fcc modeling

Published: October 19, 2011

r 2011 American Chemical Society 5298 dx.doi.org/10.1021/ef200750x | Energy Fuels 2011, 25, 5298–5319

ARTICLE

pubs.acs.org/EF

Predictive Modeling of Large-Scale Integrated Refinery Reaction andFractionation Systems from Plant Data. Part 2: Fluid Catalytic Cracking(FCC) ProcessKiran Pashikanti and Y. A. Liu*

SINOPEC/AspenTech Center of Excellence in Process System Engineering, Department of Chemical Engineering, Virginia PolytechnicInstitute and State University, Blacksburg, Virginia 24061, United States

ABSTRACT:This work presents themethodology to develop, validate, and apply a predictivemodel for an integrated fluid catalyticcracking (FCC) process. We demonstrate the methodology using data from a commercial FCC plant in the Asia Pacific with a feedcapacity of 800 000 tons per year. Our model accounts for the complex cracking kinetics in the riser�regenerator with a 21-lumpkinetic model. We implement the methodology with Microsoft Excel spreadsheets and a commercial software tool, Aspen HYSYS/Petroleum Refining from Aspen Technology, Inc. The methodology is equally applicable to other commercial software tools. Thismodel gives accurate predictions of key product yields and properties given feed qualities and operating conditions. In addition, thiswork presents the first lumped FCC kinetic model integrated with a gas plant model in the literature. We validate this work using 6months of plant data. We also perform several case studies to show how refiners may apply this work to improve the gasoline yieldand increase unit throughput. A key application of the integrated FCC model is to generate DELTA�BASE vectors for linearprogramming (LP)-based refinery planning to help refiners choose an optimum slate of crude feeds. DELTA�BASE vectorsquantify changes in FCC product yields and properties as functions of changes in feed and operating conditions. Traditionally,refiners generated DELTA�BASE vectors using a combination of historical data and correlations. Our integrated model caneliminate guesswork by providing more robust predictions of product yields and qualities. This work differentiates itself fromprevious work in this area through the following contributions: (1) detailed models of the entire FCC plant, including the overheadgas compressor, main fractionator, primary and sponge oil absorber, primary stripper, and debutanizer columns, (2) process to infermolecular composition required for the kinetic model using routinely collected bulk properties of feedstock, (3) predictions of keyliquid product properties not published alongside previous related work (density, ASTMD86 distillation curve, and flash point), (4)case studies showing industrially useful applications of the model, and (5) application of the model with an existing LP-basedplanning tool.

1. INTRODUCTION

The current economic, political, and regulatory climates placesignificant pressures on petroleum refiners to optimize andintegrate the refining process. The fluid catalytic cracking(FCC) unit is the largest producer of gasoline and light endsin the refinery.1 It plays a critical role in the profitable operationof any refinery. Plant operators can make minor adjustmentsbased on experience to improve the yield and efficiency of theFCC unit. However, major improvements must come from aconcerted effort that involves understanding the reaction chem-istry, feed characteristics, and equipment performance. In suchan endeavor, the use of rigorous simulation models is critical. Inparticular, rigorous simulation models validated with plant datacan identify key areas for process improvements.

There is significant previous work that addresses the issues ofprocess dynamics and control for the integrated FCC unit. Weparticularly note the efforts by Arbel et al.2 andMcFarlane et al.3 inthis regard. Subsequent authors4,5 use similar techniques andmodels to identify control schemes and yield behavior. However,most of the earlier work uses a very simplified reaction chemistry(yield model) to represent the process kinetics. In addition, priorwork in the literature (to our knowledge) does not connect theintegrated FCCmodelwith the complex FCC fractionation system.This work fills the gap between the development of a rigorouskinetic model and industrial application in a large-scale refinery.

2. PROCESS DESCRIPTION

The FCC unit is the primary producer of gasoline and olefins inthe refinery. Current FCC designs are based on continual improve-ments and advances in unit and catalyst design since 1940. Therearemanypopular FCCdesigns in use today, andwe choose to focuson a UOP FCC unit. The UOP design includes many features thathighlight the unique characteristics of the FCC process. Figure 1shows a general schematic of the FCC unit. We discuss the processflow and unit design in the following section.2.1. Riser�Regenerator Complex. A hot fluidized catalyst

(1000 �F+ or 538 �C+) enters the bottom riser through a standpipe,where it comes in contact with a preheated gas-oil feed. The gas-oilfeed typically consists of vacuum gas oil (VGO) from the vacuumtower, coker gas oil (CGO) from the delayed coker, and recycledproducts from the FCCmain fractionator (Figure 2). The heat fromthe hot catalyst (and any additional steam or fuel gas added to thestandpipe) is sufficient to vaporize the gas-oil feed. The componentsof the vaporized gas oil undergo several reactions over the catalystsurface: hydrocracking, isomerization, hydrogenation/dehydrogena-tion, alkylation/dealkylation, cyclization/decyclization, and conden-sation. These reactions result in components that make up the

Received: May 19, 2011Revised: August 3, 2011

5299 dx.doi.org/10.1021/ef200750x |Energy Fuels 2011, 25, 5298–5319

Energy & Fuels ARTICLE

product slate. The products typically present are dry gas (hydrogen,methane, and ethane), sour gas (H2S), liquid petroleum gas(propane, propylene, butanes, and butenes), gasoline (up to 430�F), light cycle oil (LCO), heavy cycle oil (HCO), slurry or decantoil, and coke. Properties of the feed oil and impurities present on thecatalyst significantly affect the distribution of products and theoperating profile in the riser.The catalyst travels to the top of the riser carrying heavy

components and coke deposits from preceding reactions. Thecatalyst enters a stripping zone where some steam is added tofurther crack and remove the heavy hydrocarbons from thecatalyst surface. The catalyst then enters the reactor section,where a cyclone separates the catalyst from the product vapor.The separated product vapor is sent to the main fractionationcolumn (Figure 2) that separates the product into gaseous andliquid products. The separated catalyst is piped into the regen-erator, where the coke on the catalyst is burned off.The separated catalyst typically contains about 0.4�2.5% of

coke by weight.1 Air and possibly pure oxygen (depending uponunit configuration) also enter into the regenerator through addi-tional ports. Freshmakeup coke also enters the FCCplant throughadditional ports. The coke is mostly oxidized, producing CO2 andCO as primary products and SOx and NOx as secondary products.These flue gas products are typically used in heat-integration loopsto provide steam to the plant. The catalyst is typically oxidized to alevel containing 0.05% coke by weight.1 This oxidization also heatsthe catalyst as it re-enters the riser through the standpipe.2.2. Downstream Fractionation. The effluent from the FCC

enters the main fractionator with a significant quantity of steam,as shown in Figure 2. This fractionator separates the reactoreffluent into four product groups: light gases (C1�C4), gasoline(C5+ to 430 �F or 221 �C), LCO and HCO (430�650 �F or221�343 �C), and slurry/decant oil (650+ �F or 343+ �C). Thetemperature range of these products varies in different refineries

(or different operating scenarios in the same refinery) dependingupon the product demand and current operating constraints.There are several pumparounds associated with the main frac-tionator that help control the product distribution and tempera-ture profiles. Most of the products from the main fractionatorcannot be sent directly into the product blending pool of therefinery. Additional fractionation and product isolation occur inthe gas plant associated with the FCC unit, as shown in Figure 3.The overhead vapor contains some C5 components, which mustbe recovered in the product gasoline. A portion of the LCOproduct is drawn off as sponge oil to recover gasoline in a spongeoil absorber. The liquid from the overhead condenser flows to theprimary absorber, where C3�C4 components are recovered.There is significant value in separating and isolating the C3�C4

components. These components may be sold as liquefied petro-leum gas (LPG) or serve as a valuable feedstock for otherpetrochemical processes. The FCC gas plant is responsible forthe separation of C3�C4 components and stabilization of gaso-line. The latter refers to controlling the amount of C4 componentspresent in the product gasoline.The overhead vapor from themain fractionation column enters the

wet gas compressor train.The vapor leaving the compressor train thenenters a high-pressure flash system. The vapor from the high-pressureflash enters the primary absorber. The C5 components leave with thebottom product from the primary absorber. This bottom product re-enters the high-pressure flash. The overhead vapor product enters asponge oil absorber, where it is contacted with LCO drawn off fromthe main fractionator. The overhead products of the sponge oilabsorber are H2, C1, and C2 components that can serve as feeds tomeet the energy demands of the refinery. The bottom product fromthe sponge oil absorber is recycled back to the main fractionator.The liquid product from the high-pressure flash enters the

primary stripping column. The overhead product from the strip-ping column consists mainly of C2 components. This product isrecycled back to the high-pressure flash. The bottom product fromthe column consists mainly of C3�C4 components and gasoline.This product enters the primary stabilizer (sometimes called adebutanizer), which separates most of the C3�C4 componentsinto the overhead liquid. The stabilized gasoline (containing aregulated amount of C4) leaves as the bottom product.Some FCCgas plants further separate the gasoline product leaving

the stabilizer into heavy and light gasoline. We do not includeadditional gasoline splitting in this work. In addition, most plantscontain a water wash or injection system to control the presence ofacidic compounds that lead to corrosion.Thiswater injection typicallyoccurs between the stages of the overhead wet gas compressor. Mostof this water leaves the process flow before entering the columns ofthe gas plant. Thiswaterwash has little effect on the overall simulationof the process; therefore, we do not include it in this work.

3. PROCESS CHEMISTRY

The feed to the FCC unit is a complexmixture consisting of long-chain paraffins, single- and multiple-ring cycloalkanes, and largearomatic compounds. It is impossible to list every reaction that eachindividual molecule undergoes in the FCC riser. However, we canplace each of the reactions into five different classes based on the typeof reactants andproducts, effect on catalyst activity, and contributionsto product slate. In general, catalytic cracking occurs through theformation of a carbocation (from the feed hydrocarbonmolecule) inconjunction with a catalyst acid site. This carbocation may thenundergo cracking (to produce smaller molecules), isomerization (to

Figure 1. General schematic of a typical FCC reactor�regenerator unit.

http://pubs.acs.org/action/showImage?doi=10.1021/ef200750x&iName=master.img-000.jpg&w=194&h=267



Figure 2. Downstream fractionation (main fractionator).

Figure 3. FCC gas plant section.

http://pubs.acs.org/action/showImage?doi=10.1021/ef200750x&iName=master.img-001.png&w=459&h=382




rearrange molecules), and hydrogen transfer (to produce aromaticcompounds). Table 1 gives a simplified overview of key classes ofreactions and the general formulas for reactants and products.

Themost significant classes of reactions are cracking (reaction class1), isomerization (reaction class 2), and hydrogen transfer (reactionclass 3).1,6,7 The remaining classes are undesirable and contribute tohydrogen or coke production. The acid-catalyzed cracking reactionsfrom reaction class 1 form the primary pathway for light gas and LPG(C3�C4) components and the long-chain paraffin components ofdiesel. These reactions also provide some of the lighter aromaticcomponents present in the products. When catalytic conditionsare not present (e.g., contaminated/occluded catalyst or hightemperatures), a thermal cracking process takes over, promotinglower-order cracking reactions. These cracking reactions tend toproduce very large amounts of dry gas components (C1 andC2) andresult in higher coke production.1,6 In addition, excessive thermalcracking is not an economically attractive operating scenario.

