Beruflich Dokumente
Kultur Dokumente
TIANG HOCK YU
April 2019
ii
1 DECLARATION
I hereby declare that this project report is based on my original work except forcitations and
quotations which have been duly acknowledged. I also declare that it has not been previously
and concurrently submitted for any other degree or award at UTAR or other institutions.
Signature : _________________________
ID No : 15UEB05212
Date : 15/4/2019
iii
Approved by,
Signature : _________________________
Date : 15/4/2019
iv
ACKNOWLEDGEMENTS
I would like to express my appreciation and gratitude to my FYP supervisor, Dr. Yap Yeow
Hong for his patience and wisdom in guiding me throughout the period. Without his meticulous
guidance, this project would have been a lot more difficult and time consuming. Also, I would
like to thank my dearest brother who provided invaluable comments and support when I am
doing my research. Lastly, to my friends who helped and provided encouragement throughout
the preparation of this report.
v
ABSTRACT
Catalytic converters are common fixture in almost all motor vehicle on the road for their ability
to treat the exhaust emission. Modern research and developments on catalytic converter employ
the use of computer models to help simulate various aspects of catalytic combustion that take
place in the converters, as this will provide valuable insight to develop an efficient product.
Typically, such simulation will involve modelling the ignition-extinction phenomena as well
as predicting emission for drive cycle. In this project, a simulation based on Island Model will
be developed and simulated and this will be compared with other established kinetic models.
Simulation will be done on a 5mm thin slice Pt/γ-Al2O3 Diesel Oxidation Catalyst (DOC) and
benchmarked with the results by Ye. et. al. (2012). The transient 1D-gas phase, 1D-solid phase
single channel model considers the mass and energy transport as well as the change of the
adsorbed species on the catalyst surface. The simulation results have shown that the Herz and
Marin basic model and Salomons’ compressed oxygen model gave the best similarity in terms
of both conversion and hysteresis characteristics while the Island Model provided a good
agreement in terms of conversion but not hysteresis characteristics. Fine-tuning of the effective
diffusivity of the species of the Island Model is done and its effect on the hysteresis curve is
shown to be significant. Effective diffusivity and along with other kinetic parameters should
be optimized to achieve a better agreement between experimental and simulated results.
vi
TABLE OF CONTENTS
DECLARATION ii
ACKNOWLEDGEMENTS iv
ABSTRACT v
TABLE OF CONTENTS vi
LIST OF FIGURES ix
CHAPTER 1 1
INTRODUCTION
1.1 Background 1
1.2 Problem Statement 2
1.3 Aims and Objectives 2
1.4 Scope of Study 3
CHAPTER 2 4
LITERATURE REVIEW
2.1 Diesel Oxidation Catalysts (DOCs) 4
2.2 Hysteresis 6
2.3 Global and Mechanistic Kinetics 21
2.4 Modifications to the Basic Model 27
2.5 Adsorption and Desorption Kinetics 30
vii
CHAPTER 3 40
METHODOLOGY
3.1 Single Monolith Channel 40
3.2 Heat and Mass Transfer Equations 40
3.3 Adsorption and Desorption Kinetics 43
3.4 Reaction Models Investigated 45
3.5 Hydrocarbon Oxidation 49
3.6 Coding 50
CHAPTER 4 63
CHAPTER 5 95
APPENDICES 97
REFERENCES 97
viii
LIST OF TABLES
Table 4.1.1: Summary of Hysteresis Characteristics from the simulations done (Gas Inlet
Temperature) 68
Temperature at 3mm) 73
LIST OF FIGURES
Figure 2.1.2: Schematic Diagram of the Interior of the DOC (adapted from microscopy.cz) 5
Figure 2.2.2: A Series of Ignition and Extinction Curves Obtained by Varying Gas Inlet
Temperature in Cu/Al2O3 and Pt/Al2O3 Tubular Packed Bed Reactors. (Hlaváček and
Votruba, 1979) 8
Figure 2.2.3: Extinction Curves Due to Increasing CO Concentration of the Inlet Gas.(Wicke,
Figure 2.2.4: Hysteresis Is Still Attained When Heat of Reaction Is Set to 0. (Raj, Harold and
Balakotaiah, 2015) 10
Figure 2.2.5: Simulated Curves Using the Global Kinetics by Voltz et al. (Salomons et al.,
2007) 11
1980) 13
Figure 2.2.7: Development of Clusters Based on the Previous Scheme. (Wicke, Kummann
Figure 2.2.8: Ignition Curves Obtained by Varying Ramp Up Times (Ye et al., 2012) 15
x
Figure 2.2.9: Effect of ramp up rate on width of hysteresis. (Raj, Harold and Balakotaiah,
2015) 16
Figure 2.2.10: The effect of inlet CO concentration on the hysteresis phenomena and the rate
Figure 2.2.11: The effect of inlet CO concentration on the hysteresis phenomena and the
Figure 2.2.13: Psedo Steady State Experiments on Ignition Curves Produced by Different CO
Figure 2.2.14: Snapshots of the surface during reaction at different CO concentrations. (a)
Xeo = 0.39, near oxygen poisoning transition. (b) Xco = 0.50, in steady reaction region. (c)
Xco = 0.52, near CO poisoning. (black dots = O and white dots = CO) (Ehsasi et al., 1989) 19
Figure 2.2.15: Ignition Points as a Function of Pt:Pd Ratio. (Daneshvar et al., 2017) 20
Figure 2.3.1: 𝒓𝒄𝒐𝟐, and Surface Coverage of O and CO for a CO beam Impinging on an
Figure 2.5.2: Initial Sticking Coefficient as a Function to the Substrate Temperature. (Engel,
1978) 34
Figure 2.8.1: Stencil for the Explicit Method (adapted from Wikipedia.com). 37
Figure 2.7.2: Stencil for the Implicit Scheme (adapted from Wikipedia.com). 38
Figure 2.7.3: Stencil for the Crank-Nicholson Method (adapted from Wikipedia.com). 38
Figure 3.6.7: Coding for Dynamic Constants (Heat and Mass Transfer Coefficient) 53
Figure 3.6.15: Coding for Setting up the Heat and Mass Balances 58
Figure 3.6.16: Coding to close the model function and return the calculated differential values
59
Figure 3.6.20: Coding for the compilation of data of all stages and its storage. 61
Figure 4.1.1: Conversion against inlet temperature plot from the experimental results of Ye
Figure 4.1.2: Conversion against inlet temperature plot for Voltz Model 64
Figure 4.1.3: Conversion against inlet temperature plot for Herz and Marin Basic Model 65
xii
Figure 4.1.4: Conversion against inlet temperature plot for Salomons' Compressed Oxygen
Model 65
Figure 4.1.5: Conversion against inlet temperature plot for CO Island Model 66
Figure 4.1.6: Conversion against inlet temperature plot for O Island Model 66
Figure 4.1.7: Conversion against inlet temperature plot for CO-O Island Model 67
Figure 4.1.8: Conversion against inlet temperature plot for Dumont Postulation 67
Figure 4.2.1: Conversion against solid temperature plot from the experimental results of Ye
Figure 4.2.2: Conversion against solid temperature plot for Voltz Model 69
Figure 4.2.3: Conversion against solid temperature plot for Herz and Marin Basic Model 70
Figure 4.2.4: Conversion against solid temperature plot for Salomons' Compressed Oxygen
Model 70
Figure 4.2.5: Conversion against solid temperature plot for CO Island Model 71
Figure 4.2.6: Conversion against solid temperature plot for O Island Model 71
Figure 4.2.7: Conversion against solid temperature plot for CO-O Island Model 72
Figure 4.2.8: Conversion against solid temperature plot for Dumont Postulation 72
Figure 4.3.2: Temperature plot for Herz and Marin Basic Model 75
Figure 4.4.1: Fractional Coverage of CO against time for Herz and Marin Basic Model 78
Figure 4.4.2: Fractional Coverage of O against time for Herz and Marin Basic Model 79
xiii
Figure 4.4.3: Fractional Coverage of Vacant Sites against time for Herz and Marin Basic
Model 79
Figure 4.4.4: Fractional Coverage of CO against time for Salomons' Compressed Oxygen
Model 80
Figure 4.4.5: Fractional Coverage of O against time for Salomons' Compressed Oxygen
Model 80
Figure 4.4.6: Fractional Coverage of Compressed Oxygen against time for Salomons'
Figure 4.4.7: Fractional Coverage of Vacant Sites against time for Salomons' Compressed
Oxygen Model 81
Figure 4.4.10: Fractional Coverage of Vacant Site against time for CO Island Model 83
Figure 4.4.13: Fractional Coverage of Vacant Site against time for O Island Model 85
Figure 4.4.14: Fractional Coverage of CO against time for CO-O Island Model 86
Figure 4.4.15: Fractional Coverage of O against time for CO-O Island Model 86
Figure 4.4.16: Fractional Coverage of Vacant Sites against time for CO-O Island Model 87
Figure 4.4.19: Fractional Coverage of Vacant Sites against time for Dumont Postulation 89
xiv
LIST OF SYMBOLS/ABBREVIATIONS
𝑧 Channel Length 𝑚𝑚
𝐷 Channel Width (without washcoat) 𝑚𝑚
𝑊 Washcoat Thickness 𝑚𝑚
𝐶 Average Washcoat Loading 𝑔
𝐷𝑂𝐶
𝜌 Density of Washcoat 𝑘𝑔
𝑚
𝐶 , Heat Capacity of Washcoat 𝐽
𝑘𝑔𝐾
𝑘 Heat transfer coefficient of Washcoat 𝑊
𝑚𝐾
𝑆 BET Surface Area 𝑚
𝑔
𝐷 Average Pore Diameter 𝑛𝑚
𝜀 Porosity
𝜏 Tortuosity
𝛿 Constrictivity
𝑇 Surface temperature 𝐾
𝐷 , Diffusivity of species i 𝑐𝑚
𝑠
𝑟, Rate of adsorption of species i
𝐴 Pre-exponential factor 1
𝑠
xvi
𝑘 Rate constant for OO formation for the Compressed Oxygen Model 𝑚𝑜𝑙
𝑚 𝑠
𝑘 Rate constant for surface reaction of CO and OO for the Compressed 𝑚𝑜𝑙
Oxygen Model 𝑚 𝑠
CHAPTER 1
1 INTRODUCTION
1.1 Background
Diesel engines still drive most of the heavy-duty vehicles because it can produce more power
per unit of fuel used when compared with its gasoline counterpart. Moreover, diesel fuel
requires less processing cost and are generally cheaper. Lastly, the average lifetime of a diesel
engine is almost double of a gasoline engine.
However, diesel engines emits much more soot, unburnt hydrocarbons and carbon
monoxide. This is due to the heavy nature of the diesel fuels which makes incomplete
combustion a bigger possibility. Moreover, the fuel is injected just before the power stroke
causing incomplete combustion due to the lack of oxygen. These products of incomplete
combustion are harmful to both humans and the environment.
As a result, regulations were drafted and implemented by the government to control
emission of harmful products by fossil fuelled vehicles and engines. Therefore, the pressure is
building on the automobile manufacturers to reduce the emission from diesel vehicles. The
most popular and cost effective ways is to introduce a catalytic converter at the exhaust outlet.
These converters are able to convert products of incomplete combustion into less harmful
substances which are water and carbon dioxide. The improvement of catalytic converters or
diesel oxidation catalysts (DOCs) are of a huge interest over the past few decades.
