You are on page 1of 25

Chemical Engineering and Processing 42 (2003) 67 /91

www.elsevier.com/locate/cep

Simulation and model predictive control of a UOP fluid catalytic


cracking unit
Mircea V. Cristea a,*, Şerban P. Agachi a, Vasile Marinoiu b,1
a
Faculty of Chemistry and Chemical Engineering, ‘‘Babeş-Bolyai’’ University, 11 Arany Janos Street, 3400 Cluj-Napoca, Romania
b
Control and Computers Department, ‘‘Petrol-Gaze’’ University, 39 Bucuresti Blvd., 2000 Ploiesti, Romania

Received 14 March 2001; received in revised form 9 March 2002; accepted 9 March 2002

Abstract

Based on a newly developed mathematical model, the complex dynamic simulator of an industrial Universal Oil Products (UOP)
fluid catalytic cracking unit was used to implement the model predictive control (MPC) algorithm. The simulator revealed the
multivariable, nonlinear and strong interacting feature of the process. Combined with equipment and operating constraints they put
severe limits on control performance. Different MPC schemes for the reactor and regenerator’s most important process variables
were tested and the most favorable have been presented. The constrained MPC approach using scheduled linearization to account
for non-linear behavior and a larger number of manipulated than controlled variables proved successful. Comparison with
traditional control using decentralized PID controllers revealed incentives for the multivariable model based predictive control in
maintaining controlled variables very close to their constrained limits where usually the optimum is situated.
# 2002 Elsevier Science B.V. All rights reserved.

Keywords: FCCU dynamic simulator; Complex nonlinear behavior; PID control; Model predictive control

1. Introduction adiabatic plug flow reactor model is usually used for


the riser. Two zones frequently describe regenerator
Over 60 years catalytic cracking has been one of the model: a dense bed zone (with dense phase as a CSTR
main processes in petroleum refining supporting a model but gaseous phase as a plug flow reactor model)
spectacular development [1]. The fluid catalytic cracking and an entrained catalyst zone (plug flow model) [5].
unit (FCCU) became in the last decades the testing The control system design and implementation have
bench of every advanced control method. Both acade- to solve challenging tasks. The multivariable character
mia and industry are interested in developing new of the process presenting strong interactions, the non-
control algorithms and their efficient industrial FCC linear behavior leading to the need for nonlinear control
implementation, as successful results are usually of large and the demand to operate the unit in the presence of
economic benefits [2]. The catalytic cracking process is material and operating constraints, are the main ones.
complex both from the modeling and from the control Additionally, the control system has to cope with both
point of view [3,4]. large and short time constants and to face the changing
The dynamic mathematical model development im- operating conditions, in the presence of usually unmea-
plies some assumptions taking into account specific sured disturbances. As a consequence, model predictive
aspects of the process. The complex nature of the feed control (MPC) proves to be a good candidate for
oil assumes a lumped kinetic mechanism for the treat- implementing FCCU advanced control due to its multi-
ment of the cracking process. Both reactor and regen- variable structure, direct approach of constraints and
erator mass and heat transfer are complex. The optimal character [6,7].
Based on these preliminary aspects the paper presents
* Corresponding author. Fax: /40-64-193-833.
the development of a mathematical model for a UOP
E-mail address: mcristea@chem.ubbcluj.ro (M.V. Cristea). type FCCU and the associated dynamic simulator.
1
Fax: /40-44-175-847. Different MPC schemes are investigated and tested by
0255-2701/02/$ - see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 2 5 5 - 2 7 0 1 ( 0 2 ) 0 0 0 1 7 - X
68 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

dynamic simulation revealing interesting aspects from complex succeeding to capture the major dynamic
the perspective of the industrial implementation. behavior of UOP type FCCU [8]. The model includes
the main reactor/regenerator subsystems: feed and
preheat system, reactor, regenerator, air blower, wet
2. FCCU dynamic model gas compressor and catalyst circulation lines.
Main aspects of the new model are outlined in the
The FCCU, for which the mathematical model has following.
been developed and then performed the MPC study, is
presented in Fig. 1.
The newly developed mathematical model for the 2.1. Reactor model
UOP type FCCU is based on the mechanistic Amoco
Model IV FCCU [5]. Compared with Amoco Model IV Developing the new mathematical model for the
the new mathematical model describes a different reactor implied a thorough survey, selection and then
FCCU type, both from the operation and from the synthesis, based on a large variety of models presented
construction point of view. The main new model in literature. The three-lump model has been considered
characteristics are related to the following aspects: to be adequate for the global description of the
phenomena taking place in the reactor. Reactor is
. Different geometric dimensions and relative position
divided in two parts: riser and stripper. The riser model
define the reactor and regenerator, compared with
the Model IV case. is built on the following assumptions: ideal plug flow
. The reactor model uses a Weekman kinetic scheme [9] and very short transient time (the residence time in the
for describing the cracking process. riser is very short compared with other time constants,
. The regenerator of the UOP FCCU operates in especially with the regenerator time constants [1,5,8,10]).
partial combustion mode. It is modeled by mass balance describing the gasoline
. Catalyst circulation is described including spent and and coke/gases production based on Weekman’s
catalyst valves on catalyst circulation lines. These triangular kinetic model [9]. The mixed nonlinear
valves are used as main manipulated variables for differential and algebraic system of equations also
FCCU control. accounts for the amount of coke deposited on catalyst
and for the cracking temperature dynamics [13]. The
The FCCU dynamic model has been developed on the reactor is presented in Fig. 2.
basis of reference construction and operation data from Detailed description of the reactor model is presented
an industrial unit. The described model is rather in the following.

Fig. 1. Scheme of UOP type FCCU.


M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 69

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
tc Ecf
C cat K c exp : (8)
CrgcN RTr
The fraction of coke on the spent catalyst leaving the
riser is:
Csc1 Crgc Ccat : (9)
The constant values m * and N have been used to
perform a good fit of the mathematical model with
operating data from the industrial unit.

2.1.2. Heat balance for the riser

du DHf Ff dyf
 : (10)
dza T0 (Fsc Cpc  Ff Cpf  lFf Cpd ) dza

Fig. 2. Scheme of the FCCU reactor. The amount of gases produced by cracking is
described by the equation:
2.1.1. Mass balance for the riser Fwg (F3 F4 )[C1 C2 (Tr Tref )]: (11)
Mass balance for the feed is described by the equation
Constants C1and C2 have been fitted based on data
dy f from the industrial unit.
K1 y2f [COR]F tc : (1)
dz a The stripper model is of CSTR type (mass and heat
balance) evaluating the temperature in the stripper and
Mass balance for the gasoline is described by the the fraction of coke on spent catalyst.
equation
dyg
 (a2 K 1 y 2f K3 yg )[COR]F tc ; (2) 2.1.3. Mass and heat balance for the stripper
dza
where: dTs Frgc
 (Tr Ts ); (12)
dt Wr
Ef  
K1 (u) kr1 exp ; (3) dCsc dWr 1
RT0 (1  u)  Frgc (Crgc Ccat )Fsc Csc Csc ; (13)
dt dt Wr
Eg
K3 (u) kr3 exp ; u (T T0 )=T; (4) dWr
RT0 (1  u) Frgc Fsc : (14)
dt
F f0 exp(atc [COR]za ); (5)
f0 1mCrgc : (6)
Inlet temperature in the riser T0 is determined by the 2.1.4. Pressure balance for riser bottom pressure
heat balance equation [3] determination
Frgc Cpc Treg  Ff Cpf T2  DHevp Ff
T0  : (7) rris hris
Frgc Cpc  Ff Cpfv Prb P4  ; (15)
144
The term K 1y2f [COR]
represents the kinetics of the
feed, K3yg[COR] the kinetics of the gasoline; F is a F3  F4  Frgc
rris  ; (16)
function of catalyst deactivation due to coke deposition; nris
f0 the reduction of catalyst activity due to the coke
F3  F4 Frgc
resident on the catalyst after regeneration; tc residence nris   : (17)
time in the riser; and a2 /k1/k2 fraction of feed oil that rv rpart
cracks to gasoline. This model develops the models The amount of catalyst in the riser is determined by
presented by Lee and Groves [24], Shah et al. [25] and the equation
Hovd and Skogestad [13]. The amount of coke produced
is described by the following correlation taken from Frgc Aris hris
Wris  : (18)
Voorhies and Kurihara [26]: nris
70 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