Isomerization reactions (reaction class 2) give an importantpathway for high-octane components in the gasoline. This classof reactions is critical for producing high-octane components inthe gasoline products. In addition, we find more valuableisobutene components because of the isomerization of butanes.The isoparaffins from the isomerization class of reactions alsoreduce the cloud point of the diesel product.1

Hydrogen-transfer reactions (reaction class 3) form a class ofreactions that improves gasoline yield and stability (by lowering theolefin content) but also lowers the overall octane rating of the product.These reactions produce paraffins and aromatics that have low octaneratings. In addition, we cannot recover the olefins consumed byhydrogen-transfer reactions in the LPG or the light ends of gasoline.8

Dehydrogenation (reaction class 4) is a result of the presence ofmetals, such as nickel and vanadium, on the catalyst. Themetal sites onthe catalyst promote dehydrogenation and dealkylation. These reac-tions tend to produce large amounts of H2 and paraffin components

with low octane ratings. The coking process follows a complicatedseries of reactions that include olefin polymerization and aromatic ringcondensation (reaction class 5). The coking reactions dominate whenthe unit is operating at a non-optimal temperature (typically less than850 �F or 454 �C or greater than 1050 �F or 566 �C) or when feedcontains significant amounts of residue, recycled coke, or olefins.8

4. LITERATURE REVIEW

We can divide the literature on FCC modeling into twocategories: kinetic and unit-level models. Kinetic models focus onchemical reactions taking place within the riser or reactor section ofthe FCC unit and attempt to quantify the feed as a mixture ofchemical entities to describe the rate of reaction from one chemicalentity to another. In contrast, unit-level models contain severalsubmodels to take into account the integrated nature of modernFCC units. A basic unit-level model contains submodels for theriser/reactor, regenerator, and catalyst-transfer sections. The riserrequires a kinetic model to describe the conversion of chemicalentities. The regenerator contains another kinetic model to describethe process of coke removal from the catalyst. The unit-level modelalso captures the heat balance between the riser and the regenerator.

Table 1. Key Classes of Reactions with General Formulas for Products and Reactants

description general reaction formula for reactants and products

Reaction Class 1: Cracking

paraffin cracked to olefins and smaller paraffins Cm+nH2[(m+n)+2] f CmH2m+2 + CnH2n+2

olefins cracked to smaller olefins C(m+n)H2(m+n) f CmH2m + CnH2n

aromatic side-chain scission Ar�C(m+n)H2(m+n)+1 f Ar�CmH2m�1 + CnH2n+2

naphthenes (cycloparaffins) cracked to

olefins and smaller naphthenes

C(m+n)H2(m+n) (naphthene) f CmH2m (naphthene) + CnH2n (olefin)

Reaction Class 2: Isomerization

olefin bond shift x-CnH2n f y-CnH2n (x and y are different locations of the olefin)

normal olefin to iso-olefin n-CnH2n f i-CnH2n

normal paraffins to isoparaffin n-CnH2n+2 f i-CnH2n+2

cyclohexane to cyclopentane C6H12 (naphthene) f C5H9�CH3 (naphthene)

Reaction Class 3: Hydrogen Transfer

paraffins and olefins converted to

aromatics and paraffins

CnH2n (naphthene) + CmH2m (olefin) f ArCxH2x+1 (aromatic)

+ CpH2p+2 (paraffin) (where x = m + n � 6 � p)

Reaction Class 4: Dehydrogenation and Dealkylation (Contaminated Catalyst)

metals catalyzed aromatic and light

hydrocarbon production

i-CnH2n-1 + CmHm�1 f Ar + C(n+m�6)H2(n+m�6)

n-C2H2n+2 f CnH2n + H2

Reaction Class 5: Aromatic Ring Condensation

condensation of single aromatic cores to produce multiple-ring aromatic cores Ar�CHCH2 + R1CH�CHR2 f Ar � Ar + H2

Figure 4. Lumped model from Takatsuka et al.9 VR, vacuum resid;CSO, coke slurry oil; HCO, heavy cycle oil; and LCO, light cycle oil.




4.1. Kinetic Models.We classify kinetic models according to thechemical entities that makeup the model. Typically, the entities or“lumps” are boiling point lumps or yield lumps, grouped chemicallumps, and full chemical lumps. Early kinetic models consist entirelyof yield lumps, which represent the products that the refiner collectsfrom themain fractionator following the FCCunit. Figure 4 shows atypical kinetic model based on yield lumps by Takatsuka et al.9

Many similar models have appeared in the literature. The modelsdifferentiate themselves based on their number of lumps. Modelsmay contain as few as 210 or 3 lumps11 and as many as 50 lumps.12

We note thatmodels withmore lumps do not necessarily havemorepredictive capabilities than models with fewer lumps.6

The next class of kinetic models considers both chemical-typelumps and boiling point or yield lumps. For example, Jacob et al.13

present a popular 10-lumpmodel (shown in Figure 5) that includescoke and light ends (C), gasoline (G, C5 to 221 �C), light paraffin(Pl), heavy paraffin (Ph), light naphthene (Nl), heavy naphthene(Nh), light aromatics (Al), heavy aromatics (Ah), light aromaticswith side chains (CAl), and heavy aromatics with side chains (CAh).The “l” subscript refers to “light” lumps in the boiling point rangebetween 221 and 343 �C,whereas the “h” subscript refers to “heavy”lumps that have boiling points above 343 �C.The key advantage of this lumped kinetic model is that the

composition of lumps can be measured with various experimentaltechniques. In addition, the rate constants that arise from the use ofthis model are less sensitive to changes in feed and processconditions.14 This model has served as the basis for models thatinclude more chemical types. Pitault et al.15,16 have developed a 19-lump model that includes several olefin lumps. AspenTech17,18 hasdeveloped a 21-lump model to address heavier and more aromaticfeeds, which we will use to model the reaction section of the FCCunit. We discuss this 21-lump model in a subsequent section.

Hsu et al.6 state that “lumped kinetic models developed by thetop-down route have limited extrapolative power”. To remedy thissituation,many researchers have developed complex reaction schemesbased on basic chemical principles that involve thousands of chemicalspecies. We can classify them into mechanistic models and pathwaymodels. Mechanistic models track the chemical intermediates, such asions and free radicals, that occur in the catalytic FCC process.Transition-state theoryhelps inquantifying the rate constants involvedin adsorption, reaction, anddesorptionof reactant andproduct speciesfrom the catalyst surface. Froment and co-workers19 have pioneeredthe use of such models in a refinery context and have developed amodel for catalytic cracking of VGO. Hsu et al.6 claim that using thismethod is challenging because of its large size and reaction complexity.Structure-oriented lumping (SOL) is a leading example of the

pathway-based models. Quann and Jaffe20�22 have developed aunique method for tracking molecules in the feed oil. Themethod tracks different compositional and structural attributesof a molecule (number of aromatic rings, number of nitrogensubstituents, sulfur substituents, etc.) in a vector format. Figure 6shows typical vectors for some sample molecules.After these vectors are developed for the feed oil, several rules are

used to generate reaction paths that convert the feed vectors toproduct vectors. The rate constants and activation energies for thesereactions are functions of the reaction type and the feed oil composi-tion vector. Christensen et al.23 discuss applying the SOL method todevelop a FCC kinetic model, which contains over 30 000 chemicalreactions and 3000 molecular species. The resulting model canaccurately predict product yields, composition, and quality over awide range of operating conditions. Klein and co-workers24 have alsodeveloped similar models for FCC and catalytic reforming.Figure 7 compares these kineticmodels on the basis of complexity

andmodel fidelity. The yield lumpmodels have the lowest complex-ity and require the least amount of data. Typically, the feed may be

Figure 5. Ten lump model from Jacob et al.13

Figure 6. Typical SOL vectors (from ref 20).

Figure 7. Summary of kinetic models.






treated as a single lump, and there are few reaction rates to calibrate.Chemical lumps require knowledge of the chemical type of the lump,namely, the paraffin, naphthene, and aromatic (PNA) contents ofeach boiling point range. Pathway andmechanisticmodels require thedetailed analysis of the feed data to developmolecular representation.Additionally, pathway and mechanistic models require more data tocalibrate the numerous kinetic parameters.6

4.2. Unit-Level Models. Table 2 compares a selection ofpublished work (after 1985) regarding modeling of an entire FCCunit. This table does not include work that only compares theperformance of the riser with experimental or plant data. It includesworkwhere the authors compare thepredictionsof the entireFCCunitmodel to published data, experimental data, or plant data. Thework byLee et al.,10 McFarlane et al.,3 and Arbel et al.2 provide the basis formany dynamic and process control-related models by later authors.These studies focus on optimal control strategies and the dynamicresponse of the FCC unit. There are few papers that compare thesteady-stateoperationof theFCCunitwithdetailedpredictionsof yieldand product properties with data. Notably, the work by Fernandeset al.33 follows an industrial FCC unit over the course of 3 years andgives good predictions of the performance of the unit. However, thiswork does not include any detailed predictions of product quality andcomposition. Additional work by Fernandes et al.35 shows how feedand operating conditions, such as coke composition, catalyst/oil ratio,Conradsoncarbon residue (CCR) in feed, air/oil ratio, and regeneratorcombustionmodes, can inducemultiple steady states with implicationsfor a general unit control strategy.A complete unit-level model for a FCC unit includes several

submodels of varying degrees of rigor. A modern FCC unit

involves complex kinetic, heat management, and hydrodynamicissues. Necessarily, researchers develop models that focus onparticular aspects of FCC operation. There is significantresearch36 on the topic of complex hydrodynamics in the riserand regenerator sections using computational fluid dynamics(CFD). These models often require detailed information aboutthe process that is proprietary. The focus of this paper isdeveloping a model to predict key process output variables, suchas product yields, product properties, and operating profiles ofthe FCC unit and associated gas plant. We acknowledge that thehydrodynamics and complex kinetics have significant effects onthese out variables.1 However, our goal is to develop amodel thatengineers can use and modify based on limited process data.Arandes et al.37 and Han et al.38 summarize the key submodels

required for a unit-level model that can provide necessarysimulation fidelity for this work. We briefly summarize thesesubmodels in Table 3 and refer readers to these two papers fordetailed equations and additional references.Modern FCCunits and catalyst have very high conversions in the

riser section. The conversion of feed species to product speciescompletes within the riser; therefore, we require no additionalsections for feed conversion. There are units where feed conversionmay occur in locations other than the riser,39,40 but we have chosento limit our discussion to the most common type of unit.

