ENCH 617 – MODELLING AND IDENTIFICATION FOR

ADVANCED CONTROL

1 BASIC CONCEPTS AND TERMINOLOGY

The primary focus of our work will be upon the design of process control systems for
chemical plants. The goal of the analytical methods we shall discuss is to enable us to
design, and ultimately to build, equipment that functions according to the performance
requirements specified by the customer when the project was initiated. Before introducing
some of the main concepts and methods, the term design needs to be discussed since no
single definition has achieved wide acceptance. Here, we will simply describe the design
function relevant to process control, without any claim to generality (Marlin, 2000):
Design is the procedure by which an engineer arrives at a complete control system
specification that satisfies all performance objectives.

It is important to recognize that the performance specifications are determined by the

engineer at the first stage of the design procedure based on physics, chemistry and the
marketplace, which defines product quality, demands and economics. The initial
objectives are specified independently of the solutions possible. The result of the design
is a complete specification that satisfies the objectives, if possible. The simple flash
process in Fig. 1.1 is shown as a typical starting point for the design procedure. During the
design the engineer must (a) clearly state the objectives, (b) determine needed process
modifications, (c) specify instrumentation performance and alterations, (d) define control
structure (loop pairing), (e) select algorithms and (f) define special tuning requirements.

Figure 1.1 - Example flash process.

Five important features of control system design are now discussed which distinguish it
from the topics normally treated in introductory process control courses (Marlin, 2000).
The first is the rich definition of the objectives or performance that the design is to satisfy.
Thus, design involves considerable interaction between the objective statement and design
results. The objective is usually stated to be the reduction of variability in the operation of
a process plant. However, not all variability can be eliminated, and variability is much
more important in some variables than in others. In fact, the plant is designed to provide
specific variables and systems that can be easily adjusted with minimum effect on plant
performance. For example, the cooling water, steam, electricity and fuel systems are
designed to be able to respond rapidly to demands in the plant. Thus, process control
generally moves variability from important variables to less important variables. This is
achieved by controlling the (important) controlled variables by adjusting the (less
important) manipulated variables. Therefore, the control design must conform to the
priority of variables indicated in the objective statement.

A second major feature is the large number of process design decisions that can be
considered. It is well-known fact that process dynamics have a major effect on control
system performance. Thus, process design changes would be the preferred manner for
achieving good control. When a plant is being designed initially, the engineers can make
essentially any design changes, although equipment design modifications to achieve good
dynamic performance may be prohibitively expensive when compared to the alternative of
additional instrumentation and control algorithms. Major process equipment options
require a thorough safety, reliability and cost analysis of the alternatives.

A third feature of design is the iterative nature of the decision-making process. Initial
decisions may place limitations on future decisions, and the limitations may not be easily
predicted when the initial decisions are made. Thus, the engineer must be ready to rethink
previous decisions and be willing to iterate by changing some decisions and repeating the
design.

A fourth feature is the ambiguity in determining the conclusion of the design procedure.
One would have to evaluate most or all possible designs to be sure that the final design is
the best. To respond quickly to market demands and limit total cost, the time for design is
limited and judgement must be used in deciding when the design is “good enough”. The
typical procedure is to develop approximate bounds on the achievable performance and
find a low-cost design that approaches the best performance.

Also, situations arise in which the initial objectives lead to unacceptable designs that are
very costly and unreliable; in such cases, it is the engineer’s task to alter the objective
statement to meet the initial intent (e.g., make high-quality product safely), thereby
preventing an unsatisfactory design. In fact, very restrictive objectives may not be

2
achievable in the situation defined. For example, for specified disturbances in the feed
composition and flow rate and available sensors, it may not be possible to control the
product quality of a chemical reactor. Clearly, a major change in the process design or
performance specification is required.

