Sunteți pe pagina 1din 179

Modeling and Model-Based Control of

a Three-Way Catalytic Converter


PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de
Technische Universiteit Eindhoven,
op gezag van de Rector Magnicus, prof.dr. R.A. van Santen,
voor een commissie aangewezen door het College voor
Promoties in het openbaar te verdedigen op
maandag 25 maart 2002 om 16.00 uur
door
Mario Balenovic
geboren te Sisak, Kroatie
Dit proefschrift is goedgekeurd door de promotoren:
prof.dr.ir. A.C.P.M. Backx
en
prof.dr.ir. J.C. Schouten
Copromotor:
dr.ir. J.H.B.J. Hoebink
CIP-DATA LIBRARY TECHNISCHE UNIVERSITEIT EINDHOVEN
Balenovic, Mario
Modeling and model-based control of a three-way catalytic converter / by
Mario Balenovic. Eindhoven : Technische Universiteit Eindhoven, 2002.
Proefschrift. ISBN 9038619006
NUGI 832
Trefwoorden.: uitlaatgassen / procesregeling / katalysatoren / reactiekinetiek.
Subject headings: air pollution control / predictive control / catalysts /
reaction kinetics.
Eerste promotor: prof.dr.ir. A.C.P.M. Backx
Tweede promotor: prof.dr.ir. J.C. Schouten
Copromotor: dr.ir. J.H.B.J. Hoebink
Kerncommissie:
prof.dr.ir. M. Steinbuch
prof.dr.ir. O.H. Bosgra
The Ph.D. work is supported by the Technology Foundation STW, applied
science division of NWO.
The Ph.D. work forms a part of the research program of the Dutch Institute
of Systems and Control (DISC).
Preface
Wise men say that everything sooner or later comes to an end, and so has the
writing of this thesis. At a moment like this the writer is usually expected
to put together some smart thoughts that will encourage the reader to strug-
gle through the rest of the text. Well, since it is a late Sunday evening smart
thoughts have already left me, hiding in front of the upcoming Monday. Maybe
they were also washed away by the heavy rain that caught me on the bicycle
today. Anyway, this thesis is about the catalytic converters that should help
to clean the air above us, what is of an essential importance for all the bicy-
cle riders in the Netherlands that have to manoeuvre between all those mean
dirty cars (sometimes without their lights on, thus making the cars especially
mean). I dont suppose this work will solve all the existing problems or save
the Planet by itself, but I hope it is a step in the right direction and that the
last four and half years of my life (I guess I shouldnt count all the evenings,
Saturdays, Sundays...) were not in vain. But I suppose I should stop with this
brainstorming that will turn away even the most nave reader and thank some
of the people that helped this thesis to see the daylight.
First of all I would like to thank Ton Backx for his guidance and sup-
port during my research, despite always being busy with more things than one
person should be busy with. Youve always found some time for my urgent
problems and I appreciate that.
Further, I would also like to thank Jan Harmsen, Jozef Hoebink and Jaap
Schouten for a fruitful collaboration on the project. There were always some
stupid chemical questions from my side and you always had patience to answer
them.
Im very grateful to Paul van den Bosch for accepting me into his group to
do Ph.D. and promptly solving many problems that occurred along the way.
My fellow AIOs (and some already dr.irs) Victor, Dik, Maurice, Liu Hong,
Andrei, Hardy, Patricia, Aleksandar, Ivo, Patrick and especially my (ex) room-
mates Yvo, Leo and Bart I thank for setting up a relaxed and comfortable
atmosphere within the group and all the football evenings that usually ended
up disastrously for the Dutch teams. This thesis would certainly not look the
same if Leon didnt have this great idea about making The Style File for his
generation and generations to come. I should certainly not forget Udo for
making my computer work bug-free (from time to time) and Barbara for moral
and grammatical support and all the stories about early Queen and good old
England in general.
When I started working on the project I knew that I didnt want it to be
conned only in the theoretical sphere. Im very grateful to Toon de Bie of
TNO Automotive, William van der Velden of PD&E and Will Hendriks for
2 Preface
their support to various practical tests performed during the project. I would
also like to thank Martin Votsmeier of OMG AG & Co. KG (formerly dmc
2
) for
his interest and support by making results of their engine bench tests available
to me and supplying the monolithic converter for the experminets.
If it wasnt for my rst mentor, Zdenko Kovacic, who introduced me into
the world of Science, I would probably not be doing all of this.
Some people did not have much understanding of what I was exactly doing,
but still without their support and love I would not have passed so easily
through all the obstacles, and not only those of the last four years. I am
thinking here of course of my parents and my sister.
And nally it is Monday already. I should be nishing this. But not before
one last person, who is mumbling at the moment about me going to sleep, is
mentioned. Four years ago, all alone in a little room I was dreaming about the
things that I have right now. You were so far away and everything here was
waiting for the time to pass. And then you came and the isolation, desolation
and separation all faded away in the same moment, so that I can say now that
I have the time of my life. And although you will probably not nd anything
romantic in it (certainly not if you look at the contents), I dedicate this thesis
to you Viktorija.
Mario Balenovic
Eindhoven, January 28
th
2002
Abstract
An increased concern about automotive pollution in the last 30 years has led
to very stringent emission standards. The subject of the research presented in
this thesis was the development of new control strategies for automotive three-
way catalytic converters in order to fulll future ultra-low exhaust emission
standards. More specically, the goal was to develop a model-based control
strategy that can reduce the emissions under highly dynamic operation of the
process, i.e. city driving. Also a possible improvement of the catalyst light-o
(reduction of the temperature needed for the converter to become operational)
has been studied. The main contribution of the thesis is the development of
a model-based controller on the basis of information extracted from the rst
principle modeling of the converter.
The three main parts of the research were: development of the rigorous,
rst principle model of the catalytic converter; development of the control-
oriented model of the catalytic converter and connecting it with the engine
model; development and testing of the novel model-based controller by both
simulations and experiments.
The development of the rst principle model for a catalytic converter was
based on chemical kinetic models of the reactions taking place inside the con-
verter. By adding appropriate mass transfer and energy equations a complete
converter model was obtained. The model predictions have been compared
to experimentally measured data. An improvement of the converters light-o
by means of oscillating inlet feed (oscillations of the inlet lambda value) and
secondary air injection (additional air is injected behind the engine exhaust
valves) was studied. It is shown that the light-o improvement is possible if
right operating conditions are kept.
After the converter light-o the main dynamic eect stems from oxygen
storage and release on ceria, which is placed in the washcoat of the reactor.
By properly controlling this process an extra buer can be obtained to al-
low temporary excursions of the engine lambda (air/fuel ratio) value, which
are inevitable during a dynamic operating regime. The goal of the catalytic
converter controller is to nd the optimal oxygen storage coverage and to nd
optimal trajectories to reach this steady state (fast response with a low exhaust
emission). In order to use the model information in the controller the rigorous
model had to be reduced. A simplied control-oriented model has been devel-
oped to predict the level of oxygen storage coverage on-line. It is a one state
nonlinear model with the state being the oxygen storage coverage. It was found
experimentally that in some cases a two state model which makes a distinction
between oxygen stored on the ceria surface and in the bulk can lead to a better
prediction. The model can automatically be tuned on the basis of the catalytic
4 Abstract
converter step responses during which the inlet and outlet lambda values are
measured. The information about the nonlinear process dynamics obtained
from the rst principle modeling has led to an algorithm for the extrapolation
of the control-oriented model obtained in one operating point to other operat-
ing conditions. In this way the model tuning procedure, which can also lead
to substantial exhaust emissions and cannot be performed during a standard
system operation, can be reduced. The prediction of the model was compared
to the rigorous model prediction and it was found to be quite accurate.
The model is used as an inferential sensor in the applied controller in the
engine control unit for predicting the degree of the oxygen storage coverage
that cannot be measured. The actual controller is an analytic approximation
of the developed Model Predictive Controller. The developed Model Predic-
tive Controller is capable of using the process information available from the
model to nd an optimal control behavior set by the control objectives. This
controller requires solving an optimization problem at every sampling instant,
which cannot be achieved in the engine control unit due to limited processing
capabilities. Therefore, the optimization problems for the expected operating
conditions are solved o-line and used to train a simple neural network that
emulates the Model Predictive Controller.
The controller has been tested by simulations on the rst principle model,
and by experiments performed on an engine test bench. The performed tests
simulated highly dynamic system operation, when the majority of emissions
occur, such as city driving. Due to a proper use of the model information, the
novel controller leads to the emission reduction under above given conditions.
Contents
Preface 1
Abstract 3
1 Introduction 7
1.1 Exhaust and the environment . . . . . . . . . . . . . . . . . . . 7
1.2 Legislation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Historic overview of the exhaust aftertreatment . . . . . . . . . 12
1.4 Standard air/fuel control system . . . . . . . . . . . . . . . . . 14
1.5 Scope of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 First principle modeling 23
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Mathematical model . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 Reactor model . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Kinetic model . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.3 Numerical procedure . . . . . . . . . . . . . . . . . . . . 34
2.3 Cold start . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.1 Model verication at cold start conditions . . . . . . . . 36
2.3.2 Light-o with steady inlet feed . . . . . . . . . . . . . . 38
2.3.3 Light-o with oscillatory inlet feed . . . . . . . . . . . . 43
2.3.4 Light-o under lean conditions - secondary air injection 49
2.4 Warmed-up converter . . . . . . . . . . . . . . . . . . . . . . . 52
2.4.1 Steady state operation . . . . . . . . . . . . . . . . . . . 52
2.4.2 Dynamic operation . . . . . . . . . . . . . . . . . . . . . 54
2.4.3 Motivation for control . . . . . . . . . . . . . . . . . . . 62
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3 Model-based controller 67
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Engine model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.1 Air path . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.2 Fuel path . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2.3 The complete model . . . . . . . . . . . . . . . . . . . . 76
3.3 Engine air/fuel control . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.1 IMC controller . . . . . . . . . . . . . . . . . . . . . . . 80
3.4 Control-oriented model of the catalytic converter . . . . . . . . 84
3.4.1 Model basics . . . . . . . . . . . . . . . . . . . . . . . . 85
6 Contents
3.4.2 Parameter estimation . . . . . . . . . . . . . . . . . . . 90
3.4.3 Experimental model verication . . . . . . . . . . . . . 98
3.5 Feasibility of control . . . . . . . . . . . . . . . . . . . . . . . . 104
3.5.1 Gain scheduling controller . . . . . . . . . . . . . . . . . 105
3.5.2 Inuence of the sensor oset on the control robustness -
steady vs. oscillatory . . . . . . . . . . . . . . . . . . 108
3.6 Model-based predictive control . . . . . . . . . . . . . . . . . . 111
3.6.1 Steady state optimization . . . . . . . . . . . . . . . . . 112
3.6.2 Dynamic optimization . . . . . . . . . . . . . . . . . . . 113
3.6.3 Analytic MPC approximation . . . . . . . . . . . . . . . 116
3.6.4 Simulation results . . . . . . . . . . . . . . . . . . . . . 120
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4 Experimental testing of the control system 127
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.2 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.3 Open loop tests: model evaluation . . . . . . . . . . . . . . . . 129
4.3.1 Model application range . . . . . . . . . . . . . . . . . . 131
4.3.2 Model testing . . . . . . . . . . . . . . . . . . . . . . . . 137
4.4 Closed loop tests: model-based control . . . . . . . . . . . . . . 139
4.4.1 Controller tuning . . . . . . . . . . . . . . . . . . . . . . 139
4.4.2 Experimental results . . . . . . . . . . . . . . . . . . . . 141
4.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5 Conclusions and Outlook 151
5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.1.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.1.2 Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.2.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.2.2 Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Notation 159
Bibliography 163
Samenvatting 171
Sazetak 173
Curriculum Vitae 175
1
Introduction
1.1 Exhaust and the environment
1.2 Legislation
1.3 Historic overview of the
exhaust aftertreatment
1.4 Standard air/fuel control
system
1.5 Scope of the thesis
Gasoline (spark-ignition) combustion engines are in use for more than hun-
dred years now, since the invention by German engineer N.A. Otto in 1878.
Though the engine operation principle has basically remained the same, the
engines have undergone vast improvements since. However, a perfect combus-
tion is still not obtained. Hence, together with large amounts of carbon dioxide
(CO
2
) and water (H
2
O) in the exhaust, also undesired carbon monoxide (CO),
unburned hydrocarbons (HC) and oxides of nitrogen can be found. With the
increase of health and pollution problems caused by the above mentioned ex-
haust components, concern was raised in the western society about the impact
of the automotive pollution on our present and future life. This has lead to the
introduction of pollution control and signicant advances in this eld over the
last 30 years.
1.1 Exhaust and the environment
A brief overview of major pollutants stemming from the gasoline engines is
given here. Pollutants with dierent origin, such as particulate matter stem-
ming from diesel vehicles, will not be included as this is out of the scope of this
thesis.
Carbon monoxide
Carbon monoxide is a product of a partial combustion of hydrocarbons in fuel.
It is always present when there is a lack of oxygen during combustion and thus
directly dependent on the applied engine air/fuel ratio. It can, however, also
be found in the exhaust when there is a net abundance of oxygen (i.e. lean
conditions) [21, 45]. The same applies for diesel engines, which run with a very
lean combustion mixture. These emission levels are much lower, of course, then
8 Introduction
with rich mixtures. The reason are local regions with oxygen deciency inside
the cylinder, which can occur though the total mixture is net oxidizing. Such
local rich regions may occur due to a poor fuel vaporization or late mixing. A
large CO engine-out emission can also occur unintentionally with the mixture
becoming rich during engine transients (acceleration, deceleration, gear shift-
ing). Sometimes the mixture is intentionally made richer due to for example,
torque demand or during the cold start to assure smooth combustion.
CO is the best known for its toxicity as already a couple of hundreds ppm
can cause dizziness and headaches. Several thousands ppm are usually lethal.
CO has a greater anity to bond with haemoglobin in blood than oxygen, and
reduces the supply of oxygen to the body tissue. It is a colorless and odorless
gas so the victim is often completely unaware of its presence until it is too late.
Since the local level of CO is more dangerous than the global level, the most
dangerous areas are where the trac is dense or engines are running in poorly
ventilated or conned spaces.
Hydrocarbons
The term hydrocarbons (HC) in the exhaust stands for the unburned organic
compounds that contain hydrogen and carbon. It is better to refer to these
components as unburned hydrocarbons than as unburned fuel since the ma-
jority of the exhaust hydrocarbon is a product of partial oxidation and is ligther
than the original hydrocarbons in the fuel. Note that there are more organic
compounds in the exhaust, that also contain oxygen atoms (ketons, aldehy-
des), but do not fall under the denition of hydrocarbons. A larger denition
can be Volatile Organic Compounds (VOC) that would include all carbon
containing compounds present in the gaseous state at ambient temperatures
[21]. However, due to the largest presence of hydrocarbons in the exhaust only
the term HC is usually applied. The are hundreds of hydrocarbon components
with various concentrations that can be found in the exhaust. The most im-
portant are alkanes, alkenes, alkynes and aromatics. A vast amount of various
hydrocarbons makes it very dicult to create a reliable model for a catalytic
converter as will be discussed in the next chapter.
There are several paths that cause hydrocarbons in the exhaust. The most
obvious is, as in the case of CO, a lack of oxygen when the air/fuel mixture
is rich. The other reasons that can cause hydrocarbon emissions even with
lean mixtures are [45]: crevices (piston top, threads around the spark plug),
the quench layer (due to a lower temperature of the cylinders walls), porous
deposits, absorption by oil, bulk quenching (due to regions of the charge which
are oxygen decient), late burning and problems with injectors (fuel remains
in the nozzle sac of an injector). Another source of released hydrocarbons are
so called evaporative emissions (evaporated fuel escapes from the fuel tank).
This problem will not be treated here.
1.2. Legislation 9
Various hydrocarbons (and other VOC) have dierent more or less harmful
eects on health and environment. Many irritate muscous membranes, leading
to coughing, sneezing and drowsiness. Some have a narcotic eect. Benzene,
for example, is very toxic and carcinogenic, while 80% of benzene in atmosphere
stems from the automotive exhaust. They (especially alkenes) also react with
NOx to create secondary pollution such as tropospheric ozone and photochem-
ical smog. Ozone causes to humans irritation on the eyes, nose and throat,
leads to lung problems, coughing, etc. It is also hazardous for plants.
Oxides of nitrogen
Oxides of nitrogen that are of the largest concern in the automotive applica-
tions are nitric oxide (NO), nitrogen dioxide (NO
2
) and nitrous oxide (N
2
O).
The rst two are usually understood under the term NOx. They are also a
direct product of the combustion in the engine, while N
2
O is primarily a prod-
uct of catalytic converter under some operating conditions. NOx is formed
during combustion in the enigne when oxygen reacts with nitrogen because of
a high combustion temperature. It is therefore an unwanted secondary prod-
uct of combustion. The amount of produced NOx is very dependent on the
combustion temperature (engine load). The NO
2
/NO ratio is very low for
gasoline engines, less than 2%. Some additional NO
2
can also be created by
the catalytic converter.
While NO is odorless, colorless and relatively non-toxic, NO
2
is reddish-
brown, pungent and very toxic. It aects respiratory tract, damages lung tissue
and increases airway resistance. It may also interfere with oxygen transport in
blood via reaction with haemoglobin, provoke coughing, running noses, bron-
chitis, etc. The current legislation aims at NOx emission levels, while it is
to be expected that also N
2
O will be included in the future as it is a strong
greenhouse gas. NOx is also involved in secondary pollution (smog, depletion
of the ozone layer). It is also, together with sulfur oxides, responsible for acid
rains.
1.2 Legislation
1970 is considered to be the beginning of the U.S. motor vehicle emission con-
trol programme. The U.S. Congress passed then the Clean Air Act Amend-
ments, which imposed tough, technology-forcing emission standards. Congress
required 90% reduction in hydrocarbons and CO by 1975 and approximately
90% reduction in NOx by 1976 [11]. At the time these requirements were
too stringent but were setting the pace for technology development needed to
meet such standards. In the last 30 years there were some major standard
revisions, as can be seen in gure 1.1, which led to todays Tier 1 and future
Tier 2 standards. The gure proves that the exhaust emission control is one
10 Introduction
of todays most rapidly developing technologies with the emission reduction of
more than 95% when comparing to the pre-control era. Even more stringent
emissions standards than in the rest of the U.S. are applied in California, which
has the unique authority to implement its own motor vehicle emission control
programme.
0
20
40
60
80
100
120
g
/
m
i
l
e
1
9
6
0
1
9
7
0
1
9
7
5
1
9
8
0
1
9
8
3
1
9
9
4
2
0
0
1
2
0
0
4
Model year
U.S. automobile emission standards
CO
HC x10
NOx x10
Figure 1.1: U.S. automobile emission standards. Adopted from [11]. The
standards for 2001 and 2004 represent corporate average.
In Europe, the rst base directive 70/220/EEC was introduced in 1970 to
set the emission limits for CO and HC [40]. Standards have involved since,
leading to the consolidating directive 91/441/EEC in 1991, which set EURO I
standards. EURO I was the rst mandatory European vehicle emission stan-
dard. In 1992 catalytic converters became compulsory on all new cars sold
in Europe. Also in 1992, Auto-Oil Programme, a cooperative project of the
European commissioners for the environment, industry and energy, automobile
manufacturers and oil industry trade associations was introduced. The goal of
this programme was to set a framework for the development of future emission
standards based on science, available technology, environmental need and cost
eectiveness. Euro III standard is currently enforced and Euro IV standard
will come to power in 2005 [31, 80]. The overview of EURO standards is given
in table 1.1.
Note that the U.S. and European standards should not be directly com-
pared. The goal of a standard is to limit the average g/km (or g/mile) emission
of a certain exhaust component. These gures are, however, very dependent on
the driving test cycle applied. Dierent test cycles are applied in the U.S. and
Europe (see gure 1.2 for the standard European emission test cycle). More-
1.2. Legislation 11
Year 1993/94 1996/97 2000/01 2005/06
Euro I Euro II Euro III Euro IV
CO 2.72 2.2 (2.7) 2.3 1.0
HC - - (0.341) 0.20 0.10
HC+NOx 0.97 0.5 - -
NOx - - (0.252) 0.15 0.08
Table 1.1: European standards for gasoline fueled vehicles [g/km]. The cor-
rected values for Euro II take into account the test procedure from 2000 that
eliminates the 40s pre-test idle period. Adopted from [31, 80].
over, before 2000 the emission sampling in the European cycle would start only
40s after cold start of a vehicle. Since the cold start, together with the dynamic
vehicle operation, is the largest source of emissions, such a test does not give
an accurate estimate of actual vehicle emissions. Correction gures to account
for the rst 40s emission in the EURO II standard are also given in table 1.1.
0 200 400 600 800 1000 1200
0
20
40
60
80
100
120
140
time [s]
v
e
h
ic
le

s
p
e
e
d

[
k
m
/
h
]
Figure 1.2: European emission driving test cycle: vehicle speed measured dur-
ing the test.
Simulation and experimental studies in this thesis were not based on any
standard test cycles, but rather on a highly dynamic operation under conditions
more stringent than during the test cycles. Such more dynamic test cycles may
be expected in the future legislation.
Emission standards will become even more stringent in the future leading
to the ultimate goal: Zero Emission Vehicles (ZEV). Due to the population
growth and steady growth in the number of motor vehicles such a goal is still
very legitimate despite some great results already achieved.
12 Introduction
1.3 Historic overview of the exhaust aftertreat-
ment
Automobile manufactures started considering exhaust aftertreatment systems
after engine-only measures had failed to satisfy new pollution legislation. It
was known for a long time that several precious metals and even some base
metals, placed in a converter in the engine exhaust, could convert hydrocar-
bons and CO provided that enough oxygen and high enough exhaust temper-
atures were present. The catalytic converters were therefore already consid-
ered before 1970s, when they actually entered the production, but a number
of reasons such as the exhaust harsh environment, large fuel lead contents
(that serves as engine antiknocking measure, but speeds up the catalyst de-
activation immensely), and uncertainty about precious metal application due
to their scarcity and price have prolonged their application [82]. With legisla-
tion becoming more stringent and lot of investment in the research, the above
mentioned problems were being solved one by one and the catalytic converter
has found itself as an unavoidable and reliable part of almost every automotive
system today.
Early developments
In the late 70s the catalytic converters were very simple oxidizing converters
aiming to convert CO and HC. The necessary conditions were a net oxidizing
inlet feed and suciently high temperature [95]. Since engines at that time
were still mostly tuned rich to obtain a higher torque, an oxidizing mixture at
the converter inlet was simply obtained by placing an air pump in the exhaust
manifold. Catalysts were rather simple at the time and contained typically Pt
or Pt/Pd noble metal. These noble metals are known to promote the oxidation
processes. It was then already realized that the optimal converter congura-
tion was a monolithic conguration, in contrast to packed-bed converters that
prevailed in the process industry. Monolithic converters cause much lower back
pressure than the packed bed catalysts, what is important in order to avoid
engine power losses. The problem, however, was to ensure a high conversion
with a very low residence time of the exhaust gas in the converter. Monoliths
are multi-channel structures through which the exhaust gas passes with a high
volumetric ow rate. The channel walls are coated with a high-surface porous
material (washcoat) with nely dispersed noble metal catalytic particles on it.
Washcoat is typically made of oxides of Al, Ce, Zr, etc. High washcoat surface
enables a high conversion despite low residence times. Schematic view of such
a monolith and one of its channels is given in gure 1.3.
With the increased concern about NOx emissions it was found that a Rh
based catalyst has a better capability of reducing NO than Pt or Pd based
catalysts. In the beginning was such a reducing catalyst used in a dual converter
1.3. Historic overview of the exhaust aftertreatment 13
Figure 1.3: Schematic view of a monolithic converter and one of its channels.
set-up, where the rst catalytic converter was a reducing converter and the
second an oxidizing converter. The engine was tuned rich to promote the NO
reduction in the rst converter, while an air pump was placed behind the rst
converter to ensure a net oxidizing mixture for the second catalytic converter.
Three-way catalytic converter
This development has subsequently led to the introduction of three-way cat-
alytic converters in the early 80s. These converters can simultaneously convert
all three pollutant groups. The necessary condition for an optimal conversion
is that the engine runs with stoichiometric mixture. A typical conversion of a
three-way catalytic converter is shown in gure 1.4. When the feed is net oxi-
dizing the conversion of CO and HC is promoted, while the conversion of NO
is very low, whereas the opposite happens when the feed becomes rich. These
converters typically contain Pt/Rh or Pd/Rh catalysts with various support.
A drawback of three-way catalytic converters at the time of their introduc-
tion was that a very accurate engine air/fuel control (i.e. fuel metering) was
necessary to maintain the exhaust mixture at stoichiometry. With the devel-
opment of exhaust gas oxygen (EGO) sensors, popularly called sensors, it
was possible to develop a control system that keeps the engine air/fuel ratio
at stoichiometry. Carburators, that were still used at the time for the air/fuel
mixture preparations were not up to the task of a precise fuel metering. Only
with development of more accurate fuel injectors, that were able to accurately
control the amount of the injected fuel, was the system able to obtain the nec-
essary high performance. The next section describes todays exhaust emission
control system with the three-way catalytic converter.
Though the basic function of three-way catalytic converters has not changed
in the last two decades, there were major developments in converters durabil-
ity, decreased susceptibility to poisoning, thermal stability, etc. Light-o tem-
perature (the temperature where the conversion reaches 50%) of converter is
14 Introduction
13.5 14 14.5 15 15.5
0
10
20
30
40
50
60
70
80
90
100
A/F []
T
W
C

c
o
n
v
e
r
s
io
n

e
f
f
ic
ie
n
c
y

[
%
]
CO
HC
NOx
Figure 1.4: Typical steady state conversion eciency of a three-way catalytic
converter. Adopted from [82].
constantly decreasing, together with improvements in catalystss oxygen stor-
age capability. The latter is the capability of the catalyst to store oxygen when
there is an abundance of it, i.e. lean feed, and to release it when there is a
shortage of it, i.e. rich feed. This feature is predominantly caused by ceria,
which is placed in the converter washcoat. Hence, the operating window of the
converter widens, allowing also feed oscillations around stoichiometry while
preserving a high conversion. A study of oxygen storage dynamics is one of
the main topics of this thesis. Together with the advances in the converter
performance, the last decade has also witnessed great improvements in the en-
gine electronic control system. The engine exhaust aftertreatment system is
thus becoming one of the most advanced collaborations between chemistry and
electronics.
1.4 Standard air/fuel control system
A general introduction to the operation of engine air/fuel control schemes is
given in this section. The stress here is on the introduction of control system
elements, while more detailed studies will be conducted in chapter 3. A basic
scheme of the control system is given in gure 1.5. The main control system
components are the engine and the catalytic converter, which act as the process,
fuel injectors, which are the actuators (in some modern drive-by-wire systems
the throttle can also be considered as an actuator) and various sensors (en-
gine speed, throttle position, intake manifold pressure, coolant temperature).
Among sensors, the most important are so-called sensors which directly serve
to create the feedback signal for the controller.
1.4. Standard air/fuel control system 15
air
fuel
controller
engine speed,
temperature
sensor
upstream
sensor
downstream
intake
valve
exhaust
valve
fuel
injector
fuel
puddle
spark
plug
throttle
a
i
r
f
l
o
w
e
s
t
i
m
a
t
i
o
n
catalytic
converter
air
fuel
controller controller
engine speed,
temperature
sensor
upstream
sensor
downstream
intake
valve
exhaust
valve
fuel
injector
fuel
puddle
spark
plug
throttle
a
i
r
f
l
o
w
e
s
t
i
m
a
t
i
o
n
catalytic
converter
Figure 1.5: Electronic engine control system
Before proceeding, rst the engine air/fuel ratio (A/F) and value have to
be dened. The engine air/fuel ratio is the ratio of the air and fuel mass ows
(or in-cylinder mass of air and fuel). By dividing A/F with the stoichiometric
A/F the relative air/fuel ratio, commonly called , is obtained:
A/F =
m
a
m
f
=
A/F
A/F
stoich.
(1.1)
For the stoichiometric mixture holds that = 1, for a lean mixture > 1 and
for a rich mixture < 1. Stoichiometric A/F is around 14.6, depending on the
applied fuel.
In order to calculate the exhaust value the above equation is not suitable,
as it should be expressed in terms of present exhaust gas concentrations rather
then in terms of fuel and mass ow. The following denition of the exhaust
will be applied throughout the thesis:

exh
=
2C
O
2
+C
NO
+C
N
2
O
+ 2C
NO
2
+C
CO
+ 2C
CO
2
+C
H
2
O
2C
CO
+

n
i=1
(2x +
y
2
)C
C
x
H
y
+ 2C
CO
2
+C
H
2
O
(1.2)
where n denotes the number of dierent hydrocarbon species.
16 Introduction
Equation (1.2) is valid at every position in the exhaust line (also in front
of the engine if air and fuel are expressed in terms of concentrations of the
oxidizing and reducing species). Hence, it is valid also in front of and behind
the converter. Note that this value represents the normalized air/fuel ratio,
and therefore are the products of reactions also included in the equation. It
should not be confused with the ratio of oxidizing and reducing species, which
is typically called the equivalence ratio.
The working principle of the A/F control system (gure 1.5) can now be
understood. On the basis of the measured lambda signal in the exhaust (the
depicted sensor behind the converter should be disregarded at the moment)
and estimated air ow, the necessary amount of the injected fuel is calculated
by the controller. The operating conditions, such as the engine speed, coolant
temperature and intake manifold pressure have to be known. The air ow
estimation methods and detailed system dynamics will be studied in chapter 3.
Exhaust gas oxygen sensor
The air/fuel control system was not possible before the development of exhaust
gas oxygen (EGO) sensors, which can measure the exhaust oxygen concentra-
tion on bases of which the actual exhaust can be estimated. It is stressed
here that only an estimation of the real exhaust value is possible since the
measurement is indirect, and sensor errors are hence quite common.
Figure 1.6 shows a standard EGO sensor. The sensor has two electrodes,
one placed in the exhaust and the other in the atmospheric conditions. The
electrodes are typically made of platinum, and have thus catalytic properties.
Between the electrodes a ceramic material is placed that has a capability of
transferring the oxygen ions between the electrodes. The standard ceramic
electrolyte is ZrO
2
. When the sensor is placed in the exhaust, an electromotive
force between the electrodes is created. This voltage can be measured and
theoretically agrees with the Nernst equation:
E =
RT
4F
ln
p(O
2
)
ref
p(O
2
)
ex
(1.3)
where R is the gas constant, T is the temperature in Kelvin, F is Faradays
constant while p(O
2
)
ref
and p(O
2
)
ex
are partial pressures of oxygen in the
reference gas and in the exhaust, respectively.
Since the sensor also acts as a catalytic converter, there is a large dierence
in the sensor output under the rich and lean conditions, because in rich condi-
tions all oxygen is converted on the catalytic surface and the eective oxygen
partial pressure is near zero. Therefore, a relay type sensor characteristic is
obtained, as shown in gure 1.7, with a sudden jump at the stoichiometry. For
rich exhausts the sensor electromotive force is approx. 900mV, while for lean
exhausts it is approx. 100mV. The stoichiometric sensor voltage is approx.
1.4. Standard air/fuel control system 17
Exhaust
Air
Zirconia
Electrolyte
Pt
Electrode
Figure 1.6: EGO sensor
0.94 0.96 0.98 1 1.02 1.04
0
0.2
0.4
0.6
0.8
1
lambda []
E
G
O

s
e
n
s
o
r

v
o
lt
a
g
e

[
V
]
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2
10
5
0
5
10
lambda []
O
2

p
u
m
p
in
g

c
u
r
r
e
n
t

[
m
A
]
Figure 1.7: Characteristics of EGO and UEGO sensors
450mV. For a good operation, the sensor has to reach a certain temperature
(light-o of the catalytic converter), so often heated exhaust gas oxygen sensors
(HEGO) are applied.
It is clear that this sensor gives a signal useful for control only at stoichio-
metric conditions. This is sucient for a standard control system that aims
at controlling the air/fuel ratio at stoichiometry under all conditions, but it is
not sucient for more advanced control schemes when the exact A/F control
is desirable. Such a case will be in this thesis.
Another type of sensor is universal exhaust gas oxygen (UEGO) sensor, or
also called wide range sensor. This sensor gives information about the exact
value. The working principle is shown in gure 1.8.
The UEGO sensor consists of two cells: the pumping and the detecting cell.
18 Introduction
Figure 1.8: UEGO sensor
In the rst cell, oxygen ions can be moved by applying the voltage to the ZrO
2
element. The second cell has the exhaust gas at one electrode and reference gas
at the other electrode. By comparing the voltage produced by the detecting
cell (a standard EGO sensor) and the reference voltage (450mV), a current is
produced and applied to the pumping cell. If the exhaust is rich, the pumping
cell will move oxygen into the exhaust cavity until the detecting cell detects
stoichiometry. By measuring the current needed to achieve this conditions it
can be concluded about the oxygen deciency in the exhaust, i.e. the value.
The opposite holds if the exhaust is lean. Typical sensor characteristic is shown
in gure 1.7. Note that the value predicted by the sensor is dependent on
the exhaust gas characteristic.
Stoichiometric control
Although the application of wide range sensors is slowly increasing in commer-
cial vehicles, HEGO sensors are still widely applied. Due to the relay sensor
characteristic, together with the delay in the control loop, the control sys-
tem exhibits constant oscillations of the engine lambda value. Typically a PI
controller is applied, which serves to adjust the amplitude and frequency of
oscillation.
For example: assume the process to be a pure delay, exp(sT), with a gain
k
p
. The controller is a pure integrator, k
i
/s, and the sensor an ideal relay with
levels 0 and 1 and switching at stoichiometry. The control loop exhibits then
oscillations whose period is 4T and amplitude k
p
Tk
i
/2 [33].
Because of the oxygen storage capabilities of the catalytic converter such os-
cillating behavior does not necessarily deteriorate the converters performance.
Figure 1.9 shows typical lambda signals in front of and behind the catalytic
converter during the closed loop operation measured on a 1997 production ve-
hicle. The applied sensors for data acquisition were wide range sensors, while
a HEGO sensor was applied for control. A ltering behavior of the catalytic
1.5. Scope of the thesis 19
0 2 4 6 8 10 12 14 16 18 20
0.95
1
1.05
l
a
m
b
d
a

[

]
0 1 2 3 4 5 6 7 8 9 10
0.85
0.9
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
Figure 1.9: Lambda signals in front of and behind the catalytic converter during
a closed-loop operation on a production vehicle. Dotted line - converter inlet
signal, solid line - converter outlet signal. Above - period of high conversion,
below - period of lower conversion after a fuel enrichment.
converter is obvious. When the downstream lambda signal equals stoichiom-
etry it is very likely that the conversion of all pollutants is high. When the
signal is rich it means that still some rich components exit the converter (CO,
HC), while when the signal is lean it is very likely that there is some NOx in
the exhaust. Note that in the second gure the downstream signal does not
remain at stoichiometry although the inlet signal oscillates around it, after a
disturbance took place. This means that the dynamic behavior of the catalytic
converter should probably be taken into account if a high-performance control
is desired.
More modern control schemes today include also a HEGO sensor behind the
catalytic converter, because it is less prone to thermal aging and senses more
fully equilibrated exhaust gas mixture then the sensor upstream the converter.
Hence it switches at a gas composition that is closer to the true stoichiometry.
The signal of the downstream sensor is used to correct the switch point of the
main sensor (see gure 1.10) [82].
1.5 Scope of the thesis
In the recent review [82], Shelef and McCabe wrote:
The coupling of electronics with chemistry for optimum emission
control is one research direction which will remain a fertile ground
20 Introduction
Controller
Catalyst Catalyst
HEGO
sensor
HEGO
sensor
Reference
Engine Engine
HEGO
sensor
HEGO
sensor

