Fluid Catalytic Cracking FCC Process Mod

REVIEW
pubs.acs.org/IECR
Fluid Catalytic Cracking (FCC) Process Modeling, Simulation,

and Control4
Carla I. C. Pinheiro,*,† Joana L. Fernandes,‡ Luís Domingues,† Alexandre J. S. Chambel,† Inês Grac-a,†
Nuno M. C. Oliveira,§ Henrique S. Cerqueira,# and Fernando Ramôa Ribeiro†
†
Institute for Biotechnology and Bioengineering (IBB), Department of Chemical Engineering, Instituto Superior Tecnico/Universidade
Tecnica de Lisboa, Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal
‡
Process Design and Modeling Division, IFP Energies Nouvelles Lyon, Rond-point de l’echangeur de Solaize, B.P. 3,
69360 Solaize, France
§
Centre for Chemical Processes Engineering and Forest Products (CIEPQPF), Department of Chemical Engineering,
Universidade de Coimbra, R. Sílvio Lima Polo II, 3030-790 Coimbra, Portugal
#
ATP Engenharia, Rua S~ao Jose 90/2201-C, 20010-020 Centro, Rio de Janeiro, RJ, Brazil
ABSTRACT: This paper focuses on the fluid catalytic cracking (FCC) process and reviews recent developments in its modeling,
monitoring, control, and optimization. This challenging process exhibits complex behavior, requiring detailed models to express the
nonlinear effects and extensive interactions between input and control variables that are observed in industrial practice. The FCC
models currently available differ enormously in terms of their scope, level of detail, modeling hypothesis, and solution approaches
used. Nevertheless, significant benefits from their effective use in various routine tasks are starting to be widely recognized by the
industry. To help improve the existing modeling approaches, this review describes and compares the different mathematical
frameworks that have been applied in the modeling, simulation, control, and optimization of this key downstream unit. Given the
effects that perturbations in the feedstock quality and other unit disturbances might have, especially when associated with frequent
changes in market demand, this paper also demonstrates the importance of understanding the nonlinear behavior of the FCC
process. The incentives associated with the use of advanced model-based supervision strategies, such as nonlinear model predictive
control and real-time optimization techniques, are also presented and discussed.
1. INTRODUCTION binder is responsible for the very large number of reactions involved
Growing demand for refinery products combined with the in the FCC process.11,12 Different additives may also be added to
decreasing quality of crude oils and tighter product specifications the FCC catalyst with the purpose of modifying the FCC yields
due to environmental constraints is forcing refiners to make and/or reducing pollutant emissions.13,14 The most common are
significant investments. In fact, during the past decades, new combustion promoters,15 octane and light olefins booster,13,1618
hydroprocessing units have been built, together with the revamp- SOx and NOx reducers,19,20 and metal traps.21
ing of old refining processes to meet market demands.1 In this Like all industrial processes involving heterogeneous catalysis,
context, the use of advanced process engineering tools has become the FCC process also deals with catalyst deactivation.22 Actually,
essential for refiners, not only for design but also in the tasks of a significant fraction of the FCC feedstock (usually ∼6 wt % from
process control, optimization, scheduling, and planning. Besides a typical vacuum gas oil plus residue feedstock23) is converted
the application to specific process units, these techniques are also into a mixture of compounds (called coke) that remain retained in
being used for the entire refinery supply chain.2 the catalyst structure after stripping. These compounds quickly
Fluid catalytic cracking (FCC) remains a key unit in many deactivate the acid sites of the catalyst, resulting in a significant
refineries; it consists of a three-step process: reaction, product activity loss.23,24 For that reason, the FCC unit was designed to
separation, and regeneration. In this cyclic process, gas oils from allow a continuous recirculation of catalyst between the reactor and
vacuum distillation towers and/or residues from atmospheric the regenerator, where coke is removed from the catalyst by
distillation towers are converted into lighter and more valuable combustion at high temperatures (typically 9501030 K). One
products. One of the most desired products is cracked naphtha, advantage of this continuous catalyst recirculation is that FCC units
which is the major constituent of the gasoline pool.3 Operating operate in heat balance; that is, the heat released during the burning
conditions comprise high reaction temperatures in the range of of the coke deposited on the spent catalyst is utilized to heat the
750800 K and pressures close to atmospheric conditions. FCC regenerator flue gas, to vaporize and heat the feed to the reaction
is able to process a wide variety of feedstocks and is suitable to temperature, and to heat the steam and the additional quench
operate in special campaigns46 that may also soon include
coprocessing of renewable feedstocks.79 Nowadays, more than Received: April 7, 2011
400 FCC units are operated worldwide.10 Accepted: November 17, 2011
A multicomponent catalyst that usually contains an acid USHY Revised: October 21, 2011
zeolite, an active alumina matrix, an inert matrix (kaolin), and a Published: November 17, 2011
r 2011 American Chemical Society 1 dx.doi.org/10.1021/ie200743c | Ind. Eng. Chem. Res. 2012, 51, 1–29
Industrial & Engineering Chemistry Research REVIEW
Figure 1. Scheme of some FCC unit designs: (a) UOP stacked unit; (b) model IV; (c) Exxon Flexicracking unit; (d) R2R residue unit (adapted from
Montgomery32).
streams to outlet temperatures while providing enough heat for the catalyst is frequently required. Moreover, for FCC units proces-
endothermic cracking reactions and remaining heat losses. sing feedstocks with high levels of metals,2527 it is also common
During the successive reactionseparationregeneration cy- to replace a portion of the inventory by fresh catalyst, to keep the
cles, the FCC catalyst particles may break, producing fines that amount of contaminant metals at an acceptable level. This
will result in particulate emissions. To compensate for this loss regular addition of fresh catalyst makes the FCC process one
(and hence to maintain the catalyst activity), addition of fresh of the most important markets for catalysts.12,21 As a result, the
2 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

so-called equilibrium catalyst (e-cat) circulating in the FCC unit is a Besides the developments in FCC technology toward the
heterogeneous mixture, ranging from young particles (fresh, high- conversion of heavier feedstocks, recent developments have also
activity catalyst with very low metal concentrations) to old particles been implemented to satisfy the increasing propylene demand.
(aged, low-activity catalyst with high metal concentrations). The FCC Alliance between Axens, Shaw, Total and IFP Energies
Since the first FCC unit started operation in 1942, several Nouvelles (IFPEN) have proposed since 2008 the PetroRiser
design improvements have been made.28 Indeed, almost all of the technology for the production of high yields of propylene. This
components of the FCC unit have been modified to improve new technology incorporates a second riser in the FCC complex
performance.29 The first unit in operation was model I from to which the light cracked naphtha produced in the first riser is
Standard Oil Development Co. (SOD), now ExxonMobil. This sent for further processing.36,37
unit was composed of multiple small vessels and had a catalyst Due to the diversity of physicochemical phenomena that
up-flow configuration in both the reactor and regenerator vessels. occur, FCC processes are characterized by complex steady-state
The regenerator operated at low pressures, and external cyclones and dynamic behavior, requiring detailed process models to
were used. In 1947, UOP built the first unit that used the concept adequately express the nonlinear effects and the extensive inter-
of spent catalyst stripping: the stacked FCCU (Figure 1a). This actions among variables that are observed in industrial practice.
unit had smaller and improved regenerators, where the regener- Various efforts to model the behavior of FCC units have resulted
ated catalyst was lifted to a bed cracking reactor by vaporized feed from academic efforts and have already provided important
and the spent catalyst flown by gravity to the regenerator.30,31 insights into the functioning of these important processes. In
In 1951, M. W. Kellogg introduced the Orthoflow unit, com- parallel, a significant number of such models have also started to
posed by a low elevation regenerator and a high reactor with an be used in actual plants as the basis for various routine tasks, such
internal stripper. In this model, the catalyst flow was made through as model-based control and optimization. The FCC models
internal vertical straight tubes, a standpipe, and a lift line, controlled currently available differ enormously in terms of their scope,
by plug valves. Another FCC configuration, called model IV, was level of detail, modeling hypothesis used, and solution ap-
introduced by SOD in 1952. This unit presented smaller vessels proaches. Additionally, the potential advantages of their more
arranged side by side (Figure 1b) and was operated at higher effective use are also starting to be widely recognized by the
pressures and internal velocities; catalyst flow control was done by industry. Consequently, improving the existing modeling ap-
changes in the differential pressure between the reactor and proaches and introducing further developments in this area are
regenerator (U-bend concept) and by changes in the aeration in goals with important practical implications. With the aim of
the spent catalyst entrance to the regenerator. The riser cracking making a contribution to the facilitation of this task, the main
unit was first proposed by Shell in 1957, which, together with the objective of this work is to provide a comprehensive review of
introduction of high-activity zeolite catalysts in the 1960s, defini- past modeling approaches of FCC units, both in dynamic and
tively established this configuration. Since then, all new FCC unit steady state, and to survey the various model-based approaches
designs have included riser cracking reactors.28 reported in the open literature for the supervision, control, and
The improvement of FCC catalysts (e.g., through addition of optimization of these units. The structure of the paper is as
combustion promoters) allowed further developments in the follows: Section 2 describes the various modeling approaches
FCCs’ regeneration systems, which made possible the reduction proposed, including the description of the cracking kinetics and
of coke on the regenerated catalyst to <0.1% wt. Kellogg’s the various frameworks available to model the deactivation of the
Orthoflow F process, with two stages of regeneration in the catalysts. An analysis of the existence of steady-state multiplicity
same vessel, appeared in 1973. Later, in 1978, UOP introduced a in these models and their characterization is presented in section
typical side-by-side unit with a high-efficiency regenerator unit. 3. Monitoring of these units, in terms of both process and quality
This regeneration system was designed to operate in the fast variables, is analyzed in section 4. Finally, the various approaches
fluidization regimen and was composed by a combustor and a lift developed for the control and real-time optimization of these
of small diameter that discharge the catalyst and the combustion units are described in section 5, followed by final remarks and
gases in a disengagement vessel. In 1979, Exxon introduced the conclusions.
Flexicracking unit (Figure 1c) that maintained a side-by-side
arrangement but included a riser with an elevated stripper/
disengager vessel and a lower elevation regenerator.28 2. MATHEMATICAL MODELING
During the 1980s, the increasing need to process heavier feeds The implementation and maintenance of model-based engi-
brought new developments to existing FCC designs. In 1981, neering tools are currently limited by the lack of experienced
Total Petroleum USA developed its residue FCC unit (R2R unit, manpower in the refineries.38 Consequently, it has become
now licensed by Axens/IFP and Stone & Webster), presenting a common practice for refiners to request costly hand-in-hand
side-by-side configuration with a two-stage regeneration system solutions from companies specialized in the development of
without catalyst cooling, which occurs in two separate stacked these engineering tools. Model development and process identi-
vessels, and a straight riser reactor with a proprietary feed injection fication are probably the most time-consuming steps in the
system and an internal exit separation system (Figure 1d). Further implementation of many advanced process engineering strate-
developments continued with improved designs focusing on atmo- gies at the industrial level. Advanced control strategies usually
spheric residue conversion,3234 proposed by UOP (UOP’s Residue rely on linear “black box” models from step identification tests
Processing) and Petrobras (Petrobras Advanced Converter).35 carried out on the unit. Although these models are faster to
This later technology encompasses a set of proprietary develop- develop, they are valid only in the operating region where they
ments, namely, the PASS closed cyclones system and the Ultramist were obtained, and they do not usually capture the nonlinearities
optimized feedstock injection system, combined with improve- of the industrial process. In addition, it is difficult to use
ments in the riser and optimization of the mechanical design of the simultaneously these models for plant optimization, which
equipment. requires accurate and rigorous models for a wider range of
operating conditions. The development of rigorous models is, gas oil (VGO) feeds70,71 and for coke formation,72 based on the
however, a difficult task inherent to the complexity of the single-event approach. This modeling technique aims at retaining
industrial processes. Besides, the insufficiently detailed feed the full detail of the reaction pathways of all individual feed
characterization and the nonexistence of true steady states in components and reaction intermediates, by describing the reac-
typical plants lead often to imprecise and poorly detailed models. tion network in terms of elementary steps.73 Therefore, this
The success of real time optimization applications is thus requires a very detailed characterization of feeds and products,
dependent on more accurate models and more reliable feed which is generally not available in industry, and is computation-
characterization data.38 ally more demanding than the lumped models.
The next sections consider the different mathematical models Semilumped models that consider the different reaction types
that have been applied to the FCC process aiming at the simulation, occurring in catalytic cracking have also been proposed, in which
control, and optimization of this key downstream unit. the products and reactants are grouped by common character-
2.1. Cracking Kinetics. The catalytic reactions occur on the istics such as the carbon number and types of molecules such as
surface of the catalyst particles through a positively charged paraffins and olefins.74,75 Unfortunately, studies concerning this
intermediate. This intermediate can be either a carbenium ion, type of model have been carried out mainly for model molecules
CR3+, or a carbonium ion, CR4H+ (where R represents either an and, like the single-event models, require a good characterization
alkyl group or a hydrogen atom3,32). The carbenium ion is a of the reactants and products.
tricoordinated species with three bonds, whereas the carbonium In the case of very complex feeds containing several chemical
ion is a pentacoordinated species with five bonds. These carbo- species, such as the FCC feedstock, a detailed composition can be
cations can be formed on either Brønsted or Lewis acid sites of obtained from partial analytical data through molecular recon-
the FCC catalyst39 and undergo several reactions. struction techniques. The so-called “feedstock reconstruction”
The primary reaction in catalytic cracking is the β-scission method was proposed by Verstraete et al.76 by means of an
reaction that occurs by breakage of the β CC bond to the algorithm that generates a complex mixture of molecules based
carbon with the positive charge; thus, an olefin and a new on standard petroleum analysis, which was validated for FCC
carbocation are obtained. The new carbocation is now free to gasolines,76 VGO,77 LCO,78 and VR.79 It consists of a two-step
react with another paraffin molecule and continue the β-scission approach; in the first step, called “stochastic reconstruction”, a
reaction. As the reaction proceeds, smaller olefins and paraffins large representative set of molecules is created. It is initially assumed
are produced from the original larger feedstock hydrocarbon that the oil mixtures can be described by distributions of structural
molecules. The reaction ends when the carbocation loses a blocks (polycyclic cores, rings, chains, etc.), the transformation
proton to the catalyst and is converted to an olefin or the from a set of distributions into an equimolar mixture of mole-
carbocation picks up a hydride ion from a donor, such as the cules being performed by Monte Carlo sampling. The properties
coke, and is converted into a paraffin.39,40 of the mixture are then optimized using a genetic algorithm to
A number of secondary acid-catalyzed reactions such as minimize the difference between calculated properties and
isomerization, alkylation, hydrogen transfer,41 and condensation experimental data. The second step, named “reconstruction by
also occur under FCC conditions.3,41,42 Considered together, the entropy maximization”, further improves the representative-
presence of thousands of reacting species (both in the feed and in ness of the set of constructed molecules by adjusting their molar
the products), the need for extensive analytical resources avail- fractions.
able in production facilities, and the computational effort re- The choice of the kinetic model used should be supported by
quired both in parameter estimation and in model simulation the level of detail desired, which normally depends on the type of
make difficult the development of detailed and accurate kinetic usage that the FCC model will have. It is also important to keep
rate expressions for commercial FCC reactions. To overcome in mind that it is not realistic to attempt to define a detailed
these limitations, a typical approach groups different molecules microkinetic model for gas oil industrial cracking, because it
(lumps) according to their boiling point and/or their molecular would involve hundreds of elementary steps, for which kinetic
characteristics (paraffins, olefins, naphthenes, and aromatics). parameters would need to be determined.70,80 Nevertheless, if
One of the first and most widely used catalytic cracking kinetic one intends to predict the behavior of FCC industrial units, it is
models was the three-lump model proposed by Weekman,4345 necessary to choose a kinetic model with a reasonably detailed
which mainly focused on feedstock conversion and gasoline product distribution.75
selectivity. The lumps considered were the feedstock, the gaso- 2.2. Deactivation. In addition to the already mentioned diffi-
line (C5-221 °C), and the remaining gases plus coke. To account culties in modeling the catalytic cracking kinetics, other important
for the coke on the catalyst a four-lump model was proposed by aspects are the ad/desorption steps of reactants and products on the
Lee et al.46 This lump model was extensively used by many other catalyst surface8184 and the very fast deactivation of the catalyst that
authors in their works4754 because it allowed the prediction of complicates the independent determination of the kinetic reaction
coke formation, which is essential for the simulation of the entire rates and the catalyst deactivation.58,65,85,86 Indeed, the main cause of
reactorregenerator unit. Various other authors have further FCC catalyst deactivation is the coke formation and deposition on
extended this lumping approach on the basis of the FCC product the surface of the catalyst.22,32,39,42,84,87 The first efforts in modeling
fractions.5566 coke formation were made by Voorhies:88 he proposed an empirical
Other detailed kinetic modeling approaches have also been correlation for coking in the catalytic cracking of gas oil that depends
applied to catalytic cracking. The “structure oriented lumping” on the catalyst residence time.
(SOL) approach proposed by Quann and Jaffe67,68 and Chris- Two fundamentally different approaches to modeling catalyst
tensen et al.69 was based on an extensive network of reactions deactivation have been used: the time-on-stream and the coke-on-
between lumps that are represented as an assembly of molecular catalyst functions. Nam and Kittrell89 addressed this issue and
building blocks. Froment and co-workers proposed detailed concluded that time-on-stream functions exhibit the advan-
kinetic models for the catalytic cracking of industrial vacuum tages of simultaneously allowing the deactivation mechanisms
