10 1 1 454 702

Loughborough University
Institutional Repository
Model-based approach to
plant-wide economic
control of fluid catalytic
cracking unit
This item was submitted to Loughborough Universitys Institutional Repository
by the/an author.
Additional Information:
A Doctoral Thesis. Submitted in partial fulfillment of the requirements
for the award of Doctor of Philosophy of Loughborough University.
Metadata Record: https://dspace.lboro.ac.uk/2134/8566

c Redah Mousa Alsabei
Publisher:
Please cite the published version.
This item was submitted to Loughboroughs Institutional Repository

(https://dspace.lboro.ac.uk/) by the author and is made available under the
following Creative Commons Licence conditions.
For the full text of this licence, please go to:

http://creativecommons.org/licenses/by-nc-nd/2.5/
MODEL BASED APPROACH FOR THE PLANTWIDE ECONOMIC CONTROL OF FLUID

CATALYTIC CRACKING UNIT
by
REDAH MOUSA ALSABEI
A Doctoral Thesis submitted in partial fulfilment of the requirements

for the award of the degree of Doctor of Philosophy in Chemical
Engineering
Redah Alsabei (2011)
Abstract
ABSTRACT
Fluid catalytic cracking (FCC) is one of the most important processes in the petroleum refining
industry for the conversion of heavy gasoil to gasoline and diesel. Furthermore, valuable gases such
as ethylene, propylene and isobutylene are produced. The performance of the FCC units plays a
major role on the overall economics of refinery plants. Any improvement in operation or control of
FCC units will result in dramatic economic benefits. Present studies are concerned with the general
behaviour of the industrial FCC plant, and have dealt with the modelling of the FCC units, which
are very useful in elucidating the main characteristics of these systems for better design, operation,
and control. Traditional control theory is no longer suitable for the increasingly sophisticated
operating conditions and product specifications of the FCC unit. Due to the large economic
benefits, these trends make the process control more challenging. There is now strong demand for
advanced control strategies with higher quality to meet the challenges imposed by the growing
technological and market competition.
According to these highlights, the thesis objectives were to develop a new mathematical model for
the FCC process, which was used to study the dynamic behaviour of the process and to demonstrate
the benefits of the advanced control (particularly Model Predictive Control based on the nonlinear
process model) for the FCC unit. The model describes the seven main sections of the entire FCC
unit: (1) the feed and preheating system, (2) reactor, (3) regenerator, (4) air blower, (5) wet gas
compressor, (6) catalyst circulation lines and (7) main fractionators.
The novelty of the developed model consists in that besides the complex dynamics of the reactorregenerator system, it includes the dynamic model of the fractionator, as well as a new five lump
kinetic model for the riser, which incorporates the temperature effect on the reaction kinetics;
hence, it is able to predict the final production rate of the main products (gasoline and diesel), and
can be used to analyze the effect of changing process conditions on the product distribution. The
FCC unit model has been developed incorporating the temperature effect on reactor kinetics
reference construction and operation data from an industrial unit. The resulting global model of the
FCC unit is described by a complex system of partial-differential-equations, which was solved by
discretising the kinetic models in the riser and regenerator on a fixed grid along the height of the
units, using finite differences. The resulting model is a high order DAE, with 942 ODEs (142 from
material and energy balances and 800 resulting from the discretisation of the kinetic models).
The model offers the possibility of investigating the way that advanced control strategies can be
implemented, while also ensuring that the operation of the unit is environmentally safe. All the
investigated disturbances showed considerable influence on the products composition. Taking into
account the very high volume production of an industrial FCC unit, these disturbances can have a
significant economic impact. The fresh feed coke formation factor is one of the most important
disturbances analysed. It shows significant effect on the process variables. The objective regarding
the control of the unit has to consider not only to improve productivity by increasing the reaction
temperature, but also to assure that the operation of the unit is environmentally safe, by keeping the
concentration of CO in the stack gas below a certain limit.
The model was used to investigate different control input-output pairing using classical
controllability analysis based on relative gain array (RGA). Several multi-loop control schemes
Abstract
were first investigated by implementing advanced PID control using anti-windup. A tuning
approach for the simultaneous tuning of multiple interacting PID controllers was proposed using a
genetic algorithm based nonlinear optimisation approach. Linear model predictive control (LMPC)
was investigated as a potential multi-variate control scheme applicable for the FCCU, using
classical square as well as novel non-square control structures. The analysis of the LMPC control
performance highlighted that although the multivariate nature of the MPC approach using
manipulated and controlled outputs which satisfy controllability criteria based on RGA analysis can
enhance the control performance, by decreasing the coupling between the individual low level
control loops operated by the higher level MPC. However the limitations of using the linear model
in the MPC scheme were also highlighted and hence a nonlinear model based predictive control
scheme was developed and evaluated.
Results demonstrate that using modern nonlinear model predictive control (NMPC) approaches
based on state-of-the-art optimization algorithms and software the advanced control of complex
chemical processes, such as the FCCU can be brought into the realm of possibility. The proposed
nonlinear control scheme was applied to the control of the highly nonlinear FCC process.
Simulations showed that the non-linear model-based MPC is highly effective, and coped well with
the complexity and nonlinearities of the FCC process. Both NMPC and LMPC were superior to
PID control, and it was shown that the NMPC provided better performance compared to LMPC. A
novel economic criteria-based NMPC (ENMPC) was proposed and demonstrated the benefits of
controlling the plant based on economic criteria rather than following predetermined setpoint
trajectories. The new concept of differentiating between beneficial (good) and harmful (bad)
disturbances was introduced and it was shown that regulating the plant at it predetermined
operating conditions when beneficial disturbances occur can actually decrease control performance.
The novel ENMPC scheme is inherently able to differentiate between good and bad disturbances
and exploit potential beneficial disturbances continuously adapting the plant operating conditions to
achieve economic optimisation and satisfy environmental and operating constraints.
These results have demonstrated that the control structures proposed may be applied in industry in
the form of a new scheme for controlling highly complex chemical processes with significant
economic benefits.
ii
Acknowledgments
Acknowledgments
First and foremost, I thank Almighty Allah (God) for blessing me with success and prosperity. I
would like to thank all the people who have supported me in various ways throughout the period of
this doctoral research. In particular, I would like to thank my supervisors, Prof. Zoltan Nagy and
Prof. Nassehi for their enthusiasm, guidance and continuous encouragement throughout the
research. Special recognition is also paid to their incredible patience, their many fruitful
discussions and excellent ideas.
I would like to take this golden opportunity to show my appreciation, gratitude and deep thanks to
all those people who helped me in carrying out my research and provided all kinds of support in the
Department of Chemical Engineering; from the administrative staff: Yasmin, Anna, Liz, Ann, Janey
and Becky; from support and technical staff Paul and Tony.
I would like to express sincere thanks is extended to my friends in S158: Hakem, Fahd, Saleemi, Dr.
Bahareh, Dr. Matthew and Alhajji for their support and encouragement.
Thanks and gratitude to my parents and my parents-in-law, who have been the greatest driving
force behind my PhD, and I absolutely could not have finished without their support, love and
prayers. To all my family; my brothers and sisters you have inspired me to go far and get the PhD.
Last but not least, I would like to extend my sincere gratitude to my wife, my daughters and my son,
you have always there for me and there is no way I can repay you, my thesis dedication is the
smallest gesture I can offer.
iii
Table of Contents
Contents
ABSTRACT ......................................................................................................................................... i
Acknowledgments .............................................................................................................................. iii
Contents ............................................................................................................................................. iv
List of Figures .................................................................................................................................. viii
List of Tables ................................................................................................................................... xiv
List of Nomenclature ........................................................................................................................ xv
CHAPTER 1: INTRODUCTION ....................................................................................................... 1
1.1
Background ......................................................................................................................... 1
1.2
Research aim and objectives ............................................................................................... 4
1.3
Main contribution of the thesis ........................................................................................... 5
1.4
Thesis structure ................................................................................................................... 6
CHAPTER 2: LITERATURE REVIEW ............................................................................................ 8

2.1
Fundamentals of crude oil processing ................................................................................. 8
2.1.1
2.2
Mathematical modelling concept ...................................................................................... 18
2.2.1
2.3
Process technology .................................................................................................... 10
FCC cracking kinetics modelling .............................................................................. 20
Introduction to advanced chemical process control .......................................................... 28
2.3.1
Why advanced process control? ................................................................................ 28
2.3.2
Concepts of process control ...................................................................................... 30
2.3.3
History of automatic control ..................................................................................... 31
2.3.4
Economical importance of process control ............................................................... 34
2.3.5
What is modern control theory? ............................................................................ 36
2.3.6
Advanced process control techniques ....................................................................... 37
2.3.7
Historical development of LMPC ............................................................................. 40
iv
Table of Contents
2.3.8
Industrial MPC development .................................................................................... 42
2.3.9
Unconstrained Linear Model Predictive Control (LMPC) with step response and
impulse response models........................................................................................... 46
2.3.10
2.4
Interaction analysis and multiple single-loop designs....................................................... 56
2.4.1
2.5
General tuning guidelines of MPC ............................................................................ 54
The Relative Gain Array (RGA) ............................................................................... 56
Nonlinear Model Predictive Control (NMPC) and industrial applications ....................... 59
2.5.1
Introduction ............................................................................................................... 59
2.5.2
Generic NMPC problem formulation ........................................................................ 63
2.5.3
Industrial implementation of NMPC ......................................................................... 64
2.5.4
Challenges in industrial NMPC ................................................................................ 68
2.5.5
First principle (analytical) model based NMPC ........................................................ 72
2.5.6
Nonlinear model predictive control with guaranteed stability .................................. 74
2.6
Conclusions ....................................................................................................................... 75
Chapter 3: MATHEMATICAL MODELING OF A FLUID CATALYTIC PROCESS .................. 77

3.1
Analytical mathematical model of the FCCU ................................................................... 77
3.1.1
Feed system model .................................................................................................... 79
3.1.2
Preheat system model ................................................................................................ 80
3.1.3
The reactor stripper model ........................................................................................ 86
3.1.4
Regenerator model .................................................................................................... 88
3.1.5
Air blower model ...................................................................................................... 94
3.1.6
Model of catalyst circulation lines ............................................................................ 95
3.1.7
Main fractionator model ............................................................................................ 95
3.1.8
Wet gas compressor model ....................................................................................... 98
3.2
Conclusions ..................................................................................................................... 100
Table of Contents
CHAPTER 4: Development of a Dynamic Process Simulator For The FCCU .............................. 102
4.1
Considerations of process operation ............................................................................... 102
4.1.1
4.2
Reactor-regenerator section..................................................................................... 103
Implementations and simulation procedure .................................................................... 110
4.2.1
Open loop simulations for the investigations of the dynamic behaviour of the FCCU.
................................................................................................................................. 111
4.3
Conclusions ..................................................................................................................... 123
CHAPTER 5: ADVANCED PID CONTROL OF THE FCCU PROCESS ................................... 124

5.1
Introduction ..................................................................................................................... 124
5.2
Controllability analysis and control structure selection for the FCCU ........................... 124
5.2.1
General considerations for the control of the FCCU ............................................... 124
5.2.2
Objectives for regulatory control ............................................................................ 126
5.2.3
Control structure selection for the FCCU using RGA ............................................ 127
5.3
Advanced PID control of the FCCU ............................................................................... 134
5.3.1
Design of the PID control algorithm ....................................................................... 134
5.3.2
Constrained optimization based MIMO PID tuning ............................................... 135
5.3.3
Multivariable PID control of the FCCU .................................................................. 139
5.4
Conclusions ..................................................................................................................... 151
CHAPTER 6: MULTIVARIABLE LINEAR AND NONLINEAR MPC OF THE FCCU ........... 153
6.1
Introduction ..................................................................................................................... 153
6.2
Linear MPC design and implementation......................................................................... 154
6.2.1
LMPC simulation using the linear model as plant .................................................. 154
6.2.2
Linear MPC applied to the nonlinear model-based plant simulator ........................ 165
6.3
Nonlinear model predictive control (NMPC) of the FCCU ............................................ 168
6.3.1
Efficient real-time iteration based NMPC approach ............................................... 168
6.3.2
NMPC FCCU simulation results ............................................................................. 173
vi
Table of Contents
6.4
Conclusions ..................................................................................................................... 177
CHAPTER 7: ECONOMIC OPTIMIZATION OF FCCU ............................................................. 179

7.1
Dynamic Real - time Optimization (DRTO) ................................................................... 179
7.1.1
7.2
Process optimizer for Economic NMPC of the FCCU ................................................... 183
7.2.1
7.3
Integrated Real-Time Optimization and Control..................................................... 182
One layer economic NMPC of the FCCU ............................................................... 184
Conclusions ..................................................................................................................... 208
CHAPTER 8: CONCLUSIONS AND FUTURE WORK .............................................................. 210

8.1
Conclusions ..................................................................................................................... 210
8.2
Recommendations for future work.................................................................................. 213
References ....................................................................................................................................... 215

Appendix ......................................................................................................................................... 240
List of publications ..................................................................................................................... 249
vii
List of Figures
List of Figures
Figure 1.1: Products from typical barrel of crude oil .......................................................................... 3
Figure 2.1: UOP type Fluid Catalytic Cracking Unit. ....................................................................... 12
Figure 2.2: 3-lump Model. ................................................................................................................ 23
Figure 2.3: 10-lump model................................................................................................................ 24
Figure 2.4: 4-lump Mode. ................................................................................................................. 24
Figure 2.5: 5-lump Model. ................................................................................................................ 25
Figure 2.6: 8-lump model.................................................................................................................. 26
Figure 2.7: 13-lump model................................................................................................................ 27
Figure 2.8: An example of price trends for real time minicomputers. .............................................. 29
Figure 2.9: Definition of input and output variables considered for control system design. ............ 30
Figure 2.10: Open-loop control system. ............................................................................................ 31
Figure 2.11: Closed-loop feedback control system. .......................................................................... 31
Figure 2.12: Evolution of industrial control technology. .................................................................. 32
Figure 2.13: Advanced computer control scheme. ............................................................................ 37
Figure 2.14: Common process characteristics, important in the choice of control strategy ............. 38
Figure 2.15: Classification of advanced control techniques. ............................................................ 40
Figure 2.16: The moving horizon approach of MPC techniques. ..................................................... 41
Figure 2.17: Plant with manipulated variable u, measured disturbance d and effect of unmeasured
disturbance w..................................................................................................................................... 50
Figure 2.18: Economical benefit given by the use of APC. .............................................................. 60
Figure 2.19: Typical operating regions with and without APC. ....................................................... 61
Figure 2.20: Distribution of MPC application versus the degree of process nonlinearity. ............... 62
Figure 2.21: Summary of industrial LMPC applications by areas. ................................................... 65
viii
List of Figures
Figure 2.22: Summary of NMPC applications by areas.................................................................... 67

Figure 2.23: Sequence of tasks involved in a generic NMPC application. ....................................... 69
Figure 2.24: Design effort of NMPC versus traditional control. ...................................................... 70
Figure 2.25: Classification of model types used in industrial MPC algorithms. ............................... 71
Figure 3.1: UOP fluid catalytic cracking unit. .................................................................................. 78
Figure 3.2: Feed system. ................................................................................................................... 80
Figure 3.3: Preheat systems............................................................................................................... 80
Figure 3.4: Schematic diagram of the industrial riser reactor and the stripper. ................................ 81
Figure 3.5: The five-lump kinetic model for cracking. ..................................................................... 83
Figure 3.6: Scheme of the FCCU stripper. ........................................................................................ 87
Figure 3.7: Regenerator phases. ........................................................................................................ 89
Figure 3.8: Scheme of the FCCU regenerator. .................................................................................. 90
Figure 3.9: Combustion air blower. .................................................................................................. 94
Figure 3.10: Wet gas compressor. ..................................................................................................... 98
Figure 4.1: SIMULINK block diagram used for the open-loop simulation. ................................... 112
Figure 4.2: Distribution of products and raw material, and raw material conversions in the riser at
different CTOs. ............................................................................................................................... 113
Figure 4.3: Simulation of FCCU dynamic response in the presence of step change in flow fresh feed
disturbance (F3 = 10% increase at t = 1.5 h). ................................................................................ 115
Figure 4.4: Simulation of FCCU dynamic response in the presence of step change in air flow rate
disturbance (Fair_c = 5% decrease at t = 1.5 h). ............................................................................ 116
disturbance (Fair_c = 10% decrease at t = 1.5 h). .......................................................................... 117
Figure 4.6: Simulation of FCCU dynamic response in the presence of step change in fresh feed
coke formation factor disturbance (f = 5% decrease at t = 1.5 h). .............................................. 119
ix
List of Figures
Figure 4.7: Simulation of FCCU dynamic response in the presence of step change in fresh feed
coke formation factor disturbance (f = 10% decrease at t = 1.5 h). .......................................... 121
Figure 4.8: Simulation of FCCU dynamic response in the presence of step change in slurry flow
recycle rate disturbance (F4 = 20% decrease at t = 1.5 h). ........................................................... 122
Figure 5.1: Hierarchical control system. ......................................................................................... 125
Figure 5.2: A block diagram of a PID controller. ........................................................................... 135
Figure 5.3: Fitness value for 2X2 PID controller system tuning..................................................... 138
Figure 5.4: Fitness value for 3X3 PID controller system tuning .................................................... 139
Figure 5.5: SIMULINK block diagram for the closed loop simulation with three PID
controller loops. .............................................................................................................................. 140
Figure 5.6: Performance of the regulatory controls for (Tr vs. F5), (Treg vs. svsc) and (cO2,sg vs. V14)
in the presence of step increase in the coking factor f +5%, at t = 1.5 h. .................................... 141
Figure 5.7: Dynamic behaviour of other uncontrolled process variables in the case of the
disturbance rejection scenario in Figure 5.6.................................................................................... 142
Figure 5.8: Performance of the servo control for (Tr vs. F5), (Treg vs. svsc) and (cO2,sg vs. V14) in the
presence of step increase, at t = 1.5 h. ............................................................................................. 143
Figure 5.9: Dynamic behaviour of other uncontrolled process variables in the case of the setpoint
tracking scenario in Figure 5.8. ....................................................................................................... 144
Figure 5.10: SIMULINK block diagram for the closed loop simulation with two PID
controller loops. .............................................................................................................................. 145
Figure 5.11: Model responses for the reactor Tr and regenerator cO2,sg temperatures by manipulated
svrgc & V14 with changes in the coking factor f +5%, at t = 1.5 h. ............................................... 146
Figure 5.12: Dynamic behaviour of other uncontrolled process variables in the case the scenario in
Figure 5.11. ..................................................................................................................................... 147
Figure 5.13: Model responses for the reactor Tr and regenerator Treg temperatures by manipulated
F5 & svsc with changes in the coking factor f +5%, at t = 1.5 h. .................................................. 148
List of Figures
Figure 5.14: Dynamic behaviour of other uncontrolled process variables the case scenario in
Figure 5.13. ..................................................................................................................................... 149
Figure 5.15: Performance of the servo control for (Tr vs. F5) and (Treg vs. svsc) in the presence of
step increase, at t = 1.5 h. ................................................................................................................ 150
Figure 5.16: Dynamic behaviour of other uncontrolled process variables the case scenario in
Figure 5.15. ..................................................................................................................................... 151
Figure 6.1: The 2x2 LMPC simulation results with the controlled variable response for setpoint
change for Tr and Treg at different times. ......................................................................................... 160
Figure 6.2: Manipulated variables for the 2x2 LMPC control scheme. .......................................... 161
change for Tr , Treg and Wr at different times................................................................................... 161
change for Tr , Treg and cO2,sg at different times. ............................................................................. 162
change for Tr , Treg , cO2,sg and Wr at different times. ...................................................................... 163
Figure 6.9: The non-square 4x2 MPC simulation results setpoint change for Tr and Treg at
different times. ................................................................................................................................ 164
Figure 6.10: Manipulated variables for the non-square 4x2 control scheme. ................................. 165
Figure 6.11: MPC block diagram for FCCU. .................................................................................. 166
Figure 6.12: Simulation results with the LMPC applied to the plant represented by the nonlinear
process model, in the case of setpoint changes in Tr and Treg , at t=1.5 h. ...................................... 167
Figure 6.13: Manipulated variable response for set point change in the output, for 2x2
control scheme. ............................................................................................................................... 167
xi
List of Figures
Figure 6.14: Illustration of the direct multiple shooting algorithm. ................................................ 171
Figure 6.15: Performance of the NMPC in the case of setpoint change with and
without anticipation......................................................................................................................... 176
Figure 6.16: CPU times required for the optimisation problems of the NMPC .............................. 176
Figure 6.17: Performance of the NMPC in the case of setpoint change with anticipation and a
measured disturbance (5 % step increase at t = 5000 s in f). ........................................................ 177
Figure 7.1: Illustration of D-RTO or Economically-Oriented NMPC ............................................ 180
Figure 7.2: Controlled and manipulated variables responses for NMPC in presence of f
disturbance (5 % step increase at t = 8000 s). ................................................................................. 188
Figure 7.3: Model outputs dynamic responses for NMPC (a) LCO, (b) LPG, and (c) Gasoline. ... 189
Figure 7.4: Model outputs dynamic responses for NMPC (a) cCOsg. and (b) cO2,sg. ...................... 190
Figure 7.5: Controlled and manipulated variables responses for DRTO/ENMPC of maximizing
LCO in presence of f disturbance (5 % step increase at t = 8000 s). ........................................... 191
Figure 7.6: (a) LCO yield when the economic objective is to maximize LCO and (b) LPG yield. 193
Figure 7.7: Model outputs dynamic responses for DRTO/NMPC (a) gasoline, (b) cCOsg and
(c) cO2,sg . ......................................................................................................................................... 194
Figure 7.8: Controlled and manipulated variables responses for the regulatory NMPC using the
setpoints determined by the ENMPC for the nominal raw material, in presence of f disturbance
(5 % step increase at t = 8000 s). .................................................................................................... 196
Figure 7.9: Model outputs dynamic responses for NMPC (a) LCO, (b) LPG and (c) gasoline. ..... 197
Figure 7.10: Model outputs dynamic responses for NMPC (a) cCOsg and (b) cO2,sg...................... 198
Figure 7.11: Controlled and manipulated variables responses for DRTO/ENMPC of maximizing
LPG in presence of f disturbance (5 % step increase at t = 8000 s)............................................. 199
Figure 7.12: (a) LPG yield when the economic objective is to maximize LPG, (b) LCO and (c)
gasoline. .......................................................................................................................................... 200
xii
List of Figures
Figure 7.13: Model outputs dynamic responses for DRTO/ENMPC of (a) cCOsg and (b) cO2,sg. .. 201
Figure 7.14: Controlled and manipulated variables responses for the DRTO/ENMPC of maximizing
LPG with environmental constraints in the presence of f disturbance (5 % step increase at
t = 8000 s). ...................................................................................................................................... 202
Figure 7.15: Model outputs dynamic responses for the DRTO/ENMPC (a) LPG, (b) LCO and (c)
gasoline. .......................................................................................................................................... 203
Figure 7.16: Model outputs dynamic responses for DRTO/NMPC of (a) cCOsg and (b) cO2,sg. .... 204
Figure 7.17: Controlled and manipulated variables responses for the DRTO/ENMPC of minimizing
LCO in the presence of f disturbance (5 % step increase at t = 8000 s). ..................................... 205
Figure 7.18: (a) LCO yield when the economic objective is to minimize LCO, DRTO/NMPC,
(b) Gasoline and (c) LPG. ............................................................................................................... 206
Figure 7.19: Model outputs dynamic responses for DRTO/NMPC of (a) cCOsg and (b) cO2,sg. .... 207
xiii
List of Tables
List of Tables
Table 2.1: Uses of FCC products. ..................................................................................................... 11
Table 2.2: Review of kinetic models used in FCCU with different lump components..................... 21
Table 2.3: Selected historical developments of control systems. ...................................................... 33
Table 2.4: Summary of the interpretation of the RGA. ..................................................................... 59
Table 2.5: NMPC companies and product names. ............................................................................ 65
Table 2.6: Comparison of industrial NMPC control technology. ..................................................... 66
Table 2.7: Approximate summary of NMPC applications by application areas. .............................. 67
Table 3.1: Construction and operation data. ..................................................................................... 99
Table 4.1: Process input and base case operating conditions with disturbances. ............................ 111
Table 5.1: Operating constraints and their relative importance (0: low; 10: high). ........................ 131
Table 5.2: Different control structure schemes for PID controller.................................................. 133
Table 5.3: RGA matrix of a sample 2X2 control scheme. .............................................................. 133
Table 5.4: RGA matrix of a sample 3x3 control scheme. ............................................................... 134
Table 5.5: Tuning parameters of the PID controllers. ..................................................................... 138
Table 5.6: Tested data for 3 x 3 PID controllers in MIMO loops parameters................................. 140
Table 5.7: Tested data for PID controllers in MIMO loops parameters.......................................... 145
Table 6.1: Input-output configurations tested with the MPC parameters used. .............................. 157
Table 6.2: RGA Matrix of 3X3 Control Scheme. ........................................................................... 159
Table 6.3: Tested scenarios for NMPC simulations. ....................................................................... 174
Table 7.1: Constrained and manipulated variables for application to FCC unit. ............................ 185
Table 7.2: Scenarios of the operation studies.................................................................................. 186
Table 7.3: Summary of results ........................................................................................................ 208
xiv
List of Nomenclature
LIST OF NOMENCLATURE
Alp
cross sectional area of lift pipe (ft2)
Areg
cross sectional area of regenerator (ft2)
Aris
cross sectional area of the reactor riser (ft2)
Asp
cross sectional area of catalyst circulation line (ft2)
Astripper
cross sectional area of the reactor stripper (ft2)
Asv
cross sectional area of regenerated/spent catalyst slide valve at completely open

position (in2)
a1
furnace heat loss parameter (B.t.u/s F)
a2
furnace heat loss parameter (B.t.u/s F)
aj
activity of the catalyst in reaction j
bottom product (kmol)
random fluctuating velocity of an individual particle, (m/s)
Ccat
coke fraction on catalyst at the reactor outlet (kg/kg)
cCOsg
concentration of carbon monoxide in stack gas (ppm)
CH
weight fraction of hydrogen in coke (lb/lb)
cO2,sg
concentration of oxygen in regenerator stack (mol%)
Cp
specific heat at constant pressure ([kJ/kg K)
cp(z)
average heat capacity (B.t.u/mol F)
cpair
heat capacity of air (B.t.u/mol F)
cpc
heat capacity of catalyst (B.t.u/mol F)
cpCO
heat capacity of carbon monoxide (B.t.u/mol F)
cpCO2
heat capacity of carbon dioxide (B.t.u/mol F)
cpfl
heat capacity of fresh liquid (B.t.u/mol F)
xv
cpf
heat capacity of fresh vapour (B.t.u/mol F)
cpH2O
heat capacity of steam (B.t.u/mol F)
cpN2
heat capacity of nitrogen (B.t.u/mol F)
cpO2
heat capacity of oxygen (B.t.u/mol F)
cpsv
heat capacity of slurry (B.t.u/mol F)
Cps
concentration of the reactant lump in path p (%)
Crgc
weight fraction of coke on regenerated catalyst (lb/lb)
Crw
wet gas compressor compression ratio (lb/lb)
Csc
weight fraction of coke on spent catalyst (lb/lb)
C1
wet gas production constant (mol/lb)
C2
wet gas production constant (mol/lb)
cc
coke content on catalyst (wt%)
CTO
COR catalyst-to-oil ratio (gcat/gfeed)
distillate product flow (kmol/min)
deactivation function parameter
Eoil, inlet
elevation where oil enters the reactor riser (ft)
Etap
pressure tap elevation on standpipe (ft)
Ecf
activation energy of coke formation (kJ/kg)
Ep
activation energy for reaction path p, J/mol
feed rate (kmol/min)
Fair
air flow rate into regenerator (mol/s)
Fcat
flow rate of spent or regenerated catalyst (lb/s)
Fcc
catalyst flow in the regenerator (lb/s)
Fb
effect of feed type on coke production
xvi
Fcoke
production of coke in reactor riser (lb/s)
Ff
flow of feedstocks entering the reactor (kg/s)
FH
burning rate of hydrogen (lb/s)
fof
overflow factors
fpp(x)
nonlinear valve flow rate function
frgc
force exerted by regenerated catalyst (lbf)
Frgc
flow rate of regenerated catalyst (lb/s)
Fs
flow of catalyst entering in reactor (kg/s)
fsc
force exerted by spent catalyst (lbf)
Fsc
flow rate of spent catalyst (lb/s)
Fsg
stack gas flow (mol/s)
Fsucn, comb
combustion air blower inlet suction flow (ICFM)
Fsucn, wg
wet gas compressor inlet suction flow (ICFM)
FT
air flow rate into regenerator (lb/s)
Ftset
set point to total air flow controller (lb/s)
fubend, rgc
regenerated catalyst friction factors (lb s/ft2)
fubens,sc
spent catalyst friction factors (lb s/ft2)
FV6
flow through combustion air blower suction valve (lb/s)
FV7
flow through combustion air blower vent valve (lb/s)
FV11
flow through wet gas compressor suction valve (lb/s)
FV12
flow through wet gas flare valve (lb/s)
FV13
flow through wet gas compressor anti-surge valve (lb/s)
Fwg
flow of wash oil to reactor riser (lb/s)
F1
flow of wash oil to reactor riser (lb/s)
F1set
set point to wash oil flow controller (lb/s)
xvii
F2
flow of diesel oil to reactor riser (lb/s)
F2set
set point to diesel oil flow controller (lb/s)
F3
flow of fresh feed to reactor riser (lb/s)
F3set
set point of fresh feed flow controller (lb/s)
F4
flow of slurry to reactor riser (lb/s)
F4set
set point of slurry flow controller (lb/s)
F5
flow of fuel to furnance (lb/s)
F5set
set point to wash oil flow controller (lb/s)
F6
combustion air blower throughput (lb/s)
F7
combustion air flow to the regenerator (lb/s)
F11
wet gas flow to the vapour recovery (lb/s)
acceleration due to gravity (ft/s2)
hris
height of reactor riser (ft)
Hwg
wet gas compressor head (psia)
flow coefficient for the slide valve
KK
light gas scaling constant
kcomb
combustion air blower discharge pipe flow resistance factor (lb/s psia)
kinetic constant of feedstocks cracking (s-1)
K2
kinetic constant of feedstocks cracking (s-1)
Kh
absorption coefficient of lump Fh
kij
rate constant in the four-lump model [m3/(kg s)]
ki
reaction rate constant (s-1)
k1
reaction rate constant for reaction I (s-1)
k2
reaction rate constant for the reaction II (s-1)
k3
reaction rate constant for the reaction III (mol s-1)
xviii
k6
combustion air blower suction valve flow rating (lb/s psia)
k7
combustion air blower vent valve flow rating (lb/s psia)
k11
wet gas compressor suction valve flow rating (lb/s psia)
k12
wet gas flare valve flow rating (lb/s psia)
k13
wet gas compressor anti-surge valve flow rating (lb/s psia)
k14
regenerator stack gas valve flow rating (lb/s psia)
kc
kinetic constant of coke formation in Kuriharas expression (s-1)
kp
reaction rate coefficient for reaction path p, m3/ s kgcat
kd
deactivation constant for conversion (wt%)
reflux (kmol/min)
LT
condensers liquid flow (kmol)
Lubend, rgc
length of regenerated catalyst u-bend (ft)
Lubend,s
length of spent catalyst u-bend (ft)
molecular mass of the lump (kg/kmol)
m1
polytrophic exponent
total liquid holdup (kmol)
M1
effective heat capacity of regenerator mass (B.t.u/ F)
Mcpeff
effective heat capacity of riser vessel and catalyst (B.t.u/ F)
Me
flow rate of entrained catalyst from dense bed into dilute phase (lb/s)
Mow
liquid holdup over weir (kmol)
Mrgc
inertial mass of regenerated catalyst (lb s2/ft)
Muw
liquid holdup under weir (kmol)
Msc
inertial mass of spent catalyst (lb s2/ft)
Mwi
molecular weight of species i (kg/kmol)
quantity of gas (mol)
xix
n1
exponential of the coke generation kinetics
n2
N*
integer value representing a constant for pressure drop on catalyst pipes
NC
number of components
NT
number of stages
NF
feed stage
Patm
atmospheric pressure (psia)
Pbase
combustion air blower base discharge pressure (psia)
Prb
pressure at bottom of reactor riser (psia)
Prgb
pressure at bottom of regenerator (psia)
Pvru
discharge pressure of wet gas compressor to vapour recovery unit (psia)
P1
combustion air blower suction pressure (psia)
P2
combustion air blower discharge pressure (psia)
P4
reactor pressure (psia)
P5
reactor fractionator pressure (psia)
P6
regenerator pressure (psia)
P7
wet gas compressor suction pressure (psia)
deactivation function parameterQair = enthalpy of air to regenerator (B.t.u/s)
Qc
total heat of burning carbon (B.t.u/s)
Qcatout
enthalpy of catalyst out of reactor riser (B.t.u/s)
Qcracking
heat generated from cracking (B.t.u/s)
Qe
total heat loss from regenerator to environment (B.t.u/s)
Qff
heat required to bring fresh feed to reactor riser temperature (B.t.u/s)
Qfg
enthalpy of outgoing regenerator stack gases (B.t.u/s)
Qfr
heat required to raise temperature of fresh feed (B.t.u/s)
xx
QH
enthalpy of hydrogen to regenerator (B.t.u/s)
Qin
enthalpy into regenerator, reactor (B.t.u/s)
Qloss
heat loss from furnance (B.t.u/s)
Qout
enthalpy out of regenerator, reactor (B.t.u/s)
Qrgc
enthalpy of regenerated catalyst (B.t.u/s)
Qsc
enthalpy of spent catalyst (B.t.u/s)
Qslurry
heat required to bring slurry to reactor riser (B.t.u/s)
qF
feed liquid fraction (kmol/kmol)
universal gas constant (ft3 psia/lbmol)
Rj
net production rate of component j/species l, (mol/s kg)
Ri
rate of reaction i in the kinetic model (s-1)
Rp
reaction rate for path p, (mol/s kg)
Rpm
net production rate of a lump (mol/kg s)
tc
residence time of catalyst in the reactor (s)
gas-solid phase temperature (K)
Tair
temperature of air entering regenerator (F)
Tatm
atmospheric temperature (F)
Tbase
base temperature (F)
Tbase,f
base temperature of reactor fresh feed (F)
Tcomb,d
combustion air blower discharge temperature (F)
Tcy
regenerator stack gas temperature at cyclone (K)
Tlm
furnance log mean temperature difference (F)
Tr
temperature of reactor riser (F)
Tref
base temperature for reactor riser energy balance (F)
Treg
temperature of regenerator bed (F)
xxi
Tsc
temperature of spent catalyst entering regenerator (F)
T1
temperature of fresh feed entering furnance (F)
T2
temperature of fresh feed entering reactor riser (F)
T3
furnance firebox temperature (F)
Uaf
furnance overall heat transfer coefficient (B.t.u/s)
boil up (kmol/min)
Vcomb,d
combustion air blower suction system volume (ft3)
Vcomb,s
combustion air blower suction system volume (ft3)
Vreg
regenerator volume occupied by gas (ft3)
vrgc
velocity of regenerated catalyst (ft/s)
vris
volumetric flow rate in reactor riser (ft/s)
vs
superficial velocity in regenerator (ft/s)
vsc
velocity of spent catalyst (ft/s)
V6
combustion air blower vent valve position
V7
combustion air blower vent valve position
V11
wet gas compressor suction valve position
V12
wet gas flare valve position
V13
wet gas compressor vent valve position
V14
stack valve position
catalyst hold-up (ton)
Wc
inventory of carbon in regenerator (lb)
Wr
inventory of catalyst in reactor (lb)
Wreg
inventory of catalyst in regenerator (lb)
Wris
inventory of catalyst in reactor riser (lb)
wtir
weight fraction of ten lump species (lb/lb)
xxii
WHSV
weight hourly space velocity (1/h)
XCO
molar ratio of CO to air (mol/mol)
XCOsg
molar ratio of CO to air in stack gas (mol/mol)
XCO
molar ratio of CO2 to air (mol/mol)
XCO2, sg
molar ratio of CO2 to air in stack gas (mol/mol)
XN2
molar ratio of N2 to air (mol)
XO2
molar ratio of O2 to air (mol/mol)
XO2,sg
molar ratio of O2 to air in stack gas (mol/mol)
Xv
volume fraction conversion
xi
column liquid composition (kmol/kmol)
yf
feedstocks mass fraction (kg/kg)
yg
gasoline mass fraction (kg/kg)
yi
column vapor composition (kmol/kmol)
yj
mass fraction for the jth component (kg/kg)
vertical position in regenerator (ft)
zbed
regenerator dense bed height (ft)
zcyc
height of cyclone inlet (ft)
zF
column feed mass fraction (kmol/kmol)
zlp
height of lift pipe discharge (ft)
ztop
O2, CO measured point (ft)
Vij
stoichiometric coefficient for the reaction from i to j lump determined by the

molecular weight of lumps i and j.
kij
kinetic constant for the reaction from i to j lump
weight hour space velocity
dz
dimensionless length
xxiii
the deactivation parameter for the cracking of i lump.
MWc
molecular weight of coke.
Greek notations
Dirac delta function
H crack
heat of cracking (B.t.u/s)
H fu
heat of combustion of furnance fuel (B.t.u/SCF)
HH
heat of combustion of hydrogen (B.t.u/lb)
H 1
heat of formation of CO (B.t.u/mol)
H 2
heat of formation of CO (B.t.u/mol)
Pfrac
pressure drop across reactor main fractionator (psia)
Pelb,rgc
pressure drop of the different elements of regenerated catalyst pipe (psia)
Pelb,sc
pressure drop of the different elements of spent catalyst pipe (psia)
PRR
differential pressure between regenerator and reactor (psia)
Psvrgc
pressure drop regenerated catalyst slide valve (psia)
Psvsc
pressure drop spent catalyst slide valve (psia)
Tstripper
temperature drop across reactor stripper (F)
effective void fraction in regenerator dense phase bed
apparent void fraction in regenerator dense phase bed
polytrophic efficiency
density (lb/ft3)
airg
density of air at regenerator conditions
xxiv
volume fraction of catalyst
density of catalyst in spent and regenerator catalyst circulation line (lb/ft3)
c,dilute
density of catalyst in dilute phase (lb/ft3)
c,dense
density of catalyst in dense phase (lb/ft3)
density of feedstocks (kg/m3)
density of exit gas (mol/ft3)
lift
density of catalyst in lift pipe (lb/ft3)
part
settled density of catalyst (lb/ft3)
ris
average density of material in reactor riser (lb/ft3)
vapour density of at reactor riser conditions (lb/ft3)
fb
furnance firebox time constant (s)
fill
riser fill time (s)
fo
furnance time constant (s)
catalyst residence time in riser (min)
effective coke factor for gas-oil feed (psif)
deactivation constant for coke formation (s-1)
crack ability factor for the feedstocks
relative volatilities of components
deactivation parameter in reaction j
fraction of original activity of the catalyst (dimensionless)
coke deactivation function
xxv
coke
activity function for coke formation
conv
activity function for conversion
residence time (s)
stoichiometric factor of a lump
sum of molar concentrations (ai: mol/g of gas) of all the gaseous lumps
xxvi
Chapter 1: Introduction
CHAPTER 1
INTRODUCTION
1.1 Background
Worldwide, the fluid catalytic cracker (FCC) is the workhorse of the modern refinery. Its function
is to convert heavy hydrocarbon petroleum fractions into a slate of more usable range of products,
and must adapt to seasonal, environmental and other changing demand patterns, such as for
gasoline, LPG and diesel. Furthermore, valuable gases such as ethylene, propylene and isobutylene
are produced. The performance of the FCC units plays a major role in the overall economics of
refinery plants (Alhumaizi and Elnashaie, 1997). Any improvement in operation or control of FCC
units will result in dramatic economic benefits. First commercialised over half a century ago, the
FCC is still evolving. Improvements in the technology, as well as changing feed stocks and product
requirements continue to drive this evolution. Control of the FCC has been, and continues to be, a
challenging and important problem. As will be seen, its steady state behaviour is highly nonlinear,
leading to multiple steady states and input multiplicities. In earlier years, before the development of
zeolite catalysts, the major control problem was one of stabilisation, of just keeping the unit
running. Later with zeolite catalysts, the emphasis shifted to increasing production rates in the face
of unit constraints and to handling heavier feeds (Arbel et al., 1995). New world trends in product
demands, and to meet more severe legislation about fuel compositions raised the significance of
controlling FCC product selectivity. The different product slates of the FCC process are the
consequence of the complex interplay between reactions, such as cracking, isomerisation, hydrogen
transfer, oligomerisation, etc. The complexity of gas oil mixtures, which are the typical FCC feeds,
makes it extremely difficult to characterise and describe the inherent kinetics at a molecular level.
Hence, one is forced to examine generalities rather than the details. One of the methods used to do
this, is to consider the behaviour of groups of compounds as a unit. In this way, similar components
are grouped into a few cuts or lumps. Therefore, the study of the reactions involved in the
catalytic cracking process has followed the lumping methodology (Serti-Bionda et al., 2010).
The FCC unit consists of two interconnected gas-solid fluidised bed reactors. The riser reactor,
where almost all the endothermic cracking reactions and coke deposition on the catalyst occur, and
the regenerator reactor, where air is used to burn off the coke accumulated on the catalyst. The heat
produced is carried by the catalyst from the regenerator to the reactor (Emad and Elnashaie, 1997).
Thus, in addition to reactivating the catalyst, the regenerator provides the heat required by the
endothermic cracking reactions. The region of economically attractive operational conditions is
determined by both the properties of the feed stocks, catalyst and the desired product distribution
requirements. In practice, the optimisation of the FCC unit to the desired range of products is
usually carried out by trial and error. The disadvantage of this approach is that the transition from
one state to the other must be gradual, and is not always successful, because of the complex
interactions between the two reactors. As a result, it could lead to loss of production and
consequently affect profits. Most of the economic gain from FCC control development has come
from the optimisation level, with the regulation system simply providing stable, responsive, and
safe operation. The problem is to find regulator schemes that are effective, economically justified,
related to existing practice, and able to provide an adequate operator interface when desired. Most
studies (Kurihara, 1967; Iscoll, 1970; Lee and Kugelman, 1973; Eng et al., 1974; Edwards and
Kim, 1988; McFarlane et al., 1990; Krishna and Parking, 1985; Elshishini and Elnashaie, 1990)
concerning FCC units have dealt with the process control based on a simplified reactor-regenerator
model, which in principle incorporates major observed dynamics. Any FCC control should
maintain a suitable reactor temperature distribution so as to achieve good product characteristics.
The regenerator temperature profile should also be bounded so as to prevent abnormal combustion
and excessive temperatures. At the same time, energy and material balances must be maintained
between the two parts of the unit (Lopez-Isunza, 1992).
The modelling of complex chemical systems in the simulation of process dynamics and control has
been motivated by the economic incentives for improvement of plant operation and plant design.
Presently, studies are concerned with the general behaviour of industrial FCC units; these research
efforts deal with the modelling of FCC units, which are very useful in elucidating the main
characteristics of these units for better design, operation, and control. Traditional control theory is
no longer suitable for the FCC units increasingly sophisticated operating conditions and product
specifications (Jia et al., 2003). Due to the large economic benefits, these trends make the process
control more challenging. There is now a strong demand for advanced control strategies with higher
control quality to meet the challenges imposed by the growing technological and market
competition (Liao, 2008).
Demand in the world oil markets is primarily for motor gasoline and other high quality fuels.
Gasoline and other high products can be sold at highest prices; therefore, heavy oils are less
valuable than the light fractions. Crude distillation separates crude oil into more useful fractions.
However, distillation alone cannot meet the demand for high quality fuels. Figure 1.1 shows the
range of products demanded by the market from a barrel of crude oil, compared with the products
produced by distillation alone. An important job of modern refineries is to convert oil from the
bottom of the barrel to gasoline and other marketable products. Since a typical FCC unit can
convert a large amount of feedstock into more valuable products, the overall economic benefits of a
refinery could be considerably increased if proper control and optimisation strategies are
implemented. But, analysis and control of FCC processes have been known as challenging
problems due to the following process characteristics; (1) very complicated and little known
hydrodynamics, (2) complex kinetics of both cracking and coke burning reactions, (3) strong
interactions between the reactor and the regenerator, and (4) many operating constraints. However,
the large throughput of FCCU, the change in operating conditions, and the substantial economic
benefits are the motivation behind this research (Han et al., 2000).
Figure 1.1: Products from typical barrel of crude oil
1.2 Research aim and objectives

The aim of this research project is to develop a mathematical model that can simulate the behaviour
of the FCC unit, which consists of feed and preheat system, reactor (riser and stripper), regenerator,
air blower, wet gas compressor catalyst circulation lines, and the main fractionators. The model will
be subsequently used in studies of control and economic optimisation. The developed model deals
with the complex dynamics of the reactor-regenerator system, and also includes the dynamic model
of the fractionator, as well as a detailed five lump kinetic model for the riser (namely: gas oil,
gasoline, diesel, LPG and coke). This model is able to predict and describe the compositions of the
final production rate, and the distribution of the main components in the final product. This allows
the estimation of economic factors, related to the operation of the FCCU.
Seven objectives have been identified that lead to a logical progression through the research:
a) to gain knowledge and understanding of FCC unit behaviour by developing a user-friendly,
process simulator using an object-oriented programming environment. The simulator can
be used to understand the process dynamics, and perform operator training, control
structure design, controller tuning, through a comprehensive literature review;
b) implement and evaluate the 5-lump kinetic model in the FCC global model, to simulate the
dynamic behaviour of open and closed loops using decoupled advanced PID control
algorithms;
c) design and evaluate control schemes for controlled and manipulated variables in order to
choose the best control pairs by using an analysis tool, known as relative gain array (RGA);
d) develop a PID tuning algorithm for a Multi Input Multi Output (MIMO) system by using
model based optimisation;
e) design and implement linear and nonlinear Model Predictive Control (MPC);
f) derive and evaluate different economic optimisation objectives; and
g) implement a hierarchical Real Time Optimisation (RTO) algorithm and perform evaluation
for different economic objectives.
1.3 Main contribution of the thesis

The main contributions of this thesis are summarised in the following lists:
a) An original mathematical model for the FCC process was developed based on the
momentum, mass, and energy dynamic balances. It incorporates process hydrodynamics,
heat transfer, mass transfer and catalytic cracking kinetics based on a lumping strategy,
which lumps molecules and reactions according by their boiling point and treated as
pseudo-components for a global description of the phenomena taking place in the reactor.
The process is multivariable, strongly nonlinear, highly interactive, and subject to many
operational, safety and environmental constraints, posing challenging control problems.
The model was implemented in the C programming language for efficient solution, and
compiled in Matlab/Simulink language programming, because it provides a convenient
graphical user interface. Moreover, the model has been used to study the dynamic
behaviour of the process, control the output variables and economic optimisation. The
global model of the FCCU is described by a complex system of partial-differentialequations, which was solved by discretising the kinetic models in the riser and regenerator
on a fixed grid along the height of the units, using finite differences. The resultant model is
described by a complex system of a higher order differential-algebraic-equation (DAE),
with 142 ODEs (from material and energy balances and 800 algebraic equations resulting
from the discretisation of the kinetic models).
b) A flexible process simulator was developed that can be used to show how the open and
closed loop control systems perform in the case of disturbances and model uncertainty. The
developed simulator enables engineering and technical personnel to carry out research on
the design, operation, performance, and development of a proper control system for a
modern catalytic cracking unit. It could also act as an efficient tool for training operating
personnel.
c) An automatic tool is proposed to tune PID controllers in a MIMO process based on a
Genetic Algorithm (GA). The tuning of several interacting controllers in complex industrial
plants is a challenge to process engineers and operators. The success of this task depends
on complete knowledge of plant behaviour and control requirements, which can present
strong interactions among variables, non-linearity and conflicting objectives. An advantage
of the proposed scheme is that a coupled MIMO process with several control loops can
have all the controllers tuned in a unified way as a full MIMO controller. In the
development phase, the problem of tuning n regulatory control loops was modelled as an
optimisation problem, where tuning one control loop may damage the performance of the
remaining ones. The solution of the multi-objective problem by the weighted sum approach
is possible, since each loop is locally evaluated by a function that considers the integral
squared error (ISE). It is worthwhile to note that this modelling is not restricted to PID
control. The objective function can be used as a basis for the design and tuning of general,
linear or non-linear multivariable controllers.
d) To introduce an analysis method based on the relative gain array (RGA) as a tool to help
select variable pairings for decentralised multivariable control structures.
e) To evaluate the performance of Model Predictive Control (MPC) based on linearised global
reactor-regenerator-main fractionator FCCU global model. Various square and non-square
control structures are investigated, and the tuning of the MPC is evaluated.
f) To develop an efficient real-time nonlinear model predictive control (NMPC) strategy,
based on efficient multiple shooting optimization and a real-time iteration scheme, and to
evaluate the performance of the proposed scheme in the case of the simulated complete
FCCU.
g) To develop a novel NMPC scheme based on the on line optimization of overall process
economics related objectives. To demonstrate the benefits of controlling the FCCU based
on real time economic decisions, and exploiting the beneficial effects of disturbances if
any, rather than simply tracking set point trajectories determined off line.
1.4 Thesis structure

An outline of the chapters in this thesis is as follows:
Chapter 1 gives an introduction to the problems in fluid catalytic cracking technology, together
with the recent improvements that maximise the productivity and product quality. The aims of the
current study, thesis contribution and structure of the thesis report are discussed.
Chapter 2 provides a review of relevant literature on chemical process control. The sub-sections
within this chapter consist of fundamentals of crude oil processing, introduction to the FCC and its
significance, process description, and FCC catalyst and process chemistry. The FCC unit is
discussed in its modelling and kinetics aspects. Moreover, various control tools and techniques are
discussed. The important information that can be obtained about control, and economic process
control is mentioned.
Chapter 3 presents the details of the FCC mathematical model, including the five lump kinetic
model for the riser section and the main systems: feed and preheat system, reactor (riser and
stripper), regenerator, air blower, wet gas compressor catalyst circulation lines and the main
fractionator.
Chapter 4 presents the development of a dynamic process simulator for an open-loop system with
a wide range of operating conditions and in the presence of disturbances. Also, sensitivity analyses
for composition of the products are presented. The economic aspects have been described by plant
gross profit and gasoline octane number, which are sensitive to the presence of disturbance.
Chapter 5 describes and discusses the implementation of a closed loop system using different
regulatory controller structures. A set of FCC unit dynamic simulations have been performed and
studied in response to disturbances.
Chapter 6 presents an overview of NMPC and LMPC techniques and their implementation on a
very complex, industrially relevant FCC process. Software packages for advanced control
simulations are studied and presented.
Chapter 7 introduces the novel NMPC approach for the hierarchical Real Time Optimisation
(RTO) algorithm and provides an evaluation for different economic objectives.
Chapter 8 concludes the thesis by summarising the results of all the work that has been presented,
and proposes areas of future work.
Chapter 2: Literature review
CHAPTER 2
LITERATURE REVIEW
2.1 Fundamentals of crude oil processing
The debate on fossil fuels and substitute energy sources is presently at a peak. In this respect, it can
be seen that there are significant efforts on the global scale to develop alternatives in the form of
sustainable energy sources. Yet, the currently existing alternatives to fossil fuels suffer from
drawbacks (Kirby, 2004), listed in the following:
a. Water power, hydroelectric or otherwise, is a clean, non-polluting source. However, it
relies on the presence of a sufficiently large water resource, and dam construction is not a
sustainable activity.
b. Another option involves utilising hydrogen, which is an abundant resource. Yet, currently,
requires large amounts of electrical energy for it to be generated from fossil fuels or water,
and is also problematic in its transport and storage.
c. The option of energy from the sun or wind has been subject to significant work, and as such
may meet a large part of the needs in some areas. However, this depends on climatic
conditions, and the day/night cycle, and as such availability is variable.
d. Development of cleaner derivative fuel technologies, which produce gas from coal before
burning to produce power, are said to result in less pollutants (E.ref.1).
e. The alternative of nuclear power has the benefit of zero emissions of pollutants. However,
it presents huge challenges in the safe disposal of radioactive waste.
f.
Biomass-derived bio-fuels rely on processing organic, vegetable matter. Such fuel is

blended with conventional fossil fuel-derived diesel or petrol at low concentrations. For
example, maximum 10% bio-fuel blended into petrol (E.ref. 2). A further problem lies in
the amount of resources diverted from food production to bio-fuel, which raises a number
of ethical questions.
g. The recovery of low-grade heat from ocean and geothermal sources offers a potential longterm solution through power generation using, for example, the Organic Rankine Cycle
(14% max. efficiency). Presently, such technology is in the early stages of development,
and has seen little implementation in practice.
Given these realities, it would be correct to say that there is still some time before alternative
technologies can play an effective role in meeting all current and future power generation needs.
Therefore, as yet, oil remains the prime energy source, providing humanity with its essential needs,
including food, chemical products, therapeutic drugs, clothing, warmth and facilitating movement
(Kirby, 2004).
OPEC reported that over the last forty years, fossil fuels, primarily oil, have been the main source
of energy in the world (E. ref. 3).
The organisation predicts that the dependence on oil will remain the same over the next two
decades, with a decline by 2030 from 39% in consumption to 36.5%. If the current trends in energy
policy and technology development continue unchanged then the demand for oil could rise to
118mb/d by 2030, i.e. an increase of 34mb/d in the interval 2005-2030. Given the economic crisis,
affecting the world since 2008, allied to the high price of crude oil, global consumption has fallen.
This is expected to reverse as the global economy begins to recover. The demand for crude oil will
be driven by the needs of the transport sector, with a predicted rise to 18mb/d by 2030. This will
remain so, even with the existence of alternative sources for producing transport fuels, such as tar
sand, coal, shale, and even recycled tyres and plastics. However, the process of producing petrol
and diesel from crude oil remains the least expensive (E.ref. 4).
The process of producing transport fuels and other useful products involves transporting crude oil
to the refinery where it is passed through a distillation column at atmospheric pressure, and
fractions of jet fuel, diesel, heavy and light straight run gasoline are recovered; these high quality
fuels constitute only 40% of the crude oil (E.ref. 5). The remaining 60% is composed of poor fuels
that require more processing to transform them into valuable transport fuels. The atmospheric
distillation column bottoms are then distilled under a vacuum, which produces gas oils. These
heavy hydrocarbons are then passed as feedstock to hydrocracker or fluid catalytic cracker (FCC)
units, and converted into transport fuels and light gases after sweetening.
As mentioned previously, valuable fuels are produced following vacuum distillation and cracking
of the 60% crude oil part. This percentage varies with the crude oil constituents, rising in heavier
crude oils, characterised by higher molecular weight constituents with less hydrogen molecules. In
this case, the value of cracking units, FCC and hydrocrackers, is even greater. Therefore, refineries,
in future, will require greater upgrading unit capacity, reaching 7.5mb/d in 2020, according to
OPEC (E.ref. 3). It is widely recognised that one of the most efficient way to increase productivity
and efficiency is by adapting advanced controller strategies (McFarlane et al., 1990; McFarlane et
al., 1993; Krishna and Parking, 1985; Elshishini and Elnashaie, 1990).
2.1.1 Process technology