1

2
Figure 1.10: Dual HEGO sensor control scheme.
far into the future. As on-board computational capabilities con-
tinue to increase, more sophisticated control strategies will be im-
plemented. ... the cold start period and periods of highly transient
driving will be the focus of further work. In addition, greater capa-
bility will be developed to incorporate learning and self-diagnostic
features into control strategy which will allow the system to com-
pensate for mileage induced changes in catalyst activity and in the
sensor response characteristics.
The main goal of this thesis is to investigate the dynamic behavior of the
catalytic converter and to develop a control system that can improve the sys-
tem performance under transient conditions. As it was already shown, the
converter dynamics is often neglected by the controller assuming that only by
achieving the conditions optimal for steady state one also achieves optimal dy-
namic operating conditions. This thesis will answer the question whether such
an assumption is justied. Other important questions to be answered are what
are the major dynamic eects occurring within the catalytic converter, which
dynamics have to be considered during development of the control system, what
are the possible benets of a novel control approach, which variables should be
controlled and which control objectives have to be considered.
Dynamic behavior of the catalytic converter has to be studied over the
entire operating range. This allows selection of an appropriate control strategy
that covers a wide range of driving conditions properly. The main mechanisms
responsible for the converter behavior at low temperatures have to be studied to
understand the light-o process. To achieve high performance emission control
dynamic process behavior at higher temperatures has to be studied as well.
The operating conditions of the main interest are similar to those that occur
during city driving, i.e. low converter temperature, sudden transients.
Two approaches for studying the converter dynamics were possible. The
rst approach involves experimental investigation of the converters dynamic
1.5. Scope of the thesis 21
behavior, while the second approach is a study based on the process model-
ing. The rst approach is very attractive if some basic system behavior has
to be studied in a limited time. It will, however, lead only to an observation
of the system dynamic behavior, but not to a full understanding of it. With
a process not fully understood, such as a catalytic converter, research limited
only to experiments will leave many important questions unanswered. The
modeling approach, on the other hand, gives information about process vari-
ables that cannot be measured, leads to a better understanding of the process
characteristics over a wide operating range and is very suitable for fast and
cheap development of controller prototypes. Of course, experiments may not
be neglected, as they serve to validate the model. Nevertheless, the necessary
amount of experiments is greatly reduced by applying this second approach.
Therefore, in order to study dynamics of the catalytic converter a rst
principle model was created. Such rigorous model is important for understand-
ing interactions of the various processes within the catalytic converter. This
approach also enables a better selection of control goals and possible control
strategies.
The rst principle model is based on kinetic submodels of individual re-
actions that occur inside the converter. These submodels were obtained via
transient experiments. The model allows a study of processes that are tak-
ing place on the catalytic surface, and is valid under all operating conditions
since it is based on the elementary step kinetics. Once it is understood what
needs to be controlled to yield a better performance of the system, dedicated
control-oriented model(s), which are simple enough to be used in the on-board
computer, are created and tested. The control-oriented model can be used as
inferential sensor to estimate some variables that cannot be measured. The
link between the rst principle model and control-oriented model is strong, as
the former is used to understand the process behavior under dierent operating
conditions, and the mutual interaction of the process variables. This helps to
broaden the application range of the control-oriented model. For example, its
parameters can be estimated in one operating point, and because of known
standard process behavior the model can be extrapolated to other operating
points without a need for further parameter optimization. The model is also
directly used to tune the controller to obtain the optimal performance. The
model can beforehand predict the expected future process behavior on which
basis an appropriate control action is chosen. The implementation of such a
controller is discussed in the thesis. Finally, the controller experimental veri-
cation on an engine dynamometer test bench is presented.
The thesis is divided into 5 chapters. Chapter 2 introduces the rst princi-
ple model. After setting the mathematical basis for the model and describing
kinetic submodels, a number of illustrative simulations is performed to study
the converter dynamics after the engine cold start and under standard oper-
ating conditions. The model prediction is compared to some experimentally
22 Introduction
measured data in order to judge the model accuracy. Measures to lower the
light-o temperature by introducing feed oscillations and secondary air injec-
tion are studied, and their potential benets explained on the basis of noble
metal surface dynamics. At higher temperatures oxygen storage dynamics dom-
inate the converter dynamic behavior. Various parameters of importance are
analyzed. The motivation for control is given.
Chapter 3 describes theoretical development of the controller. First, the
control-relevant engine dynamics is introduced and development of a model-
based A/F controller presented. The development of the control-oriented model
for the catalytic converter proves to be crucial for the controller feasibility.
On-line model estimation and broadening of the model application range are
introduced and discussed. The model accuracy is tested with experimental
data. After the model is found to be suciently accurate, the control scheme
is introduced. The controller uses the model as an inferential sensor to esti-
mate the controlled variable which is not measurable. A simulation study with
a simple controller proves the feasibility of control. Further, a more advanced
controller based on Model Predictive Control is introduced. The controller is
tuned o-line and a Gaussian network is used on-line as the MPC approxi-
mation. Controller performance is tested in a simulation of a highly dynamic
engine operation.
Chapter 4 presents the experimental testing of the proposed control system.
The experiments are performed on engine dynamometer test setup. The open
loop behavior of the catalytic converter is analyzed rst, and compared with
the results obtained from the modeling. After development and testing of
the control-oriented converter model, the controller is tested with few sets of
dynamic tests. The experimental results are compared with the simulation
results and the assessment of the controller performance is made.
Chapter 5 concludes the thesis and recommendations for further research
are given.
2
First principle modeling
2.1 Introduction
2.2 Mathematical model
2.3 Cold start
2.4 Warmed-up converter
2.5 Conclusions
2.1 Introduction
After introducing new stringent emission regulations, accurate modeling of cat-
alytic converters now plays an important role in eorts to reduce emissions. A
model can be applied for design and optimization of the converter, as well as
for investigation of dynamic behavior of the converter, which can then be used
to improve the existing converter control strategies. Though many eects that
are taking place inside of the converter are quite well known [24], when one
tries to convert them into a reliable and accurate mathematical model, a chal-
lenging problem arises. Not only the number of reactions occurring (which to a
large extent determines the model complexity) is quite large, but the converter
operation is also a highly dynamical process. The converter inputs (exhaust
gas concentrations, temperature and mass ow) are subject to on-going vari-
ations leading to a dynamical regime of the converter operation. The current
fuel injection control systems cause a continuously changing converter input
gas ow composition, which is a result of control system delays and a relay
type exhaust lambda sensor. These oscillations are known to inuence the con-
verter performance [10, 69, 84]. Due to a limited control system bandwidth
a sudden vehicle acceleration or deceleration can also lead to further excur-
sions of the inlet lambda value to rich or lean. Inlet gas mass ow is also
aected by these changes. Another well-known dynamic process is the con-
verter warming-up process. This process is especially signicant as 70-80% of
all harmful substances is emitted during an FTP or Euro test cycle immediately
after the engine cold start. A good converter model must therefore have the
ability to predict these dierent dynamic eects and thus incorporate dierent
mechanisms that describe them.
There are dierent types of converter models aiming at dierent applica-
tions. Only detailed, rst principle models will be discussed in this chapter,
24 First principle modeling
while simplied, control-oriented models will be the topic of the next chapter.
Some models include very simple chemical kinetic sub-models while trying to
describe ow patterns or thermal behavior in more details, usually applying
3-D models [39, 99, 100, 101]. Most converter models reported in the literature
are based on the steady state reaction kinetics with lumped surface adsorption-
desorption-reaction phenomena and an oxygen storage phenomenon included
as a separate sub-model [23, 55, 70, 75]. These models typically assume one
rate determining step and apply Langmuir-Hinshelwood type of rate equations.
This approach may fail even in the steady state because the rate determining
step does not necessarily remain the same when conditions change. Presence
of other species can also have quite a large inuence. It is well known [23, 59]
that the presence of some hydrocarbons aects the light-o temperature of
carbon monoxide. Therefore, self-inhibition and inhibition terms have to be
accounted for explicitly in reaction rate expressions. Since they do not have a
very large amount of kinetic parameters, such models can be tuned to give a
very good prediction of total emissions during some standard test cycle, but
fail to give more insight into the converter dynamic behavior which is of a great
importance for control.
The approach presented in this thesis is based on the elementary step ki-
netics of individual global reactions. Though numerically more complex, such
models can have a larger impact being able to describe in more detail processes
that are taking place on the catalytic surface. One can simply combine kinetic
models for individual global reactions without a need for further model adap-
tation. It has been shown that these models, even if based on kinetics obtained
from completely uncorrelated literature sources (even completely dierent cat-
alyst formulations have been used), can fairly well predict the global converter
behavior [10, 46, 69]. Another impact of elementary step kinetic models is
that they can also be used in dierent applications, apart from the automo-
tive catalysis. The currently applied converter model comprises kinetic models
of carbon monoxide and hydrocarbon (ethylene and acetylene) oxidation and
NO reduction on the same Pt/Rh/-Al
2
O
3
/CeO
2
catalyst obtained in recent
research [35, 36, 37, 67]. The converter model has been validated with mea-
sured data on an engine test bench. A closer look will be given to the light-o
process and converter dynamic responses to inlet lambda perturbations. This
chapter will demonstrate the application of such a model for optimization of
cold-start strategies (feed oscillations, secondary air injection) and analysis of
converter dynamic behavior that will be the basis for the control design in the
later chapters.
2.2 Mathematical model
The mathematical model of the three-way catalytic converter which will be
presented here basically contains two sub-models: kinetic and reactor model.
2.2. Mathematical model 25
The reactor model deals with the physical characteristics of the converter. It
contains mass and energy balance equations and is standard for some type of
converter, i.e. does not depend on the type of catalyst used. The kinetic model,
on the other hand, refers to the reaction rate equations of a kinetic mechanism
for the catalyst that is being used to coat the converter. It is based on the
experimentally obtained kinetic sub-models over a Pt/Rh/-Al
2
O
3
/CeO
2
cat-
alyst, and determines largely the dynamical behavior of the catalytic converter
of interest for control. The kinetic model will be more thoroughly discussed.
The representative gases for the engine exhaust are carbon monoxide, hy-
drocarbons, hydrogen, oxygen, nitric oxide, carbon dioxide and water. Since
the model for hydrogen oxidation on the applied catalyst has not been obtained,
hydrogen has not been modeled in this study, but accounted for by increasing
the reaction enthalpy of the carbon monoxide oxidation. It is assumed that
the ratio between CO and H
2
is 3:1. The selection of good representatives for
all hydrocarbons in the exhaust gas is not so straightforward. Usually two hy-
drocarbon representatives are modeled, representing slowly and fast oxidizing
hydrocarbon species. Ethylene is the representative of fast oxidizing hydro-
carbon species as it can be found in large amounts in the exhaust [49, 57],
while acetylene represents the slowly oxidizing hydrocarbons. Acetylene is a
well-known inhibitor of the converter light-o at lower temperatures [59] and
it can also be found in the exhaust in substantial quantities especially after the
cold start of an engine [20, 49, 57].
2.2.1 Reactor model
The monolithic reactor is modeled as a one dimensional adiabatic reactor, thus
ambient heat losses are not considered. The radial gas velocity distribution is
assumed to be uniform, though the gas velocity in the outer channels would
be somewhat smaller than in the inner channels due to a short divergent inlet
of the monolith. All channels are assumed to have equal diameters and the
walls of the channels are assumed to be impenetrable to gas. Under above
given conditions only one channel can be modeled as the representative for the
whole converter. Since the Taylor criterion [88] is satised the laminar ow in
the small diameter channel can be approached as plug ow. Since this is a one
dimensional model it considers only axial gradients of reactants concentrations
and temperature. Constant heat and mass transfer coecients based on the
limit values of Nusselt and Sherwood numbers for laminar ow are used to
describe mass and heat transfer from bulk gas to washcoat. The diusion in
the washcoat has been neglected [39]. More detailed modeling assumptions can
be found in [96].
The continuity equation for component i (i=CO, O
2
, NO, N
2
O, NO
2
, C
2
H
4
,
C
2
H
2
, CO
2
, H
2
O) in the bulk gas phase is given by:
26 First principle modeling

t
_
C
f,i

f
_
=
sup
m

x
_
C
f,i

f
_

f
k
f,i
a
v
_
C
f,i

C
s,i

f
_
(2.1)
and in the solid phase (washcoat):

f
4d
w
d
b

t
_
C
s,i

f
_
=
f
k
f,i
a
v
_
C
f,i

C
s,i

f
_
a
cat
(r
a,i
r
d,i
) (2.2)
The last term in the equation (2.2) accounts for the adsorption and des-
orption of the species to and from the noble metal and oxygen storage surface.
The dependent variables are expressed as
C

f
to correct for the density changes
as a function of the axial coordinate due to non-uniform temperatures. The
energy equations in the gas and solid phase are given by:

f
c
pf
T
f
t
=
sup
m
c
pf
T
f
x
a
v
(T
f
T
s
) (2.3)
(1 )
s
c
ps
T
s
t
=
s
(1 )

2
T
s
x
2
+a
v
(T
f
T
s
) +a
cat

j
(
r
H)
j
r
j
(2.4)
The reaction heat generation is accounted for by the last term of equation
(2.4). It is calculated using the rates of global reactions, r
j
. This is allowed
since the heat capacity of the reactor is much higher than the heat production
due to changing surface coverage [96]. The continuity equation for the species j
adsorbed on the noble metal surface or ceria surface can be written as follows:
L
CAP

j
t
= r
a,j
r
d,j
+

k
r
k,j
(2.5)
where k denotes a certain surface reaction which involves the species j, and
L
CAP
stands for L
CS
, L
CB
orL
NM
. The reactor parameter values used in the
simulations are shown in table 2.1. The global reactions that can occur inside
the catalytic converter can be summarized as follows (note that due to the
application of the elementary steps, more reactions are possible, i.e. reactions
between NO and hydrocarbons):
2.2. Mathematical model 27
parameter value parameter value

_
m
3
f
m
3
R
_
0.636 L [m] 0.1524
A
s
_
m
2
R

8 10
3
a
cat
_
m
2
NM
m
3
R

1.25 10
4
a
v
_
m
2
i
m
3
R

2.4 10
3

w
_
m
3
f
m
3
w
_
0.4
d
b
[m
R
] 1.037 10
3
d
w
[m
R
] 4.9 10
5
c
pf
_
Jkg
1
K
1

1.064 10
3
c
ps
_
Jkg
1
K
1

1.000 10
3

f
_
Wm
1
K
1

4.15 10
2

s
_
Wm
1
K
1

1.675

_
Wm
2
K
1

141
s
_
kgm
3
s

1.8 10
3
k
f,CO,O
2
_
m
3
f
m
2
i
s
1
_
0.22 k
f,NO
_
m
3
f
m
2
i
s
1
_
0.25
k
f,C
2
H
4
,C
2
H
2
_
m
3
f
m
2
i
s
1
_
0.16 k
f,CO
2
_
m
3
f
m
2
i
s
1
_
0.17
k
f,H
2
O
_
m
3
f
m
2
i
s
1
_
0.27 k
f,N
2
O,NO
2
_
m
3
f
m
2
i
s
1
_
0.15
Sh,Nu [] 3.66 L
NM
_
molm
2
NM

0.45 10
5
L
CS
_
molm
2
NM

2.5 10
4
L
CB
_
molm
2
NM

2.5 10
3
(
r
H)
r1
_
kJmol
1

283 (
r
H)
r2
_
kJmol
1

373
(
r
H)
r3
_
kJmol
1

1322 (
r
H)
r4
_
kJmol
1

1254
(
r
H)
r5
_
kJmol
1

98.4 (
r
H)
r6
_
kJmol
1

57.2
(
r
H)
r7
_
kJmol
1

381
Table 2.1: Reactor parameters (at 550K) used in simulations. The global
reactions r1 r7 are dened in (2.6).
2CO +O
2
2CO
2
(r1)
2CO + 2NO 2CO
2
+N
2
(r2)
C
2
H
4
+ 3O
2
2CO
2
+ 2H
2
O (r3) (2.6)
2C
2
H
2
+ 5O
2
4CO
2
+ 2H
2
O (r4)
4NO 2N
2
O +O
2
(r5)
2NO +O
2
2NO
2
(r6)
O
2
+ 2Ce
2
O
3
4CeO
2
(r7)
The last reaction involves oxidation and reduction of ceria, and is the reaction
that is most responsible for the dynamic behavior of the converter which is of
interest for control, as will be shown later in the text.
2.2.2 Kinetic model
The kinetic model, which is the basis for the reactor model, is presented in
this section. The model consists of four sub-models describing carbon monox-
ide, ethylene and acetylene oxidation and the reduction of nitric oxide. All
28 First principle modeling
kinetic sub-models have been obtained by transient kinetic experiments over
a xed bed laboratory reactor with the same Pt/Rh/-Al
2
O
3
/CeO
2
catalyst
[38]. These experiments included steps of rich and lean feeds, thus simulat-
ing dynamic cycling conditions that occur under typical converter operating
conditions. The steps were 0-100%, meaning that rst a completely rich mix-
ture would be fed through the reactor (i.e. CO and/or HC only) and then a
completely lean mixture would be fed (i.e. O
2
and/or NO). The kinetic rate
parameters have been acquired via nonlinear multi-response regression of data
with the selected kinetic models [35, 36, 37, 38, 67]. A typical model validation
experiment is shown in gure 2.1. In this experiment the model prediction of
CO oxidation in the presence of O
2
and NO is presented. The experiments
used for obtaining the kinetic models were performed in the absence of steam
and carbon dioxide in the inlet gas feed. It is well known that these gases
form a large fraction of the exhaust gas, and it was also found, [38], that these
exhaust components, especially water, have a substantial inuence on the ki-
netics. Moreover, the model prediction in cases of all components present in the
feed at all time (and superimposed dynamic variations of the inlet equivalence
ratio) showed to be less accurate. Therefore the model presented here has to
be considered as a good, qualitative, approximation of the converter behavior,
and some tests shown in this chapter will demonstrate its accuracy.
The model uses standard expressions for adsorption, desorption and reac-
tion rate calculations. The adsorption rate of a component i on the noble metal
is given by:
r
a,i
= k
a,i
L
NM
C
s,i
(2.7)
and the same expression is used for the adsorption of species (oxygen, nitric
oxide) on the ceria surface, provided that determines the fraction of the
empty ceria surface. The adsorption rate coecients are typically obtained
from the kinetic gas theory with the following equation:
k
a,i
=
1
L
NM
_
RT
2M
i
s
0
i
(2.8)
This theory has not directly been used in the kinetic studies preceding to
this work where xed rate coecients have been determined. In the model pre-
sented here the temperature dependence from the kinetic gas theory has been
preserved to extrapolate the rate coecients obtained at a narrow tempera-
ture range to a wider temperature range needed for a more complete converter
model:
k
a,i
= k
a,i0
_
T
T
0
(2.9)
where the temperature T
0
is the temperature at which the kinetic model for a
2.2. Mathematical model 29
Figure 2.1: Reactor outlet concentrations (markers, measurements; lines, model
predictions) versus time for NO reduction by CO in the presence of O
2
at a
temperature of 523K and an oscillation frequency of 1/10 Hz.
specic reaction has been obtained. The desorption rate of an adsorbed species
is proportional to the degree of surface coverage of that species and is given by:
r
d,i
= k
d,i
L
NM

i
(2.10)
The rates of reactions on the noble metal surface depend on the degrees of
surface coverages of involved species (x,y) in the following manner:
r
r
= k
r
L
NM

y
(2.11)
In the case of a reaction between species on the noble metal and ceria the
reaction rate is calculated in a similar way:
r
r
= k
r
L
NM

y
(2.12)
Note that
y
generally stands for the species on the ceria surface, while
x
generally stands for the species adsorbed on the noble metal. The noble metal
capacity L
NM
is used by denition in this case. As it will be seen in the kinetic
scheme, transfer of species from the surface of ceria to bulk and back is also
modeled as (2.12), only
x
, which represents the species in bulk ceria, replaces

x
. Another type of reactions that occurs in the kinetic model are Eley-Rideal
30 First principle modeling
type of reactions when molecules in the washcoat can adsorb on noble metal
sites already covered with oxygen. The reaction rates for those reactions are
calculated as follows:
r
r
= k
r
L
NM
C
s,x

O
(2.13)
Both desorption and reaction rate coecients are of Arrhenius type:
k
r,d
= A
r,d
exp
_

E
act
RT
_
(2.14)
The complete kinetic model is presented in table 2.2, while the accompany-
ing rate parameters are given in table 2.3.
It is clear that the given model is quite complex as all these steps are
based on the modeling of the individual reactions. It is expected that the
model can be simplied without great loss of accuracy, but this was not the
goal of this research. Steps 1 through 25 are completely related to the noble
metal surface, while steps 26 through 39 account for the inuence of ceria
storage capability. These steps involving ceria will be shown as crucial for
determination of the converter dynamic behavior. The oxidation of carbon
monoxide [67] is described with steps 1 through 5 and 26 through 29. The
rst ve steps account for the so-called monofunctional path, which includes
noble metals only. An adsorbed CO molecule can react with an adsorbed
oxygen adatom via a Langmuir-Hinshelwood type of surface reaction. Another
possibility is the adsorption of gas phase CO on an already adsorbed oxygen
adatom (Eley-Rideal-like step) and subsequent reaction leading to CO
2
. Steps
26 and 27 account for the reaction of CO adsorbed on the noble metal surface
with oxygen adsorbed on the ceria surface. This reaction has a great impact on
the converter performance when there is a shortage of oxygen on the noble metal
surface, i.e. the inlet feed is rich, because it improves CO conversion while there
is enough oxygen stored on ceria. The rate parameters in [67] were determined
at temperatures 390-430K. This is much lower than the temperatures in a
typical automotive application. The extrapolation to higher temperatures may
not always be accurate so the rate parameters obtained in the study of CO
oxidation by NO in the presence of oxygen [37] were applied. This study was
conducted at higher temperatures (525-575K). In the same study the steps 28
and 29 were added. It was observed that CO probably has to be stored on ceria
surface during rich inlet conditions, because of the increased adsorption of CO
on the ceria containing catalyst compared with a catalyst without ceria. This
eect could not be solely explained on the basis of stoichiometry. This step is
questioned in [38] because it may lead to a strange dynamic behavior of the
converter model under some conditions. Namely it can lead to the inhibition of
oxygen adsorption on the ceria surface after a rich-lean switch by the adsorbed
CO. On the other hand, this step can explain CO desorption from the catalyst
after a rich-lean transition that has been observed during engine bench testing
2.2. Mathematical model 31
1 O
2g
+ 2* 2O*
2 CO
g
+ * CO*
3 CO* + O* CO
2
+ 2*
4 CO
g
+ O* OCO* +
5 OCO* CO
2
+ *
6 NO
g
+ * NO*
7 NO* + * N* + O*
8 NO* + N* N
2
O* + *
9 2N* N
2g
+ 2*
10 N
2
O* N
2
O
g
+ *
11 N
2
O* N
2g
+ O*
12 NO
g
+ O* NO
2
*
13 NO
2
* NO
2g
+ *
14 C
2
H
4g
+ 2* C
2
H
4
**
15 C
2
H
4
** C
2
H
4
* + *
16 C
2
H
4
** + 6O* 2CO
2g
+ 2H
2
O
g
+ 8*
17 C
2
H
4
* + 6O* 2CO
2g
+ 2H
2
O
g
+ 7*
18 C
2
H
4g
+ O* C
2
H
4
O*
19 C
2
H
4
O* + 5O* 2CO
2g
+ 2H
2
O
g
+ 7*
20 C
2
H
2g
+ * C
2
H
2
*
21 C
2
H
2
* + 2* C
2
H
2
***
22 C
2
H
2
* + 3O* 2CO* + H
2
O
g
+ 2*
23 C
2
H
2
*** + 3O* 2CO* + H
2
O
g
+ 4*
24 C
2
H
2g
+ O* C
2
H
2
O*
25 C
2
H
2
O* + 2O* 2CO* + H
2
O
g
+ *
26 O
2g
+ 2s 2Os
27 CO* + Os CO
2
+ * + s
28 CO* + s COs + *
29 COs + O* CO
2g
+ * + s
30 NO
g
+ s NOs
31 NO
g
+ Os NO
2
s
32 NO* + s N* + Os
33 C
2
H
2
* + 3Os 2CO* + H
2
O + 3s
34 C
2
H
4
* + 6Os 2CO
2
+ 2H
2
O + * + 6s
35 Os + m Om + s
36 NOs + m NOm + s
37 NOs + Om NO
2
m + s
38 NO
2
s + Om NO
2
m + Os
39 NO
2
s + m NOm + Os
Table 2.2: The complete kinetic model used in the simulations of the rst
principle model. Subscript g denotes a molecule in the gas (washcoat) phase, *
a noble metal site, s a ceria site on the surface, while m denotes a ceria site in
the bulk. Steps 1,16,17,19,22,23,25,26,33 and 34 are rst order with respect to
the second reactant shown. The above kinetic model has been obtained from
various sources as referenced in the text.
32 First principle modeling
k
1f
1.01 10
5
k
2f
9 10
5
A
2b
5.5 10
9
E
2b
67.0
A
3f
2.0 10
7
E
3f
49.1
k
4f
4.61 10
3
A
4b
248 E
4b
20.3
A
5f
6.90 10
14
E
5f
118.28
k
6f
3.63 10
5
A
6b
3.04 10
10
E
6b
83.2
A
7f
2.19 10
5
E
7f
45.8
A
8f
2.16 10
5
E
8f
38.3
A
9f
4.08 10
8
E
9f
56.6
A
10f
2.71 10
6
E
10f
45.3
A
11f
4.64 10
3
E
11f
20.1
k
12f
5.85 10
2
A
12b
2.27 10
3
E
12b
28.6
A
13f
8.66 10
2
E
13f
39.6 k
13b
5.34 10
6
k
14f
1.26 10
6
A
14b
1.20 10
5
E
14b
39.7
A
15f
1.17 10
10
E
15f
84.6 A
15b
4.33 10
8
E
15b
55.3
A
16f
6.25 10
7
E
16f
81.2
A
17f
355 E
17f
11.3
k
18f
14.7 k
18b
6 10
5
A
19f
1.52 10
10
E
19f
78.7
k
20f
1.32 10
7
A
20b
1.11 10
10
E
20b
93.5
A
21f
2.50 10
9
E
21f
44.4 A
21b
2.27 10
11
E
21b
125.0
A
22f
9.35 10
11
E
22f
151
A
23f
2.25 10
5
E
23f
161
k
24f
534 k
24b
5.86
A
25f
9.73 10
3
E
25f
0.50
k
26f
33.3
A
27f
9.62 10
2
E
27f
11
A
28f
7.78 10
17
E
28f
185 A
28b
6.12 10
7
E
28b
51.1
A
29f
3.5 10
5
E
29f
46
k
30f
6.46 10
0
A
30b
2.11 10
2
E
30b
7.60
k
31f
6.46 10
0
A
31b
2.11 10
2
E
31b
7.60
A
32f
1.03 10
17
E
32f
168
A
33f
1.76 10
12
E
33f
124
A
34f
1.74 10
11
E
34f
106
A
35f
1.71 10
9
E
35f
79.7 A
35b
6.08 10
12
E
35b
123
A
36f
2.45 10
4
E
36f
22.4 A
36b
3.35 10
4
E
36b
20.24
Table 2.3: Rate parameters for the kinetic model as shown in table 2.2. The
number in subscript denotes the reaction number as in table 2.2, while f stands
for the foreward step and b stands for the backward step. If the activation
energy and pre-exponential factors are given the rate is calculated by equation
(2.14), otherwise the rate is constant unless it represents adsorption process
(2.9). Reactions 37 39 have the same rate coecients as the reaction 36.
Dimensions: k
_
m
3
mol
1
s
1

, A
_
s
1

, E
_
kJmol
1

. The above parameters


have been obtained from various sources as referenced in the text.
2.2. Mathematical model 33
(see section 2.4.2). Therefore these steps have been kept in the model, but
the parameters have been slightly adapted to overcome the above described
problem.
Steps 20-25 and 33 describe the oxidation of acetylene [36]. The primary
products of oxidation are carbon monoxide adsorbed on the noble metal surface
and water. Various acetylene species (C
2
H
2
, C
2
H
2
), having dierent
reactivity with oxygen exist on the noble metal surface and can reversibly
be converted into each other. Experiments have shown that acetylene can
also adsorb on a noble metal site already covered with an oxygen adatom. A
clear inuence of ceria on the oxydation of acetylene was also experimentally
revealed. Step 33 models that inuence. Oxidation of another hydrocarbon
representative, ethylene, is modeled by steps 14 through 19 and step 34 [35].
The rst ve steps account for the oxidation involving noble metal sites only and
are similar to those of acetylene. Again two dierent surface species (C
2
H
4
,
C
2
H
4
) had to be included to describe the experimental data, together with
the possibility of gas phase ethylene to adsorb on the oxygen covered surface.
No inuence of ceria on ethylene oxidation was observed in [35] because it
was conducted at rather low temperatures (393 and 443K). According to [19,
92], a substantial inuence of ceria on hydrocarbon oxidation can be observed
only at temperatures higher than 570K. The same was concluded in [38] when
the experiments with a complete exhaust mixture were conducted at 573K.
Step 34 involving the oxidation of C
2
H
4
with oxygen stored on ceria, was
therefore added to the model. The rate parameters for the ethylene oxidation
were also taken from [38] because of the more accurate extrapolation to higher
temperatures. In further text total surface coverage of acetylene will be denoted
simply with C
2
H
2
, and that of ethylene with C
2
H
4
.
Steps 6 through 13 describe the reduction mechanism of NO on the noble
metal, while 30-32 account for the inuence of the surface ceria. Nitrogen
containing species coming out of the converter are N
2
, N
2
O and NO
2
. The rst
one is the result of a complete reduction, and therefore a desired product. N
2
O
in the outlet is a product of partial reduction. When an NO molecule adsorbes
on a noble metal site (step 6) it needs an extra empty site for dissociation (step
7). When there are enough empty sites it is very likely that N
2
will be created
by recombination of two N* adatoms (step 9). This process becomes activated
at somewhat higher temperatures. The other path is a reaction between NO*
and N* (step 8), which is more likely if there is a lack of the empty noble metal
surface. This reaction will lead to an adsorbed N
2
O molecule which can simply
desorb (step 10) or further dissociate to create N
2
(step 11). An NO molecule
can also adsorb on a noble metal site already covered by oxygen, which can
lead to formation of nitrogen dioxide, NO
2
. It was shown in experiments that
NO can adsorb on both empty and oxygen covered ceria surface (steps 30,31).
However, ceria is not an eective NO storage because a desorption occurs even
under lean conditions. A very important step is NO bifunctional dissociation
34 First principle modeling
by help of ceria (step 32). This step proves to be very important under dynamic
operating conditions, during ceria lling and emptying, as will be shown later.
Steps 35-39 account for the transfer of species from the ceria surface to bulk.
As it was found in [37] the storage capacity of ceria bulk increases with the
temperature and is therefore modeled as an activated process (2.14). The sur-
face capacity, on the other hand is kept constant in the model. It was observed
experimentally [38, 50] that a large heat amount is released after a rich to lean
step, meaning that the oxygen storage on ceria is a very exothermal reaction.
The heat release is more notably seen at higher temperatures. Therefore, the
process of oxygen storage in the ceria bulk is modeled as a reaction rather than
diusion.
If was found during the kinetic studies that in the presence of hydrocarbons
some noble metal sites are covered with carbonaceous species and accessible
only to carbon containing components. They contribute to 10% of total noble
metal sites in simulations.
Some rate parameters (see table 2.3) used in the model do not coincide
with the parameters reported in the kinetic studies. As already mentioned, the
changes have been performed because the kinetic experiments were performed
at rather low temperatures, and thus errors can occur when extrapolating the
model to higher temperatures. The parameter update was performed in such
way that the rate value does not change signicantly at temperatures of kinetic
experiments, but at the same time that the performance of the converter at
higher temperatures (i.e. engine bench tests) can be described with sucient
accuracy. Another eect that had to be taken into account is the presence
of steam and carbon dioxide in the automotive exhaust in large quantities.
Namely, the kinetic parameters were obtained in the absence of those compo-
nents and it was shown in the conclusion of [38] that their inuence cannot be
neglected. Apparently, the presence of H
2
O and CO
2
speeds up the adsorption
of oxygen on the ceria surface but decreases the ability of NO to adsorb on
ceria. Therefore the kinetic rate parameters of forward reactions in steps 26,
30 and 31 were adapted (increased in the case of O
2
and decreased in the case
of NO).
Figure 2.2 presents a schematic view of the modeling procedure. A simple
case of CO oxidation is shown, which takes into account monofunctional and
bifunctional path (steps 1-3, 26-27).
Although all oxides of nitrogen are of interest, in many simulations only the
outlet concentration of NO will be presented, if outlet concentrations of NO
2
and N
2
O can be neglected.
2.2.3 Numerical procedure
The model variables are the concentrations of the components in the gas phase
and in the pores of the washcoat, the catalyst surface coverages and tempera-
2.3. Cold start 35
Figure 2.2: Schematic view of the mass balance model in the case of CO oxi-
dation. Only the bifunctional path is shown.
tures in gas and solid phase, all of them along the reactor axis. This leads to
a system of nonlinear, partial dierential equations (PDE) to be solved, which
is quite a complex task. To relax the problem complexity the method of lines
with the discretization in the axial direction has been applied to transform
the system into a larger set of ordinary dierential equations (ODE), which
is then solved using Backward Dierentiation Formulae (BDF) with variable
order and variable size [66]. The Jacobian of the system has been, in some
cases, analytically calculated to achieve a better numeric stability and speed
up the calculation.
2.3 Cold start
Cold start performance is one of the most important characteristics of a three-
way catalytic converter because the majority of exhaust emissions occur while
the converter has not reached the for conversion required temperature, so called
light-o temperature. Light-o temperature is dened as a temperature at
which the conversion assumes 50%. If one can reduce the light-o temperature,
overall emission reduction can be signicant. Therefore it is also very important
that the converter model can describe this phenomenon up to a satisfactory
accuracy. The model can then be used to search for measures to improve
the cold start performance of the converter. A model prediction of a light-
o test measured on an engine test bench will rst be presented. Thereafter,
the analysis of the light-o process and important factors that inuence the
36 First principle modeling
converter performance will be given.
2.3.1 Model verication at cold start conditions
The engine bench tests were performed at dmc
2
, Hanau, Germany. The tests
were performed with the monolith coated by the same catalyst powder as used
in the kinetic experiments on basis of which the kinetic model was obtained
[38]. The parameters of the reactor are given in table 2.1. The engine used in
the experiments was a BMW 1.9L gasoline engine. Two tests were performed:
light-o tests with stoichiometric and lean ( 1.15) feeds, which were kept
constant during the tests. Inlet and outlet concentrations have been measured
with gas analyzers. Also temperatures have been recorded. The temperature
sensor behind the catalytic converter has not been placed directly behind the
converter so this measurement will be disregarded (heat loss of the exhaust pipe
would have to be taken into consideration to properly interpret this signal).
The inlet ow is kept constant in both experiments at 110kg/h. It should be
noted that these light-o tests do not accurately represent typical operating
conditions for the converter because the slope of the temperature ramp is very
low. Under standard conditions the catalyst inlet temperature would reach the
typical operating conditions after only 1-2 minutes. The only adapted model
parameter was the amount of the noble metal sites, as its value was not known
in advance.
Figure 2.3 shows the inlet temperature, as well as measured and predicted
concentrations of CO, total HC and NO
x
under stoichiometric conditions. Fig-
ure 2.4 shows the same responses under lean conditions. The importance of the
converter light-o time (temperature) is clearly visible from these experiments
as there is no conversion until the light-o temperature is reached what causes
the majority of emissions during a standard driving cycle. The problem may
also be caused by a frequent start-stop driving such as city driving, where the
engine exhaust gas temperature is kept low (i.e. idle conditions) and the cata-
lyst temperature can drop below the light-o temperature. After the light-o a
very high conversion is possible if the operating conditions for the converter are
being kept at optimum (i.e. stoichiometry). Under lean conditions a very low
conversion of NO is achieved. The light-o temperature of CO in the stoichio-
metric case is at approx. 600K, while in the lean case it is reduced to approx.
570K.
The model prediction in both cases is fairly good. The predicted light-
o temperature for hydrocarbons is slightly too high in the lean case, while
slightly too low in the stoichiometric case. Interestingly, acetylene light-o
matches the light-o of hydrocarbons in the lean case, while the ethylene light-
o matches the measured light-o in the rich case. While this is probably a pure
coincidence it shows that various hydrocarbons can have very dierent behavior
and by modeling more of them more accurate models would be possible. This
2.3. Cold start 37
0 200 400 600
450
500
550
600
650
700
time [s]
t
e
m
p
e
r
a
t
u
r
e

[
K
]
0 200 400 600
0
0.2
0.4
0.6
0.8
1
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
0 200 400 600
0
1000
2000
3000
4000
time [s]
N
O

c
o
n
c
.

[
p
p
m
]
0 200 400 600
0
100
200
300
400
500
time [s]
H
C

c
o
n
c
.

[
p
p
m
]
Figure 2.3: Engine bench light-o experiment with the stoichiometric feed.
Thin line - converter inlet, thick line - converter outlet, dashed line - predicted
converter outlet.
would, however, increase the model complexity. In the case of CO both light-o
temperatures are predicted well. The conversion of NO in the stoichiometric
case was predicted to start slightly later than in the experiments while the light-
o time (50% of the conversion) was predicted well. The low NO conversion
at higher temperatures was well predicted in the lean input case. A small
peak in the NO conversion in the lean case was not predicted, however. This
peak in the NO conversion was predicted by some earlier, simpler NO models
[9, 47] meaning that still some ne tuning of NO reduction kinetic parameters
is necessary.
A perfect t between the model and experiments would hardly be expected
merely because of the approximation of total hydrocarbons by two represen-
tatives ethylene and acetylene (each was assumed to represent 50% of the hy-
drocarbons). This is clearly seen in the lean case at higher temperatures when
a fraction of hydrocarbons is not converted. This is attributed to alkanes [13],
which where not modeled in this study. The model therefore predicts a com-
plete conversion of hydrocarbons. Since this eect takes place only at very
lean conditions, it will not have much eect here as the simulated conditions
are typically around stoichiometry. If, on the other hand, a catalyst for a
lean burn engine would have to be modeled, the model would need to account
38 First principle modeling
0 200 400 600
450
500
550
600
650
time [s]
t
e
m
p
e
r
a
t
u
r
e

[
K
]
0 200 400 600
0
0.2
0.4
0.6
0.8
1
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
0 200 400 600
0
500
1000
1500
2000
2500
3000
time [s]
N
O

c
o
n
c
.

[
p
p
m
]
0 200 400 600
0
100
200
300
400
500
time [s]
H
C

c
o
n
c
.

[
p
p
m
]
Figure 2.4: Engine bench light-o experiment with the lean feed (=1.15).
Thin line - converter inlet, thick line - converter outlet, dashed line - predicted
converter outlet.
also for alkanes. Other eects that inuence the model accuracy stem from
conditions used to asses the kinetic model (eects of water and noble metal
oxidation [38]). The model thus serves as a good tool for qualitative analysis
of the converter light-o. The light-o process will be analyzed in more detail
in the following section. A model application in the assessment of cold start
control strategies will be studied in sections 2.3.3 and 2.3.4.
2.3.2 Light-o with steady inlet feed
The processes taking place on the catalytic surface will closely be investigated
in this section. Simulations with lean(=1.02), rich(=0.98) and stoichiometric
steady feeds will be compared. The inlet gas concentrations as function of
are depicted in gure 2.5. It is well known that the NO inlet concentration
increases with the increase of the engine load, so NO inlet concentration is set
to be a function of the exhaust mass ow (thus is the oxygen concentration also
a function of the exhaust mass ow). All simulations were performed with an
exhaust mass ow of 54kg/h. The inlet temperature is increased linearly from
300 to 650K in 30s after which it stays constant.
Conversions of all species during the light-o with various feeds are depicted
2.3. Cold start 39
0.94 0.96 0.98 1 1.02 1.04 1.06
0
0.5
1
1.5
2
lambda []
c
o
n
c
e
n
t
r
a
t
i
o
n

[
v
o
l
%
]
CO
O
2
0.94 0.96 0.98 1 1.02 1.04 1.06
0.05
0.1
0.15
0.2
0.25
0.3
lambda []
c
o
n
c
e
n
t
r
a
t
i
o
n

[
v
o
l
%
]
NO
HC
Figure 2.5: Inlet gas concentrations as a function of inlet value as used in
the simulations. Exhaust mass ow is 54kg/h.
in gure 2.6. As expected, only the stoichiometric feed leads to a full conversion
of all species under warmed-up conditions. The light-o occurs in a sequence.
First acetylene and CO are converted, followed by the conversion of NO and
ethylene. A lean feed benets the oxidation of CO and HC, while a rich feed
benets the reduction of NO. The light-o temperature of CO is lowered with a
leaner feed. Acetylene light-o is not inuenced by the feed composition at all,
while ethylene benets mostly from the stoichiometric feed. Ethylene light-o
occurs faster with a rich feed, but the steady state conversion with the lean feed
is higher. NO conversion with the rich feed starts slightly sooner then with the
stiochiometric feed, which also needs a much higher temperature to completely
reach the full conversion, while the lean feed does not lead to the light-o of
NO at all because the conversion remains below 20%. It is already clear that
the control of NO conversion will be extremely dicult because of the sudden
fall of the conversion eciency once the inlet feed becomes only slightly lean.
Figure 2.7 shows the gas concentration and the noble metal surface coverage
in the middle of the reactor and with the stoichiometric feed. The light-o
sequence, that can be observed also from the gas concentrations in the middle
of the reactor, can be understood on the basis of surface dynamics. It was
assumed in this study that oxygen covers most of the catalytic surface at the
start of the simulation. That assumption is reasonable, since before the cold
40 First principle modeling
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
C
O

c
o
n
v
.

[
%
]
stoich.
rich
lean
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
N
O

c
o
n
v
.

[
%
]
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
C
2
H
4

c
o
n
v
.

[
%
]
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
C
2
H
2

c
o
n
v
.

[
%
]
Figure 2.6: Light-o of all species under various steady air-fuel ratios. The
inlet temperature increases from 300 to 650K in 30s, and the exhaust mass
ow is 54kg/s.
start the catalyst had been exposed only to the ambient air for a long period of
time. After the engine start CO adsorbs on the oxygen covered surface creating
OCO* species. When the desorption and reaction of OCO* have ignited (steps
4b,5 in table 2.2), C
2
H
2
becomes the strongest competitor for the noble metal
surface and inhibits the reactions of the other species. Just when acetylene
starts to react, CO and oxygen have easier access to the noble metal surface
and CO conversion also starts. Ethylene has more diculties to reach the noble
metal, and when it does it reacts with oxygen very easily. Therefore there is
no large accumulation of ethylene observed. Though NO can reach the surface
easier than ethylene, its reduction also starts just after the majority of CO
and acetylene have reacted. The reason for that is that NO needs more empty
sites than other species. It rst has to adsorb (table 2.2, step 6) and then has
to dissociate (step 7). The dissociation process requires an extra empty site.
Though NO can also adsorb on an already oxygen covered site, this step can
lead only to formation of NO
2
, but not to NO conversion. After all the species
have been converted, oxygen slowly accumulates on the noble metal surface.
Very similar processes take place on the noble metal surface under the lean
and rich inlet feeds (gure 2.8). OCO* covers the majority of the noble metal
surface right after the start, to be succeeded later by acetylene. In the rich case
2.3. Cold start 41
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
time [s]
s
u
r
f
.

c
o
v
.

[

]
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
c
o
n
c
e
n
t
r
a
t
i
o
n

[
v
o
l
%
]
CO
NO
C
2
H
4
C
2
H
2
OCO* C
2
H
2

O*
NO
2
* NO*
CO*
Figure 2.7: Gas composition (above) and noble metal surface coverage (below)
in the middle of the reactor during the light-o test with the stoichiometric
feed.
hydrocarbons and CO cover most of the surface at higher temperatures, while
oxygen covers most of the surface in the lean case. There is almost no NO
accumulated in the lean case because oxygen is a stronger competitor under
those conditions. Also, a large increase of NO
2
* is observed during the lean
experiment. This is a strong evidence that oxygen reaches the surface before
NO. CO light-o is somewhat improved in the lean case, also due to a larger
availability of oxygen on the noble metal surface. This eect is much more
pronounced under very lean conditions as already seen in the engine bench
experiment (gure 2.4). Oxygen, however, hinders the oxidation of ethylene
under lean conditions at lower temperatures as it reaches the noble metal sur-
face easier and inhibits the adsorption of ethylene. Such eect probably causes
the reduced hydrocarbon conversion under very lean conditions as measured
by engine bench tests (gure 2.4). This cannot be predicted by the model be-
cause ethylene conversion is not inhibited at higher temperatures. Therefore,
more hydrocarbons should be modeled for the model to properly predict the
converter operation under very lean ( >1.1) conditions, such as in lean burn
or diesel engines.
Figure 2.9 shows that the inuence of initial conditions (assumed initial
surface coverage at 300K) on the light-o is negligible. The cases with the
42 First principle modeling
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
time [s]
s
u
r
f
.

c
o
v
.