2.1.5. Momentum balance for reactor and main The operating conditions are corresponding to CO
fractionator pressure determination partial combustion mode.
The regenerator model consists in mass and heat
dP5 balance equations for O2, CO, CO2 and coke, but also in
0:833(Fwg FV11 FV12 FV13 ) (19) heat balance equations for solid and gaseous phase.
dt
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi These balance equations are correlated with equations
FV12 k12 V12 P5 Patm : (20) describing entrained catalyst (bed characteristics) in the
zone above dense bed, catalyst flow and pressure in the
A constant pressure drop, DPfrac, between reactor and
regenerator.
main fractionator is considered; according to this, the
reactor pressure is computed by the equation
2.3. Model of the catalyst circulation lines
P4 P5 DPfrac : (21)
For the catalyst flow in the spent and regenerated
catalyst circulation lines (piping), a steady state behavior
is assumed. It is considered that dynamics of the lines
are very fast compared to the time constants of other
2.2. Regenerator model subsystems of the FCCU.
Spent and regenerated catalyst circulation considers a
The mathematical model for the regenerator presents single-phase flow, based on force balance [14]. For the
a higher complexity due to the importance of this system regenerated catalyst line the equation is:
in determining the time constant for the entire FCCU.
The regenerator is considered divided in two zones: a 144(P 6 P rb )zbed rc (Etap Eoil )rc DPsv;rgc
dense bed zone and a zone of entrained catalyst (the Frgc Lrgc Frgc
DPelb;rgc  0; (22)
disengaging zone) (Fig. 3). A2rgc rc
The dense bed zone consists of two phases: a bubble
phase of gaseous reactants and products moving up the and for the spent catalyst line the force balance is given
bed in plug flow and a perfectly mixed dense phase by:
containing gases and solid catalyst. Wr
Mass transfer occurs between the two phases but at 144(P 4 P 6 )(Estr Elift )rc 
Astr
regenerator temperatures the reaction rates are control-
ling, rather than mass transfer between the two phases. F L F
DPsv;sc DPelb;sc  sc sc sc 0: (23)
Since the dense phase is considered perfectly mixed, the A2sc rc
temperature is assumed uniform in the bed and the
The pressure drop on the slide valves is described by
gaseous phase in equilibrium with dense phase. Catalyst
the following equation:
is present in the zone above dense bed due to entrain-  
ment. The amount of catalyst decreases with the 50Fcat 2 144
DPsv  : (24)
regenerator height. In the entrained catalyst zone the KAsv sv rc
CO combustion is dominant (the amount of catalyst is
diminished) having an important heat contribution. Pressure drop on other pipe restrictions are given by
equations of the type:
1
DPelb  Nrc v 2 : (25)
2
Detailed presentation of models for the feed and
preheat system, regenerator, air blower, and wet gas
compressor are presented in Appendix B.

3. Results and discussion

3.1. Dynamic simulation results

A set of FCCU dynamic simulations have been


performed and studied as response to different upsets
in manipulated variables and disturbances. From this
Fig. 3. Scheme of the FCCU regenerator. set, the dynamic response to the coking rate disturbance
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 71

KC is presented. A step coking rate disturbance KC a diminished amount of CO2 production (Fig. 4(n)).
(3.2% step increase) has been applied at time t /500 s Taking into account the fact that heat generated by CO
from the beginning of the simulation. This type of formation is about three times less than CO2 heat
disturbance simulates changes in properties of the feed formation, the global effect is the reduction of net heat
oil residing in the increase of the amount of coke contribution in the regenerator with direct consequence
deposited on the catalyst. For the industrial unit, this on temperature decrease (Fig. 4(c)). The cyclone tem-
kind of coking rate change is possible to appear due to perature follows this decrease (Fig. 4(i)). New equili-
the fresh feed composition change or to the recycle brium is reached at lower reactor and regenerator
flowrate upset. The evolutions of the most representa- temperatures compared to temperature values before
tive process variables are presented in Fig. 4. The disturbance occurrence (Fig. 4(c,e)). Catalyst flowrates
dynamic responses are interpreted over two periods and coke fraction on spent and regenerated catalyst are
corresponding to the time sequence of the phenomena. also increased (Fig. 4(f,g)). The CO fraction is increased
(Fig. 4(m)), but CO2 and O2 fractions are decreased.
3.1.1. First period Results obtained by dynamic simulation present a
The increase of coke amount deposited on spent good fit with industrial operating data, simulated
catalyst evacuated from the reactor is rapid (Fig. 4(j)). variables being situated in a range corresponding to
The increased amount of coke entered in the regenerator industrial unit behavior (Table 1). Comparison between
induces, in the first time sequence, the small temperature industrial operating data and dynamic simulation results
rise in the regenerator (Fig. 4(c)), and then in the reactor has been performed for a set of data (1 month period),
(Fig. 4(e)), leading to the intensification of cracking confirming the main trends of the dynamic behavior
reactions, with direct effect on reactor pressure rise (Fig. both on short and large time scales. Obtaining a better
4(b)). As a consequence of the reactor pressure rise, the fit is still possible by increasing the complexity of the
flowrate of spent catalyst increases (Fig. 4(f)). As spent model, but also necessary, as properties of the raw
catalyst flowrate becomes higher than regenerated material is subject to changes.
catalyst flowrate (Fig. 4(g)), the reactor catalyst inven- Dynamic simulations reveal the multivariable and
tory decreases (Fig. 4(a)). The small increase in regen- nonlinear behavior of the process presenting strong
erator pressure and then in regenerator catalyst interactions. Inverse response has been noticed denoting
inventory determines a small decrease in the air entering multiple paths with opposing effect transmission. Single
the regenerator (due to the increased counter-pressure) loop decentralized control has to face strong impedi-
(Fig. 4(l)). The regenerator temperature begins to ments for such challenging interacting behavior.
decline (after a first low-amplitude peak), as a conse- The newly developed dynamic simulator offers the
quence of the increased contribution of spent catalyst possibility to study different operating regimes induced
(with lower temperature) entering the regenerator. both by design changes and by changing operation
strategies. It also proves to be a valuable tool for
3.1.2. Second period investigating the way that different control strategies
The regenerator temperature decrease induces the may be implemented and predict their results. Advanced
temperature decrease in the reactor followed by a control systems, as MPC algorithms, are based on
pressure decrease in the reactor (Fig. 4(b)). The reactor mathematical models and rely on the dynamic simula-
pressure reduction determines the decrease of spent tor.
catalyst flowrate (Fig. 4(f)), and the increase of regen-
erated catalyst flowrate (Fig. 4(g)). For the reactor, the 3.2. Model predictive control results and interpretations
consequence is the increase in catalyst inventory (Fig.
4(a)). The regenerator temperature decrease continues 3.2.1. Control scheme selection
due to the fact that net coke contribution is increased MPC, also referred as moving (receding) horizon
(the combustion air flowrate has a negligible increase). control, has become an attractive control strategy
The explanation of this net coke contribution increase is especially for linear but also for nonlinear systems
the following: both spent and regenerated catalyst subject to input, state and output constraints.
flowrates increase (small growth), but mainly, the There are some features that individualize MPC in the
fraction of coke on spent (Fig. 4(j)) and on regenerated field of control design, making it attractive. In contrast
catalyst (Fig. 4(k)), are also increasing. As a result, the to other feedback controllers that calculate the control
difference of these two fractions increases. For this action based on present or past information, MPC
reason, combustion in the regenerator is performed in a determines the control action based on the prediction of
diminished excess of oxygen (Fig. 4(h)), with direct future dynamics of the system. Due to the future
implication on the heat balance of the regenerator. The prediction, early control action can be taken accounting
equilibrium of carbon combustion is shifted to an for future behavior. In practice, most of the systems
increased amount of CO formation (Fig. 4(m)), and to have to satisfy input, state or output constraints,
72 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

Fig. 4. Simulation of FCCU dynamic behavior in the presence of a coking rate KC disturbance (3.2% step increase). (a) Reactor catalyst inventory
Wr [t], (b) reactor pressure P4 [bar], (c) regenerator temperature Treg [8C], (d) regenerator pressure P6 [bar], (e) reactor temperature Tr [8C], (f) spent
catalyst flowrate [kg/s], (g) regenerated catalyst flowrate frgc [kg/s], (h) oxygen to air molar fraction in stack gas x O2sg, (i) cyclone stack gas
temperature Tcyc [8C], (j) mass fraction of coke on spent catalyst csc [kg coke/kg catalyst], (k) mass fraction of coke on regenerated catalyst crgc [kg
coke/kg catalyst], (l) regenerator inlet air flowrate Ft [Nm3/h], (m) CO to air molar fraction in stack gas x COsg, (n) CO to air molar fraction in stack
gas x CO2sg.

resulting in limitations on achievable control perfor- by constraints that are predicted to become active in the
mance (in the extreme case affecting the stability). MPC future. The objective function specifying the desired
is able to obtain better control performance in the control performance is optimized (minimized) on-line at
presence of constraints since it is able to determine the each time step. The number of computed values in the
current control action for minimizing the errors caused manipulated variable sequence is finite (finite input
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 73

Fig. 4 (Continued)

horizon) and discrete in time , accounting for the fact cases. For the present study the Dynamic Matrix
that the involved optimization problem can be solved Control form of MPC has been employed.
with numerical methods. A time-continuous approach Based on literature survey and analysis of the current
can lead to extremely demanding numerical problems. industrial FCCU operation, a set of process variables
Multivariable controllers are often the only solution able has been selected and considered to have first role
to provide desired control performance in the presence importance in efficient and safe operation of the unit
of interactions and MPC can successfully handle such [11 /15].
74 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

Fig. 4 (Continued)

The controlled variables have been selected to provide, thermal operation for the regenerator and for the
through control, a safe and economic operation. Con- downstream units (piping and CO boiler).
trol of reactor catalyst inventory (reactor level) Wr, The manipulated variables have been chosen from the
provides stabilization of catalyst circulation. It also sets set of independent variables possible to be changed from
up a buffer for diminishing upsets in coke concentration a practical point of view. The main manipulated
deposited on the catalyst and for temperature change variables are the spent and regenerated catalyst flow-
progressing from the reactor toward the regenerator. rates that may be changed by regenerated svrgc and
Regenerator temperature, Treg, has to be maintained at spent svsc slide valve position. The preheating furnace
a certain value to allow a stable removal of coke from fuel flow, F5, is an important manipulated variable with
the catalyst. Overriding a high temperature limit pro- effective action on the thermal balance of the entire unit.
duces a permanent catalyst deactivation; a reduction The stack gas flowrate from the regenerator, changed by
under a lower limit leads to coke accumulation on the stack gas valve position V14 and the air vent flowrate,
regenerated catalyst. The reactor temperature, Tr, has to changed by air vent valve position V7, are other two
be maintained at a certain level to provide a desired manipulated variables. The wet gas suction flowrate,
maximum conversion of the feed oil. The stack gas changed by suction valve position V11, is another
oxygen concentration, x O2sg, has to be controlled in manipulated variable considered in the control schemes.
order to provide a desired coke combustion, preventing The selected disturbances reflect main upsets possible
both a thermal increase and an inefficient load of the to affect the normal operation of the unit: main
combustion air blower. Maintaining the cyclone tem- fractionator pressure upset, feed oil coking character-
perature, Tcyc, under a maximum limit, provides safe istics (coking rate) upset and ambient temperature upset

