and the elements of A and B are

Ai

k1

Q N
K in in k K in

in k ,1

nuc agg VB0 f1 VD0 VR1

i1 1 i neq

3. While N is changing, Do: a. Calculate A b. Determine N from Eq. 67 4. Exit

k1

Q N
in k

in k,i

VB0 finuc VRiagg

The Material and Energy Balances Stream data

Representation of two types of streams is included: solid liquid streams and vapor-only streams. Within each stream data specication, ows are represented as species molar ows for each phase and some redundant data are included to prevent frequent recalculation. For a system containing J species the stream specications are as shown in Table 2.

ai

2 Vr D i1 G i1 1 r r 2 1 l i
Kout

1 i n eq

(63)

bi

2 V D 1 G 1 VS 1 1 r r 2 1 l 1 2VDi Gi VSi 1 rli

K out

k1

out k ,1 k

i1 1 i neq (64)

The material balance

The material balance states simply that the molar ow in of each species equals the molar ow out of that species

k1

out k,i k

k 1

Kin

K out in n T k

k1

out Tk

n V

(68)

c i V

2 r D Gi 1 2 1 rli r 1 i 1

1 i neq (65)

F i, j Vbi , j Sj

1 i neq; j i

(66)

It may be seen that, for any one set of conditions for which the DPBE is to be solved, B is a constant, from which it follows that B1 is also a constant. On the other hand, A depends on the aggregation rate, which in turn depends on N by Eq. 50. The strategy is to invert Eq. 57 to give N B 1 AN (67)

where is a matrix of stoichiometric coefcients for the E independent solution-phase chemical reactions taking place . Note that the prowith reaction rates given by the vector duction of solid does not count as a reaction here, given that Eq. 68 counts, for example, NaCl(aq) with the same species number as that of NaCl(s). The rate of generation of solid can be deduced from the PBE, Eq. 25. The mass balance couples directly to the PBE by the suspension density MT, as given in Eq. 36, which in turn can be related to the molar ows as follows

M
M T,k 1 S

R, i Sk , i

where now A is shown explicitly to be a function of N. Equation 67 may be applied iteratively using the following algorithm: 1. Assemble B 2. Find B1
2950 November 2005

M
i

R, i Sk , i

1 L

M
i

(69) n

R, i Lk , i

where MR,i is the relative molar mass for species i and S and L are the mass densities for the solid and liquid phases, respectively.
Vol. 51, No. 11 AIChE Journal

It will usually be the case that the solution composition of all exiting liquid streams will be the same so that, if the mol fractions of each species in solution in the vessel are x, then n Lk , i xi n LTk (70)

Heat input, q Vessel size, V Selection functions yk(l) Composition of the vapor stream Vessel operating pressure

Unknowns to be determined
The model must then determine the complete specication of the product streams Solid liquid streams: (7 2J neq) Kout variables Vapor stream: 4 J , N, that is, 3 J Vessel conditions: T, x, HS, HL, MT, X E neq (note that x counts as J 1 unknowns) The rate constants, such as B0, G0, D, , S (that is, 5) The total number of unknowns is: (7 2J neq) (Kout 1) 5E

where n LTk is the total molar ow of liquid in stream k. The total ow for each stream is STk n LTk n Tk n Similarly, if the solid has mol fractions xS n Sk , i x n STk Si and the vapor mol fractions xvap n ivap xivap n vap T (73) (72) (71)

Equations available
In Table 3, the available equations for a Type I (that is, nonhigh yield) crystallization are presented. So, for example, for a system with three cations, three anions, four complexes, and water, that is, J 11 four independent reactions, E 4 two exit streams, that is, Kout 2 neq 30 the total number of equations to be solved is (7 22 30) (2 1) 5 4 186. For a Type II, or high-yield system, one of the kinetics equations would be lost, and an additional equilibrium equation (for the solid) asserted.

The volume ow rate for the solidliquid streams may be determined from

Qk

i1

M n
J R, i

Sk , i S

n Lk , i L

(74)

The energy balance

The energy balance can be written most succinctly in terms of the specic enthalpies H. The balance is

Tearing
There are many commercial thermodynamic modules that allow the solution of material and energy balance problems,8,9 so it seems sensible to use these and therefore to deconstruct the problem between the thermodynamic calculations on the one hand and the population balances on the other hand. The supersaturation ratio is calculated at a xed T, P, composition, and mass of crystal precipitated. The supersaturation ratio then allows the calculation of the kinetic parameters to be used in the solution of the population balances. In fact, this is the real strength of a model such as this. Coupling the solution to the PBE to a thermodynamic package means that the supersaturation is rigorously calculated on the basis of the comprehensive solution chemistry embodied in such a package. The solution of the population balances then allows the calculation of a new vector of number of particles (Ni) and the mass of crystal precipitated from the suspension density. Thus, for Type I systems, the iteration occurs over (neq 1) variables. For Type II systems, the algorithm still applies, but it is necessary to iterate over the growth rate. Type I Systems: NewtonRaphson Solution Algorithm 1. Estimate Mc, the total crystal mass in the crystallizer produced by crystallization, by varying Mc to match the userspecied estimate of the supersaturation ratio. Compute umc, a transform of Mc, as follows u mc ln 1
Vol. 51, No. 11

k1

n H n
Kin K out in Tk in k k1

out Tk

out vap Hk n vap q T H

(75)

It is to be expected that the specic enthalpy of the solution and of the solids in each of the exit streams will be the same as that in the vessel, that is, TSkHS n TLkHL n TkHk n (76)

In addition, it is assumed that all exit streams have the same temperature and pressure as those of the vessel
out Tk Tvap T out Pk Pvap P

(77)

Solving the Balance Equations Problem specication

A standard specication would be Complete description of the feedstreams Total ow rates of all but one of the solidliquid exit streams Crystallization kinetic equations: B0(s), f nuc(l, s), G(l, s), D(l, s), (l, x, s), S(l, s), b(l, x, s)
AIChE Journal November 2005

Mc Mcmax

(78)
2951

Table 3. Equations Available

Source Thermodynamics Equation H L( T , P , x) H S( T , P ) H vap( T , P , xvap) T ( P , x) Equilibria 54 Given 55 68 56, 69 56 70 71 72 75 76 77 73 Given 74 Instances 1 1 1 1 E n eq 5 n eq K out J K out 1 J K out K out J K out 1 K out 2( K out 1) J K out 1 K out (7 2 J n eq) ( K out 1) 5 E

Population balance Kinetics Classication Material balance Solid ow Denition of suspension density Solution composition Total ows Solid composition Energy balance Steam enthalpies Equilibration of T and P Vapor species ows Exit ows Volume ows Total

This transform prevents Mc from exceeding the maximum amount of solute available Mcmax and is used only for this purpose in the algorithm. For classied product removal, the calculated value of Mcmax is the same as that for mixed product removal because it is merely used as the upper numerical bound. 2. Estimate a starting value for Ni such that MT (suspension density) Mc/Vm. 3. Calculate MT, the suspension density, for the current set of Ni values. 4. Calculate the supersaturation ratio S for the current value of Mc, the mass of crystal precipitated. 5. Calculate the predicted value of the mass of crystal precipitated, Mc_calc from M c_calc MT Vm MT 1 c

13. Compute Mc from M c M cmax1 expumc (80)

(79)

6. Calculate the kinetic rate constants from the vector s. 7. Compute the error for RHS(i) as {[RHS(i) of Eq. 48]2}. Compute the error for RHS(neq 1) Mc Mc_calc. 8. Compute the derivatives d(Eq. 53)/dNi using nite differences. 9. Compute the derivatives d(Eq. 53)/dumc. 10. Compute corrections to the Ni and umc from the solution of Ax b where A is the the Jacobian; b is the RHS vector of errors in the material and energy balances and Mc Mc_calc; x represents corrections to the variables Ni and umc, designated Ni and umc. 11. Set 1. 12. Update Ni using Ni(k 1) Ni(k) (Ni) and umc(k 1) umc(k) (umc), where k is the iteration number.
2952 November 2005

14. Compute a new error using steps 3 8. 15. If the error was reduced, check for convergence. In this case, the tolerance is 1.0E-6. If converged, exit. 16. If the error was reduced but not converged, go to step 3 and calculate another Newton step. 17. If the error was not reduced, set /4 and go to step 12 to take a shorter step in the direction of the Newton correction. Type I Systems: Successive Substitution with Root-Bracketing Solution Algorithm. The root bracketing algorithm below is used to solve Eq. 36 for the suspension density MT. 1. Perform a mass and energy balance assuming no solid is precipitated. 0 a. Denote MT as M T . b. Calculate the kinetic rate constants from the vector s. c. Solve the PBE; note the error in Eq. 36 as 0. 2. Perform a mass and energy balance assuming solid is precipitated to form a saturated solution. max a. Denote MT as M T . b. Calculate the kinetic rate constants from the vector s. c. Solve the PBE; note the error in Eq. 36 as max 3. If 0 max 0, exit with error. 4. If 0 0, 1 0 2 max a. then set M T MT , MT MT , 1 0, 2 max 1 max 2 0 b. else set M T MT , MT MT , 1 max, 2 0 1 2 5. Set MT 0.5(M T MT ). a. Solve the material and energy balance with the mass precipitated corresponding to MT. b. Calculate the kinetic rate constants from the vector s. c. Solve the PBE; note the error in Eq. 36 as . d. If 1 0, 1 i. then M T MT, 1 2 ii. else M T MT, 2 6. Do While tol. 0
Vol. 51, No. 11 AIChE Journal

1 2 a. MT 0.5(M T MT ). b. Solve the material and energy balance with the mass precipitated corresponding to MT. c. Calculate the kinetic rate constants from the vector s. d. Solve the PBE; note the error in Eq. 36 as . e. If 1 0, 1 i. then M T MT, 1 2 ii. else M T MT, 2 7. Exit.

For type I crystallization, the choice of the most appropriate algorithm depends on the nature of the system to be modeled. For cases where the index of aggregation (Iagg) is small, where Iagg 1 (m0/B0) 0.333, it is appropriate to use the successive substitution method because this will converge economically and rapidly. For cases with higher indices of aggregation, the NewtonRaphson solver is more appropriate, given that the nonlinearity introduced by the aggregation terms becomes more signicant. This is covered in more detail in a subsequent article. Type II Systems: Successive Substitution with Root-Bracketing Solution Algorithm. The root-bracketing algorithm below is used to solve Eq. 36 for the size-independent portion of the growth rate G0, starting from a pair of rst guesses (0, G max 0 ) 1. Perform a mass and energy balance with solid precipitated to form a saturated solution. 2. Solve the PBE with G0 G 0 0 0; note the error in Eq. 36 as 0. 3. Solve the PBE with G0 G max 0 ; note the error in Eq. 36 as max. 4. Do While 0 max 0 a. G max 2G max 0 0 . b. Solve the PBE with G0 G max 0 ; note the error in Eq. 36 as max. 5. If 0 0, 0 2 max 1 0 2 max a. then set G 1 0 G 0, G 0 G 0 , , 1 max 2 0 1 max 2 b. else set G 0 G 0 , G 0 G 0 , , 0. 2 6. Set G0 0.5(G 1 0 G 0 ). a. Solve the PBE; note the error in Eq. 36 as . b. If 1 0, 1 i. then G 1 0 G0, 2 2 ii. else G 0 G0, . 7. Do While tol. 0 2 a. G0 0.5(G 1 0 G 0 ). b. Solve the PBE; note the error in Eq. 36 as . c. If 1 0, 1 i. then G 1 0 G0, 2 ii. else G 2 G , . 0 0 8. Exit.

(1) It links the size distribution from a crystallizer to the kinetics of the processes taking place in that crystallizer. (2) It couples the kinetics to the mass balance equation, by enabling calculation of the mass of solid produced. Solution of the PBE requires both conversion to a discrete form as well as numerical solution of the discrete form. Two algorithms for the solution of the resulting matrix of algebraic equations are proposed: The rst solves the population balance equations using a NewtonRaphson solver with nite-difference approximations for the derivatives, converging around a variable related to the crystal mass and Ni for each size interval. The second uses a successive substitution technique with root bracketing and iterates around the suspension density. The choice of the most appropriate algorithm depends on the nature of the system to be modeled. A similar framework is suggested for the solution for Type II crystallization, except that the iteration variable is the growth rate at a xed supersaturation ratio.

Acknowledgments
This work was supported by the EPSRC of the United Kingdom, OLI Systems Inc., and the National Research Foundation of South Africa.

Notation
a a A A A b b(l, x, s) B B0(s) c c c c* d50(c) D D(l, s) D0 DAB a coefcient, Eq. 54 component of E component of A parameter in Eq. 3 area of crystal, m2 component of B and E breakage function, m1 parameter in Eq. 3 nucleation rate, m3 s1 component of C component of E solute concentration, kg m3 equilibrium solute concentration, kg m3 corrected d50 for the hydrocyclone (Eq. 19), m shrinkage rate constant in Eq. 23, m s1 linear rate of shrinkage, m s1 rate of disappearance, m3 s1 binary diffusion coefcient (m2 s1) in calculating kc in Eq. 9, mol m1 s1 impeller diameter, m inside diameters of hydrocyclone, inlet, vortex nder, and apex, respectively (Eq. 19), cm number of independent reactions the nuclei size distribution, m1 the discrete equivalent of f nuc(l, s) component of F linear rate of crystal growth, m s1 maximum guessed value for growth rate, m s1 specic enthalpy, J mol1 Heaviside step function in Eq. 17 the distance from the bottom of the vortex nder to the top of the underow orice in the hydrocyclone (Eq. 19), cm a size class index of aggregation, Iagg 1 (m0/B0) number of species present a constant in Eq. 20 rate constants for nucleation, kg1 s1 a stream number iteration number area shape factor rate constant for breakage, m3 s1 mass-transfer coefcient, ms1

Dimp Dc, Di, Do, Du E f nuc(l, s) f i nuc F G(l, s) G max 0 H H h i Iagg J K1 ii iii iv v k nuc , k nuc , k nuc , k nuc k k ka kbreak kc

Conclusions
The modeling of a single continuous stirred tank crystallizer with any number (1) of product streams, any number of feed streams, one vapor product stream, nonrepresentative sampling, and only one solid species has been described. The mechanisms allowed are nucleation, growth, aggregation, dissolution, breakage, and classication. The model constitutes coupled material, energy, and population balance equations. The population balance equation, Eq. 25, serves two roles:
AIChE Journal November 2005

Vol. 51, No. 11

2953

kgrow kV K l l0 m m0 m3 Mc Mc_calc Mcmax MR MT n(l) neq n LT n ST n T n j Ne Ni Nimp P q q

Q Q r[n, l, s] r r3 Ri Re S Sc Sh S(l, s) t T umc V V

Vm w x x y(l, s)

rate constant for growth, ms1 volume shape factor number of streams particle size, m cut size for classication, m mass of crystal, kg zeroth moment third moment mass of crystal precipitated, kg predicted mass of crystal precipitated, kg maximum mass of solute available, kg relative molar mass suspension density, kg m3 number density, m4 number of size classes total molar ow of liquid, mol s1 total molar ow of solid, mol s1 total molar ow, mol s1 molar ow of species j, mol s1 power number (also known as Newton number) number of particles per unit volume in size class i, m3 impeller frequency, s1 pressure, Pa heat ow, W parameter determining the neness of the discretization ow rate of feed slurry to the hydrocyclone (Eq. 19), m3 h1 volumetric ow rate (L min1 in Eq. 20), m3 s1 density functional describing rate of destruction of particles, m4 s1 discretization ratio, Eq. 40 rate of destruction of third moment, s1 rate of destruction of particles in interval i; the discrete equivalent of r[n, l, s] Reynolds number supersaturation ratio Schmidt number Sherwood number selection rate constant for breakage, s1 time (s) and time dependency for breakage frequency temperature, K transform of Mc in Eq. 78 tank volume, m3 volumetric percentage of solids in the feed to a hydrocyclone, Eq. 19 molar volume of the aqueous phase from calculated from supersaturation, m3 mass fraction of solute mol fraction particle size, m selection function or grade efciency for classication

s a vector of conditions describing the solution, such as supersaturation, temperature, etc. x vector of mol fractions

Greek letters
0 p w nuc L S m L
size-independent portion of aggregation rate constant, m3 s1 aggregation rate constant, m3 s1 Dirac delta distribution diffusion layer thickness, m dened in Eq. 51 error in Eq. 36 specic power input, W kg1 weight percent solids in the feed slurry to a hydrocyclone, Eq. 20 particle size, m rate constant for secondary nucleation, m3 kg1 kinematic viscosity of the liquid, m3 s1 solids density, kg m3 molar density of particles, mol m3 liquid density, kg m3 density, kg m3

Subscripts and superscripts

agg break grow I aggregation breakage growth exponent for concentration or supersaturation in nucleation and growth rate equations size interval number inlet streams exponent for suspension density in nucleation and growth rate equations stream number liquid exponent in volume-dependent breakage rate equation nucleation, growth, and dissolution nucleation outlet streams solid total vapor vapor

i in J k L m ngd nuc out S T V vap

Literature Cited
1. Hulbert HM, Katz S. Some problems in particle technology: A statistical mechanical formulation. Chem Eng Sci. 1964;19:555. 2. Randolph AD, Larson MA. Transient and steady state size distributions in continuous mixed suspension crystallizers. AIChE J. 1962;8: 639-645. 3. Randolph AD, Larson MA. A population balance for countable entities. Can J Chem Eng. 1964;42:280-281. 4. Randolph A, Larson M. In: Mullin JW, ed. Theory of Particulate Processes: Analysis and Techniques of Continuous Crystallization. London, UK: Academic Press; 1993. 5. Ramkrishna D. The status of population balances. Rev Chem Eng. 1985;3:49-95. 6. Ramkrishna D. Population Balances. Theory and Applications to Particulate Systems in Engineering. San Diego, CA: Academic Press; 2000. 7. Hounslow MJ, Wynn E. Short-cut models for particulate processes. Comput Chem Eng. 1993;17:505-516. 8. Stream Analyzer V1.2. Morris Plains, NJ: OLI Systems Inc.; 2003. 9. Aspen Plus. Cambridge, MA: Aspen Technology Inc.; 2002. 10. Hounslow MJ, Mumtaz HS, Collier AP, Bramley AS. A micro-mechanical model for the rate of aggregation during precipitation from solution. Chem Eng Sci. 2001;56:2543-2552. 11. Bermingham SK, Kramer HJM, van Rosmalen GM. Towards on-scale crystalliser design using compartmental models. Comput Chem Eng. 1998;22:S355.

Vectors
A B C E F X l N n L (n L1 , n L2, . . . , n LJ ) n S (n S1 , n S2, . . . , n SJ) n V (n V1, n V2, . . . , n VJ ) r a vector of coefcients, Eq. 57 a matrix of coefcients, Eq. 57 a matrix of coefcients, Eq. 57 a tridiagonal matrix, Eq. 58 an upper right triangular matrix, Eq. 58 vector of reaction rates, mol m3 s1 list of particle sizes, Eq. 37 a list of particle numbers Eq. 38 matrix of stoichiometric coefcients species molar ow in the liquid, mol s1 species molar ow in the solid, mol s1 species molar ow in the vapor, mol s1 location in physical space for breakage frequency

2954

November 2005

Vol. 51, No. 11

AIChE Journal

12. Volmer M, Weber A. Keimbildung in u bersa ttigten Gebilden. Z Phys Chem (Leipzig). 1926;119:277. 13. Turnball D, Fisher JC. Rate of nucleation in condensed system. J Chem Phys. 1965;17:71. 14. Miers HA, Isaac F. The refractive indices of crystallising solutions, with special reference to the passage from the metastable to the labile condition. J Chem Soc Lond. 1906;89:413. 15. Garside J. The concept of effectiveness factors in crystal growth. Chem Eng Sci. 1971;26:1425-1431. 16. Sohnel O, Garside J. Precipitation: Basic Principles and Industrial Application. Oxford, UK: ButterworthHeinemann; 1992. 17. Puel F, Fevotte G, Klein JP. Simulation and analysis of industrial crystallisation processes through multidimensional population balance equations. Part 1: A resolution algorithm based on the method of classes. Chem Eng Sci. 2003;58:3715. 18. Garside J. Industrial crystallisation from solution. Chem Eng Sci. 1985;26:1425-1431. 19. Gahn C, Mersmann A. Brittle fracture in crystallization processes. Part A. Attrition and abrasion of brittle solids. Chem Eng Sci. 1999;54: 1273-1282. 20. Bermingham S, Verheijen P, Kramer H. Optimal design of solution crystallization processes with rigorous models. Trans IChemE. 2003; 81:893-903. 21. Eek RA, Dijkstra SJ, van Rosmalen GM. Dynamic modeling of suspension crystallizers using experimental data. AIChE J. 1995;41: 571. 22. O Meadhra R, Kramer HJM, van Rosmalen GM. A model for secondary nucleation in a suspension crystallizer. AIChE J. 1996;42:973. 23. Ottens EPK, Janse AH, de Jong EJ. Secondary nucleation in a stirred vessel cooling crystalliser. J Crystal Growth. 1972;13/14:500. 24. McCabe WL. Crystal growth in aqueous solutions. Ind Eng Chem. 1929;21:30. 25. Giulietti M, Seckler MM, Derenzo S, Re MI, Cekinski E. Industrial crystallization and precipitation from solutions: State of the technique. Braz J Chem Eng. 2001;18:423-440. 26. Mullin JW. Crystallization. Oxford, UK: ButterworthHeinemann; 1997. 27. Branson SH. Factors in the design of continuous crystallisers. Br Chem Eng. 1960;5:838. 28. Canning TF, Randolph AD. Some aspects of crystallization theory: Systems that violate McCabes Delta L law. AIChE J. 1967;13:5. 29. Abegg CF, Stevens JD, Larson MA. Crystallization size distribution in continuous crystallizers when growth rate is size dependent. AIChE J. 1968;14:118. 30. White ET, Wright PG. Magnitude of size dispersion effects in crystallization. Chem Eng Prog Symp Ser. 1971;67:81. 31. Larson MA, White ET, Ramarayanan KA, Berglund KA. Proc of 75th Annual Meeting of AIChE, Los Angeles, CA; 1982.

32. Human HJ, van Enekevork WJP, Bennema P. Industrial Crystallisation 81. Amsterdam: North-Holland; 1982. 33. Smoluchowski MV. Mathematical theory of kinetics of the coagulation of colloidal solutions. Z Phys Chem. 1917;92:129. 34. Sastry KVS. Similarity size distribution of agglomerates during their growth by coalescence in granulation or green pelletization. Int J Miner Process. 1975;2:187. 35. Bramley A, Hounslow M. Aggregation during precipitation from solution: A method for extracting rates from experimental data. J Colloid Interface Sci. 1996;183:155. 36. Prasher CL. Crushing and Grinding Process Handbook. New York, NY: Wiley; 1987. 37. Nicmanis M, Hounslow M. Finite-element methods for steady-state population balance equations. AIChE J. 1998;44:2258-2272. 38. Hounslow M, Pearson J, Instone T. Tracer studies of high-shear granulation: II. Population balance modeling. AIChE J. 2001;47:19841999. 39. Attarakih MM, Bart HJ, Faqir NM. Optimal moving and xed grids for the solution of discretized population balances in batch and continuous systems: Droplet breakage. Chem Eng Sci. 2003;58:1251-1269. 40. Plitt LR, Kawatra SK. Estimating the cut (d50) size of classiers without product particle-size measurement. Int J Miner Process. 1976; 5:364-378. 41. Wills BA. Mineral Processing Technology. Oxford, UK: Butterworth Heinemann; 1997. 42. Lynch AJ, Rao TC. Modelling and scale-up of hydrocyclone classiers. Proc of 11th Mineral Processing Congress, Cagliari, Italy; 1975. 43. Kramer HJM, Bermingham SK, van Rosmalen GM. Design of industrial crystallisers for a given product quality. J Cryst Growth. 1999; 198/199:729-737. 44. Jager J. Control of Industrial Crystallisers; the Physical Aspects. PhD Thesis. Delft, The Netherlands: Delft University of Technology; 1990. 45. Kumar S, Ramkrishna D. On the solution of population balance equations by discretizationIII: Nucleation, growth and aggregation of particles. Chem Eng Sci. 1997;52:4659-4679. 46. Kiparissides CA, Alexopoulos AH. Solution of PBEs for prediction of PSD in particulate systems: Effect of combined nucleation, growth and aggregation. Industrial Processes of Particle Formation. Proc of AIChE 2002 Annual Meeting, Indianapolis, IN, Nov. 3 8; 2002. 47. Hounslow MJ, Ryall RL, Marshall VR. A discretized population balance for nucleation, growth, and aggregation. AIChE J. 1988;34: 1821. 48. Litster J, Smit D, Hounslow M. Adjustable discretized population balance for growth and aggregation. AIChE J. 1995;41:591-603. 49. Wynn EJW. Improved accuracy and convergence of discretized population balance of Lister et al. AIChE J. 1996;42:2084-2086. 50. Hounslow MJ. A discretized population balance for continuous systems in a steady state. AIChE J. 1990;36:106-116.
Manuscript received Nov. 5, 2004, and revision received Feb. 21, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2955

Scale-up Effect of Riser Reactors: Particle Velocity and Flow Development

Aijie Yan and Jesse (Jingxu) Zhu
Dept. of Chemical and Biochemical Engineering, University of Western Ontario, London, Ontario, Canada N6A 5B9 DOI 10.1002/aic.10556 Published online August 8, 2005 in Wiley InterScience (www.interscience.wiley.com).

The inuence of riser diameter on the axial and radial particle velocity proles and ow development is studied in a 10 m high twin-riser system (76- and 203-mm ID risers). Cross-sectional average particle velocity is somewhat lower for the larger riser with a steeper radial particle velocity prole. The ow development in the riser center is nearly instant with the particle velocity remaining high. There is no signicant difference for the two risers in the center region. In the wall region, the ow development is signicantly slower and the particle velocity of the smaller riser is higher. The ow development slows down in the larger riser. In all locations measured, there was a clear dependency between the local particle velocity and solids concentration of both risers. Gas distribution and particle aggregation are considered the key factors that inuence the local hydrodynamics in the twin-riser system. 2005 American Institute of Chemical Engineers AIChE J, 51:
2956 2964, 2005

Keywords: circulating uidized bed, particle velocity prole, ow development, riser diameter, ve-ber optic probe

Introduction
Gassolid reactions take place in a wide range of chemical processes, such as circulating uidized bed (CFB) combustion and uid-catalytic cracking (FCC). Because of their use in petroleum reneries for FCC, circulating uidized beds were reinvented1,2 in the late 1970s. Since then, intensive studies have been carried out to improve these industrial processes. FCC units are used in most reneries all over the world to convert high molecular weight gas oils or residuum charge stocks into lighter hydrocarbon products in a riser reactor within a few seconds. However, a shorter and a more uniform catalyst residence time in the riser reactor would potentially lead to a better reaction performance (larger amounts of desired products and/or a higher conversion). That is, because coke deposition rapidly deactivates the catalyst leading to reduced selectivity and some desirable products (gasoline, light olens, light cycle oil) may further react to cause over-cracking. These
Correspondence concerning this article should be addressed to J. Zhu at jzhu@uwo.ca.

2005 American Institute of Chemical Engineers

non-benecial factors could be limited or avoided by having more uniform axial and radial particle ow structure in the riser, leading to shorter and more uniform solids as well as gas residence times. To apply fundamental knowledge from pilot scale risers into a commercial riser, it is necessary to achieve a clear understanding of the scale-up effects. Extensive research has been carried out on CFB riser reactors.3-17 Although there were some researches on particle velocity in the literature,3-5,7,9,12,17,18 most data were taken in risers of small diameter (0.15 m) and / or short length (7 m). Considering most industrial risers are of 1 m or larger in diameter and of 20 m or taller in height, it is not reliable to base the industrial design on the experimental results obtained in small diameter and very short risers since small diameter vessels are dominated by wall effects and the entrance and exit structures have great inuence on the short height vessels. It is very important to understand the scale-up effect on the hydrodynamics inside the risers of signicant sizes. Unfortunately, no previous studies have been published with respect to gas and particle ow over a wide range of operating conditions in risers of different diameters. Therefore, more research on the inuence of riser diameter on particle
Vol. 51, No. 11 AIChE Journal

2956

November 2005

Figure 1. Twin-riser circulating uidized bed apparatus.

velocity distribution and ow development is required in largescale CFB risers to improve and optimize the existing industrial CFBs and enhance the design for novel applications. To minimize the possibility of expensive errors in scaling-up to commercial operation, and to optimize and improve the designs of existing industrial FCC riser reactors, a good understanding of the scale-up obtained from pilot-scale risers, such as the knowledge of the particle ow distribution in CFB risers of signicant size, is necessary. This work used a twinriser system, with two 10 m long riser units having internal diameters (ID) of 76 mm (3 in.) and 203 mm (8 in.) and otherwise having identical geometries (sharing the same downcomer and air supply, using the same particles), over a wide range of operating conditions with the solids circulation rate up to 200 kg m2 s1 and supercial gas velocity up to 8.0 m/s. The inuence of riser diameter on the axial and radial particle velocity proles and ow development is carefully studied by measuring the local particle velocity with an optical-ber probe along the twin-riser system. The information obtained from this study could be benecial, especially for the FCC industry.

Experimental Apparatus
The experiments were conducted in a twin-riser circulating uidized bed system shown schematically in Figure 1, consisting of two 10 m long risers, 76- and 203-mm IDs, which use the same downcomer (storage tank) of ID 0.32 m. FCC catalyst with a mean diameter of 67 m and a particle density of 1500 kg/m3 were used. The total solids inventory was about 350 kg, equivalent to a solids level of nearly 5 m in the downcomer. The humidity level of the air was controlled between 70 and 80% to minimize the effect of electrostatics in the system. The congurations (except for the diameter) are identical for both risers. The relative dimensions of the inlet and outlet openings to the diameters are equal to the corresponding diAIChE Journal November 2005

ameter. After passing through a short inclined pipe section, where a ip valve was installed to control the solids ow rate, the solids from the storage tank entered the riser bottom at a height of 0.21 m in each riser and were accelerated by air in near-ambient conditions (at the base of the riser, the temperature is about 20C and absolute pressure is about 105.7 kPa). After initial mixing and accelerating, the gassolids suspension traveled up in the column and passed through a smooth exit into the primary cyclone for gassolids separation, and escaped solids entered into the secondary and tertiary cyclones, with the nal gassolids separation carried out by a bag lter. Two separate sets of cyclones were used for each riser, although the two risers shared the same bag lter. Only one riser operated at a given time. From the bottom of the large-capacity bag lter, collected ne particles could be returned to the downcomer. The solids ow rate measuring device located in the top portion of the downcomer sectioned the column into two halves with a central vertical plate and with two half-buttery valves xed at the top and the bottom of the two-half section. By appropriately ipping over the two valves from one side to the other, solids circulated through the system can be accumulated in one side of the measuring section for a given time period to provide the solids circulation rate. A ve-ber optic velocity probe was inserted into the column to measure the particle velocity. Each of the ve bers is a silicon optical ber of diameter 200 m. The probe consists of a horizontal cylindrical portion of diameter 2 mm and length about 0.3 m, leading to a 10 mm long tip having an oval-shaped cross section of 0.5 1.8 mm (width height). The ve-ber optic probe consists of two light-emitting bers (B and D) and three light-detecting bers (A, C, and E) arranged precisely in the same vertical line. A particle owing by the center point between any two neighboring bers will produce a reective signal to a detection ber. By counting the time difference between the two signals from AB and BC (or CD and DE), the velocity of a particle passing along the array of the ve bers can be determined. More details of the ve-ber optic probe have been presented by Zhu et al.19 The particle velocity was measured at 11 radial positions (r/R 0.00, 0.16, 0.38, 0.50, 0.59, 0.67, 0.74, 0.81, 0.87, 0.92, and 0.98) on eight axial levels (Z 1.53, 2.73, 3.96, 5.13, 5.90, 6.34, 8.74, and 9.42 m) for the 76-mm ID riser and ve axial levels (Z 1.47, 2.69, 3.91, 5.90, and 8.79 m) for the 203-mm ID riser under several operating conditions as given in Table 1. At each measurement location the sampling time was typically 30 s and the amount of sampled particles 2500. The local solids concentration was measured with a reective-type ber-optic concentration probe. The 3.8 mm diameter probe tip consisted of nearly 8000 emitting and receiving quartz bers, each having a diameter of 15 m. The active area, where the bers were located, was about 2 2 mm. More
Table 1. Operating Conditions
76-mm ID Riser U g (m/s) 5.5 3.5 5.5 8.0 5.5 G s (kg m
2

203-mm ID Riser s )
1

U g (m/s) 5.5 5.5 5.5 8.0

G s (kg m2 s1) 50 75 100 100

50 100 100 100 200

Vol. 51, No. 11

2957

details of this probe can be found in Zhang et al.,20 and more details of the concentration probe measurements from the same twin-riser system can be found in Yan and Zhu.21 The solids concentration was measured in the same positions and at the same operating conditions as the particle velocity. Two samples were taken at each location, and the total sampling time was 60 s. By combining the results from the current study with those of Yan and Zhu,21 the cross-sectional net solids uxes were obtained for each axial elevation. Those were found to be in a good agreement with the results from the solids ow rate measuring device (Gs) in the downcomer (within 10 15%). The cross-sectionally mean solids holdups were obtained by averaging the local solids holdups measured at 10 radial positions (excluding the center point) with the ber-optic concentration probe at each axial elevation.

Results and Discussion Development of radial proles of particle velocity

In our study, we used large-scale CFB risers (with diameters up to 0.2 m and bed height of 10 m to ensure that the particle ow is fully developed in the riser) with high gas velocity (up to 8 m/s) to study the inuence of bed diameter on radial proles of the particle velocities. Figure 2a shows the radial proles of particle velocity on eight axial elevations of the 76-mm ID riser under ve operating conditions. The solids circulation rates were 50, 100, and 200 kg m2 s1 and the supercial gas velocities were 3.5, 5.5, and 8.0 m/s. Radial proles of particle velocity on ve axial elevations of the 203-mm ID riser under four operating conditions are shown in Figure 2b. The solids circulation rates were 50, 75, and 100 kg m2 s1 and the supercial gas velocities were 5.5 and 8.0 m/s. In general, the particle velocities are more uniform in the upper section than in the lower section of the riser at all radial positions and higher in the center than in the wall region of the riser at all axial locations. Figure 2 also shows that the particle velocity in the center region of the riser remains nearly constant throughout the riser under each operating condition for both risers, and then decreases toward the wall. Figure 2 further shows that the particle velocity in the center region of the riser does not seem to change signicantly under the same supercial velocity within the tested range of operating conditions for both risers. As a result, the ow development is mostly represented by the increase of the particle velocity toward the riser top at r/R from about 0.50 to 1.00. In the riser bottom, the particle velocity drops signicantly toward the wall. Toward the riser top, the particle velocity in the wall region tends to increase. Increasing the supercial gas velocity Ug increases the particle velocity throughout the riser. The solids circulation rate (overall solids ux) Gs seems to have only a slight inuence on the value of particle velocity. In addition, the radial prole of particle velocity seems more uniform with lower solids ux. That is, increasing solids ux slows down the ow development. Similar phenomena were also observed in the radial particle velocity proles of the same 76-mm ID riser under higher ux operating conditions as shown in Pa rssinen and Zhu.17 Because of the limitation of the capacity of the storage tank, the solids ux in the 203-mm ID riser cannot reach a very high ux. It will be our future work to study higher ux conditions in the 203-mm ID riser after modifying the equipment to increase the capacity of the storage tank.
2958 November 2005

Figure 2. (a) Radial proles of local particle velocity under ve operating conditions on eight axial levels of the 76-mm ID riser, showing the ow development along the riser; (b) radial proles of local particle velocity under four operating conditions on ve axial levels of the 203-mm ID riser, showing the ow development along the riser.

Vol. 51, No. 11

AIChE Journal

Figure 4. Comparison of the development of radial proles of local particle velocities for the risers. Figure 3. Comparison of the typical radial proles of local particle velocity in the same ve axial levels for the risers.

Figure 2 also shows the ow development with respect to the radial proles of particle velocity under each operating condition in the 10 m high twin-riser system. At the bottom section (Z 1.53 and/or 2.73 m), the radial prole shows a at center region, turning then smoothly downward toward the wall, and having a fairly wide wall region with a velocity value of 2 m/s. This may be referred to as a horizontal S-shape. In the upper levels, the solids accelerate more in the radial region of r/R 0.0 to 0.4, leading to a combination of linear (but not at) and parabolic-shaped radial velocity prole. When the solids reach the exit section, the particles slow down somewhat, and the corresponding velocity proles become scattered as a result of the exit features as the solids pass through the 90 smooth elbow. Observed from Figure 2, it seems that the ow development in the 203-mm ID riser is considerably slower than that of the 76-mm ID riser. The shape change of the radial proles of the 203-mm ID riser also happens at a much higher level than that of the 76-mm ID riser. This is because of the scale-up effect, which will be discussed later in more detail. Figure 3 provides the typical radial proles of particle velocity for both risers at Z 5.90 m. Figure 3 also shows that increasing Ug tends to make the radial proles more uniform. There is little inuence of solids ux on the radial distributions of particle velocity, which is likely a consequence of the quicker ow development under lower ux, in agreement with the observations from Figure 2. Similar tendencies are observed in the other axial levels. Zhou et al.18 also observed that, under lower ux conditions, increasing solids circulation rate leads to steeper radial proles for particle velocity. By comparing the velocity proles at 1.53, 2.73, and 3.96 m (marked as closed symbols with connecting lines), Figure 4 shows that increasing Ug from 5.5 to 8.0 m/s (with Gs 100
AIChE Journal November 2005

kg m2 s1) causes a faster ow development. On the other hand, increasing Gs seems to slow down the ow development slightly under a constant Ug of 5.5 m/s. The ow development at higher ux is slower, as observed under higher ux operating conditions in the same riser as shown in Pa rssinen and Zhu.17 It is also observed in Figure 4 that the shape change of the radial proles of the 203-mm ID riser is at a much higher level than that of the 76-mm ID riser. When observing the radial proles of the riser from the bottom to the top, it is also seen that the ow develops rst in the riser center region, and then gradually and progressively closer to the wall as the solids pass through the riser. This can be seen more clearly in Figure 5, which plots the particle velocities in three radial regions on eight axial elevations in the 76-mm ID riser and ve axial elevations in the 203-mm ID riser (the particle velocity in each region is obtained by aver-

Figure 5. Comparison of the average particle velocities in different radial regions along the riser under all three operating conditions for the risers. Vol. 51, No. 11 2959

aging the local particle velocities in that region). In this gure, the difference in the ow development in the three radial regions is clearly revealed. Furthermore, the scale-up effect on the ow development is also clearly observed. In the center region of the riser (0.0 r/R 0.632, 40% of the crosssectional area), the particles already gained a fairly high velocity before the height of 1.5 m, whereas a maximum velocity is nally reached at the height of 3 4 m. There is no signicant difference for the two risers in this region. In the middle region (0.632 r/R 0.894, 40% of the cross-sectional area), on the other hand, the particle velocity is increasing throughout the riser, and the increase is more obvious with higher gas velocity. Meanwhile, the particle velocity of the 76-mm ID riser increases more sharply than that of the 203-mm ID riser in this middle region. That is, increasing riser diameter slows down the ow development. In the wall region (0.894 r/R 1.0, 20% of the cross-sectional area), the particle velocity remains low up to about 4 6 m and then increases slowly toward the riser top. The ow development in this region is signicantly slower than that in the two other regions. Moreover, it is also clearly shown that the particle velocity of the 203-mm ID riser increases much less than that of the 76-mm ID riser in this region. From the riser bottom to the top, the radial distribution of the 76-mm ID riser becomes uniform much earlier than that of the 203-mm ID riser. Thus, the ow development slows down with large riser. Figure 5 also indicates that increasing gas velocity signicantly enhances the ow development (that is, to cause particles to accelerate quicker and at lower levels in the riser). The same tendency was also observed under higher ux operating conditions in the same 76-mm ID riser as shown in Pa rssinen and Zhu17; however, the ow development of higher ux conditions is much slower, especially in the middle region. In Figure 5, the particle velocities from the riser bottom to the top in the center region are almost the same and change only slightly along the riser. However, the differences of particle velocities from the riser bottom to the top at the middle and wall regions are much more obvious. Figure 5 clearly shows that particle acceleration (and therefore ow development) rst starts from the center and then extends to the wall. This dictates the development of the radial proles of particle velocity. In the bottom section, only those particles in the center region had some acceleration so that the horizontal S-shape appears. In the middle section, particles in the middle radial region start to accelerate so that the particle velocity in this region increases, resulting in the combination of linear (but not at) and parabolic-shape prole. In the exit section, particles in the middle region accelerate further to catch up with those particles in the center region, and particles in the wall region also begin to accelerate, leading to a parabolic radial prole. Because the ow development slows down when the riser diameter is scaled-up, the shape change of the radial proles of the 203-mm ID riser happens at a much higher level than that of the 76-mm ID riser, as also observed in Figures 2 and 4. Figure 6, which compares the radial proles of particle velocity between the 76- and the 203-mm ID risers, indicates that there is a slight difference in the center region of radial proles of the two risers. However, it is quite different in the wall region. The radial distributions of particle velocity of the 76-mm ID riser are comparably more uniform and less sensi2960 November 2005

Figure 6. Comparison of the radial proles of local particle velocity for the risers of different diameters.

tive to the change of the axial position than that of the 203-mm ID riser. The difference is mainly a result of riser diameter difference. This may be explained by some secondary effects, such as the introduction of the solids to the side at the base of the riser, poor dispersion of solids in the riser, or the shorter L/D in the larger-diameter vessel. Further research is needed for these secondary effects. Figure 7 compares the radial solids holdup proles of the twin-riser system at Z 5.84 m. It is shown that the solids holdup of the 203-mm ID riser at each radial position is higher than that of the 76-mm ID riser under the same operating conditions and at the same axial level, especially in the wall region. In an earlier paper21 by the same authors reporting on the axial and radial solids concentration distribution, similar results can also be observed in the other axial levels. That is, under the same solids circulation rate and gas velocity, a larger riser has a larger solids concentration at each radial position, especially in the wall region. This shows a much more substantial wall effect, even in the relative magnitude, for the larger-diameter riser. A denser concentration of solids occupies the wall region of the large-diameter riser and restricts the gas ow at the wall. As a result, the particle velocity tends to be lower in the wall region for the largerdiameter riser. To maintain the cross-sectional Ug, the gas velocity has to be correspondingly higher in the center region of the large riser. A higher gas velocity in the center could also lead to higher particle velocities. Again, this veried the scale-up effect of risers on the ow development: increasing riser diameter impedes ow development. Figure 8a shows the variation of particle velocity in the wall region with the cross-sectional mean solids holdup, obtained by
Vol. 51, No. 11 AIChE Journal

are relatively higher, so that the particle velocities near the wall are lower than those of the 76-mm ID riser at the same axial level. Therefore, as shown in Figure 8, no data appeared in the range of very low cross-sectional solids holdup and high particle velocity for the wall region of the 203-mm ID riser. The particle velocity in the wall region and the cross-sectional solids holdup of the 203-mm ID riser are more clustered together than that of the 76-mm ID riser.

Comparison of the axial development of particle velocities in the twin-riser system

Figure 9 compares the axial proles of the one-dimensional (cross-sectional average) particle velocities on each axial level obtained with the ber-optic velocity (Figure 9a) and with the concentration (Figure 9b) probes under the same operating conditions for both risers. In Figure 9a, the local particle velocity was weighted with the local solids holdup in each

Figure 7. Comparison of radial proles of local solids holdup the twin-riser system at Z 5.84 m.

averaging the local solids holdups measured at 10 radial positions (excluding the center point) with the ber-optic concentration probe at each axial elevation, at each axial level of the 76-mm ID riser. For this 76-mm ID riser, with a high crosssectional mean solids concentration (riser bottom), the particle velocity at the wall region is relatively low. With a lower cross-sectional solids concentration (riser top), the particle velocities are higher at the wall region. A simple regression provides the following correlation between the particle velocity at the wall region and the cross-sectional average solids holdup of the 76-mm ID riser V p,wall 0.98 ln s 2.23 0 s 0.42, Vp 1.38 m/s (1)
Figure 8. (a) Relationship between the local particle velocity near the riser wall (r/R 0.92 and 0.98) and the cross-sectional mean solids holdup on all axial elevations under all measured operating conditions for the 76-mm ID riser; (b) relationship between the local particle velocity near the riser wall (r/R 0.92 and 0.98) and the cross-sectional mean solids holdup on all axial elevations under all measured operating conditions for the 203-mm ID riser. Vol. 51, No. 11 2961

This correlation is very close to that obtained from the higher ux conditions of the same 76-mm ID riser.17 The particle velocity in the wall region of the 203-mm ID riser could also be related to the cross-sectional solids concentration at each axial level, as shown in Figure 8b. However, the relationship between the particle velocity in the wall region and the crosssectional solids concentration of the 203-mm ID riser is different from that of the 76-mm ID riser. As mentioned earlier, because of the slow ow development of the large riser, the cross-sectional solids concentrations of the 203-mm ID riser
AIChE Journal November 2005

Figure 9. (a) Comparison of axial proles of one-dimensional particle velocity in the twin-riser system: averaged from local particle velocities (weighed by local solids concentration); (b) comparison of axial proles of onedimensional particle velocity in the twin-riser system: deduced from the cross-sectional average solids concentration. (c) Comparison of axial proles of one-dimensional particle velocity in the twin-riser system: averaged from local particle velocities measured at 10 radial positions (excluding the center point).

radial position to give the true cross-sectional average velocity, given that a higher solids population is typically owing near the wall than in the center. In Figure 9b, the cross-sectionally averaged solids concentration21 was used to calculate the average particle velocity [Gs/(p s )]. The excellent agreement (with average deviation of 10%) between Figure 9a and Figure 9b further conrms the reliability of the two opticalber probes used in this study. In Figures 9a and 9b, the cross-sectional average particle velocity of the 203-mm ID riser is lower than that of the 76-mm ID riser under the same operating conditions, although there is no signicant change for the shape of the axial proles. That is, in terms of scale, the cross-sectional average particle velocity decreases with the riser diameter, but the shape of the axial prole changes only negligibly with the riser diameter. These correspond well to the typical axial solids holdup proles (S-shape) of our previous study.21 Seen from Figures 9a and 9b, increasing gas velocity obviously increases the axial particle velocities for both risers. However, changing solids uxes appears to have no obvious inuence on the axial particle velocities for both risers. As mentioned earlier, at the same solids circulation rate and gas velocity, a larger riser has a larger solids concentration at each radial position, especially in the wall region. Thus, a larger riser has a larger cross-sectional average solids concentration at each radial position. To maintain the cross-sectional Gg, the cross-sectional average particle velocity has to be correspondingly lower in the large riser. The large difference between the average particle velocities in the two risers may not be completely attributed to the scale, but may also be explained by the steepened radial solids distribution in the larger riser. By closely analyzing the particle velocity in each cross-section (Figure 6), it is easier to see that there is not much difference in the center region between the two risers; the
2962 November 2005

203-mm ID riser has a much lower particle velocity closer to the wall. Because more particles accumulated at the wall for the large riser (as shown in Figure 7), the weighted cross-sectional average particle velocity becomes much lower in the 203-mm ID riser. To further illustrate the point, the straight mathematic average particle velocity, without taking the weight of the radial solids distribution, were plotted in Figure 9c for the two risers. Compared with Figure 9a, the difference between the mathematic mean particle velocities of the two risers shown in Figure 9c is much smaller, suggesting that the lower local particle velocity near the wall in the larger riser is only part of the reason for the larger riser to have a lower cross-sectional average particle velocity at each axial level at the same solids circulation rate and gas velocity. In other words, the scale-up effect may not be as large as shown in Figure 9a. In the study on the axial distributions of solids holdup, Yan and Zhu21 reported three axial sections along the twin-riser system: the bottom dense section (Z 3 m), a middle section, and a top exit section. Such a three-section structure can also be observed in the axial proles of the cross-sectional average particle velocity shown in Figures 9a and 9b: the bottom section (3 4 m), with particle velocity 2 m/s and no obvious solids acceleration; a middle section; and a top exit section.

Particle velocities vs. solids holdups

The local particle velocity at a given radial position of a certain axial level can also be directly related to the local solids concentration at the same position throughout the twinriser system. As shown in Figure 10, the local particle velocity decreases monotonically with the local solids concentration and such a decrease follows the same pattern in all axial sections for both risers. Such a strong dependency could be understood only by considering solids aggregation at each local position: a higher solids concentration leads to stronger particle
Vol. 51, No. 11 AIChE Journal

riser is higher than that of the 203-mm ID riser. In the wall region, the particle velocity of the 76-mm ID riser is also higher than that of the 203-mm ID riser. That is, the ow development is faster in a smaller riser diameter in these two regions. Thus, the overall ow development is slower with the large riser. These scale-up effects are the same as shown before. A clear dependency between the particle velocity and solids concentration is also shown in these gures.

Conclusions
Numerous measurements were performed in a twin-riser system to show the dependency of the particle velocity distribution and the ow development on riser diameter, using a ber-optic probe on eight axial levels in a 10 m long riser of 76-mm and 203-mm IDs. The particle velocity is more uniform in the upper section than in the lower section of the riser at all

Figure 10. Relationship between local particle velocity and local solids holdup for both risers.

aggregates, resulting in a lower effective drag between the particles and gas, and therefore a lower particle velocity under a constant gas velocity. The independence of the above variation with the axial locations shows the inherent connection between the particle concentration and velocity. As a result, a single correlation works for all axial sections in each riser: For the 76-mm ID Riser V p 1.85 lns 3.12 0 s 0.42, Vp 1.52 m/s (2) For the 203-mm ID Riser V p 2.23 lns 4.73 0 s 0.42, Vp 2.80 m/s (3) The difference between the two correlations may result from the scale-up effect as discussed above. The average local particle velocity in the three radial sections (0.0 r/R 0.632, 0.632 r/R 0.894, and 0.894 r/R 1.0) can also be related to the cross-sectional mean solids holdup at a given axial position for both risers. Figures 11a and 11b show the relationship of the average local particle velocity and the cross-sectional mean solids holdup for the 76- and the 203-mm ID risers. The trends of the three regions in the 203-mm ID riser were also represented as darker lines in Figure 11a to compare with those in the 76-mm ID riser. The difference in the ow development in the three radial regions is also clearly shown in Figure 11. Similar to the case stated before in Figure 5, the particle velocities were rst developed in the center region of the riser (0.0 r/R 0.632) and the particle velocities are the highest. In the middle region (0.632 r/R 0.894), on the other hand, the particle velocity is intermediate. In the wall region (0.894 r/R 1.0) the particle velocity remains low all over the riser. This again veried that the ow develops rst in the center region of the riser, and then gradually and progressively closer to the wall as the solids pass through the riser. Furthermore, the scale-up effect on the ow development is also clearly observed. There is no signicant difference for the two risers in the center region. In the middle region, on the other hand, the particle velocity of the 76-mm ID
AIChE Journal November 2005

Figure 11. (a) Relationship between average particle velocity in the three radial regions and crosssectional mean solids holdup for the 76-mm ID riser (dark line: trends for the 203-mm ID riser); (b) relationship between average particle velocity in the three radial regions and cross-sectional mean solids holdup for the 203-mm ID riser. Vol. 51, No. 11 2963

radial positions and is higher in the center than in the wall region of the riser at all axial locations. In the center region of the riser (0.0 r/R 0.632), the particle velocity remain high and relatively constant throughout the riser, suggesting very quick solids ow development in the riser center, and then gradually closer to the wall as the solids pass through the riser. In the middle radial region (0.632 r/R 0.894), particle velocity is increasing and continues throughout the column. In the wall region (0.894 r/R 1.0), however, the ow development is signicantly slower. Furthermore, there is no significant difference for the two risers in the center region. In the middle and wall regions, on the other hand, the particle velocity of the 76-mm ID riser is higher than that of the 203-mm ID riser. The radial particle proles change from an S-shape, in the bottom of the riser, to a combination of linear (but not at) and parabolic shape in the middle, and eventually, parabolic with some scattering at the exit. The shape change of the radial particle proles of the 203-mm ID riser happens at a much higher level than that of the 76-mm ID riser. The results also show at the same solids circulation rate and gas velocity, a larger-diameter riser has a steeper radial particle velocity prole at each axial level. Therefore, all the results show that the ow development slows down with larger-diameter riser. Increasing the supercial gas velocity Ug increases the particle velocity throughout the riser length. By increasing Ug, the ow development was shown to become faster. Increasing solids ux Gs, on the other hand, slows down the ow development. The cross-sectional average particle velocity is somewhat lower with the larger-diameter riser resulting from the scale-up effect. In all locations measured, there was a clear dependency between the local particle velocity and concentration for each riser. In addition, it is also because of the scale-up effect that the relationship between the particle velocity in the wall region and the crosssectional solids concentration, and the correlation of the local particle velocity and solids concentration of the 203-mm ID riser are different from those of the 76-mm ID riser. More research must be done to study the riser diameter effect (D), the particle to diameter effect, a change in the dispersion and ow patterns arising from an effect at the riser inlet, or the effect of the length/diameter ratio on the development of the acceleration zone. Gas radial distribution and particle aggregation are considered the key factors that affect the local hydrodynamics in the twin-riser system.

Z height from the riser bottom, m s cross-sectional mean solids holdup s local solids holdup

Literature Cited
1. Reh L. Fluidized bed processing. Chem Eng Prog. 1971;1:58-63. 2. Yerushalmi J, Turner DH, Squires AM. The fast uidized bed. I&EC Proc Des Dev. 1976;15:47-53. 3. Bader R, Findlay J, Knowlton TM. Gas/solid ow patterns in a 30.5 cm-diameter circulating uidized bed. In: Basu P, Large JF, eds. Circulating Fluidized Bed Technology II. Oxford, UK: Pergamon Press; 1988:123-137. 4. Glicksman LR. Circulating uidized bed heat transfer. In: Basu P, Large JF, eds. Circulating Fluidized Bed Technology II. Oxford, UK: Pergamon Press; 1988:13-29. 5. Hartge EU, Rensner D, Werther J. Solids concentration and velocity patterns in circulating uidized beds. In: Basu P, Large JF, eds. Circulating Fluidized Bed Technology II. Oxford, UK: Pergamon Press; 1988:165-180. 6. Kunii D, Levenspiel O. Entrainment of solids from uidized beds. I. Hold-up of solids in the freeboard. II. Operation of fast uidized beds. Powder Technol. 1990;61:193-206. 7. Nowak W, Mineo H, Yamazaki R, Yoshida K. Behavior of particles in a circulating uidized bed of a mixture of two different sized particles. In: Basu P, Horio M, Hasatani M, eds. Circulating Fluidized Bed Technology III. Oxford, UK: Pergamon Press; 1991:219-224. 8. Bi H, Zhu JX. Static instability analysis of circulating uidized beds and concept of high-density risers. AIChE J. 1993;39:1272-1280. 9. Knowlton T. Interaction of pressure and diameter on CFB pressure drop and holdup. Proc of workshop on Modeling and Control of Fluidized Bed Systems, Hamburg, Germany, May; 1995:22-23. 10. Bai D, Issangya AS, Zhu JX, Grace JR. Analysis of the overall pressure balance around a high-density circulating uidized bed. Ind Eng Chem Res. 1997;36:3898-3903. 11. Issangya AS, Bai D, Bi HT, Lim KS, Zhu J, Grace JR. Axial solids holdup proles in a high-density circulating uidized bed riser. In: Kwauk M, Li J, eds. Circulating Fluidized Bed Technology V. Beijing, China: Science Press; 1996:60-65. 12. Issangya AS, Bai D, Grace JR, Lim KS, Zhu J. Flow behaviour in the riser of a high-density circulating uidized bed. AIChE Symp Ser. 1997;93:25-30. 13. Issangya AS, Bai D, Bi HT, Lim KS, Zhu J, Grace JR. Suspension densities in a high-density circulating uidized bed riser. Chem Eng Sci. 1999;54:5451-5460. 14. Grace JR, Issangya AS, Bai D, Bi HT, Zhu JX. Situating the highdensity circulating uidized bed. AIChE J. 1999;45:2108-2116. 15. Karri SBR, Knowlton TM. A comparison of annulus solids ow direction and radial solids mass ux proles at low and high mass uxes in a riser. In: Werther J, ed. Circulating Fluidized Bed Technology VI. Frankfurt, Germany: Dechema; 1999:71-76. 16. Pa rssinen JH, Zhu JX. Axial and radial solids distribution in a long and high-ux CFB riser. AIChE J. 2001;47:2197-2205. 17. Pa rssinen JH, Zhu JX. Particle velocity and ow development in a long and high-ux CFB riser. Chem Eng Sci. 2001;56:5295-5303. 18. Zhou J, Grace JR, Brereton C, Lim CJ. Particle velocity proles in a circulating uidized bed of square cross-section. Chem Eng Sci. 1995; 50:237-244. 19. Zhu JX, Li GZ, Qin SZ, Li FY, Zhang H, Yang YL. Direct measurements of particle velocities in gassolids suspension ow using a novel 5-ber optical probe. Powder Technol. 2001;115:184-192. 20. Zhang H, Johnston PM, Zhu JX, de Lasa HI, Bergougnou MA. A novel calibration procedure for a bre optic concentration probe. Powder Technol. 1998;100:260-272. 21. Yan AJ, Zhu JX. Scale-up effect of riser reactors. (1) Axial and radial solids concentration distribution and ow development. Ind Eng Chem Res. 2004;43:5810-5819.
Manuscript received May 27, 2004, and revision received Mar. 14, 2005.

Acknowledgments
The authors gratefully acknowledge the nancial support from the Natural Sciences and Engineering Research Council of Canada and help from J. H. Pa rssinen, J. Wen, H. Zhang, and X. Zhu during the experiments.

Notation
D Gs r R Ug Vp Vp,ave Vp,wall riser diameter, m or inch solids circulation rate (solids ux), kg m2 s1 radial distance from riser axis, m radius of riser, m supercial gas velocity, m/s particle velocity, m/s average particle velocity in three radial regions, m/s particle velocity in the wall region, m/s

2964

November 2005

Vol. 51, No. 11

AIChE Journal

SEPARATIONS

Experimental and Modeling Study of Protein Adsorption in Expanded Bed

Ping Li, Guohua Xiu, and Alirio E. Rodrigues
Laboratory of Separation and Reaction Engineering, Dept. of Chemical Engineering, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, s/n 4200-465 Porto, Portugal DOI 10.1002/aic.10536 Published online July 19, 2005 in Wiley InterScience (www.interscience.wiley.com).