A fifth feature of process control design is the concurrent application of process

engineering and automatic control technologies. Automatic control principles may
indicate that the feedback dynamics of a chemical reactor should be faster; then chemical
engineering principles can be used to select a process change (e.g., increasing the
temperature or relocating the sensor). This tight coupling of process and control is the
main reason why chemical engineers must learn control and why a “control specialist”,
without understanding of the process, cannot adequately perform the control design tasks.

It must be said at this point that control design – in fact, all engineering design – is very
challenging and requires considerable practice to master. Topics covered in the
undergraduate process control course and other core elements of the chemical engineering
curriculum can be learned quickly because they presuppose limited prior knowledge and
typically involve relatively straightforward analyses. The design engineer has to master
and apply all of these technologies concurrently. Adding to the challenge is the lack of a
single, structured procedure for control design. This is to be expected, because design
involves an element of creativity in adding process or control equipment, altering
objectives and specifying control structures. Procedures for stimulating creativity cannot
be reduced to a flowchart. However, much can be presented and learned about the design
procedure. Certainly, general procedures can be applied to the tasks of collecting
information, defining objectives and evaluating common checklists of potential decisions
and outcomes. Also, typical sequences for considering control design decisions can be
explained, although the best sequence is problem-dependent. Finally, examples
demonstrating the interplay between process and control technology help the new engineer
learn how to design. The material presented in the following sections is intended to provide
some guidance on designing advanced strategies for the optimal, constrained control of
multivariate chemical processes.

1.1 Process Control: General Concepts and Terminology

It may be useful at this point to define some broad and general concepts and some of the
terminology used in the process control field (Luyben, 1990; Luyben and Luyben, 1997):

(i) Dynamics: Time-dependent behaviour of a process. The dynamic behavior with no

controllers in the system is called the open loop response. The behavior with
controllers included with the process is called the closed loop response. The open
loop time constant is a parameter which generally refers to the time required for the
process variable to accomplish 63% of its eventual change following a step change
in the controller output (e.g. valve position) with the controller operating in manual.

3
(ii) Variables:
 Manipulated Variables (MVs): Typically flow rates of streams entering or
leaving a process that we can change to control the plant.
 Controller Output (U): The signal sent by the controller to the ‘final control
element’, which in Chemical Process Control applications is almost always a flow
control valve. The units of the controller output U(t) are % in the process control
computer then are linearly converted to 4-20 mA in the D/A converter and finally
to 3-15 psig in the I/P converter. Virtually all flow control valves are pneumatic.
 Controlled Variables (CVs): Flow rates, compositions, temperatures, levels and
pressures in the process that we will try to control, either trying to hold them as
constant as possible or trying to make them follow some desired time trajectory.
 Measured or Process Variable (Y or PV): Measurement of the controlled variable
with engineering units of percent of sensor calibration scale.
 Setpoint (Ysp): The desired or target value of the measured variable, also in units
of %.
 Uncontrolled Variables: Variables in the process that are not controlled.
 Load Disturbances (DVs): Flow rates, temperatures or compositions of streams
entering (but sometimes leaving) the process. We are not free to manipulate them.
They are set by upstream or downstream parts of the plant. The control system
must be able to keep the plant under control despite the effects of these
disturbances.

(iii) Feedback Control: The traditional way to control a process is to measure the variable
that is to be controlled, compare its value with the desired value (the setpoint to the
controller), and feed the difference (the error) into a feedback controller that changes
a manipulated variable to drive the controlled variable back to the desired value (see
Fig. 1.2). Information is thus “fed back” from the controlled variable to a
manipulated variable. Action is taken after a change occurs in the process.
DV
(e.g. feed temperature, Load
feed flowrate, etc.)

TC
YSP
+ CV
+ +
(e.g. desired outlet Controller Valve Plant (e.g. hot oil
temperature, %) - temperature, C)

U MV
e
(e.g. valve stem (e.g. heat input rate, kJ/min,
(control error, %)
position, %) or steam flowrate, kg/min)
Y
(e.g. measured outlet
temperature, %) Sensor/Transmitter

TT

Figure 1.2 - Feedback control block diagram.