Table 1
Typical operating conditions and values obtained with the simulator

Process variable Measuring unit Minimum value Maximum value Nominal value Value in the simulator

Air flowrate entering regenerator Nm3/h 85 00 147 000 98 500 102 514
Air vent flowrate Nm3/h 0 5500 2500 2510
Regenerator temperature 8C 650 700 682 685.06
Cyclone temperature 8C 677 710 705 708.5
Reactor temperature 8C 490 525 515 516.99
Reactor pressure Bar 1.2 1.9 1.3 1.279
Regenerator pressure Bar 1.2 2.8 1.5 1.495
Coke on spent catalyst Mass fraction 0.009 0.014 0.012 0.01165
Coke on regenerated catalyst Mass fraction 0.002 0.0045 0.0035 0.00393
CO2 concentration in flue gas Volume fraction 0.08 0.16 0.13 0.141
O2 concentration in flue gas Volume fraction 0.001 0.008 0.0035 0.00288
CO concentration in flue gas Volume fraction 0.03 0.08 0.05 0.042
Catalyst inventory in the reactor Tons 30 60 50 55.7
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 75

[5,13]. The main fractionator pressure disturbance has 3.2.2. Different MPC control schemes results
been included in the simulation by the term DPfrac, Following the results obtained by dynamic simula-
representing the reactor/main fractionator pressure tion, the most favorable MPC control schemes, from
drop. This disturbance reveals the effect of upsets in each category, are: S1: 3/3, S5: 4 /4, S10: 5 /5. From
main fractionator operation, acting on reactor/regen- this large set of MPC dynamic simulations of FCCU,
erator system. Such main fractionator pressure upsets the representative S5: 4 /4 control scheme results are
may appear when: vapor flow is changed as a result of presented in Figs. 5 and 6.
suction flowrate change of wet gas compressor, internal As can be noticed, the S5: 4 /4 control scheme
liquid/vapor traffic of the main fractionator is changed succeeds to counteract the disturbance effects, present-
due to reboiler and condenser load upset or by pressure ing small overshoot and short settling time. This
changes induced from downstream gas recovery unit. behavior demonstrates good setpoint following capa-
city.
An increasing step disturbance has been selected (having
The superior behavior of S5: 4/4 control scheme,
/37% amplitude increase and applied at time t/500 s).
predicted by the controllability analysis based on RGA
The coking characteristic of the feed oil, coking rate KC,
values presented in Table 3, has been confirmed by the
was included as a disturbance to study the effect of
dynamic simulation results.
changes in raw material properties. It was noticed that Compared to S5: 4 /4, the S6: 4 /4 control scheme
this unmeasured disturbance has a strong effect on the has inferior control performance showing higher over-
heat balance of the entire unit. A positive step change shoot and longer response time (especially for the case
has been selected for this disturbance (having /3.2% of KC disturbance). The S7: 4/4 control scheme
amplitude increase and applied at time t/500 s). The presented unsatisfactory control performance (offset)
ambient temperature change is a continuous disturbance for all controlled variables in the case of KC disturbance.
affecting FCCU on a day-time basis. It consists in For the case of the other investigated disturbances the
combustion air flowrate change, introducing low ampli- control performances of S6: 4/4 and S7: 4 /4 control
tude upsets in the unit. This disturbance was included as schemes are not essentially affected.
a descending ramp, with negative slope (/16 8C/8 h), Compared to S1: 3/3 control scheme, S5: 4/4
applied for 1 hour between t/300 s and t/3900 s. scheme has an unimportant increase of the overshoot
The MPC of the FCCU was designed in a two-level (for the case of KC performance), but a small decrease of
control structure, acting at the top level of the hierarchic the response time can be noticed. The ability to maintain
control system by cascading the low-level regulatory the stack gas oxygen concentration at a predefined value
control loops (usually flowrate control loops). allows a more efficient FCCU operation due to better
A controllability study, based on relative gain array use of air blower capacity and to safer operation by the
(RGA), has been performed for selecting both the most control of ‘‘afterburning’’ phenomenon. Having an
efficient manipulated variables for changing the con- additional variable, compared to the 3 /3 control
trolled variables but also for determining the best MPC schemes, it may be concluded that S5: 4 /4 scheme is
control scheme, among a set of schemes of the same preferable.
dimensions. The RGA is a measure of interaction Compared to lower dimension schemes presented
between controlled variables, each of the RGA elements before, the 5/5 control schemes are characterized by
denoting the ratio between open loop and closed loop the existence of higher overshoot and a longer response
time, possibly coupled with small offset, but the control
gain in decentralized control. This controllability in-
performances are not considerably affected.
dicator, as a first filter for selecting the best control
S12: 5 /6 MPC scheme did not reveal improvements
scheme, proved to be useful not only for decen-
compared to S10: 5/5 scheme. The advantage of using
tralized control but also for the multivariable approach
a control scheme with a higher number of manipulated
[13,17]. than controlled variables will become operative when
Based on this approach, a set of control schemes has constraints on manipulated variables are imposed. The
been investigated [16,17]. They have a different number number of manipulated variable surplus may serve as a
of controlled/manipulated variables: 3 /3, 4 /4, 5/5, supply for the case of operating conditions when one or
5 /6 schemes, presented briefly in Table 2. The set of more of the manipulated variables become restricted.
MPC schemes presented in Table 2 have been tested in
the presence of the three typical described disturbances. 3.2.3. Considerations on MPC tuning
Different values have been investigated for the error It is a well-known fact that model predictive con-
diagonal weighting matrix, ywt (Gy ), and for the troller tuning, especially for the MIMO case, is difficult
manipulated-variable move diagonal weighting matrix, [18,19,21]. This aspect is unexpected if taking into
uwt (Gu ), from the MPC quadratic optimization objec- consideration the relatively large number of parameters
tive. possible to be tuned for obtaining desired control
76 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

Table 2
Tested control schemes

Control scheme (name/dimension) Controlled variables Manipulated variables MPC tuning parameters uwt and ywt

S1 3 /3 Wr Treg Tr svrgc svsc F5 uwt /[120 120 0.8], ywt/[0.1 0.2 1]
S2: 3 /3 Wr Treg Tr svrgc svsc V14 uwt /[120120480], ywt/[0.1 0.2 1]
S3: 3 /3 Wr Treg Tr svrgc svsc V7 uwt /[120120600], ywt/[0.1 0.2 1]
S4: 3 /3 Wr Treg Tr svrgc svsc V7 uwt /[7575300], ywt/[0.1 0.2 1]
S5: 4 /4 Wr Treg Tr x O2sg svrgc svsc V14 V7 uwt /[3030120120], ywt/[0.1 0.2 1 0.5]
S6: 4 /4 Wr Treg Tr x O2sg svrgc svsc F5 V7 uwt /[1501501600], ywt/[0.1 0.2 1 0.5]
S7: 4 /4 Wr Treg Tr x O2sg svrg, svsc V11 V7 uwt /[150150300600], ywt /[0.1 0.2 1 0.5]
S8: 5 /5 Wr Treg Tr x O2sg Tcyc svrgc svsc F5 V7 V11 uwt /[1501501600300], ywt /[0.1 0.2 1 0.5 0.5]
S9: 5 /5 Wr Treg Tr x O2sg Tcyc svrgc svsc F5 V7 V11 uwt /[30 30 0.2 120 60], ywt/[0.1 0.2 1 0.5 0.5]
S10: 5 /5 Wr Treg Tr x O2sg Tcyc svrgc svsc F5 V7 V14 uwt /[1501501600600], ywt /[0.1 0.2 1 0.5 0.5]
S11: 5 /5 Wr Treg Tr x O2sg Tcyc svrgc svsc V11 V7 V14 uwt /[150150300600600], ywt/[0.1 0.2 1 0.5 0.5]
S12: 5 /6 Wr Treg Tr x O2sg Tcyc svrgc svsc F5 V7 V11 V14 uwt /[1501501600300600], ywt/[0.1 0.2 1 0.5 0.5]

performance. These possible tuning parameters are: Due to these aspects, the model predictive controller
sampling time T , model horizon n , prediction horizon tuning has an iterative character and the control
p , input horizon m , error weighting matrix ywt (Gy ) and performance enhancement may be performed, in a great
manipulated variable move weighting matrix uwt (Gu ). extent, by recursive simulations. A set of recommenda-
The tuning difficulties are tied to the MIMO character- tions for MPC tuning has been specified and may be
istics of the problem and to the insufficient control of regarded as a tuning MPC guide [7,19,20]. The sampling
the tuning effect of parameter changes on the control time T is established as a trade-off between losing
performance. These difficulties become more important important dynamic information and overloading the
when nonlinear behavior of the model is present [20]. computing system; a value of T /100 s has been chosen.