Streamline DEAE is the rst-generation adsorbent developed for expanded bed adsorption (low-density base matrix with wide particle size distribution and ligand sensitive to ionic strength and salt concentration), and Streamline direct CST I is the secondgeneration adsorbent (high-density base matrix with narrow particle size distribution and ligand not sensitive to ionic strength and salt concentration). In this paper, experiments were carried out for bovine serum albumin (BSA) protein adsorption in expanded beds, where a Streamline 50 column was packed either with Streamline DEAE or with Streamline direct CST I. The hydrodynamics, BSA dynamic binding capacity, and BSA recovery in the whole expanded bed adsorption process were compared for both adsorbents. A mathematical model, in which intraparticle diffusion, lm mass transfer, liquid axial dispersion, solid axial dispersion, particle size axial distribution, and bed voidage axial variation were taken into account, was developed to predict the breakthrough curves in expanded bed adsorption. BSA breakthrough curves in expanded bed adsorption were measured for both Streamline DEAE and Streamline CST I, and compared with the predictions from this mathematical model. The effects of intraparticle diffusion, lm mass transfer, liquid and solid axial dispersion, particle size axial distribution, and bed voidage axial variation on the breakthrough curves were evaluated for expanded bed adsorption with both adsorbents. 2005 American Institute of Chemical Engineers AIChE J, 51:
29652977, 2005

Keywords: expanded bed adsorption, protein adsorption kinetics, modeling, breakthrough curves, residence time distributions

Introduction
Expanded bed adsorption is a single operation in which desired proteins are puried from particulate-containing feedstocks without the need for separate clarication, concentration, and initial purication. This technology has been widely applied to capture proteins directly from crude feedstocks, such
Correspondence concerning this article should be addressed to A. E. Rodrigues at arodrig@fe.up.pt.

2005 American Institute of Chemical Engineers

as E. coli homogenate, yeast, fermentation, mammalian cell culture, milk, animal tissue extracts, and other unclaried feedstocks, and various applications have been reported from labscale to pilot-plant and large-scale production.19 With specially designed adsorbents and columns, the adsorption behavior in expanded beds is comparable to that in xed beds.1 Streamline DEAE and Streamline SP are typical rstgeneration adsorbents, developed for expanded bed adsorption. The modied Sepharose matrices allow capture of biomolecules directly from unclaried feedstocks; the adsorbents have high binding capacities and product yields, attributed to stable
Vol. 51, No. 11 2965

AIChE Journal

November 2005

expanded beds, and long life as the result of high chemical and mechanical stability. Many theoretical and experimental researches were reported in the literature10-17 with respect to the hydrodynamics and protein adsorption kinetics in expanded beds packed with the rst-generation adsorbents. Streamline direct CST I, a second-generation adsorbent recently developed for use in expanded bed adsorption, has two special features compared with the rst-generation adsorbents: a high-density base matrix and a salt-tolerant ligand. The high-density matrix means minimizing dilution arising from biomass or viscosity and reducing dilution buffer consumption; the ligands lack of sensitivity to ionic strength means there is no need for dilution of feedstock because of high ionic strength. However, hydrodynamics and protein adsorption kinetics in expanded beds packed with the second-generation adsorbent have not been investigated in detail. In this article, experiments were carried out for protein [bovine serum albumin (BSA)] adsorption in expanded beds, where a Streamline 50 column was packed with 300 mL Streamline DEAE and with 300 mL Streamline direct CST I, respectively. The experimental results are compared for both adsorbents to give a comprehensive evaluation of the hydrodynamics, BSA dynamic binding capacity, and BSA recovery in the whole expanded bed adsorption process. The hydrodynamics and adsorption kinetics in expanded beds are more complex than that in xed beds. The liquid axial dispersion in expanded beds is more signicant than that in xed beds; because of the uidized nature of the expanded bed, adsorbent particle axial dispersion occurs. Moreover, there are variations of particle size axial distribution and bed voidage axial in expanded beds for the specially designed adsorbents with wide particle size distribution.11,15,16,18 Models available for xed beds may be not adequate to describe the hydrodynamic and adsorption behavior in expanded beds. Wright et al.14 developed a mathematical model to predict the breakthrough curve for protein adsorption in a uidized bed, where intraparticle diffusion resistance, lm mass transfer resistance, liquid axial dispersion, and adsorbent particle axial dispersion were taken into account. Later, Tong et al.19 and Chen et al.20 used this model to predict the breakthrough curves in the expanded bed adsorption. When capturing proteins in an expanded bed with a high ow velocity, the slow diffusion rate of proteins results in high intraparticle diffusion resistance, signicantly affecting the breakthrough curve. It is argued that, in this case, the particle size, characterizing the diffusion path in the adsorbent particles, should have a substantial effect on the breakthrough curves.13 Therefore, simulation results should be improved when the particle size axial distribution and bed voidage axial variations are taken into account in the model. Tong et al.17 modied the mathematical model by taking into account the particle size axial distribution in expanded beds. Following their experimental research using in-bed monitoring in expanded beds, Bruce et al.12 predicted the in-bed breakthrough curves in expanded beds by using zonally measured parameters. Li et al.21 developed a three-zone model to predict in-bed breakthrough curves and conrmed the effect of the particle size axial distribution and bed voidage axial variations on the breakthrough curves in expanded beds. Recently, Kaczmaarski et al.22 also analyzed the effects of the axial and local particle size distribution and bed voidage axial variation on the breakthrough curves in expanded beds.
2966 November 2005

Up to now, theoretical and experimental researches have focused on protein adsorption onto Streamline DEAE or Streamline SP in expanded beds. Streamline DEAE and Streamline SP are rst-generation adsorbents; the adsorbent matrix has low density and large particle diameter with wide particle size distribution (100 300 m). The matrix of the second-generation adsorbent, Streamline direct CST I, has high density and small particle diameter with narrow particle size distribution (80 165 m). In this study, BSA breakthrough curves in expanded bed adsorption are measured for both Streamline direct CST I and Streamline DEAE, and a mathematical model is developed to predict the breakthrough curves and to compare with the experimental results. The effects of intraparticle diffusion resistance, lm mass transfer resistance, liquid axial dispersion, solid axial dispersions, adsorbent particle size axial distribution, and bed voidage axial variation on the breakthrough curves will be evaluated for expanded beds packed with Streamline DEAE and with Streamline direct CST I, respectively. Although the adsorbents and the columns are designed specically for expanded beds to maintain stable bed expansion, the liquid axial dispersion is more signicant than that in xed beds. Sometimes, some inadequate operationsuch as the column not being in a vertical position, trapped air in the bottom distribution system, clogging of the bottom distribution system and pump pulsewould make the liquid axial dispersion more signicant. Therefore, it is necessary to exactly measure and predict the liquid axial dispersion in expanded beds to allow stable bed expansion during protein adsorption. Usually, the liquid axial dispersion coefcient in expanded beds is measured from residence time distribution (RTD) by using the moment method.11,23-25 Based on the experimental data of RTD curves, the mean residence time and the variance of distribution can be calculated. Then, by letting the rst absolute moment of the dispersion model equal the mean residence time, and the second central moment of the dispersion model equal the variance of distribution, the liquid axial dispersion coefcient can be obtained easily. However, the validity of the dispersion model cannot be judged by this moment analytical method because the tting degree of the calculated response curves to that measured experimentally cannot be evaluated directly. Sometimes, it will cause a signicant deviation for the estimation of the liquid axial dispersion coefcient with the baseline drift and baseline uctuation of experimental RTD curves during measurements. FernandezLahore et al.26 tted the experimental RTD curves with the theoretical model in Laplace domain using the expression originally developed by Villermaux et al.27 In this paper, an analytical solution for diract input mode, in which liquid axial dispersion, tracer intraparticle diffusion, and lm mass transfer all are taken into account, is used to t experimental RTD curves to better estimate the liquid axial dispersion coefcient.

Experimental Equipment
A pilot-scale Streamline 50 column (Amersham Pharmacia Biotech, Uppsala, Sweden) was used in all expanded bed experiments. Masterex peristaltic pumps (Cole-Parmer Instrument Co., Vernon Hills, IL) were used for buffer and feed application and to raise and lower the hydraulic adaptor of the
Vol. 51, No. 11 AIChE Journal

Streamline 50 column. A Jasco 7800 UV detector (Tokyo, Japan) equipped with a owcell was used to monitor online BSA efuent concentration from the expanded bed and tracer (acetone) efuent concentration in RTD experiments, and the absorbance signal was logged using the data-acquisition software in a personal computer.

Adsorbents
Streamline DEAE and Streamline direct CST I were purchased from Amersham Pharmacia Biotech. Streamline DEAE is a weak anion exchanger with an OCH2CH2N(C2H5)2H functional group, with the following characteristics: its matrix consists of macroporous crosslinked 6% agarose constraining crystalline quartz core materials, with a particle density of about 1200 kg/m3, a particle size distribution of 100 300 m, and a mean particle size of 200 m. Streamline direct CST I is an ion exchanger with a multimodal functional group, with the following characteristics: its matrix consists of macroporous crosslinked 4% agarose constraining stainless steel core materials, with a particle density of about 1800 kg/m3, a particle size distribution of 80 165 m, and a mean particle size of 135 m.

maintain a stable expansion. For Streamline direct CST I, the settled bed height is 15.6 cm and the settled bed voidage is 0.39; for Streamline DEAE, the settled bed height is 16.5 cm and the settled bed voidage is 0.4. The liquid axial dispersion coefcient in expanded beds is measured by the RTD method. Acetone inert tracer is used in all RTD experiments. About 50% v/v concentration of acetone (1.5 mL buffer aqueous solution) is used for the dirac input mode, and input position at the bottom of the column. Before carrying out the RTD measurement, the bed is expanded about 1 h by 20 mM phosphate buffer (pH 7.5) for Streamline DEAE, and by 50 mM acetate buffer (pH 5) for Streamline direct CST I. A dirac input of acetone sample is applied to the column at the bottom of the bed; the efuent liquid sample passes through the owcell online, where the acetone concentration is monitored by UV detector at 280-nm wavelength; and the ABS (absorbance) signal of acetone is logged by the data-acquisition software in a personal computer.

Experimental procedures for the whole expanded bed adsorption process

Expansion/Equilibration Stage. First, the equilibration buffer is pumped through the column with upward ow to the expected expansion degree. Second, the expanded bed is allowed to stabilize at this degree of expansion for about 30 40 min. Then, the liquid axial dispersion in the expanded bed is checked by the RTD method; the liquid axial dispersion coefcient should be as small as possible by avoiding inadequate operation. The adaptor will be positioned about 0.5 cm above the height to which the bed expands, to reduce the dead volume in expanded beds. Adsorption Stage. When the expanded bed is stable and equilibrated with the appropriate buffer, the process switches to feedstock application. Because of protein adsorption on adsorbents, the expanded bed height gradually drops, especially in the expanded bed of low-density Streamline DEAE; therefore, the liquid ow velocity will be increased gradually during the adsorption process to maintain a constant degree of bed expansion. The average liquid velocity was calculated from the ratio of total feed volume supplied to the column to the operation time in the loading process. The efuent stream from the top of the column will pass through the owcell, where BSA efuent concentration is monitored online by UV 7800 detector, and the ABS (absorbance) signal of BSA is logged by the dataacquisition software in a personal computer. The BSA concentration in the feed was 2 kg/m3, showing a linear relation between the UV detector signal (ABS) and BSA efuent concentration during the adsorption stage. Washing Stage. When the measurement of the breakthrough curve was completed, the process switches to the wash buffer to wash out the excess protein, other loosely bound materials, and particulates from the column in the expanded mode until the efuent absorbance reaches a relative stable value. In the expanded bed of Streamline direct CST I, because of the highly favorable adsorption of BSA, the efuent absorbance approaches the baseline, meaning almost irreversible adsorption. Elution Stage. After washing, the pump is turned off and the bed is allowed to settle. When the adsorbent has settled, the adaptor is moved down toward the surface until the edge of the
Vol. 51, No. 11 2967

Model protein
The model protein, bovine serum albumin (BSA; product number A3059, further puried fraction V, 99%), was purchased from Sigma (St. Louis, MO). The molecular weight of BSA is about 65,400 g mol1, diffusion coefcient at innite dilution in water is about 6.15 1011 m2/s, radius of gyration is 29.8 , and isoelectric point is about 4.7. A sample solution of BSA was prepared with the appropriate buffer; in the case of Streamline DEAE the buffer is 20 mM phosphate buffer (pH 7.5), a mixture of Na2HPO4/NaH2PO4, and in the case of Streamline direct CST I the buffer is 50 mM acetate buffer (pH 5), a mixture of acetate sodium and acetic acid. Distilled water was used in all experiments.

Batch adsorption experiments

Before performing adsorption isotherm experiments, the adsorbents must be saturated by phosphate buffer or acetate buffer. Adsorbents, in the amount of 0.5 mL by particle volume, are equilibrated with 30 mL of different concentrations of BSA solution about 8 h at 25C on a shaking incubator (about 30 rpm); then BSA concentration in supernatant liquid is measured by UV 7800 detector at 280 nm (using a 2-mL quartz cuvette). The adsorption capacity is calculated by mass balance. For the kinetic experiment 2 mL of adsorbent was mixed with 100 mL of BSA solution in a ask. The adsorption was carried out in the shaking incubator at 25C at 150 rpm. Every few minutes, about 2 mL of the liquid phase was aspirated using a suction tube to determine protein concentration, and the sample was immediately returned to the vessel. By this procedure, the time course of BSA concentration in the batch adsorber was determined to estimate the effective pore diffusivity of BSA in Streamline DEAE and Streamline direct CST I.

Residence time distribution (RTD) experiments

A Streamline 50 column is packed either with 300 mL Streamline DEAE or with 300 mL Streamline direct CST I to
AIChE Journal November 2005

adaptor net touches the bed. Elution buffer is then pumped through the settled bed with a downward ow to elute BSA protein. Because the BSA concentration in the efuent is very high, the efuent samples are collected by small sample tubes, and the BSA concentration in these samples is then monitored by UV 7800 detector.

Boundary Conditions

q Z

0
Z0

(2a)

Mathematical Model for the Protein Adsorption in Expanded Beds Mathematical model
The mathematical model is developed based on the work of Wright et al.,14 where intraparticle diffusion, lm mass transfer, liquid axial dispersion, and solid axial dispersion were taken into account. In addition, the particle size axial distribution and bed voidage axial variation in expanded beds are also included in the model. The material balance equation for the liquid bulk phase in an expanded bed is DL C C 3 k f Z 1 B Z 2 B Z C u B Z 2 Z Z t RZ C c rRZ 0 (1)

q Z Initial Condition t0

0
ZH

(2b)

q Z, 0 0

(2c)

The pore diffusion equation in the adsorbent is described as P c q 2c 2 c De t t r2 r r

(3)

where q is the adsorbed amount concentration in adsorbent, p is the particle porosity, and De is the adsorbate effective pore diffusivity. Boundary Conditions

where DL is the axial dispersion coefcient; C is the concentration in the uid phase; u denotes supercial velocity; c is the concentration in the adsorbent pore; Z is the axial distance from column entrance; B(Z)denotes bed voidage at the axial distance Z of the column, and thus [1 B(Z)] denotes the fractional volume taken up by the solid phase; r is the radial coordinate in the adsorbent particle; R(Z) is the radius of the adsorbent at the axial distance Z of the column; t is the time; and kf (Z) is the lm mass transfer coefcient at the axial distance Z of the column. Boundary Conditions

De

c r

rRZ

R Z D ax,S 2q k f Z C c rRZ 3 1 B Z Z 2 (3a)

c r Initial Condition t0

0
r0

(3b)

DL

C Z

u C Z0 C 0 B 0 Z0

cr 0

qr 0

(3c)

(1a) In Eq. 3, the relationship between q and c depends on the adsorption equilibrium of the selected experimental system. Based on the experimental measurements for BSA protein adsorption on Streamline DEAE and on Streamline direct CST I, the Langmuir isotherm is assigned as q (1c) q mc kd c

C Z Initial Condition t0

0
ZH

(1b)

CZ, 0 0

(4)

The mass balance for adsorbent bulk phases in the expanded bed column is described as 1 B Z q 3 2q D ax,S 2 1 B Z k Z t Z RZ f C c rRZ (2) where the q is the average adsorbent phase concentration and Dax,S is the solid axial dispersion coefcient in the expanded bed.
2968 November 2005

where qm is the adsorption capacity and kd is the dissociation constant, both of which are determined by experiments. The correlations recommended to estimate the particle size axial distribution and bed voidage axial variation in expanded beds, packed with the rst-generation adsorbents, are RZ p Z d 1.20 0.51 2 H

(5)

B 0.629 0.738 B Z
Vol. 51, No. 11

Z H

(6)

AIChE Journal

 p is an averwhere the averaged adsorbent particle diameter d aged value over the wider particle size distribution, and the bed averaged voidage B is estimated as a function of the whole expanded bed height28 B 1 (1 0)

H0 H

(7)

where 0 and H0 are the settled bed voidage and the settled bed height, respectively. The correlations given by Eqs. 5 and 6 are based on the experimental data of Bruce et al.,11 where a Streamline 50 column was packed with Streamline SP to measure both the particle size axial distribution and bed voidage axial variation in the expanded bed. Because that system is similar to our experimental setup, that is, a Streamline 50 column packed with Streamline DEAE (the same Streamline matrix), we used in the modeling the hydrodynamic results obtained by Bruce et al.11 Kaczmarski et al.22 recommended two different equations for the particle size axial distribution and bed voidage axial variation; simulated breakthrough curves using correlations from both groups11,22 are similar.

Figure 1. BSA adsorption isotherms on Streamline DEAE and on Streamline direct CST I at room temperature (25C).
Circle points: experimental data; lines: the calculated results by Langmuir equations, Eqs. 10a and 10b.

Results and Discussion Adsorption isotherm and BSA effective pore diffusivity
Based on the independent batch adsorption equilibrium experiments, as shown in Figure 1, circle points represent the experimental data at room temperature (25C). BSA protein adsorption isotherm on both Streamline DEAE and Streamline direct CST I can be approximately described by the Langmuir equation as follows. BSA Adsorption on Streamline DEAE q 92.59 c 0.065 c (10a)

Model parameters
The effective pore diffusivity De in adsorbents is determined by independent batch experiments. The liquid axial dispersion coefcient DL is measured experimentally during the expansion stage by the residence time distribution method. The adsorbent axial dispersion coefcient Dax,S is estimated by the correlation of Van Der Meer et al.,29 using experimental values for supercial velocity u, as follows D ax,S 0.04 u 1.8 m2 /s (8)

BSA Adsorption on Streamline Direct CST I q 82.15 c 0.0109 c (10b)

The WilsonGeankopolis equation,30 applicable to low Reynolds number (Eq. 9), is used to estimate the lm mass transfer coefcient kf (Z) in an expanded bed Sh 1.09 Re1/3Sc1/3 B Z 0.0015 Re 55 (9)

According to the Langmuir equation, the separation factor f can be dened as f 1 1 c 0/ k d (11)

where Re is the Reynolds number [2R(Z)u/], Sc is the Schmidt number [/(Dm)], Sh is the Sherwood number [2kf (Z)R(Z)/Dm], and Dm is the molecular diffusion coefcient.

Numerical method
The model equations are numerically solved by the orthogonal collocation method. Equations 1 and 2 are discretized at collocation points in the axial direction in the column, and Eq. 3 is discretized at collocation points in the particle radial direction, leading to a set of ordinary differential equations with initial values that are integrated in the time domain using Gears stiff variable step integration routine. To obtain a stable numerical solution for a highly favorable adsorption isotherm, 21 bed axial collocation points and 21 particle radial collocation points are used.
AIChE Journal November 2005

which characterizes the adsorption conditions. For a BSA concentration of 2 kg/m3, which will be used in the adsorption kinetics experiments in batch adsorber and in expanded bed adsorption later, the separation factors are 0.032 for BSA adsorption on Streamline DEAE and 0.0055 for BSA adsorption on Streamline direct CST I, respectively, indicating highly favorable adsorption for BSA on Streamline adsorbents, especially on Streamline direct CST I (where virtually irreversible adsorption occurs). Batch adsorption kinetic experiments are carried out to estimate BSA effective pore diffusivity in Streamline DEAE and in Streamline direct CST I; Figure 2 shows typical experimental data of the bulk concentration proles in a batch adsorber (with 100 mL of 2 kg/m3 BSA aqueous solution and 2 mL adsorbent).Then, the experimental data of the bulk concentraVol. 51, No. 11 2969

degree of expansion, the supercial liquid ow velocity for the new Streamline direct CST I is higher than that for the old Streamline DEAE, given that the particle density of the new adsorbent (1800 kg/m3) is greater than that of the old adsorbent (1200 kg/m3). When the expansion degree is equal to 2, the supercial ow velocity for Streamline DEAE is only 228 cm/h, but for Streamline CST I, it allows a higher velocity of the feedstock (up to 553 cm/h) to pass through its expanded bed. The residence time distribution method with a dirac tracer (acetone) input mode is used to estimate the liquid axial dispersion coefcient in expanded beds packed with Streamline DEAE and packed with Streamline direct CST I, respectively; the experimental data of RTD curves are shown in Figure 4, marked as circle points. Based on the experimental data of RTD curves, the mean residence time ( t m ) and the variance of distribution (2) are calculated as tm
tCdt t iC i t 0 C i t 0 Cdt

(12)

2
Figure 2. Estimation of BSA effective pore diffusivity in Streamline DEAE adsorbents and in Streamline CST I adsorbents in batch adsorber.
Circle points: experimental data; solid lines: simulation results of the pore diffusion model with De 3.2 1011 m2/s and kf 8.6 106 m/s for Streamline DEAE (a) and with De 1.7 1011 m2/s, and kf for Streamline direct CST I (b).

2 t Cdt 0 t i2C i t 2 2 t tm m C i t 0 Cdt

(13)

If we then let the rst absolute moment 1 of the dispersion model equal the mean residence time ( t m ) and the second central moment 2 of the dispersion model equal the variance of distribution (2), the liquid axial dispersion coefcient can be easily obtained. Some common calculation formulas, Eq. 14,33 Eq. 15,11 and Eq. 16,34 are summarized as follows: BD L 2 2 2 uH tm 2 BD L/ uH 3 2 BD L/ uH 2 2 1 2 BD L/ uH BD L/ uH 2 tm BD L 2 BD L 2 BDL 2 2 1 e uH/ 2 uH uH tm (14)

tion prole are tted with the simulation results of the pore diffusion model to estimate BSA effective pore diffusion coefcient in Streamline DEAE as 3.2 1011 m2/s and in Streamline direct CST I as 1.7 1011 m2/s. It should be emphasized that in the pore diffusion model, the particle size is assigned an average particle diameter over the particle size p 200 m for Streamline DEAE; d p 135 m distribution (d for Streamline direct CST I), and the particle porosity is assigned a value of 0.55, according to the published literature.31 The lm mass transfer resistance is not negligible for BSA adsorption on Streamline DEAE; in contrast, the lm mass transfer can be neglected for BSA absorption on Streamline direct CST I in the batch experimental system. A method to estimate both lm mass transfer coefcient and effective pore diffusivity from a single bulk concentrationtime curve in a batch adsorber has been reported elsewhere.32

(15)

(16)

Bed expansion and liquid axial dispersion coefcient in expanded bed

The expansion degree of expanded beds packed either with Streamline DEAE or with Streamline directs CST I is measured at various supercial ow velocities, as shown in Figure 3, where a Streamline 50 column is packed either with 300 mL Streamline DEAE or with 300 mL Streamline direct CST I; the settled bed height is 15.6 cm for Streamline direct CST I and 16.5 cm for Streamline DEAE. It is apparent that at the same
2970 November 2005

Figure 3. Relationship between bed expansion degree and supercial liquid ow velocity in expanded beds packed with Streamline DEAE and with Streamline CST I. Vol. 51, No. 11 AIChE Journal

residence time of extra-column volume and variance are tc 2 38 s and 2 c 352 s at 228 cm/h ow rate and t c 18 s and 2 2 c 126 s for 553 cm/h, respectively. If there is a baseline drift or a baseline uctuation for experimental RTD curves, the liquid axial dispersion coefcient estimated by the previous calculation method may deviate from the real value of liquid axial dispersion coefcients. Therefore, in Figure 4, the experimental data of RTD curves are tted by the analytical solutions with the dirac input of acetone tracer to conrm the calculation accuracy, and at the same time, the parametric sensitivity is analyzed. The analytical solution is given in the Appendix for reference, in which the tracer intraparticle diffusion resistance, lm mass transfer resistance, and liquid axial dispersion coefcient are all taken into account. In Figure 4, the mean residence time has been corrected, and the variance associated with extra-column volumes was negligible compared to the variance associated with the expanded bed.

BSA protein breakthrough behavior in expanded beds packed with Streamline direct CST I and with Streamline DEAE
Figure 4. Experimental data of RTD curves are tted by the analytical solution with the dirac input of acetone tracer at expanded beds packed with Streamline DEAE and packed with Streamline direct CST I.
(a) Settled bed height 16.5 cm, expanded degree as 2, super 2 0.128t; (b) cial liquid ow velocity 228 cm/h, Dpt/R settled bed height as 15.6 cm, expansion degree as 2, super 2 0.281t. cial liquid ow velocity 553 cm/h, Dpt/R

When considering the tracer intraparticle diffusion resistance, lm mass-transfer resistance, and liquid axial dispersion, one has 2 2 1 B 5 2 BD L 2 2 uR 1 P 2 2 15 H 1 D uH tm k fR B p B P P (17) In Eq. 17, estimations of acetone intraparticle diffusivity and particle porosity for acetone are described in the Appendix. The results calculated by different formulas are similar. Because of the small size of the tracer molecule, acetone, the effect of the tracer intraparticle diffusion is negligible. For Streamline DEAE, when the bed expands to about twice the settled bed height with 228 cm/h supercial ow velocity, the liquid axial dispersion coefcient is 4.45 106 m/s; for Streamline direct CST I, when the bed expands to about twice the settled bed height with 553 cm/h supercial ow velocity, the liquid axial dispersion coefcient increases to 17.2 106 m/s as a result of the liquid ow velocity increase. In the calculation, the mean residence time and variance of the residence time distribution of the sampling system and the precolumn tubing were subtracted from the measured mean residence time and variance of the overall residence time distribution to yield the corrected mean residence time and variance of the RTD of the expanded bed column. As an example, the mean
AIChE Journal November 2005

BSA Breakthrough Behavior in Expanded Beds Packed with Streamline Direct CST I. When a given volume (300 mL) of Streamline direct CST I is packed into a Streamline 50 column, the settled bed height is 15.6 cm. BSA aqueous solution with concentration 2 kg/m3 prepared with 50 mM acetate buffer (pH 5) is introduced to the column in upward ow with 181 mL/min ow velocity. The experimental data of the BSA breakthrough curve are shown in Figure 5, marked as circle points. The uniform model, where the model parameters are average values all over the column (average particle diameter and average bed voidage), is used to predict the breakthrough curve, as shown in Figure 5 (solid line). Because of the heavier adsorbent, Streamline direct CST I, with narrower particle size distribution (80 165 m), the effects of the particle size axial distribution and the bed voidage axial variation on the breakthrough curves are smaller, so the uniform model predicts the breakthrough curve in expanded beds reasonably well.

Figure 5. BSA breakthrough curve in expanded bed packed with Streamline direct CST I.
Circle points: experimental data; solid line: uniform model; dashed line: the analytical solution (Eq. 18) with irreversible adsorption (qm 82.15 kg/m3). The experimental conditions and model parameters are summarized in Table 1.

Vol. 51, No. 11

2971

Table 1. Experimental Conditions and Model Parameters Used for the Simulation of the Breakthrough Curves in Expanded Beds
Expanded Bed of Streamline Direct CST I H 0 15.6 cm B 0 0.39 p 135 m d p 0.55 C 0 2 kg/m3 u 15.37 10 4 m/s* H 32.0 cm H / H 0 2.05 B 0.7025 q m 82.15 kg/m3 k d 0.0109 kg/m3 D e 1.7 10 11 m2/s D L 17.2 10 6 m2/s k f 10.6 10 6 m/s D ax , S 3.45 10 7 m2/s Expanded Bed of Streamline DEAE H 0 16.5 cm B 0 0.4 p 200 m d p 0.55 C 0 2 kg/m3 u 7.10 10 4 m/s* H 33.5 cm H / H 0 2.03 B 0.7045 q m 92.59 kg/m3 k d 0.065 kg/m3 D e 3.2 10 11 m2/s D L 4.45 10 6 m2/s k f 6.6 10 6 m/s D ax , S 8.78 10 8 m2/s

tion on Streamline direct CST I is highly favorable, leading to almost irreversible adsorption (Figure 1); therefore, instead of the numerical solution, a simple analytical solution derived at the irreversible adsorption isotherm with qm 82.15 kg/m3 (Eq. 18), which takes into account both the intraparticle diffusion and lm mass transfer, may be used approximately to predict the breakthrough curves in the expanded bed (as shown in Figure 5, dashed line). The analytical solution is indeed close to the experimental data in expanded bed adsorption. Moreover, the analytical solution (dashed line in Figure 5) is similar to the simulated breakthrough curve from a model where only the intraparticle diffusion and lm mass transfer are signicant (dashed double-dotted line in Figure 6). This analytical solution was reported by Weber and Chakravorti35 based on the assumption of constant pattern, as follows 1 1 N pore 15 tan 1

*u is the average value during the adsorption stage, with the interval minimum maximum liquid velocity of 6.68 7.52 104 m/s for Streamline DEAE, and of 15.20 15.54 104 m/s for Streamline direct CST I.

1 5 5 ln1 3 1 3 Bi 2 3

2 1

15 ln(1 2 ) 2 (18)

The model parameters De, kf, DL, and Dax,S characterize the effects of intraparticle diffusion resistance, lm mass-transfer resistance, liquid axial dispersion, and solid axial dispersion on the breakthrough curves during expanded bed adsorption. Figure 6 demonstrates the individual contribution of each model parameter (De, kf, DL, or Dax,S) on the breakthrough curve during expanded bed adsorption. First, by controlling lm mass transfer resistance, the liquid and solid axial dispersion must be as small as possible to neglect their effects on the breakthrough curves (kf increased 10 times, DL and Dax,S decreased 10 times); the effect of the model parameter De on the breakthrough curve is demonstrated in Figure 6, represented by a dashed line. It is apparent that the contribution of BSA effective pore diffusivity to the breakthrough curves is dominant (dashed line). Then, in turn, the effects of lm mass transfer resistance, liquid axial dispersion, and solid axial dispersion are considered in the model; the simulation results are demonstrated in Figure 6, respectively. The dashed double-dotted line represents the simulation result when both the intraparticle diffusion resistance and the lm mass transfer resistance are taken into account in the model. By comparing with the simulation result (dashed line, considering only De), the effect of the lm mass transfer coefcient is not negligible for such the highly favorable protein adsorption isotherm because the breakthrough time is signicantly shortened as the result of the effect of kf. The dashed-dotted line represents the simulation results where intraparticle diffusion resistance, lm mass transfer resistance, and liquid axial dispersion are taken into account in the model, and the solid line represents the simulation results when intraparticle diffusion resistance, lm mass transfer resistance, liquid axial dispersion, and solid axial dispersion all are taken into account in the model. It is apparent that the effects of the liquid axial dispersion coefcient and the solid axial dispersion coefcient are smaller even at high liquid ow velocity (up to 553 cm/h) if the bed expansion is stable. Based on the simulation results, as shown in Figure 6, the effects of the liquid axial dispersion and solid axial dispersion on the breakthrough curves are smaller in a stable expanded bed packed with Streamline direct CST I. BSA protein adsorp2972 November 2005

where the dimensionless parameters are dened as ut B H 1

N pore

151 B De H C 1 2 C0 uR Bi kf R De

1/3

1 B q m C0

Figure 6. Contribution of each model parameter (De, kf, DL, or Dax,[in]S) to the breakthrough curve in expanded bed packed with Streamline direct CST I.
Dashed line: uniform model with De 1.7 1011 m2/s (neglect kf, DL, and Dax,S effects); dashed double-dotted line: uniform model with De 1.7 1011 m2/s and kf 10.6 106 m/s (neglect DL and Dax,S effects); dashed dotted line: uniform model with De 1.7 1011 m2/s, kf 10.6 106 m/s, and DL 17.2 106 m2/s (neglect Dax,S effect); solid line: uniform model with considering De, kf, DL, and Dax,S effects (De 1.7 1011 m2/s, kf 10.6 106 m/s, DL 17.2 106 m2/s, and Dax,S 3.45 107 m2/s). The other calculation conditions are the same as in Figure 5.

Vol. 51, No. 11

AIChE Journal

Figure 7. BSA breakthrough curve in expanded bed packed with Streamline DEAE.
Circle points: experimental data; dashed line: uniform model; solid line: modied uniform model; dashed dotted line: modied uniform model with double kf value. The experimental conditions and model parameters are summarized in Table 1.

It should be emphasized that this formula requires that 1Npore 2.5 1/Bi. The effects of particle size axial distribution, bed voidage axial variation, liquid axial dispersion, and solid axial dispersion on the breakthrough curve are smaller than those of intraparticle diffusivity and lm mass-transfer coefcient for BSA adsorption in the expanded bed packed with streamline direct CST I; therefore, the uniform model and the analytical solution give a reasonable t of experimental breakthrough curves. However, in the uniform model, the effects of the particle size axial distribution and bed voidage axial variation on the breakthrough curve are neglected, which results in predicted breakthrough curves slightly broader than the experimental data (greater amount of larger-size adsorbent at the bottom of the column and low amount of smaller-size adsorbent at the top of the column, will make the real breakthrough curve steeper). In the analytical solution, the effects of the particle size axial distribution and the bed voidage axial variation on the breakthrough curve (making the breakthrough curves slightly steeper), and the effects of the liquid axial dispersion and the solid axial dispersion on the breakthrough curve (making the breakthrough curves slightly broader) are all neglected; it happens that the sharpening effect almost equals the broadening effect for our experimental conditions, leading to an analytical solution that is closer to experimental data. BSA Breakthrough Behavior in Expanded Beds Packed with Streamline DEAE. When Streamline DEAE (300 mL) is packed into a Streamline 50 column, the settled bed height is 16.5 cm. A 2 kg/m3 BSA aqueous solution, prepared with 20 mM phosphate buffer (pH 7.5) is introduced in upward ow at the column bottom with 83.6 mL/min ow velocity. The bed is expanded to about twice the settled bed height. The experimental data of the BSA breakthrough curve are shown in Figure 7, marked as circle points. First, the uniform model is used to predict the breakthrough curve, and the simulation result (dashed line, in Figure 7) is compared with the experimental data. The simulation result does not t the experimental data very well. Based on the experimental results, at the initial breakthrough stage, the real adsorption behavior in expanded beds is better than the predicted results by the uniform model; however, there
AIChE Journal November 2005

is a tailing behavior of the breakthrough curves when the efuent concentration approaches the feed concentration. In previously published articles, the tailing behavior was explained by the presence of the dimer in BSA samples, or microporous diffusion in the macroporous adsorbent, or protein steric hindrance on active sites of the surface of the adsorbent. Until now, the explanation is still unclear for the tailing behavior of the breakthrough curves that often occur for both medium and large size protein adsorption. Because the Streamline adsorbents are macroporous, at the initial adsorption stage, the macroporous diffusion should be considered; that is, the breakthrough curves may be predicted by the macroporous diffusion model for the initial breakthrough stage. For the protein adsorption in a stable expanded bed, the slow diffusion rate of macromolecular protein in adsorbent signicantly affects the breakthrough behavior in the expanded bed. Therefore, the particle size should have an effect on the breakthrough curves as the result of particle diameter, characterizing the diffusion path in the adsorbent particle (small particle diameters having a shorter diffusion path length, leading to lower diffusion resistance than that in larger particle). Compared with Streamline direct CST I, Streamline DEAE has a wider particle size distribution (100 300 m); when the bed expands, the smaller and lighter particles move to positions at the top of the expanded bed, the larger and heavier particles to the bottom, and more adsorbents will be present at the bottom of the column. At the top zone of the column there is a small amount of adsorbent, and thus the real adsorption behavior in expanded beds is better than the predicted result by the uniform mode. The simulation result by the modied uniform model with taking into account the particle size axial distribution (Eq. 5) and bed voidage axial variation (Eq. 6) is shown in Figure 7, denoted by the solid line. Compared with the uniform model, the simulation results with the modied uniform model better describe the initial breakthrough. For such a highly favorable adsorption isotherm, BSA protein diffuses by a shrinking core mode in the Streamline DEAE, so the more signicant effect of the particle size axial distribution and bed voidage axial variation on the breakthrough curves will be observed at the end of the breakthrough curves, and not at the initial breakthrough stage. In our previous work21 for the small-size protein (lysozyme) adsorption on Streamline SP in expanded beds, where the tailing behavior of the breakthrough curves was not observed, a signicant improvement in the simulation results by the modied uniform model can be observed. In Figure 8, we give the detailed comparisons among three simulation results: the dashed line represents the uniform model, the dashed-dotted line represents the modied uniform model by taking into account the particle size distribution, and the solid lines represent the modied uniform model by taking into account both the particle size distribution and bed voidage axial variation. From Figure 8, by comparing the simulation results of the uniform model (dashed line) and modied uniform model with particle size axial distribution (dashed dotted line), the effect of the particle size distribution on the breakthrough curve is signicant; however, by comparing the simulation results of the modied uniform model with particle size axial distribution (dashed dotted line) and the modied uniform model with both particle size axial distribution and bed voidage axial variation (solid line), the effect of the bed voidage axial variation on the breakthrough curve is small. It should
Vol. 51, No. 11 2973

Figure 8. Effect of the particle size axial distribution and bed voidage axial distribution on the breakthrough curves in expanded beds.
Circle points: experimental data; dashed line: uniform model; dash dotted line: modied uniform model with considering particle size axial distribution; solid line: modied uniform model with considering both particle size axial distribution and bed voidage axial distribution. The experimental conditions and model parameters are the same as in Figure 7.

Figure 9. Efuent curves of BSA protein during adsorption, washing, and elution stages in expanded bed packed with Streamline direct CST I.
At the adsorption stage, 2 kg/m3 BSA aqueous solution, prepared with 50 mM acetate buffer, pH 5, is applied from the bottom of the expanded bed at 181 mL/min ow velocity; at the washing stage, 50 mM acetate buffer, pH 5, is applied from the bottom of the expanded bed; and at the elution stage, 50 mM acetate buffer with 1 M NaCl, pH 7, is applied from the top of the settled bed at 39 mL/min ow rate.

be emphasized that there is an effect of bed voidage axial variation on the breakthrough curve when adsorbents have larger particle diameter with wider particle size distribution. Kaczmarski and Bellot22 claimed that the bed voidage axial variation had no effect on breakthrough curves only for the p values of 50 or 150 m) case of smaller-size adsorbents (d with relatively narrow particle size distribution. It is observed that the real bed adsorption behavior at the initial breakthrough stage is still better than the predicted results even if we modify the uniform model by taking into account the particle size axial dispersion and bed voidage axial variation. Based on the theoretical analysis, as shown in Figure 6, the lm mass transfer coefcient also has an important effect on the breakthrough time for such a highly favorable adsorption isotherm. It is very important to correctly estimate kf in the simulation. From the published the papers, the lm mass transfer coefcient kf is estimated by the correlations for xed beds or their revised formula. In expanded beds, the adsorbent particles are suspended in the liquid phase and uctuate slightly up and down, which will favor the lm mass transfer; that is, the real kf value should be larger than the estimated value of kf by the previous correlations. In Figure 7, simulation results with double kf value (kf 13.2 106 m/s) will more nearly reect the real breakthrough time.

recovery proceeds at the elution stage. The detailed operation procedures have been previously described. Based on the experimental results shown in Figures 9 and 10, the comprehensive evaluations on the hydrodynamics, BSA dynamic adsorption capacity, and BSA recovery in the expanded bed adsorption process with Streamline DEAE and with Streamline direct CST I are summarized as follows: (1) For the same degree of expansion and the same expanded bed height, the high-density Streamline direct CST I allows a higher feed ow velocity (553 cm/h) to pass through the expanded bed; in contrast, a low feed ow velocity (259 cm/h) is allowed to pass through the expanded bed packed with low-density Streamline DEAE. Because of the high ow velocity, the liquid axial dispersion coefcient is correspondingly increased in expanded beds of Streamline direct CST I. Based on our experimental and theoretical research on the BSA break-

Comprehensive evaluations on the whole expanded-bed protein adsorption process with Streamline DEAE and with Streamline CST I
Experiments are carried out for the whole expanded-bed BSA protein adsorption process with Streamline direct CST I (Figure 9) and with Streamline DEAE (Figure 10). A Streamline 50 column is packed either with Streamline direct CST I or Streamline DEAE at the same amount of the adsorbents (300 mL). With the same degree of expansion (twice the settled bed height), 2 kg/m3 BSA aqueous solution is applied to the expanded beds and BSA protein is adsorbed by the adsorbents; after the adsorption stage, the bed is washed and BSA protein
2974 November 2005 Figure 10. Efuent curves of BSA protein during adsorption, washing, and elution stages in expanded bed packed with Streamline DEAE.
At the adsorption stage, 2 kg/m3 BSA aqueous solution, prepared with 20 mM phosphate buffer, pH 7.5, is applied from the bottom of the expanded bed at 83.6 mL/min ow velocity; at the washing stage, 20 mM phosphate buffer, pH 7.5, is applied from the bottom of the expanded bed; and at the elution stage, 20 mM phosphate buffer with 0.5 M NaCl, pH 7.5, is applied from the top of the settled bed at 37 mL/min ow rate.

Vol. 51, No. 11

AIChE Journal

through behavior in an expanded bed of Streamline direct CST I, the effect of the liquid axial dispersion on the breakthrough curve is also slight, even at such a high ow velocity if the bed expansion is stable. (2) At 5% BSA breakthrough point during expanded bed adsorption, BSA dynamic binding capacity on Streamline direct CST I is 34 mg (BSA)/mL of settled bed volume and BSA dynamic binding capacity on Streamline DEAE is 50 mg (BSA)/mL of settled bed volume. However, BSA binding capacity on Streamline direct CST I is not sensitive to ionic strength in feedstock, which means there is no need for dilution of feedstock arising from high ionic strength. In contrast, BSA binding capacity on Streamline DEAE is very sensitive to the ionic strength in feedstock; when the ionic strength in feedstock is increased to 50 mM, the BSA adsorption capacity decreases by half. BSA dynamic binding capacity, Q5%, is calculated from the BSA breakthrough curve at expanded bed adsorption stage; the formula is dened as

Conclusions
With the specially designed adsorbents (Streamline DEAE and Streamline direct CST I), a stable expanded bed can be formed. At the same degree of expansion, the high-density Streamline direct CST I allows a higher feed ow velocity to pass through the expanded bed; in contrast, a lower feed ow velocity is allowed to pass through the expanded bed of lowerdensity Streamline DEAE. With the high feed ow velocity, the liquid axial dispersion is more signicant in the expanded bed of Streamline direct CST I than that in the expanded bed of Streamline DEAE. In spite of the existence of intraparticle diffusion resistance, lm mass transfer resistance, liquid axial dispersion, and solid axial dispersion during expanded bed adsorption, the contribution of BSA effective pore diffusivity to the breakthrough curves is dominant. The lm mass transfer coefcient has a signicant effect on the initial breakthrough time for the highly favorable protein adsorption isotherm; liquid axial dispersion and solid axial dispersion have a lesser effect on the breakthrough curves, even at high liquid ow velocity (up to 553 cm/h for Streamline direct CST I), if the bed expansion is stable. Because of the narrow particle size distribution of Streamline direct CST I, the effects of the particle size axial dispersion and the bed voidage axial variation on the breakthrough behavior in the expanded bed are small, and thus the uniform model can be used to predict the breakthrough curves with acceptable accuracy. In contrast, because of the wide particle size distribution of Streamline DEAE, the effects of the particle size axial distribution and bed voidage axial variation on the breakthrough curves in the expanded bed should be taken into account in the model. Based on the experimental results, at 5% BSA breakthrough point during expanded bed adsorption, the BSA dynamic binding capacity on Streamline DEAE is 50 mg (BSA)/mL of settled bed volume, larger than that on Streamline direct CST I (34 mg (BSA)/mL of settled bed volume). However, the BSA binding capacity on Streamline CST I is not sensitive to ionic strength in the feedstock, which means there is no need for dilution of feedstock even at high ionic strength. In contrast, the BSA binding capacity on Streamline DEAE is very sensitive to the ionic strength in the feedstock; when the ionic strength in the feedstock is increased to 50 mM, BSA adsorption capacity decreases by half. The ligand of Streamline DEAE is very sensitive to salt concentration in the buffer, so BSA adsorbed on Streamline DEAE can be easily eluted by increasing the salt concentration to 0.5 M in 20 mM phosphate buffer, pH 7.5. BSA recovery in the whole expanded bed adsorption process reaches 91%, and a slight amount of the elution buffer is consumed. Streamline direct CST I has a multimodal ligand that is not sensitive to the salt concentration. Both salt concentration and pH value should be increased in the elution buffer; for example, 50 mM acetate buffer with 1 M NaCl, pH 7, is used as the elution buffer in this experiment, and BSA recovery in the whole expanded bed adsorption process is up to 87%, and the amount of the elution buffer is greater than that consumed for Streamline DEAE. In addition, from the BSA efuent concentration curve during the elution stage, it is observed that BSA cannot be eluted completely from Streamline CST I.
Vol. 51, No. 11 2975

Q 5%

V5%

C 0 C dV VA (19)

where V is the efuent liquid volume from expanded beds, V5% is the efuent liquid volume at 5% BSA breakthrough point, and VA is the settled bed volume of adsorbents. (3) At the washing stage, it is found that BSA efuent concentration quickly decreases and approaches the baseline for the expanded bed of Streamline direct CST I, which means almost irreversible adsorption for BSA binding to Streamline CST I. BSA adsorption on Streamline DEAE is reversible; when washing, some BSA can be desorbed from Streamline DEAE so that the efuent concentration approaches a relatively stable value. (4) The ligand of Streamline DEAE is very sensitive to salt concentration in buffer, so BSA adsorbed on Streamline DEAE can be eluted easily by increasing the salt concentration to 0.5 M in 20 mM phosphate buffer, pH 7.5. BSA recovery in the whole expanded bed adsorption process can reach 91%. From the BSA efuent concentration curve during the elution stage, it is observed that BSA can be eluted almost completely from Streamline DEAE; however, as a result of desorption during the washing stage, BSA recovery is not up to 100%a small amount of the elution buffer is consumed, as shown in Figure 10. Streamline direct CST I has a multimodal ligand that is not sensitive to the salt concentration. Therefore, it is very difcult to elute BSA from Streamline CST I in the column only by increasing the salt concentration in 50 mM acetate buffer, pH 5. Therefore, to accomplish elution of adsorbed BSA proteins, both salt concentration and pH value in acetate buffer are increased. Here, when the elution buffer is 50 mM acetate buffer with 1 M NaCl at pH 7, BSA recovery is up to 87%. From the BSA efuent concentration curve during the elution stage, it can be noted that BSA cannot be eluted completely, and the amount of the elution buffer consumed is also greater, as shown in Figure 9.
AIChE Journal November 2005

Acknowledgments
We acknowledge nancial support from Fundac a o para a Cie ncia e Tecnologia (FCT Grant SFRH/BD/6762/2001). 10.

Notation
Bi c C C0 p d DL Dm Dp De Dax,S H0 H kd kf n Pe q q qm r R Re Sc Sh t u Vc Z Biot number concentration in particle pore, kg/m3 concentration in uid, kg/m3 inlet concentration in uid, kg/m3 average diameter of adsorbent, m liquid axial dispersion coefcient, m2/s molecular diffusion coefcient, m2/s intraparticle pore diffusivity, m2/s intraparticle effective pore diffusivity, m2/s solid axial dispersion coefcient, m2/s settled bed height, m expanded bed height, m dissociation constant dened by Eq. 4, m3/kg lm mass transfer coefcient, m/s amount of sample injected, kg Peclet number adsorbed concentration in adsorbent, kg/m3 particle averaged adsorbent phase concentration, kg/m3 maximum adsorbed concentration, dened by Eq. 4, kg/m3 particle radial distance from center of particle, m radius of adsorbent, m Reynolds number Schmidt number Sherwood number time, s supercial liquid ow velocity, m/s column volume, m3 axial distance from column entrance, m 11.

12.

13.

14. 15.

16.

17.

18.

19.

20.

Greek letters
0 B p settled bed voidage, m3/m3 bed voidage of expanded bed, m3/m3 adsorbent porosity, m3/m3 liquid viscosity, Pas liquid density, kg/m3 21. 22.

23.

Literature Cited
1. Chase HA. Purication of proteins by adsorption chromatography in expanded beds. Trends Biotechnol. 1994;12:296-303. 2. Hjorth R. Expanded bed adsorption in industrial bioprocessing: Recent developments. Trends Biotechnol. 1997;15:230-235. 3. Thommes J, Bader A, Halfar M, Karau A, Kula MR. Isolation of monoclonal antibodies from cell containing hybridoma broth using a protein A coated adsorbent in expanded beds. J Chromatogr. 1996; 752:111-122. 4. Ujam LB, Clemmitt RH, Clarke SA, Brooks RA, Rushton N, Chase HA. Isolation of monocytes from human peripheral blood using immuno-afnity expanded-bed adsorption. Biotechnol Bioeng. 2003;83: 554-566. 5. Clemmitt RH, Chase HA. Direct recovery of glutathione S-transferase by expanded bed adsorption: Anion exchange as an alternative to metal afnity fusions. Biotechnol Bioeng. 2002;77:776-785. 6. Smith MP, Bulmer MA, Hjorth R, Titchener-Hooker NJ. Hydrophobic interaction ligand selection and scale-up of an expanded bed separation of an intracellular enzyme from Saccharomyces cerevisiae. J Chromatogr A. 2002;968:121-128. 7. Bai Y, Glatz CE. Capture of a recombinant protein from unclaried canola extract using Streamline expanded bed anion exchange. Biotechnol Bioeng. 2003;81:855-864. 8. Anspach FB, Curbelo D, Hartmann R, Garke G, Deckwer WD. Expanded-bed chromatography in primary protein purication. J Chromatogr A. 1999;865:129-144. 9. Jahanshahi M, Sun Y, Santos E, Pacek A, Franco TT, Nienow A, Lyddiatt A. Operational intensication by direct product sequestration 24.

25.

26.

27.

28. 29.

30. 31. 32. 33.

from cell disruptatesApplication of a pellicular adsorbent in a mechanically integrated disruption-uidised bed adsorption process. Biotechnol Bioeng. 2002;80:201-212. Chang YK, Chase HA. Development of operating conditions for protein purication using expanded bed techniques: The effect of the degree of bed expansion on adsorption performance. Biotechnol Bioeng. 1996;49:512526. Bruce LJ, Chase HA. Hydrodynamics and adsorption behaviour within an expanded bed adsorption column studied using in-bed sampling. Chem Eng Sci. 2001;56:3149-3162. Bruce LJ, Chase HA. The combined use of in-bed monitoring and an adsorption model to anticipate breakthrough during expanded bed adsorption. Chem Eng Sci. 2002;57:3085-3093. Karau A, Benken C, Thommes J, Kula MR. The inuence of particle size distribution and operating conditions on the adsorption performance in uidized beds. Biotechnol Bioeng. 1997;55:54-64. Wright PR, Glasser BJ. Modeling mass transfer and hydrodynamics in uidized-bed adsorption of proteins. AIChE J. 2001;47:474-488. Willoughby N, Habib G, Hoare M, Hjorth R, Titchener-Hooker NJ. The use of rapid on-line monitoring of products and contaminants from within an expanded bed to control separations exhibiting fast breakthrough characteristics and to maximize productivity. Biotechnol Bioeng. 2000;70:254-261. Willoughby NA, Hjorth R, Titchener-Hooker NJ. Experimental measurement of particle size distribution and voidage in an expanded bed adsorption system. Biotechnol Bioeng. 2000;69:648-653. Tong XD, Xue B, Sun Y. Modeling of expanded-bed protein adsorption by taking into account the axial particle size distribution. Biochem Eng J. 2003;16:265-272. Yun JX, Yao SJ, Lin DQ, Lu MH, Zhao WT. Modeling axial distributions of adsorbent particle size and local voidage in expanded bed. Chem Eng Sci. 2004;59:449-457. Tong XD, Dong XY, Sun Y. Lysozyme adsorption and purication by expanded bed chromatography with a small-sized dense adsorbent. Biochem Eng J. 2002;12:117-124. Chen WD, Dong XY, Sun Y. Modeling of the whole expanded-bed protein adsorption process with yeast cell suspensions as feedstock. J Chromatogr A. 2003;1012:1-10. Li P, Xiu GH, Rodrigues AE. A 3-zone model for protein adsorption kinetics in expanded beds. Chem Eng Sci. 2004;59:3837-3847. Kaczmarski K, Bellot JC. Theoretical investigation of axial and local particle size distribution on expanded bed adsorption process. Biotechnol Process. 2004;20:786-792. Palsson E, Axelsson A, Larsson PO. Theories of chromatographic efciency applied to expanded beds. J Chromatogr A. 2001;912:235248. Thommes J, Weiher M, Karau A, Kula MR. Hydrodynamics and performance in uidized-bed adsorption. Biotechnol Bioeng. 1995;48: 367-374. Fenneteau F, Aomari H, Chahal P, Legros R. Modeling of scale-down effects on the hydrodynamics of expanded bed adsorption columns. Biotechnol Bioeng. 2003;81:790-799. Fernandez-Lahore HM, Kleef R, Kula MR, Thommes J. The inuence of complex biological feedstock on the uidization and bed stability in expanded bed adsorption. Biotechnol Bioeng. 1999;64:484-496. Villermaux J, van Swaaij WPM. Modele representativ de la distribution des temps de sejour dans un reacteur semi-inni a dispersion axiale avec zones stagnantes. Application a lecoulement ruisselant dans des colonnes danneaux Raschig. Chem Eng Sci. 1969;24:10971111. McCabe WL, Smith JC, Harriott P. Unit Operations of Chemical Engineering. 4th Edition. New York, NY: McGraw-Hill; 1985. Van Der Meer AP, Blanchard CMRJP, Wesselingh JA. Mixing of particles in liquid uidized beds. Chem Eng Res Des. 1984;62:214222. Wilson EJ, Geankopl CJ. Liquid mass transfer at very low Reynolds numbers in packed beds. Ind Eng Chem Fundam. 1966;5:9-14. Boyer PM, Hsu JT. Experimental studies of restricted protein diffusion in an agarose matrix. AIChE J. 1992;38:259-272. Li P, Xiu GH, Rodrigues AE. Modeling separation of proteins by inert core adsorbent in a batch adsorber. Chem Eng Sci. 2003;58:3361-3371. Airs R. Notes on the diffusion-type model for longitudinal mixing in ow. Chem Eng Sci. 1959;9:266-267.

2976

November 2005

Vol. 51, No. 11

AIChE Journal

34. Levenspiel O. Chemical Reaction Engineering. 3rd Edition. New York, NY: Wiley; 1999. 35. Weber TW, Chakravorti RK. Pore and solid diffusion models for xed-bed adsorbers. AIChE J. 1974;20:228-238. 36. Li P, Xiu GH, Rodrigues AE. Analytical breakthrough curves for inert core adsorbent with sorption kinetics. AIChE J. 2003;49:2974-2979. 37. Xiu GH, Nitta T, Li P, Jin G. Breakthrough curves for xed-bed adsorbers: Quasi-lognormal distribution approximation. AIChE J. 1997;43:979-985. 38. Rasmuson A, Neretnieks I. Exact solution of a model for diffusion in particles and longitudinal dispersion in packed beds. AIChE J. 1980; 26:686-690.

Based on published works,36-38 the analytical solution for the RTD curve at dirac input mode, is y 1

exp

Pe 2

cos where a Pe

a2 b 2 a 2

a2 b 2 a d (A9) 2

Appendix: Analytical Solution for the Residence Time Distribution (RTD) Curve
In the mathematical model, the liquid axial dispersion, intraparticle diffusion, and lm mass transfer resistance are all taken into account. The material balance equation for uid phase is DL B 3 u C C 1 2C c PD p 2 Z B Z t B r R

2 2 Bi2 I1 Bi(I1 I2 ) Pe 3 2 2 4 (Bi I1 ) I2

(A10)

b Pe 3

Bi2 I2 2 (Bi I1 )2 I2

(A11)

rR

0 (A1)

I1

The material balance equation in pore particle is p c 2c 2 c pD p t r2 r r

I2 (A2)

Z H

sinh 2 sin 2 1 2 cosh 2 cos 2 sinh 2 sin 2 2 cosh 2 cos 2

(A12)

Bi

(A13)

0 r R

The dimensionless variables are y C C 0 c

The initial and boundary conditions for Eqs. A1 and A2 are CZ, 0 0 cr, Z, 0 0 C 0, t C 0t c t C , t with dirac input mode is limited 0
r 0

1 B p B Dp t 2 R

kf R p Dp

(A3) (A4) (A5) (A6) (A7)

BHD p 2 uR

Pe

uH B DL

c r PD p

c r

rR

k f C c rR

(A8)

where the reference concentration is the ratio between the amount injected and the uid volume in the column, C0 2 (tc n/ B V c and c Dptc/R B H /u is the space time). The tracer (acetone) intraparticle pore diffusivity is estimated approximately as Dp 1.28 109 m2/s; the lm mass transfer coefcient kf is estimated by the correlation Eq. 9; and the particle porosity for acetone is estimated by rst moment 1, equal to the mean resistance time of the experimental RTD curve, as suggested by Boyer and Hsu.31
Manuscript received Dec. 13, 2004, and revision received Feb. 22, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2977

The Fiber-Coating Model of Biopharmaceutical Depth Filtration

Glen R. Bolton, Daniel LaCasse, Matthew J. Lazzara, and Ralf Kuriyel
Millipore Corporation, Bedford, MA 01730 DOI 10.1002/aic.10541 Published online August 3, 2005 in Wiley InterScience (www.interscience.wiley.com).

Depth lters are often used in biotechnology processes to remove cell debris from cell culture solutions. A model of lter plugging was developed to allow predictions of required lter area from limited amounts of initial data. The lter microstructure was idealized as an assembly of randomly oriented, straight, cylindrical bers. It was assumed that lters plug as solids coat the surfaces of bers, making them thicker and reducing lter porosity and permeability. The CarmanKozeny equation was used to allow calculation of lter permeability as a function of lter ber radius, lter solid fraction, and volume ltered. An explicit equation for ltrate volume as a function of time during constant pressure operation was derived, and an explicit equation for pressure as a function of time during constant ow rate operation was derived. A second model that accounted for the combined effects of ber coating and surface caking was also derived. The models were tested on data from a replica cell culture uid ltered through a glass ber based depth lter and data from E. coli lysate ltered through glass ber and diatomaceous earth based depth lters. The ber coating model provided good ts of the volume vs. time data and good predictions of lter capacity from limited initial data. The combined cake-ber coating model provided improved performance over the ber coating model. The models are useful tools for sizing of depth lters for fermentation and cell culture applications and may be useful for other liquid and gas lter applications. 2005
American Institute of Chemical Engineers AIChE J, 51: 2978 2987, 2005

Keywords: bioseparations, membrane separations, depth lter, fouling, sizing

Introduction
A wide range of processes utilize lters to remove particles from aqueous solutions. Depth lters are often used in biotechnology processes to remove cell debris and other particulates. The lter area required by these applications can either be determined by lter permeability or lter capacity. Filter capacity is an indication of how much volume can be ltered before plugging occurs. Capacity during constant pressure operation is dened as the amount of uid per lter area that can be processed until the ow rate declines to a set fraction of the initial ow. Capacity under xed ow rate conditions is deterCorrespondence concerning this article should be addressed to G. Bolton at Glen_Bolton@millipore.com.