Introduction to FCC and it significance
The FCC process among those used in upgrading crude oil products is characterised by its long
term reliability and operation, and the variety of products resulting, such as high octane gasoline.
The capability of the FCC process to produce, at low cost, high yields of gasoline with high octane
number, means it plays a major role in the strategy of gasoline production in refineries (E.ref. 6).
The valuable contribution made by FCC units in gasoline production can be seen in the example of
the El Segundo refinery FCC unit operated by Chevron, producing 20 gallons of gasoline per
second, which represents the volume of 100,000 car fuel tanks (E.ref. 5).
FCC units vary quite significantly, in terms of where the cracking reaction occurs, and the mode of
control selected. In pre-1965 FCC units, the cracking reaction occurs in the dense-phase, fluidised
catalyst bed reactor, while control is established through choice of bed depth time and temperature
during the process. In time, cracking reactions in the riser became more prominent, especially after
introduction of potent zeolite catalysts; all this, required modifications to unit operation. New FCC
unit designs have contributed to fundamental changes (Saxton and Worley, 1970), where reactor
bed level was minimised; in particular, the rate of circulation of the catalyst was used to control the
cracking reaction. In the same context, older units were also modified to accommodate the
increased prominence of riser cracking reactions. In these scenarios, feed cracking reactions in the
riser have varied in proportion to those taking place in the bed reactor, through varying the
combinations of reactor and riser, between multiple or single riser feed lines, in parallel and serially
(Saxton and Worley, 1970).
It is worthwhile to mention that FCC units are also capable of producing other high value oil
derivatives, such as polypropylene, in addition to gasoline, extremely flexibly, meeting demand in
this regard. For example, the FCC unit designed and licensed by UOP has been modified according
to the target product, where the purpose of the PetroFCC unit is production of petrochemical
feedstock. The UOP PetroFCC unit has the ability to generate ethylene, propylene, benzene and
paraxylene, and butylenes, which represents a significant rise in yield in comparison to the
traditional unit (E.ref. 7). Such production capabilities are reflected in a production capacity of
10
about 28% of a total production 50.2 million tons per year of PG/CG (polymer and chemical grade)
propylene by FCC units. The yields of the process are shown in Table 2.1.
Table 2.1: Uses of FCC products.
FRACTION
USES
Gases
Plant fuel system

Further processed into petrochemical
Feedstocks
LPG (Liquid Petroleum Gas)
Treated and blended into motor Gasolines
Refined into motor gasoline
Diesel fuel
Jet fuel
Fuel oil
Residual fuel
Thermal cracking
Residual fuel
Thermal cracking
Gasoline
Naphtha
Light Cycle Oil
Heavy Cycle Oil

Clarified Slurry Oil
The process results in production, in the regenerator, of waste flue gas, composed of water vapour,
carbon dioxide, and carbon monoxide.
FCC units are the key producers of the majority of refinery-grade (RG) polypropylene as a result of
gasoline and distillate production operations. RG polypropylene is a by-product of refineries, and
annual worldwide output is 31.2 million tons. The petrochemical industry currently absorbs over
50% of this RG polypropylene output. It is predicted that output of RG polypropylene by FCC units
will rise by around 5% each year (E.ref. 8).
The present discussion has reflected on the role of FCC units in the context of oil consumption in
the future, where they are expected to play a significant role in accommodating expanding demand
for olefins, diesel, and gasoline; FCC units will represent 47% of new refinery capacity (E.ref. 3). A
technical description of the three parts of the FCC process is presented below (see Figure 2.1):
a) Pre-treatment of the feedstock: this involves the removal of sulphur as well as other
contaminants;
b) Catalyst Reaction/Regeneration: the hydrocarbon feedstock is cracked after coming in
contact with the hot catalyst. The coke formed on the catalyst surface is then burnt off in
the regenerator section; and
c) Separation stage: based on specific boiling range, separation of hydrocarbons into different
products, such as diesel, gasoline, occurs.
11
Main fractionator
Reactor
Stack gas valve
V14
Regenerator
Steam
tc
en
Sp
ly
ata
st
Wet
gas
V11
Scsc
Riser
ed
at
er st
en aly
eg t
R ca
Steam
V7
Combustion
air blower
Wet gas
compressor
svreg
Feed preheat
furnace
V6
Fresh feed
oil
Atmospheric Air
F3
F2
F4
Slurry
recycle
F1
F5
Furnace
fuel
Diesel
Wash
oil
Figure 2.1: UOP type Fluid Catalytic Cracking Unit.
While the term mature may be used to describe FCC technology, strictly, it is only true with
respect to:
a. its existence as a commercial process for a long time; and
b. multi-phase fluidisation, as the underlying technology (Sadeghbeigi, 2000).
Otherwise, the technology continues to develop, and respond to the changing demands of the
industry. In this context, understanding FCC process behaviour is key to responding to technical
challenges and market demands, including environmental protection, operating conditions, higher
hydrocarbon feeds processing, unit upgrading, refinery product derivatives, and feeds.
12
FCC unit operation may be better understood through experimental, or computer simulation studies.
In comparing both approaches, experiments are very costly, given the large number of variables
involved, e.g. process parameters, production strategy, feedstock type. Therefore, simulation offers
a more cost effective and flexible means of studying and optimising the FCC process. The main
effort lies in developing a model that accurately describes the process, after which responses can be
modelled for any change in the variables of interest, without any further work on the model
(Sadeghbeigi, 2000).
Process Description
The FCC process earns its name from the fluid-like behaviour of the very fine particle catalyst
when mixed with vapour. The constant circulation of fluidised catalyst between reaction and
regeneration zones keeps the cracking process continuous and stable. In addition, the re-circulating
catalyst provides a convenient heat transfer medium between the reactor, regenerator and the oil
feed. Current basic designs of operational FCC units include: side-by-side, Orthoflow or stacked
types (i.e. separate but adjacent reactor and regenerator vessels), and reactors mounted directly
above the regenerator, respectively (Gary and Handwerk, 1975).
Under normal FCC operational conditions, gas-oil and any other feedstocks are heated. This heat is
usually provided by the main fractionator bottoms pumparound and/or fired heaters. The FCC feed
is preheated and then channelled to the reactor riser base, meeting the regenerated catalyst. Since
the catalyst has been heated in the regenerator, this allows the feed to attain the desired temperature
for the reactor. The catalytic reaction in the riser is endothermic, with cracking taking place in the
vapour phase, and requires the feed to be vaporised. In practice, steam is the preferred means of
atomising the feed. Catalyst activity is steadily degraded due to coke deposits arising from the
cracking reactions while the catalyst promotes the reaction without it being chemically changed. It
is at the reactor stripper stage that the catalyst and cracking products are separated. The centrifugal
separation effect of cyclone stages is used to collect and return the catalyst through the diplegs and
flapper/trickle valves to the stripper. On its journey into the stripper, hydrocarbons are adsorbed on
the surface of the spent catalyst. Meanwhile, the porous catalyst is filled with hydrocarbon vapour;
it also entrains vapours as it falls into the stripper. Steam is primarily used to strip and remove the
hydrocarbons entrained by fine catalyst particles. Product vapour leaves the cyclone stage in the
main fractionator, while hot product vapour from the reactor also enters close to the main
fractionator base. Fractionation involves condensing and then vaporising the hydrocarbon
components. Vapour/liquid contact is promoted using devices such as shed decks, disk/doughnut
13
trays, and grid packing. The wet gas compressor receives unstabilised gasoline and light gases exit
at the top of the main fractionator. Further transformation then takes place in other refinery units.
The main fractionator produces:
A gas phase comprising 70-90% C3 and C4 hydrocarbons, each containing 60% propane
and 40-55% butane respectively.
A 195 C end distillation point gasoline fraction made up of paraffin (45-50%), alkenes (815%), naphtens (7-15%), and aromatics (20-30%).
A diesel fraction (distillation range: 195-420 C); di- and tri-aromatic molecules make up
the bulk of the 30-50% of the aromatics in this fraction.
A slurry above 420 C.
Constant regeneration of the catalyst is needed because of coke poisoning. This takes place in a
regenerator which, in addition to restoring catalyst activity, also supplies heat that drives the
endothermic feed cracking reaction. A blower brings in fresh air to burn off the coke, which is
made up of carbon, hydrogen, and traces of sulphurs and nitrogen. An air distributor near the
bottom of the regenerator vessel is provided for this purpose. The flue gas resulting from coke
combustion passes through cyclones found at the top of the regenerator and, in some designs, is
sent to a CO boiler, while in others a shell/tube or box heat exchanger harnesses the flue heat to
produce steam from the boiler feed water (Sadeghbeigi, 2000).
FCC catalyst
Since there are over 140 different types of catalyst formulations in the world today, it is important
that the refinery personnel involved in cat cracker operations have the fundamental understanding
of catalyst technology. This is useful in areas like trouble shooting of units. The use of additives is
likely to increase in the coming years due to the need to produce reformulated gasoline and to
reduce SOx and NOx emissions (Scherzer, 1990).
Catalyst components
FCC catalysts are in the form of fine powders with an average particle size of 75 microns. There are
four major components in the catalyst.
Zeolite
It is the key ingredient of the FCC catalyst. Its role is to provide product selectivity and much of
catalytic activity. It has a well-defined lattice structure. Its basic building blocks are silica and
alumina tetrahedral. Each tetrahedron consists of a silicon or aluminium atom at the center of the
14
tetrahedron with oxygen atoms at the corners. The activity of the zeolite comes from its acid sites.
The zeolites with application to FCC are Type X and Type Y. Virtually all of the catalysts used
today in FCC are of Type Y (Humphries and Skocpol, 2004).
Rare earth components such as lanthanum and cerium were used to replace sodium in the crystal.
They form bridges between two to three acid sites in the zeolite framework. The rare earth
exchange adds to the zeolite activity and thermal and hydrothermal stability. Low aluminium
content zeolites are called Ultra stable Y or USY because of its higher stability. They are used for
production of high-octane gasoline by raising the olefinicity (O'Connor et al., 2001).
The properties of zeolite play a significant role in the overall performance of the catalyst. The
reactor/regenerator environment can cause significant changes in chemical and structural
composition of the zeolite. For e.g., in the regenerator, the zeolite is subjected to thermal and
hydrothermal treatments. The zeolite must also retain its crystallinity against feedstock
contaminants such as vanadium and sodium. The three parameters, which govern zeolite behavior,
are Unit Cell Size, Rare earth level, Sodium content.
The Unit Cell Size (UCS)
Is a measure of aluminium sites or the total potential acidity per unit cell. The negatively charged
aluminium atoms are sources of active sites in the zeolite. Silicon atoms do not possess any activity
(Cheng et al., 1998).
Rare earth
Elements serve as a bridge to stabilize aluminium atoms in the zeolite structure. They prevent the
aluminium atoms from separating from the zeolite lattice when the catalyst is exposed to high
temperature steam in the regenerator. The rare earth increases zeolite activity and gasoline
selectivity with a loss in octane (Silverman et al., 1986). The insertion of rare earth maintain more
and closer acid sites, which promotes hydrogen transfer reactions. In addition, rare earth improves
thermal and hydrothermal stability in the zeolite.
Sodium
Is originates either from the zeolite or from the feedstock. It decreases the hydrothermal stability of
the zeolite. It also reacts with the zeolite acid sites to reduce catalyst activity. In a de-aluminated
zeolite, when the UCS is low, the sodium can have an adverse effect on the octane of gasoline. It is
attributed to the droop in the number of strong acid sites.
15
Matrix
This refers to the components of the catalyst other than the zeolite. The term active matrix means
the component of the catalyst other than the zeolite having catalytic activity.
Alumina is the source for an active matrix. The active matrix contributes significantly to the overall
performance of the FCC catalyst. The zeolite pores are not suitable for cracking of large
hydrocarbon molecules. They are too small to allow diffusion of the large molecules to the cracking
sites. An effective matrix must have a porous structure to allow diffusion of hydrocarbons into and
out of the catalyst. The active matrix pre-cracks heavy feed molecules for further cracking at the
internal zeolite sites. The result is a synergistic interaction between matrix and zeolite in which the
activity attained by their combined effects can be greater than the sum of the individual effects. An
active matrix can also serve as a trap to catch some of the vanadium and basic nitrogen. The high
boiling fraction of the FCC feed usually contains metals and basic nitrogen that poison the catalyst.
Catalysts and the Cracking Reaction
First, a long-chain hydrocarbon molecule is absorption on the porous surface of the catalyst. This
absorption weakens the bonds between the atoms of the hydrocarbon molecule which makes the
molecule more reactive (Bhattacharyya, 2010).
Inside the pores, the hydrocarbon molecule contacts the catalysts active sites. The acidity of the
active sites promotes a chemical reaction that cracks the hydrocarbon molecule apart. The cracked
molecules that leave the catalyst particle are made up of shorter-chained hydrocarbons. They may
form single bonds or rings. When these shorter-chained hydrocarbons are disengaged from the
catalyst, they can be separated into several more valuable products: olefinic gases, liquefied
petroleum gases, gasoline, and light fuel oil (Katz, 2001).
Process Chemistry
The chemistry of the fluid catalytic cracking process is complex although catalytic cracking has
been subject to long and in-depth investigation. This reaction has been described as occurring
because carbenium ion intermediates are produced on the acid sites of a catalysts surface
(Tominaga and Tamaki, 1997). Yet, the nature of these carbocations is still controversial. Meyers
(1997) proposed the description: "classical" carbenium ions and protonated cyclopropane
derivatives. In addition to C-C bond cleavage, the complex cracking process also includes other
reactions:
16
Isomerisation;
Protonation, deprotonation;
Alkylation;
Polymerisation;
Cyclisation,
Condensation (due to which coke is also formed).
Complex intra- and inter-molecular reactions distinguish the catalytic cracking process, also leading
to coke formation while the deactivating effect of coke on the catalyst holds the key to an elegant
practical process (Sadeghbeigi, 1995; Magee and Mitchell, 1993).
Equations 2.1-2.15, describe the series of complex reactions occurring to a large gas-oil molecule in
the presence of 1,200F to 1,400F (650C to 760C) FCC catalysts. The nature and strength of the
acid sites on the catalyst surface feature among the many factors affecting the distribution of
products. However, other factors, such as non-ideal mixing in the riser and poor separation of
cracked products in the reactor, cause the majority of cracking in the FCC.
In a typical FCC unit, the main reactions occurring are (Sadeghbeigi, 2000):
a) Cracking
paraffins cracked to olefins and smaller paraffins

C10H22
C4H8+C5H10
(2.2)
aromatic side-chain scission

ArC10H21
(2.1)
olefins cracked to smaller olefins

C9H18
C4H10+C6H12
ArC5H9+C5H12
(2.3)
naphthenes cracked to olefins and smaller ring compounds

Cyclo-C10H20
C6H12+C4H8
(2.4)
b) Isomerisation
olefin bond shift

1-C4H8
iso-C5H10
(2.6)
normal paraffin to iso-paraffin

n-C4H10
(2.5)
normal olefins to iso-olefin

n-C5H10
trans-2-C4H8
iso-C4H10
(2.7)
cyclo-hexane to cyclo-pentane
17
C6H12
C5H9CH3
(2.8)
c) Hydrogen transfer
Naphtene+Olefin
Aromatic+Parafin
(2.9)
cyclo-aromatisation
C6H12+3C5H10
C6H6+3C5H12
(2.10)
d) Trans-alkylation/alkyl-group transfer
C6H4(CH3)2+C6H6
2C6H5CH3
(2.11)
e) Cyclisation of olefins to naphthenes

C7H14
f)
CH3-cyclo-C6H11
(2.12)
Dehydrogenation
n-C8H18
C8H16+H2
(2.13)
g) Dealkylation
Iso-C3H7-C6H5
C6H6+C3H6
(2.14)
h) Condensation
Ar-CH=CH2+R1CH=CHR2
Ar-Ar+2H
(2.15)
2.2 Mathematical modelling concept

Given the various processes occurring in chemical plant, a multidisciplinary approach is needed in
modelling such processes. A system of equations mathematically describes the intrinsic and
interdependent relationships between the process variables, to form the computer model.
The benefits of simulation models are useful in (Tasoti, 2007):
Design of the system: including optimizing the process and its parameters, performing
dynamic analysis, and analyzing the effect of key factors, as well as unit upgrade.
Process control: analysing the possible control strategies, identifying the best control
scheme; developing predictive control model and expert system.
Potential failure mode identification: includes determining factors disturbing the process, or
those that reduce performance of the plant.
Personnel training: enabling realistic training simulation of plant start-up, operation, and
shutdown, including operation at the plant design limits.
Developing safety procedures and determining the environmental impact: allows dangerous
regimes of operation to be identified with associated risk assessment leading to accident
prevention controls, and quantifying social and economic impacts.
18
The mathematical models for a system may be described as stochastic / statistical (empirical) or
analytical, based on the type of relations existing among the system variables. A statistical model is
built on measurements, and observations of system operation, and has the advantage of:
mathematical simplicity, and so requires basic mathematics treatment
Knowledge of the system does not need to be deep, concerning the underlying phenomena,
and processes
However, statistical models also present several major disadvantages, where:
A large amount of data, obtained from experiments on the system to be modelled, is

required
It is not possible to extrapolate and extend mathematical models beyond the range of
experimental data used to build them
A unique and new model must be built to fit each new system or variant developed, using
new data derived from experiments performed on these systems
In contrast, analytical mathematical models are built on equations for the laws of conservation of
mass, energy, and momentum, as well as those laws governing the physics and chemistry of
processes occurring in the system. Such models offer a number of advantages:
The ability to extend them easily to take account of changes, i.e. their domain of validity is
extended,
Provide greater flexibility,
Analytical models can be extended to fit new systems with similar processes.
However, analytical models also present disadvantages, in that:
Knowledge of the systems underlying processes and phenomena must be quite good
Specialist skill is required to derive those equations that describe the physics and chemistry
of the system to be modeled
A model validation/verification stage, involving experiment, is needed; the experimental

data is input into the model to compare output with reality.
Additionally, the system models may reflect either the steady state, or unsteady (dynamic) state; in
the former, the model algebraic equations are independent of time, i.e. do not change with time;
while in the latter, the model equations are time-dependent, differential equations describing
continuous system variables. In addition, the relationships in either case may be linear or non-
19
linear. The analytical model of the chemical system is built by specifying appropriate energy,
momentum, and mass conservation equations coupled to the equations of chemical equilibrium and
the physical state of the system.
The statistical model is built by recording the process variables in experiments, and
applying statistical methods to determine the equations relating the process inputs to the
output
A mixed statistical and analytical model may be defined by coupling empirical and
analytical equations
2.2.1 FCC cracking kinetics modelling

Prior to the 1930s, hydrocarbon cracking was achieved by applying high temperature to effectively
break the bonds in the high molecular weight compounds resulting in lower molecular weight
hydrocarbons. However, this thermal cracking process was superseded by catalytic cracking, which
would occur rapidly, and at lower temperature.
Currently, key components of modern refinery infrastructure are fluid catalytic crackers, which
facilitate the conversion of high molecular weight hydrocarbons into desired products, LPG,
gasoline, and diesel. Given the high value products of cracking, any development that increases
FCC process yields, by as little as 1%, is welcomed (Das et al., 2003). However, such advances can
only be achieved by deep understanding of FCC operations, typically, residence times,
temperatures, catalysts, feedstocks, and catalyst-to-oil (C/O) ratio. Optimisation of unit operation
for maximum yield in particular modes of operation requires models for the reactor and reaction
kinetics that are highly accurate, regardless of the feedstock used.
Review of FCCU Modelling
The importance of the FCCU has led to the development of many models over time, based on
varying assumptions regarding reaction kinetics, and component hydrodynamics, including
generator and riser. The focus in some models has been the regenerator only (Ford et al., 1976;
Errazu et al., 1979; de Lasa et al., 1981; Guigon and Large, 1984; Krishna and Parkin, 1985; Lee et
al., 1989a), while others have been restricted to the cracking process or reactor (Weekman and
Nace, 1970; Paraskos et al., 1976; Jacob et al., 1976; Shah et al., 1977; Lee et al., 1989b; Larocca et
al., 1990; Takatsuka et al., 1987), and yet others that included both reactor and regenerator (Kumar
et al., 1995; Lee and Kugelman, 1973; McGreavy and Isles-Smith, 1986; Bozicevic and Lukec,
1987; Arandes and de Lasa, 1992; McFarlane et al., 1993; Arbel et al., 1995; Arandes et al., 2000).
These models have been summarised in Table 2.2, with a breakdown of feedstock/product
combinations with respect to each component.
20
Table 2.2: Review of kinetic models used in FCCU with different lump components.
Components
Kinetic
Models
Pl
Nl
Asl
3 Lump
4a Lump
4b Lump
5a Lump
5b Lump
6a Lump
6b Lump
10
11
12
Ash
Arh
Ph
Nh
Gasoline
LPG
Gas
Coke
Approach
R e f.
PF
Gas Oil
Gas + Coke
Gas Oil
Gas
Coke
PF
2,3
Gas Oil
LPG + Gas
Coke
PF
4,5.19
Gas Oil
Coke
PF
6,7
Coke
PF
8,9,10
Coke
PF
11
Coke
PF
12
Coke
PF
13
Coke
PF
14
LPG
Gas
LCO
Gas Oil
LPG
Gas
LCO
Gas Oil
LPG
Gas
HCO
Non-Aromatic Carbons
8c Lump
12 Lump
HCO
8b Lump
10 - Lump
Arl
Product
5
LCO
7 Lump
8a Lump
Feed
4
Aromatic carbons
LPG + Gas
G + Diesel
Propene + butene
Propane + butane
Ethene
G + Diesel
Propene + butene
Propane + butane
Ethene
Other
gases
Other
gases
LCO
LPG
Gas
Coke
PF
15
Gas Oil
LPG
Gas
Coke
PF
16
Pl
Nl
Asl
Arl
Ash
Arh
Ph
Nh
Pl
Nl
Asl
Arl
Ash
Arh
Ph
Nh
Gas + Coke
LPG
FF
Gas
Coke
FF
17,18,19,20,21
22
Where : PF: Product Focus, FF: Feed Focus, AG : Aromatic Gasoline, Arh : Heavy Aromatic Rings, Arl : Light Aromatic Rings, Ash : Heavy Aromatic Substituents, Asl : Light Aromatic Substituents, Gas : ( Light Gas,
Dry Gas, Fuel Gas ), HCO : Heavy Cycle Oil ( Ph, Nh, Ash, Arh ), HGO : Heavy Gas Oil ( Ash, Arh ), LCO : Light Cycle Oil ( Pl , Nl , Asl , Arl ), LGO : Light Gas Oil ( Asl , Arl ), LPG : Liquid Petroleum Gas, Nh :
Heavy Naphthene, Nl : Light Naphthene, Ph : Heavy Paraffin, Pl : Light Paraffin, VGO : Vacuum Gas Oil, P, O, N, A: Paraffin, Olefins, Naphthene, Aromatic.
21
A common assumption regarding the flow in the reactor riser is to consider that it follows a plug
regime (Arbel et al., 1995), allied to a uniform distribution temperature in the cross-section, and a
gradient over the riser height. Further assumptions include adiabatic conditions, quasi-steady state,
temperature uniformity irrespective of position for the two phases, no slip conditions, and oil
vapour/catalyst constant heat capacities. Another assumption is that the riser has isothermal
conditions, as in the CSTR type, despite the non-isothermal conditions existing in the riser bottom,
given the finite time for mixing (McFarlane et al., 1993). Therefore, the main difference among
these models of the riser is in the treatment of the kinetics of reaction.
The study of catalytic cracking involves two key areas: (i) catalytic cracking reaction mechanism;
and (ii) the approach chosen for describing the kinetics of complex, multi-component mixtures (the
level of kinetic expression complexity selected must be balanced with the financial, resource, and
computational cost, etc. of the model). A reduced modelling overhead may be secured by
employing correlations of cracking kinetics, given the availability of several, such as The saturates,
aromatics, resins and asphaltenes SARA (Xu et al., 2005), lumps like the 5 lump model (Bozicevic
and Lukec, 1987), and a mechanistic approach to reactions as in single-event kinetics (Feng et al.,
1993).
The Lumped Approach to Kinetic Modelling
Feedstocks passed through a cracking unit vary in component species, which are typically a huge
amount of different hydrocarbons, beyond the ability of models to account for the kinetics of each
one. Overcoming this obstacle involves the introduction of lumping, in which species are grouped
together based on their boiling range, where it is thought that their reaction kinetics will be the
same. These lumped groups represent pseudo-components that are treated as though they are
individual components. An example of lumped components is that of C6 to C8 paraffins, which
share the same physical and chemical properties. In gas oil cracking, reactions take place
sequentially along a series of paths, where the resulting primary products are lumped into pseudocomponents, such as gasoline, coke, LCO, and LPG (Gomez-Prado et al., 2006).
Kinetic Models based on Products
Blanding (1953) presented the first model of catalytic cracking kinetics using a two lump approach,
where the first lump contained all the components with boiling points above that of gasoline, and
the second lump contained those below. Nearly two decades later, a 3-lump model Figure 2.2 was
proposed; the first lump contained gas oil that was not cracked, the second grouped the components
of gasoline, and the third light gas and coke (Weekman and Nace, 1970). This model was
22
distinguished by being simple as it had three components resulting in only three reactions, and
allowed the yield of gasoline, and conversion of gasoil to be calculated at the same time (Lee et al.,
1989).
Gas + Coke
Gas oil
Gasoline
Figure 2.2: 3-lump Model.
However, this model also presented some disadvantages, including the need to calculate a new
kinetics set, whenever the feedstock changed (Weekman and Nace, 1970). This major disadvantage
was overcome with the development of a 10-lump model, which was more complex (Figure 2.3),
but allowed feedstock composition and kinetic constants to remain independent (Jacob et al., 1976).
This model was made up of lumps, in which groups with similar molecular structure were grouped
together. These lumps are the products: (i) gasoline (G) and (ii) coke with gases (C). Then the
reaction groups: (iii) naphthtenes (N), (iv) paraffins (P), (v) aromatic rings (Ar), (vi) aromatic
substituent groups (As); these reaction groups are further subdivided into light and heavy fractions,
where the former have boiling ranges below 343 C, and are denoted by the subscript, l, and the
latter have boiling ranges above 343 C, and are denoted by the subscript, h.
The 10-lump model provided a number of advantages, as it enabled estimates of the rate of light
and heavy oils, as well as gasoline production, as well as gas oil conversion to be calculated.
However, the model required the composition of feedstock to be determined, and also further data
from experiments (Jacob et al., 1976). In common with the 3-lump model, it was not possible to
independently estimate coke formation (Feng et al., 1993).
The concept of lumping feedstock components and products has become well-established and
progressed over time (Weekman and Nace, 1970). In catalytic cracking models proposed by Yen et
al. (1988) and Lee et al. (1998), the three lump model was developed into a 4-lump model (Figure
2.4) by creating two separate lumps from the coke with gas oil lump.
23
C
Pl
Ph
Nl
Nh
Asl
Ash
Arl
Arh
Figure 2.3: 10-lump model.
Gas
Gasoline
Gas oil
Coke
Figure 2.4: 4-lump Mode.
A 5-lump model was proposed (Figure 2.5), where the gas oil lump was further sub-divided into
light and heavy fractions (Corella and Frances, 1991). For the case of aromatic gas oil cracking, this
5-lump model was made simpler by Dupain et al. (2003) in modelling less reaction. The Weekman
and Nace (1970) 3-lump model was extended into a 5-lump model by dividing the gas oil lump into
three new lumps comprising paraffins, naphthtenes, and aromatics by Larocca et al. (1990). The 5lump model by Ancheyta-Juarez et al. (1999) was different in that rather than divide the gas oil
lump, they split the gas lump into two dry gas and liquefied product gas lumps. The model
describing the distribution of the products of catalytic cracking was proposed by John and
24
Wojciechowski (1975). In the model, primary and secondary products were n-butane, propylene,
and butane, and secondary products were represented by iso-butane, ethylene, ethane, methane, and
coke. Corma and Martinez-Triguero (1984) proposed modifications to the model in adding
isobutene and propane to the primary and secondary products group.
Coke
Gasoline
LCO
Gas oil
Gas
Figure 2.5: 5-lump Model.
6-lump models were proposed by Takatsuka et al. (1987), who added a new lump for the residual
oil, and Oliveira and Biscaia (1989), who focused on paraffinic gas oils and gasoline. An 8-lump
model was derived from the Ancheyta-Juarez et al. (1999) 5-lump model by Hagelberg et al.
(2002) (Figure 2.6). The gasoline lump was split into olefin, paraffin, naphthene, and aromatic
components lying in the 343C+ boiling temperature range, and also those components with boiling
temperatures in the range, 220C to 343C, comprising light naphthenes, light paraffins, and light
aromatics. The assumption was that the cracking the heavy hydrocarbon lumps would result in
gasoline, light gas oil, coke, and light gases lumps. Furthermore, the light gas oil lump would be
cracked to give gasoline, coke, and light gases. Further cracking was considered possible, where
gasoline would be converted into coke and light gases. The model is easier to deal with given the
reduced set of kinetic parameters. Additionally, data to fit the kinetic rates of the model were
obtained from short contact reactor experiments.
An advance in the area of 5-lump model was presented in the work of Dupain et al. (2006), in
taking account of thermal cracking in producing coke and gas. However, despite the simplicity of
kinetic models based on products, they are dependent on the feedstock conceptualised as one
component. Such dependence on the feedstock in the model kinetics is a significant disadvantage,
as new experiments must be performed for any change to the feedstock composition to derive the
appropriate kinetic parameters.
25
Gasoline
PG
OG
NG
Coke
D. Gas
AG
LPG
Kinetic Models based on the feed

As was discussed previously for those models based on products, a number of models have also
been proposed that are based on the feedstock, beginning with the Jacob et al. (1976) 10-lump
model, where the spectrum of products was treated in a better way. The composition of the
feedstock was studied from the perspective of its influence on the gasoline product composition.
The feedstock was considered to be a combination of napthtenes, paraffins, and aromatics, while
gasoline was considered to be composed of olefins, naphthenes, paraffins, and aromatics (Pitault et
al., 1994). This model also considered the LPG, light cycle oil, coke, and dry gas product groups.
All this led to the inclusion of 25 reactions, such as cyclisation of olefins to naphthenes, hydrogen
transfer, -scission, and condensation with hydrogen transfer between olefins and aromatics.
Data for the reaction rate constants were determined by experiments on the microactivity-test
(MAT) system. However, this may introduce errors to the model given the limitations associated
with the MAT system in the kinetic modelling of catalytic cracking (Farag, 1993). It is argued that
the time allocated (1-2 min) for catalyst contact is shorter than is the case in practice in the riser. In
addition, feed vaporisation, and the endothermic nature of cracking may lead to a significant drop
in reactor temperature, which is not accounted for (Hagelberg et al., 2002).
A 13-lump model (Figure 2.7) presented by SINOPEC (Chen and Cao, 1995) shares the same
features as the 10-lump model. Feedstock components are grouped by family, i.e. naphthtenes (N),
paraffins (P), asphaltenes (Fa), and aromatics (Ca), and also based on range of boiling temperature
denoted by l for diesel (205 - 320C),
for gas oil (320 - 500C), and
for residue (500C and
above). Products are then split into a gasoline lump (G), coke lump (Ck), and a lump including all
C1 to C4 gases (Gs) (Gomez-Prado et al., 2006).
26
In another model, the gas products lump is divided into a primary lump composed of 46.7g/gmol
average molecular weight (LPG) gases, and secondary lump composed of 18.4 g/gmol average
molecular weight dry gaseous constituents to give a 12-lump model (Olivera, 1987).
Cah
Fah
Cam
Cal
CK
Gs
Ph
Nh
Pl
Nl
Pm
Nm
These 10, 12 and 13 lump models cannot include the whole spectrum of products, and so further
treatment is needed to provide estimates of coke and light gas yields. In the case where coke is not
presented in the kinetic model component balance, catalyst coke deposits may be estimated by the
correlation given by Kambreck (1991).
Gas oil cracking was represented in a kinetic model using the lumping approach, where the gas oil
feed is converted into lower boiling point products, namely gasoline, liquid petroleum gas (LPG),
light cycle oils (LCO), coke, and light gases (LG). Further conversion of these products results in
lower cut point products, such as LPG, gasoline and LG from LCO, as well as coke (AraujoMonroy and Lopez-Isunza, 2006).
This kinetic model is distinct given that the dependence of kinetic parameters on feedstock type is
taken
care
of
through
frequency
factor
functions
representing
the
ratio
of
feed
aromatics/naphthenes. Moreover, an innovative feature is that each feedstock component is

converted into its constituent species that are adsorbed and cracked on the catalyst surface to give
the lower level component. For example, the paraffins component of gas oil, at the LCO level, is
transformed into, and adsorbed as olefins, paraffins, aromatics, and naphthtenes.
27
However, a number of disadvantages are present, where the basis of this model is claimed to be
Gates (1979) catalytic cracking reactions, but this is not so clear in the actual model. Furthermore,
light gas and LPG components are treated as lumps, rather than individually.
Feed/Product-focus Kinetic Models
Since the models discussed above focused either on the product spectrum or the feedstock, their
limitations led workers to propose models that integrated both, e.g. Gomez-Prado (2003). In the
model, four families are considered to form the feedstock, namely olefins, paraffins, aromatics, and
naphthenes. In addition to asphaltenes, these are split by boiling range into cuts comprising light
gasoline (50 to 120C), heavy gasoline (120 to 205C), diesel (205 to 320C), gasoil (320 to
500C), and residue (greater than 500C). The product spectrum is defined to include the key
refinery products of interest, i.e. gases, gasoline, and diesel, as well as coke, which represents a
major component in the model since it is the source of heat for vaporising the feed, and maintaining
the endothermic cracking reactions. The full model is made up of 294 reactions for the 32
components (Gomez-Prado et al., 2006).
2.3 Introduction to advanced chemical process control

2.3.1 Why advanced process control?
From its beginnings in antiquity until the early 1960s, the field of process control was based almost
entirely on mechanical, electrical, or pneumatic analog controllers, which were usually designed
using linear single-input, single-output considerations. Hardware limitations, economic cost, and
the dearth of applicable theory usually precluded anything more complex then these simple
schemes. Because many large-scale industrial processes are endowed by nature with large time
constants, open-loop stability, and significant damping of fluctuations through mixing and storage
tanks such simple control schemes work well for perhaps 70 percent of the control loops one might
encounter. For the remaining 30 percent more difficult control problems, most controllers were
considered marginally acceptable during this early period because there were few environmental
regulations, product specifications were quite loose, and intermediate blending tanks could cover
many of the sins of inadequate control. Thus the costs of even small sophistications in control were
high and the economic incentives for improved control were comparatively low (Agachi et al.,
2006).
28
Over the last 25 to 30 years, there has been a dramatic change in these factors. Industrial processes
are now predominantly continuous with large throughputs, highly integrated with respect to energy
and material flows, constrained very tightly by high-performance process specifications, and under
intense governmental safety and environmental emission regulations. All these factors combine to
produce more difficult process control problems as well as the requirement for better controller
performance. Significant time periods with off-specification product, excessive environmental
emissions, or process shut-down due to control system failure can have catastrophic economic
consequences because of the enormous economic multipliers characteristic of high throughput
continuous processes. This produces large economic incentives for reliable, high-quality control
systems in modern industrial plants.
Another recent development in process control is that the performance of real-time digital
computers suitable for on-line control has improved significantly, while prices have fallen
drastically. Figure 2.8 shows an example of the price trends for small minicomputers in spite of the
inclusion of more reliable electronics and increasing inflation. With the process control computer
now such a small part of the overall process capital costs, the installation of a fast minicomputer
with large amounts of storage can often be easily justified on the basis of improved safety and
manpower savings. Once in place, the computer is usually operating in a timesharing mode with
large numbers of input/output operations, so that the central processing unit (CPU) is typically in
use only about 5 percent of the time. Thus many installations have 95 percent of the computing
power of a highly capable minicomputer, programmable in a high-level languages such as Fortran,
C, Visual Basic, LabVIEW etc., already available for implementing sophisticated computer control
schemes.
Cost (thousands of $)
35
30
25
20
15
10
5
0
1965
1970
1975
1980
1985
1990
1995
Year
Figure 2.8: An example of price trends for real time minicomputers.

29
2000
2005
2010
At the same time, modern control theory is undergoing intense development, with many successful
applications in the chemical industry. Recently, a number of process control research groups have
been applying new and more sophisticated control of algorithms and schemes to simulated,
laboratory-scale, and even full-scale processes. The process control engineer must design an
economically optimal process control scheme based on a judicious comparison of the available
control algorithms. The present work offers help in this and strives to present a brief introduction to
the theory and practice of the most important modern process control strategies, using the
industrially relevant chemical processes as the subjects of control performance studies (Lu, J. 2001
and Agachi et al., 2006).
2.3.2 Concepts of process control

A control system comprises a number of elements, and is used to bring and maintain a process
variable at a specified level or trajectory-the set point or reference. Control system analysis is based
on linear system theory, whereby the cause-effect relationship between the various system elements
is assumed linear. Figure 2.9 is a schematic representation of the variables in a process. These
variables can be classified into input, and output, and further sub-divided into disturbance,
manipulated, associated, and controlled variables. An input variable is one governed by the system
environment including disturbance inputs and manipulated or control inputs, which in turn,
affect the output; these include mass or energy flows, environmental variables etc. Output variables
include technological parameters, yield, etc. Output variables comprise controlled variables that
need to be held at specific set points, and associated variables that are controlled within a specified
range (Agachi et al., 2006).
Unmeasured controlled variables
Disturbances (d)
Inputs (v)
Associated variables
PLANT
Manipulated
variables (u)
Outputs (y)
Controlled variables
Figure 2.9: Definition of input and output variables considered for control system design.
Control systems are classified into open and closed loop. Figure 2.10 shows a schematic of an
open-loop system, in which a controller or control actuator achieves the desired response. On the
other hand, the closed-loop control system utilises a measure of the output-the feedback signal-to
see how far this differs from the desired output.
30
Desired output
Controller
Process
Output
Response
Figure 2.10: Open-loop control system.
Figure 2.11 represents a simple closed-loop feedback control system. In this type of system, the
process is controlled by a function describing the relationship between the output and the reference
input. Generally, the difference between the reference input and actual process output is amplified.
This is fed back into the control system, with the objective of constantly reducing this difference.
This concept of feedback control forms an important basis of system design and analysis (Agachi et
al., 2006; Nagy, 2001).
Desired
output
Response
Comparison
Controller
Process
Output
Measurement
Figure 2.11: Closed-loop feedback control system.
2.3.3 History of automatic control

The concept of feedback control was one of the most important contributions in the history of
automatic control (Mayr, 1970). The use of feedback in order to control a system has had a
fascinating history. The first applications of feedback control rest in the development of float
regulator mechanisms in Greece in the period 300 to 1 B.C. The water clock of Ktesibios used a
float regulator. An oil lamp devised by Philon in approximately 250 B.C. used a float regulator in
the oil lamp for maintaining a constant level of fuel oil. Heron of Alexandria, who lived in the first
century A.C., published a book entitled Pneumatica, which outlined several forms of water-level
mechanisms using float regulators (Mayr, 1970).
31
Governors and
Mechanical
Controllers
Relays and Stepping Switches
Direct-connected Pneumatic
controllers
Electronic Logic Controllers
Programmable Logic
Controllers (PLCs)
Transmitter-Type Pneumatic Controllers
Electronic Analog
Controllers
Discrete Drive
Control Systems
Distributed Digital
Controls
Computer-based
Control System
Direct Digital Control (DDC)
Systems
Supervisor Computer Control
1930
1940
1950
1960
1970
1980
1990
Figure 2.12: Evolution of industrial control technology.
The lines of technological development can be divided into two separate streams, as illustrated in
Figure 2.12. The upper stream with its two branches is the more traditional one, and includes the
evolution of analog controllers and other discrete devices such as relay logic and motor controllers.
The second stream is a more recent one that includes the use of large-scale digital computers and
their mini and micro descendants in industrial process control. These streams have merged into the
current mainstream of distributed digital control systems.
The dates of several key milestones in the evolutionary process of the development of control
theory and practice are shown in Table 2.3.
32
Table 2.3: Selected historical developments of control systems.
1769
James Watt's steam engine and governor developed. The Watt steam engine is often used
to mark the beginning of the Industrial Revolution in Great Britain. During the industrial
revolution, great strides were made in the development of mechanization, a technology
preceding automation.
1800
Eli Whitney's concept of interchangeable parts manufacturing demonstrated in the

production of muskets. Whitney's development is often considered as the beginning of
mass production.
1868
J. C. Maxwell formulates a mathematical model for a governor control of a steam engine.
1913
Henry Ford's mechanized assembly machine introduced for automobile production.
1927
H. W. Bode analyzes feedback amplifiers.
1932
H. Nyquist develops a method for analyzing the stability of systems.
1934
Direct-connected pneumatic controls dominate market.
1938
Transmitter-type pneumatic control systems emerge, making centralized control rooms

possible.
1952
Numerical control (NC) developed at Massachusetts Institute of Technology for control of

machine-tool axes.
1954
George Devol develops programmed article transfer" considered to be the first industrial
robot design.
1958
First computer monitoring in electric utility
1959
First supervisory computer in refinery
1960
First solid-state electronic controllers on market.
1960
First Unimate robot introduced, based on Devo1's designs. Unimate installed in 1961 for
tending die-casting machines.
1963
First direct digital control (DDC) system installed.
1970
First programmable logic controllers (PLCs) on market.
1970
Sales of electronic controllers surpass pneumatic
1975
First distributed digital control system on market.
33
Computer-based Control System Developments

In addition to the evolution of the traditional types of control systems described above, a more
recent (and equally important) evolution of computer-based process control systems has been
taking place, as shown in the lower part of Figure 2.12. The first application of computers to
industrial processes was in the areas of plant monitoring and supervisory control. In September
1958, the first industrial computer system for plant monitoring was installed at an electric utility
power generating station. This innovation provided an automatic data acquisition capability not
available before, and freed the operator from much drudgery by automatically logging plant
operating conditions on a periodic basis. Shortly thereafter (in 1959 and 1960), supervisory
computer control systems were installed in a refinery and in a chemical plant. In these applications,
analog controllers were still the primary means of control. The computer used the available input
data to calculate control set points that corresponded to the most efficient plant operating
conditions. These set points then were sent to the analog controllers, which performed the actual
closed-loop control. The ability of supervisory control computers to perform economic optimization
as well as to acquire, display, and log plant data provided the operator with a powerful tool for
significantly improving plant operations.
The next step in the evolution of computer process control was the use of the computer in the
primary control loop itself, in a mode usually known as direct digital control, or DDC. In this
approach, process measurements are read by the computer directly, the computer calculates the
proper control outputs then sends the outputs directly to the actuation devices. The first DDC
system was installed in 1963 in a petrochemical plant. For security, a backup analog control system
was provided to ensure that the process could be run automatically in the event of a computer
failure. This proved to be a wise precaution, because this early DDC installation (as well as many
others) was plagued with computer hardware reliability problems. Despite these problems, it
demonstrated many of the advantages digital control has over analog control: tuning parameters and
set points do not drift, complex control algorithms can be implemented to improve plant operation,
and control loop tuning parameters can be set adaptively to track changing operating conditions.
2.3.4 Economical importance of process control

Control engineering is concerned with the analysis and design of goal-oriented systems. Modern
control theory is concerned with systems with the self-organizing, adaptive, robust, learning, and
optimum qualities. The control of an industrial process (manufacturing, production, and so on) by
34
automatic rather than human means is often called automation. Automation is prevalent in the
chemical, electric power, paper, automobile, and steel industries, among others. The concept of
automation is central to our industrial society. Automatic machines are used to increase the
production of a plant per worker in order to offset rising wages and inflationary costs. Thus
industries are concerned with the productivity per worker of their plant. Productivity is defined as
the ratio of physical output to physical input. In this case we are referring to labour productivity,
which is real output per hour of work. In a study conducted by the U.S. Commerce Department it
was determined that labour productivity grew with an average annual rate of 2.8% from 1948 to
1984. In order to continue these productivity gains, expenditures for factory automation in the
United States have increased from 8.0 billion dollars in 1992 to 20.0 billion dollars in 2000.
Worldwide, expenditures for process control and manufacturing plant control have grown from
22.0 billion dollars in 1992 to 45.0 billion dollars in 2000. The U.S. manufacturers currently supply
approximately one-half of worldwide control equipment (Brittain, 1977).
The transformation of the U.S. labour force in the country's brief history follows the progressive
mechanization of work that attended the evolution of the agrarian republic into an industrial world
power. In 1820 more than 70% of the labour force worked on the farm. By 1900 fewer than 40%
were engaged in agriculture. Today, less than 5% work in agriculture.
The easing of human labour by technology, a process that began in prehistory, is entering a new
stage. The acceleration in the pace of technological innovation inaugurated by the Industrial
Revolution has until recently resulted mainly in the displacement of human muscle power from the
tasks of production. The current revolution in computer technology is causing an equally
momentous social change: the expansion of information gathering and information processing as
computers extend the reach of the human brain. The work week in U.S. manufacturing industries
shortened from 67 hours in 1860 to about 39 hours in 1984 and even less nowadays. Control
systems are used to achieve (1) increased productivity and (2) improved performance of a device or
system. Automation is used to improve productivity and obtain high quality products. Automation
is the automatic operation or control of a process, device, or system. Automatic control of machines
and processes is used in order to produce a product within specified tolerances. The term
automation first became popular in the automobile industry but nowadays none of the industries
can survive to the very strict economical and ecological requirements and the process complexities
without automation. There are about 350,000 control engineers in the United States and also Japan,
and over 200,000 control engineers in the states of former Soviet Union and Eastern Europe. In the
United States alone, the control industry does a business of over 30 billion dollars per year!
35
2.3.5 Whatismoderncontroltheory?
In response to an input perturbation, a system is considered stable if it settles to a steady state;
however, it is unstable if it remains unsteady, with outputs varying. In a stable integrating system,
the output asymptotically approaches a ramp, while it changes in an exponential manner in an
exponentially unstable system.
Classical control theory only addresses single-input, single-output (SISO) systems represented by
constant coefficient linear differential equations, also described by Laplace as transforms. However,
modern control theory is able to address general multivariable systems, whether defined by
variable-coefficient linear, nonlinear, and partial differential equations, as well as by integral
equations. Moreover, optimal control theory has been developed so that control schemes can be
implemented to minimise a specified function, such as cost etc.
Furthermore, methods for process identification and state estimation have also been
developed. Process identification algorithms determine the structure of the process model
and can then provide estimates of process parameters. State estimation techniques are
applied in real-time systems to estimate those state variables that are not measured, or to
improve estimates of state-variables given measurement errors. Nagy and Agachi (2004)
describe a typical advanced control scheme as consisting of (see Figure 2.13):
a process subjected to control inputs (u), natural process disturbances (d1), and special input
disturbances (d2) used for the purpose of identification;
measurement devices: these monitor a number of system state variables or a combination of

these, with measurement error included;
state estimator: this provides a best estimate of process state (xest), based on the noisy
measurements (y) and the process model;
a controller: this takes account of the state estimates (xest), setpoints (r) and controller
parameters (K) and generates appropriate control actions;
a process identification block: its function is to identify the process model parameters ()
from user measurements of the process. Provided the parameters do not change with time,
this identification is unique. However, if process parameters vary with time, identification
has to be undertaken periodically, in order to adapt to these changes.
In most applications only a few of the components of this control structure are required.
36
Identification input
d1
d2
Natural process
disturbance
yr
Measurement
device
PROCESS
True value of output
u
xest
State
estimator
Controller
u
On-line
Controller
parameters
r
Setpoint
Controlled
parameter
computation
Process
identification
Off-line
Controller setpoint
optimizer
Figure 2.13: Advanced computer control scheme.
2.3.6 Advanced process control techniques

Key problems in Advanced Control of Chemical Processes
Figure 2.14 outlines the main features of chemical processes in relation to the challenges they pose
in terms of control. Chemical processes are mainly characterised by (Nagy and Agachi, 2004):
These characteristics have revealed the implementation of various advanced control strategis to
maximize process performance.
The chemical industry is characterized as having very dynamic and unpredictable marketplace
conditions. For instance, in the course of the last 15 years we have witnessed an enormous variation
in crude and product prices.
The demands of chemical products vary also widely, imposing different production yields. It is
generally accepted that the most effective way to generate the most profit out of the plants while
37
responding to marketplace variations with minimal capital investment is provided by the integration
of all aspects of automation of the decision making process (Garcia et al., 1989), which are:
Unmeasured
and frequently
disturbances
Nonlinear
dynamic
behavior
Multivariable interactions
between manipulated and
controlled variables
processes
Chemical
Process
Uncertain and
time-varying
parameters
Constraints on
manipulated and
Unmeasured
state variables
Dead time on
inputs and
measurements
High-order and
distributed
state variables
Figure 2.14: Common process characteristics, important in the choice of control strategy
measurement (acquiring measurements and monitoring process variables and states using
appropriate instrumentation);
control (satisfying operating criteria by manipulating degrees of freedom in the process);
optimisation (satisfying economic objectives by manipulating degrees of freedom in the

process);
logistics (maximising profits and realising company objectives through the appropriate
allocation of raw materials and the scheduling of the operating plants).
Currently, practical performance criteria must be achieved by control systems in one or more of the
following aspects:
economic: these are related to holding process variables at the targets defined by the
optimisation phase or through dynamically minimising an operating cost function.
safety and environmental: where process variables are not permitted to pass specific limits
dictated by the safety of personnel or equipment, or environmental regulations.
equipment: process operations may not exceed physical limitations of the equipment.
38
product quality: products must satisfy set specifications.
human preference: these may be dictated by the tolerance of the plant operator: e.g.
variable oscillation levels that are not tolerable.
In addition, the implementation of such integrated systems is forcing the processes to operate over
an ever-wider range of conditions. As a result, we can state the control problem that any control
system must solve as follows (Garcia et al., 1989):
Update on-line the manipulated variables to satisfy multiple, changing
performance criteria in the face of changing plant characteristics.
The whole spectrum of process control methodologies in use today is faced with the solution of this
problem. The difference between these methodologies lies in the particular assumptions and
compromises made in the mathematical formulation of performance criteria and in the selection of
a process representation. These are made primarily to simplify the mathematical problem so that its
solution fits the existing hardware capabilities. The natural mathematical representation of many of
these criteria is in the form of dynamic objective functions to be minimized and of dynamic
inequality constraints. The usual mathematical representation for the process is a dynamic model
with its associated uncertainties.
At the moment there are an important number of advanced control techniques from ad hoc
algorithms for particular systems to very general methods with wide application area and welldeveloped theory. A classification of these techniques is difficult because many of the algorithms
are very similar, being obtained from some more general methods via usually minor changes
concerning the performance criteria, optimization method, prediction horizon, constraint handling,
etc. However, all these algorithms have a common feature: all are based on a process model,
described in different ways. The proposed classification, based on this feature is presented in
Figure 2.15. According to this, the advanced control techniques can be classified first in four
conceptually different categories. The first and most important approach, the Model Predictive
Control (MPC), can be classified further, for example, according to different model types used for
prediction in the controller. This feature being usually the most significant difference among MPC
algorithms.
39
Linear
MPC
Model Predictive
Control (MPC)
Advanced
control
techniques
Optimal
Control (OC)
Nonlinear
MPC
Step Response Models
Impulse Response Models
State Space Models
Polynomial Models
(ARX, ARMAX, etc.)
Analytical Models
Artificial Neural Networks
Fuzzy Models
Linear Weighted Models
Adaptive
Control (AC)
Expert Systems based
Control (ESC)
Figure 2.15: Classification of advanced control techniques.
2.3.7 Historical development of LMPC

The current interest of the processing industry in MPC can be traced back to a set of papers, which
appeared in the late 1970s. In 1978 Richalet et al. described successful applications of "Model
Predictive Heuristic Control" and in 1979 engineers from Shell outlined "Dynamic Matrix Control"
(DMC) (Cutler and Ramaker, 1980) and reported applications to a fluid catalytic cracker (Prett and
Gillette, 1979). In both algorithms an explicit dynamic model of the plant is used to predict the
effect of future actions of the manipulated variables on the output (thus the name "Model Predictive
Control"). The future moves of the manipulated variables are determined by optimization with the
objective of minimizing the predicted error subject to operating constraints. The optimization is
repeated at each sampling time based on updated information (measurements) from the plant. Thus,
in the context of MPC the control problem including the relative importance of the different
objectives, the constraints, etc. is formulated as a dynamic optimization problem. While this by
itself is hardly a new idea (Clarke, 1994; Chang, 1961), it constitutes one of the first examples of
large-scale dynamic optimization applied routinely in real time in the process industries. The MPC
concept has a long history (Kwon, 1994). The connections between the closely related minimum
time optimal control problem and Linear Programming were recognized first by Zadeh and Whalen
in 1962. Propoi (1963) proposed the moving horizon approach, which is at the core of all MPC
40
algorithms. It became known as "Open Loop Optimal Feedback". The main idea of this approach is
presented on Figure 2.16.
Past
Future
ymax
setpoint
Predicted
output
ymin
umax
Manipulated
input
umin
k-1 k k+1 k+2
k+M
time
Control horizon (M)

Prediction horizon (P)
Figure 2.16: The moving horizon approach of MPC techniques.
The extensive work on this problem during the 1970s was reviewed in the thesis by Gutman (1982).
The connection between this work and MPC was discovered by Chang and Seborg in 1983.
Since the rediscovery of MPC in 1978 and 1979, its popularity in the Chemical Process Industries
has increased steadily. Mehra et al. (1982) reviewed a number of applications including a
superheater, a steam generator, a wind tunnel, a utility boiler connected to a distillation column and
a glass furnace. Shell has applied MPC to many systems, among them a fluid catalytic cracking unit
(Prett and Gillette, 1979) and a highly non-linear batch reactor (Garcia, 1984). Matsko (1985)
summarized several successful implementations in the pulp and paper industries Matsko (1985).
Several companies (Bayley, DMC, Profimatics, Setpoint) offer MPC software. Cutler and Hawkins
(1987) reported a complex industrial application to a hydrocracker reactor involving seven
independent variables (five manipulated two disturbances) and four dependent (controlled)
variables including a number of constraints. Martin et al. (1986) cited seven completed applications
and ten under design. They include: fluid catalytic cracker-including regenerator loading, reactor
severity and differential pressure controls; hydrocracker (or hydrotreater) bed outlet temperature
41
control and weight average bed temperature profile control; hydrocracker recycle surge drum level
control; reformer weight average inlet temperature profile control; analyzer loop control. The latter
has been described in more detail by Caldwell and Martin (1987). Setpoint has applied the MPC
technology to: fixed and ebulating bed hydrocrackers (Grosdidier, 1988); fluid catalytic crackers;
distillation columns; absorber/stripper bottom C2 composition control, polymerization reactors and
other chemical and petroleum refining operations (Meadows et al., 1995; Nagy and Agachi 1997;
Marques and Morari 1986; Matsko 1979; Prett and Gillette 1979).
In academia, MPC has been applied under controlled conditions to a simple mixing tank and a heat
exchanger (Arkun et al., 1986) as well as a coupled distillation column system for the separation of
a ternary mixture (Levien and Morari, 1987). Parrish and Brosilow (1985) compared MPC with
conventional control schemes on a heat, exchanger and an industrial autoclave Parrish and Brosilow
(1985).
2.3.8
Industrial MPC development
An excellent review of the industrial development of MPC, was presented by Allgower et al. (1999)
in their review paper. They presented the most important commercial software for industrial
implementation of MPC as follows:
IDCOM
The first description of MPC control applications was presented by (Richalet et al., 1976; Richalet
1993; Richalet et al., 1978). They described their approach as Model Predictive Heuristic Control
(MPHC). The commercial software was referred to as IDCOM, an acronym for Identification and
Command. The distinguishing features of IDCOM are:
1. finite impulse response (FIR) model for the plant;
2. quadratic performance objective over a finite prediction horizon;
3. future output behavior specified by a reference trajectory;
4. input and output constraints included in the formulation;
5. optimal inputs computed using a heuristic iterative algorithm.
The MPHC algorithm drives the predicted future output as closely as possible to a reference
trajectory, defined as a first order path from the current output value to the desired setpoint. The
speed of the desired closed loop response is set by the time constant of the reference trajectory.
42
Richalet et al. (1978) maked the important point that dynamic control must be embedded in a
hierarchy of plant control functions in order to be effective. They describe four levels of control.
Level 3 - Time and space scheduling of production
Level 2 - Optimization of setpoints to minimize costs and ensure quality and quantity of
production
Level l - Dynamic multivariable control of the plant
Level 0 - Control of ancillary systems; PID control of valves.
They pointed out that significant benefits do not come from simply reducing the variations of a
controlled variable through better dynamic control at level 1. The real economic benefits come at
level 2 where better dynamic control allows the controlled variable setpoint to be moved closer to a
constraint without violating it. This argument provides the basic economic motivation for using
MPC technology. This concept of a hierarchy of control functions is fundamental to advanced
control applications and seems to have been followed by many practitioners.
DMC
Engineers at Shell Oil developed MPC technology independently in the early 1970's, with an initial
application in 1973. Cutler and Ramaker presented details of an unconstrained multivariable control
algorithm which they named Dynamic Matrix Control (DMC) at the 1980 Joint Automatic Control
Conference (Cutler and Ramaker, 1980; Cutler et al., 1983). In a companion paper Prett and
Gillette (1979) described an application of DMC technology to a FCCU reactor/regenerator in
which the algorithm was modified to handle nonlinearities and constraints. Key features of the
DMC control algorithm include:
1. linear step response model for the plant;
2. quadratic performance objective with move suppression over a finite prediction
horizon;
3. future output behavior specified by following the setpoint;
4. optimal inputs computed as the solution to a least-squares problem.
By using the step response model one can write predicted future output changes as a linear
combination of future input moves. The matrix that ties the two together is the so-called Dynamic
Matrix. Using this representation allows the optimal move vector to be computed as the solution to
43
a least-squares problem. Feedforward control is readily included in this formulation by modifying

predicted future outputs.
The objective of a DMC controller is to drive the output as close to the setpoint as possible in a
least-squares sense with a penalty term on the input moves. This results in smaller computed input
moves and a less aggressive output response. As with the IDCOM reference trajectory, this
technique provides a degree of robustness to model error. Move suppression factors also provide an
important numerical benefit in that they can be used to directly improve the conditioning of the
numerical solution.
The initial IDCOM and DMC algorithms represent the first generation of MPC technology; they
had an enormous impact on industrial process control and served to define the industrial MPC
paradigm (Garcia and Prett, 1986).
QDMC
The original IDCOM and DMC algorithms provided excellent control of unconstrained
multivariable processes. Constraint handling, however, was still somewhat ad-hoc. Engineers at
Shell Oil addressed this weakness by posing the DMC algorithm as a Quadratic Program (QP) in
which input and output constraints appear explicitly (Garcia and Morshedi, 1986). A distinguishing
feature of QDMC over DMC is that the control moves are solved as a QP with hard constraints
rather than least squares without constraints.
The default QDMC algorithm requires strict enforcement of input and output constraints at each
point of the prediction horizon. Constraints that are strictly enforced are referred to as hard
constraints. Constraints for which violations are allowed are referred to as soft constraints. In
practice Garcia and coworkers (Garcia and Morari, 1982; Garcia and Morshedi, 1986) report that
hard output constraints are typically required to be satisfied over only a portion of the horizon
which they refer to as the constraint window. The constraint window generally starts at some point
in the future and continues on to steady state. They report that if non- minimum phase dynamics are
present, performance is improved by pushing the constraint window farther into the future. This
amounts to ignoring hard output constraints during the initial portion of the closed loop response.
The QDMC algorithm can be regarded as representing a second generation of MPC technology,
comprised of algorithms, which provide a systematic way to implement input and output
constraints. This was accomplished by posing the MPC problem as a QP.
44
IDCOM-M, SMOC, and PCT

As MPC technology gained wider acceptance, and problems tackled by MPC technology grew
larger and more complex, control engineers implementing second generation MPC technology ran
into other practical problems. The QDMC algorithm provided a systematic approach to incorporate
hard input and output constraints, but there were several obvious, limitations:
1. there was no clear way to handle an infeasible solution. For example it is possible for a
feedforward disturbance to lead to an infeasible QP; what should the control do to
recover from infeasibility?
2. the soft constraint formulation is not completely satisfactory because it means that all
constraints will be violated to some extent, as determined by the relative weights.
Clearly some output constraints are more important than others, however, and should
never be violated.
3. process models are still limited to step response models, which are incapable of
representing unstable processes.
4. the constant-disturbance output feedback is sub-optimal and unrealistic especially when
low-level PID controllers are in place.
Fault tolerance is also an important practical issue. Rather than simply turning itself off as signals
are lost, a practical MPC controller should remain online and try to make the best of the sub-plant
under its control. It also became increasingly difficult to translate control requirements into relative
weights for a single objective function. Including all the required tradeoffs in a single objective
function means that relative weights have to be assigned to the value of output setpoint violations,
output soft constraint violations, inputs moves, and optimal input target violations.
These issues motivated engineers at Shell (France), Adersa and Setpoint Inc. to develop new
versions of MPC algorithms. The version marketed by Setpoint was called IDCOM-M, while the
Adersa version was referred to as HIECON (Hierarchical Constraint Control). The IDCOM-M
controller was first described in a paper by Grosdidier et al. (1988). Distinguishing features of the
IDCOM-M algorithm include:
1. controllability supervisor to screen out ill-conditioned plant subsets;
2. multi-objective function formulation; quadratic output objective followed by a
quadratic input objective;
3. controls a single future point in time for each output, called the coincidence point,
chosen from a reference trajectory;
45
4. constraints can be hard or soft, with hard constraints ranked in order of priority.
The SMOC algorithm (Marquis and Broustail, 1988) developed at Shell France is very close to the
"modern" development of MPC with the following distinguishing features:
1. state space models are used which can represent both stable and unstable processes.
2. full state estimation known as extended close-loop observer is used for output
feedback; constant output disturbance is simply a special case.
3. a distinction is introduced between controlled variables that are in the control objective
and feedback variables that are used for state estimation, resolving the difficulty for
interdependent output variables.
4. input and output constraints are observed via a QP formulation.
The SMOC and IDCOM-M algorithms are two of several that represent a third generation of MPC
technology; others include the PCT algorithm sold by Profimatics, the RMPCT controller
developed by Honeywell, and the PFC algorithm developed by Adersa. This generation
distinguishes between several levels of constraints (hard, soft, ranked), provides some mechanism
to recover from an infeasible solution, addresses the issues resulting from a control structure that
changes in real time, uses state estimation as optimal output feedback, and allows for a wider range
of process dynamics (stable and unstable) and controller specifications.
While MPC theory has advanced to the point where nominal stability can be guaranteed with
several schemes, the majority of the current industrial MPC algorithms are based on ideas inherited
from the original DMC and IDCOM algorithms. Products sold by Adersa, Aspen Technology,
Continental Controls, Honeywell, and Pavilion Technology share such features as a finite
prediction horizon and options for impulse or step response models. The main emphasis in recent
years has been to allow for a wider range of model types, including nonlinear models, state space
models (e.g., SMOC from Shell), and better integration of the controller interface into existing
distributed control hardware (Qin and Badgwell, 2000).
2.3.9 Unconstrained Linear Model Predictive Control (LMPC) with step

response and impulse response models
Step Response Models
To derive the step response model for a general multi-input-multi-output (MIMO) it was assumed
that yl, l=1,...,ny, were stable and deal with an output vector at each time interval:
46
y(k
1)
[y1 (k
1),..., yn (k
y
1)]T
(2.16)
and can be define the state vectors

Y (k
1) [y(k
1)T , y(k )T ,..., y(k
3)T , y(k
2)T ]T for
v(k
i)
0;
(2.17)
and
Y (k ) [y(k )T , y(k
1)T ,..., y(k
2)T , y(k
1)T ]T for
v(k
i)
0;
0.
(2.18)
The output vector evolution is described by:

M S Y (k
Y (k )
1)
(2.19)
where
MS
0
0
:
0
0
I
0
0 ... ... 0 0
I 0 ... 0 0
: .
0 ... ... ... 0 I
0 ... ... ... 0 I
(2.20)
And the identity matrix I is of dimension ny x ny.