[

]
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
time [s]
s
u
r
f
.

c
o
v
.

[

]
C
2
H
2

OCO*
NO*
NO
2
* CO*
C
2
H
4

C
2
H
2

OCO*
O*
NO
2
*
Figure 2.8: Noble metal surface coverages during the light-o with rich (above)
and lean(below) feeds.
surface initially mostly covered with oxygen and acetylene are presented. In
the latter case almost no OCO* formation is observed, but this hardly inuences
the CO light-o.
The inhibition eect is studied further in gure 2.10. Nominal CO con-
version during the warming up with the stoichiometric feed is compared to a
case with acetylene amounting to only 10% of hydrocarbons, and a case for
which the NO inlet concentration was decreased with 80%. It has already ex-
perimentally been observed that the presence of NO and some hydrocarbons
increases the light-o temperature of CO [23, 38, 59]. This occurs due to sur-
face inhibition and can be well predicted by the model. Here lies the impact of
the elementary step kinetic approach because one does not have to separately
account for the inhibition process as it is inherently included in the model. The
simulation shows that the CO light-o speeds-up when either acetylene or NO
concentration in the inlet feed is reduced. The eect of acetylene reduction
is more profound, however. Ethylene conversion improves when less acetylene
is present in the feed for the same reasons as CO conversion. NO reduction,
however, does not benet ethylene conversion because of a larger concentra-
tion of oxygen that can inhibit ethylene adsorption. For the same reason the
conversion of NO reduces with the decreasing inlet concentration of NO. Lower
acetylene inlet concentration also does not benet the conversion of NO as
2.3. Cold start 43
0 10 20 30 40 50 60 70 80 90 100
0
0.5
1
1.5
2
C
O
o
u
t

[
v
o
l
%
]
init. O*
init. C
2
H
2
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
time [s]
s
u
r
f
.

c
o
v
.

[

]
C
2
H
2

O*
NO
2
*
NO*
CO*
OCO*
Figure 2.9: Comparison of the system warm-up responses with the noble metal
surface initially covered with oxygen and acetylene. Above: CO outlet concen-
tration in both cases. Below: surface coverage in the middle of the reactor in
the case of initial acetylene coverage; the case with initial oxygen coverage as
shown in gure 2.7.
it competes for the noble metal surface with ethylene whose concentration is
increased in that case. Surface inhibition phenomena is thus the most impor-
tant eect to understand the light-o process of the catalytic converter. By
modeling single reactions, i.e. only CO+O
2
, the light-o temperature cannot
correctly be estimated and a wrong picture about the converter operation can
be established.
The inuence of the exhaust mass ow is depicted in gure 2.11. The
conversions in the nominal case (54 kg/h) are compared to situations with
mass ows of 27 and 108kg/h. The conversion improvement with higher mass
ows stems from the faster converter warming up due to a higher thermal mass
entering the converter. Catalyst surface temperature at the end of the reactor,
increases much faster in the case of the high mass ow.
2.3.3 Light-o with oscillatory inlet feed
One of the most common features of todays vehicles with three-way catalytic
converters are oscillations of the inlet lambda value around stoichiometry. As
already explained in the previous chapter, these oscillations stem from the
44 First principle modeling
20 30 40 50 60 70
0
20
40
60
80
100
time [s]
C
O

c
o
n
v
.

[
%
]
20 30 40 50 60 70
0
20
40
60
80
100
time [s]
N
O

c
o
n
v
.

[
%
]
20 30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
4

c
o
n
v
.

[
%
]
20 30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
2

c
o
n
v
.

[
%
]
nominal
C
2
H
2
10%
NO 20%
Figure 2.10: Eect of acetylene (10% of total HC and nominal) and NO (20% of
nominal and nominal) concentration on the converter light-o for stoichiometric
feed.
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
C
O

c
o
n
v
.

[
%
]
54 kg/h
108 kg/h
27 kg/h
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
N
O

c
o
n
v
.

[
%
]
0 20 40 60 80 100
0
20
40
60
80
100
time [s]
H
C

c
o
n
v
.

[
%
]
0 20 40 60 80 100
300
400
500
600
700
800
time [s]
t
e
m
p
e
r
a
t
u
r
e

[
K
]
Figure 2.11: Eect of various exhaust mass ows on the converter light-o and
converter surface temperature in the middle of the reactor.
2.3. Cold start 45
switching type lambda sensor and control loop delays. These oscillations were
inevitable due to hardware restrictions, but were found by some authors [53, 83,
84, 86, 89] to be benecial for converter performance. The largest improvements
were observed under non-automotive conditions, such as with synthetic inlet
gas CO-O
2
or CO-NO mixtures. Muraki and Fujitani [64] showed that the
light-o of CO-NO reaction over a supported platinum catalyst can be reduced
by 150K by oscillations. In more complex mixture studies [65, 86, 89] the
feed oscillations reduced light-o temperatures of NO and HC. The impact
of oscillations was, however, lower then with more simple mixtures. Similar
conclusions were also drawn from the rst modeling studies with literature
based kinetic models [58, 68, 69]. The impact of oscillations is dependent on
the type of the catalyst [86]. The general conclusion from these studies is that
the benet from oscillations is the largest at lower temperatures and that the
converter light-o may be reduced by oscillations. After the light-o oscillations
have very often a detrimental eect on the converter performance.
The light-o model predictions with oscillating feeds are presented in this
section. Typical engine management systems impose lambda oscillations of
0.5-2Hz with an amplitude around 2%. The converter outlets with oscillations
present are averaged on the basis of the oscillation period. Figure 2.12 shows
that the eect of oscillations on the converter light-o is very low. Oscillations
do not deteriorate the light-o but also do not improve it. Interestingly, ethy-
lene and NO outlets oscillate without a phase lag during the light-o, indicating
that a rich feed improves conversions of both species. This is in line with the
light-o characteristics of these species under lean and rich inputs. Of course,
only rich feed does not benet ethylene conversion so periodic oxygen pulses,
stemming from lean excursions, are necessary. Test with rich bias (mean value
of oscillations is slightly shifted to rich region) are shown in gure 2.13. NO
and ethylene light-os are slightly improved in this test, as expected. It was
found that the bias should not be too large in order not to deteriorate the CO
conversion. This eect can be seen in the steady state with the bias of 0.005.
Frequency again does not substantially inuence the light-o, while larger am-
plitudes are also not benecial because of too large lean or rich excursions.
The eects of oscillations will further be discussed on a previously developed
model with a simpler NO kinetic submodel. This model was developed before
the one shown in this thesis. The NO reduction submodel consists of steps 6
to 9 in table 2.2, with step 8 leading directly to N
2
instead of N
2
O like in the
present model. Steps that describe creation of NO
2
and interactions between
NO and ceria were not included in the model. The description of the complete
kinetic model and rate parameters can be found in [4, 9]. Only surface ceria
with a xed capacity for oxygen storage is assumed to play a role, and CO
adsorption on ceria (steps 28,29) are modeled was not considered. This model,
when used, will be specied as model II. The light-o curves with steady and
oscillating feeds are shown in gure 2.14. The temperature increases from 300
46 First principle modeling
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
O

c
o
n
v
.

[
%
]
1%, 1Hz
2%,1Hz
1%,0.5Hz
stoich.
30 40 50 60 70
0
20
40
60
80
100
time [s]
N
O

c
o
n
v
.

[
%
]
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
4

c
o
n
v
.

[
%
]
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
2

c
o
n
v
.

[
%
]
Figure 2.12: Light-o characteristics of a steady, stoichiometric feed, compared
to oscillating feeds with dierent frequency and amplitude.
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
O

c
o
n
v
.

[
%
]
30 40 50 60 70
0
20
40
60
80
100
time [s]
N
O

c
o
n
v
.

[
%
]
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
4

c
o
n
v
.

[
%
]
30 40 50 60 70
0
20
40
60
80
100
time [s]
C
2
H
2

c
o
n
v
.

[
%
]
1%,0.5Hz,0.002off
1%,0.5Hz,0.005off
stoich.
Figure 2.13: Light-o improvement of C
2
H
4
and NO, resulting from feed oscil-
lations with a slightly rich bias.
2.3. Cold start 47
Ampl. 2% 2% 2% 2% 5% 1%
Freq. stoic. 0.67Hz 0.33Hz 1.33Hz 2.67Hz 0.67Hz 0.67Hz
CO 91.1 91.6 89.4 91.5 91.9 92.3 91.1
NO 92.5 93.1 91.8 93.2 93.2 93.8 92.8
C
2
H
4
125.1 114.3 115.3 113.4 112.2 118.7 111.7
C
2
H
2
95.1 95.4 93.4 95.5 95.5 94.7 95.3
Table 2.4: Light-o times [s] under oscillating and steady (stoichiometric) inlet.
Simulations performed with modelII.
to 550K in the rst 100s and then remains constant. The inlet ow is 54
kg/h. Acetylene forms 10% of hydrocarbons. Again, oscillations clearly do not
improve the light-o characteristics of CO, NO and acetylene. Ethylene light-
o is, on the other hand, notably improved with the oscillating feed. Light-o
times in cases of dierent oscillating feeds are shown in table 2.4. An oscillation
amplitude of 2% shows to be optimal for the reaction ignition, but a smaller
amplitude of 1% is better to be applied after the conversion starts. Frequency
is, interestingly, not a crucial factor for the light-o time of ethylene. Low
frequencies are somewhat less benecial for the conversion. Note in gure 2.14
that with the oscillation amplitude of 5% conversions of CO and NO have
diculties to reach 100% because ceria, with its oxygen storage properties,
does not have large enough capabilities to buer the oscillations. The eects
of ceria will be discussed thoroughly in the next section.
The benecial eect of the oscillating feed on the ethylene conversion can
be explained by gure 2.15. Noble metal surface coverage at 4.4 cm from
the reactor inlet and ethylene outlet concentration, in a time interval when
feed oscillations improve the conversion, are shown for the stoichiometric and
oscillating (A=2%, f=0.67Hz) feed. The main dierence, which also leads to
the ethylene conversion improvement, lies in the noble metal surface coverage
by oxygen. While with the stoichiometric feed oxygen accumulates on the
surface because it has a stronger anity to reach the surface than ethylene,
the oscillating feed leads to the interchanging of oxygen and ethylene on the
surface during lean and rich periods. The conversion maximum occurs at the
interchange between rich and lean feed when both components have access to
the noble metal surface. This is generally the case when an oscillating feed can
improve the conversion: if one component has a stronger ability to reach the
surface when the feed is lean, while another component occupies the surface
when the feed becomes rich, an oscillating feed can help both components to
reach the surface and improve the rate of conversion. Therefore, the conversion
of ethylene is improved mostly when the inlet feed is rich, as ethylene then has
a stronger anity than oxygen to access the noble metal, and yet there is
enough oxygen to promote the oxidation process. This is opposite to the eect
that is observed at higher temperatures where lean inlet benets the oxidation
48 First principle modeling
Figure 2.14: Light-o characteristics with steady, stoichiometric, and oscillat-
ing feeds: thin line - stiochiometric, medium thick line - A=1%, f=0.67Hz, thick
line - A=2%, f=0.67Hz, dashed line - A=2%, f=1.33Hz, dotted line - A=5%,
f=0.67Hz. Simulations with model II.
of ethylene, as intuitively expected. The light-o of other components is not
improved by oscillating the feed because the inhibiting species, C
2
H
2
and NO
on the noble metal surface, are not aected by oscillations and do not let other
species access the surface. The inhibiting eect of NO in model II is somewhat
larger that in the model used in the rest of the thesis.
Feed oscillations can thus improve the conversion in principle only at lower
temperatures by lowering the light-o temperature of a certain exhaust com-
ponent. As seen from the given examples and various experimental work, this
improvement will probably depend on the applied catalyst and in general will
not always exist. The benet of oscillations is much lower in complex exhaust
mixtures than in simple binary mixtures. On the other hand, feed oscillations
do not deteriorate light-o either, so their application at low temperatures can
be recommended. To select a proper amplitude, frequency and possible bias,
experimental work on a specic catalyst should be performed.
The benets of oscillations at low temperatures are based on noble metal
surface dynamic eects. After reactions have been ignited the origin of the
main dynamic eect is shifted to the oxygen storage capability of ceria.
2.3. Cold start 49
Figure 2.15: Eects leading to the ethylene conversion improvement with oscil-
lating feed. Above - outlet concentrations with stoichiometric and oscillating
(A=2%,f=0.67Hz) feed. Middle - surface coverage at 4.4 cm with stoichiomet-
ric feed. Below - surface coverage at the same position with the oscillating
feed. Note the horizontal scale: during the light-o of C
2
H
4
and just after the
light-o of other components. Simulations performed with model II.
2.3.4 Light-o under lean conditions - secondary air in-
jection
The engine bench experiment, discussed in section 2.3.1 gure 2.4, has shown
that by applying a very lean feed the light-o temperatures of CO and hy-
drocarbons can be reduced. However, such highly lean lambda values (>1.1)
can hardly be achieved on standard SI engines (excluding lean burn engines)
without causing driveability problems. Moreover, engines are often run rich
during the warm-up phase, though modern engines have capabilities of a lean
start. Therefore, a standard measure applied to achieve lean inlet feed to the
catalyst, even under rich engine conditions, is to apply so called secondary air
injection. Air (sometimes hot) is injected directly into the exhaust manifold,
close to the exhaust valves, where the exhaust mixture is the hottest [52, 54].
By introducing additional oxygen to the exhaust gas mixture exothermal oxi-
dation of CO and hydrocarbons can occur already in the exhaust manifold due
to the very high gas temperature. This is also benecial for the converter light
of as it increases the exhaust gas temperature.
The secondary air injection is included in this study in a simple manner
50 First principle modeling
disregarding the exothermal reaction in front of the converter as well as ex-
haust cooling by the injected air. Experiments have shown [54] that the net
product of these two eects is in favor of exhaust temperature increase, so in
a correct model the eect of secondary air would be even more benecial than
presented here. The increase of the exhaust gas mass ow, and dilution of
exhaust components with the additional air were taken into account.
Figure 2.16 compares the light-o of all species with and without the sec-
ondary air. The base engine tuning when the additional air is applied is slightly
rich, with =0.995. It will later be explained why this tuning is chosen. When
the air is not applied the tuning is stoichiometric. In section 2.3.2 it was shown
that stoichiometric tuning is the optimal light-o tuning when all components
are considered. Of course, if a rich tuning is necessary, the secondary air is
even more benecial so this case will not be shown here. The gure shows
an improvement in CO light-o, while acetylene light-o characteristic barely
changes. The CO light-o improvement can be explained on the basis of chem-
ical kinetics. One of the problems at low temperatures is that oxygen has
diculties to reach the noble metal surface because acetylene and CO have
better adsorption capabilities. By introducing more oxygen via secondary air
injection, more oxygen can reach the catalyst surface, because the adsorption
rate is proportional to the washcoat concentration of oxygen (2.7). Acetylene
oxidation is not only inhibited by the lack of oxygen, but also by a high activa-
tion energy of the oxidation rate coecient. Therefore, even if oxygen reaches
the catalytic surface the reaction will still not be activated. On the other hand,
CO is mostly inhibited by a lack of oxygen on the catalytic surface, and its own
lower ability to adsorb when acetylene is present. Therefore, an additional oxy-
gen leads to an improvement of the CO light-o. The model predicts a slightly
deteriorated light-o of ethylene and of course a very low NO conversion under
lean conditions. While the latter is obvious, the model prediction of ethylene
conversion should be taken cautiously when generalizing to other hydrocarbons
(section 2.3.1).
The improved CO light-o also leads to a faster internal heating of the
converter due to the released heat of reaction. During the warming-up of the
reactor, reactions mostly occur in the warmer, front part of the reactor, so the
reaction heat will mainly be released in the front part as well.
After the reactions have been ignited, a high inlet lambda resulting from
the secondary air injection is not favorable any more because the catalytic
surface becomes extensively covered with oxygen, and NO conversion is very
low. In this situation stoichiometric operation becomes optimal. Thus, when
the reaction ignition occurs, the inlet lambda should be decreased back to
stoichiometry. This may lead to an improvement in the conversion of all com-
ponents. However, the problem is to detect the ignition point. A possibility
would be to detect a larger temperature increase due to the released heat of
reaction, but the thermal processes are rather slow and the reactions occur
2.3. Cold start 51
30 40 50 60
0
0.2
0.4
0.6
0.8
time [s]
C
O
o
u
t

[
v
o
l
%
]
30 40 50 60
0
0.02
0.04
0.06
0.08
0.1
0.12
time [s]
N
O
o
u
t

[
v
o
l
%
]
30 40 50 60
0
0.01
0.02
0.03
0.04
0.05
0.06
time [s]
C
2
H
4
o
u
t

[
v
o
l
%
]
30 40 50 60
0
0.01
0.02
0.03
0.04
0.05
0.06
time [s]
C
2
H
2
o
u
t

[
v
o
l
%
]
stoich
sec. air
reaction starts
testing oxygen storage
Figure 2.16: Light-o with and without (stoichiometric feed) secondary air
injection. The base engine tuning during the applied air injection is = 0.995.
The outlet concentrations during air injection periods are normalized to exclude
dilution eects so the signals can directly be compared. Exhaust mass ow is
27kg/h.
mostly in the front of the reactor during the light-o. By the time the temper-
ature increase reaches the end of the reactor the detected moment for switching
back to stoichiometry would occur too late.
Another possibility to detect the reaction ignition is to use the downstream
sensor. When there is no conversion, the signals in front of and behind
the converter are equal. After the conversion start the signals are not the same
during a dynamic reactor operation. The main dynamics stems from the oxy-
gen storage on ceria and will in detail be studied in the next section. When
a switch from lean to rich is made the excess CO and HC can temporarily be
converted by oxygen stored on ceria. This can be detected by the downstream
signal which will enter the rich region slower than the signal upstream the
converter. Such lean to rich steps can be made by imposing short breaks in air
injection and with a base rich engine tuning. Once a dierence between the up-
stream and downstream signal after a switch is detected, and this dierence
is larger than some predened threshold value, the ignition has taken place and
the converter is operating. The secondary air injection can be switched o and
the stoichiometric operation applied. Also oscillating feed as discussed in the
previous section can be temporarily applied. In the case shown in gure 2.16
52 First principle modeling
0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04
0
10
20
30
40
50
60
70
80
90
100
lambda []
c
o
n
v
e
r
s
i
o
n

[
%
]
CO
NO
C
2
H
2
C
2
H
4
Figure 2.17: Steady state conversion as predicted by the model. The applied
inlet temperature is 650K, and the exhaust mass ow is 50kg/h.
the switch happens 36.5s after the cold start. Due to increased reactor tem-
perature after the early ignited CO reaction, both NO and ethylene light-os
are improved when compared to the stoichiometric case without the secondary
air.
2.4 Warmed-up converter
2.4.1 Steady state operation
A well known characteristic of a three-way catalytic converter is the optimal
conversion of all species when the reactors feed composition is around the
stoichiometry. The model prediction of steady state conversion at the inlet
temperature of 650K and exhaust mass ow of 50kg/h is shown in gure 2.17.
As expected, lean inlets favor CO and HC conversion, while NO conversion
is favored with slightly rich inlets. When the inlet becomes very rich the NO
conversion drops. With such a static conversion characteristic a straightforward
control solution is to control the engine air/fuel ratio at stoichiometry. It should
be noted, however, that the NO conversion is slightly below 100% with the
stoichiometric feed. It will be shown in the next section, that the stoichiometric
feed is not always the best solution if the dynamic behavior of the catalytic
converter is considered.
2.4. Warmed-up converter 53
0 100 200 300 400 500 600 700 800
0
0.5
1
C
O

[
v
o
l
%
]
0 100 200 300 400 500 600 700 800
0
0.1
0.2
0.3
0.4
N
O

[
v
o
l
%
]
0 100 200 300 400 500 600 700 800
0
0.02
0.04
0.06
time [s]
H
C

[
v
o
l
%
]
Figure 2.18: Measured vs. predicted outlet concentrations during a sweep test
on an engine test bench with the inlet temperature of 670K and mass ow of
110 kg/h. Inlet - thin solid line, measured outlet - dashed line, predicted outlet
- thick solid line.
The model static prediction was validated via engine bench tests with the
same engine and catalytic converter as in the light-o validation. The tests
were actually sweep tests, but the sweep rate was very slow compared to the
converter dynamics, so the tests can be considered as quasi steady state tests.
The inlet temperature was 670K and the exhaust mass ow 110kg/h. Fig-
ure 2.18 compares the model prediction with the measured outlet. The overall
qualitative prediction is good. Note that no attempt was made to adjust the
model parameters such that a better t with the data would be obtained.
The NO conversion under lean conditions is predicted a bit too low. The
prediction of CO and HC conversion under rich conditions are also quantita-
tively not very accurate. There are two major reasons for this. The reactions
of CO and HC with water (water gas shift and steam reforming) that are very
important under rich conditions, especially after the oxygen has been depleted
from ceria, are not included in the model. These reactions would certainly
improve the conversions of CO and HC as this additional reaction path would
not rely on the available oxygen. A simple stoichiometric calculation shows
that the observed CO conversion cannot be obtained without these reactions.
HC conversion, on the other hand, is already predicted as too high by the
model so an additional reaction would only increase the error. This probably
54 First principle modeling
has to be connected with the approximation of total hydrocarbons by only two
representatives and model extrapolation to higher temperatures.
2.4.2 Dynamic operation
The converter light-o is a dynamical process that is mostly determined by the
noble metal surface dynamics. During the normal operation of the converter,
above the light-o temperature, storage and release of oxygen to and from the
ceria surface becomes the main dynamic eect. It was mentioned already that
in the current engine management systems the lambda controller induces the
exhaust gas oscillations around stoichiometry. As shown in previous text, these
feed oscillations can be benecial in terms of conversion at lower temperatures
but usually deteriorate the conversion at higher temperatures. With ceria
present in the washcoat of the converter this deteriorating eect is lowered. It
is the intention here to highlight dynamic eects that occur during the feed gas
perturbations. As the current engine control systems do not take into account
the converter dynamic behavior, but only the engine dynamics to keep the
engine lambda value at the desired level, it is hoped that there is an opportunity
for further improvement of conversion by controlling the catalytic converter
itself. This idea has usually been abandoned in the past, due to the complexity
and lacking of complete understanding of the converter dynamic behavior.
The dynamic eects will be analyzed by rst looking at the processes on
the catalytic surface during transients and then at the inuence of some pa-
rameters (mass ow, oxygen storage capacity) on the converter dynamics. The
simulations have been performed by imposing step-like changes of the converter
inlet lambda signal. The lambda inlet signal was created by step responses of
a second order lter with the following transfer function:
T
f
(s) =
1
(0.08s + 1)
2
(2.15)
Direct step inputs were not applied because they would not be realistic since
the input of the catalytic converter is basically the delayed output of the engine,
where some mixing of exhaust gases may be assumed. The step response of this
lter reaches steady state after approximately 0.5s, what is in good correlation
with the response of the engine control system.
The initial conditions are kept the same in all simulations to allow easy
comparison of results. The inlet lambda changes from lean (=1.04) to rich
(=0.96) at 5s and back at 20s. The inlet temperature is kept constant at
650K, while the inlet mass ow is 54kg/h. Figure 2.19 presents the outlet
concentrations of all components during the transients. Calculated inlet and
outlet lambda responses are presented in gure 2.20 together with the coverage
of the ceria surface and bulk throughout the reactor. Coverage of the ceria
surface by oxygen and CO as well as coverage of the bulk by oxygen is presented.
2.4. Warmed-up converter 55
0 10 20 30
0
0.5
1
1.5
2
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
0 10 20 30
0
0.05
0.1
0.15
0.2
0.25
0.3
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
0 10 20 30
0
0.01
0.02
0.03
0.04
0.05
0.06
time[s]
C
2
H
4

c
o
n
c
.

[
v
o
l
%
]
0 10 20 30
0
0.01
0.02
0.03
0.04
0.05
0.06
time[s]
C
2
H
2

c
o
n
c
.

[
v
o
l
%
]
Figure 2.19: Inlet (dashed) and outlet concentrations during the lambda step
tests.
Oxides of nitrogen that are also stored on ceria are not shown because their
concentration is quite low. Figure 2.21 shows the noble metal surface coverage
of the most important components in the front and back part of the reactor.
The steady state conversions in gure 2.19 are as presented before: the lean
feed favors the conversion of carbon monoxide and hydrocarbons, while nitric
oxide is almost completely converted with the rich inlet feed.
Dynamic eects, however, are not that simple. The oxygen storage on the
ceria surface and in the bulk is completely lled before the lean-rich step. Just
after the step enough oxygen is supplied from the storage to convert both CO
and hydrocarbons completely. This eect can be seen in gure 2.20 as the
output lambda plateau and is quite well known from experiments [16, 28]. As
the level of the stored oxygen decreases the rate of oxidation also decreases
and more carbon monoxide and hydrocarbons (especially ethylene) will not be
converted. Acetylene has a stronger ability to stick on the noble metal surface
and use the oxygen from ceria, so it is still highly converted even when the
amount of available oxygen becomes very low. This also applies in the steady
state, as seen in gure 2.19. After all oxygen has been depleted the outlet
concentrations reach steady states. The NO conversion becomes high almost
instantaneously after the step though some desorption of NO stored on noble
metal and ceria can be observed. First the oxygen from the storage in the
front part of the reactor is used for the oxidation (gure 2.20), but there is an
interesting eect in the very beginning of the reactor: the oxygen from the ceria
56 First principle modeling
0 10 20 30
0
0.2
0.4
0.6
0.8
1
time [s]
O
m

c
o
v
e
r
a
g
e

[

]
0 10 20 30
0.95
1
1.05
time [s]
l
a
m
b
d
a

[

]
0 10 20 30
0
0.2
0.4
0.6
0.8
1
time [s]
O
s

c
o
v
e
r
a
g
e

[

]
0 10 20 30
0
0.2
0.4
0.6
0.8
1
time [s]
C
O
s

c
o
v
e
r
a
g
e

[

]
Figure 2.20: Lambda inlet (dashed) and outlet signals during the lambda step
tests. Coverage of ceria surface and bulk at inlet (thin line), middle of the
reactor and outlet(thick line)
0 5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
s
u
r
f
a
c
e

c
o
v
.

[

] C
2
H
4
CO*
O*
C
2
H
2
NO
2
*
0 5 10 15 20 25 30 35
0
0.2
0.4
0.6
0.8
1
time [s]
s
u
r
f
a
c
e

c
o
v
.

[

]
Figure 2.21: Noble metal coverage during the step tests in front (above) and
back part of the converter.
2.4. Warmed-up converter 57
surface is not completely used. The reason for that is that a major fraction of
the noble metal surface is covered with acetylene, because the conversion is still
low in this part of the reactor. This hinders oxygen to reach the noble metal
surface in larger quantities, and instead it adsorbs on the ceria surface, where
sites are available. Therefore an unexpected equilibrium is obtained with a
substantial fraction of the noble metal surface covered with CO and acetylene,
but also a quite full oxygen storage. It has been observed that this eect is
more remarkable at low temperatures and higher space velocities, where the
conversion is lower and it is more dicult for oxygen to reach the noble metal
surface. This eect can also contribute to the experimentally observed increase
of the oxygen storage capacity with increasing temperature [23]. The responses
of oxygen on the ceria surface and in the bulk at the same point in reactor
match quite well, meaning that the transfer from the bulk to the surface is not
rate determining. Therefore both storages act as a single, but larger storage.
After oxygen becomes depleted from the ceria surface (and from the bulk at
the same time), spill-over of CO from the noble metal to the ceria surface
occurs. Though the anity of CO for adsorption on ceria is not very high, the
spill-over occurs since there are no competitors under rich conditions. CO has
somewhat lower anity to react with oxygen from ceria than hydrocarbons,
so the breakthrough occurs earlier. This can also be seen by the large peak in
the CO* signal at the reactor end in gure 2.21. Since the CO breakthrough
occurs earlier than that of hydrocarbons, CO rst reaches the back part of
the reactor. As the conversions of all species decrease, hydrocarbons (specially
acetylene) cover most of the surface. Though not clearly visible in gure 2.19,
a decreased conversion of NO occurs in steady state during the rich inlet. The
conversion of NO rst reaches almost 100% and then drops to around 98% as
oxygen from ceria becomes depleted and CO occupies the ceria surface. This
eect has experimentally been observed [16, 50]. The most likely explanation
for this is that as the surface of ceria becomes lled with CO the bifunctional
dissociation of NO involving an empty ceria site (step 32, table 2.2) becomes
inhibited thus leading to a lower CO conversion. This eect is amplied at
higher space velocities, when the NO conversion drops more signicantly.
After the rich-lean step, the CO conversion slowly increases to almost 100%
(gure 2.19). There is a clear desorption of CO from the catalyst, that is largely
stored on the ceria surface. This desorption is related to oxygen lling of ceria
since it disappears as the oxygen breakthrough occurs. The response of the
hydrocarbons is also not immediate, but faster. This conversion dip is also
reected in the lambda signal response (gure 2.20). Hence, the outlet lambda
signal stays slightly below stoichiometry even after the inlet feed has become
lean. Such feature of the outlet lambda has also been observed experimentally
[16, 50]. In gure 2.22 a lambda step response measured at ETH in Z urich
shows the same eect. Very fast CO and NO analyzers (response time few
miliseconds) have been used in that study, and CO desorption from the catalyst
58 First principle modeling
is obvious. After the inlet step change, oxygen starts to ll the vacant ceria
sites, rst in the front part of the reactor. This leads to vacant noble metal
surface in the back of the reactor, which becomes covered by ethylene, acetylene
and CO. Since very little oxygen reaches the back part of the reactor, as most of
it adsorbs before on the ceria surface, some hydrocarbons, as well as CO desorb
from the noble metal surface. With oxygen advancing through the reactor,
more oxygen covers the noble metal surface and the conversion slowly increases.
The conversion of all reducing species is almost complete in the steady state.
A very interesting eect can be seen in the NO conversion. Though no large
adsorption of nitrogen containing species on the ceria surface is observed , the
lling of the oxygen storage clearly inuences the NO conversion (gure 2.19).
As oxygen lls the oxygen storage after the rich to lean step, vacant sites
remain available on the noble metal surface for NO to adsorb and dissociate.
Also the bifunctional-dissociation of NO plays an important role here. As the
oxygen storage becomes lled, oxygen starts to cover the noble metal surface
too, and the NO conversion decreases. A similar eect has also been observed
experimentally [16, 50]. NO conversion drops as the oxygen storage starts
being lled with oxygen and the NO breakthrough occurs before the oxygen
breakthrough. That can also be observed by the bend in the lambda signal
(gure 2.20). The oxygen coverage of the ceria surface and bulk again have a
similar form indicating that no transfer problems occur. Therefore, also here
the oxygen storage can be treated as one bigger storage.
Figure 2.23 shows the gas temperature in the middle of the reactor during
the steps. After the lean-rich step the temperature raises due to exothermal
oxidation of CO and hydrocarbons with oxygen from ceria. As oxygen becomes
depleted, conversion decreases and so does the temperature. The temperature
does not reach steady state as the step duration is too short and thermal
processes in the converter are much slower then the dynamic processes on the
catalytic surface. After the rich-lean step even larger heat release is observed.
This is due to the oxidation of ceria, which is a highly exothermal process.
The inuence of exhaust mass ow, oxygen storage capacity and inlet
signal amplitude on dynamic responses
As already mentioned, the catalytic converter operates in a very dynamical
regime with constantly changing inlet concentrations, mass ow and temper-
ature. It is interesting to see how the converter responds to inlet lambda
step changes with dierent exhaust mass ows and temperatures. Figure 2.24
presents the converter responses to the same inlet lambda changes as in the
previous cases, but with three dierent exhaust mass ows: nominal (the same
as before), two times decreased and two times increased. A very interesting fea-
ture to compare is the duration of the high conversion (outlet lambda plateau
around stoichiometry) after both steps. One may expect that if the exhaust
2.4. Warmed-up converter 59
Figure 2.22: Responses of CO and NO, measured with fast gas analyzers,
on inlet lambda step changes. Courtesy of Engine Systems Laboratory, ETH
Z urich.
0 5 10 15 20 25 30 35
718
720
722
724
726
728
730
732
734
736
time [s]
t
e
m
p
e
r
a
t
u
r
e

[
K
]
Figure 2.23: Gas temperature in the middle of the converter during the step
tests.
mass ow rate is doubled, the plateau would shorten two times, but this eect is
clearly nonlinear. The conversion of CO starts decreasing after approximately
8s in the case of the lowest mass ow, while in the case of the highest ow it
already drops after 1s. There are more reasons for such behavior. First, as the
60 First principle modeling
5 10 15 20 25 30 35
0
0.5
1
1.5
2
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
54 kg/h
108 kg/h
27 kg/h
5 10 15 20 25 30 35
0
0.1
0.2
0.3
0.4
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0
0.01
0.02
0.03
0.04
0.05
time [s]
H
C

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0.95
1
1.05
time [s]
l
a
m
b
d
a

[

]
Figure 2.24: Outlet gas concentrations and lambda signals during the step tests
with dierent exhaust mass ows.
mass ow increases, the needed transfer of oxygen from ceria would also have
to be increased to convert all the species at the same rate. Since the transfer
rate is limited by the concentration of oxygen on ceria, the conversion is likely
to be lower with the same amount of oxygen on ceria. This eect will be fur-
ther studied in the next chapter. Also, the noble metal coverage by acetylene
increases by higher mass ows, what further inhibits adsorption of CO and
ethylene and their conversion. The amount of unused oxygen from the ceria
surface in the front part of the reactor also increases because of the inhibition.
It enhances the process further. The already discussed eect of decreased NO
conversion in the steady state with the rich feed can clearly be observed when
the mass ow increases largely. This eect is enhanced by the increase of the
inlet NO with the mass ow.
Similar eects take place after the rich-lean step, which lead to faster oxygen
breakthrough in the case of higher space velocity. Note that the lower mass ow
leads to a longer CO desorption after the rich-lean step, because it takes longer
time for oxygen to ll the oxygen storage. Thus, in this case, the conversion
of CO during the transient is not favored by decreasing the mass ow. The
conversion of ethylene in the steady state is decreased with the high mass ow.
The reason is inhibition by oxygen as already observed during the light-o
tests.
Figure 2.25 presents the inuence of decreasing the oxygen storage capacity,
2.4. Warmed-up converter 61
5 10 15 20 25 30 35
0
0.5
1
1.5
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
nom.
OSC/2
5 10 15 20 25 30 35
0
0.05
0.1
0.15
0.2
0.25
0.3
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0
0.01
0.02
0.03
0.04
0.05
time [s]
H
C

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0.95
1
1.05
time [s]
l
a
m
b
d
a
o
u
t

[

]
Figure 2.25: Outlet gas concentrations and lambda signals during the step tests
with nominal and 50% decreased oxygen storage capacity.
which leads to a more linear eect than when changing the exhaust mass ow
or inlet lambda amplitude. When comparing the plateau of the lambda signal
behind the converter its length decreases by nearly the same percentage as
the oxygen storage capacity decreases. The nominal case is compared to the
situation when both surface and bulk ceria capacities are reduced by a factor
2. The decrease of the oxygen storage capacity is usually associated with the
aging of the converter. To simulate the aging process more accurately, one also
would have to take into account changes in the noble metal capacity, which
would then lead to more complex changes in the converter behavior.
The converter responses to input lambda signals of dierent amplitudes
are shown in gure 2.26. Though it might not be apparent at a rst glance,
the lambda responses are much more linear (when compared to mass ow
dierences) in the case of a rich-lean step, but also nonlinear in the case of a
lean-rich step. The outlet lambda plateau at stoichiometry, in the case of a
lean-rich step, increases more than two times when the inlet lambda is set to
0.98 instead of 0.96. When the rich inlet lambda is set to 0.99 the outlet lambda
barely moves from the stoichiometry during the step length (15s). Note that
before the rich-lean step, in the case of the inlet lambda changing from 0.99 to
1.04, the oxygen storage was not completely emptied and this breakthrough.
It can also be observed that the CO desorption depends much less on the inlet
lambda than on the mass ow.
62 First principle modeling
5 10 15 20 25 30 35
0
0.5
1
1.5
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0
0.01
0.02
0.03
0.04
time [s]
H
C

c
o
n
c
.

[
v
o
l
%
]
5 10 15 20 25 30 35
0.95
1
1.05
time [s]
l
a
m
b
d
a
o
u
t

[

]
Figure 2.26: Outlet gas concentrations and lambda signals during the step
tests with dierent inlet lambda amplitudes: 1.02-0.98-1.02(thin line), 1.04-
0.99-1.04, 1.04-0.96-1.01(thick line).
2.4.3 Motivation for control
The analysis of the dynamic responses of the converter indicates that the dy-
namics of the converter is by no means negligible and should be considered in
the control system if a tight control is desired. Only few attempts to create
converter control-oriented models [12, 76], or to actually design a converter
controller [81] can be found in the literature. As already mentioned, most of
the current control systems control the engine air/fuel ratio at stoichiometry.
In other words, such a controller is a feedforward controller for the catalytic
converter, not taking into account the converter dynamic characteristics. There
is no problem in the steady state, as stoichiometry is then optimal for converter
performance. During a highly transient operation however, i.e. acceleration or
deceleration of a vehicle, the inlet lambda value may drift from stoichiometry
due to engine performance requirements and due to bandwidth limitations of
the closed loop control. This can, in the worst case, lead to a complete lling
or emptying of the ceria surface. Other common situations are fuel enrichment,
when extra torque is desired, and fuel cut-o. These conditions may easily lead
to a completely empty or completely lled oxygen storage, respectively. If the
controller does not take into account converter dynamics, it forces the inlet
lambda back to stoichiometry (or to oscillate around stoichiometry), assuming
automatic restoring of the high conversion. It will be shown here that the con-
2.5. Conclusions 63
verter can perform very dierently, though with the same inlet lambda signal
present, when its state varies. This main state of the converter is assumed to be
the degree of the ceria coverage. Suppose the inlet lambda signal is oscillating
with amplitude of 2% and frequency of 0.67Hz around stoichiometry. This sig-
nal corresponds well to a typical inlet lambda signal, resulting from the engine
closed loop control. Three assumed initial conditions are considered: a com-
pletely empty ceria surface, a completely full ceria surface and an optimally
covered ceria surface. The word optimal should not be taken literally be-
cause no optimization has been performed to nd such surface coverage. It was
simply taken as close to 50% of the surface coverage to qualitatively show the
dierences in the three cases. Figure 2.27 shows the outlet concentrations of all
components in the three cases. Note that in the cases of full and optimal lling
the rst half-period of the inlet lambda signal is lean, while in the case of empty
surface it is set to be rich. As one would expect, the full oxygen storage favors
the oxidation of carbon monoxide and hydrocarbons, while the empty storage
increases the conversion of nitric oxide. The reactor responses are clearly not
the same, as the optimally lled reactor retains a high conversion of all com-
ponents during the test. The empty ceria surface deteriorates conversions of
hydrocarbons and carbon monoxide, when the inlet feed becomes rich, as there
is no extra oxygen to buer the conversion decrease. The outlet concentrations
of ethylene and CO are rather high even when the inlet feed becomes lean due
to desorption of the species stored on the noble metal and ceria surface. When
the surface becomes lled with oxygen, the NO conversion drops as the inlet
signal becomes slightly lean. The presented results show that a tight control of
the catalytic converter oxygen storage phenomenon could indeed lead to a con-
version improvement under highly transient operation. The obvious problem
is the impossibility of measuring the degree of the ceria surface lling, which
becomes the controlled variable in the control system. A simplied, control-
oriented model of the oxygen storage phenomenon has to be developed for the
on-line application as the inferential sensor in the control system. The model
could also be used for obtaining the optimal desired (reference) values for the
controlled variable. Considering the changing nature of the converter dynam-
ics due to the aging process, the controller (model) should have some adaptive
possibilities. The development of the simplied model, control strategy and
the controller are the topics of discussion in the next chapter.
2.5 Conclusions
A model of a three-way catalytic converter based on the elementary step ki-
netics has been presented in this chapter. The experimentally obtained (on
the same Pt/Rh/-Al
2
O
3
/CeO
2
catalyst) kinetic sub-models for the oxidation
of carbon monoxide and hydrocarbons and for the reduction of NO have been
applied. The model can predict, with a reasonable accuracy, the warming up
64 First principle modeling
0 2 4 6 8
0
0.1
0.2
0.3
0.4
0.5
0.6
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
full
empty
half filled
0 2 4 6 8
0
0.05
0.1
0.15
0.2
0.25
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
0 2 4 6 8
0
0.005
0.01
0.015
time [s]
C
2
H
4

c
o
n
c
.