2005 American Institute of Chemical Engineers

mined when the pressure increases to some set multiple of the initial value. Fouling of a lter can occur by deposition of particles inside or on top of the lter.1,2 There are four mechanistic models that are typically used to describe fouling. Complete blocking assumes that particles seal off pore entrances and prevent ow. Intermediate blocking is similar to complete blocking but assumes that a portion of the particles seal off pores and the rest accumulate on top of other deposited particles. Cake ltration occurs when particles accumulate on the surface of a lter in a permeable cake of increasing thickness that adds resistance to ow. Standard blocking assumes that particles accumulate inside lters on the walls of straight cylindrical pores. As particles are deposited, the pores become constricted and the permeability of the lter is reduced. This model was developed by Hermans and Bredee3 and has typically been applied to the fouling of microporous lters.4
Vol. 51, No. 11 AIChE Journal

2978

November 2005

Figure 1. Illustration of a lter composed of randomly oriented bers before (left) and after (right) plugging.
The lters plug as solids coat the surfaces of bers, increasing ber thickness and reducing lter permeability.

The physical assumptions in the classical blocking models are not representative of most depth lters and are therefore not widely used. The straight cylindrical pore microstructure assumed in the classical models does not represent the brous structure of most depth lters. Cake can form above depth lters but the pore blockage and pore constriction models do not accurately describe the mechanisms of depth lter fouling. Depth lters are typically fouled by various mechanisms of deposition of particles on the internal surfaces of the media.5 In this work an alternative model that treats lters as random ber matrices was developed. The model predicts a decrease in lter permeability as solids from the process stream deposit on the lter bers. This deposition increases the effective radius of the bers and decreases the lter void fraction. A combined model that accounts for cake formation and ber coating occurring simultaneously was also developed. The models provided good ts of volume vs. time data during ltration and made accurate predictions of lter capacities from limited initial data.

L F r f2 r f20

c AV 1

(2)

where rf is the ber radius, LF is the total length of all bers, c is the volume of solids per unit of ltrate volume, is the porosity of solid when deposited on the bers, and V is the volume processed, normalized by lter area A. The lter solid fraction is related to the ber radius, lter thickness LM, and other parameters.

r f20L F 0 L MA

(3)

The subscript 0 is used to indicate an initial value of a given quantity. After fouling has occurred, the new solid fraction can be related to the new ber radius by

Theory Fiber coating model

A model of internal lter plugging that treats lters as assemblies of straight cylindrical bers was developed. It assumes that lters plug as solids coat the surfaces of bers, making them thicker and reducing lter permeability. This is illustrated in Figure 1. The model was developed by rst relating the ber radius to the volume of solids removed in a manner similar to that of the standard blocking model1 L F2 r fdr f
AIChE Journal

r f2L F rf 0 L MA rf 0

(4)

Although the deposited solids are assumed to be porous in Eq. 2, Eq. 4 assumes the solids are nonporous. This assumption will be accurate if the ow through the deposited solids is minimal. The term Vmax describes the amount of solution volume that can be ltered until the lter void volume is completely lled with solids. A new term, the ber coating constant Kf, is equal to the inverse of Vmax. By material balance of solids in the lter volume, Vmax is calculated as V max
Vol. 51, No. 11

c AdV 1

(1)
November 2005

1 1 LM 1 0 Kf c

(5)
2979

Rearranging Eq. 3 to an expression for LF and substituting into Eq. 2 yields Eq. 6, which allows calculation of rf and from V and four physical constants: LM, c, , and 0. Insertion of Eq. 5 reduces this to two physical constants, Kf and 0. r f2 (1 0) c V 1 1 K fV 1 L M 0 0 r f20 0

(6)

To determine the ltrate volume as a function of percentage ux decline, such as the volume after the ux has declined 75% from its initial value, it is necessary to solve for the roots of Eq. 9. The real root of this cubic polynomial can be calculated using Cardans formula8 V 1 3 3 1 Kf (12)

Combining Eq. 6 with a theoretical model that relates hydraulic permeability to ber radius allows a lter-plugging model to be derived that relates permeability to ltrate volume. The CarmanKozeny model,

where , , and are placeholders dened, respectively, as

1 0 J 30 J0 1 J 20 J0

1 3 20 2 r f2

(7)

where is lter Darcy permeability, was developed by extending a correlation between Reynolds number and friction factor for laminar ow through pipes to ow through porous media.6 It has also been used to describe ow through brous media.7 The ow rate through a lter (dV/dt) can be calculated from Darcys Law as dV P dt LM (8)

3 2

Fiber coating model: constant ow

For ltration experiments run at a constant ux, Eq. 9 can be rearranged to obtain an explicit expression for pressure as a function of time, where V J0t

where P is the transmembrane pressure and is the solution viscosity. In this study it was found that the CarmanKozeny model was ideal for use in the ber coating model of lter plugging. As described in more detail in Appendix A, the Carman Kozeny model accurately represented both rigorous theoretical predictions and a large amount of experimental data over the relevant range of ber volume fractions.

P P0

(1 0) K fV 0 3 1 K fV

(13)

This model is similar to a model of increasing pressure drop with deposition of solids on packed spheres.5 To determine the ltrate volume as a function of relative pressure increase, such as the volume after the pressure has increased tenfold from the initial value, Eq. 12 can be used, where the parameters and will now be dened, respectively, as

Fiber coating model: constant pressure

To obtain the equations for the ber coating model, Eq. 6 can be substituted into Eqs. 7 and 8 to yield ux as a function volume for ltration experiments run at constant pressure dV P r f2 1 3 dt LM 20 2

1 0 P 0 30 P 1 P0 20 P

J 0 1 K fV 3 (1 0) 1 K fV 0

(9)

where J0 is the initial solvent ux. The equation can be integrated to obtain an explicit equation for ltrate volume as a function of time V 1 1 1 K f K f

(1 0)

(10)

where is a dimensionless parameter, dened as

2 K fJ 0t 1
2980

0 1 1 0

(11)
November 2005

Equations 10 and 13 were derived assuming that bers never overlapped and all solids deposited evenly along the bers. It is possible, however, to have bers that intersect or have a signicant portion of their surface area in contact with one another. In this case the ber radius will increase with permeate volume faster than that described by Eq. 6 because the solids will deposit on ber surfaces only where there is no contact or overlap. Equations 10 and 13 are corrected for ber overlap in Appendix B. In practice lters typically will have characteristics that are between the cases of overlapping and nonoverlapping bers. For the glass bers used in this study there will be ber contacting and but no overlap initially. Once solids are deposited, however, the coated bers will overlap at the points where the bers were initially in contact. This is illustrated in Figure 1. The sizing predictions in this study were not significantly changed by accounting for the effects of ber overlap.
Vol. 51, No. 11 AIChE Journal

Combined cake-ber coating model

Typically foulants accumulate throughout the thickness of depth lters. Large foulants such as intact cells, however, can accumulate as cakes above lters. To account for the effects of these cakes, the ber coating model was modied by adding the resistance caused by these cakes. The cake resistance R can be calculated as a function of permeate volume R 1 K cJ 0 V R0 (14)

To determine the ltrate volume as a function of relative pressure increase, such as the volume after the pressure has increased tenfold from the initial value, it is most convenient to solve Eq. 19 numerically.

Experimental Glass-ber lter

Filters were composed of glass bers with a binder resin (AP1504700, Millipore, Bedford, MA). Fiber radii were determined from scanning electron micrographs and lter permeabilities. Samples were sputter coated using a Au/Pd target before SEM analysis. Images were acquired at a 1000 magnication using a Topcon DS-130-C scanning electron microscope. The bers observed in scanning electron micrographs had radii ranging from 0.2 to 5.0 m. The lter solid fraction was determined to be 0.07 0.01 by measuring the volume displacement of the media as follows. Three 47-mm media disks were rolled up and submerged in a 20-mL graduated cylinder containing 18 mL of water. The cylinder was then put in a 1 gallon sonic cleaner (FS30H, Fisher Scientic, Hampton, NH) for 10 min to remove air bubbles and the volume displacement was measured. This was done in triplicate. The media thickness was determined to be 0.54 mm using calipers. The average hydraulic permeability of the media was 9400 L m2 h1psi. The effective ber radius was determined by tting the initial permeability of the lter using Eq. 16. The effective ber radius was 0.17 microns, which was at the low end of the range observed in lter SEMs.

where Kc is the cake ltration constant and R0 is the initial lter resistance.2 For the combined model, the lter resistance increases with permeate volume as the bers become coated with foulants and, simultaneously, the cake thickness increases with permeate volume. This is similar to recently developed models of the combined effects of the classical fouling modes.9 Higher Kf or Kc values physically correspond to a larger reduction in lter permeability for a given permeate volume. The ux can be calculated as a function of permeate volume using the following equation, where the total resistance is the sum of the resistances of the brous lter and the cake dV dt

J0 (1 0) 1 K fV (1 K fV ) 3 K cJ 0V 0

(15)

Equation 17 can be integrated to calculate time as a function of volume t 1 K cV 2 2 2Kf J0

E. coli lysate
Plugging experiments were performed with a solution of lysed Escherichia coli BL21 cells obtained from Dr. Carl Lawton of the University of Massachusetts, Lowell. The lyaste was obtained frozen at pH 7.0 in a mixture of 50 mM sodium acetate with 2 mM EDTA lysis buffer and PBS media buffer.

(1 0) ( K fV ) 2 0

(1 K fV ) 2 1

(16)

Equations 17 and 18 reduce to the equations for cake ltration when Kf is small and to the equations for ber coating when Kc is small. It is difcult to rearrange Eq. 18, a fourthorder polynomial in volume, to calculate volume as a function of time, although Eq. 18 is still useful for tting experimental data and predicting lter capacities. Typically the experimental values of time are used as inputs to plugging models to calculate volumes. The predicted volumes are then compared to the experimental values to determine the best-t parameters. The best-t parameters in Eq. 18 can be determined by using the experimental volume values as inputs and calculating the corresponding time values. The predicted times can then be compared to the experimental values and the parameters Kc and Kf can de determined. For the case where the ltration experiment is run at a constant ux, Eq. 17 can be rearranged to obtain Eq. 19, which allows explicit calculation of pressure as a function of time, where V J0t

Replica cell culture uid

A uid representing a claried solution from a mammalian cell culture vessel was made up of the following components: 13.47 g/L DMEM (D-7777, Sigma, St. Louis, MO), 3.70 g/L NaHCO3 (7412, Mallinckrodt, Hazelwood, MO), 1 g/L Pluronic F-68 (P1300, Sigma), 0.0775 g/L DNA(D6898, Sigma), 5.0 g/L soy peptone fraction IV (P0521, Sigma), 1.5 g/L dairy whey (W1500, Sigma), and 3.00 mg phosphatidylcholine (P3644, Sigma). The feed solution was used within 2 h after preparation to minimize variability.

Filter testing equipment: constant pressure

The plugging solution was contained in a pressure vessel (XX1100000, Millipore). The lter was placed in a 47-mm stainless steel holder (XX4404700, Millipore). An electronic balance aperture and a collection vessel were used to collect and measure the ltrate weight. The electronic balance consisted of a load cell (1042-3-I, Tedea Huntleigh, Covina, CA) that interfaced with a computer through a data-acquisition board. During constant pressure experiments, the system was pressurized with air and a digital pressure gauge was used
Vol. 51, No. 11 2981

P P0
AIChE Journal

(1 0) K fV 0 K cJ 0V 1 K fV 3

(17)

November 2005

Figure 2. Volume vs. time data compared to the ber coating and cake-ber coating model predictions with 10 diluted E. coli lysate run at 3 psi through AP15 glass-ber media.

Figure 3. Volume vs. time data compared to the ber coating and cake-ber coating model predictions with 100 diluted E. coli lysate run at 3 psi through AP15 glass-ber media.

(2798K12, McMaster-Carr, Atlanta, GA). The 47-mm lter disks were wetted with water and placed in holders. The holders were then connected to 600-mL feed reservoirs. The feed solutions were added to their appropriate chambers and the four chambers were pressurized to 3 psi. The lter holders were vented to remove air. The ltrate valves were then opened and data acquisition began. The lter surface area was 1.38 103 m2. Data were collected at 1-s intervals for 1 h. During constant ow experiments, a three-head peristaltic feed pump (7523-50, 77201-60, Cole Parmer, Vernon Hills, IL) was used with a pressure gauge (XXPXLGAGE, Millipore). The load cell, data acquisition, and lter holder were the same as those used in the constant pressure experiments.

does benet from the use of two tted parameters. It is possible, however, that more cake formed above the lters used with the more concentrated streams, as would be expected, resulting in the improved t of the combined model. The Kf term was near zero, indicating that the cake model alone provided the best t for the 10 diluted stream. Table 1 summarizes the residuals, an indication of quality of t, and the tted parameters from the two models for the different streams.

Constant ow rate: replica cell culture uid on glass ber

The applicability of the models to data from a constant ow experiment performed with the replica cell culture uid was tested. The uid was ltered at a constant ow rate of 2600 L m2 h1 through the AP15 glass media and the pressure was measured. The experiment was performed with the replica cell culture uid diluted four- and ten fold. The experimental ap-

Results Constant pressure: E. coli lysate on glass ber

The E. coli lysate solution was diluted 10-, 100-, and 1000fold with pH 7.2 PBS buffer (DF2314-15-0, Fisher) and ltered at a constant pressure through the glass lters. The permeate volume was measured as a function of time until the ux had decreased by 95% from its initial value. The permeate volume and time data were t using the ber coating model, Eq. 10, by varying Kf, and using the cake-ber coating model, Eq. 18, by varying Kf and Kc. The best t was determined by minimizing the sum of squared residuals (SSR), where the residual was equal to the difference between an experimental data point and a model prediction. The residual was calculated from volume values for the ber coating model, and from time values for the cake-ber coating model. The volume vs. time data are compared to model predictions in Figures 2, 3, and 4 for the different lysate dilutions. The ber coating model provided good ts of the E. coli lysate data. The solid volume fraction, determined from Eq. 12, increased from the initial value of 0.07 to 0.54 and 0.39 for the streams diluted 10- and 1000-fold, respectively. The cake-ber coating model provided better ts than the ber coating model alone. The difference between the t of the two models increased with decreasing dilution factor. The combined model
2982 November 2005

Figure 4. Volume vs. time data compared to the ber coating and cake-ber coating model predictions with 1000 diluted E. coli lysate run at 3 psi through AP15 glass-ber media. Vol. 51, No. 11 AIChE Journal

Table 1. Error of Fit and Model Parameters for the Fiber Coating and Cake-Fiber Coating Models*
Model Fiber coating model (Eq. 10) Cake-ber coating model (Eq. 18) Fiber coating model (Eq. 10) Cake-ber coating model (Eq. 18) Fiber coating model (Eq. 10) Cake-ber coating model (Eq. 18) Dilution Factor 10 10 100 100 1000 1000 Initial Flux, J 0 (m/s) 7.83 103 7.83 103 7.83 103 7.83 103 7.83 103 7.83 103 Model Fit Error, SSR 4.37 103 2.63 101 2.09 104 5.46 102 7.80 105 3.24 104 Fit Parameter Values Kf Kf Kc Kf Kf Kc Kf Kf Kc 1.25 1.00 7.78 1.39 9.36 3.01 9.87 8.16 1.02 10 1 10 8 10 4 10 1 10 3 10 2 10 2 10 2

*Data for 10,100 and 1000 diluted E. coli lysate run at 3 psi through AP15 glass-ber media.

paratus, lter area, and lter thickness were the same as described previously. The pressure vs. time data were t to the ber coating model, Eq. 13, and the cake-ber coating model, Eq. 19, by minimizing the SSR values. The initial pressure drop was too low to measure accurately. It was estimated to be 0.3 psi by dividing the ux by the lter permeability. The data and model predictions are shown in Figures 5 and 6. Table 2 summarizes the residuals and tted parameters from the two models. The ber coating model provided good ts of the pressure increase observed in the experiments for both the 4 and 10 diluted replica cell culture uid stream. The ts were improved by using the combined cake-ber coating model, although this model does benet from the use of two tted parameters. It is possible that in these experiments, similar to the observation in the E. coli lysate experiments, more cake formed above the lter used with the more concentrated stream, resulting in the improved t of the combined model.

Constant ow rate: E. coli lysate on Millistak

The applicability of the models to constant ow data for the ltration of 10 diluted E. coli lysate on commercially available brous depth lters was tested. The uid was ltered at a constant ow rate of 364 L m2 h1 through a Millistak A1HC (MA1HC23HH3, Millipore) disposable depth lter device and the pressure was measured. The lter consisted of about 3 mm of an open grade of cellulose media embedded with diatomaceous earth, followed by 3 mm of a tighter grade, followed by a thin layer of microporous membrane. The brous, cellulosebased structure of this media makes it ideal for application of the ber coating model. The solid fraction of the rst layer was determined to be 27%, and the second layer 23%, from measurements of volume displacement using water and a volumetric ask. The solid fraction of the overall lter was taken as 25%, the thickness-weighted average of the two measurements. The pressure vs. time data were t to the ber coating model, Eq. 13, by minimizing the SSR values. The best t of the

cake-ber coating model occurred with Kc near zero and was identical to the t of the ber coating model alone. The ber coating model provided a good t of the pressure increase observed in the experiment, as shown in Figure 7. The ability of the ber coating and cake-ber coating models to accurately predict lter sizing was evaluated by tting a limited amount of initial pressure data and then comparing the predicted volume corresponding to a set pressure limit to the experimental data. This is representative of scale-down studies used to estimate area requirements for large-scale lter processes. Generally only limited time and process uid are available. Extrapolating data from partially plugged lters to higher plugging values using a plugging model can save time, process uid, and allow more experiments to be performed. Parameters were t to the initial data and the volume that would be obtained when the pressure reached the nal pressure was calculated. The predictions are compared to the actual volumes at the respective nal pressures in Figure 8 for the ber coating model (Eq. 13) and Figure 9 for the cake-ber coating model (Eq. 19). The ber coating model predictions in Figure 8 were within 20% of the experimental data when more than 15% of the full pressure range was used for parameter tting. This was true for both dilutions of the replica cell culture uid run on the glass lter and the E. coli lysate run on the Millistak media. This indicates the ber coating model is a useful tool for making accurate lter sizing predictions with heterogeneous depth ltration media based on limited amounts of pressure vs. volume data. Occasionally, brous clarication media will be limited by breakthrough of impurities and not by pressure increase as a result of fouling. In this case other criteria such as permeate turbidity will be more important for sizing. The combined cake-ber coating model provided some improvement in the accuracy of the capacity predictions. The cake-ber coating model predictions in Figure 9 were up to 16% closer to the actual values than the ber coating model for the 10 diluted replica cell culture uid, and up to 4% better

Table 2. Error of Fit and Model Parameters for the Fiber Coating and Cake-Fiber Coating Model*
Model Fiber coating model (Eq. 13) Cake-ber coating model (Eq. 19) Fiber coating model (Eq. 13) Cake-ber coating model (Eq. 19) Dilution Factor 4 4 10 10 Initial Flux, J 0 (m/s) 7.25 104 7.25 104 7.25 104 7.25 104 Model Fit Error, SSR 41.6 10.7 24.1 3.43 Fit Parameter Values Kf Kf Kc Kf Kf Kc 2.92 2.76 3.75 10 4 1.72 1.64 1.49 10 4

*Data for 4 and 10 diluted replica cell culture uid run at 2600 L m2 h1 through AP15 glass-ber media.

AIChE Journal

November 2005

Vol. 51, No. 11

2983

Figure 5. Pressure vs. time data compared to the ber coating and cake-ber coating model predictions with 4 diluted replica cell culture uid run at 2600 L m2 h1 through AP15 glassber media.

Figure 7. Pressure vs. time data compared to ber coating model predictions for experiment run at 364 L m2 h1 with 10 diluted E. coli lysate on Millistak A1HC depth lter media.

for the 4 diluted replica cell culture uid. The combined model did not provide better ts of the E. coli lysate data and was not used for capacity predictions. It is likely that the cake-ber coating model will be more accurate than the ber coating model for sizing lters in processes where large foulants that accumulate above lters are present.

Conclusions
A model of internal lter plugging that assumed lters were composed of straight cylindrical bers was developed. The model relates the volume ltered to the lter solid fraction and ber thickness by assuming that all plugging solids coated the bers and increased the lter solid fraction. The Carman Kozeny equation was used with Darcys Law to calculate ux as a function of volume ltered. The ux equation was inte-

grated to derive an explicit equation for ltrate volume as a function of time when operating at constant pressure. An explicit equation for pressure as a function of time when operating at a constant ow rate was also derived. A second model that accounts for the combined effects of ber coating and cake formation was also developed. The models were tested by ltering different dilutions of an E. coli lysate through lters composed of glass bers at constant pressure. The ber coating model provided good ts of data from constant pressure experiments run with E. coli lysate on glass lters. The ber coating model made accurate lter capacity predictions using limited pressure vs. time data for constant ow fouling of glass lters by the replica cell culture uid and commercially available depth ltration media by E. coli lysate. The cake-ber coating model provided improved

Figure 6. Pressure vs. time data compared the ber coating and cake-ber coating model predictions with 10 diluted replica cell culture uid run at 2600 L m2 h1 through AP15 glassber media. 2984 November 2005

Figure 8. Accuracy of capacity predictions from the ber coating model.

Predicted nal volume normalized by actual nal volume as a function of percentage of pressure range used to determine parameter values. Data for the 4 and 10 diluted replica cell culture uid run at 2600 L m2 h1 through AP15 glass-ber media, and 10 diluted E. coli lysate on Millistak A1HC depth lter media run at 364 L m2 h1.

Vol. 51, No. 11

AIChE Journal

Literature Cited
1. Grace HP. Structure and performance of lter media. AIChE J. 1956; 2:307-336. 2. Hermia J. Constant pressure blocking ltration lawsApplication to power-law non-Newtonian uids. Trans IChemE. 1982;60:183-187. 3. Hermans PH, Bredee HL. Principles of the mathematical treatment of constant-pressure ltration. J Soc Chem Ind. 1936;55T:1-4. 4. Badmington F, Wilkins R, Payne M, Honig ES. Vmax testing for practical microltration train scale-up in biopharmaceutical processing. Biopharm. 1995;Sep.:46-52. 5. Tien C. Granular Filtration of Aerosols and Hydrosols. Stoneham, MA: Butterworth; 1989. 6. Carman PC. Fluid ow through granular beds. Trans IChemE. 1937; 15:150-166. 7. Sullivan RR, Hertel KL. The ow of air through porous media. J Appl Phys. 1940;11:761-765. 8. Nickalls RWD. A new approach to solving the cubic: Cardans solution revealed. Math Gazette. 1993;77:354-359. 9. Bolton GR, LaCasse D, Kuriyel R. Combined models of membrane fouling: Development and application to microporous and ultraporous membranes. J Membr Sci. 2005; in press. 10. Ogston AG. The spaces in a uniform random suspension of bers. Trans Faraday Soc. 1958;54:1754-1757. 11. Bolton GR, Deen WM. Limitations in the application of ber-matrix models to glomerular basement membrane. In: Layton HE, Weinstein AM, eds. Membrane Transport and Renal Physiology (IMA Series: Mathematics and Its Application). New York, NY: Springer-Verlag; 2001:141-156. 12. Jackson GW, James DF. The permeability of brous porous media. Can J Chem Eng. 1986;64:364-374. 13. Higdon JJL, Ford GD. Permeability of three-dimensional models of brous porous media. J Fluid Mech. 1996;308:341-361. 14. Clague DS, Kandhai BD, Zhang R, Sloot PMA. Hydraulic permeability of (un)bounded brous media using the lattice Boltzmann method. Phys Rev E. 2000;61:616-625.

Figure 9. Accuracy of capacity predictions from the cake-ber coating model.

Predicted nal volume normalized by actual nal volume as a function of percentage of pressure range used to determine parameter values. Data for the 4 and 10 diluted replica cell culture uid run at 2600 L m2 h1 through AP15 glass ber media.

volume vs. time ts and improved sizing predictions. The improvement was greatest for concentrated streams where it is possible cakes formed above the lters. The ber coating model and cake-ber coating models will be useful tools for the sizing of depth ltration media used in biotechnology and potentially for other types of brous lters used with both liquids and gases.

Appendix A
There are a number of theoretical models that relate lter permeability to ber radius and ber volume fraction.11 Jackson and James12 modeled the hydraulic permeability of a random brous lter using a cubic array of cylindrical rods. The resistance to ow was the sum of that arising from the rods aligned parallel and perpendicular to the direction of ow. Permeability data were reviewed for materials with a wide range of ber radii and ber volume fractions. The model t the data well for ber volume fractions 0.2. Higdon and Ford13 obtained numerical solutions to ow around three-dimensional brous media using the spectral boundary element method. They obtained results for a wide range of lter solid fractions using slender-body theory at low ber concentrations and the lubrication approximation at high ber concentrations. Their results agreed with the data reviewed in Jackson and James12 for ber volume fractions between 0.0309 and 0.865. Clague et al.14 used a LatticeBoltzman simulation method to make numerical calculations of the Darcy permeability of a disordered, three-dimensional array of cylinders. Their results (presented in their Figure 8) also t the data in Jackson and James12 and agreed with the face-centered cubic ber lattice predictions of Higdon and Ford13 for ber volume fractions between 0.05 and 0.7. The CarmanKozeny model,

Acknowledgments
The authors thank Dave Yavorsky, Carl Lawton, Ven Raman, Oleg Shinkazh, John Royce, and Matthew Tomasko for providing the lters, the replica cell culture uid, and the E. coli lysate, and Robin de la Parra for taking the SEM images.

Notation
A c Kc Kf LF LM P J rf t V Vmax total lter area, m2 volume of solids per unit of ltrate volume cake ltration constant, s/m2 ber coating constant, 1/m total ber length, m lter thickness, m pressure, kg/ms2 solvent ux, m/s ber radius, m time, s volume ltered, m3/m2 maximum volume ltered, m3/m2

Greek letters

dimensionless placeholder in Eq. 10 dimensionless placeholder in Eq. 12 dimensionless placeholder in Eq. 12 dimensionless placeholder in Eq. 12 porosity of deposited solids lter solid or ber volume fraction Darcy permeability, m2 solution viscosity, kg/ms

1 3 20 2 r f2
November 2005 Vol. 51, No. 11

(A1)
2985

AIChE Journal

one another. In this case the ber radius will increase with permeate volume faster than as described by Eq. 6 because the solids will deposit on ber surfaces only where there is no contact or overlap with another ber. The following derivation corrects the ber coating model for the effects of ber overlap both initially and as the bers grow uniformly in radius. The lter is assumed to be composed of bers that are randomly distributed with respect to one another. One consequence of the assumption of randomness is that bers in this hypothetical matrix will overlap with one another. Thus, the true solid volume fraction of the lter cannot be calculated directly using the number concentration of bers and the volume of an individual ber. For completely nonintersecting bers, the solid volume fraction varies as r 2 f and is related to 0 as
Figure A1. Hydraulic permeability as a function of ber volume fraction.
The permeabilities calculated with the CarmanKozeny model are compared to those calculated with the Jackson and James model12 and with the numerical calculations of Higdon and Ford13 and Clague et al.14

rf rf 0

(B1)

where is lter Darcy permeability, was developed by extending a correlation between Reynolds number and friction factor for ow through pipes to ow through porous media.5 It has also been used to describe ow through brous media.6 The ow through a lter can be calculated from Darcys Law: dV P dt LM (A2)

For any random-ber matrix with a nominal value of , the true solid volume fraction * can be calculated by considering the partitioning of a point-sized particle in that matrix. As a solute becomes innitely small, the fraction of the lter volume from which it is excluded is equal to the true solid volume fraction of the lter. The partition coefcient () for such a solute is given by 1 * (B2)

In this study the CarmanKozeny model predictions were compared to the numerical results of Higdon and Ford12 and Clague et al.13 The CarmanKozeny equation provided good agreement with the results from the other two studies. A comparison of the models is shown in Figure A1. The predictions of the CarmanKozeny were within 50% of the Higdon and Ford predictions for ber volume fractions between 0.1 and 0.8. This range is typical of the majority of brous media. For low ber volume fractions, between 0.05 and 0.1, the CarmanKozeny predictions were higher by less than a factor of two. In Figure A2 the predictions of the CarmanKozeny model are compared to the experimental data summarized by Jackson and James.11 The CarmanKozeny model t the majority of the experimental data. The predictions were within the scatter of the summarized data for volume fractions above about 0.05. At volume fractions 0.05 the model overpredicted the data. This indicates the CarmanKozeny model can be used as a simple mathematical ux equation that accurately represents both rigorous theoretical predictions and a large amount of experimental data.

To relate * to , we consider now the classical result of Ogston10 for the partitioning of rigid spherical solutes in random-ber matrices, wherein no specic accounting for ber intersection is made. The Ogston result relates to the solute radius rs, ber radius rf, and nominal ber solid volume fraction , as exp 1

rs rf
2

(B3)

Appendix B
Equations 10 and 13 were derived assuming that bers never overlapped and all solids deposited evenly along the bers. It is possible, however, to have lters with bers that intersect or have a signicant portion of their surface area in contact with
2986 November 2005 Figure A2. Hydraulic permeability as a function of ber volume fraction.
The permeabilities calculated with the CarmanKozeny model are compared to the experimental data summarized in the Jackson and James review.12

Vol. 51, No. 11

AIChE Journal

As rs 3 0, the Ogston result reduces to exp(). Equating Eqs. B2 and B3, 1 * exp (B4)

dV P r f2 1 * 3 ln1 * * 0 J0 2 dt LM 20 * ln1 * 0 *

1 * 1 * 0

(B9) An equation for ux as a function of volume can be obtained by inserting the equation relating true solid volume fraction to permeate volume, Eq. B7, into Eq. B9. This can be solved numerically for volume vs. time at different values of the parameter Vmax. The bers will be thicker for a given change in solid volume fraction when accounting for the ber overlap. Solids will deposit more on the ends of bers where there is no overlap than in centers where the surface area is unavailable. The ux increases with the square of ber radius in Eq. B9. For a single solid fraction value, the ux will be higher when the ber radius is larger. It will thus require a larger amount of solids to cause a similar ux decline when accounting for ber overlap. An equation similar to B9 can be derived for constant ow conditions, as follows: ln1 * P 0 * P 0 ln1 * * 0

In the brous lter the true solid volume fraction will increase as solids are deposited according to

c L M * * 0 1

(B5)

The maximum volume that can be ltered or Vmax will be described by

V max LM

1 1 * 0 c

(B6)

Inserting Eq. B6 into B5 yields V * 0 V max

1 * 0 1 *

(B10)

* 1 * 0

(B7)

The ber radius will increase faster with a given change in true solid volume fraction than an equivalent change in nominal solid volume fraction. The ber radius can be related to the true solid volume fraction by inserting Eq. B4 into Eq. 4. ln1 * rf 0 ln1 * r 0 f0

(B8)

An equation for ux as a function of true solid volume fraction can be obtained by inserting Eq. B8 for ber radius into the CarmanKozeny equation:

An explicit equation for calculation of pressure as a function of time or volume can be obtained by inserting the equation relating the true ber volume fraction to permeate volume, Eq. B7, into Eq. B10. The effect of the overlap on the predicted lter capacities in Figure 8 was determined. The predictions made using Eq. B10 were within 1% of the ber coating model for the 10 diluted replica cell culture uid run on glass lters and the E. coli lysate run on Millistak. The overlap model predictions were up to 4% better for the 4 diluted replica cell culture uid run on glass lters. The effect of overlap is not expected to signicantly alter the capacity predictions. In cases where it is important, Eq. B10 can be used to calculate lter capacities numerically.
Manuscript received Sep. 3, 2004, and revision received Mar. 7, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2987

Three-Bed PVSA Process for High-Purity O2 Generation from Ambient Air

Jeong-Geun Jee, Sang-Jin Lee, Min-Bae Kim, and Chang-Ha Lee
Dept. of Chemical Engineering, Yonsei University, Shinchon-dong, Seodaemun-gu, Seoul, 120-749, Korea DOI 10.1002/aic.10548 Published online August 4, 2005 in Wiley InterScience (www.interscience.wiley.com).

A three-bed PVSA (pressure vacuum swing adsorption) process, combining equilibrium separation with kinetic separation, was developed to overcome the 94% O2 purity restriction inherent to air separation in the adsorption process. To produce 97% and/or 99% purity O2 directly from air, the PVSA process with two zeolite 10X beds and one CMS bed was executed at 33.44 45.60 to 253.31 kPa. In addition, the efuent gas from the CMS bed to be used for O2 purication was backlled to the zeolite 10X bed to improve its purity, recovery, and productivity in bulk separation of the air. PVSA I, which made use of a single blowdown/backll step, produced an O2 product with a purity of 95.4 97.4% and a recovery of 43.4 84.8%, whereas PVSA II, which used two consecutive blowdown/backll steps, produced O2 with a purity of 98.299.2% and a recovery of 47.2 63.6%. Because the primary impurity in the O2 product was Ar, the amounts of N2 contained in the product were in the range of 4000 5000 ppm at PVSA I and several tens of ppm at PVSA II. A nonisothermal dynamic model incorporating mass, energy, and momentum balances was applied to predict the process dynamics. Using the linear driving force (LDF) model with constant diffusivity for the equilibrium separation bed and a modied LDF model with concentration dependency of the diffusion rate for the kinetic separation bed, the dynamic model was able to accurately predict the results of the experiment. 2005
American Institute of Chemical Engineers AIChE J, 51: 2988 2999, 2005

Keywords: high-purity O2, PVSA, backll step, blowdown step, CMS, zeolite 10X

Introduction
In the study of air separation, classical cryogenic distillation has been widely recognized as one of the most common methods of O2 production. However, this technology is not compatible with relatively small size [100 TPDc (tons per day contained)] O2 plants.1 Recently, the development of highly selective adsorbents such as LiX, LiAgX, and LiCaX has contributed to the possibility of manufacturing up to 200 TPDc by vacuum swing adsorption (VSA) process, at a power consumption rate competitive with that of cryogenic distillation.2 During the last 30 years, commercial applications for adCorrespondence concerning this article should be addressed to C.-H. Lee at leech@yonsei.ac.kr.

2005 American Institute of Chemical Engineers

sorptive O2 generation from ambient air using several molecular sieve zeolites have been expanded.3-7 However, because air contains a small amount of Ar, which has physical properties similar to those of O2, the product generated from the zeolite bed typically contains a substantial amount of the Ar impurity. Therefore, in the equilibrium separation process using zeolites, O2 purity is limited to about 94% purity. For this reason, O2 produced by adsorption technology typically has 90 93% purity, of which 4 5% is Ar and 2 6% N2, and is generally used in many chemical processes such as the biological treatment of wastewater, steel industries, paper and pulp industries, and glass-melting furnaces.8,9 In addition to these industries, there is a great demand for O2 with a purity level of 99% or higher in other applications, such as welding and cutting processes, plasma chemistry, ozone generation, medicine, combustion, cylinder lling, and O2 for
Vol. 51, No. 11 AIChE Journal

2988

November 2005

Table 1. Cyclic Sequences of Three-Bed PVSA Process for Air Separation and O2 Purication*
PVSA I Step 1st zeolite bed 2nd zeolite bed CMS bed [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 8 ] [ PR 1 ] [ AD 1 ] [ VU 2 ] [ BF 2 ] [ BD 2 ] [ PR 1 ] [ AD 1 ] product (O2 supply from 1st zeolite bed) [ DPE 1 ] [ PPE 2 ] [ ID ] [ VU 2 ] [ BF 2 ] [ PR 1 ] [ AD 1 ] [ BD 2 ] [ PR 1 ] [ AD 1 ] product (O2 supply from 2nd zeolite bed) [ PPE 2 ] [ DPE 1 ] [ ID ]

PVSA II Step [ 1 ] [ 2 ] [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 8 ] [ 9 ] [ 10 ] 1st zeolite bed [ PR 1 ] [ AD 1 ] [ DPE 1 ] [ VU 2 ] [ BF1 2 ] [ PPE 2 ] [ BF2 1 ] 2nd zeolite bed [ VU 2 ] [ BF1 2 ] [ PPE 2 ] [ BF2 1 ] [ PR 1 ] [ AD 1 ] [ DPE 1 ] [ VU 2 ] CMS bed [ BD2 2 ] [ PR 1 ] [ AD 1 ] [ ID ] [ BD1 2 ] [ BD2 2 ] [ PR 1 ] [ AD 1 ] [ ID ] [ BD1 2 ] product (O2 supply from product (O2 supply from 1st zeolite bed) 2nd zeolite bed)
*1, Cocurrent ow; 2, countercurrent ow; PR, pressurization step; AD, adsorption step; DPE, depressurizing pressure equalization step; VU, vacuum step; BF1, rst backll step; BF2, second backll step; PPE, pressurizing pressure equalization step; ID, idle step; BD1, rst blowdown step; BD2, second blowdown step.

breathing at high altitudes in pressurized space suits or in various laboratory applications.10 Recently, because of the reduction and sequestration of CO2, O2 inhalation combustion processes have been developed extensively for use in, among others, the iron, steel, and incineration industries.11 In these industries, O2 needs to be at least 95% pure to be used cost-effectively in fuel combustion because of the carbon tax placed on CO2 capturing. Moreover, the recommended amount of N2 impurity contained in an O2 product for controlling the amount of NOx released under the allowable exhaust standard is 1%. As a result of extensive industrial demand for high-purity O2, several adsorption processes have been developed, the patents for which involve a two-stage process consisting of zeolite beds for nitrogen removal and carbon molecular sieve (CMS) beds for the removal of argon and nitrogen impurities.12-15 Knaebel and Kandynin14 posited that an O2 product with 99.6% purity and 25% recovery could be obtained from a ve-step, four-bed PSA using two consecutive blowdown steps. They also noted that a desirable level of productivity would be about 11 Nm3 m3 h1. Hayashi et al.16 observed that an O2 recovery ratio of up to 50% was obtained during the production of O2 with a purity of 99% through a connection between the zeolite Ca-X bed and the CMS bed, each of which consisted of three beds. Recently, Jee et al.17 reported that a two-bed PSA process with CMS could purify 95% O2 to 99.8% purity with 78% recovery. However, these processes mentioned need to overcome the additional equipment costs to set up the purication unit and the signicant product loss for applications as a O2 balance unit with a cryogenic unit and a direct on-site unit. In this study, a three-bed pressure vacuum swing adsorption (PVSA) process, consisting of two zeolite 10X beds and one CMS bed, was developed to produce high-purity O2 with 1% N2 impurity from ambient air. The rst two zeolite 10X beds functioned to achieve bulk separation of the air. Thereafter, the CMS bed puried the O2 supplied from the zeolite 10X beds. To increase the purity and productivity of the PVSA process, we proposed reuxing the waste from the CMS bed into the zeolite 10X beds by using vacuum pressure in the zeolite bed. As a result, the PVSA process could be operated with one blower/compressor and one vacuum pump.
AIChE Journal November 2005

A nonisothermal dynamic model incorporating mass, energy, and momentum balances was used to predict dynamic behavior in the PVSA process, combining equilibrium separation with kinetic separation. The linear driving force (LDF) model with constant diffusivity for the equilibrium separation bed and the concentration-dependent rate model for the kinetic separation bed were used concurrently to the dynamic model.

Description of the PVSA Process

Two different three-bed PVSA cycles were proposed to produce 97 and 99% O2 generation from ambient air. The cyclic sequences of the proposed PVSA processes are represented in Table 1.18 In the case of the PVSA I process, the two zeolite beds were operated as a PVSA sequence containing pressure equalization and single backll steps from the CMS bed. Meanwhile, the CMS bed was operated as a PSA sequence with one blowdown step. As noted in Table 1, when the rst zeolite bed produced highly concentrated oxygen containing impurities such as Ar and N2, the CMS bed was pressurized by the product from the zeolite 10X bed until it reached the adsorption pressure of the bed and this pressure was maintained throughout the adsorption step of the process. During the adsorption step, Ar and N2 impurities were removed by the kinetic selectivity of CMS, although the concentration of O2 efuent from the CMS bed was much higher than that of the ambient air. The efuents from the adsorption step of the CMS bed were supplied to the second zeolite bed and regenerated by vacuum pressure, as a backll step to partially pressurize the zeolite bed with the concentrated oxygen. This step prevented any of the gas supplied by the zeolite bed from being wasted in the CMS bed. During the pressure equalization step in the zeolite bed, the CMS bed underwent the idle step for the cyclic symmetry of the process. The idle step provided the contact time for the diffusion of O2.17 Then, during the pressurization step of the second zeolite bed, the CMS bed produced puried oxygen at the blowdown step. In the case of the PVSA II process, a two-stage blowdown step of the CMS bed, followed by a fractionation of its oxygen product, was implemented to increase O2 purity to 99%. BeVol. 51, No. 11 2989

cause Ar and N2 impurities remain in the gas phase and not solid phase, most of them can be removed at the beginning of the blowdown step, but the O2 concentration of the efuent is much higher than that of ambient air. Therefore, in the PVSA II process, one more backll step by the efuent from the rst blowdown step of the CMS bed was added to the zeolite bed compared to the PVSA I process.

Mathematical Model
To understand the dynamic behaviors of a PVSA process, mathematical models were developed on the basis of the following assumptions: (1) the gas phase behaves as an ideal gas mixture, (2) radial concentration and temperature gradients are negligible, (3) thermal equilibrium between adsorbents and bulk ow is assumed, (4) the ow pattern is described by the axially dispersed plug-ow model, and (5) the pressure drop along the bed is measured using Erguns equation. In numerous studies the signicance of the radial gradient was neglected and similar assumptions in simulating the adsorption processes were also made.19,20 The component and overall mass balances for the bulk phase in the adsorption column are given by DL 2 ci uci ci 1 qi p 0 2 z z t t

Figure 1. Simulation scheme of three-bed PVSA process with two zeolite beds and one CMS bed.
(a) BF1 and (b) BF2.

n i1

(1)

1 uC C p z t

qi 0 t

(2)

Another characteristic of the adsorption process is the temperature variation caused by the heat of adsorption and desorption. To minimize the deviation in the prediction of high-purity product, the energy balance for the gas phase also includes the heat transfer to the column wall, expressed as KL T 2 T uT t g Cpg B Cps g Cpg z2 z t B

ware, shown in Figure 1a, after retaining the temporal data from the efuent stream of the zeolite 10X bed during the adsorption step, these data were used in the pressurization and adsorption steps of the CMS bed. The temporal efuent data from the end of the CMS bed during its adsorption step was used in the rst backll step of the zeolite 10X bed. The temporal data from the CMS bed during the blowdown step were applied to the second backll step of the zeolite 10X bed (Figure 1b). The two sets of boundary conditions were applied to the zeolite bed and CMS bed separately. The boundary conditions for zeolite two-bed PVSA are as follows: Boundary Conditions for Feed Pressurization (PR) and Adsorption (AD) Steps

i1

qi 2hi Hi T Tw 0 t RBi

DL (3) KL

ci z T z

uci z 0 ci z 0
z0

ci z

0
zL

(5a)

where t is the total void fraction [ (1 ) ], is the particle porosity, and B is the bed density [(1 ) p]. To consider the heat loss through a wall and the heat accumulation in the wall, another energy balance for the wall of the adsorption bed was used

g Cpg uTz 0 Tz 0
z0

T z

0 (5b)

zL

wC pwA w

Tw 2 R Bih i T T w 2 R Boh o T w T atm t

Boundary Conditions for First Backll (BF1), Second Backll (BF2), and Pressurizing Pressure Equalization (PPE) Step

(4) DL

2 2 where Aw (R Bo R Bi ). In this study, the proposed PVSA process was operated by a combination of two cyclic sequences (Table 1). Therefore, an imbalance in the pseudo-cyclic operation among these beds was inevitable, although the efuents remained systematically connected to each other. In the simulation using gProms soft-

ci z T z

uci z L ci z L
zL

ci z

0
z0

(6a)

KL

g Cpg uTz L Tz L
zL

T z

0 (6b)

z0

2990

November 2005

Vol. 51, No. 11

AIChE Journal

Table 2. Adsorption Isotherm and Diffusion Rate Parameters of N2, O2, and Ar on Zeolite 10X and CMS
Zeolite 10X Constants O2 Ar N2 Equilibrium Parameters For LRC model k 1 (mol/g) k 2 (mol g1 K1) k 3 (1/kPa) k 4 (K) k5 k 6 (K) Heat of adsorption, (kJ/mol) H 4.568 103 1.071 105 9.866 105 578.9 3.552 797.4 9.623 3.697 5.010 106 1.651 105 869.7 3.184 665.6 10.46 2.330 103 4.400 106 1.674 108 3586 1.384 142.1 16.32 Adsorption Rate Parameters For rate model D ci (cm2/s) in Eq. 14 C i (1/s) in Eq. 15 0.0153 0.0147 0.0043 0.024 0.000047 0.000090 1.527 102 3.230 105 2.290 105 966.1 1.187 106.0 13.81 2.042 102 5.300 105 23.97 105 324.6 1.646 238.2 14.23 2.363 102 6.380 105 36.10 105 1444 1.692 270.0 13.39 O2 CMS Ar N2

Boundary Conditions for Depressurizing Pressure Equalization (DPE) and Countercurrent Vacuum Regeneration (VU) Steps

ci z
z0

T z

0
z0

ci z
z L

T z

0
z L

The pressure history during a PSA experiment at the feed or bed end was tted by polynomials and used as a boundary condition for the overall mass balance.21 To consider the pressure drop effect across the bed, Erguns equation was introduced as a momentum balance22 dP au buu dz b 1.75 1 2 R p

(7) The boundary conditions for a CMS one-bed PSA are as follows: Boundary Conditions for PR and AD steps DL

(12a)

150 1 2 2 2 4Rp

(12b)

ci z T z

uci z 0 ci z 0

z0

ci z

0
zL

(8a)

KL

g Cpg uTz 0 Tz 0
z0

T z T z

where u is the interstitial velocity. The multicomponent adsorption equilibrium was predicted by the following: Loading Ratio Correlation (LRC) Model qi q miB iP ini 1 jn1 B jP jnj

0
zL

(8b) Boundary Conditions for Idle (ID), First Blowdown (BD1), and Second Blowdown (BD2) Steps

(13)

ci z
z0

T z

0
z0

ci z
z L

0
z L

where qmi k1 k2 T, Bi k3exp(k4/T), and ni k5 k6/T. In this study, the general LDF model assuming constant diffusivity was applied to the sorption rate of equilibrium separation bed (zeolite 10X) qi i q * i q i t

(9) Initial Condition for Fluid Flow c i z , 0 c * i q i z , 0 q * i (10)

15 D ei 2 rp

(14)

In the case of the kinetic separation bed (CMS), the modied LDF model with concentration-dependent diffusivity was used.23,24 qi i q * i q i t

In this study, the adsorption beds were saturated with pure oxygen before each run. Initial Condition for Heat Flow T z , 0 T atm
AIChE Journal

15 D ei 2 rp
0.5 D ei C i P ri 1 B iP i 2 (15)

(11)
November 2005 Vol. 51, No. 11

2991

Table 3. Characteristics of Adsorbent and Adsorption Bed

Characteristic Adsorbent Type Average pellet size, R p (cm) Heat capacity, C ps (J g1 K1) Pellet density, p (g/cm3) Particle porosity, Bed density, B (g/cm3) Adsorption Bed Length, L (cm) Inside radius, R Bi (cm) Outside radius, R Bo (cm) Heat capacity of column, C pw (J g1 K1) Density of column, w (g/cm3) Internal heat transfer coefcient, h i (J cm2 K1 s1) External heat transfer coefcient, h o (J cm2 K1 s1) 100 1.72 1.987 0.50 7.83 3.85 103 1.42 103 100 1.1 1.275 0.50 7.83 3.85 103 1.42 103 Pellet 0.115 1.13 1.1 0.36 0.82 Pellet 0.28 0.96 0.90 0.30 0.63 Zeolite 10X CMS

A high-performance vacuum pump (DAH-60; Ulvac Kiko Inc., Kanagawa, Japan) was applied to bring the vacuum step of the zeolite 10X bed up to a level of 33.44 45.60 kPa, depending on the applied step time. Each activated adsorbent bed was lled with pure O2 (99.9%) to prevent contamination from the outside air. Before running each experiment, the adsorption beds were vacuumed for 2 h. As an initial condition, the PVSA experiments were conducted at the bed saturated with pure O2 (99.9%) under the same level of adsorption pressure as used throughout the experiment. The temperatures of the feed, bed, and surroundings were kept in the range of 297 to 300 K during the experiments. The ternary mixture (N2/O2/Ar; 78:21:1 vol %; DaeSung Industrial Gas Co., Seoul, Korea) was used as feed gas for the PVSA experiments. Specic operating conditions such as operating pressure, feed ow rate, and each cycle/step times can be found in Table 4.

Values of the adsorption isotherm and rate parameters of N2, Ar, and O2 on CMS and zeolite 10X, presented in Table 2, were very similar to the published data.25-27

Results and Discussion Cyclic performance of PVSA I with a single blowdown/ backll step
Because the efuent from the zeolite 10X bed during the AD step at 253.31 kPa was supplied as feed gas for the PR and AD steps of the CMS bed, the CMS bed was pressurized to 162.12 kPa. The efuent from the AD step of the CMS bed was backlled into the other zeolite 10X bed. Therefore, the pressure of the zeolite bed increased slightly after the VU step. It is noteworthy that the CMS bed underwent two cycles for every one cycle of the zeolite bed because each of the zeolite beds and the CMS bed was tied to the same cycle time. To prevent confusion of the number of cycles, the cycle number of the

Experimental
Zeolite 10X (Baylith WE-G 639; Bayer AG, Leverkusen, Germany) and CMS (3A; Takeda Chemicals, Tokyo, Japan) were used as adsorbents in the PVSA experiments. Characteristics of each adsorbent are listed in detail in Table 3. Before the experimental runs, the zeolite 10X was regenerated at 613 K overnight and the CMS at 423 K. The zeolite 10X adsorption beds were made of stainless steel according to the following dimensions: length, 100 cm; ID, 3.44 cm; and wall thickness, 2.67 mm. The dimensions of the CMS bed were: length, 100 cm; ID, 2.2 cm; and wall thickness, 1.75 mm. Characteristics of the adsorption beds are listed in detail in Table 3. The PVSA apparatus is shown in Figure 2. Three resistance temperature detectors (RTD, Pt 100 ) were installed 10, 50, and 80 cm from the feed end to measure temperature variations inside the bed. The ow rate was regulated by a mass ow controller and the total amount of feed ow was measured by a wet gas meter (Sinagawa Co., Tokyo, Japan). To keep the pressure of the adsorption bed constant during the adsorption step of the zeolite 10X bed, a back-pressure regulator was installed between the zeolite 10X bed and the CMS bed. An additional back-pressure regulator was also equipped at the end of the CMS bed to keep the pressure of the adsorption bed constant during the adsorption step of the CMS bed. Bed pressure was measured by pressure transducers equipped at the top and bottom of each bed. The concentration of inuent and efuent was analyzed by a mass spectrometer (QMG 422; Balzers Instruments, Lichtenstein). This analysis was conrmed by gas chromatography (HP 5890II; HewlettPackard, Palo Alto, CA). The system was fully automated by a personal computer running a control program, and all measurements including ow rate, pressure, temperature, and concentration were saved on the computer through the use of an AD converter.
2992 November 2005

Figure 2. Apparatus for a three-bed PVSA process. Vol. 51, No. 11 AIChE Journal

Table 4. Operating Conditions for PVSA for Air Separation and O2 Purication
Adsorption Pressure of Zeolite 10X Bed (kPa) 253.31 Adsorption Pressure of CMS Bed (kPa) 162.12 174.28 131.72 PVSA II 167.19 177.32 136.79 PVSA I* (BD-PR-AD-ID-BDPR-AD-ID) PVSA II* (BD2PR-AD-ID-BD1-BD2-PR-ADID-BD1) 40-20-20-20-40-20-20-20 (Base step time) 40-30-10-20-40-30-10-20 40-10-30-20-40-10-30-20 30-20-20-20-10-30-20-20-20-10 (Base step time) 30-30-10-20-10-30-30-10-20-10 30-10-30-20-10-30-10-30-20-10

Run No. 1 2 3 4 5 6

Process PVSA I

Vacuum Pressure of Zeolite 10X Bed (kPa) 38.50 33.44 45.60 39.52 33.44 44.58

Feed Rate (LSTP/min) 2.0

*The step sequence and time for the CMS bed were presented.

PVSA process was determined by the cycle of the zeolite 10X bed. The temperature excursion in the zeolite 10X bed was around 5 K, whereas in the CMS bed it was 1 K. Because the experiment was performed under O2 saturated conditions, the temperature history of the CMS bed showed an increase during the initial cycles. The temperature cyclic steady state for the process was reached at 15 cycles. Figure 3 shows the cyclic variation of the mean concentration produced by the zeolite 10X and CMS beds at Run 1 (Table 4). The concentration variation curves also reached a cyclic steady state after nearly 15 cycles. In previous studies, the two-bed O2 PSA purier packed with CMS reached the cyclic steady state after only three cycles17 when the binary feed (O2/Ar; 95:5 vol %) was used. However, in this study, the cyclic steady state of the CMS bed was delayed because of the cyclic steady state of the zeolite 10X bed.

Figure 3. Cyclic variation of O2 purity and Ar/N2 impurities produced from the BD step of CMS bed during the cyclic operation of PVSA I at base operating condition (Run 1). AIChE Journal November 2005

As noted in Figure 3, the zeolite 10X bed, as a bulk air separator, produced 93.5% O2 with 2.5% N2 and 4% Ar impurities. The product purity was higher than that of conventional O2 VSA at similar operating conditions because the zeolite bed was partially pressurized by the BF step with the concentrated O2 from the AD step of the CMS bed. Because such product quality was supplied to the CMS bed, the purity of O2 puried through the CMS bed was about 96.8% at the cyclic steady state (Figure 3). The amount of N2 impurity signicantly decreased to the level of 4000 5000 ppm (0.4 0.5 vol %) and Ar impurity also decreased from 4 to 2.7%. The results simulated by the dynamic model corresponded with the results of the experiment in both the equilibrium and kinetic separation beds. The simulated concentration proles of each adsorbate at the gas phase of the zeolite 10X bed are presented in Figures 4a to 4c. At the end of the AD step (Figure 4a), the mass-transfer zone (MTZ) of each adsorbate became steeper as that of each gas propagated to the bed end. In addition, the concentration of Ar at the bed end became slightly higher than that of N2. At the end of the BF step after the DPE and VU steps (Figure 4b), the MTZ of N2 was observed retreating to the feed end with a broader shape than it had at the VU step and O2 and Ar accumulated in higher concentrations at the bed end. Finally, at the end of the PPE step (Figure 4c), each MTZ retreated further to the feed end bearing the concave shape of N2 MTZ and the convex shape of O2 and Ar MTZs. Figures 4d and 4e show the axial concentration proles along the CMS bed at the end of each step. As noted in Figure 4d, the O2 concentration prole at the end of the PR step was formed at the concave shape around the bed end because the Ar and N2 in the gas phase were concentrated at that location of the bed. Moreover, because a higher concentration of Ar than N2 was supplied to the CMS bed and the diffusion rate of Ar was much slower than that of N2, the concentration of Ar at this step was higher than that of N2. At the end of the AD step (Figure 4e), the gas-phase concentration of each adsorbate was nearly constant at each of the axial positions because the adsorbed phase was nearly saturated with highly concentrated O2. This result means that a tangible amount of the impurities in the gas phase was removed during the AD step time. In addition, because the
Vol. 51, No. 11 2993

of the CMS bed was tied to the VU and BF steps of the other zeolite bed. Therefore, a decrease in the AD step time of the CMS bed caused an increase in the VU step time (higher vacuum pressure in Table 4) as well as an decrease in the BF step time of the zeolite bed. As shown in Figure 5a, as the AD step time of the CMS bed decreased from 30 to 10 s (the PR step time of the CMS bed increased from 10 to 30 s), the purity of both the zeolite 10X and the CMS beds decreased linearly because of the reduced amount of backll gas in the zeolite bed and the diminished time for impurity removal in the CMS bed. The effect of the AD step time of the CMS bed on the purity of each bed was slightly greater in the zeolite bed than in the CMS bed. As shown in Figure 5b, with a decrease in the AD step time, O2 recovery substantially increased to 85% because the extended PR step time of the CMS bed led to increased adsorption pressure in the CMS bed. Compared to the change of recovery in the AD step time, the effect of the AD step time of the CMS bed on purity was relatively small because the extended VU step time in the zeolite bed could compensate somewhat for the reduced BF gas (Table 4). Figure 6 shows the axial concentration proles at the AD step in the CMS bed and at the BF step in the zeolite 10X bed. In Figure 6a, the MTZs of the components did not completely proceed to the CMS bed end within the AD step time of 10 s. Therefore, this AD step time was not sufcient to remove the
Figure 4. Simulated O2 concentration proles in the gas phase at (a) AD, (b) BF, and (c) DPE steps of zeolite 10X bed and (d) PR, (e) AD, and (f) BD steps of CMS bed during the PVSA I operation at base operating condition (Run 1).

concentration of O2 during the AD step in the CMS bed was relatively high, the efuent BF gas could improve the performance of the zeolite bed (Figure 4c). At the end of the ID step, the MTZs of N2 and Ar at the bed end increased slightly because the small amount of O2 diffused into the solid phase. Then, high-purity O2 was produced at the end of the BD step (Figure 4f). As shown in Figure 3, a greater amount of Ar than N2 was included in the product because the Ar with the slower diffusion rate remained in the gas phase.23,28 In addition, the MTZs of Ar and N2 crossed over at the vicinity of the bed end in the BD step because the desorption of N2 from the adsorbed phase was quicker than that of Ar.