Each step response coefficient Si is a vector of dimension ny
Si
[s1i , s2i ,..., sn i ]T .
(2.21)
The step response model becomes:

Y (k )
M SY (k
1)
v(k
1)
(2.22)
where S is an nny vector.

Assume next that there are nv inputs vm, m=1,...,nv. Because of linearity, the effects of the individual
inputs can simply be added up:
Y (k )
M SY (k
1)
nv
Sm
vm (k
1)
(2.23)
m 1
This can be defined as an input vector
v(k
1)
coefficient matrix:
47
[ v1 (k
1),..., vn (k
v
1)]T and the step response
s1,1,i
Si
s1,2,i
s2,1,i
:
sn ,1,i
s1,n ,i
...
(2.24)
sn
... sn
,2,i
,nv ,i
where sl,m,i is the ith step response coefficient relating the mth input to the lth output. Then the MIMO
step response model becomes:
Y (k )
M sY (k
1)
v(k
1)
(2.25)
The recursive expression (2.26) is initialized (k=0) assuming that there were no changes v(-l) cu
l=1,...,n (which put the system at steady state) and that the steady state output y 0 is known
(measured). Then the vector Y(0) can be set equal to the steady state system output y 0 repeated n
times:
^
Y (0)
[y(0)T , y(0)T ,..., y(0)T ]T .
(2.26)
The recursive formula (2.26) was proposed in the original formulation of DMC.
If all the outputs are integrating, then (2.26) holds with M S replaced by:
MI
0
0
:
0
0
I
0
0 ... ...
I 0 ...
0 ... ... ...

0 ... ... ...
0
0
0
I
0
0
: .
I
2I
(2.27)
Consequently, both the stable and integrating outputs lead to the general formula:
Y (k )
MY (k
1)
v(k
1).
(2.28)
Impulse Response Models

For stable SISO (single-input-single-output) systems equation (2.19) in an expanded form is:
y(k 1)
y(k )
y(k 1)
:
y(k n 1)
0 1 0 ...
y(k 2)
0 0 1 ...
y(k 1)
0 0 0 ...
y(k )
...
:
0 0 1
y(k n 2)
0 0 1
0
s1
s2
:
sn
48
v(k
1).
(2.29)
and by taking the differences between the adjacent rows the following can be obtaind:
0 1 0 ...
0 0 1 ...
0 0 0 ...
...
0 0 1
0 0 0
y(k )
y(k 1)
y(k 2)
:
y(k n 1)
y(k 1)
y(k )
y(k 1)
:
y(k n 2)
h1
h2
h3
:
hn
v(k
1)
(2.30)
where
y(k
i)
y(k
i) y(k
1)
(2.31)
and
hi
si
si 1;
[h1 h2 . . . hn ]T ;
(2.32)
(2.33)
are the impulse response coefficients of the system.

Analogous to the development of the step response the model can be derived as:
y(k
l)
Y (k
N Ml
l)
Y (k )
N M i 1H
v(k
i)
(2.34)
i 1
Note that
N M i 1H
hi
0
i
i
n
n
(2.35)
and therefore for l > n,

n
y(k
l)
hi
v(k
hi
v(k
i)
i)
(2.36)
i 1
n
Or,
y(k )
(2.37)
i 1
which is referred to as the input-output form of the impulse response model.

Estimation and prediction
The objective for using a model is to predict the effect of future manipulated variable moves on the
output. This will allow finding the best future moves to reach some desired output behavior. To
obtain a more general form of the model in which the measured disturbances are also accounted thy
modeled system can be represented by the block diagram in Figure 2.17. It is assumed that Pu (the
49

effect of u an y) is described by the step response coefficients and Pd (the effect of the disturbances
d on y) by a similar dynamic matrix:
Pd
w
w
+
Pu
Figure 2.17: Plant with manipulated variable u, measured disturbance d and effect of unmeasured
disturbance w.
u
S
S
Su
S
S
1
u
2
Sd
S
S
d
1
d
2
(2.38)
S
S
n 1
u
n 1
d
The straightforward generalization of (2.28) is given as follows:

Y (k | k
1)
M SY (k
1|k
1)
Su
u(k
1)
Sd
d(k
1)
(2.39)
Using this general linear model, the following operations have to be performed to obtain a p-stepahead prediction vector:
1. Preparation. Do not vary the manipulated variables for at least n time intervals
(u(-1)=u(-2)=...= u(-n)=0) and assume the measured disturbance changes are zero
(d(-1)=d(-2)=...= d(-n)=0) during that time. Then the system will be at rest at k=0.
2. Initialization (k=0). Measure the output y(0) and initialize the model prediction vector as
Y (0 | 0)
[y(0)T , y(0)T ,..., y(0)T ]T
(2.40)
Measure d(0), obtain measurement at next sampling time ( y(1) , d(1)); and set k=1.
3. Model Prediction
Y (k | k
1)
MY (k
1|k
1)
S d d(k
1)
50
S u u(k
1)
(2.41)
where
Y (k | k
1)T , y(k
1) [y(k | k
1)T ,..., y(k
1|k
2|k
1)T , y(k
1|k
1)T ]T
(2.42)
Correction:
Y (k | k )
Y (k | k
KF (y(k )
1)
y(k | k
1)
(2.43)
4. Compute the p-step-ahead prediction vector:

Y(k
1 | k)
MY(k | k )
Sd d(k )
Su u(k )
(2.44)
where
Y(k | k
1 | k )T , y(k
1) [y(k
u(k
estimated for
[ u(k )T , u(k
u(k )
2 | k )T ,..., y(k
0;
1)T ,..., u(k
5. Obtain measurements y(k
i)
1 | k )T , y(k
p | k )T ]T ;
(2.45)
m;
1)T ]T .
d(k+1) at the next sampling time, set k:=k+1 and go to
1) ,
step 3.
Least Squares Solution of the Control Problem
The algorithm presented before to compute the p-step-ahead prediction vector for m present and
future input moves allow expressing the control problem as an optimization problem of the form:
min{||
u (k )
s.t.
Y(k
[Y(k
1 | k)
1 | k)
R(k
MY(k | k )
1)] ||2
||
Sd d(k )
u(k ) ||2 }
Su u(k )
(2.46)
where
u
diag { 1u ,...,
u
m
diag { 1y ,...,
y
p
(2.47)
};
are the weight matrices in block diagonal form.

This optimization problem can be solved by the least squares algorithm. For this the problem can be
rewritten as the following set of linear equations:
51
Su
0 R(k
I
e(k
e(k
1 | k)
2 | k)
u(k) =
1)
MY(k | k )
0
Sd d(k )
E p (k 1 | k )
0
(2.48)
where:
E p (k
1 | k)
R(k
:
e(k
1) [MY(k | k )
Sd d(k )].
.(2.49)
p | k)
Ep(k+1|k) is the measurement corrected vector of future output deviations from the reference
trajectory (i.e. errors), assuming all future moves are zero. Note that this vector contains the effect
of measurable disturbances (Sdd(k)) on the prediction.
For practical choices of p n and weights this system of equations is over specified: the total
number of independent equations is greater than or equal to the dimension of the vector of
unknowns u(k). That is, the total number of controlled variable projections is larger than the total
number of manipulated variable projections.
The vector of residuals of the system (2.48) is
y
[MY(k | k )
Sd d(k )
u
Su u(k )
R(k
1)]
(2.50)
u(k )
or
y
[Y(k
1 | k)
u
R(k
1)]
(2.51)
u(k )
It should be noted that minimizing the sum of squares of the residuals of (2.48) is equivalent to
solving the optimal control problem (2.46). The solution that minimizes the sum of squares of the
residuals of (2.48) is given by
u(k )
(SuT
yT
Su
uT
) 1 SuT
yT
Ep (k
1 | k)
(2.52)
At a given interval of time k, having implemented moves in the past and having measured the most
current values of the output variables y(k ) , a set of future manipulated variable moves u(k) up to
m intervals of time is computed. This set of moves into the future could be implemented as such
52
and the solution repeated after m intervals of time have elapsed. However, using this
implementation measurement information would be used only every m intervals of time.
The output measurement y(k ) is used for the correction of the present state estimate Y(k|k). Because
at time k future measurement information is not available, we judge it to be best to omit the
correction step in the p-step-ahead predictor. However, at time k+1 new measurement information
becomes available and can be used to obtain improved state estimates Y(k+1|k+1) and output
predictions Y(k+1|k+1). Thus the available measurement information is better used if only the
moves corresponding to the present time are implemented and if the whole procedure is repeated at
the next time interval when a new measurement becomes available. That is, as a new measurement
is obtained, a new value for Y (k+1|k+1) is computed, accounting for measurement errors and
changes in disturbances. Such a controller implementation is denoted as the moving horizon or
receding horizon approach.
Moving Horizon and Closed Form Control Law
As explained above, the MPC control problem is solved at each time interval to allow the controller
to compensate effectively for disturbances on the outputs and for model errors. The resulting
control law becomes:
u(k )
[I 0 0...0] (SuT
yT
Su
uT
) 1 SuT
yT
E p (k
1 | k)
(2.53)
where only the first move (at time k) is computed. Note that the only part of the algorithm that
changes at each execution of the controller is the projected error Ep(k+1|k). Therefore, for
implementation the constant matrix pre-multiplying Ep(k+1|k) can be computed off-line:
KMPC
[I 0 0...0] (SuT
yT
Su
uT
) 1 SuT
yT
(2.54)
The on-line algorithm is summarized in the following:

1. Preparation. Do not vary the manipulated variables for at least n time intervals
(u(-1)=u(-2)=...=u(-n)=0) and assume the measured disturbance changes are zero
(d(-1)=d(-2)=...=d(-n)=0) during that time. Then the system will be at rest at k = 0.
2. Initialization (k = 0). Measure d(0) and the output y(0) and initialize the model prediction
vector as
Y (0 | 0)
[y(0)T , y(0)T ,..., y(0)T ]T .
(2.55)
53

Obtain measurement at next sampling time ( y(1) , d(1)). Set k = 1.
3. Model Prediction:
Y (k | k
1)
M SY (k
1|k
1)
Su
u(k
1)
Sd
d(k
1)
(2.56)
where the first element of Y(k|k-1), y(k|k-1), is the model prediction of the output at time k.
The value of u(k-1) should be the actual value implemented on the plant, rather than the
value computed by the controller at the last time step. This value may be different because
of actuator saturation.
Correction:
Y (k | k )
Y (k | k
KF (y(k ) y(k | k
1)
1).
(2.57)
4. Compute the reference trajectory error vector:

Ep (k
1 | k)
R(k
1)
MY (k | k )
Su d(k ).
(2.58)
5. Compute the manipulated variable move:

u(k )
KMPC Ep (k
1 | k)
(2.59)
which is implemented on the plant.

6. Obtain measurements y(k
1) ,
d(k+1) at next sampling time, set k:=k+1 and go to step 3.
The dynamic behaviour of this algorithm will be determined by the choice of adjustable parameters.
These include the weights y and u and the number of moves m. These parameters will influence,
among other things, the speed of response and specifically, the stability of the inputs and outputs
(i.e. whether the algorithm produces bounded u and/or y signals). Note that, if any of these
parameters needs to be changed on-line, it is necessary to recompute the matrix KMPC, thus
imposing additional on-line computational requirements.
2.3.10 General tuning guidelines of MPC

The most significant tuning parameters that must be selected for MPC controllers are prediction
horizon, control horizon, sampling interval, penalty weight matrices, and if the control engineer
decided to use a filter, the parameters of this.
The choice of these parameters has a profound effect on the nominal stability, robustness, and
controller performance of the MPC algorithms. In practical applications the most important criteria
54
that have to be satisfied by controllers are: stability and robustness. Thus, ideally one should know
a range of the control parameters, which provide stability and robustness, and then select from this
interval the values for the parameters, which give the best control performance in accordance to
certain control objectives. For linear systems there are developed algorithms to obtain the sufficient
conditions, which guarantee nominal stability and robustness (Garcia and Morari, 1985a; Garcia
and Morari, 1985b; Shridar and Cooper, 1997). However for nonlinear systems the known
sufficient conditions (Allgower, 1999) are usually much too strong to be met for practical
implementation and one must resort to a set of heuristics based on the extrapolation of linear
systems, simulations and experiments.
According to this, the following effects of the tuning parameters on the control performance were
observed from the author experience and the examples from the literature:
Prediction horizon. Usually longer prediction horizon leads to more aggressive control action and
faster response. With longer prediction horizons the closed-loop system is less robust to modelplant mismatch, however this effect can be reduced by including a filter in the feedback loop.
Nominal stability is strongly affected by the horizon length. Usually there is a critical minimum
horizon length to achieve a stable closed-loop system.
Control horizon. When the number of control moves is increased a more aggressive control action
can be observed. In this case the system response is faster and more sensitive to disturbances.
Additionally, increasing the control horizon leads to much more complicated optimization
problems, especially in the nonlinear case. Usually there is an upper band on the control horizon
established by the controller performance and the computational complexity of the problem.
Sampling time. To ensure good closed-loop performance, the sampling time should be small
enough to capture adequately the dynamics of the process, and at the same time large enough to
permit the feasibility of real-time implementation (the computational time necessary to solve one
open-loop control problem should be smaller than the sampling interval). Zafiriou and Morari
(1986) proposed some criteria to select the sampling interval for stable linear SISO systems
(Zafiriou and Morari, 1986). For unstable systems robustness depends on the sampling time and
there is an inverse relationship between the model error and the maximum allowable size of the
sampling time.
Weight matrices. Some attempts to obtain these parameters have been presented in the literature
(Kalra and Georgakis, 1994). The results of the author simulations show that the smaller the weight
on the control inputs the better is the control performance with regards of the controlled outputs.
55
However, to avoid aggressive control action small penalties (usually 5-15 % of the output penalty)
on the control action should be used, too. Finally there is the possibility of making the weights time
varying. Because there are no easy guidelines this is rarely pursued in practice.
Feedback filter. The use of a well-tuned filter on the feedback signal provides good disturbance
rejection and fast systems response. However, the choice and the effects of the filter are strongly
dependent on the certain system (Luenberger, 1971). An example of the effects of the Extended
Kalman Filter (EKF) used in the NMPC of a high purity distillation column is presented later.
2.4 Interaction analysis and multiple single-loop designs

2.4.1 The Relative Gain Array (RGA)
In the control of multi-parameter plant, a common technique is the Relative Gain Array (RGA),
which was first developed by Bristol (1966), with other workers (Shinskey, 1979, 1984; McAvoy,
1983) contributing to its further development and wider adoption. The relative gain array forms the
basis of the analysis, and is defined as a matrix containing all possible combinations of Single Input
Single Output (SISO) variable pairs expressed in the form of interactive measures (Bristol, 1966).
As such, considering the defined interactions for control of decentralised (multiloop SISO) systems,
a set of preferred variable pairs are indicated in the RGA. While the RGA technique was initially
developed for steady state systems, it was relatively easy to account for dynamic systems (Witcher
and McAvoy, 1977).
However, with further improvements, the RGA included other than simple interaction measures,
and also included fundamental property data, i.e. robustness against errors in the model and
uncertainties in the input, as well as stability of the closed loop (Grosdidier et al., 1985; Yu and
Luyben, 1987; Skogestad and Morari, 1987).
The RGA has been well described, in terms of properties and uses (Hovd and Skogestad, 1992),
and while it enjoys wide use, suffers a number of disadvantages, where interactions between control
loops may occur in the steady and dynamic states, but these are not captured in the RGA (Friedly,
1984). This is explained in that the RGA does not provide the true closed loop interaction measures
(Jensen et al., 1986). In addition, for greater than 2x2 systems, the RGA is not capable of providing
the response satisfying the variable pairing problem (Bristol, 1979), and is the case where there is
total dependence on other control loops of feasibility of the variable pairing (Shinskey, 1984;
Haggblom, 1994).
56
The influence of other output variables on the transfer function governing the input and output
variables (uj) and (yi) respectively has been defined in a proposed interaction measure (Bristol,
1966). This interaction measure represented by the ratio of the two transfer functions; one defines
the relation between the given two input and output variables, while others are uncontrolled, and
the other transfer function of the same two variables, but with all others controlled, denoted by
ij
where:
yi
uj
ij
gain with other loops open
uk constant,k j
uj
(2.60)
gain with other loops closed
yi
yk constant,k i
The relative gains
ij
for all possible variable pairings define a matrix, the relative gain array
(RGA), . The partial derivatives in eq. (2.60) can be related to the open-loop transfer functions of
a system. Consider a 2x2 system described by the model:
y1 (s) G
11 (s)u1 (s) G
12 (s)u2 (s)
(2.61)
y2 (s) G
21 (s)u1 (s) G
22 (s)u2 (s)
(2.62)
The transfer function between y1 and u1 with y2 uncontrolled (i.e., u2 = 0) is G11. The corresponding
transfer function when y2 is perfectly controlled is obtained by the elimination of u2 with y2 = 0.
This gives the relative gain:
11
(s )
G12 (s )G21 (s )
(2.63)
G11 (s )G22 (s )
The value for
must be computed for every possible combination of i and j. After every value is
found, the RGA matrix can then be formed:

12 1n
22 2n
21
11
RGA
(2.64)
n1
n 2 nn
57
Interpreting the RGA

There are certain important properties and guidelines used in understanding and analysing the
RGA, and what the different values of the RGA mean:
1. All elements of the RGA across any row, or down any column will sum up to one:
n
n
ij
ij
i 1
(2.65)
j 1
This makes calculating the RGA easier because:

in a 2x2 case, only 1 element must be calculated to determine all elements,
in a 3x3 case, only 4 elements must be calculated to determine all elements and so on.
2. The
ij
calculated from a steady-state matrix is dimensionless and unaffected by scaling.
3. Each of the rows in the RGA represents one of the outputs. Each of the columns represents
a manipulated variable.
If
ij
= 0: The manipulated variable (mj) will have no effect on the output or the controlled
variable (yi).
If
ij
= 1: The manipulated variable mj affects the output yi without any interaction from the
other control loops in the system. From the definition of
ij
this implies that the gain loop
with all loops open is equal to the gain loop with all other loops closed: i.e. g11 = g*11.
ij
< 0: The system will be unstable whenever mj is paired with yi, and the opposite
response in the actual system may occur if other loops are opened in the system.
f 0 <
ij
< 1: This implies that other control loops (mj- yi) are interacting with the
manipulated and controlled variable control loop.

Three different relationships based on =0.5 imply different interpretations of pairing and the RGA:
If
ij
= 0.5: The control pairing effect is equal to the retaliatory effect of other loops.
If
ij
< 0.5: The other control loops are influencing the control pair, and the influence of the
other control loops is greater than the influence of the control pair.
If
ij
> 0.5: This means that the control pair has a greater influence on the system than the
other control loops.
58
If
ij
>1: The open-loop gain of the control pair is greater than the gain with all other loops
closed: i.e. g11 > g*11. The positive value of RGA indicates that the control pair is dominant
in the system, but the other loops are still affecting the control pair in the opposite
direction. The higher the value of
ij
the more correctional effects the other control loops
have on the pair (Haggblom, 1995).

Table 2.4: Summary of the interpretation of the RGA.
Possible Pairing
ij
ij
ij
=0
Avoid pairing mj with yi
ij
=1
Pair mj with yi
ij
<0
= or < 0.5
>1
Pair mj with yi
ij
2.5 Nonlinear Model Predictive Control (NMPC) and industrial

applications
2.5.1 Introduction
During in the last decade nonlinear model predictive control (NMPC) techniques have become
increasingly important and accepted in chemical industries. The NMPC paradigm encompasses a
significant number of different approaches, each one with its own special feature, all NMPC
techniques rely on the idea of generating values for process inputs as solutions of an on-line (realtime) optimization problem using a nonlinear process model. The nonlinear dynamic model can be
used in different ways and in different phases of the control algorithm depending on the particular
nonlinear model predictive control approach. Excellent reviews have been published in the
literature recently, indicating the current and continuously increasing interest directed toward this
control strategy (Allgower et al., 1999; Qin and Badgwell, 2000; Henson, 1998; Imsland et al.,
2003; McAvoy, 2002; Nikolaou and Michael, 2001; Poloski et al., 2003; Mayne, 2000). Our
intention is not to review these reviews here. The objective of this is rather to provide a motivation
59
of the study, identify potential benchmark processes categories, and to bring up key problems that
need to be addressed in industrial NMPC applications. For details about the applications and
NMPC products the reader is referred to the aforementioned review papers and the references
therein.
In the wide variety of chemical processes nonlinearity is the rule rather than the exception. There
are processes which present many challenging control problems including: nonlinear dynamic
behaviour, multivariable interactions between manipulated and controlled variables, unmeasured
state variables, unmeasured and frequent disturbances, high-order and distributed processes,
uncertain and (variable) dead time on inputs and measurements. Further, reliable measurements of
important variables to be controlled, such as quality related variable, are often difficult to obtain online. The economic benefits of applying advanced process control (APC) approaches in chemical
industry had been widely recognized, first in academia but nowadays in industry, too. The
theoretical economical optimal operating condition of a chemical process usually lies on active
constraints. Therefore in practice the operating region has to be chosen such that constraints are not
violated even in the case of strong disturbances. Quality of control determines how close the
process can be pushed to the boundary. APC approaches allow the tighter control of process
variables, hence permitting the operation of processes closer to the limits, yielding higher profit. A
simple graphical explanation of the economical advantages of APC is shown in Figure 2.18 and
2.19. The schematic representation on Figure 2.19 shows that the optimal operating region given by
APC is usually on active constraints but provides higher quality with lower variability then typical
operating regions with classical control approaches.
Limit
Safety
Margin
= Profit
Target
No APC
APC
reduces
variation
Reduced variation
allows
operation
closer to limit
Figure 2.18: Economical benefit given by the use of APC.
60
Optimal operating region given by

APC
Typical operating
region (no APC)
dT/dt
P
Temperature
Pressure
Quality
Figure 2.19: Typical operating regions with and without APC.
A number of APC approaches and algorithms that are able to handle some of the aforementioned
process characteristics have been presented in the recent years. Many of these approaches are not
able to handle the various process characteristics and requirements met in industrial applications
resulting in a large gap between the number of industrial and academic NMPC products. Although,
it is well recognized, that the performance of a control system is mostly inherent in how
successfully it can cope with the nonlinearity of the process, chemical processes have been
traditionally controlled by using algorithms based on a linear time-invariant approximate process
model the most common being step and impulse response models derived from the convolution
integral. One reason for this is that in most LMPC applications reported the goal is largely to
maintain the process at a desired steady-state (regulator problem, disturbance rejection) (e.g. in
refinery processing), rather than moving rapidly from one operating point to another (setpoint
tracking problem). A carefully identified linear model, which usually can be identified in a fairly
straightforward manner from process test data, is sufficiently accurate in the neighborhood of a
single operating point. In addition, by using a linear model and a quadratic objective, the nominal
MPC algorithm takes the form of a highly structured convex Quadratic Program (QP), for which
reliable solution algorithms and software can easily be found. This is important because the
solution algorithm must converge reliably to the optimum in no more than a few tens of seconds to
be useful in manufacturing applications. For these reasons, in many cases a linear model will
provide the majority of the benefits possible with MPC technology (Bequette, 1991; Morari and
Lee, 1997; Lee and Cooley, 1996).
61
Nevertheless, there are cases where nonlinear effects are significant enough to justify the use of
nonlinear model predictive control (NMPC) technology. These include at least two broad categories
of applications (Qin and Badgwell, 2000); i.e., disturbance rejection control problems where the
process is highly nonlinear and subject to large frequent disturbances (pH control, bioreactor, etc.)
(Galan et al., 2000; Nagy and Agachi, 1998; De Oliveira, 1996), or setpoint tracking problems
where the operating points change frequently and incorporate a sufficiently wide range of nonlinear
process dynamics (batch process control, start-up problems, polymer manufacturing, etc.) (Muske
et al., 2000a; Muske et al., 2000b; Nagy and Agachi, 1997; Brengel and Geider, 1989).
A rough distribution of the number of MPC applications versus the degree of process nonlinearity
is shown on Figure 2.20, (Qin and Badgwell, 2000). MPC technology has not yet penetrated deeply
into areas where process nonlinearities are strong and market demands require frequent changes in
operating conditions. It is these areas that provide the greatest opportunity for NMPC applications
(Morari and Lee, 1997; Qin and Badgwell, 2000).
MPC applied
MPC not yet applied
Number of MPC applications
Refinery
Petrochemical
Chemicals
Polymers
Gas plants
Pulp & paper
Process nonlinearity
Figure 2.20: Distribution of MPC application versus the degree of process nonlinearity.
62
2.5.2 Generic NMPC problem formulation

A direct extension of the LMPC methods results when a nonlinear dynamic process model is used,
rather than the linear convolution or state space model.
Nonlinear model predictive control is an optimization-based multivariable constrained control
technique using a nonlinear dynamic process model for the prediction of the process outputs
(Allgower et al., 1999; Bequette, 1991). At each sampling time the model is updated on the basis of
new measurements and state variable estimates. Then the open-loop optimal manipulated variable
moves are calculated over a finite prediction horizon with respect to some cost function, and the
manipulated variables for the subsequent prediction horizon are implemented. Then the prediction
horizon is shifted by usually one sampling time into the future and the previous steps are repeated.
This generic moving horizon approach of model predictive control algorithms (both nonlinear and
linear) is given in Figure (2.16).
The optimal control problem to be solved on-line in every sampling time in the NMPC algorithm
can be formulated as:
min
u (t )
(x (t ), u(t ), p)
(2.67)
subject to:
x (t )
f (x (t ), u(t ), p) ,
(2.68)
y(t )
g(x (t ), u(t ), p) ,
(2.69)
x (tk )
x(tk ),
x0 ,
x (t0 )
h(x (t ), u(t ), p)
0,
(2.70)
[tk , tF ] ,
(2.71)
where is the performance objective, t is the time, tk is the time at sampling instance k, tF is the
final time at the end of prediction, x (t )
vectors with
ny
input, y(t )
p
nu
np
nx
is the nx vector of states, u(t) is the nu set of input
representing the set of all possible trajectories of each control
is the ny vector of measured variables used to compute the estimated states x(tk ) ,
is the np vector of uncertain parameters, where the set can be either defined by hard
63
bounds or probabilistic, characterized by a multivariate probability density function. The function

f :
nx
nx
g:
nx
ny
is the vector function of the dynamic equations of the system,

is the measurement equations function, and h :
nx
is the
vector of functions that describe all linear and nonlinear, time-varying or end-time algebraic
constraints for the system, where c denotes the number of these constraints. The objective function
can have the following general form:
(x (t ), u(t ), p)
(x (tF ), p) +
tF
tk
(x(t ), u(t ), p)dt
(2.72)
The form of (2.72) is general enough to express a wide range of objectives encountered in NMPC
applications (moving or shrinking horizon approach on regulation and/or setpoint tracking, direct
minimization of the operation time, optimal initial conditions, multiple simultaneous objectives,
treatment of soft constraints, etc.). For batch processes with end-point optimization the objective
usually reduces to the Mayer form ( ( )
0 ),
however the Lagrange term ( ( ) ) still may be used,
e.g. to implement soft constraints on control rate.
2.5.3 Industrial implementation of NMPC

In the past 20-30 years LMPC has become a preferred control strategy for a large number of
processes mainly due to the reasons mentioned in the introduction. While the development of
LMPC approaches and industrial products has reached the fourth generation with more than 4500
applications (see Figure 2.21), there are only few NMPC providers (Table 2.6) with rather limited
number of applications (Table 2.6).
Table 2.5 lists several commercial NMPC products and the companies supplying them. This list is
by no means exhaustive. There are several other approaches but these are either mainly applications
to a particular process or the NMPC products are too new (Fischer-Rosemount, ABB, Ipcos, etc.)
and were not included here due to lack of information. However the technology sold by the
companies in Table 2.6 is representative of the current state-of-the-art.
64
Refinig, 44%
Others, 35%
Petroch.
12%
Refinig
Petrochem
Chemicals
Pulp&Paper
Air&Gas
Utility
Mining
Food
Polymer
Furnaces
Aerospace
Automotive
Others
Figure 2.21: Summary of industrial LMPC applications by areas.
Table 2.5: NMPC companies and product names.
Company
Product Name (Acronym)
Adersa
Predictive Functional Control (PFC)
Aspen Technology
Aspen Target (Apollo)
Continental Controls
Multivariable Control (MVC)
DOT Products
NOVA Nonlinear Controller (NOVA-NLC)
Pavilion Technologies
Process Perfecter
An excellent review and description of these NMPC products is given in (Qin and Badgwell, 2000).
Table 2.6 provides information on the details of each algorithm, including the model types used,
options at each step in the control calculation, and the optimization algorithm used to compute the
solution.
65
Table 2.6: Comparison of industrial NMPC control technology.
Company
Adersa
Aspen
Technology
Continental
Controls
DOT
Products
Pavilion
Technologies
Algorithm
PFC
Aspen Target
MVC
NOVA-NLC
Process
Perfect
Model Forms1
NSS-FP
S ,I, U
NSS-NNN
S, I, U
SNP-ARX
S
NSS-FP
S, I
NNN-ARX
S, I, U
Feedback2
CD, ID
CD, ID, EKF
CD
CD
CD, ID
Rem. Ill-cond.3
IMS
IMS
IMS
SS. Opt. Obj.4
Q[I,O]
Q[I,O]
Q[I,O]
Q[I,O]
SS. Opt. Const.5
IH, OH
IH, OH
IH, OS
IH, OH, OS
Q[I,O], S
Q[I,O,M]
Q[I,O,M]
(Q,A)[I,O,M]
Q[I,O]
Dyn. Opt.
Const.7
IA, OH, OS,

R
IH, OS-l1
IH, OS
IH, OH, OS
IH, OS
Output Traj.8
S, Z, RT
S, Z, RT
S, Z, RT
S, Z, RTUL
S, Z, TW
Output Horiz.9
CP
CP
FH
FH
FH
Input Param.10
BF
MM
SM
MM
MM
NLS
QPKWIK
GRG2
NOVA
GRG2
Dyn. Opt. Obj.
Sol. Method
11
Model Form: (ARX) autoregressive with exogeneous inputs (input-output model), (FP) First-Principles,
(NSS) Nonlinear State-Space, (NNN) Nonlinear Neural Net, (SNP) Static Nonlinear Polynomial, (S) Stable,
(I) Integrating, (U) Unstable
2
Feedback: (CD) Constant Output Disturbance, (ID) Integrating Output Disturbance, (EKF) Extended
Kalman Filter
3
Steady-State Optimization Objective: (Q) Quadratic, (I) Inputs, (O) Outputs
4
Removal of ill-conditioning: (IMS) Input move suppression
5
Steady-State Optimization Constraints: (IH) Input Hard maximum, minimum, and rate of change constraints,
(OH) Output Hard maximum and minimum constraints
6
Dynamic Optimization Objective: (Q) Quadratic, (A) One norm, (I) Inputs, (O) Outputs, (M) Input Moves
7
Dynamic Optimization Constraints: (IH) Input Hard maximum, minimum, and rate of change constraints,
(IA) IH with input acceleration constraints, (OH) Output Hard maximum and minimum constraints, (OS)
Output Soft maximum and minimum constraints, (OS-l1) Output soft constraints with l1 exact penalty
treatment
8
Output Trajectory: (S) Setpoint, (Z) Zone, (RT) Reference Trajectory, (TW) Trajectory Weighting, (TRUL)
Upper and lower reference trajectories
9
Output Horizon: (FH) Finite Horizon, (CP) Coincidence Points
10
Input Parameterization: (SM) Single Move, (MM) Multiple Move, (BF) Basis Functions
11
Solution Method: (NLS) Nonlinear least squares, (QPKWIK) multi-step Newton method, (GRG2)
Generalized reduced gradient, (NOVA) Mixed complimentary nonlinear program
An approximate summary of industrial NMPC application is presented in Table 2.8 with the breakdown on
different areas of application shown on Figure 2.22.
66
Table 2.7: Approximate summary of NMPC applications by application areas.

Company
Area
Adersa
Aspen
Technology
Air and Gas
Continental
Controls
DOT
Products
Pavilion
Technologies
18
Chemicals
Total
18
15
22
15
21
Pulp and
Paper
Refining
13
13
Food
Processing
Polymers
Utilities
Unclassified
Total
36
43
93
Others 2%
Refining
Polymer,
23%
Refining
14%
Food 10%
Chemicals
Pulp&Paper
Air&Gas
Chemicals
23%
Air&Gas,
19%
Utility
Food
Polymer
Pulp&Pap.
1%
Others
Figure 2.22: Summary of NMPC applications by areas.
It is interesting to note that the distribution of NMPC applications has changed significantly
compared to LMPC applications. In refining where processes are operated around a steady-state
LMPC can do the job, and the additional burden related to NMPC is probably not necessary.
However in polymers and chemicals (e.g. pharmaceuticals) NMPC seem to be the right approach.
Half of the NMPC applications are in these two areas, while we can find only a very small
percentage of LMPC applications in these processes. The explanation for this could be in the fact
that these industries are dominated by highly nonlinear and unstable batch processes, where LMPC
67
approaches usually fail. The analysis of the LMPC approaches in these areas shows that those
applications are for continuous processes where LMPC approaches (sometimes with gain
scheduling) can work around the operating points. For example INCA (the product of IPCOS
Technology (Van Brempt et al., 2001) in the actual control calculations uses linear models carefully
identified for the operating points, approach which is however more difficult to use efficiently for
batch processes. Also for polymers and chemicals (e.g. pharmaceuticals) the objectives are in terms
of distribution control of the final product quality which is difficult to express in the LMPC
framework. An interesting observation from the analysis of the NMPC applications shows that
almost all reported applications (those which were found in the literature) are for continuous
processes and many of the presented products do not even work for discontinuous highly nonlinear
and unstable processes with large variation in process gain. The need of NMPC for continuous
polymerization processes can be explained by the frequent grade changeover operations of the
reactors in order to meet diversified demands of the market. However even in these cases relatively
simpler NMPC approaches based on linear model scheduling can work and are applied. In the case
of batch polymerization and certain fine chemical processes the physico-chemical properties of the
system (viscosity, density, heat capacity) can change dramatically during the process leading to
change in the gain sometimes even with two order of magnitudes. In these cases the controller has
to cover a wide range of operating conditions and cope with highly nonlinear process dynamics.
This motivates the use of batch polymerization (Park et al., 2003; Pan and Lee, 2003; Sekia et al.,
2001; Dulce and Silva, 2002) or fine chemical processes (Le Lann et al., 1999) as good benchmark
problems for NMPC assessments. These processes also allow a broad variation in the control
problem formulation from simple setpoint tracking to shrinking horizon online optimizing control,
or end-point performance control, product property distribution control, etc. Although the above
statistics shows the contrary, it should be mentioned however, that in the fine chemicals industry
frequent changes in products, lower production rates can make the implementation of NMPC more
difficult to be accepted, due the high rate of return needed in the case of short product lifetime.
2.5.4 Challenges in industrial NMPC

The flow of tasks which need to be performed in generic industrial NMPC applications is presented
in Figure 2.23. Each block represents its specific problem which need to be discussed in more detail
and can represent the main framework of industrial NMPC assessments. Even the link of the
NMPC to the process through the input/output devices (I/O) is an important issue where
communication protocols between process and controller or even between different task-blocks
68
within the NMPC need to be compared (OPC, DDE, UDP, TCP/IP, etc.). Obviously not all blocks
are present for all NMPC applications. However if the goal is the development of generic NMPC
tools adequate for SISO, or thin (CVs > MVs), fat (CVs < MVs), or square (CVs = MVs) MIMO
plants all components need to be considered, and those components which are not needed for a
certain application will be turned off. For example for SISO control systems the determination of
process subsets and ill-conditioning is not an issue and the corresponding modules will be bypassed in the controller.
Setpoint optimization
(e.g. steady state)
Dynamic
optimization
Remove process
ill-conditioning
NMPC
I/O: Implement MVs
Lower level
controllers
Determine controlled
process subsets
Output feedback
(State and disturbance
estimation)
I/O: Read MVs, DVs, CVs
Plant
Figure 2.23: Sequence of tasks involved in a generic NMPC application.
Some of the major challenges related to industrial NMPC applications in my opinion are:
Efficient development and identification of control-relevant model. The importance of

modeling in NMPC applications is straightforward. Unlike traditional control where
modeling and analysis represent a small part of the effort in the application development, it
is estimated that up to 80% of time and expense in designing and installing a NMPC is
attributed to modeling and system identification. Figure 2.24 schematically represents the
design effort involved in NMPC design versus traditional control.
All LMPC and most of the NMPC products use empirical models identified through plant
tests. Figure 2.25 shows the distribution of industrial MPC algorithms based on the model
type used. It is attractive to use first-principles (FP) models (in following section details for
FP), which give the most inside about the process. First principles models are globally valid
and therefore well suited for optimization that can require extrapolation beyond the range
of data used to fit the model. Despite the clear advantages of using FP models most of the
NMPC approaches are based on empirical nonlinear models. The reasons for this are in the
difficulty of the development of proper process models and in the increased computational
69
NMPC
Effort
Traditional Control
Design and
Modeling
tuning
and
of controller
identification
Modeling
and anlaysis
Control
specification
Figure 2.24: Design effort of NMPC versus traditional control.
burden required for the solution of FP models. Since the modelling is a time consuming
part of the NMPC design the choice of proper modelling environment is crucial. Almost all
industrial NMPC approaches try to beneficiate form the power of chemical modelling
software (ASPEN, gPROMS, Hysys, etc.) rather than building and solving the models from
scratch. The model identification is important even if first principle models are used. In this
case usually offline parameter identification has to be performed. It is very important to
keep in mind (however often overlooked) that models (whether empirical or FP) are
imperfect (both in term of structure and parameters). Therefore robustness is a key issue in
NMPC applications. Robustness has also been identified as one of the major deficiencies of
current NMPC products (actually none of the products presented here has any systematic
treatment of robust performance except the inherent robustness due to feedback, and
penalization of excessive control movements). How to choose the proper plant tests to
identify the best model is an important question (Gopaluni et al., 2003, Shouche et al.,
2002). For example many vendors believe that the plant test is the single most important
phase in the implementation of DMC-plus controllers. Optimal experimental design even if
it is not the, but it can be a answer/recipe to the model identification question. Additionally,
robust NMPC design can lead to significant performance improvement (Nagy and Braatz,
2003; Wang and Romagnoli, 2003). Another important problem which need to be assessed
in case of FP model based NMPC approaches, is the tradeoff between the model accuracy
(hence complexity) and computational requirement. Often the control-relevant model does
not necessarily have to be the most accurate and finding the proper balance between the
70
accuracy of the model and computational burden is usually very challenging. Including
many details into the model can lead to large number of states resulting in unobservable
models based on available measurements besides the prohibitively large computational
burden. The determination of the control-relevant model always has to be done in
conjunctions with the observer design.
Nonlinear
Aspen Target MVC

Process Perfecter
PFC
NOVA-NLC
First
Principles
Empirical
DMC plus
HIECON PMPCT
Connoisseur
SMOC
Linear
Figure 2.25: Classification of model types used in industrial MPC algorithms.
State and parameter estimation. The lack of reliable sensors is one of the major bottlenecks
in industrial NMPC applications. This problem is crucial when FP models are used
(Allgower et al., 1999). This problem needs to be assessed in detail in the project. In many
industrial applications software sensors need to be developed based on additional empirical
models (most often neural networks are used for this purpose, e.g. INCA, (Van Brempt et
al., 2001).
Managing the on-line computational load. In NMPC approaches exorbitant computational

load is involved. The use of efficient optimization approach is crucial (Biegler, 2000;
Biegler, 1984, Biegler and Rawlings, 1991; Bock et al., 1999; Hassapis, 2003). There are
other important problems here which need to be discussed such as robustness of
optimization, or the choice of proper objective functions and optimization problem setup
for efficient solution. These will be discussed in more detail in the final report.
Design as an integrated system. In industrial NMPC optimization is involved in several

steps of the controller presented in Figure 2.23 (steady-state optimization, dynamic
optimization, state estimation, and model identification). An efficient approach to design
71
model, estimator (of model parameters and states), and optimization algorithm as an
integrated system (that are simultaneously optimized) rather than independent components
could be the capstone in NMPC design.
Long-term maintenance of control system. It is clear the implementation complexity of

NMPC approaches is high. But it is important to assess how long term maintenance can be
performed and what the limits of the approach in face of changing process and operating
conditions are.
2.5.5 First principle (analytical) model based NMPC

The objective of NMPC is to calculate a set of future control moves (control horizon, Tc or M) by
minimization of a cost function, like the squared control error on a moving finite horizon
(prediction horizon, TP or P). The optimization problem is solved on-line based on prediction
obtained from a nonlinear model. One can use different empirical nonlinear models for prediction
in the controller, but the most attractive approach is the use of first principle models, which are
globally valid and therefore well suited for the optimization that can require extrapolation beyond
the range of data used to fit the model (De Oliveira and Biegler, 1994; Lee, 1998).
A general mathematical formulation of the nonlinear model predictive control problem, when the
process is described by ordinary differential equations (ODEs), is:
objective function:
min J x t , u . , P, M
(2.73)
u ( ),P ,M
constraints:
dx
dt
x (tk )
0
yp
f x, u, q, d
(2.74)
x(tk )
(2.75)
g1 x
g2 x
umin k
u k
(2.76)
(2.77)
i
i
u k
1
umax
umax k
u k
(2.78)
u k
umax
72
(2.79)
u k
u k
x min k
x k
ymin k
yp k
1
i
x max k
i
for all
1, P
(2.80)
ymax k
(2.81)
(2.82)
Where x = state variables, u = manipulated variables, q = parameters, d = measured and

unmeasured disturbances, y = output variables.
Generally, the prediction and control horizon, respectively, are considered fixed for an open loop
optimization. The objective function usually is chosen as the sum of the squares of the differences
between the predicted outputs and the setpoint values over the prediction horizon of P time steps:
tk TP
J x (t ), u( )
Q E dt
J x (k ), u( )
Qi r k
yp k
(2.83)
i 1
tk
continuous form
discrete form
Often the objective function (2.83) includes a second term, which is the squared sum of the
manipulated variable changes over the control horizon (M):
P
J x (t ), u( )
Qi r k
yp k
Ri u k
i 1
(2.84)
i 1
The second term was originally introduced by the unconstrained formulation of LMPC in which
constraints are handled artificially through the weighting factors (matrices Qi and Ri). Since the
constrained LMPC algorithms likewise the NMPC methods explicitly include constraints there is
no need of the second term in the objective function, however in many of the constrained LMPC
and NMPC applications authors use this term, too. Instead of, or additionally to the second term of
the objective function, we can assure a smooth control action by introducing another term to
minimize the deviations of the manipulated inputs from their setpoints. In this way a more general
formulation of the performance function is obtained and the optimization problem to be solved at
each sampling time can be written as follows:
P
uk
Qi r k
min
uk M 1
i 1
yp k
Ri u k
i 1
Rl' u k
u ref k
i 1
(2.85)
The predicted values of the output variables (y-p) can be considered equal to the value obtained
from the model (ym), but usually a correction is made to reduce the cumulative error effect of the
73
measurement errors and the model/plant mismatch. The correction equation usually has the
following form:
yp k
ym k
K (k
i) ym k
yp k
(2.86)
The decision variables in the optimization problem expressed by equations (2.73)-(2.82) are the
control actions, M sampling time steps into the future (control horizon). Generally, 1 M P and it
is assumed that manipulated variables are constant beyond the control horizon (equation 2.80).
Although the optimization provides a profile of the manipulated input moves over a control horizon
(M), only the first control action is implemented. After the first control action is implemented, new
measurements are obtained which are used for the compensation of plant/model mismatch and the
estimation of unmeasured state variables. Finally the prediction horizon is shifted by one sampling
time into the future and the optimization is performed again.
In the NMPC approaches absolute (2.78) and velocity (2.79) constraints as well as state- and
output-variable constraints are explicitly included.
Since a constrained nonconvex nonlinear optimization problem has to be solved on-line, the major
practical challenge associated with NMPC is the computational complexity that increases
significantly with the complexity of the models used in the controller. There has been a significant
progress in the field of dynamic process optimization. Fast on-line optimization algorithms that
exploit the specific structure of optimization problems arising in NMPC have been developed and
real-time applications have been proven to be feasible for small-scale processes. However, the
global solution of the optimization cannot be guaranteed and the development of fast and stable
optimization techniques is one of the major objectives in the NMPC research (De Oliveira and
Biegler, 1995).
2.5.6 Nonlinear model predictive control with guaranteed stability

Besides the constraints described in the previous section, another important requirement that the
nonlinear model predictive controllers have to meet (however seldom taken into consideration for
practical implementation) is that it should assure a stable closed-loop system.
The most straightforward way to achieve guaranteed stability is to use an infinite horizon cost
functional (P = M =) (Campo and Morari, 1986; Keerthi and Gilbert, 1988). In this case, from the
Bellmans principle of optimality results that the open loop input and sate trajectories obtained as
the solution of the optimization problem are equal to the closed loop trajectories of the nonlinear
74
system. Consequently, any feasible predicted trajectory goes to the origin. However, using infinite
horizon in the performance criterion leads to practically unsolvable optimization problem. To cope
with this disadvantage, and guarantee stability, besides the input and state constraints, so-called
stability constraints have to be included into the finite horizon open-loop optimization problem
(Chen and Allgower, 1996; Chen and Allgower, 1998a; Findeisen and Allgower, l999; Findeisen
and Rawlings, 1997; Sistu and Bequette, 1996).
The most widely suggested stability constraint is the terminal equality constraint, which forces the
states to be zero (equal to their steady state values) at the end of the finite horizon:
x t
TP
(2.87)
Using the terminal equality constraint to guarantee stability is an intuitive approach, however it
increases significantly the on-line computation necessary to solve the open loop optimization
problem and often causes feasibility problems (De Nicolao et al., 1996; De Nicolao et al., 1997; De
Nicolao et al., 1998).
Another approach to guarantee stability is the so-called quasi-infinite horizon nonlinear MPC
(QIHNMPC), in which the prediction horizon is approximately extended to infinity by introducing
a terminal penalty term in the objective function (Chen and Allgower, l997; Chen et al., 1997; Chen
et al., 1998; Findeisen and Allgower, 2000a; Mayne and Michalska, 1990; Michalska and Mayne,
1993). The basic idea of this approach consists of the approximation of the infinite horizon
prediction to achieve closed-loop stability, while the input function to be determined on-line is of
finite horizon only. The terminal penalty term is determined off-line such that it bounds from above
the infinite horizon objective function of the nonlinear system controlled by a local state feedback
law in a terminal region (Chisci et al., 1996; Mayne, 1996; Mayne et al., 2000; Mayne and
Michalska, 1990; Meadows et al., 1995; Meadows and Rawlings, 1993).
2.6 Conclusions
For modern refiners, the catalytic cracker is the key to profitability in that the successful operation
of the unit determines whether or not the refiner can remain competitive in today's market. The
major objective of FCC units is to convert low-value, high-boiling feedstocks into valuable
products such as gasoline and diesel. The three dominant reactions are cracking, isomerisation, and
hydrogen transfer. The catalyst acid sites have a major influence on the reaction chemistry. The
introduction of zeolites into the FCC catalyst drastically improved the performance of the catalytic
cracker reaction products. The zeolite catalysts are active and selective. The higher activity and
75
selectivity translate into more profitable liquid product yields and additional cracking capacity. The
need to produce reformulated gasoline will increase demand for the shape-selective zeolite; the
additive technology is continuously expanding.
The benefits of the new catalyst technologies is the achievement of deep bottoms cracking with low
coke; high conversion and high yields of gasoline and light olefins. These benefits have been
verified in many successful commercial applications.
Selective modifications of the unit's components (feed injection, riser termination, air distribution,
catalyst cooling, stripper design), flexibility of operations, mechanical improvements are improving
the unit's reliability, and increasing the quantity and quality of valuable products, and operating
flexibility. However, these modifications also increase the complexity of controlling the highly
interacting and strongly nonlinear FCCU.
It is well recognized that the economic effectives on the performance of the control system applied
for its operation suitable control system can increase efficiency and reduce the environmental
impact of an FCCU.
This chapter provides an overview of the development of linear and nonlinear model predictive
control strategies .The detailed evaluation of the trends of industrial applications and preparation of
the model predictive control approaches indicate that linear and nonlinear MPC are suitable
technologies for the FCCU, with significant potential to increase the profit a ability and efficiency
of the unit.
The FCCU high degree of efficiency will continue to play a key role in meeting future market
demands, in which the application of advanced control approaches will be undoubtedly a key
element.
76
Chapter 3: Mathematical Modelling of a Fluid Catalytic Process
CHAPTER 3
MATHEMATICAL MODELING OF A FLUID
CATALYTIC PROCESS
3.1 Analytical mathematical model of the FCCU
In current refineries, the FCC unit plays a prominent role, producing gasoline and diesel, as well as
valuable gases, such as ethylene, propylene and isobutylene, from feedstocks that comprise
atmospheric gas oils, vacuum gas oils and hydrocracker bottoms. The significant economic role of
the FCC unit in modern-day petroleum refining has attracted great interest in academia and industry
in terms of developing and modelling control algorithms for efficient FCC application. Figure 3.1
depicts the unit to which the mathematical model relates. The main parts of the FCC unit that have
been modelled are:
1. Feed and preheat system
2. Reactor
3. Catalyst circulation lines
4. Regenerator
5. Air blower
6. Wet gas compressor
7. Main fractionators.
The modelling of this unit is sufficiently complex to capture the major dynamic effects that occur in
an actual FCCU system; it is multivariable, strongly interacting and highly nonlinear.
The FCCU dynamic model presented in this chapter includes the main systems as well as a kinetic
model based on a five-lump system, capable of predicting the yields of valuable products. While
the five-lump kinetic model is a significant simplification of the actual cracking kinetics; it is
sufficiently complex to describe the yields of valuable products and gasoline octane value. The
resulting global model of the FCCU is described by a complex system of partial-differentialequations, which were solved by discretising the kinetic models in the riser and regenerator on a
77
fixed grid along the height of the units, using finite differences. The resulted model is a high order
DAE, with 942 ODEs (142 from material and energy balances and 800 resulting from the
discretisation of the kinetic model).
The body of literature relating to the modelling of the FCC process is quite large, reporting strong
interactions and a variety of operating, security and environmental constraints. The prospect of
realising even greater yields of high-demand products through increased production and more
stable operation have motivated the search for models that are practical and more accurate, allied to
control strategies that are flexible, better performing, and more cost effective. Ford et al. (1976)
proposed a distributed parameter model of the FCC unit regenerator based on a detailed kinetic
combustion model. This was followed by Lee and Groves (1985), who modelled an FCC unit by
employing macroscopic models to describe the reactor and regenerator. However, it was only until
McFarlane et al. (1993) described a dynamic FCC unit model with constraints that a problem which
was challenging the chemical process control community was overcome.
Main fractionator
Reactor
Stack gas valve
4
V14
7
Regenerator
Steam
t
en
Sp
ta
ca
lys
Wet
gas
V11
Scsc
V7
Combustion
air blower
ed
at
er st
en aly
eg t
R ca
Steam
svreg
Wet gas
compressor
Riser
2
Feed preheat
furnace
V6
Fresh feed
oil
Atmospheric Air
F3
F2
F4
Slurry
recycle
Figure 3.1: UOP fluid catalytic cracking unit.
78
F1
F5
Furnace
fuel
Diesel
Wash
oil
Their work proposed a distributed parameter model for the regenerator, yet used a continuous
stirred-tank reactor (CSTR) without a yield model for the reactor section. In order to achieve
effective FCC optimisation, it is quite important that a detailed yield model, which will predict
gasoline octane value and yields of valuable products, is applied, as declared in the literature in
Section 2.2. In this context, there are many kinetic models, which can be categorised as: heavy
lumped models and molecular based models. Despite being more generic, molecular models require
analysis at a molecular level. Since such an analysis is very complex, one strategy has been to lump
different groups of molecules by boiling point. These are then treated as pseudo-components,
allowing the phenomena occurring in the reactor to be described globally. Regarding modelling for
the riser section, different kinetic schemes were proposed: from three kinetic lumps, up to thirteen
or more (Jacob et al., 1976). Secchi et al. (2001) dynamically simulated the UOP stacked FCC unit
by modelling the regenerator as a bubbleemulsionfreeboard model, while the riser was
represented by a 10-lump dynamic model. This was validated against experimental data obtained
from a working industrial unit. Han and Chung (2001) also proposed a detailed dynamic FCC
process simulator, including models of catalyst liftlines, stripper, feed preheater and cyclones. The
riser reactor was described as a distributed parameter 4-lump model and the regenerator as a tworegime, two-phase model.
Accurately modelling industrial FCC units is very complicated owing to the strong interactions
between process variables, the significant uncertainty surrounding the kinetics of the cracking
reactions, coke deposition leading to catalyst deactivation in the riser reactor, and the coke burning
process in the regenerator.
3.1.1 Feed system model