2
Chapter 3: Methodology
Equations for heat and mass transfer
The models that are selected and their respective formulations and equations
Kinetic parameters that are selected.
Method used to solve the monolith system
Application of Spyder in solving the monolith system
Fine tuning strategies
Chapter 5
Conclusion of research
Recommendation for the improvement of the simulation
4
CHAPTER 2
1 LITERATURE REVIEW
DOC (Diesel Oxidation Catalysts) is one of the many parts which makes up the complex
emission control system of a diesel powered vehicle. They usually come in the form of
monolithic catalytic reactors, which is a form of packed bed reactor which consists of many
long thin hollow channels. It is made up of cordierite/γ-Al3O2 which acts as supports the
catalysts and noble metals such as platinum and palladium. A plethora of improvements and
innovations were made to the catalyst composition, where 2 or 3 metals are used as the catalyst
(Nibbelke et al., 1997).
5
Most DOCs are based on ceramic “honeycomb” monoliths even though sometimes metallic
monoliths can be seen in the market. Ceramic monoliths are better in withstanding thermal
expansion. However, its low porosity makes it undesirable for catalyst dispersion. A remedy
to this is to introduce a layer of porous material on to the surface of the monolith channel which
is normally known as washcoat.
Figure 2.1.2: Schematic Diagram of the Interior of the DOC (adapted from
microscopy.cz)
6
Washcoat acts as a catalyst support, improving its surface area, and thermal stability.
Alumina is a very common washcoat material as it fulfills all the criteria for a good support.
Alumina is porous, providing more surface area of reaction and also being thermally stable
even at high temperatures. Other supports such as zeolites and zirconia are also used. Zeolites
generally improve the cold start performance of the DOC. It aids by trapping the unburnt
hydrocarbons at lower temperatures where oxidation is not possible. When the desired
temperature is reached, the hydrocarbons are then released and oxidized by the catalyst.
Zirconia also lowers the ignition temperature of the DOC.(Russell and Epling, 2011)
2.2 Hysteresis
DOC and they showed that hysteresis is a phenomena which is mostly related to the reaction
mechanism and is an intrinsic behavior rather than a result of heat and mass limitations. (Ye et
al., 2012). Hysteresis can be observed by either changing the gas inlet temperature or inlet CO
concentration.
Figure 2.2.2: A Series of Ignition and Extinction Curves Obtained by Varying Gas Inlet
Temperature in Cu/Al2O3 and Pt/Al2O3 Tubular Packed Bed Reactors. (Hlaváček and
Votruba, 1979)
In general, there are two types of hysteresis: steady state and dynamic. (Dadi, Luss and
Balakotaiah, 2016)
Steady state hysteresis is well documented in literatures ever since the 1960s.(Razon
and Schmitz, 1986) Researchers noticed that CO oxidation can be at two stable steady states:
upper and lower, categorized based on the rate of reaction that they are having. Instead of using
hysteresis, they used the term steady state multiplicity. Most attributed this to the mathematical
nature of the heat and mass transfer solutions, which will be covered in detail on the next
section.
Dynamic hysteresis became a point of interest when researchers are investigating the
performance of catalytic converters. They noticed that during a cold-start, the conversion of
CO is low. However, when a certain inlet gas temperature is achieved, the ignition temperature,
conversion rises drastically, reaching to its maximum conversion shortly after. Then, when the
engine slows down, causing a drop in inlet gas temperature, the rate of reaction will not follow
the same path as before. Instead, the conversion will still remain until a certain temperature is
reached, the extinction temperature, conversion will often plummet or drop off at a fast rate.
The conversion will eventually plateau off at a low conversion state.
Notice that how both hysteresis are quite similar as two different states are achieved
(low and high conversion). The main difference is that steady state hysteresis is observed
during steady state experiments where the system is given sufficient time to reach steady state
while dynamic hysteresis is observed when the system is not in steady state or in a transient
state where inlet gas temperature or inlet CO concentration is varied with time and the readings
are taken instantaneously.
One of the earliest reasons that are proposed by researchers is the heat and mass transfer
limitations within the reactor. They looked into the partial differential equations which governs
the heat and mass transfer of the reactor and found out that there are certain mathematical
criteria (usually involves the external parameters of the reaction) that is needed so that state
multiplicity or hysteresis to occur. (Hlaváček and Votruba, 1979)
10
Such analysis, stability analysis, dominated the literature on this subject during 1960s
and 1970s. Weisz and Hicks (1962) first mentioned the possibility of multiple steady states in
their classic paper regarding heat and mass transfer of packed bed reactors. Their works then
provided a good foundations for researchers later on. Aris, Schimtz and Ray then provided
papers that further refines the theory.(Hlaváček and Votruba, 1979)
Fast forward to this decade, in recent studies published by the researchers of the
University of Houston, they performed simulation and it provides computational evidence that
hysteresis occur due to the delay in temperatures between the feed inlet temperature and solid
(catalyst) temperature. (Raj, Harold and Balakotaiah, 2015; Dadi, Luss and Balakotaiah, 2016)
They established that the cause of the commonly observed extinction curve is due to the heat
transfer limitations within the reactor not the exothermic nature of the reaction itself. In their
simulation, they set the heat of reaction as zero and still observe the hysteresis phenomena.
(Raj, Harold and Balakotaiah, 2015; Dadi, Luss and Balakotaiah, 2016) Raj et al explained that
when decreasing the inlet gas feed, conversion is still maintained because of the heat transfer
limitation between the monolith and the inlet gas. The surface temperature is still high enough
for reactions to occur. Extinction will eventually occur when surface temperature is low enough.
Figure 2.2.4: Hysteresis Is Still Attained When Heat of Reaction Is Set to 0. (Raj,
Harold and Balakotaiah, 2015)
However, they did not try to fit their simulations with any available ignition-extinction
curve that is obtained from experiment. Salomons et al (2007) attempted that and stated that
11
using global kinetics by Voltz could not provide a good fit for the extinction curve. They
proposed the use of mechanistic models to model hysteresis instead. (Salomons et al., 2007)
Figure 2.2.5: Simulated Curves Using the Global Kinetics by Voltz et al. (Salomons et
al., 2007)
Hysteresis is still being reported on experiments that are performed with the elimination of
mass transfer limitations, and isothermal conditions or on Pt wires, where heat and mass
transfer limitations is nonexistent.(Eigenberger, 1978; Razon and Schmitz, 1986) Therefore, it
is important that we take into consideration the effect of each elementary step has.
It is long known that CO adsorption is not inhibited by adsorbed oxygen atoms but not
vice versa. Also, at low temperatures, CO is the dominant species that is adsorbed while O2
requires temperature at the excess of 200℃ to adsorb. Finally, O2 desorption is insignificant at
temperatures below 300℃. These behaviors could justify the hysteresis phenomena.
At low gas inlet temperatures, CO is the primary species that is adsorbed but O2
adsorption is not expected because CO poisons and inhibits O2 adsorption. As the temperature
increases, CO starts to desorb, creating vacancies for O2 to adsorb. This would lead to a LH
12
mechanism reaction and CO2 is released, creating more vacancies for O2 to adsorb. At high
temperatures, the adsorption of O2 is more favorable, causing O atoms to be the dominant
species on the surface.
However, adsorbed O species does not inhibit the adsorption of CO. CO is able to
squeeze in between the spaces of the adsorbed O atoms and hence compress the O atoms around
it. This effect of compression will lower the O-O bond strength, causing the rate of reaction to
be higher.(Engel et al., 1979) All of these factors would contribute to the acceleration of the
rate of reaction causing an ignition phenomena where at certain temperature (the ignition
temperature) a sudden increase in conversion can be seen.
When the gas inlet temperature is reduced, the conversion of CO will maintain at its
maximum until a temperature way lower than the light-off temperature. This is because of the
exothermic nature of the oxidation reaction and also the high temperature favored O2
adsorption and CO desorption. The exothermic nature of the reaction enabled the surface
temperature to be sufficiently high for both CO desorption and O2 adsorption to occur and
hence maintaining the rate of reaction. (Casapu et al., 2017)
The formation of CO and possibly O islands are reported by various literatures. Engel and Ertl
(1979) first provided possibility of it by introducing the concept of cooperative adsorption.
Wicke et. al. (1980) demonstrated through computer simulation that the island
formation is possible under the assumption that both CO and O2 have the same probability to
adsorb, and the surface reaction is a LH mechanism. The left figure of Figure 2.2.7 shows the
14
initial distribution of two species which is even and random. The right figure shows the
formation of clusters (or islands) after 104 reaction steps.
Wicke et al (1980) provided a detailed explanation on how islands of CO will cause the
light-off phenomena. As clusters of CO islands grow because of the low temperature favored
CO adsorption, the chemisorbed bonding between CO and Pt active sites will decrease. This is
a well-known effect that can be proven through calorimetric and equilibrium measurements,
and also infrared spectroscopy. This is due to the increased competition of the CO molecules
for the electrons of the Pt metal. As a result, it is easier for CO molecules that is attached to an
island to desorb than the isolated ones.
The increase in the island size and also reduced reaction rate of CO oxidation causes
the rate of CO desorption to increase.(Wicke, Kummann and Schiefler, 1980) It will then reach
a point where O2 adsorption becomes more favorable due to the increasing availability of two
vacant adjacent sites or the temperature is high enough such that O2 adsorption is high along
with rate of CO desorption. The latter can also happen in experiments where CO concentration
is varied due to the exothermic nature of the reaction. As rate of oxidation increases, the
temperature of the surface rises, which then leads to ignition as O2 adsorption and CO
desorption increases.
Haaland and Williams (1982) also mentioned that the increasing CO island size due to
increased CO adsorption would cause a drop in rate of reaction during extinction as the
available species for reaction at the domains are largely reduced and also the inhibition effect
of CO for O2 adsorption.
The width of hysteresis, and its ignition and extinction point is affected by a number of factors
such as ramp up and cool down rates of inlet gas, composition and flow rate of inlet gas, size
of Pt nanoparticles, washcoat diffusivity and last but not least, the degradation or aging of
catalysts.
Increasing the ramp up rate will result in higher ignition points as it is apparent from both
Figure 2.2.8 and Figure 2.2.9. (Ye et al., 2012; Raj, Harold and Balakotaiah, 2015)
Figure 2.2.8: Ignition Curves Obtained by Varying Ramp Up Times (Ye et al., 2012)
16
Figure 2.2.9: Effect of ramp up rate on width of hysteresis. (Raj, Harold and
Balakotaiah, 2015)
This can be explained through heat transfer limitations within the monoliths. As ramp
up rate increases, the rate of temperature change of the inlet gas increases too, However, this
effect does not translate into an equal rise in rate of change of the catalyst surface temperature.
As a result, a delay can be seen in reaching ignition point. For example, when the temperature
of inlet gas is at 200℃, the surface temperature might still be at 130℃. (Raj, Harold and
Balakotaiah, 2015)
From Figure 2.2.8, it can be seen that as ramp up rate is decreased, ignition point
decreases. Also, if ramp up rate is slow enough, we could attain steady state at every point of
the measurement. Low ramp up rates will actually allow time for heat transfer to take place and
hence the delay in between gas inlet temperature and surface temperature would be less
significant.