[
v
o
l
%
]
0 2 4 6 8
0
0.2
0.4
0.6
0.8
1
x 10
4
time [s]
C
2
H
2

c
o
n
c
.

[
v
o
l
%
]
Figure 2.27: Outlet gas concentrations during inlet lambda oscillations (A=2%,
f=0.67Hz) with various initial degrees of ceria lling (empty, full, half lled).
Some concentrations are very low due to high conversion.
phase of the converter, helping to get a better insight into the processes that are
important for determining the light-o characteristics of the converter. Such
a kinetic model has to be used to describe the light-o process, as the most
important phenomenon is the competition between the species for vacant noble
metal sites. The main inhibitors are acetylene and nitric oxide, which have the
ability to easily reach the surface, but their light-o temperature is higher than
that of carbon monoxide and ethylene. The light-o temperature of the lat-
ter two species is therefore increased in the presence of the inhibiting species.
The oxygen storage and release capability of ceria is mostly responsible for the
converter dynamic behavior after the light-o, during normal operation. As
the inlet feed tends to oscillate around stoichiometry, resulting from the engine
control system, this dynamic behavior was studied in detail. Also inuences
of typically changing variables, such as mass ow, inlet gas temperature, in-
let signal amplitudes and oxygen storage capacity, on the dynamic responses
have been simulated. Since the converters behavior is highly nonlinear it is
dicult to directly control it, though the possible benets are substantial. The
developed model serves as a basis for the design of a converter controller. The
overall model accuracy is satisfactory for the model to be used as a basis for
the control system development. With further kinetic studies (detailed eects
of water, more hydrocarbons) the model can be adjusted to give even more ac-
curate prediction. The dynamic processes of interest for control, mostly related
2.5. Conclusions 65
to ceria, are (at least qualitatively) well predicted. Since the control strategy
should certainly be model-based, this rigorous model can also serve to develop
simplied, control-oriented models, which could then be used on-line in the
controller. Not only does the model save time during the development pro-
cess of the controller, but it can also give a better performance index as many
variables, which can not directly be measured, can easily be accessed from the
model. The model is then used as a soft sensor or inferential predictor for these
variables and can be included in the closed loop control system.
66 First principle modeling
3
Model-based controller
3.1 Introduction
3.2 Engine model
3.3 Engine air/fuel control
3.4 Control-oriented model of
the catalytic converter
3.5 Feasibility of control
3.6 Model-based predictive
control
3.7 Conclusions
3.1 Introduction
Although catalytic converters have been used for exhaust gas purication in
commercial vehicles for more than two decades now, a tight control of this pro-
cess has just recently become a relevant issue. The major reason for this new
development is that the exhaust emission regulations are constantly becom-
ing more stringent. Apart from converter design improvements, new control
strategies have to be developed to achieve a high performance under transient
operating conditions [82]. Current engine control systems for gasoline engines
are controlling the engine air-fuel ratio at stoichiometry. Such a controller
performs quite well under stationary conditions, but a further improvement is
possible under transient conditions if the dynamics of the catalytic converter
would be included in the control scheme. These dynamics stem mostly from the
oxygen storage and release capabilities of ceria [16, 28, 50] as it was shown in
the previous chapter. The underlying eects are quite nonlinear and not even
completely understood. This may be the reason for scarce control applications.
The dynamic behavior of the engine cannot not be disregarded when study-
ing control of the catalytic converter. These two subsystems can be considered
as two reactors in series with the engine producing the input for the catalytic
converter. A Mean Value Engine Model (MVEM) that has become widely ac-
cepted for control purposes in the last decade, will be used in all simulation
studies. On the basis of the throttle (driver) input and controller output (fuel
injection) the model calculates the inputs to the catalytic converter, exhaust
value and mass ow.
The control objective is to keep the level of oxygen stored on ceria at some
optimal level. A short introductory study to asses the advantages of such
68 Model-based controller
a control strategy was presented in the previous chapter. This can be only
achieved by using the model as an inferential sensor because the controlled
variable cannot be measured. The model accuracy and speed are thus crucial for
a high performance. The speed of calculation is of special importance because
the model based simulations have to be calculated in the vehicle on-board
computer. Therefore, a model like the rst principle model presented in the
previous chapter cannot be used for control directly. However, the information
about the process dynamics that stems from the rst principle model will be
proven as crucial for building a simpler control-oriented model. This chapter
will demonstrate the feasibility to both construct such a model, as well as to
use it for control. Namely, it is possible to control the average (a catalytic
converter is a distributed parameter process) level of oxygen stored on ceria
throughout the reactor as desired.
After developing a basic controller and demonstrating its feasibility a step
further is taken. The goal is to design a controller in an optimal fashion to min-
imize the exhaust emissions. Because of an explicit model presence (inferential
sensor), the obvious choice is to apply Model Predictive Control. Model Predic-
tive Control is at the moment the most applied advanced control methodology
in industry [27]. The controller uses a model to predict the future process be-
havior and to nd the optimal control sequence to achieve the control goals.
The control sequence is found by solving the optimization problem on-line and
applying the rst control output to the process. The optimization process is
repeated at each sampling interval, thus requiring a lot of computational power
and making this approach suitable only to slower processes. The ability of a
Model Predictive Controller to deal with constraints on both controller out-
puts and process variables as well as its high potential to eectively control
nonlinear processes makes it also very attractive for faster processes. An ap-
proximate solution is to train some nonlinear function (i.e. a neural network)
with the o-line calculated outputs of the Model Predictive Controller. The
Model Predictive Controller is thus replaced by an analytic nonlinear function
which can be calculated very fast at each sampling interval. Such an approach
has been successfully applied to solve various control problems [48, 73]. With a
proper selection of the nonlinear function (neural network), a sucient level of
complexity and enough training points, the Model Predictive Controller can be
approximated to an arbitrary accuracy [73]. A Gaussian radial basis function
network is used as the nonlinear function.
Finally, the eectiveness of the novel controller is shown by simulations of
highly transient driving cycles that can, for example, be related to the condi-
tions during city driving.
3.2. Engine model 69
3.2 Engine model
Engine models of dierent complexity and for various applications exist. The
goal of a model that would be applied for the exhaust control is to predict
the exhaust gas composition in a dynamic operating regime. Since the gas
composition is mainly a function of in-cylinder lambda value and load, the
most important goal of the model is to accurately predict the amount of fuel
and air in the engine cylinders. This can be achieved with mathematically very
complex rst principle models that perform calculations on cycle to cycle basis
[17]. This is, however, too complex and not necessary for emission control since
the time scale of interest is that of a couple of engine cycles. Therefore, so called
Mean Value Engine Models (MVEM) have drawn a large interest in the last
two decades [2, 3, 26, 41, 42]. An MVEM typically consists of two submodels,
air path and fuel path. Together these submodels predict the amount of air
and fuel in the cylinders of the engine, and thus also the exhaust lambda value.
In the model applied in this thesis no distinction was made between dierent
cylinders. A torque delivery submodel is sometimes also included if a complete
powertrain model is desired, but is not of great importance for emissions and
is not included in the model developed here therefore.
3.2.1 Air path
The goal of the air path submodel is to calculate the amount of air entering
the cylinder on the basis of external inputs that are position of the throttle,
and the engine speed. The former is inuenced by the driver and therefore can
be seen as a disturbance on the system (if no drive-by-wire exists). The model
has one state, the intake manifold pressure.
The change of air mass in the intake manifold can be written as the dif-
ference between the air mass past throttle valve that is entering the manifold,
m
at
, and the port air ow (air entering the cylinder), m
ap
:
dm
m
dt
= m
at
m
ap
(3.1)
On the other hand, the manifold mass change can be calculated by using the
ideal gas equation:
dm
m
dt
=
d
dt
_
p
m
V
m
M
a
RT
m
_
(3.2)
As manifold pressure, p
m
and temperature T
m
are the only variables on the
right hand side of (3.2) that can change in time, the manifold pressure can be
calculated by the following dierential equation:
dp
m
dt
=
RT
m
V
m
M
a
( m
at
m
ap
) +
p
m
T
m
dT
m
dt
(3.3)
70 Model-based controller
Since the last term in (3.3) is small due to slow temperature changes in the
intake manifold, the standard manifold pressure state equation (MPSE) can be
written as:
dp
m
dt

RT
m
V
m
M
a
( m
at
m
ap
) (3.4)
Hence, to have a complete air path model the port and throttle air ows have
to be calculated.
The throttle air mass ow equation is adopted from [41]:
m
at
(, p
r
) = m
at1
p
a

T
a

1
()
2
() (3.5)
where stands for the throttle angle, p
r
= p
m
/p
a
is the ratio of the manifold
pressure and the pressure in front of the throttle (atmospheric pressure less the
pressure drop of the air lter) and m
at1
is a constant for a given engine. The
functions
1
() and
2
(p
r
) are calculated as follows:

1
() = 1
1
cos() +
2
cos
2
()

2
(p
r
) =
_
1
p
n
_
p
p
1
r
p
p
2
r
, if p
r
p
c
1, if p
r
p
c
(3.6)
p
c
=
_
p
1
p
2
_ 1
p
2
p
1
p
n
=
_
p
p
1
c
p
p
2
c
Constants
1
,
2
, p
1
and p
2
can be tted for a specic engine, but their values
do not vary much for dierent engines. Authors in [41] have obtained the
following values:
1
= 1.4073,
2
= 0.4087, p
1
= 0.4404 and p
2
= 2.3143.
The throttle air mass ow equation (3.5) clearly introduces a nonlinearity
into the manifold pressure state equation since it is nonlinearly dependent on
the manifold pressure. Figure 3.1 shows plots of functions
1
() and
2
(p
r
) for
the parameters as given above.
Port air mass ow can be calculated in a simple manner by the speed-density
equation:
m
ap
(n, p
m
) =
V
d
120RT
m
e
v
p
m
n (3.7)
where V
d
stands for the engine displacement, n for engine speed (rpm) and e
v
for the volumetric eciency. The latter is a very important variable since it is
not a constant for a given engine, but rather depends on the manifold pressure
and speed. Engine management systems typically contain look-up tables to
approximately determine this quantity for given operating conditions.
3.2. Engine model 71
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
0.05
0.1
0.15
0.2
0.25
throttle angle [deg]

1

[

]
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
p
r
[]

2

[

]
Figure 3.1: Functions
1
() and
2
(p
r
) of the throttle air mass ow model.
For a four cylinder, four stroke engine the air mass induced per stroke can
be written as:
m
i
=
1
2
m
ap
n/60
=
V
d
4RT
m
(e
v
p
m
) (3.8)
The temperature in the manifold does not change rapidly, so equation (3.8)
can be written as:
m
i
= k(T
m
)m

i
(3.9)
where m

i
= e
v
p
m
and k = V
d
/(4RT
m
). m

i
is typically called normalized
engine air charge [41]. The proportionality constant k is a function of the inlet
manifold temperature.
Volumetric eciency can be calculated by integrating the p-V diagram of
an SI engine during the pumping cycle. Hendricks et al. [41] have shown that
with some minor assumptions the normalized engine air charge at a given speed
can be expressed as a linear function of intake manifold pressure:
m

i
= e
v
p
m
= s
m
(n)p
m
+y
m
(n) (3.10)
Hence, the volumetric eciency can be approximated with the above given
analytic function, what can considerably simplify the nal control algorithm
as a look-up table does not have to be applied. The only problem is that the
above expression still depends on the speed, so the engine mapping should be
performed at various speeds to obtain accurate results. The parameters s
m
and
y
m
do not vary much for dierent engines. Finally, disregarding the dependence
72 Model-based controller
of the normalized air charge on speed the port air mass ow can be expressed
as:
m
ap
(n, p
m
) =
V
d
120RT
m
(s
m
p
m
+y
m
)n (3.11)
Port air mass ow is thus linear in manifold pressure and almost linear in speed.
Since speed dynamics has an order of magnitude larger time constant than the
manifold pressure dynamics, speed will be assumed constant in most of the
simulations and only equation (3.11) will be used to calculated the port air
mass ow.
Model verication
A simple model verication test has been performed to asses the model accu-
racy. During an European emission test cycle all relevant engine signals have
been recorded. The test lasts only 20 minutes and a variety of driving condi-
tions is met. The test was performed on a chassis dynamometer with Volvo
V40, 1.8l, 4 cylinder engine vehicle at PD&E in Helmond. In order to validate
the port and throttle air ow sub-models a number of steady state points had
been selected. Measured variables that were used to validate the air ow model
include intake manifold pressure (p
m
), engine speed (n) , intake manifold tem-
perature (T
m
), throttle air mass ow ( m
at
) which is in steady state equal to
port air mass ow ( m
ap
), and throttle position (). Only typical operating
points are reached during the test, so the model cannot be validated rigorously
under all driving conditions. Such a validation would require tests on an engine
dynamometer.
Figure 3.2 shows the accuracy of the throttle air ow model (3.5,3.6). Points
with various throttle positions and intake manifold pressures have been used.
The function
2
(p
r
) has not been changed regarding the function recommended
in [41]. Optimal values (obtained by solving a nonlinear least squares problem)
for
1
and
2
were found to be 1.3813 and 0.3917 what is very close to the
literature data. The value of m
at1
was 6.53 10
5
kgK
1/2
/Pas. The average
absolute error is 3.9%, with a standard deviation of 0.059. The largest errors
were found at low throttle angles. This was expected as the position of the idle
valve was not known. A special controller controls the idle speed via the idle
valve when the throttle is closed. The problem is that the idle valve is probably
open also during standard operation, but the opening was unknown.
Figure 3.3 shows the prediction of the normalized air charge by the model.
Again, some operating conditions have not been reached during the driving
test, but the model predicts fairly well with a mean absolute error of 2.96%.
Parameters s
m
and y
m
were assumed to be functions of speed (a quadratic
dependence) as suggested in (3.10). When no speed dependency was assumed
the average error was 4.35%. At 2000 rpm s
m
=0.976 and y
m
=-0.134, while
3.2. Engine model 73
0 5 10 15 20 25 30 35 40
30
40
50
60
70
80
90
throttle pos. [%]
m
a
n
i
f
o
l
d

p
r
e
s
s
u
r
e

[
k
P
a
]
0 5 10 15 20 25 30 35 40
0
50
100
150
200
throttle pos. [%]
m
a
s
s

f
l
o
w

[
k
g
/
h
]
measured
predicted
Figure 3.2: Measured and predicted throttle mass ow as function of the throt-
tle position for various intake manifold pressures (input data points depicted
above).
at 3000rpm s
m
=0.924 and y
m
=-0.106. The obtained values correspond well
to the literature data where s
m
was reported between 0.85 and 0.96 and y
m
between -0.045 and -0.11 for various engines. The intake manifold pressure has
to be expressed in bars, as in [41], to obtain the above given values.
3.2.2 Fuel path
The fuel path sub-model calculates the amount of fuel in a cylinder. After the
fuel is injected a part of it is vaporized and goes directly into the cylinder. A
fraction of the injected fuel is converted into small fuel droplets and some of
them, instead of entering the cylinder, stick on the port wall forming a liquid
fuel lm. The fuel from the lm slowly evaporates and eventually enters the
cylinder. This is called the wall wetting phenomenon. It is widely accepted
that in order to make an accurate controller this dynamic behavior has to be
accounted for. The fraction of fuel entering the lm, which is also sometimes
called fuel puddle, is larger at low temperatures (during the cold start of the
engine). The same holds for the evaporation time constant.
In order to model this eect a model with at least one state has to be
used. The one state model is typically called Aquino model [3]. This model
was originally developed for a central fuel injection engine, but it applies also
74 Model-based controller
1000 1500 2000 2500 3000 3500 4000
40
50
60
70
80
90
speed [rpm]
m
a
n
i
f
o
l
d

p
r
e
s
s
u
r
e

[
k
P
a
]
40 45 50 55 60 65 70 75 80 85 90
20
30
40
50
60
70
80
manifold pressure [kPa]
n
o
r
m
.

a
i
r

c
h
a
r
g
e

[
k
P
a
]
measured
predicted
Figure 3.3: Measured and predicted normalized air charge as function of the
intake manifold pressure for various engine speeds (input data points depicted
above).
for modern port injection engines what led to a wide acceptance of this model
[2, 15, 32, 61]. The change of the fuel mass in the fuel lm can be expressed
as the dierence between the fuel entering the lm, X m
finj
, and evaporating
from the lm, m
ev
,:
dm
ff
dt
= X m
finj
m
ev
(3.12)
Note that the fuel injection is assumed here to be a continuous process, while it
is in fact a discrete process because of the engine operation nature. X denotes
the factor of the injected fuel entering the lm. The fuel ow into the cylinder
is:
m
fcyl
= (1 X) m
finj
+ m
ev
(3.13)
Assuming the evaporation time constant
ev
the rate of evaporation can be
calculated as:
m
ev
=
m
ff

ev
(3.14)
Given the above equations the nal model in an operating point can be ex-
pressed in Laplace domain with a lead-lag lter:
3.2. Engine model 75

M
fcyl
(s)

M
finj
(s)
=
1 + (1 X)
ev
s
1 +
ev
s
(3.15)
Though in one operating point the model is linear, it is a nonlinear model
in principle, since parameters X and
ev
depend nonlinearly on engine speed,
intake manifold pressure and temperature. The temperature dependence is
mostly connected with the coolant temperature, but it actually depends on the
temperature of port walls, which is much more dicult to measure [2]. It is
important to notice that the model (3.15) is invertible.
Some authors [71, 85, 93] have shown that the wall wetting model should
actually consists of two fuel lms with dierent evaporation time constants
(one faster and one slower evaporating puddle), thus leading to a system with
two states. Such a model has the following transfer function:

M
fcyl
(s)

M
finj
(s)
= (1 X
1
X
2
) +
X
1
1 +
ev1
s
+
X
2
1 +
ev2
s
(3.16)
The above model is thus a system with two poles and two zeros. Due to
the nature of the process complex conjugate poles and zeros do not occur. It
was shown that the model can describe the measured frequency response of
the system more accurately [85]. It has been found that for some operating
conditions, like low speed and low load, the second order model converges into
the rst order model, but in most operating points two distinct time constant
were found with an order of magnitude dierence between them. For a warm
engine the smaller time constant is typically between 50 and 100 ms, while the
larger time constant has typical values between 300 and 800 ms [71]. It was
shown in the same study that for a cold engine these time constants can increase
2-3 times. It should be noted that at low wall temperatures the main dynamic
eect is not evaporation of the fuel lm, but the liquid fuel ow due to the shear
force at the gas/liquid interface [45]. A deposit factor X between 0.04 and 0.3
and a time constant between 0.5 and 1.5s (for a warm engine; one state model)
were reported in [2]. For a cold engine these parameters also increase 2-3 times.
Moreover, it was shown in that study that the wall wetting model parameters
depend on the type of fuel used, yielding sometimes a parameter dierence
of more than 100% under the same operating conditions. A completely other
characteristic, with factor X between 0.7 and 0.9 and evaporation time constant
between 20 and 70 ms (warm engine) was found in [87]. The deposition factor X
typically decreases with increased speed and decreased manifold pressure, while
the evaporation time constant has much more complex behavior. Therefore it
can be concluded that the wall wetting parameters vary a lot from engine to
engine, in dierent operating conditions and even with various fuels. The latter,
together with the system aging process may require on-line adaptation to cope
with the problem [61, 85].
76 Model-based controller
One drawback of the two state model is that the smaller time constant is of
the same order of magnitude to the time constant of the wide range sensor,
thus making an accurate wall wetting identication rather dicult. Application
of another, faster, sensor such as a NOx sensor in [71, 85] circumvents this
problem but yields additional hardware demands. The one state model in
principle neglects the smaller time constant, sometimes by compromising with
increasing the feedthrough term. Since it was shown in many studies that a
one state model is suciently accurate for control, this model will be used in
this thesis.
3.2.3 The complete model
With the mass of air and fuel entering the cylinder known, the engine lambda
signal can be calculated:

cyl
=
m
ap
K
s
m
fcyl
(3.17)
where K
s
stands for the stoichiometric constant which is typically around 14.6.
There is sometimes also a delay present between the moment of a fuel injection
command and the actual fuel injection (T
dinj
), amounting to a fraction of the
engine cycle (engine cycle = two revolutions). This delay will be neglected in
the simulation studies. Another system delay stems from the engine operation
nature: delay between the intake and the exhaust stroke. It is assumed here
that the intake takes place at the half of the intake stroke and exhaust at the
half of the exhaust stroke, thus leading to a delay of 3/4 of an engine cycle
(T
de
). If the lambda sensor is not placed directly behind the exhaust valve, the
transport delay in the exhaust pipe, together with some mixing, has to be taken
into account. This can be modeled by a rst order lag (time constant
mix
)
with a delay (T
dp
). The total system delay is thus T
d
= T
de
+ T
dp
. Finally,
the lambda sensor dynamics can be modeled as a rst order lag (time constant

sen
). The model from the cylinder inlet to the lambda sensor is thus:

sen
(s)

cyl
(s)
=
1
(1 +
mix
s) (1 +
sen
s)
e
T
d
s
(3.18)
The complete engine model is schematically depicted in gure 3.4. The
model of torque delivery and loss is also included in this gure for the com-
pleteness of the model. This part, however, has not been included in simulations
because the speed was always assumed to be constant. The delivered torque is
typically calculated by empirical nonlinear static functions, and it depends on
engine , manifold pressure, speed and ignition timing. Delay T
T
stands for
intake to torque delay time. Power losses (friction) can be calculated also as
empirical nonlinear functions of speed and manifold pressure [41].
3.2. Engine model 77
ap m

throttle
) , (
1 m
p
) , (
2
n p
m

m
K

comm th,

wetting
wall
) , , , (
ign m
s T
p n f e
delivery torque
T

inertia
K

s T
mix
dp
e
s

+ 1
1
s
sen
+ 1
1
CATALYST
s
sen
+ 1
1
at m

device
control
loss
M
n
m
p
/
ign

) ( finj
inj
m t

load
M
sen

cat

+
+

s T
dinj
e

X 1
s
X
ev
+ 1
+
finj m

fcyl m

wetting wall
s T
c
e

fcyl m

sen

s T
de
e

ap m

throttle throttle
) , (
1 m
p
) , (
2
n p
m

m
K

comm th,

wetting
wall
) , , , (
ign m
s T
p n f e
delivery torque
T

inertia
K

s T
mix
dp
e
s

+ 1
1
s
sen
+ 1
1
CATALYST
s
sen
+ 1
1
at m

device
control
loss
M
n
m
p
//
ign

) ( finj
inj
m t

load
M
sen

cat

++
++

s T
dinj
e

X 1
s
X
ev
+ 1
+
finj m

fcyl m

wetting wall
s T
c
e

fcyl m

sen

s T
de
e

Figure 3.4: Block scheme of the complete engine model.


78 Model-based controller
Fuel path model verication
The fuel path model accuracy was assessed during the same test as the air path.
It is, however, much more dicult to validate the wall wetting model because
dynamic tests are needed. During a closed-loop system operation of the engine
the switching characteristic of the lambda sensor introduces step-like changes in
the fuel injection signal, which lead to oscillations of the engine lambda signal.
This type of operation was used as dynamic test to asses the model parameters.
The lambda sensor time constant is kept constant at 0.04s, while the complete
delay, T
d
, and mixing time constant,
mix
, are obtained by the optimization
for dierent working points. Figure 3.5 shows the model prediction in a typical
validation experiment. The drawback of these tests is that the frequency of the
input signal is constant and also rather high. Therefore, it was only possible
to obtain the values of X with some certainty. It has been observed that
the sensitivity of the parameter
ev
was very low and therefore the obtained
values can be considered very uncertain. Moreover, the estimated value of the
parameter X was very dependent on the characteristic of the injector. The
injection characteristic was not precisely known (due to a lack of steady state
data). Also, the lambda sensor was placed at the inlet of the catalytic converter,
thus not very close to the exhaust valve. That has caused a substantial eect
of exhaust gas mixing, making it more dicult to accurately asses the wall
wetting parameters. To obtain the model parameters with more accuracy a
special test should be performed, for example with a PRBNS type input signal
for the injected fuel. Since much more dynamic information would be available,
standard indentication techniques could be applied. Also, the lambda sensor
should be placed much closer to the exhaust valve to avoid the large inuence
of mixing. Some steady state tests should be performed beforehand to obtain
the exact injector characteristic. Such a test can be performed on a motor
dynamometer test cell. Due to the fact that only a part of the operating
region was covered by the above mentioned tests, and more important due to
parameter uncertainty in those tests, the available literature data was used to
build the ultimate model applied in the sequel of this thesis [2]. The tests from
this source were performed on a Volvo 2.5 liter sequentially injected engine.
Figure 3.6 shows the deposit factor X and evaporation time constant
ev
as
used in simulations. Both parameters, X and
ev
are functions of the engine
coolant temperature and increase when this temperature decreases. This was
modeled by multiplying the parameters by a temperature dependent factor.
3.3 Engine air/fuel control
Various advanced schemes for the air/fuel control of SI engines have been pro-
posed in the last decade. Before strict emission regulations the air/fuel control
was predominantly steady state. As such a control lead to a very poor transient
3.3. Engine air/fuel control 79
6 7 8 9 10 11 12
4
4.2
4.4
4.6
4.8
i
n
j
e
c
t
i
o
n

t
i
m
e

[
m
s
]
6 7 8 9 10 11 12
0.04
0.02
0
0.02
0.04
time [s]
l
a
m
b
d
a

[

]
Figure 3.5: Wall wetting model validation during a standard driving cycle.
Above: duration of injector opening. Below: predicted (thick) and measured
(thin) lambda signals (scaled such that 0=stoichiometry).
500
1000
1500
2000
2500
20
40
60
80
100
0.2
0.4
0.6
0.8
X

[

]
500
1000
1500
2000
2500
20
40
60
80
100
0.5
1
1.5
speed [rpm] man. press. [kPa]
t
a
u
e
v

[
s
]
Figure 3.6: Parameters X and
ev
as used in simulations.
80 Model-based controller
Q P
+
-
+
-
y
y
m
u r
~
P
d
Figure 3.7: Standard IMC structure.
behavior (i.e. fast throttle changes) and vast exhaust emissions, new solutions
were needed to tackle the problem arousing from the strict emission legislation.
As already pointed out earlier, the major characteristic of a three-way catalytic
converter is the optimal conversion of all harmful components at stoichiometry.
Therefore, the goal of the engine air/fuel controllers has become to keep the
exhaust composition at stoichiometry under all operating conditions. The ma-
jor problem when using a pure feedback controller comes from system delays
(engine cycle, transport to sensor). The only possibility to achieve the goal is
to apply model-based techniques that allow a larger bandwidth of the control
system to cope with fast transients [32, 43, 51, 94]. These dierent controllers
were developed in parallel to advances in engine control-oriented modeling.
Model-based control has also been applied to relay-type sensors [14], but it will
not be considered here because such a controller can only keep lambda value
at stoichiometry and always leads to oscillations of the lambda signal.
The controller used in this thesis is based on the Internal Model Controller
(IMC) [62]. Some of the most important controller features and problems will
be discussed in this section. The controllers applied for this application typi-
cally have a feedforward-feedback structure. An observer is used to estimate
the air cylinder charge. A (nonlinear) compensator is applied to cancel the
wall-wetting dynamics. In this application, the IMC controller enables perfect
control in case that a perfect model is available. This is due to the invertibility
of the wall-wetting model (rst order with a feedthrough).
3.3.1 IMC controller
The structure of an IMC is given in gure 3.7. This controller is applied to
control the fuel injection quantity assuming the amount of air in the cylinder
is known (observer). The IMC controller consists of the model (

P) in parallel
3.3. Engine air/fuel control 81
to the process (P) and the controller (Q). Note that actually both Q and

P
together constitute the controller. A disturbance on the system is denoted
with d. An extensive analysis of IMC and its applications is given in [62]. If
disturbance d is assumed to be an output disturbance, the following expression
can be written for the system output:
y =
pq
1 +q(p p)
r +
1 pq
1 +q(p p)
d (3.19)
q,p,and p represent transfer functions (in general nonlinear operators) of
the controller, process and process model. It can easily be seen that in the case
that q = p
1
, y = r and thus the perfect control is obtained. This can also be
proven to be the case for a nonlinear system [22]. The necessary condition that
needs to be satised in both cases is that the system remains stable. Though
the perfect control can hardly be achieved in a typical process (the process has
to be invertible, otherwise an unstable controller results), it is very important
for the air/fuel control as the fuel path model is nearly invertible. A standard
design of the controller is the inverse of the model at low frequencies to ensure
a good steady state tracking and disturbance rejection.
The controller has to satisfy two goals: performance and robustness. To
satisfy the rst goal the engine lambda signal should be as close as possible
to the reference signal under all conditions. Due to the model uncertainties
this is, however, not always possible. The controller has to be made robust
enough to be able to deal with those uncertainties, which can even destabilize
the system.
In the case of IMC the expressions for the sensitivity() and complementary
sensitivity () functions are as follows:
=
y r
d r
=
1 pq
1 +q(p p)
=
y
r
=
pq
1 +q(p p)
(3.20)
When the process model is perfect the above equations become very simple:
= 1 pq
= pq (3.21)
If the error bound on the process model can be quantied, this information
can be used to assess the control system robustness. The controller is usually
tuned for the nominal model, but it has to remain at least stable for all possible
process models that can actually occur. In order to develop a robust controller
a model uncertainty has to be dened. The multiplicative model uncertainty
(l
m
(i), frequency dependent) is dened as follows:
82 Model-based controller
p(i) = p(i)(1 +l
m
(i)) (3.22)
with:
|l
m
(i)|

l
m
() (3.23)
A standard tuning procedure for the controller is to nd a good rst ap-
proximation, q, assuming that the model is perfect. This rst approximation
will yield a good performance in the ideal case. The controller is then detuned
to achieve the following robust performance specication [62]:
| w| +



l
m

< 1 (3.24)
where w is a performance weight (tuning factor). By choosing w = 0, (3.24)
becomes a robust stability problem. As the uncertainty increases, the above
condition is harder to satisfy and performance is lost in order to obtain robust-
ness. The initial controller is detuned with some, typically low-pass, prelter
f in order to satisfy (3.24). The nal controller q becomes then:
q = q

f (3.25)
The lter is sometimes applied even if detuning is not necessary but the initial
controller is chosen to be a non causal inverse of the process model.
In the case of air/fuel control the process model is the fuel path model
(3.15,3.17,3.18). The initial controller is the inverse of the wall-wetting model:
q

=
1 +
ev
s
1 + (1 X)
ev
s
(3.26)
If the model is perfect this initial controller leads to a perfect engine air/fuel
control. In that case the control is in fact open-loop because the feedback signal
is equal to zero. The compensator q

simply inverts the wall wetting dynamics.


The air model estimates the amount of air entering the cylinder and in fact
scales the output of the fuel controller. If the air prediction is not correct it
can be considered as a disturbance on the system.
The lter, that can be added to improve the robustness (3.24) of the system,
is a rst order lag element:
f(s) =
1
1 +T
f
s
(3.27)
The engine controller was tested in simulation with the following nominal
parameters: X=0.3,
ev
=1s, T
d
=0.1s,
mix
=0.015s,
sen
=0.04s, w=0.4. Pa-
rameters X and T
d
were assumed to be uncertain. The uncertainties were:
|X| = 0.1, |T
d
|=0.05s. By solving the robust performance problem (3.24)
the values of the lter time constant
f
needed were found. In order to obtain
robust stability a lter was not needed. A lter with a time constant of 0.06s
3.3. Engine air/fuel control 83
was needed to satisfy the robust performance criterion. It was found that the
most crucial parameter to be correctly estimated is the process delay. A poor
estimation of the evaporation time constant,
ev
, did not lead to severe robust-
ness problems. Parameter X is more important, but still not as signicant as
the time delay. This makes controller design easier because the delay is the
parameter that can be estimated rather accurately and is not likely to change
in time. On the other hand, an inaccurate estimation of the wall wetting pa-
rameters is more likely to occur. For example, with X uncertainty of 0.2 (100%
increase) and T
d
uncertainty of 0.03s (40% decrease) the needed lter time
constant decreases to 0.05s for robust performance, while remaining zero for
robust stability.
Dynamic operation of the system is simulated by imposing fast throttle
changes. The speed is kept constant at 2000rpm. The throttle position changes
from 20 to 30 degrees and back (ramp-like change in 0.1s). The process delay,
T
d
, changes from 0.09s to 1.1s. The controller assumes a constant delay of
0.1s. A nominal case with perfectly estimated X, and a case with X=0.4 are
considered. All other parameters are assumed to be accurately modeled. The
goal is to keep the engine lambda at stoichiometry. The air cylinder charge is
assumed to be perfectly estimated. This is of course a rather strong assumption,
but the goal here is only to demonstrate the main problems for the air/fuel
control. An observer based approach with the throttle position signal as the
input and measured manifold pressure as the output could be used to suppress
the inuence of model uncertainties.
Figure 3.8 shows the throttle position, intake manifold pressure and throttle
and port air mass ows. Note that the air path subsystem is only inuenced
by the driver and is independent of the controller. A known nonlinear behavior
of the air mass ow can be observed. The throttle air ow has an overshoot
while the port air mass ow and intake manifold pressure do not exhibit such
behavior. This overshoot occurs when the throttle position suddenly changes
while the intake manifold reaction is slower (3.5). Therefore, a measurement of
the throttle air (what is typically the case) does not give accurate information
about the port air (air entering the cylinder). The manifold pressure signal
gives better information, but the problem related to making direct use of this
signal are the sensor lag and necessary signal preconditioning (ltering of the
oscillating signal). Hence, the most accurate in-cylinder air estimation can be
obtained by an observer.
Figure 3.9 shows the engine lambda response in dierent cases. If X is equal
to its nominal value, and no prelter is applied, a nearly perfect control is pos-
sible. This is of course not the case with X=0.4. The IMC controller with and
without prelter (T
f
=0.06s) is compared to a pure open loop compensation.
By applying the prelter a conservative (slow) controller is obtained leading
to a large lambda excursion from stoichiometry. With no feedback applied the
system response is much slower as the bandwidth is lower then in the IMC case.
84 Model-based controller
0 1 2 3 4 5 6 7 8 9 10
20
25
30
t
h
r
o
t
t
l
e

[
d
e
g
]
0 1 2 3 4 5 6 7 8 9 10
0.01
0.015
0.02
0.025
0.03
a
i
r

f
l
o
w

[
k
g
/
s
] throttle
port
0 1 2 3 4 5 6 7 8 9 10
70
80
90
time [s]
m
a
n
.

p
r
e
s
s
.

