You are on page 1of 17

Computers and Chemical Engineering 164 (2022) 107900

Contents lists available at ScienceDirect

Computers and Chemical Engineering


journal homepage: www.elsevier.com/locate/compchemeng

An open source fluid catalytic cracker - fractionator model to support


the development and benchmarking of process control, machine
learning and operation strategies
Omar Santander a, Vidyashankar Kuppuraj b, Christopher A. Harrison b, Michael Baldea a ,c ,∗
a
McKetta Department of Chemical Engineering, The University of Texas at Austin, Austin, TX 78712, United States
b
Marathon Petroleum Corporation, Garyville, LA 70051, United States
c
Institute for Computational Engineering and Sciences, The University of Texas at Austin, Austin, TX 78712, United States

a r t i c l e i n f o a b s t r a c t

Article history: The present study describes the capabilities of an open source integrated framework for the dynamic
Received 7 January 2022 simulation and control of a refining operation, comprising a fluidized bed catalytic cracker (FCC) and a
Revised 16 May 2022
fractionator. A detailed cracking reaction scheme that yields realistic predictions of product distributions
Accepted 21 June 2022
is used. In addition, the fractionator is rigorously modeled and solved using the dynamic stagewise adia-
Available online 27 June 2022
batic flash algorithm. A regulatory control layer is also included.
Keywords : Two simulation examples are shown for set point tracking and FCC feed quality disturbance rejection,
Fluidized bed catalytic cracking (FCC) respectively. We expect that the model can serve as a platform for multiple research efforts, including
Data generation algorithm development and benchmarking in fault detection, control and process operations. The model
Mathematical modeling can be downloaded using the following link: https://github.com/Baldea- Group/FCC- Fractionator.
Process control © 2022 Elsevier Ltd. All rights reserved.
System benchmark

1. Introduction tory control is needed. Robust dynamic models are essential to


this end. FCC is a mature technology and its fundamental ele-
Fossil fuels such as gasoline and diesel still account for most ments were developed in the 1930s and 1940s. The process con-
of the road transportation fuel used worldwide. Consequently, flu- sists of three key stages: 1-Oil cracking, 2-Catalyst regeneration
idized bed catalytic cracking (whereby gas oil is cracked into and 3-Product separation (Fig. 1). While several cracker configu-
smaller molecules such as liquified petroleum gas (LPG), naphtha, rations exist (Pinheiro et al., 2012), and there are multiple open
light cycle oil (LCO) etc.) and fractionation (whereby the products source literature models for the cracker, there are few models that
are separated) remain key technologies at the heart of refining op- consider the product separation stage. Consequently, the main dif-
erations. There are more than 400 Fluidized Bed Catalytic Cracker ference between open source models lies in the level of detail in
(FCC) units operating worldwide, with plans to build new ones still which stages 1 and 2 are described.
underway (Pinheiro et al., 2012). These findings motivate the present work, where we present a
As any other chemical or manufacturing process, fuels pro- detailed model of a cracking process including the reactor and sep-
duction has been subject to more restrictive environmental con- arator, and provide the open-source MATLAB(R) code to allow for
straints, volatile international market conditions, supply uncer- this model to be further used in the process systems community
tainty, stringent competition and fast fluctuating customer de- as a platform for other research efforts, including algorithm devel-
mand. In order to meet these challenges, the fuels production pro- opment and benchmarking in fault detection, control and process
cess efficiency must be enhanced. operations.
To realistically assess the effect of advanced process control
(APC) or to enhance the decision making process in, e.g., schedul- 2. Background and contributions of this work
ing, planning and supply chain management, detailed informa-
tion about process performance, including dynamics and regula- Available cracking process models share several common as-
sumptions. The reactor is usually modeled as a series of sub-units
(riser and stripper). In the reactor riser, fresh feed comes in con-

Corresponding author at: McKetta Department of Chemical Engineering, The tact with hot regenerated catalyst supplied from the regenerator. It
University of Texas at Austin, Austin, TX 78712, United States. is commonly assumed that the fresh feed vaporization is instan-
E-mail address: mbaldea@che.utexas.edu (M. Baldea). taneous, that the cracking reactions take place in the riser and

https://doi.org/10.1016/j.compchemeng.2022.107900
0098-1354/© 2022 Elsevier Ltd. All rights reserved.
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Nomenclature Ff g flow of flue gas


FHN flow of HN
Abbreviations Fk flash feed
AP I American petroleum institute gravity FLCO flow of LCO
CAB combustion air blower FLN flow of LN
CST continuous stirred tank FLPG flow of LPG
CV controlled variable FPF R output PFR flow
DAE differential algebraic equation Fr flow of reflux
DSAF dynamic stagewise adiabatic flash Frgc flow of regenerated catalyst
EOS equation of state Fsc flow of spent catalyst
F CC fluid catalytic cracking Fslu flow of slurry
HN heavy naphtha Fst p flow out of reactor stripper
HT C hydraulic time constant Gv superficial mass flow rate
LCO light cycle oil Hl enthalpy liquid
LN light naphtha Hv enthalpy vapor
LP G liquefied petroleum gas HD departure enthalpy
MESH mass-equilibrium-summation-enthalpy H f orm enthalpy of formation
MPC model predictive control H ig enthalpy ideal gas
MV manipulated variable Hl Fk flash feed enthalpy
NBP normal boiling point K rate constant matrix
PC pseudo component Kc controller gain
SP set point Kh aromatic adsorption constant
SRK Soave–Redlich–Kwong Ki equilibrium constant
ODE ordinary differential equation kii binary interaction parameter
PF R plug flow reactor kj reaction rate constant
PI proportional integral kj reaction rate modified constant
T BP true boiling point Kn nitrogen poisoning constant
V GO vacuum gas oil l reactor riser position
W GC wet gas compressor L reactor riser normalized position
Lclip valve position lower clip
Indices
L f ra fractionator accumulator level
i species (pseudo component, light gas or coke)
Lk liquid flow
j reaction
Lrea reactor catalyst inventory
k equilibrium stage
Mk hold up
Sets MW average molecular weight
C all species other than coke MW CST average molecular weight of gaseous mixture in
Ji reactions in which i is the reactant stripper
NC all species N nitrogen concentration
NBPhc NBP of heaviest component
Superscripts
NBPshc NBP of second heaviest component
D departure (from ideal condition)
P pressure
F feed
p controller output
f orm formation
Pc i critical pressure of species i
ig ideal gas
Pf ra fractionator overhead pressure
n time step
pnominal nominal controller output (no error)
Functions Prea reactor pressure
f () aromatic adsorption catalyst deactivation Preg regenerator pressure
g() nitrogen poisoning catalyst deactivation Qcatout entahlpy of catalyst out of the reactor
φ () coke deposition catalyst deactivation Qcracking heat generated from cracking
ν () flow of material through valve Qf f heat needed to raise the feed temperature from
700 F to reactor riser temperature
Notation Qfl heat needed to raise the feed temperature from
ar concentration vector Tpre to 700 F
aCST concentration vector in stripper Qin entahlpy into the reactor
ar i concentration for species i Qout entahlpy out of the reactor
Aj pre-exponential factor Qrgc entahlpy of regenerated catalyst
aPF R concentration vector out of PFR Qslurry enthalpy of slurry
Car aromatics concentration rˆ reaction rate vector
C pi heat capacity R ideal gas constant
e controller error rj reaction rate
Ea j activation energy Rl reactor riser length
f connection frequency Swh true weight hourly space velocity
Fa flow of air t time
Ff flow of fuel

2
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

tion, temperature dependence of the reaction rate constants etc.