The cooling rate would also have some significant effect on the extinction point. As
cooling rate is increased, the extinction point decreases. Again this is due to the heat transfer
limitation of within the monolith itself. The gas inlet temperature would be a lot cooler than
the surface temperature.
The effect of ramp up and cooling rates have no effect on the real ignition and extinction
points where it is dependent on the surface temperature. The temperatures obtained due to heat
transfer limitations are actually the apparent ignition and extinction points. Therefore, to
improve the true ignition-extinction characteristic of the catalysts, this would be a non-factor.
17
Figure 2.2.10: The effect of inlet CO concentration on the hysteresis phenomena and the
rate of oxidation of CO (Ye et al., 2012)
Figure 2.2.11: The effect of inlet CO concentration on the hysteresis phenomena and the
conversion of CO (Ye et al., 2012)
18
Hysteresis is still observed at different CO concentrations although the ignition and extinction
points did not vary significantly. From Figure 2.2.11, the conversion of the DOC shows a slight
decrease when the inlet CO concentration is reeduced while it is vice versa for the rate of
oxidation of CO.
However, looking into Figure 2.2.12, increasing the CO concentration shifts the
hysteresis curve to the left where a significantly higher ignition-extinction points are obtained.
This is different from what Ye et al (2012) observed in their experiments (difference in
conversion and reaction rates). The main reason behind this difference is the ramp up rates used
by both researchers. Salomons et al used a lower ramp up rate when compared with Ye et al
(0.133K/s vs 1.556 K/s). Ye et al described that using low ramp up rates would result in the
experiment being at a pseudo steady state rather than transient state.
19
Indeed, looking at Ye et. al.’s pseudo steady state results from Figure 2.2.13, we can
confirm that the shift produced in Salomon’s experiment is performed at a pseudo steady state
condition since both results give identical conversion (both at 100%). Increasing CO
concentration during a steady state condtion will result in more CO surface coverage
demonstrated by both experiments and simulation which is evident in Figure 2.2.14. (Ehsasi et
al., 1989) As a result, a higher temperature is needed so that sufficient amount of CO is
desorbed for ignition to occur.
c) Monolith properties
The properties of the monolith, especially washcoat properties, are rather significant on the
hysteresis behavior of the DOC. Studies have been conducted on the effect of substrate material
on the hysteresis phenomena. They found out that the ignition point would be lower if a metallic
monolith is used. (Gundlapally and Balakotaiah, 2013)
The use of bimetallic catalyst would affect the ignition point. The actual mechanism
involved is very complex and it is beyond the scope of this project. Nevertheless, we will
include the results of the simulation conducted by Daneshvar et. al. (2017) and give a brief
explanation on it.
Figure 2.2.15: Ignition Points as a Function of Pt:Pd Ratio. (Daneshvar et al., 2017)
Table 2.2.1: Properties of the bimetallic Pt-Pd catalyst. (Daneshvar et al., 2017)
Pt/Pd Pt Pd Total Pt loading Pd loading Dispersion Particle size
Mol ratio Wt% Wt% Wt% g/ft3 g/ft3 % nm
Pt 1 0 1 27.75 0 5.8 19.4
3:1 0.75 0.13 0.88 20.81 3.78 15.6 7.2
1:1 0.5 0.27 0.77 13.87 7.57 21.7 5.2
1:3 0.25 0.41 0.66 6.93 11.35 23.5 4.8
Pd 0 0.54 0.54 0 15.13 26.4 4.3
Global kinetics lumped every step into one single equation by making several
assumptions.(Salomons et al., 2007) It usually takes into account the gas phase concentration
species. Global kinetics do not take into account the surface concentration of the adsorbed
species. Two groups of researchers went on and suggested simplifications of the basic
elementary steps with the aid of experimental evidence. The two proposed mechanisms are as
below:
1. Langmuir-Hinshelwood
The mechanism proposed by Langmuir involves only adsorbed CO and O2:
𝐶 𝐶
𝑟 , =
1+𝐾 𝐶 +𝐾 𝐶
Equation 2.3.1
22
2. Eley-Rideal
This mechanism proposed that adsorbed oxygen atoms reacts with an incoming gas
phase CO.
𝐶𝑂 + 𝜃 → 𝐶𝑂. 𝜃 Reaction 1.1
𝐶𝑂 + 𝑂. 𝜃 → 𝐶𝑂 + 𝜃 Reaction 1.4
Figure 2.3.1: 𝒓𝒄𝒐𝟐 , and Surface Coverage of O and CO for a CO beam Impinging on an
Oxygen Saturated Surface on Pd (111) at T=374K.(Engel and Ertl, 1978)
The limitations of this simplified global model is due to two of its assumptions. First,
it assumed CO adsorption to be at equilibrium. Second, the surface reaction is the rate limiting
23
step. The first assumption is wrong when we are dealing dynamic hysteresis modelling, where
there is no equilibrium state when a system is transient. Also, there is no experimental evidence
which supports the assumption of equilibrium CO adsorption. The second assumption is also
erroneous as it is proven that surface reaction between the adsorbed species is not the rate
limiting step.(Herz and Marin, 1980).
Mechanistic kinetics, however, takes into account every elementary steps and hence the surface
concentration of species adsorbed. Also, assumptions made when formulating such kinetics are
kept as minimal as possible. The model by Herz and Marin (1980) is one of the earliest works
which uses this approach. In their paper, they argued that the LHWM model is not a correct
representation of this catalytic reaction. The reasons are covered in the previous section. Their
model produced a much better fit than the LH rate law. (Herz and Marin, 1980)
The biggest downfall of this method is that we need to obtain a lot of kinetic parameters
and the calculation could be resource intensive if not very tedious to formulate and key into the
program. Mechanistic calculations are often simplified using appropriate assumptions.
Salomons et. al’s simulation is one of the many simulations done in this way and often provide
satisfactory agreement with the experimental results.(Salomons et al., 2007)
With the introduction of better software capable in handling he tedious and complex
mathematics such as FOTRAN and DETCHEM. Researchers such as Chatterjee et al (2001)
are able to perform more rigorous methods to model CO oxidation. They included the rate laws
of all possible steps when simulating the reaction. A total of 61 pathways are included with
every parameter for the reaction obtained from numerous literatures.
They are able to obtain very good agreements with the experimental data at lean,
stoichiometric and rich conditions but propane oxidation at rich conditions does not fit very
well. They attributed that to the existence of greater variety of species that reside on the surface
24
of the catalyst which can reduce the oxygen coverage. This can be improved by taking into
account the products of partial oxidation of propane. (Chatterjee, Deutschmann and Warnatz,
2001)
Another new approach is to use the Monte Carlo simulation method to replace the
adsorption-desorption model, which takes into account of the average concentration of the
adsorbed species, with a two dimensional surface which is able to take into account of non-
uniform nature of the adsorbed species on the surface. (Ziff, Gulari and Barshad, 1986)
However, this method requires even more computing power when compared with the usual
method. So, we will not attempt such simulation in this project.
The exact mechanistic mechanism and rate law of CO oxidation is highly debated over the past
few decades. The basic elementary steps are:
𝐶𝑂 + 𝜃 → 𝐶𝑂. 𝜃 Reaction
12.1
𝑂 + 2𝜃 ⎯ 𝑂 . 2𝜃 Reaction 2.2
In their classical paper, Herz and Marin introduced a series of assumptions which would
later be known as the Oxygen Exclusion Model throughout numerous literatures. (Salomons et
al., 2007)
25
(1) An adsorbed CO excludes other adsorbed CO from an area on the surface equivalent in size
to the area occupied by one surface Pt atom.
(2) An adsorbed O excludes other adsorbed CO from an area on the surface equivalent in size
to the area occupied by two surface Pt atom.
(3) An adsorbed CO excludes other adsorbed O from an area on the surface equivalent in size
to the area occupied by one surface Pt atom.
(4) An adsorbed O excludes other adsorbed O from an area on the surface equivalent in size to
the area occupied by one surface Pt atom.
The first two assumptions are made based on the limiting values of surface concentrations
of CO and O being 1.0 and 0.5 respectively. The latter assumptions agree qualitatively with the
absence of inhibition by an O adlayer on CO adsorption. As a result, the amount of vacant sites
available for CO adsorption will be:
𝜃 , =1−𝜃 −𝜃
Equation 2.3.2
𝜃 , = 1 − 𝑓𝜃 − 2𝜃
Equation 2.3.3
1 − 2𝜃
𝑓=
1−𝜃
Equation 2.3.4
Most literatures before this one derived rate laws for adsorption based on the surface
concentration of either CO or O but not both. This could be the reason why they find it difficult
to fit the LH model with the experimental data.
Herz and Marin then started their postulation with the basic elementary steps that is presented
earlier:
𝐶𝑂 + 𝜃 → 𝐶𝑂. 𝜃 Reaction
12.1
26
𝑂 + 2𝜃 ⎯ 𝑂 . 2𝜃 Reaction 2.5
They are interested in reactions that occur under 573K. At this temperature, oxidation
and reduction of metal oxides by oxygen can be excluded for platinum and palladium catalyst.
Moreover, association of adsorbed oxygen is very slow, which results:
𝑂 . 2𝜃 ⎯ 2𝑂 . 𝜃 → 𝑂 . 2𝜃 2𝑂 . 𝜃
They also assumed that oxygen is in adsorption equilibrium since sticking coefficient is low
on platinum. This also results in another assumption that the rate of dissociation of O2 is very
slow when compared with the rate of desorption. Moreover, surface concentration of O2* is
assumed to be negligible since it is weakly adsorbed. As a result, oxygen adsorption and
dissociation can be assumed as 1 elementary step as O* does not need to be determined
explicitly. Then in the end,
𝐶𝑂 + 𝜃 → 𝐶𝑂. 𝜃 Reaction
12.1
𝑂 + 2𝜃 → 2𝑂. 𝜃 Reaction 2.5
𝑟 , =𝑘 𝜃 𝜃
Equation 2.3.5
The surface coverage of CO and O needed to be calculated from the following system of
ordinary differential equations:
𝜃̇ =𝑘 𝐶 𝜃 , −𝑘 𝜃 −𝑘 𝜃 𝜃
Equation 2.3.6
27
𝜃̇ = 𝑘 𝐶 𝜃 , −𝑘 𝜃 𝜃
Equation 2.3.7
𝜃 , =1−𝜃 −𝜃
Equation 2.3.8
𝜃 , = 1 − 𝑓𝜃 − 2𝜃
Equation 2.3.9
1=𝜃 , +𝜃 , +𝜃 +𝜃
Equation 2.3.10
This set of reactions and rate laws will then provide the basis for researchers later on who try
to explain on the other two phenomena which are known ever since CO oxidation experiments
are conducted: state multiplicity/hysteresis and periodic oscillations. Razon and Schmitz (1986)
described the model by Herz and Marin as the “basic model” which would then be tinkered
around to explain the aforementioned phenomena.
Herz and Marin (1980) dismissed the possibility of island formation due to the small size of
their Pt crystals (in the magnitude of nanometers). They do consider the potential effects of
boundaries of islands where islands of CO and O2 is present and reaction only occurs on the
boundaries. However it is not included since in their experiment the analyzed area is small
enough so that the effect is diminished.
