Is the Souders-Brown equation sucient for scrubber design?

An experimental investigation at elevated pressure with hydrocarbon uids

Trond Austrheim, Lars H. Gjertsen

Statoil ASA, 7005 Trondheim, Norway

Alex C. Homann
Dept. of Physics and Technology, University of Bergen, Allegt. 55, 5007 Bergen, Norway

Abstract Scrubber design practice today is largely based on experimental data generated at ambient conditions with air-water systems. This paper reports experimental results for gas scrubber performance at pressures up to 92 bar using two types of hydrocarbon liquids: Exxsol D60 and a synthesized live natural gas condensate. The scrubber has a conguration very commonly used in industry with three types of internals in series: a vane inlet, a mesh pad and a cyclone deck. Results for the separation eciency and the pressure drop of the internals separately and combined are shown graphically. The relevance of the Souders-Brown K-value, commonly used as scrubber design parameter, is elucidated both theoretically and in light of the results. The results show that the Souders-Brown value is rather good in practise

Preprint submitted to Elsevier

9 June 2007

for design of inlet vane and mesh pad, and that results for the two hydrocarbon systems agree reasonably, but also that better design rules are required for highpressure separation of hydrocarbon liquids. Key words: Gas scrubber, Souders-Brown equation, inlet, mesh pad, axial ow cyclone, high pressure, hydrocarbon uids

Introduction

While conventional oil production is expected to reach a plateau within the near future, the production of natural gas and associated products is expected to increase and become increasingly important as new gas elds are being developed and oil elds enter tail-end production. Development of oil and gas elds moves toward more marginal and remote elds worldwide, often oshore and in deep waters. This requires more costeective processing, and the trend is toward remotely controlled sub-sea processing. In addition to reducing the need for expensive top-side facilities and pumping of multiphase mixtures, sub-sea separation makes it possible to reclaim natural gas at high pressure reducing needs for recompression before pipeline transport. However, remote subsea separation requires more robust design of separation equipment. Expenses due to failures caused by faulty design and/or operation of separators are much larger. In addition to this, separators may need to be more compact for installation sub sea, since they need to be carried and
Corresponding author

submerged from ships, that can handle only a limited weight. This adds extra, rigorous demands on the quality of the design. Today, much design is based on experimental results obtained in air-water systems under ambient conditions. Design for the rigorous duties mentioned above requires data for high-pressure operation on hydrocarbon uids, and even live natural gas uids. Very little public-domain research literature is available with information about the performance of scrubbers acting on hydrocarbon uids and elevated pressures. To the authors knowledge, no tests under high-pressure conditions are reported in the public domain for conventional scrubber internals. For compact separation equipment, a few results are found. The Gasunie scrubber (Oranje, 1990) was tested with processed natural gas at operating pressures up to 40 bara, but no information was given about the liquid phase. More recently, Chin and Standbridge (2003) tested an in-line degasser using a mixture of methane and diesel up to 40 bara, but no information was given on the physical properties of the uid. In a sense their uid was live since some diesel will evaporate and some methane will dissolve in the liquid. This system, however, only covers a small part of the gas-liquid properties existing in real scrubbers. Rawlins and Ting (2002) reported results for a long-term eld test of an IRISunit (a demister mounted directly in a pipe-line, having a separation principle similar to an axial-ow cyclone) at 77 bara pressure. The IRIS was mounted in a well stream consisting of natural gas, condensate and associated water from 3

the well. Up till now, these are, to our knowledge, the most challenging conditions under which a natural gas separator has been tested in the published literature. In addition Rawlins and Ting also did some more systematical testing in a large-scale lab where processed natural gas and decane was used as the uid system. The aim of this project is to provide data generated in a standard scrubber conguration at a full range of pressures on realistic uids, and make a rst assessment on the impact on scrubber design methods. It builds on an investigation of scrubber performance at low pressures, showing that the nature of the liquid used has a signicant impact on the separator performance (Austrheim et al., 2006).

Theory, the Souders-Brown Equation and its Signicance

The most used expression for sizing of gas scrubbers is the one developed by Souders and Brown (1934) for sizing of fractionating columns. This involves an empirically quantied factor known as the Souders-Brown value, the Kvalue, or the Gas Load Factor (GLF). Here, the term K-value will be used. The basis of the Souders-Brown expression is a force balance resolved in the vertical direction on a spherical droplet in an upward owing gas in a gravity eld. When the droplet is held stationary, while moving at its terminal velocity relative to the gas ug,set , the ow force, Fr , balances the gravity force, Gd : 1 Fr = Cd Ad g u2 = Gd = d3 g(l g ), g,set 2 6 d where Cd is the drag coecient, Ad is the projected area of the droplet, (1)
d2 d 4

with dd the droplet diameter, and g and l are the densities of the gas and 4

liquid, respectively. Equation (1) can be rearranged to: ug,set g = l g 4gdd . 3Cd (2)

The right-hand-side of the equation is dened as the K-value: K= 4gdd . 3Cd (3)

Thus, if Cd is constant, designing and operating a column at a constant Kvalue means that a droplet of a given diameter will just not be transported upward and out of the column, irrespective of the uid properties. In practice Cd varies with the droplet Reynolds number, Rer =
g ug,set dd ,

except for high values of Rer , where Newtons law states that it is constant and about equal to 0.43. At low Rer , the well-known relation of Stokes states that: Cd = 24 , Rer

while a general empirical expression due to Putnam (1961), valid for Rer < 1000 is: Cd = 24 Re2/3 r 1+ . Rer 6 (4)

The validity range of the Putnam expression is sucient for most gas scrubbers in practice. The paper of Souders and Brown focuses on terminal settling in a gravity eld, and shows that a column has to be designed for a lower gas velocity at higher pressures in order to avoid ooding. To investigate the wider physical signicance of the K-value, we can study the unsteady equation of motion of a particle. If we write the equation of motion for a spherical droplet neglecting, as is 5

often done, the added mass and the history integral terms while taking into account gravity, the equation of motion for a spherical droplet is: m du 1 = Cd d2 g |ur | ur + mg dt 4 d2 I II III

(5)