Fig. 5. MPC simulation results (solid) in the presence of KC disturbance (step increase of coking rate), for S5: 4 /4 control scheme; disturbed process
without control (dashed).
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 77

Fig. 6. MPC simulation results (solid) in the presence of DPfrac disturbance (step increase of reactor /main fractionator pressure drop), for S5: 4/4
control scheme; disturbed process without control (dashed).

Table 3 zone with inefficient control effect. The prediction


RGA for S5: 4/4 control scheme horizon p is not established very long (relative to open
loop settling time) in order to prevent sluggish control
svrgc svsc V14 V7
action (having in fact a stabilization effect) and raising
Wr 0.3634 1.3981 /0.8004 0.0390 the computational load. For the control horizon the
Treg 2.0118 /0.6095 /0.2298 /0.1725 value of m /10 has been taken. The control horizon m
Tr 0.3946 0.3969 /0.9546 1.1631 is established not too long, to prevent aggressive control
x O2sg /1.7698 /0.1855 2.9848 /0.0296
action, but also not too short, to determine an inefficient
control and to provide a sufficient number of degrees of
The model horizon n is established such as Tn should freedom.
extend over the open loop response time (smaller values The diagonal error weighting matrix ywt (Gy ) was
can lead to undesired peaks appearing at Tn time determined such as the elements on the main diagonal be
horizon, when the model error first becomes significant); equal to the inverse of the maximum allowed offset of
a value of n/400 has been taken. the particular controlled variable; these values are
Both prediction horizon p and control horizon m weighted again after dynamic simulation tests (Table
have been established based on the assumptions that 2). The diagonal manipulated variable move weighting
large values lead to increased computational effort and matrix uwt (Gu ) was determined such as the elements on
short values produce ‘‘short-sighted’’ control policy. the main diagonal be equal to the inverse of the
The value of p/100, i.e. one fourth of the settling time, maximum allowed variation of the manipulated vari-
was selected for the prediction horizon. The choice of a able; these values are weighted again after dynamic
smaller p leads to short-sighted control associated with simulation tests (Table 2).
more aggressive control action. An additional conse- It is meaningful to mention that tuning was per-
quence of reducing p is that the constraint violations are formed to obtain good control performance for all cases
only checked over a short horizon, leading to a dead- of the three applied disturbances resulting in a more
78 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

conservative tuning than would have been necessary for them are strong, a multivariable control strategy can be
each of them considered individually. successful and MPC proves to be an effective one.

3.2.4. MPC versus decentralized PID control 3.2.5. MPC using model scheduling approach
To make a comparison between MPC and traditional The model used for computation of manipulated
decentralized PID control, simulations have been per- variables is a linear one, obtained by the linearization
formed involving the set of controlled and manipulated of the nonlinear model around the operating point [23].
variables of S5: 4/4 control scheme. The pairing of Results presented in previous paragraphs use such a
controlled and manipulated variables used for the PID unique model. For the elimination of errors caused by
decentralized control have been suggested by the RGA nonlinearities the authors proposed and investigated the
(Table 3): Wr /svsc, Treg /svrgc, Tr /V7 and x O2sg /V14. behavior of a control scheme using scheduled lineariza-
Anti-windup PID digital controllers have been applied tion. The FCCU linearized model is periodically up-
[22]. dated at time moments multiple of 3000 s, starting from
Tuning of the PID controllers has been made by t /1500 s. The changing model case has a roughly better
repeated simulations using an ‘‘experimental’’ type control performance, particularly for Wr controlled
method based on bringing first the system at the stability variable (affected by the lowest value in the error-
limit. Again, the tuning has been made in a way to weighting matrix) (Fig. 8). The scheduled linearization
obtain good control performance for all of the three test using a higher frequency did not reveal significant
disturbances taken into consideration. Comparative improvement for the cases of MPC control in the
results of MPC and PID control are presented in Fig. presence of the investigated disturbances. This may be
7. Results presented in Fig. 7 reveal the superior determined by keeping the operating point relatively
behavior for the case of MPC, both with respect to close to the setpoint values. As disturbance effects are
overshoot and response time. Following the performed more important, the updating of the linearized model, at
simulations it may be concluded that, as the number of higher and possibly variable frequency, may become
controlled variables is high and the interactions between necessary. Further results are under investigations.

Fig. 7. Comparative results between MPC, S5: 4/4 (solid) and PID control (dash /dotted) in the presence of KC disturbance; disturbed process
without control (dashed).
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 79

Fig. 8. MPC comparative representation for: MPC adaptive model case (dashed /dotted), MPC with unique linearized model case (solid) and case of
disturbed process without control (dashed); S10: 5 /5 control scheme in the presence of KC disturbance.

3.2.6. Constrained MPC ing to this aspect, the interest and success MPC
Among the most attractive MPC characteristics is the algorithm has gained in a large number of reported
possibility of considering constraints in a direct way. industrial applications may be explained [24]. The case
This attribute offers, while specifying FCCU operating of MPC with constraints on manipulated variables is
and material constraints, the best (in an optimal sense) investigated.
solution for the control problem. For the SISO case, To test this ability, the following potential FCCU
requiring and conforming to constraints is frequently malfunction event is simulated. One of the slide valves,
not very difficult. But for the MIMO case, where the spent catalyst slide valve svsc, presents a malfunc-
interactions are present, the aim of obtaining desired tion consisting in the impossibility of opening it over the
control performance is usually a difficult task. Accord- upper limit specified by the value svrgcsup /0.4 and
80 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

closing it under the lower limit specified by svrgcinf /0.3 Based on the present study it may be considered that
value. The position of the slide valve during nominal this way of MPC application is revealing and sustaining
operation is given by svrgc /0.35 value. This accidental the incentives of MPC algorithm from the perspective of
situation raises special problems for the operating its industrial implementation.
personnel in an industrial unit having traditional
(classical) control system. For the case of MPC system
it is sufficient to specify this constraint and keep closed
the feedback control loops until the normal operation 4. Conclusions
regime is restored.
The simulation of MPC behavior for this special The paper presents a new model and dynamic
operating condition is presented in Fig. 9, but only for simulator for the FCCU aggregate systems: reactor,
reactor catalyst inventory Wr, controlled variable; other regenerator, catalyst circulation lines, preheating sys-
tem, air blower and wet gas compressor. The nonlinear,
variables exhibit similar behavior with the uncon-
dynamic and multivariable model has been fitted and
strained case. The coking rate disturbance KC has been
then verified with a set of representative operating data
applied and the MPC with adaptive model has been
originating from an industrial FCCU, showing its
simulated. The investigated control scheme is S12: 5/6.
complex behavior as response to typical disturbances.
Vector of constraint limits imposed to the manipulated It may be observed that the disturbance most difficult to
variables is given by ulim /[0 0.3 0 0 0 0 1 0.4 1.98 0.5 1 reject proved to be the coking rate factor, KC, although
0.8]. The first six values fix the minimum limits and the disturbance considered with the highest amplitude
the last six the maximum limits allowed for the change was the reactor/main fractionator pressure
manipulated variables (in the order they are specified drop, DPfrac.
in Table 2). Investigations have been performed by simulation to
As may be observed in Fig. 9, the control perfor- reveal incentives and limitations for implementing MPC.
mance with MPC is not substantially affected by the The most favorable MPC control schemes, for each
occurred constraint. Two of the manipulated variables investigated category, are: S1: 3/3, S5: 4/4, S10: 5/
(svsc and V7) reached the lower limit values. These 5. The last one is the most profitable, due to the large
limitations do not seem to have a negative impact on the number of controlled variables. It is interesting to notice
controlled variables due to the fact that optimal strategy that S12: 5/6 control scheme (containing an extra-
succeeds to change the other manipulated variables in a manipulated variable), in its unconstrained form, does
way to provide good control performance. not bring additional quality to MPC. But when con-
The possibility may also be observed to involve a straints on manipulated variables are present, this
higher number of manipulated variables than controlled approach proves real improvements due to the ‘‘sur-
variables and the potential use of this ‘‘excess’’ of plus’’ of command able to compensate for those
command for the cases when constraints on manipu- manipulated variables limited by constraints. Compared
lated variables are present. with the traditional decentralized PID control, MPC
presents better control performance based on its multi-
variable feature, inherent prediction ability and capacity
to directly handle constraints using an even larger
number of manipulated than controlled variables. A
nonlinear MPC method has been proposed and inves-
tigated to account for process non-linearity based on
periodic updating of the linearized model used for
control action computation. This nonlinear MPC im-
plementation may lead to potential improvement by the
use of dynamic sensitivity analysis.
In practice, the MPC implementation is intended to
be performed in a two-layer structure: the layer of
decentralized PID loops stabilizing the main process
variables and the MPC layer adjusting the setpoints of
the underlying regulatory loops.
Benefits of better control performance in FCCU
operation mainly consist in the achievement of safe-
Fig. 9. Controlled variables for constrained MPC (svscinf /0.3,
svscsup /0.4), scheme S12: 5/6 in the presence of KC disturbance; keeping the controlled variables very close to the
unconstrained MPC (solid), constrained MPC (dashed /dotted), dis- constrained limits, where optimum operating conditions
turbed process without control (dashed). usually lie.
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 81