The fresh feed stream is transported through the preheat furnace; after exiting, slurry recycled from
the main fractionator bottom is mixed with it. The model presented by McFarlane et al. (1993) and
Cristea et al. (2003) assumed that, at all times, actual flows were equal to the controllers set point
by disregarding controller dynamics and those of the flow streams:
F1
F1set
(3.1)
F2
F2set
(3.2)
F3
F3set
(3.3)
F4
F4set
(3.4)
79
F1set
F3set
FC
FC
Preheater
Wash
oil
Fresh
feed
F1
V1
F5
F3
V3
F5
T1
F2set
FC
Diesel
F2
V2
F5
Gas oil
from
tankage
Figure 3.2: Feed system.
3.1.2
Preheat system model
According to McFarlane et al. (1993), the feed pre-heater is represented as a furnace with a fixed
flame temperature in the combustion chamber, regardless of spatial position. Moreover, air supply
to the pre-heater is assumed to be at the temperature of the surroundings.
Furnace
TO reactor
riser
T3
Fresh
feed
F5set
T2
FC
Fuel
gas
F4set
V5
F5
F5
FC
Slurry
recycle
V4
F5
F4
Figure 3.3: Preheat systems.
Based on the assumption that fresh feed, F3, enters the preheat furnace at temperature T1, the
dynamic energy balance returns furnace firebox and outlet temperatures using the following models
(McFarlane et al., 1993; Cristea et al., 2003):
dT3
1
F5H fu UAf Tlm Qloss

dt fb
(3.5)
80
T lm =
(T 3 - T 1) - (T 3 - T 2 )
(3.6)
ln T 3 T 1
T3 -T 2
Q loss = a 1F 5T 3 - a 2
F5
(3.7)
F5set
(3.8)
dT 2 = 1 (
T 2,ss - T 2 )
dt
fo
T 2,ss = T 1 +
(3.9)
UA fT lm
(3.10)
F3
Reactor Model
Figure 3.4 presents the two parts of the reactor: the riser and the stripper.
Gas phase
products
Reactor
Separator
Cyclone
Gas phase
product
+ catalysts
Steam
Spent catalyst
Riser
Regenerated
catalyst
Fresh
feed oil
Figure 3.4: Schematic diagram of the industrial riser reactor and the stripper.
In gasoline and olefin refinery production, FCC units are the main conversion units. FCC reactor
models typically represent the riser using lump-kinetic schemes with varying numbers of lumps to
simulate the two-phase (i.e. constant slip between phases) plug-flow transported bed. These models
are used to simulate FCC operation and predict product yields.
81
The reactor riser model

The complexity inherent in the riser is due to complicated hydrodynamics, heat transfer, mass
transfer and catalytic cracking kinetics. Moreover, influential parameters vary along the riser height
(Gupta and Rao, 2003):
Feed is injected into the riser, as are particles of hot catalyst from the regenerator; the
liquid feed vaporises and entrains catalyst and liquid drops along the riser height.
Vaporisation of liquid feed increases gas velocity.
The varying gas velocity affects axial (and radial) catalyst volume fractions.
Gas and catalyst particles exhibit significant slip between each other.
Vaporised hydrocarbons undergo cracking on the surfaces of catalyst particles, resulting in

lighter hydrocarbons and coke. The coke is deposited on the catalyst surface steadily
deactivating it. The reactant and product hydrocarbons continue to diffuse to and away
from the catalyst surface.
Heat is transferred from the hot catalyst, vaporising liquid feed and providing heat to the
endothermic cracking reactions.
The lumping approach is applied to FCC unit modelling given that it is not feasible to treat all
species and reactions, because the complex reaction network comprises a huge number of unknown
rate constants. Therefore, molecules and reactions are lumped by their chemical nature, i.e. the
number of carbon atoms, while different molecular groups are lumped by their boiling points. For
this purpose, kinetic models have been found to describe cracking behaviour adequately
(Carabineiro et al., 2004).
In modelling the riser for this study, the following assumptions were made:
Cracking reactions take place exclusively in the reactor,
The riser wall is adiabatic,
Negligible dispersion and adsorption occur inside catalyst particles,
The catalyst and coke are given the same properties,
The riser reactor is taken as being tubular, one-dimensional, and with no axial or radial
dispersions.
82
Kinetic model of riser reactor

The following representative models describing riser kinetics were investigated:
Five-lump kinetic model
When it is important to predict yields of coke, gasoline and dry gas, a five-lump kinetic model is
used. Knowledge of coke yields allows heat integration to be studied, together with air blowers and
the design and simulation of the main fractionator. While predicting yields of gasoline and dry gas
separately from other lumps, it allows the effects at main components in the FCC unit (e.g. gas
compressors) to be analysed. Therefore, this catalytic cracking kinetic model separates gasoline,
dry gas and coke from other lumps (Ancheyta-Juarez and Sotelo-Boyas, 2000).
The following scheme simplifies five-lump kinetic models (Dupain and Gamas, 2003; Dupain et
al., 2006): A number of assumptions are made in this model: coke forms at initial catalyst-oil
contact, diesel only cracks to gas, and it is not converted into gasoline where a small quantity of
uncrackable gas oil exists.
Coke
k4
k6
k1
Gasoline
k2
LCO
Gas oil
k5
k3
Gas
Figure 3.5: The five-lump kinetic model for cracking.
All the cracking reactions are assumed to occur only in the reactor riser. Sensible heat, heat from
vaporisation, and the heat driving the endothermic cracking reactions are provided by the hot
catalyst. The riser is modelled by mass balance, describing the production of gasoline, diesel,
slurry, coke and gases. The following sets of differential equations describe this model:
83
dy feed
dtc
dydiesel
dtc
(( (k1
((k1 (yslurry
dygasoline
((k2 (yslurry
dtc
dygas
dtc
dycoke
dtc
k2
((k3 (yslurry
((k 5(yslurry
k3 )(yslurry
x slurry ))
xslurry ) k4ydiesel )
xslurry ))
x slurry )
conv
(k5 (yslurry
(k6ydiesel )
coke
xslurry )CTO
)CTO
CTO
CTO
(3.12)
(3.14)
conv
k6ydiesel )
(3.11)
(3.13)
conv
k4ydiesel )
x slurry )
conv
CTO
(3.15)
coke
The catalyst-to-oil ratio (CTO) is expressed as follows:

CTO
Frgc
f3 f4
(3.16)
The activity function of coke formation describes this process:

d
coke
(3.17)
coke
dtc
Coke deposits poison the catalyst and reduce cracking activity. The activity function of catalyst
conversion can be described as a function of coke content; it is assumed to be the same for all the
reactions.
d
conv
dcc
kd
(3.18)
conv
Catalyst deactivation model

The catalyst deactivation model is used in kinetic studies conducted on catalytic cracking/pyrolysis
lumping models, and is typically based on a function dependent on time-on-stream, represented by
the catalyst residence time, or a function dependent on coke content of catalyst, represented by the
catalyst coke mass fraction. While the former catalyst deactivation functions are in use, the latter
are particularly appropriate, since a main cause of catalyst deactivation is coke (Alsabei et al.,
2008).
84
The aim in this work was to develop a coke content-dependent, catalyst deactivation function.
Among the estimated function variables, coke content of the catalyst is independent, and represents
an operating parameter, with unknown value at any particular point in time. As such, to prevent a
new unknown from being introduced, direct measurement to deduce the parameter, and
consequently the deactivation function, was performed. The function of coke content in the catalyst
was derived from least squares regression analysis, varying with the prevailing operating conditions
and properties of the feed. The catalyst deactivation function was then derived.
As shown in Eq. (3.29), the catalyst coke content (Cc) in heavy oil pyrolysis is a function varying
with feed atomic ratio C/H, temperature of reaction (T), oil gas residence time (t), catalyst residence
time (tc), and weight ratios of catalyst-to-oil (Rco) and steam-to-oil (Rso).
Least squares regression analysis is performed on experimental results to yield the parameters
needed in Eq. (3.29), resulting in Eq. (3.30). In turn, experimental data applied to the catalyst
deactivation function described by Eq. (3.31), yields the function used in this work. This function is
independent of reaction temperature, and does not vary, compared to a deactivation function based
time-on-stream, as the function already accounts for the effect of temperature (Roman et al., 2009).
b
Cc
C
a
T ct d 1
H
Cc
C
3.5248
H
Rso tcg 100
(3.19)
4.3889
T 0.2838t
0.1774
Rco
0.4795
Rso
0.3963
tc0.2276 100
Cc
exp
exp
Rco
2.55
(3.20)
(3.21)
C
H
4.3889
T 0.2838t
0.1774
Rco
0.4795
Rso
0.3963
tc0.2276
(3.22)
Assumptions regarding riser energy balance include stirred tank dynamics, negligible heat loss to
the environment, and the heat of cracking being proportional to riser temperature. Hence, the
energy balance is (Cristea et al., 2003):
MCp
dTr
eff
dt
Qin
Qout
(3.23)
where:
Q in = Q rgc + F 3Cp fl(T 2 - T base,f )
(3.24)
85
Q out = Q catout + Q slurry + Q cracking + Q ff
(3.25)
Qcatout
(3.26)
FrgcCpc (Tr
Tbase )
Q slurry = F 4[ Cp sv(T r - T ref ) + Q sr ]
(3.27)
Q ff = F 3[ Cp fv(T r - T ref ) + Q fr ]
(3.28)
F3
Q cracking
F4
H crack
(3.29)
H crack = 172,7 + 3(T r - T ref )
(3.30)
The bottom of the reactor riser pressure is needed in the force balance on the regenerated catalyst
bend. This is expressed by:
h ris
144
ris
P rb = P 4 +
(3.31)
where
ris
F 3 + F 4 + F rgc
(3.32)
ris
ris
F3 + F 4
F rgc
(3.33)
part
The inventory of the catalyst in the riser is:

W ris =
F rgcArish ris
(3.34)
ris
3.1.3 The reactor stripper model

This part of the FCC unit is represented as a continuous tank with perfect mixing, where reactions
do not occur. This is justifiable, since the catalyst and product vapour are separated immediately by
the cyclones; thus, usually, no cracking reaction occurs in this disengaging-stripping section. The
factors affecting coke yield in the riser in the form of deposits on the catalyst, are assumed
exclusively to be: weight hourly space velocity in the riser (WHSV), carbon concentration on the
regenerated catalyst, residence time of the catalyst in the riser, and the coking characteristics of the
various feeds (Cristea et al., 2003).
86
To the Main
Fractionator
P4
Tr
Ts
Steam
Fsc
Figure 3.6: Scheme of the FCCU stripper.
The production of coke is given by:
F coke
1.3557 F3
F4 FB r -1.9843
(3.35)
100WHSV
where
= W ris
60 F rgc
WHSV =
FB
and
(3.36)
3600( F 3 + F 4 )
(3.37)
W ris
F3
5F 4
F3
3F 1
F4
0.8F 2
(3.38)
1 for normal gasoil
>
1 for heavier than normal gasoil
<
1 for lighter than normal gasoil
>
Wet gas yield is given by:
87
F wg = ( F 3 + F 4)[ C 1 + C 2(T r - T ref )]
(3.39)
The coke balance in the stripper also includes any carbon on the regenerated catalyst entering the
reactor. This is expressed by:
dC sc = 1 (
dW r )
F rgcC rgc + F coke - F scC sc - C sc
dt
dt
Wr
(3.40)
In the stripper, catalyst balance is given by:

dW r =
F rgc - F sc
dt
(3.41)
The pressure drop between the main fractionator and the reactor is assumed constant; hence, the
reactor pressure is given by:
P4
P5
Pfrac
(3.42)
3.1.4 Regenerator model

For the regenerator model, the assumptions are that the catalyst phase is perfectly mixed, that coke
deposits on the catalyst consist of carbon and hydrogen, and that all the hydrogen is burned off in
the regenerator. Moreover, it is assumed that the conversion of hydrogen is complete following
these reactions:
4H
O2
2H 2O
(3.43)
where oxygen and carbon in the coke react to produce CO and CO2:
2C
O2
O2
2CO
(3.44)
CO2
(3.45)
Furthermore, carbon monoxide reacts to produce carbon dioxide:

2CO
O2
2CO2
(3.46)
Mass and heat transfer in the reactor and regenerator are complex. Hence, the regenerator is
typically described as two zones: a dense bed zone (with a dense phase and a gaseous phase), and
an entrained catalyst zone (Figure 3.8). Two phases are assumed to make up the fluidised bed:
gaseous reactants first giving a bubble phase and including products moving up the bed in plug-
88
flow; and a second, perfectly mixed, dense phase comprising solid catalyst and gas (Bollas et al.,
2007; Han and Chung, 2001).
Flue
gas
Gas
Fcc
Dilute
Fcc
Fsc
Bubble
Emulsion
Frc
Tbo Fair
Figure 3.7: Regenerator phases.
Along the bed, gas moves and mass transfer between the two phases occurs. In order to allow
modelling of the dense bed, a number of assumptions, as follow, are made:
bubbles do not contain any catalyst particles within them;
the cyclones restore all the particles of the ejected catalyst in the freeboard to the dense
bed;
perfect mixing of the catalyst particles in the dense bed occurs due to the cyclone
recycling action and circulation in the bed;
thermal equilibrium exists between the catalyst in the dense bed and the gases;
the gas densities of the bubble and emulsion phases are both the same;
gases follow a tubular flow regime in both the bubble and emulsion phases;
regenerator gas consists of oxygen, carbon monoxide, carbon dioxide, water vapour,
and nitrogen.
89
The gaseous and dense phases of the bed are assumed to be in internal equilibrium. Due to
entrainment in the flows, catalyst is to be found in the zone above the dense bed (the disengaging
zone). The quantity of catalyst decreases with vertical distance. In the presence of the catalyst, heat
is generated by reactions (3.43), (3.44) and (3.45); this influences the energy balance in the
regenerator. Only reaction (3.46) is significant in the region above the disengaging zone (the dilute
phase) since there is very little catalyst there (Han et al., 2000; Ansari and Tade, 2000).
Stack gas valve
V14
xO2sg
Tcyc
P6
Air Vent
F7
F sc
V7
Fair
F rgc
Combustion
air blower
V6
Atmospheric Air
Figure 3.8: Scheme of the FCCU regenerator.
The following equations describe the energy balance for the reactor bed (McFarlane et al., 1993;
Cristea et al., 2003):
[WregCpc
MI ]
dTreg
Qin
dt
Qout
(3.47)
where
Q in = Q air + Q H + QC + Q sc
Qout
Qfg
Qrgc
Qair
FairCpair (Tair
(3.48)
Qe
(3.49)
Tbase )
(3.50)
90
QH
FH H H
Qc
Fair (XCOsg H1
Qsc
FscCpc (Tsc
Qrgc
FrgcCpc (Treg
FH
Fsc (Csc
(3.51)
XCO2,sg H 2
(3.52)
Tbase )
(3.53)
Tbase )
(3.54)
C rgc )C H
(3.55)
Qfg [Fair (XO2,sgCpO2 XCOsgCpCO

0.5FHCpH 2O ](Tcyc Tbase )
XCO2,sgCpcO2
0.79CpN 2 )
(3.56)
The energy balance of the regenerator is expressed by:

dT reg(z)
=0
dt
(3.57)
z bed
dXCO(z)
dT reg(z)
=( H1
+
dt
dz
z bed < z
H2
dXCO2(z)
dz
1
Cp(z)
(3.58)
z_
Cp(z) = 0.79 CpN 2 + XCO(z) CpCO + XCO2(z) CpCO 2 + XO2(z) CpO 2 + ...
1
[0.5 CpH 2O F H + z Cp c M e ]
F air
z Z cyc
z =0
z < z cyc
z =1
(3.59)
The carbon balance is given by:

1
dC rgc
dW reg
=
( dW c - C rgc
)
dt
dt
dt
W reg
(3.60)
dW reg
= F sc - F sp
dt
(3.61)
dwc
FscCsc FrgcCrgc 12Fair ( XCOsg XCO2,sg )
dt
(3.62)
Mass balances of oxygen, carbon monoxide, carbon dioxide are expressed as:
dXO 2 = [100 (-0.5
k 1 - k 2)
dz
(z) C rgc - k 3 XCO(z)]
XO2(z)
91
(3.63)
dXCO(z)
= [100 k 1
dz
dXCO2 (z)
dz
=-
XO2(z)
(z) C rgc - 2 k 3 XCO(z)]
dXO2(z)
dz
(z)
- 0.5 dXCO
dz
(3.65)
XCO2(z) = XO2(0) - XO2(z) - 0.5 XCO(z)

XO2( 0) =
k1
1
0.21Fair
Fair
6.9547 e
(3.64)
(3.66)
0.25FH
(3.67)
34000
19.88
Treg (z ) 459.6
(3.68)
25000
k2 = 0.69148 e 15.06-T reg(z)+459.6
(3.69)
45000
k3 = 0.6412 P 6 e 25.55-T reg(z)+459.6
(3.70)
The output concentrations are:

100 F air XO 2
F sg
cO 2,sg =
cCO sg =
(3.71)
6
10 28 XCO
28 XCO + 44 XCO 2 + 32 XO2 + 22.12
(3.72)
The following empirical formula gives volume fractions of catalyst in the regenerator bed:
d
(z)
=0
(3.73)
(z) = 1 - e
0 z z bed
(3.74)
dz
B
(z)
dz
-1000 F air
Areg
z bed < z
e = min[1,
0.332
(z)
(3.75)
c,dilute
z cyc
max( f ,
f +
1.904 + 0.363
- 0.048
z bed
0.06 vs
2
s
)]
(3.76)
(3.77)
92
The empirical equations describing the mass flow rate of entrained catalyst exiting the bed are:
Me
Areg
(3.78)
c,dilute s
where:
c,dilute
0.582 vs
0.878
F sg + F air
2
(3.79)
1
g
(3.80)
Areg
where
520 P 6
379 14.7 (T reg + 459.6)
(3.81)
Ideal gas behaviour is assumed to give regenerator pressure balance:

dn
dP 6 = R [n dT reg +(
]
T reg + 459.6)
dt
dt
dt
V reg,g
(3.82)
dn
= F air - F sg
dt
(3.83)
V reg,g = Areg z cyc - Areg z bed (1 -
Prgb = P 6 +
PRR
(3.84)
W reg
144 Areg
P6
(3.85)
P4
(3.86)
Stack gas flows through the stack valve from the regenerator are given by:
F sg = k 14 V 14
P 6 - P atm
(3.87)
Empirically, the regenerator bed height is expressed as:

z bed = min[ z cyc, (2.85 + 0.8
W reg -
c,dilute
Areg
Areg z cyc
c,dense
) (
c,dense
c,dense
)]
(3.88)
c,dilute
where
c,dense
part
(1
(3.89)
93
3.1.5 Air blower model

A single-stage centrifugal compressor, driven by a variable speed steam turbine, is used to simulate
the air blower system. For the anti-surge control system, an atmospheric vent line and valve are
available on the discard line (Cristea et al., 2003).
Steam
Atmospheric
Air
Air Vent
V7
V6
F7
F6
P1
Electric
drive
P2
Combustion
air to the
regenerator
Blower
Figure 3.9: Combustion air blower.
Head capacity performance is given by:

9
6
2
F sucn,comb = 45000 + 1.581 10 - 1.249 10 P base
P base =
(3.90)
14.7 P 2
P1
(3.91)
Suction and discharge pressure are given by:

dP1
R (Tatm
dt
460)
29 Vcomb,s
(Fv 6
F6 )
(3.92)
dP 2 = R (T comb,d + 460) ( F 6 F V7 - F 7 )
dt
29 V comb,d
(3.93)
where
F6 =
520 29 P 1 F sucn,comb
14.7 379 60 (T atm + 460)
(3.94)
94
F7
kcomb P2
Prgb
(3.95)
F V6 = k 6 f pp(V6)
P atm - P 1
(3.96)
F V7 = k 7 f pp(V7)
P 2 - P atm
(3.97)
3.1.6 Model of catalyst circulation lines

The rate of catalyst circulation through the lines system between the regenerator and riser is
controlled by two slide valves; the pressure drop across each governs the catalyst circulation rate.
Spent and regenerated catalyst circulation is represented as a single-phase flow, driven by simple
force balances. For the purpose of simplification, factors affecting catalyst circulation in an
operational FCCU are disregarded in the model. These include carbon concentration on the
regenerated catalyst and stream injection at various points in bends (Cristea et al., 2003):
144(P6
Prb )
zbed
(Etap Eoil )
Psvrgc
Pelb,rgc
Frgc Lrgc frgc

2
Argc
(3.98)
The force balance on the spent catalyst line is expressed as:

144(P4
P6 )
(Estr Elift )
Wr
c
Astr
Psvsc
Pelb,sc
Fsc Lsc fsc

Asc2
(3.99)
where
Psv
Pelb
50Fcat
KAsvsv
]2
144
(3.100)
1 * 2
N cv
2
(3.101)
3.1.7 Main fractionator model

Distillation technology is quite prevalent in the refining industry, which counts as one of the largest
users. A key refinery process, fractionating, occurs in the distillation unit so modelling the
distillation unit is vital in predicting the composition of products derived from various gas oils
processed under different operating conditions.
Among separation processes, FCC distillation stands out as one of the most complex. Distillation
typically results in a mixture of various hydrocarbon compounds which cannot be characterised in
95
terms of individual components. This has given rise to the practice of substituting a finite number
of pseudo-components to express the composition of crude feedstocks and resulting products. Each
pseudo-component is a complex hydrocarbon mixture characterised by an average boiling point and
an average specific gravity; this is then treated as a single component (Kumar and Sharma, 2001).
The FCCU global model includes a nonlinear model representing a continuous 38-stage distillation
column with a reboiler and a total condenser; feed flow enters the column at stage 8. The following
assumptions were made with respect to the 117th order model for the main fractionator:
There are three pseudo-components (multicomponent hydrocarbon mixture) representing

gasoline, diesel and slurry(the heavy component)
relative volatilities are constant,
there is no vapour holdup,
molar flows are constant (vapour flow is the same for all stages),
Franci's Weir Formula describes liquid flow dynamics,
there is a total condenser,
energy balance and hydrodynamics are neglected.
Hence, the pressure balance for the main fractionator is expressed by:
dP 5 = 0.833(
F wg - F V11 - F V12 + F V13 )
dt
where
FV 12
k12V12 P5
(3.102)
Patm
(3.103)
Liquid flows according to Franci's Weir Formula:

M
Mow
Muw
(3.104)
The liquid flow L depends only on holdup over the weir, Mow ; total holdup is the sum of the holdup
over the weir and the holdup below the weir:
M
Mow
Muw
(3.105)
In the rectifying section, vapour flows are given by:

V
Vb
(1 gF )F
(3.106)
96
The equations of the derived model are given (Halvorsen et al., 2000; Halvorsen and Skogestad,
2003) by:
the liquid holdup at each column stage j except the feed stage is given by:
dM j
Lj
dt
Lj
Vj
(3.107)
the material balance for component i at stage j except the feed stage:
dM j x i, j
Lj 1x i, j
dt
Lj x i, j
Vj 1yi, j
Vj y i , j
(3.108)
the liquid holdup at the feed stage NF:
dM NF
dt
LNF
LNF
VNF
VNF
(3.109)
the material balance for component i at the feed stage:
dM NF x i,NF
LNF 1x i,NF
dt
LNF x i,NF
VNF 1yi,NF
VNF yi,NF
(3.110)
for the reboiler (stage j=1), the liquid holdup is:
dM1
L2
dt
V1
(3.111)
the material balance for component i in the reboiler (stage j=1):
dM1x i,1
L2x i,2
dt
V1yi,1
Bx i,1
(3.112)
for the condenser (stage j=NT)
dM NT
dt
VNT
LT
(3.113)
the material balance for component i in the reboiler (stage j=NT):
dM NT x i,NT
VNT 1yi,NT
dt
yj
Vj
LTx i,NT
Dx i,NT
(3.114)
Vapour-liquid equilibrium in each stage j for the multicomponent mixture i:

i
xj
(3.115)
1)x j
97
3.1.8 Wet gas compressor model

This component of the FCC process is represented by a single-stage centrifugal compressor. The
compressor is driven by a constant speed electric motor and is assumed to pump against a constant
pressure created by the vapour recovery unit (Cristea et al., 2003).
V13
Fv11
V12
P1
V11
P7
Main
Fractionator
From the
Reactor
Wet Gas
Compressor
Figure 3.10: Wet gas compressor.
Wet gas compressor performance is given by:
(3.116)
Where
H wg
0.0942
182.922(C rw
C rw =
P vru
P7
1)
(3.117)
(3.118)
The ideal gas law provides the relation for molar flow rate through the compressor:
F 11 =
520 F sucn,wg P 7
379 60 590 14.7
(3.119)
98
Pressure balance around the compressor is given by:

dP 7 = 5 (
F V11 - F 11)
dt
F V11 = k 11 f pp(V11)
FV 13
(3.120)
P5 - P7
(3.121)
k13V13Pvru
(3.122)
The nonlinear valve flow characteristics are expressed by:
fpp (x )
e 2 ln[0.15(1
fpp (x )
0.3x
x )]
for x
0.5
(3.123)
for x
0.5
(3.124)
The geometric and functional parameters of the model were adjusted to provide simulation outputs
in the range of the real plant data, shown in Table 3.1. The Matlab function with the implemented
model equations and use constants is in Appendix.
Table 3.1 provides data for the global model of a complex reactor-regenerator-fractionator process
plant.
Table 3.1: Construction and operation data.
Experimental data
Value
Unit of
measure
Riser diameter
1.2
Riser height
39.4
Striper diameter
2.9
Striper height
5.6
Elevation of the feed inlet in the riser
2.4
Cross sectional area of regenerated catalyst pipe
0.6
m2
Cross sectional area of spent catalyst pipe
0.7
m2
Length of regenerated catalyst pipe
7.3
Length of spent catalyst pipe
5.4
99
Experimental data
Value
Unit of measure
Cross sectional area of regenerated catalyst, slide valve at

completely open position
0.58
m2
Cross sectional area of spent catalyst, slide valve at

completely open position
0.7
m2
Regenerator diameter
8.5
Regenerator height
13.7
Height of cyclone inlet
15.2
Height of O2 measurement point
17
Elevation of the pipe for regenerated catalyst, outlet from

regenerator
12.2
Elevation of the pipe for spent catalyst, inlet in the

regenerator
18.5
Temperature of the fresh feed entering the regenerator
143
Atmospheric pressure
bar
Location of the feed stage tray
Spent catalyst valve position
55
Regenerated catalyst valve position
65
Fresh feed flow rate
170
m3/h
Slurry recycle flow rate
9.6
m3/h
Furnace fuel flow rate
34
scf/s
Gasoil density
0.8915
t/m3
Slurry density
0.968
t/m3
Gasoline density
0.7372
t/m3
Diesel density
0.862.6
t/m3
3.2 Conclusions
This chapter proposes a complex nonlinear dynamic model for the FCCU. The model describes the
7 main sections of the entire FCCU; including (1) the feed and preheating system, (2) reactor, (3)
regenerator, (4) air blower, (5) wet gas compressor, (6) catalyst circulation lines and (7) main
fractionators.
The novelty of the model consists in that besides the complex dynamics of the reactor-regenerator
system, it includes the dynamic model of the fractionator, as well as a new five lump kinetic model
for the riser which incorporates the temperature effect an reaction kinetics; hence, it is able to
100
predict the final production rate of the main products (gasoline and diesel) and can be used to
analyze the effect of changing process conditions on the product distribution.
The FCCU model has been developed on the basis which incorporates the temperature effect on
reactor kinetics reference construction and operation data from an industrial unit. The resulting
global model of the FCCU is described by a complex system of partial-differential-equations,
which was solved by discretising the kinetic models in the riser and regenerator on a fixed grid
along the height of the units, using finite differences. The resulting model is a high order DAE, with
942 ODEs (142 from material and energy balances and 800 resulting from the discretisation of the
kinetic models).
101
Chapter 4: Development of a Dynamic Process Simulator for the FCCU
CHAPTER 4
DEVELOPMENT OF A DYNAMIC PROCESS
SIMULATOR FOR THE FCCU
In this system the main operation and considerations for the FCCU are analysed. In the
development of the process simulation these aspect must be taken into consideration to provide a
tool that enables operators to analyse realistic process conditions. The following operational and
control aspects will determine the key input and output parameters used in the simulations.
4.1 Considerations of process operation

The primary purpose of FCCU is to convert the heavy oils such as vacuum gasoil, coker gasoil, etc
into lighter and more valuable products such as LPG, gasoline, distillate and diesel. This particular
unit is designed to maximize the production of LPG and gasoline while the distillate product stream
is used either in the diesel pool or the fuel oil pool. The aim of operation of the unit depends on the
specific requirements of each of these products. The control of operation of the unit depends on a
few selected parameters, which are closely monitored to achieve the optimum mix of products that
satisfy all the requirements. They also depend on the various unit constraints such as equipment
limitations, etc. The FCCU is a complex multivariate system with strong interactions between the
parameters. This interaction is a natural consequence of heat balance. This heat balance refers to a
steady state condition of the process where energy, or heat, requirements are met by the combustion
of coke. This heat gets transferred from the regenerator to the reactor by hot circulated catalyst. For
e.g., if reactor temperature is increased, at constant feed preheat temperature, then there will be a
corresponding increase in the catalyst circulation rate and in the regenerator temperature, which
will have its own effect on the yields and product quality.
The most important and common term that describes the operation of FCCU is conversion. This
gives a broad picture of the performance of the unit. It is defined as the weight percent of the
feedstock that has been converted to gasoline or lighter than gasoline products. It also includes the
coke formation during the cracking reactions.
Conversion (%) = [Feed rate (vol.) (Distillate + LCO + HCO) /Feed rate (vol.)] 100
102
The conversion is also represented as apparent conversion and true conversion based on seasonal
demands of gasoline. Normally, the end point of gasoline distillation range would be between 380
o
F and 450 oF, i.e. 193 oC to 232 oC. Any undercutting of gasoline would result in that portion of
gasoline dropping down to the LCO pool.

Hence apparent conversion is the conversion calculated before allowing for the adjustment and true
conversion is the one calculated on actual yields. It should, however, be noted that the fractionator
is operated in a fixed temperature profile and no cut point adjustments are done in the normal
operation of the unit.
Another term used for comparison of the different operating variables is the kinetic conversion.
This is defined as conversion divided by 100 minus conversion.
Kinetic conversion = Conversion / (100-conversion)
The second order values of kinetic conversion are useful in generating straight-line relationships
between the variables. This term is used generally for the process monitoring and catalyst
evaluation of the unit and is not widely used by the operating personnel in the routine operation of
the unit.
For purpose of discussion, the unit can be broadly divided into the reactor-regenerator and the rest,
i.e. fractionator, gas concentration and power recovery sections. Most of the major parameters in
these sections are adjusted independent of each other. Normally the reactor section is operated and
stabilized in a fixed pattern before adjusting the rest of the unit. Any change in the reactorregenerator operation parameters will subsequently need the fractionation and other sections to be
re-stabilized.
4.1.1 Reactor-regenerator section

The major operating parameters that are adjusted in the reactor-regenerator section include the
charge rate, combined feed temperature, combustion air rate, pressure balance, reactor temperature,
recycle ratio, catalyst inventory, fresh catalyst addition rate, dispersion steam and stripping steam
rates, etc. The parameters, which cannot be adjusted directly, include catalyst activity, catalyst
circulation rates, feed quality, conversion, regenerator temperature, etc.
Feed charge rate
The feed charge rate to the unit is normally decided on a long-term basis and is not subjected to
frequent changes during routine operation. However, the unit operates best when the feed rate is
103
close to design. The flexibility of feed rate is usually dependent on constraints of equipments and
its capacity limitations rather than the process itself. For effective unit operation, change of feed
rate involves restoration of the original conversion levels by adjusting the severity of operation.
This is achieved by maintaining the reactor temperature at the required level. The various means of
maintaining the required reactor temperature is discussed in detail later.
Combined feed temperature
Any FCCU in operation will automatically adjust its parameters to maintain its heat balance. One of
the major inputs of heat to the reactor is the feed. The temperature of the combined feed, i.e. fresh
feed and the recycle together, is directly related to the heat input to the reactor.
Any change in the feed temperature will correspondingly lead to changes in the catalyst circulation
rate and thereby conversion levels if the reactor temperature is to be maintained constant. If, for
example, the feed temperature is increased slightly, then in order to maintain the same reactor
temperature, less heat is required from the catalyst. So the catalyst circulation rate gets reduced. As
a result of that, conversion level decrease as there is less contact between feed and catalyst. Also
coke generation is reduced when catalyst circulation rate reduces. Regenerator temperature tends to
rise slightly as catalyst circulation rate reduces. If there is any limitation in the catalyst circulation
rates and in maintaining the required reactor temperature, then it is advisable to increase the feed
temperature. If, on the other hand, there is no limitation in catalyst circulation, then it is advisable
to reduce feed temperature and maintain same reactor temperature to increase conversion level.
When conversion is increased in this manner, the gas yield is also reduced.
The effects of an increase in the feed temperature are:
Reduction in catalyst to oil ratio
Reduction in coke yield
Reduction in conversion
Increase in regenerator temperature
Combustion air rate

This is the most important parameter in the regenerator and can vastly affect the operation of the
unit. The main purpose of the regenerator is to burn off all the coke that has been formed during the
course of the cracking reactions. The heat that is produced during the burning of coke in the
regenerator is used for providing the required temperature in the reactor for normal operation of the
104
unit. It is also essential to ensure that the entire coke is burnt off or else the activity of the catalyst
gets reduced and also maintaining the reactor temperatures become difficult. The air rate to the
regenerator is set to maintain a slight amount of excess oxygen in the regenerator flue gases. About
1-2% excess oxygen is maintained so as to ensure that the entire coke in the spent catalyst is burnt
off in the regenerator.
Reactor temperature
The reactor temperature is the most important parameter in the reactor structure during the normal
operation of the unit. The conversion level of the unit is dependent on the reactor temperature.
Increase in reactor temperature will result in an increase in the conversion. Higher reactor
temperature will also result in a slightly higher octane number of gasoline produced. This controller
acts on the regenerated catalyst slide valve, thereby adjusting the catalyst circulation rate to meet
the required reactor temperature. Generally, reactor temperature can also be adjusted by varying the
other parameters of heat balance in the system like the feed preheat, the coke make, recycle ratio,
airflow, etc. However, the best method of adjusting the reactor temperature would depend on
individual scenario taking into consideration the flexibility available in the feed preheat circuit,
main air blower capacity, catalyst regeneration, slide valve opening, etc.
The effects of an increase in the reactor temperature are:
Increase in catalyst to oil ratio
Increase in regenerator temperature
Increase in conversion
Increase or decrease in gasoline yield depending on over-cracking point
Increase in C4 yield and olefins
Increase in gasoline octane and bromine number
Increase in cycle oil aromaticity
Slight increase in coke yield
CO combustion
During the course of the cracking reactions in the riser, coke is formed as a by product, which gets
deposited on the catalyst. This coke is burnt off in the regenerator to evolve the heat required to
carry out the reactions in the riser. It is also required to retain its activity, as the coke deposition on
the catalyst will prevent the oil particles from coming in contact with the active sites of the catalyst.
Coke comprises of carbon and hydrogen in varying proportions. In the regenerator, the entire
105
carbon is converted to carbon dioxide, either directly from the carbon state or in steps where the
carbon is first converted to carbon monoxide and then carbon monoxide to carbon dioxide. The
hydrogen in the coke gets converted to water.
The reactions that take place in the regenerator can be summarized as given below:
C O2 CO2 heat
(4.1)
C O2 CO heat
(4.2)
CO O2 CO2 heat
(4.3)
H2 O2 H2O heat
(4.4)
Maximum heat is generated from the combustion of carbon monoxide to carbon dioxide. Flow of
air to the regenerator is increased to such an extent that complete oxidation to carbon dioxide is
ensured. This also confirms that the catalyst has been regenerated to the maximum possible extent
thus regaining its activity. CO combustion to CO2 is also promoted by adding additional agents
such as CO promoters in the fresh catalyst being added into the system. The process of regeneration
of the spent catalyst in the regenerator requires great attention because of the after effects of either
excess air or less air than what is required. They are called afterburning and behind in burning,
which is discussed below.
Behind in burning / afterburning
It is very important to burn off all the coke that is produced during the cracking reactions. The flow
of air to the regenerator has to be enough so as to ensure that all the coke has been burnt off. This
can be done by maintaining a slight excess amount of oxygen in the flue gas exiting out of the top
of the regenerator. The amount of coke burnt should be equal to the amount of coke generated
during the reactions. If the coke burnt is less than the coke being deposited on the catalyst, then
there will be net accumulation of coke. This will result in loss of activity of the catalyst. The
regenerator temperature will also start coming down slightly. The difference in temperature
between the dense phase of the catalyst and the dilute phase of the catalyst or the flue gas
temperature will come down as there is no more air available for further combustion of coke. The
sample of catalyst, if checked periodically, will tend to get darker and darker. The material balance
of the unit gets affected as a result of the lower activity of the catalyst with conversion levels
coming down. This phenomenon is called behind in burning. Care should be taken not to let the
106
unit go behind in burning. In case the unit does go behind in burning, then the flow of air should be
increased in very slow steps keeping constant watch over the regenerator temperatures. There will
be a sharp rise in the temperatures leading to a corresponding decrease in the catalyst circulation
rates. This will lead to further temperature rise in the regenerator, which in turn will have to be
compensated by decreasing the combined feed temperature to increase catalyst circulation rates.
Once all the accumulated coke in the catalyst has been confirmed as burnt off by physically
checking the sample of catalyst and verifying with the expected improvement in conversion, then
the unit can be brought back to the normal state. The catalyst absorbs about 70% of the heat that is
generated in the regenerator during the course of burning of coke. Hence it is essential that most of
the coke burning take place in the combustor zone so that the catalyst present in the dense phase
can absorb the heat that is generated. In case of excessive air to the regenerator, the velocity of
catalyst will increase resulting in transfer of the burning location from the dense phase to the dilute
phase of catalyst. When coke burning takes place in the dilute phase then the heat that is liberated
cannot be fully absorbed as there will be very less catalyst around that location. As a result the heat
liberated will increase the temperatures of the cyclones and the flue gas lines leading to possible
damage to the equipment.
Regenerator temperatures
Regenerator temperature is as such not a directly controlled parameter. It however gives a fair idea
of the operation of the unit. The change in regenerator temperature indicates the change in the
process of reaction in the riser. An increase in the regenerator temperature indicates either a heavier
feedstock, i.e. more coke formation, lower catalyst circulation rate, higher operating pressures,
higher combined feed temperatures, etc. The difference in temperatures within the dense phase and
dilute phase of catalyst in the regenerator also gives an idea of the extent of behind in burning or
afterburning.
Regenerator catalyst re-circulation
The high efficiency regenerator operation involves complete oxidation of carbon and carbon
monoxide to carbon dioxide within the dense phase of catalyst. The regenerator can be divided into
two sections; the lower regenerator and the upper regenerator connected by a riser through which
the catalyst is transported up from the lower to the upper zone. There is also another line, which
sends the catalyst from the upper regenerator section to the lower zone. The re-circulation slide
valve is located on this line. This line is utilized to send back some of the hot catalyst from the
upper zone to the lower zone to maintain the required minimum temperature so that complete
107
combustion is ensured. This is essential because if the combustion is incomplete, then the balance
coke in the catalyst will start burning in the dilute phase of the catalyst in the upper zone and in the
flue gas lines. The heat that will be liberated during that burning will not be absorbed fully as the
density of catalyst in that area is very low. This will result in the cyclone and other hardware
getting exposed to very high temperatures causing mechanical damage. This phenomenon, as
explained earlier, is called afterburning. In such cases, the re-circulation slide valve is opened so
that the hot catalyst flows back to the lower section of the regenerator, thereby increasing
temperature, which will result in more combustion taking place and the entire C and CO gets
converted to CO2.
Catalyst circulation rate
This is also not a directly controlled parameter. It depends on the overall operation of the reactor
and regeneration. The rate of catalyst circulation is dependent on the feed rate, reactor temperature,
combined feed temperature and the regenerator temperature. The catalyst circulation rate is the
main driver for maintaining the reactor temperature. It is also dependant on the pressure
differentials across the slide valves. The opening of the regenerated catalyst slide valve gives a fair
picture of the relative performance of the unit. This should be in a floating range and not too
close or too wide open.
Catalyst to oil ratio (CTO ratio)
This is similar to catalyst circulation rate in that it is not a directly controlled parameter. This is
defined as the tones per hour of catalyst circulated divided by the tones per hour of feed charge rate.
This parameter is dependent on the catalyst circulation rate and is affected by the changes in the
process parameters, which affect the circulation rate. An increase in the catalyst to oil ratio occurs
as a result of an increase in the reactor temperature, a decrease in the regenerator temperature or a
reduction in the combined feed temperature. If, for the same reactor temperature, the CTO ratio is
increased, it will lead to an increase in conversion resulting in higher gas and LPG yield, more coke
formation and a slight decrease in regenerator temperature. However, the rate of increase in
conversion with increasing CTO ratio becomes less when high levels of conversion are reached.
The relationship of kinetic conversion with CTO ratio is linear and keeps on increasing without
change.
Feed quality
The feed quality itself is not a parameter that can be controlled or changed. It keeps on changing
over a period of time due to the varying proportions and the quality of the individual components
108
like vacuum gas oil, coker gas oil, unconverted bottoms from hydrocracker unit, etc. Hence the feed
sample is tested for its properties and then the appropriate process conditions are maintained to get
the desired results from the unit.
Effect of reactor-regenerator variables on products
Dry gas
At a constant feed temperature and reactor temperature, the reactor energy requirement will remain
the same. Some operating variables can be changed without altering the energy requirements of the
reactor like regenerator temperature and feed preheat temperature. This is because the energy input
is a combination of catalyst circulation rate and temperature. To maintain a constant reactor
temperature, if the feed temperature or regenerator temperature is raised, then the catalyst
circulation rate is reduced. Although the reactor temperature doesnt go up, the riser-mix
temperature goes up because lower quantity of hotter catalyst mixing with feed. It has been
accepted that dry gas yield increases with higher reactor temperatures. It has also been noted that
the dry gas yield varies even at constant reactor temperatures. Whenever the riser-mix temperature
increases, it has been seen that dry gas yield increases. Hence increase in regenerator temperature or
feed preheat temperature also increases the dry gas yield. It goes to show that the dry gas yield is
not a function of conversion but is a thermally driven process. Decreasing the reactor temperature,
regenerator temperature or the feed temperature can reduce dry gas generation but its effect on the
other sections and products should also be considered.
LPG
Normally, the yield of C3s and C4s are dependant only on conversion levels. That is to say that the
yield of LPG will increase as conversion increases no matter how it is achieved, increasing reactor
temperature or catalyst to oil ratio, etc. The ratio of isobutane to butylenes yields drops sharply as
the reactor temperature is increased. This is because at higher reactor temperatures, higher olefins
will be generated with little effect on the C4 yield. This is of relevance to estimate the amount of
isobutane in feed to the alkylation unit.
Gasoline
At constant conversion levels, increasing the reactor temperature decreases gasoline yield. When
the reactor temperature is increased, to hold constant conversion the C/O ratio also has to be
adjusted. Hence there will be more dry gas yield and coke formation. So the liquid yield, including
gasoline, will come down. Generally it is accepted that a 10% increase in conversion or a 10oC rise
109
in reactor temperature will increase the RON of gasoline by 1.0 number from the base octane
number. As the temperatures go higher, the rate of increase in octane goes down. Since gasoline
yield and gasoline octane are both dependant on reactor temperature in different ways, an optimum
operating range has to be decided to maximize octane-barrels. The increase in octane due to higher
reactor temperature is due to the increase in the olefin content of gasoline. The olefin content of
gasoline is reflected in the bromine number. Lowering conversion level at the same reactor
temperature can reduce olefins in gasoline. Gasoline octane sensitivity, defined as the difference
between RONC and MONC is a function of reactor temperature only and is not affected by any
change in the conversion level.
Cycle oils
Like LPG, the yield of cycle oils is also a function of conversion level only.
4.2 Implementations and simulation procedure

The FCC model comprises algebraic equations coupled with ordinary differential equations and
differential algebraic equations of high order.
The set of algebraic and differential equations describing the 7 main parts of the FCCU were
implemented in simulators using the Matlab differential algebraic equation solvers. Matlab ODE
functions are used to solve the spatial ordinary differential equations for the riser, yielding the
output variables of gas temperature at the riser exit, and the distribution of product in the stream.
The spatial differential equations for the reactor riser and regenerator are solved using an Eulis
discretization method in space, and then applying adoptive predictions ODE integrators
(implemented in Matlab ODE solvers) to the discretized equations for the time-domain integration.
For the dynamic model, the solution in that the first step involves calculating the variables at the
steady state. Once this is accomplished, these values are used to define the open loop initial
conditions. The C++ programming language was used to write the code for the plant model
simulator. The Matlab MEX compiler was used to compile the C-function used in the simulation.
Although, source files can be compiled and linked within a Matlab executable shared library by
MEX, the model was run on MATLAB SIMULINK to ensure solution efficiency. The dynamic
process simulator achieves 3 key aims:
1. Exploring the dynamic behaviour of the process
2. Development and evaluation of facilitates the control schemes based on the model for the
whole process unit.
110
3. Serves as the predictive model in the nonlinear model predictive schemes developed in this
study.
4.2.1 Open loop simulations for the investigations of the dynamic behaviour of
the FCCU
The major manipulated variables (MV) in this FCC unit are: fresh feed flow, recycled slurry flow,
spent catalyst valve, regenerated catalyst valve, furnace fuel flow, air temperature, coke formation
factor and air flow rate. A set of process outputs were investigated in this section, as shown in
Table (4.1).These inputs and outputs correspond to the most frequently used manipulated and
controlled outputs, corresponding to most industrial FCC units (Arbel et al., 1996; Grosdidier et al.,
1993). Other manipulated variables and output variables can be added to the simulator according to
specific industrial practice by implementing it in a SIMULINK block diagram, as shown in Figure
(4.1).The step change analysis is shown in this chapter in Figures (4.3 to 4.8). These figures show
the widely different dynamic characteristics of the variables, including inverse response and other
important non-linearities with different disturbance scenarios (Table 4.1). All the disturbances
reflect possible upsets that affect the normal operation of the unit, and have been selected from a
practical point of view (Alsabei et al., 2008).
Table 4.1: Process input and base case operating conditions with disturbances.
Type
variable
F3
Fair_c
Manipulated
variable
svsc
svrgc
F5
psif
F4
tair
Description
Base case
operating
conditions
Flow of fresh feed to reactor riser

(lb/s)
Air flow rate into regenerator (lb/s)
Spent catalysis valve (%)
Regenerator catalyst valve (%)
Flow of fuel to furnace (lb/s)
Effective coke factor for gas-oil
feed
Flow of slurry to reactor riser (lb/s)
Temperature of air entering
regenerator (F)
111
120
69.92
0.3
0.3
32
1
3
416
Disturbance
Step change
value
+
132
108
10
73.41
76.91
66.42
62.92
1.05
1.1
3.6
0.95
0.9
2.4
5
10
5
10
20
-
Figure 4.1: SIMULINK block diagram used for the open-loop simulation.
Sensitivity analysis for composition of the products

The objective of the simulations is to investigate whether small errors in the parameters can have
consequences for products composition. In the riser section (described by the five-lump kinetic
model as described in Chapter 3), a sensitivity analysis for composition of the products at different
catalyst-raw material-ratios, i.e. the ratio of catalyst circulated to fresh feed (CTO), was performed.
The CTO is not an independent variable, and increases with an increase in reactor temperature;
moreover, it decreases with higher regenerator or combined feed temperatures. Therefore a high
sensitivity of the product composition to variations in the CTO indicates that the reactor and
regenerate temperatures must be tightly controlled for smooth operation of the FCCU.
Figures (4.2) illustrate that CTO has an important influence on the composition of coke, gases and
diesel resulting in the riser. It also has a small influence on the gasoline yield at the outlet from the
riser. The increase of the CTO leads to an increased cracking rate. The coke amount is formed very
fast during the first 0.4 seconds. During the first 0.4 seconds, a rapid decrease in the rate of
conversion of raw material is found due to coke formation. After 4 seconds, no net diesel is formed;
it only disappears. Gas, formed from both gas oil and diesel, rapidly forms within 0.4 seconds and
then starts to saturate. During the first 4 seconds, the gas is predominantly formed from gasoil, but
112
afterwards is taken over by diesel. This indicates that diesel is a stable fraction, which displays
hardly any cracking activities.
0.12
0.08
CTO=8
0.07
0.1
CTO=8
0.08
CTO=6
gas (kg/kg)
coke (kg/kg)
0.06
0.05
0.04
CTO=4
0.03
CTO=6
CTO=4
0.06
0.04
0.02
0.02
0.01
0
10
riser stationary time (s)
10

0.18
0.5
0.16
CTO=6
CTO=8
CTO=8
diesel (kg/kg)
gasoline (kg/kg)
0.4
CTO=4
CTO=6
0.14
CTO=4
0.3
0.2
0.12
0.1
0.08
0.06
0.1
0.04
0
0.02
10
90
0.9
80
0.8
70
0.7
60
0.6
0.5
0.4
0.3
CTO=6
10
conversion (%)
raw material (kg/kg)
CTO=4
CTO=8
CTO=6
CTO=4
50
40
30
20
CTO=8
10
0.2
0.1
10
10
Figure 4.2: Distribution of products and raw material, and raw material conversions in the riser at
different CTOs.
The fact that diesel is a fairly inactive phase is because the gasoil molecules are more reactive a
result of longer side-chains attached to the polycyclic aromatic cores; diesel is less reactive due the
113
short side-chains, which are attached to these cores. The raw-material conversion at different CTOs
is shown in Figure 4.2. The CTO has a small influence on the raw-material conversion; an increase
in CTO leads to approximately the same value of conversion (~ 80%).
Analysis of the dynamic response of the FCCU to various process disturbances
Dynamic simulation of the FCC process was performed according to the simulation design shown
in Table (4.1). Gas oil feed rate, air flow rate, and stem positions of the slide valves at both catalyst
transport lines were chosen as simulation variables to whose changes the dynamic responses are
demonstrated. Each simulation run started from the steady state corresponding to the base case
operating conditions, and the subsequent transient response was obtained as each simulation
variable went through a series of step changes shown in Table (4.1).
Change in fresh feed rate (F3)
Disturbance could appear due to the raw-material rate changes. Figure (4.3) shows that a 10%
increase in the fresh feed rate determines the decrease in the regenerator temperature, and also in
the reactor, after a peak.
Figure (4.3) shows the dynamic responses to the changes in fresh feed rate. After the fresh feed rate
is increased by10% at time equal to 1.5 h, the reactor temperature drops because of an increase in
heat consumption to vaporise the additional amount of fresh feed. The lowered riser temperature
(Tr) results in a considerable decrease in feed conversion, while the increased feed rate leads to a
sharp rise in coke on spent catalyst (Csc), and subsequently on regenerated catalyst (Crgc). The
increased feed rate also raises the density of hydrocarbon gases in the reactor, and so there is a
jump in gas pressure, as well as in the pressure at the reactor (not shown). This jump will trigger a
jump in spent catalyst flow rate, but has an opposite effect on regenerated catalyst flow rate.
Therefore, the catalyst holdup in the reactor steadily decreases, until a new steady state is reached,
where the two catalyst circulation rates balance each other. Regenerator temperatures (Treg) start to
decrease due to increased spent catalyst inflow from the reactor. Since the fixed air flow does not
supply enough oxygen to burn of the increased coke in the dense bed nor to sustain after-burning
reactions in the freeboard, CO concentration (cCOsg) in the stack gas steadily rises. The oxygen
deficiency in the regenerator also leads to no further rises in the temperature at the new steady state.
A 10% step decrease in the fresh feed rate at 1.5 h causes the system to move in the opposite
direction to the case of the 10% step increase, but the detailed responses show quite different
aspects due to process nonlinearity. The feed conversion increases due to increased catalyst-to-oil
ratio in the riser as well as better regeneration of catalyst in the regenerator. Contrary to the oxygen
114
deficiency encountered before, oxygen surplus prevails throughout the regenerator, because of the
decreased spent catalyst inflow from the reactor. This causes a rise in the regenerator temperature
due to after-burning of CO, and a consequent drop in the CO concentration (cCOsg) of stack gas. It
was found in the simulations described above that the step changes in fresh feed rate immediately
Treg(C)
10
15
3
2.5
2
1.5
1
0.5
gasoline
Coke
Crgc (wt%)
4
2
0
W c (ton)
cCOsg(ppm)
1.915
0.05
0
-0.05
0
-3
x 10
10
10
10
10
10
15
15
15
15
10
time (h)
6
4
2
6
5
4
0.25
0.2
15
0.45
0.4
0.35
0
850
800
750
3.5
3
2.5
LPG
P6(bar)
1.92
Frgc (kg/s)
Csc (wt%) cO2sg(mol%)
735
730
725
W sp(ton)
540
530
520
LCO
Tr(C)
alter the reactor pressure, which subsequently triggers a series of changes in the process dynamics.
15
- 10 %
0.4
0.2
0
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
+ 10 %
time (h)
Figure 4.3: Simulation of FCCU dynamic response in the presence of step change in flow fresh feed
disturbance (F3 = 10% increase at t = 1.5 h).
Change in air flow rate (Fair_c)