5. ASPEN HYSYS/PETROLEUM REFINING FCC MODEL

The Aspen HYSYS/Petroleum Refining FCC model relies ona series of submodels that can simulate an entire operating unit

Table 2. Survey of Related Published Literature for Integrated FCC Modelinga

reference application kinetics

property

predictions

fractionation

modeling validation data

integration with

production planning

10 dynamic/process control 3 lumps none none none none

3 dynamic/process control 2 lumps none none none none

2 dynamic/process control 10 lumps none none literature none

5 dynamic/process control 3 lumps none none literature none

25 steady state 10 lumps none none literature none

4 dynamic/online optimization 4 lumps none none literature none

26 dynamic/process control 10 lumps light gas

composition (C1�C4)

and RON/MON

of gasoline products

none literature none

27 dynamic 10 lumps none none industrial (dynamic) none

28 steady state/online

optimization

NA extensive properties of

all key products

none industrial, pilot plant,

and experimental

29 steady state 3 lumps none none industrial none

30 steady state 11 lumps none none industrial none

31 steady state 6 lumps light gas composition none industrial none

32 dynamic/pilot-plant

process control

2 lumps none none pilot plant none

33 steady state/dynamic 6 lumps none none industrial none

34 steady state 4 lumps none none pilot plant none

this work steady state 21 lumps light gas composition,

flash point, density

of key products,

and RON/MON

main fractionator

and associated gas plant

industrial export model to

LP-based planning tool

aRON/MON = research octane number/motor octane number.



while satisfying the riser and regenerator heat balance (Figure 8).Note that the configuration is similar to the minimum submodelslisted in Table 3 of the previous section. We summarize AspenHYSYS/Petroleum Refining submodels in Table 4 and highlightsome key features in sections 5.1�5.3.5.1. Slip Factor and Average Voidage. An important con-

cern in FCC riser submodels is how to calculate the slip factor,j,and the average voidage, ε, of the riser. The slip factor is simplydefined as the ratio between the gas velocity and catalyst particlevelocity. The slip factor plays an important part in determiningthe residence time of reactions and, thus, affects the overallconversion in the riser. Harriot describes a slip factor range of1.2�4.0 for most FCC risers but also indicates that there is noreliable correlation available for prediction.44 Previous authorshave used a variety of approaches, including constant slipfactor,45 multiple slip factors,46 and correlations.47 An alternativeapproach is to include additional momentum balance equationsfor the gas and catalyst phases.48 This approach allows users tocalculate velocity profiles for each phase and the overall pressuredrop in the riser directly.

Aspen HYSYS uses a custom correlation based on fullydeveloped flow (away from the catalyst particle accelerationzone) that accounts for various angles of riser inclination. Wepresent a similar correlation from Bolkan-Kenny et al.47 in eq 1using dimensionless Froude numbers (eqs 2 and 3) Thiscorrelation is essentially a function of the riser diameter, D,acceleration due to gravity g, superficial gas velocity, uo, andterminal settling velocity of the catalyst particle, ut.

j ¼ 1 þ 5:6Fr

þ 0:47Frt0:41 ð1Þ

Fr ¼ uoffiffiffiffiffiffigD

p ð2Þ

Frt ¼ utffiffiffiffiffiffigD

p ð3Þ

5.2. 21-Lump Kinetic Model. The 21-lump kinetic model inAspenHYSYS/PetroleumRefining is similar to the popular 10-lump

Table 3. Required Submodels for a Basic Simulation of a Complete FCC Unit

submodel purpose unit operation

riser reactor crack feed species to product species

plug-flow reactor (PFR) operating under pseudo-steady conditions

catalyst activity decay because of coke formation as result of time on stream,

coke on catalyst, and catalyst type

stripper remove adsorbed hydrocarbons on the catalyst continuously stirred tank reactor (CSTR) with well-mixed model

regenerator combust coke present on the catalyststoichiometric or partial combustion of coke

bubbling bed reactor with a dense phase and a dilute phase

feed vaporizer vaporize the feed species for input into the riser model heater with associated two-phase flash

valves control the flow and pressure drop from the riser/reactor

section to the regenerator section

typical valve equations based on pressure drop across the valve

cyclones separate solids from the hydrocarbon and effluent vapors simple component splitter

Figure 8. Overview of the major submodels that make up the Aspen HYSYS/Petroleum Refining FCC model (Adapted from ref 6).




model from Jacob et al.13 (Figure 5). The21-lumpmodel follows thesame basic structure and pathways as the 10-lump model bygrouping lumps into boiling point ranges and chemical types withineach boiling point range. In addition, the 21-lump model includes aboiling point range to deal with heavy feeds (boiling point greaterthan 510 �C) that the original 10-lump model cannot handle. Toaccount for the differences in reactivity of various aromatic com-pounds, aromatic lumps are further split into lumps containing sidechains and multiple rings separately. The 21-lump model also splitsthe original single lump for coke into two separate coke lumps.These separate lumps account for coke produced from crackingreactions (called kinetic coke) and coke produced from metalactivity (called metal coke) individually. We note that the rateequations in the kinetic network in Aspen HYSYS/PetroleumRefining are largely similar to equations in the first-order networkfor the 10-lump model. However, the rate equations in the 21-lumpmodel include additional terms to account for the adsorption of theheavy hydrocarbons (because of the extended boiling point range ofthe lumps) and the metal activity of the catalyst. Table 5 lists thekinetic lumps used in the 21-lump model.We can obtain the lump composition of the feed stock directly

via gas chromatography/mass spectrometry (GC/MS), 1H and13C nuclear magnetic resonance (NMR), high-performanceliquid chromatography (HPLC), and American Society for Test-ing and Materials (ASTM) methods. However, this is infeasibleon a regular basis for refineries given the changing nature of thefeedstock. Aspen HYSYS/Petroleum Refining includes a methodthat uses existing feed analysis to infer feed composition usingroutinely collected data. However, we have developed an alter-native scheme to infer feed composition. We detail this methodin section 8.5.3. Catalyst Deactivation. Another important consideration

in the FCC unit model is the deactivation of the catalyst as itcirculates through the unit. Previous work has used two differentapproaches to model catalyst activity: time on stream and cokeon catalyst.49 Because the 21-lump model includes discretelumps for the kinetic and metal coke, this work uses a coke-on-

catalyst approach to model catalyst deactivation. In addition, thiswork includes a rate equation in the kinetic network for cokebalance on the catalyst. The general deactivation functionbecause of coke, ϕCOKE, is given by eq 4

ϕCOKE ¼ ϕKCOKEϕMCOKE

¼ expð�aKCOKECKCOKEÞexpð�aMCOKECMCOKEf ðCMETALSÞÞð4Þ

where aKCOKE is the activity factor for kinetic coke, aMCOKE is theactivity factor for metal coke, CKCOKE is the concentration ofkinetic coke on the catalyst, aMCOKE is the activity factor formetal coke, CMCOKE is the concentration of metal coke on thecatalyst, and CMETALS represents the concentration of metals onthe catalyst.

6. CALIBRATING THE ASPEN HYSYS/PETROLEUM RE-FINING FCC MODEL

Given the variety of feedstock that the FCC unit processes, itunlikely that a single set of kinetic parameters will provideaccurate and industrially useful yield and property predictions.In addition, changes in the catalyst may significantly alter theyield distribution. Therefore, it is necessary to calibrate themodelto a base scenario. Table 6 lists the key calibration parameters forthe FCC model. We group them by their effects on the modelpredictions.

Aspen HYSYS/Petroleum Refining includes a base set ofkinetic and calibration parameters regressed for a variety of feedoils and catalyst types.We use these as a starting point to calibratethe model to our specific operating scenario. Because of thechemical nature of the feed lumping, the calibration processresults in only small changes in the values of calibration para-meters. Significant changes from the base values may result in“overcalibration” and fix the model to a particular operatingpoint. An “overcalibrated” model gives poor predictions evenwhen we make small changes to input variables. It is critical to

Table 4. Summary of Aspen HYSYS/Petroleum Refining FCC Submodels (Adapted from Ref 6)

submodel purpose unit operation considerations

riser (more than one

can be present)

convert feed to product species

using 21 lump kineticsmodified PFR

allows any angle of inclination

pressure drop is a combination of the pressure drop

because of solid and vapor phases

catalyst activity decay because of kinetic and metal coke

on the catalyst

slip factor correlations (difference between vapor and

solid velocities) to estimate species density

reactor/stripper complete feed conversion and remove

adsorbed hydrocarbons

bubbling-bed reactor

with two phases

switches to fluidized-bed reactor model for units with

low catalyst holdup

regenerator combust coke present on the catalyst bubbling-bed reactor

with two phases

kinetic models for coke combustion with air and

enriching oxygen41

regenerator

freeboard

complete combustion of coke simple plug-flow

reactor

additional kinetics to match the behavior of

industrial units42

cyclones separate solids from hydrocarbon

and effluent vapors

two-phase, pressure

drop calculation

pressure drop is a combination of the pressure drop

because of solid and vapor phases

delumper

convert lumped composition into a set of true

boiling point (TBP) pseudo-components

suitable for fractionation

carries chemical information about the kinetic lumps as

an attribute of the pseudo-component

additional delumping of light gas into C1�C4

components using known kinetics43



keep track of these changes in the calibration factors and makesure that they are reasonable. The key steps in the calibrationprocess are as follows: (1) Obtain a base or reference set ofoperating data that fully defines the operation of the FCC unitand associated product yields. Table 12 lists the relevant dataused for calibration in this work. (2) Use experimentally mea-sured chemical composition of liquid products (or estimate usingthe methods given in section 8) to calculate the expected effluentcomposition of kinetic lumps from the FCC unit. (3) Vary thereaction selectivities for reaction paths (three parameters) thatlead to coke lumps (kinetic coke and metal coke), gasoline(G lump), and VGO (PH, NH, AHs, AHr1, AHr2, and AHr2lumps), deactivation activity factors (two parameters), and cokeburn activity (one parameter), so that model predictions forkinetic lump compositions agree with measured (or estimated)kinetic lump compositions from step 2. (4) Vary the distributionselectivities (minimum of two parameters, ratio between C1 andC2 and ratio between C3 and C4) for light gases to match thetotal measured light gas composition from the dry gas and LPGstream of the refinery. (5) Once calibration is complete, verifythat overall material and energy balances hold.

In Aspen HYSYS, we can modify the parameters in steps 3 and4 concurrently to simplify the calibration process. We note that, if

the initial kinetic parameters have been regressed from amultiplevariety of sources, small adjustments to calibration parametersare enough to match typical plant operation. In our work, therange of calibration parameters is roughly on the order of0.5�1.5 times the initial calibration parameter values.

7. FRACTIONATION

The fractionation sections use standard inside-out methods50

implemented by many popular simulators, including AspenHYSYS. This method offers robust convergence and wideflexibility in specifications. The key issue in implementingfractionation models is whether to use stage efficiencies. Readersshould be careful to avoid confusion with a related concept:overall efficiency. Overall efficiency refers to the ratio of theore-tical stages used in simulations to physical stages in the actualcolumn. For example, consider the case where we model adistillation column having 20 physical stages, with the simulatorusing only 10 theoretical stages. This column has an overallefficiency of 10/20 = 0.50. Note that each stage in the simulationoperates under valid thermodynamic vapor�liquid equilibriumassumptions.

Alternatively, many simulators offer stage efficiency modelsunder the name Murphree stage efficiency, given by E in thefollowing equation:

E ¼ yn � ynþ1

y�n � ynþ1or

xn � xnþ1

x�n � xnþ1ð5Þ

where xn represents the mole fraction of a given component inthe liquid leaving stage n, xn + 1 represents the mole fraction of agiven component in the liquid leaving stage n + 1. yn and yn + 1

refer to the vapor mole fraction of a given component leaving asvapor from stages n and n + 1, respectively.