Mukesh et al (1984) formulated a set of mathematical models which takes into account of the
effect of CO islands. They pointed out that the presence of CO islands (and potentially O
islands) will limit the rate of reaction. It is believed that reactions only occur on the island
boundaries. To take into CO island boundaries, they used the following rate law:
𝑟 , ∝𝜃 𝜃
Equation 2.4.1
28
The square root term of 𝜃 is a mathematical treatment to the island boundaries. They also
assumed that there will be no O within the CO islands and vice versa. Already we can see that
this assumption is flawed as there is evidence that CO are able to adsorb into the O adlayer or
in this case O island. (Engel et al., 1979) In the end, they concluded that CO islands affects the
rate more significantly.(Mukesh et al., 1984)
However, the presence of CO islands will affect the rate of adsorption and desorption as
well. Graham and Lynch (1987) provided new rate equations that takes into account of such
effects. Then, they came up with a two models. The first model is a simple island model while
the second model takes into account the effects of increased local concentration due to island
formation. Then, it is concluded that the second model is the better because it fits the
experimental data better when compared with the other two models.
Graham and Lynch (1987) modified the rate law for adsorption and desorption of CO as:
𝑟 , ∝ 𝑎 [𝐶𝑂](1 − 𝜃 −𝜃 )
Equation 2.4.2
𝑟 , ∝𝑎 𝜃
Equation 2.4.3
𝑚𝜃 𝑚𝜃
𝑟 , =𝑘 , 𝐶 (1 − 𝜃 −𝜃 )−𝑘 , 𝜃
𝑉𝑌 𝑉𝑌
Equation 2.4.4
The modified rate law for oxygen adsorption and desorption would be:
𝑎 [𝑂 ](1 − 𝜃 − 𝜃 )
𝑟 , ∝
(1 − 𝜃 )
Equation 2.4.5
𝑎 𝜃
𝑟 , ∝
(1 − 𝜃 )
Equation 2.4.6
29
𝑚𝜃 (1 − 𝜃 − 𝜃 ) 𝑚𝜃 𝜃
𝑟 , =𝑘 , 𝐶 −𝑘 ,
𝑉𝑌 (1 − 𝜃 ) 𝑉𝑌 (1 − 𝜃 )
Equation 2.4.7
𝑚𝜃 𝜃
𝑟 , , =𝑘 , 𝜃
𝑉𝑌 (1 − 𝜃 )
Equation 2.4.8
𝑚𝜃 (1 − 𝜃 − 𝜃 ) 𝑚𝜃
𝑟 , =𝑘 , 𝐶 −𝑘 , 𝜃
𝑉𝑌 (1 − 𝜃 ) 𝑉𝑌
Equation 2.4.9
𝑚𝜃 𝜃
𝑟 , , =𝑘 , 𝜃
𝑉𝑌 (1 − 𝜃 )
Equation 2.4.10
𝑚𝜃 𝑚𝜃
𝑟 , =𝑘 , 𝐶 (1 − 𝜃 −𝜃 ) −𝑘 , 𝜃
𝑉𝑌 𝑉𝑌
Equation 2.4.11
𝑚𝜃
𝑟 , , =𝑘 , 𝜃 𝜃
𝑉𝑌
Equation 2.4.12
30
The above equation is investigated and refined by Dumont et. al. and the following is
formulated as the best fit equation to the experiment (Dumont, Poriaux and Dagonnier, 1986):
𝑚𝜃
𝑟 , =𝑘 , 𝐶 (1 − 𝜃 −𝜃 )
𝑉𝑌
Equation 2.4.13
𝑚𝜃
𝑟 , , =𝑘 , 𝜃 (1 − 𝜃 )𝜃 (1 − 𝜃 )
𝑉𝑌
Equation 2.4.14
The dimensionless kinetic parameters reported in this paper is temperature independent.
Therefore, it would not be suitable for our simulation since we are dealing with changing inlet
temperature rather than changing the inlet CO concentration as done by the researchers. So, we
need to import kinetic parameters from other models which varies with temperature.
Another key component in simulating an accurate hysteresis phenomena is the adsorption and
desorption kinetics of the model. It is key that we use the most appropriate and accurate kinetics.
31
a) CO Adsorption Kinetics
The rate of adsorption of a species onto a solid surface that is formulated by Langmuir is given
as (Langmuir, 1922):
𝑑𝑛 𝑝
𝑟, = = ∙ 𝑠(𝜃)
𝑑𝑡 2𝜋𝑀 𝑘𝑇
Equation 2.5.1
where 𝑛 is the density of the adsorption species per cm2, 𝑝 is the gas pressure of the species,
𝑚 the absorbate mass, k the Boltzmann constant, T the absolute temperature, and 𝑠(𝜃)the
probability of adsorption or simply the sticking coefficient, which is the function of surface
concentration.
The first term represents the gas flux on the surface of the solid while the second term
describes the kinetics of the adsorption. The above rate equation is based on the assumption
that desorption rate is negligible which is valid if only the surface temperature is low enough.
The initial sticking coefficient, s0, can be determined experimentally using molecular beam
experiments which can yield accurate results if s0 is close to 1, which is true for Pt surfaces.
Sticking coefficient is independent on surface temperature, between 300K and 650K, and also
the angle of incidence of the gas flux (Engel et al., 1979). Also, it is shown experimentally that
s0 is also independent on the atomic oxygen surface coverage (Gland et al., 1983) . However,
McCabe and Schmidt (1977) showed experimentally that s0 would vary between 0.24 and 0.95
when they varied the angle of incidence. Therefore, generally:
𝑠 = 𝑓(𝜃 )
𝑠(𝜃) = 𝑠 (1 − 𝜃 )
Equation 2.5.2
𝑠(𝜃) = 𝑠 (1 − 𝜃 )
Equation 2.5.3
32
where Equation 2.5.2 is used if only one free site is needed and Equation 2.5.3 is used if two
adjacent free sites are required.
However, sticking coefficient of CO are constant and decreases at a lesser rate at lower
coverage, evident from Figure 2.5.1.
The precursor state model is used to explain this behavior. This model tells us that a
molecule may diffuse into a layer below the chemisorbed species and it will stay there until a
vacant adjacent site is present so that it can chemisorb. The following is the sticking coefficient
according to the precursor state model (Kisliuk, 1957):
(1 − 𝜃)
𝑠(𝜃) = 𝑠
(1 − 𝑐𝜃)
Equation 2.5.4
,
𝑓 +𝑓 −𝑓
𝑐=
𝑓 +𝑓
Equation 2.5.5
33
where,
For c = 0, the Langmuir model is obtained. Experiments were carried out to obtain the
parameter, c, and satisfactory results were obtained where c = 0.85 for Pt (111). Yet, the model
does not agree quantitatively well with the experiments as mentioned by Engel et al (1979)
b) CO Desorption Kinetics
𝑑𝑛 −𝐸 (𝑛 )
𝑟, =− = 𝑣 (𝑛 ) exp ∙𝑛
𝑑𝑡 𝑘𝑇
Equation 2.5.6
𝐸 (𝑛 ) = 𝐸 − 𝛽𝜃
Equation 2.5.7
c) O2 Adsorption Kinetics
𝑑𝑛 𝑝
𝑟 = = ∙ 𝑠(𝜃)
𝑑𝑡 √2𝜋𝑚𝑘𝑇
Equation 2.5.1
34
However, the sticking coefficient does not follow the precursor state model. It decreases
exponentially with O2 coverage. Moreover, various experiments suggested that O2 adsorption
increases with temperature unlike the adsorption of CO. Also from Figure 2.5.2, initial sticking
coefficient decreases linearly with temperature.(Engel, 1978)
Bonzel and Ku (1973) developed the following model for the sticking coefficient:
−(𝐸 + 𝛽𝜃)
𝑠 = 𝛿𝑓(𝜃)𝑒𝑥𝑝
𝑅𝑇
Equation 2.5.8
where,
Some years later, Hopster, Ibach and Comsa (1977) developed the following model:
. )
−(3.2𝑒𝑉𝜃
𝑠 = 0.018𝑒𝑥𝑝
𝑘𝑇
Equation 2.5.9
where k is the Boltzmann constant.
This model is developed by taking into account the effects of electrostatic potential on
activation energy which then predicts the sign of effect, the exponential dependence, and
approximately, the exponent.(Hopster, Ibach and Comsa, 1977).
d) O2 desorption
O2 desorption is generally not observed at temperatures that a DOC will experience. (Herz
and Marin, 1980) Therefore, it will not be reviewed in this project.
However, simultaneous tuning of parameters to obtain the optimal fit is beyond the
scope of this project. Therefore, tuning the effective diffusivity through diffusivity constants
would be the strategy adopted.
The Python Programming Language is chosen for this task. The following are the advantages
of using Python:
For this project, we will use the Scientific Python Development Environment to compile
our codes. It is an integrated development environment where we can compile, debug and run
our program in it. Moreover, it has the necessary libraries such as NumPy and MatPlotLib that
we can use to perform our simulation.
The solutions to the partial differential equations of heat and mass transfer cannot be obtained
analytically. Therefore, we need to employ numerical methods to solve for our PDEs and ODEs.
The method we will be using is the method of lines. Method of lines generally refers to the re-
expression of PDEs as a system of ODEs. The spatial derivative of the PDE can be replaced
with algebraic expressions. As a result, we would obtain a system of ODEs with time as its
only derivative. (Hamdi et al, 2007) Derivatives can be approximated into algebraic
expressions using the finite difference method. This method can be categorized into three
methods: explicit, implicit, and Crank-Nicholson method. If the boundary conditions are
37
specified the solution of every recurrence equation can be obtained and through numerical
iterations we can obtain the whole complete solution albeit an approximated one.
Figure 2.9.1: Stencil for the Explicit Method (adapted from Wikipedia.com).
This method uses the forward difference for the time derivative at time tn and a central
difference for the space derivative at xi (forward-time-central space scheme or just simply
FTCS). A recurrence equation can be derived for a 1D heat equation
𝑈 = 𝛼𝑈
Equation 2.9.1
𝑢 −𝑢
𝑈 =
𝑘
Equation 2.9.2
𝑈 =
Equation 2.9.3
𝑢 −𝑢 𝑢 − 2𝑢 + 𝑢
=𝛼
𝑘 ℎ
Equation 2.9.4
There exist a stability criteria for explicit method to work.(Chapra and Canale, 2010) The
criteria is given as:
𝛼𝑘 1
≤
ℎ 2
Equation 2.9.5
38
Figure 2.9.2: Stencil for the Implicit Scheme (adapted from Wikipedia.com).
This method utilizes the BTCS scheme where the solution to the point j,n+1 is obtained
implicitly with the following recurrence equation:
𝑢 −𝑢 𝑢 − 2𝑢 +𝑢
=
𝑘 ℎ
Equation 2.9.6
There will be no stability condition for this method unlike the explicit method. However,
accuracy is sometimes greatly sacrificed.(Chapra and Canale, 2010)
Figure 2.9.3: Stencil for the Crank-Nicholson Method (adapted from Wikipedia.com).
This method employs a central difference for the time derivative therefore, a CTCS scheme.
The recurrence equation is:
𝑢 −𝑢 1 𝑢 − 2𝑢 +𝑢 𝑢 − 2𝑢 + 𝑢
= +
𝑘 2 ℎ ℎ
Equation 2.9.7
39
This method is unconditionally stable with an even higher accuracy when compared with the
implicit method.(Chapra and Canale, 2010)
Explicit method is chosen due to its ease of translation into Python coding. The ODEINT
function uses an explicit method by default. The stability criteria will be addressed and its
details will be covered in the methodology section.