(e.g., basic nitrogen titration simultaneous with coking) together A general equation for deactivation considering variations in
with the mechanistic kinetics of coke formation. Corella and the decay order can represent the different mechanisms of deac-
Monzon90 considered that different kinds of coke (whiskerlike, tivation for all reactions:71,94
pyrolitic, polymeric, in multilayers, etc.) can be present in the
catalyst. Although they all contribute to increased coke content dΦ
¼ αΦd ð1Þ
on the catalyst, they do not contribute equally to deactivation; dx
therefore, it is more appropriate to use time-dependent relation-
ships in the presence of multiple deactivation causes. Froment In this formula Φ represents an average value that measures the
and Bischoff91 criticize the use of time-on-stream functions and catalyst activity for all of the reactions and different mechanisms of
attribute their frequent use to the simplicity of this approach; deactivation and for all active site strengths present in the catalyst
a coke content function requires an additional rate equation for the surface, α represents the kinetic deactivation constant, and x can be
coke formation (to introduce the process time) and characteriza- either the time on stream or coke on catalyst. The decay order (d)
tion data from spent catalyst. can be related with the number of active sites involved in the steps
The use of coke-on-catalyst relationships provides additional controlling the main reaction mechanisms and the catalyst deacti-
insights into the deactivation mechanism, through the inclusion of vation for that reaction.94 This equation can be integrated for
microbalance data and coke-bed profile data.89 Furthermore, this different decay orders.
type of relationship can be used for catalyst regeneration studies, Another important aspect of FCC catalyst deactivation is the
because it accounts for the effect of nonregenerated coke at the use of selective versus nonselective deactivation models. Selec-
riser inlet.89,92 However, a function of coke content requires an tive deactivation models are more rigorous and accurate than
additional rate equation for the coke formation to introduce nonselective models, explaining the variation of product selec-
the process time and characterization data from the spent catalyst tivity with time-on-stream, coke-on-catalyst, or catalyst poison
as well.91 concentration.98 In fact, depending on the poisoned acid sites,
The two distinct approaches to model catalyst deactivation, as the overall distribution of acid sites may change, which should
a function of time-on-stream or coke-on-catalyst, can be unified affect product selectivities. In this scenario, the deactivation of
by tracking catalyst activity as a function of the decrease on the catalyst can affect to different extents the reactions in the
effective diffusivity of the reactants due to pore occlusion network considered.72,98 This means that different deactivation
(external) by coke.93 Jimenez-García et al.93 considered the pore functions (with distinct deactivation constants or both deactivation
blockage, caused by coke growth at the external surface of the constants and orders) should be used for each reaction or group of
catalyst, to be the main reason for catalyst deactivation. This reactions. When detailed analysis of the coke is available, the
situation leads to an increase of Thiele modules and conse- deactivation function may also take into account different toxicities
quently to a decrease of the effectiveness factor of each reaction. for the different types of coke present on the catalyst surface.99
This approach was applied to the simulation of both a MAT Until now, the use of selective deactivation models is very
reactor and an industrial FCC riser.93 The author’s model scarce in the literature, because it adds a significant degree of
consisted of five lumps described by a set of seven first-order complexity that requires more experimental data and could
reactions. Nonetheless, knowledge of mean free path and molar potentially result in overparametrization of the kinetic model.98
mass for each lump is required to determine the efficiency factor. To avoid this problem, four simplified scenarios were considered in
These properties are not, however, easy to determine because an experimental study of how catalyst deactivation affects each
they require the lump composition and knowledge of the reaction in the cracking network, considering a five-lump model:99
molecular diameter for each component in every lump. Simpli- (1) similar for all reactions (nonselective catalyst deactivation);
fication of lump composition may lead to a high uncertainty in (2) different for each reactant in the lumping scheme; (3) different for
the value of the efficiency factor; however, the exact composition each product in the lumping scheme; (4) different for each reactant
of each lump may be very difficult or very resource-consuming to in the lumping scheme, except for dry gas, which is considered to be
determine and, therefore, not practical. produced via thermal cracking and not affected by catalyst deactiva-
Besides these different approaches, various correlations (mainly tion. A power law (function of the residence time) was used to
exponential and power laws) are available in the literature to describe the deactivation in the catalyst. These four scenarios were
describe the deactivation rates.43,59,91,94,95 Despite their mathema- compared using experimental data obtained in a pilot plant, with
tical differences, good adjustments against experimental data have the deactivation order varying for the selective deactivation scenar-
been reported. Nevertheless, realistic comparison between the ios (2, 3, and 4). The authors concluded that, although scenario 3
proposed functions becomes difficult, due to the different experi- gave better global results, the differences between the selective or
mental conditions at which deactivation data were obtained, nonselective models were not sufficiently significant to justify the
affecting the coke formation, which is highly dependent on large number of parameters to be estimated. The deactivation
operating conditions, feedstock, and catalyst properties. function can also be dependent on the type of coke present in the
As major reasons for having significantly different equations catalyst.82 In this case more detailed characterization data on the
that adjust well to experimental data, Larocca et al.96 and Corella types of coke (CH2Cl2 soluble and insoluble) is required.42
et al.94 point out the differences in the catalyst residence times Recently, significant progress was made by a group from the
used by different researchers and the difficulty in obtaining data University of Ghent in the development of a single-event
at very low residence times in traditional bench-scale reactors. It microkinetic model (SEMK) for catalytic cracking of alkanes
has been experimentally observed by these authors that in the and mixtures of alkanes or cycloalkanes with 1-octene, which
first seconds the deactivation behavior is different, presenting a takes into account deactivation by coke formation.100102 Beir-
different decay order94,96 or a different decay coefficient in the naert et al.100 and Quintana-Solorzano et al.101,102 improved the
exponential function.96 The decay order depends on the feed- description of coke formation in terms of elementary steps and
stock and catalyst used.97 the interaction between coke formation and the actual cracking
reactions,100,101 as well as the description of the single-event considers both gas and solid continuum phases. A set of balance
kinetics in relumped form. Coke formation is indeed considered equations is written for each phase, representing conservation of
as starting from carbenium ions on the catalyst surface, and the mass, momentum, and energy inside the riser. In this approach,
concentration of the carbenium ions on the catalyst surface is the particle flow characteristics are modeled in the framework of
obtained via the single-event modeling of the catalytic cracking the kinetic theory of granular flow.113115 On the other hand, the
reactions as such. This approach presents the advantages of EulerianLagrangian approach considers the gas to be a conti-
requiring a less detailed feed characterization and allowing a fast nuum phase, whereas the particles in the solid phase are described
simulation of a riser reactor.103,104 using the Lagrangian equations of motion for each particle of the
2.3. Riser Models. The feed vaporization and the cracking system, with a prescribed set of initial conditions.115118 Previous
reactions occur inside the riser, while catalyst, hydrocarbon liquid authors have considered that the overall performance of the riser
droplets, and hydrocarbon vapors travel upward. Therefore, can be accurately predicted by using simple 1-D mass, energy, and
there are several chemical and physical complex phenomena chemical species balances.28 However, to evaluate heat transfer,
occurring simultaneously that need to be considered. chemical reaction, and geometry effects, especially at the feed
The rate of feed vaporization in the riser affects its perfor- injection area of the riser, a 3-D model is usually required.109,117,119
mance to a great extent. A liquid phase feed has a negative effect Nayak et al.54 used the EulerianLagrangian approach to
on the cracking reactions, whereas a slow vaporization leads to develop a 3-D adiabatic model of the FCC riser reactor, which
very high “effective CTO”, that is, the catalyst to vaporized feed simulates heat transfer, evaporation, and cracking. The motion of
ratio. The presence of liquid feed combined with the high the solid catalyst particles and the liquid droplets was modeled in
catalytic activity and temperature in the riser entry zone51 can the Lagrangian framework, whereas the motion of the contin-
lead to undesirable secondary cracking reactions, accelerating uous phase was described in the Eulerian framework. Although a
coke formation and decreasing gasoline yield.51,53 For this 3-D model was presented, simulations were carried out only for a
reason, extensive research has been done in the development 1-D model, to study the influence of the droplet diameter on the
of effective injection systems capable of atomizing the feed in riser performance. These simulations were used for evaluating
very fine drops, leading to fast vaporization and short and the cost benefit of new FCC nozzles and for related decision-
intimate contact between catalyst and oil.105,106 making. All of these developments addressed mainly steady-state
Although the feed vaporization section is considered to be one simulations and the study of equipment performance, design, and
of the most important parts of the riser, it is rarely modeled for optimization.
riser simulation purposes, because considering instantaneous vapor- Besides the 1-D plug flow model and the complex 3-D
ization simplifies considerably the process modeling. However, approaches, there is also the coreannulus model.47,53,120 This
modeling the vaporization is important for the optimization and conceptualization considers two zones, with a central core where
design of feed injection systems and nozzles. Theologos and gas is flowing upward at high velocity, entraining dilute solid with
Markatos107 mention that the vaporization takes place in the first a small slip velocity, generally equal to the particle terminal
1.53 m of the riser, which corresponds to 510% of the riser velocity. The second zone is a peripheral annulus where con-
total length (3040 m). On the other hand, Ali et al.48 state that centrated solid is flowing downward, with a gas velocity close to
the feed needs only approximately 3% of the mixture residence zero. Moreover, this model is based on hydrodynamic correla-
time to vaporize completely, which corresponds to 0.330 ms tions that account for the slip velocity and porosity variations
according to drop size.108 Also, the cracking reactions occur only along the radial and axial directions in the riser.
when the feed is vaporized,3,109 which justifies the assumption of Common assumptions in the 1-D model are plug flow for both
instantaneous vaporization and will not introduce a significant vapor and catalyst phases, adiabatic operation, and no mass or
error in yield or conversion calculation. heat transfer resistances between the two phases. The ratio
Vaporization modeling involves three phases: catalyst parti- between the gas and catalyst velocities, usually called the slip
cles, hydrocarbon liquid droplets, and hydrocarbon vapors. It factor (ψ = vg/vc), is commonly neglected. However, considering
also requires different balances to account for the sensible heat the existence of a slip between the gas and the solid velocities is
gain of the liquid droplets (while the temperature in the droplet is quite important, because it affects the contact time between the
below vaporization temperature), the vaporization of the liquid hydrocarbon vapors and catalyst flows and, consequently, the
droplets with mass transfer from the droplet to the gas phase, and feed conversion. Different assumptions have been considered for
the heat transfer between the catalyst and gas phases after vapor- the slip factor between the catalyst and vapor phases. Malay
ization is complete. Moreover, modeling feed vaporization implies et al.50 assumed a constant slip factor. Corella and Frances97
predicting the diameters of the liquid droplets and relates the considered the riser to consist of three or four well-mixed
vaporization rate to droplet sizes and mass transfer to the gas compartments, with variation of important quantities such as
phase, together with estimating the heat transfer coefficients the slip factor and the molar expansion factor from compartment
between the two phases. The complexity of these requirements to compartment, whereas they were assumed to be constant
explains why only a few authors have included feed vaporiza- within each compartment. Han and Chung121 wrote momentum
tion in their riser models.49,52,54,107,110,111 equations for both phases, which allow determination of the
Although various modeling approaches for the riser can be velocity profiles along the riser reactor and account for the slip
found in the literature, a one-dimensional (1-D) model is the velocity between the gas and catalyst particles. Gupta and Rao52
most commonly used method to describe industrial FCC considered the slip velocity equal to the terminal velocity of a
units.49,112 With the more complex 3-D models, two kinds of single particle.
approaches are applied to describe the riser hydrodynamics: the Apart from describing the velocity profiles for both phases,
EulerianEulerian and the EulerianLagrangian approaches, another important factor in determining the slip factor is the
both implemented by computational fluid dynamics (CFD) diameter of the solid particles. In fact, for particles with diameters
modeling techniques. The EulerianEulerian methodology typical of FCC catalysts, the slip factor approaches very rapidly 1,
which means that the gas and solid velocities are practically
coincident.121 This disagrees with typical slip factors, close to 2,
that are usually reported in the riser by many other authors.50,117,122
High slip factors, up to 4, have been experimentally observed and
can be related to the formation of clusters along the riser
reactor.122,123 A cluster is an aggregate of particles moving together
with the same velocity (Figure 2). It has a higher “free falling
velocity” than the terminal velocity of the individual particles.
Hence, when the amount of clusters increases, the slip factor
between gas and catalyst will also increase. The reported size of
FCC catalyst clusters is between 2 and 15 mm.124 However, in the
dilute region of the riser the probability of cluster formation is
smaller due to the higher velocities, and the slip factor decreases
significantly in that region. Subbarao125 has recently presented a
model for the structure fluidsolid in a riser, which includes an Figure 2. Hydrodynamic cluster formation (adapted from P€arssinen124).
equation developed for estimating cluster size.
Although there are studies that experimentally demonstrate
the formation of clusters in the riser reactor,122,124,126 this factor hydrocarbons that remain adsorbed/occluded in the catalyst
is not usually considered in modeling these sections.52,117 The pores after stripping (cat-to-oil coke), because they contribute
studies that consider the formation of hydrodynamic clusters in to increase the H/C molar ratio in coke and consequently the
1-D simulations give comparable results to 3-D simulations for heat produced in the regenerator.131,137
product yield and conversion, if an effective cluster diameter is Only a few models can be found in the literature for the
used instead of the catalyst single-particle diameter.52,117 Never- dynamic simulation of the stripper/disengager section in a FCC
theless, the 3-D simulations describe better the gassolid slip in unit.48,50,51,138141 In these studies the stripper and disengager
the riser, considering a larger range of operating conditions and are always considered as a unique section, so they are modeled
model parameters. together. The usual assumption is that the stripper/disengager
Computational fluid dynamics (CFD) simulations with a section can be modeled as a continuous stirred tank (CST)
EulerianEulerian two-fluid modeling approach of gassolid without reaction. Some authors have also proposed linear
flow in the riser have been widely conducted to predict hydro- empirical correlations to predict cat-to-oil coke as a function of
dynamics in circulating fluidized bed risers, by including cluster stripping steam flow rate.51,138,140 Other authors have proposed
diameter correlations used within the framework of the energy an empirical exponential law function of catalyst, feed, and
minimization multiscale model.127129 stripping steam flow rates.51
The behavior of catalytic cracking reactions of particle cluster Furthermore, there are steady-state studies that include the
in fluid catalytic cracking risers was recently numerically analyzed stripper/disengager vessel mass and/or energy balances. Some
using a hydrodynamics model coupled with a four-lump math- authors considered the energy balance of this vessel in their
ematical model.130 The effects of the cluster porosity, inlet gas model,92 whereas others modeled this section as a continuous
velocity, and cluster formation on cracking reactions were stirred tank in steady state, using a Stone & Webster proprietary
investigated. Distributions of temperature, gases, and gasoline model to estimate the cat-to-oil coke that remains adsorbed in
from both the catalyst particle cluster and an isolated catalyst the catalyst after stripping.142
particle are presented. Simulated results show that the reactions 2.5. Regenerator Models. In FCC modeling, studies of the
from VGO to gasoline, gas, and coke of individual particles in regenerator hydrodynamics and steady-state behavior have been
the cluster are slower than those of the isolated particles, but conducted early on by Errazu et al.143 and De Lasa and Grace.144
faster for the reaction from gasoline to gas and coke. Particle Moreover, the combustion phenomena in the regenerator, such
clustering will reduce the reaction rates from VGO to gasoline, as the CO postcombustion reaction145 and the prediction of the
gas, and coke and increase the reaction rates from gasoline to gas CO2/CO ratio in the flue gas,146 have also been a matter of
and coke.130 interest for research.
2.4. Stripper/Disengager Models. In contrast to that on the FCC regenerators usually consist of large fluidized bed
riser, research on stripper/disengager vessels published in the reactors with complex hydrodynamics, where the strongly
open literature is scarce.131 There are, nevertheless, some studies exothermic reaction of the combustion of coke on the catalyst
in cold-flow units (in the absence of reactions) using CFD for takes place. Inside the regenerator, two distinct regions can be
detailed modeling of the hydrodynamics of the stripper/disen- postulated: the dense region and the dilute region (frequently
gager vessels in the FCC unit. The main objective of these studies referred to as the freeboard). Most of the solids and gases are
is to improve the design of the internal hardware in this section, concentrated in the dense region, where almost all reactions
such as the riser termination device, the cyclones system, and occur, whereas the freeboard is a very low concentrated region,
stripper baffles132136 resulting from bubbles that erupt at the surface of the dense
Besides cold-flow modeling, only the authors that proposed fluidized bed, projecting particles into the region above. In this
models for the integrated reactorregenerator system have region, the concentration of solids decreases sharply with
described the stripper/disengager vessel as a component in their height.144 The catalyst particles carried away in the freeboard
models. This is particularly important in the dynamic models, are returned to the dense region via cyclones.
because in modern units the catalyst holdup in this vessel is high Because most FCC units have a large fluidized bed type
and contributes significantly to the transient behavior of the FCC regenerator, many of the original models for the regenerator
unit. Moreover, it is important to predict the percentage of section use the two-phase theory for modeling the dense phase
region. According to this theory,147 the gas flow is divided into