4
(iv)Feedforward Control: The basic idea is to take action before a disturbance reaches the
process. As shown in Fig. 1.3, the disturbance is detected as it enters the process and an
appropriate change is made in the manipulated variable such that the controlled variable is
held constant. Thus, we begin to take corrective action as soon as a disturbance entering
the system is detected instead of waiting (as we do with feedback control) for the
disturbance to propagate all the way through the process before a correction is made.

Figure 1.3 – Feedforward control.

(v) Stability: A process is said to be unstable if its output becomes larger and larger
(either positively or negatively) as time increases. Examples are shown in Fig. 1.4.
No real system actually does this, of course, because some constraint will be met; for
example, a control valve will completely shut or completely open, or a safety valve
will “pop”. A linear process is right at the limit of stability if it oscillates, even when
undisturbed, and the amplitude of the oscillations does not decay.

Figure 1.4 - Stability.

5
Most processes are open-loop stable or self-regulating; i.e., stable with no controllers on
the system. Fortunately, almost chemical unit operations are self-regulating. The most
common example of a non-self-regulating process is liquid level in a surge drum. Certain
chemical reactors, e.g. those in which an irreversible exothermic reaction is taking place,
may be prone to temperature runaway and are therefore open-loop unstable. Virtually all
real processes can be made closed-loop unstable (unstable when a feedback controller is in
the system) if the controller gain is made large enough. Thus, stability is of vital concern
in feedback control systems.

1.2 Process Control Laws

Several fundamental laws have been developed in the process control field as a result of
many years of experience. Some of these may sound similar to some of the laws attributed
to Parkinson, but the process control laws are not intended to be humorous.

First Law: The best control system is the simplest one that will do the job.

Complex and elegant control systems look good on paper but soon end up on “manual”
(taken out of service) in an industrial environment. Bigger is definitely not better in control
system design.

Second Law: You must understand the process before you can control it.

No degree of sophistication in the control system (from adaptive control, to expert systems,
to Kalman filters, to nonlinear model predictive control) will work if you do not know how
your process works. Many people have tried to use complex controllers to overcome
ignorance about the process fundamentals, and they have failed! Learn how the process
works before you start designing its control system.

Third Law: Liquid levels must always be controlled.

The structure of the control systems must guarantee that the liquid levels in tanks, column
base, reflux drums, etc. are maintained between their maximum and minimum values. A
common error is to develop a control structure in which tank levels are not controlled and
to depend on the operator of the plant to control tank levels manually. This increases the
workload on the operator and results in poor plant performance because of inconsistencies
among various operators concerning what should be done under various conditions.
Having an automatic, fixed inventory control structure produces smoother, more consistent
plant operation.

6
1.3 Levels of Process Control
There are four levels of process control. Moving up these levels increases the importance,
the economic impact, and the opportunities for process control engineers to make
significant contributions. The lowest level is controller tuning; i.e., determining the values
of controller tuning constants that give best control. The next level is algorithms – deciding
what type of controller to use (P, PI, PID, multivariable, model predictive, etc.).

The third level is control system structure – determining what to control, what to
manipulate, and how to match one controlled variable with one manipulated variable
(called “pairing”). The selection of the control structure makes it easy to select an
appropriate algorithm and to tune. No matter what algorithm or tuning is used, it is very
unlikely that a poor structure can be made to give effective control.

The top level is process design – developing a process flowsheet and using design
parameters that produce an easily controllable plant. The steady-state economically
optimal plant may be much more difficult to control than an alternative plant that is perhaps
only slightly more expensive to build and operate. At this level, the economic impact of a
good process control engineer can be enormous, potentially resulting in the difference
between a profitable process and an economic disaster. Several cases have been reported
where the process was so inoperable that it had to be shut down and the equipment sold to
the junk man. Chapter 10 of Svrcek et al. (2006) discusses this vitally important aspect in
more detail.