The data from Ye et. al.’s experiment on thin slice DOCs with an inlet CO concentration of
3000ppm is chosen. One of the reason being that it requires a lot less computing power and
time to perform numerical integration of the transient partial differential equation. Thin slice
DOCs do exhibit hysteresis behavior which is the key of this project. (Ye et al., 2012) Only
one data is chosen since up this time of report writing, no other researchers have performed
experiments on thin slice DOCs.
40
CHAPTER 3
METHODOLOGY
A single channel can represent the whole channel by assuming uniform inlet gas velocity and
catalyst distribution, and no heat loss along the radius of the channel. The boundaries of the
simulation would be from the centerline of the channel up to half of the substrate thickness.
The use of the 1D-2D axi-symmetric model is seen for most literatures, with gas phase
heat and mass transfer being 1 dimensional and heat and mass transfer through washcoat 2
dimensional. Experimental evidence, which justifies the use of 1D modelling for gas phase
diffusion, shows that radial gradient at gas phase is non-existent. On the other hand, most would
agree on the significance of concentration gradient that exist on the washcoat layers so that a
2D model is required. (Mukadi and Hayes, 2002) However, due to the difficulty in translating
the 2D model into a working code, simplifications are made and a 1D model is used for a
washcoat instead.
Figure 3.2.1
41
𝑣 𝑅
𝐷 , =𝐷 +
48𝐷
Equation 3.2.1
𝜕 𝑣 𝐶, 𝜕 𝑣 𝐶,
= 𝑣 𝐶, 𝑎𝑛𝑑 =0
𝜕𝑧 𝜕𝑧
𝜕𝑇 𝜕 𝜕𝑇 𝜕𝑇 4
𝜌 𝐶 = 𝐾 , −𝜌 𝐶 𝑣 − 𝑘 𝑇, − 𝑇
𝜕𝑡 𝜕𝑧 𝜕𝑧 𝜕𝑧 𝐷
Equation 3.2.2
𝑣 𝑅
𝐾 , =𝛼 +
48𝛼
Equation 3.2.3
The boundary conditions are:
𝜕𝑇 𝜕𝑇
= 0 𝑎𝑛𝑑 =0
𝜕𝑧 𝜕𝑧
𝜕𝐶 , 𝑊 𝜕 𝜕𝑠
𝜀 = 𝑘 , 𝐶, −𝐶, + 𝐷 ,, + 𝜉𝑅
𝜕𝑡 𝐴 𝜕𝑧 𝜕𝑧
Equation 3.2.4
42
𝜕 𝐶, 𝜕 𝐶,
= 0 𝑎𝑛𝑑 =0
𝜕𝑧 𝜕𝑧
𝜕𝑇 𝑊 𝜕 𝜕𝑇
𝜌 𝐶 = 𝑘 𝑇, − 𝑇 + 𝐾 , + (∆𝐻𝑅)
𝜕𝑡 𝐴 𝜕𝑧 𝜕𝑧
Equation 3.2.5
𝜕 𝑇 𝜕(𝑇 )
= 0 𝑎𝑛𝑑 =0
𝜕𝑧 𝜕𝑧
The Nusselt and Sherwood numbers are utilized to obtain the heat and mass transfer
coefficients. Their values are set to equal 4.0 since fully developed flows are assumed with the
conditions of constant wall temperature and wall flux.
𝑘 𝐷
𝑁𝑢 = = 4.0
𝑘
𝑘 𝐷
𝑆ℎ = = 4.0
𝐷
Knudsen diffusion mechanics are the dominant diffusion mechanism in the reactor washcoat.
.
𝜀 𝑇
𝐷 , = 97𝑎
𝜏 𝑀
Equation 3.2.6
43
a) CO Adsorption Kinetics
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.3.1
where,
b) CO Desorption Kinetics
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation 3.3.2
where,
c) O2 Adsorption Kinetics
1000𝑅𝑇 𝑃𝑌 , 1 − 2𝜃
𝑟 , =2 𝜎 𝑆 1− 𝜃 − 2𝜃
2𝜋𝑀 𝑅𝑇 1−𝜃
Equation 3.3.3
where,
In this project, a number of global and mechanistic model were simulated and investigated.
Voltz et al (1973) developed the global kinetic rate equation for catalytic CO oxidation.
𝑌 𝑌
𝑟 , =𝑘
𝑅(𝜃)
Equation 3.4.1
𝑅(𝜃) is a term which takes into account of the inhibition effects of CO, C3H6, and NO on the
reaction rates.
.
𝑅(𝜃) = 1 + 𝑘 𝑌 +𝑘 𝑌 × 1+𝑘 𝑌 𝑌 × [1 + 𝑘 𝑌 ]
Equation 3.4.2
The first term takes into account of the inhibition effect of CO and C 3H6 while the second term
is needed to fit the experimental data at higher concentrations of CO and C3H6.(Voltz et al.,
1973) The last term can be ignored since it takes into account of the effect of NO which is not
considered in this project.
𝐶𝑂 + 𝜃 → 𝐶𝑂. 𝜃 Reaction
12.1
𝑂 + 2𝜃 → 2𝑂. 𝜃 Reaction 2.5
𝐶𝑂. 𝜃 + 𝑂. 𝜃 → 𝐶𝑂 + 2𝜃 Reaction
2.13
The adsorption and desorption kinetics are mentioned in section 3.3. The rate of surface
reaction is given by:
1000(𝐸 − 𝛽𝜃 )
𝑟 , , = 𝑘 exp − 𝜃 𝜃
𝑅𝑇
Equation 3.4.3
A small alteration to the activation energy is made such that it will decrease with increasing
surface coverage of CO. The involved parameters, 𝛽, will be taken from Salomons’ work.
In this project, a modification of the Islands models was made. The rate constants, k, capacity
factor, Z, and gas phase concentration will be replaced with the rate constants, collision
frequency and sticking coefficient from Herz and Marin (1980). Also, oxygen desorption is
assumed to be negligible. The following equations are proposed:
CO islands
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.4.4
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation 3.4.5
1000𝑅𝑇 𝑃𝑌 , (1 − 𝜃 − 𝜃 )
𝑟 , , = 𝜎 𝑆
2𝜋𝑀 𝑅𝑇 (1 − 𝜃 )
Equation 3.4.6
47
1000(𝐸 − 𝛽𝜃 ) 𝜃
𝑟 , = 𝑘 exp − 𝜃
𝑅𝑇 (1 − 𝜃 )
Equation 3.4.7
O islands
,
𝑟 , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
Equation 3.4.4
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation 3.4.8
1000𝑅𝑇 𝑃𝑌 , (1 − 𝜃 − 𝜃 )
𝑟 , , = 𝜎 𝑆
2𝜋𝑀 𝑅𝑇 (1 − 𝜃 )
Equation 3.4.9
1000(𝐸 − 𝛽𝜃 ) 𝜃
𝑟 , = 𝑘 exp − 𝜃
𝑅𝑇 (1 − 𝜃 )
Equation 3.4.10
CO and O Islands
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.4.4
48
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation 3.4.11
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.4.12
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 exp − 𝜃 𝜃
𝑅𝑇
Equation 3.4.13
Dumont Postulation
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.4.4
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation 3.4.14
1000𝑅𝑇 𝑃𝑌 ,
𝑟 , , = 𝜎 𝑆 (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
Equation 3.4.15
1000(𝐸 − 𝛽𝜃 )
𝑟 , = 𝑘 exp − 𝜃 (1 − 𝜃 )𝜃 (1 − 𝜃 )
𝑅𝑇
Equation 3.4.16
The same elementary steps are used from the Herz and Marin basic model with an addition of
elementary steps accounting for the compressed oxygen effect:
𝑟 =𝑘 𝜃 𝜃
Equation 3.4.18
𝑟 =𝑘 𝜃 𝜃 ,
Equation 3.4.19
The rate constants for Equation 3.4.17, Equation 3.4.18, and Equation 3.4.19 follows the
Arrhenius Equation(Salomons, 2008):
𝐸
𝑘 = 𝐴 exp
𝑅 𝑇
Equation 3.4.20
𝑌 𝑌
𝑟 =𝑘
𝑅(𝜃)
Equation 3.5.1
𝑅(𝜃) is a term which takes into account of the inhibition effects of CO, C3H6, and NO on the
reaction rates.
50
.
𝑅(𝜃) = 1 + 𝑘 𝑌 +𝑘 𝑌 × 1+𝑘 𝑌 𝑌 × [1 + 𝑘 𝑌 ]
Equation 3.4.2
3.6 Coding
Step 1: Discretisation of Monolith Channel
The monolith channel is divided sections of 1mm each. Therefore, a 5mm slice would be
divided into 6 sections: 0mm, 1mm, 2mm, 3mm, 4mm, and 5mm.
Step 2: Defining Monolith Properties and Characteristics and also other Constants
The monolith properties are defined from the data provided by Ye et. al (2012).
Constants such as ideal gas constant and molecular weight are also defined at this stage.
Variables are stated. 4 types of variables are needed: gas phase concentration denoted by Y,
gas phase concentration within the solid washcoat denoted by Ys, fractional coverage denoted
by fc and temperature denoted by T. Also, a letter is used as the name of each species to simplify
the code.
Letter Species
A Carbon Monoxide
B Carbon Dioxide
C Propane
Kinetic parameters that are dependent on physical properties of the system are defined.
Figure 3.6.7: Coding for Dynamic Constants (Heat and Mass Transfer Coefficient)
54
Python is able to plot the temperature profile by just identifying the end points of each stage.
There will be six different ramp up and ramp down rates in this experiment. Six end points of
the stages are identified: 417K, 505K, 545K, 545K, 485K, 445K, and 417K.
A model is created which returns the value of the differentials. The details on the mathematics
behind it is in appendix.
The differential variable for the fractional coverage of each species will be declared later on.
57
Figure 3.6.15: Coding for Setting up the Heat and Mass Balances
59
Finally, the differential value calculated will be returned as a single array, D. This is when the
boundary condition is defined.
Figure 3.6.16: Coding to close the model function and return the calculated differential
values
Step 7: Setting up the ODE solver and storing the calculated values
An ODE solver is set up by using the defined models and initial values of the variables are
stated. The heat and mass balance equations are solved in this step.
A two dimensional array is created, first axis having the size of the maximum length of
the stage and the second axis having the size of the returned array from the odeint function. Its
function is to store all the computed values for a stage so that it can be called back for graph
plotting and to be exported to a spreadsheet. In this case, it will be sol_grid.
Another array will be created for the same purpose but this array will store all the
computed values for all the six stages. It will have another dimension with the size as the
number of stages. In this case, it will be sol_grid_2.
60
A loop is then created so that each iteration will solve a particular stage
The loop starts off by calculating the temperature slope of the stage. This slope will be
named as mp. The purpose of this slope is to be the boundary condition of our inlet gas and
solid temperature as indicated at the previous step.
Next, an array called ‘time’ is created based on the time length of each stage and each
with the number of intervals calculated as:
𝑠𝑡𝑎𝑔𝑒 𝑙𝑒𝑛𝑔𝑡ℎ
𝑑𝑒𝑠𝑖𝑔𝑛𝑎𝑡𝑒𝑑 𝑡𝑖𝑚𝑒 𝑖𝑛𝑡𝑒𝑟𝑣𝑎𝑙
Then, another loop is then created so that the odeint will solve the model from t = 0s to
t = 1s, instead of the conventional method whereby most would solve from t = 0s to t = 200s.