bubble and emulsion phases in the dense region. The emulsion
phase consists of the gas necessary to fluidize most of the solids.
The bubble phase is assumed to result from the excess gas in
relation to the minimum fluidization flow rate that passes
through the bed, as bubbles are essentially solid free. In the
two-phase theory, the bubble phase is considered in plug flow
and the emulsion phase is modeled as a continuous stirred tank
reactor (CSTR) or a plug flow reactor (PFR). Because the bubble
phase is solid free, no reactions or only gas-phase reactions are
considered in this phase.
Kunni and Levenspiel148 proposed the bubbling-bed model,
which is a three-phase model that assumes the presence of a thin
layer around the bubble, with less solid content than the emul-
sion (the cloud) and a similar zone being pulled up by the
bubble (the wake). Two-phase models neglect the cloud and the
wake, an assumption usually justified when small particles are
fluidized, as in FCC regenerators. Another model was proposed
to distinguish the grid region at the regenerator inlet.143,149 The
grid region is the gas feed zone at the bottom of the fluidized bed,
where the gas is not yet dispersed as bubbles but flows as jets that
will break down to bubbles in the highest part of the catalyst bed.
The jets are assumed to be well mixed in the radial direction and Figure 3. Fluidized bed regenerator model (adapted from Krishna
to follow a plug flow regimen in the axial direction. The emulsion et al.151).
and the bubbles in the grid region are assumed to be perfectly
mixed. The grid model was further improved by dividing the combustion in the dense region and also to afterburning reactions,15
domain into two regions: the grid region and the bubbling-bed modeling the dilute region of the regenerator can be of considerable
region.150 The bubbling-bed region was modeled with the two- interest to predict the regenerator temperature in this phase. There-
phase theory, whereas the grid region was subdivided into a dead fore, several authors have considered the freeboard and modeled it as
zone and a fully mixed zone. The dead zone includes two phases: a 1-D plug flow reactor (Figure 3).121,140,144,147,151,153155
the jet and the grid; no chemical reactions take place in either In contrast to the riser models, there are several regenerator
phase. The fully mixed zone is formed only by one gassolid models for the dynamic state when the entire FCC reac-
emulsion phase. In this zone, the porosity is the minimum torregenerator system is modeled.48,50,121,138140,154 The rea-
fluidization porosity and the gas flow is equal to the gas flow of son for this fact is that the regenerator’s dynamics contribute
the feed. The transfer line that carries the spent catalyst was also significantly to the FCC unit dynamics, because the regenerator
included in the regenerator description,151 which was modeled as presents a much higher characteristic time than the riser (of the
plug flow with chemical reaction. Simulation results showed order of some minutes to an hour, depending on the type of
that <4 wt % of the carbon on the spent catalyst removal occurs in regenerator).
this region. A detailed dynamic model of the regenerator as a standalone
Lee et al.152 applied three different models (two-phase, grid, unit was proposed by Faltsi-Saravelou and Vasalos,147 applicable
and bubbling-bed) of the dense region of a typical regenerator to particles of Geldart groups A (which is the case of FCC
and compared them with experimental results of an industrial catalyst) and B. Their model includes a rigorous description of
plant. They concluded that the bubbling-bed model gives the the hydrodynamics of a fluidized bed using the two-phase theory
smallest error in describing the experimental data. Steady-state and a detailed combustion kinetic model that considers carbon,
studies were carried out to compare the application of one- and hydrogen, and sulfur combustion. Later, Penteado et al.155
two-phase theory models to the regenerator, assuming that the proposed a dynamic model concerning only the regenerator of
dense region can be modeled as a CSTR.142,143 They concluded an industrial FCC unit. Their model includes both the dense
that there were no significant differences between the predictions region and the dilute region. The dense region is considered to
of the two models. This means that a model as simple as a CSTR have two phases: the emulsion phase and the bubble phase, both
can be used to predict the overall performance of a complex modeled as CSTRs. The bubble phase contains only gases, and
fluidized bed FCC regenerator. On the other hand, Secchi homogeneous oxidation of CO to CO2 is considered in this
et al.139 have also compared a one-phase model with the two- phase. The dilute region is modeled as a 1-D plug flow reactor.
phase model and concluded that the one-phase model does not For combustion, it is often considered that coke is composed
predict well the behavior of the regenerator and that a two-phase of only carbon and hydrogen. Moreover, no distinction is made
model should be used instead. These authors justified this between different types of coke molecules with respect to their
conclusion with the fact that, in the real plant, oxygen is not combustion. This last assumption might be subject to discussion,
homogeneously distributed in each phase. The CSTR model as it is assumed that gaseous hydrocarbons entrained with the
shows kinetic limitations to the consumption of oxygen only in catalyst (soft coke) burn in the same way as hard coke, which is a
the fluidized bed, therefore allowing larger consumption of solid deposited on the catalyst.137,156
supplementary quantities of oxygen. The kinetics of the regeneration of carbonaceous deposits on
There is more consensus in the modeling approach for the cracking catalysts has been extensively investigated since the
dilute region of the regenerator (freeboard). Due to incomplete early days of FCC. Arthur157 showed that the intrinsic carbon
burning at the catalyst site favored the formation of CO over CO2 hydrogen conversion be expressed as a function of carbon
and proposed a widely used correlation to predict the intrinsic conversion.153
molar ratio, σ, which states that σ decreases exponentially with 2.6. Fluid Catalytic Cracking ReactorRegenerator Mod-
temperature. Weisz and Goodwin158,159 presented a pioneer els. For the purposes of monitoring, controlling, and optimizing
work concerning the combustion of carbonaceous deposits on FCC units, both steady-state and dynamic models have been
porous particles. They found that the kinetics is controlled by the proposed.
rate of pore diffusion of oxygen to the coke surface. Observing 2.6.1. Steady-State Models. Corella et al.92 presented detailed
the coke combustion, they found also that the reaction kinetics steady-state mass and energy balances for a commercial FCC
were best described by a first-order rate equation. The intrinsic unit. A five-lump model is used to describe the cracking kinetics,
molar ratio CO2/CO was also a subject of their studies.146 They and eight “installation factors” are included in the heat balances
showed that this ratio is modified by further oxidation of CO to to provide a good match to particular units. This model can
CO2, catalyzed by the presence of transition metal oxides. therefore be used to study the influence of the main process
However, these early works were conducted for the regeneration variables, as well as to predict and improve the performance of
of amorphous bed catalysts at relatively low temperatures existing commercial units.
(∼600 °C). During the 1980s, different studies dealt with the Kumar et al.163 developed a steady-state process simulator
combustion of coke in zeolite catalysts, at temperatures up to (named “CATCRACK”) for a FCC reactorregenerator system.
800 °C and residual coke contents down to 0.05%.160,161 Kinetic This process simulator resulted from the integration of already
equations for both carbon and hydrogen combustion were existing models of the kinetics of cracking and reactorregenerator
proposed and fitted to experimental data. Hydrogen combustion hydrodynamics. The description of each component in the FCC
kinetics is much faster than carbon combustion kinetics at lower reactorregenerator system was based on a detailed discussion of
temperatures (600700 °C). Assuming that coke consists of existing individual models and their limitations. A 10-lump
only carbon, it was demonstrated that the intrinsic kinetic kinetics164 is used for the cracking reactions, assuming plug flow in
constant for coke burning at the reaction site was equal to the the riser and using Errazu’s grid model for the regenerator. This
global kinetic constant.145,161 The kinetic rates for coke burning model can be used in preliminary calculations for the design,
and for catalytic and homogeneous CO oxidation were obtained monitoring, and optimization of FCC units.
from experimental data on oxygen, CO, and CO2 concentrations, Another steady-state model of an industrial R2R FCC unit
assuming that coke burning and CO oxidation reactions are (Figure 1d) that can realistically predict the performance of this
additive. The rate equation obtained for carbon burning is first type of unit was presented by Vale.142 For the riser reactor, he
order in both carbon-on-catalyst and oxygen, whereas the rate considered a 1-D plug flow model with a six-lump model to
equations for CO oxidation reactions are both first order in CO describe cracking kinetics. For the regenerator, different models
and half order in oxygen. Another study, based on open literature were compared against industrial data, considering one or two
data and pilot plant studies,151 had already suggested the same phases in the dense region and different combinations of hydro-
dependences for the kinetic rate equations for carbon combus- dynamics (plug flow or CSTR). The author concluded that the
tion and CO homogeneous and catalytic oxidation (eqs 2 and 3). one-phase CSTR model gave better results.
The intrinsic molar ratio, σ, was considered to decrease expo- Gupta and Rao66 extended their riser model52 to the entire
nentially with temperature. The following stoichiometric equa- reactorregenerator system, to study the effect of feed atomiza-
tions were proposed for the overall reaction of the combustion of tion (quantified by the initial drop size produced by the feed
coke: nozzle) on the performance of the unit at constant coke yield.
They observed that for smaller drop size the overall performance
σ þ 2 x σ
CHx ðsÞ þ þ O2 ðgÞ f COðgÞ of the unit improved, presenting increased conversion and lower
2ðσ þ 1Þ 4 σ þ 1 regenerator temperatures.
1 x A series of papers focused on simulation studies using a steady-
þ CO2 ðgÞ þ H2 OðgÞ ð2Þ state model of a model IV unit (Figure 1b) that includes a three-
σ þ 1 2
lump model to describe cracking kinetics.165167 Both reactor
2COðgÞ þ O2 ðgÞ f 2CO2 ðgÞ ð3Þ and regenerator were modeled using the two-phase model for
fluidized beds from Kunii and Levenspiel.148 Their model was
Although hydrogen and water are considered in the global then used to study the effect of the main operating variables and
equation for coke combustion, no kinetic rate equation is feedstock compositions in the bifurcation behavior of the FCC
proposed to describe hydrogen burning. Neglecting hydrogen unit and its implications on gasoline yields.165,166 The effect of
combustion and assuming instantaneous combustion are com- the bubble diameter in the hydrodynamic models of the reactor
mon assumptions in the modeling of coke combustion. However, and regenerator vessels was also studied.167 Additional references
according to Faltsi-Saravelou and Vasalos,162 this reaction should also consider a modified steady-state model of an industrial FCC
be explicitly considered, because it has significant thermal effects. unit, including a riser type reactor modeled as 1-D plug flow
Wang et al.160 proposed a kinetic law for hydrogen combustion, reactor.168
which was first order both in oxygen and in hydrogen. 2.6.2. Dynamic Models. Several authors have presented dy-
A more recent experimental work has shown that the carbon namic models that describe the transient behavior of a FCC
combustion rate and the σ ratio are affected by additional unit.49,121,134,138,140,169175 A reference model for the FCC
variables apart from temperature, such as water concentration, system, cited many times, is the one proposed by McFarlane
coke conversion, and metal content. Therefore, a kinetic rate et al.171 They developed a dynamic simulator for the reactor
equation for carbon burning and a correlation for the intrinsic regenerator system and ancillary equipment (feed preheater,
molar ratio were suggested, depending on oxygen and water air blower, and wet gas compressor) of an old Exxon model IV
partial pressure and temperature. It has also been proposed that FCC unit for the AIChE industrial challenge problem. This model
represents well the dynamics of the catalyst circulation in that However, this model integrates a steady-state plug flow riser
specific unit and the interactions between the main process model with Weekman’s43,44 kinetic scheme with three lumps to
variables. Therefore, it can be used for identification, control, describe cracking kinetics.
and optimization. Nevertheless, the unit’s profit is largely depen- An enhanced model of the reactorregenerator system of a
dent on the market demand of the products; because this model R2R FCC unit was presented by Fernandes et al.180,181 to
lacks a detailed description of the catalytic cracking kinetics, it describe the steady-state and dynamic behavior of this unit. This
is consequently unable to predict the corresponding product model was later modified to describe the UOP FCC unit with a
distribution. high-efficiency regenerator, which differs mainly in the config-
Elnashaie and Elshishini174 have extended their steady-state uration of the regeneration section.180,182 The model was derived
model to a dynamic model of the riser reactorregenerator system from fundamental principles, considered different cracking ki-
to investigate both the open-loop and closed-loop behaviors of a netics (using a six-lump model adapted from Takatsuka et al.55),
feedback-controlled industrial model IV FCC unit. Zheng172 and explicitly described the freeboard region in the regenerator
developed a dynamic model of the reactorregenerator system vessel. The resulting model was fitted against industrial data for
to simulate the startup, shutdown, and routine operation of a FCC parameter estimation and validation, reaching a good agreement
unit. This author used the four-lump model to describe the cracking between simulated, industrial, and previously published results.
kinetics and assumed a steady state in the riser. The combustion A detailed description of important characteristics of these units,
kinetics are the same as in Krishna and Parkin.151 such as the hydrodynamics and the cracking kinetics, was con-
Moro and Odloak176 proposed a dynamic model of a Kellog sidered in this model, to allow its use in a large range of process
Orthoflow F converter, for use in control applications. Their optimization tasks, including model-based control.
model is based on a previous one that includes the riser, the Han and Chung121 developed a model for a FCC unit with a
reactor separation vessel, and the two regenerator stages that side-by-side reactorregenerator system with the purpose of
operate in partial combustion mode. Although the model is using it as a tool (i.e., a virtual plant) for carrying out various
capable of representing well the important dynamic aspects of process systems studies like control and optimization. Their
the system, it includes only the coke balance in the riser and no model describes a FCC unit comprising a riser, stripper/disen-
description is given for the product distribution. gager, regenerator, and catalyst transport lines with slide valves.
Arbel et al.138 presented a model for a typical side-by-side FCC Besides the mass and energy balances for the riser, the authors
unit with a bubbling bed regenerator that can predict both the also included momentum equations for the gas and catalyst
dynamic and the steady state of these units. For the riser, their particles. This allowed determining the velocity profiles along the
model assumed a pseudo steady state, plug flow, no slip velocity riser reactor and accounting for the slip velocity between gas and
between the gas and catalyst particles, constant superficial catalyst particles. A four-lump model was used to describe the
velocity, and adiabatic operation. The regenerator was modeled cracking kinetics.
considering the dense and dilute regions. The cracking kinetics A dynamic simulator of a FCC pilot plant has been used in the
were described by means of the 10-lump model proposed by Chemical Process Engineering Research Institute (CPERI) in
Jacob et al.164 Arbel et al.177179 used this model in posterior Thessaloniki, Greece.147 The simulator predicts the feed con-
studies on the state multiplicity and control problems of indus- version, coke yield, and heat of catalytic reactions in the FCC
trial FCC units. riser on the basis of semiempirical local models, based on pilot
Ali and Rohani49 and Ali et al.48 developed a dynamic model plant data. The ultimate scope of this research study was to use
that predicts both the steady-state and dynamic behavior of an the simulator in the development of model-based control
UOP stacked unit (Figure 1a). Their model includes a 1-D plug structures for the pilot plant, thus improving the process
flow riser with a four-lump kinetic model. The two-phase theory productivity for catalyst benchmarking experiments at constant
was used to model the dense region in the regenerator vessel, feed conversion and riser temperature, by allowing accurate
whereas the freeboard was neglected. This model was later targeting of prescribed operating conditions with a minimum
modified by Malay et al.,50 who included the effects of the number of experiments.
volumetric expansion of gases and the slip velocity between the
gas and solid phases in the riser. The model was developed for 3. STEADY-STATE MULTIPLICITIES IN FLUID CATALY-
design, control, and optimization applications. TIC CRACKING UNITS
For the stacked UOP FCC unit (Figure 1a), there is also a
model presented by Secchi et al.,139 in which the riser is described The possibility of the existence of multiple steady states in
FCC units has a major impact in the supervision of these systems.
as a plug flow reactor in dynamic state, however, with no overall
The causes for these behaviors are usually due to the exothermi-
mass accumulation. Slip velocity between the gas and catalyst was
city of the catalyst regeneration reactions and the strong inter-
not considered, and cracking kinetics was based on a 10-lump
actions between the reactor and the regenerator systems. Distinct
model.164 Two hydrodynamic models were proposed for the
approaches have been used to define and search for the existence
regenerator: a CSTR and a two-phase theory model. According
of steady-state multiplicity in a FCC unit.
to Secchi et al.,139 the use of the 10-lump kinetics allows the use
If one considers a generic process model of the form
of this FCC model in the development of new control strategies
needed to adjust the production on the basis of variable market dx=dt ¼ f ðx;uÞ ð4Þ
demands and to maximize the unit’s profit.
In an advanced control study, Cristea et al.175 presented a y ¼ hðxÞ ð5Þ
dynamic model for a UOP side-by-side FCC unit, with a
bubbling-bed regenerator that operates in the partial combustion where x ∈ Rn are the state variables, u ∈ Rm are the input variables,
mode. Their model is based on the work of McFarlane et al.,171 and y ∈ Rp are the output variables, a steady-state solution (uss,
using the original model equations for the ancillary equipment. xss, yss) to the above equations is denoted by the subscript ss;
thus, 0 = f(xss,uss) and yss = h(xss). In the definition of multiplicity associated with a given set of outputs. In this situation, the system
of equilibrium states, the trivial situations that originate from can possibly reach a steady state that is different from the desired
nonsquare systems should be excluded (e.g., when m > n), operating point.190
because in this case the number of solutions is generally A different line of study consists in looking for output multi-
unlimited and also occurs with linear models. Therefore, given plicities.140,165168 In this case, the study includes the influence
a known steady-state solution xss and assuming that m e n, the of the operating conditions in the number of steady states as well
above model presents input multiplicity when there is more than as their behavior. Arbel et al.177 have shown the differences
one solution to the square algebraic system between input and output multiplicities and addressed both
types of multiplicities. Due to the consecutive reactions in the
0 ¼ f ðxss ;u1 , u2;ss Þ ð6Þ regenerator, these authors reported the possibility of having more
where u1 ∈ R is a subset of the elements of u and u = {u1 ∪ u2}. If
n than three steady states, although they also conclude that in actual
m e n, the input multiplicity would be observed for multiple FCC operation there should be only three. Hernandez-Barajas
solutions of 0 = f(xss,u). These situations are of particular interest et al.140 also relate the possibility of finding more than three steady
when u1 (or u) can be manipulated for control purposes. states due to the inflections in the heat-generated curve that result
In a similar way, given a known steady-state solution uss, the from the migration from one combustion mode to another.
above model presents state multiplicity when there is more than However, in their simulation results, there is evidence of only
one solution to the square algebraic system: three possible steady states in both partial and complete combus-
tion modes. Furthermore, these authors suggest that the number
0 ¼ f ðx;uss Þ ð7Þ of steady states in FCC units is fixed and independent of the
operating conditions, at least within the typical operating range.
Finally, output multiplicity can be defined in cases when the
Elnashaie et al.168 considered the feedback effects resulting
algebraic system
from complex nonlinear interactions between the regenerator
0 ¼ f ðxss ;uss Þ ð8Þ and the reactor as the main cause for output multiplicity. After
investigation of several industrial units, these authors state that
y ¼ hðxss Þ ð9Þ the behavior of the different industrial FCC configurations differs
quantitatively, although qualitatively they are very similar. Ac-
has more than one solution. This can be caused by the presence cording to them, configurations with riser-type reactor and
of state multiplicities, particularly when y corresponds mostly to a regenerator are expected to exhibit output multiplicity, although
subset of the state vector. each part of the unit will not exhibit such behavior individually.
Because output multiplicities are easier to study and validate Also, according to Arbel et al.,177 all FCCs should exhibit multiple
than state multiplicities, their identification is much more often steady states, because they are autothermic reactors in which
carried out in practice. A common way of determining output the heat of combustion is used to heat the incoming feed to the
multiplicity in chemical systems is to plot the curves that refer to required temperature at which the reaction occurs.177,191 The
removed and generated heat. The intersections of these two high-efficiency regenerator has a behavior closer to a plug flow
curves represent the steady-state points of the system.183185 reactor than to a CSTR, and usually plug flow reactors do not
The identification of output multiplicities is important, from an show output multiplicity. However, if one considers the FCC
operational point of view, mainly due to two reasons:186 (1) system as an autothermal plug flow reactor with external heat
Having several steady states in feasible operating regions may exchange, then output multiplicity will likely occur. The interac-
result in very distinct economical returns. (2) Operating conditions between the endothermic reactions that occur in plug flow
tions can be changed in such a way that a stable steady state is lost. in the riser, together with the highly exothermic catalyst regen-
In this case, the unit may wind down to the cold steady state eration, lead to a system with multiple steady states within the
(which can be physically infeasible and imply the shutdown of the normal range of operating conditions.
unit) or else reach an unstable steady state that needs further More recently, Fernandes et al.180,181,186 presented an analysis
stabilization by the online control system. of the existence of output and input multiple steady states in a
Steady-state multiplicity is an important feature of FCC units UOP FCC unit with a high-efficiency regenerator. This study
and has been analyzed in various references.140,168170,177,186189 shows that changes in the operating conditions can lead to the
The first author to study this phenomenon was Iscol,188 in 1970. Lee possibility of finding between three and five multiple output
and Kugelman169 claimed that in a typical open-loop FCC there is steady states. However, this hypothesis is affected by unit
only a unique steady state at fixed feed rate, catalyst circulation rate disturbances and model uncertainties, such as coke, feedstock
(Fc), and air flow rate (Fair). However, their concept of uniqueness is composition, and cracking enthalpy. These authors also conclude
restricted to an operation region of practical interest. Oliveira189 also that other process inputs such as the regenerator pressure (PRG),
studied state multiplicity in FCC units, using the same model as Lee combustion air temperature (Tair), and feed temperature (Tfeed)
and Kugelman,169 and reached the same conclusion. can also influence the number (and position) of steady states.
Other authors have used the term “multiplicity of steady Figure 4 shows the effect of changing each of these operating
states” in the sense of input multiplicity. In their studies, they conditions, by fixing the remaining operating variables, in the
searched for multiple steady states by fixing process variables heat curves. Thus, changing the regenerator pressure (Figure 4a)
such as the feed flow rate, feed temperature, and combustion air will only slightly affect the curve of heat generated because higher
rate, while the catalyst circulation rate was allowed to vary to keep pressures favor the combustion rates, resulting in an increase of
the riser reactor temperature constant.170,187 Under these con- the heat generated. On the other hand, changing the combustion
ditions, two steady states were obtained. A system with input air temperature (Figure 4b) affects only the heat-removed curve,
multiplicities can be difficult to control because there exist more and, as expected, low air temperatures increase the quantity
than one set of steady-state values for the manipulated variables of heat removed, because more heat is needed to warm the
Figure 4. Effect of operating conditions in the heat generation and heat removal lines in an industrial UOP FCC unit with a high-efficiency
regenerator:186 regenerator pressure (a), combustion air temperature (b), and feed preheat temperatures (c).
combustion air inside the regenerator. Changing the feed tem- effective supervision of these units. Therefore, the identification of
perature (Figure 4c) affects both the heat-removed and heat- these model features is an important aspect of the dynamic study
generated curves. The heat-removed curve changes position, of FCC units and fundamental information for the synthesis of
because the requirements to heat and vaporize the feed flow rate adequate control structures for the corresponding industrial units.
also change. On the other hand, the feed temperature directly Fernandes et al.186 have shown that these stability problems
affects the temperature in the riser and subsequently affects the can be avoided by operating the FCC unit with a high-efficiency
formation of coke and the heat-generated curve. Furthermore, regenerator under full combustion conditions, using a 2 2
panels b and c of Figure 4 show that low feed and air tempera- control structure to monitor the riser reactor temperature and
tures can even result in the loss of the intermediate and upper the oxygen level in the flue gas. Moreover, the results obtained by
steady states. This situation would result in a set of nonpermis- Fernandes et al.180,181,186 have also shown that it is not recom-
sible operating conditions, because the only steady state available mended to use the same control structure for operation in partial
would be the cold point.186 These different heat curves and and full combustion modes, especially with traditional linear
consequently the multiple steady states can be very close to each controllers, because significant differences were found in the
other, so in industrial operation the behavior can shift from one dynamic behavior of these two cases. In this situation, the
steady state to another, and this is an incentive for choosing a operating point should be chosen in a manner such that a
robust control system. transition between modes of combustion will not occur. Given
Therefore, from a control point of view, particular care is the effects that possible perturbations in the feedstock quality and
required in the choice of manipulated variables that will allow other unit disturbances might have, this also demonstrates the
correct operation of the unit in the region of five steady states. importance of understanding the nonlinear behavior of the FCC
Depending on the choice of the control structure, input multi- unit and the possible incentives to the use of more sophisticated
plicities can also cause stability problems. An important conse- nonlinear control strategies.
quence of the input multiplicity is that there are usually points in It is possible to envision a control strategy for industry based
the operating region where the steady-state gain matrix becomes on two legs: a linear one for control around a stable steady-state
singular.192 This means that changes in the sign of the steady-state point combined with a nonlinear one for more significant
gains are possible, which poses additional difficulties, especially variations such as major disturbances, startup, or shutdown.
for linear control actions, because if the controller gains are
maintained constant, positive feedback will be introduced in the 4. MONITORING OF PROCESS AND QUALITY VARIABLES
system.189 Hence, for processes with such characteristics, non- Extensive collection and analysis of process data are required
linear control strategies can be the only feasible approaches for for the safe operation of FCC units and also to evaluate and
improve the efficiency of operation. In addition to online process Simultaneously, it provides a database for process review, which
measurements gathered by the process distributed control can be used to detect abnormal modes of operation and to flag
system (DCS), this also includes various offline measurements, developing problems for correction.
essential to evaluate the quality of the products and to under- Besides the use of linear statistical methodologies, artificial
stand their relative yields. Relevant data include (1) mass balance neural networks (ANN) have also been used to regress FCC data
data (product flows, densities, pressures); (2) heat balance data and estimate various product quality properties, using process
(temperatures, flue gas composition, air rate); (3) feedstock charac- and laboratory data. A neural network is a processing structure
teristics (density, refractive index, sulfur, metals, etc.); (4) essen- consisting of an arrangement of interconnected simple proces-
tial product properties (distillation curves, octane numbers, sulfur, sing elements (called neurons) and is characterized by parallel
viscosity); and (5) equilibrium catalyst analyses (activity, residual computation and “learning” capabilities, which can approximate
coke, metals, among others). arbitrary nonlinear mappings.198,199 ANN empirical models can
In many cases, measurements that enable the closure of the predict the regenerator average temperature, catalyst circulation
heat and material balances are available on a continuous basis rate, reactor top temperature, and product flow rates (fuel gas,
from the process DCS. Feed and product analyses are usually LPG, naphtha, LCO, HCO, and slurry).200 In addition, ANN
performed at least weekly in the laboratory.193 The e-cat analyses models can be built to estimate several product quality proper-
are usually conducted weekly. To minimize the cost associated ties such as density, sulfur content, initial boiling point, RVP,
with these analyses, the product properties are usually limited to a octane number, and product compositions (aromatic and olefin
number of well-defined key characteristics. Various methods contents).
exist to estimate hydrocarbon composition or carbon atom One important difficulty in the effective analysis of process
distributions in FCC feeds, as published by ASTM, API, and information is the relatively large volume of information that is
some oil companies.194 However, the unit performance is often available. To help in this task, Wang and McGreavy201 developed
explained adequately using just the basic analyses such as density, a data-mining system capable of classification of process data into
refractive index, or aniline point directly, without conversion into classes, corresponding to different operational modes or regions,
compositional breakdowns. The carbon atom distribution can be relying on a Bayesian automatic classification method. This
used to estimate the maximum crackability of the feedstock. All of system was tested using a FCC model provided by a dynamic
these data need to be validated through the elimination of gross simulator developed for training industrial operators. Various
errors and inconsistencies through the use of a data reconciliation cases were simulated corresponding to normal operation and
methodology195,196 before they can be used in model-based various failure conditions. A comparison of the performance of
tasks, such as advanced control and optimization and process the system with an ANN is also provided. The authors concluded
diagnosis. that, in both cases, reasonable classification for practical purposes
Due to interactions between the reactor and regenerator units, can be achieved, provided that the thresholds for discrimination
the heat balance of a FCC unit is of major importance to are adequately represented.
characterize the efficiency of the operation. The most significant The reduction of operation data to a small number of
independent variables in the heat balance of the reactor and operating regions is also considered by Sebzalli and Wang,202
regenerator are (i) the quantity of processed feed, (ii) the using principal component analysis (PCA) and fuzzy clustering.
properties of processed feed, (iii) circulating catalyst properties This technique allows the determination of the variables that
(activity and selectivity by coke), (iv) the temperature of contribute most importantly to the definition of each operating
combined feed, (v) the temperature at discharge of lift reactor, region, therefore facilitating the development of simpler but
and (vi) the relation of combined feed. The three most significant robust process monitoring strategies. PCA is also used to detect
dependent variables are regenerator temperature, catalyst-to-oil abnormal conditions in sensor validation and fault diagnosis of
(CTO) ratio, and coke yield. Often the regenerator mass balance FCC units by Pranatyasto and Qin.203 This technique is able to
provides the air/coke and hydrogen/coke ratios, whereas the identify process faults even when used in a unit equipped with
heat balance is used to calculate the catalyst/oil ratio (from model predictive control (MPC). This observation is relevant to
regenerator side balance) and the heat of cracking (reactor practical experience because with the introduction of feedback
side).197 terms in the controller, and its multivariable nature, the effects of
Unit monitoring data constitute often the basis to predict the faults in the measured and manipulated variables can become
effects of process changes from previous operating data, using closely related.
various statistical frameworks. For instance, this approach can be Han et al. considered the use of process data for parameter
used to model the effects of frequent changes in the feedstocks on estimation of a steady-state model of a FCC unit under the full
the product yields, and their properties can be predicted from and partial combustion modes.180 Depending on the type of data
existing experience. When a catalyst replacement is evaluated, the available, a total of up to 52 model parameters can be simulta-
same calculations can help to normalize the unit’s data to neously adjusted. After minimization of the plantmodel mis-
constant conditions and feed quality, unit operating conditions, match, this model is used for process optimization using an
and catalyst type. This is also true for variables such as the economic objective function. Because essentially the same de-
gasoline octane and the FCC standard conversion (defined as the tailed model can be used for data reconciliation and parameter
mass or volume percentage of VGO and/or residue that is estimation (and later process optimization), various authors
converted to products other than oils such as LCO, HCO, and advocate that these two tasks should be considered simulta-
DO). The predictions can be used in a straightforward sensitivity neously through the solution of a large online optimization
analysis to increase the profitability, for example, by variation of problem, using a global errors-in-variables approach, for which
the feed rate or other operating parameters. Consistent monitor- decomposition strategies are available to speed its solution.204,205
ing of the performance of the unit also offers the opportunity to A number of industrial reports reinforce the importance of the
test whether the catalyst in use is at its optimal activity level. parameter estimation task to the success of developing adequate
models for the application of model-based supervision strategies

