Professional Documents
Culture Documents
C
hemical process control textbooks and courses differ reactor in another facility. Process operations management
significantly from their electrical or mechanical-ori- philosophy, plant control hardware and software, process
ented brethren. It is our experience that colleagues in engineering structure, sensor selection and maintenance,
electrical engineering (EE) and mechanical engineering (ME) and the analytical laboratory vary substantially from plant
assume that we teach the same theory in our courses and to plant. These issues lead to virtually an entirely new con-
merely have different application examples. The primary trol system development for each process, a factor that sig-
goals of this article are to i) emphasize the distinctly chal- nificantly influences process control practice and hence,
lenging characteristics of chemical processes, ii) present a indirectly, process control education.
typical process control curriculum, and iii) discuss how The common characteristics that make chemical pro-
chemical process control courses can be revised to better cesses so challenging to control are noted in papers pre-
meet the needs of a typical B.S.-level chemical engineer. sented at most control research conferences and in
In addition to a review of material covered in a standard countless research proposals. It is worth reviewing these
process control course, we discuss innovative approaches problems here to understand if we are getting the major
in process control education, including the use of case stud- points across to our undergraduates. Chemical processes
ies, distributed control systems in laboratories, identifica- are usually high order, nonlinear, with multiple inputs and
tion and control simulation packages, and studio-based outputs; they have time delays, input constraints, and a lim-
approaches combining lecture, simulation, and experi- ited number of measured states. The desired properties of a
ments in the same room. We also provide perspectives on product stream are often not directly measured, so inferen-
needed developments in process control education. tial control is important. Economic objectives are domi-
nated by steady-state considerations. Large-scale processes
Chemical Engineering Curricula
are often energy integrated, causing a high degree of interac-
Chemical engineering curricula across the United States are
tion between inputs and outputs of different process units.
relatively uniform for several reasons: departmental history
Specialty chemicals and pharmaceuticals are often pro-
and culture, Accreditation Board of Engineering and Tech-
duced in batches, frequently with a single vessel serving
nology (ABET) accreditation requirements, and the success
more than one function (heater, reactor, and separator, for
of alumni in industry. Standard courses include material and
example). The same temperature controller may be re-
energy balances, thermodynamics, equilibrium stage sepa-
quired to provide cooling under some conditions and heat-
rations, transport phenomena, chemical reaction engineer-
ing under others. Often robustness, rather than nominal
ing, process dynamics and control, process design, and
performance for any particular operating condition, be-
chemical engineering laboratory. Virtually every topic cov-
comes the prime consideration.
ered in these chemical engineering curricula assumes
The proportional-integral-derivative (PID) controller is
steady-state process operation—the sole exception is the
dominant in the chemical process industry and will remain
process dynamics and control course, which must therefore
so for many reasons. One is that lower-level loops, such as
accept the burden of introducing all topics associated with
flow control, are adequately controlled by PID action. Also,
the dynamic behavior of process systems.
no explicit process model is required for tuning the two or
Characteristics of Chemical Processes three controller parameters; many commercial PID control-
A major difference between process control and device con- lers have autotuning algorithms. Cascade control is preva-
trol is the replicability of control system designs. For exam- lent, since most higher level loops cascade a setpoint to a
ple, a disk drive manufacturer can perform a single flow control loop. Feedforward and ratio control are used in
advanced control system design for this device and use that well-studied unit operations. Distributed control systems
controller in thousands of units. Each chemical process con- (DCSs) are the norm, although the hardware/communica-
trol system design project, on the other hand, tends to be tion structure is significantly different from the systems of
unique. A styrene polymerization reactor in one manufac- the late 1970s and 1980s. Most loops are sampled at a high
turing plant may have feedstocks, flow patterns, and prod- frequency relative to the process dynamics, so continuous
uct specifications that differ significantly from a similar control system design procedures can easily be used.
Bequette (bequette@rpi.edu) is with The Howard P. Isermann Department of Chemical Engineering, Rensselaer Polytechnic Institute, Troy,
NY 12180-3590, U.S.A. Ogunnaike is with DuPont Central Research and Development, Wilmington, DE 19880-0101, U.S.A.
0272-1708/01/$10.00©2001IEEE
10 IEEE Control Systems Magazine April 2001
Cascade control is worthy of further dis-
cussion, since the approach does not appear Reactor
Temperature
to be well known in other disciplines. An ex- Controller
ample of a double cascade control strategy
Tjsp TCI
to regulate temperature of a chemical reac- Reactor Feed
tor is shown in Fig. 1; the corresponding Jacket
block diagram is shown in Fig. 2. The domi- Temperature
nate time constant for the flow control loop Controller
is a few seconds, the jacket temperature loop TC2 Coolant
is a few minutes, while the reactor tempera- Return
Fsp
ture may be several minutes to several hours
(particularly for a polymerization reactor). Jacket
Notice that each control loop rejects differ- Flow Recirculation
ent disturbances. The flow control loop re- Controller Pump Reactor Product
FC
jects coolant header pressure disturbances v
and compensates for valve nonlinearities. F
The jacket temperature control loop rejects
coolant header temperature disturbances.
