Professional Documents
Culture Documents
Control Theory
Control Theory
2.6 2.10
CONTROL SYSTEMS — CASCADE LOOPS 148 GENETIC AND OTHER EVOLUTIONARY
ALGORITHMS 181
Introduction 148
Cascade System Advantages 148 Introduction 181
Components of the Cascade Loop 148 Applications 181
The Secondary Loop 148 The Concept of EA 181
Secondary Control Variables 149 Fitness Function: Encoding the Problem 181
Cascade Primary Loop 150 Model of Natural Selection 182
Cascade Application Examples 152 Genetic Algorithm 182
Cascade Controller Design and Simulation 153 Genetic Programming 183
Summary 155 Evolutionary Strategy 184
Bibliography 156 Evolutionary Programming 184
System Identification 184
2.7 Polymerization Reactor Example 185
EMPIRICAL PROCESS OPTIMIZATION 157 Complete Process Models 186
Levels of Optimization 157 Process Control Applications 186
Empirical Optimization 157 Controller Tuning 186
Optimization 157 Application of IEC in Controller Tuning 186
Providing Process Data 159 Control Structure Design 187
Search Procedure 160 Online Applications 187
Conclusions 161 Software Tools 188
References 161 Application-Oriented Systems 188
Bibliography 161 Algorithm-Oriented Systems 190
Tool Kits 190
2.8 References 191
EXPERT SYSTEMS 162 Bibliography 192
Artificial Intelligence 162
Expert Systems 162 2.11
Statistical Process Control (SPC) 165 HIERARCHICAL CONTROL 193
Artificial Neural Networks 165 Hierarchical Levels 193
Future Trends 165 History 194
Fuzzy Logic and Neural Networks 165 The Central Computer 194
Neural Networks 166 Distributed Control 196
2.33
2.31
STATE SPACE CONTROL 393
SOFTWARE FOR FUZZY LOGIC CONTROL 360
Introduction 393
Introduction 360
State Space Description 393
Principle of Fuzzy Systems 361
Control Law Design 394
Fuzzy Sets 361
State Variable Feedback 394
Fuzzy Systems 362
State Feedback with PI Control 396
Direct and Supervisory Fuzzy Control 365
Observer Design 396
Classical Fuzzy Control Algorithms 365
Full Order Observer 397
Model-Based Fuzzy Control 367
Full Order Observer for Constant Error 398
Inverse Fuzzy Model-Based Control 368
Reduced Order Observer 398
Fuzzy Model-Based Predictive Control 368
Combined Observer–Controller 400
Operating Regime-Based Modeling 368
Combined Observer-
Takagi–Sugeno Fuzzy Models 369
Controller Behavior 400
Software and Hardware Tools 370
Transfer-Function Interpretation 402
Project Editor 371
Conclusions 403
Rule Base and Membership Functions 372
Bibliography 404
Analysis and Simulational Tools 372
Code Generation and Communication
Links 372
Conclusions 372 2.34
References 373 STATISTICAL PROCESS CONTROL 405
Introduction 405
SPC and Process Control 405
2.32 Continuous Processes 406
STABILITY ANALYSIS, TRANSFER FUNCTIONS 375 What Is Statistical Control? 406
Introduction 375 SPC Tools and Techniques 406
Laplace Transforms 375 Control Charts 406
Theorems 376 Charts and Tables 407
First-Order Lag 376 Interpretation of Charts 408
Partial Fraction Expansion 377 The Purpose of the Charts 409
Z Transforms 379 Using Charts 409
96
1
PROCESS DYNAMICS
Output variable, c
Response to
the natures and the characteristics of their vehicles are dif- unit step change Rm
ferent. This is also true in industrial process control. The pilot of m
(the controller) must be correctly matched to the process that
it controls. In order to do that, the designer of a process must 0
understand the “personality” of the process.