to FCC units.206208
5. CONTROL AND OPTIMIZATION OF FLUID CATALY-

TIC CRACKING UNITS
Huge economic incentives are normally tied to operating FCC
processes at optimal conditions, because these units are respon-
sible for approximately 40% of the gasoline pool.207,209 In
practice, this implies remaining close to physical and operating
constraints of various kinds (blower capacity, limit material
temperatures, etc). Given the complexity of the process, achiev-
ing this goal while simultaneously guaranteeing the safety of
operation requires the use of elaborate supervision strategies,
making these units natural candidates for the development and
implementation of advanced control and real-time optimization Figure 5. Hierarchy of control structure in a modern processing plant
studies.198,207,210 using MPC technology (adapted from Qin and Badgwell221).
A key aspect in most of these strategies is the presence of
various layers, each addressing a specific set of operational the higher levels will depend critically on the effectiveness of this
objectives, arranged in a hierarchy of decisions, as illustrated in regulatory control system. In their study, they present a quanti-
Figure 5. The lowest level is the regulatory control (also called tative analysis of five 2 2 control structures for the partial
basic or local control), which is composed by the control loops combustion mode and three 2 2 control structures for the
that allow safe basic operation of the unit. These loops are often complete combustion mode. Some of these structures were also
linear controllers with proportionalintegralderivative (PID) previously analyzed by Hicks et al.226 and Kurihara,227 who were
actions integrated in the DCS of the plant. The remaining block two of the first authors to address the issue of loop selection for
levels in the control hierarchy address the plant optimization at regulatory control. To evaluate the controllability of the FCC
different levels, by coordinating the various loops in the regula- system using each of these control structures, the right half-plane
tory level and specifying additional degrees of freedom in the transmission zeros (RHP) and the RGA of the process transfer
process. Research studies concerning FCC control applications function matrix were determined. The control structures con-
at each level in the control hierarchy can be found in the sidered (Table 1) allowed the conclusion that for partial combus-
literature175,207,211220 and will be discussed in the next section. tion mode the most appropriate control structure is the one
5.1. Control of Fluid Catalytic Cracking Units. Besides the proposed by Hicks et al.,226 although the use of the conventional
need to operate close to process constraints, strong interaction control structure has been predominant for many years. For the
between the individual control loops and the nonlinear behavior complete combustion mode, the most appropriate structure is
of the unit constitute major challenges in the design and the conventional control structure with the pair [TRT, yO2,RG]T as
implementation of the basic regulatory layer.211 This control controlled variables.
structure has also to deal with time constants of different scales Arbel et al.178,179 also studied the problem of the selection of
and to efficiently reject unmeasured disturbances.175 A key aspect appropriate control structures, using a different approach. The
in the design of this control layer is the choice and pairing of following elements were considered: the derivation of a reduced
controlled and manipulated variables. Previous experience has mathematical model, the identification of the dominant variables
shown that a relatively small set of variables has a primordial in the system, the study of the nonlinearities (including the
importance in the safe and efficient operation of the unit.194,198,211 determination of the regions with output and input multi-
The most important controlled variables are the stripper catalyst plicities), checking the time scale of responses to set-point
level (which provides stabilization for catalyst circulation), the riser changes and disturbances, and the sufficiency of the control
temperature (which directly influences the feedstock conversion), system relative to the desired specifications and constraints. Here
the stack gas oxygen concentration (to achieve the desired coke the dominant variables are defined as the process variables that
combustion), the regenerator temperature (to allow the required tend to have a stronger impact on the overall process behavior
recovery of catalyst activity, without causing excessive irreversible and, therefore, intuitively constitute good candidates for con-
catalyst deactivation), and the cyclone temperature (due to the trolled variables, as their control helps to maintain a larger set of
afterburning effect, thus providing safe thermal operation). The process variables within their respective bounds. A good example
most important manipulated variables include the spent and of a dominant variable is the riser temperature. Partial control
regenerated catalyst flow rates, the stack gas flow rate, and the means that control of these dominant variables is sufficient to
blower air flow rate.178,198,211,222224 indirectly achieve acceptable control of the unit, whereas exact
5.1.1. Regulatory Control Structures. Several researchers have control aims at keeping the controlled variables at their set
relied on the relative gain array (RGA) technique for the points. Although Arbel et al.178 reach the same conclusions as
choice of the pairings between controlled and manipulated Hovd and Skogestad211 on the selection of the control structure
variables.211,223 Hovd and Skogestad211 presented a detailed for complete combustion mode, for partial combustion mode the
study on the selection of the structure of the regulatory control structure that seems to perform better is the riserregenerator
loops in FCC units, based on the analysis of linear models control structure (Table 1). Besides the distinct analytic meth-
derived from the nonlinear model of Lee and Groves.225 These odologies, and like the different conclusions derived by various
authors also state that a good selection of the control structure at authors on the multiplicity of steady states, these contradictory
the regulatory level is fundamental, because the performance of results can result from significant differences in the mathematical
Table 1. FCC Decentralized Control Structures211

control structure combustion mode manipulated variables, u controlled variables, y observations
conventional partial [Fc, Fair]T [TRT, ΔTRG]T Fc controls TRT; Fair controls ΔTRG
Kurihara partial T
[Fc, Fair] [TRG, ΔTRG]T Fair controls TRG; Fc controls ΔTRG
alternative Kurihara partial [Fc, Fair]T [TRG, TCY]T Fair controls TRG; Fc controls TCY
Hicks Partial [Fc, Fair]T [TRT, TCY]T Fc controls TRT; Fair controls TCY
riserregenerator partial [Fc, Fair]T [TRT, TRG]T Fc controls TRT and Fair controls TRG
conventional complete [Fc, Fair]T [TRT, yO2,RG]T Fc controls TRT; Fair controls yO2,RG
Kurihara complete [Fc, Fair]T [TRG, yO2,RG]T Fc controls TRG; Fair controls yO2,RG
riserregenerator complete [Fc, Fair]T [TRT, TRG]T Fc controls TRT; Fair controls TRG
models used in each case. This phenomenon reinforces the More recently, the application of a statistical framework to
previous observation that detailed mathematical models are characterize the dynamics of the riser temperature was
needed to capture the most important details to enable an proposed,229 with the purpose of decreasing its variability and
accurate description of the global dynamic behavior of the unit. increasing gasoline yield. The model used by the authors was
Alhumaizi and Elnashaie224 considered six different propor- based on the model proposed by Ali et al.222 and was validated
tional control loops and investigated their effect on the bifurca- with operating data from a real unit. On these bases, the authors
tion behavior and gasoline yield of an industrial model IV FCC concluded that the two main causes for large variability observed
unit (Figure 1b). This included four single-inputsingle-output were the presence of nonlinearities and chaos. After model
(SISO) and two multi-inputmulti-output (MIMO) control validation, the authors retuned the parameters of the two
structures, where the temperature of the combustion air and operating PID controllers with respect to three step distur-
the catalyst circulation rate were used as manipulated variables bances: feed flow rate, feed temperature, and saturated steam
and the reactor temperature and the dense bed regenerator flow rate. A decrease from (5 to (1.5 °C in the variability of the
temperature were the controlled variables. The global stabiliza- riser temperature as well as a relative increase of 17% in the
tion of the industrial operating point was achievable only when gasoline yield was claimed.
using the combustion air temperature as the manipulated vari- 5.1.2. Advanced Control Strategies. Besides the conventional
able. Moreover, using the regenerator dense phase temperature PID feedback controllers, more complex control strategies are
as the controlled variable, the MIMO control loops allowed the normally used in industrial FCC operation.199,217,218,230233
use of lower gains to globally stabilize the system around the Among these strategies, model predictive control (MPC) has
desired set point. They also concluded that the operating point been thoroughly studied and is well accepted in industrial
with the maximum gasoline yield needs very high values of the practice. Its multivariable characteristics, together with the
controller gains to be stable with the proposed control solutions inherent anticipative nature, and the aptitude to explicitly handle
for the FCC unit, which justifies the use of more advanced constraints in the controlled and manipulated variables give
control strategies. MPC a better control performance over conventional decentra-
Ali et al.48 and Arandes et al.154 studied the inclusion of the lized PID control.198,221
three primary PI controllers commonly used in industrial FCC At each sampling instant, the MPC algorithm attempts to
units operating in complete combustion mode to control the optimize future plant behavior by determining the best sequence
reactor temperature, catalyst bed level in the stripper. and differ- of future manipulated variable actions relative to a given perfor-
ential pressure between the reactor and regenerator. The manipu- mance objective, considering a set of constraints on the process
lated variables are, respectively, the regenerated catalyst flow rate, variables. The first input of the optimal sequence is then sent to
the spent catalyst flow rate, and the flue gas flow rate that exits the the plant, and the entire calculation is repeated in each of the
regenerator. Ali et al.222 obtained the RGA for this 3 3 control subsequent sampling intervals.221 Most MPC controllers mini-
structure and confirmed that these were the most appropriate mize a quadratic control cost function, with the future behavior of
pairings between these controlled and manipulated variables. controlled variables being specified by one of four basic alter-
Fernandes180 determined the best pairings for decentralized natives: set point, zone, reference trajectory, or funnel. Using a
control using 2 2 and 4 4 schemes, by calculation of the RGA set-point specification, any deviations from this value are pena-
matrix. For the 2 2 scheme the conventional control structure lized. If the controlled variables are specified within a zone, then
(Table 1) was obtained, whereas in the 4 4 case the best pairing penalization will occur whenever they are predicted to lie outside
was obtained with the regenerated catalyst slide valve, spent the zone with constant bounds. In the case of a reference
catalyst slide valve, stack gas slide valve, and combustion air flow trajectory, penalization in the objective function will exist when-
rate controlling the riser temperature, catalyst level in the ever the predicted value is different from the reference value.
stripper, regenerator pressure, and stack gas oxygen content, Finally, if a funnel (zone control with various upper and lower
respectively. However, to compare the dynamic responses with bounds) is used, then penalization will occur every time the
industrial data, only three of the four control loops were imple- controlled variable is predicted to lie outside the bounds.221,234
mented: riser temperature, stripper catalyst level, and regenerator One of the major requirements for the use of MPC is the
pressure loops. These three loops were “field tuned”228 because the availability of a process model; these can be either linear (LMPC)
strong interactions caused the process to become unstable when or nonlinear (NMPC). The most popular MPC applications are
controllers tuned using ZieglerNichols (ZN) rules were used based on linear models, with a quadratic objective function that
instead. Good adjustments were reported in the comparison of the can be easily solved with available optimization software, result-
responses of the controlled variables. ing in reliable and practical MPC applications. Apart from the
Table 2. FCC Control Structures for LMPC175