Effect of adsorption step time in the CMS bed on adsorption dynamics in PVSA I
As shown in Figure 4, the AD step in the CMS bed played a key role in removing the N2 and Ar impurities from the zeolite bed. In addition, because this step was connected to the zeolite bed as a BF step, the AD step time in the CMS bed contributed to the purity and recovery of the product in the process. Therefore, the effect of the AD step time of the CMS bed on O2 purity and recovery was studied at the xed AD step time of the zeolite bed, 40 s, which corresponded to the sum of the PR and AD step times of the CMS bed (Table 1). Although the AD step time of one zeolite bed was xed to 40 s, the AD step
2994 November 2005

Figure 5. Effect of PR and AD step times of CMS bed on (a) O2 purity and (b) O2 recovery at base adsorption pressure and base feed ow rate condition (Runs 1 to 3). Vol. 51, No. 11 AIChE Journal

Figure 6. Axial concentration proles along the bed in the gas phase at the end of (a) BF step of zeolite 10X bed and (b) AD step of CMS bed at base adsorption pressure and base feed ow rate condition (Runs 1 to 3).

production time (Figure 4). To improve the product purity of O2 in PVSA II, a fractionation step of the product from the CMS bed was added to PVSA I, as shown in Table 1. Unlike PVSA I, an additional BF step (BF2) was added in the zeolite 10X bed and two consecutive BD steps (BD1 and BD2) were applied to the CMS bed. The efuents from the AD and BD1 steps in the CMS bed were used to partially pressurize the zeolite 10X bed at the BF1 and BF2, respectively. In comparison to PVSA I, PVSA II was operated at the same total cycle time, adsorption pressure, and feed ow rate (Table 4), all based on the zeolite bed. Figure 7 shows the representative cyclic behaviors of the CMS and zeolite 10X beds at Run 4, comparing the experiments efuent O2 purity and Ar/N2 impurities with the calculated results. Similar to the PVSA I process, the concentration variation reached a cyclic steady state after nearly 17 cycles. However, in Run 4 of PVSA II, O2 purity from the CMS bed increased noticeably to 98.8%. Compared with PVSA Is performance in Figure 3, it is noteworthy that the amount of N2 in the zeolite bed and the amount of Ar in the CMS bed were reduced signicantly in PVSA II. In the nal product of PVSA II, the mole fraction of Ar impurity contained in the product decreased to 1.2%, whereas that of N2 was several tens of ppm. In addition, O2 purity from the zeolite 10X bed exceeded 96% with 3.6% Ar and 0.4% N2, recognized as the theoretical limitation of the adsorption equilibrium process. Figures 8a 8c show the axial concentration prole at the end of each step along the zeolite 10X bed in PVSA II. As shown in Figures 8a 8c, the MTZs of all the components in PVSA II were similar to those in PVSA I (Figures 4a 4c). However, the axial concentration amount of N2 at the bed end was less than that of Ar in the entire PVSA II process, which stands in marked contrast to PVSA I (Figures 4a 4c).

impurities of N2 and Ar in the CMS bed. Furthermore, N2 acted as the primary impurity in the product because the concentration of N2 in the bed was higher than that of Ar under that step time. However, over 20 s, all the MTZs were linear along the CMS bed, showing a lower concentration of N2 than Ar. The amount of N2 signicantly decreased with an increase in AD step time. The corresponding step in the zeolite bed is shown in Figure 6b. At the BF step time of 10 s (corresponding to the AD step time of 10 s in the CMS bed), the MTZs of O2 and Ar were unfavorable because a relatively large amount of N2 was supplied from the CMS bed. However, as the BF step time increased, the MTZ of N2 retreated to the feed end and the MTZs of O2 and Ar became favorable because the highly concentrated O2 feed backlled from the AD step of the CMS bed was supplied from the bed end. Consequently, an increased AD step time in the CMS bed led to an increase in the purity of the O2 product through the AD step of the zeolite bed, although the recovery decreased (Figure 5) because a greater amount of the gas supplied from the zeolite bed was used in the BF step.

Cyclic performance of PVSA II with double blowdown/ backll steps

As to be expected, most of the impurities of the product from the CMS bed in PVSA I were recorded at the initial period of
AIChE Journal November 2005

Figure 7. Cyclic variation of O2 purity and Ar/N2 impurities produced from BD2 step of CMS bed during cyclic PVSA operation for 99% O2 generation at base operating condition (Run 4). Vol. 51, No. 11 2995

operated under pressures ranging from adsorption pressure to medium pressure. Simultaneously, the efuent gas at this step was supplied to the zeolite bed through the BF2 step (Figure 8c). As a result, high-purity O2 of 99% could be produced through the BD2 step of the CMS bed (Figure 8f) and the bed was kept clean for the next step.

Effect of adsorption step time in the CMS bed on adsorption dynamics in PVSA II
The effect of AD step time in the CMS bed on O2 purity and recovery in PVSA II was studied to compare to the results with those of PVSA I in Figure 5. As shown in Figure 9a, when the AD step time of the CMS bed increased from 10 to 30 s, the purity of both the zeolite 10X and CMS beds increased linearly, similar to the results noted in Figure 5a. Consequently, oxygen with 99.2% purity was produced at the AD step time of 30 s because a greater amount of impurities had been removed from the CMS bed and a greater amount of O2 had been supplied to the zeolite bed through the BF1 step. Therefore, the O2 purity of the zeolite 10X bed also increased to 97% under this AD step time. However, as shown in Figure 9b, the recovery diminished as the PR step time decreased because the decreased PR step time of the CMS bed led to a lower adsorption pressure in the bed. Compared with the results of PVSA I, the deviation of the

Figure 8. Simulated O2 concentration proles in the gas phase at (a) AD, (b) PPE, and (c) BF2 steps of zeolite 10X bed and (d) PR, (e) AD, and (f) BD2 steps of CMS bed during the PVSA II operation at base operating condition (Run 4).

At the end of the PR and AD steps in Figure 8a, the MTZs of N2 in PVSA II propagated to the bed end more slowly than those in PVSA I. This was mainly caused by the second backll gas, which was composed of highly concentrated O2. In Figure 8b, like Figure 4c, the concave shape of axial N2 concentration at the bed end was formed at the end of the PPE step even though the bed had for the most part been regenerated through previous regeneration steps such as VU and BF1. This concave shape had a detrimental effect on the production of high-purity O2 because the N2 impurity near the bed end was contained in the product at the AD step of the next cycle. However, the concave shape of N2 MTZ and the convex shape of O2 MTZ at the bed end disappeared at the end of the BF2 step (Figure 8c) because the high-purity O2 from the CMS bed was reintroduced at the bed end of the zeolite 10X bed. Figures 8d 8f show the axial concentration prole at the end of each step along the CMS bed in PVSA II. Compared to the axial concentration proles of PVSA I in Figure 4d, the rollups of N2 and Ar at the PR step occurred at the bed end as shown in Figure 8d. In addition, because the amount of N2 impurity supplied from the zeolite 10X bed was noticeably smaller (Figure 8a), the concentration proles of N2 throughout the steps were much lower than those in Figure 4. Therefore, as shown in Figure 8e, the CMS bed in PVSA II was saturated with a higher concentration of O2 at the end of the AD step than was that in PVSA I. Moreover, by introducing the two-stage blowdown step, considerable amounts of Ar and N2 were removed from the CMS bed through the BD1 step, which was
2996 November 2005

Figure 9. Effect of PR and AD step times of CMS bed on (a) O2 purity and (b) O2 recovery at base adsorption pressure and base feed ow rate condition (Runs 4 to 6). Vol. 51, No. 11 AIChE Journal

zeolite bed had conditions favorable to the effective supplying of better feed to the CMS bed. Figure 11 shows the axial concentration proles of the BD1 and BD2 steps in the CMS bed of PVSA II. At the end of the BD1 step (Figure 11a), the O2 mole fraction in the gas phase was in the range of 9798% and the N2 mole fraction was 0.005%. The lower mole fraction of impurities at the end of the BD1 step was achieved with an increase in AD step time because the AD step in the CMS bed worked to remove the impurities. After the BD2 step, as shown in Figure 11b, the O2 mole fraction exceeded 99.5% and almost the same concentrations of impurities remained regardless of AD step time conditions in the CMS bed. Therefore, the CMS bed was ready to purify feed under more favorable conditions in the PVSA II process. Work is currently under way to optimize the process by using a parametric study and to assess economical analysis by using the pilot-plant study.

Conclusions
The three-bed PVSA process was rened to surpass 94% O2 purity from ambient air using adsorption technology. By using two zeolite 10X beds and one CMS bed, the PVSA process
Figure 10. Axial gas phase concentration proles along the bed at the end of (a) BF1 and (b) BF2 steps of zeolite 10X bed at base adsorption pressure and base feed ow rate condition (Runs 4 to 6).
BF2 step time was xed to 10 s.

simulated recovery in the PVSA II is slightly higher than that in the PVSA I. Because one more backll step and one more blowdown step were applied to the PVSA II, respectively, more errors in these steps, especially velocity, might be accumulated in the simulation. It is clear that PVSA II can produce a higher purity of O2 than PVSA I under the same conditions in cyclic time, pressure, and feed ow rate. However, the recoveries of Runs 4 and 5 were about 10 20% lower than the corresponding Runs 1 and 2 in PVSA I as a result of the BD1 step in the CMS bed and the BF2 in the zeolite bed. In the case of Run 6, it is noted that O2 with 99.2% purity could be produced at a higher recovery than the corresponding Run 3 in PVSA I. Figure 10 shows the axial concentration proles of the BF1 and BF2 steps in the zeolite bed. Similar to the results in Figure 6b, as the BF step time increased, the MTZs of each component retreated to the feed end (Figure 10a). However, unlike the results in Figure 6b, the MTZs of O2 and Ar maintained their favorable shape even at a BF1 step time of 10 s and the MTZs of Ar crossed near the center of the bed. This was because the BF2 step following from the CMS bed led to a more highly saturated O2 condition throughout the bed. When the same BF2 step time was applied to the above three BF1 conditions, the MTZ of each adsorbate further retreated to the feed end (Figure 10b). The BF1 step of 30 s showed the lowest MTZ of Ar and the highest MTZ of O2 at the BF2 step. Moreover, because the convex shape of the O2 MTZ and the concave shape of the N2 MTZ at the end of the PPE step (Figures 4c and 8b) disappeared at the end of the BF2 step, the
AIChE Journal November 2005

Figure 11. Axial gas phase concentration proles along the bed at the end of (a) BD1 and (b) BD2 steps of CMS bed at base adsorption pressure and base feed ow rate condition (Runs 4 to 6).
BD1 and BD2 step times were xed to 10 and 30 s, respectively.

Vol. 51, No. 11

2997

could produce O2 with a purity 94% directly from the ambient air. In addition, the nonisothermal dynamic model successfully predicted the cyclic behaviors in the experiment. In the PVSA I process, which adopted a single stage of the blowdown/backll step, the cyclic steady state was reached after about 14 15 cycles. In the case of the bulk air separator, the zeolite 10X bed, O2 with a purity of 90.293.7% was produced, whereas the impurities of N2 and Ar were 2.5 6.3 and 3.5 4.2%, respectively. This O2 product was further puried to 95.4 97.4% with a recovery of 43.4 84.8% in the CMS bed. The amount of N2 in the product was 4000 5000 ppm during the PVSA I process. The effect of the PR and AD step times in the CMS bed on the performance of the process was analyzed at the xed AD step time of the zeolite 10X bed. As a consequence, the increased AD step time of the CMS bed contributed to an increase in O2 purity because the AD step of the CMS bed acted as an impurity-removal step and a greater amount of backll gas was supplied to the zeolite 10X bed. However, the excessive AD step time of the CMS bed led to a signicant decrease in recovery because the backll and vacuum step times of the zeolite bed increased simultaneously. In the PVSA II process, two-stage blowdown/backll steps were adopted to increase O2 purity by fractionating the O2 product from the BD step of the CMS bed. The zeolite 10X bed produced O2 with a purity of 95.297.1%. This process overcame the purity limit of common adsorption processes using zeolites because of the two-stage backll steps supplied from the CMS bed. Thus, PVSA II could produce an O2 purity of 98.299.2% with a recovery of 47.2 63.6% through the CMS bed. Also, the amount of N2 impurity decreased to 100 ppm. Similar to PVSA I, the increased AD step time of the CMS bed led to an increase in O2 purity and a decrease in O2 recovery in the PVSA II process. However, because more highly concentrated O2 was supplied from the zeolite bed in PVSA II rather than in PVSA I, the CMS bed in PVSA II could be kept fairly clean after the production step, regardless of the applied AD step time in the CMS bed. Therefore, the CMS bed had more favorable conditions for the purication of feed in PVSA II. In the near future, further results will be presented of the parametric study to directly produce a maximum of 99.8% O2 from air by using the proposed PVSA processes.

L bed length, cm n dimensionless equilibrium parameter for Langmuir Freundlich model P total pressure, kPa Pr reduced pressure q, q*, q amount adsorbed, equilibrium amount adsorbed, and average amount adsorbed, respectively, mol/g qm equilibrium parameter for LangmuirFreundlich model, mol/g R gas constant, J mol1 K1 Rp radius of pellet, cm RBi, RBo inside and outside radius of the bed, respectively, cm t time, s Tatm temperature of atmosphere, K T, Tw pellet or bed temperature and wall temperature, respectively, K u interstitial velocity, cm/s yi mole fraction of species i in gas phase z axial distance in bed from the inlet, cm

Greek letters
particle porosity , t voidage of adsorbent bed and total void fraction, respectively g, p, B, w gas density, pellet density, bulk density, and bed wall density, respectively, g/cm3 LDF coefcient, s1 viscosity, Pas

Subscripts
B i p g s w bed component i pellet gas phase solid phase wall

Literature Cited
1. Kumar R. Vacuum swing adsorption process for oxygen productionA historical perspective. Sep Sci Technol. 1996;31:877-893. 2. Rege SU, Yang RT. Limits for air separation by adsorption with LiX zeolite. Ind Eng Chem Res. 1997;36:5358-5365. 3. Rei G. Status and development of oxygen generation processes on molecular sieve zeolites. Gas Sep Purif. 1994;8:95-99. 4. Kawai M, Kaneko T. Present state of PSA air separation in Japan. Gas Sep Purif. 1989;3:2-6. 5. Wilson SJ, Webley PA. Cyclic steady-state axial temperature proles in multilayer, bulk gas PSAThe case of oxygen VSA. Ind Eng Chem Res. 2003;41:2753-2765. 6. Jiang L, Biegler LT, Fox VG. Simulation and optimization of pressure-swing adsorption systems for air separation. AIChE J. 2003;49:1140-1157. 7. Mendes AMM, Costa CAV, Rodrigues AE. Oxygen separation from air by PSA: Modeling and experimental resultsPart I: Isothermal operation. Sep Purif Technol. 2001;24:173-188. 8. Yang RT. Gas Separation by Adsorption Processes. Boston, MA: Butterworth; 1987. 9. Ruthven DM, Farooq S, Knaebel KS. Pressure Swing Adsorption. New York, NY: VCH; 1994. 10. Zayaraman A, Yang RT, Cho SH, Bhat SG, Choudary VN. Adsorption of nitrogen, oxygen and argon on Na-CeX zeolites. Adsorption. 2002; 8:271-278. 11. Ko D, Siriwardane R, Bigler RT. Optimization of a pressure-swing adsorption process using zeolite 13X for CO2 sequestration. Ind Eng Chem Res. 2003;42:339-348. 12. Miller GW, Theis CF. Secondary oxygen purier for molecule sieve oxygen concentrator. U.S. Patent No. 4 813 979; 1989. 13. Miller GW, Theis CF. Molecular sieve oxygen concentrator with secondary oxygen purier. U.S. Patent No. 4 880 443; 1989.

Acknowledgments
The nancial support of the Carbon Dioxide Reduction & Sequestration R&D Center (C002-0103-001-1-0-0) is gratefully acknowledged.

Notation
AW cross-sectional area of the wall, cm2 B equilibrium parameter for LangmuirFreundlich model, kPa1 ci i component concentration in bulk phase, mol/cm3 Cpg, Cps, Cpw gas, pellet, and wall heat capacity, respectively, J g1 K1 De effective diffusivity dened by solid diffusion model, cm2/s DL axial dispersion coefcient, cm2/s hi internal heat-transfer coefcient, J cm2 K1 s1 ho external heat-transfer coefcient, J cm K1 s1 average heat of adsorption, J/mol H k parameter for Langmuir and LRC models KL axial thermal conductivity, J cm1 s1 K1

2998

November 2005

Vol. 51, No. 11

AIChE Journal

14. Knaebel KS, Kandynin A. Pressure swing adsorption system to purify oxygen. U.S. Patent No. 5 226 933; 1993. 15. Rege SU, Yang RT. Kinetic separation of oxygen and argon using molecular sieve carbon. Adsorption. 2000;6:15-22. 16. Hayashi S, Kawai M, Kaneko T. Dynamics of high purity oxygen PSA. Gas Sep Purif. 1996;10:19-23. 17. Jee JG, Kim MB, Lee CH. Pressure swing adsorption processes to purify oxygen using a carbon molecular sieve. Chem Eng Sci. 2005; 60:869-882. 18. Lee CH. Apparatus for producing oxygen and method for controlling the same. PCT/KR2004/001209; 2004. 19. Jee JG, Lee JS, Lee CH. Air separation by a small-scale two-bed medical O2 PSA. Ind Eng Chem Res. 2001;40:3647-3658. 20. Kikkinides ES, Yang RT. Effects of bed pressure drop on isothermal and adiabatic adsorber dynamics. Chem Eng Sci. 1993;48:1545-1555. 21. Jee JG, Park H, Haam S, Lee CH. Effects of nonisobaric and isobaric steps on O2 pressure swing adsorption for an aerator. Ind Eng Chem Res. 2002;41:4383-4392. 22. Yang J, Park MW, Chang JW, Ko SM, Lee CH. Effects of pressure drop in a PSA process. Korean J Chem Eng. 1998;15:211-216.

23. Bae YS, Lee CH. Sorption kinetics of eight gases on a carbon molecular sieve at elevated pressure. Carbon. 2005;43:95-107. 24. Bae YS, Moon JH, Ahn H, Lee CH. Effect of adsorbate properties on adsorption mechanism in a carbon molecular sieve. Korean J Chem Eng. 2004;21:712-720. 25. Ruthven DM, Raghavan NS, Hassan MM. Adsorption and diffusion of nitrogen and oxygen in a carbon molecular sieve. Chem Eng Sci. 1986;41:1325-1332. 26. Chen YD, Yang RT, Uawithya P. Diffusion of oxygen, nitrogen, and their mixtures in carbon molecular sieve. AIChE J. 1994;40: 577-585. 27. Sorial GA, Granville WH, Daly WO. Adsorption equilibria for oxygen and nitrogen gas mixtures on 5 molecular sieves. Chem Eng Sci. 1983;38:1517-1523. 28. Qinglin H, Farooq S, Karimi IA. Binary and ternary adsorption kinetics of gases in carbon molecular sieves. Langmuir. 2003;19: 5722-5734.

Manuscript received Nov. 23, 2004, and revision received Mar. 10, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

2999

PROCESS SYSTEMS ENGINEERING

New Numerical Method for Solving the Dynamic Population Balance Equations
Q. Hu
V & R Limited, P. O. Box 1149, London, ON N6P 1R2, Canada

S. Rohani and A. Jutan

Dept. of Chemical & Biochemical Engineering, University of Western Ontario, London, ON N6A 5B9, Canada DOI 10.1002/aic.10585 Published online July 19, 2005 in Wiley InterScience (www.interscience.wiley.com).

A new numerical scheme is proposed for solving general dynamic population balance equations (PBE). The PBE considered can simultaneously include the kinetic processes of nucleation, growth, aggregation and breakage. Using the features of population balance, this method converts the PBE into a succession of algebraic equations which can be solved easily and accurately. The new method is free from stability and dispersion problems of general numerical techniques. Some benchmark problems with analytic solutions were tested. In all cases tested this method gave accurate results with very few computational requirements. For nucleation and size-independent growth without aggregation and breakage, the numerical method gives exactly the same result as analytic solution. 2005 American Institute of Chemical Engineers AIChE J, 51: 3000 3006, 2005

Introduction
The population balance equation (PBE) has been used to model a variety of particulate systems. Analytical solutions to the PBE can only be obtained for very few special cases. Numerical solution of the PBE remains a considerable challenge due to aggregation and breakage terms. Most of the existing numerical methods are tailored to handle specic applications and lack generality (Raphael et al., 1995). The method of weighted residuals with global functions is one of the most popular methods to solve the PBE (Ramkrishna, 1985). In the method of weighted residuals, the solution is approximated by a linear combination of a series of chosen basis functions, whose unknown coefcients are determined by satisfying the PBE to dene a residual. The idea of weighted residuals is to nd the coefcients that force the residuals to be orthogonal to a chosen set of weighting functions. The method of moments is equivalent to the method of weighted residuals if the weighting functions are chosen to be
Correspondence concerning this article should be addressed to S. Rohani at rohani@eng.uwo.ca.

2005 American Institute of Chemical Engineers

polynomials. Since polynomials weights are often a poor choice for population balances on semi-innite intervals, and not all PBE formulations can be reduced to moment equations, the method of moments is not recommended. A limitation of global functions is that they cannot always capture the features of the solution, especially when there are sharp changes and discontinuities in the solution. Discontinuities arise in crystallizer systems with product classication or nes destruction. Finite-element methods approximate the solutions with local functions which can be tailored to handle discontinuities and sharp changes. Gelbard and Seinfeld (1978) considered orthogonal collocation and spline collocation on nite elements to solve population balance equations with nucleation, growth and aggregation. The semi-innite particle size domain is truncated at some large value, and then the nite domain is divided into elements. Lower order polynomials are used to approximate the solution on each elements. Nicmanis and Hounslow (1998) solved the steady-state PBE with aggregation, breakage, nucleation and growth using collocation and Galerkin methods. The method is based on an error estimate of the second moment. Raphael et al. (1995) used orthogonal collocation to solve the problem of isoelectric precipitation of sunower protein. Bennett and RoVol. 51, No. 11 AIChE Journal

3000

November 2005

hani (2001) solved a PBE by combining Lax-Wendroff and Crank-Nicholson methods. Rigopoulos and Jones (2003) used linear elements to represent the solution of dynamic PBE. The method is expected to be faster than higher-order nite element collocation methods, but it is less accurate. An alternative numerical technique for solving PBE is discretization method, which divides the particles into discrete but contiguous size ranges (Baterham et al., 1981; Hounslow et al., 1988; Kostoglou and Karabelas, 1994). This method requires a uniform particle distribution, which is unrealistic for processes such as crystallization or coagulation of aerosols where agglomeration exists. Kumar and Ramkrishna (1997) extended this method for solving PBE for breakage and aggregation of particles. The method combines the features of discretization technique with the method of characteristics. Detailed reviews of previous work on solving PBEs have been made by Kostoglou and Karabelas (1994), Vanni (2000), and Lee (2001). Most of the earlier numerical methods are difcult to implement, computationally demanding, or lacking in accuracy. They usually are tailored to solve individual problems. Their predictions of moments are subject to errors. The objective of this paper is to present a new technique for solving dynamic PBE for nucleation, growth, breakage and aggregation, all processes occurring simultaneously. The approach makes use of the properties of population balance. The new method is free from problems due to stability and dispersion of the numerical solutions. The proposed approach has been tested on a number of benchmark problems. Comparison between the analytic solutions and the numerical solutions indicates the distributions can be accurately predicted by the proposed method. In addition, the proposed method is easy to implement with few computational requirements.

Figure 1. Population balance.

where (v, v) is the aggregation kernel. The birth and death rates of particles due to breakage can be written as (Prasher, 1987)

B br v

v , v S v n v , t d v

(4)

D br v S v n v , t

(5)

where (v, v) is the breakage function and S ( v ) is the rate of breakage for particles of volume v. Equation 1 is a rst order hyperbolic differential equation for which very few analytical solutions have been found.

Population Balances
In this work, we consider the general PBE of the form (Randolph and Larson, 1988) n v , t G v n v , t B nuc v B agg v D agg v t v B br v D br v (1)

Formulation of the Algorithm Approximation of solution

In the absence of aggregation and breakage, a representation of population balance is shown in Figure 1, where population balance distributions at time t and t t are demonstrated. The particles grow into the volume range [ v ( t t ), v v ( t t )] from volume range [ v ( t ), v v ( t )] over the time interval t . n ( v ( t ), t ) and n ( v ( t t ), t t ) represent the population density at time t and t t , respectively. The population balance implies (Hu et al., 2004) n v t , t v t n v t t , t t v t t (6)

where n ( v , t ) is the population density of particles of volume v and time t , G ( v ) is growth rate for particles of volume v , B nuc ( v ) is the nucleation rate of particles of volume v , B agg ( v ) and D agg ( v ) are the birth and death rates of particles of volume v due to aggregation, and B br and D br are the birth and death rates of particles of volume v due to breakage. B agg ( v ) and D agg ( v ) can be written as (Hulburt and Katz, 1964) 1 2

However, in the presence of nucleation, aggregation and breakage, the Eq. 6 is not satised anymore. In this case, we assume that n v t , t v t n v t t , t t v t t v t v t t (7)

B agg v

v v , v n v v t n v , t d v

(2)

The last term in Eq. 7 is included due to nucleation, aggregation and breakage, and ( v ( t )) is a parameter to be determined. On the basis of the denition of growth rate, we have v t t v t G v t t (8)

D agg v n v , t
AIChE Journal

v , v n v , t d v

(3) which implies

November 2005

Vol. 51, No. 11

3001

vt t 1

Gv v

t vt
vvt

(9)

Using the Taylor series, we have nvt t, t t nvt, t t nv, t t v

Thus, given the population density at time t , the population density at time t t can be solved by Eq. 16. For numerically solving the PBE (Eq. 1), the innite domain of the particle volume v must be truncated to a nite upper limit. The error in the i th moment of the solution due to domain truncation is

v t t v t
vvt

(10)

M ie

vi nvdv

(17)

vmax

With Eq. 8 and Eq. 10 can be written as nvt t, t t nvt, t t nv, t t v

G v t t
vvt

(11)

Substituting Eq. 9 and Eq. 11 into Eq. 7 yields nvt, t nvt, t t nv, t t v Gv v

nvt, t t nvt, t t v

Gv v vvt

G v t t
vvt

t
vvt

where vmax is the upper limit of the nite domain. In practice, the population density n ( v ) tends to zero at sufciently large particle volumes, so vmax can be chosen to be sufciently large such that M e i is negligibly small. In this method, the truncated domain v [0, vmax] is partitioned into N discrete and contiguous elements, that is, we subdivide it into subintervals with common endpoints, called nodes. The time step is set to t . We use ( v j , i , n j , i ) to represent a node on the particle distribution prole on the v n -coordinate plane. The index j denotes the time j t , and the index i 0, 1, . . . , N indicates the series number of the nodes on the particle distribution prole. Note that only the initial truncated domain (at t 0) needs to be partitioned, and the nodes on the v n -coordinate plane at other time can be approximated successively using Eq. 8 and Eq. 16 which imply v j1,i v j,i G v j,i t n j,i v j,i t Gv 1 t v vvj,i (18)

G v t t 2
vvt

v t t which can be written as nvt, t t nvt, t t nvt, t Gvt t v Gv nvt, t t v nv, t t v

(12)

n j1,i

(19)

vvt

vt

for i 0, 1, . . . , N and j is a nonnegative integer.

vvt

Gv v vvt

Gvtt
vvt

(13)

Approximation of birth and death terms In Eq. 19, the term ( v j , i ) still needs to be solved. We use
cubic spline interpolation to nd additional nodes we may need in this section, while ( n j ,1 n j ,0 )/( v j ,1 v j ,0 ), ( n j , N n j , N 1 )/( v j , N v j , N 1 ) are used as the end slopes for the cubic spline. Gauss-Legendre quadrature is used to perform integrations over each element [ v j , i , v j , i 1 ] (Hildebrand, 1956). An approximation to the integral

If t 3 0, Eq. 13 becomes nv, t Gv nv, t Gv nv, t v t v v which is equivalent to Eq. 1 if (14)

v B nuc v B agg v D agg v B br v D br v

(15)

f x dx

therefore, we can use Eq. 7 to solve Eq. 1. Using Eq. 8 and Eq. 9, Eq. 7 can be changed to n v t , t v t t nvt t, t t Gv 1 t v vvt

i1

w
N

N,i

f z N,i

(20)

(16)

is obtained by sampling f ( x ) at the N unequally spaced abscissas z N ,1 , z N ,2 , . . . , z N , N , where the corresponding weights are w N ,1 , w N ,2 , . . . , w N , N . The abscissas and weights for GaussLegendre quadrature can be computed analytically for small N . For N 5
Vol. 51, No. 11 AIChE Journal

3002

November 2005

Case Studies
f x dx w 1f z 1 w 2f z 2 w 3f z 3 w 4f z 4 w 5f z 5 (21) The method was tested for six cases of the PBE, where analytical solutions are available. The numerical results were compared with analytic solutions. The CPU time reported corresponds to a desktop computer with 1.6 GHz Intel Pentium 4 processor.

where 1 1 322 13 70 245 14 70, w1 z1 21 900 z2 1 1 245 14 70, w2 900 322 13 70 21 z 3 0, 1 21 1 21 w3 128 225 1 322 13 70 900

Aggregation and size-dependent growth

By letting the nucleation rate B nuc ( v ), breakage function (v, v), and the rate of breakage S ( v ) to be zero, the PBE for aggregation and size-dependent growth is n v , t G v n v , t 1 t v 2 vtnv, tdv nv, t

z4

245 14 70, 245 14 70,

w4

v v , v n v

v , v n v , t d v

(24)

The initial distribution is z5 w5 1 322 13 70 900 nv, 0 v N0 exp v0 v0

(25)

For ease of reading the earlier list the notation z i and w i instead of z N , i and w N , i , has been used respectively. To apply the rule over the interval [ a , b ], use the change of variable t ab ba x 2 2 and dt ba dx 2

and size-dependent growth rate is G v 1v (26)

Case A1. For the size-independent aggregation kernel

v , v 0

(27)

1 then the relationship b a f ( t ) dt 1 f (( a b )/ 2 (( b a )/ 2) x )(( b a )/ 2) dx is used to obtain the quadrature formula

the analytical solution has been derived by Ramabhadran et al. (1976) nv, t
2 M0 M0 exp v M1 M1

ba f t dt 2

i1

ab ba z N,i w N,if 2 2

(28)

where M 0 ( t ) and M 1 ( t ) are the rst two moments M 0 t

If there is no aggregation and breakage, the solution Eq. 19 becomes n j1,i n j,i B nuc v j,i t Gv 1 t v vvj,i

nv, tdv

2N0 2 0N 0t

(29)

(22) M 1 t

v n v , t d v N 0v 0exp1 t

(30)

then the accuracy of the solution can be increased by simply selecting a small t . Furthermore, if the growth rate is volumeindependent, that is, G / v 0, (Eq. 22) can be simplied to n j1,i n j,i B nuc v j,i t (23)

which indicates that accurate solution can be obtained. This is a very useful property for some batch crystallization processes.
AIChE Journal November 2005

The constant N 0 , v 0 , 1 and 0 were set to 1. The truncation point (vmax) was chosen as 10, and the initial truncated domain was evenly partitioned into 20 discrete elements with common ends. The time step t was chosen to be 0.02. The problem (Eq. 24) was solved using the proposed method. The particle density distributions for the initial time, t 0.5, t 1 and t 2 are shown in Figure 2. The solid lines represent the analytical solution, while the symbols are the
Vol. 51, No. 11 3003

Figure 2. Aggregation with constant kernel (v, v) 1, and size-dependent growth rate G(v) v.

Figure 3. Aggregation with sum kernel (v, v) v v, and size-dependent growth rate G(v) v.

nodal values obtained by the proposed method. The numerical and analytical solutions are in excellent agreement. Case A2. For the size-dependent aggregation kernel

nv, t

2 M0 M1

1 M 0/ N 0

exp

N0 v sinh M1

M0 N0 v N0 M1

v , v 1 v v
The analytical solution is (Ramabhadran et al., 1976)

(31)

(36) where M 0 and M 1 are dened in Eq. 29 and Eq. 30, respectively. All constants and truncation are the same as in Case A1. Figure 4 compares the numerical solution with the analytical solution. The solids lines represent the analytical solution, while the symbols are the nodal values. The agreement is quite good. Errors in the rst moment of the numerical solution and CPU time are recorded in Table 1. The accuracy of the numerical solutions is satisfactory, as shown in Table 1, despite the use of a fairly coarse grid.

nv, t

2 M0 M0 2N0 M 0 N0 v exp 1 v I1 2 1 M1 M1 M0 N0 M1

M0 v M1

M0 N0

(32) where I 1 is the modied Bessel function of the rst kind of order one, and M 0 t N 0exp

1 N0 v0 1 exp1 t 1

(33) (34)

M 1 t N 0v 0exp1 t

The constant 1 was set to 1. The truncation point vmax was selected to be 30. The domain [0, vmax] was evenly partitioned into 30 elements. The time interval was 0.02. Comparison of simulation with the analytical solution is shown in Figure 3. The solids lines represent the analytical solution, and the symbols are the numerical results. Again, the numerical solution overlaps the analytical solution excellently. Case A3. If the initial distribution is nv, 0 v N 0v 2 exp v0 v0

(35)
Figure 4. Aggregation with constant kernel (v, v) 1, size-dependent growth rate G(v) v, and initial distribution (Eq. 35). Vol. 51, No. 11 AIChE Journal

and (v, v) and G ( v ) are the same as in Case A1, then the analytical solution is (Ramabhadran et al., 1976)
3004 November 2005

Table 1. Aggregation and Size-Dependent Growth Rate

Case A1 A2 A3 Time 1 1 1 t 0.02 0.02 0.02 No. of Elements 20 30 20 Error in M 1 % 0.15 0.46 0.28 CPU (s) 34 76 34

Breakage
The PBE for pure breakage is obtained by setting the growth rate G ( v ), the nucleation rate B nuc ( v ), and the aggregation kernel (v, v) to zero, that is nv, t Svnv, t t

v, vSvnv, tdv (37)

Case B1. If the probability that a particle breaks up into two pieces is independent of both the size of the object and of the pieces, then

Figure 6. Breakage with breakage rate proportional to the particle volume S(v) v2.

v , v
Sv v For the initial distribution

2 v

(38) (39)

analytical solutions, and the symbols represent the numerical solutions. The numerical solutions match the analytical solutions very well. Case B2. In this case, the rate of breakage is proportional to the size, and

n v , 0 expv the analytical solution is (Ziff and McGrady, 1985) n v , t 1 t 2expv1 t

(40)

v , v

2 v

(42) (43)

Sv v2 (41)

For the exponential initial distribution (Eq. 40), the analytical solution is (Ziff and McGrady, 1985) n v , t exptv2 v1 2t1 v (44)

Figure 5 shows the initial distribution and the distributions at t 0.5, and t 1 for this case. The solid line represents the

Distributions at different moments for this case are shown in Figure 6. The numerical solutions are in good agreement with the analytical solutions. Parameters used for numerical solution, errors in the rst moment and CPU time were summarized in Table 2.

Nucleation and growth

For simultaneous nucleation and size independent growth the PBE is nv, t nv, t G B 0 v t v (45)

Table 2. Breakage Figure 5. Breakage with a binary breakage function (v, v) 2/v, and size-dependent breakage rate S(v) v. AIChE Journal November 2005
Case B1 B2 Time 1 1 t 0.02 0.02 No. of Elements 50 50 Error in M 1 % 0.50 0.55 CPU (s) 80 80

Vol. 51, No. 11

3005

methods can also be applied. Accuracy of the numerical solution can be simply improved by smaller time step and more partition elements of particle size domain without any stability problems, which makes this method particularly useful for developing hierarchical solution for particulate processes. A compromise between computational time and accuracy can be easily achieved by choosing the node resolution.

Literature Cited
Baterham, R. J., J. S. Hall, and G. Barton, Pelletizing Kinetics and Simulation of Full Scale Balling Circuits, in Proc. 3rd Int. Symp. on Agglomeration, Nurnberg, Germany (1981). Bennett, M., and S. Rohani, Solution of Population Balance Equation with a New Combined Lax-Wendroff Cranck-Nicholson Method, Chem. Eng Sci., 56, 6623 (2001). Gelbard, F., and J. H. Seinfeld, Numerical Solution of the Dynamic Equation for Particulate Systems, J. Comput. Phys., 28, 357 (1978). Hildebrand, F. B., Introduction to Numerical Analysis, McGraw-Hill, New York (1956). Hounslow, M. J., R. L. Ryall, and V. R. Marshall, A Discretized Population Balance for Nucleation, Growth, and Aggregation, AIChE J., 34, 1821 (1988). Hu, Q., S. Rohani, D. X. Wang, and A. Jutan, Nonlinear Kinetic Parameter Estimation for Batch Cooling Seeded Crystallization, AIChE J., 50, 1786 (2004). Hulburt, H. M., and S. Katz, Some Problems in Particle Technology. A Statistical Mechanical Formulation, Chem. Eng Sci., 19, 555 (1964). Kostoglou, M., and A. J. Karabelas, Evaluation of Aero Order Methods for Simulating Particle Coagulation, J. Colloid Interface Sci., 163, 420 (1994). Kumar, S., and D. Ramkrishna, On the Solution of Population Balance Equations by DiscretizationIII. Nucleation, Growth and Aggregation of Particles, Chem. Eng Sci., 52, 4659 (1997). Lee, M. H., A Survey of Numerical Solutions to the Coagulation Equation, J. Phys. A: Math. Gen., 34, 10219 (2001). Nicmanis, M., and M. J. Hounslow, Finite-Element Methods for SteadyState Population Balance Equations, AIChE J., 44, 2258 (1998). Prasher, C. L., Crushing and Grinding Process Handbook, Wiley, New York (1987). Ramabhadran, T. E., T. W. Peterson, and J. H. Seinfeld, Dynamic of Aerosol Coagulation and Condensation, AIChE J., 22, 840 (1976). Ramkrishna, D., The Status of Population Balances, Rev. Chem. Eng., 3, 49 (1985). Randolph, A. D., and M. A. Larson, Theory of Particulate Processes, 2nd ed., Academic Press, New York (1988). Raphael, M., S. Rohani, and F. Sosulski, Isoelectric Precipitation of Sunower Protein in a Tubular Precipitator, Can. J. Chem. Eng., 73, 470 (1995). Rigopoulos, S., and A. G. Jones, Finite-Element Scheme for Solution of the Dynamic Population Balance Equation, AIChE J., 49, 1127 (2003). Vanni, M., Approximate Population Balance Equations for AggregationBreakage Processes, J. Colloid Interface Sci., 221, 143 (2000). Ziff, R. M., and E. D. McGrady, The Kinetics of Cluster Fragmentation and Depolymerisation, J. Phys. A: Math. Gen., 18, 3027 (1985).
Manuscript received Feb. 17, 2004, and revision received Feb. 15, 2005, and nal revision received May 3, 2005.

Figure 7. Nucleation and growth.

where B 0 and G are constant. The analytical solution for this model is (Hounslow et al., 1988) nv, t v B0 u t G G

(46)

where u is the unit step function. With G 1 and B 0 1, comparison between the numerical solution with the analytical solution is shown in Figure 7, where the solid line represents the analytical solution, and the symbols are for the numerical solution. It should be noted that the proposed method provides absolutely accurate solution in this case and is better than any other existing numerical methods. The CPU time was only 0.02 s.

Conclusions
In this work, a new numerical method for solving dynamic PBE was proposed. The new method is able to accurately predict the solution of PBE involving any combination of particulate system subprocess, that is, nucleation, growth, aggregation and breakage, with reasonable computational requirement. The proposed approach is free from stability and dispersion problems of other numerical methods such as niteelement method. The numerical scheme requires a partition of the particle-size domain, but there are no restrictions on the location of nodes when using this method. An evenly spaced partition was used in this method, although other partition

3006

November 2005

Vol. 51, No. 11

AIChE Journal

Integration of Data Uncertainty in Linear Regression and Process Optimization

Marco S. Reis and Pedro M. Saraiva
GEPSI-PSE Group, Dept. of Chemical Engineering, University of Coimbra, Po lo IIPinhal de Marrocos, 3030-290 Coimbra, Portugal DOI 10.1002/aic.10540 Published online August 9, 2005 in Wiley InterScience (www.interscience.wiley.com).

Data uncertainties provide important information that should be taken into account along with the actual data. In fact, with the development of measurement instrumentation methods and metrology, one is very often able to rigorously specify the uncertainty associated with each measured value. The use of this piece of information, together with raw measurements, shouldin principlelead to more sound ways of performing data analysis, empirical modeling, and subsequent decision making. In this paper, we address the issues of using data uncertainty in the task of model estimation and, when it is already available, we show how the integration of measurement and actuation uncertainty can be achieved in the context of process optimization. Within the scope of the rst task (model estimation), we make reference to several methods designed to take into account data uncertainties in linear multivariate regression (multivariate least squares, maximum likelihood principal component regression), and others whose potential to deal with noisy data is well known (partial least squares, principal component regression, and ridge regression), as well as modications of previous methods that we developed, and compare their performance. MLPCR2 tends to achieve better predictive performance than all the other tested methods. The potential benets of including measurement and actuation uncertainties in process optimization are also illustrated. 2005 American Institute of
Chemical Engineers AIChE J, 51: 30073019, 2005

Keywords: measurement uncertainty, multivariate least squares, maximum likelihood principal component regression, partial least squares, principal component regression, optimization under uncertainty

Introduction
The large amounts of industrial and laboratorial data generated in the chemical process industries and stored in databases do have a substantial potential to set the ground for further process improvement and optimization. Noting that this potential is not always being fully developed and that the goals that were present at the conception of such databases are often not being achieved, numerous efforts have been made and documented in the literature toward a more effective use of these
Correspondence concerning this article should be addressed to M. S. Reis at marco@eq.uc.pt.

2005 American Institute of Chemical Engineers

information resources, that is, in the elds of process monitoring,1,2 fault detection and diagnosis,3 and data mining.4 However, quite often these approaches do not explicitly and quantitatively take into account data quality, or do so only in an implicit or tacit way. Following the efforts undertaken in the metrology eld, with respect to the characterization and quantication of measurement uncertainty, in a rigorous and normalized approach,5 we believe it is quite appropriate and timely to develop and apply methods that explicitly and consistently take into account this important piece of information. Measurement uncertainty is a well-dened quantity and there are well-documented standardized procedures that assist its specication or estimation. Basically, uncertainty is dened as a parameter associated with the result of a measurement
Vol. 51, No. 11 3007

AIChE Journal

November 2005

that characterizes the dispersion of the values that could reasonably be attributed to the measurand. 5 The standard uncertainty u (to which we will often refer simply as uncertainty) should be expressed in terms of a standard deviation of the values obtained under the same experimental conditions, and can be obtained either from the analysis of collected data (the so-called Type A evaluation) or through other adequate means (Type B evaluation). The availability of raw values, along with their associated uncertainties, implies that we should not have only one data table available for analysis, but in fact two (one with raw values and another one with the corresponding measurement uncertainties). Therefore, with this additional information at our disposal, we should be able to take advantage of it through its integration into our data analysis tasks. For instance, data reconciliation6-8 is designed to handle noisy measurements, to adjust raw data in some optimal way, so that it conforms to conservation laws and other constraints. The fact that the objective function to be minimized consists of quadratic terms involving the inverse of variance covariance matrices of measurements7 indicates that uncertainty information is in fact being considered in data reconciliation. However, there are many application scenarios where no conservation laws are available to perform preliminary data reconciliation, such as the analysis of spectra, microarray data, and laboratorial data sets. Furthermore, uncertainty-based methods can be applied to data sets after reconciliation or ltering. Sometimes it happens that uncertainty associated with measurements is sufciently small for the techniques that disregard it completely or treat it in a very simplied way (such as assuming homoscedastic behavior), still holding as adequate. However, these are tacit assumptions, quite often not veried or clearly stated. The main purpose of this article is to bring the issue of data uncertainty into the priorities for the data analyst, which should explicitly address it in a preliminary phase, as well as to present, develop, and test procedures that do exploit and take advantage of data uncertainty information. In particular, we will address the use of uncertainty information in two different tasks: model estimation and process optimization. In the next section, we refer several methodologies with the potential of integrating uncertainty in the estimation of parameters for a multivariate linear model. This type of model is widely used in the analysis of industrial data sets, and its prediction ability, when parameters are estimated by different methodologies, is thus an important issue in practical applications. In the following section, a complementary situation regarding the use of uncertainties, that is, when a model is considered to be known, is illustrated under the context of process optimization. Then, in the fourth section, we present two case studies that provide the ground for comparison among all the methods referenced in the second section and another case study that illustrates the methodology presented in the third section. We end this paper with a discussion section, where some computational issues are addressed (fth section) and some nal conclusions are drawn. Apart from the comparative study undertaken in the case study section, new methods (unc-PLS3, unc-PLS4, and unc-PLS5) are also presented and tested. Methods MLMLS, unc-PLS1, unc-PLS2, MLPCR2, rMLS, and rMLMLS are carefully described elsewhere.9 The formulations presented in the third section provide also a contribution to the explicit consideration of measurement uncertainties for process optimization.
3008 November 2005

Table 1. Formulation of the Optimization Problems Underlying OLS and MLS Methods
OLS OLS arg min b

b b 0 b p T

MLS

MLS arg min b

b b 0 b p T

n i1 n i1

[y(i) y (i)]2

[y(i) y (i)]2 2 se (i)

(1)

(2)

MLMLS b

MLMLS arg max b b 1 n ln2 2

i1

b b 0 b p T

ln i

1 2

i1

[y(i) y (i)]2 2 i

(3)

Measurement Uncertainties in Model Estimation

This section is devoted to the description of four groups of multivariate linear regression methods that have the potential to accommodate measurement noise information, either explicitly or implicitly. As already mentioned, our focus on multivariate linear regression methods arises from the quite widespread use of this type of approaches in the development of input/output models for industrial and/or laboratorial applications. The several methodologies here addressed are combined under four separate groups, according to their afnity: ordinary least squares (OLS), ridge regression (RR), principal component regression (PCR), and partial least squares (PLS, also referred to as projection to latent structures). These four basic methods do not explicitly incorporate measurement uncertainty information, so that several alternatives already developed are also presented, as well as other recent modications that we propose here and do take uncertainty information explicitly into consideration.

OLS group
Ordinary least squares (OLS) and multivariate least squares (MLS)10,11 parameter estimates for a linear regression model are the solutions of the optimization problems formulated in Eqs. 1 and 2 of Table 1. OLS tacitly assumes a homoscedastic behavior (that is, with constant variance) for the noise error term in the standard linear regression model. On the other hand, MLS is built on an error in variables (EIV) functional relationship relating true values of both the input and output variables, which are then affected by zero mean random errors with a given covariance structure (presumed to be known). In the denominator of Eq. 2 we can nd a term, s 2 e (i), that results from the summation of the uncertainties associated with the response to those arising from the propagation of uncertainties of the predictors to the response, according to a formula derived from error propagation theory10,12:
Vol. 51, No. 11 AIChE Journal

Table 2. Formulation of the Optimization Problems Underlying RR, rMLS, and rMLMLS
RR RR arg min b

b b 0 b p T

rMLS

rMLS arg min b

b b 0

rMLMLS

rMLMLS arg min b

b b 0 b p T

n i1

b p T

n i1 n i1

(y(i) y (i))2 (y(i) y (i))2 2 se (i)

j1 p

p j1

b(j)2

b(j)2

ln(se (i))

i1

(5)

(6)

(y(i) y (i))2 2 se (i)

j1

b(j)2

(7)

2 se i uy i 2

j1

b uXi, j
p 2 j p

j2 kj1

b b cov i, i
p j k j k

(4)

where uX(i, j) and uy(i) are the uncertainties associated with the ith observation of the jth input and output variables, respectively, and j(i) is the random error affecting the ith measure j represents the coefcient of the linear ment of variable j; b regression model associated with variable j. The method whose objective function is presented in Table 1, Eq. 3, is derived from the analysis of the Berkson case (controlled regressors with error) within the scope of EIV models13,14 and under the assumption of Gaussian errors. The objective function arises from the maximization of the resulting likelihood function, and we included this approach in our present study given both the similarity between the quadratic functional part of its objective function and the one underlying MLS, and its simplicity. Because the solution for the Berkson case formulation is sometimes similar to MLS,14 we make reference to the above formulation maximum likelihood multivariate least squares (MLMLS), to stress the statistical origin of the underlying objective function.

regressed onto. These predictors are orthogonal and thus the collinearity problem is overcome if we disregard the linear combinations with small variability explanation power.19 After developing MLPCA, which estimates the PCA subspace in an optimal maximum likelihood sense, when data are affected by measurement errors with a known uncertainty structure,20 Wentzell et al.21 applied it in the context of developing a PCR methodology that incorporates measurement uncertainties (MLPCR). As in PCR, MLPCR consists of rst estimating a PCA model, now using MLPCA, to calculate the scores through nonorthogonal (maximum likelihood) projections to the estimated MLPCA subspace (instead of the PCA orthogonal projections), and then applying OLS to develop a nal predictive model. This technique makes use of the available uncertainty information in the former phases (estimation of a MLPCA model and calculation of its scores), but not during the stage at which OLS is applied. Therefore, Mart nez et al.10 proposed a modication to the regression phase, to make it consistent with the efforts of integrating uncertainty information carried out in the initial stages, which consists of replacing OLS by MLS (we will call this modication MLPCR1). To implement MLS in the second phase, estimated score uncertainties for the ith observation need to be calculated, being given by the diagonal elements of the following matrix10 Z i P T diaguXi, : 1 P 1 (8)

RR group
A well-known characteristic of the OLS method is the fact that the variance of its parameter estimates increases when the input variables become more correlated. Computational simulations showed us that the same applies to MLS. One possible way to address this issue consists of enforcing an effective shrinkage in the coefcients under estimation, following a ridge regression (RR) regularization approach. It basically consists of adding an extra term to the objective function that penalizes large solutions (in a square norm sense). Optimization formulations underlying RR estimates,15,16 as well as those proposed for its counterparts based on MLS and MLMLS, rMLS and rMLMLS, respectively (standing for ridge MLS and ridge MLMLS), are presented in Table 2. where diag is an operator that converts a vector into a diagonal matrix, and P is the matrix of maximum likelihood loads. In our study, we will compare these algorithms based on OLS and MLS (MLPCR and MLPCR1, respectively), with the one obtained when we use the MLMLS algorithm instead of MLS, in the second phase of MLPCR (MLPCR2).

PLS group
PLS17,18,22-27 is a widely used algorithm in the chemometrics community that also adequately handles noisy data with correlated predictors in the estimation of a linear multivariate model. As in PCR, PLS nds a set of uncorrelated linear combinations of the predictors, belonging to some lower-dimensional subspace in the X-variables space, where y is to be regressed onto. However, in PLS, this subspace is the one that, while still adequately covering the X-variability, provides a good description of the variability exhibited by the Y-variable(s). Here we will make reference to a pair of classes of PLS algorithms, one implemented from raw data and another based on covariance matrices.
Vol. 51, No. 11 3009

PCR group
PCR17,18 is another methodology that handles collinearity among predictor variables. It uses those uncorrelated linear combinations of the input variables that most explain input space variability [from principal components analysis (PCA)] as the new set of predictors, where the response is to be
AIChE Journal November 2005

3010

Table 3. PLS1 as a Succession of Optimization Subproblems (First Column) and Its Counterparts That Make Use of Information Regarding Measurement Uncertainties

November 2005

Vol. 51, No. 11

AIChE Journal

Table 4. SIMPLS Algorithm43

S X TX s X Ty for a1, . . . ,A r 1 st left singular vector of s r r /( r T Sr ) 1/2 R [R,r] P [ P , Sr ] s [ I P ( P TP ) 1P T ]s end T XR

PLS algorithms implemented directly from raw data

The algorithmic nature of PLS22,26 can be translated into the solutions of a succession of optimization subproblems,17,18,23 as presented in the rst column of Table 3 for one of its common versions, relative to the case of a single response variable (PLS1). However, if besides having available raw data, [X y], we also know their respective uncertainties, [uX uy], then one way to incorporate this additional information into a PLS algorithm is through an adequate reformulation of the optimization subtasks. Therefore, we have modied the objective functions underlying each optimization subproblem to incorporate measurement uncertainties, but still preserving the successful algorithmic structure of PLS. Such a sequence of optimization subproblems is presented in the second and third columns of Table 3, where MLS and MLMLS replace OLS in several algorithmic stages, giving rise to the uncertainty-based equivalents unc-PLS1 and unc-PLS2, respectively.

base of a(s; S). The relevancy of S and s for PLS provided the motivation to direct some efforts toward the incorporation of uncertainty information in the computation of better estimates for both of these matrices. The reason that we have not called them estimates so far is explained by the lack of a consistent statistical population model underlying PLS.24,31,32 However, when we now say that our goal is to calculate better covariance matrices, this implies that some goodness criteria must be assumed. Therefore, to give a step forward toward the integration of measurement uncertainties in our analysis, one should postulate a statistical model to provide an estimation setting for the covariance matrices S and s. For the sake of the present work, we consider the following latent variable multivariate linear relationship for Z [xT y]T, which has the ability to incorporate heteroscedastic measurement errors with known uncertainties (these uncertainties are considered by now to be independent of the true levels for the noiseless measurands) Z k Z A l k m k (10)

where Z is the (m 1) 1 vector of measurements, Z is the (m 1) 1 mean vector of x, A is the (m 1) a matrix of model coefcients, l is the a 1 vector of latent variables, and m is the (m 1) 1 vector of measurement noise. This model is still incomplete because we need to provide it with the probability density functions assumed for each random component l k iid MNa 0, l , m k id MNm 1 0, m k lk and m j
are independent @ k , j

PLS algorithms implemented from covariance matrices

There are several alternative ways to develop a PLS model, most of them leading to very similar or even exactly the same results. In fact, Helland25 has shown the equivalence between two of such algorithms (one based on orthogonal scores and another using orthogonal loadings instead), both of them based on available raw data matrices for the predictors and response variables. Another class of PLS methods that encompasses the so-called SIMPLS, developed by Sijmen de Jong (see Table 4), or the approach presented by Kaspar and Ray,28 built on previous work from Ho skuldsson,29 consists of algorithms entirely based on data covariance or cross-product matrices. For the single response case, a SIMPLS solution provides exactly the same results as Svant Wolds orthogonalized PLS algorithm, leading to only minor differences when several outputs are considered. Matrices S and s in Table 4 do play a central role in PLS. Theoretical analysis of this algorithm25,30 leads to the conclusion that the calculated vector of coefcients, when a PLS a latent variables are considered, , is given by
a T T PLS Va Va SVa 1 Va s

(11)

where MN stands for multivariate normal distribution, i is the covariance matrix of the latent variables, m(k) is the covariance matrix of the measurement noise at time k, given by 2 m(k) diag[ m (k)]. Thus, for estimating the covariance matrix, we assume a multivariate behavior for Z that can be adequately described by propagation of the underlying variation of p latent variables, plus added noise in the full variable space. This model and the calculation details associated with the estimation of the unknown parameters are fully described elsewhere.33 It can be shown that the probability density function of Z, under the conditions stated above, is a multivariate normal distribution with the following form Z k id MNm 1 Z , Z k where Z k l m k l A lA T (13) (12)

(9)

where Va [v1, v2, . . . , va] is any (m a) matrix whose columns span the following Krylov subspace, a(s; S), that is, the subspace generated by the rst a columns of the Krylov sequence, {s, Ss, . . . , Sa1s}. Thus, matrices S and s dene the structure of the relevant Krylov subspace where the PLS solution will lie. In fact, the columns of the PLS weighting matrix W, which dene the subspace of the full predictor space with maximal covariance with the response, do form an orthogonal
AIChE Journal November 2005

With the raw measurements (Z) and the associated uncertainties [from which we can calculate m(k)], it is possible to estimate Z and l by maximizing the likelihood function. Matrix Z(k) l m(k) is the estimate of the covariance matrix for noisy measurements at time step (k), but because PLS is based on S and s, it requires single estimates for the population parameters (and not one per time step k). Thus, we , maintain the estimate of the covariance of noiseless data, l
Vol. 51, No. 11 3011

but average out the heteroscedastic square uncertainties, to come up with a single term, m, leading to m Z l (14)

With the estimate of Z, we can nally calculate the esti (1 : m, 1 : m), s Z (1 : m, m 1). mates for S and s: S Z The algorithm that consists of implementing the SIMPLS algorithm with these matrices as inputs will be referred to here as unc-PLS3. In the present context, we use the full measurement space to estimate Z (a m) because we want the relevant subspace for prediction to be dened by the PLS algorithm itself, and not by a previous estimation step. In the prediction phase, when new values for the predictors become available along with their measurement uncertainties, and the goal is to predict what the value of the response variable would be, we add an additional calculation step before applying the uncPLS3 regression vector (calculated in the estimation phase). This step consists of projecting the new multivariate observation in the full X-space into the subspace that is relevant for predictions (that is, the one spanned by the columns of the weighting matrix, W in PLS or R in SIMPLS). The availability of the associated uncertainties leads to a generally nonorthogonal projection methodology that consists of estimating the projected points using a maximum likelihood approach, just as the one adopted in MLPCA.21 In the present study, we also tested an algorithm that implements the same nonorthogonal projection operation, but using the weighting matrix provided by PLS (a hybrid version of the classic PLS because it contains a projection step that incorporates measurement uncertainty), herein referred to as unc-PLS4. For the sake of completeness, we also introduced another methodology, based on the same weighting matrix as unc-PLS3, but that bypasses the nonorthogonal projection step, designated as unc-PLS5.

Figure 1. Schematic representation of measured quantities [as seen by an external operator and marked with a tilde ()] and the quantities that are actually involved in the underlying process.

