Petroleum Refining

Process Control and

Real-Time Optimization ROBERT E. YOUNG

TRANSFORMING OIL BOILING INTO MANAGING MOLECULES

efineries are a complex network of processes that convert crude oil into finished

R petroleum products such as gasoline, diesel, jet fuel, heating oil, fuel oil, lubricants, asphalt,
coke, wax, and chemical feed stocks. Crude oil and these products are complex mixtures
of hydrocarbon molecules as described in “What Is Crude Oil?” and “What is Gaso-
line?” Examples of simple hydrocarbon molecules are shown in Figure 1. The
boiling point curves in Figure 2 [1] characterize the composition of a typical
crude oil without having to determine the precise molecular composition.
While crude oil naturally contains some gasoline-type molecules,
Figure 2 shows there are not enough gasoline molecules to meet
demand. In simple terms, refineries are processes that convert
the crude oil curve to the product demand curve. Expanding
on this simple idea, refineries separate hydrocarbons into
similar oil fractions, convert low-value molecules into
higher value molecules, and blend hydrocarbon frac-
tions into product streams. Additionally, the con-
version step can include processing to remove
environmentally undesirable components such
as sulfur or hydrogen sulfide (H2 S).

REFINERY PROCESS OVERVIEW

Refining processes can be classified into
three major groups: separations, conver-
sions, and blending. Distillation is the
most common separation process in
refining. Conversion includes differ-
ent types of reaction process, such as
cracking, hydroprocessing, reform-
ing, isomerization, and alkylation.
Cracking reactions break large mol-
ecules into smaller ones, while iso-
merization and alkylation reactions
rearrange and combine smaller
molecules into larger, more valu-
able molecules. Reforming reac-
tions rearrange the structure of
molecules without changing the
number of atoms to create higher

COURTESY OF EXXON MOBIL CORPORATION

1066-033X/06/$20.00©2006IEEE DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 73

value gasoline blending components. Hydroprocessing The flow sheet in Figure 3 illustrates the network of
reactions use hydrogen to remove contaminants such as process units in a typical fuels refinery. The process vessels
sulfur and nitrogen from the hydrocarbons. Most of these and piping connections at the ExxonMobil Torrance Refin-
reactions require a catalyst and are carried out in a variety ery are shown in Figure 4. U.S. refinery process capacities
of different reactor configurations including fluidized bed are available online from the Department of Energy [7].
reactors, fixed-bed reactors, and liquid-contacting reactors. Crude unit capacities range from 1.7 KBD (thousand bar-
Refineries use a variety of conversion processes and pro- rels per day) to 580 KBD, where one barrel equals 42 gal.
cessing capacity to convert crude oil fractions into products. As an example, the ExxonMobil Torrance Refinery has a
The ability of a refinery to convert crude oil fractions to crude unit capacity of 149.5 KBD, fluidized catalytic
products is its conversion capacity. Refineries that lack con- cracker (FCC) capacity of 95 KBD, and a hydrocracker
version capacity must sell intermediate products to other capacity of 20 KBD.
refiners instead of converting higher molecular weight oil The first refinery process is the crude unit, the only
fractions into finished products. There are currently about purely separations unit. The crude unit often includes two
153 operating refineries in the United States, with the large distillation columns for separating the crude oil into
newest being 25 years old [6]. Over their life span, these fractions that can be further processed. Several smaller dis-
refineries have undergone design modifications reflecting tillation units are also used to separate the lighter hydro-
changing crude supplies and product specifications as well carbons into purified gas streams. Conversion units
as new regulatory requirements. Some refineries are capable downstream of the crude unit include a reformer, FCC,
of producing lubricants, wax, or other specialty products, coker, hydrocracker, and alkylation unit. Each conversion
whereas some companies integrate petrochemical complex- unit also possesses the ability to separate the reaction
es with refineries to gain economic efficiencies. products into blending component streams.

What Is Crude Oil?