T temperature (Oliveira and Biscaia, 1989; Corella and Frances, 1991; Sugungun
Tacc fractionator accumulator temperature et al., 1998; Ancheyta-Juarez et al., 1999; Hagelberg et al., 2002;
Tamb ambient temperature Dupain et al., 2003; Corma et al., 2005; Cerqueira et al., 1997;
tc catalyst residence time in riser T. Takatsuka and Hashimoto, 1987; Pitault et al., 1994; Jacob et al.,
Tc i critical temperature of species i 1976; den Hollander et al., 2001; Fusheng et al., 2016; Hernandez-
T f ra fractionator overhead temperature Barajas et al., 2009; Shayegh et al., 2012; Xu and Chu, 2006). Intu-
Thnt HN 98% cut point itively, the accuracy of the overall process model increases when a
THT temperature of heavy tail detailed reaction network that explicitly accounts for temperature
Tlcot LCO 98% cut point dependence, catalyst deactivation and provides a realistic descrip-
Tpre preaheater output temperature tion of product distribution is used.
Trea reactor temperature The stripper/disengager (the reactor subunit in which the spent
Treg regenerator temperature catalyst is separated from the cracked gases) is usually modeled as
Tr i reduced temperature a continuously stirred tank (CST) assuming perfect mixing, but in-
Uclip valve position upper clip cluding dynamics (Ali et al., 1997; Malay et al., 1999; Secchi et al.,
Vk vapor flow 2001; Hernandez-Barajas et al., 2006; Bollas et al., 2007). Similar
Vstripper stripper volume to the riser, detailed hydrodynamic studies can be carried out us-
V al ve valve position ing CFD tools (Gao et al., 2008; T. Gauthier, 2000; Wiens, 2010).
V al venominal nominal valve position The regenerator is a complex unit where air is used to burn
x∗ process states around valve the coke deposited on the catalyst as part of the cracking reac-
xi liquid composition tions. Regenerated catalyst is constantly recirculated to the reactor
xhc composition of heaviest component riser. The regenerator is typically modeled as a two region unit (Ali
xshc composition of second heaviest component et al., 1997; Malay et al., 1999; Han and Chung, 2001b; Arandes
yi vapor composition et al., 20 0 0; McFarlane et al., 1993; Secchi et al., 2001): a dense
yst p composition out of reactor stripper region, where most of the exothermic combustion reactions take
Z compressibility factor place, and a dilute region, consisting mainly of exhaust gases (CO
zk,i flash feed composition and CO2 ) where combustion reactions still occur due to catalyst
entrainment. Furthermore, it is common to divide the dense re-
Greek
gion into a bubble phase and a dense phase (a higher density with
α linear mapping between controller output and valve
respect to bubble phase is due to the substantial presence of solid
position
catalyst). The bubble phase is characterized by reactants and prod-
β coke deposition constant
ucts (gases) moving in plug flow and the dense phase is character-
γ coke deposition exponential power
ized by perfect mixing of the catalyst and the gas phase. Mass and
 maximum absolute SP change at time step (n)
energy transfer are typically considered within the fluidized bed.
P pressure drop
Fractionation is the process in which valuable products (e.g.
t discretization time step
LPG, naphtha) are obtained by separating the cracked oil vapors
 catalyst void fraction
coming from the reactor according to their normal boiling point
η relative valve position power
(NBP). Open source literature references on fractionator modeling
ρ density of gaseous mixture
are scarce (Chung and Riggs, 1995; Huang, 20 0 0). To the knowl-
ρc density of catalyst
edge of the authors, there is no open source literature reference
τCST stripper time constant
that rigorously considers and integrates the FCC-Fractionator pro-
τI controller integral time constant
cess. Motivated by the above, we propose an open source inte-
τk hydraulic time constant
grated model. The main contributions of this work are summarized
τrea reactor time constant
as follows:
υi molar volume
φl i fugacity liquid • Development of a rigorous integrated open source FCC-
φv i fugacity vapor Fractionator dynamic model that can be used for improved de-
ωi acentric factor cision making, machine learning, data generation/analysis etc.
• Incorporation of an industry-relevant regulatory control struc-
ture that for the key process variables
that the residence time of the feed is sufficiently small to allow
• Consideration of relevant process disturbances (e.g. feed quality,
for describing the dynamics using a quasi-steady-state assumption
ambient temperature) that enhance the model prediction capa-
(Pinheiro et al., 2012; Arbel et al., 1995; Cristea et al., 2003; Yescas
bilities
et al., 1995; Ellis et al., 1998; Zheng, 1994). Therefore, the reac-
• A computational implementation in MATLAB(R) is made avail-
tor riser is usually modeled as a (1-D) process unit at steady state.
able.
There are more complex models that describe in detail the riser
by means of computational fluid dynamics (CFD) (Pinheiro et al., The paper is organized as follows: in Section 3 the FCC-
2012; Lan et al., 2009; Lopes et al., 2011; Das et al., 2003) or as Fractionator model and regulatory control structure are described,
a sequence of subunit stages (e.g mixing and reaction zones) (Han in Section 4 the algorithm used to simulate the dynamic inte-
and Chung, 2001a; Sildir et al., 2015). grated model is explained, in Section 5 the results of the integrated
The kinetics of the cracking reactions are complex. Initial stud- model responses to set point tracking and disturbance rejection are
ies represented the kinetics using empirical correlations for the shown, Section 6 presents conclusions and recommendations.
distribution of products, accounting for a limited number of ag-
gregate species (“lumps”) that represent chemical species within, 3. Model description
e.g., certain molecular weight ranges (Weekman, 1968; Lee et al.,
1989). Later works increased the model complexity by accounting The Fluid Catalytic Cracking-Fractionator dynamic model
for more lumps, as well as taking into account catalyst deactiva- is based on the FCC Amoco model type IV presented by

3
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 1. FCC-Fractionator process diagram. The green (dashed) arrows represent the input of material, whereas the blue (solid) arrows represent the output of material. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