The air flow rate is the most important parameter in the regenerator, and can vastly affect the
operation of the unit. Figure (4.4) shows the dynamic responses to the changes in the air flow rate
115
to the regenerator. A 5% step increase in air flow causes a sharp increase in the regenerator pressure
(P6), as well as in the regenerated catalyst flow rate (frgc) to the riser bottom. On the other hand, the
spent catalyst flow rate (not shown) has inverse response behaviour of initial decrease followed by
ultimate increase. This behaviour can be explained by a steady increase in the reactor pressure due
to increasing static pressure exerted by the rising catalyst holdup in the reactor. The increased air
flow rate also accelerates coke burning, and thus raises temperature in every part of the unit and
reduces the CO concentration (cCOsg) in the stack gas. It also causes the coke concentrations on
both spent and regenerated catalyst to steadily drop to their respective steady state values. The
1
0.95
0.9
0.02
0.01
0
Treg(C)
0
-3
x 10
10
10
10
10
15
15
15
15
10
4
3
2
5.5
3.2
3
2.8
0.25
0.2
15
0.45
0.4
Frgc (kg/s)
10
850
800
750
cO2sg(mol%)
15
LPG
gasoline
W c (ton)
3
2.5
2
10
Csc (wt%)
1
0.5
0
W sp(ton)
2
1.9
1.8
735
730
725
LCO
535
530
525
Coke
Crgc (wt%) cCOsg(ppm)
P6(bar)
Tr(C)
conversion of feed is increased by the elevated riser temperature and cleaner regenerated catalyst.
15
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
0.2
0.18
0.16
time (h)
time (h)
-5%
+5%
disturbance (Fair_c = 5% decrease at t = 1.5 h).
116
A 5% step decrease in the air flow rate at time equal to 1.5 h brings about the state of oxygen
deficiency in the regenerator. A large amount of coke cannot be burnt, resulting in gradual
accumulation on catalyst surfaces in both vessels over a long time. Such a long response time is
characteristics of the composition dynamics in the regenerator that has large mass and thermal
holdups and strong interactions with the reactor. CO concentration (cCOsg) also shows a
considerable increase, because of poor afterburning reactions in the regenerator. This oxygen
deficiency lowers the temperatures of both vessels as well as the conversions, but gasoline yield is
0.02
0.01
0
0.5
0.45
0.4
Treg(C)
0
-3
x 10
10
10
10
10
15
Frgc (kg/s)
10
15
15
15
10
time (h)
4
3
2
5.5
5
4.5
3.2
3
2.8
0.25
0.2
15
15
- 10 %
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
800
780
760
cO2sg(mol%)
15
Csc (wt%)
1.2
1
0.8
gasoline
W c (ton)
3
2.5
2
10
W sp(ton)
1
0.5
0
740
730
720
LCO
2
1.9
1.8
LPG
535
530
525
Coke
Crgc (wt%) cCOsg(ppm)
P6(bar)
Tr(C)
significantly increased.
0.2
0.18
0.16
+ 10 %
time (h)
disturbance (Fair_c = 10% decrease at t = 1.5 h).
117
Figure (4.5) shows the dynamic responses to the changes in the air flow rate to the regenerator. A
10% step increase in air flow causes an effect on air blower efficiency, because of its limitations.
This can be clearly noted as there are no significant changes in the variables; the conversion of feed
is also affected.
A 10% step decrease in the air flow rate at time equal to 1.5 h resulted in similar behaviour as seen
for 5% step degrees as shown in figure (4.4).
Change in coke formation factor (f)
Heat is required to make the cracking reaction work and achieve satisfactory conversion. The
quality of the FCC feedstock impacts the concentration of coke on the catalyst entering the
regenerator. Coke formation is a necessary by product of the FCC operation, the heat released from
burning coke in the regenerator supplies the heat for the reaction.
Figure (4.6) shows the dynamic responses to the (5%) increase in f, fresh feed coke formation
factor, at time equal to 1.5 h.
This simulation run with a 5% increase represents a realistic unmeasured disturbance to the feed
composition. The increase in f affects the cracking reactions, and coke is deposited on the
catalyst, reducing catalyst activity. This led to a rapid increase in coke deposition in the riser and
concentration of coke on spent catalyst (Csc). This additional coke is transported to the regenerator
on the spent catalyst resulting in higher combustion rates in the regenerator, and thus an increase in
regenerator temperature (Treg). However, the initial increase in combustion rates is insufficient to
burn off all additional coke arriving from the reactor, and so the concentration of carbon on
regenerated catalyst (Crgc) increases. Consumption of oxygen increases, while total air flow rate to
the regenerator is held constant by the total air controller, so excess oxygen concentration in the
stack gas (cO2,sg) decreases and stack carbon monoxide concentration (cCO,sg) peaks and begins to
decline. As regenerator temperature increases, combustion rates increase and more of the coke
transported from the reactor is consumed. The concentration of carbon on regenerated catalyst
peaks and begins to decline. It eventually reaches steady state at a level only slightly higher than the
initial concentration. Taking into account the fact that the heat generated by CO formation is about
three times less than CO2 heat formation, the global effect is to reduce the net heat contribution in
the regenerator with the consequence that the temperature decreases in the last part of the
simulation.
During regeneration, the coke level on the catalyst is typically reduced (Crgc). The carbon also
decreases during the process (Wc). Moreover, the catalyst inventory stand pipe entering the
118
regenerator (Wsp) increases, and the regenerator pressure (P6) is also increased. The pressure has
little effect on conversion, so it is not adjusted to change product yields, but is important for
catalyst circulation phenomena.
From the regenerator, the catalyst flows down through a transfer line commonly referred to as a
standpipe, where (frgc) is the catalyst circulation rate. The amount of carbon that remains on the
regenerated catalyst is an operating variable, and more carbon on the regenerated catalyst results in
less conversion.
Increasingly hotter regenerated catalyst is transported back to the riser, affecting its energy balance.
Treg(C)
5
10
10
10
10
15
Frgc (kg/s)
10
15
15
15
15
10
time (h)
6
4
2
5.5
5
4.5
3.1
3.05
3
0.25
0.2
15
0.45
0.4
0.35
0
820
800
780
cO2sg(mol%)
0
-3
x 10
10
Csc (wt%)
0.02
0.01
0
W sp(ton)
Crgc (wt%)
1.2
1
0.8
gasoline
3
2.5
2
W c (ton)
4
2
0
Coke
cCOsg(ppm)
P6(bar)
1.922
1.92
1.918
740
730
720
LCO
535
530
525
LPG
Tr(C)
As riser temperature (Tr) increases, wet gas production increases due to higher cracking rates.
15
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
0.2
0.18
0.16
0
-5%
5
+5%
time (h)
Figure 4.6: Simulation of FCCU dynamic response in the presence of step change in fresh feed coke
formation factor disturbance (f = 5% decrease at t = 1.5 h).
119
With this magnitude of disturbance, the carbon on the regenerated catalyst can be reduced by
increasing the regenerator temperature. It is important, however, to keep a constant percentage of
carbon on the regenerated catalyst, because carbon yields increase in the reactor when the
regeneration is incomplete. If this additional carbon is not burned off in the regenerator, the carbon
yields will again increase when the catalyst passes through the reactor. This can lead to carbon
runaway, i.e. more carbon is produced than burned off (Sadeghbeigi, 2000; Bollas et al. 2007).
The composition of the products and raw material along the riser is also influenced in both either a
positive or negative way by this disturbance; the increase in the coking rate leads to an increase in
the flow of the regenerated catalyst, which then increases the catalyst-to-oil ratio, is not an
independent variable. It increases with an increase in reactor temperature and decreases with higher
regenerator temperature; this then leads to a decrease in gasoline and coke yield, but an increase in
LCO and LPG yield. The disturbance would be considered positive, if the objective is to have more
LCO or LPG than gasoline, and negative if gasoline is the desired product.
A 5% step decrease in coke formation factor, f, at 1.5 h causes the system to move in the opposite
direction to the case of the 5% step increase, but the detailed responses show quite different aspects
due to process non-linearity with inverse responses.
The dynamic responses to the disturbance test of (10%) increase in f (fresh feed coke formation
factor) at time equal to 1.5 h, is shown in Figures (4.7). These figures show the widely different
dynamic characteristic of the variables, including inverse response and other important nonlinearities.
However, it can be noted in the scenario of (10%) decrease in coke formation factor, f, which the
gasoline products increase hugely. This determines two facts, in that the disturbance could be a
positive sign, and that feed quality affects the gasoline production.
120
1.2
1
0.8
0.02
0.01
0
0.5
0.45
0.4
Treg(C)
10
10
10
10
15
Frgc (kg/s)
10
15
15
15
15
10
850
800
750
cO2sg(mol%)
0
-3
x 10
10
5
4
3
Csc (wt%)
W c (ton)
Coke
gasoline
4
3
2
5.5
5
4.5
W sp(ton)
1
0.5
0
Crgc (wt%)
cCOsg(ppm)
P6(bar)
1.92
1.918
1.916
740
720
700
3.2
3.1
3
LCO
535
530
525
0.4
0.2
0
15
LPG
Tr(C)
15
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
0.2
0.18
0.16
time (h)
time (h)
- 10 %
+ 10 %
Figure 4.7: Simulation of FCCU dynamic response in the presence of step change in fresh feed coke
formationfactordisturbance(f = 10% decrease at t = 1.5 h).
Change in slurry recycle rate (F4)

Dynamic responses are shown in Figure (4.8). Slurry recycling could appear as a result of change
in the raw-material fresh feed rate, or because of changes in the valve position on the slurry
circulation line. Heavier feeds than gas oil result in deposits of greater quantities of coke upon
cracking. A given change, i.e.(20% increase in flow rate of slurry at time equal to 1.5 h, is
expected to have a greater effect on system response than an equal size change in gas oil feed.
121
As demonstrated in this run, small changes in slurry flow rate produce significant temperature
responses, and therefore slurry flow rate is a useful manipulated variable. The reduction in slurry
flow rate causes an immediate and rapid drop in the concentration of coke on spent catalyst (Csc).
With lower Csc, less coke arrives at the regenerator on spent catalyst, reducing the rate of
combustion reactions occurring in the bed. Regenerator temperature (Treg) drops slightly, excess
oxygen (cO2,sg) rises and concentration of carbon monoxide in the stack gas (cCO,sg) falls. The flow
rate of regenerated catalyst (Frgc) falls a small amount, for reasons described previously; this leads
to increases of carbon inventory in the regenerator (Wc) and inventory of catalyst in regenerator
stand pipes (Wsp). With falling regenerator bed temperature and flow rate of regenerated catalyst,
reactor riser temperature (Tr) also increases after initial falls caused by the sudden drop in slurry
15
1.919
1.9188
1.9186
0.98
0.96
0.94
gasoline
Coke
Crgc (wt%)
2.65
2.6
2.55
W c (ton)
1
0.5
0
10
8
6
0.46
0.44
0.42
0
-3
x 10
10
10
0
-3
x 10
10
10
10
15
15
15
15
10
time (h)
810
800
790
3.1
3.05
3
0.2
0.19
0.18
15
15
- 20 %
10
15
10
15
0
-3
x 10
10
15
10
15
10
15
10
15
10
15
3.5
5.15
5.1
5.05
LPG
cCOsg(ppm)
cO2sg(mol%) Frgc (kg/s)
10
Csc (wt%)
W sp(ton)
732
730
728
LCO
P6(bar)
Tr(C)
532
531.5
531
Treg(C)
flow rate.
0.19
0.18
0.17
5
+ 20 %
time (h)
Figure 4.8: Simulation of FCCU dynamic response in the presence of step change in slurry flow recycle
rate disturbance (F4 = 20% decrease at t = 1.5 h).
122
Disturbance in slurry recycle flow rate leads to a decrease in the flow of the regenerated catalyst,
which decreases the catalyst-to-oil ratio, then leading to a small increase in LCO yield, but
decreases gasoline or LPG yield.
In figure (4.8), it can clearly be observed that a 20% decrease in slurry recycle flow rate, F4, has an
opposite effect on all the other variables. In addition, the same disturbance caused nonlinear
behaviour and inverse response for most of the variables.
4.3 Conclusions
Cracking reactions are endothermic, and therefore, to achieve higher productivity, the reactor
should be operated at as high a temperature as possible. Assuring the regeneration of the catalyst
(the burning of the deposited coke), the regenerator, as with the reactor, must be operated as close
as possible to its metallurgical limit in terms of temperature. Restrictions concerning the
concentration of carbon monoxide in the stack gas are very important from an environmental point
of view. A dynamic simulator was developed, which implemented the detailed dynamic model for
an FCC process and the model solver (these were presented in chapter 3). The correlations were
developed for the thermodynamic properties and transport parameters contained in the model either
by literature survey, or by nonlinear regression of literature data. The steady-state behaviour was
investigated and the results showed good match with those in the literature, and real plant data.
All the investigated disturbances showed considerable influence on the products composition.
Taking into account the very high volume production of an industrial FCCU, these disturbances can
have a significant economic impact. The fresh feed coke formation factor is one the most important
disturbances analysed. It shows significant effect on the process variables. Moreover, the objective
regarding the control of the unit has to consider not only to improve productivity by increasing the
reaction temperature, but also to assure that the operation of the unit is environmentally safe, by
keeping the concentration of CO in the stack gas below a certain limit.
The dynamic simulations show the nonlinear and multivariable behaviour of the FCC process. The
simulation results also show that the FCC model is sufficiently complex to capture the major
dynamic effects that occur in an actual unit. The process outputs exhibited a series of complex
dynamic behaviours, such as inverse response behaviour, variations different time scales, impulse
response, indicating difficulties in controller designs.
The model offers the possibility of investigating the way that advanced control strategies can be
implemented, while also ensuring that the operation of the unit is environmentally safe.
123
Chapter 5: Advanced PID Control of the FCCU Process
CHAPTER 5
ADVANCED PID CONTROL OF THE FCCU
PROCESS
5.1 Introduction
Over the course of six decades, significant developments have been achieved in catalytic cracking,
which is a key petroleum refining process. As a result, the fluid catalytic cracking unit (FCCU) has
been the test bed for many new control schemes, and both the industry and academics have worked
steadily to develop and implement efficient algorithms for this purpose. The FCC process presents
many challenges due to its (i) complex and poorly understood hydrodynamics, (ii) complicated
reaction kinetics for cracking and coke burning, (iii) significant reactor/regenerator interactions, and
(iv) multiple constraints in operation (Elamurugan and Kumar, 2010). Moreover, the process is
characterised by behaviour in the steady and dynamic states that is highly nonlinear, which results
in many potential states and input levels, etc. Prior to the introduction of zeolite catalysts,
stabilising and maintaining the process was the key control objective. However, with the use of
zeolite catalysts attention focused on raising production output, and coping with heavy feedstocks.
As a desirable product, reformulated gasoline has added the challenge of controlling the product
chemistry, which is a complicated undertaking as those variables that can be manipulated are very
few compared to the large number of actual process variables (Han et al., 2001).
5.2 Controllability analysis and control structure selection for the FCCU
5.2.1 General considerations for the control of the FCCU
In general, the key objective of controlling the FCCU is to keep the process running within
acceptable safety, environmental, and staff workload limits, etc, while simultaneously guaranteeing
economic plant operation. Typically, a conventional hierarchical control system satisfies these
criteria, where each level performs a specific task or tasks (see Figure 5.1). In the schematic not all
the logic control functions are presented, e.g. startup/shutdown and safety systems, as these do not
feature in normal plant operation. There are various ways for the concept to be implemented in a
control system, but more importantly the idea of distributing control to the different levels is key.
124
This chapter discusses the regulatory control level, which is the lowest in the hierarchy, and solely
tasked with maintaining smooth running of the process. In contrast, economic optimisation is
handled by higher levels in the hierarchy. Typically, such a control system is both decentralised and
cascaded, where specific measurements for set points are acquired in the implementation of the
former, and then set point values commanded by the upper levels or the human operator, in the
latter. Cascaded loops may also be found within the regulatory control level, in that a valve position
is cascaded into a measurement of flow, in which case, flow is a manipulated variable not the
position of the valve. Henceforth, regulatory control system and lower-level control system are
used interchangeably.
Plant wide
optimization
Supervisory control
Set points
Regulatory control
Measurements
Process
Manipulated
variables
Figure 5.1: Hierarchical control system.
125
Within this control hierarchy, an intermediate or supervisory level may hold a multivariable,
process model-based system that computes optimised plant operating parameters according to a
specific objective. This supervisory level is characterised by the capability to handle constrained
manipulated and controlled variables, especially in avoiding saturation of the manipulated
variables. This preventative action may require coordination between control loops through
assigning set values, or perhaps directly introducing more degrees of freedom.
The control hierarchy upper or top level is concerned with optimisation across the entire process
plant, which is typically in the steady state and taking place regularly off-line. Even though this top
level is where the most significant economic gains are realised, proper functioning of the lower
levels is necessary for this to happen. This fact is often not properly understood.
5.2.2 Objectives for regulatory control

A control system regulating process plant should satisfy several criteria or objectives:
1. Control quality needs to be at a level where operating personnel are able to maintain the
plant operating in safety, without need for action on part of the control system higher
hierarchical levels. This enables simple recovery from failures, where no expensive backup
is needed for the higher levels.
2. The system needs to be easily understood and tuned, which is usually solved by using
control loops that are simple and decentralised. However, where significant interactions
exist, multivariable control may be introduced.
3. At the top level, the system should as far as practicable use simple models, in order to
enhance reliability and reduce cost in formulating and implementing a complex dynamic
plant model, which will have a computation overhead applied at the top level. This is
resolved by ensuring that the regulatory control level is at the lower level of the system
hierarchy, which has the advantage of minimising model uncertainty and facilitating local
linearization.
4. The system should allow the upper levels in the hierarchy to have longer sampling times,
which would lessen the computational burden at those levels. Distinct sampling times
should be used, with faster times at the lower levels, and so higher levels will not wait for
responses from the lower levels.
126
Therefore, based on the criteria or objectives above, a set of four further objectives for the
regulatory control may be formulated:
5. Rapid control should be possible, particularly in the case of some variables.
6. The control system should operate the process closely following the set values commanded
by the hierarchy top levels. These set values sent to the bottom level represent manipulated
variables, where the aim is to move the process to these new values directly with minimum
interaction, avoiding the need to map interactions and process dynamics in a model residing
at the top levels.
7. The system should effectively reject local perturbations, which is an extension of the
second objective, and aims at maintaining the controlled parameters at the set points. The
system needs to include extra degrees of freedom or additional manipulated variables
(unused) that are directly changed by the top levels in the system hierarchy.
8. The system must be robust, i.e. it should not be unnecessarily limited in performance due
to, for example, significant sensitivity to perturbations, or have RGA elements. This
remaining control problem is that which is seen by the top level, and contains manipulated
inputs, both used and unused, as well as lower level set values. The unnecessary
performance limitations refer those which are not part of the original control problem
formulation in the absence of the control system lower level (Hovd and Skogestad, 1993).
The first three objectives (5,6,7) above relating to rapid control, set point following, and
perturbation rejection are discussed here, and relate to system controllability at the lower level. The
criterion of simplicity is satisfied by the system lower level being completely decentralised.
5.2.3 Control structure selection for the FCCU using RGA

The criteria or objectives presented in the previous section are satisfied primarily by choosing the
appropriate control strategy or structure, and then deciding the following:
Outputs y: the regulatory system controlled variables, both primary and secondary, or
control objectives, are chosen. Primary outputs can normally be chosen easily, as they
represent key process variables, or those needed to achieve a control objective. Such
variables may need rapid control, as previously discussed in the system objectives (5), and
include liquid levels, and specific pressures and temperatures to be maintained within strict
limits. On the other hand, secondary outputs are variables, which can normally be measured
127
easily, but are not essential to control. They are chosen to satisfy some of the objectives (1,
2, and 4) discussed previously. Such variables are usually pressures and temperatures at
specific points in the process plant. Hence, the choice of regulatory control system output
variables is closely linked to the problem of measurement choices.
Inputs u: regulatory control system manipulated variables are chosen. The chosen inputs
represent a subset of the whole range of manipulated inputs, while any unused variables
represent manipulated inputs that can be used ether by the top control hierarchy, or the
plant operating personnel.
Pairing: the controlled and manipulated variables selected are paired to achieve
decentralised control. Such pairing will affect the nature of system interactions and
disturbances. Moreover, in this decentralized control system, the tolerance to failure in one
or many loops is also determined by pairing.
Selection of the outputs

Generally several variables are controlled in an FCCU. However, in this work controlling two or
three variables at the regulatory level will be considered, which represent the most important output
parameters and also exemplify the difficulties of the multivariable controller design for interacting
systems. These parameters are: Tr, Treg, and cO2,sg (Grosdidier et al., 1993). The justification for
trying to control these three variables is as follows:
The product distribution is determined by the reaction conditions inside the riser, which are
therefore very important for the economic performance of the FCC process. There is then an
incentive to control both Tr and Treg which are directly related to the riser outlet and inlet
temperature, respectively. A secondary controlled variable is not necessarily less important than a
primary controlled variable. For example, although Tr is classified as a secondary controlled
variable because it in itself is not very interesting, the importance of Tr as a controlled variable
comes from the close connection between Tr and conversion. The need to avoid afterburning should
be obvious and makes it reasonable to control directly cO2,sg.
Based on this discussion, the primary controlled variables considered in this work is the cO2,sg.,
whereas the secondary variables are Tr and Treg , for which measurements are generally available.
128
Independent variables (manipulating inputs and disturbances)

The following nine independent variables were considered:
svrgc = regenerated catalyst slide valve position.

svsc = the spent catalyst slide valve position.
Fair_c = flow rate of air to the regenerator
F5 = furnace fuel flow rate
F3 = fresh feed gas oil flow rate
F4 = slurry recycle flow rate
V14= stack gas valve
tair = air temperature
f = feed oil composition; here represented by the coke production rate factor
The feed oil composition may be adjusted by changing the ratio between recycled and fresh feed.
There are actually several additional manipulating variables, but we have assumed that these are
already used by the regulatory control level to control holdups and pressure. These variables
include:
Fsc = spent catalyst flow rate
Ffg = flue gas flow rate
Wwg = wet gas compressor throughput, as the wet gas compressor is situated downstream of the
distillation column receiving the reaction products.
Fsc is used to control the catalyst holdup in the separator, whereas F3 is used to control the
regenerator pressure P6. Wwg indirectly controls the separator pressure P4. In practice the pressures
P6 and P4 may have to be adjusted when the catalyst slide valve position svsc is changed, in order to
avoid reversal of the catalyst flow (Fsc).
All the nine independent variables above may be used as manipulating variables (us) for control,
but in most cases we will only use two: the regenerated catalyst flow rate, Frgc and the air flow rate
Fair_c .The remaining independent variables may then be regarded as disturbances (ds) in the
regulatory control system. The variables f , F5, F4 and F3 are all related to the oil feed. The
129

variable f is the coke production rate factor and depends on the feed composition. Immediately
downstream of the FCC there is a distillation column which separates the products from the
cracking reactions. The heavy fraction from the distillation column, slurry, has a large coke
producing tendency. The coke production rate factor can therefore be changed indirectly by
changing the amount of slurry, which is recycled to the riser. The air temperature tair is generally a
disturbance since there is usually no air preheater in the system.
Consideration of typical operational constraints for the FCCU
The constraints that apply to plant operation must be carefully considered in making choices
regarding the manipulated input and controlled output variables. For the FCC, one or more
constraints determine the optimal operating point, and so the preferred control structure is the one
that operates the plant nearest to these constraints. The plant optimal operating point will vary
according to the properties of the feedstock, and the product range aimed for, which obviously
influences the relative importance of the different constraints. This encourages the development of
a different control structure for each optimal operating state. Yet, it may not be practical to
formulate a different control structure for different operating conditions. Control constraints that are
commonly encountered, may include:
Flue gas minimum oxygen concentration (cO2,sg): The concentration of oxygen in flue gas
needs to be minimum in that complete combustion will have taken place previously within
the dense regenerator bed, where CO is converted into CO2 (complete combustion mode).
This is an important constraint that ensures that no afterburning takes place, as there is
insufficient oxygen in the flue gas.
Wet gas compressor maximum capacity: The FCC unit wet gas compressor compresses the
output products ready to be taken for gas treatment downstream.
Air blower maximum capacity (FT): The blower capacity needs to be at the maximum
required to supply air for combustion within the regenerator.
Implications for Regulatory Control

For the purpose of operating the plant close to the constraints, it is worthwhile to choose the
constrained outputs (measurements) to be controlled by the regulatory system. Therefore, in the
complete combustion mode, flue gas oxygen concentration, cO2,sg, may be adopted as a controlled
variable.
130
At the same time, in the control system, it is wise to avoid those manipulated inputs that may cause
constraints to be easily reached. For example, wet gas production is affected significantly by feed
flow rate F3, and so in a process state nearing the wet gas compressor constraint, feed flow rate F3,
should not be adopted as a manipulated variable. A similar scenario involves the air flow rate Fair_c
and air blower capacity constraint. However, a number of workers (Pohlenz, 1963; Hicks et al.,
1966) have proposed air flow rate, Fair_c, as a manipulated variable. Therefore, the control system
should prevent the process from violating the air blower capacity constraint. This may be achieved,
for example, by modifying the feedstock or the conditions of reaction leading to less coke
formation. FCC constraints were treated in detail by Grosdidier et al. (1993), of which a key
selection is presented in Table 5.1.
Table 5.1: Operating constraints and their relative importance (0: low; 10: high).
No.
Process variable
Notation
Constraints
Importance
Tr
Tr < 537 C
10
1.
Temperature of reactor riser
2.
Temperature of regenerator bed
Treg
720 C < Treg < 745 C
10
3.
Concentration of O2 in stack gas
cO2,sg
cO2,sg > 2 %
4.
Concentration of CO in stack gas
cCO,sg
cCO,sg > 350 ppm
9.5
5.
Differential pressure between
dprr
0.15 bar < dprr < 0.2 bar
regenerator and reactor
Controller and manipulated variable pairing

The complexity of manipulated and controlled variable pairing is such that systematic mathematical
analysis is needed to account for more than physical or process parameters. A well-configured
control system would be capable of keeping the process within specified constraints, and rapidly
restore it to its nominal operating state. In the context of choosing the appropriate control structure,
related to multiple single-loop controllers (see Chapter 2), and determining interaction among the
variables, the relative gain array (RGA) and variations of it (Wolff et al., 1992; Skogestad and
Wolff, 1992) is commonly applied. As RGA is a linear approach, the process model equations,
which are non-linear, need to be linearised for different points of operation, as illustrated below.
The FCC unit model may be considered to be composed of vector differential-algebraic-equations
(DAEs), as follows:
131
dx
dt
y
f (x , u )
(5.1)
y(x, u)
(5.2)
with state (x), input (u) and output (y) vectors expressed by
x
x1,..., x n
(5.3)
u1,..., um
(5.4)
y1,..., yk
(5.5)
Linearizing the system in the operating point (xo, uo) gives the following linear model
dx
dt
Ax
Cx
Bu
(5.6)
Du
f
x
f
u
(5.7)
y
x
C
xo,uo
yo,uo
y
u
D
xo,uo
(5.8)
yo,uo
The RGA matrix for a square (n x n) plant can be calculated using the following formula:
RGA
C A
C A
(5.9)
where the symbol is the Hadamard product (element by element product) and A, B, C, and D are
coefficient matrixes from the linear state space model.
For independent control loops, pairs where the input and output relative gain is close to 1, are the
best. In contrast, pairs that exhibit relative gains that are negative in the steady state must be
avoided (Bristol, 1966). Such a pairing will cause whole system or single loop instability in an
integral control scheme or on removal of that specific loop (Grosdidier et al., 1985). Furthermore,
process plants that exhibit high values of RGA, especially at high frequency, suffer inherently poor
controllability, regardless of the control system used (Skogestad and Morari, 1987b). Moreover,
132
implementing decentralised control, where input and output pairs have low RGA values (compared
to 1) in the bandwidth region, signals potential problems in stability (Hovd and Skogestad,
1992a,b).
Table 5.2 shows different control structures which were considered for control implementation
From the RGA analysis the best results were obtained with control schemes 1 and 2 (Table 5.2) for
which RGA matrix values close to 1 were obtained. Table 5.3 gives a suggested pairing for scheme
1 Tr-F5, Treg-svsc. In Table 5.4 the suggested pairing for scheme 2 Tr-F5, Treg-svsc, cO2,sg-V14 is
given. The results obtained with the other schemes from Table 5.4 were not conclusive since there
are small and negative values, indicating that those configurations potentially would not work.
These points out strong couplings between some of the control loops especially when three or more
control loops are considered.
Table 5.2: Different control structure schemes for PID controller.
Control scheme
(2x2)
(3x3)
(4x4)
(5x5)
(6x6)
Manipulated
F5,
F5, svsc,
F3, F4,
F3, svsc, F5,
F3, F4, svrgc,
inputs
svsc
V14
VB, V11
V14, svrgc
svsc, LT,VB
Controlled
Tr,
Tr, Treg,
Tr, P6,
Tr, Treg,
variables
Treg
cO2,sg
X(1), X(82)
P5, T3, P6
X(1), X(82), Wr,

Tr, Treg, Wreg
Table 5.3: RGA matrix of a sample 2X2 control scheme.
Controlled
Manipulated Variables
variable
F5
svsc
Tr
Treg
133
Table 5.4: RGA matrix of a sample 3x3 control scheme.
Controlled
variable
F5
svsc
V14
Tr
-0
Treg
-0
1.1297
-0.1297
cO2,sg
-0.1297
1.1297
5.3 Advanced PID control of the FCCU

5.3.1 Design of the PID control algorithm
In the domain of controlling FCC units and achieving specific objectives, many approaches quite
advanced techniques. A block diagram of a PID controller is presented in Figure 5.2, which
illustrates this generic feedback controller, commonly used in industrial and process control
applications. In this controller, the values of the measured process variable and the commanded set
value are compared revealing a difference termed the "error", which the controller aims to reduce to
a minimum through appropriate control inputs. The PID or three term algorithm involves
proportional (P), integral (I), and derivative (D) constant values, which may be explained
heuristically in that P is based on present error, I on past error accumulation, while D predicts
future errors, according to the present rate of change (Taguchi and Araki, 2000). The process is
adjusted by applying control actions, based on the weighted sum of these three constants, to a
controlled element; for example, a valve or heating circuit power supply.
The PID controller does not require any prior knowledge of the process to be controlled, which is a
major advantage (Bennett, 1993). The controller is tuned for the particular process by adjusting the
values of the three parameters, and controller response may be defined according to responsiveness
to error, set point overshoot, and degree of oscillation. However, despite its advantage in general
applicability, PID control may not provide optimised operation or assure process stability.
The PID algorithm allows one or two of the three parameters to be disabled simply by assigning a
zero value to the relevant constant, and hence may become a PD, PI, I or P controller accordingly.
Due to the sensitivity of derivative action (D) to noise, it is quite common to encounter PI
controllers. However, the integral (I) is important to enable command values to be achieved.
134
Figure 5.2: A block diagram of a PID controller.
The simulator described in the previous chapter was used to conduct simulations for closed-loop
FCCU control. This used the MatLab-Simulink interface to apply an advanced PID algorithm
(discrete position algorithm with backward approximation and anti-windup) that facilitates stability
of the closed loop. Anti-windup in the algorithm was the main reason for the improved PID
controller performance, and overcomes the physical limits associated with control elements. For
example, a valve cannot exceed its fully closed or fully open positions, as this is physically
impossible. The effect on control is severe, as the integral action may lead to instability; when the
control element is saturated, simply by the fact that the output cannot change regardless of the
input; this breaks the feedback loop. This unstable operating mode will cause a drift, involving high
values, and even after the control element is desaturated, the system may take even longer to be
restored to a stable state. Moreover, the control element may also experience, high and low value
swings before recovery. Anti-windup, also termed tracking or back calculation, ensures that the
integral is maintained at an appropriate value, once the control element saturates, which enables the
controller to act properly, once again, after the error state changes (Bohn and Atherton, 1995).
5.3.2 Constrained optimization based MIMO PID tuning

The PID controller is in general a single input single output (SISO) controller which has a set of
tuning parameters, which can be used to fine-tune the closed-loop response for good performance
and stability. Tuning the multiple interacting PID loops is generally very difficult due to the high
level of coupling between the loops.
135
In practice often the loops appear to be tuned independently, however when they are activated
simultaneously the process can exhibit poor control performance or can be destabilized. To take the
coupling between the loops into account the controllers should be tuned simultaneously. This work
proposes a method for the tuning of the multiple PID control loops based on the use of a
constrained optimization problem, which minimizes the overall integral square error (OISE) of the
control loops defined as,
OISE
NL
tf
i 1
(ei (t ))2 dt
(5.10)
where ei (t) is the error function in the case of loop i , N L is the number of PID loops and t f is the
time horizon on which the error is computed. From the objective function the control effort was
omitted since the primary objective was to evaluate whether it is possible to find a set of tuning
parameters, which would provide suitable control performance in a given input-output structure.
However once a particular control architecture would be found, in a practical framework the control
efforts could also be included in the objective function (5.10). The optimization problem for the
tuning is expressed as follows
min OISE
(5.11)
Ki ,Ti
subject to:
kmin,i
ki
kmax,i
Tmin,i
Ti
Tmax,i ,i
(5.12)
1,2, , N L
(5.13)
where K i and Ti are the proportional gain and integral time, respectively, for the PID controller of
loop i , and the constraints (5.12) and (5.13) sets the minimum and maximum limits on the tuning
parameters. This is a highly nonlinear and nonconvex optimisation problem which may exhibit
multiple local minimums, has the solution generally represents a complex numerical problem.
Additionally due to the presence of constraints and input saturations the computations of the
gradients for gradient-based optimisation approaches may be difficult. In this chapter a Genetic
Algorithm-based optimisation framework is proposed for the tuning of MIMO interaction PID
control systems.
The main advantages of the genetic algorithm (GA) include that it represents a global optimisation
approach (although it does not guarantee that the final global optimum can be achieved within finite
136
number of iterations) and in the large numerical stability of the approach due its derivative-free
nature (Schaffer, 1984; Kursawe, 1990; Hajela et al., 1992; Murata, 1997; Srinivas and Deb, 1994;
Fonseca and Fleming, 1993; Horn and Nafploitis, 1993; Deb, 2001; Carlos et al., 2002; Goldberg,
1989; Holland, 1975; Petry et al., 1994; Davidor, 1991; Karr, 1992; Kacprzyk, 1995; Linkens and
Nyongesa, 1995a; Linkens and Nyongesa, 1995b; Surmann et al., 1993).
The GA typically uses binary digits to form character strings to represent codes for potential
(candidate) solutions in an optimisation. This character string, consisting of binary digits, is termed
a chromosome, mirroring the terminology used in genetics. The candidate solution indicated by
the chromosome is termed an individual, where the GA is concerned with many individuals
gathered in a population. The individuals in the population are iteratively modified and updated,
in each step termed a generation, where the optimisation proceeds through a search for the best
solutions. A fitness function within the GA assesses the individuals, and reveals the extent to which
they are satisfactory; in addition, an objective function may be used for comparison in a classical
approach to optimisation. The fitness function enables the GA concept of survival of the fittest to
be applied, where the algorithm seeks to achieve maximum fitness values rather than the classical
optimisation approach of seeking to minimise some objective function. As such, specifying the
fitness function is a vital aspect in GA design, leading to output of optimised solutions, and
appropriate algorithm performance. In each iteration (generation), the GA evaluates the candidate
solutions (individuals) by computing their fitness values, and then generating new individuals based
on the old information.
The generation of new solutions is quite distinct in Gas compared to approaches in numerical
optimisation. First, an initial population composed of individuals is generated, which describes the
primary estimates regarding the optimisation. In each iteration step, the GA evaluates a group of
solutions, as opposed to one solution, as is the case in gradient descent and similar approaches. The
candidate solutions (individuals) are assessed by computing their fitness values, with new ones
being formulated to form part of the subsequent generation. Given its distinct approach, the GA
does not involve calculating the Hessian or Jacobian matrices, which may produce an illconditioned problem for a stiff and highly dimensional FCC unit model, negatively affecting the
optimisation. In contrast, simulation has demonstrated that even with a small population, GA
convergence may be rapid; 2x2 and 3x3 systems converged over 10 generations only to a global
optimum (see Figures 5.3 - 5.4). The resulting tuning parameters for the individual PID loops are
presented in Table 5.5. Tracking the evolution of fitness values (average and best) over successive
generations reveals that within the initial generation, tuning parameters were quite close to the best,
137
and so no significant improvements were detected. This was because the initial tuning parameters
had been obtained using standard (Ziegler-Nichols) methods for SISO systems applied to the
individual controllers. Over the ten generations, there was a sharp decrease in mean fitness values,
in parallel with convergence; this indicates that the parameters represent the global optimum, where
all the individuals (candidate solutions) had been converging to the best solution, over successive
generations. Furthermore, this demonstrated that for the system, individual and independent tuning
of the controllers may result in achieving the same level of performance as simultaneous tuning of
interacting controllers. In industry, the norm in the FCC unit is that PID controllers are tuned
individually.
Table 5.5: Tuning parameters of the PID controllers.
Controller
Parameters
Type
K1
T1 [s]
K2
T2[s]
K3
T3[s]
Td [s]
Tsamp [s]
OISE
3x3
0.0009
5.9
0.001
5.5
0.003
0.8
100
1.07 E5
2x2
0.0367
5.0
0.0004
63.953
100
8.081 E4
3.5
x 10
Best fitness
Mean fitness
3
Fitness value
2.5
1.5
0.5
5
Generation
Figure 5.3: Fitness value for 2X2 PID controller system tuning.
138
10
x 10
Best fitness
Mean fitness
8
7
Fitness value
6
5
4
3
2
1
0
5
Generation
10
Figure 5.4: Fitness value for 3X3 PID controller system tuning
5.3.3 Multivariable PID control of the FCCU

3x3 structure
The MIMO closed loop system designed and implemented in Matlab/Simulink environment with
the most common control scheme using PID controller is shown in Figure 5.5. Process simulations
with three independent control loops were carried out. Based on the RGA results in Table 5.4 the
following control pairings were used: Tr-F5, Treg-svsc, cO2,sg-V14. The control performance was
evaluated in the case of regulatory (disturbance rejection) and servo control (set point tracking).
The regulatory control has been investigated in the case of a +5% disturbance in the feed oil
composition represented by the coke production rate factor f and in the case of a setpoint change
as indicated in Table 5.6.
139
Figure 5.5: SIMULINK block diagram for the closed loop simulation with three PID controller loops.
Table 5.6: Tested data for 3 x 3 PID controllers in MIMO loops parameters.
Loop
Controlled
Manipulated
variables
variables
Tr - Treg -cO2,sg
Disturbance
F5 - svrgc - V14
Control scenario
Servo, C
Regulatory,%
Tr541.6
+5
Treg737.7
cO2,sg 4
In the case of the regulatory control scenario considered the +5% increase in coke formation factor
in the feed oil occurs at time 1.5 h. The control system is able to maintain the controlled outputs
within small variations, compared to the uncontrolled case (shown in Chapter 4). Variations are
within 1-1.5 C for the reactor temperature and bellow 4 C for the regenerator temperature,
whereas for the uncontrolled case these are around 5 and 8 C, respectively. Whilst the CO2
emission reaches similar maximum levels in the case of the controlled scenario as for the
uncontrolled case, the control systems eventually brings the emission level down , below 4%.
140
Similar variations can be observed in the case of the CO emission, too. The variations in the
product composition (Figure 5.7) are also relatively small. The settling time however is
considerably long. It takes more than 6 hours for the control systems to bring the process back in
the original operating point. Such a long period of operation with product specifications outside the
required boundaries may lead to considerable economic effects. These results indicate that although
the proposed control structure is able to eventually suppress the effect of the investigated
disturbance on the process relatively well, this takes a significant amount of time, hence if the
process is subject to frequent disturbances this may lead to poor overall control performance and
potentially considerable economic and environmental losses.
Tr(C)
533
532
531
530
10
Trgc (C)
Tset,r
12
Trgc
14
Tset,reg
CO2sg (mol%)
730
F5 (scf/s)
svsc (%)
735
0
4
3
2
10
CO2sg
12
14
CO2set,sg
10
12
14
10
12
14
10
12
14
10
12
14
60
40
20
0.3
0.25
0.2
V14 (%)
Tr
0.64
0.62
0.6
0.58
0.56
0.54
0.52
time (h)
Figure 5.6: Performance of the regulatory controls for (Tr vs. F5), (Treg vs. svsc) and (cO2,sg vs. V14) in the
presenceofstepincreaseinthecokingfactorf +5%, at t = 1.5 h.
141
CCosg (ppm)
10
12
14
10
12
14
10
12
14
10
12
14
LCO (mol%)
0.21
0.2
0.19
Gasoline (mol%)
0.18
0.46
0.44
0.42
0.4
LPG (mol%)
0.18
0.17
0.16
time (h)
Figure 5.7: Dynamic behaviour of other uncontrolled process variables in the case of the disturbance
rejection scenario in Figure 5.6.
In the case of the servo control scenario (tracking control), a positive setpoint change was applied
to al three controlled variables at the same time. Figure (5.8) shows the results of the closed loop
simulation indicating that whilst the control system eventually brings all three process outputs to
the new setpoints this takes a variably long time. The reactor temperature is brought to the new
setpoint within 2 hours whereas the regenerator temperature and cO2 concentration takes
considerably longer, about 6 hours for Treg and 10 hours for cO2. The effect of operating the system
at the new setpoint values, on the product composition and CO emissions are shown on Figure 5.9.
.
142
Tr(C)
540
Tr
530
Trgc (C)
10
Tset,r
12
14
740
Trgc
CO2sg (mol%)
730
0
10
Tset,reg
12
14
5
4
CO2sg
CO2set,sg
10
12
14
10
12
14
10
12
14
10
12
14
F5
60
40
20
svsc
0.4
0.3
0.2
V17
0.65
0.6
0.55
time (h)
Figure 5.8: Performance of the servo control for (Tr vs. F5), (Treg vs. svsc) and (cO2,sg vs. V14) in the
presence of step increase, at t = 1.5 h.
143
CCosg (ppm)
3
2
1
0
10
12
14
10
12
14
10
12
14
10
12
14
LCO (mol%)
0.3
0.2
Gasoline (mol%)
0.1
0.5
0.45
0.4
0.35
LPG (mol%)
0.24
0.22
0.2
0.18
0.16
time (h)
Figure 5.9: Dynamic behaviour of other uncontrolled process variables in the case of the setpoint
tracking scenario in Figure 5.8.
2 x 2 Control Systems
The output variables considered in this study, Tr and Treg, are strongly coupled, and a control that
stabilizes one of the parameters may be able to stabilize inherently the other, too. The manipulated
variables were chosen based on their behaviour and ability to stabilize the system as shown in the
simulation results using the SISO PID control systems. Two 2x2 control structures were evaluated
in the case of regulatory control and setpoint tracking as shown in Table (5.7). In the first structure
evaluated the manipulated inputs svrgc and V14 were tested for the control of the reactor
temperature and cO2 concentration. Wolff et al. (1992) also proposed a similar control architecture
144
based on their analysis of the disturbance sensitivities for the FCCU system studied. The closedloop simulation model was implemented using two separate PID control loops as shown in Figure
(5.10) and the parameters were tuned using a tuning approach based on a Genetic Algorithm (GA).
Figure 5.10: SIMULINK block diagram for the closed loop simulation with two PID controller loops.
Table 5.7: Tested data for PID controllers in MIMO loops parameters.
Loop
Controlled
Manipulated
variables
variables
Disturbance
Tr - cO2,sg
svrgc V14
Tr -Treg
F5 - svsc
Controller problem
Servo, C
Regulatory, %
+5
Tr
541.6
+5
Treg 737.7
The simulation results are shown in Figures (5.11 5.16). Both control structures provide excellent
performance. The plots clearly show that the Tr , Treg and oxygen concentration can be controlled
and brought to their steady states or new setpoints. Usually temperature control systems in large
unit operations have a large time constant, but in the riser, the gas oil and catalyst flow rates are
145
very large, residence times are short, the cracking reactions are quick, therefore the riser outlet
temperature reveals a quick response to the regenerated catalyst valve opening.
534
Tr(C)
Tr
Tset,r
532
530
10
12
14
CO2sg (mol%)
4
CO2sg
CO2set,sg
3.5
3
2.5
10
12
14
10
12
14
10
12
14
svrgc
0.3
0.25
0.2
V14
0.64
0.62
0.6
time (h)
Figure 5.11: Model responses for the reactor Tr and regenerator cO2,sg temperatures by manipulated
svrgc & V14 withchangesinthecokingfactorf +5%, at t = 1.5 h.
146
CCosg (ppm)
6
4
2
0
10
12
14
10
12
14
10
12
14
10
12
14
LCO (mol%)
0.22
0.21
0.2
Gasoline (mol%)
0.19
0.44
0.42
0.4
LPG (mol%)
0.185
0.18
0.175
0.17
time (h)
Figure 5.12: Dynamic behaviour of other uncontrolled process variables in the case the scenario in
Figure 5.11.
147
532.5
Tr
Tr(C)
532
Tset,r
531.5
531
10
Trgc (C)
734
12
Trgc
14
Tset,reg
732
730
728
0
10
12
14
10
12
14
10
12
14
F5
60
40
20
svsc
0.3
0.25
0.2
time (h)
Figure 5.13: Model responses for the reactor Tr and regenerator Treg temperatures by manipulated F5
& svsc withchangesinthecokingfactorf +5%, at t = 1.5 h.
148
LPG (mol%)
Gasoline (mol%)
LCO (mol%)
CO2sg (mol%)
CCosg (ppm)
4
2
0
10
12
14
10
12
14
10
12
14
10
12
14
10
12
14
6
4
2
0.21
0.2
0.19
0.44
0.43
0.42
0.18
0.17
0.16
time (h)
Figure 5.14: Dynamic behaviour of other uncontrolled process variables the case scenario in figure
5.13.
149
545
Tr(C)
540
535
530
Tr
0
10
Tset,r
12
14
745
Trgc (C)
740
735
Trgc
730
725
Tset,reg
10
12
14
10
12
14
10
12
14
80
F5
60
40
20
svsc
0.35
0.3
0.25
time (h)
Figure 5.15: Performance of the servo control for (Tr vs. F5) and (Treg vs. svsc) in the presence of step
increase, at t = 1.5 h.
150
1
0.5
0
LPG (mol%)
Gasoline (mol%)
LCO (mol%)
CO2sg (mol%)
CCosg (ppm)
10
12
14
10
12
14
10
12
14
10
12
14
10
12
14
6
4
2
0.4
0.2
0
1
0.5
0
0.25
0.2
time (h)
Figure 5.16: Dynamic behaviour of other uncontrolled process variables the case scenario in Figure
5.15.
5.4 Conclusions
The dynamic simulator was used to study the closed-loop dynamic behaviour of the process, using
advanced PID controllers for the multivariable process plant. A simple RGA matrix based
controllability analysis was performed in order to study the interdependence among the process
variables. The PID algorithm was tried for different control structures. The results revealed the
strong couplings among some of the control loops, which may lead to difficulties in tuning the
individual control loops. A genetic algorithm based tuning approach is proposed to simultaneously
151
identify the controller parameters for the interacting SISO PID control loops. Sample 2x2 and 3x3
control schemes were evaluated for disturbance rejection and set point tracking. The results
demonstrate that as the number of independent control loops increases, the performance of the PID
controller scheme strongly deteriorates, as indicated, for example, by significantly increased
settling time. While the control system is able to stabilize the system, the decreased performance in
the case of frequent disturbances may lead to significant economic losses or negative environmental
impact, providing incentives towards the implementation of more advanced multivariable control
systems.
152
Chapter 6: Multivariable Linear and Nonlinear MPC of the FCCU
CHAPTER 6
MULTIVARIABLE LINEAR AND NONLINEAR
MPC OF THE FCCU
6.1 Introduction
FCC unit control has been investigated by many workers, including the application of multivariate
and nonlinear predictive control approaches (Alvarez-Ramirez et al., 1996; Aguilar et al., 1999;
Abou-Jeyab and Gupta, 2001; Kiss and Szeifert, 1997; Balchen et al., 1992; Grosdidier et al.,
1993). A multivariable control scheme was proposed for an industrial FCC unit with controls in the
reactor and regenerator; it was concluded that this offered a robust solution resulting in good
dynamic performance (Grosdidier et al. 1993). Temperature control of FCC units was also achieved
using a state-space model based predictive controller, with the non-linear optimisation problem
solved in real-time. Simulation studies demonstrated good closed loop performance. However, the
model-based control schemes are characterised by high computational costs, due to the large
numbers of process variables and the complex interactions between the controlled and manipulated
variables (Balchen, 1992). Non-linear control, accommodating uncertainty, was also applied to the
temperature control in the FCC regenerator and riser (Alvarez-Ramirez et al., 1996; Aguilar et al.,
1999). The closed loop controllers developed using this concept were practical, and regulated the
temperature near the set point. Hovd and Skogestad (1993) reported that the key to satisfactory
control of the FCC units is in choosing the appropriate controlled variables for the linear model
predictive control scheme investigated. It was demonstrated that control of FCC units presents a
non-linear, multi-variable, complex dynamic problem, where the control loops frequently interact
strongly. At the same time, FCC unit operation takes place in the presence of constraints, and so the
control strategy must be designed to accommodate these complex input-output interactions, and to
demonstrate robustness in face of operating non-linearity, and errors in models.
In the practical control of FCC units, as mentioned previously in Chapter 5, generally the preferred
control loops are decentralised, temperature in the reactor and regenerator being the main
parameters that need to be closely controlled (Christensen et al., 1999; Avidan and Shinnar, 1990).
Temperature as an independent variable plays a critical role in adjusting the activity of the catalyst
153
in the steady state, and influences the product properties, such as octane number of gasoline.
Moreover, process efficiency, conversion, and product yield, as well as plant safety are affected by
the temperatures in the system. Therefore, in the key parts of the FCC unit, i.e. reactor, regenerator
and riser, temperature needs to be controlled for optimal operation, and kept within the plants
design limits.
This meant that a new design and implementation of linear and non-linear controllers was needed to
handle the challenging control problem for the FCC process. Furthermore, the strong nonlinearity
of the FCC process and the presence of large number of operational, safety and environmental
constraints require the use of a multivariate model-based controller. In this context, model
predictive control (MPC) offers a promising solution to the FCC unit control, as it accommodates
multiple inputs and outputs, constraints, and is an optimisation based strategy (Emad and Elnashaie,
1997; Karla and Georgakis, 1994).
Based on these preliminary aspects, two main types of model predictive control (MPC) approaches
are investigated in this chapter: (i) linear model predictive control (LMPC) and (ii) nonlinear model
predictive control (NMPC). The former approach is also investigated under the realistic scenario
which considers model-plant mismatch, by using the linearized model in the LMPC scheme and
using the complex first-principles nonlinear model for the plant simulator.
6.2 Linear MPC design and implementation

6.2.1 LMPC simulation using the linear model as plant
Linear Model Predictive Control (LMPC) is one of the most widely used advanced control methods
in the chemical industry. Also referred as moving (receding) horizon control, has become an
attractive control strategy especially for linear but also for nonlinear systems subject to input, state
and output constraints. There are some features that individualize MPC in the field of control
design making it attractive. In contrast to other feedback controllers that calculate the control action
based on present or past information, MPC determines the control action based on the prediction of
future dynamics of the system. Due to the future prediction, early control action can be taken
accounting for future behaviour. In practice, most of the systems have to satisfy input, state or
output constraints, resulting in limitations on achievable control performance (in the extreme case
affecting the stability). The main idea of MPC is to calculate a set of future manipulated variable
154
moves (on the control horizon), which is calculated to minimize an objective function based on the
sum of squares of the differences between model predicted outputs and a desired output variable
trajectory over a prediction horizon (p). The mathematical formulation of the LMPC problem is
given in detail in Chapter 2.
As highlighted in Chapter 2, the tuning of a MPC is challenging, particularly in the multi-inputmulti-output (MIMO) configuration, which is the most commonly used structure for MPC
applications (Garcia et al., 1989; Semino and Scali, 1994; Lee and Yu, 1994). Potential tuning
variables include the input and prediction horizons, designated m and p respectively, as well as
sampling time, Ts The MIMO features of the problem present the difficulties in tuning, where
changes in the variables influence control performance. The challenges are more significant, where
model non-linearities exist (Chen and Allgower, 1997). The nature of these challenges dictates an
iterative approach to MPC tuning, with further improvement in controller performance achieved
through recursive simulations. The task of MPC tuning has been greatly facilitated with publication
of MPC tuning guidelines (Karla and Georgakis, 1994; Semino and Scali, 1994; Chen and
Allgower, 1997). In this case, choice of sampling time, Ts, is a compromise between potentially
exceeding the computational capacity of the system, on one hand, and missing key data for the
dynamic process, on the other.
Taking the existing tuning guidelines into account for the FCCU process a sampling time of Ts=100
s was selected. Similarly, since high values for prediction and control horizons (p and m
respectively) result in greater computation, while small values leads to control that is shortsighted, therefore, p = 20 was chosen, which with the sampling time of 100 s corresponds to a
quarter (2000 s) of the open loop setting time of the process. Smaller values would result in a short
term control view, and so more controlling effort; moreover, for these smaller values of p,
constraint checking takes place over a small horizon, creating a dead zone where control is
inefficient. At the same time, larger values of p (compared to the settling time) result in slow (but
stabilising) control responses, which come at a greater computation cost. The value selected for the
control horizon, m is 5, which was found to provide a suitable compromise between computational
effort and enough degrees of freedom to stabilize the process and achieve the desired control
performance. Values less than this resulted in inefficient control action with degrees of freedom
that are insufficient, while larger values result in aggressive control actions.
Dynamic simulations were used to investigate various MPC schemes resulting in interesting results
with practical implications in the industrial setting.
155
Control scheme selection