Where u is the particles absolute velocity, and ur its velocity relative to the gas. The terms are from left to right: I: particle mass times acceleration, II: uid drag force and III: body force (due to gravity). The drag, II, tends to make the droplets follow the gas stream, and can thus be said to oppose separation. The two other terms can be seen as forces that will make the particle move relative to the gas, and therefore as separating forces that may separate the particles from the gas stream. We consider two types of separation equipment based on gravity and impaction, respectively: In separation in a gravity eld, term I is normally zero, and the separating force, III, depends only on particle mass and the acceleration due to gravity, but not the uid velocity. This is the situation scaled correctly by K, as shown above. In separation based on impaction or centrifugation, on the other hand, the gravitational term, III, is often negligibly small and term I constitutes the separating force. In cyclones, for example, where the particle rotates with the same tangential velocity as the gas, v , the magnitude of the acceleration
du dt

is

2 v . r

In fact, in all such equipment

du dt

is approximately proportional

to the uid velocity squared. We thus see that the variation of the separating force with gas velocity is quite dierent for the two types of separation equipment, and that the Souders-Brown value is not relevant for impaction equipment at all. 6

In practice, when designing a column to avoid that the upward velocity entrains droplets, the recommended K-value is K < 0.1 m/s for low-pressure applications; often a safety margin of 50% is added for vessels without internals. For increasing pressures the critical K-value has been seen to decline. This is not surprising, since increasing pressure in oil/gas applications is often accompanied by a decrease in interfacial tension and thereby a decrease in the droplet sizes. Gas Processors Suppliers Association (GPSA) (1998) recommend (for separators with a mesh pad) decreasing the K-value with 25% for 85 bar pressure. Obviously, if a separator is equipped with separation equipment, e.g. a mesh pad or cyclones, it can operate at a higher K-value than that required if it simply separates by settling under gravity against the ow in an empty separation space. In fact, the K-value at which a given separator can operate without entrainment can be seen as a measure of its compactness, and in practice some scrubbers can operate at K-values up to 0.3 m/s (Gjertsen et al., 2003).

Experimental Facilities

The experimental rig used for this work is specially designed for scrubber testing with live natural gas uids, but can in theory be used for all kinds of uids. Separation eciency and pressure drops can be determined for various types of separation equipmentboth total scrubber congurations and individual internalsover a wide range of gas ows and liquid loads at pressures up to 100 bara and temperatures ranging from 40 to +50 C. 7

The dierence between a hydrocarbon model uid (nitrogen/Exxsol) and a natural gas system was investigated. These are described in more detail below. Three dierent operating pressures were used: 20, 50 and 92 bara. All experiments were carried out in the temperature range 2025 C.

A 3-D sketch of the scrubber investigated is shown in Figure 1. The inlet vane distributes the incoming two-phase ow over the cross-section, and separates some of the liquid from the gas. The mesh pad may act as a further separator at low liquid loading, or as a coalescer for the cyclone deck at high loadings, the gas loading dominates in determining how the vane pack acts. The deck of axial ow cyclones (AFCs), or once-through cyclones, working in parallel is the last separation step.

Fig. 1. Diagram of the scrubber conguration investigated here. At the bottom the vaned inlet, in the middle the mesh pad, and at the top the cyclone deck consisting of two cyclones working in parallel

3.1 The rig

The rig has been designed to generate data under actual eld conditions, and is built inside a container in which both the atmosphere and the process itself are temperature controlled by a climate control system. The test rig is remotely operated with a PC-based control system. The rig is built as a closed circulation loop. Liquid is initially charged and gas subsequently used to pressurize the rig. Liquid can also be charged under pressure by use of a piston pump. The rig is designed to operate at pressures from 1 to 100 bara. A gas blower with an adjustable speed motor circulates the gas in the loop at owrates ranging from 0 to 60 m3 /hr at 100 bara. The gas blower is submerged in a temperature-controlled glycol bath which provides control of the gas temperature in the loop. Liquid can be injected into the gas stream through either two injection points in the inlet piping or through a nozzle within the test scrubber at rates ranging from 0.05 to 1.0 m3 /hr. A sketch of the main process can be seen in Figure 2. The liquid captured in the test scrubber can be drained to two separate drain tanks, while the entrained (or lost or overhead) liquid is captured by a bulk scrubber, which is a scrubber equipped with a mesh pad, of twice as large a diameter as the test scrubber. If the gas ow is high or the droplets are very small, some liquid might escape this bulk scrubber as well, the remaining liquid will then be separated by a tangential cyclone scrubber and a large lter coalescer further downstream. The captured and overhead liquid fractions are measured by liquid accumulation measurements using dierential pressure cells detecting the liquid level in 9

the drain tanks, bulk scrubber, tangential cyclone scrubber and lter coalescer.
Gas circulator Tangential Bulk scrubber cyclone w/meshpad scrubber w Test scrubber w
dP

FT

Filter Coalescer
dP

dP

w Liquid flow w
2 3

Drain tanks

Main gas flow Gas booster Liquid reservoir Gas bottles

dP

w
dP FT dP

Liquid Piston Pump

Fig. 2. A sketch of the experimental rig used. The main gas ow-loop is marked by a thickened line. The inlet (49.24 mm ID) and drainpipes on the test scrubber (150 mm ID) are visible, and the 5 window sections are indicated by w. Other internal diameters: Bulk scrubber: 298.7 mm; tangential cyclone scrubber: 202.7 mm; lter coalescer: 460 mm; drain tank 2: 298.6 mm; drain tank 3: 295.4 mm

The circulated gas and liquid rates are continuously monitored by Coriolis meters. The test scrubber is equipped with ve window sections at dierent heights. The height of the test scrubber is two meters from the inlet to the top. This design oers the possibility of testing total scrubber congurations with a large range of distances between the internals. Figure 3 shows the scrubber arrangement used. A cyclone deck consisting of two cyclones working in parallel is placed in the top of the test scrubber. Below the deck is a mesh, or mesh pad, which in turn is placed above an inlet 10

vane. In contrast to a real scrubber, the drain pipe from the cyclone deck does not extend down to the liquid sump in the bottom of the test scrubber, but penetrates the scrubber wall and the liquid is thereby collected in a separate drain tank.