Detailed mathematical model of the UOP FCCU may Ft air flowrate into regenerator (Nm3/h)
be found in Appendix B. F3 fresh feed flowrate (lb/s, kg/s)
F4 slurry recycle flowrate (lb/s, kg/s)
hris height of the riser (ft, m)
Appendix A: Nomenclature K flow coefficient for the slide valve (0.7)
Kc reaction rate constant for coke production
Argc cross sectional area of regenerated catalyst (s 1)
pipe (ft2) kr1 reaction rate constant for the total rate of
Aris cross sectional area of reactor riser (ft2) cracking of the feed oil (s 1)
Asc cross sectional area of spent catalyst pipe kr3 reaction rate constant for the rate of
(ft2) cracking gasoline to light gases and coke
Astr cross sectional area of reactor stripper (ft2) (s 1)
Asv cross sectional area of regenerated/spent k12 wet gas V12 valve flow rating (mol/s psia1/2,
catalyst slide valve at completely open kg/s (N/m2)1/2)
position (in2) Lrgc length of regenerated catalyst pipe (ft, m)
[COR] catalyst to oil ratio Lsc length of spent catalyst pipe (ft, m)
Ccat mass fraction of coke produces in the riser m manipulated variable (input) horizon
Cpc heat capacity of catalyst (Btu/lb/F, J/kg/K) m* factor for the dependence of the initial
Cpd heat capacity of steam (Btu/lb/F, J/kg/K) catalyst activity on Crgc
Cpf heat capacity of the feed (Btu/lb/F, J/kg/K) n model horizon
Cpfv heat capacity of feed vapor (Btu/lb/F, J/kg/ N exponent for the dependence of Ccat on Crgc
K) N* integer value representing a constant for
Crgc coke fraction on regenerated catalyst (lb pressure drop on catalyst pipes
(crgc) coke/lb catalyst, kg coke/kg catalyst) p prediction horizon
Csc (csc) coke fraction on spent catalyst in the stripper Patm atmospheric pressure (psia, N/m2)
(lb coke/lb catalyst, kg coke/kg catalyst) Prb pressure at the bottom of the riser (psia, N/
Csc1 coke fraction on spent catalyst at riser outlet m 2)
(lb coke/lb catalyst, kg coke/kg catalyst) P4 reactor pressure (psia, N/m2)
C1 wet gas production constant (mol/lb feed, P5 main fractionator pressure (psia, N/m2)
mol/kg feed) P6 regenerator pressure (psia, N/m2)
C2 wet gas production constant (mol/lb feed/F, R universal gas constant (ft3 psia/lb mol/R, J/
mol/kg feed/K) mol/K)
Ecf activation energy for coke formation (Btu/ sv spent/regenerated catalyst slide valve posi-
mol, kJ/mol) tion (0 /1)
Ef activation energy for cracking the feed (Btu/ svsc spent catalyst slide valve position (0 /1)
mol, kJ/mol) svscinf spent catalyst slide valve lower limit con-
Eg activation energy for cracking gasoline (Btu/ straint (0.3)
mol, kJ/mol) svscsup spent catalyst slide valve higher limit con-
Elift elevation of the pipe for spent catalyst, inlet straint (0.4)
in the regenerator (ft, m) svrgc regenerated catalyst slide valve position
Eoil elevation of feed inlet in the riser (ft, m) (0 /1)
Estr elevation of the pipe for spent catalyst outlet t time (s)
from the reactor (ft, m) tc catalyst residence time in the riser (s)
Etap elevation of the pipe for regenerated catalyst, T sampling time (s)
outlet from the regenerator (ft, m) Tcyc regenerator stack gas temperature at cyclone
Fcat flowrate of spent or regenerated catalyst (F, K)
(t/min) Tr temperature of reactor riser outlet (F, K)
Ff total feed flowrate (lb/s, kg/s) Tref base temperature for energy balance (F, K)
Frgc (frgc) regenerated catalyst flowrate (lb/s, kg/s) Treg temperature of regenerator bed (F, K)
Fsc (fsc) spent catalyst flowrate (lb/s, kg/s) Ts temperature of stripper outlet (F, K)
FV11 flow through wet gas compressor suction T0 temperature of the feed entering the riser
valve V11 (mol/s, molg/s) after mixing with the catalyst (F, K)
FV12 flow through valve V12 (mol/s, molg/s) T2 furnace outlet temperature of the feed (F, K)
FV13 flow through valve V13 (mol/s, molg/s) ulim vector of constraints imposed to the manip-
Fwg wet gas production in the reactor (mol/s, ulated variables ([0 0.3 0 0 0 0 1 0.4 1.98 0.5 1
molg/s) 0.8])
82 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

uwt (Gu ) diagonal weighting matrix for the manipu- Appendix B: FCCU model description
lated variable move, in the optimization
index The newly developed mathematical model for the
v catalyst velocity in spent/regenerated pipe (ft/ UOP type FCCU is based on the mechanistic Amoco
s, m/s) Model IV FCCU [5]. Compared with Amoco Model IV,
vris volumetric flowrate in the riser the new mathematical model describes a different
(ft3/s, m3/s) FCCU type, both from the operation and from the
V14 position of the stack gas valve (0 /1) construction point of view. Different geometric dimen-
V7 position of the air vent valve (0 /1) sions and relative position define the reactor and
V11 position of the wet gas compressor suction regenerator in this case, compared with the Model IV
valve (0 /1) case. The regenerator of the presented UOP FCCU
V12 position of the flare valve (0 /1) operates in partial combustion mode. The reactor model
Wr inventory of catalyst in the reactor (stripper) uses a Weekman kinetic model [9]. Catalyst circulation
(lb, kg) is described including spent and catalyst valves on
Wris inventory of catalyst in the riser (lb, kg) catalyst circulation lines. These valves are used as
x O2,sg molar ratio of O2 to air in stack gas (mol O2/ main manipulated variables for FCCU control.
mol air) The FCCU dynamic model has been developed on the
x COsg molar ratio of CO to air in stack gas (mol basis of reference construction and operation data from
CO/mol air) an industrial unit. The described model is rather
x CO2,sg molar ratio of CO2 to air in stack gas (mol complex succeeding to capture the major dynamic
CO2/mol air) behavior of UOP type FCCU [8]. The unit consists of
yf mass fraction of feed oil the following parts: feed and preheat system, reactor,
yg mass fraction of gasoline regenerator, air blower, wet gas compressor and catalyst
ywt (Gy ) diagonal weighting matrix for the error, in circulation lines [5,14].
the optimization index
za dimensionless distance along riser
B.1. Feed and preheat system
zbed dense bed height (ft, m)
a catalyst deactivation constant (s 1)
The feed and preheat system is presented in Fig. A1.
DHevp heat of vaporizing the feed oil (Btu/lb,
The total feed flow F3 enters the preheat furnace at T1
kJ/kg)
temperature and is heated by means of the gaseous fuel
DHf heat of cracking (Btu/lb, kJ/kg)
having F5 flowrate. The feed preheat dynamic behavior
DPelb,sc pressure drop on different elements of spent
is described by the following mass and energy balance
catalyst pipe (psia, N/m2)
equations:
DPelb,rgc pressure drop on different elements
of regenerated catalyst pipe (psia, dT3 1
N/m2)  (F5 DHfu UAf Tlm Qloss ) ; (A1)
dt tfb
DPfrac pressure drop across reactor main fractio-
nator (psi, N/m2) (T3  T1 )  (T3  T2 )
Tlm    ; (A2)
DPsv pressure drop on regenerated/spent catalyst T  T1
ln 3
slide valve (psi) T3  T2
DPsv,sc pressure drop on spent catalyst slide valve
Qloss a1 F5 T3 a2 ; (A3)
(psia, N/m2)
DPsv,rgc pressure drop on regenerated catalyst slide
valve (psia, N/m2)
f0 initial catalyst activity at riser inlet
l ratio of mass flowrate of dispersion steam to
mass flowrate of feed oil
rc density of catalyst in the dense phase (lb/ft3,
kg/m3)
rpart settled density of catalyst (lb/ft3, kg/m3)
rris average density of material in the riser (lb/ft3,
kg/m3)
rv vapor density at riser conditions (lb/ft3, kg/
m 3)
u dimensionless temperature in the riser
Fig. A1. Feed and preheat system.
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 83

dT2 1
 (T2;ss T2 ) ; (A4)
dt tfo
UAf Tlm
T2;ss T1  : (A5)
F3 Cpf

B.2. Reactor and main fractionator model

Developing the new mathematical model for the


reactor implied a thorough survey, selection and then
synthesis, based on a large variety of models presented
in literature. The three-lump model has been considered
to be adequate for the global description of the
phenomena taking place in the reactor. Reactor is
divided in two parts: riser and stripper. The riser model
is built on the following assumptions: ideal plug flow
and very short transient time (the residence time in the
riser is very short compared with other time constants,
especially with the regenerator time constants Fig. A2. FCCU reactor.
[1,5,8,10].). It is modeled by mass balance describing
the gasoline and coke/gases production based on The term K 1y2f [COR] represents the kinetics of the
Weekman’s triangular kinetic model [9]. The mixed feed, K3yg[COR] the kinetics of the gasoline; F is a
nonlinear differential and algebraic system of equations function of catalyst deactivation due to coke deposition;
also accounts for the amount of coke deposited on f0 the reduction of catalyst activity due to the coke
catalyst and for the cracking temperature dynamics [13]. resident on the catalyst after regeneration; tc residence
The reactor is presented in Fig. A2. time in the riser; a2 /k1/k2 fraction of feed oil that
cracks to gasoline. This model develops the models
presented by Lee and Groves [24], Shah et al. [25] and
B.2.1. Mass balance for the riser
Hovd and Skogestad [13]. The amount of coke produced
Mass balance for the feed is described by the equation
is described by the following correlation taken from
dy f Voorhies and Kurihara [26]:
K1 y2f [COR]Ftc : (A6) sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dz a tc Ecf
C cat K c exp : (A13)
Mass balance for the gasoline is described by the CrgcN RTr
equation
The fraction of coke on the spent catalyst leaving the
dy g riser is:
 (a2 K1 y2f K3 yg )[COR]Ftc (A7)
dz a Csc1 Crgc Ccat : (A14)
where: The constant values m * and N have been used to
Ef perform a good fit of the mathematical model with
K1 (u) kr1 exp ; (A8) operating data from the industrial unit.
RT0 (1  u)
Eg
K3 (u) kr3 exp ; (A9)
RT0 (1  u) B.2.2. Heat balance for the riser
u (T T0 )=T0 ;
du DHf Ff dyf
F f0 exp(atc [COR]za ); (A10)  : (A15)
dza T0 (Fsc Cpc  Ff Cpf  lFf Cpd ) dza
f0 1mCrgc : (A11)
The amount of gases produced by cracking is
Inlet temperature in the riser T0 is determined by the described by the equation:
heat balance equation [3]
Fwg (F3 F4 )[C1 C2 (Tr Tref )]:: (A16)
F C T  Ff Cpf T2  DHevp Ff
T0  rgc pc reg : (A12) Constants C1and C2 have been fitted based on data
Frgc Cpc  Ff Cpfv from the industrial unit.
84 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