) does not correspond to the exact true value lated variables (Z of the manipulation action over the process. In fact, because of actuation noise, there is also here another uncertainty source to be taken into account. Considering that we want to drive the process in such a way as to minimize some relevant cost function, , we propose the following formulation that incorporates measurement and actuation uncertainties, in the calculation of the adequate values for the manipulated variables to be specied externally, ). As when a given measurement for the load is acquired (L often happens in the formulation of optimization problems under uncertainty, the objective function constitutes an expected value for the performance metric, taken over the space of uncertain parameters: Formulation I min E L, Z, Y
Z

Measurement Uncertainties in Process Optimization

In the previous section we have addressed the explicit incorporation of measurement uncertainty in statistical model development. We now move to a different working scenario, where an appropriate model is already available and our goal is to use it for process optimization, but also taking into account information regarding measurement and actuation uncertainties. In particular, we address the problem where one wants to optimize an objective function (such as maximizing some prot metric or minimizing a cost function), for a given measurement of the vector of load variables (L), by manipulating another set of variables (M). However, because of the presence of uncertainties, the following issues do arise: and the outputs Y ) Measured quantities (that is, the loads L are affected by measurement noise, with statistical characteristics dened by their associated uncertainty L L L Y Y Y (15)

s.t. gY, L, Z 0 L LL Z ZZ Y Y Y where E{ } is the expectation operator, E

(16)

j d

(17)

with quantities marked with a tilde accent () being the values actually available, whereas L and Y are the corresponding true, but unknown, values for these quantities (Figure 1). Similarly, the set-point that we specify for the manipu3012 November 2005

T T T T [ L , Z , Y ] and j() provide the joint probability density function for the uncertain quantities . The available model is represented by g(Y, L, Z) 0, and we will assume here that the uncertainty associated with its parameters is negligible (if not, such uncertainties can also be incorporated into our problem formulation34). In Formulation I, we assume that the relevant quantities for evaluation of the performance metric are the values of L and Z that really affect the process, as well as the measured value of the output. We point out that these assumptions do not necessarily hold in every situation. For instance, sometimes the performance metric should be calculated with the true value ] (Formulation II, see below), as is of the output, Y, instead of Y the case when output measurements become available with much less uncertainty in a subsequent stage (such as from

Vol. 51, No. 11

AIChE Journal

off-line laboratory tests). Other times, only measured values should be used because no better measurements or reconciliation procedures can be adopted. The correct formulation is therefore case dependent, and should be tailored to each particular situation. Formulation II min E L, Z, Y
Z

s.t. gY, L, Z 0 L LL Z Z Z

(18)

In our case studies section we will also present the results obtained for the situation where uncertainties are not at all taken into account, and thus where the manipulated variable values are found by solving the following problem Formulation III , Z , Y min L
Z

, L , Z 0 s.t. gY

(19)

identity matrix of dimension a). The dimension of the input space is set equal to 10 and that of the output space equal to 1 (m 10, k 1). Rows of the P matrix form an a-orthonormal set of vectors with dimension m. The same applies to matrix Q, which consists of an a-orthonormal set of vectors with dimension k. Each element of matrices E and F of random errors is drawn from a normal distribution with zero mean and standard deviation given by the uncertainty level associated with that specic variable (column of X or Y) for a particular observation (row). These uncertainties were allowed to vary, and this variation is characterized by the heterogeneity level (HLEV), which measures the degree of variation or heterogeneity of uncertainties from observation to observation: HLEV 1 means a low variation of the noise uncertainty or standard deviation from observation to observation, whereas HLEV 2 means a highly heteroscedastic behavior for the noise uncertainties. More specically, for variable Xi the uncertainties along the observation index are randomly generated from a uniform distribution centered at u (Xi) (the average uncertainty for a given variable), with range given by R(HLEV) K2(HLEV) u (Xi), where K2 0.01 (if HLEV 1; low heterogeneity level) or K2 1 (if HLEV 2; high heterogeneity level), that is, ( X i) u X i k U u

Case Studies
In this section we present the results reached from comparative analysis encompassing all the methods mentioned above (PLS, unc-PLS1, unc-PLS2, unc-PLS3, unc-PLS4, unc-PLS5, RR, rMLS, rMLMLS, PCR, MLPCR, MLPCR1, MLPCR2, OLS, MLS, and MLMLS), and illustrate the implementation of the approach treated in the third section under a realistic simulation scenario, using a model estimated from a real paper pulp pilot digester. Case studies 1 and 2 provide different contexts to set a ground for the comparison study among the multivariate linear regression methods. In both of them, a latent variable model structure is adopted to generate simulated data, given that this kind of model structure is quite representative of data collected from many real industrial processes because the number of inner sources of variability that drives process behavior is usually of a much smaller dimensionality than the number of measured variables.35,36 The latent variable model used has the following form X 1 n T X TP E T Y 1 n Y TQ F

R ( HLEV ) R(HLEV) ,u (Xi ) 2 2

In the present study, u (Xi) was kept constant at 0.5 times the theoretical standard deviation calculated for each noiseless variable.

Case study 1: complete heteroscedastic noise

With the goal of evaluating overall performance of the methods under different uncertainty structures for the measurements errors, the following sequence of steps was adopted: (1) We set the tuning parameters for each method and for each set of conditions (number of latent dimensions for PLS and PCR methods, and ridge parameter for RR methods). Regarding PLS and PCR methods, we did set a 5. As for ridge methods, we selected our ridge parameter using crossvalidation and the generation of a logarithmic grid in the range of plausible values (the criterion used in cross-validation is RMSEPW). This procedure is repeated 10 times, and the median of the best values is chosen as the tuning parameter to be used in our simulations. Variables are auto-scaled in all methods, except for OLS, MLS, and MLMLS. (2) For each scenario of HLEV (1 or 2), two noiseless data sets are generated according to the latent variable model presented above: a training or reference noiseless data set and a test noiseless data set, both with 100 multivariate observations. Furthermore, a random sequence of uncertainties (noise standard deviations) for all the observations belonging to each variable is generated according to HLEV. (3) Zero-mean Gaussian noise, with standard deviation given by the uncertainties calculated in (2), is generated and added to the noiseless training and testing data sets, after which a model is estimated according to each linear regression method (using the training data set) and its prediction perforVol. 51, No. 11 3013

(20)

where X and Y are the m 1 and k 1 vectors with the column averages of X and Y; 1n is an n 1 vector of ones; X is the n m matrix of input data; Y is the n k matrix of output data; T is the n a matrix of latent variables that constitute the inner variability source, structuring both the input and output data matrices; E and F are n m and n k matrices of random errors; and P and Q are a m and a k matrices of coefcients. The model used in our simulations consists of ve latent variables (a 5) that follow a multivariate normal distribution with zero means and a diagonal covariance (Ia, that is, the
AIChE Journal November 2005

Figure 2. Results for number of losses, ties, and wins for each method, under the simulation scenario with heterogeneity level (HLEV) 1 [using root mean square error of prediction (RMSEP)].

mance evaluated (using the test data set). This process of noise addition, followed by parameter estimation and prediction, is repeated 100 times, and the corresponding performance metrics saved for future analysis. Performance metrics used for prediction assessment are the square root of the weighted mean square error of prediction in the test set (RMSEPW), where the weights are the result of combining the predictor and response uncertainties, and the more familiar root mean square error of prediction (RMSEP)

RMSEPWi

1 n
n k1

[ y(k) y (k)]2 uy(k) [uX(k, :)*2 ]T B*2

2

i 1100 (21)

RMSEP i

1 n
n k1

[ y(k) y (k)]2

i 1100

(22)

where n is the number of observations in the test set. At the end of the simulations, we do have 100 values for the above metrics available for comparing the performances achieved by the different methods, under a given noise structure scenario. To take into account both the individual variability of the performance metrics for the different methods, as well as their mutual correlations, we based the comparison strategy in paired t-tests among all the different combinations of methods. Therefore, for each simulation scenario, paired t-tests were used to determine whether method A is better than method B (a Win for method A), performs worse (a Loss), or if there is no statistical signicant difference between both of methods A and B (a Tie), for a given signicance level (we used 0.01). For the sake of simplicity, we will only present here the number of wins, losses, and ties that each method obtained for each simulation scenario. Figure 2 presents the comparison results for the scenario
3014 November 2005

HLEV 1, using RMSEP as performance metric (because the trends for RMSEP and RMSEPW do not differ signicantly, only those for the more familiar RMSEP are presented). Examining rst the performance of the methods belonging to the same group, we can see the following for this simulation scenario: OLS Group. MLS performs worse than OLS and MLMLS shows the best performance among the three methods. In general terms, comparing all the methods where MLS and MLMLS have similar roles (such as unc-PLS1/unc-PLS2, rMLS/rMLMLS, MLPCR1/MLPCR2), the second version never resulted in worse results and, as a matter of fact, almost always signicantly improved them. RR Group. Both rMLS and rMLMLS conducted to improved results with respect to those obtained by RR. PCR Group. MLPCR does not improve over PCR predictive results, but MLPCR2 leads to an improvement. PLS Group. Methods unc-PLS3 and unc-PLS5, both using uncertainty-based estimation of the relevant covariance matrices for PLS, present the best performance. Their similar performance results can be explained by the fact that, under mild homoscedastic situations and if the variables present approximately equal uncertainties associated with them, the orthogonal and nonorthogonal projections almost coincide. The same applies for the comparison of PLS and unc-PLS4, both using PLS weighting vectors but different projection strategies. Comparing the results obtained for all the methods against each other, we can see that MLPCR2 is the one that presented the best overall performance, followed by PCR, MLPCR, uncPLS3, and unc-PLS5. Figure 3 summarizes the results obtained for condition HLEV 2. A comparison of performances regarding methods within the PLS group shows that those methods that estimate the covariance matrices using uncertainty information (uncPLS3, unc-PLS5) present better performance then their counterparts that use the same projection strategies (unc-PLS4, PLS, respectively). However, looking now to the methods that differ
Vol. 51, No. 11 AIChE Journal

Figure 3. Results for number of losses, ties, and wins for each method, under the simulation scenario with HLEV 2 (using RMSEP).

only on the projection methodology, we can see that those that are based on orthogonal projections achieve better results that those based on nonorthogonal maximum likelihood projections. This result is quite interesting and will be further discussed below. In the PCR group we can see that all the methods perform quite well. As for the remaining groups of methods, the trends mentioned for HLEV 1 remain roughly valid. MLPCR2 continues to be the method with the best overall performance, followed by MLPCR and a group of methods that include MLPCR1, PCR, and unc-PLS5.

Case study 2: handling missing data

In this second case study, we analyze the prediction performance of the several methods when missing data are present (both in model estimation and in prediction), and a very simple strategy for handling missing data is adopted: mean substitution. For uncertainty-based methods, one also has to specify the associated uncertainty, and the values we have considered here are the standard deviations of the respective variables during normal operation. Other more sophisticated methodologies for missing data imputation during model estimation are also available for regression methods (especially PLS and PCR37), as well as methods for handling missing data once we have already available an estimated model.38 Analogous approaches can also be developed for the uncertainty-based techniques that require only the estimated value and the respective uncertainty to ll existing blanks. However, the aim of this study is to assess the extent to which one can easily handle missing data in model estimation and prediction (that is, with minimum assumptions regarding missing values and the least modication over standard procedures), taking advantage of the possibility of using uncertainty information. That being the case, we decided to keep the same replacement strategy among all methods, so that the real advantage of handling such an additional piece of information, provided by measurement uncertainties, can be easily evaluated and compared with the current alternatives.
AIChE Journal November 2005

Because our focus here is related with the evaluation of the methods regarding prediction when missing data is present, we adopted a simulation structure which is now different from that of case study 1. For each simulation the following steps are repeated and the corresponding results saved: (1) Generate a new latent variable model (matrices Q and P) and noiseless data to be used for model estimation and prediction assessment. Also generate measurement uncertainties to be associated with each nonmissing value, according to the value of HLEV used in each simulation study. (2) Generate a new missing data mask that removes (on average) a chosen percentage of the data matrix [X Y]. We used a target percentage of 20%, both for the reference and test data sets. (3) Generate and add noise to the noiseless data that were not removed, according to the measurement uncertainties generated in (1). (4) Replace missing data with column means for the data set used to estimate the model, and calculate the associated uncertainties using the columns standard deviations, for the same data set. (5) Estimate models using the data set constructed in (4). (6) For the test data set, do the same operation as in (4). (using the same values for the input values and uncertainties) and calculate the predicted value for the output variable. Calculate overall performance metrics (RMSEPW and RMSEP). The results obtained with HLEV 1 are presented in Figure 4, where we can see that within the PLS group methods unc-PLS5 and unc-PLS3 lead to improved predictive performances, but now with unc-PLS3 presenting better results that unc-PLS5, that is, the nonorthogonal projection seems to bring some added value when missing data are present, under homoscedastic scenarios. In the PCR group, all MLPCR methods outperform the conventional PCR. As for the other groups, results obtained follow the same trends veried when no missing data were present. In global terms, MLPCR2 presents the
Vol. 51, No. 11 3015

Figure 4. Results for number of losses, ties, and wins for each method, under the simulation scenario with HLEV 1 and 20% of missing data (using RMSEP).

best overall performance, followed by MLPCR1, MLPCR, and unc-PLS3. By analyzing the results for HLEV 2 (Figure 5), we can also see that unc-PLS3 and unc-PLS5 still show the best predictive performance within the PLS group, but now with unc-PLS3 presenting lower scores relatively to the previous scenario (HLEV 1), a result that is consistent with what was veried in case study 1. In the global comparison, after MLPCR2 we can nd MLPCR1 and MLPCR. Therefore, under the conditions adopted for this simulation study, we can conclude that MLPCR methods tend to have the best overall performance in the presence of missing data. We point out that when adopting a methodology that integrates data uncertainty, one follows the same calculation pro-

cedure adopted for the situation where no data are missing, simply replacing the missing elements with rough estimates that will be properly weighted by the algorithms, according to their associated uncertainties. However, if we do have available better estimates, such as those arising from more sophisticated imputation techniques, one can also integrate them as well, without any further changes.

Case study 3: process optimization under data uncertainty

This case study illustrates the integration of measurement uncertainties in process optimization decision making. The problem we address herein consists of calculating the values

Figure 5. Results for number of losses, ties, and wins for each method, under the simulation scenario with HLEV 2 and 20% of missing data (using RMSEP). 3016 November 2005 Vol. 51, No. 11 AIChE Journal

Table 6. Solutions Obtained under Formulations I, II, and III, and Their Associated Average Costs
Average Cost ($) Solutions 7.16 S 1602.0 H 7.83 II S 1184.2 H 5.38 III S 1274.6 H I Formulation I 10.80 11.16 25.46 Formulation II 5.93 5.40 8.17

Figure 6. Cost function for deviations of total yield (TY) from its target value (52%), for S 20 and H 1000.

In this example, EA is assumed to be a load variable, and thus our optimization goal consists of calculating the S and H values that minimize expected cost in the presence of uncertainties for both measurements and process actuations. Formulations I, II, and III hold for this example, with L EA, Z [S H], and Y TY (Table 5). We further assumed that the vector of uncertain quantities, [EA, S, H, TY]T, follows a multivariate normal distribution with zero mean and diagonal covariance given by diag22 22 502 42 (25)

) to minimize a for the manipulated variables to be specied (Z cost function, when measurements for the loads become avail ). This particular case study is based on the following able (L model, developed for a batch paper pulp pilot digester39 TY 55.2 0.39 EA 324/ EA log10 S 92.8 log10 H/EA log10 S (23)

This model relates pulp total yield (TY) with effective alkali (EA, a measure of the joint concentration of Na2OH and Na2S, the active elements in the cooking liquor), suldity (S, the percentage of Na2S in the cooking liquor), and H factor (H, a function of the temperature prole across the batch). We consider the situation where a cost function (L) penalizes deviations from a target value for TY (52%): the penalty for lower values is attributed to ber losses, and that for higher values to deterioration in other pulp properties. Our cost function also considers the cost of S and H (proportional to their respective magnitudes). As an example, Figure 6 illustrates the shape of the assumed cost function for S 20 and H 1000 S TY sp TY H 100 100 4 500 L TY 2 S H TY sp 75 2 100 100 4 500 100

d d

TY TY sp TY TY sp (24)

where diag stands for the operator that converts a vector into a diagonal matrix with its elements along the main diagonal. To illustrate the implementation of the formulations above referred, let us consider that the observed value for EA is 15 ). Table 6 summarizes the results obtained for the manipEA and H ) and the average cost obtained with ulated variables (S the objective function assumed under formulations I and II, with a third degree specialized cubature being used for estimation of expected values.40 From Table 6 we can see that under the simulation conditions considered here, and assuming that the relevant objective function is the one associated with formulation I, the optimal solution obtained when one disregards measurement and actuation uncertainties (formulation III) corresponds to an average cost increased by 136%. If the relevant objective function were the one corresponding to problem formulation II, the average cost increase would be 51%. It should also be noticed that the , H ) decision space, location of the optimal solution in the (S found if one ignores uncertainties, is quite distant from the true one. The cost associated with the nonconsideration of these types of uncertainties decreases when their magnitude becomes smaller. Figure 7 presents the results obtained for three alternative problem formulations, when the covariance matrix for uncertain quantities is multiplied by a monotonically decreas-

Table 5. Optimization Formulations I, II, and III as Applied to Case Study 3

Formulation I min E EA, S, H, TY
S,H S,H

Formulation II min E EA, S, H, TY s.t. g ( TY , EA , S , H ) 0 EA EA EA S S S H H H

S,H

Formulation III , S , H , TY min EA , EA , S , H 0 s.t. gTY

s.t. g ( TY , EA , S , H ) 0 EA EA EA S S S H H H TY TY
TY

AIChE Journal

November 2005

Vol. 51, No. 11

3017

Figure 7. Behavior of average cost (formulation I), corresponding to solutions for the three alternative problem formulations, using 0.9i.

ing shrinkage factor, 0.9i. As expected, the differences arising from the solutions associated with such three optimization formulations tend to vanish when measurement and actuation uncertainties decrease. Furthermore, the average cost also decreases because of the improved quality of information obtained from measurement devices and the better performance of nal control elements, as one moves across the several simulation scenarios considered here.

Discussion
Results presented in the previous section highlight not only the potential of using all the information that is available (data and associated uncertainties), but also the difculty that such a task may encompass, with respect to model estimation. In fact, we have come across with some unexpected results and relevant issues have been identied and merit being discussed here. First of all, we stress the fact that, even though simulation results are strictly valid within the conditions established, they can provide useful guidelines for real processes that present structural similarities with them. The fact that classical methods do not make explicit use of uncertainty information may not be very relevant if it represents just a small part of the global variability exhibited by variables. Therefore, uncertainty-based methods presented here are expected to bring potentially more added value only under contexts where uncertainty is quite high (noisy environments) or experiments have large variations. In other words, these methods should complement their classical counterparts, depending on the noise characteristics that prevail in measured data. Still regarding model estimation, we have found some convergence problems in MLMLS, something that is not unusual in approaches based on numerical optimization of a nonlinear objective function. However, problems in MLPCR2 arising from the nonconvergence of MLMLS are usually rare. From the experience that we have gathered so far, no limitations were found regarding the implementation of MLPCR2 in the analysis of real industrial data. The poor performance of MLS under the scenarios considered here, where predictors are
3018 November 2005

strongly correlated, may indicate that the inversion operation undertaken at each iteration is interfering with its performance (the matrix to be inverted in this method becomes quite illconditioned under collinear situations of the predictors). Results obtained for the ridge regularization of MLS (rMLS) show an effective stabilization of this operation. As for PLS methods, the extensive solution of small optimization problems can make unc-PLS1 and unc-PLS2 more prone to numerical convergence problems than the original PLS method, something that does not occur with the remaining uncertainty-based PLS methods (unc-PLS3, unc-PLS4, and unc-PLS5), given that they are based on the estimation of covariance matrices and projection operations. Quite interesting is the fact that, when comparing under heteroscedastic situations (Figure 3) PLS methods that adopt the same estimation procedure for the covariance matrices but differ in the projection phase (as happens with pairs PLS/unc-PLS4, unc-PLS3/unc-PLS5), one can see that the use of uncertainty-based maximum-likelihood nonorthogonal projections seems to be detrimental for prediction with respect to orthogonal projections. In fact, a separate simulation study showed evidence toward a reduced variance of the orthogonal projection scores, when compared to the one exhibited by maximum likelihood projection scores. Apparently, for heteroscedastic scenarios, oscillations in the nonorthogonal projection line may also bring some added variability to the scores, other than the one strictly arising from variability attributed to noise sources. This increased dispersion in the reduced space of the scores, usually the one relevant for prediction purposes, can increase prediction uncertainty arising from poorly estimated models, something that is in line with the results presented in Figure 3. Finally, there are also some approximations considered in the methods that may interfere with their predictive performance and should be considered in future developments. That is, methods unc-PLS1 and unc-PLS2 neglect uncertainties in the load vectors and MLPCR1/MLPCR2 do assume the score uncertainties to be independent. We emphasize that, although we have focused here on steady-state applications, our approaches can also be used under the context of dynamic models, that is, through the consideration of lagged variables41-44 (the PLS methods based on the uncertainty-based estimation of covariance matrices, however, do need some modications to cope with the noise correlations appearing with the use of lagged variables). For such situations, one may also consider uncertainty descriptions connected with robust control methodologies, such as H-innity approaches.

Conclusions
In this paper we address the importance of specifying measurement uncertainties and how this information can be used in two distinct tasks: model estimation and process optimization. With respect to model estimation, under the conditions studied method MLPCR2 presented the best overall predictive performance. In general, those methods based on MLMLS present improvements over their counterparts based on MLS. We have also illustrated the potential advantage of using measurement and actuation uncertainties in process optimization problem formulations and solutions. Our study points out the relevance of not neglecting measurement/manipulation uncertainties
Vol. 51, No. 11 AIChE Journal

when addressing both on-line and off-line process optimization. Future work will address the application of uncertaintybased methods in real industrial contexts, using the guidelines extracted from the results achieved in our comparative study presented herein, regarding the most adequate methods to be adopted for a certain noise/data structure scenario.

Acknowledgments
The authors gratefully acknowledge FCT (Fundac a o para a Cie ncia e Tecnologia, Portugal) for nancial support through research project POCTI/EQU/47638/2002.

Literature Cited
1. MacGregor JF, Kourti T. Statistical process control of multivariate processes. Control Eng Pract. 1995;3:403-414. 2. Wise BW, Gallagher NB. The process chemometrics approach to process monitoring and fault detection. J Process Control. 1996;6: 329-348. 3. Venkatasubramanian V, Yin RRK, Kavuri SN. A review of process fault detection and diagnosis. Parts IIII. Comput Chem Eng. 2003; 27:Part I: 293-311, Part II: 313-326, Part III: 327-346. 4. Wang XZ. Data Mining and Knowledge Discovery for Process Monitoring and Control. London: Springer-Verlag; 1999. 5. ISO. Guide to the Expression of Uncertainty. Geneva, Switzerland: International Organization for Standardization; 1993. 6. Albuquerque JS, Biegler LT. Data reconciliation and gross-error detection for dynamic systems. AIChE J. 1996;42:28412856. 7. Crowe CM. Data reconciliationProgress and challenges. J Process Control. 1996;6:89-98. 8. Romagnoli JA, Sanchez MC. Data Processing and Reconciliation for Chemical Process Operation. Vol. 2. San Diego, CA: Academic Press; 2000. 9. Reis MS, Saraiva PM. A comparative study of linear regression methods in noisy environments. J Chemomet. 2004;18(12):526. ` , Riu J, Rius FX. Application of the multivariate least 10. Mart nez A squares regression method to PCR and maximum likelihood PCR techniques. J Chemomet. 2002;16:189-197. 11. R o FJ, Rio J, Rius FX. Prediction intervals in linear regression taking into account errors in both axes. J Chemomet. 2001;15:773-788. 12. Lira I. Evaluating the Measurement Uncertainty. Bristol, UK: Institute of Physics Publishing; 2002. 13. Mandel J. The Statistical Analysis of Experimental Data. New York, NY: Wiley; 1964. 14. Seber GAF, Wild CJ. Nonlinear Regression. New York, NY: Wiley; 1989. 15. Draper NR, Smith H. Applied Regression Analysis. 3rd Edition. New York, NY: Wiley; 1998. 16. Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. New York, NY: Springer-Verlag; 2001. 17. Jackson JE. A Users Guide to Principal Components. New York, NY: Wiley; 1991. 18. Martens H, Naes T. Multivariate Calibration. Chichester, UK: Wiley; 1989. 19. Martens H, Mevik B-H. Understanding the collinearity problem in regression and discriminant analysis. J Chemomet. 2001;15:413-426. 20. Wentzell PD, Andrews DT, Hamilton DC, Faber K, Kowalski BR. Maximum likelihood principal component analysis. J Chemomet. 1997;11:339-366.

21. Wentzell PD, Andrews DT, Kowalski BR. Maximum likelihood multivariate calibration. Anal Chem. 1997;69:2299-2311. 22. Geladi P, Kowalski BR. Partial least-squares regression: A tutorial. Anal Chim Acta. 1986;185:1-17. 23. Haaland DM, Thomas EV. Partial least-squares methods for spectral analysis. 1. Relation to other quantitative calibration methods and the extraction of qualitative information. Anal Chem. 1988;60:1193-1202. 24. Helland IS. Some theoretical aspects of partial least squares regression. Chemomet Intell Lab Syst. 2001;58:97-107. 25. Helland IS. On the structure of partial least squares regression. Commun Stat Simulat Comput. 1988;17:581. 26. Ho skuldsson A. Prediction Methods in Science and Technology. Ventura, CA: Thor Publishing; 1996. 27. Wold S, Sjo stro m M, Eriksson L. PLS-regression: A basic tool of chemometrics. Chemomet Intell Lab Syst. 2001;58:109-130. 28. Kaspar MH, Ray WH. Partial least squares modelling as successive singular value decompositions. Comput Chem Eng. 1993;17:985-989. 29. Ho skuldsson A. PLS regression methods. J Chemomet. 1988;2:211228. 30. Phatak A. Evaluation of Some Multivariate Methods and Their Applications in Chemical Engineering. PhD Thesis. Waterloo, Ontario, Canada: Dept. of Chemical Engineering, University of Waterloo; 1993. 31. Helland IS. Rotational symmetry, model reduction and optimality of prediction from the PLS population model. Proc of 2nd Int Symp on PLS and Related Methods, Capri, Italy, October; 2001. 32. Helland IS. Partial least squares regression. Encyclopedia of Statistical Sciences. 2nd Edition. Hoboken, NJ: Wiley; 2002. 33. Reis MS, Saraiva PM. Heteroscedastic latent variable modelling with applications to multivariate statistical process control. Accepted for publication in Chemomet Intell Lab Syst. 34. Rooney WC, Biegler LT. Design for model parameter uncertainty using nonlinear condence regions. AIChE J. 2001;47:1794-1804. 35. Burnham AJ, Macgregor JF, Viveros R. Latent variable multivariate regression modeling. Chemomet Intell Lab Syst. 1999;48:167-180. 36. MacGregor JF, Kourti T. Multivariate statistical treatment of historical data for productivity and quality improvements. Proc of Foundation of Computer Aided Process Operations (FOCAPO 98), Snowbird, UT, July 510; 1998. 37. Walczak B, Massart DL. Dealing with missing data. Chemomet Intell Lab Syst. 2001;58:Part I: 15-27, Part II: 29-42. 38. Nelson PRC, Taylor PA, MacGregor JF. Missing data methods in PCA and PLS: Score calculations with incomplete observations. Chemomet Intell Lab Syst. 1996;35:45-65. 39. Carvalho MGV, Martins AA, Figueiredo MML, Kraft pulping of Portuguese Eucalyptus Globulus: effect of process conditions on yield and pulp properties. Appita. 2003. 267. 40. Bernardo FP, Pistikopoulos EN, Saraiva PM. Integration and computational issues in stochastic design and planning optimization problems. Ind Eng Chem Res. 1999;38:3056-3068. 41. Ku W, Storer RH, Georgakis C. Disturbance detection and isolation by dynamic principal component analysis. Chemomet Intell Lab Syst. 1995;30:179-196. 42. Ricker NL. The use of biased least-squares estimators for parameters in discrete-time pulse-response models. Ind Eng Chem Res. 1988;27: 343-350. 43. Shi R, MacGregor JF. Modeling of dynamic systems using latent variable and subspace methods. J Chemomet. 2000;14:423-439. 44. de Jong S, Wise BW, Ricker NL. Canonical partial least squares and continuum power regression. J Chemomet. 2001;15:85-100.
Manuscript received May 25, 2004, and revision received Mar. 7, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

3019

Design and Testing of a Control System for Reverse-Flow Catalytic Afterburners

Miguel A. G. Hevia, Salvador Ordo n ez, and Fernando V. D ez
Departamento de Ingenier a Qu mica y Tecnolog a del Medio Ambiente, Universidad de Oviedo, Julia n Claver a 8, 33006 Oviedo, Spain

Davide Fissore and Antonello A. Barresi

Dipartimento di Scienza dei Materiali ed Ingegneria Chimica, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy DOI 10.1002/aic.10573 Published online August 11, 2005 in Wiley InterScience (www.interscience.wiley.com).

The design and testing of a control system for reverse-ow catalytic afterburners based on a two-variable bifurcation map is focused on. The combustion of lean methaneair mixtures is considered as the test reaction and a bench-scale apparatus, with a special temperature-control system based on dynamic compensation of the thermal losses to achieve adiabatic operation, is used for validation purposes. The aim of the control system is to avoid both catalyst overheating and reaction extinction when the adiabatic temperature increases and the ow rate of the feed changes. Stability maps of the reactor are obtained by means of numerical simulations, showing the values of the operating parameters (switching time), which allows fulllment of the operating constraints (catalyst maximum temperature and methane conversion) when the inlet concentration and ow rate change. This system was realized and tested experimentally, mainly for inlet concentration changes, proving to be effective in all cases investigated. 2005 American Institute of Chemical Engineers AIChE J, 51: 3020 3027, 2005 Keywords: reverse-ow reactors, autothermal combustion, reaction extinction, adiabatic reactor

Introduction
The reverse-ow catalytic reactor (RFR) is a device in which the feed ow direction is periodically reversed, thus giving rise to unsteady-state operation. When the combustion of cold and lean volatile organic compound (VOC) mixtures in air is carried out, the reversal of the ow direction traps the heat of reaction inside the bed, whose ending parts act as regenerative heat exchangers, thus allowing autothermal combustion and eliminating the need for auxiliary fuel to sustain the reaction. Extensive investigations about the RFR, including both numerical simulations and exper-

Correspondence concerning this article should be addressed to A. A. Barresi at antonello.barresi@polito.it.

2005 American Institute of Chemical Engineers

imental analysis, were performed in the past 30 years and were reviewed, for example, by Matros and Bunimovich.1 In addition to the intrinsically dynamic behavior of the RFR, one must deal with unexpected external perturbations (in the feed concentration, composition, temperature, and ow rate), which may lead to reactor extinction (and thus later to reactant emissions) or catalyst overheating (and thus subsequent deactivation). To avoid these problems it is necessary to implement some closed-loop control strategy. Few papers have appeared in the literature concerning this topic. Van de Beld and Westerterp2 discussed different possible solutions to avoid reactant emissions in a RFR in the case of temporary reduction of the concentration of the feed, but none of these proved to be effective: (1) Increasing of the feed temperature requires a high energy input and, for control purposes, it is too slow because the entire bed has to be heated by the hotter feed.
Vol. 51, No. 11 AIChE Journal

3020

November 2005

(2) Adding combustibles to the feed seems the most logical method to use because it is simple to implement, although a reactive combustible, when added to the feed, will already react in the inlet part of the catalyst bed and the increase of the maximum temperature in the center of the bed is initially small. Moreover, the ignition temperatures of the original pollutants and the supporting fuel must be almost equal; otherwise, the reactor may establish the temperature prole corresponding to the component with the lower ignition temperature and will not be able to oxidize the component with the higher ignition temperature.3-5 (3) An electrical heating device in the center of the reactor may increase the temperature in the hottest zone of the bed, even very rapidly, although the heat must be transferred from the heating device to the catalyst, thus introducing an additional heat transport resistance. Experimental results of Cunill et al.6 show that this device is effective, given that the maximum temperature in the reactor is now a function of the electrical power and almost independent of the mixture components and composition; thus a mixture of contaminants with very different ignition temperatures can be burned completely, provided that the electrical power is sufciently high. If a heating support is required for an extended period of time the electrical heating device may become too expensive; in this case, adding combustibles in the central part of the reactor is a better choice, even if lack of good distribution of the combustible over the entire cross-sectional area may cause difculties. (4) Permanent addition of a component difcult to oxidize, fed in a concentration sufcient to ensure the ignition, may be an alternative solution: the temperature level in the reactor will be high and all other less refractive contaminants will certainly be converted. The main drawback is that this is a rather energy-consuming solution; furthermore, an increase in the inlet concentration may be difcult to handle because the catalyst may become overheated. (5) Supply of a hot gas in the center of the reactor is an inefcient way to increase the temperature because the heat capacity of the gas is low compared to that of the bed. Finally, Barresi and Vanni7 investigated the possibility of avoiding reaction extinction by acting on the switching time. They considered a one-point control strategy, with a temperature measurement located at either end of the active portion of the bed: ow direction is changed when temperature at the controller located close to the inlet drops below a certain set point. Their simulations provided evidence that at a higher set point the maximum temperature is also increased, althoughat least for large portions of inert material conversion cannot be arbitrarily improved by this strategy; this is explained by the shorter periods necessary for higher set points, resulting in a stronger washout effect. If the inlet concentration is lower than the minimum adiabatic temperature rise that is required to allow for autothermal operation, the change of the switching time may not be sufcient and the system evolves toward the extinction following a Zeno trajectory.8 Besides the problem of reaction extinction, the temperature in the reactor should not exceed the maximum allowable catalyst temperature, which may occur when the concentration of contaminants becomes too high during a signicant period of time. A short peak of high concentration will not be a problem because the heat capacity of the system is high and it will take time before the maximum temperature exceeds the limit. Furthermore, the RFR exhibits some self-control with respect to
AIChE Journal November 2005

the inlet concentration. For a low concentration, the reaction takes place in the central part of the reactor, so that part of the catalyst is used to store heat and to preheat the cold feed. If the inlet concentration is increased, the temperature and concentration proles will move in the direction of the inlet and outlet of the reactor and a temperature plateau will develop. For higher inlet concentrations full conversion is already obtained in the rst layers of the catalyst bed and the length of the heat exchanger equals that of the inert sections and remains constant. From this point, the maximum temperature will increase very rapidly with the inlet concentration. Different technical solutions have been proposed in the literature9: (1) Dilution of the feed with additional air has the main drawback that the total ow rate increases signicantly, thus requiring additional energy costs to overcome the higher pressure drop; moreover, there is the problem of obtaining reliable and inexpensive on-line measurements of the inlet concentration.2 (2) Increase of the switching time may lower the maximum temperature, particularly when high inert fraction is used.3,6 (3) Heat recovery by internal heat exchange was discussed by Grozev and Sapundzhiev10 and by Nieken et al.,3,4 who indicated that the switching period has to be increased, so that the reaction front approaches the center of the reactor before a noticeable inuence on the maximum temperature can be noticed. However, no remarkable reduction of the maximum temperature is possible through intermediate cooling, irrespective of the switching period used, although the energy recovery increases. Sapundzhiev et al.11 used an internal heat exchange to preheat the cold feed. External cooling restricts the region of operating variables that allows for autothermal behavior as combustion can now be quenched if the coolant removes more heat than is generated.12 (4) Cold gas injection into the middle of the packed bed may have an effect similar to that of intermediate cooling through heat exchange, leading to a signicant decrease of the temperature in the middle of the reactor, whereas the maximum temperature is more or less unaffected.4 (5) Hot gas withdrawal from the center of the reactor allows for almost complete energy recovery, but again the maximum temperature cannot be lowered sufciently: sidestream withdrawal just shifts the reaction zone toward the center of the reactor.4 (6) Structured xed bed. If the packed bed is composed of portions with high effective conductivity and of portions of low conductivity then it depends on the position of the temperature fronts whether the maximum temperature is lowered or increased. If the fronts are always in the portion with high axial conductivity, the maximum temperature will be low and the efciency of the heat recovery will be weak; if the fronts stay in the portion with low conductivity, the opposite is true. Because the position of the fronts can be shifted by the amount of the hot gas withdrawn, an interesting possibility exists to operate the reactor in either of the two regimes. To exploit this possibility a structured xed bed is needed, where the packed bed is composed of three portions: an inner portion with high effective conductivity support and two outer portions with low conductivity support.4 Budman et al.13 considered the application of a conventional PID controller with anti-windup: exit concentration, the controlled variable, is used to infer the maximum temperature (because thermocouple locations in the bed are xed, it may be
Vol. 51, No. 11 3021

extremely difcult to evaluate the maximum temperature, given that the temperature proles move during the operation of the reaction). The same authors also considered a feedforward regulation of the exit concentration through knowledge of the inlet concentration. The idea behind this strategy is that if the inlet concentration is known, it is possible to evaluate the optimal combination of switching time and cooling rate to obtain full conversion and safe operation in the reactor. In this regard, a number of simulations have been performed to identify the parameters region for safe operation. In this work a control system was used to control the reverse-ow operation in a catalytic afterburner: the switching time was varied to avoid both catalyst overheating and reaction extinction. By means of numerical simulations, a stability map of the reactor is obtained, showing the values of the switching time that allow fulllment of the operating constraints (catalyst maximum temperature and methane conversion) when the inlet concentration and ow rate change. To apply this control strategy we need to know both the inlet ow rate (which is easy to do and inexpensive) and the inlet concentration. Because on-line measurements may be difcult and expensive, a soft sensor based on high-gain techniques can be used to estimate the inlet concentration from some temperature measurements.14 The article is structured as follows: in Section 2 the detailed model used in the control system is described. In Section 3 the experimental apparatus is introduced and the mathematical model validated. In Section 4 the control algorithm is introduced and validated by means of simulations and of experiments.

Energy Balance for the Gas Phase T G ha v T S T G TG k eff 2T G v t Gc p,G x 2 x Gc p,G 4hi T T W Gc p,GD R,i G (3)

Mass Balance for the Solid Phase k G, ja v y S, j y G, j S 1

k1

R
NR k j,k

k, j

(4)

with j 1 . . . nr. Energy Balance for the Solid Phase ha v TS S 2T S T T G t Sc p,S x 2 Sc p,S 1 S 1 c p,S
nr NR

i1

R H
k i,k k f,i k1

(5)

Energy Balance for the Reactor Wall 4 W 2T W TW D R,ih i T G T W 2 2 t c p , W W x 2 c p,W W D R ,e D R,i (6) Adiabatic operation is considered because the particular device adopted guarantees pseudoadiabatic behavior, as will be shown in the following. For the catalytic part of the reactor, a rst-order rate equation was considered in the mass balance: R k, j k ry S, j k exp Ea /RTS yS , j k k S 1 RT G (7) (8)

The Model
A one-dimensional heterogeneous model was used to model the RFR. Pressure loss inside the system was neglected and plug ow, with dispersive transport of mass and energy, was assumed for the gas phase; the ideal gas law was used. The transient term was taken into account in the gas-phase equations as well as in the energy equation for the solid phase, whereas the solid catalytic surface was considered in pseudosteady-state condition. The effect of the intraparticle mass transport was included in the model by means of the effectiveness factor. The effect of the reactor wall on the thermal balance of the system is taken into account by means of a further energy balance equation for the wall. Thus, the dynamics of the adiabatic process can be described by the following set of partial differential-algebraic equations (PDAE): Continuity Equation for the Gas Phase cG c v t x G

i1

k a y
nr G,i v

S,i

y G,i

(1)

Continuity Equation for Component j in the Gas Phase y G , j k G , ja v y G, j 2 y G, j D eff y S, j y G, j 2 v t x x c G y G, j with j 1 . . . (nr 1).
3022 November 2005

i1

a y kc
nr G,i v G

S,i

y G,i

(2)

whereas for the inert sections the reaction rate was set equal to zero. Similarly, the solid physical properties density, specic heat, and thermal conductivitywere set equal to the values either of the catalyst or of the inert sections, depending on the axial position in the reactor. If the physical and transport properties of the catalyst and of the inert sections are different, adequate boundary conditions (that is, identity in the heat and mass uxes) must be specied at the boundary surface. Conventional Danckwerts boundary conditions were assumed in x 0 and x L. Initially, the gas-phase temperature was considered constant along the reactor and equal to the inlet value, and the solid temperature was considered constant and equal to the preheating value. Transport and dispersion parameters were evaluated according to previous works on the same subject.15 Further details can be found in a previous publication.16 Table 1 shows the values of the main parameters and operating conditions used in the simulation of the RFR. The inuVol. 51, No. 11 AIChE Journal

Table 1. Main Operating Parameters Considered in the Simulations of the RFR

Preheating temperature Pellet diameter Catalyst density, S Catalyst specic heat, c p , S Catalyst porosity Catalyst tortuosity Catalyst thermal conductivity, eff Bed void fraction, Frequency factor, k Activation energy, E att Inlet gas temperature 650 K 24 mm 2170 kg m3 848.1 J kg1 K1 0.519 2 0.472 W m1 K1 0.36 1.0886 105 mol kg1 s1 Pa1 1.12504 105 J mol1 298 K

ence of the temperature and composition on the density and on the specic heat of the gas was taken into account. The domain of the spatial variable x was discretized on a grid of equally spaced points; 101 points are enough to ensure a grid-independent solution. As a consequence, the PDAE system (Eqs. 1 6) was transformed into a differential algebraic equations (DAE) problem. The integration in time of the differential part of the system was performed by implementing the Fortran routine LSODES, from the package ODEPACK.17 Both relative and absolute tolerance were set to the square root of the working machine precision. After a transient period, the solution of the system evolves toward a pseudosteady state (PSS), given that the behavior of the reactor (temperature and concentration proles) is the same within every cycle.

The Bench-Scale Rig and Model Validation

The design of the experimental reactor to carry out the model validation (and subsequent test of the control system) was mainly aimed to reproduce as fairly as possible the behavior of large-scale industrial reactors. The major drawback found when working on the fulllment of this goal, apart from the inuence of the wall thermal inertia that is taken into account in our model, is the difculty in obtaining operation regimes close to adiabaticity with small-sized reactors. This is because the higher ratio of external surface to volume, which is characteristic of small reactors (low diameters), causes the heat losses to be relatively more important with regard to the overall thermal balance of the system. Thus, whereas in large-industrial reactors adiabatic behavior can be assumed, it is not possible to do the same with small lab-scale or even pilot-scale reactors and it is necessary to use heat-transfer coefcients to take into account the heat losses.18,19 In this work, this problem is solved using a special device with its own control system, which acts over the temperature outside the reactor. Such a control system is able to cause such outer temperature to dynamically follow the axial inner temperature in every axial position, thus eliminating the driving force to the heat transmission in the radial direction, and therefore the heat losses. Because axial temperature proles are not uniform and change with time as a consequence of unsteady-state operation, the heating system was divided into seven sections or band heaters, each of them independently controlled by different PID controllers. The central band heater is 0.10 m long and the others are 0.065 m long: this was the best compromise between efciency and complexity of the
AIChE Journal November 2005

device. To follow the dynamic of the temperature in the reactor, cooling air has to be supplied to cool down reactor surroundings when the bed temperature decreases, thus avoiding further heating arising from the thermal inertia of the band heater. This system is currently undergoing a patent application20 and has been described in detail in a previous paper.16 The reactor is a 316 stainless steel tube with 0.496 m effective (lled with packing) length, an inner diameter of 5.17 102 m, and a wall reactor thickness of 1.15 103 m. The packing includes catalyst and inert material. The latter consists of two sections of glass spheres (4 103 m diameter) at both ends of the bed, each one taking up 24.8% of the effective length. A commercial catalyst was used composed of a mixture of metal oxides supported on -alumina, forming spheres with an average diameter of 4 103 m. The kinetics of methane combustion over this catalyst was studied in an isothermal xed-bed reactor,21 from which it was determined that a pseudo-rst-order law accurately tted the results. Analysis of the gas streams was carried out by means of an HP 6890 GC equipped with an HP-5 capillary column and ame ionization detector (FID). Several experiments were carried out, changing the values of the main operating parameters, such as switching time, inlet concentration, and surface gas velocity to test the adequacy of the model to simulate the behavior of the reactor. Figure 1 shows an example of the solid temperature proles obtained for various values of the operating parametersswitching time, inlet concentration, and surface gas velocitywhen the PSS has been reached (the temperature proles at the middle of a semicycle are shown). The experimental values are compared to those obtained by simulation of the adiabatic reactor, also proving that the dynamic compensation of the heat losses is effective in achieving adiabatic operation and avoiding overcompensation. The agreement between the experimental measured values of the temperature and the simulated values is fairly good. It is important to highlight that no tting parameters were used in the simulation of the reactor so that the model can be considered fully predictive.

The Control Algorithm

By means of numerical simulation a stability map of the reactor can be obtained that delimits regions where maximum temperature (or conversion) is higher or lower than a certain value. The full map is a function of three parameters: the switching time (which will be the manipulated variable) and the inlet concentration and surface velocity (which are the disturbances). Figure 2 (top graph) shows a section of this map, obtained with a value of the surface velocity of the gas equal to 0.14 m s1, whereas Figure 2 (bottom graph) shows a section of the map, obtained with a value of the inlet concentration equal to 4000 ppmV. Because of process constraints on outlet reactant conversion and maximum temperature on the solid, a well-dened operating region is obtained as a function of the three parameters: the disturbances and the manipulated variable. In both sections of the stability map shown in Figure 2 the region corresponding to methane conversion 99.95% and maximum solid temperatures 650C is evidenced. In the denition of the operating region it is also possible to take into account modeling errors by dening a percentage uncertainty over these curves. Until the point corresponding to the operating conditions (inlet concentration, inlet surface velocity, and switching peVol. 51, No. 11 3023

riod) is in the optimal region no control action is taken; if the inlet concentration increases too much, as well as if the inlet surface velocity increases too much, the control system has two possibilities: it may increase the switching time (because the maximum temperature decreases slightly when the switching time is increased) or, if this is not sufcient, it may dilute the feed to remain in the optimal working region. If the inlet concentration decreases too much, as well as if the inlet surface velocity decreases, a control action is taken just when the outlet conversion starts decreasing (because of the thermal inertia of the system, the reactor is able to sustain periods of low inlet concentration, with no decrease in the outlet conversion): either the switching time is reduced, if this can increase the conversion, or auxiliary fuel is added when the inlet concentration, or the inlet surface velocity, is lower than the minimum value that allows adiabatic operation. In the example considered in Figure 3 (top graph) a reduction of the switching time is sufcient to avoid any conversion decrease (bottom graph) after the reduction in the methane concentration of the feed. The subsequent increase in the inlet

Figure 1. Simulated (lines) axial temperature proles and actual measurements (symbols) in the middle of a cycle, when the PSS was reached, for various values of the switching time (graph A, uG,0 0.143 m s1, yCH4,0 3500 ppmV), gas velocity (graph B, yCH4,0 3500 ppmV, tc 600 s), and inlet methane concentration (graph C, uG,0 0.143 m s1, tc 600 s).

Figure 2. Example of stability map of a RFR (top graph: uG,0 0.14 m s1; bottom graph: yCH4,0 4000 ppmV).
The solid line separates the region where extinction occurs from the region where stable operation can be obtained; dotted lines separate the regions where conversion is larger (top part) or lower than the specied value; dashed lines separate the regions where maximum solid temperature is higher (top part) or lower than the specied value.

3024

November 2005

Vol. 51, No. 11

AIChE Journal

Figure 3. Bottom graph: maximum solid temperature and outlet methane conversion in a RFR when the inlet concentration is varied according to the top graph.
Variation of the switching time is considered to maintain the system in the optimal region. The behavior with (solid line) and without (dashed line) control action is compared. Disturbance and control action are evidenced on the stability map (top graph).

Figure 4. Bottom graph: maximum solid temperature and outlet methane conversion in a RFR when the inlet surface velocity is varied according to the top graph.
Variation of the switching time is considered to maintain the system in the optimal region. The behavior with (solid line) and without (dashed line) control action is compared. Disturbance and control action are evidenced on the stability map (top graph).

concentration does not require any control action as the system remains in the optimal working region. In the example considered in Figure 4 (top graph) the switching time in increased to face against the catalyst overheating resulting from the increase of the surface velocity (bottom graph). It is important to stress that this control system based on the model of the process is able to avoid any control action when this is useless and it is able to fulll any constraints on process operation, being sufcient to indicate them in the working map. The performance of the control system was also experimentally tested. The requirements considered were 1 ppmV for the semicycle averaged outlet concentration, and 650 C for the maximum temperature, which was experimentally demonstrated to be the maximum temperature that the catalyst can endure without deactivation. The changes of the inlet concentrations, the times at which they were introAIChE Journal November 2005

duced, and responses of the control system are included in Table 2. The negative perturbations of the inlet concentration endanger the stability of the reactor and can lead to an increase of the outlet concentration and, eventually, to the extinction. By contrast, the risk of positive perturbations is
Table 2. Perturbations, Time at Which They Were Introduced, and Responses of the Control System
Perturbation y G ,0 10 6 initial y G ,0 10 6 nal t , perturbation* t c initial t c nal 3500 3050 9900s 990s 150s Perturbation 3500 3950 4400s 120s 990s

*Time elapsed from the beginning of the experiment to the perturbation.

Vol. 51, No. 11

3025

Conclusions
A control system based on a detailed model was designed for a catalytic afterburner. A detailed one-dimensional model was used to simulate the operation in a bench-scale reactor with an innovative system for compensation of thermal losses. The control strategy, based on the appropriate choice of the switching time according to the operating conditions (inlet gas velocity, concentration), was demonstrated to be effective in preventing both catalyst overheating and reaction extinction (and thus reactant emissions), avoiding any control action when this is useless, even if the inlet ow rate and concentration changes, but remaining in the optimal working region. When acting on the switching time is not sufcient, maximum temperature is decreased by dilution of the feed, which has proved to be the simplest and more effective control action to achieve the desired result and unitary conversion is guaranteed by injection of auxiliary fuel. The pumping costs associated with the air dilution and those attributed to the auxiliary fuel, which also generates additional carbon dioxide emissions, are reduced because this measure will be adopted just when large perturbations occur. To apply this control strategy we need to know both the inlet ow rate (which is easy to do and inexpensive) and the inlet concentration. Because on-line measurements may be difcult

Figure 5. Experimental test of the performance of the control system for a negative inlet concentration perturbation (see Table 2).
Comparison between experimental value of the outlet methane concentration (symbols) and simulated values (lines) when no control action is undertaken (top graph) and when the control system is working (bottom graph).

the overheating. This is why the interesting variable to follow in the former case is the outlet concentration, whereas in the latter one is the maximum temperature. Four experimental runs, with and without applying the control system for a negative and a positive perturbation, were carried out, and the control system performed as theoretically expected for both types of disturbances in all cases. Figure 5 shows the concordance with the corresponding simulated results. Moreover, in this gure it can be observed that the outlet concentration in the uncontrolled response increases with time, whereas in the controlled one tends to zero. In this way, the diminution of the switching time is in practice demonstrated to be an effective measure to avoid the outlet concentration increase that takes place as a consequence of a negative perturbation. The results for a positive perturbation are shown in Figure 6, pointing out that, although instant maxima appear in the controlled response in the rst minutes after the perturbation, the control system eventually dampens the temperature increase, otherwise occurring in the uncontrolled response.
3026 November 2005

Figure 6. Comparison between the simulated (top graph) and the experimental values (bottom graph) of the maximum temperature (among the seven axial positions in which the temperature is measured) of the reactor to a positive inlet concentration perturbation (see Table 2) with and without control system. Vol. 51, No. 11 AIChE Journal

and expensive, a soft sensor based on high-gain techniques can be used to estimate the inlet concentration from some temperature measurements.

Literature Cited
1. Matros YS, Bunimovich GA. Reverse-ow operation in catalytic reactors. Catal Rev Sci Eng. 1996;38:1-68. 2. Van de Beld L, Westerterp KR. Operation of a catalytic reverse ow reactor for the purication of air contaminated with volatile organic compounds. Can J Chem Eng. 1997;75:975-983. 3. Nieken U, Kolios G, Eigenberger GA. Fixed-bed reactors with periodic ow reversal: Experimental results for catalytic combustion. Catal Today. 1994;20:335-350. 4. Nieken U, Kolios G, Eigenberger GA. Control of the ignited steady state in autothermal xed-bed reactors for catalytic combustion. Chem Eng Sci. 1994;49:5507-5518. 5. Cittadini M, Vanni M, Barresi AA, Baldi G. Development and design of a forced unsteady-state reactor through numerical simulation. In: Proceedings of the 10th European Symposium on Computer Aided Process Engineering (Computer-Aided Chemical Engineering Series). Vol. 8. Amsterdam: Elsevier; 2000:697-702. 6. Cunill F, Van de Beld L, Westerterp KR. Catalytic combustion of very lean mixtures in a reverse ow reactor using an internal electrical heater. Ind Eng Chem Res. 1997;36:4198-4206. 7. Barresi AA, Vanni M. Control of catalytic combustors with periodical ow reversal. AIChE J. 2002;48:648-652. 8. Mancusi E, Russo L, di Bernardo M, Crescitelli S. Zeno trajectories in a non-smooth model of a controlled reverse ow reactors. Proc of European Symposium on Computer-Aided Chemical Engineering (ESCAPE-14), Lisbon, Portugal, May 19 23; 2004. 9. Eigenberger G, Nieken U. Catalytic combustion with periodical ow reversal. Chem Eng Sci. 1988;43:2109-2115. 10. Grozev GG, Sapundzhiev CG. Modelling of reversed ow xed bed reactor for catalytic decontamination of waste gases. Chem Eng Technol. 1997;20:378-383. 11. Sapundzhiev C, Chaouki J, Guy C, Klvana D. Catalytic combustion of natural gas in a xed bed reactor with ow reversal. Chem Eng Commun. 1993;125:171-186. 12. Purwono S, Budman H, Hudgins RR, Silveston PL, Matros YS. Runaway in packed bed reactors operating with periodic ow reversal. Chem Eng Sci. 1994;49:5473-5487. 13. Budman H, Kzyonsek M, Silveston P. Control of nonadiabatic packed bed reactor under periodic ow reversal. Can J Chem Eng. 1996;74: 751-759. 14. Edouard D, Schweich D, Hammouri H. Observer design for reverseow reactor. AIChE J. 2004;50:2155-2166. 15. Van de Beld L. Air Purication by Catalytic Oxidation in an Adiabatic Packed Bed Reactor with Periodic Flow Reversal. PhD Thesis. Enschede, The Netherlands: University of Twente; 1995. 16. Fissore D, Barresi AA, Baldi G, Hevia MAG, Ordo n ez S, D ez F. Design and testing of small scale unsteady-state afterburners and reactors. AIChE J. 2005;51:1654-1664. 17. Hindmarsh AC. ODEPACK, A Systematized Collection of ODE Solvers. Stepleman RS, et al., eds. (Vol. 1 of IMACS Transactions on Scientic Computation). Amsterdam: North-Holland; 1983. 18. Van de Beld L, Borman RA, Deckx OR, Van Woezik BAA, Westerterp KR. Removal of volatile organic compounds from polluted air in a reverse ow reactor: An experimental study. Ind Eng Chem Res. 1994;33:2946-2956. 19. Ben-Tullilah M, Alajem E, Sheintuch M. Flow-rate effects in ow reversal reactors: Experiments, simulations, approximations. Chem Eng Sci. 2003;58:1135-1146. 20. D ez F, Vega A, Ordo n ez S, Hevia MAG, Fissore D, Cittadini M, Vanni M, Barresi AA, Baldi G. Dispositivo para el control de ujo de calor a trave s de la pared en equipos pequen os. Spanish Patent Application No. P200400625; 2004. 21. Hurtado P, Ordo n ez S, Vega A, D ez FV. Methane catalytic combustion over commercial catalysts in presence of ammonia and hydrogen sulphide. Chemosphere. 2004;65:681-689.
Manuscript received Sep. 9, 2004, revision received Feb. 18, 2005, and nal revision received Apr. 26, 2005.

Acknowledgments
This work was nancially supported by the European Community (Contract ENV4-CT97-0599). The nancial support of Italian and Spanish Ministries of University and Research (Project Integrated Actions Italy Spain) is also gratefully acknowledged.

Notation
av c cp Deff dP DR Ea Hf h k k keff kG kr L NR nr Pr R R ReP Sc T t tc u v x y z external particle surface area per unit volume of reactor, m1 molar concentration, mol m3 specic heat at constant pressure, J kg1 K1 effective mass dispersion coefcient, m2 s1 pellet diameter, m reactor diameter, m activation energy, J kmol1 molar enthalpy of formation, J mol1 gas-solid heat transfer coefcient, J m2 K1 s1 preexponential factor, s1 frequency factor, mol kg1 s1 Pa1 effective heat dispersion coefcient, J m1 K1 s1 gassolid mass transfer coefcient, mol m2 s1 kinetic constant, s1 total reactor length, m number of reactions number of components in the mixture Prandt number ideal gas constant, J K1 mol1 reaction rate, s1 particle Reynolds number Schmidt number temperature, K time, s switching time, s surface velocity, m s1 interstitial velocity, m s1 axial reactor coordinate, m molar fraction nondimensional axial reactor coordinate, z x/L

Greek letters
st bed void fraction effectiveness factor thermal conductivity, J m1 K1 s1 thermal conductivity of the stagnant gas, J m1 K1 s1 stoichiometric coefcient density (or apparent density for the solid), kg m3

Subscripts and superscripts

0 e G i max S W inlet conditions external value gas phase internal value maximum value solid phase or solid surface reactor wall

Abbreviations
RFR reverse-ow reactor PSS periodic steady state

AIChE Journal

November 2005

Vol. 51, No. 11

3027

REACTORS, KINETICS, AND CATALYSIS

Stationary Transversal Hot Zones in Adiabatic Packed-Bed Reactors

G. Viswanathan and D. Luss
Dept. of Chemical Engineering, University of Houston, Houston, TX 77204

A. Bindal and J. Khinast

Dept. of Chemical and Biochemical Engineering, Rutgers University, Piscataway, NJ 08854 DOI 10.1002/aic.10542 Published online August 3, 2005 in Wiley InterScience (www.interscience.wiley.com).

A pseudo-homogeneous model predicts that a stationary, stable, transversal nonuniform temperature pattern cannot form in any cross section of an adiabatic packed-bed reactor used to conduct a single reaction, if its dynamics depends only on the concentration of a limiting reactant and temperature. This is due to the fact that the transversal heat dispersion in the reactor is larger than that of the limiting reactant. This conclusion is proved for a shallow packed-bed reactor described by a pseudo-homogeneous model. Extensive numerical simulations showed that a two-phase model of an adiabatic packedbed reactor exhibits the same behavioral features. A stable, stationary hot zone can form in the cross section of the reactor only under the unrealistic assumption that the transversal species dispersion exceeded that of the temperature. 2005 American Institute
of Chemical Engineers AIChE J, 51: 3028 3038, 2005

Keywords: packed-bed reactor, hot spot, neutral stability, temperature pattern, nonuniform perturbations

Introduction
Localized hot zones have been reported to exist in packedbed reactors. They have in general a deleterious impact on the performance of the reactor, that is, decreasing the yield of the desired products. Moreover, they may initiate undesired results, leading to runaway reactions. Puszynski and Hlavacek1 observed a periodic generation of a downstream traveling hot region during CO oxidation in a packed-bed reactor. Wicke and Onken2,3 observed a similar periodic, traveling temperature wave during the oxidation of either CO or ethylene in a packed-bed reactor. They pointed out that these periodic temperature waves form, when the catalytic reaction rate in the upstream section of the reactor is oscillatory. Rovinski and Menzinger4 observed a periodic sequence of traveling concentration pulses when they conducted a BelousovZhabotinski
Correspondence concerning this article should be addressed to D. Luss at dluss@uh.edu.