Cyclone Deck
250 mm

dP dP

Nozzle Drain pipe

865 mm

dP

Mesh
150 mm

dP
485 mm

Inlet Vane Inlet Pipe

Fig. 3. The setup in the test scrubber (ID 150 mm), the right-hand gure shows the pressure drops measured

The setup for the pressure drop measurements is shown in the right-hand gure. The pressure dierential cells are in contact with the system pressure through 1/4 tubing that penetrate the scrubber wall. This tubing extends one cm into the process vessel and thereby through any possible liquid lm on the wall, except in the cyclone drain chamber, where the 1/4 pipe is connected through the top plate of the cyclone deck. Measurements were carried out over the mesh pad and the cyclone deck. Cyclone deck pressure drop measurements were divided in two parts: (1) Pressure drop from inlet underneath to the drain chamber and 11

(2) Pressure drop from the drain chamber to the outlet. The sum of these two should be equal to the total pressure drop as can be seen from Figure 3. To minimize the inuence of any dynamical contribution to the measured pressure above the cyclone deck, a vortex breaker was installed to attenuate a strong swirl that was observed during experiments in another rig.

3.2 Internals

The internals used for droplet separation, i.e. the cyclones, mesh pad and inlet vane are of the same type as those used in another rig operating at low pressure, and have been described in detail in another paper Austrheim et al. (2006). In summary it can be said that the two cyclones in the deck were designed in accordance with the cyclones used in the work of Verlaan (1991). They have an internal diameter of 5 cm, and a height of 25 cm. They are of the axial-ow type with swirl vanes (exit angle 45 ), and are equipped with vertical slits in the wall for improved liquid separation. The cyclones were modied with a vortex nder as described in Austrheim et al. (2006). The mesh pad, constructed in stainless steel, AISI 304, by Costacurta S.p.A Vico is identical to the Style A investigated by Brunazzi and Paglianti (1998). The wire diameter is 0.27 mm and the void fraction 0.98. The mesh pad is made by knitting wires to form a two-layer mesh that is rolled up spirally to form a cylindrical pad. The inlet vane distributes the incoming gas through a series of vanes at the 12

sides over the column cross-section, as shown in Figure 4.

Fig. 4. Detail of the vaned inlet. The liquid-laden gas enters through the inlet to the left and exits though the vanes on the sides. Some of the liquid is separated by the vanes

3.3 Fluids and uid properties

Since the uids used are an important distinguishing feature of this study, they are described in some detail here. In this and the following sections we use the term uid for the liquid and gas phases together. Two dierent uid systems have been used in the tests. The simplest system consisted of Exxsol D60 as liquid and nitrogen as gas. The rig was lled with Exxsol D60 to the required level and then nitrogen was used to pressurize the rig. Independently of the pressure, the Exxsol D60 mainly remains in the liquid phase while the nitrogen mainly remains in the gas phase. The more complex natural gas uid system consisted of a mixture of methane, 13

ethane and pentane. The rig was initially lled with pentane to a minimum level and then a mixture of premixed methane and ethane gas was used to bring the rig to the required pressure. The gas and liquid phase were then circulated until equilibrium was established. The composition of each phase could be calculated on basis of the initial amount of liquid pentane and the actual temperature and pressure at equilibrium knowing the total volume of the rig. The total volume of the rig was determined by adding a precisely known amount of nitrogen gas to the rig and then measuring the ensuing increase in pressure.

3.3.1 Properties for the Nitrogen-Exxsol D60 Test Fluid Exxsol D60 is not an exact composition since it is a paran distillation cut, but the composition is quite constant, and the supplier therefore gives typical values for its properties. In Tab. 1 a typical composition of the nitrogen/Exxsol system is given. The dierent fractions are groups of components with the specied number of carbon atoms that have been lumped to one fraction with a characteristic molecular weight. This composition has been used as a basis for calculations of the uid properties of the Exxsol/nitrogen mixture used. The gas density was calculated using the Soave-Redlich-Kwong equation of state (EOS) Soave (1972) with the volume correction suggested by Peneloux and Evelyne (1982). The density of nitrogen with a small amount of dissolved Exxsol turned out to be almost linear in pressure at 20 C, ranging from 1.14 kg/m3 at 1 bara to 113.67 kg/m3 at 100 bara. For the liquid density, measurements on Exxsol D60, saturated with nitrogen at operating temperature and pressure, were carried out in a high-pressure 14

Table 1 The uid (liquid and gas combined) composition of the nitrogen/Exxsol uid Component Fraction [mol%] N2 C9 C10 C11 C12 C13 33.034 0.832 13.517 28.066 19.376 5.175 Mol. wt. [g/mol] 28 121 134 147 161 175 781 792 796 810 825 Liq. Dens. [kg/m3 ]

density cell supplied by Anton Paar. The density variations within the pressure operating range were less than 1%. When increasing the pressure from 1.17 bara, the density drops o from its initial value of 784.4 kg/m3 , since more gas dissolved in the liquid, but when the pressure exceeds 10 bara, the density is more dominated by the pressure and increases smoothly with increasing pressure to 790.6 kg/m3 at 100.0 bara. The gas and liquid viscosity calculations are based on the corresponding states principle in the form suggested by Pedersen and Fredenslund (1987). The gas viscosity was calculated to increase almost linearly with pressure at 20 C from 0.0178 cp at 1 bara to 0.0203 at 100 bara. The supplier gives a typical value of 1.58 cP for the liquid viscosity of pure Exxsol D60 at 25 C and atmospheric pressure, which is approximately 20% higher than the calculated 15

value. The uncertainty is likely due to both uncertainties in the prediction model, the characterization of the heavy C9-C13 components, the molar composition of the actual sample and the fact that the calculations take into account that nitrogen is dissolved in the Exxsol. The liquid viscosity was calculated to vary little with pressure at 20 C being 1.34 cp at 1 bara, going through a shallow maximum of 1.43 cp at about 30 bara, and decreasing to 1.30 cp at 100 bara. It was found to be quite a strong function of temperature, being equal to 1.25 cp at 1 bara and 25 C. The interfacial tension may be crucial to the performance of the rig, and varied quite substantially with the pressure. The calculated interfacial tension is shown in Figure 5. The calculations are based on the simple procedure that Weinaug and Katz (1943) used for a Methane-Propane mixture, the accuracy of which depends on the accuracy of the density calculations. Since the uncertainty in the density calculations can be quite large, the interfacial tension should ideally be measured. High-pressure interfacial tension measurements are very challenging, however, and therefore only a few measurements were performed, for pressures ranging from 2 to 6 bara. These measurements show an interfacial tension between 24 and 25 mN/m at 20 C, so the calculations seem to under-predict the interfacial tension by up to 10% in the low-pressure range.