This method will allow us to avoid the stability issues that would be eminent in our case as the
differential in length is small. Having a small differential length means we need to have a
61
smaller differential in time to fulfill the stability criteria. So, instead of solving for a big
timeframe i.e. from 0 to 200s, we need to solve it at a smaller timeframe i.e. from 0 to 1s. The
disadvantage of this method is that we need more time to compute the solution as we need to
solve for 1800 time steps for our case (total time involved is 1800s).
The initial condition will be declared as y0. Then, the solution will be stored in sol_grid
array. The solution will also be stored in the y0 array which will then be used as the initial point
of the next solution. After that, sol_grid will be resized such that the excess parts of the array
will be removed.
Finally, with the pd.DataFrame function, we can export the array into a spreadsheet to
save the computed data externally. Also, the sol_grid will be stored into sol_grid_2. Then, the
sol_grid array needs to be reset at the end of this loop.
When the loop has finished every calculation, sol_grid_2 will be converted into a two
dimensional array and exported as a spreadsheet and graph plotting will be initiated.
Figure 3.6.21: Coding for the compilation of data of all stages and its storage.
62
The required graphs are plotted by using the plotting function and are exported as a PNG image
file. An array ‘timespace’ is created to become the x axis of certain plots.
CHAPTER 4
Figure 4.1.1: CO Conversion against inlet temperature plot from the experimental
results of Ye et. al.(Ye et al., 2012)
Figure 4.1.2: CO Conversion against inlet temperature plot for Voltz Model
V
65
Figure 4.1.3: CO Conversion against inlet temperature plot for Herz and Marin Basic
Model
Figure 4.1.4: CO Conversion against inlet temperature plot for Salomons' Compressed
Oxygen Model
66
Figure 4.1.5: CO Conversion against inlet temperature plot for CO Island Model
Figure 4.1.6: CO Conversion against inlet temperature plot for O Island Model
67
Figure 4.1.7: CO Conversion against inlet temperature plot for CO-O Island Model
Figure 4.1.8: CO Conversion against inlet temperature plot for Dumont Postulation
68
Generally, for all models other than the Dumont Postulation, agreement with the experimental
results on the conversion of CO is within the range of ±5%. However, only the Herz and Marin
Basic Model, and Salomons’ Compressed Oxygen Model showed good fit to the overall
hysteresis curve. Meanwhile, both O Island Model and CO-O Island Model showed hysteresis
albeit the width of hysteresis is very small (around 10℃ or less). Finally, Voltz Model, CO
Island Model and Dumont Postulation failed to show hysteresis.
Table 4.1.1: Summary of Hysteresis Characteristics from the simulations done (Gas
Inlet Temperature)
Figure 4.2.1: CO Conversion against solid temperature plot from the experimental
results of Ye et. al.(2012)
Figure 4.2.2: CO Conversion against solid temperature plot for Voltz Model
70
Figure 4.2.3: CO Conversion against solid temperature plot for Herz and Marin Basic
Model
Figure 4.2.4: CO Conversion against solid temperature plot for Salomons' Compressed
Oxygen Model
71
Figure 4.2.5: CO Conversion against solid temperature plot for CO Island Model
Figure 4.2.6: CO Conversion against solid temperature plot for O Island Model
72
Figure 4.2.7: CO Conversion against solid temperature plot for CO-O Island Model
Figure 4.2.8: CO Conversion against solid temperature plot for Dumont Postulation
73
Generally, the same observations and conclusions can be made with this plot. However, it can
be seen that the simulated result gives a solid ignition temperature 20 to 40℃ lower than the
experimental value. This could mean that heat transfer between the gas and solid washcoat is
not simulated accurately in our models. More evidence can be found when we look at the
temperature plot in the next section.
The change in temperature profile of the inlet, outlet, and 1mm, 2mm, 3mm of the solid is
plotted against time and will provide us key information about the accuracy of the simulation.
Figure 4.3.3: Temperature plot for Herz and Marin Basic Model
Generally, the outlet temperature is lower or almost the same as the inlet temperature for all
simulated models. This is not true as catalytic combustion is an exothermic reaction, so the gas
coming out of the monolith should be a lot hotter than the inlet. This is evident from Figure
3.6.9. Hence, the temperature plots exposed a flaw in this simulation which is poorly simulated
heat and mass transfer characteristics. Better correlations needed to be chosen instead.
Moreover, the fact that the solid wall temperature does not agree well with the
experimental value (20 to 40℃ lower), implies that maybe the heat of reaction of the surface
reaction should be higher than the one used in the simulation (-282.55 kJ/mol) . Further
research needs to be done in selecting better parameters.
Figure 4.4.1: Fractional Coverage of CO against time for Herz and Marin Basic Model
79
Figure 4.4.2: Fractional Coverage of O against time for Herz and Marin Basic Model
Figure 4.4.3: Fractional Coverage of Vacant Sites against time for Herz and Marin Basic
Model
80
Figure 4.4.5: Fractional Coverage of O against time for Salomons' Compressed Oxygen
Model
81
Figure 4.4.6: Fractional Coverage of Compressed Oxygen against time for Salomons'
Compressed Oxygen Model
Figure 4.4.7: Fractional Coverage of Vacant Sites against time for Salomons'
Compressed Oxygen Model
82
Figure 4.4.10: Fractional Coverage of Vacant Site against time for CO Island Model
84
Figure 4.4.13: Fractional Coverage of Vacant Site against time for O Island Model
86
Figure 4.4.14: Fractional Coverage of CO against time for CO-O Island Model
Figure 4.4.15: Fractional Coverage of O against time for CO-O Island Model
87
Figure 4.4.16: Fractional Coverage of Vacant Sites against time for CO-O Island Model
88
Figure 4.4.19: Fractional Coverage of Vacant Sites against time for Dumont Postulation
90
The reason why Dumont Postulation failed to give the expected conversion values are clear
when we look at Figure 4.4.17 where CO coverage is 1.0 at throughout the simulation. This
would let us draw the following possibilities:
The kinetic parameters can be fine-tuned to provide a better fit with the hysteresis curve.
The fractional coverage behaviour in the O Island Model and CO-O Island Model tells us
that ignition is not happening in the same manner observed in the fractional coverage of the
CO and O in both Herz and Marin Basic Model and Salomons’ Compressed Oxygen Model.
First of all, the time at which ignition occur is different for the models. For the lesser models
(O Island Model and CO-O Island Model), it is apparent from Figure 4.4.11 and Figure 4.4.14
that ignition is happening at around t = 200s while for the greater models (Herz and Marin
Basic Model and Salomons’ Compressed Oxygen Model), it is happening at around t = 400s
based on Figure 4.4.1 and Figure 4.4.4. However, one similarity between the models is the
minimum CO coverage attained in the process. For the lesser models, 0.7 is the minimum
coverage while for the greater models, 0.65. A strong hypothesis that we can draw from this
observation is that coverage of CO and O atoms can tell us the information about the state of
conversion of the models since all models reported similar maximum conversion and this is
correlated with similar maximum/minimum CO and O coverage. A possible suggestion to
improve the hysteresis behaviour of the lesser models is to try to delay the ignition by tuning
the kinetic parameters of the elementary steps such as pre-exponential factors, activation
energies and effective diffusivity of the molecules(Lundberg et al., 2016).
Also, looking into the fractional coverage of compressed oxygen in Figure 4.4.6, we
realised that only 0.008% is covered by compressed oxygen. This indicates that only a small
amount of CO will adsorb on sites that is preoccupied with an O atom. However, considerable
amounts of CO should be able to adsorb onto the O ad layer.(Engel et al., 1979; Mongeot et
al., 1998) Improvements can still be made in Salomons’ Compressed Oxygen Model by tuning
the parameters for Equation 3.4.17, Equation 3.4.18, and Equation 3.4.19.
91
Figure 4.5.1: Conversion Plot of CO-O Island Model after increasing kd by 10 times
92
Figure 4.5.2: Conversion Plot of CO-O Island Model after increasing kd by 2 times
Figure 4.5.3: Conversion Plot of CO-O Island Model after reducing kd by 10 times
93
Figure 4.5.4: Conversion Plot of CO-O Island Model after reducing kd by half
As observed in Figure 4.5.1 and Figure 4.5.2, increasing the bulk diffusion coefficient constant
of the species leads to the model behaving more like the Voltz Model. In other words, the
model is showing the characteristic curves of a global kinetic model. Similarities between
Figure 4.5.1 and Figure 4.5.2, and Figure 4.5.5 can be observed.
94
Figure 4.5.5: Simualtion done by Daneshvar et. al. using Global Kinetic Model
(Daneshvar et al., 2017)
On the other hand, reducing the diffusivity constant of the species would result in a
more pronounced hysteresis character as evident in Figure 4.5.3 and Figure 4.5.4. These
observations suggest that hysteresis of the conversion is not only dependent on the mechanics
of the reaction, it is also strongly influenced by the washcoat properties. Further research needs
to be done on this subject.
95
CHAPTER 5
5.1 Conclusion
In this project, simulation of CO oxidation in a thin slice (5mm) DOC has been performed.
Voltz Model, Herz and Marin Basic Model, Salomons, Compressed Oxygen Model, CO
Island Model, O Island Model, CO-O Island Model, and Dumont Postulation are investigated.
The following conclusions are drawn:
Herz and Marin Basic Model and Salomons’ Compressed Oxygen Model gives the
best results with the latter being marginally better in terms of ignition and extinction
temperature.
The Island Models (with an exception of Dumont Postulation) gives good fit for
conversion but cannot simulate a good hysteresis curve.
The temperature profile of the models does not match with experimental data. This
might be due to poor heat and mass transfer modelling or incorrect value of heat of
reaction for catalytic combustion chosen.
Increasing the diffusivity constant of the species will lead the CO-O Island Model to
behave like a global model i.e. no hysteresis is present while reducing it will result in
hysteresis. This leads us to hypothesise that hysteresis might be due to the mass
transfer limitation of the washcoat.
96
5.2 Recommendation
Below are several recommendations that can be implemented in future research.
To use other programs such as ANSYS and DETCHEM where the CFD is used and a
better approximation could be achieved with models of higher dimensions.
To fine tune effective diffusivity of the species.(Lundberg et al., 2015)
To implement the use of mechanistic kinetics for propylene oxidation.
Robustness and stability of the better models (Herz and Marin Basic Model, and
Salomons’ Compressed Oxygen Model) can be tested using experimental data from
other researchers.
97
REFERENCES
Bonzel, H. P. and Ku, R. (1973) ‘On the kinetics of oxygen adsorption on a Pt(111) surface’,
Surface Science, 40(1), pp. 85–101. doi: 10.1016/0039-6028(73)90053-8.
Casapu, M. et al. (2017) ‘Origin of the Normal and Inverse Hysteresis Behavior during CO
Oxidation over Pt/Al2O3’, ACS Catalysis, 7(1), pp. 343–355. doi: 10.1021/acscatal.6b02709.
Chapra, S. C. and Canale, R. P. (2010) Numerical Methods for Engineers. 7th edn. McGraw-
Hill Education.