2005 American Institute of Chemical Engineers

reaction in a catalytic packed-bed reactor under excitable conditions. Sheintuch research group5-7 conducted extensive theoretical analyses of the formation of these traveling hot zones. A uniform temperature is expected to exist at any cross section of adiabatic packed-bed reactors. There have been several attempts to predict the conditions that may lead to evolution of a stable, nonuniform transversal temperature eld in such a reactor. Matros8 reported formation of hot zones in a packed-bed reactor in which the catalyst was not uniformly packed. This type of hot zone formation may be circumvented by improving the packing of the bed and will not be addressed here. Barkelew and Gambhir9 reported the formation of small clumps of molten catalyst (clinkers) during hydrodesulfurization in trickle-bed reactors. Wicke and Onken2,3 noted that the temperature at two locations in the same cross section of the reactor was different, that is, transversal (normal to the ow direction) temperature patterns existed. When local hot regions exist next to the reactor walls, they may decrease its strength, leading to severe safety problems. A thorough understanding and ability to predict when such local hot zones form are
Vol. 51, No. 11 AIChE Journal

3028

November 2005

paramount for a rational reactor design and operation that circumvent this undesired behavior. Several studies attempted to predict the conditions leading to formation of the transversal hot zones. Viljoen et al.10 pointed out that an exothermic chemical reaction may generate various convection and temperature patterns in a porous media. Stroh and Balakotaiah,11 Nguyen and Balakotaiah,12,13 and Subramanian and Balakotaiah14 all showed that spatiotemporal ow and temperature patterns may evolve in packed-bed reactors, when the feed ow rate is rather low. Schmitz and Tsotsis15 found that stationary patterns can form in a chain of interacting catalysts (cell model) only when the rate of species exchange exceeded that of heat exchange. Balakotaiah et al.16 showed that the uniform stable state in an adiabatic packed-bed reactor may become unstable, leading to evolution of a nonuniform state. They showed that the larger the diameter of the reactor, the larger the number of possible stationary nonuniform states that may form. Yakhnin and Menzinger17 pointed out that the analysis of Balakotaiah et al.16 was for a case in which the transversal effective heat dispersion was lower than that of the species dispersion. This condition is similar to that enabling formation of stationary patterns in two variable reaction diffusion systems, if the diffusion of the inhibitor is larger than that of the activator.18,19 However, in practice the transversal dispersion of heat in a packed-bed reactor is always larger than that of the species dispersion. The goal of this study is to determine under what conditions and reaction rates stable, stationary, nonuniform temperature states exist in packed-bed reactors, and to analyze the qualitative features and stability characteristics of the branches of the nonuniform states.

1 2 1 r 2 r r r r 2

z0

(5)

The corresponding boundary conditions are Df , a f , a Cf vCin Cf z

Tf vCp f Tin Tf z Ts 0 z Tf 0 z Tf 0 r

(6)

Cf 0 z Cf 0 r

Ts 0 z Ts 0 r

zL

(7)

rR

(8)

Because the characteristic reaction time is very small relative to the others in the system the solid reactant concentration may be described by a pseudo-steady-state equation (that is, ignoring the time derivative in Eq. 4). For certain kinetics this enables one to obtain an explicit solution for Cs that can be substituted into the rate expression. For example, for a rstorder reaction Cs

k E 1 exp ka v RTs

x C in C / C in

Cf

(9)

Mathematical Model
We use a heterogeneous or two-phase model to describe an adiabatic packed-bed reactor. The corresponding energy and limiting species balances for both the gas and the solid phases are as follows C p f Tf Tf 2 Tf Cp f v f , a 2 t z z
2 f , Tf hav Ts Tf (1)

This substitution reduces the model to three equations for Tf, Cf, and Ts. By introducing the following dimensionless variables

T T in / T in
E R T in Stm vL/kav Pefh, a
h Pes ,a

HCin Cp f Tin

Sth vL/hav Pefm , a vL/Df , a Pefh,

h Pes ,

2 Cf Cf Cf 2 Cf kav Cf Cs v Df , a 2 Df , t z z (2)

vL f , a /Cp f

vL 1 s , a /Cp f

vR2 Lf , /Cp f

C p s 1

Ts 2T s 2 T s 1 s,a 1 s, t z2 ha v T s T f Hr Cs , Ts dC s Cs , Ts kav Cf Cs r dt (3)

2 Pefm , vR /LDf ,

vR2 Ls , /Cp f (10)

Le 1 (4) the model becomes

Cp s Cp f

where Tf and Ts are the uid- and solid-phase temperatures, Cf and Cs are the uid- and solid-phase concentrations, v is the supercial gas velocity, r (Cs, Ts) is the reaction rate, and
AIChE Journal November 2005

1 2 f f 1 f 1 2 Sth s f h f Pef , Pefh, a 2

Vol. 51, No. 11

(11)
3029

1 2 xf xf xf 1 1 2 s , xf 2 m xf Pef , Pefm , a 1 1 2 s 1 s 2 s 2 St h s f h h Le Pes Pes , , a s , xf where, for a rst-order reaction,

(12)

, x Da exp

1 x 1

(23)

The dimensionless quantities in the pseudo-homogeneous model are (13) Le Cp /Cp f

h Pe ,a

vL , a /Cp f
m Pe ,a

s s 1 s, x f Da Da s 1 exp Stm s 1
exp

vL D , a (24)

1 xf Lk e v

h Pe ,

vR2 L , /Cp f

m Pe ,

vR2 LD ,

Da The corresponding boundary conditions are 1 f f Pefh, a f 0 f 0 1 xf xf Pefm , a xf 0 xf 0 s 0 s 0 s 0

(14)

0 (15)

(16)

(17)

When the transport resistances between the solid and uid phase are small, the reactor may be described by the simpler pseudo-homogeneous ( )model 1 1 2 1 2 h , x h Le Pe , Pe , a 2

(18) 1 2 x x 1 x 2 x m , x m Pe , Pe , a 2 subject to boundary conditions 1 h Pe , a 0 0 1 x x m Pe , a x 0 x 0

(19)

(20)

(21)

Vortmeyer and Schaeffer20 derived a relation for the effective thermal conductivity in this model. Recent studies by Dommeti et al.15 and Balakotaiah and Dommeti21 indicate that this relation is not always valid. Steady-state numerical solutions were obtained by discretizing the spatial derivatives by nite differences in all three directions (102 grid points in the axial direction and 60 100 grid points in the radial and azimuthal directions). In some cases, global orthogonal collocation was used to discretize the radial and azimuthal directions (10 25 points in each direction). The steady-state solutions were computed either by a sparse inexact NewtonKrylov subspace method using ILU preconditioned BiCGSTAB22 or by a NewtonRaphson iteration procedure using a nonlinear equation solver (NLEQ), direct (LAPACK), and sparse (MA28) solver.23 The bifurcation diagrams were computed using pseudo-arc length continuation.24 The branches of the 2-D and 3-D states and of the codimension-1 were calculated using the Library of Continuation Algorithms (LOCA) software.25,26 Dynamic simulations were conducted using the Linear Implicit Extrapolator (LIMEX).27,28 The 1-D calculations were computed using the LAPACK option. The 2-D and 3-D calculations were performed using either the direct linear solver (MA28) or the sparse iterative linear solver (GMRES/BiCGSTAB). The sparse iterative solver in the linear step usually converged in two to three iterations. Both the steady-state and dynamic simulations of the full model are rather demanding. Thus, we rst studied a simplied version of the full model, that of a shallow reactor, to gain insight into the structure of the bifurcation diagrams of the states with the nonuniform transversal temperature, and their stability features and transitions. A similar approach was used by Balakotaiah et al.29 in their study of pattern formation in a shallow monolith reactor. This shallow reactor model was obtained by a LiapunovSchmidt reduction30 of the full model. Details of the reduction procedure used in this case are reported by Viswanathan.31 We denote by f, xf, and s the spatially averaged quantities in the shallow reactor model, that is

(22)

fd

xf

x fd

sd

(25)

where, for a rst-order irreversible reaction,

3030 November 2005

The shallow-reactor two-phase model is

Vol. 51, No. 11 AIChE Journal

df 1 1 2 f Sth s f 0 d Pefh, f 1 dx f 1 2 xf xf s , xf 0 d Pefm ,

(26)

(27)

1 1 ds 2 s Sth s f s , xf 0 h d Le Pes , (28) with (s, xf) dened by Eq. 14. The corresponding boundary conditions are f 0 xf 0 s 0
Figure 1. The rst six transversal eigenmodes.

(29)

The pseudo-homogeneous model can be reduced in a similar way. The reduced model was discretized using a second-order central difference scheme. To circumvent the singularity at the center, the grid points were placed at the radial positions j (2j 1)/(2N 1) @ j 1, N.32 After discretization, the reduced model is similar to a cell model in a circular cross section, with the diffusion terms (in the continuous model) mimicked by exchange coefcients among the cells.

x 0

j 1

(32)

Linear stability analysis predicts that a uniform steady state is stable with respect to uniform perturbations, if the two eigenvalues of 1 Le Le 1 , x ss

Stationary nonuniform temperature in a shallow reactor

We consider here the evolution and dynamics of stationary temperature patterns in an adiabatic shallow packed-bed reactor ( )using a pseudo-homogeneous model. This limiting model provides important insight about the dynamic features of the more complex models, the analysis and simulations of which are much more demanding. The mathematical model of this two variables system is 1 1 d 2 , x 0 h d Le Pe , dx 1 2 x x , x 0 m d Pe ,

, x ss

Le x , x ss 1 x , x ss

(33)

have a negative real part. In a packed-bed reactor, Le 1 and it can be proven that the stability condition det1 0 (34)

(30)

implies that in this system tr(1) 0. Thus, condition 34 is necessary and sufcient for stability and no Hopf bifurcation can occur from a state satisfying Eq. 34. Next we consider a small nonuniform spatial perturbation of the form

(31)

1J m mn e im mn , J e im 2 m mn

(35)

with (, x) dened by Eq. 23. The corresponding boundary conditions are

where m and n are the azimuthal and radial mode numbers and eim and Jm(mn) are the corresponding eigenfunctions. Because of the no-ux boundary condition, mn satises the relation dJ m mn d

Table 1. First Nine Eigenvalues Satisfying Eq. 36

No. 1 2 3 4 5 6 7 8 9 m 1 2 0 3 4 1 5 2 0 n 1 1 1 1 1 2 1 2 2

mJ m mn mnJ m1 mn 0
1

(36)

mn
1.8412 3.0542 3.8317 4.2012 5.3176 5.3314 6.4156 6.7061 7.0155

The rst nine eigenvalues mn of the spatial perturbations are reported in Table 1. Although all the perturbations with m 0 have no azimuthal dependency, all the others depend on both and . Figure 1 describes the rst six eigenmodes. Linear stability analysis of a uniform solution of the steady state of the pseudo-homogeneous model following a small perturbation of mn, dened by Eq. 35, shows that it is stable to the inhomogeneous perturbations, if both eigenvalues of
Vol. 51, No. 11 3031

AIChE Journal

November 2005

 2 1 mn 1

(37)

have a negative real part, where

1 h Le Pe , 0 Pe

0 1
m ,

2 mn h Pe ,

(38)

The neutral stability curve, at which a transition from a uniform to a nonuniform state occurs, is obtained by a simultaneous solution of the uniform steady-state solutions of Eqs. 30 32 and
0 det

(39)

Condition 39 requires that

2 mn h Pe ,

Le det1 Pe

, x ss 2 mn h ,

1 x
, x ss

(40)

The effective heat diffusion in a pseudo-homogeneous model of a packed-bed reactor is always larger than that of the species dispersion, that is, 1. The lefthand side of Eq. 42 is positive when det(1) 0. Thus, the condition for a transition from a uniform to a nonuniform state of det( ) 0 is not satised if 1. This leads to the important conclusion that for reactions described by the above model a bifurcation to a stationary nonuniform state from a uniform state, stable to uniform perturbations, cannot occur for the realistic cases of 1. However, when 1, an unstable bifurcation to a nonuniform state may occur from the unstable branch of uniform states for which det(1) 0. Extensive numerical calculations showed that all the solutions on a branch of nonuniform states are unstable whenever the branch emanates and terminates on the branch of unstable uniform states. Thus, stable nonuniform states do not exist for the realistic case of 1. A stable bifurcation to a nonuniform state may occur from the stable branch only for 1. The two-phase shallow reactor model ( )consists of three state variables. Its stability analysis is similar, but more intricate. Assuming a small spatial perturbation of the form

where
m Pe , , h D Pe , Cp ,

(41)

1J m mn e im 2J m mn e im 3J m mn e im

(43)

The above condition may be rewritten as

2 2 2 mn mn mn Le det 1 h h h Pe Pe Pe , , ,

Pe

2 mn h ,

Pe
2 mn h ,

the uniform steady-state solution is stable with respect to this perturbation if all eigenvalues of
2

Le det1

, x ss

2 2 mn 2

(44)

(42)

have a negative real part, where (the subscript denotes two-phase)

1 1 Sth 0 Sth Le

1 1 xf f , x f , s ss Le xf

f , x f , s ss

1 Sth Le s

1 s

Sth

f , x f , s ss

f , x f , s ss

(45)

and

3032

November 2005

Vol. 51, No. 11

AIChE Journal

1 1 Pefh, 0 0

0 1 1 Pefm , 0

0 0 1 1 h Le Pes ,

(46)

The neutral stability curve is obtained by a simultaneous solution of the uniform steady-state solutions of Eqs. 26 29 and
0 det

(47)

Unfortunately, we were unable to derive in this case a bound on Pefm f , , h D Pef , f , Cp f (48)

above which a stable uniform state cannot bifurcate to a nonuniform state. Extensive calculations indicated that for the kinetic model used here a uniform steady state may bifurcate only to a stationary nonuniform state. However, we are unable to prove this. Before proceeding to the regular reactor case, we shall present some numerical results describing the impact of on the evolution of the nonuniform states for the two-phase . In all the numerical simulations, unless otherwise stated, we used the following set of parameters 0.36 Le 1416
Figure 2. Neutral stability curves corresponding to the rst transversal mode (mn 1.8412) for a two-phase model of a shallow packed-bed reactor (L dp) for various in (a) the R/dp vs. Da and (b) R/dp vs. f,exit plane.

0.4
Sth 300 Pefh, a
h Pes , a 0.25

15
Stm 200 Pefm , a 5.0 Pefh, 2.0 R/dp 2 L/dp L dp

L dp

L dp L dp

L dp R/dp 2 L/dp (49)

h Pes , 1.0

We used L dp in all the shallow reactor simulations and L 30dp in all the regular reactor simulations. Figure 2 shows the neutral stability curve of the bifurcation to the rst mode (mn 1.8412) for the shallow reactor. Figure 2a shows that for 1 the stability curve is bounded between the Da values corresponding to the ignition (Dai) and extinction (Dae). However, this graph does not show from which branch of uniform states the bifurcation occurs. Thus, from now on we report the stability curves in the plane of R/dp vs. f,exit, the efuent temperature, because these plots show from which branch of uniform solutions the bifurcation occurs. Figure 2b shows that for all 1, the bifurcation occurs from the branch of unstable uniform states because the efuent temperature is bounded between the ignition and extinction temperaAIChE Journal November 2005

tures (i and e). However, for 0.165 and R/dp 7.5 one bifurcation to nonuniform temperature state occurs from a stable, uniform high-temperature state ( e), and a second one from an unstable, uniform temperature state (e i). For 0.075 and R/dp 6.5 one bifurcation is from a stable, uniform high-temperature state and the second from a stable, uniform low-temperature state. Figure 3 is a bifurcation diagram of the uniform and nonuniform states of this reactor. The nonuniform states were obtained by perturbing the uniform ones close to the bifurcation points by the rst mode and numerical continuation to compute the branch. The transition from uniform to nonuniform states occurred at a pitchfork bifurcation, at which a pair of nonuniform branches emerged. The nonuniform states bifurcating from a uniform unstable branch are unstable. In all the cases we studied, the pitchfork bifurcation from the branch of uniform stable states is subcritical, that is, the bifurcating nonuniform states are unstable. (In all the bifurcation diagrams shown here, except for Figure 10, dashed lines represent unstable transversal nonuniform states and solid segments on these branches represent stable transversal nonuniform states.) In the case shown in Figure 3 for 1.5, all the states on the two nonuniform branches are unstable. In contrast, some of
Vol. 51, No. 11 3033

Figure 3. Dependency of the uniform and nonuniform states corresponding to the rst mode on Da for a two-phase model of a shallow packedbed reactor (L dp) for 0.165 and 1.5.
Dotted line denotes unstable uniform state; dashed line denotes unstable nonuniform state; solid line denotes stable states.

Figure 4. Neutral stability curves corresponding to the third transversal mode (mn 3.8317) for a two-phase model of a shallow packed-bed reactor (L dp) for various .

Stationary nonuniform temperature in packed-bed reactors

We consider here the formation of nonuniform transversal temperature patterns in an adiabatic packed-bed reactor using a two-phase model. At the neutral stability curve a transition from a uniform to a nonuniform state occurs following a nonuniform perturbation. It is obtained by solving simultaneously the transversally uniform steady-state form of equations (Eqs. 1117) and the eigenvalue problem

the nonuniform states for 0.165 are stable. The nonuniform states are rather similar in shape to the rst mode, that is, half of the cross section is hot and half is cold. The neutral stability curve in Figure 2 indicates that for 0.075 and R/dp 6.5 the branch of the nonuniform state emanates from a stable, uniform ignited state and terminates at a stable, uniform extinguished state. In the special case of mode 1, the two states (on either branches of the nonuniform states) are mirror images of each other and have the same efuent temperature. Thus, the two states are represented by a single branch in Figure 3. Figure 4 shows the neutral stability corresponding to the third mode (mn 3.8317), which has no azimuthal dependency and has the shape of a target. Again, bifurcation from stable uniform states do not occur for values of 1. However, bifurcations from stable uniform states occur for sufciently small values of . Figure 5 is a bifurcation diagram of the uniform steady states and the stationary nonuniform ones corresponding to the third mode. The states corresponding to 1.5 emanate from unstable, uniform states and all the states on this branch are unstable. For 0.165 the branch of nonuniform state bifurcates (subcritically) from a stable, uniform ignited state and terminates at an unstable, uniform state. Some of the nonuniform states on this branch are stable (solid line in Figure 5), whereas the others are unstable. The nonuniform states were rather similar in shape to the third mode, that is, had the shape of a target. Two branches of nonuniform states exist in this case. This occurs because the pitchfork bifurcation leads to formation of two nonuniform states that have different efuent temperatures.
3034 November 2005

Figure 5. Dependency of the uniform and nonuniform states corresponding to the third mode on Da for a two-phase model of a shallow packedbed reactor (L dp) for 1.5 and 0.165.
Dotted line denotes unstable uniform state; dashed line denotes unstable nonuniform state; solid line denotes stable states.

Vol. 51, No. 11

AIChE Journal

2 1 1 1 2 1 1 mn Sth 3 1 1 2 h h Pef , a Pef , 2 2 1 1 2 2 2 mn 2 m m Pef , a xf Pef ,

(50)

f , x f , s ss

3 2
f , x f , s ss

(51)

2 3 1 1 2 3 mn Sth 3 1 2 h h Le Pes , a Pes ,

xf

2
f , x f , s ss

3 3
f , x f , s ss

(52)

subject to the boundary conditions 1 1 1 Pefh, a 1 2 2 Pefm , a 3 0

(53)

1 2 3 0 1 2 3 0

(54)

(55)
Figure 6. Neutral stability curves corresponding to the rst transversal mode (mn 1.8412) for a two-phase model of a packed-bed reactor (L 30dp) for various in (a) the R/dp vs. Da and (b) R/dp vs. f,exit plane.

Linear stability predicts the nature of the bifurcating nonuniform solutions. When a single negative eigenvalue becomes positive upon crossing the neutral stability curve, a stationary nonuniform state evolves, whereas if a pair of complex eigenvalues crosses the imaginary axis an oscillatory nonuniform state evolves. Extensive numerical simulations suggest that this three-variable model leads only to bifurcations to a stationary nonuniform state, although we could not prove this. Similarly, we were unable to derive a criterion predicting the impact of the ratio between the effective transversal dispersion of heat and mass on the stability of the uniform steady state from which the nonuniform states can bifurcate. Figure 6 shows the neutral stability curve of the bifurcation to the rst mode (mn 1.8412). Figure 6a shows that for all 1 the stability curve is bounded between the Da values corresponding to the ignition (Dai) and extinction (Dae). Figure 6b shows that for all 1 the nonuniform states bifurcate from the branch of unstable uniform states because their efuent temperature is between the ignition and extinction temperatures (i and e). This indicates that, as predicted by the shallow reactor model, no bifurcation from a stable uniform steady-state solution can occur under the practical case of 1. However, for values of 0.1 and 0.075 bifurcations to nonuniform temperature state occur from the stable, uniform high-temperature states ( e). The qualitative behavior of this model is similar to that of the shallow reactor shown in Figure 2. However, a nonuniform state bifurcated from stable uniform states for 0.165 for the shallow reactor. No such bifurcation occurred in the regular packed-bed reactor model. Numerical simulations indicated that when the neutral
AIChE Journal November 2005

stability curve has a fold in the (R/dp vs. Da) plane, such as for 0.075 in Figure 6a, then a bifurcation can occur from a stable uniform state. The neutral stability curves for the higher modes have a shape similar to that for mode 1. For example, Figure 7 shows the neutral stability for nonuniform states corresponding to the third mode (mn 3.8317) in Figure 1. Again, bifurcation from stable uniform states do not occur for values of 1. However, bifurcations from stable uniform state occur for sufciently small values of . Comparison of Figures 6b and 7 shows that the neutral stability curve for the higher mode are shifted to higher values of R/dp. The fact that the neutral stability curves for higher modes are shifted to higher values of R/dp implies that the larger the reactor diameter, the larger the number of possible stationary nonuniform states. The bifurcation diagram of the uniform steady state and of the nonuniform ones corresponding to the third mode are shown in Figure 8. The states corresponding to 1.5 emanate from unstable, uniform states and all the states on this branch are unstable. This occurs for all the states satisfying the realistic condition that 1. For 0.075 the branch of nonuniform state bifurcates (subcritically) from a stable, uniform ignited state and terminates at an unstable, uniform state. Some of the nonuniform states on that branch are stable
Vol. 51, No. 11 3035

Figure 7. Neutral stability curves corresponding to the third transversal mode (mn 3.8317) for a two-phase model of a packed-bed reactor (L 30dp) for various .

(solid line in Figure 8), whereas the others are unstable. The nonuniform states are rather similar in shape to the third mode. Two branches of nonuniform states exist in this case. This occurs because the pitchfork bifurcation leads to formation of two nonuniform states with different efuent temperatures. The solutions corresponding to mode 3 have no azimuthal dependency and have the shape of a target at any cross section of the reactor. Figure 9 describes the temperature proles within such a reactor of such a stable nonuniform state. It was computed for 0.075. Thus, it corresponds to the con-

Figure 9. Reactor temperature of stable nonuniform states in a packed-bed reactor predicted by a two-phase model and corresponding to the third mode for Da 134 and 0.075 when (a) f,exit 0.237 and (b) f,exit 0.325.

dition that the effective heat diffusivity is smaller than that of the limiting reactant. The branches of the nonuniform states corresponding to the second mode (mn 3.0542) are shown in Figure 10 for 1.5, that is, for a realistic situation, in which the heat dispersion exceeds that of the limiting reactants. In this case the two branches of the nonuniform states emanate from an unsta-

Figure 8. Dependency of the uniform and nonuniform states corresponding to the third mode on Da for a two-phase model of a packed-bed reactor (L 30dp) for 1.5 and 0.075.
Dotted line denotes unstable uniform state; dashed line denotes unstable nonuniform state; solid line denotes stable states.

Figure 10. Dependency of the uniform and nonuniform states corresponding to the second mode on Da for a two-phase model of a packed-bed reactor (L 30dp) for 1.5.
Dotted line denotes unstable uniform state; dashed line denotes unstable nonuniform state; solid line denotes stable states.

3036

November 2005

Vol. 51, No. 11

AIChE Journal

ble uniform steady state, located on the intermediate branch of solutions. Detailed linear stability calculations showed that all the states on the two branches were unstable and there was no change in stability at the limit points on the branches of these states.

Conclusions and Remarks

A pseudo-homogeneous model predicts that a stable, stationary nonuniform temperature pattern cannot form in any cross section of an adiabatic packed-bed reactor used to carry out a single reaction, having a conventional rate expression that accounts only for reactant surface adsorption, desorption, and temperature. This is explained by the fact that the transversal heat dispersion in the reactor is larger than that of the limiting reactant. This conclusion is true for all the possible modes. This conclusion was proved for a shallow packed-bed reactor described by a pseudo-homogeneous model. Extensive numerical simulations showed that both a regular and shallow packed-bed reactor described by a two-phase model exhibited the same qualitative behavioral features. A stable, stationary pattern may form in the cross section of the reactor only under the assumption that the transversal species dispersion exceeds that of the temperature. As indicated by Schmitz and Tsotsis14 and Yakhnin and Menzinger,17 this condition is not satised in packedbed reactors and is the same as that generating a Turing pattern in reaction diffusion systems. This study shows that the analysis of pattern formation in a shallow packed-bed reactor provides useful insight and guidance about pattern evolution and stability in packed-bed reactors. Moreover, it drastically decreases the numerical effort needed to conduct such studies. This nding should be used in future studies of pattern formation involving different classes of reactions. Our study is the rst to present the branches of stationary transversal pattern states in a 3D packed-bed reactor. We found that all the stationary patterned states on the branch were unstable, when the branch bifurcated from two states that were unstable to uniform disturbances. Stable stationary states were obtained only when the branch bifurcated from a uniform state stable to uniform disturbances. Only some of the states on the branch of nonuniform states were stable. The analysis of the pattern formation in an adiabatic packedbed reactor exploits the fact that the concentration and temperature satisfy the same (no-ux) boundary condition at the reactor walls. Consequently, the radial perturbations of all the variables can be expressed by the same Bessel functions. A much more intricate analysis is required when heat transfer occurs at the reactor walls. In this case different Bessel functions describe the concentration and temperature disturbances and the concentration and temperature patterns are no longer similar. It is important to conduct future analysis of this case to gain insight into the impact of the heat losses on the pattern evolution and stability. This study, however, does not provide an answer to the important question: Which reaction mechanism and operating condition can lead to transversal pattern formation in adiabatic packed-bed reactors? Several industrial reports and laboratory experiments mentioned in the introduction reported formation of hot zones in packed-bed reactors. This study suggests that a bifurcation to a stable nonuniform state may occur only if the
AIChE Journal November 2005

reaction rate dependsin addition to the temperature and surface adsorption and desorption of reactants on other rate processes, such as a periodic variation of the surface catalytic activity33or impact of subsurface adsorption desorption of reactants.34,35 The pattern formation for these more complex kinetic mechanisms will be addressed in a future publication. Our analysis is of a bed that is uniformly packed with no internal ow obstructions. Clearly, internal obstruction of the ow36 or nonuniform packing of the bed8 may also generate hot zones.

Acknowledgments
We gratefully acknowledge support of this research by grants from the National Science Foundation, the Welch Foundation and the San Diego Supercomputing Center. We thank Dr. Andrew Salinger for helpful discussions and advice.

Notation
av C Cp dp D Da E h J k k L Le Le N Pe Pe r R r R Stm Sth t T v x z specic surface area, m2/m3 concentration, mol/m3 specic heat capacity, J kg1 k1 diameter of particle, m species diffusion coefcient, m2/s Damko hler number activation energy, J/mol interfacial heat transfer coefcient, W K1 m2 Bessel function of rst kind interfacial mass transfer coefcient, m/s intrinsic reaction rate constant, 1/s reactor length, m rst Fre chet derivative Lewis number, dened by Eq. 10 Lewis number, dened by Eq. 24 ratio of transversal heat to mass dispersions radial grid points transversal perturbation matrix Peclet number, dened by Eq. 10 Peclet number, dened by Eq. 24 radial coordinate, m reactor radius, m reaction rate, mol m3 s1 dimensionless reaction rate universal gas constant, J mol1 K1 Stanton number for mass, dened by Eq. 10 Stanton number for heat, dened by Eq. 10 time, s temperature, K linear velocity, m/s conversion, dened by Eq. 10 axial coordinate, m

Greek letters
i H
adiabatic temperature rise, dened by Eq. 10 dimensionless activation energy, dened by Eq. 10 bed voidage dimensionless axial coordinate dimensionless temperature, dened by Eq. 10 thermal conductivity, W/(m.K) transversal eigenmode number dimensionless radial coordinate eigenvalue dimensionless time, dened by Eq. 10 azimuthal coordinate ith component of nonuniform perturbation heat of the reaction, J/mol

Other
averaged quantity

Vol. 51, No. 11

3037

Subscripts
a f in m n s ss axial uid phase inlet azimuthal mode number radial mode number pseudo-homogeneous solid phase transversally uniform steady state two-phase transversal 17. 18. 19. 20. 21. 22.

Superscripts
e h i m extinction mass ignition mass shallow reactor 23.

24.

Literature Cited
1. Puszynski J, Hlavacek V. Experimental study of ignition and extinction waves and oscillatory behavior of a tubular non-adiabatic xedbed reactor for the oxidation of CO. Chem Eng Sci. 1984;39:681-691. 2. Wicke E, Onken HU. Periodicity and chaos in a catalytic packed bed reactor for CO oxidation. Chem Eng Sci. 1988;43:2289-2294. 3. Wicke E, Onken HU. Bifurcation, periodicity and chaos by thermal effects in heterogeneous catalysis. In: Markus M, Muller SC, Nicolis G, eds. From Chemical to Biological Organization. Berlin/New York, NY: Springer-Verlag; 1988:68-81. 4. Rovinski AB, Menzinger M. Self-organization induced by the differential ow of activator and inhibitor. Phys Rev Lett. 1993;70:778-781. 5. Sheintuch M, Shvartsman S. Spatiotemporal patterns in catalytic reactors. AIChE J. 1996;42:1041-1068. 6. Sheintuch M. Pattern selection in a general model of convection, diffusion, and catalytic reaction. Physica D. 1997;102:125-146. 7. Sheintuch M, Nekhamkina O. Pattern formation in models of xed-bed reactors. Catal Today. 2001;70:369-382. 8. Matros YS. Unsteady Processes in Catalytic Reactors. Amsterdam: Elsevier; 1985. 9. Barkelew CH, Gambhir BS. Stability of trickle-bed reactors. ACS Symp Ser. 1984;237:61-81. 10. Viljoen HJ, Gatica JE, Hlavacek V. Bifurcation analysis of chemically driven convection. Chem Eng Sci. 1990;45:503-517. 11. Stroh F, Balakotaiah V. Modeling of reaction induced ow maldistributions in packed beds. AIChE J. 1991;37:1035-1052. 12. Nguyen D, Balakotaiah V. Flow maldistributions and hot spots in down-ow packed bed reactors. Chem Eng Sci. 1994;49:5489-5505. 13. Nguyen D, Balakotaiah V. Reaction-driven instabilities in down-ow packed beds. Proc R Soc Lond A Phys Sci 1995;450:1-21. 14. Subramanian S, Balakotaiah V. Analysis and classication of reaction driven stationary convective patterns in a porous medium. Phys Fluids. 1997;9:1674-1695. 15. Schmitz RA, Tsotsis TT. Spatially patterned states in systems of interacting catalyst particles. Chem Eng Sci. 1983;38:1431-1437. 16. Balakotaiah V, Christaforatou EL, West DH. Transverse concentration 25. 26.

27. 28.

29. 30. 31. 32. 33. 34. 35. 36.

and temperature non-uniformities in adiabatic packed bed catalytic reactors. Chem Eng Sci. 1999;54:1725-1734. Yakhin V, Menzinger M. On transverse patterns in catalytic packed bed reactors. Chem Eng Sci. 2001;56:2233-2236. Turing A. The chemical basis for morphogenesis. Philos Trans R Soc London B Biol Sci 1952;237:37-72. Segal LA, Jackson JL. Dissipative structure: An explanation and an ecological example. J Theor Biol. 1972;37:545-559. Vortmeyer D, Schaeffer RJ. Equivalence of one- and two-phase models for heat transfer processes in packed beds: One dimensional theory. Chem Eng Sci. 1974;29:485-491. Balakotaiah V, Dommeti SMS. Effective models for packed bed catalytic reactors. Chem Eng Sci. 1999;54:1621-1638. van der Vorst HA. BI-CGSTAB: A fast and smoothly converging variant of BI-CG for the solution of nonsymmetric linear systems. SIAM J Sci Stat Comput. 1992;13:631-644. Nowak U, Weimann L. A family of Newton codes for systems of highly nonlinear equationsAlgorithm, implementation, application. Technical Report TR90-10. Berlin-Dahlem, Germany: Konrad-ZuseZentrum fu r Informationstechnik; 1990. Keller HB. Numerical solutions of bifurcation and nonlinear eigenvalue problems. In: Rabinowitz PH, ed. Applications to Bifurcation Theory: Proceedings of an Advanced Seminar. New York, NY: Academic Press; 1977:159-385. Salinger AG, Burroughs EA, Pawlowski RP, Phipps ET, Romero LA. Bifurcation tracking algorithms and software for large scale applications. Int. J. Bif. and Chaos. 2005;15:1015-1032. Burroughs EA, Ramero LA, Lehoucq RB, Salinger AG. Large scale eigenvalue calculations for computing the stability of buoyancy driven ows. Sandia Technical Report 2001-0113. Albequerque, NM: Sandia National Laboratories (SAND); 2003. Deuhard P, Hairer E, Zugck J. One-step extrapolation methods for differential-algebraic systems. J Numer Math. 1987;51:501-516. Ehrig R, Nowak U, Oeverdieck L, Deuhard P. Advanced extrapolations methods for large scale differential algebraic problems. In: Bungartz H-J, Durst F, Zenger Chr, eds. High Performance Scientic and Engineering Computing. (Lecture Notes in Computational Science and Engineering). Vol. 8. Munich, Germany: Springer-Verlag; 1999:233244. Balakotaiah V, Gupta N, West DH. Transport limited pattern formation in catalytic monoliths. Chem Eng Sci. 2002;57:435-448. Golubitsky M, Schaeffer DG. Singularities and Groups in Bifurcation Theory. Vol. I. New York, NY: Springer-Verlag; 1984. Viswanathan GA. Transversal Temperature Patterns in Packed Bed Reactors. PhD Dissertation. Houston, TX: University of Houston; 2004. Middya U, Luss D. Impact of global interaction on pattern formation on a disk. J Chem Phys. 1995;102:5029-5036. Krischer K, Eiswirth M, Ertl G. Oscillatory CO oxidation on Pt(110): Modeling of temporal self-organization. J Chem Phys. 1992;96:91619172. Ivanov EA, Chumakov GA, Slinko MG, Bruns DD, Luss D. Isothermal sustained oscillations due to the inuence of adsorbed species on the catalytic reaction rate. Chem Eng Sci. 1980;35:795-803. Sales BC, Turner JE, Maple MB. Oscillatory oxidation of CO over a Pt catalyst. Surf Sci. 1981;103:54-74. Jaffe SB. Hot spot simulation in commercial hydrogenation processes. Ind Eng Chem Process Des Dev. 1976;15:410-416.

Manuscript received Dec. 23, 2004, and revision received Mar. 7, 2005.

3038

November 2005

Vol. 51, No. 11

AIChE Journal

CFD Simulation of UV Photocatalytic Reactors for Air Treatment

Fariborz Taghipour and Madjid Mohseni
Dept. of Chemical and Biological Engineering, University of British Columbia, Vancouver, BC V6T 1Z4, Canada DOI 10.1002/aic.10538 Published online August 3, 2005 in Wiley InterScience (www.interscience.wiley.com).

A photocatalytic reactor was simulated through computational uid dynamics (CFD) with surface reaction for trichloroethylene (TCE) oxidation at various pollutant concentrations, ow rates, and reactor lengths. The results were compared with those from experiments. The experimental work involved using a differential photoreactor for kinetics studies and an annular ow photoreactor for overall removal investigations under various conditions. The modeling predictions agreed closely with the experimental data within the range in which results were examined. The modeling results indicated signicant radial TCE concentration gradient and nonuniform ow distributions in the annular photoreactor. CFD was applied to predict the performance of a number of UV photocatalytic reactor design concepts, to study the impacts of some design parameters on the reactor efciency. The modeling results demonstrated that under similar ow rate conditions, the thickness of the contaminated air layer owing over the photocatalyst surface could substantially inuence the reactor performance. Thinner contaminated air layers provided more uniform radial concentration distribution of TCE and improved the reactor performance. 2005 American Institute of Chemical Engineers AIChE J, 51: 3039 3047, 2005 Keywords: CFD, photocatalysis, trichloroethylene (TCE), UV, air treatment

Introduction
Heterogeneous photocatalysis, using semiconductor particles such as TiO2 in combination with ultraviolet (UV) irradiation, has proved to be very effective for the abatement of contaminated air streams.1-3 This effectiveness along with the increasingly stringent environmental regulations has intensied research toward the development of efcient photoreactors for large-scale applications. To that end many researchers have investigated different photoreactor congurations for air treatment,1,4-6 and have attempted to understand the fundamentals of photoreactor design and development using mathematical modeling.2,5 More recently, computational uid dynamics (CFD) has been used as a tool to simulate photoreactor performance.7,8 CFD, which uses the conservation of mass and momentum equations to model uid ow, is a valuable tool for simulating
Correspondence concerning this article should be addressed to F. Taghipour at fariborz@chml.ubc.ca.

2005 American Institute of Chemical Engineers

reacting ows with high precision. It is often superior to simplied hydrodynamic models, such as plug ow or completely mixed, which do not usually reect the real ow pattern inside the reactors. By incorporating detailed description of ow in complex reactor geometries using CFD, the use of plug-ow or completely mixed assumptions and correlations for mass transfer can be avoided. In simulating reactive ows, in addition to mass and momentum equations, the mixing, transport, and production/consumption of chemical species should also be considered through species transport equations, which represent the conservation of each species in the system. The system of equations is solved in the computational domain using one of the numerical techniques (such as nite-volume). CFD has been applied to modeling uid ow and overall performance of single-phase and multiphase chemical reactors,9-12 as well as photoreactors. In photoreactors with volumetric reactions, such as UV disinfection or oxidation reactions, reaction kinetics is usually a function of radiation uence rate. As a result, radiation models describing the uence rate distribution are required for considering the radiation effect on reactor performance. In modeling photoreactors, the reaction
Vol. 51, No. 11 3039

AIChE Journal

November 2005

kinetics and radiation distribution are integrated with the reactor hydrodynamics to predict reactor performance. Kucuk Unluturk et al.13 performed CFD modeling of the UV dose distribution in a thin-lm UV reactor for processing of apple cider, and observed reasonable agreement between simulation and experimental UV dose values. Pareek et al.7 conducted CFD modeling of a three-phase photocatalytic reactor using an EulerianEulerian approach and reported on the model adequacy for predicting the experimental results and reactor performance. Models for simulating the performance of ozone-UV reactors14 and UV disinfection reactors15 have been developed, taking into account parameters such as kinetics, hydrodynamics, and uence rate within the reactor. The results indicated that decomposition of contaminants is inuenced by both reactor hydrodynamics and uence rate distribution. CFD has also been applied to simulate surface reactions in catalytic reactors. Endo et al.16 presented a CFD model for the rate of carbon nanotubes production by catalytic decomposition of xylene in a chemical vapor deposition reactor. Using gas-phase and surface reactions, the authors reported good agreement of the model with the experimental data. Seo et al.17 carried out CFD simulation of the surface reactions of a catalytic heat exchanger and found reasonable agreement between the modeling and experimental results. Heitsch18 performed CFD analysis of catalytic surface reactions in a catalytic recombiner and compared the results with the reported test data. Finally, in a study of photocatalytic destruction of gas-phase vinyl chloride (VC), Mohseni and Taghipour8 demonstrated that both CFD modeling and experimental approaches showed similar trends with respect to VC removal as a function of VC loading rate. The primary objectives of this study were: to evaluate CFDs capability in modeling the performance of UV photocatalytic reactors for use in air treatment; to use CFD to provide a better understanding of the ow characteristics and their impacts on the overall reactor performance; and to explore CFD application as a tool for improving photoreactor design. The performance of a UV photocatalytic reactor was predicted by CFD and the results were compared with those from experiments under various pollutant concentrations, gas ow rates, and reactor lengths. CFD was also applied to predict the performance of a few UV photocatalytic reactor design concepts to study the impacts of some design parameters on the reactor efciency.

P g t Conservation of Species Yi Y i i Ri Y i J t

(2)

(3)

is velocity, P is pressure, is stress In Eqs. 13, is density, v i is diffusion ux of tensor, g is gravitational force, and J species i. In the species equation (Eq. 3), in addition to convective and diffusive terms, the reactive term also appears. In this equation, Ri is the rate of production or depletion of the compounds by chemical reaction. Reactions may occur at the photoreactor wall surface and/or in the homogeneous phases. For surface reaction, the concentration of the species on reacting surface is based on a balance between the convection/diffusion of each species to/from the surface and the consumption/production rate at the surface: i n J R is (4)

where n is a unit vector normal to the surface and R s i is the rate of the production/depletion of species i arising from surface reaction. R is M w,i

r 1

R
NR

s i,r

(5)

s i where Mw,i is the molar weight of the species i and R , r is the rate of production/depletion of species i in surface reaction r. Rs i is computed as the sum of the reaction sources over NR reactions in which the species i is involved. The general form of the rth surface reaction can be considered as

i1

k f,r i,rM i O

i1

M
N i,r

(6)

CFD Modeling
The transport and chemical reactions were modeled by solving mass, momentum, and species conservation equations, using Fluent 6.0 CFD software.19 The simulation of the system was performed with a three-dimensional, steady-state, laminarow model including the photocatalytic surface reaction.19 The general forms of the governing equations for modeling the system are as follows: Conservation of Mass 0 t Conservation of Momentum
3040 November 2005

where N is the total number of species in the system, v i , r is the stoichiometric coefcient for reactant i in reaction r, v i , r is the stoichiometric coefcient for product i in reaction r, Mi is the symbol denoting species i, and kf,r is the forward rate constant for reaction r. The molar rate of production/depletion of species i in reacs i tion r, R , r , is given by

is,r R i,r i,r k f,r

j1

[C ]
Nr j,r

r j,

(7)

(1)

where Nr is the number of chemical species in reaction r, Cj,r is the molecular concentration of each reactant and product species j in reaction r, and j , r is the forward rate exponent for each reactant and product species j in reaction r. In the vapor phase, trichloroethylene (TCE) absorbs UV
Vol. 51, No. 11 AIChE Journal

Figure 1. Glass plate differential photoreactor.

slightly (1000 M1 m1) at a wavelength of 254 nm,20 and thus the homogeneous photolytic reaction in the photoreactors irradiated by 254-nm lamps was insignicant. This was further substantiated in our research, where photolytic experiments conducted with no photocatalyst present in the photoreactor, showed that UV photolysis provided 5% TCE removal from the contaminated air. Given this insignicant contribution, heterogeneous photocatalysis on the wall surface (R s i ) was the primary reaction that took place in the system. Thus, the following one-step reaction on the surface of the photocatalyst was considered C2HCl3 gas phase 3 C2HCl3 surface (8)

C2HCl3 surface OH 3 products CO2H2OHClintermediates (9) The overall degradation of TCE, involving adsorption to the photocatalyst surface followed by the oxidation reaction, was assumed to follow a rst-order kinetics. This assumption was supported by the experimental results obtained from the differential glass plate photoreactor (refer to Experimental section). Thus, the molar rate of TCE degradation in the system s i (R , r in Eq. 7) is given by is,r k C2HCl3 R (10)

number for the reactor at the highest air ow rate of 3 104 m3 s1 was about 500, which is within the laminar regime. The three-dimensional physical domain of the photoreactor was discretized with approximately 42,000 hexahedral and tetrahedral cells. Increasing the number of cells to 100,000 through solution-adaptive mesh renement changed the concentration results by 3%, indicating a reasonable sensitivity of the solution to cell size. Physical and chemical properties of the species (that is, TCE and air as contaminant-carrying medium) were specied for calculating the coefcients of the governing equations. The no-slip boundary condition was selected at the surfaces of the lamp and the reactor. Constant uid velocity with TCE mass fractions and pressure were specied as boundary conditions for the inlet and outlet ows, respectively. A segregated implicit solution algorithm was used to numerically solve the governing equations. The relative error between two successive iterations was specied with a convergence criterion of 104 for each scaled residual component of mass, velocity, and TCE concentration.

Experimental
Two types of photoreactor setups were used for the photocatalytic experiments involved in this research (complete descriptions are presented elsewhere8). Figure 1 shows the schematic of the glass-plate differential photoreactor used to determine the rate of TCE photocatalytic oxidation. It consisted of a 25-mm-wide aluminum reactor designed to allow the contaminated air to ow over the TiO2 photocatalyst (Degussa P-25, Degussa AG, Du sseldorf, Germany), that was coated on glass plates and placed in the reactor. The reactor was covered by quartz window; gaskets between the quartz window and the aluminum block created a ow passage of 25 mm (width) 3 mm (height) above the photocatalyst glass plates. The amount of TiO2 coated onto glass slides was about 16 g m2 and the
Vol. 51, No. 11 3041

where [C2HCl3] is the molar concentration of TCE. The laminar nite rate model, in which the transport equations are solved for species mass fractions with the predened chemical reaction mechanism, was chosen for reaction modeling. The reaction rates appear as reactive terms in the species conservation equations. Models of this type are suitable for simulating a broad range of reacting systems. The Reynolds
AIChE Journal November 2005

Figure 2. Setup of the annular photoreactor for experimental analysis of TCE photocatalysis.
Annular photoreactor was used for the CFD analysis.

total length of TiO2-coated slides was 22.5 mm. Two lowpressure monochromatic germicidal lamps (G10T51/2/L, Light Sources Inc., Orange, CT) with 254-nm output, provided UV illumination to the photocatalyst. The UV irradiance was regulated at 30 W m2 by adjusting the distance between the lamps and the catalyst surface. UV irradiance was measured by a radiometer (IL1700, SED240 detector, International Light Inc., Newburyport, MA). The annular ow photoreactor setup (Figure 2), on which the CFD analysis was conducted, consisted of a glass annular reactor, a mass ow controller, and a syringe pump for delivering TCE. The annular reactor (custom-made, Trojan Technologies Inc., London, Ontario, Canada) was equipped with a UV lamp and a quartz sleeve, which was made of diffused quartz to allow UV to penetrate through and irradiate/activate the TiO2 photocatalyst (Degussa P-25, Degussa AG) coated on the outside wall of the reactor. Dimensions for the annular reactor were 0.035 m OD, 0.024 m ID (also the OD of the quartz sleeve) and 0.29 m height, with the effective volume of 9 105 m3. The UV irradiance in the annular reactor and at the surface of the photocatalyst was constant at 73 W m2 for all experimental runs. TCE-contaminated air entering the photoreactor was generated using the syringe pump that allowed for the introduction of pure TCE (Fisher certied ACS) into a stream of clean air (pressurized building air) controlled by the mass ow controller (Figures 1 and 2). The retention times and inlet concentrations were adjusted using the mass ow controller for air and the injection rate of the syringe pump, respectively. Covering different portions of the UV lamp in the annular photoreactor allowed one to vary the effective photocatalyst surface area and photoreactor length. This provided an opportunity to run the experiments with different photoreactor lengths while maintaining the same inlet concentrations. Gas samples were taken from the inlet and outlet streams and were analyzed for the primary contaminant (that is, TCE). Stainless steel tubing (1.6 mm diameter) was used for online samplings from the inlet and outlet streams. Gas analysis was
3042 November 2005

conducted using a gas chromatograph equipped with mass spectrometer detector (GC/MS, Saturn 2200, Varian Associates, Palo Alto, CA) and a megabore capillary column (CPSil-8 CP5860). The injector temperature was 200C. The initial column temperature was 50C, held constant for 120 s, after which it was increased to 110C at a rate of 0.67C/s. Gas samples were injected into the GC/MS using a six-port sampling/injection valve (Valco Instruments Co., Inc., Houston, TX). A minimum number of three replicate samples were injected.

Results and Discussion Kinetics of TCE photocatalytic oxidation

The TCE photocatalytic oxidation rate was directly calculated from the change in TCE concentration and gas retention time in the differential glass-plate photoreactor.8 The effect of mass transfer was investigated by studying the oxidation rate at various air ow rates. At ow rates above 1.4 104 m3 s1, the reaction rate remained constant, indicating that the process was not mass transfer limited. This was in agreement with the results obtained by Keshmiri et al.21 using the same experimental setup and conditions. To ensure gas-phase mass transfer did not inuence the process in further experiments, the oxidation rate was determined at a ow rate of 1.7 104 m3 s1. Differential mode operation (that is, very low conversion) of the glass-plate photoreactor was conrmed by monitoring the fractional changes of pollutant concentrations that were 10% for all the experiments. Also, control experiments were conducted without a photocatalyst and under direct photolytic conditions. The results showed no signicant oxidation of TCE during the photolysis experiments, indicating that the removals obtained during the photocatalytic oxidation tests were primarily attributed to the presence of TiO2 and reactions involving OH radicals generated on the surface of the photocatalyst. Figure 3 shows that the photocatalytic oxidation of TCE on TiO2 photocatalyst followed rst-order kinetics with respect to TCE concentrations up to 1.6 g m3. These ndings agree with
Vol. 51, No. 11 AIChE Journal

Figure 3. Rate of TCE photocatalytic oxidation as a function of concentration.

Error bars represent the standard deviation for a minimum replication of three samples.

Figure 4. Modeling and experimental results of the concentrations of TCE along the length of the photoreactor (ow rate 1.7 104 m3 s1; inlet concentration 0.5 g m3).
() Represents experimental data; solid line represents CFD simulation. Error bars represent the standard deviation for a minimum replication of three samples.

those of other researchers who reported rst-order oxidation kinetics for relatively low concentrations of TCE. Because the concentration gradients along the reactor and in reactor cross section are negligible throughout the differential reactor, the rate constant of the surface reaction could be obtained directly from the TCE concentration in the reactor inlet. The kinetic information obtained from the differential reactor was then corrected for the irradiance before being used for the CFD modeling of the annular reactor. This was because the irradiance at the catalyst surface was different for the two reactors as a result of different reactor geometries and congurations. The correction for irradiance from the differential to the annular photoreactor was made based on separate experiments that showed a relatively linear relationship between the reaction rate and irradiance. When the reaction rate constant was corrected for the UV-irradiance difference in the annular and differential reactors, the surface rate constant for the annular photoreactor was 1.2 102 m3 m2 s1. The rst-order dependency of the reaction rate on UV irradiance obtained in this work and used for CFD modeling of the annular photoreactor is in agreement with the literature. Upadhya and Ollis22 reported that for systems with relatively high photo-efciency, the rate of TCE photocatalysis follows a rst-order irradiance dependency, which is believed to be the result of fairly high gas-phase TCE concentration (between 0.1 and 1.6 g m3) and high photo-efciency in the photoreactor.
2,22

reactor. The higher rate near the inlet was accurately predicted by CFD, and is the result of both the higher concentration of TCE in these areas and the rst-order kinetics behavior for TCE photocatalysis. Figure 5 shows the effect of ow rate on TCE degradation at an inlet concentration of approximately 0.5 g m3. The experimental data indicated that increasing the ow rate (or decreasing the retention time) reduces the percentage of TCE removed, which was predicted by CFD simulation. The relationship between TCE removal and ow rate was to some extent different from that predicted by the rst-order oxidation kinetics. Although rst-order kinetics behavior predicts an exponential drop in TCE removal with increasing ow rate, a nearly linear relationship was observed at low ow rates. This behavior, observed experimentally and predicted by CFD simulation, could be the result of mass-transfer limitations. At low ow

Comparison between the modeling and experimental results

CFD results were compared to those obtained experimentally for various reactor lengths (0 0.25 m), TCE concentrations (0.08 0.67 g m3), and ow rates (3 1053 104 m3 s1). Figure 4 compares the CFD-simulated and experimental results for TCE concentrations along the length of the annular photoreactor during photocatalytic oxidation (ow rate 1.7 104 m3 s1, inlet concentration 0.5 g m3). As expected, the TCE concentration decreases along the photoreactor. The results show a higher initial oxidation rate near the inlet of the photoreactor, which then gradually decreases as the concentration of the pollutant decreases along the length of the
AIChE Journal November 2005

Figure 5. Modeling and experimental results of TCE removal at various ow rates (inlet concentration 0.5 g m3).
() Represents experimental data; solid line represents CFD simulation. Error bars represent the standard deviation for a minimum replication of three samples.

Vol. 51, No. 11

3043

lation did not account for homogeneous reactions such as those initiated by photolysis, it showed some deviations from the experimental data at low TCE concentrations (Figure 6). Overall, Figures 4 through 6 provide an evaluation of the CFD simulation of the photoreactor under various operating conditions. The results demonstrate close agreement between the modeling prediction and the experimental data within the range of reactor lengths, contaminant concentrations, and ow rates that were examined. Such close agreements between CFD predictions and experimental results indicate that CFD could become a valuable tool to understand the system and enhance the design of the photoreactors for optimum photocatalyst activities. Preliminary investigations in this area are presented below.
Figure 6. Modeling and experimental results of TCE removal at various concentrations (ow rate 1.7 104 m3 s1).
() Represents experimental data of percentage removal; (f) represents experimental data of amount removed; solid line represents CFD simulation of percentage removal; dashed line represents CFD simulation of amount removed. Error bars represent the standard deviation for a minimum replication of three samples.

CFD modeling of different reactor designs

A number of attempts were made to examine the potential for using CFD simulation to improve the photoreactor performance. Different photoreactor congurations and designs were evaluated by studying their effects on reactor hydrodynamics and overall TCE removal efciency. These results were then compared to the performance of the prototype photoreactor evaluated experimentally in this research. Figure 7, Design A, shows the CFD modeling of TCE concentration along the length of the annular prototype photoreactor and the TCE deposition rate on the photocatalyst surface. Clearly, the concentration is higher on the left-hand side of the UV lamp, opposite the inlet. This could be explained by the nonuniform ow distribution in the reactor. On the opposite side of the inlet, the ow rate is considerably higher because the inlet position leads the ow to one side of the reactor. This results in a higher gas velocity, which shortens residence time and produces a higher TCE concentration in these regions. The deposition rate on the surface of the photocatalyst is also higher on the left-hand side, where the gas ow rate is higher. In an attempt to inuence these reactor hydrodynamics, the reactor inlet position was changed from perpendicular (Design A) to parallel (Design B) to the reactor length. As shown in Figure 7, Design B produces signicant changes in the concentration distribution along the reactor. Also, the deposition rate at the surface of the photocatalyst (that is, along the reactor wall) became more uniform (Figure 7, Design B). The overall reactor performance, however, was not signicantly improved, increasing only slightly from 67 to 69% (Table 1). This could be the result of a balance between the two factors associated with Design B. On the one hand, less mixing and lower diffusion of TCE to the surface of photocatalysts have a negative impact on the overall degradation of TCE (disadvantages of conguration B to A). On the other hand, more uniform ow distribution and optimum use of the photocatalyst in all regions of the photoreactor bring about a positive impact on TCE degradation (advantages of conguration B to A). Despite their similar performances, Design B has some advantages for the long-term operation of the photoreactor. This design results in a more uniform shear stress on the photocatalyst surface, which could be important if photocatalyst stability is an issue, such as in situations where the coated photocatalyst is prone to shear stress and abrasion. Furthermore, Design B provides a more uniform deposition/adsorption, which could be of consideration during the treatment of some organic compounds (such as
Vol. 51, No. 11 AIChE Journal

rates, there would be a limited degree of radial mixing along the photoreactor that, in turn, prevents TCE molecules that are farther away from the surface boundary layer from reaching the surface of the photocatalyst for adsorption and subsequent degradation. At relatively high gas velocity, increasing the ow rate will induce some degree of turbulence and mixing within certain regions of the photoreactor. Such a phenomenon would increase the diffusion rate of the chemicals to and from the photocatalyst surface. Thus, with reduced mass-transfer limitation at higher ow rates, the percentage removal and reactor behavior approach an exponential shape and the removal rate is less dependent on ow rate. The removal of TCE at various concentrations is illustrated in Figure 6. Both experimental data and CFD modeling indicate the percentage removal of TCE remains constant at relatively high concentrations (0.2 g m3). Such behavior can be explained based on the rst-order kinetics for the oxidation of TCE. For rst-order kinetics, the rate of oxidation increases with concentration, resulting in higher amounts of TCE removed from the system (Figure 6). This, in turn, corresponds to a relatively constant percentage removal of TCE throughout the range of concentrations between 0.2 and 0.7 g m3. At concentrations 0.2 g m3, however, there is a discrepancy between the CFD modeling and experimental results; the model does not predict the high degree of removal observed experimentally. This difference was likely the result of degradation caused by photolysis that is generally insignicant, especially at higher concentrations. However, when the concentration of TCE decreases and the photocatalytically initiated surface reactions become less pronounced, the small amount of removal obtained by photolysis could become more signicant. This hypothesis was further examined using separate photolytic experiments with the glass-plate photoreactor at TCE inlet concentrations of 0.1 g m3. About 4 5% TCE removal was observed, indicating the potential, albeit small, impact of photolysis to the overall removal of TCE. Given that CFD simu3044 November 2005

Figure 7. CFD modeling of contours of mass fraction and surface deposition rate (kg m2 s1) of TCE for the cross-ow inlet (A) and parallel-ow inlet (B) photoreactors (ow rate 1.33 104 m3 s1; inlet concentration 0.54 g m3).
Inlet is on top of the photoreactor (Z direction). There is no photocatalyst in the areas where the deposition rate is zero (at the top and bottom of the photoreactors).

aromatics) that form polymeric intermediates that inactivate the photocatalyst. Finally, uniform ow through the reactor results in a more consistent use of the photocatalyst, which could also be regenerated more effectively/consistently upon inactivation. In Figure 8, Design A depicts CFD modeling of mass fraction of TCE at the prototype photoreactor cross section, located 0.085 m from the reactor inlet (ow rate 1.33 104 m3 s1, inlet concentration 0.54 g m3). The results show that TCE concentration is not symmetric in the reactor cross section and that the overall concentration is higher at the opposite side of the inlet (the left-hand side), where there is a higher gas ow rate and shorter local residence time. Design A also shows a radial concentration gradient between the outer and inner surfaces of the reactor with lower concentration at the outer surface of the reactor, where the photocatalyst is present in the form of coating on the reactor wall. This concentration gradient is anticipated because TCE degradation takes place only on the surface of the TiO2 photocatalyst. Therefore, changing the contaminated air layer owing over the photocatalyst surface, by altering the distance between the lamp sleeve and the photocatalyst surface, could provide more uniform radial concentration distribution and potentially improve reactor performance. Two other designs were simulated by changing the diameter of the photoreactor sleeve (from 23 to 27.5 and 14 mm) to

increase/decrease the air layer by a factor of 2 (from 4.5 mm in Design A, to 9.0 mm in Design C, and to 2.25 mm in Design D). In all the design alternatives, the inlet ow rate and TCE concentration remained the same (ow rate 1.33 104 m3 s1, inlet concentration 0.54 g m3). Also, the photoreactor diameter and, in turn, the photocatalyst surface area, remained unchanged. Figure 8 illustrates that a thicker air layer (Design C) creates a nonuniform concentration distribution, but with a thinner air layer (Design D), the concentration distribution is more uniform in the photoreactor cross section. In addition, as the air layer in the photoreactor decreases (Designs C, A, and D, respectively), the radial concentration gradient becomes more uniform. This results in an increase in the diffusion rate of the chemicals to the photocatalyst surface and, thus, enhancement in the efciency of the catalyst for eliminating the contaminant of interest. The CFD modeling results (Table 1) predict an improvement in the photoreactor performance (from 67 to 75% degradation) as a result of reducing the air layer by a factor of 2 (from 4.5 mm in Design A to 2.25 mm in Design D). This improvement, predicted by CFD, was substantiated by preliminary experiments that showed the rate of TCE removal increased by almost 150% when the air layer was decreased by a factor of 3 in a separator at-plate reactor conguration. Although this experiment was performed in a different reactor conguration, the preliminary test proved that smaller air lay-

Table 1. Reactor Performance at Various Conceptual Designs*

Design A B C D Inlet Inlet Inlet Inlet Description cross ow, air layer 4.5 mm parallel ow, air layer 4.5 mm cross ow, air layer 9.0 mm cross ow, air layer 2.25 mm TCE Mass Fraction Outlet/Inlet 1.5 104/4.4 104 1.4 104/4.4 104 1.9 104/4.4 104 1.1 104/4.4 104 TCE Removal (%) 67 69 56 75

*The performance is reported in terms of the outlet/inlet TCE mass fraction and the TCE removal (ow rate 1.33 104 m3 s1).