Most processes contain resistance, capacitance, and dead-
time elements, which determine their dynamic and steady-state 0 Time, t
responses to upsets. Before the control of processes is dis-
cussed, these three elements of the “personalities” of processes
will be described. c, Head
To describe the personalities of processes, block diagrams
m, flow Capillary
are used (Figure 2.1b). The two main symbols used in any
block diagram are a circle and a rectangular box. The circle Physical diagram
represents algebraic functions such as addition or subtraction
and is always entered by two lines but exited by only one.
The rectangular box always represents a dynamic function, m c
R
such as a controller, where the output is the controlled variable
(c) and is a function of both time and of the input or manipulated Block diagram
variable (m). Only one line may enter and only one line may
leave a rectangular block. The “system function,” the “person- c = Rm
ality” of the process component, is placed inside the block, and Where c = Output variable (head)
the output is determined by product of the system function and R = Resistance
the input. m = Input variable (flow)
FIG. 2.1c
Resistance-Type Processes Physical example of a resistance element and its block diagram.
u0 u1 u2 q
KV c
h
u2
Kb P
w
u0
Kb C, Capacitance
T
The steady-state gain
of the process is
Response
Kp = B/A
of output
Output variable, c
(c)
R Mo = B
0.632 R Mo
Unit step-change (A) m, flow
in input flow (Mo)
at time = 0 c, head
A
0 1 2 3 4
Time, t/T R q
m c Physical diagram
R
Ts + 1
Block diagram
FIG. 2.1h
A single time-constant process consists of a capacitance and a resistance element. The time it takes for the controlled variable (c) to reach
63.2% of its new steady-state value is the value of the time constant.
would eventually rise to the steady-state height of c = Rm Step response of a first-order system
in the tank. 2
Response y(t)
In order to develop the overall system function, one has 1.5
to combine the capacitance element of Figure 2.1f with a
1
resistance element. If the input variable is the inflow (m) and
the output variable is the level (c), the tank capacitance equals 0.5
the difference between inflow (m) and outflow (q): 0
0 1 2 3 4 5 6
C ( dc/dt ) = m − q 2.1(2) tc
Step response of a first-order system with a 1 minute dead time
In the fluid resistance portion of the system, the output (the 2
head c) equals the product of the outflow (q) and the resistance
Response y (t)
1.5
(R). Because c = qR, therefore q = c/R. Substituting c/R for q
in Equation 2.1(2) and multiplying both sides by R gives 1
0
The unit of R is time divided by area and the unit of C 0 1 2 3 4 5 6
is area; therefore, the product RC has the unit of time. This Time in minutes
time (T) is called the time constant of the process. It has been
found experimentally that after one time constant the value FIG. 2.1i
Step response of a first-order system with and without dead time.
of the output variable (c) of a single time-constant process
will reach 63.2% of its final value. of such a system with a = 1 and b = 2 to a unit step change
Process elements of this description are common and are (value 1 for t ≥ 0 and 0 for t < 0) in the input is shown in
generally referred to as first-order lags. The response of a first- the top panel of Figure 2.1i.
order system is characterized by two constants: a time constant This result can be demonstrated from the solution of the
T and a gain K. The gain is related to the amplification asso- differential equation in Equation 2.1(4). The solution of this
ciated with the process and has no effect on the time charac- differential equation is important because it provides insight
teristics of the response. The time characteristics are related into the operation of the system and nature of the control
entirely to the time constant. The time constant is a measure problem.
of the time necessary for the component or system to adjust The solution of any linear differential equation can be
to an input, and it may be characterized in terms of the capac- broken into the sum of two parts: the homogeneous and the
itance and resistance (or conductance) of the process. particular solution. The homogeneous solution is that expres-
In their responses (Figure 2.1h), two characteristics dis- sion that when substituted for y(t) in the left-hand side of
tinguish first-order systems: Equation 2.1(4) produces the value zero. The particular or
steady-state solution is that part of the total solution that is
1. The maximum rate of change of the output occurs
strictly due to the forcing input, u(t).
immediately following the step input. (Note also that if
To establish the homogeneous solution, assume the solu-
the initial rate were unchanged the system would reach
tion to be of the form
the final value in a period of time equal to the time
constant of the system.)