3.3.2 Fluid Properties for the Synthetic Natural Gas The nitrogen/Exxsol D60 model system can be considered as two separate phases where only small amounts of liquid are dissolved in the gas phase and vice versa. The natural gas system on the other hand, is a live gas, a 16

Interfacial Tension [mN/m] .

24 23 22 21 20 19 18 17 16 15 0 10 20 30 40 50 60 70 80 90 100 20oC 25oC

Pressure [bara]

Fig. 5. The calculated interfacial tension of Exxsol D60 and nitrogen

three-component system in two-phase equilibrium. The physical properties of the natural gas system are much more dependent on pressure than the nitrogen/Exxsol system where only the gas density varies signicantly with pressure. For the tests with synthetic natural gas, mixtures of methane, ethane and pentane were used. The rig was lled with an initial amount of liquid pentane at 20 C, then a premixed gas consisting of 85 mol% methane and 15 mol% ethane was used to pressurize the rig. The rig was operated at three dierent pressures, 20.1, 50.2 and 92 bara, when the synthetic natural gas was used. All tests were performed at temperatures within the range 20-23 C. The total uid compositions were calculated with the SRK EOS Soave (1972) with Peneloux volume correction Peneloux and Evelyne (1982) for each operating pressure. Since all the pentane is lled initially and only methane and ethane is added, the uid composition varies with pressure, the uid becoming lighter with increasing pressure, as seen in Tab. 2. In Figure 6 the phase envelopes for the uid compositions corresponding to the three dierent test 17

pressures are shown, and the operating points are indicated.

Table 2 The calculated uid compositions for the three dierent operating pressures at 21 C Fluid Composition Methane [mol%] Ethane [mol%] N-Pentane [mol%]
160 140 120 20.1 bara 50.2 bara 92 bara Critical point

20.1 bara 45.6 8.0 46.4

50.2 bara 64.6 11.4 24.0

92 bara 73.6 13.0 13.4

Pressure [bara]

100 80 60 40 20 0 -50 Operating points

50

100

150

200

Temperature [oC]

Fig. 6. The gure shows the calculated phase envelopes for the three uid compositions given in Tab. 2. The operating points are also indicated

The calculated composition was used as a basis for uid properties calculations. In order to verify the uid composition calculations, a gas sample from the rig was taken at 92 bara and analyzed in a gas chromatograph. The analysis was then compared to the calculated gas composition at the actual pressure and temperature for the sample. The calculations were in good accordance with the analysis, the relative deviations in the methane, ethane and n-pentane mol percentages being 0.6, -3.2 and -5.4%, respectively. Due to the large amount of uids required, completely pure uids could not 18

be used, and some trace elements were present, nitrogen and propane being the most abundant with 0.87 and 0.23 mol%, respectively. These have been ignored, and compositions normalized to methane, ethane and pentane. The uid properties of the synthetic natural gas were calculated using the same methods as described for the nitrogen/Exxsol D60 uid. They are listed in Tab. 3.
Table 3 The calculated gas and liquid properties at gas-liquid equilibrium at 20 C for the three uid compositions given in Tab. 2 Calculated Fluid Properties Gas Density [kg/m3 ] Liquid Density [kg/m3 ] Gas viscosity [cP] Liquid Viscosity [cP] Interfacial Tension [mN/m] 20.1 bara 17.2 602.2 0.011 0.207 11.4 50.2 bara 45.2 551.3 0.012 0.150 6.8 92 bara 97.0 469.5 0.015 0.096 2.2

The properties in Tab. 3 are based on calculated compositions, and thus contains two uncertainties: in the composition calculation and in the properties calculations. To assess these, the same pentane to methane/ethane ratio as in Tab. 2 was added to pressurized sample cylinders, and the liquid and gas density of these samples measured with the Anton Paar high-pressure density cell. A comparison of the measured densities with the calculated properties in Tab. 3 includes errors from both the composition and the properties calculations. In addition, the actual gas composition in the sample cylinders was analyzed in a Gas Chromatograph (GC), and the analyzed composition used as 19

input in a second density calculation. Comparing this second calculation with the gas density in Tab. 3 eliminates most of the uncertainty in the composition calculation. All of the comparisons were favorable, the relative deviations never exceeding 2.5% for the gas phase and 3.5% for the liquid phase. Measurements of the other uid properties i.e. viscosity and interfacial tension, are not available. However, despite the simplicity of the model of Weinaug and Katz (1943), it should be suitable for this mixture. Weinaug and Katz developed the model by use of methane and propanetwo components that have very much the same chemical properties as the components used in this project. Also, the density calculations have been seen to be fairly good. The density calculations are used as input in the interfacial tension calculations.

Results and Discussion

In this section separation eciency for the inlet vane/mesh pad and the cyclones is given. The gas and liquid loads have been varied systematically. Cyclone decks are normally delivered with cyclones of xed sizes that operate in parallel, the number of the cyclones varying with the scrubber cross-sectional area. The liquid load to the cyclones has therefore been expressed as the liquid rate per cyclone. The size of the inlet sections and the mesh pads, however, do vary with the cross-sectional area of the vessel and the liquid load in these experiments has therefore been expressed as the volumetric liquid content in the gas. The gas load has been systematically varied so that the same K-values has been tested for the dierent test pressures and uid systems. Some of the liquid at the scrubber inlet will be separated by the inlet vane, 20

some will impinge on the wall to form a lm that drains and some will coalesce in the mesh pad and drain as large droplets by means of gravity against the upward owing gas. It is not possible to dierentiate between the dierent mechanisms and the combined eciency of the inlet vane and the mesh pad is therefore given instead. The combined eciency is hereafter referred to as the primary separation eciency.