AIChE Journal

November 2005

Vol. 51, No. 11

3045

Figure 8. CFD modeling of contours of mass fraction of TCE at the reactor cross section (0.085 m from the inlet) for the 4.5-mm (A), 9-mm (B), and 2.25-mm (C) air layer photoreactors (ow rate 1.33 104 m3 s1; inlet concentration 0.54 g m3).
Cross ow inlet is from the right-hand side of the photoreactor, resulting in higher ow rate and concentration on the left-hand side (see Figure 7A for the overall geometry).

ers can increase the removal rate. Although narrowing the air layer will increase the pressure drop, it is anticipated that the greater reactor performance will compensate for this disadvantage. Reaching an optimal air layer might be achieved by considering a balance between the highest possible removal rates at an appropriate pressure drop.

Conclusions
Photocatalytic oxidation of TCE followed rst-order oxidation kinetics for the range of TCE concentrations between 0.1 and 1.6 g m1. CFD modeling of a photocatalytic reactor with surface reaction for TCE oxidation at various reactor lengths, pollutant concentrations, and ow rates predicted the experimental data well. CFD modeling of the annular photocatalytic reactor provided detailed information on concentration gradients of the contaminant in the reactor and provided accurate analysis and explanation of the experimental results. Improvement in photoreactor performance could be achieved by using the information provided by modeling. CFD modeling demonstrated that under similar ow rate conditions, the thickness of contaminated air ow (the contaminated air layer owing over the photocatalyst surface) inuences the reactor performance. A thinner contaminated air layer provided more uniform radial concentration distribution of TCE and improved the reactor performance, within the range of air layers (2.259.0 mm) and ow rates (1.33 104 m3 s1) examined.

j , r forward rate exponent for each reactant and product species j in reaction r Cj,r molecular concentration of each reactant and product species j in reaction r, kg m2 s1 g gravitational force, m s2 i diffusion ux of species i, kg m2 s1 J k rst-order surface reaction rate constant, m3 m2 s1 Mi symbol denoting species i Mw,i molar weight of the species i, kg kmol1 N total number of species in the system P pressure, N m2 r photocatalytic reaction rate, g m2 min1 Ri rate of production or depletion of species i arising from volumetric reactions, kg m3 s1 Rs rate of surface production or depletion of species i, kg m2 i s1 s 2 Ri , r rate of molar production or depletion of species i, kmol m s1 v velocity, m s1 Yi species i

Literature Cited
1. Alfano OM, Behnemann D, Cassano AE, Dillert R, Goslich R. Photocatalysis in water environments using articial and solar light. Catal Today. 2000;58:199-230. 2. Dibble LA, Raupp GB. Fluidized-bed photocatalytic oxidation of trichloroethylene in contaminated air streams. Environ Sci Technol. 1992;26:492-495. 3. Mohseni M, David A. Gas phase vinyl chloride (VC) oxidation using TiO2 based photocatalysis. Appl Catal B Environ. 2003;46:219-228. 4. Raupp GB, Nico JA, Annangi S, Changrani R, Annapragada R. Twoux radiation-eld model for an annular packed-bed photocatalytic oxidation reactor. AIChE J. 1997;43:792-801. 5. Hossain MM, Raupp GB. Radiation eld modeling in a photocatalytic monolith reactor. Chem Eng Sci. 1998;53:3771-3780. 6. Lim TK, Kim SD. Trichloroethylene by photocatalysis in annular ow and annulus uidized bed photoreactors. Chemosphere. 2004;54:305312. 7. Pareek VK, Adesina AA. Analysis of photocatalytically reactive systems: Kinetic modeling and reactor design via computational uid dynamics. In: Nalwa HS, ed. Handbook of Photochemistry and Photobiology. Stevenson Ranch, CA: American Scientic Publishers; 2003:345-412. 8. Mohseni M, Taghipour F. Experimental and CFD analysis of photo-

Acknowledgments
The authors acknowledge nancial support from the Natural Sciences and Engineering Research Council of Canada (NSERC), Canada Foundation for Innovation (CFI) New Opportunity Fund, and British Columbia Knowledge Development Fund (BCKDF).

Notation
density, kg m3 stress tensor, N m2 v i , r stoichiometric coefcient for reactant i in reaction r v i , r stoichiometric coefcient for product i in reaction r

3046

November 2005

Vol. 51, No. 11

AIChE Journal

9. 10. 11. 12. 13. 14.

catalytic gas phase vinyl chloride (VC) oxidation. Chem Eng Sci. 2004;59:1601-1609. Hjertager LK, Hjertager BH, Solberg T. CFD modelling of fast chemical reactions in turbulent liquid ows. Comput Chem Eng. 2002;26: 507-515. van Baten JM, Ellenberger J, Krishna R. Scale-up strategy for bubble column slurry reactors using CFD simulations. Catal Today. 2003;79/ 80:259-265. Marchisio DL, Barresi AA. CFD simulation of mixing and reaction: The relevance of the micro-mixing model. Chem Eng Sci. 2003;58: 3579-3587. Gamwo IK, Halow JS, Gidaspow D, Mosto D. CFD models for methanol synthesis three-phase reactors: Reactor optimization. Chem Eng J. 2003;93:103-112. Kucuk Unluturka S, Arastoopour H, Koutchma T. Modeling of UV dose distribution in a thin-lm UV reactor for processing of apple cider. J Food Eng. 2004;65:125-136. Taghipour F, Elyasi S, Sozzi A. Simulation of ozone-UV reactors for water treatment (Water Environment Management Series). Proc of 2nd IWA Leading-Edge Conf on Water and Wastewater Treatment Technologies, Prague, Czech Republic, 2004. Water Intelligence Online IWA Publishing. London: International Water Association; 2005.

15. Taghipour F. Development of a CFD-based model for photo-reactor simulation. Parallel Computational Fluid Dynamics. London, UK: Elsevier Science; 2004:243-250. 16. Endo H, Kuwana K, Saito K, Qian D, Andrews R, Grulke EA. CFD prediction of carbon nanotube production rate in a CVD reactor. Chem Phys Lett. 2004;387:307-311. 17. Seo YS, Yu SP, Cho SJ, Song KS. The catalytic heat exchanger using catalytic n tubes. Chem Eng Sci. 2003;58:43-53. 18. Heitsch M. Fluid dynamic analysis of a catalytic recombiner to remove hydrogen. Nucl Eng Des. 2000;201:1-10. 19. Fluent Users Manual. Lebanon, NH: Fluent Inc.; 2002. 20. Hubrich C, Stuhl F. The ultraviolet absorption of some halogenated methanes and ethanes of atmospheric interests. J Photochem. 1980; 12:93-107. 21. Keshmiri M, Mohseni M, Troczynski T. Development of novel TiO2 sol-gel-derived composite and its photocatalytic activities for trichloroethylene oxidation. Appl Catal B Environ. 2004;53:209-219. 22. Upadhya S, Ollis DF. A simple kinetic model for the simultaneous concentration and intensity dependence of TCE photocatalyzed destruction. J Adv Oxidat Technol. 1998;3:199-202.
Manuscript received Sep. 12, 2004, and revision received Mar. 1, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

3047

MATERIALS, INTERFACES, AND ELECTROCHEMICAL PHENOMENA

Characterization and Drainage Kinetics of Colloidal Gas Aphrons

Divesh Bhatia, Gaurav Goel, Sidhartha K. Bhimania, and Ashok N. Bhaskarwar
Dept. of Chemical Engineering, Indian Institute of Technology, Delhi, Hauz Khas, New Delhi, 110 016, India DOI 10.1002/aic.10552 Published online September 9, 2005 in Wiley InterScience (www.interscience.wiley.com).

The variation of air holdup and stability of colloidal gas aphrons (CGAs) with stirring time, surfactant concentration [6, 8.1, 10, and 12 mM sodium lauryl sulfate (SLS)], and stirring speed (3500 7000 rpm) was investigated, and an empirical correlation between the air holdup and stirring time was obtained. A rst-order model for the drainage of CGAs was proposed and it correlated very well with the experimental data. A population balance model was proposed and correlated to the rst-order drainage model to obtain an expression for the size distribution of CGA bubbles. A narrower size distribution at large stirring times was observed, which was correlated to the stability of CGAs. The stabilization of CGA dispersion with attainment of equilibrium size distribution at longer stirring times, analogous to homogenization of emulsions, was observed and an empirical expression for this time was reported. Mixing of two oppositely charged CGAs, generated using SLS (anionic) and cetyl trimethyl ammonium bromide (CTAB) (cationic), showed no effect on their stability, air holdup, and drainage kinetics. 2005 American Institute of
Chemical Engineers AIChE J, 51: 3048 3058, 2005

Keywords: colloidal gas aphrons, drainage kinetics, size distribution, oppositely charged CGAs, stability; air hold-up

Introduction
Sebba1 rst described colloidal gas aphrons (CGAs) as microfoam with colloidal properties. In a later study, Sebba2 dened CGAs as microbubbles created by intense stirring (5000 10,000 rpm) of a surfactant solution. Characterization of CGAsthat is, the study of effects of various parameters, such as surfactant concentration, salt concentration, stirring time, pH, and temperature on air holdup and stability of CGAswas done by Jauregi et al.3 A review of the literature on CGAs was presented by Jauregi and Varley.4 An updated summary of the literature on stability and drainage of CGAs is given in Table 1.

Theory Structure and properties of CGAs

The term colloidal was used because of the small size of the bubbles (10 100 microns in diameter). Sebba2 proposed that CGAs consisted of a gaseous inner core surrounded by a thin aqueous surfactant lm composed of two surfactant layers and an outer electrical double layer that stabilized the structure, as shown in Figure 1. However, there is no real conclusive evidence available for this structure. CGAs are gas microbubbles and thus dynamic systems, continuously undergoing changes resulting from creaming, bubble breakage, coalescence, and disproportionation. Therefore, they possess a limited stability. Some of the salient features of CGAs can be summarized as follows: (1) Larger specic surface area and higher stability than those of normal bubbles because of their small size, like charges, and an electrical bilayer. (2) Flow properties similar to those of water and a high gas
Vol. 51, No. 11 AIChE Journal

Correspondence concerning this article should be addressed to A. N. Bhaskarwar at ashoknb@chemical.iitd.ernet.in.

2005 American Institute of Chemical Engineers

3048

November 2005

Table 1. Summary of the Open Literature on Stability and Drainage of CGAs

Drainage Velocity Bubble Size Stability and Air Holdup Yes

Author(s) Jarudilokkul et al.5

Surfactant* Tween 20 (n)

Parameters Studied Initial protein concentration, pH, protein/CGA volume ratio, surfactant concentration, rpm, stirring time, NaCl concentration pH, surfactant concentration, salt concentration, temperature

Conclusions Stability increases with increasing surfactant concentration and stirring time. Separation decreases with increasing surfactant concentration. pH should be such that electrostatic interactions favor separation. Aphron diameters could be predicted using models proposed for bubble breakage in agitated tanks and liquid drainage in foams and CGA dispersions. Drainage rate was predicted using foam drainage model modied for CGAs. Electron microscopy and X-ray diffraction techniques indicate the existence of an interface of many layers. Speed of stirring and pH had little effect on separation parameters. Concentration of protein and surfactant in the initial mixture had an important effect on separation. The optimum protein recovery was 95% without signicant loss of lysozyme activity. Surfactant and salt concentration had maximum effect. Empirical equation for air holdup and half-life determined. Power required for generating CGAs decreases with increase in gas holdup. Bubble size in the range 30300 m. Mean diameter for Tergitol was reported to be less than that for SDBS and HTAB. Increase in surfactant concentration reduced the mean diameter for all three surfactants. Increase of ionic strength reduced mean diameter for ionic surfactants, but it had no effect on the nonionic surfactants. Drainage of CGAs was shown to be consistent with hindered rising of 35 m spherical bubbles with 0.75 m shells in downward owing solution. The assumption of no coalescence may not hold good at later intervals. Ascertaining CGA dispersionfoam interface provides another limitation.

Jauregi et al.6

AOT (a)

Yes

Yes

Yes

Jauregi and Varley4

AOT (a)

Stirring time, ionic strength, initial mass of protein, pH, protein/CGA volume ratio, magnetic stirrer speed AOT concentration, temperature, stirring time, salt concentration, pH Drainage time, type of surfactant, surfactant concentration, ionic strength Yes

Yes

Jauregi et al.3

AOT (a)

Yes

Chaphalkar et al.7

HTAB (a) SDBS (c) Tergitol 15-S-12 (n)

Yes

Amiri et al.8

TTAB (c)

Yes

Yes

*n, nonionic surfactant; c, cationic surfactant; a, anionic surfactant.

content (55 65%), making them one of the lightest compressible liquids at ordinary temperature and pressure conditions. (3) Hydrophobicity of the encapsulating surfactant shells, enabling them to collect oil globules, protein molecules, and so forth. These features of CGAs provide applications such as the removal of naphthalene from a contaminated soil matrix,9 separation of ne bers from a lean slurry of cellulosic pulp in a otation column,10 and extraction of copper from aqueous acidic solution in LIX 622lkerosene (mixture of 5-dodecylsalicyal-dioxime and tridecanol in kerosene),11 to cite but a few.

bubbles or CGAs. Characterization of CGAs should involve considerations of the stability of the dispersion of microbubbles and the air holdup. Stability. The stability of CGAs, measured in terms of half-life (tdh), is dened as the time taken by the CGAs dispersion bulk liquid interface to reach half its nal height. The aphrons phase separates easily from the bulk-liquid phase because of its buoyancy. Figure 1 schematically shows the formation and drainage of CGAs as well as the state at half-life of CGAs. Air Holdup. Air holdup is dened as the volume percentage of entrapped air in the CGA dispersion, expressed as V To x 100 V To (1)

Characterization of CGAs
According to Sebba,2 the formation of CGAs occurs as follows. The surface waves, generated as the result of intense stirring, strike against the bafes. They entrap a thin lm of atmospheric air between the liquid and the surface of bafes, and reenter the solution. This lm of air breaks into microAIChE Journal November 2005

where VTo is the initial volume of CGA dispersion (m3) and x is the initial volume of surfactant solution (m3).
Vol. 51, No. 11 3049

Figure 1. CGA diagrams.

3050

November 2005

Vol. 51, No. 11

AIChE Journal

Models for liquid drainage from CGAs

Model 1. The rate of drainage of liquid from CGAs is dened as dVa dtd

n / t d v xn v in D B dR 0

R1

(2)

where td is drainage time (s), and Va is the volume of CGA dispersion at time td (m3). A rst-order model is proposed for liquid drainage wherein the rate of drainage at a particular time is proportional to the volume of CGA dispersion still present at that time. Thus, the rst-order drainage equation is dVa mVa dtd
1

where n is the number of aphrons per unit length; vx is the velocity of an aphron in the vertical direction; vi dL/dtd, where L is the diameter of the aphron; D is the number of aphrons collapsing in unit time per unit length; and B is the number of aphrons forming in unit time per unit length. Because the subregion R1 is arbitrary, the integrand must vanish identically. D, B, and vi are all equal to zero for our system because of the assumptions (3), (4), and (5), respectively. Thus, the population balance becomes n v xn 0 td x

(7)

(3)

where m is the rate constant for drainage (s ). From the volume balance on CGA dispersion bulk liquid system, assuming no air escapes from the system, we obtain V To V L V a (4)

where n n(x, td, L). Suppose the aphrons are moving at their terminal velocity and Stokes law is applicable for individual aphrons of different sizes, even if in the presence of other sized aphrons. The velocity of the aphrons is v x L gL 2 cL 2 18

(8)

where VL is the volume of surfactant solution in the bulk-liquid phase collected at the bottom at time td, (m3). By combining Eqs. 3 and 4, we obtain d V To V L mVTo VL dt d

(5)

where is the difference between the densities of the aphrons and the surfactant solution and c (g/18) is a constant for the system. Substituting the value of vx(L) from Eq. 8 into Eq. 7, we obtain L 2n n c 0 td x It has been assumed that the diameter L does not vary with respect to time, and thus for rising aphrons with x. Thus, n n cL 2 0 td x

The initial condition (a) for Eq. 5 is VL 0 at td 0, and the solution of Eq. 5 subject to initial condition (a) is VTo VL mtd ln VTo

(6)

Model 2. Consistent with actual observations, a foam layer of constant thickness is assumed to be present at the top of the CGA dispersion. Because of buoyancy, CGAs rise to the aphronfoam interface, where they release most of the associated surfactant solution that drains down to enter the bulksolution pool at the bottom. The following assumptions have been made while performing a population balance on the aphrons in the CGA dispersion: (1) Aphrons are spherical and they are numerous enough to approximate their size distribution by a continuous function. (2) Each aphron has an identical trajectory in the aphronphase space. (3) There is no breakage of aphrons. (4) No new aphrons are formed in the system. (5) The size of a particular aphron does not change with time. A population balance for aphrons in a xed subregion R1 of the phase space is
AIChE Journal November 2005

(9)

Boundary conditions (b) and (c) for Eq. 9 are, respectively, n(td 0) n0 and n(x 0) 0. Solving Eq. 9 using Laplace transformation with boundary conditions (b) and (c), we obtain n n0 0 td x cL 2

n0

td

x cL 2

Let, at a particular time td, a bubble of diameter L reach a height x. Equation 8 would then yield
Vol. 51, No. 11 3051

x ct d

(10)

V L V To 1 0

1 VG

(15)

and the above solution becomes n n0 n0 0 L L L L (11) (12)

where 0 is the initial air holdup and VG is the volume of gas in the CGA dispersion. Equations 14 and 15, with the assumption of constant gas holdup with drainage time, yield dV L 1 0 g dt d 108 0

Consider an equivalent cylinder of diameter Le and height L having the same volume as that of a spherical aphron of diameter L. Thus,

L 4n 0 L dL

(16)

L L L 6 4
3 2 e

For monodisperse aphrons in CGA dispersion (that is, n0(L) n0 a constant), Eq. 14 leads to dV G gn0 dt d 108

(13)

That is, Le (2/6)/L. The volume of aphrons of sizes between diameter L and L dL leaving the aphronfoam interface in time dtd is given by
L dV G dtd

L4 dL

and further, in view of Eqs. 8 and 10, we obtain n0 H dV G dt d 30 td

2 vx LLe nH, td , LdL 4

5/ 2

18 g

3/ 2

where H is the height of the aphronfoam interface. Thus, the total volume of aphrons leaving the aphronfoam interface is

Using Eq. 15 and the denition of c, we obtain dV L 1 0 n 0 H dt d 30 0 td

dV G

vx LL nH, td , LdL dtd 4

2 e

5/ 2

c 3/ 2

(17)

That is, dV G dt d

Thus, Eq. 17 represents the liquid-drainage rate for a monodisperse CGA dispersion. To obtain n0 as a function of the aphron diameter, we equate the value of dVL/dtd from Eq. 6 to that from Eq. 16: mV Toe mtd 1 0 g 108 0

vx LL2 e nH, td , LdL 4

L 4n 0 L dL

By inserting the values of vx(L) and Le from Eqs. 8 and 13, respectively, we obtain

that is,

g dV G dt d 108

L nH, td , LdL
4

H/ctd

L 4n 0 L dL

108 mV To 0 mtd e 1 0 g

Recalling the values of n from Eqs. 11 and 12, we obtain dV G g dt d 108

L4 n0 LdL

L4 n0 LdL

g 108

It is assumed that L4n0(L) is a slowly varying function of time. Taking the derivative with respect to td, and substituting the value of td from Eq. 10, we obtain n 0 L Thus, n 0 L a b 7 exp L L2 3888 2m 2V To 0H e 18mH/gL 1 0 g 2 2 L7

(18)

L4 n0 LdL (14)

The volume of liquid drained is

3052 November 2005 Vol. 51, No. 11

(19)
AIChE Journal

Figure 2. Half-life vs. stirring time for various SLS concentrations and stirring speeds.

where a and b are constants for a given system given by 3888 2m 2V To 0H a 1 0 g 2 2 b 18 m H g

Experimental Approach
The experimental setup consists of a 5-L beaker with two vertical inverted L-shaped bafes as shown in Figure 1. The surfactant solution is stirred in the beaker with a spinning disc at a constant stirring speed (3000 rpm). The initial surfactantsolution level was 2 cm above the spinning disc. To study the effects of parameterssurfactant concentration, stirring speed, and time of stirringnumerous sets of experiments were performed using surfactant concentrations below the critical micelle concentration (CMC) of SLS (6 mM), equal to the CMC (8.1 mM), and above the CMC (10 and 12 mM). One preliminary exploratory experiment was run for a long enough time such that there was no further change in the air holdup, that is, the steady state was achieved. Experiments were performed for various stirring speeds, ranging between 3000 and 7000 rpm. For the above sets of experiments performed, samples (100 mL) of the stirred CGAs dispersions were taken out into measuring cylinders at different stirring-time intervals, and the rise of the (CGA-dispersion)(bulk-liquid) interface with drainage time was noted. In a separate set of experiments, CGAs were prepared from an anionic surfactant (8.1 mM SLS) and a cationic surfactant (2 mM CTAB) simultaneously, in two separate CGA generators. Methyl red dye was added to the SLS-based CGA generator, which provided a yellow color to the CGAs. Equal volumes (50 mL each) of the two oppositely charged CGAs were then stirred together on a magnetic stirrer for 30 s, and then poured into a 100-mL measuring cylinder. Simultaneously, 100-mL samples of SLS- and CTAB-based CGAs were separately charged to similar measuring cylinders, and the rise of the (CGA-dispersion)(bulk-liquid) interface with drainage time was recorded for the three cases.

(20)

(21)

By inserting the aphron-size distribution function n0(L) from Eq. 19 into Eq. 16, we obtain dV L 1 0 ac bctd/H e dt d 0 12 b which leads to 1 0 aH 1 e bctd/H 0 12 b 2

VL

(22)

We therefore deduce the following equation: dV L bc dt d H

1 0 aH VL 12 b 2 0

(23)

Thus, Eq. 23 predicts a rst-order liquid-drainage rate for the CGAs with the aphron size distribution represented by Eq. 19.

Figure 3. Air holdup vs. stirring time for various SLS concentrations and stirring speeds. AIChE Journal November 2005 Vol. 51, No. 11 3053

duplicates and the average experimental error in measurements was estimated to be within 7%.

Air holdup
Figure 3 shows the semilog plots of percentage air holdup vs. stirring time (nondimensionalized with unit time) for four different surfactant (sodium lauryl sulfate) concentrations (6, 8.1, 10, and 12 mM), and each at three to four stirring speeds (3800 6500 rpm). Figure 3 shows that as the time of stirring increases, the percentage air holdup increases irrespective of the surfactant concentration and stirring speed. This is because more air is entrapped with time into the system until the steady-state dispersion is reached. At low stirring speeds (3000 rpm), there was no observable CGA formation. With an increase in stirring speed beyond 3000 rpm, for a xed time of stirring and surfactant concentration, the air holdup increased, which can be attributed to the fact that with an increase in the stirring speed, the number of surface waves striking the bafes per unit time increased, thereby entrapping more air microbubbles in the CGA dispersion during the same time period.

Figure 4. Best-t line for air holdup fraction vs. ln(stirring time fraction).

Results and Discussion Stability

Figure 2 shows the plots of half-life times vs. stirring time for different surfactant concentrations (sodium lauryl sulfate, 6 10 mM), each at three to four stirring speeds ranging from 3600 to 6500 rpm. For short stirring times, the (CGA-dispersion)(bulk-liquid) interface was not clear, indicating a broad size distribution of aphrons. Data for such short times are therefore not being reported here. Figure 2 implies that the size distribution of aphrons improves with time, surfactant concentration, and stirring speed, seen in a general increase in the half-life time. The concentration range used by Jauregi et al.,3 who reported an increase in half-life with concentration, was very large (0.134 mM for AOT). In contrast, the concentration range in the current experiments was much smaller (6 10 mM for SLS). As a result, there was only a limited increase in the half-life, that is, about 50 60 s. The increase in the stability of CGAs with surfactant concentration perhaps results from the increased repulsive forces between aphrons at higher concentrations of surfactant either in the surfactant shells or in the bulk-liquid phase.13 Some of these experiments were run in

Empirical correlation for air holdup

Figure 3 shows that for a given surfactant concentration, the semilog plots of air holdup vs. stirring time (nondimensionalized with unit time) are linear with almost equal slopes, and with increasing intercepts at higher stirring speeds. In an effort to bring all experimental data points onto a single curve, air holdup was scaled with the maximum air holdup (max) for a particular stirring speed, and stirring time (t) was scaled with the time required to attain steady state (ts). The data for four different surfactant concentrations (that is, 6, 8.1, 10, 12 mM SLS) mapped onto a single line of the form: / max k lnt/ts p (24)

Figure 4 shows that the nal air holdup correlation for the SLSwater system is

Figure 5. Variation of max and ts with stirring speed. 3054 November 2005 Vol. 51, No. 11 AIChE Journal

Figure 6. Comparison between experimental air holdup values and the values obtained from empirical correlation.

Figure 7. Plots of ln[(VTo VL)/VTo] vs. td at 6 mM SLS and 5700 rpm.

/ max 0.2407 lnt/ts 0.9669

(25) max 0.004s 44.017 and ln ts 2.0858 ln s 20.74 (27) (26)

with a correlation coefcient of R2 0.9275. This correlation has an implicit dependency of max and ts on the surfactant concentration and the stirring speed. Two out of the four surfactant concentrations used were above the CMC of SLS (8.1 mM), and one concentration was below the CMC. Figure 5 shows the variation of max and ts, respectively, with the stirring speed for all four surfactant concentrations. The corresponding R2 values of 0.6784 and 0.6959 imply that the surfactant concentration, within the range used, does not have a signicant effect on the air holdup. This observation is supported by Matsushita et al.,14 who have reported no increase in air content for concentrations of CTMAB 0.5 g L1. Interestingly, however, we nd from our experiments for SLS that a concentration 25% lower than the CMC also leads to only a slight decrease in air holdup. The empirical dependency of max and ts on the stirring speed, as obtained from Figure 5, is

where s is stirring speed (rpm). By combining Eqs. 2527, we obtain the empirical dependency of air holdup on the stirring time and stirring speed as 0.004 s 44.017 0.2407 ln

t s
2.0858

4.025

(28)

Figure 6 shows a comparison of the experimental values of air holdup with the calculated values from Eq. 28. An excellent

Table 2. Variation of Rate Constant with Stirring Speed and Stirring Time
6 mM Stirring Speed (rpm) 5700 Stirring Time (min) 6 10 14 20 24 28 8 12 16 20 25 Negative of Slope (Eq. 6), m 0.0021 0.0011 0.0007 0.0007 0.0007 0.0008 0.0012 0.0007 0.0007 0.0006 0.0007 Correlation Coefcient 0.72 0.91 0.95 0.92 0.98 0.99 0.8 0.94 0.99 0.98 0.99 Stirring Speed (rpm) 3600 Stirring Time (min) 12 16 20 25 30 36 10 14 18 23 29 35 6 10 14 18 6 8 12 16 8.1 mM Negative of Slope (Eq. 6), m 0.0032 0.0032 0.0026 0.0016 0.0015 0.0012 0.002 0.0013 0.0012 0.0009 0.0008 0.0008 0.0014 0.0007 0.0006 0.0007 0.0015 0.0011 0.0008 0.0013 Correlation Coefcient 0.99 0.95 0.97 0.99 0.96 1 1 0.97 0.99 0.99 0.99 0.99 0.98 1 0.99 1 0.98 0.97 1 0.54 Stirring Speed (rpm) 4800 Stirring Time (min) 8 12 16 21 26 29 6 10 14 18 22 26 30 6 10 14 18 22 26 10 mM Negative of Slope (Eq. 6), m 0.0026 0.0012 0.0009 0.001 0.0007 0.0006 0.0015 0.001 0.0008 0.0007 0.0007 0.0007 0.0007 0.0013 0.0007 0.0006 0.0007 0.0006 0.0006 Correlation Coefcient 0.48 0.63 0.68 0.92 0.98 0.95 0.54 0.87 0.93 0.83 0.86 0.98 0.99 0.8 1 1 0.97 0.99 1

6000

5230

5500

5750

6000

6500

AIChE Journal

November 2005

Vol. 51, No. 11

3055

Figure 10. Comparison of experimental and model-predicted liquid drainage rate. Figure 8. Equilibrium time vs. stirring speed.

agreement between the two sets of values (R 0.8352) implies that Eq. 28 can express the experimental results very well within the surfactant (SLS) concentration range explored.
2

Liquid drainage from CGAs

There was a rise in the (CGA-dispersion)(bulk-liquid) interface with time as a result of the creaming, coalescence of CGAs, and the accompanying liquid drainage. Figure 7 shows the plots of ln[(VTo VL)/VTo] vs. td, for 6 mM SLS and for different stirring times at a speed of 5700 rpm. The plots for various other stirring speeds and surfactant concentrations were similar and are summarized in Table 2 in terms of the slopes and correlation coefcients. The high correlation-coefcient values show that the experimental data correlated well with the rst-order drainage kinetics represented by Eq. 6. The values of the drainage-rate constant varied from 0.0032 to 0.0006, decreasing with increasing stirring speeds and stirring times, as shown in Table 2. This indicates an increase in the stability of CGA dispersion, which can be attributed to the attainment of a narrower size distribution of CGAs in the dispersion. The convergence of m values to a constant value of 0.0007 with increasing stirring times indicates that an equilibrium size distribution will be attained for all stirring speeds above the critical stirring speed for CGA generation. However, the time required to attain this equilibrium size distribution, te, will be larger for lower stirring speeds. It decreases exponentially with

stirring speed as shown in Figure 8. The stirring time was scaled with te, and the drainage-rate constant m was plotted against this nondimensional stirring time. A fourth-degree polynomial t was used for these data as shown in Figure 9. The best-t values of n0 were obtained by tting the experimental liquid-drainage rates to Eq. 17 for various values of surfactant concentrations, stirring times, and stirring speeds. Figure 10 shows a comparison of the experimental values of drainage rates with the predicted values from Eq. 17. Figure 10 also shows that the model predicts a faster decrease in drainage rate with time in contrast to the experimental data. The comparisons for other stirring speeds, surfactant concentrations, and stirring times yielded similar trends. The experimentally measured surfactant-solution volume was next used in Eq. 22 to nd the best-t values of a and b for various values of surfactant concentrations, stirring times, and stirring speeds. These were further used to calculate n0(L) from Eq. 19. Figure 11 shows the variation of n0 with the aphron diameter for 6 mM SLS and different stirring times at a speed of 5700 rpm. Figure 11 also shows that the size distribution of aphrons becomes narrower with increasing stirring time, further consolidating the analogous notion of homogenization. The variation of n0(L) for various other stirring speeds and surfactant concentrations yielded similar results. The increase of stability with stirring time can thus be explained by a narrower size distribution of aphrons at longer stirring times. At short stirring times, the size distribution is broad, leading to

Figure 9. Plot of m vs. nondimensional stirring time for various stirring speeds. 3056 November 2005

Figure 11. Aphron size distribution for 6 mM, 5700 rpm, and various stirring times. Vol. 51, No. 11 AIChE Journal

Figure 12. Comparison between parameters a and b obtained by curve tting and model.

a hazy interface between dispersion of CGAs and the bulk surfactant solution. The values of a and b obtained by tting the experimental data to Eq. 22 were compared with the values calculated using Eqs. 20 and 21. Figure 12 shows that the values of a and b obtained using Eqs. 20 and 21 are in agreement with those obtained by tting the experimental data to Eq. 22 to within an order of magnitude. The size distribution of aphrons, represented by Eq. 19, can therefore broadly explain the rst-order drainage kinetics of CGAs represented by Eq. 23.

Mixing of oppositely charged CGAs

Visual Observation. A 50-mL solution of SLS-based CGAs, colored yellow with methyl red dye, was charged to a 100-mL measuring cylinder. A CTAB-based solution of CGAs (50 mL) was then slowly added to the cylinder. It was observed that the whole dispersion turned lighter yellow, as a result of dilution, indicating the mixing of the oppositely charged CGAs. Drainage-Rate Measurements. Figure 13 shows ln[(VTo VL)/VTo] vs. td plots for the drainage data obtained from experiments similar to those described earlier. The following observations may be made: (1) The drainage curves overlap, indicating practically no change in the drainage kinetics on mixing of the oppositely

charged CGAs, in contrast to oppositely charged foams that are known to quickly destroy each other.2 (2) Upon mixing of oppositely charged CGAs, no appreciable changes in stability and air holdup are observed, as seen in Table 3. (3) The foam formed above the mixed (oppositely charged) CGAs, however, collapsed much faster than that above the dispersions of individual CGAs, indicating the instability of foam resulting from the oppositely charged surfactants. This feature may be useful in quickly destroying the creamed foam formed in applications of CGAs, such as for separation of oils, pigments/dyes, and proteins, from aqueous emulsions, dispersions, or solutions.

Conclusions
Characterization of CGAs has been done, and an empirical correlation for the variation of air holdup with stirring time and stirring speed has been developed. First-order kinetics for drainage of CGAs has been proposed and the resultant expression was validated experimentally. A new population balance model has been developed, which for the rst-order drainage kinetics yielded an expression for the size distribution of aphrons. This size distribution has been used to explain the effect of stirring time on the stability of CGAs. Convergence of the drainage-rate constant values indicates that an equilibrium size distribution of aphrons for the dispersion is attained at a nite time te for all stirring speeds above the critical speed for formation of CGAs. An empirical correlation for te was found to be an exponentially decreasing function of the stirring speed. For the rst time, the behavior upon mixing of two oppositely charged CGAs has been studied and the results surprisingly show no effect on drainage and stability, except in the tail-end foam stage.

Table 3. Effect of Mixing of Oppositely Charged CGAs on Stability and Air Holdup
Air Holdup (%) Half-life (s) 194 229 223

Figure 13. Plot of ln[(VTo VL)/VTo] vs. drainage time for SLS, CTAB, and their mixture. AIChE Journal November 2005

SLS CTAB Mixture

59.25 65.9 64.23

Vol. 51, No. 11

3057

Notation
B number of aphrons forming in unit time per unit length, s m dVG volume of aphrons leaving the aphronfoam interface in time dtd, m3 L dV G volume of aphrons between diameter L and L dL leaving the aphronfoam interface in time dtd, m3 D number of aphrons collapsing in unit time per unit length, s1 m1 H height of the aphronfoam interface, m L diameter of the aphron, m Le diameter of cylinder of height L and having the same volume as a sphere of diameter L, m m rate-constant for drainage, s1 n number of aphrons per unit length, m1 n0 number of aphrons per unit length at td 0, m1 r rate of drainage of liquid from CGAs, m3 s1 s stirring speed, rpm t stirring time, min td drainage time, s tdh half-life of CGAs, s te time required for attaining equilibrium size distribution, min ts time required for attaining steady state for a given stirring speed, min t nondimensionalized stirring time [t/(1 min)], dimensionless vx velocity of an aphron in the vertical direction, ms1 Va volume of CGA dispersion at time td, m3 VG volume of gas in the CGA dispersion at time td, m3 VL volume of surfactant solution in bulk-liquid phase collected at the bottom at time td, m3 VTo total initial volume of CGA dispersion, m3 x vertical coordinate, m x initial volume of surfactant solution, m3 percentage air holdup of CGAs, dimensionless max maximum air holdup for a given stirring speed, dimensionless 0 percentage air holdup of CGAs at td 0, dimensionless difference in densities of the aphrons and the surfactant solution, kg/m3
1 1

Literature Cited
1. Sebba F. MicrofoamsAn unexploited colloid system. J Colloid Interface Sci. 1971;35:643-646. 2. Sebba F. Foams and Biliquid Foams: Aphrons. 1st Edition. Chichester, UK: Wiley; 1987. 3. Jauregi P, Gilmour S, Varley J. Characterisation of colloidal gas aphrons for subsequent use for protein recovery. Chem Eng J. 1997; 65:1-11. 4. Jauregi P, Varley J. Colloidal gas aphrons: A novel approach to protein recovery. Biotechnol Bioeng. 1998;59:471-481. 5. Jarudilokkul S, Rungphetcharat K, Boonamnuayvitaya V. Protein separation by colloidal gas aphrons using nonionic surfactant. Sep Purif Technol. 2004;35:23-29. 6. Jauregi P, Mitchell GR, Varley J. Colloidal gas aphrons (CGA): Dispersion and structural features. AIChE J. 2000;46:24-36. 7. Chaphalkar PG, Valsaraj KT, Roy D. A study of the size distribution and stability of colloidal gas aphrons using a particle size analyzer. Sep Sci Technol. 1993;28:1287-1302. 8. Amiri MC, Woodburn ET. A method for the characterization of colloidal gas aphron dispersions. Trans IChemE. 1990;68A:154-160. 9. Roy D, Kongara S, Valsaraj KT. Applications of surfactant solutions and colloidal gas aphron suspensions in ushing naphthalene from a contaminated soil matrix. J Hazard Mater. 1995;42:247-263. 10. Hashim MA, Gupta BS. The application of colloidal gas aphrons in the recovery of ne cellulose bres from paper mill wastewater. Bioresource Technol. 1998;64:199-204. 11. Save SV, Pangarkar VG, Kumar SV. Liquidliquid extraction using aphrons. Sep Technol. 1994;4:104-111. 12. Jauregi P, Varley J. Colloidal gas aphrons: Potential applications in biotechnology. Trends Biotechnol. 1999;17:389-395. 13. Noble M, Brown A, Jauregi P, Kaul A, Varley J. Protein recovery using gasliquid dispersions. J Chromatogr B Biomed Sci Appl. 1998; 711:31-43. 14. Matsushita K, Mollah AH, Stuckey DC, Cerro CD, Bailey AI. Predispersed solvent extraction of dilute products using colloidal gas aphrons and colloidal liquid aphrons: Aphron preparation, stability and size. Colloids Surf. 1992;69:65-72.
Manuscript received Apr. 6, 2004, and revision received Mar. 4, 2005.

3058

November 2005

Vol. 51, No. 11

AIChE Journal

ENVIRONMENTAL AND ENERGY ENGINEERING

Numerical Study on the Underground Coal Gasication for Inclined Seams

Lanhe Yang
College of Resources and Geosciences, China University of Mining and Technology, Xuzhou, Jiangsu Province, 221008, China DOI 10.1002/aic.10554 Published online July 22, 2005 in Wiley InterScience (www.interscience.wiley.com).

According to the characteristics for combustion and gasication reactions occurring in the gasication gallery, the mathematical functional relationship between the chemical reaction rate and every inuencing factor is studied. The dynamic nonlinear coupling mathematical models on underground coal gasication of inclined seams are established. The determination methods of major model parameters are introduced. Additionally, the control volume method is adopted to nd the numerical solution to the mathematical models. The patterns of development and variation for temperature eld, concentration eld and pressure eld in gasication panel are studied. On the basis of the model test, calculation results are analyzed. From the distribution of temperature eld, its calculation value is a little higher than the experimental one, with the relative error of every measuring point virtually within 17%. Research shows that, the experiment value of gas heat value and calculated value take on a good conformity; due to the inuence of temperature, in the high temperature zone, the change gradient of the experiment value for concentration eld of gas compositions is greater than that of the calculation value. The simulated results indicate that the relative error of the pressure eld calculation is 4.13%12.69% and 8.25%17.47%, respectively, 7 h and 45 h after the ignition. The drop rate for the uid pressure is 6.01% and 10.91%, respectively. Research shows that the simulated values conform with experimental values comparatively well, which demonstrates that the numerical simulation on the three elds in underground coal gasication is correct. 2005 American Institute of Chemical Engineers AIChE J, 51: 3059 3071, 2005 Keywords: underground coal gasication, mathematical models, temperature eld, concentration eld, pressure eld

Introduction
Field test and laboratory model test verify that the ways of the underground coal gasication is closely related to the occurrences of coal seams. It is easy for the channel gasication to be formed in the gently-inclined coal seams during the underground gasication, while in the steep coal seams, the percolation-patterned gasication is in a primary position.1 Research shows that in the underground gasication of inclined
Correspondence concerning this article should be addressed to L. Yang at Lhyang2053@sina.com.

2005 American Institute of Chemical Engineers

coal seams, the heat institution and uid mechanics institution of gasication process are ever-changing. Its thermal conditions and dynamics conditions fall between those of gentlyinclined coal seams gasication and those of steep coal seams gasication.1,2 With the enlargement of the channel diameter, the podzolization of the wall plane for the channel produces gradually, which, in turn, affects the temperature eld needed for keeping the normal gas production, leads to the beginning of the combustion gasication effect weakening for coal seams, and worsens the gasication conditions gradually. However, as the process of gasication occurs, the effect of high temperature and gravitation cause the coal seams on the roof of the channel to continually inbreak. This leads to the gradual
Vol. 51, No. 11 3059

AIChE Journal

November 2005

change from free gasication channel to percolation-patterned porous loose channel. Thus, the specic surface area for the reaction between gas and solid carbon increases, improving the gas production conditions.13 For the development and exploration of the underground coal gasication technique, experiment certainly is an important means, but, recently, people have tended to adopt the method of the combination of experiment and mathematical models. Because the underground coal gasication process is rather complicated, it is comparatively difcult to establish and solve its mathematical models. In spite of those, a number of studies have been made by many scholars home and abroad.4 13 At one time, a large amount of experimental and theoretical research on the three elds (temperature eld, concentration eld and pressure eld) was carried out in countries such as the Former Soviet Union, the United States, China, Germany and France, resulting in great progress.14 20 While their research results are conned to the general analysis method, experimental method or comparatively simple 1-D, 2-D steady or unsteady-numerical simulation. The dynamic coupling mathematical models on the underground gasication of the inclined coal seams are established in this paper. Whats more, the numerical analysis is made. The calculation results are checked against the model test.

diffuse to the surface of the carbon where reaction has not yet occurred. Theory proves that, under the underground combustion and gasication condition, the diffusion mass transfer of the gas passing through the surface grieshoch is the major factor controlling the carbon reaction rate. The gasication reaction is controlled by the diffusion process. The diffusion rate of the mole number for gas compositions equals the dynamic reaction rate of coal surface.21 On the basis of the above, we can obtain every chemical reaction rate R j . The inuence of temperature on the reaction rate is mainly reected on the constant of reaction rate, the value of K . According to literature,21 the Arrhenius formula is K A cexpE/RT (1)

According to the Eq. 1 and the research ndings22 of V. Fredersdorff and M. A. Elliott, we can obtain R 1 A cexpE1 /RT PO2 R 2 A cexpa/T bT PCO2 R 3 A cexpE3 /RT PH2O where K f expG/RT where (5) (2) (3)

1 P P Kf CO H2

(4)

The Chemical Reactions in the Underground Coal Gasication Process

During the underground coal gasication, the following seven kinds of chemical reactions mainly take place in the gasication channel 1 C O2 O CO2 2CO 2 C CO2 O CO H2 3 C H2 O O CH4 4 C 2H2 O
R 1 5 CO O2 O CO2 2
5

R1

H1 393.8 MJ/kmol H2 162.4 MJ/kmol H3 131.4 MJ/kmol H4 74.9 MJ/kmol

2 R 4 A cexpE4 /RT PH 2

1 P Kf CH4

(6) (7) (8)

R2

R 5 A cexpE5 /RT PCO R 6 A cexpE6 /RT PH2

R3

R4

R7 where

K1 K 1

1/ 2 1/ 2 PH K2 K3 PH2PCO2 K 2 K 3 PCOPH2O 2

K 2 PCO K3 PH2

(9)

H5 285.1 MJ/kmol

K 1 a 1T b1expE1 /RT
R 1 H2 O 6 H2 O2 O 2
6

(10) (11) (12) (13) (14)

H6 242.0 MJ/kmol

K 2 a 2T b2expE2 /RT K 2 a 2expE2 /RT K 3 a 3expE3 /RT

CO2 H2 7 CO H2 O O

R7

H7 41.0 MJ/kmol

The above reactions are the main basis for the conservation of mass equation of the chemical reactions. (1), (2), (3) and (4) reactions take place on the wall plane of the coal seams, while (5), (6) and (7) reactions occur at the gaseous stage. During the underground gasication, due to the high temperature, the coal seams crack. The combustion reactions must cause the O2 to
3060 November 2005 Vol. 51, No. 11

K 3 a 3expE3 /RT K 1 a 1 T

(15)
AIChE Journal

sion, we can obtain the conservation equation of the compositions yi yi yi yi 1 yi uy i D D D t x x x r r r r Si i 1, 2, . . . , 7 (17)

Figure 1. Model gasier.

where K j is the equilibrium constant of chemical reaction j , K f is the equilibrium constant expressed in terms of the partial fugacity of every composition; E j is the activation energy of chemical reaction j , R is universal gas constant; G is the free enthalpy of standard formation when the temperature stands at T , P i is the partial pressure of composition i in the mixed gas.

where y i is the mole fraction of the composition i (O2, CO2, CO, H2O, H2, CH4, N2); u is only taking the gas owing speed along the axial direction into consideration; D is regarding the diffusion rate of every composition along the axial and radial direction the same, therefore, D eff , i D ; S i is the generating rate of the composition i in the chemical reaction. From the earlier seven kinds of chemical reaction equations, we can obtain S 1 R5 R6 S2 2 R5 S 3 2 R5 S4 2 R6 S 5 2 R6 S6 S7 0 (18) (19) (20) (21) (22) (23)

Coupled Mathematical Models Assumed conditions

In the course of the combustion and gasication for the coal seams, various complex physical and chemical reactions occur in the gasier. Meanwhile, there are energy and mass transfer between gas phase and solid phase. In this paper, in order to simplify the calculation, the following assumptions are made: (1) The gasier itself is in a stable working state. Major physical and thermodynamic parameters, such as coefcient of heat conductivity, specic heat and coefcient of heat exchange do not vary with time,23 (2) Thermal diffusion and pressure diffusion will be ignored,24 (3) The effect of thermal resistance will be ignored; (4) The change in the mass of gas current in the oxidation zone due to the chemical reactions will be ignored;25 (5) For the inclined coal seams, since its thickness is much smaller than its slant height and strike length, temperature conduction and the movement of the mixed gas in the coal layer can be simplied into a 2-D problem (Figure 1). In the multicomponent gas, take a control body with an encompassed area of F , in which every component occupies the same area. Tracing the changes of the control body with the movement gives the changes of various parameters with the time. If a certain characteristic variable of the control body is , the total of the in the control body will be . Then, in the cylinder-coordinate system, we have

Initial conditions and boundary conditions: (1) Boundary Conditions Measuring and analyzing the gas compositions at the inlet of the gasication gallery, we can obtain the concentration of gasication agent at the inlet, that is At x 0, yi yi 0 , 0 r r1 , t0 (24)

where y i 0 is the mole fraction of gasication agent at the inlet of gallery. Because there is little change in the gradient of concentration for every gas composition at the outlet of gasication gallery, the gradient of concentration for the gas composition i at the outlet can be regarded as zero, that is At x L, yi 0, x 0 r r1 , t0 (25)

x , r , t dF

(16)

Conservation equation of the compositions

In the process of underground coal gasication, there are mainly seven kinds of compositions in the product gas, in the paper, only the balance of these seven kinds of compositions is taken into consideration, which are, in turn, O2 (1), CO2 (2), CO (3), H2O (4), H2 (5), CH4 (6) and N2 (7). According to the document,26 considering axial ow and axial and radial diffuAIChE Journal November 2005

where L is the length of gallery, the subscript 0 of the variable denotes the known value of parameter. At r r 1 , that is, at the boundary of the wall plane for the coal seams, the boundary conditions are D y1 r1 R1 0 r 2 (26)

D
Vol. 51, No. 11

y2 r1 R1 R2 R7 0 r 2

(27)
3061

y3 r1 2 R2 R3 R7 0 r 2 y4 r1 R 3 R 7 0 r 2

(28)

TS S T S T g t

i, j1

m H RHT
7 i j

(38)

(29)

y5 r1 R 4 R 3 R 7 0 r 2 y6 r1 R4 0 r 2 2 y7 0 D r

(30)

where m i is the mass owing quantity of the composition i , H j is the reaction heat of the chemical reaction j , is the coefcient of heat convection, T g is the temperature of gas phase, RHT is the quantity of heat exchange by radiation,26 that is
4 4 Tg RHT F 1 q T S

(39)

(31)

where is Stefan-Baltzmann constant, q is the coefcient of heat radiation, F 1 is the area of heat radiation,26 (32) F1 2

(2) Initial Conditions Before the gasication, the content for the composition i in the gasication agent is known, that is t 0, y i y i0, 0 r r 1, 0xL

r x , t dx

(40)

Initial conditions
(33) According to the measured results for the temperature of coal seams before ignition, the temperature for the solid phase can be known when t 0, T S T S 0, 0 z L, 0 r r1

Conservation equations of energy solid phase

Considering the interior heat conduction gives the temperature eld equation of coal seams26,27

SC S

1 TS TS TS TS S S S QS t x x r r r r (34)

where T S 0 is the known temperature of the solid phase.

Gas phase
According to the documents,28,29 we can obtain the conservation equation of energy for gas phase

where T S is the temperature of solid phase, C S is the specic heat of solid phase, Q S is the heat losses of solid phase, S is the coefcient of heat conductivity for the solid phase, S is the density of solid phase. Equation 34 must meet the following boundary conditions and initial conditions: According to the measurement results for the temperature eld in the gasier, the temperature of the outer coal seam with a radius of r 0 , and temperature at the inlet of gasication gallery can be known. Because there is little change in the gradient of temperature for coal seams at the outlet of gasication gallery, it can be regarded as zero, that is At r r0 , At x L, At x 0, TS TS , TS 0, x TS T0 , 0 x L, r1 r r0 , r1 r r0 , t0 t0 t0 (35)

jH i m i g T g SH Q g
gC gi Tg Tg u gC gi (41) t x

i, j1

(36)

where j is the coefcient of weighing and measuring for the chemical reaction j , H i is the enthalpy of formation for the gas composition i , SH is the heat of formation for the gas phase, C gi is the specic heat of the composition i for the gas phase, g is the coefcient of heat conductivity for the gas phase, g is the density of the mixed gas, Q g is the heat losses of the gas phase where j g D . 21 Simplifying Equation 41 gives

(37)

where T S is the temperature of the outer coal seam with a radius of r 0 , assumed to be a constant. At the gas-solid interface where r r 1 , because there is heat convection between gas and solid and heat and mass transfer, the following conditions should be met
3062 November 2005

i, j1

m T C T Tt C T T m T
7 g g i g gi g gi g g j i g

Vol. 51, No. 11

i1

H m Q R H R H
7 i i g 5 5 6

(42)

AIChE Journal

Boundary conditions
Measuring the gas temperature in the gasication gallery, the gas temperature at the inlet and temperature gradient for the gas at the outlet can be known, that is At x 0, At x L, Tg Tgb , Tg 0, x 0 r r1 , 0 r r1 , t0 t0 (43)

Gas state equation

In the process of underground coal gasication, the gas state equation can be shown as follows P g CRT g where C is overall mole concentration of the gas. (44) (49)

Unied form of the conservation and movement equations

The earlier conservation and movement equations can be written into a unied form, that is divu grad S t

Wall plane boundary

In the proximity of wall plane, there is similarly heat convection and heat radiation between heat current and wall plane, whose boundary conditions can be shown as

(50)

At r r1 ,

Tg TS Tg RHT, r

0 x L,

t0 (45)

At r 0,

Tg Tgp ,

0 x L,

t0

(46)

where is the dependent variable; and S indicate the coefcient of exchange and source corresponding to , respectively. The equation can be regarded as being made up of four items, which are transient item, convection item, diffusion item and source item, in turn.

Initial conditions
According to the measured results for the gas temperature in the gasication gallery before ignition, the initial temperature of the gas can be known, that is, when t 0, T g T g 0, 0 x L, 0 r r1 (47)

Numerical Calculation
In view of the nonlinearness for the earlier control equations and the strong coupling among equations, so it is difcult to solve them through analysis method, making us have no choice but to employ the numerical solution. In this article, the control volume method26,30 is adopted, which is a discretization method belonging to the nite difference method in the form, but being not fundamentally different from nite element method in methodology. The method aims at the integral equilibrium in the control volume, uses knot to represent control volume. The discretization of the domain to be solved usually includes two kinds, namely, even grid and uneven grid. On the basis of the characteristics of the problem to be addressed in the paper, every variable is the function of time for its spatial distribution. Hence, in the process of solution, the grid will be divided according to the even grid. The discretization of Eq. 50 depends on the following two basic assumptions: (1) The function value is evenly distributed in each grid, that is, the function value on the node of the grid represents its value everywhere within the grid, or we can say, the size of the grid determines the spatial resolution ratio of the function. (2) The function value is evenly distributed on each interface for each grid, that is, the function value at an arbitrary point of the boundary of the grid can represent the function value at that boundary. Usually, the crossover point between the ligature of the two neighboring points on both sides of the boundary for a certain grid and the boundary is selected as the representative point of that boundary. Integrate Eq. 50 on the grid centered around point P , we can obtain the commonly-used form of the difference equation F P t
Vol. 51, No. 11

where T g 0 is the initial temperature of the gas.

Equation of the gas phase movement

As stated earlier, in the process of the underground coal gasication, the ow of the mixed uid in the gallery can be regarded as the seepage movement of the gas in the porous media.1 Thus, in so doing, the equation of the gas movement can be obtained 1 Pg ac Pg Wg t a 0 b 0u

(48)

where P g is the uid pressure; a c is the coefcient of conduction pressure; W g is the source-sink item, depending on the generating rate of the gas phase; parameters a 0 and b 0 are determined by the following two formulas, respectively r g, a 0 / ge b 0 d / gn e rg

where is the coefcient of movement viscosity, g is the gravitational acceleration, e is the permeability, r g is the weight degree, is the coefcient of geometrical shape for the media particles, d is the average compromising particle size, n is the porosity.
AIChE Journal November 2005

u grad A S F
e,w n,s

(51)

3063

where F indicates the area of the grid, varying with time, under this circumstance, u takes the velocity of the uid relative to the moving boundary of the grid. The absolute value of A indicates the length of a certain interface for the grid, whose direction is that of the outer normal for the interface, indicates taking a sum of the four boundaries for the grid.

Discretization of the transient item

The expression of the discretization for the transient item is

tt

F Pdt F P F P0 t

(52)

where superscript 0 denotes the state of the point P at the previous time, t t ; t is time step size.

Figure 2. Grid points and grid interface.

Discretization of the source item

Assuming adopting the fully implicit form when making the integration of the source item in Eq. 51 relative to time, that is For the convenience of the writing, the discretized expression of the general ux item at the e plane can be written as A e Je Ce e De E P S F Pdt S F P t (53) where C e uA e D e eA e/ x e (58) (59) (57)

tt

If the source item is a constant one, its discretized expression will be the form of Eq. 53. If the source item is the function of the variable , generally the linearization method will be adopted to further discretize the source item, that is, write the source item S in the general form of the differential Eq. 50 personally as S S C S P (54)

where S C and S P are constants relative to , actually they can either be a real constant or the function of the previous iterative value * for the variable .

C e indicates the convection (or owing) intensity at the e plane, whose sign is determined by the direction of u , D e indicates the diffusion intensity at the e plane, always to be positive. Normally, the exchange coefcient e at the interface takes the arithmetic average of the exchange coefcient for two neighboring grid points. In the even grid system 1 E 2 P

Discretization of the total ux item

The item in the of the Eq. 51 combines the inuence of the convective ux, and the diffusion ux, known as the total ux item. Write the total ux as J J u grad (55) e (60)

Discretized equation
On the basis of the numerical model established earlier for the 2-D problem, the author developed the generallyused computer program used to solve for the mathematical models on the 2-D nonlinear dynamic reaction ow in the underground coal gasication, and realized the calculation of the models in this article. The general form and its corresponding coefcients for the discretized (nite difference) equations of the control differential, Eq. 50, in 2-D unsteady problem, are as follows a P P a E E a W W a N N a S S b C
Vol. 51, No. 11

In view of the similarity for the principles and treatment method in which the total ux item is discretized at each interface, now we will take the calculation of the total ux for an arbitrary grid interface (for example, e plane) as an example (Figure 2), so as to illustrate the discretization method of the total ux item. At the e plane of the grid P , the general ux item is J e J Ae /Ae ue e
3064

d dx
e

(56)

(61)

November 2005

AIChE Journal

where a E D eA P e Ce , 0 aW Dw APw Cw , 0 aN Dn APn Cn , 0 aS Ds APs Cs , 0 aP 0 P 0F /t bC SC F aP 0P 0 aP aE aW aN aS aP 0 SP F

(62)

In the coefcient expressions, a I indicate coefcients of difference equation corresponding to different grid points ( I E , W , N , S ); b C is constant term, P 0 is the uid density of the point P at the previous time, C k and D k are the convection intensity and diffusion intensity of the k interface, respectively, P k is the ratio of the two; their denition expressions are C k uA k kA k Dk k Ck Pk Dk

Figure 3. Relation between the specic heat and temperature of coal.

Specic heat of solid phase, CS

The experimental research shows that the specic heat for solid substances containing carbon is linked to its characteristic composition and temperature, whose relationship is shown in Figure 3. In the process of inversion calculation, C S will be determined according to the Figure 3.