The process parameters (variables) that are seen to play a key role in proper operation of the FCC
unit were identified from the published literature, and by analysing the commercial process
(Rhemann et al., 1989; Yang et al., 1996). These process variables are controlled to ensure that the
process is run safely and economically. The amount of catalyst or catalyst inventory in the reactor,
Wr, is controlled to stabilise circulation of the catalyst, and also creates a buffer preventing
disturbances due to coke deposition on the catalyst, as well as helping to stabilise the temperature
from reactor to regenerator. In the regenerator, a controlled temperature regime is necessary, with
Treg at a set value, for the process of coke burn-off and consequent catalyst regeneration to achieve
stability. Exceeding the limit of temperature permanently deactivates the catalyst, while a lower
temperature prevents proper coke removal, and results of coke accumulation and gradual decrease
in performance with time. The catalytic cracking reactions in the reactor also require good control
of temperature, Tr, for efficient feedstock conversion. Another controlled process variable, stack gas
oxygen concentration, cO2,sg, is key to burning the coke, and as such maintaining the thermal
balance, and maximising the combustion air blower load efficiency.
The manipulated process variables are independent variables selected for the practical reason that
they can be modified, where the main ones are for catalyst, preheating furnace fuel, stack gas, and
air vent flow rates. The catalyst flow, spent and regenerated, is controlled by slide valve positions,
svrgc and svsc respectively. The flow of fuel to the preheating furnace, F5, is a key variable in
maintaining the FCC unit thermal balance. The flow of regenerator stack gas is controlled by the
position of the stack gas valve, V14. Similarly, the flow from the air vent is controlled by the
position of the air vent valve position, V7. The variables that represent potential sources of
disturbances are ambient temperature, and the coking characteristics of the feedstock, i.e. rate of
coking.
MPC is inherently a multivariate control scheme which can automatically deal with interactions
between multiple inputs and outputs. Therefore in principle it is not necessary to choose input and
output variables that correspond to a close to diagonal RGA matrix, representing decoupled noninteracting control loops, when MPC is applied directly at the basic level of the typical control
hierarchy. However in most practical situations MPC is implemented on top of the basic control
system which consists typically of single-input-single-output (SISO) control loops mainly based on
PID controllers. Therefore in a practical industrial scenario before the implementation of the MPC
the first step is to select the control loops at the basic control level.
156
The controllability study is the first step in control scheme selection, and has shown its
effectiveness in both decentralised and multivariable control approaches (Hovd and Skogestad,
1991; Cristea and Agachi, 1996). In this step identifying the best input-output pairs with the highest
sensitivities but minimal interactions, can significantly improve the control performance of the
higher level MPC. In this context the RGA analysis facilitates the selection of the best manipulated
variables influencing the controlled variables, and hence the most appropriate MPC scheme.
Based on this approach, a set of control schemes has been investigated. They have a different
number of controlled/manipulated variables for square control structure (same number of inputs and
outputs): 2x2, 3x3, 4x4, and non-square control structures (different number of manipulated inputs
and controlled outputs): 2x4, schemes, presented briefly in Table 6.1.
Table 6.1: Input-output configurations tested with the MPC parameters used.
No Control
scheme
Controlled
variables
Manipulated variables
MPC tuning parameters

(duration, prediction and
control horizons and u )
2x2
Tr - Treg
F5 - svsc
[8000 s, Ts = 40, p = 20,

m = 5, u = 0.8]
3x3
Tr - Treg - Wr
F5 - V14 - svsc
[10000 s , (Ts = 80, p = 20,

m = 2), u = 0.8]
4x4
Tr - Treg - cO2,sg - Wr
svrgc - V14 - V7 - svsc
[10000 s , (Ts = 80, p = 20,

m = 5), u = 0.8]
4x2
Tr - Treg
F5 - svrgc - svsc- V14
[8000 s , (Ts = 80, p = 20,

m = 5), u = 0.8]
A linear model was used for the computation of the manipulated variables, obtained by the
linearization of the non-linear model around the nominal operating point (Patwardhan et al., 1990).
The RGA analysis was performed for all configurations. The result for control scheme 1 is given in
Chapter 5, whereas the RGA matrix for the control structures 2 and 3 are given in Tables 6.2 and
6.3.
Representative results from the control schemes given in Table 6.1 are presented in Figures 6.16.10. Note that the input and output variables represented in the figures are deviation variables
compared to their steady state values. This is due to the fact that the LMPC control design is
157
generally based on the linear model identified around a particular operating point. The controller
was tested in different setpoint changes. Figure 6.1 and Figure 6.2 indicate that the 2x2 control
scheme is able to provide a very good setpoint tracking, for a 10C and 15C change in the Tr and
Treg setpoints, respectively. The new setpoints are achieved quickly with no or minimal overshoot.
Additionally, the MPC provides a good decoupling between the two control loops, shown by the
relatively small disturbance in one of the controlled output when the setpoint changes for the other
controlled outputs. This is mainly due to the fact that the input-output pairs are well-correlated
according to the RGA analysis presented in Chapter 5. When the setpoint corresponding to any of
the controlled outputs changes, both manipulating inputs must react, on one hand, to bring one of
the outputs to the new setpoint value, and on the other hand to eliminate the effect of changes and
maintain the other output at its original setpoint value. The control actions are smooth, indicating
the control-change weight coefficient was chosen properly to achieve fast but stable closed-loop
performance.
Figures 6.3 and 6.4 illustrate the performance of the 3x3 control scheme, which includes as an
additional control output the inventory of catalyst in the reactor (Wr) and as an additional input the
stack valve position (V14). The setpoint tracking performance of the LMPC is very good, the new
setpoints are achieved very fast, with no overshoot and with minimal interactions. It is interesting to
observe that increasing the number of manipulating inputs the MPC controller is able to provide a
significantly better decoupling of the controlled outputs compared to the 2x2 case. When the
setpoint changes for the Tr or the Treg, practically no disturbance can be observed in the other
process outputs.
Similar observations are found by analyzing the results of the other 3x3 control structure on Figures
6.5 and 6.6. These results compared to the simulation results with the PID controllers for the same
structure presented in Chapter 5, also indicate a significantly better performance. However, note
that in the LMPC simulations no model-plant mismatch is assumed and the plant is represented by
the linearized process model.
The results of the 4x4 control scheme are presented in Figures 6.7 and 6.8. These results also
indicate that with the increase in the number of manipulating inputs the controller is able to provide
better decoupling. In this control structure the fuel flow to the furnace (F5) was eliminated from the
manipulating inputs and the combustion air blower vent valve (V7) and the regenerated catalyst
valve (svrgc) were added instead. This, however, has led to a decrease in the controller performance
compared to the 3x3 structure studied before. Although the observed interactions between the
158
controlled variables are minimal some of the process outputs settle slower to their new setpoint
value (e.g. Treg). This indicates that the F5 is an important manipulating input, which should be kept
in the MPC control scheme.
The 4x4 control scheme indicates the ability of this approach to maintain the stack gas oxygen
concentration at a predefined value, which allows more efficient FCCU operation due to better use
of air blower capacity, and leads to safer operation through control of the afterburning
phenomenon.
One of the advantages of the MPC approach is that different number on manipulating and
controlled variables can be used (non-square system) for control. This is an important property,
given the previous observations that by increasing the number of manipulating inputs the
decoupling between the controlled outputs can be improved. The simulations for the non-square
4x2 MPC scheme (4 inputs and 2 outputs) did reveal improvements compared to 2x2 or even to the
4x4 schemes, in providing better control performance. The advantage of using a control scheme
with a higher number of manipulated than controlled variables is especially important when
constraints on the manipulated variables are imposed. The surplus manipulated variables may be
advantageous when one or more of the manipulated variables become restricted.
These results demonstrated the benefits of the model based multivariate control and the advantages
of using excess of manipulated inputs. However the judicious selection of the manipulated inputs is
emphasized, as it was indicated that the F5 is an important parameter, which should be kept
amongst the set of manipulated inputs to achieve the best control performance.
Table 6.2: RGA Matrix of 3X3 Control Scheme.
Controlled
variable
F5
svsc
V14
Tr
-0
Treg
-0
-0
Wr
-0
159
Table 6.3: RGA Matrix of the 4x4 control scheme.
Controlled
variable
svrgc
V7
V14
svsc
Tr
-0
-0
-0
Treg
-0
cO2,sg
-0
Wr
-0
-0
Plant Outputs
12
10
Tr (C)
Tr (C)
8
6
4
2
0
20
Treg (C)
Treg (C)
15
10
-5
1000
2000
3000
Time 4000
(sec)
1000
2000
3000
4000
5000
6000
7000
8000
5000
6000
7000
8000
Time (s)
Figure 6.1: The 2x2 LMPC simulation results with the controlled variable response for setpoint change
for Tr and Treg at different times.
160
Plant Inputs
120
F5 (SCF/s)
100
F5
80
60
40
20
0
0.2
svsc
svsc
-0.2
-0.4
-0.6
1000
2000
1000
3000
2000
3000
Time4000
(sec)
4000
5000
6000
5000
7000
6000
7000
8000
8000
Time (s)
Figure 6.2: Manipulated variables for the 2x2 LMPC control scheme.
Plant Outputs
Tr (C)
40
Tr (C)
20
0
-20
Treg (C)
Treg (C)
15
10
5
0
10
Wr
Wr
20
0
-10
0
0.2
0.2
0.4
0.4
0.6
0.6
0.8
0.8
Time (sec)
1
1.2
1.2
1.4
1.4
1.6
1.6
1.8
1.8...........24
x 10 4
x10
Time (s)
for Tr , Treg and Wr at different times.
161
Plant Inputs
10
F5
F5 (SCF/s)
20
0
-10
0
svsc
svsc
-1
-2
2
V14
V14
1.5
1
0.5
0
0
0.2
0.4
0.2
0.4
0.6
0.6
0.8
0.8
Time (sec)
1
1.2
1.2
1.4
1.6
1.4
1.6
Time (s)
1.8
1.8........
.24
x 10
x104
Plant Outputs
Tr (C)
Tr (C)
15
10
5
Treg (C)
Treg (C)
0
20
10
0
cO2,sg
Co2,sg (%)
-10
5
-5
0 0
0.2
0.2
0.4
0.4
0.6
0.6
0.8 Time (sec)

11
0.8
1.2
1.2
1.4
1.4
1.6
1.6
1.8
1.8... ........22
44
Time (s)
x x10
10
for Tr , Treg and cO2,sg at different times.
162
Plant Inputs
F5 (SCF/s)
F5 (SCF/s)
200
100
0
svsc
svsc (%)
-100
0.5
0
-0.5
V14 (%)
V14
-1
2
0
-1
0.2
0.2
0.4
0.4
0.6
0.6
0.8
1
0.8 Time (sec)
1
1.2
1.2
1.4
1.4
1.6
1.6
Time (s)
1.8
1.8........... 2
44
xx10
10
Plant Outputs
Tr (C)
Tr (C)
10
5
Treg (C)
Treg (C)
20
0
-20
cO2,sg
CO2,sg
20
0
-20
W
Wr r
20
0
-20
0
1000
1000
2000
2000
3000
3000
4000
4000
Time5000
(sec)
5000
6000
6000
7000
7000
8000
8000
9000
10000
9000.... 10000
Time (s)
for Tr , Treg , cO2,sg and Wr at different times.
163
Plant Inputs
svrgc
svrgc
2
1
V14
V14
0.5
0
-0.5
V
V77
0.5
0
-0.5
svsc
svsc
1
0
-1
0
1000
1000
2000
2000
3000
3000
4000
4000
Time5000
(sec)
5000
6000
6000
7000
7000
8000
8000
9000
10000
9000.. .. 10000
Time (s)
Plant Outputs
12
8
Tr (C)
Tr (C)
10
6
4
2
0
15
Treg (C)
Treg(C)
10
-5
0
1000
1000
2000
2000
3000
3000
Time4000
(sec)
4000
5000
5000
6000
6000
7000
7000
8000
8000
Time (s)
Figure 6.9: The non-square 4x2 MPC simulation results setpoint change for Tr and Treg at different
times.
164
Plant Inputs
F5
F5
0.4
0.2
svrgc
svrgc
0
-2
-4
svsc
svsc
1
0.5
V14
V14
1
0
-1
0
1000
1000
2000
2000
3000
3000
Time4000
(sec)
4000
5000
5000
6000
6000
7000
7000
8000
8000
Time (s)
Figure 6.10: Manipulated variables for the non-square 4x2 control scheme.
6.2.2 Linear MPC applied to the nonlinear model-based plant simulator

In real model-based control application a mismatch always exists between the actual process and
the model. The mismatch may be structural or parametric; the former is when process and model
differ, whereas the latter is when parameter values differ between model and actual process. As the
difference increases, the response of the closed loop will deteriorate, in terms of properties
important to industrial implementation, namely overshoot and rise time. With a proper model,
signifying little mismatch, the process exhibits minimal overshoot, and short rise time, but in the
presence of mismatch and large perturbations, these cannot be maintained.
This study aimed to simulate the realistic scenario when model-plant mismatch exists, due to the
application of the linear model predictive control to a real nonlinear plant. In this study the realplant is represented by the detailed first-principles nonlinear model described in Chapter 3, whereas
the model used in the controller is a linear model obtained via linearization of the nonlinear model.
The Simulink block diagram with the implementation of this control scenario is shown in Figure
165
6.11 the 2x2 LMPC scheme with Tr and Treg as the controlled outputs and F5 and svsc as the
manipulated inputs.
The result obtained in the case step change in the setpoint is presented in Figure 6.12. The MPC
tuning was performed by choosing the output and input weights based on the maximum allowable
deviation of the corresponding variables, followed by a trial and error refining step.
A very good control performance was achieved when the setpoint change was small as indicated in
Figure 6.12. However if the magnitude of the setpoint change increases by 5 C the LMPC fails to
control this strongly nonlinear plant. This is expected since the predictions are based on the linear
model obtained in the original operating point and as the process deviates from this operating point
the model-plant mismatch increases and the control performance deteriorates. Generally these
deviations and increasing model-plant mismatch may lead to stability problems in practical
implementations. One approach to improve robust performance is to use different linearized models
in the LMPC controller obtained in different operating points, and automatically choosing the
corresponding model based on the operating region of the FCCU.
Figure 6.11: MPC block diagram for FCCU.
166
535
Tr (C)
534
533
532
Tr
531
10
Tset,r
12
14
733.6
Trgc (C)
732.6
731.6
730.6
729.6
Treg
0
10
Tset,reg
12
14
time (h)
Figure 6.12: Simulation results with the LMPC applied to the plant represented by the nonlinear
process model, in the case of setpoint changes in Tr and Treg , at t=1.5 h.
F5 (SFC/s)
80
60
40
20
10
12
14
10
12
14
svsc (%)
0.4
0.3
0.2
0.1
time (h)
Figure 6.13: Manipulated variable response for set point change in the output, for 2x2 control scheme.
167
6.3 Nonlinear model predictive control (NMPC) of the FCCU

In this section, an efficient algorithm is presented and applied for the non-linear MPC of the FCCU.
Chemical processes are typically non-linear, and as such, control system performance is generally
governed by its ability to handle process non-linearity. Linear MPC schemes where linear models
are used for prediction are in common use in various sectors, and so control issues, such as stability
have been fully addressed. In contrast, non-linear MPC schemes are relatively new, appearing in
the last twenty years, but with a growing base of applications (Agachi et al., 2006).
6.3.1 Efficient real-time iteration based NMPC approach

The optimal control problem to be solved on-line in every sampling time in the generic NMPC
algorithm (Nagy et al., 2007) can be formulated as:
Problem P1 (tk ) :
min F
u (t )
, tk
(x (t ), u(t ); )
(6.1)
subject to:
x (t )
f (x (t ), u(t ); ) ,
(6.2)
y(t )
g(x (t ), u(t ); ) ,
(6.3)
x(tk ),
x (tk )
h(x (t ), u(t ); )
where
x0 ,
x (t0 )
0,
(6.4)
[tk , tkF ] ,
(6.5)
is the performance objective, t is the time, tk is the time at sampling instance k , tkF is
the final time at the end of prediction, t0
is the nu set of input vectors, y(t )
u(t )
ny
is the n x vector of states,
is the ny vector of measured variables used to
compute the estimated states x(tk ) with initial values x0 ,

uncertain parameters, where the set
nx
0 is the initial time, x (t )
is the n vector of possible
can be either defined by hard bounds or probabilistic,
characterized by a multivariate probability density function. The function f :
nx
nx
is
the twice continuously differentiable vector function of the dynamic equations of the system,
g:
nx
ny
is the measurement equations function, and h :
nx
is the vector
of functions that describe all linear and nonlinear, time-varying or end-time algebraic constraints
for the system, where c denotes the number of these constraints.
168
The objective function in the case the typical receding horizon NMPC with the aim to minimize
deviations from the setpoints can be expressed as of can have the following general form:
tkF
(x (t ), u(t ); )
tk
(y
ref
Qy
ref
2
Qu
du
dt
)dt
Q
(6.6)
where, y ref and u ref are the output and input references, respectively, and Qy, Qu , Q u are weighting
matrixes.
Efficient optimization via direct multiple shooting
The optimal control problem P1 (tk ) is an infinite dimensional problem, which except in a few very
simple cases is impossible to solve. The main idea behind direct methods is based on transforming
the original infinite dimensional problem P1 (tk ) into a finite dimensional nonlinear programming
problem (NLP) by formulating a discrete approximation to it that can be handled by conventional
NLP solvers (Biegler, 2000). The time horizon t
(stages)
tk
[tk ,i , tk ,i 1 ]
k ,i
tk ,0
tk ,1
...
tk ,i
tk ,i
,
...
tk ,N
,Np
0,1,
with
discrete
uk (t ) is then parameterized into a piecewise representation uk ,i (t, pk ,i ) for t
control parameter vectors pk ,0 , pk ,1, , pk ,N
time
steps
tkF . The infinite dimensional continuous control input
[tk , tkF ] is divided into N p subintervals
with pk ,i
np
k ,i
with N p local
and the optimization problem is solved
with the local control parameter vectors being the decision variables. In most of the practical
application piece-wise constant or piece-wise linear control parametrization is used. In the simplest
case, using piecewise constant parameterization, we have uk (t, pk ,0 , , pk ,N
[uk , uk 1,
, uk
Np 1
Problem P1 (tk ) gives the general discrete time formulation of problem P1 (tk ) :
Problem P1 (tk ) :
uk ,uk
min
1
, ,uk
k Np
Np
k Np
(x k
Np
; )
(x j , u j ; )} ,
(6.7)
j k
subject to:
Gk (xk , uk ; )
0,
(6.8)
H k (xk , uk ; )
0,
(6.9)
169
].
where N p is the number of stages in the prediction horizon [tk , tkF ] , Gk :
nx
nx
ny
corresponds to all equality constraints resulted from the algebraic equations (6.3) of the model or
nx
from the discretized model equations, H k :
c 2nu
is the vector function of all
inequality constraints (6.5), including the constraints on the inputs considering. We consider here
that the set of possible inputs is given as hard bounds. It is assumed that the vector functions G and
H are twice continuously differentiable. Sequential Quadratic Programming (SQP) is generally
considered as the most efficient numerical method available to solve nonlinear optimization
problems (NLP). It has been shown that SQP requires the fewest function evaluations to solve
NLPs (Binder et al., 2001). SQP is a quasi Newton method that treats nonlinear optimization
problems by solving a sequence of local linear-quadratic approximations. The Lagrangian of the
optimization problem is approximated quadratically, typically by applying a numerical update
formula. Constraints are approximated linearly. SQP methods generally apply the equivalent of
Newton steps to the optimality conditions of the NLP problem achieving fast rate of convergence.
Several optimisation approaches have been proposed for the solution of the optimal control
problem (6.7)-(6.9), including sequential approaches (Hicks and Ray, 1974; de Olivieara and
Biegler, 1994), simultaneous approaches (Tsang et al., 1975; Biegler, 2000; Cuthrell and Biegler,
1989) and hybrid approaches such as direct multiple shooting (Bock and Plitt, 1984; Bock et al.,
2000a; 2000b; Diehl, 2001; Diehl et al., 2002; E. ref 9).
Direct multiple shooting is one of the most efficient optimisation approaches available for the
solution of the NMPC problems hence it was selected as the method applied in this work.The direct
multiple shooting procedure consists of dividing up the time interval [tk , tkF ] into M subintervals
[ i,
i 1
] via a series of grid points tk
tkF . Note that the grid points do
not necessary correspond to the discretization points ( N p ) in the definition of problem P1 (tk ) . Using
a local control parameterizations a shooting method is performed between successive grid points
(see Figure 6.14). The solution of the ODE on the M intervals are decoupled by introducing the
initial values
of the states at the multiple shooting nodes
as additional optimization variables.
The differential equations and cost on these intervals are integrated independently during each
optimization iteration, based on the current guess of the control, and initial conditions
. The
continuity/consistency of the final state trajectory at the end of the optimization is enforced by
adding consistency constraints to the nonlinear programming problem. The additional interior
170
boundary conditions, are incorporated into one large nonlinear programming problem (NLP) to be
solved, which is given in a simplified form below (Nagy et al., 2007):
Problem P2 :
min (v; )
subject to
G(v; )
H (v; )
0
0
(6.10)
where the optimization variable v contains all multiple shooting state variables and controls
v
[ 0 , u0 , 1, u1, ,
M 1
, uM 1,
].
(6.11)
The discretized initial value problem and continuity constraints are included in the equality
constraints:
x(tk )
x i (ti 1 ; i , ui )
g ( i , ui ; )
G (v; )
i 1
0,
(6.12)
and the inequality constraints are given by H (v; )
H ( i , ui ; )
0 , for i
idea of the direct multiple shooting algorithm is illustrated in Figure 6.14.
Figure 6.14: Illustration of the direct multiple shooting algorithm.
171
0,1, , M . The main
Real-time NMPC algorithm

The solution of problem P2 requires a certain, usually not negligible, amount of computation time
k
, while the system will evolve to a different state. In this case the optimal feedback control
u (tk )
[u0|t , u1|t ,
k
, uN
|t
p k
] computed in moment tk corresponding to the information available up to
this moment, will no longer be optimal. Computational delay
has to be taken into consideration
in real-time applications. In the approach used here, in moment tk , first the control input from the
second stage of the previous optimization problem, u1|t , (which corresponds to the first stage of
k 1
the current optimization) is injected into the process, and then the solution of the current
optimization problem is started, with fixed u0|t
the remaining period of t

tk
tk
(tk
u1|t . After completion, the optimization idles for

k 1
, tk 1 ) , and then at the beginning of the next stage, at moment
t , u1|t is introduced into the process, and the algorithm is repeated. This approach
k
requires real-time feasibility for the solution of each open-loop optimization problems (
t ).
Initial value embedding strategy

The initial values embedding strategy originally proposed by Bock et al. (2000b), can significantly
enhance computational performance. The approach is based on the fact that optimization problems
at subsequent sampling times differ only by the initial values that are imposed through the initial
value constraints. Accepting an initial violation of these constraints, the solution trajectory of the
previous optimization problem can be used as an initial guess for the current problem. Since in the
direct multiple shooting approach the decision variables include both the control input and the
initial values of the states and the discretization points, for this approach to work efficiently the
entire decision vector (control plus states) has to be initialized with the solution of the previous
optimisation problem. Furthermore, all derivatives as well as an approximation of the Hessian
matrix are already available for the solution trajectory from the previous step, and these can be used
in the new problem. In this way the first QP solution can be performed without any additional ODE
solution, leading to a computationally highly efficient approach (Diehl et al., 2002).
172
Solution tool for efficient BNMPC implementation

The solution of problem P2 can be performed using various optimization packages. The NMPC
tool, OptCon, (Nagy et al., 2004) based on the sequential-quadratic-programming (SQP) type
optimizer HQP (Franke et al.), which is used in conjunction with the implicit differential-algebraicequation (DAE) solver, DASPK, for robust and fast solution of the model equations (Li and
Petzold, 1999), has been successfully applied in various applications including the control of an
industrial pilot scale polymerization reactor (Nagy et al., 2007). OptCon includes a number of
desirable features. In particular, the NMPC is based on first-principle or grey-box models, and the
problem formulation is performed within Matlab (The Mathworks Inc.), which is the most widely
used modeling and control environment for control engineers. The model used in the controller has
to be developed in the form of Simulink (The Mathworks Inc.) mex S-function using C
language. The S-function interface of the optimization tool provides convenient and fast
connectivity with Matlab. The NMPC implementation offers an efficient approach based on the
direct multiple shooting algorithm, that exploits the special structure of optimization problem that
arise in NMPC. The software uses low rank updates of the approximation of the Lagrangian
Hessian of the nonlinear subproblems combined with a sparse interior point algorithm for the
efficient treatment of the linear-quadratic subproblems in the nonlinear SQP iterations. Bounds and
inequality constraints are handled using a barrier method and line search is used for global
convergence of the SQP iterations. OptCon implements the real-time approach, and also provides
an OPC connectivity, making it appropriate for rapid real-time prototyping of NMPC algorithms in
industrial environment.
6.3.2 NMPC FCCU simulation results

The complex dynamic model of the FCCU was used in the NMPC algorithm. The 22 control
structure is presented as one of the simulations performed. The two controlled variables are: reactor
temperature Tr and regenerator temperature Treg. The two manipulated inputs are: fuel flow F5 and
slide valve positions svsc on the spent catalyst circulation line. The coking rate factor f change
(5% step increase at moment t = 5000 s) has been selected as a typical disturbance. The control of
presented variables is important for the efficient and safe operation of the unit, and has direct
impact on the products yield. The reactor temperature has to be maintained at a certain level to
provide a desired maximum conversion of the feed oil. Proper reactor and regenerator temperature
173
control also means good management of thermal energy. A set of tested scenarios is presented in
Table 6.3.
Table 6.3: Tested scenarios for NMPC simulations.
Case
Disturbance
NMPC tuning
parameters (duration,
prediction and control
horizon and weight)
Step increase
Step decrease
Tr551, at 2000 s
Tr547, at 4000 s
[8000 s ,(Ts = 100, p = 5,
Treg729, at 3000 s
Treg727, at 4000 s
m = 5), Qu = 10]
+ 5 %, at 5000
Tr546.6, at 2000 s
[8000 s , (Ts = 100, p =
Treg741, at 3000 s
N/A
Setpoint change
10, m = 5), Qu = 10]
Two different scenarios are considered:

(i)
NMPC without anticipation in this case the setpoint change is treated as a

disturbance which happens unexpectedly. The NMPC will react only after the setpoint
change has been applied to the process.
(ii)
NMPC with anticipation in these simulations it is assumed that the moment when
the setpoint change will occur is known beforehand, and is programmed in the NMPC
controller. In this case the NMPC can take into account during the prediction the future
changes in the setpoints and hence can react before the actual change happens, to
achieve optimal performance.
Figure 6.14 illustrates the performance of the NMPC with both anticipation and no anticipation
scenarios. The first setpoint change in Tr and Treg, respectively, are considered known a priori and
were programmed in the NMPC from the beginning of the simulation. Therefore the controller can
react anticipating the change in the future resulting in very good setpoint tracking performance. The
second simultaneous setpoint change at 4000 s, are considered as disturbance, in the sense that they
are not programmed in the NMPC. This could correspond to a realistic scenario when sudden
change in the operating conditions have to be applied immediately e.g. due to safety considerations.
It can be seen that in this case the NMPC reacts with delay, since the change of the setpoint is only
174
detected at the next sampling time after the change occurred. Due to the real time iteration scheme,
according to which the control action is implemented with one sampling period delay, the NMPC
practically can only react to the unanticipated setpoint change after two sampling periods.
The CPU times corresponding to the NMPC calculations in each iterations and the number of SQP
iterations are presented in Figure 6.16. It can be observed that the maximum number of iterations
(set to 5 SQP iterations in these studies) is achieved when the setpoint changes occur, however the
real-time iteration scheme and initial value embedding strategy allow the convergence of the
optimisation problem over several iterations without considerable decrease in control performance.
The maximum total computational time (approximately 65 s) is well below the sampling time of
100 s. The results demonstrate that the efficient optimisation scheme and tool used here can
guarantee the real-time feasibility of the NMPC implementation, with very good control
performance even for a model with a large number of ODEs.
Additionally the benefits of designing control systems in which a sepoint change can be
programmed in with anticipation are highlighted. Since the control performance can be
significantly improved if the predictive controller can take a future setpoint change into account, it
is important to design the control systems to allow specifying the setpoint change and the time
when that must occur, and operators need to be aware about the benefits of programming the future
change into the controller in advance.
Figure 6.17 illustrates the results for case 2, considering NMPC with anticipation for setpoint
change and in the case of a measured disturbance scenario. The setpoint change in the Tr is very
well followed by the NMPC, however the control performance degrades when the change in Treg
occurs. Careful examination of the control variables show that one of the manipulated variables
(svsc) saturates in this case, which causes the decreased tracking performance. At t= 5000 s a 5%
increase in f occurs. It is assumed that the disturbance is measured and the NMPC is able to reject
the effect of the disturbance. The controller reacts to the disturbance as can be observed from the
variation in the manipulated inputs, but no deviations in the controlled output is observed,
indicating the excellent disturbance rejection ability of the NMPC system.
The NMPC provides stable closed-loop response in both simulate scenarios with very good setpoint
tracking and disturbance rejection performances. The NMPC is able to provide a very good
decoupled response with minimal disturbance in the controlled outputs when the setpoint change
occurs for one of the outputs only.
175
Tr (C)
556
Without anticipation
546
536
With anticipation
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
time (s)
5000
6000
7000
8000
Treg (C)
732
727
722
F5
100
50
0
svsc
1
0.5
0
CPU time per optimisation (s)
Figure 6.15: Performance of the NMPC in the case of setpoint change with and without anticipation.
100
80
60
40
20
0
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
time (s)
5000
6000
7000
8000
SQP No of iterations
4.5
4
3.5
3
2.5
2
Figure 6.16: CPU times required for the optimisation problems of the NMPC
176
Tr (C)
546.6
536.6
526.6
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
time (s)
5000
6000
7000
8000
Treg (C)
742
732
722
F5
100
50
0
svsc
1
0.5
0
Figure 6.17: Performance of the NMPC in the case of setpoint change with anticipation and a
measured disturbance (5 % step increase at t = 5000 s inf).
6.4 Conclusions
The FCC unit is the most complex and challenging unit in a modern refinery. The success of any
advanced control application (linear or nonlinear) on this unit depends on the ability to deal with
the operating constraints, and to address the economic objectives.
A comprehensive analysis of several multivariate linear MPC (LMPC) schemes is presented. The
results demonstrated the importance of including the fuel flow to the furnace among the
manipulated input in a multivariate control scheme. The potential benefits of using non-square
control structures are also demonstrated. The robust performance of the LMPC is evaluated in the
case of model-plant mismatch, when the plant is represented by the complex nonlinear model,
whilst using linear model for prediction in the MPC.
177
To further improve the control performance an efficient nonlinear model predictive control
approach (NMPC) is implemented to the FCCU. The approach used is based on a real-time
iterations scheme with initial value embedding, and is based on a highly efficient multiple shooting
optimisation approach to achieve real-time feasibility. The efficient NMPC approach is evaluated
for setpoint tracking and disturbance rejection. For setpoint tracking two different scenarios, with
and without anticipation are considered. The results demonstrate that superior performance of the
NMC scheme and highlight the importance of the implement setpoint changes into the controller in
advance whenever this is possible to achieve better control performance.
Both NMPC and LMPC were superior, and it was also shown that NMPC provided better
performance compared to LMPC. The performance advantage could be noted in terms of
overshoot, settling time, and maintaining the controlled variables within the limiting constraints of
the plant. Furthermore, the potential for practical implementation of the NMPC approach in the
industrial context of complex chemical processes has been demonstrated. These results have
demonstrated that the control structures proposed may be applied in industry in the form of a new
scheme for controlling highly complex chemical processes.
178
Chapter 7: Economic Optimization of FCCU
CHAPTER 7
ECONOMIC OPTIMIZATION OF FCCU
7.1 Dynamic Real - time Optimization (DRTO)
In the process industry, a main challenge is to ensure optimal plant operation regardless of
constantly varying materials and conditions, whether composition of feedstocks, rates of
production, prices of material inputs and outputs, or energy availability. In this complicated
context, the concept of Real-time Optimisation (RTO) was introduced in an attempt to ensure that a
process is optimised for given economic constraints. The process is optimised through RTO, from
the perspective of target profit or cost, and so profit-seeking enterprises can make sure that their
plant is achieving optimal profitability, even as operating parameters and materials are subject to
change. In typical form, RTO assumes that transient perturbations may be disregarded over a
sufficiently large interval, as the process achieves the steady state. A typical RTO system consists
of parts for detecting the steady state, acquiring and validating data, updating the process model,
solving the optimisation problem, and applying operating policies (Adetola, 2008).
Given the surge in interest in Dynamic Real-Time Optimization (DRTO) or Non-linear Model
Predictive Control (NMPC) for economic process optimisation (Backx et al., 2000; Helbig et al.,
2000; Engell, 2007; Rawlings and Amrit, 2008; Zavala, 2008), efficient means to solve such
optimisation problems have witnessed significant progress. Economic optimisation control schemes
have advanced beyond the usual implementation of NMPC based on quadratic cost criteria that are
minimised in terms of the deviation from a steady state set value. In economic NMPC all the
dynamic degrees of freedom are considered in ensuring maximum economic profit over a specific
time interval. A further point is that maximum profit is achieved at high frequency sampling rates,
compared to the low sampling rates characteristic of conventional steady-state optimisation, and
only during the process steady state. Therefore, DRTO effectively manages perturbations, and
ensures that these are exploited optimally for maximum profit, compared to simple compensation
for deviation from a steady state set value. However, sufficient research on DRTO and NMPC from
an economic perspective has not been performed.
179
The strategy of DRTO based on an economic perspective was introduced (Helbig and Marquardt,
2007; Kadam and Marquardt, 2007), and is shown in Figure 7.1. Compared to the conventional
approach, DRTO does not distinguish between dynamic and steady states, but engages in directly
optimising the process according to economic criteria over the prediction horizon. Such a strategy
involves calculating a solution to the NMPC problem utilising an economic objective function
derived from first-principle dynamic models. In this context, process control actions are generated
based on an economic objective; this particular control structure was first proposed by Morari et al.
(1980).
Scheduling
scheme
Data Reconciliation/
State & Parameter
Estimation
State &
Parameter
D- RTO
Economically Oriented NMPC
Max: Predicted Profit in Future
s.t. Dynamic First Principle
Model
Measurement
Figure 7.1: Illustration of D-RTO or Economically-Oriented NMPC
Over recent years, a number of workers have further refined DRTO or NMPC oriented towards
profit maximisation. Such a control scheme was applied to a fluidized bed catalytic cracker (Zanin
et al., 2000), while an MPC scheme integrating RTO was also proposed (Adetola and Guay, 2010).
180
Further progress took the form of a controller based on a receding horizon applied to constrained
nonlinear process plant was also proposed (Angeli and Rawlings, 2010). For economicallyoptimised NMPC, an infinite-horizon formulation was proposed (Wurth et al., 2009). In an
approach utilising a search technique for the general extremum, closed loop stability was studied
with regard to optimisation from the economic perspective (Krstic and Wang, 2000). With the
assumption of significant duality, an NMPC with an economic orientation was analysed in terms of
nominal stability property (Diehl et al., 2010). Moreover, it was reported that multiple solutions
result from formulations utilising linear economic objective functions, where a number of degrees
of freedom remain to be optimised, and may be used to enhance operability through computing a
second optimisation problem (Huesman et al., 2008). On the other hand, for efficient process
economic optimisation, it has been reported that, at times, rapid achievement of the steady state
may not be beneficial. Moreover, a turnpike-like trend results from an economic objective
formulation (Rawlings and Amrit, 2008). While it seeks to follow a constant path, in the final part
of the horizon, the trendline departs from the constant. Such a trend has also been previously
reported in the economics literature, as being a feature of economic optimisation problems (Carlson
et al., 1991).
A number of workers have addressed the problem of FCC process optimisation, where Rhemann et
al. (1989) demonstrated an advanced control approach with systematic optimisation. In another
study, a fitting model was used to derive five independent variables (preheat and reactor
temperatures, and rate of feedstock, and HCO supply, as well as slurry recycling), where the
generalised reduced gradient (GRG) technique was used to solve the economic objective function.
Subsequently, optimised set values were input to the control system. For a residual catalytic
cracking (RCC) unit, model-predictive control, with on-line optimisation, was applied (Cutler and
Perry, 1983). This comprised a 40 equation, 20 independent variables, black box system for which
a regression was performed using a large dataset covering different plants. The successive quadratic
programming (SQP) technique was utilised to both formulate and solve the objective function
containing constraints on variables; as a result, optimised set values for the process modelpredictive controller were generated. The regression equation aspect of the optimisation scheme,
means that not all the process behaviour is captured, which necessitates frequent updates of the
model. An empirical model for predicting cracking products yield was combined with energy and
mass balance, at macroscopic level, to form a Model IV FCC unit optimisation model (Ellis et al.,
1998). Subsequently, a performance comparison was made between on-line and off-line
optimisation, in terms of varying product pricing structure, and feedstock composition. However,
181
the work is not of direct industrial value, since the Model IV FCC unit lacks riser or slide valves,
and is an outmoded plant type, which has been superseded by the upgraded FCC unit.
7.1.1 Integrated Real-Time Optimization and Control

The traditional approach to optimisation and control involves layers, in which the upper supervisory
or optimisation layer is separate from the lower control layer; in this scheme, optimised control
values are generated in the real-time optimisation (RTO) layer, and then passed down to the control
layer for implementation. A steady state model is the basis for the RTO, and as such executed once
the process achieves the steady state. In its separate layer, process control values are generated
utilising a different model and frequencies. In this context, these two separate layers do not process
the same data, which creates conflict and so predicted operating values may not be optimal;
integrating both layers are still subject of research, and workers have attempted to tackle this
systematically. The theory for such an integrated approach, in regard of performance and stability
analysis, is well established. At present, two major classes have been proposed based on either
model type or RTO execution frequency (Adetola, 2008).
1. Single level strategy: This approach involves the simultaneous solution of both the control and
economic optimisation problems; this is achieved by including the economic objective term in the
standard MPC objective function, and so a single algorithm MPC solves both problems (de
Gouvea and Odloak, 1998; Zanin et al., 2002). This non-linear objective function contains both
constraints that are steady state and dynamic, which must be resolved. In order to achieve the
desired performance and stability, tuning of the relevant weighting factors is needed to properly
implement this extended controller. The weighting factor values are determined by a significant
amount of simulation (Zanin et al., 2002), which differs from a linear MPC scheme for which
tuning is subject to specific guidelines. The single layer strategy allows rapid responses to changes
in process variables, more so than in a two layer implementation. The disadvantage is that this
strategy is only suitable for processes with short settling time, to which a linear model can be
applied.
2. Two level strategy: This represents the classical hierarchical control structure, which aims to take
into consideration the process dynamic and non-linear behaviour. Typically, this approach is based
on the idea of two levels; on the upper level, the economic optimisation as a dynamic process is
addressed, while on the lower level, the non-linear MPC computations are performed (Kadam et al.,
2002; Kadam et al., 2003; Zhu et al., 2004). Given the existence of two distinct layers, their
182
solution cycles need not follow the same sample intervals; hence, the RTO may be performed only
when a disturbance or change occurs in the process conditions. As such, the RTO may be triggered
by a disturbance sensitivity analysis, which detects persistent disturbances having high sensitivities
(Kadam et al., 2002; Kadam et al., 2003). The RTO algorithm is based on dynamic models, in
which appropriate set points are recalculated whenever economic benefits can be secured. A key
factor in the success of this approach is the interaction of these two layers. The potential drawbacks
to this approach revolve around the varying dynamic models that are applied.
The decomposition of the real-time economic optimisation of the process into a single layer or two
has been proposed by many workers (Mohsen et al., 2011; Glauce et al., 2010; Wurth et al., 2009;
Ioannis, 2005; Haruo et al., 2004; Zanin et al., 2002; Zanin et al., 2000; Loeblein and Perkins,
1999; Miriam et al., 1998; Robert et al., 1998).
7.2 Process optimizer for Economic NMPC of the FCCU

The FCC unit economic optimisation has a simple objective, to achieve the maximum profit from
the process (Han et al., 2004). Therefore, the objective function seeks to maximise the difference
between material and operating costs, and the market value of the products yielded by the process;
this is given by:
N ps
Nfs
max f x, u, p F
w
F
w
C
x, u, p
pMF pMF
kFJ kFJ
opr
u
p 1
k 1
where
represents the product streams (LPG, gasoline, HCO and LCO);
feedstock streams (gas oil and slurry recycles); and
(7.1)
represents the
represents the operating cost (cost of
utility consumption and generation, and others).

A different objective function, which maximised the difference in feedstock and products value,
was proposed (Umesh and Armando, 1998). In a simplification of the function, utility costs,
comprising fuel, power consumption, etc, were considered negligible. However, the disadvantage in
neglecting these terms is that a false premise will result, where using more power to produce a
greater amount of products will be considered an optimum, while this is not actually the case, as
there is a rise in cost associated, for example, with driving compressors at the maximum.
Furthermore, there is an implicit assumption that the market has unlimited capacity to absorb
whatever quantity of product is produced. Another assumption is that any unused feedstock is fully
183
recovered, and maintains its value. With the exception of wash oil and diesel feedstocks, the
resulting cost function is given by:
C F3 Fugo v3 Fgl vgl Fgs vgs
(7.2)
where F3 flow of gas-oil feed, gal/s, Fgl flow of gasoline, gal/s, Fgs flow of lighter gas, gal/s, Fugo
flow of unused (uncracked) gas-oil, gal/s. v3 value of gas-oil feed, $/gal, vgl value of gasoline, $/gal,
vgs value of lighter gas, $/gal.
Costs with respect to the manipulated variables and utilities can also be included for a more detailed
representation of the actual costs.
7.2.1 One layer economic NMPC of the FCCU

In a hierarchical dynamic real-time optimisation and control scheme, the upper level is typically
that of the optimisation algorithm, where the FCC optimum economic operating values are
determined. The solution of the optimisation problem posed is achieved using the SQP algorithm,
which incorporates an economic objective function, the process constraints set, and four decision
variables addressed by the optimisation engine. The decision variables selected were the fuel flow
rate, spent catalyst valve position, and the temperatures in the riser and regenerator. The existing
typical constraints for an FCCU were described comprehensively by McFarlane et al. (1993). In
this work, the manipulated variables and constraints selected for optimisation are presented in Table
7.1. Values were assigned to the different parameters, where the combustion air blower (operating
range 60 70 kg/s) was at maximum as indicated by a 100% open suction valve V6; valve positions
for the spent catalyst and regenerator were 0.3% open at the beginning; in operation, the wet gas
compressor suction valve V11 was 95% open. The chosen combustion constraints followed the
proposal by McFarlane et al. (1998), and comprised the choice of regenerator temperature and
oxygen content of stack gas, or stack gas CO concentration of 350 ppm maximum. In the initial
simulations, the stack gas CO concentration was used as a constraint in the DRTO. The variable
process parameters were given by explicit expressions, following which the objective function was
computed.
184
Table 7.1: Constrained and manipulated variables for application to FCC unit.
Constrained and manipulated variables
Min
Max
Fuel flow rate, SCF/s
10
F5
70
Spent catalyst valve position, %
0.1
svsc
0.8
cO2,sg
350
685
Treg
760
Oxygen concentration in stack gas valve, ppm

Regenerator temperature, C
In this work, simplified economic objective function was used, mainly focusing on optimising the
product distribution, rather than taking production rates and cost of raw material or operational
costs into account. The aim of this analysis was to show the advantage of adopting an economic
NMPC strategy in order to adapt the operating conditions of the FCCU to obtain products that were
adapted according to the varying trends in demand due to environmental, seasonal, or other
changes. This objective function to be minimized is given by:
+
(7.3)
(7.4)
(7.5)
where w1,2,...
cost related weighting coefficient. Three different scenarios were considered,
corresponding to the scenarios in Table 7.2, represented by,

(7.6)
(7.7)
(7.8)
The objective function presented here is typical of what is found in practice, in that a system of
multiple variables requires real-time optimisation to achieve particular objectives that may vary,
during the production, while being constrained in specific aspects. Two manipulated inputs were
used in the control schemes evaluated, comprising of the fuel flow rate, u1=F5 (SCF/s), and spent
catalyst valve position, u2=svsc (%). The regenerator and reactor temperature are analysed during
185
the economic control scheme, since these represent the main controlled process outputs for the
other control approaches, and their variations has a significant impact on the overall performance of
the plant. Note however, that in the case of economic NMPC these outputs are not directly
regulated.
A set of four cases were formulated for the optimisation of a FCC unit with a daily production of
46,000 barrels with the objective of profit maximisation. The optimisation cases were used to
compare the performance of control algorithms for different market price scenarios (Table 7.2). The
benchmark chosen for comparison was the first case, corresponding to a classical NMPC
implementation which uses the advanced control algorithm to track predetermined but constant
operating setpoints for the reactor and regenerator temperatures, similarly as described in Chapter
6. Although these values would in general be determined to achieve acceptable profit and
performance corresponding to a particular raw material and operating conditions, this regulatory
scheme is unable to adapt the operating conditions to achieve continuous profit optimisation.
Table 7.2: Scenarios of the operation studies
Case
Algorithms
Objective
NMPC
Benchmark study
+ 5%
Maximize LCO
+ 5%
Maximize LPG
+ 5%
Minimize LCO
+ 5%
2.1
DRTO/NMPC
2.2
NMPC
3.1
DRTO/NMPC
3.2
4
Disturbance
DRTO/NMPC with environmental constraint

DRTO/NMPC
The NMPC controller adopts the same tuning parameters and constraints as the DRTO/NMPC
algorithm except for the weight of the economic components. The initial steady state of the FCC
was given by: uss = [32 0.3] and for the reactor and regenerator temperatures, yss = [531.6 729.8]
were used, which were also the setpoint values for the first, benchmark regulatory NMPC scenario.
The prediction horizons used in the algorithm were: m = 3 and p = 3. The constraints on the
manipulated inputs considered were: umax = [70 0.8], umin = [10 0.1]. In addition, the sampling
period is equal to 100 s. The disturbance comprised 5% increase in the coking formation factor for
all cases.
186
Case 1: Benchmark study using NMPC for regulatory control

This case presents the benchmark study when NMPC is used to regulate the reactor and regenerator
temperature at their fixed setpoint values, and the control objective is to minimize any deviation
from these setpoint values in the case of any process disturbance. The simulations performed are
illustrated below in Figures 7.2 - 7.4. In Figure 7.2, the effect of the disturbance on the systems
behaviour can be clearly seen. It is also clear that the controller can restore the temperatures by
manipulating the inputs. The effects of corrective action can be seen as a decrease in LCO yield
from 0.19 % to 0.17 % (Figure 7.3 (a)) and at the same time increasing LPG and gasoline
percentage (Figure 7.3 (b and c)). Although this might be favoured, depending on the product
requirement, but an undesirable significant increase in CO and decrease in O2 amounts does not
make this operating scenario feasible (Figure 7.4.(a and b)), due to
potential violation of
environmental constraints. Although this simulation scenario demonstrated the remarkable

performance of the NMPC approach to keep the temperatures within very small variation around
the setpoint (below 0.2 C compared to 5 C in the open loop simulations shown in Chapter 4), the
results emphasize several key observations:
(i) Good regulatory performance (maintaining the setpoint at its desired values) does not
guarantee optimal operation of the plant in all disturbance scenarios. As it can be seen, in
these results, despite the tight control of the process temperatures significant variations in
the product distribution are caused by the disturbance related to the change in the quality of
raw material.
(ii) Depending on the actual economic objective under which the FCCU operates at the
moment when a disturbance occurs, from the economic operation point of view these
disturbance can be differentiated as beneficial (good) and harmful (bad) disturbances,
despite the fact that they both may cause deviation of the control system from its original
setpoints. Note however that the same distrurbance may be considered good or bad
depending of the economic objective under which the plant operates at the moment when
the disturbance occurs. For example, if the FCCU operates under the economic objective
expressed simply as to maximize the LPG concentration, a 5% increase in the coking
factor, can be considered as a beneficial or good disturbance as it causes an increase in
the LPG concentration (also shown in the case of the uncontrolled open loop simulations in
Chapter 4). In this case, although the disturbance creates a deviation of the process
temperatures from their original setpoints using control effort and utility costs to regulate
the process at the original setpoint may actually reduce the economic objective than doing
187
no control at all. Certainly the process must be controlled under all conditions, but the
control scheme should offer the possibility to adapt the operating conditions and potential
exploit the effect of beneficial disturbance to achieve economically optimal operation,
while maintain an operational, safety and environmental constraints.
(iii) Additional constraints e.g. related to environmental considerations should also be taken into
account to achieve a globally optimal performance. Operating conditions which may
maximize a particular economic objective related to the product composition but violate
e.g. the furnace capacity to burn all resulting CO and hence violates environmental
constraints are not acceptable.
In the following several economic NMPC scenarios are presented taking the aforementioned
observations into account.
Tr (C)
531.9
531.6
531.3
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
729.9
729.8
729.7
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
40
30
20
10
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
1
0.5
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.2: ControlledandmanipulatedvariablesresponsesforNMPCinpresenceof f disturbance

(5 % step increase at t = 8000 s).
188
0.195
a
LCO (wt %)
0.19
0.185
0.18
0.175
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.2
LPG (wt %)
0.195
b
0.19
0.185
0.18
0.175
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.445
Gasoline (wt %)
0.44
0.435
0.43
0.425
0.42
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.3: Model outputs dynamic responses for NMPC (a) LCO, (b) LPG, and (c) Gasoline.
189
x 10
cCOsg (ppm)
2.5
2
1.5
1
0.5
0
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
3.5
cO2,sg (mol %)
2.5
2
1.5
1
0.5
0
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.4: Model outputs dynamic responses for NMPC (a) cCOsg. and (b) cO2,sg.
Case 2: Economic NMPC (ENMPC) for maximization of LCO yield

This case presents a dynamic real time optimization strategy (DRTO), using the economic NMPC
(ENMPC) concept presented before implemented to the simulated FCCU. First the ENMPC
simulation is performed (case 2.1 in Table 7.2), and the results are shown in Figures 7.5 - 7.7, while
the second (case 2.2 in Table 7.2), uses the typical regulatory NMPC bring the system to and keep
at the optimal temperature profiles obtained by the ENMPC, but without taking into account the
potential beneficial effect of a subsequent disturbance in the coking factor, as illustrated in Figures
7.8 7.10. Optimal conditions were obtained using the DRTO/ENMPC algorithm. The conditions
at the optimal state were then used as the initial conditions for studying the effect of disturbances
by using the NMPC technique for comparison with the DRTO/ENMPC.
190
In this case, the economic objective function is to maximise LCO concentration based on seasonal
demand, especially in winter, where it is more valuable than other products. The tuning parameters
adopted in the (DRTO/NMPC) algorithm are as follows: m = 3 and p = 3; the cost related weighting
coefficient is
; and the output constraints are defined as follows: ymax = [555
760] and ymin = [525 685].
Tr (C)
554
532
520
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
753
730
717
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
100
50
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
1
0.5
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.5: Controlled and manipulated variables responses for DRTO/ENMPC of maximizing LCO in
presenceoff disturbance (5 % step increase at t = 8000 s).
In Figure 7.5, the reactor temperature can be seen to be adjusted by varying the parameters of heat
balance in the system manipulated by the fuel flow rate F5 controller and the svsc valve. The
controller finds the new operating conditions which require increased operating temperature both in
the reactor and regenerator, required to maximize the LCO concentration. The regenerator bed
temperature need not be kept at the setpoint, but must be maintained within limits. Moreover,
catalyst circulation decreases with increasing bed temperature since less catalyst is needed to
191
maintain the riser outlet temperature on target. A higher bed temperature, therefore, decreases the
catalyst/oil ratio, which usually has undesirable effects on the product yields.
On the other hand, a low dense bed temperature is also undesirable, because it can force the
regenerated catalyst slide valve to open fully, and can thereby limit the processing capacity of the
unit. The F5 saturates during the control, indicating that the maximum fuel flow is required to raise
the temperature to the economic optimum. Note that in these simulations the costs of utilities such
as associated to the increase in F5 are neglected. After the process is brought to the new operating
conditions that correspond to the particular fresh feed quality a disturbance in the raw material
quality is simulated as a 5% increase in the coking factor. The ENMPC automatically adapted the
operating conditions (e.g. increase mainly the regenerator temperature) to maintain economically
optimal operations. This illustrates the working of the ENMPC, since when the coking factor of the
raw material increases more coke is formed and hence higher temperature is needed to regenerated
the catalyst. This new condition is automatically detected and found by the proposed ENMPC
scheme.
The simulation results can be divided into two groups i.e. before and after disturbance at 8000s. For
instance, in Figure 7.6 (a), the first part shows a sharp increase in LCO yield with a peak value of
(0.39 %) at 2000s followed by a small decrease. This corresponds to the optimal LCO
concentration corresponding to the raw material quality considered initially. In the second part after
the disturbance developed, the LCO slightly increases and stabilizes at 0.365. The DRTO/ENMPC
technique shows better performance for maximizing the LCO yield (0.365), as shown in Figure 7.6
(a), which is a significant increase compared with the previous LCO yield (0.17 %) obtained in the
bachmark simulation when the process was regulated at the original steady state temperature
setpoints (Figure 7.3 (a)). Figure 7.6 (b) also shows an increase in LPG yield before the
disturbance, yet the yield was reduced slightly after the disturbance, but still remained greater than
case 1 (0.196).
192
0.4
LCO (wt %)
0.35
a
0.3
0.25
Part one (optimal 1)
Part two (optimal 2)
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.35
LPG (wt %)
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.6: (a) LCO yield when the economic objective is to maximize LCO and (b) LPG yield.
Figure 7.7 (a) shows a significant decrease in gasoline yield in both parts, where the initial value
was (0.42). This indicates that the operating conditions obtained from the optimization are not
suitable for gasoline production, however this is expected as the objective focused only on
maximizing the LCO content.
Figure 7.7 (b and c) illustrates an increase in CO emission (500 ppm), which exceeded the allowed
emission limits (350 ppm). The controller was able to bring the CO to a lower level, however in a
practical scenario temporary violation of the environmental constraint would have occurred
resulting in the emission of a significant amount of CO gases into the environment. This
emphasizes the importance of considering environmental constraints in the economic optimisation.
193
Gasoline (wt %)
0.5
0.4
0.3
0.2
0.1
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
700
cCOsg (ppm)
600
500
400
300
200
100
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
cO2,sg (mol %)
4
3
2
1
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.7: Model outputs dynamic responses for DRTO/NMPC (a) gasoline, (b) cCOsg and (c) cO2,sg .
194
The O2 level rapidly decreases as the controller pushes more catalyst into the regenerator to
maintain the regenerator temperature at the desired level. The oxygen concentration falls slightly
when the disturbance is introduced, but returns to the pre-disturbance level after the controllers
corrective action. This is the most important parameter in the regenerator and can vastly affect the
operation of the unit. The main purpose of the regenerator is to burn off all the coke that has been
formed during the course of the cracking reactions. The heat that is produced during the burning of
coke in the regenerator is used to maintain the required temperature in the reactor for normal
operation of the unit. It is also essential to ensure that all the coke is burnt off, or else the activity of
the catalyst is reduced, and maintaining the reactor temperatures becomes difficult. The air rate to
the regenerator is set to maintain a slight amount of excess O2 in the regenerator flue gases. About
1-3% excess oxygen is maintained so as to ensure that all the coke in the spent catalyst is burnt off
in the regenerator.
For comparison the results with regulatory NMPC (case 2.2) are shown in Figures 7.8-7.10. The
steady state optimal controlled variables obtained from the DRTO/ENMPC algorithm were used as
the setpoint temperature values for the regulatory NMPC. Figure 7.8 demonstrates the behaviour of
the controlled and manipulated variables when they attempt to track the new setpoint values. The
reactor temperature appears to follow the setpoint even in the presence of the disturbance. The
regenerator temperature in the first part, before the disturbance, was brought to the new optimal
steady state by the controller, but as the disturbance proceeded, it increased slightly, because more
carbon was produced in the bed. Figure 7.9 (a) shows an increase in the LCO (0.33), and comparing
it with the previous LCO yield (0.37) produced from DRTO/ENMPC proved that the proposed
ENMPC approach gave an improvement in the LCO yield of 4 %, which is of significant value
from the economic point of view, especialy considering the high-volume production of the FCCU.
The results illustated that in this distrubance and objective scenario it is more economic to find a
new opearting condition than to regulate the process at the previously deermined optimal setpoint
values. The ENMPC approach was able to exploit the potentia beneficial effect of the distrubance
and increased the LCO concentraion even further compared to the optimal value with the original
raw material, whereas forcing the FCCU to operate at or close to the previously determined optimal
temperature leads to an actual decrease in the the LCO concentration. Furthermore the ENMPC
approach kept the CO emissions at a much lower level after the disturbance than the NMPC. Figure
7.9 (b) shows better LPG yield (0.2773) than that produced with DRTO/ENMPC, however this was
not considered in the optimisation.
195
Tr (C)
554
532
520
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
753
730
717
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
100
50
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
1
0.5
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.8: Controlled and manipulated variables responses for the regulatory NMPC using the
setpointsdeterminedbytheENMPCforthenominalrawmaterial,inpresenceof f disturbance (5 %
step increase at t = 8000 s).
Figure 7.9 (c) also shows that the gasoline yield decreased to 0.2076, from its initial value 0.4275
and the DRTO/ENMPC value 0.1874. Figure 7.10 (a and b) shows an increase in CO emission (675
ppm) with decrease in oxygen concentration in the first part. In the second part, the O2 stabilises at
(1.347 %), while the CO decreases and stabilizes at (104 ppm), which is still significantly higher
than in the case of the ENMPC. In the next case, discussed below, a constraint on CO emissions
will be applied, to investigate the ability of the ENMPC to consider environmental constraints in
the control problem.
196
0.35
LCO (wt %)
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.35
b
LPG (wt %)
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.5
Gasoline (wt %)
c
0.4
0.3
0.2
0.1
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.9: Model outputs dynamic responses for NMPC (a) LCO, (b) LPG and (c) gasoline.
197
800
cCOsg (ppm)
600
400
200
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
3.5
cO2,sg (mol %)
2.5
2
1.5
1
0.5
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.10: Model outputs dynamic responses for NMPC (a) cCOsg and (b) cO2,sg.
Case 3: Maximizing LPG yield using ENMPC with environmental constraints