McFarlane et al. (1993) and its MATLAB implementation de- 3.1.1. Kinetic model
veloped by Ali (2004). To make the model predictions more The model is based on the results of Xu and Chu (2006) which
accurate and realistic, several major modifications are made: were further expanded to include pseudo-components and a more
detailed representation of the distribution of light gases. Sim-
• The reactor model and kinetic scheme are expanded to reflect ilar to the original model (Xu and Chu, 2006), in the kinetic
a much wider range of products scheme introduced here, the feed is treated as one of the lumped
• A rigorous dynamic fractionator model is included species. A representation of the kinetic scheme is shown in Fig. 2.
• A regulatory control structure is incorporated It is assumed that feed vacuum gas oil (VGO), with an aver-
The FCC-Fractionator system consists of the following units: age molecular weight of 446.00 g/mol, is cracked into lighter
preheater, reactor, fractionator, wet gas compressor (WGC), regen- molecules (pseudo-components, light gases) and coke. The four
erator, combustion air blower (CAB), lift air blowers, catalyst circu- heavier pseudo-components (PC1-PC4) are cracked further, produc-
lation lines and process controllers. The flowsheet of the process is ing lighter molecules and coke. Pseudo-component 5 (PC5 or C5+)
shown in Fig. 1. and the light gases are cracked into lighter molecules, but without
generating coke.
All reactions in the kinetic scheme are assumed to be elemen-
3.1. Reactor modeling
tary irreversible first order reactions (while there are some works
in which a subset of the cracking reactions is considered second
The reactor model is based on the work by
order ( Hagelberg et al., 2002; Ancheyta-Juarez et al., 1999), there
McFarlane et al. (1993). The original model was extended by incor-
exists fairly broad agreement that first order reactions are suf-
porating a detailed kinetic scheme in which pseudo-components
ficient to accurately describe the cracking kinetics (Xu and Chu,
and light gases (e.g. methane, ethane) were used to describe the
20 06; Dupain et al., 20 03; den Hollander et al., 2001; den Hollan-
final product (liquefied petroleum gas (LPG), light naphtha (LN),
der et al., 1999). Consequently, any cracking reaction (j) involving
heavy naphtha (HN), light cycle oil (LCO) and slurry) compo-
species (i) is modeled as:
sition and properties. Furthermore, the kinetic model includes
dependence on reactor conditions (temperature, pressure, catalyst r j = −kj ρc (ρ ar i )/ ∀ j : j = 1 . . . 50, i ∈ Ji (1)
residence time). The model of McFarlane et al. (1993) was also
expanded to take into account a riser section (described as a where the set Ji encompasses the reactions (j) in which (i) is a re-
steady state plug flow reactor (PFR)) and a stripping section (mod- actant.
eled as a continuous stirred tank (CST)). We discuss below the It is commonly assumed that rate constants (kj ) decay with
modifications made to the model of McFarlane et al. (1993), and time due to catalyst deactivation (Corella and Frances, 1991;
we kindly direct the reader to the original paper for details on the Ancheyta-Juarez et al., 1999; Dupain et al., 2003; Corma et al.,
initial implementation. 2005; den Hollander et al., 2001). Catalyst deactivation can in prin-

4
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 2. Schematic of kinetic model. Each colored arrow represents the irreversible reaction in which the reactant (i) is cracked into a lighter species (i’). Note that there is a
different color for each subset of reactions that share the same reactant (i).

ciple be caused by three different phenomena: aromatic adsorp- By combining (1), (2) and (7), the reaction rate can be ex-
tion, coke deposition and nitrogen poisoning. Therefore, reaction pressed in vector form as:
rate constants are expressed as: 
rˆ = f (Car )φ (tc )g(N )(ρc / )(Prea /RTrea )(1/ ar i )Kar (8)
kj = k j f (Car )φ (tc )g(N ) (2) i∈Nc
da da da
where rˆ = [ dt1 dt2 . . . dt11 ]T is the vector of reaction rates, ar is the
k j = A j exp(−Ea j /RTrea ) (3) vector of concentrations (moles of i/Kg gas) and K is the rate con-
stant matrix. In the present work, aromatic compounds and nitro-
where k j represents the Arrhenius reaction constant. Aromatic ad-
gen are not considered in the kinetic model, so the vectorial kinetic
sorption, coke deposition and nitrogen poisoning effects are re-
Eq. (8) is simplified to:
spectively described by the functions f (Car ), φ (tc ) and g(N ) as fol-

lows: rˆ = φ (tc )(ρc / )(Prea /RTrea )(1/ ar i )Kar (9)
f (Car ) = 1/(1 + KhCar ) (4) i∈Nc

The values and units of the parameters of the kinetic model


above can be found in Appendix A, along with the structure of K.
φ (tc ) = 1/(1 + β tc γ ) (5)
3.1.2. Reactor mass and energy balances
For the purpose of defining the mass balance, the reactor is
g(N ) = 1/(1 + Kn N ) (6) modeled as two sequential sections: i) the riser, in which the
cracking reactions take place (which is represented as a PFR) and
Assuming an ideal gas behavior, the average molecular weight
ii) the stripper, where the gaseous cracked components are sep-
of the fluidized mixture can be obtained by:
arated from the catalyst. The stripper is modeled as a well-mixed

MW = 1/ ar i (7) continuously stirred tank (CST), and it is assumed that no reactions
i∈Nc
take place in this section.

5
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

The energy balance is assumed to follow the CST dynamics as in bi = 0.08664RTc i /Pc i (19)
McFarlane et al. (1993) where the enthalpy of the input and output
streams, and cracking reactions, is considered.
Riser model The riser is modeled by the continuity equation in αi = (1 + (0.48 + 1.574ωi − 0.176ωi2 )(1 − Tr 0i .5 ))2 (20)
which the kinetic model (9) is incorporated. In addition, given the where Tr i = T /Tc i is the reduced temperature. The SRK EOS can also
fast dynamics of the riser and the comparatively much larger time be written in polynomial form as:
constant of the fractionator (to be discussed later), a steady state
behavior is assumed. Consequently, the riser model is given by: 0 = Z 3 − Z 2 + Z (Ai − Bi − B2i ) − Ai Bi (21)

da r i
Gv = rj i ∈ Ji (10)
dl Ai = αi ai P/R2 T 2 (22)
By combining (9) and (10), the following vector-form equa-
tion is obtained:
dar  Bi = bi P/RT (23)
= φ (tc )(Prea /Swh RTrea )(1/ ar i )Kar (11)
where Z = PV/RT is the compressibility factor. By using mixing
dL
i∈Nc
rules (Sandler, 2006), the polynomial equation for a multicompo-
where L = l/Rl and Swh = Gv  /ρc Rl represent the normalized posi- nent mixture is:
tion and true weight hourly space velocity respectively (Rl is the
length of the riser). 0 = Z 3 − Z 2 + Z (A − B − B2 ) − AB (24)
Stripper model The output stream of the reactor riser serves as
a feed stream for the reactor stripper. It is assumed that the feed
A= α am P/R2 T 2 (25)
stream is well mixed, and that the spent catalyst settles in this
section. Hence, the stripping section is modeled as:
daCST B = bm P/RT (26)
= (aPF R − aCST )/τCST aPF R , aCST ∈ C (12)
dt
The time constant (τCST ) is computed as: 
α am = xi xi α aii (27)
τCST = (PreaVstripper MW CST )/(RTrea FPF R ) (13)
i∈C i ∈C
Energy balance The energy balance follows the model developed
in McFarlane et al. (1993). It is summarized here for completeness: α aii = (αi ai αi ai )0.5 (1 − kii ) (28)