[
k
P
a
]
Figure 3.8: Throttle position, throttle and port air mass ow and intake man-
ifold pressure during the air/fuel controller simulation.
The IMC controller thus retains a good open loop behavior when the model
is correct, while having a fast settling time, due to the feedback, when the
model is not correct. A possible destabilization with a badly estimated delay
is also shown (the model assumes T
d
=0.15s). In this case the added prelter
stabilizes the system (removes the steady state oscillations), while the perfor-
mance is arguably not improved with the lter presence, because large lambda
excursions during transients will probably lead to more exhaust emissions than
small oscillations in the steady state. For previously stated reasons it will be
assumed in further simulations that the delay is estimated within a 0.03s error
bound.
3.4 Control-oriented model of the catalytic con-
verter
The catalytic converter model as presented in chapter 2 is too complex to be
used on-line in an on-board computer. As already concluded, the catalytic
converter controller has to be model-based since the desired controlled variable
is the degree of ceria coverage by oxygen containing species. This variable will
in further text be called relative oxygen coverage of ceria (ROC). It cannot
be measured and therefore a model has to be used as an inferential sensor to
3.4. Control-oriented model of the catalytic converter 85
0 1 2 3 4 5 6 7 8 9 10
0.9
0.95
1
1.05
1.1
1.15
l
a
m
b
d
a

[

]
no error
IMC
IMC+filter
feedforw.
0 1 2 3 4 5 6 7 8 9 10
0.9
0.95
1
1.05
1.1
1.15
time [s]
l
a
m
b
d
a

[

]
IMC
IMC+filter
Figure 3.9: Engine lambda signals. Above: ideal model case, IMC controller,
IMC controller with a prelter with T
f
=0.06s and feedforward control. Below:
IMC controller with and without prelter with T
f
=0.06s in the case when the
assumed delay in the model is 0.15s.
estimate this variable. The only available measurements are the lambda signals
in front of and behind the catalytic converter.
Oxygen storage based models have already been used by some authors to
obtain simple control-oriented models, which could also be used for the on-
board diagnostics [12, 76]. The on-board diagnostics tries to estimate on-line
what the converter eciency is and when the converter should be replaced
[44]. The model that will be used here is similar to the model of Brandt et al.
[12], but with additional features included to account for the eects of exhaust
mass ow and inlet signal amplitudes. The eects of temperature variations
in the reactor will also be addressed. The model basically has to link the
measured lambda signals upstream and downstream the converter with the
non measurable degree of ceria coverage (ROC).
3.4.1 Model basics
The following assumptions are made during the model development:
- Only oxygen storage and release capabilities of ceria are responsible for the
dynamic behavior of the converter. Reactions that take place on the
noble metal surface only are assumed to be instantaneous.
86 Model-based controller
- The distributed oxygen storage lling along the reactor axis can be lumped
into a single (concentrated) variable.
- Lambda sensors in front of and behind the converter are ideal.
- Though it was shown that CO can adsorb on the ceria surface, only the ad-
sorbed oxygen and oxides of nitrogen are taken into account as oxygen
containing components when calculating the relative oxygen coverage of
ceria. The reason for this assumption is that CO from ceria cannot be
used for the oxidation when the inlet is rich and therefore does not t
into the role of the oxygen storage. ROC takes into account the com-
plementary ceria surface coverage by CO. The availability of ceria for
CO storage is eectively the same as additional oxygen available for CO
oxidation.
In the following discussion only CO and O
2
will be assumed in the exhaust
feed (see gure 2.2). Moreover, the reactor length is assumed to be almost
zero thus making it sucient to model only one point in the reactor. The
outlet concentrations of CO and O
2
can be expressed, with some abuse of the
notation, in the following manner (for simplicity the inputs and outputs are
expressed as concentrations while they are in fact ows):
C
COout
= C
COin
r
CO(NM)
r
CO(OSC)
C
O
2
out
= C
O
2
in
r
O
2
(NM)
r
O
2
(OSC)
(3.28)
The above relation states that both CO and oxygen can be removed from the
exhaust by surface reactions (denoted by the rst reaction rate, (NM)) and
ceria oxidation and reduction (denoted by the second reaction rate, (OSC)).
Since surface reactions are assumed to be immediate, they will occur before
oxygen storage related reactions and therefore only the excess of oxygen (lean
input) or CO (rich input) has to be considered. For example, if the inlet
composition is 1[vol%] CO and 0.7[vol%] of O
2
the net inlet concentration is
0.2[vol%] of O
2
, and that is the input for the oxygen storage-based model.
Thus, the above relations can be rewritten for a rich inlet feed as:
C
COout
= C
COex
r
CO(OSC)
C
O
2
out
= 0 (3.29)
and for a lean inlet feed:
C
COout
= 0
C
O
2
out
= C
O
2
ex
r
O
2
(OSC)
(3.30)
The subscript ex denotes the excess of some component remaining after a
surface reaction has been completed.
3.4. Control-oriented model of the catalytic converter 87
The rate of excess CO (O
2
) disappearance under rich (lean) conditions is
proportional to the oxygen storage emptying (lling). The oxygen storage
lling rate can be expressed as:
r
O
2
(OSC)

d
dt
=
1
L
OSC
k
fill
C
O
2
,s
(1 ) (3.31)
and emptying rate:
r
CO(OSC)

d
dt
=
1
L
OSC
k
emp

CO
(3.32)
The above equations are based on exact equations used in a single reactor
point in the rst principle model. This model is already nonlinear because of
the multiplication of the state with the input. Moreover, in (3.32) the input is
not directly the CO concentration but CO noble metal surface coverage. This
can be another source of nonlinearities because the mutual interactions on the
noble metal surface are very nonlinear.
The value calculation was dened in (1.2). It can be simplied by grouping
the species into oxidants (O), reductants (R) and products (P):
=
O +P
R +P
, P O, P R (3.33)
It will be assumed that all oxidants can oxidize ceria and all reductants can
reduce ceria. Since only excess of reductants or oxidants is necessary for the
model input, the following holds for lean and rich cases:

lean

O
ex
P
+ 1 (3.34)

rich

1
R
ex
P
+ 1
1
R
ex
P
(3.35)
Concentrations of products P (CO
2
, H
2
O) are much higher than concentrations
of reactants and therefore small uctuations of P due to various reactions can
be neglected. Equation (3.35) is nonlinear, but can be approximated around
stoichiometry with the 1
st
order Taylor expansion as shown above. Since the
input to the system is the excess of oxidants or reductants it is clear that
lambda excess, 1, is an accurate approximation of the true system inlet
(under rich conditions the inlet (R) is taken to have a negative value as it
empties the storage). The largest error occurs if the inlet becomes very rich
because the approximation in (3.35) does not hold. For simplicity, excess
value will in further text be denoted also by , but it should be remembered
that this does not directly represent the normalized air/fuel ratio.
88 Model-based controller
The nal model which holds in both the lean and rich region (taking into
account the analysis in (3.29) and (3.30)) becomes:

out
=
in
k
d
d
dt
(3.36)
Theoretically, only one scaling parameter k
d
is sucient; this can simply be
derived from the stoichiometry. Due to the nonlinearities in lambda signals and
some other eects that will be discussed in the next chapter it will sometimes
be necessary to introduce two parameters depending upon the inlet signal being
rich or lean.
The previous analysis was based on a reactor with its length approaching
zero. The analysis that has lead to the expression for the outlet lambda value
(3.36) is also valid for the complete reactor. In that case the local variable
is replaced with that denotes the relative oxygen coverage of ceria for the
complete reactor. The expressions for the oxygen storage lling and emptying
rate become, however, much more complex. The expressions (3.31) and (3.32),
which have linear dependence on the relative oxygen coverage of ceria, become
nonlinear even when two small reactors are connected in series. The complete
reactor is a distributed parameter system that is modeled as a series of small
reactors. The global reaction rate can therefore be expressed as:
d
dt
= k
gr

in
f() (3.37)
Function f() is nonlinear and k
gr
is a scaling factor. f() also depends on
the inlet feed and there will in fact be two functions: f
L
() for lean inputs
and f
R
() for rich inputs. The inlet lambda value has replaced the oxygen
gas concentration and CO noble metal coverage that were used as inputs by
the local models (3.31) and (3.32). While in the lean case this assumption is
straightforward, in the rich case it has to be additionally assumed that the inlet
signal is a good representation of surface coverages. By introducing (3.37) into
(3.36), where replaces , the following expression for the outlet lambda value
is obtained:

out
=
in
(1 k
d
k
gr
f()) (3.38)
The expression (1 k
d
k
gr
f()) is bounded in the interval [0,1] since the outlet
lambda cannot have the opposite sign of the inlet lambda. It also cannot
exceed the inlet lambda in amplitude. The rst assumption is not completely
true during a rich-lean step, as shown in section 2.4.2. This problem will
be addressed later. The second assumption can also be violated if a large
desorption of species from the catalytic surface occurs.
Since a function bounded in [0,1] can also be represented by (1 f()),
with f() [0, 1], the two scaling parameters are not necessary to be included
3.4. Control-oriented model of the catalytic converter 89
in the model. Therefore, k
gr
=
1
k
d
is assumed. The nal model becomes thus
very simple:
d
dt
=
1
k
d

in
f() (3.39)

out
=
in
k
d
d
dt
=
in
(1 f()) (3.40)
The process model is basically a nonlinear integrator. The integrator gain
is the inverse of the scaling parameter k
d
, which gives an indication of the
oxygen storage capacity. It also depends on the applied exhaust mass ows
because a higher mass ow will ll up the oxygen storage faster. The only
state, , represents thus the mean relative oxygen coverage of ceria throughout
the converter and the system output,
out
, is dependent on the state derivative
and process input (feedthrough). The function f() represents, in fact, the
relative conversion. For a lean inlet this function is decreasing, meaning that
the oxygen storage rate is the largest when the storage is empty, while for a rich
inlet the function is increasing. These two functions, as well as the parameter
k
d
, have to be determined by process identication. The model assumes that
the storage can only be lled with a lean inlet and emptied with a rich inlet.
When function f() becomes zero the outlet lambda equals the inlet lambda,
and the system is in steady state. This is in line with the simulation results
with the rst principle model. Experimental results in the next chapter will
show however that even this assumption, with a real sensor applied, does not
always hold.
The above model assumes that when the inlet feed becomes stoichiometric,
the system is immediately in steady state because the storage is neither lled
nor emptied. This is in principle true apart from conditions with very low ceria
coverage. It was shown in section 2.4.2 (gure 2.20) that after a rich to lean
step the outlet stays rich for some time even though the inlet is lean. This is a
consequence of CO desorption from the ceria surface and a slight HC desorption
from the noble metal surface. The control-oriented model (3.39,3.40) cannot
describe this behavior as the inlet and outlet lambda cannot have dierent
signs. A two state model with one state being ceria bulk and the other state
ceria surface (which can store both oxygen and CO) could be used. This model
would however be much more complicated to estimate and to be used later for
control. Moreover, the bulk capacity is much larger than the surface capacity.
The net eect of the storage of both oxidants and reductants on ceria is that
the observed oxygen storage capacity increases. Thus, it can still be modeled
with one capacity but with an additional function to account for the desorption
eect:
90 Model-based controller
d
dt
=
1
k
d
(
in
f() +g()) (3.41)
The expression for
out
, (3.40), remains the same, accounting for the new
expression for d/dt. The function g() forms the autonomous part of the
model that is activated only when the inlet is stoichiometric or lean. Since the
function values are positive, it becomes possible to model rich outputs with
lean inputs.
3.4.2 Parameter estimation
Since the catalytic converter changes its behavior with time, i.e. ages, the
model parameters should be adapted on-line. Therefore a simple parameter
estimation algorithm has to be available. The parameters and functions to be
estimated are k
d
, f() and g(). The method applied here is a two step method.
As will be shown, this method is very eective since the required testing time
is very short. A long test procedure causes larger exhaust emissions and it is
more dicult to perform during standard operation of the system.
From (3.40) it follows:
k
d
d = (
in

out
)dt (3.42)
If a test begins with empty oxygen storage ( = 0) and ends after T
ss
seconds
with a completely lled oxygen storage ( = 1) the following holds:
k
d
=
_
T
ss
0
(
in
(t)
out
(t)) dt
_
1
0
d
=
_
T
ss
0
(
in
(t)
out
(t)) dt (3.43)
The same holds if the oxygen storage is initially lled and empty at the end
of the test. The most obvious choices of the inlet signal for the parameter
estimation are (at least two) step tests, lean to rich and rich to lean. For use
on a digital computer the above equation can simply be approximated by the
following sum:
k
d
=
N

k=1
(
in
(k)
out
(k)) T
s
(3.44)
where N is the total number of samples, T
s
sampling time (N =
T
ss
T
s
), and k
the current sample.
The estimation procedure has two parts in which the same data has to be
used. Therefore this procedure in principle is an o-line estimation, but due
3.4. Control-oriented model of the catalytic converter 91
to its simplicity in can be used on-line in todays on-board computers. The
rst step is the determination of the capacity factor k
d
from one step response
by using (3.43). By using the same data, equation (3.42) and with known k
d
,
function (t) can be determined and used to estimate the function f() (g() is
here neglected). It follows from (3.40):
f() = 1

out

in
(3.45)
By calculating the above function for all samples, F(k) is obtained. This
can be approximated by a simpler function

f((k)) in a least squares manner:
min.
N

k=1
_
F(k)

f(k)
_
2
(3.46)
Since the function

f((k)) is a general nonlinear function a piecewise linear
function is selected. This function can be written as a linear combination of
triangular basis functions:

f() =

n
i=1
b
i
()f
i

n
i=1
b
i
()
(3.47)
where:
b
i
() =
_

_
0, if <
i1
, i 2

i1

i1
, if
i1
<
i
, i 2
1

i

i+1

i
, if
i
<
i+1
, i n 1
0, if
i+1
, i n 1
(3.48)
Parameters
1,2..n
and f
1,2..n
, where n is the number of basis functions, are
tuning parameters. If basis functions (
i
) are xed then the algorithm (3.46)
has an analytic solution since (3.47) is linear in parameters f. Details on the
least squares algorithm will be shown in section 3.6.3.
Experience has shown that good results are obtained by predening the
basis functions. Usually a piecewise function with ve points is enough. One
point is always predened as f
L
(1) = 0 and f
R
(0) = 0. Other points of interest
are = 0 for f
L
and = 1 for f
R
. Function f
L
() is typically very close to 1
for small , and f
R
() is close to 1 for large . Therefore it is very important to
select a point where the function begins to fall. That can be done on the basis
of data, F((k)). The remaining two points (if a ve point function is selected)
can be chosen by a linear sectioning of the remaining variable range.
In the case of a rich-lean step the function g() has also to be determined.
Since this function only describes the cases when inlet and outlet lambda signals
have dierent signs, the data set to train this function is dened as follows
(assuming that
in
> 0):
92 Model-based controller
G(k) =
_

out
, if
out
< 0
0, if
out
0
(3.49)
F(k) =
_
1, if
out
< 0
1

out

in
, if
out
0
The function to g() that has to be estimated is also taken as a piecewise
linear function and the same least squares algorithm that was used to nd

f()
is applied. This function does not have to be as accurately estimated as

f(),
so fewer points (2 or 3) are usually enough. After determination of G(k), F(k)
is found as shown in (3.49).
The control-oriented model prediction is compared to the rst principle
model developed in the previous chapter. Figure 3.10 compares the outlet
lambda signals of both models. The predicted relative oxygen coverage of
ceria is also compared with the mean coverage of the ceria surface and bulk
throughout the reactor. This latter quantity was calculated by assuming that
besides oxygen (and oxides of nitrogen) that are stored on the ceria surface and
in the bulk, also the ceria surface not covered by CO attributes to the oxygen
storage capacity. Therefore the mean coverage is calculated by the following
expression:

fp
= w
1

bulk
+w
2

surf(O
2
)
+w
2

surf(CO)
(3.50)
where
fp
is the overall average relative oxygen coverage of ceria calculated by
the rst principle model;
bulk
and
surf(O
2
)
are mean coverages of bulk and ce-
ria surface by oxygen and oxides of nitrogen, while
surf(CO)
is complementary
coverage of the ceria surface by CO. The latter is calculated by subtracting the
average CO coverage from 1. w
1
and w
2
are weighting parameters calculated
from the ratio of bulk and surface coverage.
The t between the control-oriented and the rst principle model is rather
good. The prediction of the outlet lambda signals is very accurate, while an
error up to 5.5% occurs in the prediction of ROC. The main error stems from
some eects not taken into account by the control-oriented model. Namely,
the degree of ceria coverage slightly deviating from 1 under lean steady state
conditions. For example, for the inlet lambda of 1.04 the steady state cover-
age is approximately 98.2%. This number increases with increasing lambda as
the storage release equilibrium becomes shifted more into the storage direc-
tion. Therefore, the oxygen storage is actually slightly emptied as the inlet
lambda shifts to rich even before it actually enters the rich region. This ef-
fect contributes to the above mentioned modeling error. Another error source
is the simplication of rather complicated nonlinear dynamics of the catalytic
converter.
3.4. Control-oriented model of the catalytic converter 93
0 2 4 6 8 10
0.04
0.02
0
0.02
0.04
time [s]
l
a
m
b
d
a

[

]
0 2 4 6 8 10
0.04
0.02
0
0.02
0.04
time [s]
l
a
m
b
d
a

[

]
0 2 4 6 8 10
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
0 2 4 6 8 10
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
Figure 3.10: Comparison of control-oriented and rst principle model outputs
during inlet lambda steps. Solid line: rst principle model, dotted line: control-
oriented model, dashed line: model input.
As already discussed in the previous chapter, not all oxygen stored on the
ceria surface is used during the rich step due to surface inhibition eects. In
the given case this amount is close to 14%. The control-oriented model assumes
that when steady state is reached in the rich region the ROC becomes zero.
The amount of unused oxygen does not have any inuence on the dynamic
response. In order to compare the outputs of both models the ceria coverage
for the rst principle model (
fp
) had to be rescaled as shown in gure 3.11.
This gure also shows the relationship between the mean values of surface and
bulk coverages. Responses are very similar for all variables except for CO
desorption from the ceria surface after the rich-lean step, which is independent
(proceeds faster) of the oxygen storage on the surface and in the bulk.
Model in a wide operating range
Model parameters are subject to slow changes in time due to aging of the
catalyst. These changes can be accounted for by periodic parameter adaptation
and should not pose a serious problem for the control system. It is more
important to account for fast changing variables such as exhaust mass, inlet
lambda signals of various amplitudes and changing reactor temperatures.
It was experimentally observed that by adapting only the scaling factor, k
d
,
94 Model-based controller
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
bulk
surf. O
2
surf CO
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
no scaling
scaled
Figure 3.11: Above: comparison of average surface and bulk ceria coverage.
Below: scaling of ROC.
the inuence of a changing space velocity cannot completely be accounted for
[76]. The function f() has to be adapted also when the space velocity changes.
The reason for this comes directly from the underlying kinetics. The adsorption
rate of oxygen on the ceria surface at some point in the reactor depends on the
oxygen concentration in the washcoat and the fraction of empty ceria surface as
shown in (3.31). The reaction rate between oxygen stored on the ceria surface
and CO (or hydrocarbon) absorbed on the noble metal surface depends on the
latter surface coverage and the fraction of the ceria surface covered by oxygen
(3.32).
If the function f
R
() would not be dependent on the inlet mass ow, it
would mean that for a given relative ceria coverage the conversion would not
change with a changed mass ow (3.40). So, if the mass ow increases with
some factor, the reaction rate would also have to increase with the same factor
to retain the same conversion. Provided that the analysis of a single reactor
point can be applied to the complete reactor behavior, and that the noble metal
surface coverage does not change considerably, ceria coverage has to change
with the same factor as the mass ow to retain the same conversion as in the
nominal case (3.32). The same reasoning holds for the oxygen adsorption on
the ceria surface.
If the inlet lambda signal amplitude changes from the nominal value (the
value at which the estimation was done), the function f
R
() also has to be
3.4. Control-oriented model of the catalytic converter 95
adapted. Because the inlet concentrations of reducing components increase
with a richer inlet lambda, the reaction rate should also have to increase for
the conversion to remain the same. Since this is not the case assuming the
inlet lambda does not inuence the surface coverage, f
R
() has to be adapted
for dierent inlet lambda amplitudes. Because the adsorption rate depends
explicitly on the oxygen concentration, f
L
() does not have to account for the
inlet amplitude variations.
The function g() depends on the exhaust mass ow. If the ow increases
the eect of CO desorption will reduce because the surface capacity is constant.
The model parameters for dierent operating conditions can therefore be
recalculated from the values estimated in a nominal case by the following ex-
pressions:
k
d
=
m
exn
m
ex
k
dn
g() =
m
exn
m
ex
g
n
()
f
L
_
1
m
ex
m
exn
+
m
ex
m
exn
_
= f
Ln
() (3.51)
f
R
_

m
ex
m
exn

in

inn
_
= f
Rn
()
The subscript n denotes the nominal case.
Without the above adaptation the tests for parameter estimation would
have to be performed in many operating points. This would lead to a large time
consumption of the model estimation procedure. Such a long procedure would
also lead to large exhaust emissions because the engine would have to run rich
or lean for a longer period of time. With (3.51) included, the procedure has to
be undertaken only in few operating points, mainly at dierent temperatures.
This mapping holds for most of the operating conditions. In some cases, i.e.
very high space velocities, it cannot be directly applied. This will thoroughly
be discussed later.
Figures 3.12 and 3.13 show the model t during the lambda step tests at
dierent exhaust mass ows. In the case of two times decreased ow the model
t is very good for the lean to rich step and slightly worse for the rich to lean
step. It is more dicult to correctly predict the latter because of the interaction
between CO desorption from the catalytic surface and oxygen/NO adsorption.
The case with no adaptation of functions f
L
and f
R
(3.51) is also shown. It
is clear that the adaptation is necessary for a good prediction of the outlet
lambda signal. This is not so important for the prediction of ROC as small
errors in functions f
L
and f
R
do not imply large errors in ROC. It will be shown
in section 3.6 that a good prediction of the outlet lambda is important for a
good tuning of the controller.
The prediction in the case of two times increased exhaust mass ow is also
better with the adaptation of f
L
() and f
R
(). It has to be noted, however,
96 Model-based controller
0 5 10 15 20 25 30
0.05
0
0.05
l
a
m
b
d
a

[

]
inlet
f.p. model
c.o. model
c.o. model2
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
f.p. model
c.o. model
c.o. model2
Figure 3.12: Comparison of control-oriented and rst principle model outputs
during inlet lambda steps with m
ex
=27 kg/h. Model2 is the model without
the adaptation of f
L
() and f
R
().
0 5 10 15 20 25 30
0.05
0
0.05
l
a
m
b
d
a

[

]
inlet
f.p. model
c.o. model
c.o. model2
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
f.p. model
c.o. model
c.o. model2
Figure 3.13: Comparison of control-oriented and rst principle model outputs
during inlet lambda steps with m
ex
=108 kg/h. Model2 is the model without
the adaptation of f
L
() and f
R
().
3.4. Control-oriented model of the catalytic converter 97
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1

f
L
27 kg/h
27kg/happrox.
108 kg/h
108 kg/happrox.
54 kg/h
Figure 3.14: Function f
L
() with various mass ows. Comparison of directly
estimated f
L
() with f
L
() obtained by (3.51).
that in the case of the positive step the direct application of (3.51) does not
yield good results because of the saturation eect. Function f
L
() typically
decreases for increasing exhaust mass ows. This means that also the break-
through ROC (relative ceria coverage at which a breakthrough of oxygen starts,
i.e. f
L
<0.98) decreases. Above some mass ow this value starts decreasing
much slower, however. The reason is that not only oxygen/NO adsorption on
the ceria surface but also the desorption of CO are responsible for the oxy-
gen breakthrough. Even though merely adsorption is not able to prevent the
breakthrough, with the additional CO desorbing this eect is slowed down at
higher mass ows. Therefore the saturation-like eect takes place. The re-
sponse shown in gure 3.13 is obtained with saturation taking into account.
The nominal and adapted functions f
L
() and f
R
() are shown in gures
3.14 and 3.15. The adapted nominal function is compared with the functions
obtained by a direct parameter estimation from the responses at given mass
ows. The agreement is satisfactory and it is clear that the adaptation is
necessary to obtain a good model. The worst match is, as expected, in the case
of the lean input with a 2 times increased mass ow, because the saturation
eect was not taken into account.
Another problem is the dependence of the observed oxygen storage capac-
ity on the space velocity. This variable increases with a decreased mass ow
because more oxygen can be used during the rich step. In the previous chapter
it was shown that because of the surface inhibition not all oxygen from ceria
is used during the rich inlet. With a reduced mass ow this inhibition is lower
and thus the observed oxygen storage capacity increases. This increment was
6% for the halved mass ow, while for the increased mass ow the decrease
98 Model-based controller
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1

f
R
27 kg/h
27kg/happrox.
108 kg/h
108 kg/happrox.
54 kg/h
Figure 3.15: Function f
R
() with various mass ows. Comparison of directly
estimated f
R
() with f
R
() obtained by (3.51).
of the observed oxygen storage capacity was 12%. It is dicult to know the
exact capacity dependence on mass ow in advance. In the rst principle model
these dierences are typically up to around 10% when the mass ow changes
two times. This can be used to approximately account for this phenomenon.
However, it would be helpful to include a couple of operating points with the
same temperature and dierent mass ows during the parameter estimation to
asses the correct characteristic for a given converter.
Figure 3.16 shows the model prediction in the step test with a smaller input
amplitude. The negative step is well predicted, with the largest error resulting
from the slight increase in the oxygen storage capacity that was not accounted
for. The adaptation (3.51) was again necessary to improve the t. The posi-
tive step response was not so well predicted largely due to a high inuence of
the peak caused by the NO breakthrough. This peak is so large because the
NO conversion drops very fast, while oxygen breakthrough increases slowly.
Because the model is based on the oxygen response this eect is not well pre-
dicted. In a real system it can be expected that also NO conversion decreases
more slowly, but typically before the oxygen outlet increases [16]. The latter is
correctly predicted by the rst principle model.
3.4.3 Experimental model verication
An initial experimental model verication is presented in this section. The
experimental data was provided by dmc
2
, Hanau, Germany. Converter step
responses have been recorded in dierent operating points, i.e. dierent exhaust
mass ows with the same inlet temperature of 620K. Such a selection was made
3.4. Control-oriented model of the catalytic converter 99
0 5 10 15 20 25 30
0.02
0.01
0
0.01
0.02
l
a
m
b
d
a

[

]
inlet
f.p. model
c.o. model
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
f.p. model
c.o. model
Figure 3.16: Comparison of control-oriented and rst principle model outputs
during inlet lambda steps with smaller amplitude.
to verify the assumptions about the dependence of f() on the exhaust mass
ow. More rigorous model validation will be presented in chapter 4. The
used catalyst was a Pt/Rh/CeO
2
catalyst on an engine bench equipped with
a BMW engine. It should be noted that the applied catalyst diered from
the one used for building the kinetic model. For the validation of the control-
oriented model parameter estimation this does not matter, however, since the
developed methods are catalyst independent. First, the responses of a fresh
and aged catalyst are compared. After that, the model prediction is compared
with the responses of the fresh catalyst.
Fresh vs. aged catalyst
A periodic on-line model adaptation is primarily needed due to the changes
of the process dynamic behavior because of aging. Dynamic responses of the
converter with a fresh and a strongly aged catalyst are shown in gure 3.17. In
both cases the applied mass ow is 50 kg/h. The conditions are similar but not
completely the same as can be observed from the inlet signals. For a qualitative
comparison the tests are satisfactory, however. The conversion characteristic of
the converter changes dramatically with aging. Mostly aected are steady state
conversions under rich conditions as the conversions of all components (CO, HC
and NO) drop. Under lean conditions the dierence is smaller with somewhat
100 Model-based controller
0 50 100 150 200
0
0.2
0.4
0.6
0.8
1
1.2
time [s]
C
O

c
o
n
c
.

[
v
o
l
%
]
0 50 100 150 200
0
500
1000
1500
2000
time [s]
N
O

c
o
n
c
.

[
v
o
l
%
]
0 50 100 150 200
0
100
200
300
400
500
time [s]
H
C

c
o
n
c
.

[
v
o
l
%
]
0 50 100 150 200
0.97
0.98
0.99
1
1.01
1.02
1.03
1.04
time [s]
l
a
m
b
d
a

[

]
freshin
freshout
agedin
agedout
Figure 3.17: Lambda step responses with a fresh and a strongly aged catalytic
converter.
lower conversion of HC. Another dierence is in the dynamic response what
can most notably be seen in the lambda responses. The oxygen storage and
release capability of the aged converter has greatly been reduced as there is
only a small dierence between the inlet and outlet lambda signals after the
steps. Moreover, the lambda plateau is completely gone because there is no
high conversion of CO and HC after the lean-rich step and NO after the rich-
lean step. Note that it is better to discuss the dynamic responses on the basis of
lambda signals than actual analyzer signals because analyzers that were used
have a slow dynamic response with time constants of several seconds. The
aging of a catalytic converter is in principle even more complicated as it was
shown in [90]. The oxygen storage capacity increases with initial aging when
compared to a completely fresh converter. This only indicates again that it
is very dicult to accurately describe the dynamics of the converter without
some on-line identication tests. On the other hand, when the oxygen storage
becomes deactivated the oxygen storage controller does not have a big chance
to succeed due to a low capacity. At the same time, the catalytic converter
eciency may not drop severely enough for a replacement to become necessary.
Some further reference to aging and oxygen storage capacity will be given in
chapter 4 but it will not be studied in detail in this thesis.
The dynamic response of the fresh catalyst in gure 3.17 is similar to that
in the simulations with some striking dierences when the lambda signals are
3.4. Control-oriented model of the catalytic converter 101
closely inspected: the plateau is not at = 1, there is a steady state bias both
in the rich and the lean area and there is a lambda peak in the signal of the
downstream sensor after the lean to rich step.
Data preprocessing
One of the major problems when applying strategies that use pre- and post-
converter lambda sensors is the dierent characteristic of the two sensors. By
denition, the lambda signals upstream and downstream of the catalytic con-
verter have to be the same in the steady state. Dierent sensor characteristics
of wide range sensors were studied by Germann et al. [29]. The authors have
used static experiments with predened feeds. With lean feeds applied, the
sensor outputs upstream and downstream the converter matched, but a large
mismatch was observed with rich feeds applied. The authors concluded that the
dierences in the sensor characteristics were probably due to dierent coe-
cients of diusion for various exhaust components through the porous catalytic
layer of the sensor. Since composition of the exhaust mixture changes behind
the converter this can lead to dierent sensor characteristics. The largest dis-
turbance is likely to be caused by hydrogen as it can diuse easier and the
sensor can feel more hydrogen than the exhaust actually contains. Since the
sensor characteristic is dependent on the actual exhaust gas composition the
sensor is tuned to accurately represent the engine-out . This will inevitably
lead to steady state errors in both rich and lean region when the same sensor is
used to measure the signal behind the catalytic converter. The errors in the
rich region are larger because the hydrogen concentration is higher. Since the
sensor itself is a small catalytic converter it is also aging and its characteristic
changes. This can lead to a sensor bias as observed in gure 3.17 where the
plateau is shifted to approximately 1.009. This bias can occur in both sensors in
front of and behind the converter and is not a result of exhaust characteristics.
Even more complex sensor behavior is observed during step tests. While
with lean mixtures a small steady state error occurs, with rich inlet applied the
post-cat lambda sensor has a typical response with overshoot and a steady state
error. Such a response cannot be explained solely by the converter dynamics
[29] and is not supported by accompanying measurements of separate exhaust
components where no peak is observed in responses of CO, hydrocarbons or
NO behind the converter. Moreover, such a response peak would mean that
the outlet temporary becomes richer than in steady state, which would be
quite dicult to explain. It is more likely that due to gas composition changes
during the step, lambda sensor characteristic also changes as a result. Another
experimental study with fast exhaust analyzers applied, [50], showed that the
peak in the outlet lambda response coincides with the depletion of oxygen
from ceria. At that time the CO outlet practically reaches steady state. The
transient response of hydrocarbons takes much more time, and is probably
102 Model-based controller
not completely related to the oxygen storage and release only. It is a bit more
dicult to do the same analysis here because the used analyzers are rather slow,
as already mentioned. It is known that water-gas shift and steam-reforming
reactions become more important after most of the oxygen from ceria has been
used [98]:
CO +H
2
O CO
2
+H
2
(3.52)
C
x
H
y
+xH
2
O xCO + (x +
y
2
)H
2
(3.53)
The steam reforming reaction (3.53) could explain the long transient of hy-
drocarbons. A product of this reaction is hydrogen, which can deceive the
sensor, which would then predict a richer feed than it actually is. The highest
steam reforming inuence is expected when oxygen from ceria becomes de-
pleted. This can form an outlet hydrogen peak and that could explain the
lambda sensor peak. The outlet hydrogen concentration that is higher than
the inlet concentration was indeed predicted in [90]. With the steam reforming
reaction becoming less intense the hydrocarbon concentration relaxes to the
steady state and so does the lambda sensor signal behind the converter. In
reality the water-gas shift and steam reforming reactions do not change the
lambda value, so the lambda sensor would actually have to reach the steady
state when oxygen from ceria is depleted, i.e. at the time of the post-catalyst
lambda sensor peak. This whole hypothesis cannot be proven at the moment
because neither hydrogen nor water signals have been measured. The rst
principle model does not account for water-gas shift and steam reforming re-
actions as there are no elementary step kinetic data available. Further studies
are therefore needed. When the exact cause of sensor errors becomes known,
it will be possible to make a soft sensor which would correct for the errors of
the real sensor.
A simple solution is exploited here. The outlet lambda signal after the rich
step is rescaled in such a manner that the peak value coincides with the inlet
lambda signal. Since that is the moment of the oxygen storage depletion the
outlet lambda value in the rest of the step response is set to be equal to the
inlet value. In some cases there is also a small steady state dierence in the pre-
and post-converter lambda signal with lean inlet mixtures. The outlet lambda
signal is then also rescaled in such a way that the steady state values of both
sensors match. The oset present in both inlet and outlet signals was removed.
This rescaling mostly aects sensor outlets close to steady states. During a
standard controller operation the catalyst outlet is kept close to stoichiometry
(high conversion) so the eect of sensor errors may not be of a very high
importance. The sensor characteristic will further be addressed in section 4.3.
3.4. Control-oriented model of the catalytic converter 103
0 20 40 60 80 100 120 140 160 180 200
0.03
0.02
0.01
0
0.01
0.02
0.03
time [s]
l
a
m
b
d
a

[

]
inlet
outlet
model
outlet w.o.scaling
Figure 3.18: Model predictions vs. measurements with the inlet ow of 50
kg/h. The model parameters were estimated on the same data set. Sensor
outputs with and without scaling are shown.
Model prediction
Figure 3.18 shows the agreement between measurements and prediction for the
data set on which the parameter estimation was performed (ow=50kg/h, inlet
temperature 620K). The estimation of the model parameters was performed on
the rescaled lambda outlet signal. A comparison with the raw signal is also
shown in the same gure. The obtained t is quite satisfactory.
The estimated model was used to predict the converter step responses at
the same inlet temperature and inlet mass ows of 80 and 110 kg/h. The model
adaptation (3.51) was used. Figure 3.19 shows the t between the model and
experiments. The adaptation of f
R
() leads to a very good t during the lean to
rich step, which is in line with the simulation results from the previous section.
The adaptation of f
L
() appeared, however, not to be necessary. This is also in
line with the observed saturation in simulations, as the values of the nominal
function f
L
() were already quite low (fast breakthrough). This implies that the
parameter estimation should always be performed at lower mass ows, outside
of the saturation area, because if the estimation is performed with saturation
activated it is hard to determine at which (lower) mass ow the saturation will
become deactivated.
The oxygen storage capacity was almost the same at 50 and 80 kg/h mass
ows. However, at 110 kg/h the observed capacity has decreased by 20%,
104 Model-based controller
0 20 40 60 80 100 120 140 160 180 200
0.03
0.02
0.01
0
0.01
0.02
0.03
l
a
m
b
d
a

[

]
inlet
outlet
model
outlet w.o.scaling
0 20 40 60 80 100 120 140 160 180 200
0.03
0.02
0.01
0
0.01
0.02
0.03
time [s]
l
a
m
b
d
a

[

]
flow = 80kg/h
flow =110 kg/h
Figure 3.19: Model predictions vs. measurements with inlet ows of 80 and
110 kg/h. The model parameters were estimated at 50kg/h. Sensor outputs
with and without scaling are shown.
what is qualitatively in line with the rst principle model predictions. It re-
mains, however, a very nonlinear eect dicult to quantitatively account for
in advance.
The increased temperature also leads to an increase in oxygen storage ca-
pacity. It increases 10% when the inlet temperature increases from 620K to
720K at the inlet mass ow of 80kg/h. The model predictions are not shown
but there is a somewhat larger mismatch than previously because the inuence
of the temperature on f() is not yet included in the model. The reaction rates
are somewhat higher, as one could expect at a higher temperature.
3.5 Feasibility of control
Some basic foundations of control of a catalytic converter will be given in this
section. A simple, gain scheduling controller will be developed to test whether
desired goals can be met by using inferential control (model used as the sensor).
The most important goal is that the not measured controlled variable, relative
oxygen coverage of ceria, stays (follows) at the desired value under all operating
conditions. The control scheme is a cascade system with the inferential oxygen
storage controller in the outer loop and a standard model-based air/fuel engine
controller (section 3.3) in the faster inner loop of the control system, as shown
3.5. Feasibility of control 105
control IMC Engine
Eng_mod
Catalyst
_ref c_m
c_est
_est
e_m
e_est
e_err
Cat_mod
- -
- -
Fuel
control IMC Engine
Eng_mod
Catalyst
_ref c_m
c_est
_est
e_m
e_est
e_err
Cat_mod
- -
- -
Fuel
Figure 3.20: Cascade inferential oxygen storage controller.
in gure 3.20. Such a control scheme is natural since the controlled variable of
the engine controller, air to fuel ratio, is the input for the catalytic converter.
Moreover, the inner loop is usually much faster then the outer loop, so its
dynamics can in that case be neglected during tuning of the catalytic converter
controller.
3.5.1 Gain scheduling controller
The outer loop of the controller calculates the reference lambda signal in order
to keep the relative oxygen coverage of ceria at the desired value. Since this
variable is not measurable, the model is used for prediction, as shown in g-
ure 3.20. This signal is then used as calculated feedback signal. Since the inner
closed loop has a faster dynamics, it can be assumed that the engine reaches
the desired air/fuel ratio instantaneously, i.e.
e
=
ref
. A possible deviation
of the engine air/fuel ratio can be considered as system disturbance. The pro-
cess delay, which always exists in the path between the fuel injector and the
entrance to the catalytic converter, should not be neglected however. Hence
the controlled process very much reects the catalytic converter dynamics and
can (for one operating point) be given by the following transfer function:
G
p
(s) =
K
P
s
e
T
d
s
(3.54)
The process gain highly depends on the operating conditions, as K
P
=
f
d
()/k
d
. The controller can assume a very simple P structure:

ref

ref

est
=
K
c

c
s + 1
(3.55)
Since the model is controlled, the controller gain can be adapted in such a
way that a smooth and fast transition is achieved. The system bandwidth is
limited by the transport delay, but large amplitudes of the controller output
106 Model-based controller
could also cause driveability problems. The controller output (
ref
) is limited
between 0.9 and 1.1. The gain of the controller is limited at high frequencies
by the low-pass lter (time constant
c
) to achieve a high frequency noise
attenuation. The reference following has a zero steady state oset because of
the integrating behavior of the process. A known standard drawback of the P
controller is its incapability to remove a steady state error in the presence of
system disturbances. Due to the engine controller this problem does not occur
as the steady state air/fuel ratio value equals the reference value and thus no
steady state disturbance exists.
The controllers reference following and disturbance tracking capabilities
have been tested. Figure 3.21 shows the performance of the control system dur-
ing tracking of the reference relative oxygen coverage of ceria, which changes
from 1 to 0.5. This test resembles the condition after a fuel cut-o when the
oxygen storage is lled with oxygen and a fast system response is needed to re-
store the optimal system performance. The second test presented in gure 3.22
is the reference tracking from 0 to 0.5. This test can for instance simulate the
system performance after a longer period with a rich feed (acceleration), which
has completely depleted oxygen from the storage. Again a fast and accurate
response is required for a high performance of the system. The reference tra-
jectory is followed quite fast. The control-oriented model predicted the relative
oxygen coverage with an error within 5%, which was expected on the basis of
dynamic predictions shown in the previous section. This steady state error is,
however, not as large a problem as the possibility of drift that can not be ob-
served and is the consequence of the integrating process behavior. Such a drift
can for example occur due to a small bias in the sensor signal. The outlet error
can only be detected by the dierence between the converter outlet lambda
value and the model predicted outlet lambda value. The controller should also
be able to distinguish between the error imposed through disturbances not
taken into account (i.e. dierence between the measured and actual lambda
signal) and errors that occur due to modeling errors. Accurate modeling has
therefore a great impact on the control system performance. Imposing step-
wise changes of the throttle position, while the relative oxygen level should be
maintained at the desired level (here 0.5), tests disturbance rejection capabili-
ties of the controller (gure 3.23). The control system behavior is fairly good,
however with a clear steady state error again. This follows not only from mod-
eling errors that were already discussed before, but also from the distributed
parameter nature of the catalytic converter. Together with the averaged ceria
coverage, the distribution of oxygen throughout the reactor can assume an im-
portant role when trying to exactly control the relative oxygen coverage. This
is however not taken into account in the control-oriented model as the model
would have to be much more complex and the parameter estimation much more
time consuming.
3.5. Feasibility of control 107
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.2
0.4
0.6
0.8
1
R
O
C

[

]
ref.
c.o. model
f.p. model
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.9
0.92
0.94
0.96
0.98
1
1.02
time [s]
l
a
m
b
d
a

[

]
ref.
exhaust
Figure 3.21: ROC reference following during transition from the completely
lled to half lled oxygen storage. Above: reference, rst principle and control-
oriented model ROC. Below: Reference and exhaust lambda.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
0.2
0.4
0.6
0.8
R
O
C

[

]
ref.
c.o. model
f.p. model
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
ref.
exhaust
Figure 3.22: ROC reference following during transition from the empty to half
lled oxygen storage. Above: reference, rst principle and control-oriented
model ROC. Below: Reference and exhaust lambda.
108 Model-based controller
0 1 2 3 4 5 6 7 8 9 10
15
20
25
30
35
t
h
r
o
t
t
l
e

[
d
e
g
]
0 1 2 3 4 5 6 7 8 9 10
0.4
0.5
0.6
R
O
C

[

]
0 1 2 3 4 5 6 7 8 9 10
0.9
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
ref.
exhaust
Figure 3.23: Disturbance (throttle changes) response of the system. Above:
Throttle position. Middle: ROC in rst principle model. Below: Reference
and exhaust lambda.
3.5.2 Inuence of the sensor oset on the control ro-
bustness - steady vs. oscillatory
Steady feed
It was shown in section 3.4.3 that wide range lambda sensors usually have a
steady state oset that can change during the system operation. This can lead
to a drift of the controlled variable because it is in principle unobservable until
some disturbance occurs. Figure 3.24 shows the cases presented in gures 3.21
and 3.22, but with an oset of 0.005. Thus, when the controller assumes that
the system is settled in a steady state with a stoichiometric lambda, ROC is
slowly drifting toward the upper limit with the lambda of 1.005.
A striking observation in gure 3.24 is that the slopes of oxygen storage
lling are not the same. The reason is the distributed parameter nature of the
catalytic converter. The controller brings in both cases ROC close to 0.5 but
in the rst case (with initially empty oxygen storage) the oxygen coverage is
mostly in the front part of the converter, while in the second case in the back
part of the converter. NO has a lower ability than oxygen to be stored on ceria
and it usually prots from oxygen storage on ceria to occupy the noble metal
surface, dissociate and subsequently react with CO or HC. When the front
part of the converter is already lled with oxygen a part of NO can neither be
3.5. Feasibility of control 109
0 1 2 3 4 5 6 7 8 9 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
R
O
C