(63)

Coefcient of heat convection,

Gas-solid phase reactions mainly occur in the pores of solid phase, the heat of the heterogeneous reaction is applied to the solid phase. When the oxidation reaction, the solid-phase transfers heat to gas phase, while reduction, dry and distillation, the gas phase transfers heat to solid phase. There are heat-transfer by convection and heat-transfer by radiation in the gas-solid heat exchange. When the temperature is high, heat-transfer by radiation will play a major role. Additionally, when the diameter of the particle is comparatively big, the heat conduction of the particle itself has to be taken into consideration. In the gas-solid heat transfer, the temperature of every point is shown in Figure 4. The quantity of heat transfer is calculated as follows31 Q F S T g T S where (66)

k e , w , n , s indicates the grid interface between two grid points, A k is the length of the k interface, k is the distance between two grid points on the vertical k interface. The distribution of the variable in the control volume takes the power function, so, for different circumstances, the form ) may be selected various forms, selecting power function A ( P law form in the article, that is 0, 1 0.1 P 5 AP (64)

is Peclet number, indicating the relative intensity of where P convection and diffusion; the operator [[ ]] denotes selecting the maximum in the square brackets. According to the difference equations established, we can obtain the simultaneous algebraic equation set of the control equations. On the basis of this, it is comparatively easier to adopt the separated iterative solution.

1 1 1 L S S K

(67)

Model Parameters Specic heat of gas phase, Cgi

The specic heat of the gas phase mixture normally depends on its thermodynamics state. The relational form for the specic heat used in this article is as follows i B iT n1 C iT n 2 C gi A (65)

2 1 1 R2 Rr RK K

(68)

3 2 2 3 S S T K T KT K1 T KT K1 T K1

(69)

i, B i, C i , n 1 and n 2 are coefcients of the expression, where A which vary with the kinds of substances and are irrelevant to temperature, whose specic values can be referred to in reference 26.
AIChE Journal November 2005

N u g DK

(70) (71)
3065

3 2 2 3 S 0 0 T K 1 T K1 T g T K1 T g T g

Vol. 51, No. 11

Figure 6. Cutaway view of the gasier model. Figure 4. Temperature distribution for gas-solid heat transfer.

Coefcient of heat conductivity, S

Under the condition of the combustion and gasication for the coal layers, the total coefcient of the heat conductivity for the media also consists of heat conduction and heat radiation, whose expression is as follows1

1 , 1 1 1 S g

S 0.7 0.8

(72)

In the earlier formula, L and S are the coefcient of heat convection, and that of heat-exchange by radiation through ash dreg layers, respectively; K and S are the coefcient of heat convection and that of heat-exchange by radiation through reduced lm, respectively; S and g is the radiation rate of solid phase and gas phase, respectively; N u is Nulet dimensionless number; and 0 are the thickness of the ash dreg layer and that of the reduced lm, respectively; F S is the specic surface area of the solid particle; D K is the diameter of the particle; R r is the radius of the reaction core, R K is the radius of the particle; T K is the temperature of reaction core; and T K 1 is the temperature of the solid particle.

S 1 T 2 T

(73)

where 1 ( T ) is the function of the coefcient of heat conductivity for coal mass; 2 ( T ) is the function of the coefcient of heat-exchange by radiation among coal chunks.

Other parameters
Other parameters in the mathematical model, such as the coefcient of diffusion, the enthalpy of formation and the coefcient of viscosity, are all functions of temperature under the nonisothermal condition, whose specic calculation method can be referred to in the relevant documents.3236 Because temperature has an important effect on model parameters, in the process of inversion calculation, the values of parameters shall be initially determined, based on the relationship between each parameter and temperature and temperature values at different nodes. Then through positive calculation, we will judge the reasonability of the model parameters selected. If the errors between calculated values and experimented values are within tolerance, the parameters selected are right, otherwise, further adjustments or correction should be made on every parameter till comparatively ideal calculated results have been obtained.

Permeability, e
Experimental research shows that, under the nonisothermal condition, the permeability of the porous media is a function of temperature.1,2,29 With the movement of the ame working face, in the process of combustion and gasication, since there are great temperature differences for the coal layers at various sections, great changes will take place in the permeability. Under the nonisothermal condition, the measured results of the permeability for the coal layers media used in the experimental model are shown in Figure 5. The input of this parameter is mainly based on the experimental data shown in Figure 5.

Model Test
The overall size of the model gasier body is 7.35 m long by 1.45 m high by 1.0 m thickness, consisting of base and lid, as shown in Figure 6. Under the bottom of the gasier was installed a line of hydraulic jacks, which makes the gasier rest at any angle. The hearth is so spacious that it can be lled with other materials, which are used to simulate coal seams with different thickness. On the sides of the gasier there are a number of gas inlets, outlets, and slip casting holes, used to simulate various gasiers and study different modes of air pumping and gasication channel with different lengths. A circular air (steam) injection pipeline is equipped on the platform in hope for supplying the air (steam) in the same or counter direction and with moving pumping points. The gasier adopts hydraulic sealing technology. On the upside of the
Vol. 51, No. 11 AIChE Journal

Figure 5. Changing curve of permeability and temperature. 3066 November 2005

Figure 7. System of model experiment.

gasier is installed a piece of board pushed by hydraulic push rod, which imposes a certain pressure on the coal seams through the board in the course of gasication, in order to simulate the pressure on the top of the coal seams. The structure of the gasier is made up of three layers. The outer one is sealant made of thick steel plate, with the middle one thermal insulating made of vermiculite, the inner one ame retardant coating. The size of simulated coal seams in the gasier is 6.7 m 1.3 m 0.45 m, with a dip angle of about 45. The type of coal is gas-fat coal. The data of the proximate analysis can be consulted in the document.1 In the process of coal injection, use as much lump coal as possible, so we can keep the natural state of the coal seams at best. The interstices will be lled with small pieces of coal. Finally, smear the interstices with the mixture of cement and coal powder. The gasication panel is 6.0 m 1.1 m 0.42 m. The initial equivalent diameter of the gasication gallery is 80 mm. The system of the model test is shown in Figure 7. The pipeline system is designed as an armillary circuit. Through the reversal valve, we can pump the air or steam negatively and positively. Coal gas cleaner is mainly to remove the tar from the gas and lower the temperature of the resulting gas. In order to measure the temperature distribution in the gasier accurately, we deploy the temperature-measuring points intensively. In the gasication panel, 17 rows of temperature-measuring points are buried, with 7 in each row. The number totals

Figure 9. Experiment value and calculation value of the concentration of O2 at various times of gasication.

119. The temperature-measuring elements adopt the strictlystandardized NiCr-NiSi thermal couples. The automatic data acquisition will be made regularly, and the data will be displayed on the screen of the computer and recorded. The resulting gas will be analyzed with gas-phase chromatographer, which can tell the contents of different compositions. Every bore hole deployed at the sides of gasier along the direction of gasication gallery also has the monitoring function. In the experiment, the real-time monitor has been made against the pressure and gas compositions in the gasier, so as to learn the state of changes for the pressure eld and concentration eld at different places and various times within the gasication gallery. The operating conditions: the gasication agent, air; the pressure at the inlet, 2550 2600 Pa; the owing capacity at the inlet, 2530 m3/h. The results of the experiment are shown in Figures 8 15.

Analysis
The calculation results of numerical simulation are shown in Figures 8 15. From Figure 8, it can be shown that the calcu-

Figure 8. Experiment value and calculation value of the temperature eld at various times of gasication. AIChE Journal November 2005

Figure 10. Experiment value and calculation value of the concentration of CO2 at various times of gasication. Vol. 51, No. 11 3067

Figure 11. Experiment value and calculation value of the concentration of CO at various times of gasication.

Figure 13. Experiment value and calculation value of the concentration of CH4 at various times of gasication.

lated values of temperature eld virtually conform with the measured ones. Except the measuring points in the combustion zone, where the relative error between the calculated values and measured ones of temperature is rather high (certain points, over 20%), the relative error of other points is no more than 17%, majority of which is within 14%, completely meeting the precision requirement of numerical simulation on the temperature eld. Judging from the distribution of temperature eld, the calculated values are a bit bigger than the measured ones. The reasons are the following: rst of all, the term of heat losses in the mathematical model is determined on such conditions as the calculation formula for the coefcient of heat transfer with the composite structure and xed setting (temperature, wind velocity). In the process of experiment, natural ventilation or forced ventilation makes the coefcient of heat losses increase, which contributes to the slight drop in the measured temperature. Then, during calculation, the coefcient of heat conductivity for coal seams is not that of body coal but the equivalent coefcient of heat conductivity for coal seams taking the inuence of convection and heat radiation into consideration. The coal seams in the model test are piled up selecting various sizes of coal chunks. Though the interstices are lled with pulverized coal, its pores are far bigger than

those of the natural ones. Therefore, heat transfer is no longer a single form of heat conduction. With the effect of convection intensifying, the coefcient of apparent heat conduction for the coal seams heightens, which inuences the calculation results precision. Because the reaction rate heightens rapidly with the rise in temperature, which leads to the increase in the reaction conversion rate, as a result, the inuence of the temperature causes the calculated values of the concentration for various compositions in gas to be a little bigger than the experimental ones; the change gradient of the composition concentration for experimental value in high-temperature zone is bigger than that of the calculated value (Figures 9 13). According to Figure 14, with the prolonging of gasication time and the increase in the gasication channel length, the heating value of gas increases gradually, and there is virtually no uctuation for it. Therefore, the experimental values and the calculated ones take on a good conformity. However, beyond the reduction zone, the extent of increase for the heating value decreases. The inuence of temperature eld on the heating value of gas is remarkable. 7 h after the ignition, in spite of the

Figure 12. Experiment value and calculation value of the concentration of H2 at various times of gasication. 3068 November 2005

Figure 14. Experiment value and calculation value of the gas heat value at various times of gasication. Vol. 51, No. 11 AIChE Journal

In short, the simulated results indicate the calculated values can conform with the measured values, which shows that the establishment of mathematical models on heat and mass transfer in the process of underground coal gasication, the determination of parameters, the analysis and treatment of boundary conditions and the solution method are correct. This will provide necessary theoretical evidence and scientic guidance for the further comprehensive quantitative study and production practice of underground coal gasication.

Conclusion
(1) According to the features of gasication process, on the basis of the model test, the nonlinear coupled mathematical models on the underground coal gasication for inclined coal layers are established. The simulated results demonstrate the calculated values can conform with the measured ones comparatively well, which proves that the numerical simulation on the temperature eld, concentration eld and pressure eld is reasonable in the underground coal gasication under the experimental condition. (2) The numerical simulation results show that, in high temperature zone, the calculated value of the temperature eld is a bit bigger than the measured one; the change gradient of the measured value of the concentration for various compositions in the gas is bigger than that of the calculated value; temperature eld has a great inuence on the heating value of the gas. The heating value of the gas increases gradually with the prolonging of gasication time. The calculated value of heating value is basically the same with the experimental one. Apparently, the ideal temperature eld is conducive to the improvement on the gas quality and the stability of the gasication process. (3) According to the calculation results, the relative error between the calculation value and measurement one of the uid pressure and its drop rate increases gradually with the process of gasication. The change of the ow condition in gasication channel is mainly responsible for the change of the calculation error. (4) The numerical calculation results basically reect the real patterns of variation for the temperature eld, concentration eld and pressure eld in the underground coal gasication for the inclined coal seams, which will provide necessary scientic theoretical evidence for the further quantitative study on the underground coal gasication process and the predictions of its variation patterns.

Figure 15. Pressure eld in the gasication tunnel 7 h and 45 h after the ignition.

comparatively long gasication channel, due to the short period of time and the low temperature in the oven, the heating value of gas is comparatively low; 45 h after the ignition, the temperature of the oven increases. Although the length of the gasication channel is relatively short, the heating value of the outlet gas is higher than that of previous one. Therefore, maintaining an ideal temperature eld and comparatively longer gasication channel is conducive to the stability and improvement of the heating value for the gas. Figure 15 demonstrates that the simulated calculation values of the gas pressure basically conform with the measured ones. The relative error at the second time between calculated values and measured ones is 4.13%12.69%, with the average drop rate of uid pressure 6.01%. From Figure 15, we can know that, 45 h after the commencement of gasication, the error of the simulated calculation for pressure is 8.25%17.47%, with the average drop rate of the gas pressure along the gasication channel 10.91%. It can be concluded that, with the prolonging of gasication time, the extent of drop for the uid pressure rises and the calculation error increases. The major reasons are the following: at the very beginning of the experiment, the gas moves along the free gasication channel with little resistance on the uid. The uctuation for the curve of the experiment is small, so the drop rate and the calculation error are low, but the calculation value is comparatively high. With the development of the gasication process, the coal layers over the gasication channel, due to the effect of high temperature, expands, splinters and falls onto the gasication channel because of deadweight, which lls the bottom of the gasication channel with loose coal chunks. Thus, the free gasication channel transforms into a percolation-patterned one. The seepage movement of the gas will go on in the porous media. The resistance on the gas movement increases by a large margin, so does the extent of the pressure drop, which also tends to be stable. The curve between the experiment and calculation takes on a good conformity. Considering the changes in movement conditions, in the preliminary determination of major model parameters, such as the coefcient of conduction pressure, permeability, the involvement of personal or empirical factors in the consideration of relevant calculation coefcients results in a certain error in the initial parameters calculation, which causes oscillatory occurrence in the value of initial numerical simulation.
AIChE Journal November 2005

Acknowledgments
This work was supported by the National Natural Science Foundation of China (Ratication No.: 59906014), and by the National Natural Science Foundation of China (Ratication No.: 50276066, 20207014). The technical contributions of Professor Yu Li and Dr. Wang Zaiquan are gratefully acknowledged by the author. The author also gratefully thanks Professor Xiang Youqian, Wang Jialian, and Gao Wenjun, for helpful discussion. Support received for this research from the Engineering Research Centre (ERC) for Clean Coal Technology of the China University of Mining and Technology (CUMT) is gratefully acknowledged.

Notation
a coefcient in Eq. 3 a m coefcients in Eqs. 10 15 ( m 1, 2, 2, 3, 3, 1) a I coefcients of difference equation (61) ( I E , W , N , S )

Vol. 51, No. 11

3069

a0 ac Ac A Ae Ae Ak Ai

) A(P A ( P k ) b bh bC b0 i B C Ce Ck i C C gi CS d D De D eff , i Dk DK e e F F1 FS F g Hi J Je K Kj Kf L mi n n1 n2 n Nu P Pi Pg Pk Q Qg QS r r0 r1 rg R RHT Rj RK Rr s SC Si

SH

parameter in Eq. 48, determined by a 0 / ge rg the coefcient of conduction pressure, m2/s coefcient in Eqs. 1 4 and Eqs. 6 8 length vector of a certain interface for the grid, m length of the e plane, m length vector of the e plane, m length of the k interface ( k e , w , n , s ), m coefcient of the expression (65), which varies with the kinds of substances form function form function of the k interface ( k e , w , n , s ) coefcient in Eq. 3 coefcients in Eqs. 10 and 11 ( h 1, 2) constant term of difference Eq. 61 parameter in Eq. 48, determined by b 0 d / gner g coefcient of the expression (Eq. 65), which varies with the kinds of substances overall mole concentration of gas, mol/m3 convection (or owing) intensity at the e plane in Figure 2 convection intensity coefcient of the expression (Eq. 65), which varies with the kinds of substances specic heat of the composition i for the gas phase, J/kg K the specic heat of solid phase, J/kg K average compromising particle size, m diffusion rate of every composition along the axial and radial direction, m2/s diffusion intensity at the e plane in Figure 2 effective diffusion coefcient of the composition i , m2/s diffusion intensity diameter of the particle, m interface between two grid points permeability, m2 encompassed area of a control body, m2 area of heat radiation, m2 specic surface area of the solid particle, m2/m3 area of the grid, m2 gravitational acceleration, m/s2 enthalpy of formation for the gas composition i , J/mol vector of total ux item vector of general ux item at the e plane in Figure 2 rate constant equilibrium constant of chemical reaction j the equilibrium constant expressed in terms of the partial fugacity of every composition length of gallery, m mass owing quantity of the composition i , kg/m3 interface between two grid points coefcient of the expression (Eq. 65) coefcient of the expression (Eq. 65) porosity Nulet dimensionless number Peclet number partial pressure of composition i in the mixed gas, Pa uid pressure, Pa ratio of the C k and D k quantity of heat transfer, W heat losses of the gas phase, J/m3 s heat losses of solid phase, J/kg s r -coordinate distance between boundary of the outer coal seam and x -coordinate axis, m radius of gallery, m weight degree, N/m3 universal gas constant, J/kmol K quantity of heat exchange by radiation, W chemical reaction rate, mol/m3 s radius of the particle, m radius of the reaction core, m interface between two grid points constant relative to in Eq. 54 generating rate of the composition i in the chemical reaction, mol/m3 s heat of formation for the gas phase, J

SP S t T T0 Tg T gb T gp T g0 TK T K1 TS T S0 T S u w Wg x yi y i0

constant relative to in Eq. 54 source item time, s temperature, K temperature of solid phase at the inlet of gallery, K temperature of gas phase, K temperature of the gas at the inlet of the gallery, K temperature of the gas at wall plane boundary, K initial temperature of the gas at the inlet of the gallery, K temperature of reaction core, K temperature of the solid particle, K temperature of solid phase, K known temperature of the solid phase, K temperature of the outer coal seam, K velocity of the uid, m/s interface between two grid points source-sink item x -coordinate mole fraction of the composition i mole fraction of gasication agent at the inlet of gallery

Greek letters
L K S
coefcient of heat convection, W/m2 K coefcient of heat convection through ash dreg layers, W/m2 K coefcient of heat convection through reduced lm, W/m2 K coefcient of heat-exchange by radiation through reduced lm, W/m2 K coefcient of heat-exchange by radiation through ash dreg layers, W/m2 K coefcient of weighing and measuring for the chemical reaction j radiation rate of solid phase, W/m3 K4 radiation rate of gas phase, W/m3 K4 thickness of the ash dreg layer, m thickness of the reduced lm, m distance between two grid points on the vertical k interface, m distance between two grid points, m function of the coefcient of heat conductivity, W/m K function of the coefcient of heat-exchange by radiation, W/m K coefcient of movement viscosity, m2/s coefcient of geometrical shape for the media particles dependent variable exchange coefcient at the e plane in Figure 2, determined by Eq. 60 exchange coefcient at the grid point E in Figure 2 exchange coefcient at the k interface ( k e , w , n , s ) exchange coefcient at the grid point P in Figure 2 coefcient of exchange corresponding to the variable characteristic variable of the control body variables of difference Eq. 61 ( I E , W , N , S ) previous iterative value for the variable activation energy of chemical reaction j , J/mol free enthalpy of standard formation, J/mol reaction heat of the chemical reaction j , J/mol time step size, s Stefan-Baltzmann constant, W/m2 K4 coefcient of heat radiation coefcient of heat conductivity for the gas phase, W/m K coefcient of heat conductivity for the solid phase, W/m K uid density, kg/m3 density of the mixed gas, kg/m3 uid density of the point P at the previous time, kg/m3 density of solid phase, kg/m3 selecting the maximum in the square brackets Hamilton operator

S j
S g 0 k x 1( T ) 2( T ) e E k P I * Ej G Hj t q g S g P0 S [[ ]]

Subscripts
b c C eff f boundary coefcient constant effective partial fugacity

3070

November 2005

Vol. 51, No. 11

AIChE Journal

g i j k p P P0 q r S 0

gas composition i chemical reaction j k interface wall plane grid point P state of the point P at the previous time heat radiation radius solid variable outer coal seam known value

15. 16. 17. 18. 19. 20.

Literature Cited
1. Yang LH. Study on the seepage combustion method in underground coal gasication. 2nd ed. Xuzhou: China University of Mining & Technology Press; 2004. 2. Yang LH, Song QY, Li YJ. Underground coal gasication project (1st ed). Xuzhou: China University of Mining & Technology Press; 2001. 3. Liang J. Study on the stability and controlling technology of underground gasication in steep coal seam (Dissertation), China University of Mining and Technology; 1997. 4. Liu SQ, Liang J, Yu L. Reaction character and inuence factors of CO2 in underground coal gasication. J of China University of Mining & Technology. 2000;29:606 609. 5. Debelle B, Malmendier M. Modeling of ow at Thulin underground coal gasication experiments. Fuel. 1992;71:95104. 6. Martin JM, Layne AW, Siriwardane HJ. Thermo-mechanical modeling of ground movements associated with large underground coal gasication cavities in thin coal seams. Proc. 10th UCG Symp., Fallen Leaf Lake, California; (April)1984;295307. 7. pe E. Two-stage underground coal gasication. Coal Chemical Industry. 1993;6:61 63. 8. y e o BM. Increasing efciency of underground coal gasication. Collections of Translated Works on Mining Technology. 1988; 9:1719. 9. Gunterman K. The comprehensive experimental model and mathematical model on the underground coal gasication. Collections of Translated Works on Mining Technology. 1988;9:6 10. 10. William BK, Robert DG. Underground coal gasication: the state of the art. 1st ed. New York: Am In Ch Eng; 1983. 11. Dufaux A. Modeling of the UCG process at Thulin on the basis of thermodynamic equilibriums and isotopic measurements. Fuel. 1990; 69:624 632. 12. Advani SH, Lee JK, Chen KS. Geomechanical modeling associated with underground coal gasication process. Proceedings of the Twelfth Annual Underground Coal Gasication Symposium, Washington, DC, (August)1986;272278. 13. Guntrmann K, Gudenau HW, Franke FH. An integrated UCG-simulation model of laboratory work and mathematical modeling. Proceedings of the Twelfth Annual Underground Coal Gasication Symposium, Washington, DC, (August)1986;207216. 14. Yeary D, Riggs JB. Experimental study of lateral cavity growth mechanisms. Proceedings of the Twelfth Annual Underground Coal 21.

22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36.

Gasication Symposium. Washington, DC, (August)1986; pp158 164. Yu F. The structural dynamic simulation on underground coal gasication. Collections of Translated Works on Mining Technology. 1992; 13:13. Liu SQ, Liu JH, Yu L. Environmental benets of underground coal gasication. J of Environ Sci. 2001;9:5154. Liu BY, Qiu P. Study on the underground coal gasication technology. Clean Coal Technol. 2003;9:2329. Guo CW. A review of the study on mathematical models on underground coal gasication. Mining World. 1994;15:25. Liu SQ, Liang J, Yu L. Study on the abatement technology of CO2 discharged from UCG. Proceedings of the Second International Symposium on Clean Coal Technology. Beijing, 1999;541545. Liu SQ, Liang J, Chang J, Yang Z, Yu L. UCG model test of Huating coal with oxygen-steam as gasication agent. J of Southeast University (Natural Science Edition). 2003;33:355358. Liu SQ. Study on the regularity of CO2 production in the process of underground coal gasication and on the disposal method of CO2 (Dissertation). China University of Mining & Technology; 2000;73 76. Guo CW. Study on long-tunnel large-section two-stage underground coal gasication (Dissertation). China University of Mining & Technology; 1993; pp109 114. Xiang YQ. Mathematical model, modeling calculation and optimal operation for coal gasication. Coal Conversion. 2002;25:60 64. Wu RY. Coal gasication. 1st ed. Xuzhou: China University of Mining & Technology Press; 1988. Yang LH. Numerical simulation on non-steady state temperature eld in the underground coal gasication. J of China University of Mining & Technology. 2000;29;140 143. Fan WC, Wan YP. The model and calculation of owing and combustion (2nd edition). Hefei: China University of Science & Technology Press; 1992. Zhou LX. Combustion theory and chemical uid mechanics. 1st ed. Beijing: Science Press; 1986. Qin KF, Fan JK. Combustion uid mechanics. 1st ed. Beijing: Hydroelectricity Power Press; 1991. Gao J R. Momentum, heat, and mass transmission principle. 1st ed. Chongqing: Chongqing University Press; 1987. Chen CK. Computation uid mechanics. 2nd ed. Chongqin: Chongqing Press; 1992. Xiang YQ, Hedden K. The determination of dynamic data for solid fuel in the process of conversion. Gas & Heat. 1986;6:4 11. Ruprecht P. A computer model of entrained coal gasication. Fuel. 1988;67:739 742. Fu WB, Zhang YL, Wang DA. Combustion science. 1st ed. Beijing: Higher Education Press; 1989. Zhao JX., Yi Y. Numerical and experimental study on the turbulence combustion model. J of Engg Thermophysics. 1994;15:99 104. Tao WQ. The progress of calculation heat transfer in modern times 1st ed. Beijing: Science Press; 2000. Bu XM, Peng WW, Xiang YQ. The study of the mathematical models on the pressurization gasication in xed beds. Coal Chemical Industry. 1993;17:6 15.

Manuscript received Jun. 8, 2004, and revision received Feb. 14, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

3071

Measurement of Flow Field in Biolm Reactors by 3-D Magnetic Resonance Imaging

Kevin P. Nott, Frank P. Heese, and Laurance D. Hall
Herchel Smith Laboratory for Medicinal Chemistry, University of Cambridge School of Clinical Medicine, University Forvie Site, Robinson Way, Cambridge, CB2 2PZ, U.K.

Lynne E. Macaskie and Marion Paterson-Beedle

School of Biosciences, University of Birmingham, Edgbaston, Birmingham, B15 2TT, U.K. DOI 10.1002/aic.10537 Published online August 29, 2005 in Wiley InterScience (www.interscience.wiley.com).

3-D Magnetic resonance imaging (MRI) was used to measure the ow eld of water in a packed-bed column containing Serratia sp. biolm supported on polyurethane foam, and subsequently to follow a reaction which precipitates lanthanum phosphate on the biolm. Sensitizing the MR image contrast to the uid ow along the axis of the bioreactor provided better image-contrast between the foam and uid compared to that based on MR signal intensity alone. After reaction, that same velocity contrast effectively dened the difference between blocked and unblocked regions by distinguishing between regions of ow and no ow. Data acquired during progressive blockage of reactors challenged at two different ow rates accord with reactor theory; thus, the faster ow rate replenished the reactants uniformly, whereas at the slower ow rate the reactants were concentration limited. MRI velocimetry was used to generate data that can be used to model reactors where the efciency is progressively compromised by blockage due to precipitation. 2005 American Institute of Chemical Engineers
AIChE J, 51: 30723079, 2005

Keywords: MRI, serratia, bioreactor, ow, blockage

Introduction
With increasing environmental legislation concerning the release of industrial wastewaters, there is a growing need for efcient and economical remediation methods to treat such uids. It is encouraging, therefore, that there is an increasing body of evidence to suggest that microorganisms can be used for the removal of heavy metals from industrial efuents which can be detrimental to health.1,2 Thus, an effective method has been developed that can typically remove 8595% of heavy metal ions by passage through a bioreactor containing immobilised Citrobacter biolm;3 a cell-bound phosphatase liberates inorganic phosphate from an organic donor molecule, which then promotes the formation of cell-bound metal phosphate.4 A simple model has been derived from the Michaelis-Menten
Correspondence concerning this article should be addressed to K. P. Nott now at kevin.nott@oxinst.co.uk.

equation to relate the efciency of removal of metal ions5,6 from a uranium mine wastewater semiquantitatively to the input ow rate, total enzyme loading and bioreactor activity.6 However, that model assumes ideal plug-ow behavior and is limited to bulk parameters that describe the overall reactor performance, and cannot account for the spatial and temporal changes in ow eld induced by blockage formation. Although assumption of ideal plug-ow behavior is generally fundamental to xed-bed bioreactor modeling, that assumption is difcult to prove experimentally within the 3-D dynamic system of an operational ow-through reactor. Use of controlled blockage deposition does provide a tool whereby these factors can be investigated quantitatively, but that black box approach based on analysis of input- and exit-solutions, provides no spatio-temporal information, and, hence, is of little use in process optimization. It is known that MRI is a powerful tool that can measure, noninvasively, the velocity of opaque uids in three dimenVol. 51, No. 11 AIChE Journal

3072

November 2005

sions, and, hence, it is invaluable for many process engineering applications.79 Increasingly, MRI is also being used for environmental engineering and biotechnology applications.10,11 Of specic relevance to this study, MRI has previously been used to visualise the formation of biolm as well as the ow velocity of uid in a bioreactor cell;1216 it has also been used to follow the absorption of heavy metals in alginate, immobilised cells and algal biosorbents.1720 As part of a broad ranging study using noninvasive imaging techniques to investigate the function of a biolm bioreactor, it has already been demonstrated that MRI of the aqueous medium can visualize biolm on polyurethane reticulated foam, as well as the removal of Cu2/La3 ions by biosorption and inorganic precipitation.21 This is possible because the MR properties of water protons are dependent on the concentration and molecular environment of the water molecules. Thus, heavy metal ions (La3) induce a concentration-dependent relaxivity, or in the case of paramagnetic ions (Cu2), a susceptibility effect which alters the MR properties of vicinal water molecules, and thereby provides image contrast with the bulk of the aqueous medium. Extensive phosphatase-mediated bioaccumulation of LaPO4 was observed, but no evidence for deposition of Cu3(PO4)2 was obtained by X-ray diffraction and proton-induced X-ray emission analyses;21 those observations were conrmed by MRI. Since Cu2 was not taken up into the biolm it was concluded that Cu2 could be added as an MRI contrast agent to increase signal-to-noise from the perfusate during experiments to monitor the biodeposition of LaPO4, which shows as regions of low signal intensity because of the lack of water. On that basis, it was decided that this is a good model system for a noninvasive study of the function of a biolm reactor by MRI velocimetry; in turn, that prompted the present study, which seeks to answer the following questions: (1) Can MRI quantitate in three dimensions the entire 3D ow velocity eld of uid owing through the reactor? (2) If so, are such data sensitive to the progressive deposition of metal as it is removed from the wastewater stream? (3) What are the effects of the air bubbles entrained in the support matrix on the uid ow, and is it possible to distinguish them from the metal precipitate? (4) Can MRI be used to generate data to rene the existing enzyme kinetic-based model, and hence help the design of a more efcient reactor?

Figure 1. Gradient echo imaging sequence with interleaved Pulsed Field Gradient (PFG) pair.
Each radio frequency (RF) pulse tips the magnetisation of the water protons through a ip angle , and its magnitude and phase are recorded at the end of the echo time (TE). TE is set to reect the spin-spin relaxation time ( T 2 ) of the protons; the shorter the T 2 , the shorter must be TE. Successive RF pulses are separated by the repetition time (TR), which is long enough for the magnetisation to relax fully before the next RF pulse is applied. 3-D spatial encoding is achieved by using a xed z gradient whilst incrementing the x and y gradients; thus, the total scan time is N x N y TR, where N x and N y are the number of pixels in the x and y directions respectively. Velocity encoding is achieved by application of a bipolar gradient pair, in this example in the z direction parallel to the bulk ow.

where is the magnetogyric ratio of the 1H nucleus and is the phase offset caused by any magnetic eld inhomogeneity. Therefore, a series of images can be acquired with different gradient strengths to obtain a value of ow velocity for each pixel. Velocity images can be acquired sequentially for all three orthogonal ( x , y , z ) directions, where required, to provide the complete 3-D ow eld.

Experimental Preparation of Reactor Packed-Bed Systems for Metal Removal by Serratia Biolm
Citrobacter sp. NCIMB 40259, now reassigned to the genus Serratia by modern molecular and biochemical methods,26 was used under licence from Isis Innovation (Oxford, U.K.). The conditions for controlled growth of Serratia sp. biolm have been established.4 The biolm was grown on polyurethane reticulated foam (Filtren TM 30 cubes, 1 cm3) supplied by Recticel (Belgium) in an air-lift fermenter in continuous culture, as described previously.4 The reticulated foam cubes, coated with biolm of average thickness of 26 m27 were withdrawn after six days of continuous culture. The phosphatase specic activity of the cells from the fermenter was determined by the release of p-nitrophenol (PNP) from p-nitrophenyl phosphate (PNPP); one unit is dened as nmol PNP liberated per min per mg bacterial protein.4 In all experiments resting cells were used since no further growth medium was provided; effectively the biolm behaves as an immobilising matrix for the phosphatase enzyme. All measurements were made using cylindrical glass columns (16 mm dia., 70 mm length) packed, without compression, with 88, 0.125 cm3 cubes of polyurethane foam with
Vol. 51, No. 11 3073

Flow Imaging
The theory of MRI and its application to the study of a biolm reactor has been described previously.22 MRI of water is based on the spatial localization of the nuclear magnetic resonance (NMR) signal in a homogeneous magnetic eld via the application of linear magnetic eld gradients. If, in addition to those spatial encoding gradients, a pair of gradient pulses with strength g , both with duration and separated by are applied in a particular direction, the phase of the MRI signal is sensitised to ow in that direction;2325 a typical pulse sequence as used in this study is shown in Figure 1. The phase , of the signal at each pixel of the resulting pulsed eld gradient (PFG) velocity-encoded MR image is related to velocity v, by the equation

g v
AIChE Journal November 2005

immobilised Serratia sp. biolm21 (made by dividing 1 cm3 cubes obtained from the fermenter into eight).

MRI Velocity Measurements

All 1H MRI images were acquired using a Bruker MSL-100 imaging console (Bruker Analytische Messtechnik GmbH, Karlsruhe, Germany) operating under TOMIKON software (Bruker Medizintechnik GmBH., Karlsruhe, Germany) connected to an Oxford Instruments 2 Tesla, 31 cm horizontal bore super-conducting magnet (Oxford, U.K.). The 11 cm internal dia. gradient set and radio-frequency (RF) probe (internal dia. 5.5 cm) were built in house. Each axis of the gradient set was powered by two Techron ampliers (Models 7560 and 7570, Elkhart, IN, U.S.A.) giving gradient strengths up to 195 mT/m. The data were transferred to a Linux workstation and processed into images using CaMReS software written by Dr. N. J. Herrod. 3-D dimensional ow encoded images (256 64 64 pixels) were acquired using a PFG gradient echo sequence (TR/TE/ 100 ms/7.5 ms/90) with 312.5 m isometric spatial resolution in a scan time of 6 min 50 sec; a schematic is given in Figure 1. A total of ten images was acquired, with ve different velocity encoding gradient strengths (117, 19.5, 0, 19.5, 117 mT/m), with and without ow, giving a total scan time of 1.1 h for each velocity direction either transverse ( x and y ) or parallel ( z ) to the bore of the magnet. The uid ow was provided by an electromagnetically driven gear pump (Integral Series 120 pump, Micropump Inc., Vancouver, WA) which was used to replenish water in a constant head gravitational ow vessel held at a given height in order to provide the reactor with steady-state ow. The velocity encoded images acquired without ow were used to correct for errors caused by imperfections in the motion encoding gradients. Initially x , y , and z velocity measurements were carried out on a bioreactor containing foam cubes with biolm, using a 20 mM CuSO4 solution to enhance the signal-to-noise of the experiment and consequently the accuracy of the velocity measurement. Concentrations normally toxic to the biolm could be used since no biomass growth was required and the probe solution was recycled without any consequence to the remaining experiment. The velocity-encoded images were processed using a Bayesian statistical method for analysis in order to maximize accuracy over a wider range of velocities using fewer gradient steps.28

Figure 2. Continuous experiment carried out in situ in the magnet.

Once the scanner had been tuned with the reactors inside the RF probe, the experimental conditions could be controlled sequentially. Initially, the challenge solution was pumped through two reactors at different ow rates (11 and 128 ml/h, termed slow and fast) for a 2 h period; that solution was then displaced by 1 mM aqueous CuSO4 for measurement of the ow velocity over the next 2.3 hr. Subsequently, this uid was displaced using a reservoir of challenge solution and the slow and fast ows of challenge solution restored for a further dened period (ca. 20 h) before a second set of velocity measurements based on the CuSO4 solution. This cycle was repeated two more times to effectively create a 4-D data set of the time dependence of the 3-D spatial effects.

Bioreactor Challenge Experiments

Metal accumulation was achieved using two fresh columns side-by-side inside the magnet as shown in Figure 2, and challenged with a uid containing La (NO3)3 6H2O (1 mM; for lanthanum phosphate precipitation onto the biomass), CuSO4 5H2O (1 mM; not precipitated appreciably onto the biomass21), and sodium glycerol-2-phosphate (5 mM; phosphate donor) in sodium citrate buffer (2 mM; to prevent metal hydrolysis and maintain metal solubility in the input solution), pH 6.0. Two different ow rates (11 and 128 ml/hr, hereafter termed slow and fast ow) were generated by an external peristaltic pump (Watson-Marlow, 505U, Wilmington, MA); these conditions were chosen, using data from outow analyses in preliminary tests, to give contrasting results; in the absence of obstructions those would have Reynolds numbers ( N Re) of
3074 November 2005

0.25 and 2.87, respectively. The Peclet numbers for the reactors at the slow and fast ow rates are ca. 22 and 261, respectively, both calculated from the equation Pe UL / D a where: U is the average velocity; L the pore size of the foam; and, D a is the dispersion coefcient.29 Initially the columns were challenged for two hours to allow La3 and Cu2 ions to be backgroundabsorbed onto the biolm. Subsequently, MRI velocity measurements were made under constant head gravitational ow; 1 mM CuSO4 solution was used in order to maximise the signalto-noise and consequently the accuracy of the measurement, without being toxic to the phosphatase enzyme (necessary for the next metal accumulation phase); the efuent was not recycled to prevent recirculation of any biolm cells which had sloughed off. The ow rate used was over 100 times faster than that for the La3 supplemented solution, in order to produce velocities which were measurable by MRI, but also too high for the metal deposition reactions to occur during the ow residence time. The ow rate was measured gravimetrically at the outlet pipe, and was maintained at approximately 3120 ml/h in both reactors for all the ow measurements; this corresponds to a Reynolds number ( N Re) of 70 without any obstructions. Importantly, since the ow data are acquired over many hours it was important to ensure that the ow is steady and non turbulent to enable measurement of velocity. Subsequently, the reactors were challenged for three longer periods (each ca. 20 h) each followed by MRI velocity measurements to spatially probe the progressive blockage (256 128 64 pixels for both reactors, giving a total acquisition time 2.3 h). The time-dependent precipitation of lanthanum phosphate was also monitored by analysis of bioreactor outows using spectrophotometry (see later).
Vol. 51, No. 11 AIChE Journal

Spectrophotometric Analyses of La3, Cu2 and Inorganic Phosphate

The La3, Cu2 and phosphate contents of the outows of the reactors were measured as described by Yong and Macaskie,30 Nott et al.,21 and Yong and Macaskie,31 respectively.

Environmental Scanning Electron Microscopy

Samples of Serratia sp. biolm grown on polyurethane reticulated foam with and without lanthanum phosphate deposits were visualized using environmental scanning electron microscopy (ESEM) (FEI Philips FEG ESEM XL30) in their hydrated state. The samples containing lanthanum phosphate were taken from a slow reactor.

Magnitude Data Analysis

Whereas the phase of the MRI signal was sensitized to velocity, the magnitude of the MRI signal acquired simultaneously, is primarily dependent on water proton density. Hence, the blockage could be quantitated using the magnitude of the MRI image acquired without either ow or ow encoding, by segmenting high signal intensity from the challenge solution from that of any obstructions, such as the foam, biolm, precipitate and/or air bubbles (low signal intensity). The cut off intensity was chosen to be 25% of the maximum signal intensity from the challenge solution, and each data set was referenced against the rst, which had no appreciable blockage despite the two hours exposure to the La3 supplemented solution.
Figure 3. Sagittal and transverse MRI images.
Sagittal and transverse images (taken at the position of the dotted line) of water in a reactor containing biolm immobilised on foam cubes. The pair of upper traces was taken from a 3-D data set of signal magnitude and shows the spatial distribution of water. The image intensity in the lower three traces has been sensitised to the ow velocity in the x -, y - and z -directions.

Results
Figure 3 shows sagittal and transverse slice images taken from the 3-D data sets of signal magnitude (related to water distribution), plus the x , y , z vectorial velocity taken from three separate phase encoded 3-D maps of uid ow through a reactor before treatment. The heavy Cu2 doping (20 mM) dominates the relaxation behavior of water protons in both the biolm and surrounding uid, hence, there is little contrast in the signal magnitude image compared to a previous study;21 in fact the upper image resembles that of reactor with no biolm and visualizes the small and large voids associated with the foam and air bubbles, respectively. Although the x and y velocities were too low to be measured accurately, and the foam structure is not clearly discernible than for z , the z velocity image is very sensitive to the foam structure. The highest z velocities are found in the gaps between the foam cubes which disperse the high velocity jet as it enters the bioreactor; the uid then follows the path of least resistance and channels between the foam cubes and down the sides. Since neither the x nor y components provided additional information, all subsequent velocity measurements were carried out only in the z -direction, parallel to the direction of ow; this reduced the total scan time for the ow measurements from 6.8 to 2.3 h. Figure 4 shows specimen data of the concentrations of lanthanum, copper and phosphate ions and the pH during challenge by an aqueous solution of lanthanum nitrate (1 mM), copper sulphate (1 mM) and sodium glycerol-2-phosphate (5 mM) in a sodium citrate buffer (2 mM), pH 6.0 at the two different ow rates; samples were taken from the outow before each imaging experiment. As was found previously,21 under both conditions the concentration of copper ions in the outow was 1 mM indicating little or no copper removal. In contrast, the majority of the La3 was removed by both the fast and slow ow reactors. The pH in both reactors increased in the latter half of the experiment. Thus, the slow
Vol. 51, No. 11 3075

Flow Data Analysis

In general, 15% of the length of the MR images at either end of the reactor was unusable for quantitative analysis of the ow data because the magnetic eld homogeneity in those regions was poor. In addition, it is apparent from a preliminary study using sagitally orientated slices (Figure 3) that the available gradient linearity (magnetic eld strength vs. position) was poor. Thus, although the transverse image dimensions were circular across the bioreactor, the MRI sagittal image along its length was distorted towards the inlet and outlet. Consequently, the cross-sectional area, required for calculation of the average z velocity (including zeros), had to be determined separately from each slice using the following procedure programmed using Matlab (Mathworks Inc., Natick, U.S.A.): (1) The centers of the circular transverse images at each end of the usable region (central 70% of the length) were dened manually. (2) An imaginary straight line was drawn between them to dene the center of the intervening slices. (3) For each transverse slice, starting from the estimated center, the size of a circle was increased until it had the largest radius (that is, the maximum distance) which included a non zero velocity pixel. (4) For each of the resultant circular regions-of-interest (ROIs) including areas of zero velocity, the average z velocity V avg was calculated; separately the areas of zero velocity were excluded for estimation of the non stagnant average z velocity V ns. (5) The blockage fraction (BF) was calculated using the equation, 1 V avg/ V ns.
AIChE Journal November 2005

fast reactors this was more evenly dispersed along the length of the column. This is to be expected since the bulk of the substrate cleavage, and hence metal precipitation, would occur near the reactor inlet at slow ow rates; in contrast, at faster ows more unreacted substrate can pass to the columns distal areas, thereby giving a more even distribution of precipitate and hence less local blockage. Figure 5 shows ESEM micrographs of Serratia sp. biolm containing lanthanum phosphate deposits taken from a slow reactor. The blockage fraction accumulated during the experiment was measured at three different times by thresholding the magnitude MR-images of both the fast and slow ow reactors. The results in Figure 6 show that even in the early stages precipitation was more uniform in the fast ow reactor (Figure 6a), whereas the slow ow reactor showed preferential blocking at the inlet

Figure 4. Phosphate-release and metal-removal by reactors.

Challenged at (a) 128 ml/h (fast), and (b) 11 ml/h (slow) ow with aqueous La3 (1 mM) and Cu2 (1 mM) in the presence of glycerol-2-phosphate (5 mM) and citrate buffer (2 mM), pH 6.0. F Residual Cu2. Residual La3. Phosphate released. E pH of outow solution. The arrow indicates the time (57.1 h) at which the concentration of free phosphate species were estimated.

reactor gave 1.65 mM phosphate in the outow (at 57.1 h) and a higher pH upshift (6.00 to 6.88) whereas the fast reactor gave 0.69 mM phosphate in the outow (at 57.1 h) and, a smaller pH upshift (6.00 to 6.45). This can be explained by the reaction of the PO3 ion generated by the system, that is, 4 cleaved from the disodium glycerol 2-phosphate, with water initially to form HPO2 4 and then H2PO4 , as follows
3 2 H2O 7 HPO4 OH PO4 2 H2O 7 H2PO4 OH HPO4

(1) (2)

The basic dissociation constants for Eqs. 1 and 2 are K b 1 2.22 10 2 and K b 2 1.58 10 7 , respectively.32 In this case K b 1 K b 2 , therefore, essentially all the OH arises from the rst reaction. The fractions () of HPO2 4 and H2PO4 at pH 6.45 and 6.88 were determined using the plot of distribution of various phosphate species () as a function of pH.33 The estimated concentrations of the phosphate species ([phosphatecontaining species] total phosphate concentration) in the outow of the slow reactor were [HPO2 4 ] 0.54 mM and [H2PO4 ] 1.11 mM, and in the fast reactor were [HPO2 4 ] 0.11 mM and [H2PO4 ] 0.58 mM. Thus, the slow reactor, with a higher pH in the outow, contains a higher concentration of HPO2 species in solution than does 4 the fast reactor. During metal accumulation, all the slow reactor columns showed a heavy deposition near the inlet, whereas with the
3076 November 2005

Figure 5. ESEM micrographs.

(a) Serratia sp. biolm on reticulated polyurethane foam (2136 magnication), and samples of Serratia sp. biolm on reticulated foam containing lanthanum phosphate deposits taken from: (b) the middle (2000 magnication), and (c) the bottom (1,600 magnication) of a slow reactor.

Vol. 51, No. 11

AIChE Journal

poor on a pixel-to-pixel basis; however, using the pixels collectively increases the signal-to-noise and, as a consequence, the condence of the measurements as a whole.

Discussion
These experiments demonstrate the potential utility of MRI not only in this specic application, but for many other processes which take place in columnar reactors, for which the resultant efciency is changed by blockage; for example, ltration. The image data can provide signicant information regarding the spatial distribution of blockage caused a by

Figure 6. Blockage fraction within the reactor as a function of time.

Each of the 2-D slice images was taken from a 3-D magnitude image after 57.1 h of ow either at (a) 128 ml/h (fast), and (b) 11 ml/h (slow). The blockage fractions were calculated by thresholding the signal intensities for scans measured at 17.6, 37.1 and 57.1 h after initiation. All of the data are in good accord with reactor theory: thus, the fast reactor is more efcient with the reactants distributed uniformly, whereas the slow reactor is concentration limited.

region, which was particularly pronounced after 57 h (Figure 6b). The fast ow rate was chosen, by prior experiment, as that which gave a small amount of La3 in the outow solution (that is, the threshold value at which not all of the La3 was removed within the ow residence time). Thus, La3 was removed to 95% at the slow ow rate (Figure 4b), but only ca. 90% at the fast ow rate (Figure 4a); the corresponding phosphate concentrations found in the ow were ca. 1.7 mM and 0.7 mM; at the slow ow rate more glycerol 2-phosphate cleavage occurred within the ow residence time. Figure 7 shows plots of the values of the non stagnant average z velocity ( V ns), average z velocity ( V avg), and blockage fraction (BF) for the central section of the fast ow reactor; there was little variation in the same region of the slow reactor for further analysis (blockage was largely conned to the post-inlet region). In theory, V avg should be the same for each transverse slice across the reactor; however, in practice this was not the case, probably due to the presence of the air bubbles which distorted the image, especially the phase data used to calculate the ow velocity. However, since V avg and V ns are both calculated using the same region of interest, each acts as an internal reference for the other, and, hence, these uctuations cancel out because the air bubbles are stable throughout each 2.3 h experiment. The data show clearly the development of blockages with time in the 7 mm and 28 mm regions (Figure 7c). These data are further quantied in Figure 8, which shows the correlation between blockage fraction and non stagnant average z velocity for the data given in Figure 7, with pooled data from two replicate experiments. The curve was tted to the equation BF 1 V avg/ V ns; hence, the intercept on the V ns axis when BF 0 equates to V avg and the value of 0.47 cm/s is the same as that obtained by volumetric analysis. The wide spread of data is because the signal-to-noise of MRI is
AIChE Journal November 2005

Figure 7. Sample data from a representative MRI velocimetry experiment for the fast reactor.
Plots of: (a) Nonstagnant average z velocity, V ns; (b) average z velocity, V avg, and (c) blockage fraction, BF ( 1 V avg/ V ns), calculated using each slice, transverse to the direction of the ow, along the length of the reactor taken from the 3-D data set. Although in theory, V avg in (b) should be the same for each slice along the reactor, in practice this was not the case. However, V ns which was calculated in (a) using the same region of interest, showed similar uctuations which cancel out for the calculation of BF shown in (c).

Vol. 51, No. 11

3077

of the blockage. In the present study, comparison of the blockage fraction across the fast ow reactor calculated from the velocity data (Figure 7c) with that from the equivalent region of the MR magnitude image (between the dotted lines in Figure 6a) shows that although the precise values are different, the peaks and troughs are similar. If the overall bulk ow rate through the reactor vessel is constant throughout the duration of the experiment, then the average velocity V avg, should remain the same for each transverse slice across the reactor. However, local blockage within any one region will cause the ow to be constricted to smaller areas causing the non stagnant average z velocity, V ns, to increase. This is clearly demonstrated in Figure 8 which shows a progressive increase in ow velocity and a quantitative relationship with blockage.
Figure 8. Correlation between non stagnant average z velocity and blockage fraction.
This correlation is based on the combined data from two independent experiments at the fast ow rate.

Conclusions
This study clearly demonstrates that MRI can quantitate the ow eld in a biolm reactor, distinguishing the lower velocity ow within the polyurethane reticulated foam cubes from the higher velocity ow between them, and along the inner edges of the reactor tube. Flow imaging is also sensitive to the buildup of metal precipitate, and, hence, enables quantitation of the blockage fraction; however, it is not possible to distinguish air bubbles from the precipitate by ow data alone since both regions are characterised by zero ow. The experimental data already available from this study are reminiscent of text book theory, and, hence, in principle, could be used to model and subsequently to optimize the efciency of this biolm reactor for scaleup. Further data could be acquired to rene the current model, developed from applied enzyme kinetics, which assumes ideal plug ow, or that the deviations from ideal behavior are constant throughout, and that the value of the K m of enzymes in immobilized bacteria ( K m apparent) is a constant.5,6,34 That the former is not the case is clearly demonstrated by the radial proles taken along the biolm reactor with increased blockage; the adjacent local ow rate is increased, and, hence, the amount of substrate which ows past the enzyme is a variable parameter, the value of which depends on the extent of local blockage and the exact position of each enzyme molecule in the reactor. Future models need to take this into account and further renements of the MRI technique and subsequent data analysis may be required for future studies of such systems.

progressive reaction, which could not otherwise be gathered from analysis of the bulk outow (the black box approach) or be visualized by eye through the opaque precipitate. In the case of an enzymatic reaction the local ow residence time is critical since the reaction is time dependent; static ow in stagnant areas wastes capacity, whereas rapid ow in others would allow insufcient time to retain V max (maximum reaction velocity) locally. The overall bioreactor activity is dened to relate to the concentration of substrate and available biomass,5,6 but only at a constant ow rate which maintains the local substrate concentration high enough to maintain V max throughout. Analysis of the concentration of La3, Cu2, and phosphate ions as well as the pH of the outow solution from the fast and slow ow reactors showed some differences (Figure 4); however, one cannot predict from those data how efcient the reactors will be in the longer term. Even after the rst day, the signal magnitude of the MR image reveals that the fast reactor is more efcient overall since the reactants, and, hence, as a consequence the precipitate, are both more uniformly distributed; however, in contrast the slow reactor is concentration limited since it cannot replenish the reactants fast enough along the length, which leads to blockage at the inlet. That this combination of results is in accord with reactor theory, is well illustrated in Figure 6. The MR images also reveal the presence of air bubbles which cannot be observed directly by eye, yet are clearly distinguished as zero signal intensity regions compared to the grey scale appearance of the precipitate. In contrast to thresholding of the magnitude images which can be subjective, MRI quantitation of ow through the reactors not only denes the difference between blocked and unblocked regions by distinguishing regions of ow and no-ow, but the resultant velocity contrast is also sensitive to smaller obstructions, such as the foam. Areas of apparent zero velocity include features, such as air bubbles; however, it is important to note that since those images may also include areas where the signal-to-noise is not sufcient to support the Bayesian analysis of the velocity encoded images (that is, below the signal-to-noise threshold), they will tend to overestimate extent
3078 November 2005

Acknowledgments
This work was supported by the U.K. Biotechnology and Biological Sciences Research Council (Grant Nos. EO9214 and 6/E1464), and by the Herchel Smith Endowment. The authors thank Recticel (Belgium) for the biolm support. KPN, FPH and LDH would like to thank Dr. N. J. Herrod and Dr. D. Xing for the computer software and facilities, respectively; and Richard Smith, Simon Smith and Cyril Harbird for supply and maintenance of the MR, and other hardware.

Literature Cited
1. Alexander M. Biodegradation and bioremediation. San Diego: Academic Press, 1999. 2. Valls M, de Lorenzo V. Exploiting the genetic and biochemical capacities of bacteria for the remediation of heavy metal pollution. FEMS Microbiology Reviews. 2002;26:327338. 3. Macaskie LE. An immobilized cell bioprocess for the removal of heavy metals from aqueous ows. J of Chem Technol and Biotechnol. 1990;49:357379.

Vol. 51, No. 11

AIChE Journal

4. Finlay JA, Allan VJM, Conner A, Callow ME, Basnakova G, Macaskie LE. Phosphate release and heavy metal accumulation by biolm-immobilized and chemically-coupled cells of a Citrobacter sp. pre-grown in continuous culture. Biotechnol and Bioeng. 1999;63:87 97. 5. Macaskie LE, Empson RM, Lin F, Tolley MR. Enzymatically-mediated uranium accumulation and uranium recovery using a Citrobacter sp immobilized as a biolm within a plug-ow reactor. J of Chem Technol and Biotechnol. 1995;63:116. 6. Macaskie LE, Yong P, Doyle TC, Roig MG, Diaz M, Manzano T. Bioremediation of uranium-bearing wastewater: biochemical and chemical factors inuencing bioprocess application. Biotechnol and Bioeng. 1997;53:100 109. 7. Gladden LF. Nuclear-magnetic-resonance in chemical-engineering principles and applications. Chem Eng Sci. 1994;49:3339 3408. 8. Gladden LF, Alexander P. Applications of nuclear magnetic resonance imaging in process engineering. Measurement Sci and Technol. 1996; 7:423 435. 9. Gladden LF. Applications of in situ magnetic resonance techniques in chemical reaction engineering. Topics in Catalysis. 1999;8:8795. 10. Lens PNL, Hemminga MA. Nuclear magnetic resonance in environmental engineering: principles and applications. Biodegradation. 1998;9:393 409. 11. Van As H, Lens P. Use of 1H NMR to study transport processes in porous biosystems. J of Ind Microbiol and Biotechnol. 2001;26:4352. 12. Lewandowski Z, Altobelli SA, Majors PD, Fukushima E. NMR imaging of hydrodynamics near microbially colonised surfaces. Water Sci and Technol. 1992;26:577584. 13. Lewandowski Z, Altobelli SA, Fukushima E. NMR and microelectrode studies of hydrodynamics and kinetics in biolms. Biotechnol Progress. 1993;9:40 45. 14. Lewandowski Z, Stoodley P, Altobelli SA, Fukushima E. Hydrodynamics and kinetics in biolm systemsrecent advances and new problems. Water Sci and Technol. 1994;29:223229. 15. Lewandowski Z, Stoodley P, Altobelli SA. Experimental and conceptual studies on mass transport in biolms. Water Sci and Technol. 1995;31:153162. 16. Hoskins BC, Fevang L, Majors PD, Sharma MM, Georgiou G. Selective imaging of biolms in porous media by NMR relaxation. J of Magnetic Resonance. 1999;139:6773. 17. Nestle N, Kimmich R. Heavy metal uptake of alginate gels studied by NMR microscopy. Colloids and Surfaces A: Physiochem and Eng Aspects. 1996;115:141147. 18. Nestle N, Kimmich R. NMR microscopy of heavy metal absorption in calcium alginate beads. Appl Biochem and Biotechnol. 1996;56:9 17. 19. Nestle N, Kimmich R. Susceptibility NMR microimaging of heavy

20. 21. 22. 23. 24. 25. 26.

27. 28. 29. 30.

31. 32. 33. 34.

metal uptake in alginate biosorbents. Magnetic Resonance Imaging. 1996;14:905906. Nestle N, Kimmich R. NMR imaging of heavy metal absorption in alginate, immobilised cells, and kombu algal biosorbents. Biotechnol and Bioeng. 1996;51:538 543. Nott KP, Paterson-Beedle M, Macaskie LE, Hall LD. Visualisation of metal deposition in biolm reactors by three-dimensional magnetic resonance imaging (MRI). Biotechnol Letts. 2001;23:1749 1757. Paterson-Beedle M, Nott KP, Macaskie LE, Hall LD. Study of biolm within a packed-bed reactor by three-dimensional magnetic resonance imaging. Methods in Enzymology. 2001;337:285305. Caprihan A, Fukushima E. Flow measurements by NMR. Physics Reports. 1990;198:195235. Pope JM, Yao S. Quantitative NMR imaging of ow, Concepts in Magnetic Resonance. 1993;5:281302. Fukushima E. Nuclear magnetic resonance as a tool to study ow. Annual Reviews of Fluid Mechanics. 1999;31:95123. Pattanapipitpaisal P, Mabbett AN, Finlay JA, Beswick AJ, PatersonBeedle M, Essa A, Wright J, Tolley MR, Badar U, Ahmed N, Hobman JL, Brown NL, Macaskie LE. Reduction of Cr(VI) and bioaccumulation of chromium by Gram positive and Gram negative microorganisms not previously exposed to Cr-stress. Environ Technol. 2002;23: 731745. Allan VJM, Callow ME, Macaskie LE, Paterson-Beedle M. Effect of nutrient limitation on biolm formation and phosphatase activity of a Citrobacter sp. Microbiology. 2002;148:277288. Xing D, Gibbs SJ, Derbyshire JA, Fordham EJ, Carpenter TA, Hall LD. Bayesian analysis for quantitative NMR ow and diffusion imaging. J of Magnetic Resonance, Series B. 1995;106:19. Fogler HS. Elements of Chemical Reaction Engineering (3rd edition). New Jersey: Prentice-Hall International; 1999:882 883. Yong P, Macaskie LE. Bioaccumulation of lanthanum, uranium and thorium, and use of a model system to develop a method for the biologically-mediated removal of plutonium from solution. J of Chem Technol and Biotechnol. 1998;71:1526. Yong P, Macaskie LE. Removal of the tetravalent actinide thorium from solution by a biocatalytic system. J of Chem Technol and Biotechnol. 1995;64:8795. Skoog DA, West DM, Holler FJ. Fundamentals of Analytical Chemistry (7th edition). New York: Saunders College Publishing; 1996. Peters DG, Hayes JM, Hieftje GM. Chemical Separations and MeasurementsTheory and Practice of Analytical Chemistry. Philadelphia: Saunders Company, 1974. Yong P, Macaskie LE. Effect of substrate concentration and nitrate inhibition on product release and heavy metal removal by a Citrobacter sp. Biotechnol and Bioeng. 1997;55:821 830.

Manuscript received May 28, 2004, and revision received Feb. 25, 2005.

AIChE Journal

November 2005

Vol. 51, No. 11

3079