dTrea 
τrea = Qin − Qout (14) bm = xi bi (29)
dt
i∈C

Qin = Qrgc + Q f l (15) where kii represents an empirical binary interaction parame-
ter. Since this work is based on theoretical pseudo-components,
the effect of kii is neglected. The equilibrium constants are ob-
Qout = Qcatout + Qslurry + Qcracking + Q f f (16) tained by implementing the fugacity (φ − φ ) method as follows
(Sandler, 2006):
The default configuration of our model assumes that slurry is
not recirculated (implying Qslurry = 0) since it is an unusual indus- Ki = φl i / φv i ∀ i ∈ C (30)
trial practice (performed, for example, at start-up). Nonetheless, where φl i and φv i represent the fugacity coefficients of component
this feature can easily be included in the code provided along with i in the liquid and vapor phases, respectively. The fugacity coeffi-
the paper. Case in point, Appendix D details an additional simula- cients are computed using the SRK EOS closed form equation:
tion in which slurry recirculation is present.
A detailed description of the numerical integration strat- lnφi = (bi /bm )(Z − 1 ) − ln(Z − B )
egy used to solve the reactor energy and mass balance equa- 
− (A/B )((2 α aii /α am ) − (bi /bm ))ln(1 + B/Z ) (31)
tions (along with those of the other units in the system) is pro-
i ∈C
vided later in the paper.
where Z represents one of the roots of Eq. (24). The roots of the
3.2. Fractionator modeling polynomial equation can be computed using any general polyno-
mial roots solver. The smallest root represents the liquid phase
The fractionator model comprises the thermodynamic model compressibility factor (Zl ) whereas the largest root represents the
and the mass, equilibrium, summation and heat (enthalpy) (MESH) vapor phase compressibility factor (Zv ).
equations. We also discuss the numerical approach that was used The departure enthalpy is computed using the SRK EOS as fol-
for solving the model. lows (Sandler, 2006):
H D = RT (Z − 1 ) + (1/(BRT /P ))ln(1 + B/Z )
3.2.1. Thermodynamic model
d(A(RT )2 /P )
The Soave–Redlich–Kwong cubic equation of state (SRK EOS), × [T − (A(RT )2 /P )] (32)
was selected to describe phase equilibrium at the desired operat- dT
ing conditions (Asp, 2005). For a pure component (i), the SRK EOS where the appropriate compressibility factor is used to compute
can be written as (Sandler (2006): the liquid or vapor departure enthalpy (HlD or HvD respectively).
Finally, the liquid or vapor enthalpies (Hl or Hv respectively) are
P = (RT /υ i − bi ) − (ai αi /υ i (υ i + bi )) (17)
computed as follows:
 T

ai = 0.42747R 2
Tc 2i /Pc i (18) Hiig = Hif orm + C pi (T )dT (33)
T re f

6
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 3. Fractionator diagram: LCO: light cycle oil, HN: heavy naphtha, LN: light naphtha, LPG: liquefied petroleum gas.

  the material and energy balances for a general stage (k) are estab-
Hlig = xi Hiig or Hvig = yi Hiig (34)
lished as follows:
i∈C i∈C
dMk
= Lk−1 + Vk+1 − Vk − Lk (36)
D
Hv = Hv + Hv ig
or Hl = HlD + Hlig (35) dt

dMk xk,i
3.2.2. MESH equations = Lk−1 xk−1,i + Vk+1 yk+1,i − Vk yk,i − Lk xk,i (37)
dt
The fractionator schematic diagram is shown in Fig. 3. It fea-
tures two pump-arounds (stage 20 to 18 and stage 11 to 9), a con-
denser (accumulator) and two side product streams – heavy naph- dMk Hl k
= Lk−1 Hl k−1 + Vk+1 Hv k+1 − Vk Hv k − Lk Hl k (38)
tha (HN) and light cycle oil (LCO). The following assumptions are dt
made for modeling the fractionator:
• Each tray and the accumulator are considered ideal equilibrium yk,i = Kk,i (Tk , Pk , xk,i )xk,i ∀i (39)
stages (stages are numbered from top to bottom, with the ac-
cumulator being stage 1 and the bottoms being stage 20) 
• Vapor hold up is negligible 1= yk,i (40)
• Pressure drop (P ) per stage is constant (Luyben, 2012) i∈C
• The dynamics of the liquid phase on each stage k can be de-

scribed using a stage-specific but time invariant hydraulic time 1= xk,i (41)
constant, τk (Huang, 20 0 0) i∈C

A schematic diagram that shows the flows of material around a The MESH Eqs. (36)–(41) are appropriately modified in the
general stage is shown in Fig. 4. With this flow structure in mind, model to account for the required flow structure (feed, side or

7
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

ynk,i+1 = Kk,i (Tkn+1 , Pkn+1 , xnk,i+1 )xnk,i+1 ∀i (49)

n+1 n+1
Hl nk +1 = HlD k + Hlig k (50)

n+1 n+1
Hv nk +1 = HvD k + Hvig k (51)


1= ynk,i+1 (52)
i∈C


1= xnk,i+1 (53)
i∈C

Mkn+1 = τk Lnk +1 (54)


This set of nonlinear equations is solved sequentially for each
stage (k) at each time step n to approximate the dynamic behavior
of the column. The computation of the enthalpy (50) and (51) fol-
lows the developments of the previous Section 3.2.1. The column
model can be solved starting either from the bottom (k = 20) or
from the top (k = 1). To avoid introducing artificial delays in the
Fig. 4. General stage diagram. The green dashed arrows represent the input flow of dynamics of the process (particularly for columns with a large
material to stage k, whereas the solid blue ones represent the output flow of mate- number of equilibrium stages), at each time step n the column
rial from stage k. (For interpretation of the references to color in this figure legend,
solution sequence is reversed with respect to the previous time
the reader is referred to the web version of this article.)
step (n − 1) solution sequence (e.g. at step n − 1 the solution se-
quence is from top to bottom and at step n the solution is pur-
pump-around streams etc.). The dynamics of the liquid phase sued from bottom to top). If this measure were not implemented,
are modeled using the hydraulic time constant (HTC) approach the last stage of the solution (and the corresponding column stage)
(Huang, 20 0 0): would constantly be affected by the sequential computation of the
1 process, creating an artificial time delay.
Mk = Lk ∀k (42)
τk
3.3. Regulatory control structure
3.2.3. Solution of the MESH equations
A set of PI controllers were modeled to simulate the regulatory
To solve the fractionator model, we used the dynamic stagewise
control layer. Controllers were tuned using the Tyreus–Luyben tun-
adiabatic flash (DSAF) algorithm (Chung and Riggs, 1995; Huang,
ing rules (Seborg et al., 2016). Where relevant, valve models were
20 0 0). The algorithm considers the column as a stack of single-
also incorporated along with the controllers. The regulatory control
stage flash units; each stage is solved sequentially as a dynamic
structure is summarized in Table 1.
adiabatic flash. A schematic representation of a general stage (k)
A mapping exists between every manipulated variable (MV) and
regarded as a adiabatic flash is shown in Fig. 5. Note that the liquid
its valve position. The valve position spans values from 0 to 100%
and vapor feeds to stage (k) are combined into a single pseudo-
(from fully closed to fully open, respectively). The ranges of the
feed whose properties can be computed as:
positions of some of the valves are smaller than this range to rep-
Fk = Lk−1 + Vk+1 (43) resent realistic operation. The generic relationship (mapping) be-
tween MV, valve position and controller output for a valve is given
by:
zk,i = (Lk−1 xk−1,i + Vk+1 yk+1,i )/Fk ∀i (44)  t
p (t ) = pnominal + Kc (e(t ) + e(t  )/τI dt  ) (55)
0
Hl Fk = (Lk−1 Hl k−1 + Vk+1 Hv k+1 )/Fk (45)
The discretized (using backward finite differences) MESH equa- p∗ (t ) = max(Lclip , p (t )) (56)
tions of the adiabatic flash (k), and HTC relationship for a given
time step n, where n is defined as the positive integer that maps
the time (t) with t (t = nt ), are a set of nonlinear equations as p(t ) = min(Uclip , p∗ (t )) (57)
follows:
(Mkn+1 − Mkn )/t = Fkn+1 − Vkn+1 − Lnk +1 (46)
Valvenominal = α pnominal (58)