4.1 Performance of the Inlet Vane and Mesh Pad

The inlet vane and the mesh pad are expected to do the main separation, while the cyclone deck should separate any remaining liquid in the gas stream from the mesh pad. The combined eciency of the inlet vane and mesh pad is therefore referred to as the primary separation eciency. The primary separation eciency of the scrubber was tested by injecting liquid in the inlet pipe two meters upstream of the inlet vane, and was calculated as: =1 Q(carry-over) + Q(drain tank 2) Q(injected) (6)

The reason for injecting the liquid pipe two meters, i.e. approximately 40 diameters, upstream of the inlet vane was to allow the ow pattern to be fairly developed by the time the liquid reached the inlet vane, so that the liquid distribution at the inlet vane would be comparable to conditions in a process plant, where, for example, some of the liquid may be owing in the form of a lm on the inner surface of the inlet tube. The K-value in Equation (3) was varied identically for all three test pressures and both uid systems when testing the primary separation performance. The K-value was controlled by adjusting the cross-sectional supercial velocity in 21

the scrubber. In Figure 7 the liquid concentration, or liquid loading, has been kept constant at 0.2 vol% in all experiments (thus the mass fraction of liquid in the stream decreases with increasing pressure) and the resulting primary separation eciency has been plotted as a function of the K-value.
100 90 80

Efficiency [%]

70 60 50 40 30 20 10 0 0.05
N2/Exxsol 20 bar 50 bar 92 bar Natural gas 20 bar 50 bar 92 bar

0.1

0.15

0.2

0.25

0.3

K-value [m/s]

Fig. 7. The primary separation eciency (combined eciency of the inlet vane and mesh pad) as function of the K-value for two dierent uid systems with 0.2 vol% liquid at three pressures

The primary eciency is above 95% for all experiments below K = 0.14 m/s. Most experiments in this region have eciency close to 100% but some of the experiments, especially the 92 bara natural gas case show lower eciency. The uncertainties for the primary separation eciency at these loadings were typically below 0.5%. During these experiments below K = 0.14 m/s the mesh pad was operated below ooding conditions so the measured carry-over is most probably related to small droplets that penetrate the mesh pad. The sudden break in the curves is due to ooding of the mesh pad. The breaks in the curves generally occur at similar K-values, although at slightly lower K-values for the nitrogen/Exxsol system than for the natural gas system, even 22

though the physical properties of the latter system are more severe in terms of separation eciency. There are two possible reasons for this slight dierence: The K-value over-compensates for the liquid density so that the resulting supercial gas velocity is reduced too much when experiments with two dierent liquid densities are compared. It is not clear how the K-value exactly relates to the phenomenon of ooding in the mesh pad anyway (see section 2), although it is likely to be a relevant parameter, since ooding is liquid build-up in the mesh, which depends on the gas lift. Since the supercial gas velocity is less at the same K-value in the natural gas cases, the absolute amount of liquid is also less, since the experiments are compared in terms of equal liquid concentration and not absolute liquid rate. The ooding velocity in packed columns decreases with increasing liquid load Sherwood et al. (1938) and this has been found to be valid also for mesh pads Brkholz (1989), but is not accounted for in the expression u for K. The deviation in break point in the curves in Figure 7 is probably a combination of the two mentioned eects. The deviation is better illustrated by the pressure drop, as will be shown below. Figure 8 shows the primary separation eciency as a function of liquid loading for a xed K-value at K = 0.15 m/s. The results show that the primary eciency increases with increasing liquid load (but also an increasing absolute liquid carry-over). Verlaan (1991) found that the eciency of the vaned inlet itself is likely to decrease with increasing load, so this eect must be mainly due to eects in the mesh pad. During the nitrogen/Exxsol experiments, the condition of the mesh pad was 23

100 95

Efficiency [%]

90 85 80 75 70 0 0.2 0.4 0.6 0.8 1 1.2 1.4 20 bar 50 bar 92 bar

Liquid Concentration [vol%]

Fig. 8. The primary separation eciency as a function of liquid loading at K = 0.15 m/s.

well above the ood point at this K-value, while during the natural gas experiments the ood point generally occurred at higher K-values. The primary eciency results at K = 0.15 m/s with natural gas were therefore very inuenced by how far the ooding process had developed. The results at this K-value are therefore not included in the plot. The pressure drop over the mesh pad is shown in Fig 9. This is very much related to the amount of liquid that builds up in the mesh pad, which again depends on the gas lift. Therefore it is relevant to plot the results against the K-value. At the lowest K-values the pressure drop is dependent on the gas velocity and density. As the K- value increases, a sudden increase in pressure drop can be seen. This increase occurs when liquid starts to build up in the mesh pad, so the sharp break in the pressure drop curve is actually a good way of identifying the ood point. The pressure drop in the nitrogen/Exxsol D60 uid system is much higher than the corresponding pressure drop in the natural gas uid system. The 24

dierence between the two uid systems is mainly the lower liquid density, liquid viscosity and surface tension in the latter system. The explanation for the lower pressure drop in the natural gas system is probably a combination of less liquid hold up in the mesh pad and that less energy is required to lift the less dense liquid. For instance, the natural gas condensate density at 92 bara is only 60% of the liquid density for Exxsol D60. When the mesh is completely ooded, the increase in pressure drop ceases, and the further development of the pressure drop is the competing eects of: on the one hand the gas lift becoming so strong that the amount of liquid build-up in the mesh decreases due to entrainment and, on the other hand, the increasing ow force. The ood point for the nitrogen/Exxsol experiments occur at approximately the same K-valueindependent on operating pressure. This shows that the K-value can be a reasonable way of pressure scaling for simple model uid systems where the physical uid properties, except for the gas density, do not change much. However, the results also show that scaling with the K-value on basis of experiments on simple gas-liquid model systems, even when the model liquid is a hydrocarbon-based one as here, is less than optimal when designing separators for natural gas uids. Reasons for the mismatch between the ood points between the two dierent systems are discussed above, and include over-compensation of the liquid density in the K-value, dependency on actual liquid rate and lack of connection between ood point and other physical uid properties than the densities. We can thus make some concluding remarks about the primary separation e25

9 8

Pressure Drop [mbar]

7 6 5 4 3 2 1 0 0.07

N2/Exxsol 20 bar 50 bar 92 bar Natural gas 20 bar 50 bar 92 bar

0.09

0.11

0.13

0.15

0.17

0.19

0.21

0.23

0.25

K-value [m/s]

Fig. 9. The pressure drop over the mesh pad as ctions of the K-value for the two dierent uid systems at a constant liquid loading of 0.2 vol%

ciency and its scaling. We showed in Section 2 that the K-value is not relevant for describing the separation in internals based on inertial separation, such as the inlet vane and mesh pad. However, as we demonstrated in an earlier paper Austrheim et al. (2006), the limitation in eciency under these present, realistic, operating conditions is likely to be re-entrainment rather than insucient separation of small droplets. Containment involves the upward transport of re-entrained droplets from the vane and mesh pad against the force of gravity, a process described correctly by the K-value. Design is, indeed, often done on basis of the K-value. The design criterion normally handled for scrubbers with these two types of internals is K 0.15 m/s. Figures 7 and 9 show that this normally handled design criterion is rather good. All the curves break at a K-value of around 0.15 m/s. Nevertheless, studying the plots in detail reveals that designing on basis of the K-value does not fully account for the dierences between the two uid systems or the dierences between the pressures. Additionally, Figure 8 reveals considerable 26

dierences in the separation eciency with diering liquid loads at a constant K-value of 0.15 m/s, showing that design on basis of the K-value does not account for the eect of liquid load suciently well.