[

]
Figure 3.24: ROC control with an unforeseen lambda sensor oset of 0.005.
converted nor adsorbed in the back part of the converter as all reductants are
already used by oxygen in the front part of the converter. There is therefore
a NO breakthrough and lower slope of ROC. In the second case NO is fully
converted so ROC increases faster.
If the downstream sensor has the same oset as the upstream sensor this
error may not be observed in the steady state. When two sensors have dierent
osets the error can be observed because in the steady state the inlet lambda
has to be equal to the outlet. Downstream sensor errors, discussed earlier, can
also help to detect an oset.
The problem is that the equilibrium point (i.e. half lled oxygen storage)
is unstable as any input that is not precisely equal to stoichiometry will lead
to drift. If the drift is constant the feed will stay constantly rich or lean, and
therefore the steady state will only be reached when the upper or lower limit
is reached. This is of course not acceptable. Therefore, it is very important
to accurately estimate the sensor oset. If a disturbance occurs when ROC is
close to a limit, the inlet lambda will depart from stoichiometry and the outlet
lambda will do the same according to (3.40), because f() < 1. If the model
lambda prediction is compared to the measured signal it can also be concluded
whether the controlled variable has drifted from the set point (assuming that
the model is accurate).
A simple method to estimate the oset of the downstream sensor can be
exploited during the parameter estimation. As shown before, after a lean to rich
step the downstream lambda stays at stoichiometry, if the converter functions
properly so that all exhaust species are converted, for a period of time. By
averaging the measured downstream sensor signal at the plateau the estimation
of the sensor bias can be obtained.
The real problem for the control is the upstream sensor bias, as this signal
110 Model-based controller
is used as a direct input to the model. If the sensor signal is denoted with
sen
and the actual signal with
in
(that is after subtracting 1), the oset can be
written as
off
=
in

sen
. By using (3.39) the error of the estimated ROC
can be calculated as (g() is neglected):

off
(t) = (t)
est
(t) =
off
(t
0
) +
_
t
t
0
1
k
d

off
f((t)) (3.56)
If the downstream sensor is assumed to be accurate and the model correct
the actual ROC can be estimated from (3.40) since f() is invertible. The model
estimation of ROC can thus be corrected. If
off
(t
0
) = 0, (3.56) can be used
to estimate
off
. The above assumption can be met if the calculation of (3.56)
starts when the oxygen storage is empty or full so the initial estimation of ROC
is accurate. is estimated during an acting disturbance with:
(t) = f
1
((t)) (3.57)
where
f((t)) =

in
(t)
out
(t)

in
(t)
(3.58)
If the same linear lter is applied to both inlet and outlet lambda signals
the above equation still holds. Therefore, to reduce the noise inuence, both
signals can be integrated during the lambda excursion. The sensor low pass
characteristic does not inuence the estimation. Care has to be taken that the
inlet signal does not change sign during the calculation. By integrating (3.56)
and calculating (3.57)
off
can be estimated.
This idea is similar to the algorithm proposed in [81] where a recursive
Markov estimate method is used to asses the parameters of both model and
sensor oset. The model in that case is rather simple, however. The problem
arises because f((t)) in (3.56) is in principle not known. However, if the
converter operates in the region where the conversion is high, f() 1 the
algorithm can be applied. A possible solution is to impose small disturbances
on the controller output to excite the system. The disturbances will, however,
also create emissions, certainly if the process operates in the low conversion
region. Therefore the disturbances should not be applied all the time, but only
when the sensor bias is suspected and until it has been properly estimated. If
both sensor bias and model are not correctly estimated the model should be
updated via the procedure described in section 3.4.2. A method like the one
applied in [81] could be used to estimate the parameter k
d
and
off
on-line.
Oscillatory feed
Systems that run with oscillating engine lambda, i.e. with a switch type lambda
sensor, have a property that the oxygen storage is constantly lled or emptied.
The steady state relative oxygen coverage calculated from the control-oriented
3.6. Model-based predictive control 111
model is obtained when the amount of oxygen stored during the lean half cycle
equals the amount of oxygen emptied during the rich half cycle:
_
t
Rb
t
Ra
1
k
d

in
(t)f
R
((t))dt =
_
t
Lb
t
La
1
k
d
(
in
(t)f
L
((t)) +g((t))) dt (3.59)
where t
Ra
, t
Rb
, t
La
and t
Lb
are the time instances when respectively rich and
lean half cycles start and end.
From (3.59) it follows that even in the case of an unforeseen sensor bias
the steady state will not necessarily be at a completely lled or completely
emptied oxygen storage. For example, if a positive bias exists, such that in a
steady feed case ROC would drift to the full oxygen storage, the oscillating feed
can cause a steady state at a lower ROC, provided that f
R
((t)) > f
L
((t))
during rich and lean half cycle respectively. The above condition is of course
only a sucient condition that steady state does not reach the upper limit. An
example is given in gure 3.25. Relative oxygen coverage in two cases with the
same inlet but dierent mass ows are compared. The initial condition in both
cases is the same. The applied inlet lambda signal is 1.005+0.02sin(20.667t),
thus with a bias of 0.005, oscillation amplitude of 2% and frequency 0.667Hz.
Exhaust mass ows of 50 and 150kg/h are compared. In the case of the lower
mass ow the oscillations can not reduce the drift. On the other hand, when
the mass ow increases, f
L
() drops faster for close to 1 (see gure 3.14) so
an equilibrium is established. It should be noted that though this feature of
oscillations has a favorable impact for the control of ROC, the overall system
performance is deteriorated by the oscillations in the case of a high mass ow.
The oscillations act as a disturbance on the system, and with f
L
() smaller more
NO emissions are produced, as can be seen in gure 3.25. It basically follows
the standard control system trade-o: with the increased system robustness
the performance is decreased. The best solution is to very accurately estimate
the bias and avoid oscillations. In order to increase the robustness oscillations
with a small amplitude can be applied in the steady state to counteract very
small biases that may remain.
3.6 Model-based predictive control
The controller considered in section 3.5 was a simple gain scheduling P con-
troller, which due to the integrating process behavior and no steady state dis-
turbances performs well in tracking of a ROC reference and disturbance rejec-
tion. The real goal of the control system, however, is to reduce the exhaust
emissions. It was shown in section 2.4.3 that there is a correlation between the
control of oxygen storage and high conversion. On the other hand, if only a
good and fast response of ROC is desired, it could produce a very aggressive
112 Model-based controller
0 1 2 3 4 5 6 7 8 9 10
0.7
0.75
0.8
0.85
0.9
0.95
R
O
C

[

]
0 1 2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
time [s]
N
O
o
u
t

[
v
o
l
%
]
Figure 3.25: Above: ROC calculated by the rst principle model (not rescaled)
with the inlet = 1.005 + 0.02sin(20.667t), and with mass ow of 50kg/h
(solid) and 150kg/h (dashed). Below: NO outlet concentration in the cases
above.
inlet lambda control which could accurately control ROC but still produce emis-
sions. The goal is therefore to also include the model prediction of emissions in
the controller tuning procedure. This leads to an MPC based controller that
is presented in this section. This controller, thus, replaces the gain scheduling
controller in the block schema of gure 3.20. The controller solves two prob-
lems: nds an optimal steady state and an optimal control sequence to reach
this steady state. Furthermore, to apply the controller in an on-board computer
a complex on-line optimization has to be avoided. To overcome implementation
problems stemming from the limited computing power, the MPC controller is
approximated with a computationally less involving Gaussian network that is
subsequently used as the on-line controller.
3.6.1 Steady state optimization
The rst problem that has to be solved is to nd the optimal oxygen storage
coverage for given operating conditions. The optimal point can be dened as
the point where the conversion remains the highest if some perturbations of

in
occur. A measure of emissions is the outlet lambda predicted by the model
(
out
). Recall that this signal is zero if the outlet is at stoichiometry, which
3.6. Model-based predictive control 113
means only CO
2
and H
2
O can be found in the outlet and therefore there are
no harmful emissions. On the other hand, if the signal is negative, some CO
and/or HC must be present, while if the signal is positive some O
2
and/or
NO must be present in the exhaust. In the latter case only NO is of concern
since O
2
emission is of course not harmful. It was shown in the previous
chapter that NO emission during a rich-lean step usually precedes the oxygen
emission and thus when some oxygen is observed in the outlet there is a high
probability that NO is also present (see gures 2.19 and 2.20). The problem is
that the concentration of NO is much smaller than the concentration of oxygen
and sometimes lambda sensors do not even properly sense the amount of NO
in the outlet [34]. Modern lambda sensors are more likely to equilibrate NO
successfully, but still have diculties to equilibrate NO
2
[91].
The optimal steady state ROC is in general dierent for various exhaust
mass ows. Large variations between neighboring steady state points can, how-
ever, sometimes lead to higher exhaust emissions than with one sub-optimal,
but xed steady state relative oxygen coverage of ceria. These emissions are
associated with
in
moving away from the stoichiometry in order to ll/empty
the oxygen storage up to desired level. The optimal steady state points can
be found by solving an optimization problem. Let fl = [
1
, . . . ,
n
] represent
the vector of n distinct exhaust mass ows. The corresponding vector of opti-
mal steady state oxygen storage coverages,
ss
, can be obtained by solving the
following problem:
min
fl
Q
ss
() = C
T
C +W
ss

n1
i=1
(
i+1

i
)
2
where C
i
=

out
(
i
)

p
in

out
(
i
)

n
in

(3.60)
The steady state optimization problem determines the oxygen storage lev-
els at which some xed positive
p
in
and negative
n
in
inlet perturbations cause
minimal emissions, taking into account that the distance between the neigh-
boring steady state points should not be too large. The weight W
ss
is the
tuning factor. The optimal steady state points for the exhaust mass ows not
included in vector fl can be obtained by interpolation. Note that due to the
model dependence on
in
the optimal point will also depend on the selection
of positive and negative perturbations in (3.60).
3.6.2 Dynamic optimization
The goal of the controller is to drive the system as fast as possible to the desired
steady state with minimum possible emissions. This goal is very close to the
standard objective of optimal control where the system has to be brought as
soon as possible to equilibrium with minimal possible usage of energy.
114 Model-based controller
Optimizer Process
Observer
Reference
x
est
u y
Optimizer Process
Observer
Reference
x
est
u y
Figure 3.26: Basic structure of a MPC controller.
Model Predictive Control
With the process model available on-line and enough computing power the
optimal control sequence can also be calculated on-line. Such a nite horizon
optimization problem is solved at every sampling interval in a Model Predictive
Controller (MPC). Figure 3.26 depicts a basic scheme of an MPC [63]. At every
sampling interval the optimizer calculates the optimal control sequence for the
next N
CH
(control horizon) samples on the basis of state measurements and
observations, and predicted future states in the next N
PH
(prediction horizon)
samples. This is a nite horizon problem with N
CH
and N
PH
as tuning factors.
In a linear time-invariant case without constraints an innite horizon problem
would yield an LQ controller, or LQG controller if states are estimated with
a Kalman lter like in gure 3.26. However, the power of the MPC is its
possibility to deal with constraints and to apply a nonlinear process model in
the optimization. A typical problem to be solved can be thus expressed as:
min
u(k...k+N
CH
1)
N
PH

i=1
x(k +i|k)
T
R
i
x(k +i|k) +
N
CH

i=1
u(k +i 1)
T
Q
i
x(k +i 1)
(3.61)
subject to constraints on inputs and states:
u
min
u(k +j) u
max
, for j = 0, . . . , N
CH
1
x
min
x(k +j|k) x
max
, for j = 1, . . . , N
PH
where R
i
and Q
i
are weighting matrices that are typically considered as tuning
parameters. x(k + i|k) denotes the state prediction by the model at i + k-th
sampling interval on the basis of the current state x(k). u is the optimization
variable. Constraints can be imposed on both states and process inputs. Some-
times is the weight put on u instead of u in problem (3.61). This is a very
3.6. Model-based predictive control 115
ecient way to reduce sudden changes in the controller output and to enforce
an accurate steady state tracking.
MPC operates in a receding horizon fashion as only the rst sample of
the calculated control sequence, u(k), is applied to the process. The whole
procedure is repeated at the next sampling interval. Though stability of the
MPC is much more dicult to asses, its application is quite straightforward
and the approach is in principle the same for all types of systems, such as
multivariable, non-minimum phase, systems with delay, etc. Further references
can be found in [1, 27].
MPC control of oxygen storage
The following nite horizon optimization problem is solved by the oxygen stor-
age controller:
min

IN
(1...N
CH
)
Q
d
=
N
PH

i=1
_
(
ss
(i))
2
+W
d2

out
(i)
2

+
N
CH

i=1
W
d1

IN
(i)
2
subject to q
n

IN
(1 . . . N
CH
) q
p
(3.62)
Note that the solution to (3.62), which is a sequence of catalyst inlet
values, depends on the exhaust mass ow as both
ss
and process parameters
are functions of it. W
d1
and W
d2
are tuning factors whose larger values in most
cases imply slower reference tracking and less emissions during that transient.
Should a special attention be given to any exhaust component, W
d2
can be set
to be a function of
out
. For example, if a greater reduction of NO emission
is desired W
d2
can be increased for a larger positive
out
. Moreover, W
d2
is also set to be a function of exhaust mass ow applied, since the emission
increase due to higher ows is not directly included in the calculated
out
. The
constraints are set on the allowed rich and lean excursions of engine to avoid
any driveability problems.
In the controller calculation, (3.62), the tracking error is weighted against
the controller output as in any typical optimal controller. However, the term
that gives a special weight to the predicted emission is added. Moreover, a
standard version of the controller has the parameter W
d1
=0 and thus only the
weight on the emissions is set. In other words, the control is allowed to be
aggressive (fast tracking) as long as it does not lead to emissions. It was found
that in the case of a very small oxygen storage capacity and high reaction
rates (f() 1) it is better to also include the weight on the controller output
(W
d1
= 0) because the obtained control can be too aggressive. This will be
discussed in the next chapter. The controller applied in simulations in this
chapter has W
d1
=0.
The applied process model is like in (3.54) the control-oriented model of the
catalytic converter (3.39,3.40,3.41) and the engine control model represented
116 Model-based controller
by a transport delay, because of a larger inner-loop bandwidth. A case when
the more accurate inner-loop dynamics has to be included in the tuning proce-
dure will be shown in next chapter. The sampling time for optimization is set
to be lower than for the engine control to reduce the number of optimization
variables. Again, this is allowed due to slower dynamics of the catalytic con-
verter. If the sampling time of 0.1s is chosen and inner loop delay also amounts
to approximately 0.1s, the engine dynamics can simply be modeled by the unit
delay.
Solution method
The dynamic optimization problem in this study is solved by the Sequential
Quadratic Programming (SQP) algorithm in Matlab [30, 60]. SQP is an e-
cient algorithm to solve a general constrained nonlinear optimization problem:
min
x
n
f(x) (3.63)
subject to equality constraints, G
i
(x) = 0, i = 1, . . . , m
e
, and inequality con-
straints, G
i
(x) 0, i = m
e
+1, . . . , m. x
n
is the vector of design parameters
(optimization variable). Objective function, f(x), returns a scalar value.
Without entering the details on the algorithm, as those can be found in any
textbook on optimization, the general idea of SQP is to transform the problem
(3.63) into a local QP-based subproblem that is solved at every iteration. The
solution of the QP subproblem then gives the search direction for the global
solution.
3.6.3 Analytic MPC approximation
Since the optimization problem (3.62) is too complex to be solved at every
sampling interval in the on-board computer, another approach is needed to im-
plement the advantages of MPC in a computationally less demanding fashion.
If it is assumed that the rst control action
IN
(1)
x
calculated by (3.62) is
a unique and continuous function of state x, it may be approximated to an
arbirtary accuracy by some universal approximator such as a neural network
[73]. For a linear system this uniqueness and continuity can be proved [48].
The state in this case is the error signal e =
ss
, whereas the exhaust mass
ow m
ex
is the measured parameter. Let x = [e m
ex
]
T
X. The goal of the
approximation is to nd a function
in
(x) such that for a given accuracy :
sup
X
|
IN
(1)
x

in
(x)| < (3.64)
The approximator used in this study is a linear combination of weighted Gaus-
sian radial basis functions. This network has been selected because of a simple
3.6. Model-based predictive control 117
tuning algorithm (linear least squares) and because of its straightforward struc-
ture. The clear structure of the network with a direct relationship between local
behavior and one specic Gaussian function representing this behavior enables
corrections of network inaccuracies only within regions where those inaccuracies
occur. The network is given with:

in
(x) =

n
i=1
y
i

i
(x)

n
i=1

i
(x)
(3.65)

i
(x) = exp
_

|x c
i
|
2

i
_
Figure 3.27 shows a schematic view of a Gaussian radial basis function net-
work. It was shown in [79] that a Gaussian network can be used as an universal
approximator if n is large enough. The necessary number of parameters can
be derived on the basis of smoothness characteristics of functions to be ap-
proximated. This network is also identical to a class of fuzzy systems that
uses center-average defuzzication [56, 74]. One problem of such a network
is commonly known as the curse of dimensionality. Though the network is
rather simple to use, if the number of states increases it is not guaranteed that
the number of network parameters will not increase exponentially in order to
obtain the desired accuracy. Therefore, for larger problems it may be better
to investigate the application of dierent neural approximators [73]. It will be
shown that in this problem the number of network parameters can be kept at
a reasonable level.
Network estimation
A Gaussian network is used because of the tuning algorithm simplicity. If xed
nodes c
i
= [c
e
i
c
m
ex
i
]
T
are chosen together with xed variances
i
, the function
becomes linear in parameters y
i
. Hence, a batch least squares problem of tting
the function to the data can be solved analytically. For some state x
k
function
(3.65) can be written as:

in
(x
k
) = (
k
)
T
(3.66)

k
=
1

n
i=1

i
(x
k
)
_

1
(x
k
) . . .
n
(x
k
)

T
= [y
1
. . . y
n
]
T
The data points resulting from (3.62) are stored in = [
1
IN
(1) . . .
N
IN
(1)]
T
.
The estimated parameter vector

can be determined as follows:

=
_

_
1

T
(3.67)
=
_
(
1
)
T
. . . (
N
)
T

T
118 Model-based controller
e
m
ex
x
c
1,
,
1
c
2,
,
2
c
n-1,
,
n-1
c
n,
,
n

n-1

n
y
1
y
2
y
n-1
y
n
+
+
/

in
e
m
ex
x
c
1,
,
1
c
2,
,
2
c
n-1,
,
n-1
c
n,
,
n

n-1

n
y
1
y
1
y
2
y
2
y
n-1
y
n
y
n
++
++
//

in
Figure 3.27: A schematic view of a Gaussian radial basis function network.
The controller tuning procedure can now be summarized. After the converter
model has been obtained by some identication procedure the outputs of the
Model Predictive Controller (3.62) for dierent working points are calculated
o-line. The neuro-fuzzy controller (3.65) is then tuned according to (3.67).
Care has to be taken to choose the centers of nodes, c
i
, and variances
i
well. If
the t is not satisfactory the whole procedure can be repeated for an increased
number of network parameters, n. Note that if all basis functions
i
are not
suciently excited, the matrix
T
can become singular. Therefore the grid
of initial states x
k
has to be suciently dense. On the other hand, the number
of training points should be kept as low as possible to reduce the optimization
time and make this method applicable for the on-board computer. Together
with the automatic identication (section 3.4.2) the controller can be updated
periodically to compensate for changes of the converter dynamics due to the
catalyst aging process.
Stability
Stability of a nonlinear model based MPC is in general very dicult to prove.
However, when MPC is approximated with some analytic function, like in this
case, the stability analysis can be performed by analyzing the nal controller
only. Such an analysis allows the judgement whether the nal control system
is stable, but will not lead to necessary conditions of stability, i.e. MPC tuning
parameters which would assure that a stable controller is obtained.
3.6. Model-based predictive control 119
G
p
(s)

y
2
u
1
y
1
e
1
u
2
e
2
+
-
+
+
G
p
(s)

y
2
u
1
y
1
e
1
u
2
e
2
+
-
+
+
Figure 3.28: System with a linear direct path and a nonlinear feedback con-
troller for which the circle criterion has been derived, [97].
Since the controller is a nonlinear, static, function of the controlled vari-
able ROC (exhaust mass ow will be assumed constant as this is an external
parameter), a well known circle criterion can be used [97]. A system for which
the circle criterion is derived is shown in gure 3.28. It has linear dynamics in
the direct path and a nonlinear static element in the feedback path. Since not
only the controller but also the dynamics of the catalytic converter is nonlinear,
the mentioned criterion can only be used in a simplied (linearized case), and
thus global stability is not proved but rather estimated. Because exhaust mass
ow is assumed to be constant, the only remaining source of nonlinearities are
functions f() and g(). Since g() is active only with positive inputs when
is close to 0, and otherwise equals zero, it will be neglected in the analysis. It
is also assumed that f() = 1, what is the worst case scenario since then the
process gain is the largest. Thus, the process transfer function becomes, like in
(3.54), G
p
(s) =
K
P
s
e
T
d
s
, with K
P
= 1/k
d
.
The circle criterion [97] states that the closed loop system is L
2
stable if
the following condition is satised:
inf

Re(G
p
(i)) >
1
b
(3.68)
where b represents the maximum allowed gain of the controller such that
(e
2
) (0, be
2
) (see gure 3.28). In the case of the catalytic converter con-
troller the stability is guaranteed if
in
(x) (0, bx). There are more cases of
the circle criterion (if the plant would have pole(s) with a positive real part),
but the above mentioned applies in the studied case. Also note that the given
condition is only a sucient stability condition.
The allowed gains are calculated for various mass ows. Due to a simple
process characteristic, the allowed controller gain for the catalytic converter
model can simply be calculated. By using (3.41,3.51,3.54) and the model pa-
rameter values applied throughout this chapter the following is obtained:
120 Model-based controller
b =
10.26
m
ex
T
d
(3.69)
where the exhaust mass ow is expressed in kg/h, and the time delay in seconds.
Hence, the allowable controller gain is reduced with an increase of both exhaust
mass ow and process time delay. For example, with a delay of 0.1s and exhaust
mass ow of 54kg/h the maximal gain permitted is 1.9, which is very unlikely
to cause any stability problem. However, with a time delay of 0.2s and exhaust
mass ow of 108kg/h, the allowable controller gain drops to 0.32 and stability
becomes an issue in that case.
Stability of the controller obtained after tuning of the network can simply
be assessed by checking whether the network with inputs equal to the basis
functions centers, c
i
, satises the criterion. The network outputs are in those
cases equal to the node weights, y
i
, since the basis functions are set up in such
a way that only one basis function is active with x = c
i
. Stability does not
pose a serious problem with low exhaust mass ows and small time delays, as
can be expected. On the other hand, with a lower oxygen storage capacity of
the converter stability problems are more likely to occur. Care should be taken
that the controller gain does not change sign with respect to the MPC output.
Due to network approximation errors in the low gain region a small error can
lead to a positive feedback and subsequently to an unstable system. Whether
this happens can also be checked by looking at the node weights. If instabillity
in the network is observed, then a specic node weight can be corrected. This
will only aect the local performance.
3.6.4 Simulation results
The MPC outputs were calculated at 348 points. N
CH
and N
PH
were selected
as 49 and 60 respectively. The small dierence between the two horizons is
allowed in this case because the process has integrating behavior and reaches
steady state already after N
CH
samples, when the input becomes zero. The
dierence between the two is still applied basically to introduce a termination
penalty. This penalty is set to enforce a small steady state tracking error.
Note, however, that the real steady state error during the control is not directly
connected with the steady state error in a single optimization when the process
is far from steady state. This follows from the receding horizon principle of
operation.
The amount of optimization variables was 13, as the variables were not
allowed to change at every sampling interval. This is a standard measure in
MPC to reduce the needed computational time. The sampling time was set
to 0.1s. The Gaussian controller in this study consists of 204 radial bases
functions, and thus so many parameters have to be estimated from the MPC
outputs. The number of parameters is kept as low as possible to shorten the
3.6. Model-based predictive control 121
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4
0.1
0.05
0
0.05
0.1
flow =40 kg/h
c
o
n
t
.

o
u
t
.

[

]
MPC
network
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.2 0.3 0.4
0.1
0.05
0
0.05
0.1
flow=100 kg/h
error []
c
o
n
t
.

o
u
t
.

[

]
Figure 3.29: Fit between MPC and Gaussian network controller outputs for
exhaust mass ows of 40kg/h and 100kg/h
tuning procedure and limit the size of the controller. Figure 3.29 shows the t
between the MPC and Gaussian controller outputs for exhaust mass ows of
40kg/h and 100kg/h.
Performance of the controller is tested by two sets of randomly chosen
dynamic tests. The rst principle model of the catalytic converter was used to
simulate the process. In the rst test the oxygen storage is initially empty (i.e.
after fuel enrichment), while in the second test the oxygen storage is initially
full (i.e. after fuel cut-o). The performance of the novel MPC-based network
controller is compared with a conventional = 1 engine controller that does
not take into account the dynamics of the catalytic converter, and the gain
scheduling controller presented in section 3.5. The latter assumes the desired
steady state for ROC of 0.5. The MPC-based controller has calculated the
optimal steady state to be around 0.39, by solving the steady state problem
(3.60). It slightly depends on the exhaust mass ow, but changes are less than
1%.
Figure 3.30 shows the throttle position, desired and actual catalyst inlet
(engine outlet) values during the test with initially empty oxygen storage
(test 1). Catalyst outlet concentrations of CO, NO and hydrocarbons, as well
as averaged ceria coverage throughout the converter are depicted in gure 3.31.
The same plots for initially full oxygen storage (test 2) are given in gure 3.32
and gure 3.33. The tests are highly dynamical because during such engine
122 Model-based controller
0 2 4 6
15
20
25
30
time [s]
t
h
r
o
t
t
l
e

a
n
g
l
e

[
d
e
g
]
0 2 4 6
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
0 2 4 6
0.92
0.94
0.96
0.98
1
1.02
1.04
1.06
time [s]
l
a
m
b
d
a

[

]
ref.
eng. lam.
0 2 4 6
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
n. network
gain sch.
=1
Figure 3.30: Throttle position, reference and actual engine values for initially
empty oxygen storage.
operation most of the emissions occur.
There is a clear dierence between the reference and actual engine values,
which is caused by the inaccurate engine model in the IMC controller. Since the
disturbances on the process are large, these inaccuracies play a more important
role. The engine controller used in simulations has a perfect estimation of
the air dynamics, but errors in wall wetting estimation amount up to 30%,
depending on the operating point. Due to a relatively low oxygen storage
capacity this engine control loop errors have also a larger impact on the catalyst
controller performance making it less optimal.
In test 1 the MPC-based catalyst controller tries to ll the oxygen storage
up to the desired level as soon as possible yet not leaving too much of NO
unconverted during the lean excursion. Since the = 1 controller does not
try to ll the oxygen storage large emissions of CO and hydrocarbons occur.
NO emissions are kept low, because the empty oxygen storage promotes the
NO conversion. The NO emissions produced by the gain scheduling controller
are much larger than emissions produced by the MPC-based controller. The
reason for this is that the latter takes into account the emissions predicted
by the model, and adapts the steady state reference point as well as reduces
the controller gain when the danger of emissions exists. This also shows that
the widely accepted idea that 50% of oxygen storage coverage is the most
advantageous may not always be correct.
3.6. Model-based predictive control 123
0 2 4 6
0
0.5
1
1.5
2
2.5
time [s]
C
O
o
u
t

[
v
o
l
%
]
0 2 4 6
0
0.01
0.02
0.03
0.04
time [s]
N
O
o
u
t

[
v
o
l
%
]
0 2 4 6
0
0.01
0.02
0.03
0.04
time [s]
H
C
o
u
t

[
v
o
l
%
]
0 2 4 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
time [s]
R
O
C

[

]
Figure 3.31: Converter outlet emissions and average degree of ceria lling with
initially empty oxygen storage. Thick line - Gaussian controller, thin line - gain
scheduling controller, dashed line - = 1 controller.
0 2 4 6
15
20
25
30
time [s]
t
h
r
o
t
t
l
e

a
n
g
g
l

[
d
e
g
]
0 2 4 6
0.9
0.95
1
1.05
time [s]
l
a
m
b
d
a

[

]
ref.
eng. lam.
0 2 4 6
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
0 2 4 6
0.9
0.95
1
1.05
time [s]
l
a
m
b
d
a

[

]
n. network
gain sch.
=1
Figure 3.32: Throttle position, reference and actual engine values for initially
full oxygen storage.
124 Model-based controller
0 2 4 6
0
0.002
0.004
0.006
0.008
0.01
0.012
time [s]
C
O
o
u
t

[
v
o
l
%
]
0 2 4 6
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
time [s]
N
O
o
u
t

[
v
o
l
%
]
0 2 4 6
0
0.2
0.4
0.6
0.8
1
x 10
4
time [s]
H
C
o
u
t

[
v
o
l
%
]
0 2 4 6
0.2
0.4
0.6
0.8
1
time [s]
R
O
C

[

]
Figure 3.33: Converter outlet emissions and average degree of ceria lling with
initially full oxygen storage. Thick line - Gaussian controller, thin line - gain
scheduling controller, dashed line - = 1 controller.
In test 2 the novel catalyst controller tries to empty the oxygen storage
to the desired level as soon as possible not producing too much of CO and
hydrocarbon emissions during the transient. The = 1 controller produces
much more NO emissions and somewhat less CO and, interestingly, more hy-
drocarbon emissions. The former is obvious because an almost completely lled
oxygen storage promotes CO conversion, while the latter is caused by a partial
inhibition of the ethylene conversion by oxygen. This eect, however, is negli-
gible in the total gure since the emission level is very low. The emissions of
CO and HC produced by the MPC-based controller are somewhat larger then
the emissions produced by the gain scheduling controller, since the steady state
of the former is at a lower ROC. The relative level of emissions, however, is not
very high, when compared with test 1.
The observed errors in reference tracking of relative oxygen level on ceria
stem from model inaccuracies, the dierence between actual and measured
converter inlet signal and disturbances. The model inaccuracies also stem
from the fact that the oxygen storage capacity dependence on the exhaust mass
ow was not taken into account during the simulations. This can for example
be observed in test 1 where the initial ROC is slightly above 0, as at higher
exhaust mass ows there is more unused oxygen from ceria with a rich input.
A fair comparison of the controllers is the summing of emissions during
3.7. Conclusions 125
the two tests. The average CO emission is reduced by 75% when using the
MPC-based controller in comparisson with the = 1 controller, while the
hydrocarbon emission is reduced by 80%. The NO emission is reduced by 89%.
Similar results have also been obtained by other tests and with dierent models
[5, 8]. The NO emission reduction of the MPC based controller compared
with the gain scheduling controller is 40% (the emission reduction of the gain
scheduling controller with respect to the = 1 controller is 83%). The emission
of HC and CO is almost the same with both controllers. The MPC-based
controller manages to better use the process nonlinearities in order to yield a
better performance of the system as expected.
3.7 Conclusions
Development of a model-based controller for a three way catalytic converter has
been presented in this chapter. The engine dynamics of interest for control and
the model-based air/fuel ratio controller have been studied. The entire control
system was obtained after the introduction of the control-oriented model for
the catalytic converter and inferential oxygen storage controller.
First, the feasibility of the model-based oxygen storage controller was in-
vestigated. The rigorous, rst principle model based on the elementary step
kinetics was used as the basis for the development of a simplied, control-
oriented model. The latter can be identied with a simple lambda-step test.
Relations that can broaden the operating range of the model with respect to
various exhaust mass ows and inlet lambda signal amplitudes have been found
on the basis of the underlying kinetics. It was also shown that the model can
accurately predict the experimentally measured data. Special attention has to
be given to the lambda sensor characteristics as the downstream sensor exhibits
both static and dynamic inaccuracies. This matter will further be addressed
in the next chapter.
A cascade control system, with an engine IMC controller in the inner loop,
is used for control of the relative level of oxygen stored on ceria. Since it is
not measurable, the control variable (ROC) is estimated by the model. The
insights in the converter dynamics obtained via rst principle modeling prove
to be very important for development of the control-oriented model whose
accuracy is crucial for the performance of the control system.
For computational reasons an analytic approximation of the Model Predic-
tive Controller is proposed as the controller. The controller is a network of
Gaussian radial basis functions, which is trained with o-line computed Model
Predictive Controller outputs. The Model Predictive Controller rst deter-
mines the optimal steady state points and then subsequently optimal trajecto-
ries that guide the system to the desired steady state with the lowest possible
corresponding exhaust emissions. Such a controller can lead to a substantial
emission reduction during dynamic engine operation. The largest improvement
126 Model-based controller
that this controller oers compared to other, more classic control strategies is
the full use of the available knowledge on the nonlinear process behavior to ob-
tain an optimal control throughout the complete region of operation. A draw-
back of the controller is the large computational power needed to calculate the
MPC outputs. Therefore, practical aspects of the controller implementation,
such as nding the fastest optimization procedure to shorten the controller
training time, have to be studied further.
4
Experimental testing of the control
system
4.1 Introduction
4.2 Experimental setup
4.3 Open loop tests: model
evaluation
4.4 Closed loop tests:
model-based control
4.5 Conclusions
4.1 Introduction
This chapter describes experiments conducted for the validation of the cat-
alytic converter controller introduced in the previous chapter. The tests were
performed on an engine dynamometer test bench at TNO Automotive, Delft.
The applied engine was a Volvo 5 cylinder, 2.0 liter gasoline engine with the
same underoor Pt/Rh/-Al
2
O
3
/CeO
2
catalyst as applied in the light-o and
sweep tests presented in chapter 2.
The rst goal was to asses the quality of the control-oriented model by
numerous dynamic tests in various operating conditions. The emphasis was
on behavior of the oxygen storage capacity at dierent temperatures, exhaust
mass ows and inlet signals. It will be shown that the actual dynamic
behavior of the studied catalyst is somewhat more complex than anticipated
during the simulation studies. Namely, oxygen storage capacity seems not
only to be inuenced by the temperature and gas mass ow, but also by the
input signal. Initial experimental verication of the control-oriented model
was already performed in section 3.4.3 on a dierent catalyst. Some of the
observed phenomena, that will be presented in this chapter, were not observed
in these initial measurements. Some additional features will therefore be added
to the control-oriented model to account for these observed eects. Necessary
changes to the rst principle model to account for those eects will also be
analyzed.
The main goal of the experiments was to assess the performance of the
complete control system under various operating conditions and to obtain an
estimate of possible benets of the novel control approach [6]. The applied
128 Experimental testing of the control system
controller is a cascaded system with the engine air/fuel controller in the inner
loop and catalyst inferential oxygen sensor controller in the outer loop. The
latter is a MPC based Gaussian network controller. Since the engine controller
is also a model-based controller, as shown in the previous chapter, to tune
the engine controller both static (air path) and dynamic (fuel path) open-loop
tests have to be performed. As the wall wetting parameters for the given engine
were known from a previous study [18], only static tests were performed for
assessment of the air path model. The catalyst controller is somewhat more
complex than the one presented in the previous chapter due to a very low
oxygen storage capacity. The impact thereof will be discussed as the engine
dynamics cannot be neglected during the tuning of the controller.
The controller was tested during three dierent highly dynamic test cycles
and its performance is compared to the performance of a simple stoichiometric
engine air/fuel controller that does not take catalyst dynamics into account.
The chosen dynamic tests can for example represent city driving as most of the
emissions occur during transients, as shown in the previous text. A discussion
of the impact of the storage capacity on the control performance and the true
benets of the control is given at the end of the chapter.
4.2 Experimental setup
Experiments were performed on an engine dynamometer test setup, with max-
imal power of 220kW and maximal torque 450Nm. The applied engine was a
2.0 liter, 5 cylinder gasoline Volvo engine. Figure 4.1 shows the outline of the
test setup.
The controller was programmed in a MatrixX rapid prototyping system
that was connected to the AC100 unit which served as the actual controller.
The communication with the ignition module was accomplished via CAN that
sends the desired ignition point and fuel injection time (duration of opening of
a certain fuel injector) and receives the measured engine speed signal.
Measured signals included the throttle position, intake manifold pressure,
the coolant temperature, exhaust gas temperatures in front of and behind the
converter, lambda sensor signals in front of and behind the converter and in
some tests exhaust emissions measured by analyzers (CO, CO
2
, NOx, HC and
O
2
).
The lambda sensor in front of the converter was placed immediately down-
stream of the exhaust manifold, to reduce the delay time in the air/fuel control
loop. During the open loop tests for assessment of the converter dynamics, a
lambda sensor was also placed directly in front of the converter. The applied
sensors were heated NTK UEGO sensors, coupled with the MO-1000 system
for control and data acquisition [72]. The sensor has a measurement range
0.7 2.2 with a response time (0-90%) of 0.1s.
4.3. Open loop tests: model evaluation 129
Ignition module
User interface:
MatrixX
AC100
5 cylinder Volvo engine
+
Catalytic converter
CAN
Ignition point
Fuel injection time
Engine speed
Cylinder ID
Sensor signals
Ethernet
Fuel
Ignition
C
r
a
n
k
/
c
a
m
s
i
g
n
a
l
Ignition module
User interface:
MatrixX
User interface:
MatrixX
AC100 AC100
5 cylinder Volvo engine
+
Catalytic converter
5 cylinder Volvo engine
+
Catalytic converter
CAN
Ignition point
Fuel injection time
Engine speed
Cylinder ID
CAN
Ignition point
Fuel injection time
Engine speed
Cylinder ID
Sensor signals
Ethernet
Fuel
Ignition
C
r
a
n
k
/
c
a
m
s
i
g
n
a
l
Figure 4.1: Outline of the experimental setup.
The HC and NOx analyzers have 90% rise time of 1.1 and 1.4s respectively.
Since the response of the converter is quite fast, as will be seen in the next sec-
tion, these analyzers are too slow to be used for exact emission measurements,
but can be used as a good emission indication when comparing dierent control
strategies. The response of the other analyzers was even slower, so they will
not be considered in the analysis of the results. Simultaneous measurements
in front of and behind the catalytic converter were not possible.
4.3 Open loop tests: model evaluation
The control-oriented model was introduced in section 3.4.1. The model struc-
ture and estimation introduced in the previous chapter will not be changed.
A larger emphasis will be given to the model nonlinearities in the complete
operating range and some eects concerning oxygen storage capacity that were
not observed in the simulations and experiments shown in chapter 3.
The model, accounting for the parameter dependencies on operating condi-
tions, can be rewritten as:
d
dt
=
1
k
d
(T, m
ex
)
(
in
f
d
(, T, m
ex
,
in
) +g
d
(, T, m
ex
))

out
=
in
k
d
(T, m
ex
)
d
dt
(4.1)
130 Experimental testing of the control system
0 2 4 6 8 10
0.98
0.99
1
1.01
1.02
1.03
time [s]
l
a
m
b
d
a

[

]
input
output
scaled
0 2 4 6 8 10
0.96
0.97
0.98
0.99
1
1.01
1.02
1.03
time [s]
l
a
m
b
d
a