C
rude oil is a complex mixture of hydrocarbons with varying because the nitrogen poisons some catalysts used to convert
small amounts of sulfur, nitrogen, salt, and metals such as nick- less desirable components in the crude oil into more valuable
el, vanadium, and copper. Hydrocarbons are molecules that have products. Finally, crudes with higher total acid content are
carbon atoms as their backbone and hydrogen attached to the car- more difficult to process because the higher acid content
bon backbone. The hydrocarbons in crude oil are typically paraf- requires exotic metallurgy, more expensive equipment, and
fins, napthenes, and aromatics. Olefins are not typically found in higher maintenance cost to achieve good reliability.
crude oil but are produced in the refining process. Examples of A distillation or boiling-point characterization is also useful
these hydrocarbon classes are shown in Figure 1, recognizing that for understanding the composition of a particular crude oil or
much larger and more complex molecules are found in crude oil. hydrocarbon mixture. A boiling-point curve shows the volume
Crude oil is primarily produced from oil fields in the earth but fraction of the crude that boils away as a function of tempera-
is also recovered from sources such as tar sands. Crude oil is ture as shown in Figure 2 [1]. These boiling-point curves indi-
transported from the oil fields to refineries by a combination of cate the types of molecules present in the crude because the
marine tankers and pipelines. Supply-chain optimization is per- components in the mixture boil off in the order of their pure
formed by crude traders, transporters, suppliers, and refiners. component boiling points. However, because crude oil and
The most common characterization of crude oils is based most petroleum products are complex mixtures of thousands
on American Petroleum Institute (API) gravity, sulfur and nitro- of individual components, a boiling-point curve is insufficient
gen content, and total acid content (TAN), where API gravity is to describe the molecular structure of the mixture. In the past
a petroleum-specific measure of density that is inversely pro- when refiners had only this limited information to base deci-
portional to specific gravity. API gravity increases for lighter sions on, they earned the nickname “oil boilers.”
crudes and decreases for denser crudes. The API gravity of While useful in general comparisons, these properties do not
water is 10◦ API. Light crudes contain lower molecular weight, provide a basis for billion-dollar decisions. For example, with
higher value products, while more dense heavy crudes contain clean fuels as a critical business driver, low-sulfur crude oils that
material that requires more processing to become gasoline, jet contain hard-to-remove sulfur-containing hydrocarbons can pro-
fuel, or diesel. Sulfur content generally runs from 0.1% to duce higher sulfur gasoline [2]. Advances in laboratory assay
greater than 5%. Crude oil with more than 0.5% total sulfur or analyses along with compositional modeling now permit
high hydrogen sulfide concentration (called sour) requires sig- improved characterization of crude oils and other petroleum mix-
nificant additional processing to make clean products that meet tures on a molecular level. Access to the molecular fingerprint of
environmental specifications. Sweet crudes contain less sulfur. petroleum mixtures including crude oil is changing the practice
High amounts of total nitrogen in crude oil are undesirable of oil boiling to managing molecules.

74 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2006

What Is Gasoline?

G
asoline is one of several thousand products produced from The RVP of gasoline indicates how easily the liquid fuel vapor-
crude oil. Transportation fuels, such as gasoline, jet fuel, and izes at 38 °C. Higher RVP is economically attractive because more
diesel, account for well over half the products produced by components can be used in gasoline blending, which improves
refineries. In 2004, gasoline alone comprised 46.8% of the prod- engine starting. Lower RVP prevents engine vapor lock and
ucts produced [3]. Gasoline, which contains hydrocarbons with reduces evaporation of gasoline to the atmosphere. RVP specifica-
between 5–12 carbon atoms, is a blend of hydrocarbons that tions are changed in the summer months because the warmer
boil between 38–205 °C. temperatures add to vapor lock and evaporation issues, resulting in
Early gasoline specifications focused on engine perfor- summer and winter gasoline specifications.
mance, specifically, octane number and Reid vapor pressure More recently, federal and state regulations to improve air
(RVP). The octane number is a measure of how well a fuel can quality and provide cleaner fuel are the primary drivers for gaso-
be compressed without igniting prior to introducing a spark. If the line specifications, bringing about boutique gasoline products as
fuel ignites before the spark plug fires, the engine is said to shown in Figure A [4], [5]. These regulations limit the amounts of
knock. Heptane, a seven-carbon paraffin, compresses poorly sulfur, benzene, and other aromatics in gasoline and, in some
and was designated an octane number of zero. Conversely, iso- locations, mandate or restrict the addition of oxygenated additives
octane, an eight-carbon branched paraffin, performs much better such as MTBE or ethanol. Currently, many refiners are working on
and was assigned an octane number of 100. Consequently, a clean fuels projects to reduce the sulfur in diesel fuels.
fuel with 87 octane performs like a mixture of 87% iso-octane Finally, another key component in gasoline, not part of the
and 13% heptane. Modern engines have sensors to measure refining process, is additives. Fuel additives improve engine per-
engine performance and are designed not to knock by changing formance by preventing buildup of combustion products such as
the ignition control. However, using fuel with a low octane num- carbon deposits and gum in fuel injectors and engine valves.
ber can result in the ignition control advancing the timing, lead- Quality base-fuel stock and fuel additives help keep consumers’
ing to lower horsepower. vehicles performing well.