(xnk,i+1 − xnk,i )/t = (Fkn+1 (zk,i


n+1
− xnk,i+1 ) − Vkn+1 (ynk,i+1 − xnk,i+1 ))/Mkn+1 (47)
Valve(t ) = α p(t ) (59)

n+1
(Hl nk +1 − Hl nk )/t = (Fkn+1 (Hl Fk − Hl nk +1 ) − Vkn+1 (Hv nk +1 − Hv nk +1 ))/Mkn+1
MV (t ) = (Valve(t )/Valvenominal )η ν (x∗ (t )) (60)
(48)

8
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 5. Adiabatic flash diagram. The green dashed arrow represents the input flow of material, whereas the solid blue arrows represent the output flow of material. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 1
Control structure (Valve 5 (V5 ) is used as a flare valve as shown in Fig. 1).

# Controller Controlled variable (CV) Valve Manipulated variable

1 T C1 Tpre (Preh. output stream) V1 (Preh. temperature) Ff (Flow of fuel)


2 T C2 Trea (Reactor temperature) V2 (Reactor temperature) Frgc (Flow of regen. cat.)
3 LC1 Lrea (Reactor inventory) V3 (Reactor inventory) Fsc (Flow of spent. cat.)
4 T C3 Treg (Regenerator temperature) V6 (Regenerator temperature) Fa (Flow of air)
5 PC1 Preg (Regenerator pressure) V7 (Regenerator pressure) Ff g (Flow of flue gas)
6 PC2 Pf ra (Overhead fractionator pressure) V4 (Frac. pressure) FLPG (Flow of LPG)
7 LC2 L f ra (Accumulator level) V9 (Accumulator level) FLN (Flow of LN)
8 T C4 T f ra (Overhead fractionator temperature) V8 (Fractionator temperature) Fr (Flow of reflux)
9 T C5 Thnt (HN 98% cut point (temperature)) V10 (HN flowrate) FHN (Flow of HN)
10 T C6 Tlcot (LCO 98% cut point (temperature)) V11 (LCO flowrate) FLCO (Flow of LCO)

Eq. (55) is the time domain expression of the PI controller. Table 2


Operating region.
Eqs. (56) and (57) are saturation relations that bound the variation
of the controller output ( p (t )). Eqs. (58) and (59) linearly map (by # Controller CV Min Max  Unit
using factor α ) the controller output with the corresponding valve 1 T C1 Tpre 609 629 2.5 F
position. Eq. (60) relates the MV and valve position, where η is a 2 T C2 Trea 962 975 2 F
positive integer and ν (x∗ ) is a function of the process states (x∗ (t )) 3 LC1 Lrea 94,500 98,500 500 lb
around that valve that determines the flowrate of material. 4 T C3 Treg 1242 1256 2 F
5 PC1 Preg 27.5 29.2 0.5 psig
The proposed regulatory control structure follows industrial
6 PC2 Pf ra 24.6 26 0.5 psig
practice and regulates the most relevant operating variables. 7 LC2 L f ra 55 75 5 kmol
Nonetheless, this is not the only feasible control structure, and 8 T C4 T f ra 242.33 253.13 2 F
we invite the readers to consider new regulatory and/or advanced 9 T C5 Thnt 520 542 5 F
10 T C6 Tlcot 740 761 5 F
process controll structures and designs. To this end, we note that
the operating region that the model and its control structure were
tested in is summarized in Table 2.
The process can be simulated outside this operating region. reach undesirable conditions (e.g. fractionator column trays losing
However, there is a possibility that for a particular set of operating liquid and becoming “dried up”). In addition, the () in Table 2 is
conditions one or more valves reach saturation or that the process defined as the maximum absolute change that the respective vari-

9
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 6. Crude oil true boiling point (TBP) curve.

able can undergo at a given time step n that is, the rate of change 3.4. Preheater modeling
limit for the variable. The inclusion of this rate of change follows
from industry operating practices, whereby the rate of change of The preheater model is presented in Appendix C, which also de-
specific variables is limited when traversing the operating region. scribes the way the impact of the feed quality (API) and ambient
In addition, the rate of change is an important consideration when temperature on system performance is modeled.
setting up advanced control structures such as model predictive
control (MPC). 4. FCC-fractionator model integration
The reactor inventory and accumulator level (Lrea and L f ra re-
spectively) can be mapped (e.g. linearly) to their corresponding The full model of the system is a large-scale, highly nonlinear
processing unit % level. The selection of Lrea (lb) and L f ra (kmol ) as and stiff system of differential algebraic equations (DAEs). With the
CV follows from the reactor inventory balance and the accumulator intent of developing a robust simulation tool, we employ a sequen-
mass balance respectively. tial solution strategy.
The quality (“98% cut point”) of the product streams (HN and To this end, the integrated model was divided into two sub-
LCO) is regulated by the temperature controllers T C5 and T C6 re- models. The first submodel contains all the FCC units (preheater,
spectively. These controllers implicitly regulate the composition of reactor, regenerator and CAB), the WGC and their respective con-
the HN and LCO streams (by considering the normal boiling point trollers, whereas the second submodel contains the remaining unit
(NBP) of the heavy tail of each stream respectively). In this context, (i.e., the fractionator and its controllers). The WGC was included in
the heavy tail is defined as the last 2% of evaporation volume that the FCC subsystems since it leads to faster solutions (otherwise the
contains the heaviest components in a given stream. To exemplify integration step for the fractionator must be reduced to guarantee
this concept, Fig. 6 illustrates the true boiling point (TBP) curve of numerical stability, increasing the communication frequency be-
a given crude oil (Watkins, 1979). The figure shows that each frac- tween the submodels and ultimately increasing the solution time).
tion of the crude oil (e.g. fuel gas, naphtha) has an associated a The information communicated between the submodels con-
crude volume % and a “cut point” temperature (the temperature sists of i) the feed to the fractionator (reactor stripper output) and
value at which the dashed green line corresponding to each frac- ii) the LPG feed to the WGC (since slurry is not being recirculated
tion intersects with the TBP curve). Then, the last 2% volume is to the reactor). To guarantee the numerical stability of the inte-
the “98% cut point” where 98% of the crude oil has been distilled. grated model, first order filters are applied to the data exchanged
This concept is applied to the fractionator products (HN and LCO) at both connection points i) and ii). For node (i) the temperature,
to define their quality. pressure, flowrate and composition are filtered and for node (ii)
For simplicity, the fractionator was designed (details of the just the flowrate and temperature are filtered.
steady state solution can be found in Appendix B) so that it con- The sequential solution consists of dynamically integrating each
tains two pseudo components in each heavy tail for HN and LCO submodel separately while keeping the connecting information
streams. The NBP for each heavy tail can be computed as: fixed (constant) for a very short period of simulation time ( f =
10 s). Once a submodel is integrated, the connection information
is updated and the other submodel is integrated considering this
THT = (N P Bhc xhc + N P Bshc xshc )/0.02 (61)
new updated information. Then, the cycle starts again moving for-
ward to the next time step. The key element of this approach is
where the subscripts “hc” and “shc” stand for heaviest compo- the magnitude of the time step selected, which is very small com-
nent and second heaviest component, respectively. This computa- pared to the time constant of each subsystem so as to represent
tion was included in the HN and LCO controllers (T C5 and T C6 re- the overall process dynamics correctly. A schematic diagram sum-
spectively) so as to regulate the 98% cut points, that is, the com- marizing the sequential solution can be found in Fig. 7.
position of the HN and LCO streams. Once the HN and LCO qual- The submodels and the sequential approach were coded in
ity requirements are defined, the quality of the remaining streams MATLAB(R). The FCC submodel was solved using the stiff DAE rou-
leaving the fractionator is defined as well. tine (ode15s). The dynamic equations of the reactor along with