The stripper model is of CSTR type (mass and heat The regenerator is considered divided in two zones: a
balance) evaluating the temperature in the stripper and dense bed zone and a zone of entrained catalyst (the
the fraction of coke on spent catalyst. disengaging zone) (Fig. A3).
The dense bed zone consists of two phases: a bubble
phase of gaseous reactants and products moving up the
B.2.3. Mass and heat balance for the stripper
bed in plug flow and a perfectly mixed dense phase
containing gases and solid catalyst.
dTs Frgc
 (Tr Ts ); (A17) Mass transfer occurs between the two phases but at
dt Wr regenerator temperatures the reaction rates are control-
  ling, rather than mass transfer between the two phases.
dCsc dWr 1
 Frgc (Crgc Ccat )Fsc Csc Csc ; (A18) Since the dense phase is considered perfectly mixed, the
dt dt Wr
temperature is assumed uniform in the bed and the
dWr gaseous phase in equilibrium with dense phase. Catalyst
Frgc Fsc : (A19)
dt is present in the zone above dense bed due to entrain-
ment. The amount of catalyst decreases with the
regenerator height. In the entrained catalyst zone the
CO combustion is dominant (the amount of catalyst is
B.2.4. Pressure balance for riser bottom pressure diminished) having an important heat contribution. The
determination operating conditions are corresponding to CO partial
combustion mode.
rris hris The model consists in mass and heat balance equa-
Prb P4  ; (A20) tions for O2, CO, CO2 and coke, but also in heat balance
144
equations for solid and gaseous phase. These balance
F3  F4  Frgc equations are correlated with equations describing
rris  ; (A21)
nris entrained catalyst (bed characteristics) in the zone above
dense bed, catalyst flow and pressure in the regenerator.
F3  F4 Frgc
nris   : (A22)
rv rpart
The amount of catalyst in the riser is determined by B.3.1. Heat balance
the equation The dense phase of the bed is assumed perfectly mixed
due to the intense circulation of the catalyst. It is also
Frgc Aris hris considered that the entire amount of hydrogen deposited
Wris  : (A23)
nris on the catalyst is burned in the regenerator. Partial
combustion mode is considered compared to Model IV
FCCU [5].

B.2.5. Momentum balance for reactor and main


fractionator pressure determination

dP5
0:833(Fwg FV11 FV12 FV13 ) (A24)
dt
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
FV12 k12 V12 P5 Patm : (A25)
A constant pressure drop, DPfrac, between reactor and
main fractionator is considered; according to this, the
reactor pressure is computed by the equation
P4 P5 DPfrac : (A26)

B.3. Regenerator model

The mathematical model for the regenerator presents


a higher complexity due to the importance of this system
in determining the time constant for the entire FCCU. Fig. A3. FCCU regenerator.
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 85

The heat balance of the dense bed is described by the dXO2


[100(0:5k1 k2 )rB (z)Crgc k3 XCO (z)]
equations: dz
dTreg XO2 (z)
[Wreg Cpc M1 ] Qin Qout ; (A28)  ; (A41)
dt ns
Qin Qair QH QC Qsc
(A29) dXCO (z) X (z)
Qout Qfg Qrgc Qe ; [100k1 rB (z)Crgc 2k3 XCO (z)] O2 ; (A42)
dz ns
Qair Fair Cpair (Tair Tbase ); QH  FH DHH ; (A30)
QC Fair (XCO;sg DH1 XCO2 ;sg DH2 ); (A31) dXCO2 (z) dXO2 (z) dXCO (z)
 0:5 ; (A43)
Qsc Fsc Cpc (Tsc Tbase ); (A32) dz dz dz

Qfg [Fair (XO2 ;sg CpO2 XCO;sg CpCO XCO2 ;sg CpCO2 XCO2 (z)XO2 (0)XO2 (z)0:5XCO (z); (A44)
0:79CpN2 )0:5FH CpH2 O ](Tcyc Tbase ); (A33)
1
Qrgc Frgc Cpc (Treg Tbase ); (A34) XO2 (0) (0:21Fair 0:25FH ); (A45)
Fair
FH Fsc (Csc Crgc )CH : (A35)
100Fair XO2
The heat balance in the disengaging zone is described CO2 ;sg  ; (A46)
by: Fsg

dTreg (z)
0 05z5 zbed ;
dz
  (A36)
dTreg (z) dX (z) dX (z) 1
 DH1 CO DH2 CO2 zbed Bz 5zcyc ;
dz dz dz Cp(z)

10 6 × 28XCO
Cp(z)0:79CpN2 XCO (z)CpCO XCO2 (z)CpCO2 CCO;sg  : (A47)
28XCO  44XCO2  32XO2  22:12
XO2 (z)CpO2 [0:5CpH2 O FH dz Cpc Me ] Volume fraction of catalyst is given by the following
equations:
1
 ; dz 0 z]zcyc ; drB (z)
Fair 0; rB (z)1o e 05z 5zbed ; (A48)
dz
dz 1 zBzcyc : (A37) drB (z) 1000Fair rB (z)
 zbed Bz5 zcyc ; (A49)
Carbon balance: dz Areg ns rc;dilute
 
  o e min 1; max o f ; o f
dCrgc 1 dWc dWreg
 Crgc ; (A38)
dt Wreg dt dt 
1:904  0:363ns  0:048n 2s
 ; (A50)
dWreg zbed
Fsc Frgc ; (A39)
dt o f  0:3320:06ns : (A51)

dWc The mass flow of entrained catalyst leaving the dense


(Fsc Csc FH ) bed is described empirically by the following equations:
dt
[Frgc Crgc 12Fair (XCO;sg XCO2 ;sg )]: (A40) Me Areg ns rc;dilute (A52)
where:
Mass balance on oxygen, carbon monoxide and
carbon dioxide are given by: rc;dilute  10:582ns ; (A53)
86 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

Fsg  Fair 1 Fsucn;comb 48 000


ns  ; (A54) qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 rg Areg
 1:58110 9 1:24910 6 P 2base ; (A62)
520P6
rg  : (A55) 14:7P2
379 × 14:7(Treg  459:6) Pbase  ; (A63)
P1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
F7 kcomb P2 Prgb Fair ; (A64)
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
B.3.2. Pressure balance FV6  k6 fpp (V6 ) Patm P1 ; (A65)
It is assumed an ideal gas behavior for regenerator pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gases. The regenerator pressure is described by the FV7  k7 fpp (V7 ) P2 Patm : (A66)
equations given below:
 
dP6 R dT dn
 n reg (Treg 459:6) ; (A56)
dt Vreg;g dt dt
dn B.5. Wet gas compressor
 Fair Fsg ; (A57)
dt Wet gas compressor is of centrifugal type (Fig. A5). It
Vreg;g Areg zcyc Areg zbed (1o e ); (A58) is driven by an electric motor. It is assumed that the wet
gas compressor is pumping against a constant pressure
Wreg in the downstream vapor recovery unit.
Prgb P6  ; (A59)
144Areg Wet gas compressor equation is described below:
DPRR  P6 P4 : (A60) qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Fsucn;wg 4353:5 1:36610 8 0:1057H 2wg ; (A67)
The bed height is described by the empirical equation
   Hwg 182:922(C 0:0942
rw 1) ; (A68)
W  rc;dilute Areg zcyc
zbed  min zcyc ; 2:850:8ns  reg
Areg rc;dense Pvru
Crw  : (A69)
  P7
rc;dense
 ;
rc;dense  rc;dilute The suction pressure of the wet gas compressor is
rc;dense rpart (1o f ): (61) described by the equations:
dP7
5(FV11 F11 ); (A70)
dt
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
FV11 k11 fpp (V11 ) P5 P7 ; (A71)
B.4. Air blower model
fpp (x)e 2ln[(0:15)(1x)] x 0:5
The air blower is a centrifugal compressor driven by a ; (A72)
fpp (x)0:3x x 50:5
steam turbine (Fig. A4). A head-capacity performance
equation describes suction flowrate as a function of FV13 k13 V13 Pvru : (A73)
discharge pressure with suction at normal atmospheric
pressure:

Fig. A4. Regenerator combustion air blower. Fig. A5. Wet gas compressor.
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 87