Montana North Dakota

Washington Minnesota

Maine
Wisconsin
South Dakota Vt.
Oregon N.H
Wyoming Michigan
Idaho New York
Mass.
Nebraska Iowa
R.l
Pennsylvania Conn.
Illinois Indiana Ohio
Nevada Md N.J.
Utah W. Va. Del.
Colorado Kansas
Missouri Kentucky Virginia

California Tennessee North Carolina

Arizona Oklahoma Arkansas
New Mexico South Carolina

Alabama Georgia
Texas Mississippi
Florida
RFG - North RFG w/Ethanol Loulsiana
RFG - South NV CBG
Oxygenated Fuels 7.2 RVP Exxon Mobil
CA CBC 7.0 RVP As of January, 2005
RFG/CA CBG 7.8 RVP, MTBE-No Increase
AZ CBG 7.8 RVP
Oxy Fuels/7.8 RVP 7.0 RVP, 30 ppm S This Map is not Intended to Provide Legal Advice or to be Used as Guidance for State and/or
Oxy Fuels/7.0 RVP 300 ppm S Federal Fuel requirements, Including but not Limited to Oxy Fuel or RFG Compliance
Requirements. Exxon Mobil Makes no Representations or Warranties, Express or
Conventional
Otherwise, as to the Accuracy or Completeness of this Map.

K.W. Gardner
G010080

FIGURE A Boutique gasoline fuel specifications. These specifications vary across the United States to meet local environmental
requirements. This variation increases both the cost of producing fuels and the complexity of ensuring an adequate supply of fuels to
each location. (Reprinted with permission from Exxon Mobil Corporation.)

DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 75

 lyst, replacing the catalyst involves lost processing and
expensive maintenance. Many of these processes are
Refining processes can be classified licensed to refining operating companies by process engi-
neering companies [8], [9].
into three major groups: separations,
conversions, and blending. REFINERY PROCESS CONTROL AND OPTIMIZATION
Refinery control and optimization systems are organized
in a hierarchical structure typical of many large-scale sys-
tems as shown in Figure 5. This hierarchical structure is
The refinery process flow includes several hydropro- common throughout the petroleum and chemicals indus-
cessing units to remove contaminants either before or after try [10]. At the lowest level, the basic flow, pressures, and
the conversion units, depending on whether the feed is temperature controls are implemented in distributed con-
being treated to avoid catalyst deactivation or the product trol systems (DCSs) and programmable logic controllers
is being treated to meet specifications. Catalysts in the con- (PLCs). These systems are designed to collect and record
version units may contain significant amounts of expen- sensor measurements from the refinery processes, com-
sive rare metals such as platinum or may be sulfuric or pute control signals for the manipulated variables such as
hydrofluoric acid, which require special handling. While control valves and solenoids, and provide the man-
the metals in the catalysts are reclaimed to make new cata- machine interface to the process including generating

Paraffins
H H
H H C H H C H
H H H H H H H H C H H H H H H H H H H H
H C C C H H C C C C H H H H C C C C C C C H H C C C C C H
H H H H H H H H C C C H H H H H H H H H H H H
H H H H C H
H

Propane n-Butane Isobutane n-Heptane iso-Octane

(2,2,4 Methyl Pentane)

Naphtenes (Cycloparaffins) Olefins

H2 H2
C C
H H H H H H
H2 C C H2 H2 C C
C H3 C C H C C C H
C C H H H H H H
H2 C C
H2 H2 C H3
C
H2
Cyclopentane 1,2-Dimethylcyclohexane Ethylene Propylene

Aromatics

H H CH3
C C C
H C C H H C C H H C C H

H C C H H C C H H C C H
C C C
H CH3 CH3
Benzene Toluene Para-Xylene

FIGURE 1 Common hydrocarbon molecules. These molecules represent various classes of hydrocarbons that make up crude oils and petro-
leum products processed in refineries.

76 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2006

 When the crude oil feed changes composition frequently or the refinery product
demand and pricing fluctuate, a predictive controller based on linear models must
compute targets that trade off the quality of separation for energy savings.

alarms for abnormal situations. To get a sense of scale, a model, which is often a linear model identified from
moderately large refinery system has on the order of process data collected during planned dynamic tests. While
100,000 sensors and 10,000 actuators. Also implemented some predictive controllers are implemented on the DCS,
at this level are advanced regulatory controls such as cas- the computing requirements frequently exceed the capabil-
cade controllers, ratio controllers, constraint controllers, ity of the DCS due to controller size as defined by the num-
and sequencing controls. The design of the distributed ber of controlled and manipulated variables and prediction
control systems allows execution of these controllers horizon length. As a result, predictive controllers are often
nominally once per second. implemented using vendor software on general purpose
At the next level, multivariable predictive controllers computers that communicate with the DCS. An overview
(MPCs) broaden the scope of the control problem. Imple- of the industrial implementation of predictive controllers in
mentation of MPCs permit control of highly interactive the process industry is provided in [11].
multivariable processes that are subject to multiple con- Another advantage of predictive controllers is that, at
straints. These controllers require a dynamic process each execution, optimal steady-state control and