The E factor violates vapor�liquid equilibrium constraintsand can predict unusual and unphysical solutions for stage-by-stage simulation models. Both Kister50 and Kaes51 advise againstthe use of the stage efficiencymodels. They warn that simulationsusing these factors may lose predictive abilities and may notconverge robustly. In our work, we use the rigorous stage-by-stage models for all fractionators with the overall efficiencyconcept. Kaes51 has documented the relevant overall efficienciesthat are reasonable for modeling columns in the FCC gas plant.Table 7 shows the number of theoretical stages and efficienciesfor FCC fractionation. We obtain the overall efficiency as theratio of the number of theoretical stages to actual physical stagesin the column. For example, the main fractionator columntypically has 30�40 physical stages, and we find that 12�16theoretical stages are sufficient for modeling purposes. Hence,the overall efficiency ranges from about 40 to 50%. We calculateoverall efficiencies for other columns given in Table 7 using a

Table 5. Summary of 21-Lump-Kinetics (Adapted fromRef 6)

boiling

point range lumps

<C5 light gas lump

C5 to 221 �C gasoline

221�343 �C(VGO)

light paraffin (PL)

light naphthene (NL)

light aromatics with side chains (ALs)

one-ring light aromatics (ALr1)

two-ring heavy aromatics (ALr2)

343�510 �C(heavy VGO)

heavy paraffin (PH)

heavy naphthene (NH)

heavy aromatics with side chains (AHs)

one-ring heavy aromatics (AHr1)

two-ring heavy aromatics (AHr2)

three-ring heavy aromatics (AHr3)

510+ �C(residue)

residue paraffin (PR)

residue naphthene (NR)

residue aromatics with side chains (ARs)

one-ring residue aromatics (ARr1)

two-ring residue aromatics (ARr2)

three-ring residue aromatics (ARr3)

cokekinetic coke (produced by reaction scheme)

metal coke (produced by metal activity on the catalyst)

Table 6. Key Calibration Parameters for the FCC Model

parameter class calibration parameters

overall reaction selectivity

selectivity to C (coke lump)

selectivity to G (gasoline lump)

selectivity to L (VGO lump)

distribution of light gas components (C1�C4) selectivities to C1�C4 light gases

deactivation factors accounting for the metal content and activity of the equilibrium catalyst (ECAT)

equipment and process conditions activity for CO/CO2 generation from coke combustion in the regenerator



typical range for the number of physical stages from variousprocess design data.

We can actually develop the initial model for the fractionationwithout connecting it to the FCC model. Here, we follow theprocess of “back-blending”, as shown in Figure 9, to recover thereactor effluent (or fractionator feed) from a known set ofproduct yield data.51 This process requires that we know theyields and composition of all of the key products from the FCCplant, the feed rate to the reactor, and additional inputs (such assteam) to the reactor. We then use the composition data of thelight products and the distillation curves of the liquid products toreconstruct a reactor effluent as the fractionator feed. We feedthis effluent into the initial fractionation model and recover theproducts that are “back-blended”. There are two advantages tothis process. First, we can verify that the fractionation modelaccurately reflects plant operation. We verify the fractionationmodel through accurate predictions of product yields, goodoverlap between plant and model distillation curves of liquidproducts, agreement of plant and model gas compositions (drygas and LPG), and small deviations between the temperatureprofiles of plant and model columns. Second, this process canshorten the model development time because we can work onmodeling the FCC unit and the fractionation units at thesame time.

In this work, calibrating the fractionation section refers to theprocess of adjusting the number of theoretical stages in each zone(in the case of themain fractionator) or the number of theoreticalstages between feed points. We use a set of basic initializationspecifications and efficiencies given in Table 7 to solve thecolumn models. Typically, we only need to add or remove a few

stages to calibrate the columns and achieve agreement with theplant operating profile. Once we converge the column modelsusing the basic initialization specifications, we change (especiallyfor the main fractionator) to specifications based on cut point andstage temperature. Kaes51 describes a similar process. We sum-marize the initial and final specifications in Table 8.

8. MAPPING FEED INFORMATION TO KINETIC LUMPS

Aspen HYSYS/Petroleum Refining includes a method toconvert limited feed information (distillation curve, density,viscosity, refractive index, etc.) into kinetic lumps for use in theunit-level FCC model. In this section, we present an alternativemethod based on data andmethods available in the literature.Weextend the method based on work by Bollas et al.52 to infer thekinetic lump composition from limited process data. Thismethod uses techniques to normalize the distillation curve, cutthe distillation curve into boiling point lumps, and infer thecomposition of the each of these boiling point lumps. We havedeveloped all of these techniques into spreadsheets using Micro-soft Excel. These spreadsheets are available for the interestedreader to download without charge on our group website (www.design.che.vt.edu).8.1. Fitting Distillation Curves. Distillation curves for FCC

feedstock can be limited. Because of the nature of the feedstock,complete true boiling point (TBP) analysis without ASTMD2887/SimDist methods is frequently not possible. Manyrefiners still use a limited ASTM D1160 distillation method toobtain some information about the distillation curve. Table 9shows a typical ASTM D1160 analysis for a heavy FCCfeedstock.This D1160 curve does not contain enough information to

convert it into a TBP curve using standard ASTM correlations.We must fit these data to a reasonable model to obtain estimatesfor the missing data points. Sanchez et al.53 have evaluatedseveral different types of cumulative probability distributionfunctions to fit distillation curves of crude oils and petroleumproducts. They conclude that the cumulative β function (withfour parameters) can represent a wide range of petroleumproducts.53 We use this method to extend the measured partialdistillation curve.

Table 7. Theoretical Stages and Efficiency Factors for FCCFractionation

fractionator theoretical stages overall efficiency (%)

main fractionator 12�16 40�50

primary absorber 6�10 20�30

primary stripper 12�15 40�50

secondary absorber 3�6 20�25

gasoline stabilizer 25�30 75�80

LPG (C3/C4) splitter 25�30 75�80

Figure 9. “Back-blending” products to reconstitute FCC reactor effluent.




The β cumulative density function is defined as

f ðx,R, β,A,BÞ ¼Z x e B

A

1B� A

� �ΓðR þ βÞΓðRÞΓðβÞ

x� AB� A

� �R � 1 B� xB� A

� �β � 1

ð6Þwhere R and β refer to the positive-valued parameters thatcontrol the shape of the distribution, Γ refers to the standard Γfunction, A and B parameters set lower and upper bounds onthe distribution, respectively, and x represents normalizedrecovery.We normalize all of the temperatures between 0 and 1 using

the following equation:

θi ¼ Ti � T0

T1 � T0ð7Þ

where T0 and T1 are reference temperatures. For this work, wechoose T0 = 250 �C and T1 = 650 �C. Then, we apply thecumulativeβ functionwith each normalized recovery, xi, and initialvalues for R, β, A, and B parameters. If we choose good estimatesfor parameters, then the output of the β function must be close tothe corresponding recovery for each xi. We define the followingerror terms:

RSS ¼ ∑n

i¼ 1ðxexp, i � xiÞ2 ð8Þ

AAD ¼ 1n ∑

n

i¼ 1absðxexp, i � xiÞ ð9Þ

where xexp,i represents the recovery measured in the distillationcurve and xi is the output of theβ function. RSS is the sum of least-squares residuals or errors, and AAD represents the averageabsolution deviation. We now use the SOLVER method inMicrosoft Excel to obtain optimized values of R, β, A, and B.Figure 10 shows how this fit compares to the result using a log-

normal distribution53 (with two fitting parameters), instead ofthe β function. Using the β function, we can generate thetemperatures and recoveries needed for the conversion to TBPusing standard ASTM methods.8.2. InferringMolecular Composition.Asmentioned earlier, we

must also be able to infer the PNA composition of each boiling pointrange, given certainmeasured bulk properties to completelymap feedinformation to kinetic lumps. The American Petroleum Institute

(API) (Riazi�Daubert)54,55 is a popular chemical compositioncorrelation that takes the form

% XP or % XN or % XA ¼ a þ bRi þ cVGC0 ð10ÞwhereXP andXN represent themole compositionof paraffins (P) andnaphthenes (N), respectively, Ri is the refractive index, and VGC0 isthe either the viscosity gravity constant (VGC) or viscosity gravityfactor (VGF). The parameters a, b, and c take on different values foreach molecule type. Using the Riazi55 correlation does not givesufficiently accurate predictions for molecular compositions for thiswork. We note that this correlation encompasses a wide molecular-weight range of 200�600.55

We present an alternate correlation in eqs 11 and 12. Ourcorrelation extends the original correlation from Riazi54,55 byincluding specific gravity (SG) as an additional parameter andproviding different sets of correlation coefficients (a, b, c, and d)for different boiling point ranges

% XP or % XA ¼ a þ bSG þ cRi þ dVGC0 ð11Þ

% XN ¼ 1� ðXP þ XAÞ ð12Þwhere XP, XN, and XA represent the mole composition ofparaffins (P), naphthenes (N), and aromatics (A), respectively,Ri is the refractive index, and VGC0 is the either the VGC or VGF.The parameters a, b, c, and d can take on different values fordifferent molecule types and boiling point ranges.We used a total of 233 different data points containing

laboratory-measured chemical composition and bulk propertyinformation (distillation curve, density, refractive index, andviscosity) for light naphtha, heavy naphtha, kerosene, diesel,and VGO. These data points come from various plant measure-ments made over the 6month course of this study and a variety oflight and heavy crude assay data (spanning several years)available to the refinery.

Table 8. Initialization and Final Specifications

column initial specifications final specifications

main fractionator

all pumparound rates and return

temperatures (or temperature changes)

column overhead temperature

cut point for naphtha draws

draw rates for all products pumparound duties

bottom temperature bottoms temperature

condenser temperature condenser temperature

primary absorber none none

primary stripper none none

secondary absorber none none

gasoline stabilizerreflux ratio (around 2.0)

overhead draw rate

gasoline n-butane fraction or Reid vapor pressure (RVP) in bottoms

column overhead temperature or C5+ content in overhead

LPG stabilizerreflux ratio (around 3.0)

overhead draw rate

reboiler temperature or bottom temperature

fraction C4 in the column overhead

Table 9. Typical Distillation Curve Collected from ASTMD1160

recovery temperature (�C)