control scheme
(name/dimension) controlled variables manipulated variables MPC tunning parameters
S1: 3 3 Wr, TRG, TRT FRC, FSC, FPF Uwt = [120, 120, 0.8]
Ywt = [0.1, 0.2, 1]
S2: 3 3 Wr, TRG, TRT FRC, FSC, FS Uwt = [120, 120, 480]
Ywt = [0.1, 0.2, 1]
S3: 3 3 Wr, TRG, TRT FRC, FSC, FV Uwt = [120, 120, 600]
Ywt = [0.1, 0.2, 1]
S4: 3 3 Wr, TRG, TRT FRC, FSC, FV Uwt = [75, 75, 300]
Ywt = [0.1, 0.2, 1]
S5: 4 4 Wr, TRG, TRT, yO2,RG FRC, FSC, FS, FV Uwt = [30, 30, 120, 120]
Ywt = [0.1, 0.2, 1, 0.5]
S6: 4 4 Wr, TRG, TRT, yO2,RG FRC, FSC, FPF, FV Uwt = [150, 150, 1, 600]
Ywt = [0.1, 0.2, 1, 0.5]
S7: 4 4 Wr, TRG, TRT, yO2,RG FRC, FSC, FMF, FV Uwt = [150, 150, 300, 600]
Ywt = [0.1, 0.2, 1, 0.5]
S8: 5 5 Wr, TRG, TRT, yO2,RG TCY FRC, FSC, FPF, FV, FMF Uwt = [150, 150, 1, 600, 300]
Ywt = [0.1, 0.2, 1, 0.5, 0.5]
S9: 5 5 Wr, TRG, TRT, yO2,RG, TCY FRC, FSC, FPF, FV, FMF Uwt = [30, 30, 0.2, 120, 60]
Ywt = [0.1, 0.2, 1, 0.5, 0.5]
S10: 5 5 Wr, TRG, TRT, yO2,RG, TCY FRC, FSC, FPF, FV, FS Uwt = [150, 150, 1, 600, 600]
Ywt = [0.1, 0.2, 1, 0.5, 0.5]
S11: 5 5 Wr, TRG, TRT, yO2,RG, TCY FRC, FSC, FMF, FV, FS Uwt = [150, 150, 300, 600, 600]
Ywt = [0.1, 0.2, 1, 0.5, 0.5]
S12: 5 6 Wr, TRG, TRT, yO2,RG, TCY FRC, FSC, FPF, FV, FMF, FS Uwt = [150, 150, 1, 600, 300, 600]
Ywt = [0.1, 0.2, 1, 0.5, 0.5]
reduced computational effort required, this allows a desired ability to more accurately model the performance of control
steady state to be maintained, using a linear model accurate in valves.235 The primary costs to switch from an existing controller
the neighborhood of a single operating point.198 For processes will be the engineering time to convert the closed-loop model,
with significant nonlinearities or that need to be operated over commission the controller, and train the operators.
larger variable ranges, nonlinear MPC has the potential to The application of linear MPC to industrial FCC units is
provide better control performance and to allow the incorpora- considered by Caldwell and Dearwater,236 describing its integra-
tion of more complex performance objectives, for example, tion with the existing regulatory control structure, in order to
related to the economical optimization of the process unit. This create a flexible structure capable of supervising different operat-
possibility is discussed in a separate section. ing modes of the unit. This solution is able to consider both
An alternative to the control hierarchy, presented in Figure 5, complete and partial combustion modes in the regenerator, using
is the development of the adaptive dynamic matrix controller an implementation partitioned on two control modules: regen-
(ADMC).235 This controller uses the DMC algorithm and erator loading control (regulating the combustion in the regen-
operates at a high frequency on an open-loop model of the erator) and severity control (regulating the reaction temperature,
process. ADMC eliminates the need for PID controllers by manipulating the feed rate, and addressing various operational
including the regulatory loops (which are usually under PID constraints). The loading control uses a SISO MPC with
control). This elimination increases the signal-to-noise ratio of feedforward inputs for the severity control. This last module is
the controlled variables. Further advantages offered by ADMC based on MIMO MPC, incorporating a priority structure for the
include the ability to operate directly on a constraint and the selection of the manipulated variables used. The nonlinearity of
the model and the effects of plant mismatch are addressed by model proposed by Cristea et al.175 extended to include the
control variable linearization and use of correction terms in the main fractionators.219 Both MPC controllers outperformed the
model, based on predicted trends. The controlled variables PID loop strategy in rejecting disturbances and tracking set-
considered can also be selected within a group, according to point changes, with the NMPC solution yielding better results
the most binding constraints. A single control move is calculated than LMPC.
in each predictive horizon, because the authors found no Dong and Riggs218 focused on the detection of unmeasured
improvement in the control performance relative to this tuning disturbances using a χ2 test and observed that the performance of
parameter. A rigorous steady-state model is also used to identify MPC applications in FCC can be compromised by the effect of
the priority selection structure for the manipulations and the unmeasured disturbances in the data used to identify the MPC
unconstrained input targets. model. The usefulness of the proposed method was evaluated
Kalra and Georgakis237,238 presented a study on the perfor- using the detailed dynamic model presented by Han and
mance of a LMPC algorithm for the control of a FCC unit. They Chung51 to represent the real process behavior. The authors
first considered the performance of the MPC algorithm based on arrived at the conclusion that the proposed asymptotical detec-
models obtained from step tests, using the nonlinear model tion method successfully identified most of the corrupted
proposed by McFarlane et al.171 Four specific models were data. This enhanced the accuracy of the MPC model, relative
derived at four different throughputs. Contrarily to what was to the dynamic model taken as the “real” response. Additionally,
expected, the authors report in the first paper237 that linear MPC the method was shown to be insensitive to the level of sensor
performed better at the higher throughput point, when using a noise verified.
model derived from step tests relative to a lower throughput Oliveira and Biegler193 studied the effect of active constraints
operating point. Consequently, in their following paper model on the closed-loop stability of MPC controllers. The authors
identification was done using pseudorandom binary sequences showed that active output constraints can modify the feedback
(PRBS) as process inputs,238 because the authors considered the law of the predictive controller in forms that can completely
possibility that the step tests did not adequately excite all of the overshadow the effect of the tuning parameters and cause closed-
dynamic modes of the process. The authors concluded that in loop instability. A similar effect can result from the presence of
specific circumstances the step test methodology might not be a active input constraints in open-loop unstable processes. This
safe approach to derive the required dynamic models. effect is common both to LMPC and to NMPC and is especially
Comparison studies between the performance of MPC algo- important for FCC units that frequently operate closer to
rithms and PID decentralized loops were done by several constraints, where the optimum operating points are located.238
authors.154,219,222,239241 Jia et al.240 used a subspace identifica- This behavior was demonstrated with the Lee and Kugelman169
tion method (N4SID) available in Matlab to derive a linear step model. To overcome this difficulty, Oliveira and Biegler193
model based on the nonlinear dynamic model previously pre- proposed the reformulation of the problematic constraints as
sented by Ali et al.222 and Malay et al.50 The authors demon- soft constraints, using an exact penalty approach that is able to
strated that MPC provides better performance when compared retain the stability properties of the unconstrained system.
to the conventional regulatory control, because the PI controllers Due to their accentuated nonlinear behavior, FCC units were
were not able to bring the system to the desired operation point among the process applications that motivated the development
in some scenarios, despite the use of integral action. of NMPC algorithms. Oliveira189 used the model of Lee and
Cristea et al.175 tested different dynamic matrix control Kugelman to assess the performance of a state-space NMPC
(DMC) based MPC schemes, selected on the basis of a con- methodology, based on the efficient computation of model
trollability study that included a RGA analysis, and compared the sensitivities. This solution was able to provide close control of
results to decentralized PID structure. The different schemes are the unit over a wide range of operating conditions (including
presented in Table 2. For this purpose, a first-principles model zones of open-loop instability), without changes in the tuning
was derived for the UOP Amoco model IV with the main parameters. The accentuated nonlinear behavior of the simple
fractionators feed preheat system and combustion air blower. process model considered is clearly visible from the profiles
The riser kinetic scheme was the three-lump kinetic model associated with the case studies presented and from the signifi-
proposed by Weekman and Nace.44 The best control schemes cant changes in the process gains in the ranges studied.
for each category were S1 (3 3), S5 (4 4), and S10 (5 5). Another NMPC algorithm with modified state estimation was
The S12 (5 6) control scheme proved to be useful only when proposed by Ali and Elnashaie222 for stabilizing the operation of
constraints in the manipulated variables were used, due to the an industrial model IV unit (Figure 1b) around an unstable high
loss in the number of degrees of freedom. Compared to the gasoline yield operating steady state. The authors used the
decentralized control, the MPC showed improved performance nonlinear model previously proposed by Elnashaie and
in both reducing overshoot and settling time. The authors also Elshishini174 for the prediction of the outputs. The use of NMPC
tackled the issue of linearization errors compromising the LMPC coupled with state estimation allowed operation of the FCC unit
performance by using a model scheduling approach. This at the unstable region and revealed it to be better than a
approach consists of updating the linear models with a given conventional PI controller.
frequency (in this case, every 3000 s). Ansari and Tade242 proposed a constrained nonlinear model
A comparison of the performance of conventional decentra- predictive controller strategy for the reactorregenerator system
lized PID control with advanced control strategies using the of an FCC unit, based on the GMC algorithm. The objective
model developed for a UOP type unit showed that the decen- function penalized absolute deviations from the reference trajec-
tralized approach was outperformed by the advanced control tories, together with deviations in the maximum rates of changes
strategy in both the overshoot and response time.217 A similar toward constraint bounds. An approach based on weighted least-
study, focusing on the performance of nonlinear MPC to linear squares to find optimal parameter change was used to update model
MPC and decentralized PID control, was done using the FCC parameters, to reduce the plant/model mismatch. The resulting
nonlinear optimization problems were solved with the SQP control (FLC). This control strategy aims at emulating a human
algorithm of the Matlab Optimization Toolbox. This strategy expert, in this case with the knowledge of the human operator
was tested in a real-time application, and the results were transformed into a set of fuzzy linguistic rules, to approximate a
compared with a linear MPC controller; the performances decision. This approach has also been used to control FCC
of both control strategies were satisfactory, although the non- units.153,197 Taskin et al.247 applied this methodology, choosing
linear GMC controller presented better results. According to 10 input variables and implementing 30 rules. The authors
the authors, the use of nonlinear MPC control allowed the solu- argued that the results are in agreement with empirical FCC
tion of smaller optimization problems. A strategy based on a data and concluded that additional proper and accurate results
unique dynamic model also reduced the need for maintenance can be expected by increasing the number of input variables,
and process re-identification efforts often necessary with linear increasing the data used to derive the rules for each variable.
controllers. More recently, Cristea et al.217 compared the FLC with ANN-
Harnischmacher and Marquardt243 used block-structured based MPC control and classical regulatory control, using anti-
models (consisting of nonlinear static and linear dynamic windup PID digital controllers. The FLC approach considered
elements) to develop a NMPC solution with lower computa- triangular and trapezoidal membership functions to “fuzzify”
tional cost. Block-structured models provide approximations of a both manipulated and controlled variables and yielded shortened
more rigorous nonlinear model, establishing a compromise settling times and overshoots.
between the evaluation of costly nonlinear dynamic models Additional advanced control strategies studied include the use
and simpler linear models with poorer prediction abilities. On of coordination between control loops (such as cascade control)
the basis of the structures of discrete-time Hammerstein and and the use of state and parameter estimation to drive process
Uryson models, the authors derived a methodology for the models that require detailed plant information. A strategy based
efficient calculation of the sensitivity information needed for in a MIMO cascade control system was proposed by Alvarez-
the solution of the corresponding optimal control problem. Also, Ramirez et al.215 to regulate the gasoline composition in the FCC
to avoid the disadvantages of nonlinear models in MPC applica- products slate. Because this work is based on the same nonlinear
tions and the shortcoming of linear models, various authors model as the one used by Hovd and Skogestad,211 they used the 2
considered the use of ANN to obtain empirical nonlinear 2 Hicks control structure for the regulatory control level of the
dynamic models.199,221,232 FCC unit. To design the multivariable temperature regulation loop,
Nagy et al.199 compared the use of an off-line trained ANN linear inputoutput models were obtained, from a step identifica-
MPC scheme to an adaptive neural network MPC scheme, to tion test in the nonlinear model. In the proposed MIMO cascade
control an Amoco model IV FCC unit. They concluded that the control system the slave controller is the loop that regulates the riser
latter had better performance, based on the value of the integral temperature and the master is the one that uses the riser tempera-
of the square error (ISE). Nonetheless, both outperformed the ture set point to regulate the gasoline composition.
3 3 decentralized PI structure proposed by the same authors. Control studies that include estimation techniques are able to
Alaradi and Rohani232 used feedforward neural networks for reduce plant/model mismatches by accounting for modeling
the steady-state and dynamic identification and control of a FCC errors and uncertainty in the process parameters, as model
unit. To provide training and testing data, the authors used the parameters are updated online.209,248,249 Aguilar et al.209 devel-
model of a FCC unit developed by Ali et al.222 A back-propagation oped a robust PID controller to control the riser reactor
algorithm with momentum term and adaptive learning rate was temperature. Uncertainties related to heat generation by the
used for training the identification networks and for neurocontrol cracking reactions were addressed using a proportional-derivative
of the process. Noise-free simulations showed that the adaptive reduced order observer that produced estimates for the heat of
neurocontroller performed well in the presence of set-point and reaction. This control strategy showed enhanced closed-loop per-
disturbance changes. Also in this case, the adaptive neurocon- formance, in comparison to ZieglerNichols (ZN) and internal
troller outperformed the PID controllers.232 model control (IMC) tuning methodologies.
Vieira et al.244 implemented and evaluated the performance of Huang et al.249 used a heuristic extended Kalman filter
a neural network-based MPC applied to a Kellogg Orthoflow F (HEKF) for fault identification in a model IV FCC unit.
unit. The neural network model was obtained from simulated data This study was developed on the basis of the FCC model of
using the nonlinear model proposed by Moro and Odloak176 and McFarlane et al.171 The HEKF differs from the EKF by using
introduced into constrained MPC. Overall, simulation experiments pseudomeasurements, obtained by interpolation between two
confirmed good regulatory and tracking properties of the proposed measurements, effectively reducing the sampling period and the
control system. Moreover, a comparison to the DMC algorithm linearization error. Huang et al.249 compared the EKF to the
showed that the neural network MPC provided comparable and HEKF, in single-fault and multifault scenarios. The fault scenar-
smooth performance. The behavior of the neural network MPC ios included variations in coke factor of the feed, heat transfer
within noisy measurements proved to be adequate for implementa- coefficient in the feed preheat furnace, catalyst particle density,
tion in an industrial environment. ambient temperature, and combustion air blower capacity.
Santos et al.245 compared MPC based on ANN with a Huang et al.249 observed that the EKF failed to converge for all
regulatory control scheme, using the model proposed by Moro of the multifault scenarios, whereas the HEKF was capable of
and Odloak176 as a substitute for the “real” process, and noted the obtaining the true estimate for all of the scenarios in a faster and
superiority of the former. However, and despite its potential more accurate way. They concluded that the use of HEKF is
advantages, the use of ANNs is constrained by the large training appropriate for large-scale nonlinear model-based state and
data sets required and the fact that, being black-box models, no parameter estimation, especially for use in fault detection, isola-
fundamental knowledge of the process is directly obtained.246 tion, and identification of complex systems.
Besides model predictive control, other advanced control 5.2. Supervision and Optimization Strategies. As in many
strategies have been studied. One such strategy is fuzzy logic other industrial processes, the most interesting operating
Figure 6. Schemes of (a) two-layer approach, (b) one-layer approach, and (c) two-layer approach with two-stage MPC (adapted from Ying et al.261 and
Tatjewski262).
conditions for FCC units can change significantly in response to characterized by the simultaneous solution of the economical and
varying market demands for products, variations in feed quality control aspects of the supervision task in a single stage, adopting
and availability, and changes in equipment configurations, re- an optimal control perspective. Diverse RTO studies based on
quiring a timely adaption of the process to remain competitive. the one-layer methodology are also available.254,255,257260 A
Consequently, and for the reasons discussed in the previous hybrid implementation of the two-layer approach that is also
sections, FCC units are recurrently considered ideal candidates available on commercial MPC solutions considers the existence
for the application of advanced supervision and real time of an additional stage in the MPC layer, by including a steady-
optimization (RTO) techniques, which complement the remain- state controller in cascade with the MPC controller (Figure 6c).
ing control layers in a hierarchy of decision levels250 (Figure 5). This is introduced to facilitate the adaptation of the optimal set
Generally, the RTO term is associated with strategies that points determined to actual process conditions, avoiding large
explicitly address the online economic optimization of a process changes in the process variables, eliminating offsets, and guaran-
plant, using an economic objective function together with a teeing the feasibility of the specifications used by the MPC con-
process model and operational limits as constraints.251 troller.261 The extra function introduces also an additional
Two major approaches have been used to implement the RTO opportunity of including economic objectives in the MPC layer,
functionality; these are illustrated in Figure 6. The two-layer by considering either additional quadratic approximation terms or
approach, also termed layered optimization and control, corre- direct minimization of a nonlinear objective, and the use of extra
sponds currently to the most common implementation of manipulated variables in nonsquare configurations.246,255
advanced control and optimization (Figure 6a). It decomposes The major advantages of the two-layer approach are related to
the supervision problem into control (dynamic) and optimiza- the functional separation achieved, which simplifies the solution
tion (static) components, each implemented by a specialized of each optimization problem, therefore becoming less demand-
algorithm and process model. Various authors have presented RTO ing in terms of computational power and creating the possibility
studies of FCC units based on this approach176,194,228,252256 The of using specialized (robust) components. However, this also
alternative methodology integrates the control and optimization involves additional coordination between the decision layers;
problems into a single-level strategy (Figure 6b), requiring the because the optimization is based on steady-state models, it
solution of a dynamic optimization problem. This approach is also creates the potential for suboptimal operation and data
inconsistencies, when the process spends a large fraction of time The regulatory layer was based on the DMC algorithm, whereas
in a transient regimen. Using an integrated approach can provide the optimization layer solves a linear programming (LP) problem
additional economic benefits, with faster adaptation times to new to perform steady-state economic optimizations. These two func-
process conditions. On the other hand, this typically involves tions are executed sequentially and at the same frequency. Con-
higher computational effort and more severe consequences straints were also considered as penalties in the objective function.
associated with the nonconvergence or down-time of the opti- A nonlinear dynamic model is used as a reference in the design
mizers. Comparisons of these two methodologies are available in of the control algorithm. This model compared satisfactorily
various references.209,227,231,232,234,263 with the plant data for open-loop changes in the air flow and
The first successful applications of optimizing control to the regenerated catalyst valve opening. Therefore, for some of the
industrial FCC units using MPC technology were described by controlled variables, the model step responses could be used
Prett and Gillette.264 The optimization level was accomplished directly in the DMC, avoiding extensive identification testing in
using linear programming, by considering a linearization of a the plant.
steady-state mechanistic process model. The optimum operating Yang et al.194 proposed a two-level control structure for the
point was sent to the DMC controller as new reference values, control of a FCC reactorregenerator system. The lower level
using a two-layer approach. After this controller was able to move control is based on a DCS regulatory control system, with some
the unit to the new steady state, the previous procedure was of set points regulated by a second level. For the upper level a
repeated. Modifications were required in the basic DMC algo- predictive control algorithm based on a transfer function matrix
rithm to prevent the violation of absolute input constraints when model of the FCC unit was used. The model was obtained from
this combined strategy was implemented. real plant step test data. Recognizing that when input variables
McFarlane and Bacon252,265 developed an adaptive optimizing reach their limits, they might no longer be suitable for use as
control strategy for nonlinear, multivariable, constrained pro- manipulated variables, these authors developed explicit solutions
cesses and applied it to a model IV FCC unit (Figure 1b), where of the unconstrained predictive control problem for various
the optimization objective was the maximization of the FCC degrees of freedom, including the situation when more con-
profit, using a process model proposed by Kurihara.227 The trolled variables are available than manipulations. A priority
authors considered a procedure based on adaptive multivariable matrix is also established, defining selection priorities that
internal model control (MIMC), where the internal control law depend on the influence that each manipulated variable has on
was updated online on the basis of information provided by a each controlled variable. The control system will prefer the use of
dynamic model. The steady-state optimization relied on sec- controlled variables with highest priority. If one of these variables
tional (or sequential) linear programming (SLP). Both dynamic is already at one of its constraint limits, the controller will select
information for control and steady-state information for optimi- the next variable with higher priority for the given output and so
zation were obtained from the original process model. This on. Yang et al.194 added a third level to this control structure. The
corresponds to a two-layer approach, as the controller runs more third-level function aims to determine at regular times the
frequently than the optimizer, which feeds data to it. The optimum operation point (set points of the controlled variables)
controller performed satisfactorily both for set-point changes and send it to the multivariable predictive controller (two-layer
and for unit disturbances. approach).
Grosdidier et al.253 presented a paper on the implementation Lid and Strand257 described an implementation of a real-time
of advanced computer controls in an industrial FCC unit optimizer (RTO) in the control system of a residue fluid catalytic
operating at Neste Oy’s Porvoo refinery. This was done in a cracking (RFCC) unit at Statoil Mongstad refinery. The models
three-level hierarchy control system that included regulatory, used by the optimizer were regressed from a sequence of plant
constraint, and optimization modules. The lowest level of this experiments, along with data from ordinary operation. The
hierarchy consisted of six PI flow and temperature controllers optimization problem was solved using a commercial sequential
that regulate the FCC unit operation and are included in the DCS quadratic programming (SQP) routine, subject to process con-
of the plant. The set points of these DCS controllers are the straints and downstream processing capacity. The optimization
manipulated variables of the multivariable reactorregenerator system was implemented in a plant computer along with model-
controller. This second level of control was implemented with predictive control (MPC) applications, both interacting with the
constrained linear MPC. The MPC algorithm tries to keep all of real-time process database. On the basis of preliminary estimates
the manipulated variables at their ideal resting values (IRV), and optimization runs during commissioning, the project pay-
calculated by the third-level optimization control module. back time was two months.
Khandalekar and Riggs228 applied nonlinear process model- Using the constrained nonlinear PMBC controller pre-
based constraint control (PMBC) to a dynamic simulator of a viously developed by Khandelakar and Riggs,228 Ellis et al.266
model IV FCC unit. The nonlinear model used by the controller also considered the optimization of a model IV unit. The main
was composed of approximated macroscopic models of the differences of this work included a 10-lump kinetic model in
reactor and regenerator, based on the model proposed by Lee the nonlinear model164 and the use of SQP software for the
and Groves.225 These models were parametrized to facilitate solution of the constrained steady-state optimization problem
their regular online update. The nonlinear PMBC was based on (two-layer approach).
generic model control (GMC), and its performance was com- Gouvêa and Odloak258 compared the performance of the two-
pared to conventional PI control. An online steady-state optimi- layer and the one-layer approaches for the LPG production
zer was placed on top of the PMBC, using an unconstrained maximization in the FCC converter. The authors concluded that
optimization algorithm. Process constraints were included as the one-layer approach was able to assimilate changes in the
penalties in the objective function. economic objectives much more quickly than the two-layer
Moro and Odloak 176 proposed a two-step control struc- approach and provided a smoother and more stable dynamic
ture that includes both regulatory and optimization functions. response. They also observed that the one-layer approach could
Table 3. Summary of RTO Studies Based on One-Layer and Two-Layer Approaches

author year approach description
Prett and Gillette264 1989 two-layer steady-state (SS) model; DMC controller; SLP optimization
McFarlane and Bacon252 1989 two-layer dynamic model; adaptative MIMC; SLP optimization
Grosdidier253 1993 two-layer SS model; constrained linear MPC
Khandalekar and Riggs228 1995 two-layer dynamic model; PMBC based on GMC; SS optimizer;
unconstrained optimization algorithm
Moro and Odloak176 1995 two-layer nonlinear dynamic model; DMC controller; LP based on DMC; SS optimization
Abou-Jeyab and Gupta268 1996 one-layer linear step model based; SMPC; LP problem
Yang et al.194,256 1996, 1998 one-layer nodel from real plant step test; transfer function matrix; MPC
Lid and Strand257 1997 one-layer industrial application at Statoil Montgomery Plant; model
from unit operational data and step tests; SQP optimizer; MPC
Ellis et al.266 1998 two-layer PMBC controller based on GMC; SQP; constrained SS optimization
Loeblein and Perkins267 1999 one-layer MPC integrated with inline optimization; unconstrained MPC based on
a linear-time invariant state space model
Zanin254 2001 one-layer industrial application at Petrobras; constrained optimization; SQP solver
Ansari and Tade242 2000 one-layer RTO/MPC, NLP problem; generic model control (GMC), SQP solver
Zanin et al.250 2005 two-layer proprietary model SimCraqOTL; proprietary advanced control SICON;
NLP optimization problem; SQP solver
Souza et al.260 2010 one-layer SRTO/linear MPC, QP problem
become more sensitive to plant/model mismatch errors, because The constrained optimization problems were solved using SQP
these are reflected not only in the set points to where the process and a process model based on the work of Moro and Odloak.176
is driven but also on the form of the overall closed-loop response. The main goal was to maximize the LPG production in the FCC
On the other hand, the discontinuous application of the optimi- unit. The authors showed that the economic returns achieved with
zation level in the two-layer approach was found to produce the improved two-layer approach match reasonably well the results
strong control actions in the first instants before plant stabiliza- possible with the integrated one-layer methodology, providing a
tion, suggesting the introduction of a filter to attenuate abrupt significant improvement to the performance of the original two-
set-point changes. layer supervision scheme.
Loeblein and Perkins259,267 described an integrated model A similar methodology to integrate the results of the RTO
predictive regulatory control and online optimization analysis of layer in the control structure is described by Souza et al.260
a FCC reactorregenerator system. The MPC algorithm was Instead of calculating the nonlinear cost function, the authors
based on a linear-time invariant state space model, considering used its reduced gradient; this leads to the solution of a QP
both the conventional and the riserregenerator control struc- problem instead of a NLP problem, very similar to the scope of a
tures. An evaluation of the economic performance of each conventional MPC controller. An iterative algorithm is described
control structure was developed, both alone and as part of a to guess the correct active set in the calculation of the gradient of
combined real-time optimization strategy. The unconstrained the economic objective function. The results are comparable
MPC control law was used to fully characterize the variances of with the ones obtained by Zanin254 regarding the economic
the constrained variables and, consequently, the necessary op- performance and stability of the combined methodology.
erational distance from the existing hard constraints. Although An alternative solution to integrate the economic optimization
the conventional control structure showed better dynamic is described by Zanin et al.250 in the context of the implementa-
economic performance at the regulatory control level than the tion of RTO in a different commercial FCC unit. Five degrees of
riserregenerator control structure, the economic performance freedom are considered at the RTO level, consisting of three
of the integrated optimization and regulatory control system was manipulated variables (feed temperature, suction pressure of the
better for the riserregenerator control structure. These results wet gas compressor, and pressure difference between the reactor
indicate a clear advantage in the simultaneous design of the and regenerator) and two controlled variables (riser outlet and
online optimization and regulatory control systems, considering regenerator second-stage dense phase temperatures). A two-level
the economic performances of both levels in an integrated MPC controller is used to interface these decisions with the
framework.259 linear MPC controller (Figure 6c), using a quadratic program
Zanin254 and Zanin et al.255 carried out a comparative study that penalizes differences between the actual and desired refer-
between the behaviors of the two-layer and one-layer approaches ence input profiles and the sizes of the predicted input moves
and proposed an approach to facilitate the integration of steady- together with a quadratic penalty for violation of the output
state optimization results in the MPC layer. This involved the constraints. An increase of approximately 6% in the unit profit
addition of various terms to the objective function, which was registered during the commissioning period of the integrated
included the calculation of an economic cost function using optimization system.
the final predicted values of the input and output variables at the All of these studies (Table 3) reinforce the conclusion that the
end of their horizons (assumed at steady state). This scope is success of advanced supervision strategies depends critically on
similar to the one previously reported on the basis of a Petrobras the following factors: (i) good integration between the various
industrial unit, with more manipulated and controlled variables.258 supervision layers, minimizing potential inconsistencies between
process information available at the various decision levels; balance, improving the use of the regenerator heat will therefore
(ii) effective usage of process information for adaptation of the serve to increase the unit’s energy efficiency.
models to the actual operating conditions; and (iii) detailed Increasing productivity, improving energy efficiency, and
process models, able to capture the major nonlinearities and the minimizing operating costs are also paramount to the profit-
interactions among the model variables. ability of FCC units. An advanced control application could
reduce the standard deviation of controlled variables by at least
50%. This allows the controller to keep the controlled variables
6. FINAL REMARKS closer to the constraints, without jeopardizing on-spec produc-
tion. A larger implementation of advanced process control,
Historically, the fluid catalytic cracking unit has expanded its modeling, and simulation techniques will help refiners to run
role from a gasoline machine to a LPG and propene producer and the FCC at optimized operating conditions, limiting its operational
a residue upgrader. FCC technology advances played an essential costs. These technologies can also help to improve economics by
part in refinery profit. Recent FCC challenges are related to optimizing maintenance activities and monitoring catalyst activity.
environmental regulations and the drive to produce cleaner fuels. As illustrated by the multiple applications considered, accurate
It also seems that in the coming decades the FCC will take on two modeling of FCC units remains a key component for the success of
additional roles: it may (partially) use feeds from renewable applied optimization strategies. The best process models for this
sources and be combined with a CO2 capture process to alleviate purpose must have a high level of detail (to capture nonlinear
growing concerns over energy consumption and global warming. behavior and variable interaction) and must be highly adaptive (to
With combined efforts by refiners and technology developers, translate accurately the state of the real process). Control and
the process will continue to demonstrate superior adaptability to online optimization using nonlinear lumped kinetic models con-
the changing market requirements. sidering deactivation as a function of coke content and nature (or
The importance of catalyst formulations (e.g., additives, new location) will play an even more significant role in the future. Given
or modified zeolites, functional ingredients) will continue to be a the increasing robustness and performance of nonlinear optimiza-
major aspect. Process configurations and hardware improve- tion codes, it is natural to expect that the powerful one-layer control
ments such as feed injectors, riser and stripper internals, and and RTO approach will be found more often in the future as the
termination devices will continue to be developed. These mod- preferred implementation of the hierarchical chain of control and
ifications, combined with advanced control techniques, are RTO in FCC units.
essential to sustain both the yield and the quality of FCC gasoline To benefit from these developments, great use of advanced
with the future feedstock scenario. Refineries equipped to process process simulation and optimization environments is required, to
heavy crudes have, so far, reported better refining margins because assist in the life cycle of the process models, in order to effectively
they can take advantage of less expensive opportunity crudes. counter the limited availability of specialized manpower, a factor
Residue fluid catalytic cracking is an important component in the that often constitutes a major limitation in the deployment
upgrading of such crudes, with unit profitability depending upon of model-based strategies for process improvement. Commit-
the extent to which heavy hydrocarbons in the feed are cracked into ment from management and a strong execution plan are needed
valuable products. Technology advances in catalysts, hardware, to capture the benefits and to increase the number of successful
and monitoring will improve residue FCC operations. Although projects. Competent engineers will be required to execute
the coprocessing of renewable feeds with FCC gas oils can be and maintain RTO applications, to improve model accuracy
performed in the existent FCC infrastructures with minimal through technology, and to improve the quality of online data.
changes, the use of renewable feeds alone may require significant The main benefits that can be obtained through the implemen-
modifications in the FCC unit. tation of all these integrated advanced technologies and strate-
Feed characterization is a key area for improvement. Predict- gies are improvement in product quality, energy savings,
ing unit dynamics caused by feed changes is important in improved operational stability, increased yields of the most
optimizing unit operation, especially for atmospheric distillation valuable products, and improved protection of the process from
residues. Product recycles and multiple reaction sections seem to violating operational constraints. All of these potential benefits
be the most prevalent technology trends. Improving feed in- have not yet been captured by the current typical hierarchy of
jectors, riser termination and catalyst separation devices, strip- applications.
pers, and regenerator components are good revamp options
for existing units. As feeds get heavier, the trend toward a higher ’ AUTHOR INFORMATION
stripper residence time and, thus, increased mass transfer between
entrained hydrocarbons and steam will continue. Moreover, the Corresponding Author
role of the regenerator is expected to evolve because of CO2 *Tel: +351 218 417 887. Fax.: +351 218 419 198. E-mail: carla.
emission reduction requirements. pinheiro@ist.utl.pt.
Several of today’s FCC units were built in the 1970s, when
efficient energy use was not of primary concern. These units have ’ ACKNOWLEDGMENT
been revamped to improve energy performance. The coke yield
L.D., A.J.S.C., and I.G. thank the Fundac-~ao para a Ciência e
of a FCC unit is dictated by the energy needs of the unit. The heat
Tecnologia (FCT) for financial support (ref SFRH/BD/60668/
of reaction from catalyst coke combusted in the regenerator is
2009, ref PTDC/EQU-ESI/73458/2006, and ref SFRH/BPD/
used to heat and vaporize fresh and recycled feed, heat atomizing
74457/2010, respectively).
steam, provide the heat required for endothermic cracking
reactions, heat the combustion air to regenerator flue gas
temperature, make up for heat lost to the surroundings, and ’ DEDICATION
4
produce stripping steam. Because the FCC unit operates in heat In memory of Prof. Fernando Ramôa Ribeiro, recently deceased.
’ ABBREVIATIONS VGO vacuum gas oil