Dadi, R. K., Luss, D. and Balakotaiah, V. (2016) ‘Dynamic hysteresis in monolith reactors
and hysteresis effects during co-oxidation of CO and C2H6’, Chemical Engineering Journal.
Elsevier B.V., 297, pp. 325–340. doi: 10.1016/j.cej.2016.03.139.
Daneshvar, K. et al. (2017) ‘Experimental and modeling study of CO and hydrocarbons light-
off on various Pt-Pd/Γ-Al2O3diesel oxidation catalysts’, Chemical Engineering Journal.
Elsevier B.V., 323, pp. 347–360. doi: 10.1016/j.cej.2017.04.078.
Dumont, M., Poriaux, M. and Dagonnier, R. (1986) ‘ON SURFACE REACTION KINETICS
IN THE PRESENCE OF ISLANDS’, Surface Science, 169(2–3), pp. 307–310. doi:
10.1016/0039-6028(86)90598-4.
Engel, T. (1978) ‘A molecular beam investigation of He, CO, and O2 scattering from
Pd(111)’, The Journal of Chemical Physics, 69(1), p. 373. doi: 10.1063/1.436363.
Engel, T. et al. (1979) ‘Elementary Steps in the Catalytic Oxidation of Carbon Monoxide on
Platinum Metals’, Advances in Catalysis, Volume 28, pp. 1–78. Available at:
http://www.sciencedirect.com/science/article/B7CS3-4S875SR-
6/2/84e0400997c8fec81806548661ee3d5b.
Gland, J. L. et al. (1983) ‘Carbon monoxide oxidation on the Pt ( 111 ) surface : Temperature
programmed reaction of coadsorbed atomic oxygen and carbon monoxide Carbon monoxide
oxidation on the Pt ( 111 ) surface : Temperature programmed reaction of coadsorbed atomic
oxygen and car’, 963(111). doi: 10.1063/1.444801.
Hlaváček, V. and Votruba, J. (1979) Hysteresis and Periodic Activity Behavior in Catalytic
Chemical Reaction Systems, Advance in Catalysis. doi: 10.1016/S0360-0564(08)60054-1.
99
Hopster, H., Ibach, H. and Comsa, G. (1977) ‘Catalytic oxidation of carbon monoxide on
stepped platinum(111) surfaces’, Journal of Catalysis, 46(1), pp. 37–48. doi: 10.1016/0021-
9517(77)90133-6.
Kisliuk, P. J. (1957) ‘The sticking probability of gases chemisorbed on the surfaces of solids’,
J.Phys.Chem.Solids, 3, pp. 95–101.
Langmuir, I. (1922) ‘The mechanism of the catalytic action of platinum in the reactions 2Co
+ O2= 2Co2 and 2H2+ O2= 2H2O’, Transactions of the Faraday Society, 17, p. 621. doi:
10.1039/tf9221700621.
Lundberg, B. et al. (2016) ‘DOC modeling combining kinetics and mass transfer using inert
washcoat layers’, Applied Catalysis B: Environmental. Elsevier B.V., 191, pp. 116–129. doi:
10.1016/j.apcatb.2016.03.024.
Mukadi, L. S. and Hayes, R. E. (2002) ‘Modelling the three-way catalytic converter with
mechanistic kinetics using the Newton – Krylov method on a parallel computer’, 26, pp. 439–
455.
Mukesh, D. et al. (1984) ‘Island models and the catalytic oxidation of carbon monoxide and
carbon monoxide-olefin mixtures d. mukesh’, 138, pp. 237–257.
Nibbelke, R. H. et al. (1997) ‘Kinetic Study of the CO Oxidation over Pt / γ -Al 2 O 3 and Pt
/ Rh / CeO 2 / γ -Al 2 O 3 in the Presence of H 2 O and CO 2’, 373, pp. 358–373.
Raj, R., Harold, M. P. and Balakotaiah, V. (2015) ‘Steady-state and dynamic hysteresis
effects during lean co-oxidation of CO and C 3 H 6 over Pt / Al 2 O 3 monolithic catalyst’,
CHEMICAL ENGINEERING JOURNAL. Elsevier B.V., 281, pp. 322–333. doi:
100
10.1016/j.cej.2015.06.057.
Razon, L. F. and Schmitz, R. A. (1986) ‘Intrinsically Unstable Behavior during the Oxidation
of Carbon Monoxide on Platinum’, Catalysis Reviews, 28(1), pp. 89–164. doi:
10.1080/03602458608068086.
Russell, A. and Epling, W. S. (2011) ‘Diesel oxidation catalysts’, Catalysis Reviews - Science
and Engineering, 53(4), pp. 337–423. doi: 10.1080/01614940.2011.596429.
Salomons, S. et al. (2007) ‘On the use of mechanistic CO oxidation models with a platinum
monolith catalyst’, Applied Catalysis B: Environmental, 70(1–4), pp. 305–313. doi:
10.1016/j.apcatb.2006.01.022.
Sjoblom, J., Ab, S. and Creaser, D. (2014) ‘Parameter Estimation of a DOC from Engine Rig
Experiments with a Discretized Catalyst Washcoat Model Björn Westerberg’. doi:
10.4271/2014-01-9049.
Voltz, S. E. et al. (1973) ‘Kinetic Study of Carbon Monoxide and Propylene Oxidation on
Platinum Catalysts’, Industrial & Engineering Chemistry Product Research and
Development, 12(4), pp. 294–301. doi: 10.1021/i360048a006.
Wicke, E., Kummann, P. and Schiefler, J. (1980) ‘Unstable and Oscillatory Behaviour in
Heterogeneous Catalysis’, 797(1962), pp. 315–323.
Ye, S. et al. (2012) ‘Catalyst “light-off” experiments on a diesel oxidation catalyst connected
to a diesel engine-Methodology and techniques’, Chemical Engineering Research and
Design. Institution of Chemical Engineers, 90(6), pp. 834–845. doi:
10.1016/j.cherd.2011.10.003.
Ziff, R. M., Gulari, E. and Barshad, Y. (1986) ‘Kinetic phase transitions in an irreversible
surface-reaction model’, Physical Review Letters, 56(24), pp. 2553–2556. doi:
10.1103/PhysRevLett.56.2553.
101
APPENDICES
For CO:
𝜕𝐶 , 𝜕 𝜕𝐶 , 𝜕 𝑣𝐶 , 4
= 𝐷 , , − − 𝑘 , 𝐶 , −𝐶 ,
𝜕𝑡 𝜕𝑧 𝜕𝑧 𝜕𝑧 𝐷
Equation A.1
The following assumptions are made:
𝐷 , , and 𝑣 are assumed to be constant throughout the channel length. This assumption
is valid since we are considering a very small channel length where the decrease in both
velocity and dispersion coefficient is negligible.
𝜕𝐶 , 𝜕 𝐶 , 𝜕 𝐶 , 4
=𝐷 , , −𝑣 − 𝑘 , 𝐶 , −𝐶 ,
𝜕𝑡 𝜕𝑧 𝜕𝑧 𝐷
Equation A.2
𝜕𝐶 , 𝐶 ,, −𝐶 ,,
=
𝜕𝑡 𝑑𝑡
Equation A.3
102
𝜕 𝐶 , 𝐶 , , − 2𝐶 , , + 𝐶 ,,
=
𝜕𝑧 𝑑𝑧
Equation A.4
𝜕𝐶 , 𝐶 , ,−𝐶 ,,
=
𝜕𝑧 𝑑𝑧
Equation A.5
where,
Then substituting these into Equation A.2, the PDE is then transformed into:
,, ,, , , ,, ,,
=𝐷 , ,
, , ,,
−𝑣
− 𝑘 , 𝐶 , −𝐶 ,
Equation A.6
Since the concentration of CO is given as ppm in the experiment, it is necessary to change the
units of concentration in the equation as well. Equation A.6 is divided by the total concentration
of the gas phase.
,, ,, , , ,, ,,
=𝐷 , ,
, , ,,
−𝑣
− 𝑘 , 𝐶 , −𝐶 ,
103
This would lead to the mass balance equation to be in terms of mol fraction, the unit of ppm.
,, ,, , , ,, ,,
=𝐷 , ,
, , ,,
−𝑣
− 𝑘 , 𝑌 , −𝑌 ,
Equation A.7
O2 balance
,, ,, , , ,, ,,
=𝐷 , ,
, , ,,
−𝑣
− 𝑘 , 𝑌 , −𝑌 ,
Equation A.8
CO2 balance
,, ,, , , ,, ,,
=𝐷 , ,
, , ,,
−𝑣
− 𝑘 , 𝑌 , −𝑌 ,
Equation A.9
104
𝜕𝐶 , 𝑊 𝜕 𝜕𝐶 ,
𝜀 = 𝑘 , 𝐶 , −𝐶 , + 𝐷 , + 𝜉𝑅
𝜕𝑡 𝐴 𝜕𝑧 𝜕𝑧
Equation A.10
The following assumptions are made:
𝜕𝐶 , 𝑊 𝜕 𝐶 ,
𝜀 = 𝑘 , 𝐶 , −𝐶 , +𝐷 , , + 𝜉𝑅
𝜕𝑡 𝐴 𝜕𝑧
Equation A.11
The reaction term of CO is the net change in CO due to adsorption and desorption.