Gasoline Diesel and Home Heating Oil

Lubricating Oils Asphalt
Jet Fuel
Volume Fraction

Crude Supply
Product
Demand

180 300 400 500 650 1,000 1,300

Normal Boiling Point (°F)
Paraffins

Naphthenes
Benzenes

Benzo-Thiophenes
S S S S

Naphthalenes
Di-Benzothiophenes
S S S S

Phenanthrenes
Pyrenes
Chrysenes

Benzo-Pyrenes

FIGURE 2 Boiling-point curves for typical crude oil and refinery products. Refineries convert crude oil molecules into product components
such as those listed above. Hydrocarbons typical of the boiling-point ranges are shown beneath these curves. The conversion process can
operate more effectively with a thorough understanding of the underlying molecular structures. (Reprinted with permission from Exxon Mobil
Corporation.)

DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 77

 Gas Gas Fuel Gas
Plant
Isobutane
Liquified Petroleum
Gasoline Gas (LPG)

Gasoline
Reformate (Regular/Premium)
Hydrotreater Reformer
Naphtha
Jet Fuel
Crude Oil Crude
Kerosene
Unit Hydrotreater Blending
Diesel Fuel
Propylene/
Gas Oil Butene Alkylate
FCC Alkylation
Heating Oil
Gasoline

Gas Oil
Gas Fuel Oil
Kerosene

Coker Hydrocracker Asphalt

Hydrocrackate

Coke

FIGURE 3 Process units and interconnecting flows in a typical fuels refinery. The crude unit separates the crude oil into streams to be converted
and purified into blending components. Each process is a complex combination of reaction and separation operations needed to produce fuels
such as gasoline, diesel, jet fuel, fuel oil, and heating oil.

manipulated variable targets are determined. Since these straints associated with reaction rates, product proper-
controllers usually employ linear dynamic models, a lin- ties such as octane number and Reid vapor pressure, and
ear program is used to evaluate the steady-state operat- the thermodynamics that determine the equilibrium in
ing targets. While linear models are useful for small separation processes. As pointed out [10], these nonlin-
regions around the current operating point, the target earities can cause the predictive control to target a
calculations over the operating range of most refineries steady-state operating point that is not optimal when the
involve significant nonlinearities. Examples of common second and third levels of the control and optimization
nonlinearities include tiered pricing discontinuities, con- hierarchy are collapsed into one level. When significant
nonlinearities exist, a nonlinear optimization program is
needed to compute the steady-state predictive control
target values.
When the feed or process changes that excite these
nonlinear process behaviors occur frequently, offline
optimization may not be sufficient to keep the operation
near the most cost-effective operation. In these cases, a
real-time optimizer (RTO) is often implemented to deter-
mine the optimal operating point under changing condi-
tions. RTO implementations make up the third level of
the control structure hierarchy shown in Figure 5. The
RTO is typically a nonlinear program minimizing cost or
maximizing profit subject to constraints derived from
steady-state mass and energy conservation balances,
reaction kinetic relationships, thermodynamic equilibri-
um equations, physical property relationships, and phys-
FIGURE 4 The ExxonMobil Torrance Refinery located in Torrance,
ical equipment constraints. The mass and energy
California. The vessels and piping comprise a complex network of
material and energy flows that convert crude oil into products. conservation balances relate energy requirements to pro-
(Reproduced with permission from Exxon Mobil Corporation and duction, permitting the value of production to be opti-
Joe Carson.) mized relative to the cost of energy.

78 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2006

 With tighter emission requirements on fuels and the changing world supply
of crude oil, more detailed methods for characterizing refinery feed
and product streams are needed to make the complicated decisions
required to control and optimize refinery operations.