[

]
0 2 4 6 8 10
0.98
0.99
1
1.01
1.02
1.03
time [s]
l
a
m
b
d
a

[

]
0 2 4 6 8 10
0.97
0.98
0.99
1
1.01
1.02
1.03
time [s]
l
a
m
b
d
a

[

]
Figure 4.2: Measured and scaled dynamic responses at an inlet temperature of
750K and exhaust mass ow of 63 kg/h (left) and 390 K and 28 kg/h (right).
The subscript d denotes the dependency of a given parameter (function) on
direction of the inlet signal (lean or rich).
The parameter estimation was presented in section 3.4.2. Again, as during
the experimental model verication in section 3.4.3 the lambda sensor signal
behind the converter had to be preconditioned before applying the estimation.
The same static scaling of the downstream sensor response was used. A typical
step response of the catalytic converter, with and without the preconditioning,
is shown in gure 4.2. With a lean inlet only a static correction of the down-
stream signal is sucient to match the inlet and the outlet signal in steady
state. After a lean-rich step the static correction is performed until the peak
in the downstream signal is detected. The outlet signal is assumed to equal
the inlet signal after the peak. As already discussed, this sensor behavior is
probably caused by the sensor sensitivity to hydrogen in the exhaust. The peak
after a lean-rich step is possibly formed because of water-gas shift (3.52) and
steam-reforming (3.53) reactions which have hydrogen as a product.
The correction factor is nearly constant at higher temperatures. At lower
temperatures, however, the correction after a lean-rich step is dierent because
the above mentioned reactions are activated only above approximately 670K
[98]. Therefore, at lower temperatures there is no overshoot in the response,
but still a steady state error is present, as can be seen in gure 4.2.
4.3. Open loop tests: model evaluation 131
4.3.1 Model application range
To obtain an accurate model, dependence of k
d
and f
d
on temperature, ex-
haust mass ow and relative oxygen coverage of ceria has to be known (4.1).
The theoretical dependence of f
L
(), f
R
() and k
d
on exhaust mass ow was
shown in (3.51). These functions are valid at one temperature, and in most
conditions. The nominal function is temperature dependent and should there-
fore be estimated at several temperatures. It was found in this study that the
nominal function undergoes the largest changes at lower temperatures, after
the light-o, and then remains almost constant above some higher, saturation
temperature. In case of the catalytic converter presented here the full conver-
sion starts around 570K, while the saturation temperature is around 690K.
The model for temperatures for which the estimation was not performed is
obtained by interpolation. Care has to be taken such that the obtained model
inaccuracies remain suciently small.
The adaptation (3.51) for functions f
L
() and f
R
() was found to be valid at
lower temperatures (apart from the fact that f
R
appears not to be dependent on
the inlet lambda), but only the adaptation for the function f
L
() was found to
be necessary at temperatures above the saturation temperature. The function
f
R
was constant for various exhaust mass ows. The values of f
R
() were very
close to 1 (100% conversion) for as low as 0.1, which means that the conversion
has reached the upper saturation due to a high converter temperature. This
very high conversion is reached already at high mass ows and therefore does
not change much at lower mass ows.
Note that this saturation is not the same as the eect observed in section
3.4.2. The saturation was imposed on the function f
L
() in that case. It was
caused by the stored CO on the ceria surface and it was a lower saturation (the
function was not decreased as much as expected at high exhaust mass ows).
The problem of estimating functions f
L
() and f
R
(), with a limited number of
operating points where measurements were obtained, now arises because it is
very important to perform the identication with an active exhaust mass ow,
i.e. where the saturation is not present. Only then the extrapolating relations
(3.51) can be applied. If the estimation is performed inside of the saturation
region for any of the functions, the extrapolation is not allowed as it will not
lead to an accurate model. After the performed identication the obtained
functions can be checked, and if saturation is suspected the identication has
to be performed in another point at lower (lower saturation) or higher (upper
saturation) exhaust mass ow. This idea is depicted in gure 4.3 in the case of
function f
L
().
Oxygen storage capacity
Another parameter that has to be estimated correctly is the oxygen storage
capacity (OSC) and hence k
dn
. As already shown, OSC increases with tem-
132 Experimental testing of the control system
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1

f
L
active
region
Figure 4.3: Eect of saturation in the case of function f
L
(). Dashed line
represents a function in the active region, while the saturation is represented
with hard constraints.
350
400
450
500
550
600
20
40
60
80
100
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
temperature [K]
exhaust flow [kg/h]
r
e
la
t
iv
e

O
S
C

[

]
Figure 4.4: Estimated surface oxygen storage capacity as a linear function of
ow and temperature.
perature but it is also a function of the exhaust mass ow. It was observed
that by increasing the exhaust mass ow the observed available oxygen from
ceria decreases. This eect was observed in both simulations and experiments
presented in the previous chapter. Figure 4.4 shows the estimated k
dn
as a
function of the exhaust mass ow and temperature. A nearly linear depen-
dence on both variables was applied. The estimation of k
dn
was performed
during lean to rich steps.
4.3. Open loop tests: model evaluation 133
0 1 2 3 4 5 6
0.97
0.98
0.99
1
1.01
1.02
1.03
la
m
b
d
a

[

]
0 1 2 3 4 5 6
0.98
0.99
1
1.01
1.02
1.03
time [s]
la
m
b
d
a

[

]
Figure 4.5: Rich to lean step responses after a short (above) and long (below)
rich period. Inlet - thin line, outlet - thick line.
It was observed that the capacity based on rich to lean transitions was al-
ways larger then the observed capacity after lean to rich transitions. Moreover,
it was depending on the duration of the previous rich period. The longer the
rich period, the larger the observed oxygen storage capacity was. On the other
hand, the capacity did not seem to be dependent on the length of the lean
input before the lean to rich step.
This eect could be due to existence of various layers of ceria, namely a
layer closer to the surface and layers deeper in the bulk. Such an assumption is
already included in the rst principle model by incorporating bulk and surface
ceria, but dierent oxygen storage capacities were not observed. Figure 4.5
shows rich to lean steps under same operating conditions after a short and a
long period in the rich region. It seems that the transfer from the ceria surface
to the bulk is rather fast, while vice versa is not. Therefore, if the bulk becomes
completely depleted after a long rich step, the observed OSC during lean steps
increases. The surface layer that is involved during a rich step is apparently
stable and does not depend on the length of the previous lean period. Oxygen
from the ceria bulk has to be used after the surface oxygen has been depleted
but this process proceeds at such a slow rate that it cannot be observed with
lambda sensors, and is of no importance for control.
The above hypothesis should yet be checked with more detailed tests. In
the following section modeling results are presented with the updated rst
principle model that can account for this eect. This idea has been used to
add the secondary (bulk) storage to the control-oriented model, which has led
to satisfactory modeling results for control purposes. The updated model is
obtained by combining the following expressions for the scaling factor with a
134 Experimental testing of the control system
lean input (k
L
) with (4.1):
k
L
= k
R
+k

(4.2)
where k
R
is the scaling factor during rich transients that is proportional to the
surface oxygen storage capacity (oxygen stored in the bulk ceria is not used
fast enough during rich steps), while k

is proportional to the amount of empty


ceria sites in bulk freed before the current lean transition. The relative change
of the bulk ceria covered by oxygen () during a lean step (lling) is calculated
by the following relation:
d
dt
=
d
dt
(4.3)
This relation is applied because during a lean step it appears that the additional
available oxygen storage does not aect the storage rate, f
L
(). It is assumed,
therefore, that during a transient with lean input the inlet oxygen equally uses
both surface and bulk sites, as the transition from the surface to the bulk is
fast. Hence the lling rates of both surface and bulk ceria are the same.
The emptying of the bulk ceria starts after the surface storage has been
depleted and is modeled as a rst order system with a time constant T
b
and
an adaptive gain K
b
:
T
b
d
dt
+ = K
b
() (4.4)
It was observed that emptying of the bulk ceria is faster in the beginning
and proceeds at a much slower rate as layers closer to the surface have been
depleted. Therefore the adaptive gain was used.
Inherent characteristic of the model is the prediction of a perfect match of
inlet and outlet lambda when the oxygen storage is either completely lled or
empty. This implies that the system response should be static if stoichiometry
is not crossed. While this is nearly true under lean conditions it was presented
in [77] that under rich conditions not only the oxygen storage phenomenon de-
termines the dynamic behavior. The additional dynamics stems most probably
from water gas shift and steam reforming reactions, as already discussed. In
gure 3.17 is shown that the hydrocarbon response slowly raises even after the
oxygen storage has been depleted. This can also be related to a decrease of
the peak measured by the downstream sensor (decrease of the downstream
hydrogen concentration). But also after the steady state has been reached and
a new transient in the rich region is applied, the response is still not immediate
[77]. This means that the reactions with steam have to be linked to some other
type of storage-release mechanism, which was not studied here. Though these
reactions should not aect the lambda value, they clearly inuence the sensor
response.
4.3. Open loop tests: model evaluation 135
A rst order lter with a zero was used in [77] to describe this behavior. It
was found here, however, that only a rst order lter is adequate to describe
rich to lean transitions when surface oxygen storage is nearly or completely
empty:
T
r
d

out
dt
+

out
=
out
(4.5)
where
out
is calculated by (4.1) and

out
represents the actual outlet lambda
signal. The lter time constant, T
r
is zero if the oxygen storage is lled above a
certain level, but also has to be reduced as bulk ceria storage becomes depleted
(see gure 4.5). It can be concluded that the behavior of the catalytic converter
is much more complex under rich conditions. In order to obtain the accurate
model more detailed studies should be conducted. The desired operation of
the catalytic converter is, however, where the oxygen storage is approximately
half-lled and that is where the model must be most accurate. Therefore, such
a rather simplied model under rich operating conditions is allowed.
Simulation analysis of oxygen transfer in ceria
The assumption that dierent rates of oxygen transfer between the ceria sur-
face and the bulk can cause variations in the observed oxygen storage capacity,
depending on the inlet direction, was tested on the rst principle model. The
rst principle model is the same as presented in chapter 2. Only a few parame-
ters have been adapted to account for the investigated eect. No attempt was
made to t the rst principle model to the measured data, but rather to see
whether the assumption that was made can explain the observed eect.
The rates of oxygen transfer from the ceria surface (store) to the bulk and
back (release) are calculated in the following way (see table 2.3):
r
store
= k
35f
L
CT

O

r
release
= k
35b
L
CT

O
(4.6)
where L
CT
denotes the total ceria capacity including the surface and the bulk.
Rate coecient k
35b
was decreased with a factor 100 to simulate the hy-
pothesis that the availability of oxygen stored in bulk for the reaction with
CO and HC at the interface between the ceria surface and noble metal is low.
The storage coecient k
35f
was not changed as the transfer of oxygen from the
surface to bulk is assumed high. The simulation results at standard conditions
(inlet temperature 650K, exhaust mass ow 54kg/h) are shown in gure 4.6.
The ceria surface capacity was increased two times to simplify the readability
of the results. In order to obtain the desired eects the adsorption of CO on
ceria (k
28f
in table 2.3) was decreased with a factor 100. It was necessary to
136 Experimental testing of the control system
0 2 4 6 8 10
0
0.5
1
1.5
C
O
o
u
t

[
v
o
l
%
]
0 2 4 6 8 10
0
0.2
0.4
0.6
0.8
O
2
o
u
t

[
v
o
l
%
]
short
long
0 2 4 6 8 10
0.95
1
1.05
l
a
m
b
d
a

[

]
outlet
inlet
0 2 4 6 8 10
0.95
1
1.05
l
a
m
b
d
a

[

]
short
long
inlet
0 10 20 30 40
0
0.5
1
time [s]
c
e
r
i
a

c
o
v
.

[

]
0 2 4 6 8 10
0
0.5
1
time [s]
c
e
r
i
a

c
o
v
.

[

]
surface
bulk
Figure 4.6: Eect of a slow bulk transition. Left - lean-rich step (note the time
scale on the gure below). Right - rich-lean step. The duration of the rich
input preceeding the lean step is 10s (short) and 40s (long).
establish the equilibrium of CO stored on the surface of ceria under rich con-
ditions at a lower coverage, to allow oxygen to reach the surface from the bulk
even after the near steady state has been reached. The simulated rich to lean
steps were in cases of 10s and 40s of the rich feed preceding the rich-lean step.
In the latter case the oxygen storage capacity, that was estimated on the bases
of the observed lambda signals, was 17% larger than in the rst case. Since the
lling of the oxygen storage, both surface and bulk, is fast, the oxygen storage
capacity observed after a lean to rich transition appears to be constant and
lower than after a rich-lean transition. Note that the lambda signal has not
reached steady state at 10s after the lean to rich step. This shows that there is
still a low rate oxidation of CO and HC occuring, but such an eect cannot be
observed experimentally due to the sensor noise and known inaccuracies in the
rich region. The plots of surface and bulk ceria coverage give further support
to the idea.
The performed simulations thus support thus the hypothesis about the ori-
gin of the oxygen storage capacity variations, but more dedicated kinetic tests,
also at higher temperatures, should be conducted to completely understand the
process. Moreover, it should be understood whether this is a feature present
only in some catalyst formulations, or whether it is a general eect which is
more or less present in every catalytic converter. The kinetic experiments for
4.3. Open loop tests: model evaluation 137
0 10 20 30 40 50 60 70
0.95
1
1.05
l
a
m
b
d
a

[

]
0 10 20 30 40 50 60 70
0.94
0.96
0.98
1
1.02
1.04
time [s]
l
a
m
b
d
a

[

]
Figure 4.7: Model prediction of converter step responses: above - ow 28 kg/h,
temperature 600K; below ow 63 kg/h, temperature 750K. Inlet - dotted line,
outlet - dashed line, prediction - solid line.
model estimation were performed at lower temperatures where the eects of
the bulk ceria are smaller.
4.3.2 Model testing
Model predictions of the converter step responses are presented in gures 4.7
and 4.8. Figure 4.7 shows the model prediction at 1200rpm, 40kPa and 80kPa
intake manifold pressure, while gure 4.8 shows the model prediction at 1200rpm
and 80kPa intake manifold pressure but at a lower reactor temperature. The
downstream signal shown in the gures is a raw signal measured by the
sensor. Note that the signal preconditioning was only necessary during the es-
timations, as only the inlet signal is needed to calculate the model prediction.
The reactor temperature is calculated by taking the average value between the
measured inlet and outlet temperature.
The model has been estimated under dierent operating conditions than
those presented in the above mentioned gures, so this is in fact a model
validation. The predictions of responses used for the parameter estimation (not
shown here) are very accurate as may be expected. The model (functions f
L
()
and f
R
()) was estimated at 630 and 710K and interpolated between those
temperatures. Above 710K the functions f
L
() and f
R
() were not dependent
on temperature. The estimation of oxygen storage capacity required, however,
138 Experimental testing of the control system
0 10 20 30 40 50 60 70
0.95
1
1.05
l
a
m
b
d
a

[

]
0 10 20 30 40 50 60 70
620
640
660
680
700
720
time [s]
t
e
m
p
e
r
a
t
u
r
e

[
K
]
Figure 4.8: Model prediction of the converter step responses (above, line de-
scription as in gure 4.7) and converter temperature (below). Exhaust ow -
63 kg/h.
more estimation points as the general dependency on temperature and exhaust
mass ow was not known. Since it was found that a linear dependency suces
(gure 4.4), less points for estimation can be used.
The model predicts the recorded process behavior fairly well under dierent
operating conditions, after both shorter and longer rich steps, and therefore is
suitable to be used for control in a wide operating range. Most of the prediction
errors occur during rich to lean steps, as can be seen in the rst positive step
response in gure 4.8, due to a rather simplied model of the bulk ceria inu-
ence. Figure 4.8 presents also the inuence of the reactor temperature on the
rich overshoot measured by the downstream sensor, as it becomes remarkable
at temperatures above approx. 670K. Note that the rst lean to rich step does
not exhibit overshoot, while the second does. That is in line with the activa-
tion of water gas shift and steam reforming reaction [98], which were already
discussed earlier in this section (gure 4.2).
4.4. Closed loop tests: model-based control 139
4.4 Closed loop tests: model-based control
4.4.1 Controller tuning
The cascade control scheme introduced in the previous chapter is also applied
here (gure 3.20). The outer loop is the catalytic converter controller, while
the model-based engine controller is in the inner loop. The Internal Model
Controller (IMC) controls the engine air/fuel ratio.
The air/fuel ratio controller consists of the wall wetting compensation and
intake air prediction. The wall wetting model is a rst order model (3.15),
whose parameters were obtained in a previous study [18]. The cylinder air
quantity is predicted on the basis of the throttle signal. If a perfect (fast)
measurement of the intake manifold pressure would be available, the speed-
density equation (3.7) could directly be applied to calculate the intake air
ow. Apart from speed, which is measured, the volumetric eciency, e
v
, has
to be known. Rather than applying approximation (3.10) a more accurate
map depending on the intake manifold and speed was created beforehand.
A problem with a direct application of the speed-density formula is that the
intake manifold pressure signal is very noisy, due to the pumping operation of
the engine. Therefore, a low pass lter has to be applied to lter the noise
what directly leads to a reduced speed of the system response. By using the
throttle position sensor, which typically has a fast response and a low noise
level, prediction of the air mass ow (intake manifold pressure) can be made
on the basis of the throttle position and speed. It is possible to use the complete
air path model that was presented in section 3.2.1, but the intake air can also
be approximated with a rst order lter with a variable gain and a variable
time constant, which are stored as a map in the engine management system:
m
ap
(, n) =
K

(, n)
1 +T

(, n)s
(4.7)
The same type of formula can be used to rst predict the intake manifold pres-
sure and subsequently the intake air on the basis of the speed-density equation.
This allows the comparison of the measured and predicted intake manifold pres-
sure and on-line model adaptation if necessary.
The bandwidth of the inner loop is typically larger than the bandwidth
of the outer loop. Therefore the inner loop dynamics, apart from the transfer
delay, can usually be neglected during the tuning of the catalytic converter con-
troller. Such a procedure was applied in the previous chapter. Here, however,
the oxygen storage capacity of the catalytic converter is very low so the engine
dynamics cannot be neglected during the tuning of the catalyst controller. The
inner loop (engine+controller) model used during the tuning of the catalytic
converter controller is a rst order lter with a delay:
140 Experimental testing of the control system

e
(s)

ref
(s)
=
e
T
e
s
1 +
e
s
(4.8)
The model of the catalytic converter (4.1) connected to the above inner control
loop model was the process model applied during the calculation of the MPC.
The steady state optimization problem was the same as in the simulations
(3.60). The dynamic optimization problem (3.62), on the other hand, included
both weighting on the predicted inlet and outlet . With W
d1
= 0, a very
aggressive control is allowed when the conversion is high. This may even lead
to stability problems since the conversion functions (f
L
(), f
R
()) of the applied
converter are close to 1 in a wide operating range. By including the weight on
the inlet signal, an aggressive input is extra penalized, but the optimization
problem becomes more complex.
The controller tuning, both steady state and dynamic, was performed for
two temperatures, 610 and 690K. This temperature range is crucial as most
changes of the model dynamics occur here. For a higher accuracy more tem-
peratures can be selected, but this will increase the computational problem
signicantly.
The nal controller is the Gaussian radial basis function network, intro-
duced in section 3.6.3. The application of the controller is somewhat more
complex, however. The rst dierence is that the engine value becomes
an additional state of the system, and thus x = [e m
ex

e
]
T
X. This,
of course, means that the number of necessary optimization problems to be
solved increases. The controller still has the form shown in (3.65) but it is now
a nonlinear PD controller as the engine lambda signal is in fact the deriva-
tive of ROC. The temperature should in principle also be included in vector x
what would lead to an even more complex, four dimensional problem. Since
the temperature dependence of the controller is very low at higher tempera-
tures, the tuning was performed at the two above mentioned temperatures, and
the controller output was subsequently obtained by interpolation. Below the
lower and the upper temperature, the controller output was assumed not to be
dependent on the temperature.
Figure 4.9 shows the block diagram of the nal controller applied in the
experiments. Figure 4.10 shows the scheme of the tuning procedure. Note
that there are two delays present: rst the delay from the fuel injection to
the rst lambda sensor placed behind the engine, T
e
, and a second delay from
the rst to the second lambda sensor placed behind the catalytic converter,
T
c
. To increase the system bandwidth the second delay is placed after the
catalyst model and is thus placed out of the control loop. This is allowed since
the controller uses the model merely for the estimation of the oxygen storage
coverage. Moreover, the second delay was very large because the catalyst was
placed rather far from the engine in order to keep the temperature during the
experiments somewhat lower. However, when predicted and measured lambda
4.4. Closed loop tests: model-based control 141
contr. IMC Engine
Eng_mod
Catalyst
_ref c
c_est
_est
e
e_est
e_err
- -
- -
m_fi
sensor
Cat_mod
sensor
e_ref
T
cat flow
Figure 4.9: Scheme of the controller applied in the experiments.
Controller
_ref
e_ref
T
cat
flow
Steady state
optimization
Dynamic
optimization
1
1+T
e
s
1
1+T
e
s
e
-Tes
e
-Tcs
d/dt=.
Catalyst model
e

Inner control loop


Process model
c_est
c
+
-
Figure 4.10: The controller tuning scheme.
signals behind the converter have to be compared the second delay has to be
included, as shown in gure 4.10.
4.4.2 Experimental results
The controller has been tested by three test cycles shown in gure 4.11. The
engine speed was kept constant at 1200rpm, while throttle transients created
intake manifold transients. Test 1 and 3 have low initial temperatures of 640
and 630K respectively. Test 2 simulates a high load operation so its initial
temperature is higher, 770K. Tests 1 and 2 start with initially fully covered
oxygen storage, while test 3 starts with initially optimal coverage of ceria. Such
142 Experimental testing of the control system
0 5 10 15 20 25 30 35 40 45 50
50
55
60
65
70
M
A
P

[
k
P
a
]
0 5 10 15 20 25 30 35 40 45 50
60
70
80
90
100
M
A
P

[
k
P
a
]
0 5 10 15 20 25 30
40
60
80
100
time [s]
M
A
P

[
k
P
a
]
Figure 4.11: Intake manifold pressure signals during the applied tests: test
1-above, test 3-below.
highly dynamic tests have been chosen as they represent a driving pattern that
can resemble city driving. It was already shown in the previous text that almost
all emissions, after the catalyst light-o, occur as a result of engine transients.
Moreover, the exhaust temperature is lower during city driving due to frequent
stops. Therefore such dynamic tests should actually be considered as emission
test cycles that try to asses the performance of the system under city driving
conditions.
The catalyst controller is compared with a = 1 controller that only tries
to keep the engine lambda value at stoichiometry without considering the dy-
namics of the catalytic converter (the outer control loop is disconnected). The
air/fuel ratio controller was slightly detuned so considerable lambda distur-
bances were present during the throttle steps. A delay of approximately 0.1s
was present in the hardware communication, such that a response to a sudden
throttle change could not be immediate. This delay was partially compensated
for by the intake air prediction on the basis of the throttle position signal, but
a perfect compensation was not possible.
Test 3 will be analyzed in more detail. Figure 4.12 shows the inlet and outlet
lambda signals of the converter in both cases. It is clear that the emission level
is quite high, largely due to the small oxygen storage capacity of the applied
catalytic converter (it is much smaller than the capacity of current production
catalysts), as the main goal of the study was to show possible benets of the
catalyst control in comparison with a no-control case. The relative oxygen
4.4. Closed loop tests: model-based control 143
0 5 10 15 20 25 30
0.95
1
1.05
1.1
l
a
m
b
d
a

[

]
0 5 10 15 20 25 30
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
Figure 4.12: Converter inlet (thin line) and outlet (thick line) lambda signals
during test 3. Above: converter controller, below: = 1 controller.
coverage is shown in gure 4.13. Note that this signal was used as the feedback
signal only in the catalyst control case. It is clear that the coverage remains
in the vicinity of the desired steady state level in the control case, while it can
deviate greatly in the no-control case. Deviations toward the full oxygen storage
increase the level of NOx emissions, while deviations toward the empty storage
increase the level of HC and CO emissions. This can be seen in gure 4.14
where the NOx and HC emissions are presented for both cases. Response
times of NOx and HC analyzers are around 1s. CO emissions have also been
measured but the analyzers time constant was very large and the results should
be taken with caution. The HC and NOx emissions in tests 1 and 2 are shown
in gures 4.15 and 4.16. The emission levels from all tests are summarized in
table 4.1. The signal in the case of the = 1 controller stays at higher levels
than in the control case what suits mostly CO oxidation. As a consequence, the
CO level is higher with the catalyst controller. This will be further discussed in
the following section. Despite this fact, the accumulated emission level of HC
is decreased with the catalyst controller. Together with a large improvement in
NOx emissions (which can be expected by examining the mean values during
the tests) this shows the usefulness of the catalytic converter controller. Note
that various tests have dierent contribution to the total emission level: though
relative CO emissions increase signicantly in test 2 the overall CO level in test
2 is very low as well as its contribution to the total gure.
144 Experimental testing of the control system
0 5 10 15 20 25 30
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
R
O
C

[

]
Figure 4.13: Relative oxygen coverage during test 3. Thick line: converter
controller; thin line: = 1 controller.
0 5 10 15 20 25 30
0
500
1000
1500
2000
N
O
X
o
u
t

[
p
p
m
]
0 5 10 15 20 25 30
0
100
200
300
400
time [s]
H
C
o
u
t

[
p
p
m
]
Figure 4.14: Measured NOx and HC emission during test 3. Thick line: con-
verter controller; thin line: = 1 controller.
4.4. Closed loop tests: model-based control 145
0 5 10 15 20 25 30 35 40 45 50
0.2
0.4
0.6
0.8
1
R
O
C

[

]
0 5 10 15 20 25 30 35 40 45 50
0
500
1000
1500
2000
N
O
x
o
u
t
[
p
p
m
]
0 5 10 15 20 25 30 35 40 45 50
0
200
400
600
800
time [s]
H
C
o
u
t

[
p
p
m
]
Figure 4.15: Relative oxygen coverage and NOx and HC emission during test
1. Thick line: converter controller; thin line: = 1 controller.
0 5 10 15 20 25 30 35 40 45 50
0.2
0.4
0.6
0.8
1
R
O
C

[

]
0 5 10 15 20 25 30 35 40 45 50
0
500
1000
1500
2000
N
O
x
o
u
t

[
p
p
m
]
0 5 10 15 20 25 30 35 40 45 50
0
100
200
300
400
500
time [s]
H
C
o
u
t

[
p
p
m
]
Figure 4.16: Relative oxygen coverage and NOx and HC emission during test
2. Thick line: converter controller; thin line: = 1 controller.
146 Experimental testing of the control system
NOx HC CO
cat.cont.

=1
Test 1 -13.1% -8.7% -4.9% 0.57 0.61
Test 2 -33.6% +5.2% +54.5% 0.55 0.71
Test 3 -36.6% -7.9% +23.8% 0.54 0.66
Total -32.4% -4.3% +14.6%
Table 4.1: Emission levels during tests with the catalytic converter controller
compared with emissions of the = 1 controller and average values during
tests.
Figure 4.17: A simplied conversion characteristic (f
d
()) and the inuence of
the disturbance ratio on the conversion.
4.4.3 Discussion
The main question that arises is when does the application of the catalytic con-
verter control signicantly reduce the emissions. Figure 4.17 shows a simplied
diagram in which the conversion characteristic of the converter during both
lean and rich excursions is plotted. The conversion eciency is basically given
with f
d
(), see equation (3.40). A very important variable is the ratio of the
disturbance amplitude and the oxygen storage capacity, which will be called
disturbance ratio. If the disturbance ratio has a high value (which is the case
in this study) the overall eect of the control may not be great because larger
disturbances cannot be fully buered with the given limited oxygen storage
capacity. This eect can be seen after the large disturbance at 17s in test 3
(see gures 4.12, 4.13).
A low disturbance ratio (which is common for todays catalytic converters)
4.4. Closed loop tests: model-based control 147
0 5 10 15 20
15
20
25
30
35
40
45
50
t
h
r
o
t
t
l
e

a
n
g
l
e

[
d
e
g
]
0 5 10 15 20
0.9
0.95
1
1.05
1.1
time [s]
l
a
m
b
d
a

[

]
Figure 4.18: Throttle position and engine lambda signals (solid line - Gaussian
controller; dashed line - = 1 controller) during test C.
will very likely lead to a much higher improvement in the control case. In
that case a disturbance cannot produce a large emission level if the system is
controlled and the only danger is that the system drifts away from the desired
coverage level. That condition may occur if no control is applied. Moreover,
the controller oers a fast and accurate transition after a fuel cut-o and fuel
enrichment (see section 3.6.4).
To support the above discussion an additional simulation on the system
presented in chapter 3 was conducted. This simulation will be called test C,
while tests presented in section 3.6.4 will be called test A (with initially empty
oxygen storage) and test B (with initially full oxygen storage). The initial
oxygen storage lling in test C is the optimal coverage. The applied dynamic
test is shown in gure 4.18. This test is similar to the experimental tests applied
in this chapter. The engine signals produced by the converter controller and
= 1 controller are also shown. The engine controller in both cases was slightly
more detuned than in section 3.6.4, producing even more severe disturbances
to the system. The disturbances are thus more similar to those applied in
the experiments. Figure 4.19 presents the emissions during the test and the
relative coverage of ceria, calculated by the rst principle model. As expected,
when the coverage of ceria reaches levels below the optimal coverage, a rich
inlet produces CO and HC emissions. The opposite holds for the emission of
NO. The ceria coverage is below the optimal value during the largest part of
the test with the = 1 controller, while above the optimal value at the end of
the test.
148 Experimental testing of the control system
0 5 10 15 20
0
0.005
0.01
0.015
0.02
time [s]
C
O
o
u
t

[
v
o
l
%
]
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
time[s]
N
O
o
u
t

[
v
o
l
%
]
0 5 10 15 20
0
0.2
0.4
0.6
0.8
1
1.2
x 10
4
time [s]
H
C
o
u
t

[
v
o
l
%
]
0 5 10 15 20
0.1
0.2
0.3
0.4
0.5
0.6
0.7
time [s]
R
O
C

[

]
Figure 4.19: Converter outlet emissions and average degree of ceria lling dur-
ing test C. Solid line - Gaussian controller; dashed line - = 1 controller
When only test C is accounted for the catalytic converter controller reduces
the emission of CO by 76% and HC by 40%. The emission of NO is increased by
37% because the ceria coverage is mostly below the optimal in the case of = 1
controller. Thus, with the controller applied, some emission peaks can still
occur due to the limitations of the oxygen storage/release rate. Though these
peaks are inevitable, they are much lower if the system stays in the optimal
point. Due to a low disturbance ratio, disturbances of approximately 10%
cannot drive the system far from the optimum to cause large emissions when
the converter controller is applied. This was not the case in the experiments.
When the total emissions produced in test C are added to tests A and B the
case with the catalytic converter controller leads to an improvement of 75%
for CO, 80% for HC and 84% for NO. The total emissions are predominantly
determined by tests A and B. The contribution of test C to the total emissions
with the catalytic converter controller is 1.3% for CO, 5.6% for HC and 35%
for NO, while with the = 1 controller 1.3% for CO, 1.9% for HC and 4.1%
for NO. The contribution in the case of the catalytic converter controller is
larger because the controllable emissions (full, empty oxygen storage) are much
smaller with respect to the uncontrollable emissions (large disturbances, limited
system bandwidth).
The catalytic converter model can also be used in advance to properly
select a catalytic converter for some application by estimating the emission
level on the basis of a known disturbance ratio. The disturbance ratio can
4.5. Conclusions 149
be assessed by knowing the accuracy of the A/F controller and anticipated
excursions from the desired value during transients. By properly controlling
the system a catalyst with a lower oxygen storage capacity can be applied
without increasing the emission level.
4.5 Conclusions
Experimental tests with the model-based controller for a catalytic converter
were presented in this chapter. Since the controlled variable the relative degree
of ceria coverage by oxygen has to be estimated by the model, the accuracy
of the model is crucial for a good control performance. Parameters of the
nonlinear model were estimated on the basis of the converter step responses.
For a model to be useful for control, it has to have a wide range application
and should therefore contain the parameters dependence on temperature and
exhaust mass ow. The crucial parameter in the model is the oxygen storage
capacity, which is a function of both above mentioned parameters.
The MPC based controller is tuned o-line by solving the static and dy-
namic optimization problems in various operating points. A Gaussian network
is used to approximate the MPC on-line. The controller was tested with vari-
ous dynamic tests and its performance compared to the stoichiometric engine
lambda controller. It was shown that by accurately controlling the level of
stored oxygen on ceria one can obtain improved performance of the exhaust
system. Though the oxygen storage capacity of the converter was very low,
and therefore accurate control proved to be very dicult, the controller lead
to signicantly improved NOx emission while not deteriorating HC emission.
More extensive tests are necessary to fully validate the proposed controller.
Namely, tests with various catalysts have to be performed. The converter
dynamics has to be further investigated, mainly the behavior after rich steps
and dependence of the oxygen storage capacity on various parameters. This
should be coupled with the investigation of improved correction models for the
downstream lambda sensor. The controller tuning is currently performed o-
line and is rather complex. Parametric optimization techniques, should lead to
a much shorter and more ecient controller tuning process.
150 Experimental testing of the control system
5
Conclusions and Outlook
5.1 Conclusions 5.2 Outlook
5.1 Conclusions
Three-way catalytic converters have been improved a lot since their introduc-
tion in the late 1970s. Though their main purpose has remained the same, to
simultaneously convert CO, HC and NO into CO
2
, H
2
O and N
2
, the system
performance, durability, and resistance to poisoning have been constantly im-
proving. However, in order to meet highly stringent future emission legislation,
the system has to be accurately controlled. The still remaining main sources
of pollution, engine cold start and transient operation (sudden acceleration or
deceleration), have been addressed in this thesis.
The rst question to be answered by the thesis was whether the dynamics
of the catalytic converter may be neglected during the design of the control
system. Further questions were related to the determination of major dynamic
eects within the converter, control system design and assessment of benets
of the novel control system. The detailed process model was created to obtain
a better insight into the process behavior. All of these questions have been
covered in the thesis. The main results are summarized in the sequel.
Dynamic behavior of the catalytic converter has been studied during the
cold start and dynamic operation by means of the developed rst principle
model. Possible (simple) cold start strategies were assessed. The rigorous, rst
principle model was also used to develop a simpler, black box, control-oriented
model. On the basis of this model, a model-based controller was created, which
can improve the systems performance under dynamic operation.
5.1.1 Modeling
The rst principle model is developed in order to study dynamic behavior of the
catalytic converter. With help of the model, many processes that take place on
the catalytic surface can be observed and studied, what is very dicult directly
152 Conclusions and Outlook
by means of experiments. The basis for the converter model was an earlier de-
veloped elementary step kinetic model for the oxidation of carbon monoxide,
hydrocarbons ethylene and acetylene and the reduction of nitric oxide. The
elementary step kinetic model is suited for a study of dynamic processes on
the catalytic surface, as it directly calculates mass balances on the catalyst,
and moreover it is valid in a wide operating range. The model was validated
with light-o and lambda sweep engine bench tests. The monolithic converter
was coated with the same Pt/Rh/-Al
2
O
3
/CeO
2
catalyst as used in the ki-
netic experiments. The model can describe, with a reasonable accuracy, the
performed experiments. Some model shortcomings which have to be addressed
in the future work will be discussed in the following section.
The general conclusion is that the dynamic behavior of the catalytic con-
verter at low temperatures is dominated by processes on the noble metal, while
at higher temperatures by processes involving oxygen storage on ceria. An-
other conclusion is that when describing light-o of the converter one should
always consider the complete exhaust mixture. Studying only single reactions,
and concluding on the basis of it about the global process behavior may be
very dangerous. The reason for this is that the light-o is predominantly de-
termined by inhibition processes. Acetylene is a very strong inhibitor, as it
can cover the largest part of the noble metal surface at low temperatures, due
to its strong adsorption capabilities, not permitting the other components to
react. The activation energy of acetylene oxidation is much higher than, for
example, the activation energy of carbon monoxide oxidation, and therefore the
conversion of carbon monoxide cannot start before the conversion of acetylene
starts. Nitric oxide has also inhibiting capabilities, but less pronounced than
acetylene.
Some simple measures, such as feed oscillations and secondary air injection,
can in some cases improve the light-o. Oscillating feed can improve the light-
o when the inhibiting component changes with the inlet feed composition.
It mostly aects the components that are the last to be converted (ethylene,
nitric oxide), as feed oscillations can lead to surface coverage oscillations and
thus improved conversion. Secondary air injection helps to obtain very lean
conditions upstream of the converter, although the engine runs rich or at stoi-
chiometry. While improvement for engine-rich conditions is rather obvious, an
improvement occurs even in engine-stoichiometric conditions as large oxygen
partial pressure increases the oxygen coverage of the noble metal surface at
low temperatures and improves mainly the CO conversion. By switching the
secondary air o, after the reaction has been ignited, a better conversion of
all components can be achieved. This is obtained by applying the downstream
lambda sensor and short rich pulses.
Oxygen, and to a lower extent NO, can be stored during lean transients
on the ceria surface, and can also subsequently reach deeper ceria bulk levels.
These stored components can later be used for oxidation of CO and HC on
5.1. Conclusions 153
the noble metal surface under rich conditions. The main conclusion is that
the catalytic converter should always be considered as a dynamic element.
Describing and understanding the oxygen storage phenomenon is important for
the controller development, as it is the main dynamic feature of the catalytic
converter under standard operating conditions (after the light-o). With a fast
transfer of oxygen between ceria surface and bulk, the system response is as
if there was only one, larger storage. The shapes of lean to rich and rich to
lean step responses are dierent due to a large CO desorption after a rich to
lean step, which can produce a rich converter output with a lean input. Due
to the noble metal surface inhibition in the front part of the reactor, not all
oxygen is being used when the input is rich. The availability of oxygen from
the storage depends on the operating conditions (temperature, exhaust mass
ow). A number of dynamic simulations (inlet lambda steps) was performed in
order to study the process behavior, with the stress on the inuence of various
parameters such as the exhaust mass ow, oxygen storage capacity (aging)
and the inlet lambda amplitude. It was found that the lambda plateau, i.e.
duration of a high conversion after a step, depends in various ways on all the
above mentioned parameters.
5.1.2 Control
The conversion of the catalytic converter depends on the oxygen storage lling
level. The same lambda input to the converter can produce very dierent
outputs if the storage is almost completely lled (large NO emission during a
lean excursion), completely empty (large CO and HC emissions during a rich
excursion) or half lled (good conversion under all conditions). This is the
basis for the controller development. The rst principle model serves as a good
tool for the initial controller testing and development of the control-oriented
model.
The developed control-oriented model predicts the relative degree of ceria
coverage by oxygen. It is a one-state nonlinear model, whose parameters can
be estimated on-line. The estimation is based on step responses of the catalytic
converter. A computationally simple method for the parameter estimation was
developed. Relations that can broaden the operating range of the model with
respect to the exhaust mass ow and inlet lambda amplitude were developed
on the basis of the rst principle model. The accuracy of the control-oriented
model was assessed by comparing its prediction to the prediction of the rst
principle model. The oxygen storage dynamics can be accurately predicted by
the simple model. Such a validation is only possible by using the rst principle
model since the degree of ceria coverage cannot be measured. The model can
also quite accurately predict experimentally measured data.
A cascade control system, with an Internal Model Controller for the engine
air/fuel ratio, is used to control the degree of ceria coverage. The controller
154 Conclusions and Outlook
uses the model of the catalytic converter as the inferential sensor because the
controlled variable cannot be measured directly. The accuracy of the model is
crucial for the controller performance. The insights into the converter dynamics
obtained via the rst principle model are very important therefore.
Analytic, model-based predictive controller is applied. The main feature
of the controller is that it includes the information on the emissions available
from the model during the controller tuning. The on-line controller is based
on a Model Predictive Control (MPC) system. First a steady state problem is
solved in order to determine the optimal steady state ceria coverage under given
operating conditions. After that the controller nds an optimal trajectory to
reach this steady state. A direct on-line application of MPC is dicult due
to computational requirements, so the controller is approximated o-line with
a Gaussian network, which is then applied on-line. The network training is
computationally very simple, but a large number of optimization problems
have to be solved to nd the MPC outputs that cover the complete operating
range of the process. Simulations of extreme dynamic conditions have shown
that the controller can lead to a signicant emission reduction.
The controller was also tested experimentally on an engine dynamometer
test bench. Open loop tests have shown that the transfer of oxygen from ceria
bulk to the surface is much slower than predicted by the rst principle model.
An additional state in the control-oriented model had therefore to be added.
This eect can also be described by the rst principle model after adaptation
of some kinetic parameters. The applied controller was slightly more complex
than the one used in the simulations because of a low oxygen storage capacity of
the available catalytic converter. Hence, the engine lambda value was included
as an additional controller input making the nal Gaussian network a nonlinear
PD controller. The controller was tested with three sets of highly dynamic tests.
Its performance has been compared to the performance of the stoichiometric
controller. Though the oxygen storage capacity of the converter was very low,
and therefore accurate control very dicult, the controller lead to a signicantly
improved NOx conversion while it did not deteriorate the HC conversion. The
benet of the controller increases signicantly with an increase of the ratio
between the oxygen storage capacity and disturbance amplitude. By applying
the controller, a catalytic converter with a lower oxygen storage capacity can
be applied without increasing the emission level.
The control ideas presented in this thesis are protected by a patent appli-
cation [7].
5.2. Outlook 155
5.2 Outlook
5.2.1 Modeling
Though quite extensive, the kinetic model is not complete yet. Some very
important issues, such as inuence of steam in the feed and constant presence
of oxygen in the exhaust were not considered during the development of the
kinetic model [38]. The accurate modeling in the presence of steam would
lead to a better understanding of the converter behavior under rich conditions,
water gas shift and steam reforming reactions. A kinetic model for the oxidation
of hydrogen should therefore also be included in the model. This is of special
interest for control because hydrogen leads to errors in the downstream lambda
sensor. Steam presence also inuences oxygen storage dynamics, by promoting
the adsorption of oxygen on ceria, and negatively inuencing the capability of
NO to adsorb on ceria. This eect was qualitatively included in the model,
but a more accurate quantitative kinetic study is necessary. Understanding of
the interplay between the reactions involving the oxygen storage and reactions
with steam is also of a great importance for improving the accuracy of the
control-oriented model.
An investigation of oxygen storage behavior at higher temperatures is also
recommended. It is known that by increasing the catalyst temperature the
ceria bulk starts playing a more important role. The observed oxygen storage
thus increases, but a precise quantication of this eect is still not available.
This would be of great importance for control, because it would reduce the
number of needed operating points for the model estimation. The transfer be-
tween the ceria bulk and surface should further be studied, since it was shown
by experiments in section 4.3.1 that the transfer of oxygen to the bulk is ap-
parently faster than the oxygen release from the bulk to the surface. This was,
however, not observed in experiments performed on another catalyst presented
in section 3.4.3. A study should therefore be performed with dierent catalysts
and their behavior has to be compared.
The kinetic models were obtained at rather low temperatures (around the
light-o). The model extrapolation to higher temperatures should yet be
checked.
In order to expand the model to other operating conditions, such as lean
burn and diesel conditions, more hydrocarbons should be taken into account.
Some hydrocarbons (alkanes) have diculties to be oxidized under oxygen rich
conditions, leading to a decreased hydrocarbon conversion under these condi-
tions. The current model cannot predict this phenomenon.
A further issue to be considered is the model complexity. The model is
already rather complex, and this complexity will be even greater with the
addition of the above mentioned reactions. This leads to large computation
times, especially at higher temperatures and under dynamic conditions. Since
most of the rate parameters in the model have exponential characteristics,
156 Conclusions and Outlook
the dierential equations becomes very sti at higher temperatures and hence
very dicult to solve. The research to tackle this problem can go in two
directions. The rst possibility is to perform a model reduction, i.e. to nd
out which parts of the kinetic model can be neglected under certain operating
conditions. One can assume, for example, that dierent acetylene and ethylene
species currently existing in the model can be replaced, at higher temperatures,
with single acetylene and ethylene species. Another possibility is to assume a
quasi steady state on the noble metal surface at higher temperatures. This is
allowed since the processes on the noble metal surface are much faster then
the processes involving ceria or thermal processes inside of the converter. This
steady state can be approximated with a neural network whose inputs may
be gas concentrations, temperature and concentrations on the ceria surface,
for each point in the reactor. The network outputs may be the degrees of
the noble metal coverage by dierent components. The neural network has
to be tuned once for a certain model. In this manner a large number of sti
dierential equations will be replaced with algebraic equations which are very
easy to solve. The simulation time will probably decrease drastically.
The ultimate idea is to create a fast and reliable model that can be applied
not only for controller design, but also for extensive testing of the system
performance what greatly decreases the need for expensive experiments.
5.2.2 Control
As already mentioned, the downstream lambda sensor exhibits errors in both
the rich and lean region. These errors are much larger in the rich region, where
due to the sensor sensitivity to hydrogen even dynamic errors occur. It is there-
fore of a great importance to make an accurate compensation for these sensor
errors. This is directly linked to studying the eects of steam and hydrogen
oxidation mentioned in the previous section. By fully understanding these pro-
cesses and sensor operation, a soft sensor can be made in order to estimate
the real downstream lambda signal on the basis of sensor measurements, op-
erating conditions (temperature, ow rate) and the process model. This can be
achieved with a nonlinear observer where the actual lambda is the state that
has to be estimated and linked via the nonlinear dynamics with the output,
which is then compared to the actually measured signal. The sensor model will
be an extension to the converter model, which incorporates the oxygen storage,
water gas shift and steam reforming dynamics.
Controller tuning is performed o-line because the on-board computer can-
not solve the necessary optimization problems on-line in every sampling inter-
val. The o-line optimization has to be performed in the complete operating
range of the process to train the Gaussian network controller. Though it ulti-
mately leads to a controller that can be calculated fast enough, the optimiza-
tion procedure may still be too involving. For an MPC with a linear model
5.2. Outlook 157
and linear constraints, a multi-parametric quadratic program can be solved in
which inputs are treated as optimization variables, and states are treated as
parameters [78]. The solutions for this problem are linear proles of all inputs,
which are valid in closed regions. The regions boundaries are given with a
set of linear inequalities depending on the optimization parameters. The linear
proles and corresponding regions can be obtained with an ecient algorithm.
Not only the on-line controller calculation becomes very simple in this case and
no neural network has to be trained o-line, but the optimization has to be
performed only once for every region. The number of regions depends on the
process model and the applied constraints. Such an approach is not very suit-
able for processes with many states, but it can be very suitable for automotive
applications where the number of states is often small. It can, however, not be
applied to the problem of controlling the catalytic converter because the pro-
cess model is nonlinear. For general nonlinear convex optimization problems
the optimal solution can be approximated in some parametric region by solving
optimization problems at the boundaries of the region. This idea is similar to
the one applied in the linear case. Only now the solution is an approximation
of the real solution. The solution of the problem is to nd the region bound-
aries which will guarantee that the error lies within acceptable bounds. The
error bound on the approximated solution can be found, but in general involves
solving a nonconvex parametric problem [25]. Further research into the area of
nonlinear parametric optimization is necessary.
The control ideas presented in this thesis can be extended to the area of
lean burn and diesel engines. The problem there is that the engine runs always
with an excess of oxygen, making it very dicult for the catalytic converter
to reduce NO. One solution to this problem is to apply so-called NOx storage
catalytic converters. These converters can store NO when the inlet feed is lean.
Short rich pulses are applied to purge the storage once it is lled. Due to a
constant dynamic operation of the system, it is expected that a similar model-
based control strategy can be applied. Kinetic studies should be conducted
to study the additional relevant reactions (reaction paths not occurring in the
three-way catalytic converter) in order to build a rst principle model. A
simplied control-oriented model should then be created on the basis of the
rigorous model. This model will comprise oxygen and NOx storage dynamics
and will be the basis for the design of the controller.
158 Conclusions and Outlook
Notation
a
cat
washcoat surface area / m
2
NM
m
3
R
a
v
geometric surface area / m
2
i
m
3
R
A
i
preexponential factor / s
1
c
p
specic heat / Jkg
1
K
1
C concentration / molm
3
f
d
b
monolith channel diameter / m
R
d
w
washcoat thickness / m
R
E activation energy / Jmol
1
k rate coecient
k
f
mass transfer coecient / m
3
f
m
2
i
s
1
L
k
capacity of catalyst phase / molm
2
NM
M molar mass / kgmol
1
r reaction rate / molm
2
NM
s
1
R gas constant / Jmol
1
K
1
s
o
sticking probability on a clean surface
T temperature / K
T time delay / s
(
r
H) reaction enthalpy / Jmol
1
A/F air to fuel ratio
p pressure /Pa
m mass /kg
V volume /m
3
n engine speed / rpm
X fuel deposit factor
i