2. The actual response obtained, when the time lapse is yh (t ) = Ke pt 2.1(5)
equal to the time constant of the system, is 63.2% of
the total response. with u(t) = 0 in Equation 2.1(4). Substituting Equation 2.1(5)
into the left-hand side of Equation 2.1(4) imposes the fol-
These two characteristics are common to all first-order processes. lowing condition on p:
TZ Water outlet:
FIG. 2.1o
Process variables in a simple water cooler.
Steam
N n = nc − n p + 2 2.1(10)
Variables:
W E Altitude 1 where n = number of chemical degrees of freedom; nc =
Latitude 1 number of components; and np = number of phases.
Longitude 1
S 3 For example, if the process is a boiler producing satu-
Equations: rated steam (Figure 2.1s), the number of components is one
None –0
Degree of freedom: 3 (H2O), the number of phases is two (water and steam), and,
therefore, the number of degrees of freedom is n = 1 − 2 +
FIG. 2.1q 2 = 1. Consequently, only one variable can be controlled:
The degrees of freedom of an airplane. temperature or pressure, but not both. If a boiler produces
Ps = Pressure
Controlled
Ws = Steam flow variable
(c)
Saturated
vapor
c)
Ma iable
r – tion
r–
v ar
nip (m)
FIG. 2.1s
b=
(e = D e v i a
ula
A saturated steam boiler has only one degree of freedom.
ted
superheated steam, the number of degrees of freedom is two,
and therefore both temperature and pressure can be indepen- Manipulated
dently controlled. m variable (m)
When the process is more complex, such as is the case
LC Set point
of binary distillation, the calculation of the degrees of free- (r)
dom also becomes more involved. Figure 2.1t lists 14 vari-
ables of this process, not all are independent. Since there are
two components and two phases at the bottom, feed and
Controlled variable
overhead, Gibbs’s law states that only two of the three variables (c)
(pressure, temperature, and composition) are independent. R Uncontrolled
Therefore, the number of independent variables is only 11. The load variable (u)
number of defining equations is two (the conservation of
mass and energy), and therefore, the number of degrees of FIG. 2.1u
freedom for this process is 11 − 2 = 9. Consequently, not The addition of a controller closes the automatic control loop.
more than nine automatic controllers can be placed on this
process.
2
CLOSING THE LOOP
FIG. 2.1v
Control loops respond to upsets differently as a function of the loop gain and of the phase lag contributions of both the process and the
controller.
c
processes is usually due to the presence of capacitance ele-
ments. For example, in the case of the process shown in
Figure 2.1w, the process causes the output (c = outflow from
the tank) to lag behind and oscillate with a smaller amplitude
than the input of the process (m = inflow to the tank).
of controller and process reaches unity, the process becomes the loop will cycle when the load drops, and if the loop was
unstable and undamped oscillations (cycling) will occur. tuned at low loads, the loop will not be able to hold the process
Therefore, it is not possible to tightly control fast (high-gain) on set point (will be sluggish) when the load rises.
processes without cycling. It is easier to obtain tight control One way to compensate for this effect is to install a
on slow, low-gain processes, because the use of high-gain control valve in the loop (an equal percentage control valve),
controllers does not destroy stability. which increases its gain as the load rises. When the gain of
The loop gain is the product of all the gains in the loop, the process drops, the gain of the valve increases and the
including sensor, controller, control valve, and process. In a total loop gain remains relatively unaffected.
properly tuned loop, the product of all these gains is 0.5.