10
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 7. FCC-Fractionator sequential simulation approach. The red arrows represent the exchange of updated information at every communication time period ( f ). The dashed
green arrows represent the input of material whereas the (solid) blue arrows represent the output of material. (For interpretation of the references to color in this figure leg-
end, the reader is referred to the web version of this article.)

Fig. 8. CVs of FCC. The dashed black lines represent the set points, whereas the continuous blue lines represent FCC variable values. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

the other FCC subsystems were integrated simultaneously. At each namic profiles are shown. Nevertheless, the model is capable of
time step (the integration step in ode15s is variable) the PFR model predicting flue gas composition, column hold-ups and internal
was solved using a fourth order Runge–Kutta method. The fraction- flows, stage compositions etc. Readers are invited to develop and
ator submodel was solved implementing the special algorithm de- simulate scenarios that cover their own use cases and research
scribed in the previous section. The discretization time (t) was needs.
set equal to the communication time period (t = f = 10 s).
The proposed approach led to a stable and fast dynamic solu- 5.1. Set point tracking
tion that can simulate FCC-Fractionator operation seven to eight
times faster than real time and at least two times faster than the The first simulation consisted of two sequential set point
simultaneous solution (when implementing Runge–Kutta methods changes for the preheater (from 616 ◦ F to 618.5 ◦ F) and reactor
or ODE MATLAB subroutines) using an Intel i5-7300U, 2.6 GHz (from 969 ◦ F to 971 ◦ F) temperatures at minutes 60 and 240, re-
computer. spectively. The simulation started from the nominal steady state
At every 10 seconds of simulation time, FCC-Fractionator infor- (the initial steady state is the same for all simulations) and the
mation is available (which is a desired feature when generating simulation time was five hundred minutes. Fig. 8 shows the CV
data) but since this frequency is much higher than real process profiles for the FCC, whereas Fig. 9 shows the CV profiles for the
measurement frequency, the default rate of reporting information fractionator.
is at every simulated minute (every six 10-second solution cycles). Figs. 8 and 9 show that the CVs are tightly controlled (in some
This 60 second sampling frequency is typically appropriate when cases the ordinate range is very small so as to illustrate change
developing advanced controllers such as MPC. However, the sam- in the process variable values). In addition, the regulatory layer is
pling frequencies, solution periods etc. can be adjusted by the user able to track the set point changes correctly and efficiently.
in the provided code. Figs. 10 and 11 show the MV profiles for the FCC and fraction-
ator respectively. The effect of the set point changes can be ob-
5. Sample FCC-fractionator simulation results served in two time intervals (i) from minute 60 to 239 and ii)
from minute 240 to 500). For the first time interval, the fuel flow
In this section, the results for two simulation scenarios (set to the preheater is increased in order to raise the preheater out-
point tracking and disturbance rejection) are presented. In both put stream temperature (Tpre ). This in turn reduces the circulation
cases, just the CVs, MVs and their respective valve positions dy- of regenerated catalyst (to keep the reactor temperature at the de-

11
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 9. CVs of fractionator. The dashed black lines represent the set points, whereas the continuous blue lines represent fractionator variable values. (For interpretation of
the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 10. MVs of FCC. The continuous blue lines represent FCC variable values. (For interpretation of the references to color in this figure legend, the reader is referred to the
web version of this article.)

Fig. 11. MVs of fractionator. The continuous blue lines represent fractionator variable values. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)

sired set point (SP)) which also reduces the circulation of spent In this case, the flow of regenerated catalyst is increased so as
catalyst (to keep the reactor catalyst inventory at its desired value). to increase the reactor temperature. Therefore, all the actions de-
Since the flow of spent catalyst is reduced, the flow of air to the scribed previously for the first time interval occur in the reverse
regenerator from the CAB is also reduced (to keep regenerator tem- direction (the flows of spent catalyst, air from CAB and flue gas are
perature at its set point), which in turn reduces the flow of flue gas increased). The higher reactor operating temperature results in the
(to keep the regenerator pressure). The effect of the preheater set VGO and the heavy pseudo components being cracked into lighter
point to the fractionator is very small since the reactor tempera- gases. Then, the hotter reactor vapor stream flows to the fraction-
ture is tightly controlled (as it can be observed in Fig. 8). ator column. As the light gas flow starts to increase in the column,
On the other hand, the set point change in reactor tempera- the reflux flow is increased (to keep the overhead temperature at
ture (second time interval) has a profound impact on both units. the desired operating value) as a consequence, the light naphtha

12
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 12. Valves of FCC. The continuous blue lines represent FCC variable values. (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)

Fig. 13. Valves of fractionator. The continuous blue lines represent fractionator variable values. (For interpretation of the references to color in this figure legend, the reader
is referred to the web version of this article.)

(LN) flow is reduced (to keep the level of the accumulator at the In addition, Figs. 15 and 16 show the CV profiles for the FCC and
desired operating value). At the same time, the flowrates of LCO fractionator respectively.
and HN are reduced (to maintain the quality of the streams at the The disturbance is successfully rejected and the process is well
target value). Finally, the flow rate of the light gas leaving the accu- controlled. The quality controller for the LCO stream (T C6 ) has the
mulator is higher, incrementing the flow of gas through the WGC. longest settling time since the disturbance (which results in the
The valve opening profiles for this simulation study are shown need to process a heavier feed) directly affects the heavier pseudo
in Figs. 12 and 13 for the FCC and fractionator, respectively. As ex- component distribution.
pected, there is a direct correlation between the MVs and their Figs. 17 and 18 show the MV profiles for the FCC and fraction-
corresponding valve openings. The valves follow the qualitative be- ator, respectively. In order to control the preheater output tem-
havior of the respective MVs (in the case of the valves related to perature, the fuel flow rate is incremented to account for a heav-
the flow of regenerated and spent catalyst through the circulation ier feed. Then, to control the reactor temperature, the flow of re-
lines, these are dictated by the U-bend dynamic model based on generated catalyst is increased, which also leads to an increase
force balances in which the pressures of the units (reactor and re- in the flow rate of spent catalyst to maintain the reactor cat-
generator) along with their catalyst inventory holdups affect the alyst inventory. The increase of the flow rate of spent catalyst
flow of material through the U-bend lines. A detailed description along with a higher rate of coke formation, lead to an increase
can be found in McFarlane et al. (1993). in the flow of air to the regenerator in order to keep its operat-
ing temperature, which also causes an increase in the flow of flue
5.2. Disturbance rejection gas (to maintain the regenerator pressure as its target value). In
terms of the fractionator, the heavier pseudo component distribu-
The second study consists on analyzing the regulatory control tion leads to a decrease the production rate of the lightest product
layer performance in the presence of a feed quality disturbance (LPG) while increasing the production rate of the (heavier) liquid
(step change from 25 API to 20 API as displayed in Fig. 14) occur- products.
ring after 60 min of operating at steady state. Similarly, 500 min of Furthermore, Figs. 19 and 20 show the valve position profiles
simulation are shown. Fig. 14 displays the feed quality disruption. that correspond to the MV profiles. As discussed previously, the