B.6. Model of the catalyst circulation lines a2 furnace heat lost parameter (Btu/
s, J/s)
For the catalyst flow in the spent and regenerated Ccat mass fraction of coke produces in
catalyst circulation lines (piping), a steady state behavior the riser
is assumed. It is considered that dynamics of the lines is CCO,sg concentration of carbon monox-
very fast compared to the time constants of other ide in stack gas (ppm)
subsystems of the FCCU. Equations such as Eq. (A77) CH mass fraction of hydrogen in coke
have been used to fit the pressure drops in the model [COR] catalyst to oil ratio
with data from the industrial unit. CO2,sg concentration of oxygen in stack
Spent and regenerated catalyst circulation considers a gas (% mol)
single-phase flow, based on force balance [14]. For the Cp (z ) average heat capacity (Btu/mol/F,
regenerated catalyst line the equation is: J/mol/K)
144(P 6 P rb )zbed rc (Etap Eoil )rc DPsv;rgc Cpair heat capacity of air (Btu/mol/F, J/
mol/K)
Frgc Lrgc frgc Cpc heat capacity of catalyst (Btu/lb/
DPelb;rgc  0; (A74)
A2rgc rc F, J/kg/K)
CpCO heat capacity of carbon monoxide
and for the spent catalyst line the force balance is given
(Btu/mol/F, J/mol/K)
by:
CpCO2 heat capacity of carbon dioxide
Wr (Btu/mol/F, J/mol/K)
144(P 4 P 6 )(Estr Elift )rc 
Astr Cpf heat capacity of the feed (Btu/lb/
F, J/kg/K)
F L f
DPsv;sc DPelb;sc  sc sc sc 0: (A75) Cpfv heat capacity of feed vapor (Btu/
A2sc rc lb/F, J/kg/K)
The pressure drop on the slide valves is described by Cpd heat capacity of steam (Btu/lb/F,
the following equation: J/kg/K)
  CpN heat capacity of nitrogen (Btu/
50Fcat 2 144 mol/F, J/mol/K)
DPsv  : (A76)
KAsv sv rc CpO2 heat capacity of oxygen (Btu/mol/
F, J/mol/K)
Pressure drop on other pipe restrictions are given by
Csc coke fraction on spent catalyst in
equations of the type:
the stripper (lb coke/lb catalyst,
1 kg coke/kg catalyst)
DPelb  Nrc v 2 : (A77)
2 Csc1 coke fraction on spent catalyst at
riser outlet (lb coke/lb catalyst, kg
coke/kg catalyst)
Crgc coke fraction on regenerated cat-
alyst (lb coke/lb catalyst, kg coke/
kg catalyst)
Appendix C: Nomenclature Crw wet gas compressor compression
ratio
Areg cross sectional area of regenerator C1 wet gas production constant (
(ft2, m2) mol/lb feed, mol/kg feed)
Aris cross sectional area of reactor C2 wet gas production constant
riser (ft2, m2) (mol/lb feed/F, mol/kg feed/K)
Astr cross sectional area of reactor Ecf activation energy for coke for-
stripper (ft2, m2) mation (Btu/mol, KJ/mol)
Argc cross sectional area of regener- Ef activation energy for cracking the
ated catalyst pipe (ft2, m2) feed (Btu/mol, KJ/mol)
Asc cross sectional area of spent Eg activation energy for cracking
catalyst pipe (ft2, m2) gasoline (Btu/mol, KJ/mol)
Asv cross sectional area of regener- Elift elevation of the pipe for spent
ated/spent catalyst slide valve at catalyst, inlet in the regenerator
completely open position (in2/m2) (ft, m)
a1 furnace heat lost parameter (Btu/ Eoil elevation of feed inlet in the riser
ft3/F, J/m3/K) (ft, m)
88 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

Estr elevation of the pipe for spent Kc reaction rate constant for coke
catalyst outlet from the reactor production (s1)
(ft, m) k1 reaction rate constant (s1)
Etap elevation of the pipe for regener- kr1 reaction rate constant for the
ated catalyst, outlet from the total rate of cracking of the feed
regenerator (ft, m) oil (s1)
Fair air flowrate into regenerator k2 reaction rate constant (s1)
(mol/s, molg/s) kr2 reaction rate constant for the rate
Fcat flowrate of spent or regenerated of cracking of feed oil to gasoline
catalyst (lb/s, kg/s) (s1)
Fcoke production of coke in the riser (lb/ k3 reaction rate constant (mol air/s
s, kg/s) mol CO)
Ff total feed flowrate (lb/s, kg/s) kr3 reaction rate constant for the rate
FH burning rate of hydrogen (lb/s, of cracking gasoline to light gases
kg/s) and coke (s1)
fpp(x) nonlinear valve flowrate function k6 combustion air blower suction
Frgc regenerated catalyst flowrate (lb/ valve flow rating (lb/s psia1/2, kg/s
s, kg/s) (N/m2)1/2)
Fsc spent catalyst flowrate (lb/s, kg/s) k7 combustion air blower vent valve
Fsg stack gas flowrate (mol/s, molg/s) flow rating (lb/s psia1/2, kg/s (N/
Fsucn,comb combustion air blower suction m2)1/2)
flow (ICFM, m3/s) k11 wet gas compressor suction valve
Fsucn,wg wet gas compressor inlet suction flow rating (mol/s psia1/2, kg/s (N/
flow (ICFM, m3/s) m2)1/2)
ffrgc friction constant for pipe-regen- k12 wet gas V12 valve flow rating
erated catalyst flow (lbf s/ft2, N s/ (mol/s psia1/2, kg/s (N/m2)1/2)
m 2) k13 wet gas V13 valve flow rating
ffsc friction constant for pipe-spent (mol/s psia1/2, kg/s (N/m2)1/2)
catalyst flow (lbf s/ft2, N s/m2) k14 regenerator stack gas valve flow
FV6 flow through combustion air rating (mol/s psia1/2, kg/s (N/
blower suction valve V6 (lb/s, kg/ m2)1/2)
s) Lrgc length of regenerated catalyst
FV7 flow through combustion air pipe (ft, m)
blower vent valve V7 (lb/s, kg/s) Lsc length of spent catalyst pipe (ft,
FV11 flow through wet gas compressor m)
suction valve V11 (mol/s, molg/s) M polytropic exponent
FV12 flow through valve V12 (mol/s, m* factor for the dependence of
molg/s) the initial catalyst activity on Crgc
FV13 flow through valve V13 (mol/s, MI effective heat capacity of regen-
molg/s) erator mass (Btu/F, KJ/K)
Fwg wet gas production in the reactor Me flowrate of entrained catalyst
(mol/s, molg/s) from dense bed (lb/s, Kg/s)
F1 oil flowrate (lb/s, kg/s) N exponent for the dependence of
F2 oil flow rate (lb/s, kg/s) Ccat on Crgc
F3 fresh feed flowrate (lb/s, kg/s) N* integer value representing a con-
F4 slurry recycle flowrate (lb/s, kg/s) stant for pressure drop on cata-
F5 furnace fuel flowrate (scf/s, m3/s) lyst pipes
F6 combustion air blower through- n quantity of gas (mol)
put (lb/s, kg/s) Patm atmospheric pressure (psia, N/m2)
F7 combustion air flow to the re- Prb pressure at the bottom of the riser
generator (lb/s, kg/s) (psia, N/m2)
F11 wet gas flow to vapor recovery Prgb pressure at the bottom of the
unit (mol/s, molg/s) regenerator (psia, N/m2)
hris height of the riser (ft, m) Pvru discharge pressure of wet gas
K flow coefficient for the slide valve compressor in vapor recovery
(0.7) unit (psia, N/m2)
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 89

P1 combustion air blower suction Ts temperature of stripper outlet (F,


pressure (psia, N/m2) K)
P2 combustion air blower discharge Tsc temperature of spent catalyst en-
pressure (psia, N/m2) tering regenerator (F, K)
P4 reactor pressure (psia, N/m2) T0 temperature of the feed entering
P5 main fractionator pressure (psia, the riser after mixing with the
N/m2) catalyst (F, K)
P6 regenerator pressure (psia, T1 temperature of the fresh feed
N/m2) entering the furnace (F, K)
P7 wet gas compressor suction pres- T2 furnace outlet temperature of the
sure (psia, N/m2) feed (F, K)
Qair enthalpy of air to regenerator T2,ss steady state furnace outlet tem-
(Btu/s, J/s) perature of the feed (F, K)
QC total heat of burning coke (Btu/s, T3 furnace firebox temperature (F,
J/s) K)
Qcatout enthalpy of catalyst out of the tc catalyst residence time in the riser
riser (Btu/s, J/s) (s)
Qe total heat lost from regenerator to UAf furnace overall heat transfer
environment (Btu/s, KJ/s) coefficient (Btu/s/F, J/s/K)
Qfg enthalpy of stack gas exiting the Vcomb,d combustion air blower discharge
regenerator (Btu/s, J/s) system volume (ft3, m3)
QH enthalpy of hydrogen to regen- Vcomb,s combustion air blower suction
erator (Btu/s, J/s) system volume (ft3, m3)
Qin enthalpy into regenerator, reac- Vreg,g regenerator volume occupied by
tor(Btu/s, J/s) gas (ft3, m3)
Qout enthalpy out of regenerator, re- v catalyst velocity in spent/regener-
actor(Btu/s, J/s) ated pipe (ft/s, m/s)
Qloss heat loss from furnace (Btu/s, J/s) vrgc velocity of regenerated catalyst
Qrgc enthalpy of regenerated catalyst (ft/s, m/s)
(Btu/s, J/s) vris volumetric flowrate in the riser
Qsc enthalpy of spent catalyst (Btu/s, (ft3/s, m3/s)
J/s) vs superficial velocity in the regen-
R universal gas constant (ft3 psia/lb erator (ft/s, m/s)
mol 8R, J/mol 0K) vsc velocity of spent catalyst (ft/s, m/
svsc spent catalyst slide valve position s)
svrgc regenerated catalyst slide valve vslip slip velocity (ft/s, m/s)
position V6, V7, V8, V9, V11, position of the corresponding
Tair temperature of air entering re- V12, V13, V14 valves (0 /1)
generator (F, K) Wc inventory of carbon in the regen-
Tatm atmospheric temperature (F, K) erator (lb, kg)
Tbase base temperature (F, K) Wr inventory of catalyst in the reac-
Tcomb,d combustion air blower discharge tor (stripper) (lb, kg)
temperature (F, K) Wreg inventory of catalyst in the re-
Tcyc regenerator stack gas temperature generator (lb, kg)
at cyclone (F, K) Wris inventory of catalyst in the riser
Tdiff temperature difference between (lb, kg)
cyclone and regenerator bed XCO molar ratio of CO to air (mol CO/
temperature (F, K) mol air)
Tlm furnace logarithmic mean tem- XCO,sg molar ratio of CO to air in stack
perature (F, K) gas (mol CO/ mol air)
Tr temperature of reactor riser outlet XCO2 molar ratio of CO 2 to air (mol
(F, K) CO2/mol air)
Tref base temperature for energy bal- XCO2,sg molar ratio of CO2 to air in stack
ance (F, K) gas (mol CO2/ mol air)
Treg temperature of regenerator bed XN molar ratio of N2 to air (mol N2/
(F, K) mol air)
90 M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91