The reactor temperature controller rejects Coolant Make-Up
reactor feed, temperature, and concentra-
tion disturbances and compensates for Figure 1. Double cascade control strategy to regulate temperature in a jacketed,
changes in the rate of heat transfer due to stirred tank reactor.
fouling, etc. This approach has many of the
benefits of a state feedback strategy used in other disci- dated and a new constrained optimization problem is
plines, without sensitivity to model uncertainty. solved at time step k+1. Multivariable systems with con-
The most commonly used advanced control scheme is straints and time delays are handled naturally by MPC. MPC
model predictive control (MPC). The basic idea behind MPC has been particularly successful in the petroleum refining
is illustrated for a single-input, single-output process in Fig. industry where large-scale, interacting, constrained sys-
3. Here, an open-loop optimal control problem is solved at tems are the norm. Time constants and sample times are
time step k. The least-squares objective function to be mini- large, so computation time to solve large-scale constrained
mized is based on the residuals between the model predic- systems is not an issue. When linear models and quadratic
tions and the desired setpoint profile over a horizon of P objective functions are used, the optimization problem re-
time steps. The decision variables are the next M control sults in a quadratic program (QP); there are a number of ro-
moves; note that the control moves can be constrained. bust QP codes available. For a tutorial overview of MPC, see
Only the first control move is actually implemented and the Rawlings [1]. Again, we should stress that it is common that
next process measurement is obtained. The model is up- flow rates are the manipulated inputs used by MPC strate-
Coolant
Flow Temperature Reactor
Disturbance Disturbance Disturbance
Tjsp Fsp F Tj
Tsp + + +
TC1 TC2 FC FP JP RP
+ + +
− − −
Flow Process Recirculating Reactor
Jacket Fluid Process
Measured Flow Process
Graduate Education
The focus of this article has been on undergraduate educa-
tion. Since most chemical engineering departments have a
single faculty member with expertise in systems and con-
trol, rarely is more than one graduate-level process control
Figure 5. LabView interface for the Rensselaer prototype
process control experiment. Each control loop can be placed in course taught on a frequent basis. The available selection
either manual or automatic (PID) mode. of graduate-level process control textbooks is limited [32]-
[34]. Graduate students conducting research in process
ing lecture and discussion periods and swivel in their chairs control generally take several systems and control theory
to perform simulations and conduct experiments on the courses in electrical engineering departments. Special top-
countertops behind them. During the problem-solving peri- ics in chemical process control are normally covered in
ods, the instructor and teaching assistant move around the course notes and instructor handouts. MPC is probably the
room answering questions and generating discussion. Most most covered special topics process control course; several
of the problems have been solved in two-person groups; MPC textbooks/monographs are currently in preparation.
however, the space could handle a group size of three. Al- Nonlinear control is probably the next most widely taught
though it is conceivable that 20 copies of a single experi- special topics course. A monograph on nonlinear process
ment could be used so that all groups are working on the control has been published [35] but does not appear to be
same problem, this is not economically attractive. It is more widely used in these courses. As plantwide control begins to
attractive to have roughly five copies of an experiment; receive more attention, the monograph by Luyben et al. [36]
groups working with an experiment during one period may will probably be the text of choice.
be doing simulations or a detailed design project during the It is widely recognized that graduate students need more
next period. practical control experience, so there is a move to develop
A prototype chemical process control experiment, experiments for graduate control courses. An example from
shown in Fig. 4, mimics the behavior of a typical chemical a graduate-level multidisciplinary control laboratory at the
process. Fresh feedwater, regulated with a control valve, University of Delaware is presented by Gatzke et al. [37].
flows into a vessel containing an electric heater. A concen-
trated salt solution from a reservoir then mixes with the Summary
heated feedwater in a mixing tank that contains a tempera- The primary purpose of this article is to provide a sum-
ture probe. The outlet from the tank discharges through a mary of chemical process control education and practice
conductivity sensor into a sink. The objective of the experi- for our colleagues in other engineering disciplines. We
ment is to regulate three measured process variables have presented a typical process control curriculum, out-
(level, temperature, and conductivity) at desired setpoint lined some of the distinctly challenging characteristics of
values by manipulating three input variables (freshwater chemical processes, and discussed recent and ongoing de-
flow rate, concentrated salt solution flow rate, and heater velopments in process control education. We consider
power) via feedback control. The experimental apparatus control education to be an area where “continuous im-
is benchtop scale (with a “footprint” of roughly 3 ft2), so provement” is important and look forward to discussions
that it can be used in the studio classroom. The experiment based on this article and education sessions at future con-
was designed to have time constants that are roughly 20-30 trol conferences.
s; the time scale is slow enough for students to observe the
physical changes, yet fast enough for a number of experi- Acknowledgment
ments to be conducted during an interactive session. Na- Support from a Curriculum Development grant from Procter
tional Instruments hardware and software (LabVIEW) is & Gamble is gratefully acknowledged.