Example 1.1:
Figure 1.5 is a simplified process flow diagram (PFD) describing a process configuration
and its control system. (The attached tables from Appendix A of Ogunnaike and Ray
(1994) define the symbols commonly employed in PFDs as well as process and
instrumentation diagrams or P&IDs.) Two liquid feeds are pumped into a reactor, in which
they react to form products. The reaction is exothermic, and therefore heat must be
removed from the reactor. This is accomplished by adding cooling water to a jacket
surrounding the reactor. The reactor effluent is pumped through a preheater into a
distillation column that splits it into two product streams.

Traditional steady-state design procedures are used to specify the various pieces of
equipment in the plant:

Fluid mechanics: pump heads, rates, and power; piping sizes; column tray layout
and sizing; heat-exchanger tube sizing and shell side baffling and sizing.
Heat Transfer: reactor heat removal; preheater, reboiler, and condenser heat
transfer areas; temperature levels of steam and cooling water.

7
 Chemical kinetics: reactor size and operating conditions (temperature, pressure,
catalyst, etc.)
Thermodynamics and mass transfer: operating pressure, number of plates and
reflux ratio in the distillation column; temperature profile in the column;
equilibrium conditions in the reactor.

But what procedures do we use to decide whether and how to regulate this plant via Model
Predictive Control (MPC)? The remainder of our studies will be aimed at understanding
the dynamics of the process and control systems so that we can develop and design plants
that operate more efficiently and safely, produce higher-quality products, are more easily
controlled, and are more environmentally friendly.

Even in this simple plant, with a minimum of instrumentation, 10 control loops are
required. We will find that most chemical engineering processes are multivariable. The
key to any successful control system is understanding how the process works.

Figure 1.5 - Typical chemical plant and regulatory control system.

8
1.4 Feedback Control Algorithms
The part of the control loop on which we will be spending the remainder of this chapter is
the controller. The job of the controller is to compare the process signal from the
transmitter with the setpoint signal and to send out an appropriate signal to the control
valve. In this section we describe what kinds of action standard commercial controllers
take when they detect a difference between the desired value of the process variable (the
setpoint) and the actual value.

Analog controllers use continuous electronic or pneumatic signals. The controllers see
transmitter signals continuously, and control valves are changed continuously. Digital
computer controllers are discontinuous in operation, looking at a number of loops
sequentially. Each individual loop is polled every sampling period. Choice of the sampling
period is an important engineering decision which will be discussed in Chapter 3. The
analog signals from transmitters must be sent through analog-to-digital (A/D) converters
to get the information into the computer in a form that it can use. After the computer
performs its calculations in some control algorithm, it sends out a signal that must pass
through a digital-to-analog (D/A) converter and a “hold” that sends a continuous signal to
the control valve. These sampled-data systems are covered in detail in Part VII of
Stephanopoulos (1984) and in Chapters 14 and 15 of Luyben (1990).

Three basic types of controllers are commonly used for continuous feedback control. The
details of construction of the analog devices and the programming of the digital devices
vary from one manufacturer to the next, but their basic functions are essentially the same.

1.4.1 Proportional Action

A proportional-only feedback controller changes its output signal, U, in direct proportion
to the error signal e, which is the difference between the setpoint signal Ysp and the process
variable signal Y coming from the A/D converter.

U (t )  bias  KC (Y sp (t )  Y (t )) (1.1)

The engineering units of U, Y and Ysp are percent of scale. The bias signal is a constant
(percentage units also) and is the value of the controller output when there is no error. A
more practical interpretation of the bias is that it is the last manually-entered value of the
controller output before the controller was switched from manual to automatic mode.
(When a controller is on manual, the console operator directly specifies the controller
output signal, U. When it is in automatic mode, the operator enters the setpoint and the
corresponding controller output is automatically calculated using an equation such as
(1.1)). KC is called the controller gain or proportional gain. For example, if the gain is 1,
an error of 10% of scale will change the controller output by 10% of scale. Figure 1.6a
sketches the action of a proportional controller for given error signals e.