DRTO/ENMPC of the FCCU without environmental constraint
Nowadays, the LPG product is one of most valuable products on world energy markets. In this
case, the economic objective function aims to maximise LPG yield by using the proposed
DRTO/ENMPC in the first scenario (case 3.1), and applying the environmental constraint on the
CO emissions in the second scenario (case 3.2).
In the simulations carried out in both cases, the tuning parameters adopted in the (DRTO/ENMPC)
algorithm were as follows: m = 3 and p = 3, and the cost related weighting coefficient is
. The constraints on the manipulated inputs considered here are: umax = [70 0.8],
umin = [10 0.1]. The output constraints are defined as follows: ymax = [560 760] and ymin = [525 685].
198
Figure 7.11 shows how the controlled and manipulated variables react to find the optimal operating
condition to achieve maximum LPG yield.
Tr (C)
554
532
520
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
753
730
717
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
100
50
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
1
0.5
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.11: Controlled and manipulated variables responses for DRTO/ENMPC of maximizing LPG
inpresenceoff disturbance (5 % step increase at t = 8000 s).
In this case both manipulated inputs quickly reach their maximum values, indicating that the best
operating condition to increase LPG production requires operating the unit close to the operating
constraints. Figure 7.12 (a) indicates the significant increase in the LPG yield during the initial part
of the simulation when the ENMPC is activated.
199
0.35
a
LPG (wt %)
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.4
LCO (wt %)
0.35
b
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Gasoline (wt %)
0.5
0.4
0.3
0.2
0.1
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.12: (a) LPG yield when the economic objective is to maximize LPG, (b) LCO and (c) gasoline.
A quick increase in LPG yield to the value of 0.29 at 2000 s followed by a small decrease can be
observed, which corresponds to an approximately 14% increase in the LPG yield, that can
correspond to considerable economic benefits. After the 5% disturbance in the coking factor occurs
200
a further increase can be observed and the LPG yield stabilises at 0.285. The limited variation after
the disturbance is due to the fact that the manipulated inputs were saturated hence limiting the
ability of the ENMPC to react to the disturbance. Nevertheless the ENMPC approach was able to
recognize the beneficial effect of the disturbance and allowed both the reactor and regenerator
temperatures to increase resulting in an additional increase in LPG yield.The first part in Figure
7.12 (b) shows a sharp increase in LCO yield, followed by a small decrease. In the second part,
after the disturbance, the LCO yield slightly increases and stabilizes at 0.375 %. The
DRTO/ENMPC technique, in this case, shows better performance not only for maximizing LPG,
but also the LCO yield has increased more than the previous case 2 (0.365) as shown in Figure
7.6(a). Figure 7.12 (c) shows a significant decrease in gasoline yield in both parts of the simulation,
before and after the disturbance, from the initial value 0.42 to 0.148, however this parameter was
not considered in the optimisation.
800
a
cCOsg (ppm)
600
400
200
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
3.5
cO2,sg (mol %)
2.5
2
1.5
1
0.5
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.13: Model outputs dynamic responses for DRTO/ENMPC of (a) cCOsg and (b) cO2,sg.
201
Figure 7.13 (a) illustrates an increase in CO emission (681 ppm) as shown in the first part, which
temporarily exceeds the allowed emission limits (350 ppm). Subsequently the controller was able to
bring the CO to a lower level, however the temporary violation of the constraints could results in a
significant amount of CO release into the atmosphere. The O2 level, shown in Figure 7.13 (b),
rapidly decreases to maintain the regenerator temperature at the desired level. The O2 concentration
goes down slightly when the disturbance began at 8000s and stabilizes at 2.48 %.
DRTO/ENMPC of the FCCU with CO environmental constraint

In this section, the same approach as in the previous case (case 3.1) was used, aiming to maximize
the LPG yield by applying an environmental constraint for CO emissions (to keep CO
concentration below 350 ppm). Figure 7.14 shows how the controllers corrective action with the
temperatures worked to achieve the optimal operating condition values, while maintaining the
environment constraints.
Tr (C)
554
532
520
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
753
730
717
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
100
50
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
1
0.5
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.14: Controlled and manipulated variables responses for the DRTO/ENMPC of maximizing
LPG with environmental constraints inthepresenceoff disturbance (5 % step increase at t = 8000 s).
202
It can be clearly seen in Figures 7.15 (a, b and c) for LPG, LCO and gasoline yields respectively,
that these were identical to the earlier values, especially in the second part. The first parts show that
slight damping occurred, because of the controllers corrective actions to maintain the CO
constraint limit, as shown in Figure 7.16 (a).
0.35
a
LPG (wt %)
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.4
LCO (wt %)
0.35
0.3
0.25
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Gasoline (wt %)
0.5
0.4
0.3
0.2
0.1
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.15: Model outputs dynamic responses for the DRTO/ENMPC (a) LPG, (b) LCO and (c)
gasoline.
203
However, the re-circulation slide valve for spent catalyst, svsc, opened fully, as a control action, to
allow more deposited catalyst to enter the upper regenerator section. After that, some of the hot
catalyst from the upper zone was taken to the lower zone in the regenerator to maintain the required
minimum temperature so that complete combustion is ensured. This is essential, because if the
combustion is incomplete, then the coke balance in the catalyst will start burning in the dilute phase
of the catalyst in the upper zone and in the flue gas lines. The heat that will be liberated during that
burning will not be absorbed fully, as the density of catalyst in that area is very low.
This will result in a cyclone effect, and other hardware being exposed to very high temperatures
causing mechanical damage. This phenomenon, as explained earlier, is called afterburning. In such
cases, the re-circulation slide valve is opened so that the hot catalyst flows back to the lower section
of the regenerator, thereby increasing temperature, which will result in more combustion taking
place and all the C and CO is converted into CO2.
700
cCOsg (ppm)
600
500
400
300
200
100
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
3.5
cO2,sg (mol %)
3
2.5
2
1.5
1
0.5
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.16: Model outputs dynamic responses for DRTO/NMPC of (a) cCOsg and (b) cO2,sg.
204
Case 4: DRTO/ENMP of the FCCU to minimize LCO yield to gain in Gasoline yield
In this case, the economic objective function is to minimize LCO based on seasonal demand. The
cost related weighting coefficient is
, and the constraints on the manipulated
inputs, considered here are: umax = [70 0.8], umin = [10 0.1]. The output constraints are defined as
follows: ymax = [555 760] and ymin = [525 685], and the sampling period is equal to 100 s. The
coking formation factor disturbance has been applied by 5% increase at t = 8000s. Figure 7.17
shows how the controlled and manipulated variables react to find the optimal operating condition
values to fulfil the economic objective function goal.
Tr (C)
539
526
513
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
Treg (C)
730
717
704
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
F5 (SCF/s)
40
30
20
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
4
x 10
svsc (%)
0.4
0.3
0.2
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.17: Controlled and manipulated variables responses for the DRTO/ENMPC of minimizing
LCO in the presence of f disturbance (5 % step increase at t = 8000 s).
205
Similarly to the previous case the manipulated inputs are saturated but this time at their minimum
limits. This limits the performance of the ENMPC approach to achieve global optimum
nevertheless the simulation results even under these constrained conditions show significant
economic benefits.
The DRTO/NMPC technique appears to show the LCO yield being minimized to be able to
maximize the gasoline yield.
0.25
LCO (wt %)
0.2
0.15
0.1
0.05
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.65
Gasoline (wt %)
0.6
0.55
0.5
0.45
0.4
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
0.18
LPG (wt %)
0.16
0.14
0.12
0.1
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.18: (a) LCO yield when the economic objective is to minimize LCO, DRTO/NMPC, (b)
Gasoline and (c) LPG.
206
Figure 7.18 (a), indicates that a sharp drop in the LCO yield to 0.092 is obtained, which is a
significant decrease compared with the LCO yield in case 1 (a decease with 17 % compare to the
benchmark values (Figure 7.3 (a)). In Figure 7.18 (b), the first part shows a sharp increase in
gasoline yield to 0.59 at 3000 s followed by a small decrease. In the second part, the gasoline
decreases and stabilizes at 0.57. Figure 7.18 (c) shows a sharp decrease in LPG yield, before the
disturbance, and the yield slightly increased, after the disturbance (to 0.11). However, this value is
still lower than case 1 (0.19).
0.5
cCOsg (ppm)
0.4
0.3
0.2
0.1
0
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
cO2,sg (mol %)
0.2
0.4
0.6
0.8
1
time (s)
1.2
1.4
1.6
1.8
2
4
x 10
Figure 7.19: Model outputs dynamic responses for DRTO/NMPC of (a) cCOsg and (b) cO2,sg.
Figure 7.19 (a) illustrates a quick decrease in CO emission (to 0.015 ppm), as revealed in the first
part, and continues to be much below the maximum emission limits. The O2 level in Figure 7.19 (b)
rapidly increases as the controller pushes less catalyst in the regenerator to maintain the regenerator
temperature at the desired level, and consequently the reactor temperature. The oxygen
concentration goes down slightly when the disturbance is introduced and stabilizes at (4.85 %).
207
However, the excess O2 (4.85 %) in the stack gas is economically wasteful, since energy is
expended to blow air into the regenerator vessel.
Table 7.3 summarises the results of cases obtained using RTO for different economic objective
functions and NMPC controller that have been considered in this chapter.
Table 7.3: Summary of results
Cases
Variable
Case 4
3.2
Case 3
3.1
2.2
Case 2
2.1
Case 1
Tr (C)
Treg (C)
LPG (%)
LCO (%)
Gasoline
cCOsg
cO2,sg
(%)
(ppm)
(%)
Initial
531.6
729.8
0.176
0.193
0.427
0.489
3.3
Optimal1
531.6
729.8
0.196
0.1777
0.4333
2.3e+4
0.0204
Initial
531.6
729.8
0.176
0.193
0.427
0.489
3.3
Optimal1
550.66
748.71
0.2706
0.356
0.1934
0.4512
3.4
Optimal2
551.2
751.6
0.267
0.365
0.187
0.489
3.315
Initial
551.2
751.6
0.176
0.193
0.427
0.489
3.3
Optimal1
550.32
749.5
0.279
0.343
0.197
0.451
3.4
Optimal2
551.2
751.6
0.277
0.336
0.207
104.8
1.347
Initial
531.6
729.8
0.176
0.193
0.427
0.489
3.3
Optimal1
553.28
751.9
0.283
0.363
0.166
1.85
2.8
Optimal2
555.1
756.4
0.285
0.374
0.148
3.832
2.484
Initial
531.6
729.8
0.176
0.193
0.427
0.489
3.3
Optimal1
553.26
751.9
0.283
0.363
0.166
1.82
2.8
Optimal2
555.1
756.4
0.285
0.374
0.148
3.832
2.484
Initial
531.6
729.8
0.176
0.193
0.427
0.489
3.3
Optimal1
526
725
0.112
0.086
0.572
0.0152
4.73
Optimal2
516.7
715.2
0.115
0.093
0.59
0.021
4.852
7.3 Conclusions
The varying conditions on the market have encouraged the chemical processing industry to
maximise the capabilities of the available plants. The productivity, and hence profitability, of high
capital cost plant must be linked more closely to market demand, which can be achieved by
208
designing control strategies which are able to take these demands into account and set the operating
conditions to provide optimal performance under changing market requirements.
In this work, a single layer, simplified, DRTO/ENMPC controller was studied. The controller
proposed combines both on-line optimisation and control of multiple parameters with constraints.
This allows simultaneous solution of both the control and optimisation problems as one. Initial
results revealed that this optimising controller can bring a highly disturbed process to a stable state,
as well as successfully controlling it at a new operating state, despite constraints, that satisfies
specific economic criteria.
Optimisation of plant during dynamic operation creates new opportunities for maximising plant
profitability and associated productivity. The novel concept of beneficial (good) and harmful (bad)
disturbances is introduced and it is shown that the proposed approach is able to differentiate
between these two scenarios, and potentially exploit the benefits of good disturbances, while
maintaining the plant within the operating constraints. An additional benefit of the economic
NMPC-based process simulator is its potential to be used as a practical tool for training plant
operating personnel in the new plant operation regimes. Plant operating personnel can use the
simulation tool to understand how the optimal operating conditions are achieved under different
disturbances in the context of new operating constraints for the key process parameters, enhancing
process understanding.
209
Chapter 8: Conclusions and Future Work
CHAPTER 8
CONCLUSIONS AND FUTURE WORK
This chapter provides an overview of the conclusions that can be drawn from the results of the
work presented in this thesis. Some aspects of the work that require further development and
improvement, which may form the basis for future investigations, are proposed.
8.1 Conclusions
The aim of this research project was to develop a mathematical model that can simulate the
behaviour of the FCC unit, which consists of feed and preheat system, reactor (riser and stripper),
regenerator, air blower, wet gas compressor, catalyst circulation lines, and the main fractionators.
The model was subsequently used in studies of control and economic optimisation. The developed
model is able to describe the complex dynamics of the reactor-regenerator system, and also
includes the dynamic model of the fractionator, as well as a detailed five lump kinetic model for the
riser (with components gas oil, gasoline, diesel, LPG and coke). This model is able to predict and
describe the compositions of the final production rate, and the distribution of the main components
in the final product. This allows the estimation of economic factors, related to the operation of the
FCCU. This aim has been attained through successful achievements of the relevant objectives listed
in Chapter 1 and individually presented in Chapters 2 to 7. The overall aim was to provide a
systematic and comprehensive framework for robust economic control for optimal system
operations. The main conclusions of the chapters and thesis are presented below.
In addition to an overview of some aspects of the advanced control processes as they currently
stand, Chapter 2 also presented a comprehensive literature review of the development of linear and
nonlinear model predictive control strategies. The detailed evaluation of the trends of industrial
applications and preparation of the model predictive control approaches indicate that linear and
nonlinear MPC are suitable technologies for the FCCU, with significant potential to increase the
profitability and efficiency of the unit. This is part of the objective to gain knowledge and
understanding of FCC unit behaviour, as well as its characteristics, so that these can be accounted
for in the process controller design. The work has produced a compilation of literature methods for
improving FCC unit productivity and efficiency that have been in use from 1941 to the present.
210
An original mathematical model for the FCC process was developed based on the momentum,
mass, and energy dynamic balances, and incorporates process hydrodynamics, heat transfer, mass
transfer and catalytic cracking kinetics based on a lumping strategy. Molecules and reactions are
lumped according by their boiling point and treated as pseudo-components for a global description
of the phenomena taking place in the reactor.
The FCCU mathematical model includes seven main components of the overal system: (1) feed and
preheat system, (2) reactor (riser and stripper), (3) regenerator, (4) air blower, (5) wet gas
compressor, (6) catalyst circulation lines, and (7) the main fractionator. The novelty of the model
consists in that besides the complex dynamics of the reactor-regenerator system, it also includes the
dynamic model of the fractionator, as well as a model capable of predicting the yields of valuable
products and gasoline octane value, and a kinetic model for the riser section, which also
incorporates the detailed effect of temperture on the product distribution. Different kinetic models
where studied starting with the three lumped kinetic model, to a complex model with 13 lumped
components. From the presented models, the five-lump (namely: gas oil, gasoline, diesel (LPG),
gases and coke) kinetic model for the riser section was chosen to model the product composition in
the FCC riser, since it corresponds to the available literature data and provides sufficient
complexity to describe the yields of the key valuable products and gasoline octane value.
The dynamic simulator was used to study the dynamic behaviour of the process. A set of dynamic
simulations were performed to study the dynamic response of the FCCU to disturbances. The
selected disturbances reflect possible upsets that affect the normal operation of the unit, and were
selected from a practical point of view, after several discussions and close cooperation with refinery
personnel. Simulations demonstrate that the process is multivariable, strongly interacting and
highly nonlinear. Inverse response behaviour has also been observed, indicating difficulties in the
controller design. The simulation study showed that the developed simulator is sufficiently complex
to capture the major dynamic effects that occur in an actual FCCU system and is able to depict the
main dynamic characteristics of a typical commercial FCC process. The simulator includes all
important economic, environmental and safety constraints of the overal system.
The developed simulator enables engineering and technical personnel to carry out research on the
design, operation, performance, and development of a proper control system for a modern catalytic
cracking unit. It could also be an efficient tool for training operating personnel.
211
As the catalytic cracking is achieved through endothermic reactions, operating the reactor at the
highest possible temperature, allows greater productivity. At the same time, catalyst regeneration
through coke burning also means that both reactor and regenerator should be operated at the
highest temperature allowed by their metallurgical limits. Stack gas carbon monoxide
concentration is also restricted due to environmental issues. A detailed dynamic FCC model and
solver were implemented in the dynamic simulator (as described in Chapter 3). The model
correlation equations relating to the transport and thermodynamic properties were derived from the
published literature, directly or through a non-linear regression. The products composition was
found to be significantly influenced by the disturbances applied in this study. Therefore, the
disturbances have a huge impact on process economics due to the FCC units high production
volume. Among the key disturbances studied is fresh feed coke formation factor, which has a large
effect on the process variables. The FCC unit control must not only act to raise productivity, but
must also ensure that the process is safe for the environment; stack gas CO concentration must be
maintained below specific limits.
In applying advanced PID controllers to the multivariable process, the closed-loop dynamic
behaviour was studied using the dynamic simulator. Analysis of the process controllability was
performed using a simple RGA matrix, with the aim of exploring the interactions between the
different process variables.The PID control approach was applied for several control structures,
which demonstrated that between some loops there were strong couplings. This could complicate
the tuning of individual loops. A genetic algorithm (GA) based optimsation approach was
implemented to determine the tuning parameters for the interacting PID control loops. This
approach was used to evaluate 2x2 and 3x3 control schemes in terms of setpoint tracking and
distrubance rejection performance. The results of the simulation show that increasing the number
of independent loops causes the PID controller performance to degrade significanty, as manifested
in higher settling interval. It was found that the control system stabilised the process, yet in the
presence of many disturbances, resulted in non-optimal operation resulting in negative
environmental effects or economic losses. Therefore, applying advanced multivariable control
systems to the control of the FCC unit is justified by the potential economic benefits to be
achieved.
The benefits of model predictive control as the most commonly used multivariate control approach
were demonstrated for the control of the FCCU. A detailed analysis of based on Linear MPC was
212
first performed which showed to advantages of using sets of control inputs and outputs based on
relative well decoupled systems, which can offer suitable performance of the lower level basic
control system, in a real industrial implementation. The study also indicated that the fuel flow in the
preheater is a key manipulated input in the control of the reactor temperature which is one of the
most important performance influencing process parameters. The benefits of non-square control
architectures with larger number of manipulated variables than controlled outputs was shown to
provide better decoupling between the controlled outputs than typical square control structures
(same number of inputs and outputs). Simulation results also indicated the limited applicability of
the linear MPC when applied to the nonlinear plant.
An efficient real-time nonlinear model predictive control (NMPC) strategy, based on efficient
multiple shooting optimization was developed using a real-time iteration scheme. The performance
of the proposed scheme was evaluated for the FCCU process. Based on this efficient approach a
single layer, DRTO/ENMPC controller was investigated, which combines real-time optimisation
with constrained multivariable control. Hence, this controller is able to simultaneously handle the
process control and economic optimisation of the plant. The primary results show that such a
controller is capable of stabilising a process and operating it within specified constraints to achieve
defined economic objectives. Several simulated case studies have demonsted that applying a realtime model-based online optimisation scheme based on economic criteria can provide optimal
process performance in case of changing quality of the raw material, while maintaining the process
within operational and environmental constraints.
8.2 Recommendations for future work

The following aspects of the work presented in this thesis could be incorporated in future work:
1. The present work permits the study of some interesting aspects regarding the modelling and
control of the catalytic cracking unit. One of these trends could be the improvement of the
present model by applying a more complex kinetic model for the riser section, together
with a more detailed model for the main fractionator and then model validation and
optimization based on experimental data. This would allow an even more comprehensive
investigation of the effects of changed operating conditions on the product distribution.
2. Another important development could be along the line of studies on thermal integration
applied to the existing catalytic cracking model and NMPC control; results could lead to
efficient use of raw material and energy, increased economic efficiency to reduce energy
213
and raw material consumption, reduction of operating costs or improvement of profit

margins of the process.
3. Plant-wide control is one of the advanced trends being applied by refiners. These systems
can look at large amounts of data, and are expected to make more consistent and logical
decisions than people could or would concerning unit or plant-wide profitability. The
presented study only used a 2x2 system for the economic control to provide a proof of
concept of the benefits of the ENMPC approach. Future studies incorporating a more
comprehensive plant-wide control scheme with multiple inputs and outputs in the ENMPC
algorithms may provide additional benefits of the control scheme.
4. Evaluation of an output feedback scheme in the control approach applied that incorporates
a state and parameter estimation step would significantly improve the robustness of the
control scheme. In the present study state feedback was assumed, which is a realistic
assumption considering that the main state variables modelled are measurable in the real
plant, however the employment of state and parameter estimation approach (e.g. extended
Kalman filter, or moving horizon estimation) would allow the use of more complex process
models and confer adaptability to changind process conditions.
5. More detailed economic objectives could be included in the economic optimisation, taking
into account production rates, operating and utility costs and other aspects that affect the
profit of the plant.
As a final conclusion, it is important to emphasise that FCCU control continues to be a vital field
with significant research problems, and a nonlinear model-based predictive control scheme that
allows the incorporation of economic objectives has significant potential benefits to provide a
positive step-change in the improvement of the current operating practices for industrial FCCUs.
214
References
References
1.
Abou-Jeyab, R.A., and Y.P. Gupta. 2001.Constrained multivariable control of uidized

catalytic cracking process using linear programming. Trans. I.Chem. Eng. A, 79:274-282.
2.
Adetola, V.A. 2008. PhD thesis, Integrated real-time optimization and model predictive
control under parametric uncertainties, Queens University, Kingston, Ontario, Canada.
3.
Adetola, V. and M. Guay. 2010. Integration of real-time optimization and model

predictive control. Journal of Process Control 20:125-133.
4.
Agachi, S.P., Z.K. Nagy, M.V. Cristea, and A.I. Imre-Lucaci. 2006. Model Based
Control, Wiley-VCH Verlag GmbH & Co. KGaA, 2006.
5.
Aguilar, R., J. Gonzalez, M. Barron, and J. AlvarezRamirez. 1999. Control of a uid

catalytic cracking units based on proportional-integral reduced order observers, Chem.
Eng. J. 75:77-85.
6.
Alhumaizi, K. and S. Elnashaie. 1997. Effect of control loop configuration on the

bifurcation behavior and gasoline yield of industrial fluid catalytic cracking (FCC) units,
Mathematicaland Coputer Modelling 25(3):36-56.
7.
Allgoewer, F., T.A. Badgwell, J.S. Qin, J.B. Rawlings and S.J. Wright. 1999. Nonlinear
predictive control and moving horizon estimation An introductory overview, in P.M.
Frank (Ed.), Advances in Control, Highlights of ECC99, Springer, 391-449.
8.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2008. Process Simulators Based Control Design
for a Fluid Catalytic Cracking Unit'', 18th International Congress of Chemical and
Process Engineering, Prague, Czech Republic.
9.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Mathematical Modelling and Control of a
UOP Fluid Catalytic Cracking Unit. 3rd Saudi International Conference, University of
Surrey, UK, 32.
10. Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Multivariate Control Structure Synthesis
for fluid catalytic cracking unit relative gain array. Chemical Engineering, D.B.Das, V.
Nassehi, L. Deka, EUROSIS-ETI. 7th International Industrial Simulation Conference.
Loughborough. ISBN 978-90-77381-4-89.251-256.
215
References
11. Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Process simulators based control design for
Universal Oil Products FCCU'', Chemical Engineering, 8th World Congress of Chemical
Engineering (WCCE 8), Montral, Canada, 2009.
12. Alvarez-Ramirez, J., R. Aguilar, and F. Lopez-Isunza. 1996. Robust regulation of
temperature in reactor-regenerator uid catalytic cracking units. Ind. Eng. Chem. Res.
35:1652-1659.
13. Ancheyta-Juarez, J. and R. Sotelo-Boyas. 2000. Estimation of kinetic constants of a fivelump model for fluid catalytic cracking process using simpler sub-models, Energy &
Fuels, 14: 1226-1231.
14. Ancheyta-Juarez, J., F. Lopez-Isunza and E. Aguilar-Rodriguez. 1999. 5-Lump kinetic
model for gas oil catalytic cracking. Applied Catal. A: Gen.177: 227-235.
15. Angeli, D. and J. Rawlings. 2010. Receding horizon cost optimization and control for
nonlinar plants, in Proceedings of 8th IFAC Symposium on Nonlinear Control Systems
(NOLCOS), Bologna, Italy.
16. Ansari, R.M. and M.O. Tade. 2000. Constrained nonlinear multivariable control of a Fluid
Catalytic Cracking Unit, Journal of Process Control. 10:539-555.
17. Arandes, J., M. Azokoiti, J. Bilbao, and H. de Lasa. 2001. Modeling FCC units under
steady and unsteady-state conditions. Can. J. Chem. Eng.78:1025-1031.
18. Arandes, J.M. and H.I. de Lasa. 1992. Simulation and multiplicity of steady states in
fluidized FCCUs. Chem Eng Sci. 47: 2535-2540.
19. Arandes, J.M., M.J. Azkoiti, J. Bilboa, and H.I. de Lasa. 2000. Modelling FCC units
under steady and unsteady state conditions. Can J Chem Eng.78:111-123.
20. Araujo-Monroy, C., and F. Lopez-Isunza. 2006. Modeling and simulation of an industrial
fluid catalytic cracking riser reactor using a lumping-kinetic model for a distinct
feedstock. Ind. Eng. Chem. Res. 45:120-128.
21. Arbel, A., Z. Huang, I. H. Rinard, R. Shinnar, and A.V. Sapre. 1995. Dynamic and control
of fluidized catalytic crackers. 1. Modeling of the current generation of FCC's, Ind. Eng.
Chem. Res. 34(4):12281243.
22. Arkun, Y., J. Hollett, W.M. Canney, and M. Morari. 1986. Experimental study of internal
model control, Ind. Engng Chem. Process Des. Dev., 25:102-108.
216
References
23. Avidan, A.A., and R. Shinnar. 1990. Development of catalytic cracking technology: a
lesson in chemical reactor design. Ind. Eng. Chem. Res. 29:931-942.
24. Backx, T., O. Bosgra, and W. Marquardt. 2000. Integration of model-predictive control
and optimization of processes. In IFAC Symposium Advanced Control of Chemical
Processes, ADCHEM, 249-260. Pisa, Italy.
25. Balchen, J.G., D. Lungquist, and S. Strand. 1992. State-space predictive control. Chem.
Eng. Sci., 47:787-798.
26. Bennett, S. 1993. A history of control engineering, 1930-1955. IET.48.
27. Bequette, B.W. 1991. Nonlinear control of chemical processes - A review. Ind. Eng.
Chem. Res. 30:1391-1413.
28. Bhattacharyya, A.A., G.M. Woltermann, J.S. Yoo, J.A. Karch, and W.E. Cormier. 1988.
Catalytic SOx Abatement: The Role of Magnesium Aluminate Spinel in the Removal of
SOx from Fluid Catalytic Cracking (FCC) flue gas, Ind. Eng. Chem. Res. 27(8):13561360.
29. Biegler, L.T. 2000. Efficient solution of dynamic optimization and NMPC problems. In F.
Allgwer and A. Zheng (Eds.), Nonlinear Predictive Control, Bagel, Birkhauser.
30. Biegler, L.T. 1984. Solution of dynamic optimization problems by successive quadratic
programming and orthogonal collocation, Comp. Chem. Eng. 8:243.
31. Biegler, L.T. and J.B. Rawlings. 1991. Optimisation approaches to nonlinear model
predictive control, Proceedings of Conf. Chemical Process Control, South Padre Island,
Texas, 543-571.
32. Binder, T., L. Blank, H.G. Bock, et al. 2001. Introduction to model based optimization of
chemical processes on moving horizons, In M. Groetschel, S.O. Krumke and J. Rambau
(Eds.), Online optimization of large scale systems, Berlin, Springer.
33. Blanding, F. 1953. Reaction Rates in Catalytic Cracking of Petroleum. Ind. Eng.Chem.
45:1186-1197.
34. Bock, H.G. and K. Plitt. 1984. A multiple shooting algorithm for direct solution of
optimal control problems. Proc. 9th IFAC World Congress, Budapest, Hungary.
217
References
35. Bock, H.G., I. Bauer, D. Leineweber, and J. Schloder. 1999. Direct multiple shooting
methods for control and optimization in engineering, In Scientific computing in chemical
engineering II, Volume 2, (F. Keil, W. Mackens, H. Vos, J. Werther, Eds.), Springer.
36. Bock, H.G., M. Diehl, D. Leineweber and J. Schlder. 2000a. A direct multiple shooting
method for real-time optimization of nonlinear DAE processes. in F. Allgwer and A.
Zheng (Edits.), Nonlinear Predictive Control, Birkhuser, 245.
37. Bock, H.G., M. Diehl, J. Schlder, F. Allgwer, R. Findeisen, and Z. Nagy. 2000b. Realtime optimization and nonlinear model predictive control of processes governed by
differential-algebraic equations. Proc. Int. Symp. Adv. Control of Chemical Processes
(ADCHEM), Pisa, Italy, 695.
38. Bohn, C. and D.P. Atherton. 1995. An analysis package comparing PID anti-windup
strategies. 15(2):34-40.
39. Bollas, G.M., A.A. Lappas, D.K. Iatridis, and I.A. Vasalos. 2007. Five-lump kinetic
model with selective catalyst deactivation for the prediction of the product selectivity in
the fluid catalytic cracking process. Research and Technology Hellas (CERTH). Catalysis
Today. Thermi-Thessaloniki, Greece. 127: 3143.
40. Bozicevic, J. and D. Lukec. 1987. Dynamic mathematical model for a fluid catalytic
cracking process, Trans Inst Meas Control 9(1): 8-12.CPCL, 1999, FCC Operating
Manual.
41. Brengel, D.D., W.D. Geider. 1989. Multistep nonlinear predictive controller, Ind. Eng.
Chem. Res., 20:1812-1822.
42. Bristol, E.H. 1966. On a new measure of interactions for multivariable process control.
IEEE Trans. Autom. Control, AC-11:133-134.
43. Brittain, J.E. 1977. Turning Points in American Electrical History, IEEE Press, New
York, Sect. II-E.
44. Caldwell, J.M. and G.D. Martin. 1987. Online analyzer predictive control, Sixth Annual
Control Expo Conf., Rosemont, Illinois.
45. Campo, P.J. and M. Morari.1986. Inf-No rm formulation of model predictive control
problems. Proc. Am. Control Conf., Seattle, Washington. 339-343.
218
References
46. Carabineiro, H., C.I.C. Pinheiro, F. Lemos, and F. Ramoa Ribeiro. 2004. Transient
microkinetic modelling of n-heptane catalytic cracking over H-USY zeolite, Chemical
Engineering Science. 59:1221-1232.
47. Carlson, D.A., A.B. Haurie, and A. Leizarowitz. 1991. Innite horizon optimal control.
Springer Verlag.
48. Chang, S.S.L. 1961. Synthesis of Optimum Control Systems. McGraw-Hill, New York.
49. Chen H., C.W. Scherer, and F. Allgower. 1997. A game theoretic approach to nonlinear
robust receding horizon control of constrained systems. In Proc. Amer. Contr. Conf.,
3073-3077, Albuquerque.
50. Chen J. and H. Cao. 1995. Catalytic Cracking Technology and Engineering of Fluidized
Catalytic Cracking. China Petrochemical Press.788-864.
51. Chen, H. and F. Allgower. 1997. A quasi-infinite horizon nonlinear predictive control
scheme for stable systems: Application to a CSTR, In Proc. Int. Symp. Adv. Control of
Chemical Processes, ADCHEM. Banff, Canada. 471-476.
52. Chen, H. and F. Allgower. 1998. Nonlinear model predictive control schemes with
guaranteed stability, In C. Berber, R. and Kravaris, editor, Nonlinear Model Based
Process Control, 465-494. Kluwer Academic Publishers, Dodrecht.
53. Chen, H. and F. Allgower. 1996. A quasi-infinite horizon predictive control scheme for
constrained nonlinear systems, In Proc. 16th Chinese Control Conference, Qindao.309316.
54. Chen, H., C.W. Scherer and F. Allgower. 1998. A robust model predictive control scheme
for constrained linear systems, In 5th IFAC Symposium on Dynamics and Control of
Process Systems, Korfu. DYCOPS-5, 60-65.
55. Cheng, W.C., G. Kim, A.W. Peters, X.Zhao, and K. Rajagopalan. 1998. Environmental
FCC technology. Catal. Rev. Sci. Eng. 40: (1-2) 39-79.
56. Chisci, L.L., A. Lombardi, and E. Mosca. 1996.
Dual-receding horizon control of
constrained discrete time systems, European Journal of Control. 2: 278-285.

57. Christensen, G., M.R. Minas, K.J. Hickey, and S.F. Jae. 1999. Future directions in
modeling the FCC process: An emphasis on product quality. Chemical Engineering
Science, 54(13): 2753-2764.
219
References
58. Clarke, D.W. 1994. Advances in model-based predictive control. In D.W. Clarke, editor,
Advances in Model-Based Predictive Control. Oxford University Press. 3-21.
59. Corella, J., E. Frances. 1991. In: M.L. Occelli (Ed.), Fluid Catalytic Cracking II (ACS
Symposium Series 452). American Chemical Society, Washington, DC.452:165.
60. Corma, A., and J. Martinez-Triguero. 1994. Kinetics of gas oil cracking and catalyst
decay on SAPO-37 and USY Molecular Sieves. App. Catal. 118:153-162.
61. Cristea, M.V. and S.P. Agachi. 1996. Controllability analysis of a model IV FCCU, in:
12th International Congress of Chemical and Process Engineering CHISA, 10.
62. Cristea, M.V., S.P. Agachi, and V. Marinoiu. 2003. Simulation and Model Predictive
Control of a UOP fluid catalytic cracking unit, Chemical Engineering and Processing.
42:67-91.
63. Cuthrell, J. and L. Biegler. 1989. Simultaneous optimization and solution methods for
batch reactor profiles. Comp. Chem. Eng. 13(1/2):49-62.
64. Cutler, C.R. and B.L. Ramaker. 1980. Dynamic Matrix Control, a computer control
algorithm, Proceedings of Joint Automatic Control Conference. San Francisco. 1: Paper
No. WP5-B.
65. Cutler, C.R. and R.B. Hawkins. 1987. Constrained multivariable control of a
hydrocracker reactor, Proc. Am. Control Conf., Minneapolis, Minnesota, 1014-1020.
66. Cutler, C.R. and R.T. Perry. 1983. Real Time Optimization with Multivariable Control is
Required to Maximize Profits. Comp. Chem. Eng. 7:663-1983.
67. Cutler, C., A. Morshedi, and J. Haydel. 1983. An industrial perspective on advanced
control, in AIChE Annual Meeting, Washington, D.C.
68. Das, A., E., Baudrez, B. Marin, and G. Heynderickx. 2003. Three-dimensional simulation
of a fluid catalytic cracking riser reactor. Ind. Eng. Chem. Res. 42:2602-2617.
69. Davidor, Y. 1991. Genetic Algorithms and Robotics:
a heuristic strategy for
optimization. Singapore: World Scientific.