A typical RTO can be characterized by a large number es. The impacts of operating strategies on blending compo-
of equations and variables with only a few decision vari- nents and products are evaluated using multiperiod linear
ables. The RTO dimensionality gets large with the number programs. Because of the broad scope of the refinery plan-
of balances needed to describe the mass and energy con- ning optimization application, decisions based on this appli-
servation in process equipment multiplied by the large cation occur on a day-to-week time scale. Multiperiod
number of components in the process streams. It is com- nonlinear solvers can also be applied to optimally schedule
mon to find RTO applications with less than 100 decision refinery maintenance activities that may occur only once
variables, more than 10,000 total variables, more than every several years [13].
100,000 constraint equations, and more than several hun- Finally, while not included in the control and optimiza-
dred thousand nonzero Jacobian-matrix entries. Since the tion hierarchy, refineries use PLCs to implement safety
underlying models are steady-state models, these applica- and protective systems. While refinery processes operate
tions operate on the same time
scale as the process steady-state
response time with often several
hours between implemented solu- Refinery-Wide Planning Optimization Product Values
Single- or Multiperiod Linear Program with Operating Costs
tions. The modeling complexity
Linear Model of Refinery Operation Inventories
and difficulties parameterizing Operating Limits
Executed Offline
these large models have motivated
some practitioners to develop (Often Manual)
methods for augmenting the non-
Real-Time Optimization
linear reactor kinetic models with Product Values
High-Dimension Steady-State Fundamental Operating Costs
the linear dynamic models used in
Mass/Energy Balance Models Operating Limits
the predictive controller to collapse Multiple-Hour Execution Period
the MPC and RTO levels into a sin-
gle application [12].
Finally, at the fourth level of the Multivariable Predictive Controllers Relative Product Values
hierarchy, refinery-wide planning Identified Models of Interactive and Operating Costs
and scheduling optimization is Multivariable Processes over Time Horizon Dynamic Response
used to determine optimal targets 1-min Execution Period Weightings/Penalties
for the individual refinery process-
ing units to create the pool of
hydrocarbon materials needed to Base Control System
meet product demand volumes and Flow, Pressure, Temperature, Analyzer Performance-Based
specifications. Traditionally, this Controllers Tuning Parameters
PID, Ratio, Cascade Algorithms
optimization is based on a linear
1-s Execution Period
program with a profit objective
function and constraints relating
product yield shifts to changes in
key operating variables. In addition Refinery Instrument System
to planning overall refinery opera-
tion, these linear programs are used
FIGURE 5 Control-system hierarchy illustrating how refineries simultaneously achieve control
to evaluate both capital-investment objectives on multiple time scales. Fast control actions are required to operate the refinery with-
decisions and crude oil and inter- in safe operating limits. Refinery-performance optimization for changes in market conditions,
mediate hydrocarbon feed purchas- feed quality, and equipment performance is applied at slower time scales.

DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 79

daily within safe operating envelopes, accidents might cally removed at the top of a common distillation column.
occur if an upset or disturbance moves the process outside In crude distillation, the towers have several side draws to
the safe operating envelope. Protective systems monitor produce oil fractions (known as cuts) that are different
process operation and, independent of the process control from the overhead and bottoms products. Also,
system, sequence the safe shutdown of processes when an pumparound circuits at various locations in the tower
excursion from safe operating limits is detected. These sys- remove energy from the tower. The crude oil feed is often
tems prevent the exposure of personnel to unsafe condi- preheated with energy transferred from the pumparounds.
tions and protect the refinery from large financial losses A fired-heater burning fuel gas or fuel oil is often used to
resulting from equipment damage. further heat the feed before it is charged to the bottom of
the tower.
CONTROL OF A REFINERY CRUDE UNIT The energy added to the feed by the preheat exchangers
The crude unit is normally two distillation towers that and fired heater provide the driving force for separating
operate at different pressures with downstream towers to the crude oil into different hydrocarbon cuts, as shown in
purify gas streams, as shown in Figure 6. The atmospheric Figure 7. The energy vaporizes the feed entering the tower.
tower runs just above atmospheric pressure, while the vac- This vapor travels up the distillation tower, contacting
uum tower operates under a vacuum allowing the heavier with the liquid flowing back down the tower. The liquid
molecules to boil at lower temperatures. A detailed flow down the tower is the vapor from the tower overhead
description of distillation and an introduction to distilla- condensed by transferring the heat of vaporization energy
tion control schemes is provided in [14]. to the air or cooling water through more heat exchangers.
One characteristic that separates refinery distillation in The contacting liquid and vapor streams equilibrate ther-
general and crude units in particular from other distilla- modynamically on each tray in the distillation tower, caus-
tion applications is the use of side draws and pumparound ing the molecules with the lower boiling points to move up
circuits. Typical distillation applications produce products the tower and those with higher boiling points to move
only from the top and bottom of the tower. Energy is typi- down the tower.

Cooling
Water
Gas
Crude Oil Plant

Top
Naphtha
Pumparound

Vacuum
System

Mid Kerosene
Pumparound

Light Vacuum
Gas Oil

Bottom Atmospheric
Pumparound Gas Oil
750 – Heavy Vacuum
800 °F Gas Oil

750 to
Steam 850°F Steam
Preheat Heater < 0.1 atm
Exchangers Heater
Atmospheric Coker Feed
Crude Tower Vacuum
Crude Tower

FIGURE 6 Typical crude unit process flow illustrating the two distillation units and the key heat exchangers, fired heaters, and pumparound
circuits. The refinery can trade off energy usage and separation quality by solving the multivariable control and optimization problems.