0 (initial point) 253

10 355

50 453

73 (end point) 600



We use Microsoft Excel and the SOLVER method to fit valuesfor the parameters a, b, c, and d that minimized the sum of squaresresidual between the measured % XP and % XA and calculated %XP and % XA.We calculate % XN by difference, as shown in eq 12.We show the results of our data regression with the associatedAAD in Tables 10 and 11. Figures 11�13 comparemeasured andcalculated molecular compositions.We can now use the two methods that we have developed to

propose a technique to use limited feed information to inferlumped composition. This technique is similar to the one givenby Bollas et al.52 However, we make several changes to accountfor limited data sets. We outline the technique in the followingsteps (changes from the procedure by Bollas et al.52 are indicatedwith an asterisk): (1) Use the β distribution method to extendpartial ASTM D1160 distillation curves.* (2) Convert ASTMD1160 to a TBP curve using standard API correlations54 (notethat we offer a spreadsheet to perform the standard correlations).* (3) Using the 50% point of the TBP, estimate theWatson factor(KW). Set the 50% TBP temperature as an initial guess for themean-average boiling point (MeABP). (4) Use the definition ofKW to create the specific gravity distribution of the fraction. (5)Calculate pseudo-component molecular weight using the cor-relation by Riazi.55 (6) Use densities and mole weights tocalculate volume-, cubic-, molar-, and mean-average boilingpoint of the total fraction.55 (7) If the MeABP from step 7 isclose to the MeABP assumed in step 3, go to step 8. Otherwise,assume a new value for MeABP and go back to step 4. (8)Assign a lump to every boiling point range in the kineticlumping.* (9) Calculate the boiling point, molecular weight,density, volume, and weight and molar concentrations of eachlump. (10) Use Goosen’s correlation to estimate the refractiveindex of each lump.56 (11) Use correlations from Riazi55 toestimate the viscosity of the lump.* (12) Calculate the relevantVGF or VGC55 for the lump.* (13) Use correlations (with anappropriate choice for the set of correlation coefficients)proposed in the preceding section to identify the PNA com-position of the lump.* (14) If required, use correlations fromRiazi55 to estimate the number of aromatic rings in eacharomatic fraction.*We have found that this technique can provide reasonable

estimates of kinetic lump composition. It is difficult to justify amore sophisticated scheme given the limited amount of dataavailable. Some refiners also make bulk chemical compositionmeasurements of the feed, which includes a measurement of the

total aromatic content. The sum of the aromatic kinetic lumpsgenerated from the above technique generally agrees with themeasured aromatic content.8.3. Convert Kinetic Lumps to Fractionation Lumps. A

related problem is the conversion of kinetic lumps back tofractionation lumps required to build rigorous fractionationmodels. For our models, Aspen HYSYS gives a method totransition the kinetic lumps to boiling-point-based pseudo-components typically used to model petroleum fractionation.We also propose an alternative technique that can provide similarresults using methods developed earlier in this section. Essen-tially, we must convert the kinetic lumps back into a TBP curve.The key steps in converting the kinetic lumps to boiling pointpseudo-components are as follows: (1) Using the “back-blending”concept from the previous section, develop a FCC effluent TBPcurve from a reference set of product yields. These yields includeall liquid products, such as light and heavy naphtha, light andheavy cycle oil or diesel, slurry, or decant oil. (2) Fit a cumulativeβ distribution to this “back-blended” reference TBP curve andobtain the best values for the cumulative β distribution fit. Wecalculate this initial set of parameters only once. (3) Run themodel to obtain the product distribution in terms of kineticlumps. (4) Apply steps 3�13 from the kinetic lumping proce-dure developed in section 8.2 in reverse; i.e., we obtain the 50%TBP point for each boiling point range from the known PNAdistribution of the kinetic lumps involved. (5) Because we knowinitial and final boiling points for all of the kinetic lumps (bydefinition), use these points in conjunction with calculated 50%TBP points to generate an updated FCC effluent TBP curve. (6)Fit a new cumulative β distribution to the updated FCC effluentTBP curve using the initial set of cumulative distribution para-meters as a starting guess. (7) Cut this new TBP curve intopetroleum pseudo-components using methods commonlyavailable in process simulations. Riazi55 discusses severalstrategies to cut a TBP curve into pseudo-components sui-table for fractionation models.

Figure 10. Comparison between using the β distribution and log-normal distribution to fit the same distillation data.

Table 10. Coefficients for Paraffin Content in PetroleumFractions

paraffin (vol %)

a b c d AAD

light naphtha 311.146 �771.335 230.841 66.462 2.63

heavy naphtha 364.311 �829.319 278.982 15.137 4.96

kerosene 543.314 �1560.493 486.345 257.665 3.68

diesel 274.530 �712.356 367.453 �14.736 4.01

VGO 237.773 �550.796 206.779 80.058 3.41

Table 11. Coefficients for Aromatic Content in PetroleumFractions

aromatic (vol %)

a b c d AAD

light naphtha �713.659 �32.391 693.799 1.822 0.51

heavy naphtha 118.612 �447.589 66.894 185.216 3.08

kerosene 400.103 �1500.360 313.252 515.396 1.96

diesel 228.590 �686.828 12.262 372.209 4.27

VGO �159.751 380.894 �150.907 11.439 2.70




9. OVERALL MODELING STRATEGY

This work relies primarily on data collected while the refineryis in regular operation. Related work in integrated FCCmodelingoften relies on pilot-plant and experimental data. It is moredifficult to produce a predictive model with plant operation dataalone. The nature of plant operation means that there may beabrupt changes in feed quality or operating parameters, poormeasurements because of poorly calibrated or failing sensors, andinconsistent data. Fernandes et al.33 have encountered similarissues in the validation phase of their work. We outline thefollowing strategy and our specific implementation in Figure 14:(1) Obtain data on a continuous basis from the plant over anumber of months. (a) Reconcile data from multiple sources(DCS, inventory, etc.) (Table 12). (b) Check the consistency ofthe data by ensuring mass balance and enthalpy balance. (c)Accept a data set when it is consistent. (d) Track variation in thedata set to ensure that there are multiple operating scenarios(Figure 15). (2) Use the first accepted data set to develop aninitial model for the FCC unit and fractionation section. (3)Calibration: (a) The most basic calibration is to introduce a

selectivity calibration factor for classes of the reactions in a kineticnetwork. (b) It is typically sufficient to vary the calibrationselectivity factors to match plant performance using the firstaccepted data set. (c) The user may introduce additional factorsto account for significant changes in the catalyst behavior of theunit. (d) The yield results from the initial model calibrationshould be within 1�2% of the actual plant yield. (4) Validation:(a) Use the subsequently accepted data sets to verify and trackthe performance of the unit and fractionation sections with themodel. (b) Make sure to examine the yield of the FCC unitindependently from the fractionation section. (c) It is typicallypossible to predict yields of key products on a feed-normalizedmass basis with AAD of less than 2�3%. (5) Case studies: (a)The model is calibrated with a finite amount of plant data;therefore, it may not be meaningful to study changing operatingparameters of the FCC over a very wide range. However, casestudies on the fractionation section can take on wide ranges.(b) Recalibrate the model when significant process changes occur.

10. RESULTS

We evaluate the model using over 6 months of operating datafrom a commercial FCCunit in the Asia Pacific with a feed capacityof 800 000 tons per year operating under a maximum diesel andgasoline plan. Figure 16 shows a process flow diagram for theentire process. The evaluation of the model includes comparisonsof overall reactor yield, light and heavy product composition, andoperating profiles for key equipment in the gas plant.We note that,in general, the model can accurately predict the product yield andcomposition over a variety of different feed conditions.

The most important prediction is the distribution of overallproduct yields from the FCC unit. A validated prediction of theoverall product yields allow for the refiner to use the model tostudy different kinds of feedstock and operating conditions.Table 13 shows the results for product yields. The mostimportant and valuable products are LPG, gasoline, and diesel.We use operating data from the BASE run to calibrate the model.In terms of the overall yield, the largest errors in the BASE caseappear with prediction of LPG and slurry. The AAD for theproduct overall validation cases (VALID-1�VALID-6) is 0.96%.The AAD is much lower than the previous AAD standard of 5%for yield predictions in the plant.

Figure 12. Comparison of calculated and measured naphthene contentin all fractions.

Figure 11. Comparison of calculated and measured paraffin content inall fractions.

Figure 13. Comparison of calculated andmeasured aromatic content inall fractions.






Another set of key indicators are the product properties of theliquid fuel from the FCC. The properties of interest to refinersare density, flash point (volatility), RON/MON (for gasoline),sulfur content, and aromatic content. This is one of the areaswhere our model is different from other published work de-scribed earlier. We discuss a method to transition from kineticlumping to fractionation lumping in section 8. Not only does thismethod allow the user to observe the results directly, we can alsosee the effect of the reactor conditions on fractionated properties.Using the results from the fractionator model, we can calculatethe distillation curves of the liquid products. Figures 17 and 18show the distillation curves for one of the validation cases. Ingeneral, the model predicts key points from the ASTM D86curve (5 and 95%) within plant tolerance. Further refinement ofthis prediction requires accurate measures of the pumparoundrates and the heat duty for each pumparound in the mainfractionator. These data are not routinely measured.

We can use the predicted ASTM D86 curves to calculateseveral other properties of interest. There are several methods tocalculate the flash point and other volatility properties in usingthe distillation curve and density. In Figure 21, we compare ourpredictions using the API flash point correlations54 to themeasure data. We note good agreement for the flash point. Inaddition, Figures 19 and 20 show the prediction of the densitiesfor gasoline and diesel. We also see good agreement between themeasured and predicted results for density.

Roughly 20�25% of the product in this FCC is LPG, whichprimarily consists of propane, propylene, butanes, and butenes. Thepresence of significant amounts (greater than 0.5%) of C5+ productsin LPG indicate that the fractionation process is not operating well.Therefore, the prediction of the composition of the all of the gas andLPG products is essential to validate the model. Tables 14 and 15compare the operating data and model predictions for LPG and dry

gas. The AAD for the predictions of mole compositions in LPG anddry gas are 1.2 and1.8%, respectively.Wenote that there is oftenmoresignificant error in the prediction of hydrogen and nitrogen.

We also apply the model to predict all temperature profiles ofcolumns for each validation case and compare the results to plantoperation. We find good agreement between plant measure-ments for all columns, with the exception of the debutanizercolumn (T302) (see Figure 24). This column is very sensitive tothe LPG composition in the model. We recall that the BASEcalibration case shows error in matching the LPG yield from theplant. It is possible to improve this prediction by includingcatalyst-specific parameters in the kinetic model to match theplant performance. However, we avoid this procedure at thistime; therefore, we can provide a more broadly useful model.Figures 22�26 compare model and plant values for temperatureprofiles for a single validation case (VALID-4).

11. APPLICATIONS

Refiners are very interested in obtaining optimal operatingconditions that maximize the yield of a profitable product slate.However, unlike traditional chemical plants, the FCC unitgenerates several products that have different profit margins.Furthering complicating matters is that these profit margins maychange depending upon refinery constraints, market conditions,and government regulations. Therefore, it is critical to under-stand how to manage the FCC unit under different operatingscenarios. We consider common scenarios in FCC operation:improving the gasoline yield, increasing unit throughput anddealing with sulfur constraints.

Figure 14. Specific implementation of the overall modeling strategy.Figure 15. Tracking aromatic content in the feed to ensure multipleoperating scenarios.

Table 12. Routinely Monitored Properties Used for Model Development and Calibration

feed products FCC fractionation

flow rateyield composition

(for light products)

temperatures

(feed, riser outlet, regenerator

bed, and flue gas)

temperature profile

distillation curves pressure profile

specific gravity densitypressure differential between riser/

reactor and regenerator

draw rates

Conradson carbon residue (CCR) RON/MON pumparound flow rates and duties

sulfur content (S) flash point set points (usually temperatures)

metal content (Fe, Na, Ni, and V) sulfur content steam usage

saturates, resins, aromatics, and asphaltenes (SARA) main air blower flow rate





11.1. Improving the Gasoline Yield. The gasoline yield istypically a complex function of the temperature, pressure, feedquality, and catalyst/oil ratio.8 We consider the case where thefeed quality is fixed. An easily manipulated operating variable isthe riser outlet temperature (ROT). Allowing the ROT toincrease improves the gasoline yield by promoting crackingand aromatic chain scission reactions that increase the yield of

C5+ components. We compute the gasoline yield at varioustemperatures and show the results in Figure 27. The currentROT is 510 �C and is marked with a yellow square. The ROT

Figure 16. Overall Aspen HYSYS model of the FCC unit and associated gas plant.