What makes tuning difficult is that the process gain often Feedback Control
varies with process load. For example, in heat transfer pro-
cesses, when the heat load is low and the heat transfer surface Two concepts provide the basis for most automatic control
available to transfer the heat is large, the transfer of heat is strategies: feedback (closed-loop) control and feedforward
performed efficiently; therefore, under those conditions, this (open-loop) control. Feedback control is the more commonly
process is a high-gain process. used technique of the two and is the underlying concept on
As the load rises, the same heat transfer process becomes which most automatic control theory used to be based. Feed-
a low-gain process because the fixed heat transfer area back control maintains a desired process condition by mea-
becomes less and less sufficient to transfer the heat. There- suring that condition, comparing the measurement with the
fore, as shown in Figure 2.1x, the gain of a heat transfer desired condition, and initiating corrective action based on
process (Gp) drops as the load rises. the difference between the desired and the actual conditions.
Tuning such a system can be a problem because in order The feedback strategy is very similar to the actions of a
to arrive at an overall loop gain of 0.5, the controller should human operator attempting to control a process manually.
apply a high gain when the load is high and a low gain when Consider the control of a direct contact hot water heater. The
the load is low. Standard controllers cannot do that because operator would read the temperature indicator in the hot water
they have been tuned to provide a single gain. Therefore, if line and compare its value with the temperature desired
the loop was tuned (controller gain was selected) at high loads, (Figure 2.1y). If the temperature was too high, he would
reduce the steam flow, and if the temperature was too low,
he would increase it. Using this strategy, he would manipulate
Gv the steam valve until the error is eliminated.
An automatic feedback control system would operate in
much the same manner. The temperature of the hot water is
measured and a signal is fed back to a device that compares
the measured temperature with the desired temperature. If an
Load
error exists, a signal is generated to change the valve position
Valve gain
m
Load in such a manner that the error is eliminated.
(Gv) (u)
m1 The only real distinction between the manual and auto-
+ +
Gc Gp matic means of controlling the heater is that the automatic
controller is more accurate and consistent and is not as likely
Controller
For stable
Process
gain
control
Gc ⫻ Gv ⫻ Gp ⫻ Gs = 0.5 tain the essential elements of a feedback control loop.
Load Load
e
c
+ –
b Sensor Desired
Set gain temperature = To
point (r) (Gs) Temperature
Gs indicator
Hot water
Amplitude
initiate and follow changes demanded in the operating point. 0.8
First, it is desirable that the output follow the desired behav- 0.4
ior. Note that in all the subplots in Figure 2.1m, the response
to a unit step change approaches the steady-state value of 1, 0.2
which corresponds to the magnitude of the input step
0
response. Because the output response eventually reaches the 0 1 2 3 4 5 6
steady-state value, the steady-state error, or the difference Time (sec)
between the desired final output and the actual one, is zero.
Second, almost always, the steady-state error should be FIG. 2.1z
zero to a step input, or constant targets, as inputs. In some The step response shown has a 4.2-second “2% settling time.”
cases, such as the case of a ramp input, it also is desirable for
the steady-state error to be zero or nearly so. There may be an As was shown earlier, the dynamic behavior of many
upper limit on the magnitude that is tolerable when no distur- processes can be modeled adequately as first- or second-order
bances are present. However, in the presence of disturbances systems. Thus, the PID is an ideal tool with which to control
the steady-state error can become larger. such systems. Furthermore, the PID is easily understood and
Third, the speed of response is important. From the dis- tuned by trained operators.
cussion in connection with equation 2.1(7), viz. the solution Consider the feedback control system with a plain pro-
of the differential equation, the steady-state is attained as the portional-only controller shown in Figure 2.1aa. Assume that
homogeneous portion of the solution of the differential equa- the process to be controlled is a static system with a gain Kp.
tion approaches zero. A control system can affect the rate at For a proportional-only controller, the controller output is the
which this happens. If the response of the system is sluggish, product of the error signal (e = r − c) and of the proportional
then the output (control action) of the controller is not changing gain Kc. That is,
enough in magnitude in response to the difference between the
desired and actual output. By changing the parameters of the
controller, the magnitude of the control action and the speed u = Kc e 2.1(11)
of response can be increased in response to control errors.