XO2 molar ratio of O2 to air(mol O2/ rB volume fraction of catalyst (lb/ft3,


mol air) kg/m3)
XO2,sg molar ratio of O2 to air in stack rc,dilute density of catalyst in the dilute
gas (mol O2/ mol air) (disengaging) phase (lb/ft3, kg/m3)
yf mass fraction of feed oil rc,dense density of catalyst in the dense
yg mass fraction of gasoline phase (lb/ft3, kg/m3)
z distance along riser (ft, m) rg density of exit gas (lb/ft3,
za dimensionless distance along kg/m3)
riser rpart settled density of catalyst (lb/ft3,
zbed dense bed height (ft, m) kg/m3)
zcyc height of cyclone inlet (ft, m) rris average density of material in the
ztop height of O2 and CO measure- riser (lb/ft3, kg/m3)
ment point (ft, m) rv vapor density at riser conditions
a catalyst deactivation constant (lb/ft3, kg/m3)
(s 1) tfb furnace firebox constant (Btu/F,
DHf heat of cracking (Btu/lb, KJ/kg) J/K)
DHevp heat of vaporizing the feed oil tfo furnace time constant (s)
(Btu/lb, KJ/kg)
DHfu heat of combustion of furnace
fuel (Btu/SCF, KJ/m3)
DHH heat of combustion of hydrogen
(Btu/lb, KJ/kg)
DH1 heat of formation of CO (Btu/ References
mol, J/mol)
DH2 heat of formation of CO2 (Btu/ [1] A.A. Avidan, R. Shinnar, Development of catalytic cracking
mol, J/mol) technology. A lesson in chemical reactor design, Ind. Eng. Chem.
DPfrac pressure drop across reactor main Res. 29 (1990) 931 /942.
[2] Advanced Process Control Handbook VII, Hydrocarbon Proces-
fractionator (psi, N/m2)
sing, 1992, May, pp. 122 /126.
DPRR pressure difference between re- [3] A. Arbel, Z. Huang, I. Rinard, R. Shinnar, Dynamics and control
generator and reactor (psi, N/m2) of fluidized catalytic crackers. 1. Modeling of the current
DPsv pressure drop on regenerated/ generation of FCCUs, Ind. Eng. Chem. Res. 34 (1995) 1228 /
spent catalyst slide valve (psi) 1243.
DPsv,rgc pressure drop on regenerated [4] A. Arbel, Z. Huang. I. Rinard, R. Shinnar, Partial Control of
FCC Units: Input Multiplicities and Control Structures, AIChE
catalyst slide valve (psia, N/m2) Annual Meeting, 1993, St. Louis, MO.
DPelb,rgc pressure drop on different ele- [5] R.C. McFarlane, R.C. Reineman, J.F. Bartee, C. Georgakis,
ments of regenerated catalyst pipe Dynamic simulator for a model IV fluid catalytic cracking unit,
(psia, N/m2) Computers Chem. Eng. 17 (1993) 275 /300.
DPsv,sc pressure drop on spent catalyst [6] E.A. Emad, S.E.H. Elnashaie, Nonlinear model predictive
control of industrial type IV fluid catalytic cracking units for
slide valve (psia, N/m2) maximum gasoline yield, Ind. Chem. Eng. Res. 36 (1997) 389 /
DPelb,sc pressure drop on different ele- 398.
ments of spent catalyst pipe (psia, [7] L. Karla, C. Georgakis, Effect of process nonlinearity on
N/m2) performance of linear model predictive controllers for environ-
mentally safe operation of a fluid catalytic cracking unit, Ind.
oe effective void fraction in regen-
Eng. Chem. Res. 33 (1994) 3063 /3069.
erator dense phase bed [8] M.V. Cristea, S.P. Agachi, Dynamic simulator for a UOP model
of apparent void fraction in regen- fluid catalytic cracking unit, Studia Universitatis ‘‘Babeş-Bolyai’’,
erator dense phase bed Ser. Chemia 42 (1997) 97 /102.
f0 initial catalyst activity at riser [9] V.W. Weekman, Jr., D.M. Nace, Kinetics of catalytic cracking
inlet selectivity in fixed, moving, and fluid bed reactors, AIChE J. 16
(1970) 397 /404.
u dimensionless temperature in the [10] S.M. Jacob, B. Gross, S.E. Voltz, V.W. Weekman, Jr., A lumping
riser and reaction scheme for catalytic cracking, AIChE J. 22 (1976)
l ratio of mass flowrate of disper- 701 /713.
sion steam to mass flowrate of [11] H. Rhemann, G. Schwartz, T. Badgwell, M. Darby, D. White,
feed oil On-line FCCU advanced control and optimization, Hydrocarbon
Processing, 1989, June.
hp polytropic efficiency [12] G. McDonald, B. Harkins, Maximizing Profits by Process
rairg density of air at regenerator con- Optimization, in: NPRA Annual Meeting, 1987, March 29 /31,
ditions (lb/ft3, kg/m3) San Antonio, Texas.
M.V. Cristea et al. / Chemical Engineering and Processing 42 (2003) 67 /91 91

[13] M. Hovd, S. Skogestad, Controllability analysis for the fluid [19] D. Semino, C. Scali, A method for robust tuning of linear
catalytic cracking process, in: AIChE Annual Meeting, 1991, quadratic optimal controllers, Ind. Eng. Chem. Res. 33 (1994)
November 17 /22, Los Angeles. 889 /895.
[14] M. Morari, I. Huq, S.P. Agachi, J. Bomberger, B. Donno, A. [20] H. Chen, F. Allgower, Nonlinear model predictive control
Zengh, Modeling and control studies on model IV FCCUs, Joint schemes with guaranteed stability, NATO Advanced Study
Research between Chevron Research and Technology Company Institute on Nonlinear Model Based Process Control, 1997, 10 /
and California Institute of Technology, 1993. 20 August, Antalya, Turkey, pp. 1 /28.
[15] S. Yang, X. Wang, C. McGreavy, A multivariable coordinated [21] J.H. Lee, Z.H. Yu, Tuning of model predictive controllers for
control system based on predictive control strategy for FCC robust performance, Comp. Chem. Eng. 18 (1994) 15 /37.
reactor /regenerator system, Chem. Eng. Sci. 51 (1996) 2977 / [22] K.J. Astrom, T. Hägglund, Automatic Tuning of PID Control-
2982. lers, Instrument Society of America, 1988.
[16] M.V. Cristea, V. Marinoiu, S.P. Agachi, Multivariable model [23] A.A. Patwardhan, J.B. Rawlings, T.F. Edgar, Nonlinear model
based predictive control of a UOP fluid catalytic cracking unit, in: predictive control, Chem. Eng. Comm. 87 (1990) 123 /141.
2nd Conference on Process Integration, Modeling and Optimiza- [24] J. Richalet, Industrial applications of model based predictive
tion for Energy Saving and Pollution Reduction, 1999, 31 May /2 control, Automatica 29 (1993) 1251 /1274.
June, Budapest, pp. 223 /228. [25] E. Lee, F.R. Groves, Jr., Mathematical model of the fluidized bed
[17] M.V. Cristea, S.P. Agachi, Controllability analysis of a model IV catalytic cracking plant, Trans. Soc. Comp. Simulation 3 (1985)
FCCU, in: 12th International Congress of Chemical and Process 219 /236.
Engineering CHISA, 1996, Prague, p. 10. [26] Y.T. Shah, G.P. Huling, J.A. Paraskos, J.D. McKinney, A
[18] C.E. Garcia, M.P. Prett, M. Morari, Model predictive control: kinematic model for an adiabatic transfer line catalytic cracking
theory and practice */a survey, Automatica 25 (1989) 335 / reactor, Ind. Eng. Chem., Process Des. Dev. 16 (1977) 89 /
348. 94.