Table 13. Product Yield Results (AAD = 0.96%)

yield mass percent (%) model plant model plant model plant

VALID-1 VALID-2 VALID-3

gasoline 43.3 41.9 43.3 44.2 40.1 39.5

diesel 24.6 23.7 21.6 22.0 25.6 25.2

LPG 18.5 20.1 17.9 19.9 19.1 21.1

dry gas 4.9 4.4 5.0 4.2 4.7 4.1

slurry 1.4 4.0 5.5 3.8 4.5 3.9

coke 7.3 5.9 6.7 6.0 6.0 6.3


gasoline 41.5 41.2 44.1 44.2 40.8 41.2

diesel 24.7 24.6 20.8 20.9 24.3 24.5

LPG 19.3 21.6 17.8 20.6 18.6 20.2

dry gas 4.8 3.8 4.7 4.3 5.3 4.4

slurry 3.9 3.9 6.5 3.9 5.1 4.0

coke 5.7 4.8 6.0 6.2 5.9 5.6

Figure 17. ASTMD86 distillation for the product diesel from the mainfractionator (VALID-1).

Figure 18. ASTM D86 distillation for the product gasoline from thedebutanizer column (VALID-1).

Figure 19. Gasoline density comparison.

Figure 20. Diesel density comparison.








that leads to the highest yield of gasoline is roughly 530 �C. Doesthis mean that we should allow the ROT to increase to 530 �C?To answer this question, we plot the yields of the other valuableproducts from the FCC in Figure 28.

Figure 28 shows that, while the gasoline yield reaches themaximum at a ROT of 530 �C, the yields of other valuableproducts (i.e., diesel) drop significantly. In addition, the yield offuel/dry gas (light gases) rises quickly. This indicates that we are“overcracking” the feed. The high temperature accelerates theproduction of C1�C2 components (i.e., fuel/dry gas) throughthe catalytic and thermal cracking pathway. This is clearly anundesired result. Dry gas is not of significant value and can easilyoverload the overhead wet gas compressor. In addition, Figure 29

Figure 21. Diesel flash point comparison.

Table 14. Comparison of LPG Composition (AAD = 1.2%)

LPG mole percent (%) model plant model plant model plant


C3 13.9 15.5 13.9 14.9 14.7 13.3

C3d 36.6 38.3 35.1 35.9 38.3 38.4

NC4 4.5 5.3 4.1 5.6 4.0 5.6

IC4 17.5 17.1 16.9 18.8 16.1 18.0

IC4d 12.8 13.1 12.1 12.8 11.5 13.4

T-2-C4d 6.0 6.0 5.5 6.1 5.3 6.1

C-2-C4d 4.4 4.7 4.0 5.0 3.9 4.7


C3 14.2 13.2 15.6 12.2 15.5 13.0

C3d 34.5 39.0 35.9 41.7 37.0 39.4

NC4 4.3 4.9 4.5 3.4 4.5 4.5

IC4 16.6 18.4 18.2 18.0 17.5 18.6

IC4d 12.3 13.1 13.1 13.1 12.7 13.2

T-2-C4d 5.7 6.1 6.0 5.7 6.0 6.3

C-2-C4d 4.1 4.8 4.5 4.8 4.5 4.6

Table 15. Comparison of Dry Gas Composition (AAD =1.8%)

dry gas mole percent (%) model plant model plant model plant


H2 24.3 29.9 23.1 31.8 24.7 29.3

N2 21.0 20.1 19.5 16.7 19.7 19.1

CO 1.6 1.6 1.5 2.0 1.6 1.8

CO2 1.8 1.8 2.2 1.6 1.1 1.8

C1 24.8 23.0 24.5 24.8 25.6 23.1

C2 10.9 10.2 12.1 9.9 11.2 10.3

C2d 11.7 10.5 12.3 10.5 13.0 11.8


H2 20.5 28.2 21.6 27.5 20.8 28.1

N2 19.7 22.5 19.7 20.3 18.9 19.8

CO 1.6 1.7 1.6 1.7 1.5 1.4

CO2 1.7 2.0 1.8 2.0 3.6 1.6

C1 27.7 21.4 26.6 23.1 24.5 23.6

C2 10.6 10.5 11.7 10.1 11.7 10.3

C2d 13.8 11.6 12.9 11.2 11.9 11.2

Figure 22. Main fractionator temperature profile.

Figure 23. Primary absorber temperature profile.

Figure 24. Debutanizer temperature profile.







shows the coke yield on the catalyst as a function of the ROT.The amount of coke present on the catalyst leaving the riser is astrong function of the ROT. Regenerating the catalyst withhigher coke deposits increases the utilities required to regeneratethe catalyst to the same level of activity. These side effects shrinkthe acceptable range of values for the ROT.We can combine the results from these figures and consider

scenarios where a refiner wants to maximize multiple products.For example, the refiner may want to maximize the production ofgasoline and diesel or maximize the production of gasoline andLPG depending upon external constraints. We can easily use themodel to generate a case study, as shown in Figure 30. This figureshows that there are different optimum ROT values for differentscenarios. The maximum gasoline and diesel production occurs inthe range of 505�510 �C (confirming the assertion of the refinerwhere these data are obtained), whereas themaximum for gasolineand LPG production occurs in the range of 530�540 �C.This example shows the importance of a model that accounts

for all products, including light gases as a distinct lump. Inaddition, the integrated heat balance between the riser andregenerator allows us to provide useful estimates for the cokeyield. We have not included the effect of these process changeson the downstream fractionation unit in this study. However, wenote that there are often significant equipment and processconstraints (a prime example is the wet gas compressor) thatrestrict the acceptable range for the ROT.

11.2. Increasing the Unit Throughput. Let us consideranother scenario where we want to increase the throughput ofthe unit. The refiner typically wants to process the largest volumeof feedstock possible. Ideally, we would like the FCC to maintaina similar mass yield of the most valuable product (i.e., gasoline).Figure 31 shows the mass yield of gasoline as a function of thefeed rate to the unit. The mass yield decreases almost linearlywith an increasing feed rate. How can we explain this phenom-enon? Figure 31 also shows the catalyst/oil ratio as a function ofthe increasing feed rate. We note that the catalyst/oil ratio alsodecreases linearly.The decreased catalyst/oil ratiomeans that there is less contact

time between the catalyst and the feed oil. Lower contact time willresult in fewer species cracking and, subsequently, reduce thegasoline yield. However, we must not confuse this effect with“overcracking” described in the previous case study. Figure 31 alsoillustrates the difference between “overcracking” and a reducedcatalyst/oil ratio. We note that yield of light products (dry gas andLPG) does not increase. This indicates that high-temperaturethermal or catalytic cracking is not taking place.Let us now consider the scenario where we want to increase or

maintain the gasoline yield that corresponds to the base unitthroughput. We will allow the ROT to increase while alsoincreasing the feed rate to the unit. Figure 32 shows the effectof the increasing feed rate and ROT. We note that the gasolineyield increases with a rising ROT. However, once we reach theROT of 540 �C, the gasoline yield drops quickly. This occursbecause we have passed the “overcracking” peak for thisparticular feed.11.3. Sulfur Content in Gasoline. The sulfur content in

gasoline is an important regulatory constraint for refiners. Many

Figure 25. Sponge oil absorber temperature profile.

Figure 26. Primary stripper temperature profile.

Figure 27. Gasoline yield profile as a function of the ROT.

Figure 28. Yields of key products as functions of the ROT.







schemes are in use to reduce the sulfur content in refineryproducts. In the case of the FCC unit, a significant portion of thesulfur in the feed leaves the process as dry or sour gas. However,the remaining sulfur leaves through the key liquid products.Sadeghbeigi1 and Gary et al.7 indicate that hydrotreating the

feed significantly reduces the sulfur content in the non-slurryproducts. However, there may be an economic disadvantage inhydrotreating the feed to the FCC unit. In addition, low sulfurconstraints may result in an excess of low value resid feeds in therefinery. Often, the refiner looks for ways to blend this high-sulfurresid feed into processing units that can tolerate higher sulfur. Inboth cases, we need to understand how the changes in feed sulfuraffect the sulfur distribution in the products (Figure 33).Let us consider the situation where a cheaper feedstock,

vacuum residue (VR), is available. The refiner may want tomaximize the profitability of the unit by blending in VR with theexisting VGO feed. Currently, 5.7 wt % of the feed to the FCCunit is VR-type feed. We would like to know how much VR wecan blend into the VGO feed while meeting the sulfur constraintof stabilized gasoline.To study this question, we must also consider that the sulfur

content in the feed VGO is changing as well. We vary both thesulfur content in the feed VGO and the amount of VR that isblended. Figure 34 shows the results of the case study.We vary the feed ratio of VR from 0 to 11.3% wt% and the

associated sulfur content in the VGO. The corresponding sulfurlimit for FCC gasoline in this refinery is 800 parts per million by

weight (ppmwt). We use the model to predict the sulfur contentin different cases of the VR ratios and sulfur content in VGO. Wenote that, for the base case of 0.71 wt % sulfur in feed VGO, wecould blend in more than 10% VR while still meeting the sulfurconstraint. However, if sulfur in the VGO increases to 0.78 wt %,we cannot blend inmore than 4.5 wt % VR if we want to meet thesulfur constraint.We note that all of the above case studies and scenarios are

limited to the FCC unit and the associated fractionation system.Modern refineries are highly integrated, and changes that appearbeneficial in one plant may not benefit another plant in therefinery. One way to apply these models in a larger context (in anexisting refinery process) is through the linear program (LP) forrefinery planning.

12. REFINERY PLANNING

We briefly alluded to the complex nature of managing a FCCunit in the previous section. The typical refinery has many unitsin addition to the FCC (such as catalytic reforming and

Figure 29. Coke yield as a function of the ROT.

Figure 30. Maximizing production of key products as a function ofthe ROT.

Figure 31. Mass yield and catalyst/oil ratio as function of the feed rate.

Figure 32. Gasoline yield as a function of the feed rate.

Figure 33. Scenario of the feed sulfur change.








hydroprocessing) that have their own product distribution andassociated profit margins. It is difficult to produce high profitmargins dealing with each unit individually when the actualrefinery process is highly integrated. The refiner needs methodsto optimize feeds to each unit and related products on a refinery-wide scale.

Refiners have typically solved this problem using LP methods,which have been used extensively in refineries since 1950. Garyet al.7 state that “a site-wide model of the refinery is, therefore,usually required in order to properly determine refinery economics”.

LP involves the maximization of a linear objective function ofmany variables subject to linear constraints on each variable.57 Inthe context of a refinery, the objective function can refer tooverall profit generated from processing a particular set of crudeoils. The variables that affect this objective function are typicallythe amounts of different crude oils purchased. The goal is todetermine an optimal set of crude oils that maximize the profitmargin of the refinery. This scenario is an example of crude oilevaluation. Refiners typically use LP methods in other scenariosas well. Prominent examples are product blending (where two ormore products from different units are mixed to form a singleproduct) and production planning (determining the most profit-able distribution of products while meeting site constraints).