4.2 Performance of the Cyclone Deck

If the load, for which the inlet vane and the mesh pad are designed, is exceeded the cyclone deck should separate the remaining liquid in the gas stream before the gas exits the scrubber. We measured the liquid load to the mesh pad in volume percent liquid. This is a convenient measure for designers, since mesh pads vary in size with the scrubber cross-sectional area. Cyclones, however, are normally of a xed size (often 50 or 80 mm diameter), and only their number vary with the crosssectional area of the scrubber. For this reason the liquid load to the cyclones can more conveniently be measured as the liquid ow rate per cyclone, and this is the measure we use in this section. The cyclone separation performance was tested by introducing the liquid through the nozzle below the two cyclones (Figure 3). The Delavan Spray Technology nozzle employed in this study was a single phase nozzle (type 1/4 BN6) that produces a solid cone spray pattern; it was placed around 150 mm below the cyclone deck. The gas and liquid ow is assumed to be evenly distributed to the cyclone deck, due to the small cross-sectional area of the scrubber. A certain fraction, the size of which depended on the gas load and pressure, of the liquid injected through the nozzle separated before it reached the cyclone 27

deck. For this reason the rate of liquid injection had to be adjusted to obtain a xed liquid ow reaching the cyclone deck at the varying gas loads and pressures. A computer program continuously calculated the liquid rate to the cyclone based on the measured liquid rate to the drain tank and the measured liquid carry-over rate from the test scrubber (see Figure 2). The separation eciency of the cyclone was also calculated on basis of this information as: cyclone = (1 Ql,co ) 100% (Ql,drain + Ql,co (7)

where Ql,drain is the liquid rate that is drained from the cyclones and Ql,co is the total liquid rate measured in the bulk scrubber, tangential cyclone and the lter coalescer. The separation eciency at the three pressures with a constant liquid ow to the cyclones of 45 l/hr per cyclone is plotted against the supercial gas velocity in the cyclone in Figure 10.
100

95

Efficiency [%]

90 N2/Exxsol 20 bar 50 bar 80 92 bar Natural gas 20 bar 75 50 bar 92 bar 70 0 1 2 3 4 5 6 7

85

Superficial gas velocity [m/s]

Fig. 10. The cyclone eciency when the liquid ow to the cyclones was kept constant at 45 l/hr per cyclone

For all cases, the eciency decreased with increasing gas ow. If the cause of liquid loss was due to limited capture eciency of droplets around the cut-size 28

of the cyclones, the eciency would be expected to increase with gas velocity in the cyclones, not decrease. This would indicate that liquid re-entrainment from the cyclones is the cause of liquid loss. For 20 and 50 bara, the performances for the two uid systems were much the same. When the pressure was increased to 92 bara, however, the eciency was radically lower for the natural gas uid, indicating that the dierent physical properties of the two uid systems is crucial for the performance. The dierences in physical properties are mainly in the interfacial tension, the liquid density and the liquid viscosity. In those cases where the natural gas case has slightly better eciency than the corresponding nitrogen/Exxsol case, the gas density of the natural gas is lower, which might cause the dierence.
2 The dynamic gas pressure, (1/2g vg ), increases if the supercial gas velocity is

kept constant while the pressure (and hence gas density) is increased. The gas ow capacity of the rig is related to the gas dynamic pressure and therefore the cyclones were generally tested at lower velocities at high pressure than at low pressure, as can be seen in the gure. 3 m/s supercial gas velocity was the only velocity that was tested at all three pressures. At this velocity, the K-values in the vessel were 0.11 and 0.26 m/s at 20 and 92 bara, respectively for the nitrogen/Exxsol, while it was 0.12 and 0.34 m/s for the corresponding natural gas experiments. In Figure 11, the same results are plotted against the K-value. It is clear from this plot that the K-value does not describe the performance of the cyclone deck well. We will discuss this issue further toward the end of this section. The amount of liquid inuences the performance and therefore another set 29

100

95

Efficiency [%]

90 N2/Exxsol 20 bar 85 50 bar 92 bar 80 Natural gas 20 bar 50 bar 92 bar 70

75

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

K-value in cyclone [m/s]

Fig. 11. The cyclone eciency for constant liquid ow of 45 l/hr per cyclone plotted against the K-value

of experiments were carried out wherein the liquid load was varied while the supercial gas velocity was kept constant at 3 m/s. The result is plotted in Figure 12.
100
N2/Exxsol 20 bar 50 bar 92 bar Natural gas 20 bar

95

Efficiency [%]

90

85

50 bar 92 bar

80

75

70 0 20 40 60 80 100 120 140

Liquid flow per cyclone [l/hr]

Fig. 12. The cyclone performance at varying liquid ows. The supercial gas velocity was constant, 3 m/s, for all experiments in the plot

In all cases except those below 20 l/hr, the eciency can be seen to drop o with increasing liquid ow. At the low liquid ows, a larger fraction of the liquid was observed to be distributed as a ne mist and the separation eciency 30

of small droplets may inuence the eciency in addition to re-entrainment. Also in these data, the eciency is much lower for natural gas than for nitrogen/Exxsol at 92 bara. Some results at high liquid loadings were corrupted due to capacity problems in the liquid drainpipe from the cyclone deck. These are not shown. The pressure drop measurements are shown in Figure 13 in the form of the Euler number, Eu = cyclone body.
10

p , 1/2v 2

where v is the supercial axial velocity in the

9.5

Euler number[-]

8.5

7.5 0 0.1 0.2 0.3 0.4 0.5 0.6

N2/Exxsol 20 bar 50 bar 92 bar Natural gas 20 bar 50 bar 92 bar

0.7 0.8

Liquid Concentration in Cyclone Inlet [vol%]