13
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 14. The red line indicates the process feed quality step disturbance. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)

Fig. 15. CVs of FCC. The dashed black lines represent the set points, whereas the continuous blue lines represent FCC variable values. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

Fig. 16. CVs of fractionator. The dashed black lines represent the set points, whereas the continuous blue lines represent fractionator variable values. (For interpretation of
the references to color in this figure legend, the reader is referred to the web version of this article.)

14
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 17. MVs of FCC. The continuous blue lines represent FCC variable values. (For interpretation of the references to color in this figure legend, the reader is referred to the
web version of this article.)

Fig. 18. MVs of fractionator. The continuous blue lines represent fractionator variable values. (For interpretation of the references to color in this figure legend, the reader is
referred to the web version of this article.)

Fig. 19. Valves of FCC. The continuous blue lines represent FCC variable values. (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)

15
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Fig. 20. Valves of fractionator. The continuous blue lines represent fractionator variable values. (For interpretation of the references to color in this figure legend, the reader
is referred to the web version of this article.)

relationship between the valves and MV dynamic profiles is clear, Acknowledgments


nevertheless, the force balance model should be considered for the
flow of catalyst in the circulation lines and the valve position of the Funding support from Marathon Petroleum Corporation is ac-
corresponding controllers as described in the previous simulation knowledged with gratitude.
scenario. Finally, additional simulations demonstrating the multi-
variate nature and nonlinear behavior of the process are detailed Supplementary material
in Appendix E.
Supplementary material associated with this article can be
6. Conclusions found, in the online version, at doi:10.1016/j.compchemeng.2022.
107900
This paper describes the development of a large scale FCC-
Fractionator open source dynamic model. The FCC section signif- References
icantly expands on an available literature model and considers
Ali, E., 2004. FCC Simulink module. http://faculty.ksu.edu.sa/Emad.Ali/Pages/
all relevant subsystems (regenerator, preheater, reactor etc.) and
SimulinkModule.aspx, last accessed 11/15/2017.
their interactions. The reactor model comprises a riser and strip- Ali, H., Rohani, S., Corriou, J., 1997. Modelling and control of a riser type fluid cat-
per, along with a detailed cracking scheme. A pseudo component/ alytic cracking (FCC) unit. Chem. Eng. Res. Des. 75 (4), 401–412.
Ancheyta-Juarez, J., Lopez-Isunza, F., Aguilar-Rodrıiguez, E., 1999. 5 lump kinetic
light gases approach was considered to successfully represent reac-
model for gas oil catalytic cracking . Appl. Catal., A 177 (2), 227–235.
tor cracking product distribution. A key contribution to this work Arandes, J.M., Azkoiti, M.J., Bilbao, J., de Lasa, H.I., 20 0 0. Modelling FCC units under
is that it incorporates a detailed model of a fractionator which is steady and unsteady state conditions. Can. J. Chem. Eng. 78 (1), 111–123.
integrated with the reactor. The fractionator is rigorously modeled Arbel, A., Huang, Z., Rinard, I.H., Shinnar, R., Sapre, A.V., 1995. Dynamic and control
of fluidized catalytic crackers. 1. Modeling of the current generation of FCC’s.
(taking into account the pseudo components/light gases approach) Ind. Eng. Chem. Res. 34 (4), 1228–1243.
leading to realistic product distribution (e.g LPG, LN, HN, LCO) and Bollas, G., Vasalos, I., Lappas, A., Iatridis, D., Voutetakis, S., Papadopoulou, S., 2007.
stream properties. A regulatory control layer is implemented to Integrated FCC riser-regenerator dynamics studied in a fluid catalytic cracking
pilot plant. Chem. Eng. Sci. 62 (7), 1887–1904.
regulate process performance and to guarantee product quality. Cerqueira, H., Biscaia, E., Falabella, E., Aguiar, S., 1997. Mathematical modeling of
A sequential numerical solution approach is proposed leading deactivation by coke formation in the cracking of gasoil . Stud. Surf. Sci. Catal.
to fast and stable solutions. Two simulation examples were shown 111, 303–310.
Chung, C.-B., Riggs, J.B., 1995. Dynamic simulation and nonlinear model based prod-
demonstrating that the dynamic model yields realistic predictions. uct quality control of a crude tower. AlChE J. 41 (1), 122–134.
As a consequence, we believe that this open source model can be Corella, J., Frances, E., 1991. On the kinetic equation of deactivation of commercial
use for multiple purposes in research and/or teaching on topics cracking (FCC) catalysts with commercial feedstocks. Stud. Surf. Sci. Catal. 68,
375–381.
such as data generation/analysis, advanced process control, deci-
Corma, A., Melo, F.V., Sauvanaud, L., 2005. Kinetic and decay cracking model for a
sion making such as production planning and fault detection. Fi- microdowner unit. Appl. Catal., A 287 (1), 34–46.
nally, it can be modified or extended to account for specific user Cristea, M.V., Şerban, P.A., Marinoiu, V., 2003. Simulation and model predictive con-
trol of a UOPfluid catalytic cracking unit. Chem. Eng. Process. 42 (2), 67–91.
requirements.
Das, A.K., Baudrez, E., Marin, G.B., Heynderickx, G.J., 2003. Three-dimensional sim-
ulation of a fluid catalytic cracking riser reactor. Ind. Eng. Chem. Res. 42 (12),
Declaration of Competing Interest 2602–2617.
den Hollander, M., Makkee, M., Moulijn, J., 1999. Fluid catalytic cracking (FCC): ac-
tivity in the (milli)seconds range in an entrained flow reactor. Appl. Catal., A
Authors declare that they have no conflict of interest.
187 (1), 3–12.
Dupain, X., Gamas, E., Madon, R., Kelkar, C., Makkee, M., Moulijn, J., 2003. Aromatic
CRediT authorship contribution statement gas oil cracking under realistic FCCconditions in a microriser reactor. Fuel 82
(13), 1559–1569.
Ellis, R.C., Li, X., Riggs, J.B., 1998. Modeling and optimization of a model iv fluidized
Omar Santander: Methodology, Investigation, Formal analy-
catalytic cracking unit . AlChE J. 44 (9), 2068–2079.
sis, Software, Writing – original draft. Vidyashankar Kuppuraj: Fusheng, O., Yongqian, W., Qiao, L., 2016. A lumped kinetic model for heavy oil cat-
Conceptualization, Methodology, Data curation, Validation, Writing alytic cracking FDFCC process. Pet. Sci. Technol. 34 (2), 192–199.
Gao, J., Chang, J., Xu, C., Lan, X., Yang, Y., 2008. CFD simulation of gas solid flow in
– review & editing. Christopher A. Harrison: Conceptualization,
FCC strippers. Chem. Eng. Sci. 63 (7), 1827–1841.
Methodology, Data curation, Validation, Writing – review & editing, Hagelberg, P., Eilos, I., Hiltunen, J., Lipiäinen, K., Niemi, V., Aittamaa, J., Krause, A.,
Supervision, Project administration, Funding acquisition. Michael 2002. Kinetics of catalytic cracking with short contact times . Appl. Catal., A 223
Baldea: Conceptualization, Methodology, Validation, Formal analy- (1), 73–84.
Han, I.-S., Chung, C.-B., 2001. Dynamic modeling and simulation of a fluidized
sis, Writing – review & editing, Supervision, Project administration, catalytic cracking process. Part I: process modeling. Chem. Eng. Sci. 56 (5),
Funding acquisition. 1951–1971.