A key issue in using LP methods is that the relationshipsbetween variables must be linear. In other words, all of theequations used in the model must be linear with respect to thevariables involved. At first, this requirement appears very con-fining. In fact, the FCC and gas plant models developed inprevious sections of this work are highly nonlinear. However, it isimportant to note that many units in the refinery have a smallwindow of operating conditions during regular operation of therefinery. This allows us to linearize highly nonlinear processesaround the regular operating window of the refinery.

That being said, modern LP software, such as Aspen PIMS,includes many tools to deal with nonlinear relationships. AspenPIMS uses techniques such as “recursion” (a form of successiveLP, where the linear model runs many times with differentcoefficients to approximate nonlinear behavior) and nonlinearprogramming (NLP) techniques. These techniques can alleviatemany problems that frequently arise, especially in productblending and property estimation, with refinery planningmodels.The focus of our application study is to improve an existing LPmodel for the FCC unit alone; therefore, we do not considermore sophisticated techniques to deal with nonlinear behavior.

Figure 35 represents a highly simplified view of a FCC unit.We can consider the FCC unit as a black box that convertsdifferent types of feed into products with varying profit margins.

The LP model expects that the profits or values of the productsare readily available. If we consider that only straight-run VGOenters the unit at fixed operating conditions (riser temperature,catalyst/oil ratio, etc.), we can represent the yield of the unit as

1:0 ðnormalized feed rateÞ ¼ ∑N

i¼ 1yieldi ð13Þ

where we know all terms on the right-hand side to be fixedconstants. The yield coefficients, yieldi, correspond to eachmeasured product of the FCC.

We consider the above equation to represent the base yield of theunit. In Aspen PIMS and other similar LP software, the base yield iscalled the base vector.We typically encode theBASEvector in a formshown in Table 16. The negative signs arise from moving all of theterms from the right-hand side of the equation to the left-hand side.

This base vector is sufficient tomodel a FCCunit that processes asingle type of feed at fixed operating conditions. However, mostFCCunits do not operate in this fashion. They acceptmultiple feedswith varying compositions and may operate at different conditions.To account for variations in feed composition, the concept of theDELTA vector is useful. Every attribute (SG, CCR, sulfur content,etc.) of the feed that can affect the yield of the unit has its ownDELTAvector. TheDELTAvector can be thought of as a slope thatmodifies the base yield of each product. If we consider the specificgravity (SPG) of the feed as an attribute that can change the productyields, we can now rewrite the yield equation as

1:0 ¼ ∑N

i¼ 1yieldi

þ ∑N

i¼ 1ðyield modifier or DELTAÞiSPG ð14Þ

where the SPG of the feed is a known quantity, yieldi and DELTAcoefficients are known for each product i. The products typically aredry gas, LPG, gasoline, diesel, and resid/coke/loss. Note the valuesof the DELTA coefficients correspond to the units of measurementof the particular feed attribute (in this case, SPG). Table 17 givessample BASE andDELTA vectors for a typical gasoline-maximizingFCC unit.

Refiners can typically obtain the base yield of the FCC unit byaveraging the measured yields over some period of time. TheDELTA vectors often come from estimations, internal refinerycorrelations, or published correlations.7,58,59 Previous work by Liet al.60 uses correlations from Gary and Handwerk7 to generateFCC DELTA�BASE vectors. These vectors are then combinedwith a blending and crude distillation unit model. This process

Figure 34. Blending in varying amounts of residue feed.

Figure 35. Simplified view of the FCC unit for a LP application.





results in two significant problems. The first problem is that thetrue yield of the FCC unit is not available to LP (only averagedyields). This leads to situations where the LPmodel can optimizethe product distribution based on poor yield information. Thesecond problem is that the DELTA vectors are fixed to particularcorrelations or estimates. These correlations or estimates maynot accurately predict changes in the yield when feed composi-tion changes.

We overcome these problems using the detailed FCC modeldeveloped in this work. We have shown that the FCC model canpredict yields accurately for varying process conditions. To applythe FCC model into the refinery LP, we must first convert thelarge nonlinear model into a linear yield model. We can then use

coefficients from this linear yield model directly in the LP for therefinery. We show the process for generating the linear yieldcoefficients in Figure 36. We have found that 4�5% is a reasonablevalue for CHANGE % (variable perturbation) for most of theimportant feed attributes in the FCC process. For example, togenerate the DELTA vector for sulfur content (SUL), we will firstrun the model at the base conditions and record these yields as theBASE vector. Next, we perturb the SUL variable by 5% and recordthe perturbed product yields. We divide the difference in base yieldsand perturbed yields by the change in the perturbed value to obtainthe DELTA vector corresponding to the SUL variable.

It is important to note that the process in Figure 36 essentiallygenerates an approximation to the Jacobianof thenonlinear FCCunitmodel. If we consider that the vector y represents the model outputs,then the y vector represents the base case in our planning scenario andthe Δx vector represents the change in model inputs from the basecase. We then have a matrix ofΔy/Δx, which represents the changefrom the base condition as a function of the selected feed attributes(or possibly process conditions). Equation 15 illustrates the connec-tion between the Jacobian and DELTA�BASE vectors.

y1y2lym

266664

377775 ðPREDICTIONÞ �

y1y2lym

266664

377775 ðBASEÞ

þ

Δy1Δx1 3 3 3

Δy1Δxn

l 3 3 3 lΔymΔx1 3 3 3

ΔymΔxn

2666664

3777775 ðDELTA� BASEÞ

Δx1l

Δxn

2664

3775 ðDELTAÞ

ð15ÞTable 18 shows the existing base and DELTA vectors for the FCCunit. The base vectors come from averaged yields of the FCC unitduring the previous quarter (ending December 2008). The DELTA

Table 16. Sample Base Vector with Typical Yields for aGasoline-Maximizing FCC Unit

row product BASE

1 feed 1.00

2 dry gas �0.04

3 LPG �0.18

4 gasoline �0.40

5 diesel �0.30

6 loss and coke �0.08

Table 17. BASE andDELTAVectors with Typical Yields for aGasoline-Maximizing FCC Unit

row product BASE SPG

1 feed 1.00

2 dry gas �0.04 �0.01

3 LPG �0.18 0.02

4 gasoline �0.40 0.01

5 diesel �0.30 �0.01

6 loss and coke �0.08 �0.02

Figure 36. Process to generate DELTA�BASE vectors.

Table 18. Existing DELTA�BASE Vectors for the FCC Unit(Normalized to a Feed Rate of 1.0)

row product BASE SPG CON SUL

1 feed 1.00

2 sour gas �0.0065 �0.0003 �0.0004 �0.0082

3 dry gas �0.0394 �0.0011 �0.0014 0.0000

3 LPG �0.1740 0.0025 0.0041 0.0000

4 gasoline �0.3929 0.0098 0.0081 0.0000

5 diesel �0.2899 �0.0057 �0.0033 0.0000

6 slurry �0.0381 �0.0032 �0.0038 0.0082

7 coke �0.0544 �0.0020 �0.0034 0.0000

8 loss �0.0048 0.0000 0.0000 0.0000

Table 19. DELTA�BASE Vectors Generated Using aRigorous Model

row product BASE SPG CON SUL

1 sour gas �0.00439 0.00068 0.0001 �0.0057

2 dry gas �0.02527 0.00069 0.00033 0.00025

3 LPG �0.19386 0.02213 0.00271 0.00164

4 gasoline �0.4421 0.09480 0.00621 0.00330

5 coke �0.06218 �0.05913 �0.00453 0.00038




vectors come from internal refinery correlations. TheDELTAvectorsrefer the specific gravity of the feed (SPG),Conradson carbon residue(concarbon) in the feed (CON), and sulfur in the feed (SUL). Wenote that this particular set of BASE and DELTA vectors does notaccurately reflect the operation of the unit. As shown earlier in thiswork, the actual gasoline yield of the FCC unit ranges from 42 to46%. The LP model underestimates the gasoline yield. In addition,because the FCC unit is the most significant producer of gasoline inthe refinery, using the LP in crude selection context can lead to non-optimal crude selection.

Table 19 shows the DELTA�BASE vectors that we generateusing the procedure in Figure 36. The new BASE vectoraccurately reflects the current base gasoline and LPG yields ofthe FCC unit. In addition, as a consistency check, we note thatthe SUL coefficient for the sour gas (row 1) has a negativecoefficient. This indicates that sour gas increases as the sulfur inthe feed increases. A similar consistency test with the CONcoefficient and coke (row 5) shows the same result. We can usethe LP model optimally, knowing that the LP model does notunderestimate key product yields.

The advantage of this method is that the LP model now reflectsthe actual capabilities of the unit and not the perceived capabilitiesbased on historical data or correlations. In addition, if the rigoroussimulation is updated alongside plant retrofits, we canmodify the LPmodel quickly to track these retrofits. Theworkflow thatwedescribein Figure 36 is easy to integrate into existing process simulation andLP software. Aspen HYSYS/Petroleum Refining includes tools toautomate the workflow and export the updated DELTA�BASEvectors to Aspen PIMS (LP software) directly. This automationallows for quick updates of the LP model to accurately reflect unitperformance.

13. CONCLUSION

In this work, we have developed a model for a FCC unit thatincludes a significant implementation of the associated gas plantusingAspenHYSYS. The key highlights of this work are as follows:(1) brief summary of the existing literature for modeling a typicalFCC unit, (2) description of the Aspen HYSYS FCC model and21 lump kinetics, (3) technique to fill out partial distillation curvesusing statistical functions, (4) regression of parameters for a newPNA correlation for petroleum fractions, (5) technique to infermolecular composition of the FCC feedstock from routineanalysis, (6) strategy to develop reasonable process models usingindustrial plant data, (7) application of the model to a large-scalerefinery process showing less than 2.0% AAD for key productyields and satisfactory predictions of product composition andproduct quality (composition/distillation data, density, and flashpoint), (8) case studies that use the model to investigate indust-rially useful changes in operation, and (9) strategy to transferresults from this model into a LP-based refinery planning tool.

Earlier work in this area has focused mostly on isolated parts(kinetic model, riser/regenerator, and gas plant) of the FCCprocess. In this work, we show how to use routinely collectedplant data with well-known commercial software tools to presentan integrated process model that includes both reaction andfractionation systems. An integrated model allows users toidentify opportunities to improve yield, increase profitability,and monitor the unit for predictable operation. This approach iscritical for modern refineries that have increasingly complexprocess flows and require engineers to examine the performanceof refinery units holistically.

’AUTHOR INFORMATION

Corresponding Author*Telephone: (540) 231-7800. Fax: (540) 231-5022. E-mail:[email protected].

’ACKNOWLEDGMENT

We thank Mr. Stephen Dziuk of Aspen Technology for hisvaluable review comments. We thank Alliant Techsystems, AspenTechnology, China Petroleum and Chemical Corporation(SINOPEC), Milliken Chemical, Novozymes Biological, andMid-Atlantic Technology, Research and Innovation Center(MATRIC) for supporting our educational programs in compu-ter-aided design and process system engineering at Virginia Tech.