Fig. 13. The cyclone pressure drop as function of liquid concentration, or liquid loading, in the cyclone inlet

The general tendency is that the pressure drop is slightly larger in the nitrogen/Exxsol experiments than in the natural gas experiments. This is probably due to the higher liquid density in the Exxsol system, which makes it more energy consuming to carry the liquid through the cyclone. There is a weak tendency for the pressure drop to increase as the liquid load increases. Figure 11 shows the K-value to be unsuitable for design of the cyclone deck. 31

The same remarks made about the relevance of the K-value to the separation eciency ofand re-entrainment fromthe inlet vane and mesh pad apply to the cyclone deck. However, our investigation indicates that another design strategy may be suitable. Stepping back to the more general case, one can distinguish three strategies for designing cyclone demisters (or any inertial-impaction internal for demisting), based on scaling, each with their region of relevance: (1) Considering the separation eciency for small droplets close to the cyclone cut size. This is based on keeping the Stokes number constant: Stk = (l g )d2 vg 18g D

where D is the cyclone diameter and vg is some characteristic gas velocity, for instance the axial supercial velocity in the cyclone body. We cannot go into details about the physical signicance of this here, but only state that in a cyclone it is a measure for the movement of the particle relative to the gas in the centrifugal eld, and refer to reference Homann and Stein (2002) for more information. If l g and g do not vary much with pressure, designing for a constant Stokes number in a given cyclone implies keeping the gas velocity constant. Some cyclone vendors focus on this. (2) Considering the upward transport of re-entrained droplets from the cyclone installation against gravity. For this, as we have discussed, the Kvalue is the relevant design and scaling parameter. This strategy is relevant if re-entrainment limits separation eciency, and the re-entrained droplets are so large that their transportation in the scrubber against gravity, rather than the actual shearing o of liquid from the wall of the 32

cyclones, limits the degree of carry-over. (3) Considering the degree of re-entrainment from the cyclone installation. Assuming that re-entrainment is governed by the friction between the gas and a liquid lm on the cyclone wall, the frictional stress acting at the interface, i can be written in terms of the velocity of the gas owing over the lm, vg and the friction factor at the interface, fi : i = fi
2 g v g 2

(8)

where we have assumed that the liquid velocity is low compared to the gas velocity. If conditions are such that fi is constant, then operating the cyclonefor instance at varying pressuresfor constant re-entrainment
2 involves keeping the gas dynamic pressure, or twice its value, vg con-

stant. Such a criterion is handled, for instance, in Perry et al. (1997) and by Brunazzi et al. (2003), and it is recommended in the NORSOK
2 standard NORSOK (2001) that vg should be kept below 6000 Pa for

scrubbers containing an inlet vane. Such a strategy can be relevant if, for instance, re-entrainment limits separation eciency and re-entrained droplets, once formed, are easily carried with this gas from the cyclone installation due to their small size and/or the ow-conguration downstream of the cyclone installation. The dierent scaling rules thus focus on dierent considerations, and may lead to very dierent cyclone scaling. Consider an example: a cyclone has been tested in a laboratory at ambient conditions and was found to perform at its optimum when the supercial gas velocity was 10 m/s. How should the gas velocity be scaled with increasing pressure? Or said with others words: How many cyclones are required to operate in parallel in order to handle a certain 33

gas volume at a certain pressure? In Figure 14 the supercial gas velocity is plotted as function of the gas density for all the scaling rules mentioned above. As can be seen, scaling to keep the gas dynamic pressure constant and to keep the K-value constant results in approximately the same velocity, while Stokesian scaling to keep the cut size the same as under laboratory conditions results in a supercial gas velocity three times higher than in the lab when the gas reaches a density of 100 kg/m3 (approximately 100 bara operating pressure). Choosing Stokesian scaling over one of the other scaling rules may result in 60 times as high required gas velocity compared to scaling on basis of constant gas dynamic pressure or K-value.
100

Superficial velocity [m/s]

10

0.1 0 10

Const K-value (for the natural gas fluid) Const. gas dynamic pressure Const superficial velocity Const Stk (HP-rig fluid) 20 30 40 50 60 70 80 90 100

Gas

density[kg/m3]

Fig. 14. The gure shows how the supercial gas velocity changes with changing gas density when the dierent scaling methods are used. The bases for all the curves are a supercial gas velocity of 10 m/s at 1 kg/m3 gas density and 1.8105 kg/ms gas viscosity.

The rst of the three strategies mentioned above are for the case where cyclone eciency is limited by the escape of small droplets below the cyclones cut size. This is not the case in this present investigation and this strategy therefore is not relevant here. It is, however, the relevant strategy in the majority of 34

demisting and dedusting cyclone technology applications and we are therefore required to mention it for completeness of our discussion. Keeping the K-value constant implies that the gas velocity should be lowered with increasing pressure. However, the results in Figure 11 showed that the performance data were not brought onto one line when plotting them against the K-value. We may therefore infer that a better strategy in required. This only conrms what we suspected, since scaling on basis of the K-value is relevant for the settling of droplets against gravity, not for the separation of inertial separation equipment. The strategy of keeping the shear force, as given in Equation (8) constant at increasing pressure is probably a better way of scaling since the eciency seems to be determined by re-entrainment rather then insucient separation of small droplets. However, the method outlined above only accounts for changes in the gas density, while changes in the physical liquid properties in the liquid lm are not accounted for. The lower the density, viscosity and interfacial tension of the liquid in the lm, the easier the lm would be expected to be ruptured by a given shear force acting from the gas owing above it. Hence, a scaling method that accounts for both the changing shear force due to the ow of the gas and the changing liquid lm properties is required.