70. De Gouvea M.T. and D. Odloak. 1998. One-layer real time optimization of LPG
production in the fcc unit: Procedure, advantages and disadvantages. Computers &
Chemical Engineering, 22:S191-S198.
220
References
71. De Lasa, H.I., A.F. Errazu, E. Barreiro, and S. Solioz. 1981. Analysis of fluidized bed
catalytic cracking regenerator models in an industrial scale unit. Can J Chem Eng. 59:
549553.
72. De Nicolao G., L. Magni, and R. Scattolini. 1996. On the robustness of receding horizon
control with terminal constraints, IEEE Trans. Automat. Contr. 41(3):451-453.
73. De Nicolao G., L. Magni, and R. Scattolini. 1997. Stabilizing receding-horizon control of
nonlinear time-varying systems, In Proc. 4rd European Control Con- ference ECC'97.
Brussels.
74. De Nicolao, G., L. Magni, and R. Scattolini. 1998. Stability and Robustness of Nonlinear
Receding Horizon Control, In Proceedings of International Symposium on Nonlinear
Model Predictive Control: Assessment and Future Directions (F. Allgower and A. Zheng,
Eds.). Ascona, Switzerland.77-90.
75. De Oliveira, N.M.C. and L.T. Biegler. 1995. An extension of Newton-type algorithms for
nonlinear process control, Automatica. 31(2): 281-286.
76. De Oliveira, N.M.C., and L.T. Biegler. 1994. Constraint handling and stability properties
of model-predictive control. AIChE J. 40(2):1138-1155.
77. De Oliveira, S.L. 1996. Model predictive control for constrained nonlinear systems, PhD
thesis, Swiss Federal Institute of Technology (ETH), Zurich, Switzerland.
78. Deb, K. 2001. Multi-objective optimization using evolutionary algorithms. Chichester,
UK: Wiley.
79. Diehl, M. 2001. Real-Time Optimization for Large Scale Nonlinear Processes. PhD
Thesis, University of Heidelberg.
80. Diehl, M., H.G. Bock, J.P. Schlder, R. Findeisen, Z. Nagy, F. Allgwer. 2002. Real-time
optimization and nonlinear model predictive control of processes governed by differential
algebraic equations. Journal of Process Control 12:577-585.
81. Diehl, M., R. Amrit, and J. Rawlings. 2010. A lyapunov function for economic optimizing
model predictive control. Accepted for Publication, IEEE Trans. on Auto.Cont.
82. Dulce, C.M. Silva, and Nuno M.C. Oliveira. 2002. Optimization and nonlinear model
predictive control of batch polymerization systems, Computers and Chemical
Engineering, 26:649-658.
221
References
83. Dupain, X., E.D. Gamas, R. Madon, C.P. Kelkar, M. Makkee, and J.A. Moulijn. 2003.
Aromatic gas oil cracking under realistic FCC conditions in a microriser reactor. Fuel.
82(13):1559-1569.
84. Dupain, X., M. Makkee, and J.A. Moulijn. 2006. Optimal conditions in fluid catalytic
cracking: a mechanistic approach. Applied Catalysis. 297:198-219.
85. Dupain, X., R. Krul, C. Schaverien, M. Makkee, and J. Moulijn. 2006. Production of
clean transportation fuels and lower olefins from fischer-tropsch synthesis waxes undel
fluid catalytic cracking conditions: the potential of highly paraffinic feedstock for FCC.
Applied Catalysis B: Enviromental. 63(3-4):277-295.
86. E. ref 1. www. gasification. org/what is gasification/environmental_benefits. aspx.
87. E.ref 2. www. labbiokraftstoffe.de/downlods/PDF/fachinformation/suedzuckerbernardd.pdf.
88. E. ref 3. World Oil Outlook 2007, O PECS ecretariat2, 007 [date accessed2: 5/02/2008].
89. E. ref 4. www.chevron.com/products/prodserv/fuels/bulletin/motorgar/3%5Frefining%2
Dtesting/ Default. asp.
90. E. ref 5. www. chevron. com/products/about/Richmond/ about/what we do. Asp.
91. E. ref 6. www.uop.com/objects/1%20Emerg% 201nterlock%2OSystemp.dfl.
92. E. ref 7.www. uop. com/PetroFCC.
93. E. ref 8. www. maiglobal. com/news/wpss. pdfJ.
94. E.ref 9. Franke, R., E. Arnold and H. Linke. HQP: A solver for nonlinearly constrained
large-scale optimization. http://hqp.sourceforge.net.
95. Edwards, W.M., and H.N. Kim. 1988. Multiple steady states in FCC unit operations.
Chem.Eng.Sci. 43:1825.
96. Elamurugan, P., and D.D. Kumar. 2010. Modeling and Control of Fluid Catalytic
Cracking Unit in Petroleum Refinery. International Journal of Computer Communication
and Information System (IJCCIS) 2(1):976-1349.
97. Ellis, R.C., X. Li, and J.B. Riggs. 1998. Modelling and Optimization of a Model IV
Fluidized Catalytic Cracking Unit. AIChE Journal 44(9):2068-2079.
222
References
98. Ellis, R.C., X. Li, and J.B. Riggs. 1998. Modelling and optimization of a model IV
fluidized catalytic cracking unit. American Institute of Chemical Engineers Journal
44:2068.
99. Elshishini, E.E. and S.H. Elnashaie. 1990. Digital Simulation of Industrial Fluid Catalytic
Cracking Units; Bifurcation and its Implications. Chem .Eng. Sci. 45:553.
100. Emad, E.A. and S.E. Elnashaie. 1997. Nonlinear model predictive control of industrial
type IV fluid catalytic cracking units for maximum gasoline yield, Ind. Chem. Eng. Res.
36:389-398.
101. Eng, M.T., T.W. Cadman, and A.M. Kugelman. 1974. Nonlinear feedforward control of a
fluidized catalytic cracking process. ISA Trans.13:232.
102. Engell, S. 2007. Feedback control for optimal process operation. Journal of Process
Control 17:203-219.
103. Errazu, A.F., H.I. de Lasa, and F. Sarti. 1979. A fluidized bed catalytic cracking
regenerator model grid effects, Can J Chem Eng. 57: 191-197.
104. Feng, W., E. Vynckier, and G. Froment. 1993. Single-event kinetics of catalytic cracking.
Ind. Eng. Chem. Res. 32:2997-3005.
105. Findeisen, R. and F. Allgower. 1999. A procedure to determine a region of attraction for
discrete time nonlinear systems, Technical Report #99.14, Automatic Control Laboratory,
Swiss Federal Institute of Technology (ETH), Zurich, Switzerland.
106. Findeisen, R., and F. Allgower. 2000(a). Nonlinear model predictive control for index-one
DAE systems, In Nonlinear predictive control, (F. Allgower and A. Zheng, Eds.),
Birkhauser.
107. Findeisen, R., and J.B. Rawlings. 1997. Suboptimal infinite horizon nonlinear model
predictive control for discrete time systems, Technical Report # 97.13, Automatic Control
Laboratory, Swiss Federal Institute of Technology (ETH), Zurich, Switzerland,.
Presented at the NATO Advanced Study Institute on Nonlinear Model Based Process
Control.
108. Fisher, J.N. and S.L. Shah. 1986. Interaction analysis in multivariable control systems,
AIChE J. 32:959-970.
223
References
109. Fonseca, C.M. and P.J. Fleming. 1993. Genetic algorithms for multi-objective
optimization: formulation, discussion and generalization. In Proceedings of the Fifth
International Conference on Genetic Algorithms, San Mateo, CA: Morgan Kaufmann,
416-423.
110. Ford, W.D., R.C. Reineman, I.A. Vasaloc, and R.J. Fahrieg. 1976. Modeling catalytic
cracking regenerators. NPRA Annual Meeting.
111. Friedly, J.C. 1984. Use of the Bristol array in designing noninteracting control loops, A
limitation and extension, Ind. Eng. Chem. Process Des. Dev. 23:469-472.
112. Fujiwara, M., D.L. Ma, T. Togkalidou, D.K. Tafti and R.D. Braatz. 2002. Identification of
pharmaceutical crystallization processes. Proc. of the IFAC World Congress on Automatic
Control, Elsevier Science Ltd., Oxford, UK, Paper T-Fr-A11.
113. Galan, O., J.A. 2000. Romagnoli, and A. Palazoglu, Robust H control of nonlinear plants
based on multi-linear models: an application to a bench-scale pH neutralization reactor,
Chem. Eng. Sci. 55:4435-4450.
114. Garcia, C.E. and A.M. Morshedi. 1986. Quadratic programming solution of dynamic
matrix control (QDMC), Proc. Am. Control Conf., San Diego, California; also Chem.
Engng Commun. 46:73-87.
115. Garcia, C.E. and M. Morari. 1982. Internal model control-l. A unifying review and some
new results, Ind. Engng Chem. Process Des. Dev. 21:308-323.
116. Garcia, C.E. and M. Morari. 1985a. Internal model control-2. Design procedure for
multivariable systems, Ind. Engng Chem. Process Des. Dev. 24:472-484.
117. Garcia, C.E. and M. Morari. 1985b. Internal model control-3. Multivariable control law
computation and tuning guidelines, Ind. Engng Chem. Process Des. Dev. 24:484-494.
118. Garcia, C.E., D.M. Prett, and M. Morari. 1989. Model Predictive Control: Theory and
practice - A survey, Automatica. 25:335-347.
119. Garcia, C.E. 1984. Quadratic dynamic matrix control of nonlinear processes. An
application to a batch reaction process, AIChE Annual Meeting, San Francisco,
California.
120. Gary, J.H. and G.E. Handwerk. 1975. Petroleum refining Technology and Economics,
Marcel Dekker, INC., New York. 86-87.
224
References
121. Gates, B. 1979. Chemistry of catalytic processes, McGraw-Hill book company, Gomez
Prado J., UMIST dissertation, 2003.
122. Goldberg, D.E. 1989. Genetic Algorithms in Search, Optimization and Machine Learning.
Addison-Wesley.
123. Gomez Prado J., UMIST dissertation, 2003.
124. Gomez-Prado J., N. Zhang and C. Theodoropoulos. 2006. Integrated feedstock
characterization, kinetic modelling and process dynamics for FCC units. Conference
proceedings of the 17th International Congress of Chemical and Process Engineering
CHISA 2006, Prague, Czech Republic.
125. Gopaluni, R.B., R.S. Patwardhan and S.L. Shah. 2003. The nature of data pre-filters in
MPC relevant identification - Open- and closed-Loop issues. Automatica 39(9):16171626.
126. Gros, S., B. Srinivasan and D. Bonvin. 2009. Optimizing control based on output
feedback. Comp. & Chem. Eng. 33(1):191-198.
127. Grosdidier, P.M. Morari, and B.R. Holt. 1985. Closed loop properties from steady state
gain information, Ind. Eng. Chem. Fundam. 24:221-235.
128. Grosdidier, P., A. Mason, A. Aitolahti, P. Heinonen, and V. Vanhamaki. 1993. FCC unit
reactor-regenerator control. Computers Chem. Engng. 17(2):165-179.
129. Grosdidier, P., B. Froisy, and M. Hammann. 1988. The IDCOM-M controller, In T.J.
McAvoy, Y. Arkun, and E. Zafiriou, edit. Proceedings of the 1988 IFAC Workshop on
Model Based Process Control. Oxford, Pergamon Press. 31-36.
130. Guigon, P., and J.F. Large. 1984. Application of the Kunii-Levenspiel model to a multistage baffled catalytic cracking regenerator, Chem Eng J. 28:131-138.
131. Gupta, A. and D. Rao. 2003. Effect of feed atomization on FCC performance: simulation
of entire unit, Chemical Engineering Science. 58:4567-4579.
132. Hagelberg, P. and I. Eilos, J. Hiltunen, K. Lipiinen and V.M. Niemi, J. Aittamaa and
A.O.I. Krause. 2002. Kinetics of catalytic cracking with short contact times. Applied
Catal.223: 73-84.
225
References
133. Haggblom, K.E. 1994. Modeling of control structures for partially controlled plants,
Prepr. IFAC Workshop on Integration of Process Design Control (IPDC94). Baltimore,
Maryland, USA.183-188.
134. Hajela, P., E. Lee, and C.Y. Lin. 1992. Genetic algorithms in structural topology
optimization. In Proceedings of the NATO Advanced Research Workshop on Topology
Design of Structures, Sesimbra, Portugal, 117-134.
135. Halvorsen, I.J. and S. Skogestad. 2000. Distillation Theory, In: Encyclopedia of
Separation Science. D. Wilson (Editor), Academic Press.1117-1134.
136. Halvorsen, I.J. and S. Skogestad. 2003. Minimum Energy Consumption in
Multicomponent Distillation. Part 1- Vmin Diagram for a Two-Product Column,
Industrial and Engineering Chemistry Research. 42(3):594-604.
137. Han, I.S., C.B. Chung, and J.B. Riggs. 2000. Modeling of a fluidized catalytic cracking
process, Computers Chemical Engineering. 24:1681-1687.
138. Han, I.S. and C.B. Chung. 2001. Dynamic modelling and simulation of a fluidized
catalytic cracking process. Chem. Eng. Sci. 55:1951-1971.
139. Han, I.S. J. B. Riggsa, and Chang-Bock Chungb. 2004. Modeling and optimization of a
fluidized catalytic cracking process under full and partial combustion modes. Chemical
Engineering and Processing. 43:1063-1084.
140. Hassapis, G. 2003. Implementation of model predictive control using real-time
multiprocessing computing, Microprocessors and Microsystems.27:327-340.
141. Heidarinejad, M., J. Liu, and P.D. Cristofides. 2011. Economic model predictive control
of nonlinear process system using lyapunov thechniques. AICHE Journal, DOI
10.1002/aic.12672.
142. Helbig, A., O. Abel, and, W. Marquardt. 2000. Structural concepts for optimization based
control of transient processes. In F. Allgwer and A. Zheng (eds.), Nonlinear Model
Predictive Control, Progress in Systems and Control Theory 26:295-311. Birkhauser,
Basel.
143. Henson, A.M. 1998. Nonlinear model predictive control: current status and future
directions, Comp. Chem. Eng. 23:187-201.
226
References
144. Hicks, G. and W. Ray. 1971. Approximation methods for optimal control synthesis. Can.
J. Chem. Eng. 49:522-528.
145. Hicks, R.C., G.R. Worrell, and R.J. Durney. 1966. Atlantic Seeks Improved Control;
Studies Analog-Digital Models. Oil and Gas J. 24:97.
146. Holland, J.H. 1975. Adaptation in natural and artificial systems. Ann Arbor: M I T Press.
147. Horn, J., and N. Nafploitis. 1993. Multiobjective optimization using the niched Pareto
genetic algorithm. Technical Report 93005, IlliGAL, University of Illinois, UrbanaChampaign.
148. Hovd, M. and S. Skogestad. 1992. Simple frequency dependent tools for control system
analysis, structure selection and design, Automatica, 28:989-996.
149. Hovd, M. and S. Skogestad. 1993. Procedure for regulatory control structure selection
with application to the FCC process. AIChE Journal. 39(12):1938-1953.
150. Hovd, M., and S. Skogestad. 1991. Controllability analysis for the fluid catalytic cracking
process, in: AIChE Annual Meeting, 17:/22, Los Angeles.
151. Hovd, M., and S. Skogestad. 1992a. Simple Frequency-Dependent Tools for Control
System Analysis, Structure Selection and Design, Automatica 28: 989.
152. Hovd, M., and S. Skogestad. 1992b. Controllability Analysis for Unstable Processes,
IFAC Workshop on Interactions between Process Design and Process Control, Pergamon
Press, London, 49.
153. Huang, R., E. Harinathc, and L.T,. Biegler. 2011. Lyapunov stability of economically
oriented NMPC for cyclic processes, Journal of Process Control 21:501-509.
154. Huesman, A.E.M., O.H. Bosgra, and P.M. den Hof. 2008. Integrating MPC and RTO in
the process industry by economic dynamic lexicographic optimization; an open-loop
exploration. In Proceedings of the AICHE Annual Meeting. Philadelphia, USA.
155. Humphries, A., and R.C. Skocpol. 2004. Fluid Cracking Catalysts design technology,
C.PG. Smit, PG. Connor, (Editors), Catalyst Courier. 33:2004.
156. Imsland, L., R. Findeisen, E. Bullinger, F. Allgower, and B. A. Foss. 2003. A note on
stability, robustness and performance of output feedback nonlinear model predictive
control, Journal of Process Control.13:633644.
227
References
157. Iscol L. 1970.The Dynamics and Stability of a Fluid Catalytic Cracker. Proc. 1970 JACC,
session paper 23-B, 602.
158. Jacob, S., B. Gross, S. Voltz, and V. Weekman. 1976. A lumping and reaction scheme for
catalytic cracking. AIChEJ. 22 (4):701-713.
159. Jacob, S.M., B. Gross, S.E. Voltz, and V.M. Weekman. 1976. A lumping and reaction
scheme for catalytic cracking, AIChE J. 22(4):701-713.
160. Jia C., S. Rohani, and A. Jutan. 2003. FCC unit modeling, identification and model
predictive control- a simulation study, Chemical Engineering and Processing 42: 311325.
161. Jianping, Gao., Rohit Patwardhan, K. Akamatsu, Y. Hashimoto, G. Emoto, L. Shah
Sirish, and Biao Huang. 2003. Performance evaluation of two industrial MPC controllers,
Control Engineering Practice, in press.
162. John, T.M. and B.W. Wojciechowski. 1975. On identifying the primary and secondary
products of the catalytic cracking of neutral distillates, J. Catal., 37:240-250.
163. Kacprzyk, J. 1995. A modified genetic algorithm for multistage control of a fuzzy system.
In Proceedings EUFIT95, 1: Aachen, Germany, 463 6.
164. Kadam, J., W. Marquardt, M. Schlegel, T. Backx, O. Bosgra, P.-J. Brouwer, G.
Dunnebier, D. Hessem, A. Tiagounov, and S. de Wolf. 2003. Towards integrated
dynamic real-time optimization and control of industrial processes, in I. E. Grossmann, C.
M. McDonald (Eds.): FOCAPO, Proceedings of the Con- ference. 593-596.
165. Kadam, J. V. and W. Marquardt. 2007. Nonlinear model predictive cotnrol of the
hashimoto simulated moving bed process, in R. Findeisen, F. Allgwer, and L. Biegler,
edits, Assessment and Future Directions of Nonlinear Model Predictive Control,
Springer, 419434.
166. Kadam, J., M. Schlegel, W. Marquardt, R. Tousain, D. Hessem, J. Berg, and O. Bosgra.
2002. A two-level strategy of integrated dynamic optimization and control of industrial
processes - a case study, in J. Grievink, J. v. Schijndel (Eds.): European Symposium on
Computer Aided Process Engineering. 12:511-516.
228
References
167. Kalra, L., and C. Georgakis. 1994. Effect of Process Nonlinearity on the Performance of
Linear Model Predictive Controllers for the Environmentally Safe Operation of a Fluid
Catalytic Cracking Unit, Ind. Eng. Chem. Res. 33:3063-3069.
168. Kambreck, F. 1991. Continuous mixtures in fluid catalytic cracking and extensions. Mobil
workshop on chemical reaction in complex mixtures, Van Nostrand Reinhold, New York.
169. Karla, L., and C. Georgakis. 1994. Effect of process nonlinearity on performance of linear
model predictive controllers for environmentally safe operation of a fluid catalytic
cracking unit. Ind. Eng. Chem. Res. 33:3063-3069.
170. Karr, C.L. 1992. An adaptive system for process control using genetic algorithms .In
IFAC Artificial Intelligence in Real-Time Control, Delft , The Netherlands.329-34
171. Katz, R.N. 2001. Advanced ceramics: Zeolites equal more miles per barrel, ceramic
Industry. 151(12):16-20.
172. Keerthi, S.S. and E.G. Gilbert. 1988. Optimal infinite-horizon feedback laws for a general
class of constrained discrete-time systems: Stability and moving-horizon approximations,
J. Opt. Theory and App. 57 (2): 265-293.
173. Kirby, A. 2004. What to use when the oil runs out. BBC news online environment
correspondent at http: //news. bbc. co. uk/l/hi/sci/tech/3623675. stm (date accessed:
10/03/2008).
174. Kiss, P., and F. Szeifert. 1998. Coordinated control of a uid catalytic cracking unit.
Chem. Eng. Tech. 21:515-521.
175. Kookos, I.K. 2005. Real time regulatory control structure selection based on economics.
Ind. Eng. Chem. Res. 44:3993-4000.
176. Krishna, A.S. and E.S. Parkin. 1985. Modelling the regenerator in commercial fluid
catalytic cracking units, Chem Eng Prog. 81(4):57-62.
177. Krstic, M. and H. Wang. 2000. Stability of extrmum seeking feedback for general
nonlinear dynamic systems, Automatica 36:595-601.
178. Kumar, S., A. Chadha, R. Gupta, and R. Sharma. 1995. A process simulator for an
integrated FCC-regenerator system, Ind Eng Chem Res. 34:3737-3748.
229
References
179. Kumar, V. and A.A. Sharma. 2001. A crude distillation unit model suitable for online
applications, Fuel Processing Technology.73:1-21.
180. Kurihara H. 1967.Optimal Control of Fluid Catalytic Cracking Processes. Sc.D. Thesis,
Massachusetts Institute of Technology.
181. Kursawe, F. 1990. A Variant of evolution strategies for vector optimization. In Parallel
Problem Solving From Nature, I (PPSN-I), Dortmund, Germany, 193-197.
182. Kwon, W. H. 1994. Advances in predictive control: Theory and applications. In 1st Asian
Control Conference, Tokyo.
183. Larocca, M., S. Ng, and H. Lasa. 1990. Fast catalytic cracking of heavy gas oils:
modelling coke deactivation, Ind Eng Chem Res. 29:171-180.
184. Le Lann M.V., M. Cabassud, and G. Casamatta. 1999. Modeling, optimization and
control of batch chemical reactors in fine chemical production, Annual Reviews in
Control. 23:25-34.
185. Lee J.H. and B. Cooley. 1996. Recent advances in model predictive control and other
related areas. In J.C. Kantor, C.E. Garcia, and B. Carnahan, editors, Fifth International
Conference on Chemical Process Control - CPC V, 201-216, American Institute of
Chemical Engineers.
186. Lee J.H. 1998. Modeling and identification for nonlinear model predictive control:
requirements, current status and future research needs, In Proceedings of International
Symposium on Nonlinear Model Predictive Control: Assessment and Future Directions
(F. Allgower and A. Zheng, Eds.). Ascona, Switzerland. 91-107.
187. Lee L., Y. Chen, T. Huang, W. Pan. 1989. Four-lump kinetic model fro fluid catalytic
cracking process. The Canadian Journal of Chemical Engineering.67:615-619.
188. Lee, E. and F.R. Groves. 1985. Mathematical model of the fluidized bed catalytic
cracking plant, Transaction of the Society for Computer Simulation. 3(3):219-236.
189. Lee, J.H. and Z.H. Yu. 1994. Tuning of model predictive controllers for robust
performance, Comp. Chem. Eng. 18:15-37.
190. Lee, L., S. Yu, C. Cheng, and W. Pan. 1989a. Fluidized-bed catalyst cracking regenerator
modelling and analysis, Chem Eng J. 40:71-82.
230
References
191. Lee, L., Y. Chen, T. Huang, and W. Pan. 1989b. Four-lump kinetic model for fluid
catalytic cracking process. Can J Chem Eng 67:615-619.
192. Lee, W. and A.M. Kugelman. 1973. Number of steady-state operating points and local
stability of open-loop fluid catalytic cracker, Ind Eng Chem process Des Dev. 12(2):197204.
193. Levien, K.L. and M. Morari. 1987. Internal model control of coupled distillation columns.
AIChE Journal 33(1):8398.
194. Liao, Bo. 2008. Time optimal control of fluid catalytic cracking unit. Ph.D
thesis. University of Toronto, Canada.
195. Linkens, D. A. and H. O. Nyongesa. 1995b. Genetic algorithms for fuzzy control - part 2:
online system development and application. IEE Proceedings on Control Theory
Applications 142(3):177-185.
196. Linkens, D. A. and H. O. Nyongesa. 1995a. Genetic algorithms for fuzzy control part1:
offline system development and application. IEE Proceedings on Control Theory
Applications 142(3):161-175.
197. Lopez-Isunza, F. 1992. Dynamic modelling of an industrial fluid catalytic cracking unit,
Computers and Chemical Engineering 16(1):S139-S148.
198. Lu, J. Z. 2001. Challenging control problems and emerging technologies in enterprise
optimization, Proceedings of the 6th IFAC Symposium on Dynamic and Control of
Process Systems (DYCOPS-6.23-34.
199. Luenberger, D.j. 1971. An introduction to observers. IEEE Trans. Automat. Contr.
16:596-602.
200. Magee, J.S. and M.M. Mitchell. 1993. Fluid catalytic cracking: science and technology.
Elsevier. Amsterdam; New York.
201. Marques, D. and M. Morari. 1896. Model predictive control of gas pipeline networks,
Proc. Am. Control Conf., Seattle, Washington.349-354.
202. Marquis, P. and J.P. Broustail. 1988. SMOC, a bridge between state space and model
predictive controllers: Application to the automation of a hydrotreating unit, In T.J.
McAvoy, Y.Arkun, and E. Zafiriou, editors, Proceedings of the 1988 IFAC Workshop on
Model Based Process Control. Oxford, Pergamon Press. 37-43.
231
References
203. Matsko, T. N. 1985. Internal model control for chemical recovery. Chem. Engng Prog.
81(12):46-51.
204. Mayne, D.Q. 2000. Nonlinear model predictive control: Challenges and opportunities, In
F. Allgower and A. Zheng, editors, Nonlinear Predictive Control. Birkhauser, Basel.
205. Mayne, D.Q. and H. Michalska. 1990. Receding horizon control of nonlinear systems,
IEEE Trans. on Automatic Control. 35(7):814-824.
206. Mayne, D.Q., J.B. Rawlings, C.V. Rao, P.O. Scokaert. 2000. Constrained model
predictive control: Stability and optimality, Automatica. 36:789-814.
207. Mayne, D.Q. 1996. Nonlinear model predictive control: An assessment, In J.C. Kantor,
C.E. Garcia, and B. Carnahanr editors, Fifth International Conference on Chemical
Process Control -CPC V. 217-231, American Institute of Chemical Engineers.
208. Mayr, O. 1970. The origins of feedback control, M.I.T. Press, Cambridge, Mass.
209. McAvoy, T. 2002. Intelligent control applications in the process industries, Annual
Reviews in Control. 26:75-86.
210. McAvoy, T.J. 1983. Interaction Analysis, Instrument Society of America, Research
Triangle Park, USA.
211. McFarlane, R.C. and D.W. Bacon. 1989. Adaptive optimizing control of multivariable
constrained chemical processes. 1. Theoretical development. Industrial and Engineering
Chemistry Research 28:1828.
212. McFarlane, R.C. and R.C. Reineman, J.F. Bartee, and C., Georgakis. 1993. Dynamic
simulator for a model IV Fluid Catalytic Cracking Unit, Computers Chemical
Engineering, 17(3):275-300.
213. McFarlane, R.C., R.C. Reineman, J.F. Bartee and C. Georg akis. 1990. Dynamic
simulation for a model IV fluid catalytic cracking unit. Presented at AIChE Meeting,
Chicago Ill., paper 24d, 14-16.
214. McGreavy, C.and P.C. Isles-Smith. 1986. Modelling of a fluid catalytic cracker,
Transactions of the Institute of Measurement and Control 8: 130-136.
215. Meadows, E. S. and J. B. Rawlings. 1993. Receding horizon control with an infinite
horizon, In Proc. Amer. Contr. Conf. San Francisco. 2926-2930.
232
References
216. Meadows, E. S., M. A. Henson, J. W. Eaton, and J. B. Rawlings. 1995. Receding horizon
control and discontinuous state feedback stabilization, Int. J. Contr. 62(5):1217-1229.
217. Meyers, R. A. 1997. Handbook of petroleum refining processes. 2nd edn. New York:
McGraw-Hill.
218. Michaska, H., D.Q. Mayne. 1993. Robust receding horizon control of constrained
nonlinear systems, IEEE Transactions on Automatic Control. 38(11):1623-1633.
219. Miriam Tvrzskh de GouvSa, and Darci Odloak. 1998. One-layer real time optimization of
LPG production in the FCC unit: procedure, advantages and disadvantages. Computers
chum .Engng 22:S191-S198.
220. Mohsen, M.S. and O.R. AI-Jayyousi. 1999. Brackish water desalination: an alternative for
water supply enhancement in Jordan, Desalination 124:163-174.
221. Morari, M. and J.H. Lee. 1997. Model predictive control: Past, present and future. In
Proc. PSE'97-ESCAPE-7 Symposium, Trondheim.
222. Morari, M., G. Stephanopoulos, and Y. Arkun. 1980. Studies in synthesis of control
structures for chemcial processes, part 1, AIChE J. 26:220-232.
223. Murata, T. 1997. Genetic algorithms for multi-objective optimization. Ph.D. Thesis,
Osaka Prefecture University, Osaka, Japan.
224. Muske, K.R., J.W. Howse, G.A. Hansen and D.J. Cagliostro. 2000b. Model-based control
of a thermal regenerator. Part 2: control and estimation, Comp. Chem. Eng., 24:25072517.
225. Muske, K.R., J.W. Howse, G.A. Hansen, and D.J. Cagliostro. 2000a. Model-based control
of a thermal regenerator. Part 1: dynamic model, Comp. Chem. Eng. 24:2519-2531.
226. Nagy, Z. and S. Agachi. 1997. Model Predictive Control of a PVC Batch Reactor, Comp.
Chem. Eng. 21(6):571-591.
227. Nagy, Z. K. 2001. Model Predictive Control techniques for chemical processes. PhD
Thesis. University of Clujnapoca.
228. Nagy, Z. K. and S. P. Agachi. 2004. Model based control of chemical processes. Press
Universitara Clujean.
233
References
229. Nagy, Z.K. and R.D. Braatz. 2003. Robust nonlinear model predictive control of batch
processes, AIChE J. 49(7):1776-1786.
230. Nagy, Z.K. and F. Allgower. 2007. A nonlinear model predictive control approach for
robust end-point property control of a thin film deposition process. International Journal
of Robust and Nonlinear Control 17:1600-1613.
231. Nikolaou, M. 2001. Model Predictive Controllers: A Critical Synthesis of Theory and
Industrial Needs. Advances in Chemical Engineering Series.
232. O'Connor, P., P. Imhof, and S.J. Yanik. 2001. Catalyst assembly technology in FCC. In
Fluid Catalytic Cracking V. Materials and Technologh Innovations; Occelli, M.L.,
O'Connor, P., Eds.; Studies in Surface Science and Catalysis; Elsevier.134.
233. Oliveira, L. 1987. Estimate of parameters and Evaluation Model Catalytic cracking. MSc
Thesis (in Portuguese), Univ. Federal do Rio de Janeiro, Brazil.
234. Oliveira, L. and E.C. Biscaia. 1989. Catalytic cracking kinetic models parameter
estimation and model evaluation. Ind. Eng. Chem. Res. 28:264-271.
235. Oliveira, L.L., E.C. Biscaia. 1989. Ind. Eng. Chem. Res. 28:264.
236. Pan Y., J.H. Lee. 2003. Recursive data-based prediction and control of product quality for
a PMMA batch process, Chemical Engineering Science. 58:3215- 3221.
237. Paraskos, J.A., Y.T. Shah, J.D. McKinney, and N.L. Carr. 1976. A kinematic model for
catalytic cracking in a transfer line reactor, Ind Eng Chem Process Des Dev. 15(1):165169.
238. Park Myung-June, Hyun-Ku Rhee. 2003. Property evaluation and control in a semibatch
MMA/copolymerization reactor, Chemical Engineering Science.58:603-611.
239. Parrish, J.R. and C.B. Brosilow. 1985. Inferential control applications, Automatica.
21:527-538.
240. Patwardhan, A.A., J.B. Rawlings, and T.F. Edgar. 1990. Nonlinear model predictive
control, Chem. Eng. Comm. 87:123-141.
241. Petry, E.E., B.P. Buckles, D. Prabhu, R. George, and R. Srikanth. 1994. Fuzzy clustering
with genetic search. In Proceedings of the First IEEE Conference on Evolutionary
Computation,1: 46-50.
234
References
242. Pitault, I., D. Nevicato, M. Forissier, and J. Bernard. 1994. Kinetic model based on a
molecular description for catalytic cracking of vacuum gas oil. Chem. Eng. Sci.
49(24A):4249-4262.
243. Pohlenz, J. B. 1963. How Operational Variables Affect Fluid Catalytic Cracking. Oil Gas
J. 61 :( 13), 124-143.
244. Poloski, A. P., and J. C. Kantor. 2003. Application of model predictive control to batch
processes, Computers and Chemical Engineering. 27:913-926.
245. Prett, D.M. and R.D. Gillette. 1979. Optimization and constrained multivariable control of
a catalytic cracking unit, AIChE National Meeting, Houston; Texas; also Proc. Joint Aut.
Control Conf., San Francisco, California.
246. Lee, W. and A.M., Kugelman. 1973. Number of steady-state operating points and local
stability of open- loop fluid catalytic cracker. Ind. Eng. Chem. Process Des. Dev. 12:
197-204.
247. Qin S.J. and T. Badgewell. 2003. A survey of industrial model predictive control
technology. Control Engineering Practice 11:733-764.
248. Qin, S.J. and T.A. Badgwell. 2000. An overview of nonlinear model predictive control
applications. In F. Allgower and A. Zheng, editors, Nonlinear Predictive Control,
Birkhauser, Basel.
249. Rahul B. Kasat, Santosh K. Gupta. 2003. Multi-objective optimization of an industrial
fluidized-bed catalytic cracking unit (FCCU) using genetic algorithm (GA) with the
jumping genes operator.Computers and Chemical Engineering 27:1785-1800.
250. Rawlings, J. and R. Amrit. 2009. Optimizing process economic performance using model
predictive control, in L. Magni, D. M. Raimondo and F. Allgwer, edits, Assessment and
Future Directions of Nonlinear Model Predictive Control, Springer, 119-138.
251. Rawlings, J.B. and R. Amrit. 2008. Optimizing process economic performance using
model predictive control. In Assessment and Future Directions of Nonlinear Model
Predictive Control. Pavia, Italy.
252. Rhemann, H., G. Schwartz, T. Badgwell, M. Darby, and D. White. 1989. On-line FCCU
advanced control and optimization, Hydrocarbon Processing 64-71.
235
References
253. Richalet, J., A. Rault, J.L. Testud, and J. Papon. 1978. Model Predictive Heuristic
Control: application to industrial processes, Automatica. 14:413-428.
254. Roman, R., Z. Nagy, M. Cristea, and S. Agachi. 2009. Dynamic modelling and nonlinear
model predictive control of a Fluid Catalytic Cracking Unit, 2009, Computers and
Chemical Engineering. 33(3):605-617.
255. Sadeghbeigi, R. 1995. Fluid catalytic cracking handbook. Houston, Tex.: Gulf Pub. Co.
256. Sadeghbeigi, R. 2000. Fluid Catalytic Cracking Handbook - Design, Operation and
Troubleshooting of FCC Facilities, second edition. Gulf professional publishing.
257. Saxton, A.L. and A.C. Worley. 1970. Modern catalytic cracking design, Oil and Gas J.
68:82-99.
258. Schaffer, J.D. 1984. Some experiments in machine learning using vector evaluated
genetic algorithms. Ph.D. Thesis, Vanderbilt University, Nashville, TN.
259. Scherzer, J. 1990. Octane-enhancing, zeolitic FCC catalysts: scientific and technical
aspects / Julius Scherzer M. Dekker, New York.
260. Secchi, A.R., M.G. Santos, G.A. Neumann, and J.O. Trierweiler. 2001. A dynamic model
for a FCC UOP stacked converter unit, Computers and Chemical Engineering, 25:851858.
261. Sekia, Hiroya, Morimasa Ogawab, Satoshi Ooyamab, Kouji Akamatsub, Masahiro
Ohshimac, and Wang Yang. 2001. Industrial application of a nonlinear model predictive
control to polymerization reactors, Control Engineering Practice. 9:819-828.
262. Semino, D. and C. Scali. 1994. A method for robust tuning of linear quadratic optimal
controllers. Ind. Eng. Chem. Res. 33:889-895.
263. Serti-Bionda, K., Z. Gomzi, and M. Muzic. 2010. Modeling of Gas Oil catalytic cracking
in a fixed bed reactor using a five-lump kinetic model, Chemical Engineering
Communications 197(3):275-288.
264. Shah, Y.T., G.T. Huling, J.A. Paraskos, and J.D. McKinney. 1977. A kinematic model for
an adiabatic transfer line catalytic cracking reactor. Ind Eng Chem Process Des Dev.
16(1):89-94.
265. Shinskey F.G. 1979. Process, Control Systems, 2nd ed, McGraw Hill, New York.
236
References
266. Shinskey F.G. 1984. Distillation Control, 2nd ed, McGraw Hill, New York.
267. Shouche Manoj S., Hasmet Genceli, and Michael Nikolaou. 2002. Effect of on-line
optimization techniques on model predictive control and identification (MPCI),
Computers and Chemical Engineering.26:1241-1252.
268. Shridar, R. and D.J. Cooper. 1997. A tuning strategy for unconstrained SISO model
predictive control, Ind. Eng. Chem. Res.36:729-746.
269. Silverman, B.W. 1986. Density estimations for statistics and data analysis, Mnonogr. Stat.
Appl. Prob. 26:175. Chapman and Hall, London.
270. Silverman, L.D., W.S. Winkler, J.A. Tiethof, A. Witoshkin. 1986. NPRA Annual
Meeting, Los Angeles, March 23-26; AM-86-62.
271. Skogestad, S. and M. Morari. 1987. Implications of large RGA elements on control
performance, Ind. Eng. Chem. Res.26, 2323-2330.
272. Skogestad, S., and E.A. Wolff. 1992. Controllability Measures for Disturbance Rejection.
IFAC Workshop on Interactions between Process Design and Process Control, Pergamon
Press, London, 23-29 (1992).
273. Skogestad, S., and M. Morari. 1987a. Effect of Disturbance Direction on Closed Loop
Performance. Ind. Eng. Chem. Res. 26:2029.
274. Skogestad, S., and M. Morari. 1987b. Implications of Large RGA Elements on Control
Performance. Ind. Eng. Chem. Res. 26(11):2323.
275. Srinivas, N., and K. Deb. 1994. Multi-objective optimization using nondominated sorting
in genetic algorithms. Evolutionary Computation 2:221.
276. Surmann, H., A. Kanstein and K.Goser. 1993. Self-organizing and genetic algorithms for
an automatic design of fuzzy control and decision systems. In Proceedings of First
European Congress on Intelligent Techniques and Soft Computing, 2:1097-1104.
277. Taguchi, H. and M. Araki. 2000. Two-Degree-off freedom PID controllers - Their
functions and optimal tuning. In IFAC Digital Control: Past, Present and Future of PID
Control. Terrassa, Spain.
278. Takatsuka, T., S. Sato, Y. Morimoto, and H. Hashimoto. 1987. Int. Chem. Eng. 27:107.
237
References
279. Takatsuka, T., S. Sato, Y. Morimoto and H. Hashimoto. 1987. A reaction model for
fluidized-bed catalytic cracking of residual oil. Int Chem Eng. 27:107-116.
280. Tasoti, V. 2007. Enhancement of environmental security by mathematical modelling and
simulations of processes, NATO Security through Science Series, 47-56.
281. Tominaga, H., and M. Tamaki. 1997. Chemical reaction and reactor design. J. Wiley. Chic
ester, England; New York.
282. Tsang, T., D. Himmelblau, and T. Edgar. 1975. Optimal control via collocation and nonlinear programming. Int. J. Contr. 21:763-768.
283. Umesh K. Chitnis and Armando B. Corripio. 1998. On-line optimization of a model IV
fluid catalytic cracking unit, ISA Transactions 37: 215-226, USA.
284. Van Brempt, Wim Ton Backx, Jobert Ludlage, Peter Van Overschee, Bart De Moor, and
R. Tousain. 2001. A high performance model predictive controller: application on a
polyethylene gas phase reactor. Control Engineering Practice 9:829-835.
285. Wang, D. and J.A. Romagnoli. 2003. Robust model predictive control design using a
generalized objective function. Comp. & Chem. Eng. 27:965-982.
286. Weekman, V. and D. Nace. 1970. Kinetics of catalytic cracking selectivity in fixed,
moving and fluid bed reactorsA. IChEJ. 16:397-404.
287. Weekman, V.W. and D.M. Nace. 1970. Kinetics of catalytic cracking selectivity in fixed,
moving and fluid bed reactors, AIChE J. 16: 397-404.
288. Witcher, M.F. and T.J. McAvoy. 1977. Interacting control systems - Steady state and
dynamic measurement of interaction, ISA Transactions 16:35-41.
289. Wolff, E. A., S. Skogestad, M. Hovd, and K. W. Mathisen. 1992. A Procedure for
Controllability Analysis, IFAC Workshop on Interactions between Process Design and
Process Control, Pergamon Press, London, 127-132.
290. Wrth, L., J. B. Rawlings, and W. Marquardt. 2009. Economic dynamic real-time
optimization and nonlinear model predictive control on innite hoirzons, in International
Symposium on Advanced Control of Chemical Process, Istanbul, Turkey.
291. Wrth, L., J.B. Rawlings, and W. Marquardt. 2009. Economic dynamic real-time
optimization and nonlinear model predictive control on infinite hoirzons, in: Proceedings
238
References
of the International Symposium on Advanced Control of Chemical Process, Istanbul,

Turkey, 2009.
292. Xu, Ch., J. Gao, S. Zhao, and Sh. Lin. 2005. Correlation between feedstock SARA
components and FCC product yields. Fuel 84:669-674.
293. Yang, S., X. Wang, C. McGreavy. 1996. A multivariable coordinated control system
based on predictive control strategy for FCC reactor-regenerator system, Chem. Eng.
Sci.51:2977-2982.
294. Yen, L. C., R. E. Wrench, and A. S. Ong. 1988. Reaction Kinetic Correlation Equation
Predicts Fluid Catalytic Cracking Coke Yields. Oil & Gas J. 86:67.
295. Yu, C.C., and W.L. Luyben. 1987. Robustness with respect to integral controllability, Ind.
Eng. Chem. Res. 26:1043-1045.
296. Zafiriou, E., and M. Morari. 1986. Design of robust digital controllers and sampling-time
selection for SISO systems, Int. J. Control 44(3):711-735.
297. Zanin, A.C., and M. Tvrzsk de Gouva, and D. Odloak. 2002. Integrating real-time
optimization into the model predictive controller of the FCC system. Control Engineering
Practice 10:919-931.
298. Zanin, A.C., M.T. Gouva, and D. Odloak. 2002. Integrating real-time optimization into
the model predictive controller of the FCC system. control engineering practice 10:819.
299. Zanin, A., M. Gouvea, and D. Odloak. 2000. Integrating real-time optimization into the
model predictive controller of the FCC system. Cont. Eng. Practice 10:819-831.
300. Zavala, V. M., E. M. Constantinescu, T. Krause, and M. Anitescu. 2009. On-line
economic optimization of energy systems using weather forecast information. Journal of
Process Control 19:1725-1736.
301. Zavala, V. 2008. Computational Strategies for the Optimal Operation of Large-Scale
Chemical Processes. Ph.D. thesis, Carnegie Mellon University.
302. Zhang, W.D., Y.X. Sun, and X. M. Xu. 1996. Modified PID controller based on H infinity
theory. IEEE Int. Conf. on Ind. Tech. 9-12.
239
Appendix
Appendix
The following MATLAB codes for main model which contains the main FCC Unit (7) parts with 5lump riser model.
function [sys,
X0]=fccmodcol2(t,X,u,flag,disturban
ce);
persistent ft xO2sg xCOsg xCO2sg
fsc frgc cO2sg cCOsg yhcofinal
ylcofinal yglnfinal
%
Constants
Qfr
dtstrp
k12
patm
dpfrac
astr
eloi
estrp
feed system constants
f1set = 13.8;
f2set = 0.0;
%f3set = 126.0;
%f4set = 5.25;
309;
10;
0.5;
15;%14.7;
9;
70.9;%60;
11.5;%124.5;
26.2;%130;
wet-gas compressor constants
pvru = 101.;
k11 = 1.5;
k13 = 0.01;
%
=
=
=
=
=
=
=
=
ubend constants
preheat system constants
%f5set
taufo
taufb
UAf
temph1
DHfu
a1
a2
=
=
=
=
=
=
=
=
roc
areaur
areaus
lur
lus
furgc
fusc
R
vslip
g
Asvsc
Asvrgc
34.0;
60.0;
200.0;
25.0;
460.9;
1000.0;
0.15;
200.0;
%
reactor and fractionator
constant
%psif
Aris
hris
rov
ropart
c1
c2
tref
mcpeff
cpc
tbase
cpfl
tbasef
t4
cpsv
cpsl
Qsr
cpfv
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
= 1.0;
12.3;%9.6
129.4;%60;
0.57;
68;
7.12e-3;%8.84e-3;
3.8e-5;%4.0e-5;
999;%999;
10000;
0.31;
959;
0.82;
700;
572;
0.80;
0.80;
412.0;
0.81;
%
ch
mh
cpc
Qe
areg
zlp
ztop
zcyc
dH1
dH2
cpH2O
cpN2
cpCO
cpCO2
cpO2
dHH
240
= 45;
= 6.7;%3.7;
= 8.2;%5.2;
= 44;%56;
= 30.2;%56;
= 17;
= 47;
= 10.73;
= 2.2;
= 32.2;
=930;%749;
= 930;%533;
Regenerator constans
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
0.075;
2;
0.31;
556;
706.5;%590;
11;
106;%52;%47;
103;%50;%45;
46368;
169080;
8.62;
7.22;
7.28;
11;
7.62;
60960;
Appendix
cpair
tair
mi
fof
zsp
hsp
asp
k14
alp
taufill
hlift
etap
elift
%
7.08;
416;
200000;
424;
13;
20;
7;
1.1;
8.73;
40;
34;
40.3;%155;
61;%134;
% Relative volatilities:
alpha=[alpha1 alpha2...alphaNC]
alpha=[2.0 1.5 1.0];
%F=0.33;
%
state variables
for i=1:82
x(i)=X(i);
end
for i=83:123
M(i)=X(i);
end
T2
= X(124);
T3
= X(125);
Tr
= X(126);
Csc = X(127);
Wr
= X(128);
P5
= X(129);
P7
= X(130);
Treg = X(131);
Crgc = X(132);
Wreg = X(133);
Wc
= X(134);
Wsp = X(135);
P6
= X(136);
n
= X(137);
P2
= X(138);
vsc = X(139);
vrgc = X(140);
P1
= X(141);
fcol = X(142);
= 250;
= 25;
= 10;
= 10;
= 40;
= 75;
= 190;
= 40;
= 1000;
= 1.39;
= 1;
= 5000;
= 6100;
= 5950;
= 5;
= 225;
= 200;
= 0.01;
= 0.005;
=
=
=
=
=
=
=
=
=
Algebraic equations
if abs(flag) == 1
airblowers
k6
k7
%k8
k9
kca
tatm
tcombd
vcombs
vcombd
kavg
polef
samin
samax
sb
klift
tliftd
vliftd
kcla
kila
V6
V7
V8
V9
V11
V12
V13
Vlift
dpset
=
=
=
=
=
=
=
=
=
=
=
=
=
1.0;
0.0;
0.0;
0.0;
0.95;
0;
0.0;
0.42;
-3.37;
%intrari
f3
f4
svsc
svrgc
LT
fair_c
f5set
psif
%coloana
NT=41;
% Location of feed stage (stages
are counted from the bottom):
NF=21;
% Number of components
NC=3;
VB
V14
=
=
=
=
=
=
=
=
u(1);
u(2);
u(3);
u(4);
u(5);
u(6);
u(7);
u(8);
= 0.4;
= 0.612;
%disturbances
%
%
241
if t>30000
dpfrac=10.45; %13; %10; %9.75;
Appendix
% end
%
%
if t>=10000
%
psif = 0.5;
%
end
%
%if t>=300 & t<=3900
% tatm=tatm+30/28800*(t300);
%elseif t>3900
% tatm=tatm+30*3600/28800;
%end
epsf
= 0.332+0.06*vs;
rocdens = ropart*(1-epsf);
zbed
=
min(zcyc,(2.85+0.8*vs+(Wregrocdil*areg*zcyc)/...
(areg*rocdens))*1/(1rocdil/rocdens));
pblp
= P6;
me
= areg*vs*rocdil;
roag
= 29*P6/R/(Trdtstrp+459.6);
%
2.2
feed system equations
f1 = f1set;
f2 = f2set;
%f3 = f3set;
%f4 = f4set;
%
2.3
f5
ubend equations
computing fsc
fsc
= vsc*areaus*roc;
Msc
= (Wr +
areaus*lus*roc)/g;
deltapsc
= 144*(p4-pblp) +
Wr/astr+(estrp-elift)*roc;
deltapsvsc =
((50*0.453*60/0.7/900)^2)*((fsc/svs
c/Asvsc)^2)/roc;
fmicsc
= (deltapsc 144*deltapsvsc)*areaus vsc*lus*fusc;
Preheat system computing
= f5set;
dtin = T3-temph1;
dtout = T3-T2;
tln
= (dtindtout)/(log(dtin/dtout));
Qloss = a1*f5*T3-a2;
%
2.5
computing frgc
frgc
= vrgc*areaur*roc;
2.4 General equations

vris = (f3+f4)/rov+frgc/ropart;
roris = (f3+f4+frgc)/vris;
prb
= P5+dpfrac+roris*hris/144;
% regenerator computing used in

ubend and other
prgb = P6 + Wreg/144/areg;
f7
= kca*sqrt(max(0,P2-prgb));
rog =
520*P6/(379*14.7*(Treg+459.6));
Mrgc
= (max(0,Wsp) +
areaur*lur*roc)/g;
deltaprgc
= 144*(P6prb)+(max(0,Wsp))/asp + (etapeloi)*roc;
deltapsvrgc =
((50*0.453*60/0.7/900)^2)*((frgc/As
vrgc/svrgc)^2)/roc;
fmicrgc
= (deltaprgc 144*deltapsvrgc)*areaur vrgc*lur*furgc;
p4
= P5+dpfrac;
dprr = P6-p4;
fsg = k14*V14*sqrt(max(0,P6-patm));
% regenerator equations using total
air flowrate
% ft
% fair
ft
fair
vs
rocdil
=
=
=
=
= f7;
= f7/29;
f7 + fair_c;
ft/29;
0.5*(fsg+fair)/rog/areg;
-0.878+0.582*vs;
%
2.6 wet-gas compressor
equations
crw
hwg
242
= pvru/P7;
= 182922.1*(crw^(0.0942)-1);
Appendix
fsucwg = 11600+sqrt(max(0,1.366e+80.1057*hwg^2));
f11
= 2.636e-6*fsucwg*P7;
fV13
= k13*V13*pvru;
ycoke0=0;
ficoke0=1;
teta0=0;
step=0.01;
2.7 airblowers equations

No_of_elemnts = fix(tc/step);
if V6 > 0.5
fpp6 = exp(2*log(0.15)*(1V6));
else
fpp6 = 0.3*V6;
end
if V7 > 0.5
fpp7 = exp(2*log(0.15)*(1V7));
else
fpp7 = 0.3*V7;
end
yhco
ylco
ygln
ygas
ycoke
ficoke
=
=
=
=
=
=
zeros(1,No_of_elemnts);
yhco(1)=yhco0;
ylco(1)=ylco0;
ygln(1)=ygln0;
ygas(1)=ygas0;
ycoke(1)=ycoke0;
ficoke(1)=ficoke0;
Pbase = 14.7*P2/P1;
Fsucn_comb =
45000+sqrt(max(0,1.581e+91.249e+6*(Pbase)^2));
f6
=
0.0451*P1*Fsucn_comb/(tatm+460);
fV7 = k7*fpp7*sqrt(max(0,P2patm));
fV6 = k6*fpp6*sqrt(max(0,patmP1));
t=0;
timp(1) = 0;
cont = 1;
while t <= tc
cc=0.68*(1-exp(-30*t));
ficonv=exp(-(kd*cc));
dyhco=((-(k1+k2+k3)*(yhco(cont)xhco))*ficonv-(k7r*(yhco(cont)xhco))*ficoke(cont))*CTO;
dylco=((k1*(yhco(cont)-xhco)k4*ylco(cont))*ficonv(k8*ylco(cont))*ficoke(cont))*CTO;
dygln=(k2*(yhco(cont)xhco))*ficonv*CTO;
dygas=(k3*(yhco(cont)xhco)+k4*ylco(cont))*ficonv*CTO;
dycoke=(k7r*(yhco(cont)xhco)+k8*ylco(cont))*ficoke(cont)*C
TO;
dficoke=-alfa*ficoke(cont);
%reactorul cu riser cu 5 componenti

COR=frgc/(f3+f4);
tc=Aris*hris/vris;
k1=1.54;
k2=5;
k3=0.55;
khco=7.09;
k4=0.35;
k5=0;
k6=0;
k7r=0.25;
k8=0;
kd=6.25;
alfa=18.5;
xhco=0.19;
yhco(cont+1)=yhco(cont)+step*dyhco;
ylco(cont+1)=ylco(cont)+step*dylco;
ygln(cont+1)=ygln(cont)+step*dygln;
ygas(cont+1)=ygas(cont)+step*dygas;
ycoke(cont+1)=ycoke(cont)+step*dyco
ke;
ficoke(cont+1)=ficoke(cont)+step*df
icoke;
CTO=COR;
yhco0=0.95;
ylco0=0.035;
ygln0=0.01;
ygas0=0;
t=t+step;
243
Appendix
timp(cont+1) = t;
cont = cont + 1;
end
fV11 = k11*fppwg*sqrt(max(0,P5P7));
fH
= fsc*(Csc-Crgc)*ch;
epse =
min(1,max(epsf,epsf+(1.904+0.363*vs
-0.048*vs^2)/zbed));
yhcofinal=yhco(end);
ylcofinal=ylco(end);
yglnfinal=ygln(end);
ygasfinal=ygas(end);
ycokefinal=ycoke(end);
ficokefinal=ficoke(end);
timp=timp(end);
frhco=yhcofinal/(yhcofinal+ylcofina
l+yglnfinal+ygasfinal);
frlco=ylcofinal/(yhcofinal+ylcofina
frgas=ygasfinal/(yhcofinal+ylcofina
frglngas=(yglnfinal+ygasfinal)/(yhc
ofinal+ylcofinal+yglnfinal+ygasfina
l);
frtotal= frhco+frlco+frglngas;
quf=frhco+frlco+frglngas-frgas;
zF=[frglngas frlco frhco];
z
= 0;
treg = Treg;
xCO = 0;
xCO2 = 0;
%xO2 = 1/fair*(0.21*fair-0.25*fH);
xO2 = max(0,1/fair*(0.21*fair0.25*fH));%% s-a modificat de mine
stepz = 0.2;
while z < ztop
if z<zcyc
deltaz = 1;
else
deltaz = 0;
end
cpz =
0.79*cpN2+xCO*cpCO+xCO2*cpCO2+xO2*c
pO2+...
fb
= (psif*f3 + 3*f4 + 2*f1 0.8*f2)/(f3 + f4);
wris
= frgc*Aris*hris/vris;
whsv
= 3600*(f3+f4)/wris;
taur
= wris/60/frgc;
fcoke
=
(1.3557*(f3+f4)*fb*taur^(1.98431))/(100*whsv);
fwg
= (f3+f4)*(c1+c2*(Trtref));
Qrgc
= frgc*cpc*(Treg-tbase);
Qrin
= Qrgc+f3*cpfl*(T2-tbasef);
Qcatout = frgc*cpc*(Tr-tbase);
Qslurry = f4*(cpsv*(Tr-tref)+Qsr);
Qff
= f3*(cpfv*(Tr-tref)+Qfr);
dhcrack = 172.7+3*(Tr-tref);
Qcracking = (f3+f4)*dhcrack;
Qrout
=
Qcatout+Qslurry+Qcracking+Qff;
Qr
= Qrin-Qrout;
tsc
= Tr-dtstrp;
fV12 = k12*V12*sqrt(P5-patm);
(0.25*cpH2O*fH+deltaz*cpc*me)/fair;
k1 = 6.9547*exp(19.8834000/(treg+459.6));
k2 = 0.69148*exp(15.0625000/(treg+459.6));
k3 = 0.6412*P6*exp(25.5545000/(treg+459.6));
%
if z<=zbed
rob = 1-epse;
else
rob = (1-epse)*exp(1000*fair/(areg*vs*rocdil)*(zzbed));
end
dxO2z = (100*(-0.5*k1k2)*rob*Crgc-k3*xCO)*xO2/vs;
dxCOz = (100*k1*rob*Crgc2*k3*xCO)*xO2/vs;
dxCO2z = -dxO2z - 0.5*dxCOz;
if z <= zbed
dtregz = 0;
else
dtregz =
(dH1*dxCOz+dH2*dxCO2z)/cpz;
if V11 <= 0.5

fppwg = 0.3*V11;
else
fppwg =
exp(2*log(0.15)*(1-V11));
end
244
Appendix
end
xO2
xCO
xCO2
treg
=
=
=
=
MDs=0.1; MBs=0.1;
% Nominal holdups - these are
rather small
Ds=0.165; Bs=0.165;
% Nominal flows
MB=X((NC-1)*NT+1);
MD=X(NC*NT) ;
% Actual reboiler
and condenser holdup
D=Ds+(MD-MDs)*KcD ;
% Distillate
flow
B=Bs+(MB-MBs)*KcB ;
%
Bottoms flow
x=X(1:(NC-1)*NT);
% Liquid compositions from btm to
top
M=X((NC-1)*NT+1:NC*NT);
% Liquid hold up from btm to top
xO2+dxO2z*stepz;
xCO+dxCOz*stepz;
xCO2+dxCO2z*stepz;
treg+dtregz*stepz;
if xCO < 0
xCO = 0;
end
if xO2 < 0
xO2 = 0;
end
if xCO2 < 0
xCO2 = 0;
end
if z > zcyc
rob = 0;
tcyc = treg;
end
z = z + stepz;
% Rearrange elements of composition

vector (x) for later use
Iu=[1:NT]'*ones(1,NC1)+NT*ones(NT,1)*[0:NC-2];
x=(x(Iu))';
% THE MODEL
end
xO2sg
xCOsg
xCO2sg
= xO2;
= xCO;
= xCO2;
% Vapour-liquid equilibria
(multicomponent ideal VLE,
Stichlmair-Fair, 'Distillation', p.
36, 1998)
y=(alpha(1:NC1)'*ones(1,NT).*x)./(ones(1,NC1)'*(1+(alpha(1:NC-1)-1)*x));
% Vapor Flows assuming constant
molar flows
i=1:NT-1;
V(i)=VB*ones(1,NT-1);
i=NF:NT-1;
V(i)=V(i) + (1quf)*fcol*0.4535/178;
% Liquid flows are given by
Franci's Weir Formula
L(i)=K*Mow(i)^1.5
% Liquid flow L(i) dependent only
on the holdup over the weir Mow(i)
% M(i)= Mow(i) + Muw(i) (Total
holdup = holdup over weir + holdup
below weir)
Kuf=21.65032/60;
% Constant above feed
Kbf=29.65032/60;
% Constant below feed
%Kuf=19/60;
%pt. dpfrac=9.0
% Constant above feed
%Kbf=27/60;
%pt. dpfrac=9.0
% Constant below feed
Muw=0.1;
% Liquid holdup under weir (Kmol)
Qfg
=
(fair*(xO2sg*cpO2+xCOsg*cpCO+xCO2sg
*cpCO2+0.79*cpN2)+0.5*cpH2O*fH)*(tc
yc-tbase);
Qsc
= fsc*cpc*(tsc-tbase);
Qc
=
fair*(xCOsg*dH1+xCO2sg*dH2);
Qh
= fH*dHH;
Qair
= fair*cpair*(tair-tbase);
Qoutreg = Qfg+Qrgc+Qe;
Qinreg = Qair+Qh+Qc+Qsc;
Qreg
= Qinreg-Qoutreg;
lsp
= max(0,Wsp)/roc/asp;
d
= min(3,hsp-lsp);
fsp
=
max(0,fof*sqrt(asp)*(zbed-zsp)4925-20*(3-d)); % - 27;
cO2sg
= 100*fair*xO2sg/fsg;
cCOsg
=
1e+6*28*xCOsg/(28*xCOsg+44*xCO2sg+3
2*xO2sg+22.12);
vregg
= areg*zcyc-areg*zbed*(1epse);
%fcol= f3+f4+frgc-fsc-fwg;
%coloana
KcB=10; KcD=10;
% controller gains
245
Appendix
i=2:NF;
L(i)= Kbf*(M(i)Muw).^1.5;
% Liquid flows below
feed (Kmol/min)
i=NF+1:NT-1; L(i)= Kuf*(M(i)Muw).^1.5;
% Liquid flows above
feed (Kmol/min)
L(NT)=LT;
% Condenser's liquid flow
(Kmol/min)
% Time derivatives from material
balances for
% 1) total holdup and 2) component
holdup
dxdt=(dMxdt-x.*(ones(NC1,1)*dMdt))./(ones(NC-1,1)*M');
% Rearrange elements of composition
vector (dxdt) for later use
Ix=[1:(NC-1)*NT]';
W=dxdt';
dxdt=W(Ix);
dT2
= (temph1+UAf*tln/f3T2)/taufo;
dT3
= (f5*DHfu-UAf*tlnQloss)/taufb;
dTr
= (Qrin-Qrout)/mcpeff;
dCsc = (frgc*Crgc+fcoke-fsc*CscCsc*(frgc-fsc))/Wr;
dWr
= frgc-fsc;
dP5
= 0.833*(fwg-fV11-fV12+fV13);
dP7
= 5*(fV11-f11);
dTreg = (QinregQoutreg)/((Wreg+max(0,Wsp))*cpc+mi)
;
dCrgc = (fsc*Csc-fH(max(0,fsp)*Crgc+12*fair*(xCOsg+xCO
2sg))-Crgc*...
(fsc-max(0,fsp)))/Wreg;
dWreg = fsc-max(0,fsp);
dWc
= fsc*Csc-fH(max(0,fsp)*Crgc+12*fair*(xCOsg+xCO
2sg));
dWsp = max(0,fsp)-frgc;
dP6
= (R*(n*(QinregQoutreg)/((Wreg+max(0,Wsp))*cpc...
+mi)+(Treg+460)*(fair-fsg)))/vregg;
dn
= fair-fsg;
dP2
=
R*(tcombd+460)/29/vcombd*(f6-fV7f7);
dvsc = fmicsc/Msc;
dvrgc = fmicrgc/Mrgc;
dP1
= R*(tatm + 460)*(fV6f6)/(29*vcombs);
dfcol = f3+f4+frgc-fsc-fcol;
%sys = [dxdt;dMdt'];
sys =
[dxdt;dMdt';dT2;dT3;dTr;dCsc;dWr;dP
5;dP7;dTreg;dCrgc;dWreg;...
dWc;dWsp;dP6;dn;dP2;dvsc;dvrgc;dP1;
dfcol];
% Column
j=2:NT-1;
dMdt(j) = L(j+1)
- L(j)
+ V(j-1)
- V(j);
for i=1:NC-1;
for j=2:NT-1;
dMxdt(i,j) = L(j+1)*x(i,j+1) L(j)*x(i,j)
+ V(j-1)*y(i,j-1)
- V(j)*y(i,j);
end
end
% Correction for feed stage
% The feed is assumed to be mixed
into the feed stage
dMdt(NF) = dMdt(NF)
fcol*0.4535/178;
dMxdt(:,NF)=dMxdt(:,NF)+(fcol*0.453
5/178)*zF(1:NC-1)';
% Reboiler (assumed to be an
equilibrium stage)
dMdt(1) = L(2) - V(1)
- B;
i=1:NC-1;
dMxdt(i,1)= L(2)*x(i,2) V(1)*y(i,1) - B*x(i,1);
% Total condenser (no equilibrium
stage)
dMdt(NT) = V(NT-1) - LT - D;
i=1:NC-1;
dMxdt(i,NT)= V(NT-1)*y(i,NT-1) LT*x(i,NT) - D*x(i,NT);
%i=1:123;
%X =
[x;M;T2;T3;Tr;Csc;Wr;P5;P7;Treg;Crg
c;Wreg;Wc;Wsp;P6;n;P2;vsc;vrgc;P1;f
col];
% Compute the derivative for the

mole fractions from d(Mx) = x dM +
M dx
246
Appendix
elseif flag == 0
sys = [142 0 142+11 8 0 0];
9.787644439610779e-002
9.514324975575179e-002
9.321328685875070e-002
9.185882971957438e-002
9.091210762040128e-002
9.025217387431267e-002
8.979299824849162e-002
8.947390960606379e-002
8.925236108228883e-002
8.909862784488616e-002
8.899199577167828e-002
8.891805499293592e-002
8.886679315082846e-002
9.271398485781029e-002
9.511705265421500e-002
9.620994638509704e-002
9.617527103715355e-002
9.519883982471469e-002
9.344428455716700e-002
9.104331179342076e-002
8.809543157234483e-002
8.467221514355509e-002
8.082300408083755e-002
7.658048860952456e-002
7.196550956157456e-002
6.699093505992447e-002
6.166468549993386e-002
5.599205504002732e-002
4.997748248122757e-002
4.362590131499029e-002
3.694376848976960e-002
2.993984307260207e-002
2.262576268252438e-002
9.284641325253319e-002
1.099044817624081e+000
1.099044817624081e+000
1.099044817624082e+000
1.099044817624082e+000
1.099044817624083e+000
1.099044817624083e+000
1.099044817624084e+000
1.099044817624084e+000
1.099044817624084e+000
1.099044817624084e+000
1.099044817624084e+000
1.099044817624084e+000
1.099044817624083e+000
1.099044817624083e+000
1.099044817624082e+000
1.099044817624082e+000
1.099044817624081e+000
1.099044817624081e+000
1.099044817624080e+000
1.099044817624080e+000
1.023048680803940e+000
X0 = [
4.102805762039235e-001
5.250768045383215e-001
6.090295678210853e-001
6.647184711852767e-001
6.999136517544166e-001
7.218309205578514e-001
7.355425780052245e-001
7.442398214671327e-001
7.498489420584803e-001
7.535247153536314e-001
7.559668699297101e-001
7.576074667588276e-001
7.587190211488416e-001
7.594769446441061e-001
7.599961569859209e-001
7.603530400594605e-001
7.605989356947257e-001
7.607686491186187e-001
7.608859237989325e-001
7.609670311072776e-001
7.610231583520370e-001
7.935100634028763e-001
8.202657175072821e-001
8.418073423207758e-001
8.589595228917157e-001
8.726250679955578e-001
8.836454796758728e-001
8.927370216494378e-001
9.004757439591793e-001
9.073081306860253e-001
9.135718798231480e-001
9.195182605182517e-001
9.253321212605720e-001
9.311482385285077e-001
9.370639780922759e-001
9.431487644117131e-001
9.494509921460599e-001
9.560029811652185e-001
9.628244826189277e-001
9.699251398615792e-001
9.773062150484678e-001
1.530244978499652e-001
1.541263538337487e-001
1.459528257717734e-001
1.344903117907614e-001
1.233999670224613e-001
1.141560945015914e-001
1.070097723447256e-001
1.017107298678100e-001
247
Appendix
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
1.023048680803940e+000
9.529993506207503e-002
7.826027736455383e+002
1.659564334971949e+003
9.932070027620795e+002
3.901281291711348e-003
2.225473104525494e+005
1.579229601661189e+001
1.557926799156491e+001
1.064425988191907e+003
1.952728693846411e-003
3.927528423494652e+005
1.845380965498859e+003
3.243834313719916e+003
2.786505547810026e+001
2.566492705837050e+002
3.479018022391441e+001
4.055562514970572e+000
4.963524272053554e+000
1.492154586174961e+001
8.300000000000543e+001
P5
= X(129);
P7
= X(130);
Treg = X(131);
Crgc = X(132);
Wreg = X(133);
Wc
= X(134);
Wsp = X(135);
P6
= X(136);
n
= X(137);
P2
= X(138);
vsc = X(139);
vrgc = X(140);
P1
= X(141);
fcol = X(142);
p4
= P5+dpfrac;
dprr = P6-p4;
if t==0
ft
=7.002416469083545e+001;
xO2sg = 9.426727601699798e-002;
xCOsg = 4.259398111910952e-008;
xCO2sg = 9.308933947905657e002;
fsc = 1.496502568024175e+003;
frgc = 1.496502568024172e+003;
cO2sg = 9.426727601699826e+000;
cCOsg = 4.079815564345754e-002;
yhcofinal= 1.899999980733520e001;
ylcofinal = 7.899168031516896e002;
yglnfinal = 5.383451451188887e001;
frglngas = 0.56952817544156;
frlco = 0.17262037126428;
frhco = 0.25785145329416;
quf = 0.93122462693013;
ygasfinal = 6.607188611783442e-
];
02;
ycokefinal = 3.430893910662157e-
elseif flag == 3
02;
for i=1:82
x(i)=X(i);
end
for i=83:123
M(i)=X(i);
end
T2
= X(124);
T3
= X(125);
Tr
= X(126);
Csc = X(127);
Wr
= X(128);
end
sys=[X];
sys =
[X;ft;xO2sg;xCOsg;xCO2sg;fsc;frgc;c
O2sg;cCOsg;yhcofinal;ylcofinal;ygln
final];
else
sys = [];
end
248
List of Publications
LIST OF PUBLICATIONS
Conference papers:
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2008. Process simulators based control design for a fluid
catalytic cracking unit. 18th International Congress of Chemical and Process Engineering, Prague, Czech
Republic.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Process simulators based control design for universal oil
products FCCU. Chemical Engineering, 8th World Congress of Chemical Engineering (WCCE 8),
Montral, Canada.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Multivariate control structure synthesis for fluid catalytic
unit Using Relative Gain Array. Chemical Engineering, D.B. Das, V. Nassehi, L. Deka, EUROSIS-ETI
2009, 7th International Industrial Simulation Conference, Quality Hotel, Loughborough, 251-256, ISBN
978-90-77381-4-89.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Mathematical modelling and control of a UOP fluid
catalytic cracking unit. Chemical Engineering, Saudi Students Club & Schools in UK and Ireland, 3 rd
Saudi International Conference, University of Surrey, UK, 32.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2010. Linear model predictive control of a fluid catalytic
cracking unit. Chemical Engineering, Saudi Students Club & Schools in UK and Ireland, 4 th Saudi
international conference, University of Manchester, UK.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2010. Model based control of a UOP fluid catalytic cracking
unit. Chemical Engineering Conference, Loughborough University, UK.
Conference Presentations:
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2008. Process Simulators Based Control Design For a Fluid
Catalytic Cracking Unit. 18th International Congress of Chemical and Process Engineering, Prague, Czech
Republic.
249
List of Publications
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Process simulators based control design for universal oil
products FCCU. Chemical Engineering, 8th World Congress of Chemical Engineering (WCCE 8),
Montral, Canada.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Multivariate control structure synthesis for fluid catalytic
unit using relative gain array. Chemical Engineering, D.B. Das, V. Nassehi, L. Deka, EUROSIS-ETI
2009, 7th International Industrial Simulation Conference, Quality Hotel, Loughborough, 251-256, ISBN
978-90-77381-4-89.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2009. Mathematical modelling and control of a UOP fluid
catalytic cracking unit. Chemical Engineering, Saudi Students Club & Schools in UK and Ireland, 3rd
Saudi International Conference, University of Surrey, UK, 32.
Poster Presentations:
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2010. Linear model predictive control of a fluid catalytic
cracking unit. Chemical Engineering, Saudi Students Club & Schools in UK and Ireland, 4th Saudi
international conference, University of Manchester, UK.
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2010. Model based control of a UOP fluid catalytic cracking
unit. Chemical Engineering Conference, Loughborough University, UK.
Journal Paper:
Alsabei, R., Z.K. Nagy, and V. Nassehi. 2011. Economic optimization based model predictive control of a
fluid catalytic cracking unit. Chemical Engineering and Processing (under review).
Awards and Prizes:
Gold poster award in Engineering at 4th Saudi International Conference, University of Manchester, UK,
2010.
Award of excellence from Saudi Cultural Bureau in London, UK, 2011.
250

10 1 1 454 702

Uploaded by

Copyright:

Available Formats

10 1 1 454 702

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

10 1 1 454 702

Uploaded by

Copyright:

Available Formats

What is the thesis about?

What is the thesis about?

What are the main sections of the FCC unit modelled?

What are the main sections of the FCC unit modelled?

Loughborough University

Metadata Record: https://dspace.lboro.ac.uk/2134/8566

This item was submitted to Loughboroughs Institutional Repository

For the full text of this licence, please go to:

MODEL BASED APPROACH FOR THE PLANTWIDE ECONOMIC CONTROL OF FLUID

REDAH MOUSA ALSABEI

A Doctoral Thesis submitted in partial fulfilment of the requirements

Redah Alsabei (2011)

Research aim and objectives ............................................................................................... 4

Main contribution of the thesis ........................................................................................... 5

Thesis structure ................................................................................................................... 6

CHAPTER 2: LITERATURE REVIEW ............................................................................................ 8

Fundamentals of crude oil processing ................................................................................. 8

Mathematical modelling concept ...................................................................................... 18

Process technology .................................................................................................... 10

FCC cracking kinetics modelling .............................................................................. 20

Introduction to advanced chemical process control .......................................................... 28

Why advanced process control? ................................................................................ 28

Concepts of process control ...................................................................................... 30

History of automatic control ..................................................................................... 31

Economical importance of process control ............................................................... 34

What is modern control theory? ............................................................................ 36

Advanced process control techniques ....................................................................... 37

Historical development of LMPC ............................................................................. 40

Industrial MPC development .................................................................................... 42

Interaction analysis and multiple single-loop designs....................................................... 56

General tuning guidelines of MPC ............................................................................ 54

The Relative Gain Array (RGA) ............................................................................... 56

Nonlinear Model Predictive Control (NMPC) and industrial applications ....................... 59

Generic NMPC problem formulation ........................................................................ 63

Industrial implementation of NMPC ......................................................................... 64

Challenges in industrial NMPC ................................................................................ 68

First principle (analytical) model based NMPC ........................................................ 72

Nonlinear model predictive control with guaranteed stability .................................. 74

Chapter 3: MATHEMATICAL MODELING OF A FLUID CATALYTIC PROCESS .................. 77

Analytical mathematical model of the FCCU ................................................................... 77

Feed system model .................................................................................................... 79

Preheat system model ................................................................................................ 80

The reactor stripper model ........................................................................................ 86

Regenerator model .................................................................................................... 88

Air blower model ...................................................................................................... 94

Model of catalyst circulation lines ............................................................................ 95

Main fractionator model ............................................................................................ 95

Wet gas compressor model ....................................................................................... 98

Conclusions ..................................................................................................................... 100

Considerations of process operation ............................................................................... 102

Reactor-regenerator section..................................................................................... 103

Implementations and simulation procedure .................................................................... 110

Conclusions ..................................................................................................................... 123

CHAPTER 5: ADVANCED PID CONTROL OF THE FCCU PROCESS ................................... 124

Introduction ..................................................................................................................... 124

General considerations for the control of the FCCU ............................................... 124

Objectives for regulatory control ............................................................................ 126

Advanced PID control of the FCCU ............................................................................... 134

Design of the PID control algorithm ....................................................................... 134