80 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2006

 The quality of the separation is measured by the loss of fidelity, the number of equations can be reduced to
amount of overlap between product cuts. In Figure 7, the about 100,000.
cut between the kerosene and the gas oil has little or no In crude unit operation, multiple models are needed
overlap because there is good separation. However, the to achieve the objectives at different control and opti-
heavy naphtha molecules overlap significantly with the mization levels. These models include a linear dynamic
light kerosene molecules from a poor separation. A good model for the model predictive controller, a steady-state
separation costs more in energy but results in higher quality fundamental model in the RTO application, and a
products, creating an optimization opportunity. steady-state linear model in the refinery-wide planning
In the atmospheric tower, the major manipulated vari- linear program (LP). While derivation of these individ-
ables are the feed rate, the furnace outlet temperature, the ual models from a single parent model could improve
tower pressure, the three pumparound flows, and the five consistency among the solutions, these models are typi-
product draw rates. The control objectives are to meet the cally created and maintained separately. Consistency
separation specifications of the downstream processing checks are needed to avoid conflicting model character-
units while minimizing energy costs and staying within istics that lead to oscillations between solutions found
regulatory and environmental limits. These limits include from each of the different models.
crude charge rate and furnace heating and emission limits.
Changing any of the draw rates impacts the quality of sep- CONTROL OF REFINERY CONVERSION UNITS
aration throughout the tower. For example, drawing more As discussed earlier, many of the refinery processes con-
kerosene pulls more heavy molecules, changing the liquid vert heavier less-valuable crude oil molecules into product
flow to the sections below the kerosene draw. Changing molecules. While reforming, isomerization, and alkylation
the liquid flow in any section of the tower changes the con- reactions are essential for modifying the quality or blend-
tacting between the liquid and vapor in the tower as well ing properties of different hydrocarbon molecules, crack-
as the equilibria throughout the tower, thus changing the ing reactions are critical for converting large molecules
quality of the separation throughout the tower. Changes to into smaller hydrocarbons that can be used in common
pumparound flows, however, affect only the separations fuels. Thermal cracking reactions that use heat and no cat-
above the pumparound because the pumparound removes alyst are still used with large molecules that do not lend
energy from the tower and puts it into the crude oil feed. themselves to catalytic cracking reactions. Catalytic crack-
This energy is no longer available higher up in the tower to ing in either fluidized catalytic crackers (FCCs) or hydroc-
effect better product separation. rackers is more popular because the catalyst can selectively
Due to the highly interactive and multivariable nature of produce the desired molecules. The differences in the reac-
the control problem and since the input-output responses for tor configurations result in interesting differences in the
the crude unit are often close to linear, model predictive con- control and optimization applications.
trollers are effectively used to control crude units.
When the crude oil feed changes composition
frequently or the refinery product demand and 1,200
pricing fluctuate, a predictive controller based
on linear models must compute targets that 1,000 Smaller Overlap
trade off the quality of separation for energy Better Separation
savings. For example, one could choose to 5%
Boiling Point (°F)

800 95%
increase a pumparound flow to reduce the fur-
nace fuel usage at the expense of overlap in a Overlap
600
product cut because the downstream require- Vacuum Tower Feed
5%
ments can be met at lower cost. In this case, real- 95% Gas Oil
400 5% 95%
time optimization can be used to characterize
the crude oil composition and model the separa- Kerosene
tion and heat transfer processes to compute tar- 200
Naphtha
gets and optimize unit profits.
The number of equations and variables in a 0
crude-unit RTO is a function of the number of 0 20 40 60 80 100
components in the crude oil, the number of sep- Crude Oil Fraction
aration stages in the towers, and the number of
heat exchangers. Without using model reduc- FIGURE 7 Distillation curves for the naphtha (green), kerosene (blue), and gas
oil (red) overlayed on a typical crude distillation curve. Significant overlap
tion, a typical crude-unit RTO is a nonlinear pro-
between the 95% point of the naphtha and the 5% point of the kerosene indi-
gram with about 500,000 equations for often less cates poor separation. Negligible overlap between the kerosene and the gas oil
than 20 decision variables. Without a significant indicates good separation.

DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 81

FLUIDIZED CATALYTIC the coke is burned off the catalyst, regenerating the cata-
CRACKER CONTROL AND OPTIMIZATION lyst. Cracking reactions in the FCC are endothermic, mean-
Fluidized catalytic cracking units are responsible for most ing that energy must be added to crack the hydrocarbons.
of the conversion of heavy oils into gasoline [8]. While This energy is generated in the process of burning off the
there are many FCC designs, a typical unit has a reactor carbon on the catalyst particles. The energy in the reactor
vessel and a regenerator vessel, followed by a main col- products is used to preheat the oil feed to reaction temper-
umn that separates the products, as shown in Figure 8. A atures. The primary manipulated variables are the reactor
fluidized catalyst stream is circulated between these two riser temperature, the catalyst-to-oil ratio in the reactor
vessels. Depending on the sulfur and nitrogen content, feed, air flow to the regenerator to control catalyst activity,
the gas oil feed may be hydrotreated before being fed to catalyst circulation rate to control the space velocity or
the FCC to prevent deactivating (also called poisoning) contact time, and reactor pressure. Additionally, the FCC
the catalyst. The gas oil feed stream is added to the cata- main column is a distillation tower similar to the atmos-
lyst stream flowing out of the regenerator into the reactor pheric tower in the crude unit. The control objectives are to
riser. The cracking reactions occur in the riser. The prima- maximize the conversion of the feed oil to hydrocarbons
ry reactions break paraffin, napthene, and aromatic mole- boiling at less than 220 ◦ C (430 ◦ F) without excessive over-
cules into smaller similar molecules, olefin molecules, cracking to light gaseous hydrocarbons and coke. The con-
and carbon called coke. The catalysts used for cracking trol system must also maintain adequate catalyst inventory
are subject to deactivation because of the generation of in both vessels and regulate the regenerator temperature to
coke. The coke produced in a FCC coats the catalyst sur- control the coke burn.
face, thus blocking access for reactions with other mole- A multivariable predictive control design is often used to
cules. Secondary isomerization and ring formation achieve these control objectives with the manipulated
reactions also occur. Depending on the operating condi- inputs. Gasoline production economics provide a basis for
tions, secondary cracking occurs, producing light hydro- selecting the controller tuning parameters. Since the regen-
carbon gas molecules and more coke. erator maintains stable catalyst activity, linear dynamic FCC
After leaving the riser, all of the hydrocarbons are MPC models do not change significantly over time. This fea-
stripped away from the catalyst with steam in the reactor. ture makes it unnecessary to re-identify models on a fre-
The catalyst then circulates back to the regenerator, where quent schedule to maintain good controller performance.

Cooling Water
Gas
Plant

Gasoline
Flue Gas

Reactor

Regenerator
Light Cat Gas Oil

Air
Riser
Heavy Cat Gas Oil

Fresh Catalyst
Gas Oil Spent Catalyst
Feed Cat Slurry Oil
Main Column

FIGURE 8 Simplified flow diagram of a fluidized catalytic cracking unit that converts gas oil into lighter hydrocarbons. The reactions occur in
the riser after the catalyst is mixed with the gas-oil feed. The products are separated from the catalyst in the reactor. The coke produced as
a by-product of the reactions is burned off the spent catalyst in the regenerator providing energy needed for the reactions in the riser. Multi-
variable predictive controllers are often used to control the mass and energy interactions among these vessels and the heat exchange
between main column and the gas-oil feed. Real-time optimizers are often used to account for the nonlinear reaction kinetic effects.