Fourth, the physical limitations of the plant constrain the The closed-loop response is the relationship between the
ability of the controller to respond to input command output (controlled variable), c, and the reference or set point
changes. Another measure of the controller’s speed is the input, r. This relationship is
settling time. The settling time is defined as the time after
which the control system will remain within a given percentage
of the desired final value when there are no outside distur- Kc K p
c= r 2.1(12)
bances. Figure 2.1z illustrates a “2% settling time,” meaning 1 + Kc K p
the time it takes for a step response to approach the final
steady-state value within 2%.
Note that if r is a constant, say one, the controlled output is
Lastly, note that (Figure 2.1l) the step change responses
less than one. Thus, there is a nonzero steady-state error to
of a second-order system all have an overshoot, when the
constant inputs. This is not surprising because if r = c, then
damping ratio of the system is less than one. Overshoot is
e = r − c = 0 and the output of the controller would also be
defined as the percentage by which the peak response value
exceeds the steady-state value (peak value of step response −
steady-state value)/(steady-state value). A small overshoot can
r + e u c
be acceptable, but large overshoots are not. PID Plant
−
The PID Controller
zero. This produces a contradiction in that c would be forced Controller response with and without derivation action
to zero, which, in general, is not the value of r. 1
Therefore, the plain proportional controller reduces but 0.8
does not eliminate the error. Note from Equation 2.1(12) that
as the controller gain increases, the controlled output, c, 0.6
Controller output
approaches the referenced input r more and more closely. 0.4
Thus, the steady-state error, e, becomes smaller as Kc is made
0.2
larger. But since Kc can never be infinite, the error is never
zero. 0
−0.2
The Derivative Mode To better understand the effect of
−0.4
derivative action, consider the situation in which the control-
ler shown in Figure 2.1aa is proportional plus derivative (PD). −0.6
In this case, the controller output is given by −0.8
0 1 2 3 4 5 6 7 8 9 10
Time
de(t )
u(t ) = e(t ) + Td 2.1(13)
dt FIG. 2.1cc
if the proportional gain Kc is 1 and the derivative gain is Td , The “anticipation” of the PD controller can be seen by noting the
controlled variable response of a plain proportional controller
which is called the derivative time. Figure 2.1bb illustrates
(solid line) and that of a PD controller (dashed line) to the same
the controlled variable’s step response, the error and deriva-
step upset.
tive of the error signal, when the set point (reference) is r = 1.
The vertical lines in the top two plots of Figure 2.1bb
are in locations where the controlled variable signal and the
the input around in time to stop the system from exceeding
error signal have local maximums and minimums or points
the desired value.
where the derivative is zero. Note that the controlled variable
This is seen in the error signal that has not changed sign
response has exceeded the set point (target value) of one. The
until the controlled variable output has exceeded the set point.
excess overshoot is due to the presence of a certain momen-
Note also that the derivative is zero at the peak values and
tum in the response of the system; the controller did not turn
of opposite sign to the value of the error signal. When the
error is added to a constant times the derivative, the result is
Step response a signal that changes sign earlier in time, that is, before the
1.5
output has actually reached the steady-state value. Figure 2.1cc
1 illustrates this.
The PD controller is used in applications where overshoot
0.5
cannot be tolerated, such as pH neutralization. The reduction
0 or elimination of overshoot can often be accomplished with-
0 1 2 3 4 5 6 7 8 9 10
Error signal
out significantly affecting the settling time of the closed-loop
system. The primary disadvantage of derivative mode is its
1 sensitivity to noise. Noise is generally of high frequency, and
0.5
differentiating just amplifies it. The controller output can
become cyclic or unstable, which can have a detrimental
0 effect on the longevity of actuators such as valves or motors.
−0.5
0 1 2 3 4 5 6 7 8 9 10
Derivative of error signal Integral Mode Nearly all controllers have some form of inte-
gral action. Integral action is important because it corrects
2
based on the accumulated error, which is the area under the
1 error curve. If the error goes to zero, the output of the integrator
0 is the constant area that had accumulated up to that point.