9
Figure 1.6 – Action of a feedback controller: (a) Proportional; (b) Integral; (c) Ideal
derivative.

Some instrument manufacturers use an alternative term, proportional band (PB), instead
of gain. The two are related by

10
 100%
PB  (1.2)
KC

The higher or “wider” the proportional band, the lower the gain, and vice versa. Thus, a
wide PB is a low gain, and a narrow PB is a high gain. Some control engineers prefer to
use proportional band because it has a straightforward physical meaning which K C lacks:
For proportional-only controllers, it is the percentage amount by which Y has to change in
order to cause U to move from 0 to 100% or 100% → 0%. In other words, the band is the
change in measurement (% of sensor calibration scale) needed to cause the valve stem to
move “full-stroke”. The proportional gain KC is always positive and, as can be deduced
from (1.1), is dimensionless, i.e. possesses engineering units of % / %.

The control engineer must specifying the correct action for the controller by choosing the
appropriate sign in a digital controller (see Eq. (1.1)). A positive sign results in the
controller output decreasing when the process variable increases. This increase-decrease
action is characteristic of a reverse-acting controller. With a negative sign, the controller
output increases when the process variable increases, and this implies a direct-acting
controller. The correct sign depends upon the action of the transmitter (which is usually
direct), the action of the valve (air-to-open or air-to-close), and the effect of the
manipulated variable on the controlled variable. Each loop should be examined closely to
make sure the controller gives the correct action.

For example, suppose we are controlling the process outlet temperature of a heat exchanger
as sketched in Fig. 1.7. A control valve on the steam to the shell side of the heat exchanger
is manipulated by a temperature controller. To decide what action the controller should
have we first look at the valve. Since this valve puts steam into the process, we should
want it to fail shut. Therefore, we choose an air-to-open (AO) control valve.

Steam Supply,
q(t)

U
TC Ysp

TT

Cool Oil Inlet Heat Hot Oil Exit

F, Ti(t) Exchanger F, T(t)

Trap

Figure 1.7 - Steam heat exchanger.

11
Next we look at the temperature transmitter. It is direct-acting (when the process
temperature goes up, the transmitter output signal, Y goes up). Now if Y increases, we want
to have less steam. This means that the controller output must decrease since the valve is
AO. Thus, the controller must be reverse-acting and employ the positive sign in Eq. (1.1).

If we were cooling instead of heating, we would want the coolant flow to increase when
the temperature increased. But the controller action would still be reverse because the
control valve would be an air-to-close valve, since we want it to fail wide open.

As a final example, suppose we are controlling the base level in a distillation column with
the bottoms product flow rate. The bottoms product control valve would be AO because
we want it to fail shut and prevent the loss of base level in an emergency. (A consideration
of lesser importance is that damage to the bottoms pump would be avoided if it does not
run dry.) The level transmitter signal increases if the level increases. If the level goes up,
we want the bottoms flow rate to increase. Therefore the base level controller should be
“increase-increase” (direct-acting).

1.4.2 Integral Action (Reset)

Proportional action moves the control valve in direct proportion to the magnitude of the
error. Integral action moves the control valve based on the time integral of the error, as
sketched in Fig. 1.6b.
t
KC
U (t )  bias 
TI  e(t ) dt
0
(1.3)

where TI is the integral time or reset time with units of minutes.

If there is no error, the controller output does not move. As the error goes positive or
negative, the integral of the error drives the controller output either up or down, depending
on the action (reverse or direct) of the controller. Most controllers are calibrated in minutes
(or minutes/repeat, a term that comes from the test of putting into the controller a fixed
error and seeing how long it takes the integral action to ramp up the controller to produce
the same change that a proportional controller would make when its gain is 1; the integral
thus repeats the action of the proportional controller).