4.3 Total Scrubber Eciency

The total scrubber eciency for varying gas loads with 0.2 vol% liquid is shown in Figure 15. All experiments show eciencies higher than 85%. The trend is that increasing operating pressure is accompanied by decreasing eciency. 35

For instance, with natural gas at 92 bara, the liquid carry-over is an order of magnitude larger than the carry-over at 20 bara, at the same K-value of 0.26 m/s.
100 95

Efficiency [%]

90 85 80 75 70 0.05

N2/Exxsol

Natural gas

20 bar 50 bar 92 bar

20 bar 50 bar 92 bar

0.1

0.15

0.2

0.25

0.3

K-value [m/s]
Fig. 15. The total scrubber eciency with 0.2 vol liquid concentration in inlet pipe at varying gas loads

At low K-values (below mesh ooding velocities) the total scrubber eciency and the primary eciency in Figure 7 are much the same. This shows that the majority of the small droplets that penetrate the mesh pad also penetrate the cyclones. The total eciency of this standard scrubber conguration with K = 0.15 m/s, which is the maximum value recommended by the NORSOK-standard for this type of scrubber, is shown in Figure 16. At this K-value the eciency does not depend signicantly on liquid concentration. However, this behaviour should not be extrapolated to liquid concentration levels higher than the tested range, as other eects related to maximum liquid rate in cyclone deck can play an important role. The eciency at this K-value, is very inuenced by the fact that the operating conditions is near ooding conditions for the mesh pad. 36

Especially the low liquid concentrations in the gure are sensitive to small changes in the process and this is probably the explanation for the spread in the results for 92 bara nitrogen/Exxsol. At higher K-values the process conditions are more stable.

At the most severe condition tested, K = 0.26 m/s, the total scrubber eciency was also more or less independent of the liquid loading, except for the highest loading tested (1.0 vol%), which was probably due to the problems with draining tube capacity mentioned before, rather than a real eect. The eciency was about 85%, which, together with the earlier mentioned primary separation eciency under the same conditions of about 1/3, shows that the cyclones separate roughly 50% of the liquid under these conditions.

100 99 98

Efficiency [%]

97 96 95 94 93 92 91 90 0 0.2 0.4 0.6 0.8 1 1.2

N2/Exxsol Natural gas

20 bar 50 bar 92 bar

20 bar 50 bar 92 bar

Liquid concentration [vol%]

Fig. 16. The total scrubber eciency at the maximum K-value recommended by the NORSOK-standard (K = 0.15 m/s)

37

Conclusions

The atypical behaviour of the natural gas system below mesh ooding in Figure 7 is quite signicant for scrubber design. Figure 5 and Tab. 3 show that the interfacial tension of the natural gas system is considerably lower than that of nitrogen/Exxsol at 92 bara. It is possible that a signicant fraction of the liquid is in the form of very ne droplets, below the mesh cut-size of 15 m.

The results show that the Souders-Brown value is rather good in practise for design of inlet vane and mesh pad, and conrm the design criterion of K 0.15 m/s normally handled for scrubbers with these two types of internals. However, the results do show a signicant dierence in the separation for the two types of liquids used, and also a signicant inuence of the pressure (Figure 7). Furthermore, the K-value does not account for the eect of liquid loading as indicated by both the eciency and the pressure drop plots (Figures 8 and 9).

The cyclones perform very dierently at high pressure for the two uids used, even though they were both hydrocarbon-based. The data given in this paper show that the interfacial tensions dier very widely between the two uid systems at 92 bara pressure, they are around 16 mN/m for Exxsol and 2 mN/m for natural gas. This may well account for much of the dierence in performance. We note that the surface tension of water is around 72 mN/m. Figures 10, 11 and 14 illustrate that scaling rules for cyclone design are at present not sucient. 38

Acknowledgement

Financial support from the Research Council of Norway through the HiPGaS programme, and the industrial sponsors Statoil AS, ConocoPhillips, Norsk Hydro AS, Vetco, FMC Kongsberg Subsea and Aker-Kvrner is highly appreciated.

References Austrheim, T., Gjertsen, L. H., Homann, A. C., 2006. An experimental investigation of scrubber internals at conditions of low pressure. Submitted. Brunazzi, E., Paglianti, A., 1998. Design of wire mesh mist eliminators. AIChE Journal 44, 505512. Brunazzi, E., Paglianti, A., Talamelli, A., 2003. Simplied design of axial-ow cyclone mist eliminators. AIChE Journal 49, 4151. Brkholz, A., 1989. Droplet separation. VCH Publishers, iSBN 0-89573-879-1. u Chin, R. W., Standbridge, D. I., 2003. Development and installation of an inline deliquidiser. In: SPE Annual Technical Conference. No. SPE 84061. Gas Processors Suppliers Association (GPSA), 1998. Engineering data book. Gjertsen, L. H., Lkken, T. V., Marheim, N. M., Ophaug, J., 2003. Separation eciency of the troll kollsnes separators and the improvement in their performance. In: GPA 82.nd Annual Convention Presentations. New Orleans. Homann, A. C., Stein, L. E., 2002. Gas Cyclones and Swirl Tubes: Principles, Design and Operation. Springer Verlag, Berlin. NORSOK, 2001. Process Systems, P-100 Rev. 2, Norwegian Technology Centre. 39

Oranje, L., 1990. Cyclone-type separators score high in comparative tests. Oil & Gas Journal 88, 5459. Pedersen, K. S., Fredenslund, ., 1987. An improved corresponding states A model for the prediction of oil and gas viscosities and thermal conductivities. Chem. Eng. Sci. 42, 182186. Peneloux, A., Evelyne, R., 1982. A consistent correction of the Redlich-KwongSoave volumes. Fluid Phase Equilibria 8, 723. Perry, R. H., Green, D. W., Maloney, J. O., 1997. Perrys Chemical Engineers Handbook, seventh Edition. McGraw-Hill. Putnam, A., 1961. Integratable form of droplet drag coecient. ARS Journal 10, 14671468. Rawlins, H., Ting, F., 2002. Testing of an in-line rotary separator (iris) at the chevron f.ramirez gas production facility. In: 52nd Annual Laurance Reid gas conditioning conference. Sherwood, T. K., Shipley, G. H., Holloway, F. A. L., 1938. Flooding velocities in packed columns. Industrial & Engineering Chemistry 30, 768769. Soave, G., 1972. Equilibrium constant for a modied Redlich-Kwong equation of state. Chem. Eng. Sci. 27, 11971203. Souders, M., Brown, G. G., 1934. Design of fractionating columns. i. entrainment and capacity. Industrial & Engineering Chemistry 2, 98103. Verlaan, C. C. J., 1991. Performance of novel mist eliminators. Ph.D. thesis, Delft University of Technology, iSBN 90-370-0054-1. Weinaug, C. F., Katz, D. L., 1943. Surface tension of methane-propane mixtures. Industrial and Engineering Chemistry 35, 239246.

40