ADMC adaptative dynamic multivariable controllers VR vacuum residue
ANN artificial neural networks W catalyst inventory (level)
API American Petroleum Institute Y controlled variable matrix
ASTM American Society for Testing and Materials ZN ZieglerNichols
CFD computational fluid dynamics
CPERI Chemical Process Engineering Research Institute Superscripts and Subscripts
CST continuous stirred tank air air to regenerator
CSTR continuous stirred tank reactor c catalyst
CTO catalyst-to-oil CY cyclones
DCS distributed control system Feed feed stream
DMC dynamic matrix control MF main fractionators (top)
DO decanted oil PF preheating furnace fuel
e-cat equilibrium catalyst r stripper (historically reactor)
EKF extended Kalman filter RC regenerated catalyst
F flow rate RG regenerator
FCC fluid catalytic cracking RT reactor section (riser + disengager + stripper)
FCCU fluid catalytic cracking unit S stack gas
FLC fuzzy logic control SC spent catalyst
GMC generic model control SS steady state
HCO heavy cycle oil V regenerator air nypass
HEKF heuristic extended Kalman filter wt weighting matrix
IFP Institut Franc-ais du Petrole
IFPEN Institut Franc-ais du Petrole Energies Nouvelles ’ REFERENCES
IMC internal model control
IRV ideal resting values (1) Fatih, B. Challenges Facing the Global Refining Industry. Oil Gas
Process. Rev. 2005, 9.
ISE integral of the square error
(2) Shah, N. K.; Li, Z.; Ierapetritou, M. G. Petroleum Refining
LCO light cycle oil Operations: Key Issues, Advances, and Opportunities. Ind. Eng. Chem.
LMPC linear model predictive control Res. 2010, 50, 1161.
LP linear programming (3) Sadeghbeigi, R. Fluid Catalytic Cracking Handbook: Design,
LPG liquefied petroleum gases Operation, and Troubleshooting of FCC Facilities, 2nd ed.; Gulf Profes-
MIMC multivariable internal model control sional Publishing: Houston, TX, 2000.
MIMO multi-inputmulti-output (4) Fernandez, M. L.; Lacalle, A.; Bilbao, J.; Arandes, J. M. Recycling
MPC model predictive control Hydrocarbon Cuts into FCC Units. Energy Fuels 2002, 16, 615.
N4SID numerical algorithms for subspace state space system (5) Gilbert, W. R.; Baptista, C. A.; Pinho, A. R. Fluid Catalytic
identification Cracking VII Materials, Methods and Process Innovations; Elsevier:
Amsterdam, The Netherlands, 2007; Vol. 166.
NMPC nonlinear model predictive control
(6) Corma, A.; Melo, F. V.; Sauvanaud, L.; Ortega, F. J. Different
NLP nonlinear programming Process Schemes for Converting Light Straight Run and Fluid Catalytic
PCA principal component analysis Cracking Naphthas in a FCC Unit for Maximum Propylene Production.
PFR plug flow reactor Appl. Catal. A: Gen. 2004, 265, 195.
PI proportiona-integral (7) Corma, A.; Huber, G. W.; Sauvanaud, L.; O'Connor, P. Proces-
PID proportionalintegralderivative sing Biomass-Derived Oxygenates in the Oil Refinery: Catalytic Crack-
PMBC process model-based constraint control ing (FCC) Reaction Pathways and Role of Catalyst. J. Catal. 2007,
PRBS pseudo-random binary sequences 247, 307.
QP quadratic programming (8) Domine, M. E.; Veen, A. C. van; Shuurman, Y.; Mirodatos, C.
RFCC residue fluid catalytic cracking Coprocessing of Oxygenated Biomass Compounds and Hydrocarbons
for the Production of Sustainable Fuel. ChemSusChem 2008, 1, 179.
RGA relative gain array
(9) Grac-a, I.; Ribeiro, F. R.; Cerqueira, H. S.; Lam, Y. L.; Almeida, M.
RHP right half-plane B. B. de. Catalytic Cracking of Mixtures of Model Bio-Oil Compounds
RTO real-time optimization and Gasoil. Appl. Catal. B: Environ. 2009, 90, 556.
RVP Reid vapor pressure (10) Hemler, C. L.; Smith, L. UOP Fluid Catalytic Cracking Process.
SCP sequential linear programming In Handbook of Petroleum Refining Processes; Meyers, R. A., Ed.; McGraw-
SEMK single-event microkinetic Hill Professional Publishing: New York, 2004; p 848.
SISO single-inputsingle-output (11) Montgomery, J. The Grace Davison Guide to Fluid Catalytic
SLP sequential linear programming Cracking Part Two; W. R. Grace & Co.: Baltimore, MD, 1996.
SMPC simplified model predictive control (12) Biswas, J.; Maxwell, I. E. Recent Process- and Catalyst-Related
SOD Standard Oil Development Co. Developments in Fluid Catalytic Cracking. Appl. Catal. 1990, 63, 197.
(13) Wallenstein, D.; Harding, R. H. The Dependence of ZSM-5
SOL structure-oriented lumping
Additive Performance on the Hydrogen-Transfer Activity of the REUSY
SQP sequential quadratic programming Base Catalyst in Fluid Catalytic Cracking. Appl. Catal. A: Gen. 2001,
SRTO simplified real-time optimization 214, 11.
SSM steady state models (14) Cheng, W.-C.; Kim, G.; Peters, A. W.; Zhao, X.; Rajagopalan,
T temperature K.; Ziebarth, M. S.; Pereira, C. J. Environmental Fluid Catalytic Cracking
U manipulated variable matrix Technology. Catal. Rev. 1998, 40, 39.

(15) Carvalho, M. C. N. A. De; Morgado, E.; Cerqueira, H. S.; (38) Moro, L. F. L. Process Technology in the Petroleum Refining
Resende, N. S. De; Schmal, M. Behavior of Fresh and Deactivated Industry Current Situation and Future Trends. Comput. Chem. Eng.
Combustion Promoter Additives. Ind. Eng. Chem. Res. 2004, 43, 3133. 2003, 27, 1303.
(16) Degnan, T. F.; Chitnis, G. K.; Schipper, P. H. History of ZSM-5 (39) Bartholomew, C. H.; Farrauto, R. J. Fundamentals of Industrial
Fluid Catalytic Cracking Additive Development at Mobil. Microporous Catalytic Processes, 2nd ed.; Wiley-AIChE: Hoboken, NJ, 2005.
Mesoporous Mater. 2000, 3536, 245. (40) Corma, A.; Orchilles, A. Current Views on the Mechanism of
(17) Blasco, T.; Corma, A.; Martínez-Triguero, J. Hydrothermal Catalytic Cracking. Microporous Mesoporous Mater. 2000, 3536, 21.
Stabilization of ZSM-5 Catalytic-Cracking Additives by Phosphorus (41) Sedran, U. A. Laboratory Testing of FCC Catalysts and
Addition. J. Catal. 2006, 237, 267. Hydrogen Transfer Properties Evaluation. Catal. Rev. 1994, 36, 405.
(18) Caeiro, G.; Magnoux, P.; Lopes, J. M.; Ribeiro, F. R.; Menezes, (42) Guisnet, M.; Magnoux, P. Organic Chemistry of Coke Forma-
S. M. C.; Costa, A. F.; Cerqueira, H. S. Stabilization Effect of Phosphorus tion. Appl. Catal. A: Gen. 2001, 212, 83.
on Steamed H-MFI Zeolites. Appl. Catal. A: Gen. 2006, 314, 160. (43) Weekman, V. W. Model of Catalytic Cracking Conversion in
(19) Stockwell, D. M.; Kelkar, C. P. Reduction of NOx Emissions Fixed, Moving, and Fluid-Bed Reactors. Ind. Eng. Chem. Process Des. Dev.
from FCCU Regenerators with Additives. Stud. Surf. Sci. Catal. 2004, 1968, 7, 90.
149, 177. (44) Weekman, V. W.; Nace, D. M. Kinetics of Catalytic Cracking
(20) Roncolatto, R. E.; Cardoso, M. J. B.; Lam, Y. L.; Schmal, M. Selectivity in Fixed, Moving, and Fluid Bed Reactors. AIChE J. 1970,
FCC SOx Additives Deactivation. Ind. Eng. Chem. Res. 2006, 45, 2646. 16, 397.
(21) Habib Jr., E. T.; Zhao, X.; Yaluris, G.; Cheng, W. C.; Boock, (45) Nace, D. M.; Voltz, S. E.; Weekman, V. W. Application of a
L. T.; Gilson, J.-P. Advances in Fluid Catalytic Cracking. In Zeolites for Kinetic Model for Catalytic Cracking. Effects of Charge Stocks. Ind. Eng.
Cleaner Technologies, Vol. 3 of Catalytic Science Series; Guisnet, M., Gilson, Chem. Process Des. Dev. 1971, 10, 530.
J.-P., Eds.; Imperial College Press: London, U.K., 2002; pp 105130. (46) Lee, L.-S.; Chen, Y.-W.; Huang, T.-N.; Pan, W.-Y. Four-Lump
(22) Cerqueira, H. S.; Caeiro, G.; Costa, L.; Ribeiro, F. R. Deactiva- Kinetic Model for Fluid Catalytic Cracking Process. Can. J. Chem. Eng.
tion of FCC Catalysts. J. Mol. Catal A: Chem. 2008, 292, 1. 1989, 67, 615.
(23) Kouwenhoven, H. W.; Kroes, N. de. Preparation of Zeolite (47) Bolkan-Kenny, Y. G.; Pugsley, T. S.; Berruti, F. Computer
Catalysts. Stud. Surf. Sci. Catal. 2001, 137, 673. Simulation of the Performance of Fluid Catalytic Cracking Risers and
(24) Forzatti, P.; Lietti, L. Catalyst Deactivation. Catal. Today 1999, Downers. Ind. Eng. Chem. Res. 1994, 33, 3043.
52, 165. (48) Ali, H.; Rohani, S.; Corriou, J. P. Modeling and Control of a
(25) Lappas, A. A.; Nalbandian, L.; Iatridis, D. K.; Voutetakis, S. S.; Riser-Type Fluid Catalytic Cracking (FCC) Unit. Trans. Inst. Chem. Eng.
Vasalos, I. A. Effect of Metals Poisoning on FCC Products Yields: 1997, 75, 401.
Studies in an FCC Short Contact Time Pilot Plant Unit. Catal. Today (49) Ali, H.; Rohani, S. Dynamic Modeling and Simulation of a
2001, 65, 233. Riser-Type Fluid Catalytic Cracking Unit. Chem. Eng. Technol. 1997,
(26) Pinto, F. V.; Escobar, A. S.; Oliveira, B. G. de; Lam, Y. L.; 20, 118.
Cerqueira, H. S.; Louis, B.; Tessonnier, J. P.; Su, D. S.; Pereira, M. M. (50) Malay, P.; Rohani, S.; Milne, B. J. The Modified Dynamic
The Effect of Alumina on FCC Catalyst in the Presence of Nickel and Model of a Riser Type Fluid Catalytic Cracking Unit. Can. J. Chem. Eng.
Vanadium. Appl. Catal. A: Gen. 2010, 388, 15. 1999, 77, 169.
(27) Occelli, M. L. Vanadium Resistant Fluid Cracking Catalysts. (51) Han, I.-S.; Chung, C.-B. Dynamic Modeling and Simulation of a
Stud. Surf. Sci. Catal. 1996, 100, 27. Fluidized Catalytic Cracking Process. Part I: Process Modeling. Chem.
(28) Murcia, A. A. Numerous Changes Mark FCC Technology Eng. Sci. 2001, 56, 1951.
Advance. Oil Gas J. 1992, 90, 68. (52) Gupta, A.; Rao, D. S. Model for the Performance of a Fluid
(29) Murphy, J. R. Evolutionary Design Changes Mark FCC Pro- Catalytic Cracking (FCC) Riser Reactor: Effect of Feed Atomization.
cess. Oil Gas J. 1992, 90, 49. Chem. Eng. Sci. 2001, 56, 4489.
(30) Fletcher, R. P. The History of Fluidized Catalytic Cracking: (53) Deng, R.; Wei, F.; Liu, T.; Jin, Y. Radial Behavior in Riser and
A History of Innovation: 19422008. In Innovations in Industrial and Downer during the FCC Process. Chem. Eng. Process. 2002, 41, 259.
Engineering Chemistry: A Century of Achievements and Prospects for the (54) Nayak, S. V.; Joshi, S. L.; Ranade, V. V. Modeling of Vaporiza-
New Millennium. Vol. 1000 of an American Chemical Society Publication; tion and Cracking of Liquid Oil Injected in a Gas-Solid Riser. Chem. Eng.
Flank, W. H., Abraham, M. A., Matthews, M. A., Eds.; Oxford University Sci. 2005, 60, 6049.
Press: Oxford, U.K., 2009; pp 189249. (55) Takatsuka, T.; Sato, S.; Morimoto, Y.; Hashimoto, H. A
(31) Avidan, A. A. Origin, Development and Scope of FCC Cata- Reaction Model for Fluidized-Bed Catalytic Cracking of Residual Oil.
lysis. In Fluid Catalytic Cracking: Science and Technology. Vol. 76 of Studies Int. Chem. Eng. 1987, 27, 107.
in Surface Science and Catalysis; Magee, J. S., Mitchell, M. M., Jr., Ed.; (56) Oliveira, L. L.; Biscaia, E. C. Catalytic Cracking Kinetic Models.
Elsevier: Amsterdam, The Netherlands, 1993; pp 139. Parameter Estimation and Model Evaluation. Ind. Eng. Chem. Res. 1989,
(32) Montgomery, J. The Grace Davison Guide to Fluid Catalytic 28, 264.
Cracking Part One; W. R. Grace & Co.: Baltimore, MD, 1993. (57) Liguras, D. K.; Allen, D. T. Structural Models for Catalytic
(33) Otterstedt, J. E.; Gevert, S. B.; JA € Ås, S. G.; Menon, P. G. Fluid Cracking. 1. Model Compound Reactions. Ind. Eng. Chem. Res. 1989,
Catalytic Cracking of Heavy (Residual) Oil Fractions: a Review. Appl. 28, 665.
Catal. 1986, 22, 159. (58) Corella, J.; Frances, E.; Bartholomew, C. H.; Butt, J. B. On the
(34) O’Connor, P. Verlaan, J. P. J.; Yanik, S. J. Challenges, Catalyst Kinetic Equation of Deactivation of Commercial Cracking (FCC)
Technology and Catalytic Solutions in Resid FCC. Catal. Today 1998, 43, 305. Catalysts with Commercial Feedstocks. Stud. Surf. Sci. Catal. 1991,
(35) Leite, L. F.; Fusco, J. M.; Ramos, J. G.; Medeiros, J.; Torem, M. A. 68, 375.
Novel Fluid Catalytic Cracking Technology Excellence in Heavy Feedstock (59) Pitault, I.; Nevicato, D.; Forissier, M.; Bernard, J.-R. Kinetic
Processing. In Proceedings of the 17th World Petroleum Congress, Rio de Model Based on a Molecular Description for Catalytic Cracking of
Janeiro, Brazil, 2002; World Petroleum Congress: London, England. Vacuum Gas Oil. Chem. Eng. Sci. 1994, 49, 4249.
(36) Wang, G.; Xu, C.; Gao, J. Study of Cracking FCC Naphtha in a (60) Cerqueira, H. S.; Biscaia, E. C., Jr.; Falabella, E.; Aguiar, S.;
Secondary Riser of the FCC Unit for Maximum Propylene Production. Bartholomew, C. H.; Fuentes, G. A. Mathematical Modeling of Deac-
Fuel Process. Technol. 2008, 89, 864. tivation by Coke Formation in the Cracking of Gasoil. Stud. Surf. Sci.
(37) Li, C.; Yang, C.; Shan, H. Maximizing Propylene Yield by Two- Catal. 1997, 111, 303.
Stage Riser Catalytic Cracking of Heavy Oil. Ind. Eng. Chem. Res. 2007, (61) Sugungun, M. Kinetic Modeling of FCC Process. Catal. Today
46, 4914. 1998, 43, 315.