𝜉𝑅 = −𝑟 , +𝑟 ,
Equation A.12
Generally, the reaction term is expressed as:
−𝑟 , +𝑟 ,
1000𝑅𝑇 1000(𝐸 − 𝛽𝜃 )
=− 𝜎𝑆 𝐶 , 𝑓 (𝜃 , 𝜃 ) + 𝑘 𝑒𝑥𝑝 − 𝜃
2𝜋𝑀 𝑅𝑇
Equation A.13
where,
𝑓(𝜃 ) is a function that is dependent on the reaction model used. For all of our models, they
will be in the form of:
𝑓 (𝜃 , 𝜃 ) = 1 − 𝜃 −𝜃
Equation A.14
105
Similar to the one dimensional gas phase mass balance, we need to change the concentration
term into mol fractions
With the ideal gas assumption, the total concentration of gases in the washcoat is given by:
𝑃
𝐶 =
𝑅𝑇
Equation A.15
Also,
𝐶 ×𝑌 =𝐶
Equation A.16
Therefore,
𝑃
𝐶 =𝑌
𝑅𝑇
Equation A.17
−𝑟 , +𝑟 ,
1000𝑅𝑇 𝑃 𝑃 1000(𝐸 − 𝛽𝜃 )
=− 𝜎𝑆 𝑌 𝑓(𝜃 , 𝜃 ) + 𝑘 𝑒𝑥𝑝 − 𝜃
2𝜋𝑀 𝑅𝑇 𝑅𝑇𝑠 𝑅𝑇
Equation A.18
𝜕𝐶 , 𝑊 𝜕 𝐶
𝜀 = 𝑘 , 𝐶 , −𝐶 , +𝐷 ,
𝜕𝑡 𝐴 𝜕𝑧
1000𝑅𝑇 𝑃
− 𝜎𝑆𝐶𝑂 𝑌𝐶𝑂,𝑠 𝑓 (𝜃 , 𝜃 )
2𝜋𝑀𝐶𝑂 𝑅𝑇𝑠 1 𝐶𝑂 𝑂
𝑃 1000(𝐸2 − 𝛽𝜃𝐶𝑂 )
+ 𝑘2 𝑒𝑥𝑝 − 𝜃𝐶𝑂
𝑅𝑇 𝑅𝑇
Equation A.19
106
𝜕𝐶 , 𝐶 ,, −𝐶 ,,
=
𝜕𝑡 𝑑𝑡
Equation A.20
𝜕 𝐶 , 𝐶 , , − 2𝐶 ,, +𝐶 ,,
=
𝜕𝑧 𝑑𝑧
Equation A.21
𝐶 ,, −𝐶 ,,
𝜀
𝑑𝑡
𝑊
= 𝑘 , 𝐶 , ,, −𝐶 , ,,
𝐴
𝐶 , , − 2𝐶 ,, +𝐶 ,,
+𝐷 , ,
𝑑𝑧
1000𝑅𝑇 𝑃
− 𝜎𝑆 𝑌 , (1 − 𝜃 −𝜃 )
2𝜋𝑀 𝑅𝑇
1000(𝐸 − 𝛽𝜃 )
+ 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation A.22
1 𝑊
= 𝑘 , 𝑌 , ,, −𝑌 , ,,
𝜀 𝐴
𝑌 , , − 2𝑌 ,, +𝑌 ,, 1000𝑅𝑇
+𝐷 , , − 𝜎𝑆 𝑌 , (1
𝑑𝑧 2𝜋𝑀
1000(𝐸 − 𝛽𝜃 )
−𝜃 − 𝜃 ) + 𝑘 𝑒𝑥𝑝 − 𝜃
𝑅𝑇
Equation A.23
107
𝑌 ,, −𝑌 ,,
𝑑𝑡
1 𝑊 𝑌 , , − 2𝑌 ,, +𝑌 ,,
= 𝑘 , 𝑌 , ,, −𝑌 , ,, +𝐷 , ,
𝜀 𝐴 𝑑𝑧
1000𝑅𝑇
−2 𝜎𝑆 𝑌 𝑓 (𝜃 , 𝜃 )
2𝜋𝑀
Equation A.24
Basic Model 1 − 2𝜃
1− 𝜃 − 2𝜃
1−𝜃
CO Island Model 1 − 2𝜃
1 − 1 − 𝜃 𝜃 − 2𝜃
(1 − 𝜃 )
O Island Model 1 − 2𝜃
1 − 1 − 𝜃 𝜃 − 2𝜃
(1 − 𝜃 )
Applying the same operation on the CO2 solid phase mass balance:
𝑌 ,, −𝑌 ,,
𝑑𝑡
1 𝑊
= 𝑘 , 𝑌 , ,, −𝑌 , ,,
𝜀 𝐴
𝑌 , , − 2𝑌 ,, +𝑌 ,,
+𝐷 , ,
𝑑𝑧
1000𝐸 𝑌 𝑌
+ 𝑘 exp − 𝑓 (𝜃 , 𝜃 ) + 𝑘 ,
𝑅𝑇 𝑅(𝜃)
Equation A.25
Basic Model 𝜃 𝜃
CO Island Model 𝜃
𝜃
(1 − 𝜃 )
O Island Model 𝜃
𝜃
(1 − 𝜃 )
𝜕𝑇 𝜕 𝜕𝑇 𝜕𝑇 4
𝜌 𝐶 = 𝐾 , −𝜌 𝐶 𝑣 − 𝑘 𝑇, − 𝑇
𝜕𝑡 𝜕𝑧 𝜕𝑧 𝜕𝑧 𝐷
Equation A.26
The following assumptions are made:
Then,
𝜕𝑇 𝜕 𝑇 𝜕𝑇 4
𝜌 𝐶 =𝐾 , −𝜌 𝐶 𝑣 − 𝑘 𝑇, − 𝑇
𝜕𝑡 𝜕𝑧 𝜕𝑧 𝐷
Equation A.27
Rearranging yields:
𝜕𝑇 𝐾 , 𝜕 𝑇 𝜕𝑇 4
= −𝑣 − 𝑘 𝑇, − 𝑇
𝜕𝑡 𝜌 𝐶 𝜕𝑧 𝜕𝑧 𝐷 𝜌 𝐶
Equation A.28
Using the explicit method the following is obtained:
𝜕𝑇 𝑇,, −𝑇,,
=
𝜕𝑡 𝑑𝑡
Equation A.29
𝜕 𝑇 𝑇, , − 2𝑇 , , + 𝑇 , ,
=
𝜕𝑧 𝑑𝑧
Equation A.30
𝜕𝑇 𝑇, ,−𝑇,,
=
𝜕𝑧 𝑑𝑧
Equation A.31
110
Equation A.32
𝜕𝑇 𝑊 𝜕 𝜕𝑇
𝜌 𝐶 = 𝑘 𝑇 −𝑇 + 𝐾 , + (∆𝐻𝑅)
𝜕𝑡 𝐴 𝜕𝑧 𝜕𝑧
Equation A.33
The following assumptions are made:
Then,
𝜕𝑇 𝑊 𝜕 𝑇
𝜌𝐶 = 𝑘 𝑇 −𝑇 +𝐾 , + (∆𝐻𝑅)
𝜕𝑡 𝐴 𝜕𝑧
Equation A.34
Rearranging yields:
𝜕𝑇 𝑊 𝑘 𝐾 , 𝜕 𝑇 1
= 𝑇 −𝑇 + + (∆𝐻𝑅)
𝜕𝑡 𝐴 𝜌𝐶 𝜌𝐶 𝜕𝑧 𝜌𝐶
Equation A.35
Using the explicit method the following is obtained:
𝜕𝑇 𝑇,, −𝑇,,
=
𝜕𝑡 𝑑𝑡
Equation A.36
111
𝜕 𝑇 𝑇, , − 2𝑇 , , + 𝑇 , ,
=
𝜕𝑧 𝑑𝑧
Equation A.37
𝑇,, −𝑇,, 𝑊 𝑘 𝐾 , 𝑇, , − 2𝑇 , , + 𝑇 , , 1
= 𝑇, , −𝑇, , + + (∆𝐻𝑅)
𝑑𝑡 𝐴 𝜌𝐶 𝜌𝐶 𝑑𝑧 𝜌𝐶
Equation A.38
Equation A.39
where
−∆𝐻 , is the heat of reaction of the surface reaction with CO, J/mol
𝑟 is the rate of reaction of the surface reaction with C3H6 given by:
𝑌 𝑌
𝑟 =𝑘 ,
𝑅(𝜃)
Equation A.40
𝑹(𝜽) is a term which takes into account of the inhibition effects of CO, C3H6, and NO on the
reaction rates.
.
𝑅(𝜃) = 1 + 𝑘 𝑌 +𝑘 𝑌 × 1+𝑘 𝑌 𝑌 × [1 + 𝑘 𝑌 ]
Equation A.41
−∆𝐻 , is the heat of reaction of the surface reaction with C3H6 , J/mol
Then,
𝑇,, −𝑇,, 𝑊 𝑘 𝐾 , 𝑇, , − 2𝑇 , , + 𝑇 , ,
= 𝑇,, −𝑇,, +
𝑑𝑡 𝐴 𝜌𝐶 𝜌𝐶 𝑑𝑧
𝐻𝐴 1000𝐸 𝑌 𝑌
+ 𝑘 exp − 𝑓 (𝜃 , 𝜃 ) × −∆𝐻 , +𝑘 ,
𝜌𝐶 𝑅𝑇 𝑅(𝜃)
× −∆𝐻 ,
Equation A.42
𝑓 (𝜃 , 𝜃 ) is dependent on the reaction model and is available in Table A.2
5
113
Monolith Properties
Parameter Symbol Value Unit Ref.
Channel Length 𝐿 5 𝑚𝑚 Ye, 2012
Channel Radius 𝐷 410 𝜇𝑚 Ye, 2012
Washcoat Thickness 𝑊 100 𝜇𝑚 Ye, 2012
Catalyst Loading 𝐿 2.0 × 10 𝑚𝑜𝑙 Ye, 2012
𝑔𝐷𝑂𝐶
Density of Washcoat 𝜌 1300 𝑘𝑔 Ye, 2012
𝑚
BET Surface Area 𝑆 160 𝑚 Ye, 2012
𝑔
Average Pore Diameter 𝐷 12.25 𝑛𝑚 Ye, 2012
Porosity 𝜀 0.55 Mukadi, 2002
Fluid Properties
Parameter Symbol Value Unit Ref.
Ideal Gas Constant 𝑅 8.314 𝐽 Raj 2015
𝑚𝑜𝑙𝐾
114
All the equations below are obtained from ‘Introduction to Catalytic Combustion’ by Hayes
and Kolaczkowski (1997)
Density of air
𝑃 × 𝑀𝑟
𝜌 =
1000𝑅 𝑇
Equation C.1
Specific mass heat capacity of air
1
𝐶 , = 28.09 + (0.1965 × 10 )𝑇 + (0.4799 × 10 )𝑇 − (1.965 × 10 )𝑇
0.02896
Equation C.2
Thermal conductivity of air
𝑘 , = 1.679 × 10 + (5.973 × 10 )𝑇
Equation C.3
Thermal diffusivity of air
𝑘 ,
𝛼=
𝜌 𝐶 ,
Equation C.4
Specific mass heat capacity of solid phase
𝐶 , = 948 + 0.2268𝑇
Equation C.5
𝑘 , = 0.9558 − (2.09 × 10 )𝑇
Equation C.6
Viscosity of air
Equation C.7
Reynolds number
𝜌 𝑣 𝐷
𝑅𝑒 =
𝜇
Equation C.8
Prandtl number
𝐶 , 𝜇
𝑃𝑟 =
𝑘 ,
Equation C.9
Graetz number
𝐷
𝐺𝑧 = 𝑅𝑒 × 𝑃𝑟 ×
𝐿
Equation C.10
Nusselt number for constant wall temperature
.
48.2 1000
𝑁𝑢 = 3.657 + 8.827 × 𝑒𝑥𝑝 − ×
𝐺𝑧 𝐺𝑧
Equation C.11
Nusselt number for constant wall flux
.
60.2 1000
𝑁𝑢 = 4.364 + 13.18 × 𝑒𝑥𝑝 − ×
𝐺𝑧 𝐺𝑧
Equation C.12
118
Nusselt number
𝑁𝑢 + 𝑁𝑢
𝑁𝑢 = +1
2
Equation C.13
Heat transfer coefficient of air
𝑁𝑢 × 𝑘 ,
ℎ =
𝐷ℎ
Equation C.14
Diffusion coefficient of species A in air
1 1
+
𝑀𝑟 𝑀𝑟
𝐷 , , = (1.013 × 10 )𝑇
𝑃 𝑉 , + 𝑉 ,
Equation C.15
Convective mass transfer coefficient of species A in air
𝑆ℎ × 𝐷 ,
𝑘 , =
𝐷
Equation C.16
Knudsen diffusion coefficient of species A
𝑇
𝐷 = 97𝑟
𝑀𝑟
Equation C.17
Diffusion coefficient of species A in air in solid phase
1 1
𝑀𝑟 + 𝑀𝑟
𝐷 , , = (1.013 × 10 )𝑇
𝑃 𝑉 , + 𝑉 ,
Equation C.18
119
Diffusion in pores
1
𝐷 , =
1 1
𝐷 +𝐷
, ,
Equation C.19
Effective Diffusivity
𝜙𝛿𝐷 ,
𝐷 =
𝜏
Equation C.20
Heat of Reaction of C3H6
Equation C.21
120