82 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2006

 Real-time optimizers are often used on FCC units. The optimization. The examples presented highlight progress
cracking reaction rates are sensitive to reactor riser temper- toward capturing the highest value of every molecule at
ature, following an exponential Arrhenius expression. every point in the refining complex.
Higher reactor riser temperature results in higher conver-
sion and more preheat energy. Higher riser temperature AUTHOR INFORMATION
also leads to higher reactor temperature and more coke to Robert E. Young ([email protected]) is the
burn to maintain catalyst activity, and potentially more advanced control section supervisor at ExxonMobil’s Tor-
over-cracking of valuable gasoline molecules to light gas rance Refinery in Torrance, California. He received his B.S.
and coke. Because of these nonlinear temperature depen- degree in chemical engineering from the University of Texas,
dencies, determining the optimal operating conditions Austin, and his M.S. and Ph.D. degrees from University of
requires a nonlinear program such as a RTO [10]. The California, Santa Barbara. He has led and implemented
development of a fundamental model for an FCC RTO process control projects in the areas of linear and nonlinear
based on the molecular structure of the feeds and products multivariable predictive control, polymerization reactor con-
is given in [15]. The dimension of the FCC model can be trol, process control system replacement, and manufacturing
reduced by assuming that some of the FCC vessels contain execution systems. He is an active AIChE Computing and
well-mixed homogeneous mixtures. However, even System Technology (CAST) division member, currently serv-
though the total number of components is less than the ing as a 2007 meeting program coordinator and a division
number of components in a crude oil stream, model reduc- director, as well as a member of IEEE and SIAM. He can be
tion is required to reduce the model dimension for success- contacted at ExxonMobil Refining and Supply—Torrance
ful FCC RTO applications. The RTO application has the Refinery, 3700 W 190th St., Torrance, CA 90504 USA.
fidelity to obtain the information needed to trade off crack-
ing conversion and energy to obtain an optimal yield for a REFERENCES
given product demand at the lowest cost. [1] D. Kushnerick and C. Kennedy, “Application of compositional modeling
in molecule management,” presented at NPRA Plant Automation and Deci-
This same strategy can be applied to the other refinery sion Support Conf., PD-04-191, San Antonio, TX, 2004.
conversion units. Hydrocrackers can crack heavier vacuum [2] “Staying the course… Yesterday, Today, and Tomorrow,” Exxon Mobil
tower gas oil and more aromatic gas oils from the FCC and Investors Meeting, Exxon Mobil Annu. Rep., 2001.
[3] Energy Information Administration, Office of Oil and Gas, U.S. Dept. of
coker. A coker is a thermal cracking process unit that con- Energy, “U.S. Refinery yield” Mar. 15, 2006 [Online]. Available:
verts oil from the vacuum tower bottom into coker naph- http://tonto.eia.doe.gov/dnav/pet/pet_pnp_pct_dc_nus_pct_a.htm
tha, coker gas oil, and coke. Unlike the FCC, because [4] K.W. Gardner, “U.S. gasoline requirements map,” presented at Exxon
Mobil Fuels Marketing, Fairfax, VA, Jan., 2005.
hydrocrackers are fixed-bed reactors, hydrocracker models [5] Office of Transportation and Air Quality, “Study of unique gasoline fuel
cannot be spatially lumped, resulting in significantly blends (“boutique fuels”), effects on fuel supply and distribution and poten-
increased model dimensionality. With judicious model tial improvements,” U.S. Environmental Protection Agency, Washington
DC, Oct. 2001. [Online]. Available: http://www.epa.gov/otaq/
development and model reduction methods, hydrocracker regs/fuels/p01004.pdf
models can be developed for use with the same control [6] “NPRA United States refining and storage capacity report,” National
and optimization structure used in the FCC application. Petrochemical and Refiners Association, Washington, DC, Jan. 2005.
[Online]. Available: http://www.npra.org/publications/statistics/
Cyclic processing and complex thermal cracking reactions RC2005.pdf
have limited RTO applications for cokers. However, MPC [7] Energy Information Administration, Office of Oil and Gas, U.S. Depart-
can be applied in both applications to improve the perfor- ment of Energy, “Refinery capacity data historical” May 13, 2005 [Online].
Available: http://www.eia.doe.gov/oil_gas/petroleum/data_publica-
mance of these other conversion units. tions/refinery_capacity_data/refcap_historical.html
[8] J.H. Gary and G.E. Handwerk, Petroleum Refining, Technology and Econom-
CONCLUSIONS ics. New York: Marcel Dekker, 2001.
[9] 2004 Refining Processes Handbook, Hydrocarbon Processing. Houston, TX:
With tighter emission requirements on fuels and the Gulf Publishing, 2004.
changing world supply of crude oil, more detailed meth- [10] O. Rotava and A.C. Zanin, “Multivariable control and real-time
ods for characterizing refinery feed and product streams optimization—An industrial practical view,” Hydrocarbon Process., vol. 84,
no. 6, pp. 61–71, June 2005.
are needed to make the complicated decisions required to [11] S.J. Qin and T.J. Badgwell, “An overview of industrial model predictive
control and optimize refinery operations. Changes in ana- control technology,” in Chemical Process Control-V, J.C. Kantor, C.E. Garcia
lytical measurement technology provide additional data and B. Carnahan, Eds. Tahoe, CA: AIChe, 1997, pp. 232–256.
[12] J. Jackson, M. Piatt, C. Timmons, P. Fountain, and R. Wagler, “Integra-
for fingerprinting the molecular structure of these hydro- tion of off-line process models with on-line control and optimization to
carbon mixtures. Integrating this molecular characteriza- improve process performance,” in Proc. 2002 Computer Conf, National Petro-
tion of the materials with advances in compositional chemical and Refiners Association, Austin, TX, Nov. 2002, paper CC-02-14.
[13] V. Bizet, I. Grossmann, and N. Juhasz, “Optimal production and sched-
modeling methods, model reduction techniques, optimiza- uling of a process with decaying catalyst,” AIChE J., vol. 51, no. 3,
tion, and equation-solving methods, along with continued pp. 909–921, 2005.
implementation of model predictive control, improves [14] H. Kister, Distillation Operation. New York: McGraw-Hill, 1990.
[15] G. Christensen, M. Apelian, K. Hickey, and S. Jaffee, “Future directions
operational effectiveness at all levels of the refinery from in modeling the FCC process: An emphasis on product quality,” Chem. Eng.
the individual process level to the refinery-wide planning Sci., vol. 54, no. 13-14, pp. 2753–2764, 1999.

DECEMBER 2006 « IEEE CONTROL SYSTEMS MAGAZINE 83