(62) Ancheyta-Juarez, J.; Lopez-Isunza, F.; Aguilar-Rodríguez, E. (85) Landeghem, F. Van; Nevicato, D.; Pitault, I.; Forissier, M.;
5-Lump Kinetic Model for Gas Oil Catalytic Cracking. Appl. Catal. A: Turlier, P.; Derouin, C.; Bernard, J.-R. Fluid Catalytic Cracking:
Gen. 1999, 177, 227. Modelling of an Industrial Riser. Appl. Catal. A: Gen. 1996, 138, 381.
(63) Hagelberg, P.; Eilos, I.; Hiltunen, J.; Lipi€ainen, K.; Niemi, V. M.; (86) Delattre, C.; Forissier, M.; Pitault, I.; Schweich, D.; Bernard, J.-R.
Aittamaa, J.; Krause, A. O. I. Kinetics of Catalytic Cracking with Short Improvement of the Microactivity Test for Kinetic and Deactivation
Contact Times. Appl. Catal. A: Gen. 2002, 223, 73. Studies Involved in Catalytic Cracking. Chem. Eng. Sci. 2001, 56, 1337.
(64) Dupain, X.; Gamas, E. D.; Madon, R.; Kelkar, C. P.; Makkee, (87) Rawet, R.; Cerqueira, H. S.; Pinto, J. C. The Influence of
M.; Moulijn, J. A. Aromatic Gas Oil Cracking under Realistic FCC Covariances during Laboratory Evaluation of FCC Catalysts. Appl.
Conditions in a Microriser Reactor. Fuel 2003, 82, 1559. Catal. A: Gen. 2001, 207, 199.
(65) Corma, A.; Melo, F. V.; Sauvanaud, L. Kinetic and Decay (88) Voorhies, A. Carbon Formation in Catalytic Cracking. Ind. Eng.
Cracking Model for a MicroDowner Unit. Appl. Catal. A: Gen. 2005, Chem. 1945, 37, 318.
287, 34. (89) Nam, I. S.; Kittrell, J. R. Use of Catalyst Coke Content in
(66) Gupta, A.; Rao, D. S. Effect of Feed Atomization on FCC Deactivation Modeling. Ind. Eng. Chem. Process Des. Dev. 1984, 23, 237.
Performance: Simulation of Entire Unit. Chem. Eng. Sci. 2003, 58, 4567. (90) Corella, J.; Monzon, A. Modeling of the Deactivation Kinetics
(67) Quann, R. J.; Jaffe, S. B. Structure-Oriented Lumping: Describ- of Solid Catalysts by Two or More Simultaneous and Different Causes.
ing the Chemistry of Complex Hydrocarbon Mixtures. Ind. Eng. Chem. Ind. Eng. Chem. Res. 1988, 27, 369.
Res. 1992, 31, 2483. (91) Froment, G. F. Bischoff, K. B.; Wilde, J. D. Chemical Reactor
(68) Quann, R. J.; Jaffe, S. B. Building Useful Models of Complex Analysis and Design, 3rd ed.; Wiley: Hoboken, NJ, 2010.
Reaction Systems in Petroleum Refining. Chem. Eng. Sci. 1996, 51, 1615. (92) Corella, J.; Garciadopico, M.; Frances, E. An Advanced Model
(69) Christensen, G.; Apelian, M. R.; Hickey, K. J.; Jaffe, S. B. Future for a whole FCCU Based on the Kinetics and on Basic Principles. Part I:
Directions in Modeling the FCC Process: An Emphasis on Product The Model in Detail. In Recent Advances in FCC Technology Symposium,
Quality. Chem. Eng. Sci. 1999, 54, 2753. Atlanta, GA, 1994; AIChE Spring Meeting, April 1721, 1994. AIChE: USA.
(70) Feng, W.; Vynckier, E.; Froment, G. F. Single Event Kinetics of (93) Jimenez-Garcia, G.; Quintana-Sol orzano, R.; Aguilar-Lopez, R.;
Catalytic Cracking. Ind. Eng. Chem. Res. 1993, 32, 2997. Maya-Yescas, R. Modelling Catalyst Deactivation by External Coke
(71) Froment, G. F. Modeling of Catalyst Deactivation. Appl. Catal. Deposition during Fluidized-bed Catalytic Cracking. Int. J. Chem.
A: Gen. 2001, 212, 117. Reaction Eng. 2010, 8, note S2.
(72) Moustafa, T. M.; Froment, G. F. Kinetic Modeling of Coke (94) Corella, J.; Bilbao, R.; Molina, J. A.; Artigas, A. Variation with
Formation and Deactivation in the Catalytic Cracking of Vacuum Gas Time of the Mechanism, Observable Order, and Activation Energy of
Oil. Ind. Eng. Chem. Res. 2003, 42, 14. Catalyst Deactivation by Coke in the FCC Process. Ind. Eng. Chem.
(73) Quintana-Solorzano, R.; Thybaut, J. W.; Galtier, P.; Marin, Process Des. Dev. 1985, 24, 625.
G. B. Simulation of an Industrial Riser for Catalytic Cracking in the (95) Gianetto, A.; Farag, H. I.; Blasetti, A. P.; Lasa, H. I. de. Fluid
Presence of Coking using Single-Event MicroKinetics. Catal. Today Catalytic Cracking Catalyst for Reformulated Gasolines. Kinetic Model-
2010, 150, 319. ing. Ind. Eng. Chem. Res. 1994, 33, 3053.
(74) Pinheiro, C. I. C.; Lemos, F.; Ramôa Ribeiro, F. Dynamic (96) Larocca, M.; Lasa, H. I.; De; Farag, H. I.; Ng, S. Cracking
Modelling and Network Simulation of n-Heptane Catalytic Cracking: Catalysts Deactivation by Nickel and Vanadium Contaminants. Ind. Eng.
Influence of Kinetic Parameters. Chem. Eng. Sci. 1999, 54, 1735. Chem. Res. 1990, 29, 2181.
(75) Carabineiro, H.; Pinheiro, C. I. C.; Lemos, F.; Ramôa Ribeiro, (97) Corella, J.; Frances, E. Analysis of the Riser Reactor of a Fluid
F. Transient Microkinetic Modelling of n-Heptane Catalytic Cracking Catalytic Cracking Unit. In Fluid Catalytic Cracking II: Concepts in
over H-USY Zeolite. Chem. Eng. Sci. 2004, 59, 1221. Catalyst Design; Occelli, M. L., Ed.; ACS Symposium Series 452;
(76) Hudebine, D.; Verstraete, J. J. Reconstruction of Petroleum American Chemical Society: Washington, DC, 1991; Vol. 452, 165.
Feedstocks by Entropy Maximization. Application to FCC Gasolines. (98) Corella, J. On the Modeling of the Kinetics of the Selective
Oil Gas Sci. Technol. 2011, 66, 437. Deactivation of Catalysts. Application to the Fluidized Catalytic Crack-
(77) Verstraete, J. J.; Revellin, N.; Dulot, H.; Hudebine, D. Molec- ing Process. Ind. Eng. Chem. Res. 2004, 43, 4080.
ular Reconstruction of Vacuum Gasoils. Prepr. Pap.Am. Chem. Soc., (99) Bollas, G. M.; Lappas, A. A.; Iatridis, D. K.; Vasalos, I. A. Five-
Div. Fuel Chem. 2004, 49, 20. Lump Kinetic Model with Selective Catalyst Deactivation for the
(78) Hudebine, D.; Verstraete, J. J. Molecular Reconstruction of Prediction of the Product Selectivity in the Fluid Catalytic Cracking
LCO Gasoils from Overall Petroleum Analyses. Chem. Eng. Sci. 2004, Process. Catal. Today 2007, 127, 31.
59, 4755. (100) Beirnaert, H. C.; Alleman, J. R.; Marin, G. B. A Fundamental
(79) Verstraete, J. J.; Schnongs, P.; Dulot, H.; Hudebine, D. Molec- Kinetic Model for the Catalytic Cracking of Alkanes on a USY Zeolite in
ular Reconstruction of Heavy Petroleum Residue Fractions. Chem. Eng. the Presence of Coke Formation. Ind. Eng. Chem. Res. 2001, 40, 1337.
Sci. 2010, 65, 304. (101) Quintana-Solorzano, R.; Thybaut, J. W.; Marin, G. B.; Lødeng,
(80) Surla, K.; Guillaume, D.; Verstraete, J. J.; Galtier, P. Kinetic R.; Holmen, A. Single-Event Microkinetics for Coke Formation in
Modeling using the Single-Event Methodology: Application to the Catalytic Cracking. Catal. Today 2005, 107108, 619.
Isomerization of Light Paraffins. Oil Gas Sci. Technol. 2011, 66, 343. (102) Quintana-Solorzano, R.; Thybaut, J. W.; Galtier, P.; Marin,
(81) Lee, C. K.; Ashtekar, S.; Gladden, L. F.; Barrie, P. J. Adsorption G. B. Single-Event Microkinetics for Coke Formation during the
and Desorption Kinetics of Hydrocarbons in FCC Catalysts Studied Catalytic Cracking of (Cyclo)Alkane/1-Octene Mixtures. Catal. Today
using a Tapered Element Oscillating Microbalance (TEOM). Part 1: 2007, 127, 17.
Experimental Measurements. Chem. Eng. Sci. 2004, 59, 1131. (103) Quintana-Solorzano, R.; Thybaut, J. W.; Galtier, P.; Marin,
(82) Barrie, P. J.; Lee, C. K.; Gladden, L. F. Adsorption and G. B. Simulation of an Industrial Riser for Catalytic Cracking in the
Desorption Kinetics of Hydrocarbons in FCC Catalysts Studied using Presence of Coking using Single-Event MicroKinetics. Catal. Today
a Tapered Element Oscillating Microbalance (TEOM). Part 2: Numer- 2010, 150, 319.
ical Simulations. Chem. Eng. Sci. 2004, 59, 1139. (104) Jimenez-Garcia, G.; Quintana-Solorzano, R.; Maya-Yescas, R.
(83) Puente, G. De la; A vila, A.; Chiovetta, G.; Martignoni, Improving Accuracy in the Estimation of Kinetic Frequency Factors
W. P.; Cerqueira, H. S.; Sedran, U. A. Adsorption of Hydrocarbons on from Laboratory Data To Model Industrial Catalytic Cracking Risers.
FCC Catalysts under Reaction Conditions. Ind. Eng. Chem. Res. 2005, Ind. Eng. Chem. Res. 2011, 2736.
44, 3879. (105) Vieira, R. C.; Pinto, J. C.; Biscaia, E. C.; Baptista, C. M. L. A.;
(84) Guisnet, M.; Ribeiro, F. R. Deactivation and Regeneration of Cerqueira, H. S. Simulation of Catalytic Cracking in a Fixed-Fluidized-
Zeolite Catalysts; Imperial College Press: London, U.K., 2011. Bed Unit. Ind. Eng. Chem. Res. 2004, 43, 6027.

(106) Chen, Y.-M. Fluid Catalytic Cracking Technology Recent Eulerian Approach for Circulating Fluidized Beds. Powder Technol.
Advances and New Challenges. Oil Gas Process. Rev. 2006, 19. 2011, 208, 98.
(107) Theologos, K. N.; Markatos, N. C. Advanced Modeling of (130) Shuyan, W.; Yurong, H.; Huilin, L.; Ding, J.; Lijie, Y.; Wentie,
Fluid Catalytic Cracking Riser Type Reactors. AIChE J. 1993, 39, 1007. L. Simulation of Performance of Cracking Reactions of Particle Clusters
(108) Buchanan, J. S. Analysis of Heating and Vaporization of Feed in FCC Risers. Ind. Eng. Chem. Res. 2008, 47, 4632.
Droplets in Fluidized Catalytic Cracking Risers. Ind. Eng. Chem. Res. (131) Cerqueira, H. S.; Baptista, C. M. L. A. Feedstock Effect on
1994, 33, 3104. FCC Catalyst Stripping. Stud. Surf. Sci. Catal. 2004, 149, 287.
(109) Gupta, R. K.; Kumar, V.; Srivastava, V. K. Modeling of Fluid (132) Gauthier, T.; Bayle, J.; Leroy, P. Fluidization Phenomena and
Catalytic Cracking Riser Reactor: A Review. Int. J. Chem. Reactor Eng. Technologies. Gas Sci. Technol.Rev. IFP 2000, 55, 187.
2010, 8. (133) Hedrick, B. W.; Koebel, J. P.; Cetinkaya, I. B.; Puppala, K.
(110) Martignoni, W. P.; Lasa, H. I. de. Heterogeneous Reaction Improved Catalyst Stripping from Cold Flow Modelling. In Akzo-Nobel
Model for FCC Riser Units. Chem. Eng. Sci. 2001, 56, 605. Catalyst Symposium Ecomagic, Noordwijk, The Netherlands, 2001;
(111) Araujo-Monroy, C.; Lopez-Isunza, F. Modeling and Simula- Akzo-Nobel Catalysts: The Netherlands.
tion of an Industrial Fluid Catalytic Cracking Riser Reactor Using a (134) Rall, R. R.; DeMulder, B. New Internal for Maximizing
Lump-Kinetic Model for a Distinct Feedstock. Ind. Eng. Chem. Res. 2006, Performance of FCC Catalyst Strippers. In 12th Annual Stone & Webster
45, 120. Refining Seminar, San Francisco, CA, 2000.
(112) Ahari, J. S.; Farshi, A.; Forsat, K. A Mathematical Modeling of (135) Gao, J.; Chang, J.; Xu, C.; Lan, X.; Yang, Y. CFD Simulation of
the Riser Reactor in Industrial FCC Unit. Petroleum Coal 2008, 50, 15. Gas Solid Flow in FCC Strippers. Chem. Eng. Sci. 2008, 63, 1827.
(113) Benyahia, S.; Arastoopour, H.; Knowlton, T. M.; Massah, H. (136) Wiens, J. S. Experimental and Modeling Study of a Cold-Flow
Simulation of Particles and Gas Flow Behavior in the Riser Section of a Fluid Catalytic Cracking Unit Stripper; University of Saskatchewan:
Circulating Fluidized Bed using the Kinetic Theory Approach for the Saskatoon, Canada, 2010.
Particulate Phase. Powder Technol. 2000, 112, 24. (137) Koon, C.; Akbar, F.; Hughes, R.; Tyagi, Y.; Castrodiaz, M.;
(114) Neri, A.; Gidaspow, D. Riser Hydrodynamics: Simulation Martin, S.; Hall, P.; Snape, C. Development of an Experimental Protocol
Using Kinetic Theory. AIChE J. 2000, 46, 52. to Evaluate FCC Stripper Performance in Terms of Coke Yield and
(115) Lan, X.; Xu, C.; Wang, G.; Wu, L.; Gao, J. CFD Modeling of Composition. Chem. Eng. Res. Des. 2000, 78, 738.
Gas-Solid Flow and Cracking Reaction in Two-Stage Riser FCC (138) Arbel, A.; Huang, Z.; Rinard, I. H.; Shinnar, R.; Sapre, A. V.
Reactors. Chem. Eng. Sci. 2009, 64, 3847. Dynamic and Control of Fluidized Catalytic Crackers. 1. Modeling of
(116) Lopes, G. C.; Rosa, L. M.; Mori, M.; Nunhez, J. R.; Martignoni, the Current Generation of FCC’s. Ind. Eng. Chem. Res. 1995, 34, 1228.
W. P. Three-Dimensional Modeling of Fluid Catalytic Cracking Industrial (139) Secchi, A.; Santos, M.; Neumann, G.; Trierweiler, J. A
Riser Flow and Reactions. Comput. Chem. Eng. 2011, DOI: 10.1016/j. Dynamic Model for a FCC UOP Stacked Converter Unit. Comput.
compchemeng.2010.12.014. Chem. Eng. 2001, 25, 851.
(117) Das, A. K.; Baudrez, E.; Marin, G. B.; Heynderickx, G. J. (140) Hernandez-Barajas, J.; Vazquez-Roman, R.; Salazar-Sotelo, D.
Three-Dimensional Simulation of a Fluid Catalytic Cracking Riser Multiplicity of Steady States in FCC Units: Effect of Operating Condi-
Reactor. Ind. Eng. Chem. Res. 2003, 42, 2602. tions. Fuel 2006, 85, 849.
(118) Behjat, Y.; Shahhosseini, S.; Marvast, M. A. Simulation Study (141) Bollas, G. M.; Vasalos, I. A.; Lappas, A. A.; Iatridis, D. K.;
of Droplet Vaporization Effects on Gas-Solid Fluidized Bed. J. Taiwan Voutetakis, S.; Padopoulou, S. Integrated FCC Riser-Regenerator
Inst. Chem. Eng. 2011, 42, 419. Dynamics Studied in a Fluid Catalytic Cracking Pilot Plant. Chem.
(119) Theologos, K. N.; Lygeros, A. I.; Markatos, N. C. Feedstock Eng. Sci. 2007, 62, 1887.
Atomization Effects on FCC Riser Reactors Selectivity. Chem. Eng. Sci. (142) Vale, H. Development of a Simulator for a Complete R2R
1999, 54, 5617. Catalytic Cracking Unit. M.Sc. Thesis, Instituto Superior Tecnico, 2002.
(120) Derouin, C.; Nevicato, D.; Forissier, M.; Wild, G.; Bernard, J.-R. (143) Errazu, A. F.; Lasa, H. I. de; Sarti, F. A Fluidized Bed Catalytic
Hydrodynamics of Riser Units and Their Impact on FCC Operation. Ind. Cracking Regenerator Model Grid Effects. Can. J. Chem. Eng. 1979,
Eng. Chem. Res. 1997, 36, 4504. 57, 191.
(121) Han, I.-S.; Chung, C.-B. Dynamic Modeling and Simulation of (144) Lasa, H. I. de; Grace, J. R. The Influence of the Freeboard
a Fluidized Catalytic Cracking Process. Part II: Property Estimation and Region in a Fluidized Bed Catalytic Cracking Regenerator. AIChE J.
Simulation. Chem. Eng. Sci. 2001, 56, 1973. 1979, 25, 984.
(122) Fligner, M.; Schipper, P.; Sapre, A. V.; Krambeck, F. Two (145) Morley, K.; Lasa, H. I. de. On the Determination of Kinetic
Phase Cluster Model in Riser Reactors: Impact of Radial Density Parameters for the Regeneration of Cracking Catalyst. Can. J. Chem. Eng.
Distribution on Yields. Chem. Eng. Sci. 1994, 49, 5813. 1987, 65, 773.
(123) Harriott, P. Chemical Reactor Design; CRC Press: New York, (146) Weisz, P. Combustion of Carbonaceous Deposits within
2003. Porous Catalyst Particles III. The CO2/CO Product Ratio. J. Catal.
(124) P€arssinen, J. H. Hydrodynamics of Low-Flux and High-Flux 1966, 6, 425.
Circulating Fluidized Beds. Acta Polytech. Scand., Mech. Eng. Ser. 2002, (147) Faltsi-Saravelou, O.; Vasalos, I. A. FBSim: A Model for
161, 119. Fluidized Bed Simulation I. Dynamic Modeling of an Adiabatic
(125) Subbarao, D. A Cluster Model for Mass Transfer in Risers. Sci. Reacting System of Small Gas Fluidized Particles. Comput. Chem. Eng.
Technol. 2008, 3, 131. 1991, 15, 639.
(126) Xu, J.; Zhu, J.-X. Visualization of Particle Aggregation and (148) Kunii, D.; Levenspiel, O. Fluidized Reactor Models. 1. For
Effects of Particle Properties on Cluster Characteristics in a CFB Riser. Bubbling Beds of Fine, Intermediate, and Large Particles. 2. For the Lean
Chem. Eng. J. 2011, 168, 376. Phase: Freeboard and Fast Fluidization. Ind. Eng. Chem. Res. 1990,
(127) Wang, J.; Ge, W.; Li, J. Eulerian Simulation of Heterogeneous 29, 1226.
GasSolid Flows in CFB Risers: EMMS-Based Sub-Grid Scale Model (149) Gupta, R. K.; Kumar, V.; Srivastava, V. K. Modeling and
with a Revised Cluster Description. Chem. Eng. Sci. 2008, 63, 1553. Simulation of Fluid Catalytic Cracking Unit. Rev. Chem. Eng. 2005, 21, 95.
(128) Shah, M. T.; Utikar, R. P.; Tade, M. O.; Pareek, V. K.; Evans, (150) Maciel Filho, R.; Batista, L. M. F. L.; Fusco, M. A Fast
G. M. Simulation of GasSolid Flows in Riser Using Energy Minimiza- Fluidized Bed Reactor for Industrial FCC Regenerator. Chem. Eng. Sci.
tion Multiscale Model: Effect of Cluster Diameter Correlation. Chem. 1996, 51, 1807.
Eng. Sci. 2011, 1. (151) Krishna, A. S.; Parkin, E. S. Modeling the Regenerator in
(129) Shuai, W.; Huilin, L.; Guodong, L.; Zhiheng, S.; Pengfei, X.; Commercial Fluid Catalytic Cracking Units. Chem. Eng. Prog. 1985,
Gidaspow, D. Modeling of Cluster Structure-Dependent Drag with 81, 57.