The basic purpose of integral action is to drive the process back to its setpoint when it has
been disturbed. A proportional controller would not usually return the controlled variable
to the setpoint when a load or setpoint disturbance occurs. This permanent error (Ysp – Y)
is called the steady-state error or offset. Integral action reduces the offset to zero.

12
Integral action usually degrades the dynamic response of a control loop. It makes the
control loop more oscillatory and moves it toward instability. Nevertheless, integral action
is usually needed if it is desirable to have zero offset.

1.4.3 Derivative Action

The purpose of derivative action (also called rate or pre-act) is to anticipate where the
process is heading by looking at the time rate of change of the measured variable (its
derivative). See Fig. 1.6c. If we were able to take the derivative of the error signal (which
we cannot do perfectly in practice), we would have ideal derivative action:

d e(t )
U (t )  bias  K C TD (1.4)
dt

where TD is the derivative time (minutes). In theory, derivative action should always
improve dynamic response, and it does in many loops. In others, however, the problem of
noisy signals (fluctuating Y) makes the use of derivative action undesirable.

1.4.4 Proportional-Integral-Derivative (PID) Control

The three actions just described are used individually or combined in commercial
controllers. Probably 10% of all controllers are proportional-only (P-only) as per Eq. (1.1),
85% are PI (proportional-integral)

 1
t

U (t )  bias  K C e(t ) 
 TI 
0
e(t ) dt 

(1.5)

and 5% are PID (proportional-integral-derivative)

 1
t
d e(t ) 
U (t )  bias  K C e(t ) 
 TI 
0
e(t ) dt  TD
dt 
 (1.6)

As stated above, the main limitation of P-only control is steady-state offset: lim e(t )  0.
t 
sp
For example, when an operator makes a step change in setpoint Y (t), the measured
variable Y(t) will not converge to the new setpoint. Offset is undesirable except in level
control applications. PI controllers are able to eliminate offset, but this comes at the price
of a reduction in controller robustness and the need to specify two tuning parameters (KC
and TI) instead of one. Derivative action cannot be used when Y is noisy. Since
thermocouple outputs are relatively smooth, PID control is normally attempted only in
temperature control loops and only if the deadtime and/or dominant time constant is large.

13
1.5 Conclusion

In this chapter we have attempted to convey three basic notions:

 The dynamic response of a process is important and must be considered in the process
design.
 The process itself places inherent restrictions on the achievable dynamic performance
that no amount of controller complexity and elegance can overcome.
 The choices of the control system structure, the type of controller, and the tuning of the
controller are all important engineering decisions.

1.6 References

Luyben, W.L. (1990). Process Modeling, Simulation and Control for Chemical Engineers
(2nd Edition). McGraw-Hill, New York.
Luyben, W.L. and M.L. Luyben (1997). Essentials of Process Control. McGraw-Hill,
New York.
Marlin, T.E. (2000). Process Control: Designing Processes and Control Systems for
Dynamic Performance (2nd Edition). McGraw-Hill, New York.
Ogunnaike, B.A. and W.H. Ray (1994). Process Dynamics, Modeling and Control. Oxford
University Press, New York.
Seborg, D.E., Mellichamp, D.A., Edgar, T.F. and F.J. Doyle III (2016). Process Dynamics
and Control (4th Edition). John Wiley & Sons, New York.
Stephanopoulos, G. (1984). Chemical Process Control: An Introduction to Theory and
Practice. Prentice-Hall, Englewood Cliff, NJ.
Svrcek, W.Y., Mahoney, D.P. and B.R. Young (2013). A Real-Time Approach to Process
Control (3rd Edition). John Wiley and Sons, Chichester.

14
Figure 1.8 – Revised flash process and instrumentation (cf. Fig. 1.1).

AC ≡ Analyzing (e.g. composition) controller

FC ≡ Flow controller
LC ≡ Level controller
PC ≡ Pressure controller
RSP ≡ Remote setpoint
TC ≡ Temperature controller

15
16
17
18
19
20