’NOMENCLATUREA, B, R, and β = fitting parameters for cumulative β distributiona, b, c, and d = fitting parameters for PNA correlationAAD = average absolute deviationaKCOKE = activity factor because of kinetic cokeaMCOKE = activity factor because of metal cokeC1 = methaneC2 = ethaneC3 = propane and propyleneC4 = butanes and butenesC5 = pentanes and pentenesCCR and CON = Conradson carbon residue (wt %)CGO = coker gas oilCKCOKE = kinetic coke on catalyst (kg of kinetic coke/kg of

catalyst)CMCOKE = metal coke on catalyst (kg of metal coke/kg of

catalyst)CMETALS = metal composition on catalyst (ppm metals/kg of

catalyst)D = riser diameter (m)ε = voidage factorE = Murphree stage efficiency factorϕCOKE = total coke deactivation functionϕKCOKE = deactivation function because of kinetic cokeϕMCOKE = deactivation function because of metal cokeFr = Froude numberFrt = particle Froude numberg = acceleration due to gravity (9.81 m/s2)HCO = heavy cycle oilj = slip factorKW = Watson K factorLCO = light cycle oilMeABP = mean average boiling point temperature (K)MON = motor octane numberPNA = paraffins, naphthenes, and aromaticsθ = normalized temperatureRi = refractive indexRON = research octane numberRSS = sum of least squaresSG and SPG = specific gravitySUL = sulfur content (wt %)T0 = lower reference temperature (�C)T1 = upper reference temperature (�C)TBP = true boiling pointuo = superficial gas velocity (m/s)ut = terminal catalyst particle settling velocity (m/s)



VGC = viscosity gravity constantVGF = viscosity gravity factorVGO = vacuum gas oilx = normalized liquid recovery% XA = mole composition of aromaticsxexp = normalized experimental liquid recoveryxn = mole fraction of liquid leaving stage n% XN = mole composition of naphthenes% XP = mole composition of paraffinsyieldi = yield coefficients for the LP modelyn = mole fraction of vapor leaving stage n

’REFERENCES

(1) Sadeghbeigi, R. Fluid Catalytic Cracking Handbook: Design,Operation and Troubleshooting of FCC Facilities; Gulf Publishing Com-pany: Houston, TX, 2000.(2) Arbel, A.; Huang, Z.; Rinard, I. H.; Shinnar, R.; Sapre, A. V. Ind.

Eng. Chem. Res. 1995, 34, 1228–1243.(3) McFarlane, R. C.; Reineman, R. C.; Bartee, J. F.; Georgakis, C.

Comput. Chem. Eng. 1993, 3, 275–300.(4) Chitnis, U. K; Corripio, A. B. ISA Trans. 1998, 37, 215–226.(5) Khandalekar, P. D.; Riggs, J. B. Comput. Chem. Eng. 1995, 19,

1153–1168.(6) Hsu, C. S; Robinson, P. R. Practical Advances in Petroleum

Processing; Springer: New York, 2006; Vol. 1 and 2.(7) Gary, J. H.; Handwerk, G. E. Petroleum Refining Technology and

Economics, 4th ed.; Marcel-Dekker: New York, 2001.(8) Raseev, S. D.; Thermal and Catalytic Processing in Petroleum

Refining; CRC Press: Boca Raton, FL, 2003.(9) Takatsuka, T.; Sato, S.; Morimoto, Y.; Hashimoto, H. Int. Chem.

Eng. 1987, 27, 107–116.(10) Lee, E.; Groves, F. R., Jr. Trans. Soc. Comput. Simul. 1985,

2, 219–236.(11) Blanding, F. H. Ind. Eng. Chem. 1953, 45, 1193–1197.(12) Gupta, R. K.; Kumar, V.; Srivastava, V. K. Chem. Eng. Sci. 2007,

62, 4510–4528.(13) Jacob, S. M.; Gross, B.; Voltz, S. E.; Weekman, V. W. AIChE J.

1976, 22, 701–713.(14) Oliviera, L. L.; Biscasia, E. C., Jr. Ind. Eng. Chem. Res. 1989, 28,

264–271.(15) Pitault, I.; Nevicato, D.; Forissier, M.; Bernard, J. R. Chem. Eng.

Sci. 1994, 49, 4249–4262.(16) Van Landeghem, F.; Nevicato, D.; Pitault, I.; Forissier, M.;

Turlier, P.; Derouin, C.; Bernard, J. R. Appl. Catal., A 1996, 138,381–405.(17) AspenTech. Aspen RefSYS Option Guide; AspenTech:

Cambridge, MA, 2006.(18) AspenTech. Aspen Plus FCC User’s Guide; AspenTech:

Cambridge, MA, 2006.(19) Froment, G. F. Catal. Rev.�Sci. Eng. 2005, 47, 83.(20) Quann, R. J.; Jaffe, S. B. Ind. Eng. Chem. Res. 1992, 31, 2483.(21) Quann, R. J.; Jaffe, S. B. Chem. Eng. Sci. 1996, 51, 1615.(22) Quann, R. Environ. Health Perspect. Suppl. 1998, 106, 1501.(23) Christensen, G.; Apelian, M. R.; Hickey, K. J.; Jaffe, S. B. Chem.

Eng. Sci. 1999, 54, 2753–2764.(24) Klein, M. T.Molecular Modeling in Heavy Hydrocarbon Conver-

sions; CRC Press: Boca Raton, FL, 2006.(25) Kumar, S.; Chadha, A.; Gupta, R.; Sharma, R. Ind. Eng. Chem.

Res. 1995, 34, 3737–3748.(26) Ellis, R. C.; Li, X.; Riggs, J. B. AIChE J. 1998, 44, 2068–2079.(27) Secchi, A. R.; Santos, M. G.; Neumann, G. A.; Trierwiler, J. O.

Comput. Chem. Eng. 2001, 25, 851–858.(28) Mo, W.; Hadjigeorge, G.; Khouw, F. H. H.; van der Werf, R. P.;

Muller, F. Hydrocarbon Asia 2002, October, 30–42.(29) Elnashaie, S. S. E. H.; Mohamed, N. F.; Kamal, M. Chem. Eng.

Commun. 2004, 191, 813–831.

(30) Rao, R. M.; Rengaswamy, R.; Suresh, A. K.; Balaraman, K. S.Trans. IChemE: Part A 2004, 82, 527–552.

(31) Araujo-Monroy, C.; Lopez-Isunza, F. Ind. Eng. Chem. Res. 2006,45, 120–128.

(32) Bollas, G. M.; Vasalos, I. A.; Lappas, A. A.; Iatridis, D. K.;Voutetakis, S. S.; Papadopoulou, S. A. Chem. Eng. Sci. 2007, 62,1887–1904.

(33) Fernandes, J. L.; Pinheiro, C. I. C.; Oliveira, N. M. C.; Inverno,J.; Ribeiro, F. R. Ind. Eng. Chem. Res. 2008, 47, 850–866.

(34) Shaikh, A. A.; Al-Mutairi, E. M.; Ino, T. Ind. Eng. Chem. Res.2008, 47, 9018–9024.

(35) Fernandes, J. L.; Pinheiro, C. I. C.; Oliveira, N. M. C.; Neto,A. I.; F. Ramoa, R. Chem. Eng. Sci. 2007, 62, 6308–6322.

(36) Chang, S. L.; Zhou, C. Q. Comput. Mech. 2003, 31, 519–532.(37) Arandes, J. M.; Azkoti, M. J.; Bilbao, J; de Lasa, H. I. Can. J.

Chem. Eng. 2000, 78, 111–123.(38) Han, I. S.; Riggs, J. B.; Chung, C. B. Chem. Eng. Process. 2004,

43, 1063–1084.(39) Paraskos, J. A.; Shah, Y. T.; McKinney, J. D.; Carr, N. L. Ind.

Eng. Chem. Process. Des. Dev. 1976, 15, 165–169.(40) Shah, Y. T.; Huling, G. P.; Paraskos, J. A.; McKinney, J. D. Ind.

Eng. Chem. Process. Des. Dev. 1977, 16, 89–94.(41) Arandes, J. M.; Abajo, I.; Fernandez, I.; Lopez, D.; Bilbao, J. Ind.

Eng. Chem. Res. 1999, 38, 3255–3260.(42) De Lasa, H. I.; Grace, J. R.Can. J. Chem. Eng. 1979, 25, 984–990.(43) Rice, N. M.; Wojciechowski, B. W. Can. J. Chem. Eng. 1991,

69, 1100–1105.(44) Harriot, P. Chemical Reactor Design; Marcel Dekker: New York,

2003.(45) Malay, P.; Milne, B. J.; Rohani, S. Can. J. Chem. Eng. 1999,

77, 169–179.(46) Corella, J.; Frances, E. Fluid Catalytic Cracking II: Concepts in

Catalyst Design; American Chemical Society (ACS): Washington, D.C.,1991; ACS Symposium Series, Vol. 452, Chapter 10, pp 165�182.

(47) Bolkan-Kenny, Y. G.; Pugsley, T. S.; Berutti, F. Ind. Eng. Chem.Res. 1994, 33, 3043–3052.

(48) Han, I. S.; Chung, C. B. Chem. Eng. Sci. 2001, 56, 1951–1971.(49) Froment, G. F.; Bischoff, K. B.; Wilde, J. D. Chemical Reaction

Analysis and Design, 3rd ed.; Wiley: Hoboken, NJ, 2010.(50) Kister, H. Z. Distillation Design; McGraw-Hill, Inc.: New York,

1992.(51) Kaes, G. L. Refinery Process Modeling: A Practical Guide to Steady

State Modeling of Petroleum Processes; The Athens Printing Company:Athens, GA, 2000.

(52) Bollas, G.M.; Vasalos, I. A.; Lappas, A. A.; Iatridis, D. K.; Tsioni,G. K. Ind. Eng. Chem. Res. 2004, 43, 370–3281.

(53) Sanchez, S.; Ancheyta, J.; McCaffrey, W. C. Energy Fuels 2007,21, 2955–2963.

(54) Daubert, T. E.; Danner, R. P. API Technical Data Book—Petroleum Refining, 6th ed.; American Petroleum Institute (API):Washington, D.C., 1997.

(55) Riazi, M. R. Characterization and Properties of Petroleum Frac-tions, 1st ed.; American Society for Testing andMaterials (ASTM):WestConshohocken, PA, 2005.

(56) Goosens, A. G. Ind. Eng. Chem. Res. 1997, 36, 2500.(57) Bazaraa, M. S.; Jarvis, J. J.; Sherali, H. D. Linear Programming

and Network Flows; John Wiley and Sons: Hoboken, NJ, 2009.(58) Xu, C.; Gao, J.; Zhao, S.; Lin, S. Fuel 2005, 84, 669–674.(59) Ancheyta-Juarez, J.; Murillo-Hernandez, J. A. Energy Fuels

2000, 14, 373–379.(60) Li, W.; Chi-Wai, H.; An-Xue, L. Comput. Chem. Eng. 2005, 29,

2010–2028.

fcc modeling

Documents

integrated fcc model

gas plant model

commercial fcc plant

model accounts

lumpkinetic model

lumped fcc kinetic model

fcc product yields

entire fcc plant