(152) Lee, L.-S.; Yu, S.-W.; Cheng, C.-T. Fluidized-Bed Catalyst (176) Moro, L. F. L.; Odloak, D. Constrained Multivariable Control
Cracking Regenerator Modelling and Analysis. Chem. Eng. J. 1989, 40, 71. of Fluid Catalytic Cracking Converters. J. Process Control 1995, 5, 29.
(153) Zwinkels, M.; Nougier, L. FCC Regenerator Simulation (177) Arbel, A.; Rinard, I. H.; Shinnar, R.; Sapre, A. V. Dynamics and
Model. Ph.D. Thesis, Institute Franc-ais du Petrole (IFP), 1997. Control of Fluidized Catalytic Crackers. 2. Multiple Steady States and
(154) Arandes, J. M.; Azkoiti, M. J.; Bilbao, J.; Lasa, H. I. de. Instabilities. Ind. Eng. Chem. Res. 1995, 34, 3014.
Modelling FCC Units under Steady and Unsteady State Conditions. (178) Arbel, A.; Rinard, I. H.; Shinnar, R. Dynamics and Control of
Can. J. Chem. Eng. 2000, 78, 111. Fluidized Catalytic Crackers. 3. Designing the Control System: Choice
(155) Penteado, J. C.; Rossi, L. F. S.; Negr~ao, C. O. R. Numerical of Manipulated and Measured Variables for Partial Control. Ind. Eng.
Modeling of a FCC Regenerator. In 17th International Congress of Mechanical Chem. Res. 1996, 35, 2215.
Engineering (COBEM 2003), S~ao Paulo, Brazil, 2003; Associac-~ao Brasileira (179) Arbel, A.; Rinard, I. H.; Shinnar, R. Dynamics and Control of
de Ciências Mecânicas (ABCM): S~ao Paulo, Brazil. Fluidized Catalytic Crackers. 4. The Impact of Design on Partial
(156) Guisnet, M.; Magnoux, P. Deactivation by Coking of Zeolite Control. Ind. Eng. Chem. Res. 1997, 36, 747.
Catalysts. Prevention of Deactivation. Optimal Conditions for Regen- (180) Fernandes, J. L. Nonlinear Modelling of Industrial Fluid
eration. Catal. Today 1997, 36, 477. Catalytic Cracking Processes for Model-Based Control and Optimiza-
(157) Arthur, J. R. Reactions Between Carbon and Oxygen. Trans. tion Studies. Ph.D. Thesis, Instituto Superior Tecnico, 2007.
Faraday Soc. 1951, 47, 164. (181) Fernandes, J. L.; Verstraete, J. J.; Pinheiro, C. I. C.; Oliveira,
(158) Weisz, P.; Goodwin, R. Combustion of Carbonaceous Deposits N. M. C.; Ramôa Ribeiro, F. Dynamic Modelling of an Industrial R2R
within Porous Catalyst Particles I. Diffusion-Controlled Kinetics. J. Catal. FCC Unit. Chem. Eng. Sci. 2007, 62, 1184.
1963, 2, 397. (182) Fernandes, J. L.; Pinheiro, C. I. C.; Oliveira, N. M. C.; Ramôa
(159) Weisz, P.; Goodwin, R. Combustion of Carbonaceous Depos- Ribeiro, F. Model Development and Validation of an Industrial UOP
its within Porous Catalyst Particles II. Intrinsic Burning Rate. J. Catal. Fluid Catalytic Cracking Unit with a High-Efficiency Regenerator. Ind.
1966, 6, 227. Eng. Chem. Res. 2008, 47, 850.
(160) Wang, G.; Lin, S.; Mo, W.; Peng, C.; Yang, G. Kinetics of (183) Levenspiel, O. Chemical Reaction Engineering, 3rd ed.; Wiley:
Combustion of Carbon and Hydrogen in Carbonaceous Deposits on Zeolite- Hoboken, NJ, 1998.
Type Cracking Catalysts. Ind. Eng. Chem. Process Des. Dev. 1986, 25, 626. (184) Fogler, H. S. Elements of Chemical Reaction Engineering, 4th
(161) Morley, K.; Lasa, H. I. de. Regeneration of Cracking Catalyst ed.; Goodwind, B. M., Ed.; Prentice Hall: Upper Saddle River, NJ, 2006.
Influence of the Homogeneous CO Postcombustion Reaction. Can. J. Chem. (185) Lemos, F. Lopes, J. M.; Ribeiro, F. R. Reactores Químicos
Eng. 1988, 66, 428. (Chemical Reactors); IST Press: Lisbon, Portugal, 2002.
(162) Faltsi-Saravelou, O.; Vasalos, I. A. FBSim: A Model for (186) Fernandes, J. L.; Pinheiro, C. I. C.; Oliveira, N. M. C.; Neto,
Fluidized Bed Simulation II. Simulation of an Industrial Fluidized A. I.; Ramôa Ribeiro, F. Steady State Multiplicity in an UOP FCC Unit
Catalytic Cracking Regenerator. Comput. Chem. Eng. 1991, 15, 647. with High-Efficiency Regenerator. Chem. Eng. Sci. 2007, 62, 6308.
(163) Kumar, S.; Chadha, A.; Gupta, R.; Sharma, R. CATCRAK: (187) Arandes, J. M.; Lasa, H. I. de Simulation and Multiplicity of
A Process Simulator for an Integrated FCC-Regenerator System. Ind. Steady States in Fluidized FCCUs. Chem. Eng. Sci. 1992, 47, 2535.
Eng. Chem. Res. 1995, 34, 3737. (188) Iscol, L. The Dynamics and Stability of a Fluid Catalytic Cracker.
(164) Jacob, S. M.; Gross, B.; Voltz, S. E.; Weekman, V. W. A In Joint Automatic Control Conference, Atlanta, GA, 1970; ASME: New York.
Lumping and Reaction Scheme for Catalytic Cracking. AIChE J. 1976, (189) Oliveira, N. M. C. Newton-Type Algorithms for Nonlinear
22, 701. Constrained Chemical Process Control. Ph.D. Thesis, Carnegie Mellon
(165) Elshishini, S. S.; Elnashaie, S. S. E. H. Digital Simulation of University, 1994.
Industrial Fluid Catalytic Cracking Units: Bifurcation and its Implica- (190) Sistu, P. B.; Bequette, W. Model Predictive Control of
tions. Chem. Eng. Sci. 1990, 45, 553. Processes with Input Multiplicities. Chem. Eng. Sci. 1995, 50, 921.
(166) Elshishini, S. S.; Elnashaie, S. S. E. H. Digital Simulation of (191) Lovo, M.; Balakotaiah, V. Multiplicity Features of Adiabatic
Industrial Fluid Catalytic Cracking Units II. Effect of Charge Stock Autothermal Reactors. AIChE J. 1992, 38, 101.
Composition on Bifurcation and Gasoline Yield. Chem. Eng. Sci. 1990, (192) Koppel, L. B. Input Multiplicities in Nonlinear, Multivariable
45, 2959. Control Systems. AIChE J. 1982, 28, 935.
(167) Elshishini, S. S.; Elnashaie, S. S. E. H.; Alzahrani, S. Digital (193) Oliveira, N. M. C.; Biegler, L. T. Constraint Handling and
Simulation of Industrial Fluid Catalytic Cracking Units III. Effect of Stability Properties of Model-Predictive Control. AIChE J. 1994,
Hydrodynamics. Chem. Eng. Sci. 1992, 47, 3152. 40, 1138.
(168) Elnashaie, S. S. E. H.; Mohamed, N. F.; Kamal, M. Simulation (194) Yang, S. H.; Wang, X. Z.; McGreavy, C. A Multivariable
and Static Bifurcation Behavior of Industrial FCC Units. Chem. Eng. Coordinated Control System based on Predictive Control Strategy for
Commun. 2004, 191, 813. FCC Reactor-Regenerator System. Chem. Eng. Sci. 1996, 51, 2977.
(169) Lee, W.; Kugelman, A. M. Number of Steady State Operating (195) Narasimhan, S.; Jordache, C. Data Reconciliation and Gross
Points and Local Stability of Open-Loop Fluid Catalytic Cracker. Ind. Error Detection: An Intelligent Use of Process Data; Gulf Professional
Eng. Chem. Process Des. Dev. 1973, 12, 197. Publishing: Houston, TX, 1999.
(170) Edwards, W. M.; Kim, H. N. Multiple Steady States in FCC (196) Bagajewicz, M. Smart Process Plants: Software and Hardware
Unit Operations. Chem. Eng. Sci. 1988, 43, 1825. Solutions for Accurate Data and Profitable Operations: Data Reconciliation,
(171) McFarlane, R. C.; Reinman, R. C.; Bartee, J. F.; Georgakis, C. Gross Error Detection, and Instrumentation Upgrade; McGraw-Hill Pro-
Dynamic Simulator for a Model IV Fluid Catalytic Cracking Unit. fessional: New York, 2009.
Comput. Chem. Eng. 1993, 17, 275. (197) Stratiev, D.; Dinkov, R. Evaluation of FCC Unit Process
(172) Zheng, Y.-Y. Dynamic Modeling and Simulation of a Catalytic Variables Impact on Yield Distribution and Product Quality. Part I.
Cracking Unit. Comput. Chem. Eng. 1994, 18, 39. Evaluation of FCC Unit Variables Impact on Yield Distribution.
(173) Maya-Yescas, R.; Bogle, D.; Lopez-Isunza, F. Approach to the Petroleum Coal 2007, 49, 71.
Analysis of the Dynamics of Industrial FCC Units. J. Process Control 1998, 8, 89. (198) Agachi, P. S-.; Nagy, Z. K.; Cristea, M. V.; Imre-Lucaci, A .
(174) Elnashaie, S. S. E. H.; Elshishini, S. S. Digital Simulation of Model Based Control: Case Studies in Process Engineering; Wiley-VCH
Industrial Fluid Catalytic Cracking Units IV. Dynamic Behaviour. Verlag GmbH: Weinheim, Germany, 2006.
Chem. Eng. Sci. 1993, 48, 567. (199) Nagy, Z. K.; Agachi, P. S-.; Bodizs, L. Adaptive Neural Network
(175) Cristea, M. V.; Agachi, P. S-.; Marinoiu, V. Simulation and Model Based Nonlinear Predictive Control of a Fluid Catalytic Cracking
Model Predictive Control of a UOP Fluid Catalytic Cracking Unit. Unit. In 10th European Symposium on Computer Aided Process Engineering
Chem. Eng. Process. 2003, 42, 67. (ESCAPE-10), Florence, Italy, 2000; Elsevier: The Netherlands.

(200) Perez, J. R.; Lopez, R.; Dassori, C. G.; Ranson, A. Artificial Neural (223) Hovd, M.; Skogestad, S. Controllability Analysis for the Fluid
Networks Applied to the Operation of a Fluid Catalytic Cracking (FCC) Catalytic Cracking Process. In 1991 AIChE Annual Meeting, Los Angeles,
Unit. In SPE Latin American and Caribbean Petroleum Engineering Conference, CA; AIChE: New York, 1991.
Buenos Aires, Argentina, 2001; Society of Petroleum Engineers: U.K. (224) Alhumaizi, K.; Elnashaie, S. S. E. H. Effect of Control Loop
(201) Wang, X. Z.; McGreavy, C. Automatic Classification for Configuration on the Bifurcation Behavior and Gasoline Yield of
Mining Process Operational Data. Ind. Eng. Chem. Res. 1998, 37, 2215. Industrial Fluid Catalytic Cracking (FCC) Units. Math. Comput. Modell.
(202) Sebzalli, Y. M.; Wang, X. Z. Knowledge Discovery from 1997, 25, 37.
Process Operational Data using PCA and Fuzzy Clustering. Eng. Appl. (225) Lee, E.; Groves, F. Mathematical Model of the Fluidized Bed
Artif. Intell. 2001, 14, 607. Catalytic Cracking Plant. Trans. Soc. Comput. Simul. 1985, 2, 219.
(203) Pranatyasto, T. N.; Qin, S. J. Sensor Validation and Process (226) Hicks, R. C.; Worrel, G. R.; Durney, R. J. Atlantic Seeks
Fault Diagnosis for FCC Units under MPC Feedback. Control Eng. Pract. Improved Control Studies Analog-Digital Models. Oil Gas J. 1966, 24, 97.
2001, 9, 877. (227) Kurihara, H. Optimal Control of Fluid Catalytic Cracking
(204) Crowe, C. Data Reconciliation — Progress and Challenges. Processes. Ph.D. Thesis, Massachusetts Institute of Technology, 1967.
J. Process Control 1996, 6, 89. (228) Khandalekar, P. D.; Riggs, J. B. Nonlinear Process Model
(205) Albuquerque, J. S.; Biegler, L. T.; Kass, R. E. Inference in Based Control and Optimization of a Model IV FCC Unit. Comput.
Dynamic Error-in-Variable-Measurement Problems. AIChE J. 1997, Chem. Eng. 1995, 19, 1153.
43, 986. (229) Ramachandran, R.; Rangaiah, G. P.; Lakshminarayanan, S.
(206) Camolesi, V. J.; Moro, L. L. F.; Zanin, A. C. Implantac-~ao de Data Analysis, Modeling and Control Performance Enhancement of an
um Otimizador em Tempo Real (RTO) no Conversor de uma Unidade Industrial Fluid Catalytic Cracking Unit. Chem. Eng. Sci. 2007, 62, 1958.
de Craqueamento Catalítico. SBA: Controle Automac-~ao Soc. Brasileira (230) Chu, J.-Z.; Tsai, P.-F.; Tsai, W.-Y.; Jang, S.-S.; Shieh, S.-S.; Lin,
Automatica 2008, 19, 128. P.-H.; Jiang, S.-J. Multistep Model Predictive Control Based on Artificial
(207) Aguilar, R.; Maya-Yescas, R. Feedback Regulation of Tem- Neural Networks. Ind. Eng. Chem. Res. 2003, 42, 5215.
perature in FCC Regenerator Reactors. Petroleum Sci. Technol. 2004, (231) Pandimadevi, G.; Indumathi, P.; Selvakumar, V. Design of
22, 201. Controllers for a Fluidized Catalytic Cracking Process. Chem. Eng. Res.
(208) Llanes, J. M.; Miranda, M.; Mullick, S. Use Modeling to Fine- Des. 2010, 88, 875.
Tune Cracking Operations. Hydrocarbon Process. 2008, September, 123. (232) Alaradi, A.-A.; Rohani, S. Identification and Control of a Riser-
(209) Aguilar, R.; Poznyak, A.; Martínez-Guerra, R.; Maya-Yescas, R. Type FCC Unit Using Neural Networks. Comput. Chem. Eng. 2002, 26, 401.
Temperature Control in Catalytic Cracking Reactors via a Robust PID (233) Biagiola, S. I.; Figueroa, J. L. Application of State Estimation
Controller. J. Process Control 2002, 12, 695. Based NMPC to an Unstable Nonlinear Process. Chem. Eng. Sci. 2004,
(210) Han, I.-S.; Riggs, J. B.; Chung, C.-B. Modeling and Optimiza- 59, 4601.
tion of a Fluidized Catalytic Cracking Process under Full and Partial (234) Ferramosca, A.; Limon, D.; Gonzalez, A. H.; Odloak, D.;
Combustion Modes. Chem. Eng. Process. 2004, 43, 1063. Camacho, E. F. MPC for Tracking Zone Regions. J. Process Control 2010,
(211) Hovd, M.; Skogestad, S. Procedure for Regulatory Control 20, 506.
Structure Selection with Application to the FCC Process. AIChE J. 1993, (235) First Adaptive Multivariable Controller. Hydrocarbon Process.
39, 1938. 2006, January, 29.
(212) Han, I.-S.; Riggs, J. B.; Chung, C.-B. Multivariable Control of a (236) Caldwell, J. M.; Dearwater, J. G. Model Predictive Control
Fluidized Catalytic Cracking Process under Full and Partial Combustion Applied to FCC Units. In Chemical Process Control —CPCIV; Arkun, Y.;
Modes. J. Chem. Eng. Jpn. 2002, 35, 830. Ray, W. H. Eds.; CACHE-AICHE: Padre Island, Texas, 1991; pp 319334.
(213) Souza, J. A.; Vargas, J. V. C.; Meien, O. F. von; Martignoni, (237) Kalra, L.; Georgakis, C. Effect of Process Nonlinearity on the
W. P.; Ordonez, J. C. The Inverse Methodology of Parameter Estimation Performance of Linear Model Predictive Controllers for the Environ-
for Model Adjustment, Design, Simulation, Control and Optimization of mentally Safe Operation of a Fluid Catalytic Cracking Unit. Ind. Eng.
Fluid Catalytic Cracking (FCC) Risers. J. Chem. Technol. Biotechnol. Chem. Res. 1994, 33, 3063.
2009, 84, 343. (238) Kalra, L.; Georgakis, C. The Effects of Operational Character-
(214) Zabiri, H.; Samyudia, Y. A Hybrid Formulation and Design of istics of Catalytic Cracking Reactors on the Closed-Loop Performance
Model Predictive Control for Systems under Actuator Saturation and of Linear Model Predictive Controllers. Comput. Chem. Eng. 1996,
Backlash. J. Process Control 2006, 16, 693. 20, 401.
(215) Alvarez-Ramirez, J.; Valencia, J.; Puebla, H. Multivariable (239) Cristea, M. V.; Agachi, P. S-.; Marinoiu, V. Simulation and
Control Configurations for Composition Regulation in a Fluid Catalytic Model Predictive Control of a UOP Fluid Catalytic Cracking Unit.
Cracking Unit. Chem. Eng. J. 2004, 99, 187. Chem. Eng. Process. 2003, 42, 67.
(216) Kookos, I. K. Real-Time Regulatory Control Structure Selec- (240) Jia, C.; Rohani, S.; Jutan, A. FCC Unit Modeling, Identifica-
tion Based on Economics. Ind. Eng. Chem. Res. 2005, 44, 3993. tion and Model Predictive Control, a Simulation Study. Chem. Eng.
(217) Cristea, M. V.; Agachi, P. S-. Comparison between Different Process. 2003, 42, 311.
Control Approaches of the UOP Fluid Catalytic Cracking Unit. In 17th (241) Iancu, M.; Cristea, M. V.; Agachi, P. S-. MPC vs PID. The Advanced
European Symposium on Computer Aided Process Engineering (ESCAPE- Control Solution for an Industrial Heat Integrated Fluid Catalytic Cracking
17), Bucharest, Romania, 2007; Elsevier: The Netherlands. Plant. In 21st European Symposium on Computer Aided Process Engineering
(218) Dong, H.; Riggs, J. B. Detection of Unmeasured Disturbances (ESCAPE-21), Chalkidiki, Greece, 2011; Elsevier: The Netherlands.
in Model Predictive Control (MPC) Plant Test Data. Ind. Eng. Chem. (242) Ansari, R. M.; Tade, M. O. Constrained Nonlinear Multi-
Res. 2008, 47, 332. variable Control of a Fluid Catalytic Cracking Process. J. Process Control
(219) Roman, R.; Nagy, Z. K.; Cristea, M. V.; Agachi, P. S-. Dynamic 2000, 10, 539.
Modelling and Nonlinear Model Predictive Control of a Fluid Catalytic (243) Harnischmacher, G.; Marquardt, W. Nonlinear Model Pre-
Cracking Unit. Comput. Chem. Eng. 2009, 33, 605. dictive Control of Multivariable Processes using Block-Structured
(220) Gonzalez, A. H.; Odloak, D. A Stable MPC with Zone Models. Control Eng. Pract. 2007, 15, 1238.
Control. J. Process Control 2009, 19, 110. (244) Vieira, W. G.; Santos, V. M. L.; Carvalho, F. R.; Pereira, J. A. F.
(221) Qin, S. J.; Badgwell, T. A. A Survey of Industrial Model R.; Fileti, A. M. F. Identification and Predictive Control of a FCC Unit
Predictive Control Technology. Control Eng. Pract. 2003, 11, 733. using a MIMO Neural Model. Chem. Eng. Process. 2005, 44, 855.
(222) Ali, E. E.; Elnashaie, S. S. E. H. Nonlinear Model Predictive (245) Santos, V. M. L.; Carvalho, F. R.; Souza, M. B. de, Jr. Predictive
Control of Industrial Type IV Fluid Catalytic Cracking (FCC) Units for Control based on Neural Networks: An Application to a Fluid Catalytic
Maximum Gasoline Yield. Ind. Eng. Chem. Res. 1997, 36, 389. Cracking Industrial Unit. Braz. J. Chem. Eng. 2000, 17.

(246) Souza, R. G. de; Souza Jr., M. de; Bhaya, A. Analise e Controle

de uma Unidade FCC do tipo Kellog Orthoflow F. In V Simposio
Brasileiro de Automac-~ao Inteligente, Canela, RS, Brasil, November 2001.
(247) Taskin, H.; Kubat, C.; Uygun, O.; Arslankaya, S. FUZZYFCC:
Fuzzy Logic Control of a Fluid Catalytic Cracking Unit (FCCU) to
Improve Dynamic Performance. Comput. Chem. Eng. 2006, 30, 850.
(248) Aguilar, R.; Gonzalez, J.; Alvarez-Ramirez, J.; Barron, M.
Control of a Fluid Catalytic Cracking Unit Based on Proportional-
Integral Reduced Order Observers. Chem. Eng. J. 1999, 75, 77.
(249) Huang, Y.; Reklaitis, G. V.; Venkatasubramanian, V. A Heuristic
Extended Kalman Filter Based Estimator for Fault Identification in a Fluid
Catalytic Cracking Unit. Ind. Eng. Chem. Res. 2003, 42, 3361.
(250) Zanin, A. C.; Niederberger, J.; Camolesi, V. J.; Moro, L. F. L.
Industrial Implementation of Real Time Optimization in an FCC Converter.
In 2nd Mercosur Congress on Chemical Engineering, Rio de Janeiro, Brazil, 2005;
http://www.enpromer2005.eq.ufrj.br/nukleo/pdfs/0371_a371_enpromer_
2005.pdf.
(251) Adetola, V.; Guay, M. Integration of Real-Time Optimization
and Model Predictive Control. J. Process Control 2010, 20, 125.
(252) McFarlane, R. C.; Bacon, D. W. Adaptive Optimizing Control
of Multivariable Constrained Chemical Processes. 1. Theoretical Devel-
opment. Ind. Eng. Chem. Res. 1989, 28, 1828.
(253) Grosdidier, P.; Mason, A.; Aitolahti, A.; Heinonen, P.; Vanham€aki,
V. FCC Unit Reactor-Regenerator Control. Comput. Chem. Eng. 1993, 17,
165.
(254) Zanin, A. C. Implementac-~ao Industrial de um Otimizador em
Tempo Real (Industrial Implementation of a Real-Time Optimizer). Ph.
D. Thesis, Escola Politecnica da Universidade de S.Paulo, 2001.
(255) Zanin, A. C.; Gouvêa, M. T. de; Odloak, D. Integrating Real-
Time Optimization into the Model Predictive Controller of the FCC
System. Control Eng. Pract. 2002, 10, 819.
(256) Yang, S. H.; Chung, P. W. H.; Wang, X. Z.; McGreavy, C.
Multi-Objective Constraint Control for FCC Reactor-Regenerator
System. Comput. Chem. Eng. 1998, 22, S831.
(257) Lid, T.; Strand, S. Real-Time Optimization of a Cat Cracker
Unit. Comput. Chem. Eng. 1997, 21, S887.
(258) Gouvêa, M. T. de; Odloak, D. One-Layer Real Time Optimi-
zation of LPG Production in the FCC Unit: Procedure, Advantages and
Disadvantages. Comput. Chem. Eng. 1998, 22, S191.
(259) Loeblein, C.; Perkins, J. D. Structural Design for On-Line Process
Optimization: II. Application to a Simulated FCC. AIChE J. 1999, 45, 1030.
(260) Souza, G. de; Odloak, D.; Zanin, A. C. Real Time Optimiza-
tion (RTO) with Model Predictive Control (MPC). Comput. Chem. Eng.
2010, 34, 1999.
(261) Ying, C.-M.; Joseph, B. Performance and Stability Analysis of
LP-MPC and QP-MPC Cascade Control Systems. AIChE J. 1999, 45,
1521.
(262) Tatjewski, P. Supervisory Predictive Control and On-Line Set-
Point Optimization. Int. J. Appl. Math. Comput. Sci. 2010, 20, 483.
(263) Tatjewski, P. Advanced Control and On-Line Process Opti-
mization in Multilayer Structures. Annu. Rev. Control 2008, 32, 71.
(264) Prett, D. M.; Gillette, R. D. Optimization and Constrained
Multivariable Control of a Catalytic Unit. In Joint Automatic Control
Conference, San Francisco, CA, 1980; ASME: New York.
(265) McFarlane, R. C.; Bacon, D. W. Adaptive Optimizing Control
of Multivariable Constrained Chemical Processes. 2. Application Stud-
ies. Ind. Eng. Chem. Res. 1989, 28, 1834.
(266) Ellis, R. C.; Li, X.; Riggs, J. B. Modeling and Optimization of a
Model IV Fluidized Catalytic Cracking Unit. AIChE J. 1998, 44, 2068.
(267) Loeblein, C.; Perkins, J. D. Structural Design for On-Line
Process Optimization: I. Dynamic Economics of MPC. AIChE J. 1999,
45, 1018.
(268) Abou-Jeyab, R. A.; Gupta, Y. P. Control of the Fluidized
Catalytic Cracking Process Using a Simplified Model Predictive Con-
troller. Ind. Eng. Chem. Res. 1996, 35, 3581.

Fluid Catalytic Cracking FCC Process Mod

Uploaded by

Copyright:

Available Formats

Fluid Catalytic Cracking FCC Process Mod

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fluid Catalytic Cracking FCC Process Mod

Uploaded by

Copyright:

Available Formats

What is the FCC process and what are its main steps?

What is the FCC process and what are its main steps?

What catalyst is used in the FCC process and what are its main components?

What catalyst is used in the FCC process and what are its main components?

REVIEW

Fluid Catalytic Cracking (FCC) Process Modeling, Simulation,

2 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

region. According to this theory,147 the gas ﬂow is divided into

models for the application of model-based supervision strategies

5. CONTROL AND OPTIMIZATION OF FLUID CATALY-

Table 1. FCC Decentralized Control Structures211

Table 2. FCC Control Structures for LMPC175

Table 3. Summary of RTO Studies Based on One-Layer and Two-Layer Approaches

’ ABBREVIATIONS VGO vacuum gas oil

23 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

24 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

25 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

26 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

27 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

28 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

(246) Souza, R. G. de; Souza Jr., M. de; Bhaya, A. Analise e Controle

29 dx.doi.org/10.1021/ie200743c |Ind. Eng. Chem. Res. 2012, 51, 1–29

You might also like