1
Greek symbols
heat transfer coecient / Wm
2
i
K
1
throttle angle / rad
monolith converter void fraction / m
3
f
m
3
R

w
washcoat porosity / m
3
f
m
3
w

sup
m
supercial mass ow / kgm
2
R
s
1
fractional surface coverage
empty surface fraction
160 Notation
local fractional ceria surface coverage
local fractional ceria bulk coverage
global relative ceria coverage - (ROC)
global relative uncovered bulk ceria
thermal conductivity / Wm
1
K
1
normalized air/fuel ratio
density / kgm
3
f
time constant / s
frequency / s
1
Subscripts and abbreviations
NM noble metal
CS ceria surface
CB ceria bulk
CT total ceria capacity (surface + bulk)
OSC oxygen storage capacity
ROC relative ceria coverage by oxygen ()
a adsorption
d desorption
r reaction
f bulk gas phase
s surface(washcoat) phase
w washcoat
a air
f fuel
ff fuel lm
t throttle
p port
m intake manifold
in inlet (input)
out outlet (output)
cyl cylinder
inj injection
ev evaporation
e engine
c catalyst
n nominal
R reactor
R rich
L lean
Notation 161
d direction
ex exhaust
ex excess
fill lling
emp emptying
gr global reaction
ss steady state
sen sensor
mix mixing
fp rst principle model
Superscripts
b backward
f forward
k k-th data point
162 Notation
Bibliography
[1] F. Algoewer, T.A. Badgwell, J.S. Qin, J.B. Rawlings, and S.J. Wright.
Nonlinear predictive control and moving horizon estimation - an intro-
ductory overview. In P.M. Frank, editor, Advances in Control, Highlights
of ECC99, pages 391449, 1999.
[2] G. Almkvist and S. Eriksson. A study of air to fuel transient response
and compensation with dierent fuels. SAE Paper No. 941931, 1994.
[3] C.F. Aquino. Transient A/F characteristics of the 5 liter central fuel
injection engine. SAE Paper No. 810494, 1981.
[4] M. Balenovic, A.C.P.M. Backx, and J.H.B.J. Hoebink. On a model-based
control of a three-way catalytic converter. SAE Paper No. 2001-01-0937,
2001.
[5] M. Balenovic and T. Backx. Model-based predictive control of a three-
way catalytic converter. Proc. IEEE Conference on Control Applications,
Mexico City, pages 626631, 2001.
[6] M. Balenovic, T. de Bie, and T. Backx. Development of a model-based
controller for a three-way catalytic converter. SAE Paper No. 2002-01-
0475, 2002.
[7] M. Balenovic, J.M.A. Harmsen, A.C.P.M. Backx, J.H.B.J. Hoebink, and
J.C. Schouten. Autonoom mobiel voertuig. Nederlandse octrooiaanvrage
1017481.
[8] M. Balenovic, J.M.A. Harmsen, A.C.P.M. Backx, and Hoebink J.H.B.J.
Advanced modeling and control of a three-way catalytic converter. Jour-
nal A, vol. 42, no. 1, pages 2530, 2001.
[9] M. Balenovic, J.M.A. Harmsen, J.H.B.J. Hoebink, and A.C.P.M. Backx.
Low temperature dynamic characteristics of a three-way catalytic con-
verter. Proc. IFAC Workshop on Advances in Automotive Control, pages
323328, 2001.
[10] M. Balenovic, A.J.L. Nievergeld, J.H.B.J. Hoebink, and A.C.P.M. Backx.
Modeling of an automotive exhaust gas converter at low temperatures
aiming at control application. SAE Paper No. 1999-01-3623, 1999.
[11] B.I. Bertelsen. The U.S. motor vehicle emission control programme. Plat-
inum Metals Review, 45, No.2, pages 5059, 2001.
164 Bibliography
[12] E.P. Brandt, Y. Wang, and J.W. Grizzle. Dynamic modeling of a three-
way catalyst for SI engine exhaust emission control. IEEE Trans. on
Control Systems Technology 8, pages 767776, 2000.
[13] R. Burch and T.C. Watling. Kinetics of the reduction of NO by C
3
H
6
and C
3
H
8
over Pt based catalysts under lean-burn conditions. Stud. Surf.
Sc. Catal. 116, pages 199211, 1998.
[14] S.B. Choi and J.K. Hedrick. An observer-based controller design method
for improving air/fuel characteristics of spark ignition engines. IEEE
Trans. on Control Systems Technology 6, No. 3, pages 325334, 1998.
[15] S.B. Choi, M. Won, and J.K. Hedrick. Fuel-injection control of S.I. en-
gines. Proc. 33rd Conference on Decision and Control, Lake Buena Vista,
FL, pages 16091614, 1994.
[16] S.J. Cornelius. Modelling and Control of Automotive Catalysts. PhD
thesis, Sidney Sussex College, University of Cambridge, March 2001.
[17] E.W. Curtis, C.F. Aquino, D.K Trumpy, and G.C. Davis. A new port
and cylinder wall wetting model to predict transient air/fuel excursions
in a port fuel injected engine. SAE Paper No. 961186, 1996.
[18] A.W.B. de Bie. Validation of a wall-wetting model for MPI petrol engines.
TNO report 00.OR.VM.050.1/TDB, TNO Automotive, Delft, 2000.
[19] A.F. Diwell, R.R. Ramaram, H.A. Shaw, and T.J. Truex. The role of
ceria in three-way catalysts. Stud. Surf. Sc. Catal. 71, pages 139152,
1991.
[20] M.C. Drake, R.M. Sinkevitch, A.A. Quader, K.L. Olson, and T.J. Cha-
paton. Eect of fuel/air ratio variations on catalyst performance and
hydrocarbon emissions during cold-start and warm-up. SAE Paper No.
962075, 1996.
[21] P. Eastwood. Critical Topics in Exhaust Gas Aftertreatment. Research
Studies Press LTD., Baldock, Herfordshire, England, 2000.
[22] C.G. Economou and M. Morari. Internal model control. 5. Extension to
nonlinear systems. Ind. Eng. Chem. Process Des. Dev. 25, pages 403411,
1986.
[23] J.M. Evans, G.P. Ansell, C.M. Brown, J.P. Cox, D.S. Lafyatis, and P.J.
Millington. Computer simulation of the FTP performance of 3-way cat-
alysts. SAE Paper No. 1999-01-3472, 1999.
[24] R.J. Farrauto and R.M. Heck. Catalytic converters: state of the art and
perspectives. Catalysis Today 51, pages 351360, 1999.
Bibliography 165
[25] A.V. Fiacco and J. Kyparisis. Computable bounds on parametric solu-
tions of convex problems. Math. Prog. 40, pages 213221, 1988.
[26] J. Fiaschetti and N. Narasimhamurthi. A descriptive bibliography of SI
engine modeling and control. SAE Paper No. 950986, 1995.
[27] C.E. Garcia, D.E. Prett, and M. Morari. Model predictive control: The-
ory and practice - a survey. Automatica 25, pages 335348, 1989.
[28] H.J. Germann, C.H. Onder, and H.P. Geering. Fast gas concentration
measurements for model validation of catalytic converters. SAE Paper
No. 950477, 1995.
[29] H. Germann, S. Taglaiferri, and H.P. Geering. Dierences in pre- and
post-converter lambda sensor characteristics. SAE Paper No. 960335,
1996.
[30] P.E. Gill, W. Murray, and M.H. Wright. Practical Optimization. Aca-
demic Press, London, 1981.
[31] P. Greening. European vehicle emission legislation - present and future.
Topics in Catalysis 16/17, Nos. 1-4, pages 514, 2001.
[32] L. Guzzela, M. Simmons, and H.P. Geering. Feedback linearizing air/fuel-
ratio controller. Control Eng. Practice 5, No. 8, pages 11011105, 1997.
[33] L. Guzzela. Models and modelbased control of IC-engines - a nonlinear
approach. SAE Paper No. 950844, 1995.
[34] D.R. Hamburg, J.A. Cook, W.J. Kaiser, and E.M. Logothetis. An en-
gine dynamometer study of the A/F compatibility between a three-way
catalyst and an exhaust gas oxygen sensor. SAE Paper No. 830986, 1983.
[35] J.M.A. Harmsen, J.H.B.J. Hoebink, and J.C. Schouten. Transient ki-
netic modelling of the ethylene and carbon monoxide oxidation over a
commercial automotive exhaust gas catalyst. Ind. Eng. Chem. Res. 39,
pages 599609, 2000.
[36] J.M.A. Harmsen, J.H.B.J. Hoebink, and J.C. Schouten. Acetylene and
carbon monoxide oxidation over a commercial exhaust gas catalyst: ki-
netic modeling of transient experiments. Chem. Eng. Sc. 56, pages 2019
2035, 2001.
[37] J.M.A. Harmsen, J.H.B.J. Hoebink, and J.C. Schouten. Kinetic mod-
elling of transient NO reduction by CO in the presence of O
2
over an
automotive exhaust gas catalyst. Stud. Surf. Sc. Catal. 133, pages 349
356, 2001.
166 Bibliography
[38] J.M.A. Harmsen. Kinetic modeling of the dynamic behaviour of an auto-
motive three-way catalyst under cold-start conditions. PhD thesis, Eind-
hoven University of Technology, September 2001.
[39] R.E. Hayes and S.T. Kolaczkowski. Mass and heat transfer eects in
catalytic monolith reactors. Chem. Eng. Sci. 49, pages 35873599, 1994.
[40] W. Hecq. Contribution of fossil fuels and air pollutants emissions in
Belgium since 1980. the role of trac. Stud. Surf. Sc. Catal. 116, pages
522, 1998.
[41] E. Hendricks, A. Chevalier, M. Jensen, S.C. Sorenson, D. Trumpy, and
J. Asik. Modelling of the intake manifold lling dynamics. SAE Paper
No. 960037, 1996.
[42] E. Hendricks and S.C. Sorenson. Mean value modeling of spark ignition
engines. SAE Paper No. 900616, 1990.
[43] E. Hendricks, T. Vesterholm, and C. Sorenson. Nonlinear closed-loop SI
engine control observers. SAE Paper No. 920237, 1992.
[44] J.S. Hepburn, D.A. Dobson, C.P. Hubbard, S.O. Guldberg, E. Thanasiu,
W.L. Watkins, B.D. Burns, and H.S. Gandhi. A review of a dual EGO
sensor method for OBD-II catalyst eciency monitoring. SAE Paper No.
942057, 1994.
[45] J.B. Heywood. Internal Combustion Engine Fundamentals. McGraw-Hill,
1988.
[46] J.H.B.J. Hoebink, J.M.A. Harmsen, M. Balenovic, A.C.P.M. Backx, and
J.C. Schouten. Automotive exhaust gas conversion: from elementary step
kinetics to prediction of emission dynamics. Topics in Catalysis 16/17,
Nos. 1-4, pages 319328, 2001.
[47] J.H.B.J. Hoebink, R.A. v. Gemert, J.A.A. v.d. Tillaart, and G.B. Marin.
Competing reactions in three-way catalytic converters: modelling of the
NOx conversion maximum in the light-o curves under net oxidising con-
ditions. Chem. Eng. Sc. 55, pages 15731581, 2000.
[48] P. Hoekstra, T.J.J. van den Boom, and M.A. Botto. Design of an analytic
constrained predictive controller using neural networks. Proc. American
Control Conference, 2001.
[49] R. Impens. Automotive trac: Risks for the environment. Stud. Surf.
Sc. Catal. 30, pages 1130, 1987.
Bibliography 167
[50] R.A. Jackson, J.C. Peyton Jones, J. Pan, and J.B. Roberts. Chemical
aspects of the dynamic performance of a three-way catalyst. SAE Paper
No. 1999-01-0312, 1999.
[51] V.K. Jones, B.A. Ault, G.F. Franklin, and J.D. Powell. Identication and
air/fuel ratio control of a spark-ignition engine. IEEE Trans. on Control
Systems Technology 3, pages 1421, 1995.
[52] H. Katashiba, R. Nishiyama, Y. Nishimura, Y. Hosoya, H. Arai, and
S. Washino. Development of an eective air-injection system with heated
air for LEV/ULEV. SAE Paper No. 950411, 1995.
[53] E. Koberstein and G. Wannemacher. The A/F window with three-way
catalysts. Kinetic and surface investigations. Stud. Surf. Sci. Catal. 30,
pages 155172, 1987.
[54] K. Kollmann, J. Abtho, W. Zahn, H. Bischof, and J. G ore. Secondary
air injection with a new developed electrical blower for reduced exhaust
emissions. SAE Paper No. 940472, 1994.
[55] G.C. Koltsakis, P.A. Konstantinidis, and A.M. Stamatelos. Development
and application range of mathematical models for 3-way catalytic con-
verters. App. Catal. B: Environmental 12, pages 161191, 1997.
[56] Z. Kovacic, M. Balenovic, and S. Bogdan. Sensitivity-based self-learning
fuzzy logic control for a servo system. The IEEE Control Systems Mag-
azine 18, No. 3, pages 4151, 1998.
[57] S. Kubo, M. Yamamoto, Y. Kizaki, S. Yamazaki, T. Tanaka, and
K. Nakanishi. Speciated hydrocarbon emissions of SI engine during cold
start and warm-up. SAE Paper No. 932706, 1993.
[58] A.B.K. Lie, J.H.B.J. Hoebink, and G.B. Marin. The eects of oscillatory
feeding of CO and O
2
on the performance of a monolithic catalytic con-
verter of automobile exhaust gas: A modeling study. Chem Eng. J. 53,
pages 4754, 1993.
[59] G. Mabilon, D. Durand, and Ph. Courty. Inhibition of post-combustion
catalysts by alkynes: A clue for understanding their behaviour under real
exhaust conditions. Stud. Surf. Sc. Catal. 96, pages 775788, 1995.
[60] Matlab. The Optimization Toolbox. The MathWorks Inc., 1996.
[61] P.E. Moraal. Adaptive compensation of fuel dynamics in an SI engine
using a switching ehgo sensor. Proc. 34th Conference on Decision and
Control, New Orleans, LA, pages 661666, 1995.
168 Bibliography
[62] M. Morari and E. Zariou. Robust Process Control. Prentice Hall Inter-
national Editions, 1989.
[63] M. Morari. Model predictive control: multivariable control technique
of chioce in the 1990s? Advances in Model-Based Predictive Control,
editor: D.Clarke, Oxford University Press, pages 2237, 1994.
[64] H. Muraki and Y. Fujitani. NO reduction by CO over noble-metal cat-
alysts under cycled feedstreams. Ind. Eng. Chem. Prod. Res. Dev. 25,
pages 414419, 1986.
[65] H. Muraki, H. Shinjoh, H. Sobukawa, K. Yokota, and Y. Fujitani. Be-
haviour of automotive noble metal catalysts in cycled feedstreams. Ind.
Eng. Chem. Prod. Res. Dev. 24, pages 4349, 1985.
[66] NAG. Fortran Library Manual, mark 15, vol. 1. Oxford, 1997.
[67] R.H. Nibbelke, A.J.L. Nievergeld, J.H.B.J. Hoebink, and G.B. Marin.
Development of a transient kinetic model for the CO oxidation by O
2
over a Pt/Rh/CeO
2
/-Al
2
O
3
three-way catalyst. Appl. Catal. B: Envi-
ronmental 19, pages 245259, 1998.
[68] A.J.L. Nievergeld, J.H.B.J. Hoebink, and G.B. Marin. The performance
of a monolithic catalytic converter of automobile exhaust gas with oscil-
latory feeding of CO, NO and O
2
: a modeling study. Stud. Surf. Sci. Cat.
96, pages 909918, 1995.
[69] A.J.L. Nievergeld, E.R. van Selow, J.H.B.J. Hoebink, and G.B. Marin.
Simulation of a catalytic converter of automotive exhaust gas under dy-
namic conditions. Stud. Surf. Sc. Catal. 109, pages 449458, 1997.
[70] S.H. Oh and J.C. Cavendish. Transient modeling of 3-way catalytic con-
verters: Response to step changes in feedstream temperature as related
to controlling automobile emissions. Ind. Eng. Chem. Prod. Res. Dev.
21, pages 2937, 1982.
[71] C.H. Onder, M.R. Roduner, C.A. amd Simons, and H.P. Geering. Wall-
wetting parameters over the operational region of a sequential fuel-
injected SI engine. SAE Paper No. 980792, 1998.
[72] Micro Oxivision. Air fuel ratio analyzer. Installation, operation and
maintenance guide, NGK Spark Plug CO. LTD., 1987.
[73] T. Parisini and R. Zoppoli. A receding-horizon regulator for nonlinear
systems and a neural approximation. Automatica 31, pages 14431451,
1995.
[74] K.M. Passino and S. Yurkovich. Fuzzy Control. Addison Wesley, 1998.
Bibliography 169
[75] K.N. Pattas, A.M. Stamatelos, P.K. Pistikopoulos, G.C. Koltsakis, P.A.
Konstantinidis, E. Volpi, and E. Leveroni. Transient modeling of 3-way
catalytic converters. SAE Paper No. 940934, 1994.
[76] J.C. Peyton Jones, R.A. Jackson, J.B. Roberts, and P. Bernard. A sim-
plied model for the dynamics of a three-way catalytic converter. SAE
Paper No. 2000-01-0652, 2000.
[77] J.C. Peyton Jones, J.B. Roberts, J. Pen, and R.A. Jackson. Modeling
the transient characteristics of a three-way catalyst. SAE Paper No.
1999-01-0460, 1999.
[78] E.N. Pistikopoulos, V. Dua, N.A. Bozinis, A. Bemporad, and M. Morari.
On-line optimization via o-line parametric optimization tools. Comput-
ers and Chemical Engineering 24, pages 183188, 2000.
[79] R.M. Sanner and J.J.E. Slotine. Gaussian networks for direct adaptive
control. IEEE Trans. on Neural Networks 3, pages 837863, 1992.
[80] R.A. Searles. Auto emissions after 2000: The challenge for the catalyst
industry. Stud. Surf. Sc. Catal. 116, pages 2332, 1998.
[81] E. Shafai, C. Roduner, and H.P. Geering. Indirect adaptive control of a
three-way catalyst. SAE Paper No. 961038, 1996.
[82] M. Shelef and R.W. McCabe. Twenty-ve years after introduction of
automotive catalysts: what next? Catalysis Today 62, pages 3550,
2000.
[83] H. Shinjoh, H. Muraki, and Y. Fujitani. Periodic operation eects on au-
tomotive noble metal catalysts - reaction analysis of binary gas systems.
Stud. Surf. Sc. Catal. 30, pages 187198, 1987.
[84] P.L. Silverston. Automotive exhaust catalysis under periodic operation.
Catalysis Today 25, pages 175195, 1995.
[85] M.R. Simons, E. Shafai, and H.P. Geering. On-line identication scheme
for various wall-wetting models. SAE Paper No. 980793, 1998.
[86] S. Tagliaferri, R.A. Koeppel, and A. Baiker. Behavior of non-promoted
and ceria-promoted Pt/Rh and Pd/Rh three-way catalysts under steady
state and dynamic operation of hybrid vehicles. Ind. Eng. Chem. Res.
38, pages 108117, 1999.
[87] S. Takahashi and T. Sekozawa. Air-fuel ratio cotrol in gasoline engines
based on state estimation and prediction using dynamic models. Proc.
IEEE, pages 217222, 1995.
170 Bibliography
[88] G.I. Taylor. Dispersion of soluble matter in solvent owing through a
tube. Proc. Roy. Soc. A219, page 186, 1953.
[89] K.C. Taylor and R.M. Sinkevitch. Behaviour of automobile exhaust cat-
alysts with cycled feed streams. Ind. Eng. Chem. Prod. Res. Dev. 22,
pages 4551, 1983.
[90] J.R. Theis. An engine test to measure the oxygen storage capacity of a
catalyst. SAE Paper No. 961900, 1996.
[91] J.A. Thomas, R.E. Soltis, and W.E. Leisenring. Laboratory and engine
studies of the eect of NOx on the response of heated exhaust gas oxygen
sensors. SAE Paper No. 1999-01-1079, 1999.
[92] A. Trovarelli. Ceria: catalytic properties of ceria and CeO
2
-containing
materials. Cat. Rev. Sci. Eng. 38, pages 439520, 1996.
[93] R.C. Turin, E.G.B. Casartelli, and H.P. Geering. A new model for fuel
supply dynamics in an SI engine. SAE Paper No. 940208, 1994.
[94] R.C. Turin and H.P. Geering. Model-based adaptive fuel control in an SI
engine. SAE Paper No. 940374, 1994.
[95] M.V. Twigg. Critical topics in exhaust gas aftertreatment. Platinum
Metals Review, 45, No.4, pages 176178, 2001.
[96] E.R. van Selow. Design of a monolithic reactor for automobile exhaust
gas conversion with improved light-o. Graduate Report UCB-ITS-PRR-
92-9, Institute for Continuing Education, Eindhoven University of Tech-
nology, 1996.
[97] M. Vidyasagar. Nonlinear System Analysis. Prentice Hall Inc., 1993.
[98] B.I. Whittington, C.J. Jiang, and D.L. Trimm. Vehicle exhaust catalysis:
I. the relative importance of catalytic oxidation, steam reforming and
water-gas shift reactions. Catalysis Today 26, pages 4145, 1995.
[99] L.C. Young and B.A. Finalayson. Mathematical models of the mono-
lith catalytic converter: Part I Development of model and application of
orthogonal collocation. AIChE Journal 22, pages 331342, 1976.
[100] L.C. Young and B.A. Finalayson. Mathematical models of the monolith
catalytic converter: Part II Application to automotive exhaust. AIChE
Journal 22, pages 343353, 1976.
[101] K. Zygourakis. Transient operation of monolith catalytic converters: A
two-dimensional reactor model and the eects of radially nonuniform ow
distributions. Chem. Eng. Sc. 44, No. 9, pages 20752086, 1989.
Samenvatting
Toegenomen zorg over de lucht vervuiling door autos heeft in de laatste 30
jaar tot strenge emissie standaarden geleid. Het onderwerp van het onderzoek
dat in dit proefschrift gepresenteerd wordt is het ontwerpen van nieuwe regel-
strategieen voor drie-weg katalysatoren, die in autos worden ingebouwd, om
te voldoen aan de strenge emissie standaarden. Specieker, het doel is om een
modelgebaseerde regel strategie te ontwikkelen, die in staat is om de emissies
tijdens een hoog dynamische operatie van het proces te verminderen. Zon
geval van dynamische operatie is bijvoorbeeld het rijden in de stad. Mogelijke
verbetering van de katalysator light-o is ook onderzocht. De hoofdcontributie
van dit proefschrift is de ontwikkeling van een modelgebaseerde regelaar op
basis van informatie verkregen uit een rigoureuze, gedetailleerde proces model.
Het onderzoek omvat drie hoofddelen: de ontwikkeling van het rigoureuze,
gedetailleerde proces model van de katalysator; de ontwikkeling van het regel-
gericht model van de katalysator en verbinding daarvan met het verbrand-
ingsmotor model; de ontwikkeling en beproeving van de nieuw modelgebaseerde
regelaar door simulaties alsmede experimenten op een motorproefstand.
De ontwikkeling van het rigoureuze model voor een katalysator is gebaseerd
op kinetische modellen van reacties die in de katalysator plaatsvinden. Het
volledige model is verkregen door de toevoeging van de geschikte vergelijkingen
voor massa overdracht en energie uitwisseling. De voorspelling van het model
is vergeleken met experimentele data. Mogelijke verbetering van de light-o
van de katalysator door de toepassing van oscillerende voeding (oscillaties van
het ingang lambda signaal) en lucht inspuiting in de uitlaat manifold van de
motor is bestudeerd. De conclusie is dat een light-o verbetering mogelijk is
als het proces op de juiste manier bestuurd wordt.
Na de katalysator light-o is de zuurstof opslag op cerium het belangri-
jkste dynamische verschijnsel. Cerium is opgenomen in de washcoat van de
katalysator. Als dit proces goed geregeld wordt ontstaat een extra buer zodat
een korte excursie van de lucht/brandstof verhouding in de motor is toeges-
taan. Anders veroorzaakt zon excursie, die onvermijdelijk is tijdens het nor-
male dynamische operationele regime van de motor, emissies. Het doel van de
katalysator regelaar is om een optimale zuurstof opslag bedekking te bepalen
en bovendien optimale trajectorien te vinden om deze bedekking te bereiken.
Het is belangrijk dat het systeem een snelle responsie met lage uitlaat emissies
heeft. Om de informatie die van het model afkomstig zijn in de regelaar te kun-
nen gebruiken, moet het rigoureuze model gereduceerd worden. Een vereen-
voudigd, regelgericht model dat on-line de zuurstof opslag bedekking voorspelt
is ontwikkeld. Het model is nietlinear met een toestand, de zuurstof opslag be-
dekkingsgraad. Soms wordt een betere model voorspelling bereikt als het model
172 Samenvatting
twee toestanden bevat: de zuurstof opslag bedekkingsgraad op het oppervlak
en in de bulk van ceria. De parameters van het model kunnen automatisch
geschat worden via een algoritme dat gebruik maakt van stap responsies van
de katalysator. Ingang voor het algoritme zijn gemeten ingangs en uitgangs
lambda signalen. Het regelgerichte model is bepaald in een werkpunt. Om het-
zelfde model in een breder, niet-lineair, werkgebied te kunnen gebruiken is een
speciale extrapolatie procedure ontwikkeld op basis van het rigoureuze model.
Op deze wijze kan de modelschattingsprocedure, die tot behoorlijke uitlaate-
missie leidt en die niet tijdens de standaard operatie van het systeem uitgevoerd
kan worden, korter en dus meer ecient worden. De voorspelling van het regel-
gerichte model is vergeleken met het rigoureuze model en de afwijkingen zijn
gering.
Het model wordt door de regelaar in de motor management systeem ge-
bruikt om de zuurstof opslag bedekking, die niet gemeten kan worden, te
schatten. De regelaar is een analytische benadering van de ontwikkelde Model
Predictive Controller. De ontwikkelde Model Predictive Controller is in staat
om de proces informatie afkomstig van het model te gebruiken en het optimale
regelsysteem gedrag en de gestelde doelen te bereiken. Om deze regelaar toe te
kunnen passen moet echter een optimalisatie probleem binnen de sample tijd
van de besturing computer worden opgelost. Wegens beperkt rekenvermogen
van het motor management systeem kan zon probleem niet on-line opgelost
worden. Daarom zijn de optimalisatie problemen voor het hele proces werkge-
bied o-line opgelost. De resultaten worden gebruikt om een eenvoudig neuraal
netwerk te trainen. Het netwerk simuleert de Model Predictive Controller.
De regelaar is getest met behulp van simulaties op het rigoureuze model
alsmede met experimenten op een motor proefstand. De uitgevoerde testen
hebben een hoog dynamische operatie van het systeem. Zulke omstandigheden,
met name het rijden in de stad, lijden tot de hoogste uitlaatemissies. Door het
correct gebruik van de model informatie, leidt de nieuwe regelaar tot signicante
reductie van de uitlaatemissies onder de bovengenoemde omstandigheden.
Sazetak
Povecana zabrinutost zbog automobilskog zagadjivanja u posljednjih 30 godina
dovela je do vrlo striktnih standarda za automobilske stetne ispusne plinove.
Predmet istrazivanja prikazanog u ovoj doktorskoj disertaciji bio je razvoj novih
regulacijskih strategija za katalizatore ugradjene u osobne automobile sa ben-
zinskim motorima, pomocu kojih ce novi, ultra niski, standardi za automobilske
ispusne plinove biti zadovoljeni. Tocnije, cilj je bio razvoj regulacijskog sustava
temeljenog na modelu procesa koji doprinosi smanjenju stetnih ispusnih plinova
za vrijeme dinamicki vrlo intenzivne operacije sustava, na primjer za vrijeme
gradske voznje. Moguce ubrzavanje ukljucenja katalizatora u rad (smanjenje
temperature katalizatora potrebne za visoku konverziju) takodjer je razma-
trano. Glavni doprinos ove disertacije je razvoj sustava regulacije temeljenog
na pojednostavljenom modelu procesa koji je razvijen na osnovi detaljnog,
zikalnog modela procesa.
Istrazivanje je bilo podijeljeno u tri glavna dijela: razvoj detaljnog, zikalnog
modela procesa; razvoj pojednostavljenog modela procesa primjerenog regu-
laciji i njegovo povezivanje s modelom motora; razvoj i testiranje novog regu-
latora temeljenog na modelu procesa kroz simulacije i eksperimentalno.
Razvoj detaljnog modela katalizatora temeljen je na kemijskim kinetickim
modelima pojedinih reakcija koje se odvijaju unutar katalizatora. Dodava-
njem odgovarajucih jednadzbi za maseni i toplinski transfer unutar reaktora,
dobiven je kompletni model katalizatora. Odzivi (predvidjanja) modela uspo-
redjeni su sa eksperimentalno mjerenim vrijednostima. Model je upotrijebljen
za ispitivanje moguceg ubrzanja ukljucenja katalizatora pomocu oscilirajucih
koncentracija ulaznih plinova (osciliranje ulaznog lambda signala) i dodatnog
ubrizgavanja zraka u ispusnoj komori motora. Pokazano je da je poboljsanje
moguce ako je proces reguliran u odredjenim radnim uvjetima.
Nakon ostvarenja potrebne temperature za ukljucenje katalizatora u rad,
osnovni fenomen koji diktira dinamiku procesa je prihvacanje i otpustanje
kisika od strane cerija. Cerij je jedan od kemijskih elemenata smjestenih na
stjenkama katalizatora (eng. washcoat). Ako se navedeni proces dobro regulira,
stvara se dodatni buer koji dozvoljava privremena odstupanja omjera zraka
i goriva u motoru od stohiometrijskog omjera bez veceg utjecaja na stetne
ispusne plinove. Regulator ima za cilj odrediti optimalnu kolicinu pohran-
jenog kisika i trajektorije upravljacke velicine (omjer zraka i goriva) pomocu
kojih se ta kolicina postize. Zahtjeva se brz odziv procesa uz minimalne stetne
ispusne plinove. Da bi se omogucilo koristenje modela procesa za regulaciju, de-
taljni model mora se reducirati. Razvijen je pojednostavljeni model prikladan
za regulaciju koji u realnom vremenu predvidja kolicinu pohranjenog kisika
u katalizatoru. Model je nelinearan i sadrzi jednu varijablu stanja (kolicina
174 Sazetak
pohranjenog kisika). U nekim ekperimentima pokazalo se da model sa dvije
varijable stanja, u kojem se radi eksplicitna razlika izmedju kisika pohranjenog
na povrsini i kisika u unutrasnjosti sloja cerij oksida, moze dovesti do boljih
rezultata. Model se automatski podesava algoritmom na osnovi mjerenih om-
jera zraka i goriva (lambda vrijednosti) ispred i iza katalizatora, za vrijeme
odziva na step pobudu. Spoznaje o nelinearnoj dinamici procesa temeljene na
detaljnom modelu procesa dovele su do algoritma kojim se parametri pojedno-
stavljenog modela dobiveni u jednoj radnoj tocki podesavaju pri ekstrapolaciji
modela u druge radne tocke. Na taj nacin skracena je procedura estimacije
parametara procesa koja dovodi do povecanja koncentracije stetnih ispusnih
plinova i cesto ne moze biti provedena za vrijeme standardne operacije procesa.
Odzivi pojednostavljenog modela usporedjeni su sa odzivima detaljnog modela
te su se razlike pokazale zanemarivima.
Model je upotrijebljen u regulatoru kao programski sensor koji predvidja
nemjerljivu kolicinu kisika pohranjenu u katalizatoru. Kao regulator upotrije-
bljena je analiticka aproksimacija MPC-a (Model Predictive Controller). Razvi-
jeni MPC koristi informacije iz modela za ostvarivanje optimalnog ponasanja
regulacijskog kruga koje zadovoljava postavljene ciljeve regulacije. Takav re-
gulator zahtijeva rjesavanje kompleksnog optimizacijskog problema u proces-
nom racunalu automobila za vrijeme svakog intervala diskretizacije, sto se u
danasnjim racunalima ne moze postici. Zbog toga se gore navedeni problem
rjesava o-line za sve ocekivane radne tocke procesa. Dobivena rjesenja ko-
riste se za ucenje jednostavne neuronske mreze koja emulira MPC u realnom
vremenu.
Regulator je testiran u simulacijama koristeci detaljni model procesa, kao i
eksperimentalno na dinamometarskom sustavu za terecenje motora s unutarn-
jim izgaranjem. Obavljeni testovi su simulirali vrlo dinamicnu operaciju sis-
tema koja dovodi do povecanih stetnih ispusnih plinova (gradska voznja). Zbog
odgovarajuce upotrebe informacija proizaslih iz modela, razvijeni regulator
dovodi do smanjene emisije stetnih ispusnih plinova pri navedenim uvjetima.
Curriculum Vitae
Mario Balenovic was born in Sisak, Croatia, in 1974. After nishing Mathe-
matical High School in 1992 (rst three years in Sisak, the senior year as an
exchange student in Knoxville, TN, USA) he enrolled the Faculty of Electrical
Engineering and Computing in Zagreb, Croatia. He graduated in 1997 with
the emphasis on scientic research under supervision of dr.sc. Zdenko Kovacic
at the Department of Control and Computer Engineering in Automation. The
title of the graduation thesis was Application of a reference model and a sen-
sitivity model for learning (adjustment) of fuzzy controller parameters. After
graduation he worked for two months as a research assistant at the same de-
partment.
From 1997 until 2001 he was a Ph.D. student at Control Systems Group,
Department of Electrical Engineering, Eindhoven University of Technology.
The work performed during this period has led to this thesis.

S-ar putea să vă placă și