16
O. Santander, V. Kuppuraj, C.A. Harrison et al. Computers and Chemical Engineering 164 (2022) 107900

Han, I.-S., Chung, C.-B., 2001. Dynamic modeling and simulation of a fluidized cat- Pinheiro, C.I.C., Fernandes, J.L., Domingues, L., Chambel, A.J.S., Graça, I.,
alytic cracking process. Part II: property estimation and simulation. Chem. Eng. Oliveira, N.M.C., Cerqueira, H.S., Ribeiro, F.R., 2012. Fluid catalytic cracking
Sci. 56 (5), 1973–1990. (FCC) process modeling, simulation, and control. Ind. Eng. Chem. Res. 51 (1),
Hernandez-Barajas, J.R., Vazquez-Roman, R., Felix-Flores, M., 2009. A comprehensive 1–29.
estimation of kinetic parameters in lumped catalytic cracking reaction models. Pitault, I., Nevicato, D., Forissier, M., Bernard, J.-R., 1994. Kinetic model based on a
Fuel 88 (1), 169–178. molecular description for catalytic cracking of vacuum gas oil. Chem. Eng. Sci.
Hernandez-Barajas, J.R., Vazquez-Roman, R., Salazar-Sotelo, D., 2006. Multiplicity 49 (24, Part A), 4249–4262.
of steady states in FCC units: effect of operating conditions. Fuel 85 (5), Sandler, S.I., 2006. Chemical, Biological and Engineering Thermodynamics, fourth ed.
849–859. John Wiley & Sons, NY.
den Hollander, M., Makkee, M., Moulijn, J., 2001. Development of a kinetic model for Seborg, D.E., Edgar, T., Mellichamp, D., Doyle, F., 2016. Process Dynamics and Control
FCC valid from ultra-short residence times. Stud. Surf. Sci. Catal. 134, 167–185. , fourth ed. John Wile y & Sons, NY.
Huang, H., 20 0 0. Simulation and Control of Complex Distillation Processes. Texas Secchi, A., Santos, M., Neumann, G., Trierweiler, J., 2001. A dynamic model for a FCC
Tech University, Lubbock, USA. UOP stacked converter unit. Comput. Chem. Eng. 25 (4), 851–858.
Hysys 2004.2, Tutorial & Applications, Aspen Technology, Cambridge, USA, 2005. Shayegh, F., Farshi, A., Dehgan, A., 2012. A kinetics lumped model for VGO catalytic
Jacob, S.M., Gross, B., Voltz, S.E., Weekman Jr., V.W., 1976. A lumping and reaction cracking in a fluidized bed reactor. Pet. Sci. Technol. 30 (9), 945–957.
scheme for catalytic cracking. AlChE J. 22 (4), 701–713. Sildir, H., Arkun, Y., Canan, U., Celebi, S., Karani, U., Er, I., 2015. Dynamic modeling
Lan, X., Xu, C., Wang, G., Wu, L., Gao, J., 2009. CFD modeling of gas-solid flow and optimization of an industrial fluid catalytic cracker . J. Process Control 31,
and cracking reaction in two-stage riser FCC reactors. Chem. Eng. Sci. 64 (17), 30–44.
3847–3858. Sugungun, M., Kolesnikov, I., Vinogradov, V., Kolesnikov, S., 1998. Kinetic modeling
Lee, L.-S., Chen, Y.-W., Huang, T.-N., Pan, W.-Y., 1989. Four-lump kinetic model for of FCC process. Catal. Today 43 (3), 315–325.
fluid catalytic cracking process . Can. J. Chem. Eng. 67 (4), 615–619. Gauthier, P.L.T., Bayle, J., 20 0 0. FCC: fluidization phenomena and technologies. Oil
Lopes, G., Rosa, L., Mori, M., Nunhez, J., Martignoni, W., 2011. Three-dimensional Gas Sci. Technol. 55 (2), 187–207.
modeling of fluid catalytic cracking industrial riser flow and reactions. Comput. Takatsuka, Y.M.T., Sato, S., Hashimoto, H., 1987. A reaction model for fluidized bed
Chem. Eng. 35 (11), 2159–2168. catalytic cracking of residual oil. Int. Chem. Eng. 27, 107–116.
Luyben, W.L., 2012. Effect of tray pressure drop on the trade-off between trays and Watkins, R.N., 1979. Petroleum Refining Distillation , second ed. Gulf Publishing Co,
energy. Ind. Eng. Chem. Res. 51 (26), 9186–9190. TX.
Malay, P., Rohani, S., Milne, B.J., 1999. The modified dynamic model of a riser type Weekman, V.W., 1968. Model of catalytic cracking conversion in fixed, moving, and
fluid catalytic cracking unit. Can. J. Chem. Eng. 77 (1), 169–179. fluid-bed reactors. Ind. Eng. Chem. Process Des. Dev. 7 (1), 90–95.
McFarlane, R., Reineman, R., Bartee, J., Georgakis, C., 1993. Dynamic simulator for a Wiens, J.S., 2010. Experimental and Modeling Study of a Cold-Flow Fluid Catalytic
model IV fluid catalytic cracking unit. Comput. Chem. Eng. 17 (3), 275–300. Cracking unit Stripper. University of Saskatchewan, Saskatoon, Canada.
Xu, O., Su, H., Mu, S., Chu, J., 2006. 7 lump kinetic model for residual oil catalytic Yescas, R.M., Bogle, D., Isunza, F.L., 1995. Steady state and dynamic simulation of
cracking. J. Zhejiang Univ. Sci. A 7, 1932–1941. FCC reactor units. IFAC Proc. Vol. 28 (9), 153–158.
Oliveira, L.L., Biscaia, E., 1989. Catalytic cracking kinetic models. Parameter estima- Zheng, Y.-Y., 1994. Dynamic modeling and simulation of a catalytic cracking unit.
tion and model evaluation. Ind. Eng. Chem. Res. 28 (3), 264–271. Comput. Chem. Eng. 18 (1), 39–44.

17

You might also like