−1 Consider the feedback system illustrated in Figure 2.1aa.
−2 The task of the integral term in the PID algorithm is to find
0 1 2 3 4 5 6 7 8 9 10
the manipulated variable (the input to the plant) needed to
FIG. 2.1bb drive the steady-state error to zero when the set point (refer-
The step response of a PD controller showing the responses of the ence input) is constant.
controlled variable (top), the error (center), and the derivative of When the error is zero, both the proportional term and
the error (bottom). the derivative term contribute nothing to the controller output.
Only the integral term provides any input to the controller control applications. In many cases disturbances cannot be
output; only the integrator drives the manipulated variable to accurately measured, and therefore pure feedforward cannot
compensate for the area under the past error curve. be used. The main limitation of feedforward is due to our
In summary, the PID controller produces an output inability to prepare perfect process models or to make per-
defined as fectly accurate measurements.
Because of these limitations, pure feedforward would
de(t ) accumulate the errors in its model and would eventually “self-
∫
1
u(t ) = K p e(t ) + e(τ )dτ + Td 2.1(14)
dt
destruct.” The main limitations of feedback control are that
Ti
feedback cannot anticipate upsets but can only respond to
them after the upsets have occurred, and that it makes its
where Kp is the proportional gain; Td is the derivative time;
correction in an oscillating, cycling manner.
and Ti is the integral time.
It has been found that combining feedback and feedfor-
The integral time can be viewed as the amount of the
ward is desirable in that the imperfect feedforward model
time it takes for the integral component to make the same
corrects for about 90% of the upset as it occurs, while feed-
contribution as the proportional term. If the integral time is
back corrects for the remaining 10%. With this approach, the
short, the integral contribution to the PID output is large and
feedforward component is not pushed beyond its abilities,
too much integral gain (Ti too small) can cause the system
while the load on the feedback loop is reduced by an order
to become unstable.
of magnitude, allowing for much tighter control.
Feedforward Control
Feedforward Response
Feedforward control is another basic technique used to com-
Ideally the feedforward correction would be so effective that
pensate for uncontrolled disturbances entering the controlled
a disturbance would have no measurable effect on the con-
process. Both feedback and feedforward control are dis-
trolled variable, the process output. As an example, con-
cussed in detail in Section 2.8 and therefore only an intro-
sider a first-order system in which there is a measurable
duction is given here. In this technique the control action is
disturbance. Suppose that a process disturbance occurs at time
based on the disturbance input into the process without con-
t = 5 seconds, as shown in the top segment of Figure 2.1ee, and
sidering the condition of the process. In concept, feedforward
causes the PID controller to generate a corrective action as
control yields much faster correction than feedback control
shown in lower part of Figure 2.1ee. Note that while the
does, and in the ideal case compensation is applied in such
controller will eliminate the disturbance, it will do that only
a manner that the effect of the disturbance is never seen in
after it has occurred.
the controlled variable, the process output.
A skillful operator of a direct contact water heater could
use a simple feedforward strategy to compensate for changes Process output with disturbance
in inlet water temperature by detecting a change in inlet water 1.5
temperature and in response to that, increasing or decreasing
Amplitude
1.5
Amplitude
Hot water
1
0.5
Steam Water 0
0 1 2 3 4 5 6 7 8 9 10
heater
Steam
Cool water valve Temperature FIG. 2.1ee
indicator If a step disturbance occurs at t = 5, the controlled variable of a
first-order process responds to that upset as shown in the top portion
FIG. 2.1dd of the figure. The bottom part shows the response of a feedback
The concept of feedforward control implemented by a human PID controller to such an upset, which generates the manipulated
operator. variable.
d(t) (Figure 2.1ff), the system exactly compensates for the dis-
Disturbance turbance. In practice, exact compensation is not possible.
However, much of the effect of a disturbance can be mitigated
Feedforward by a judiciously designed feedforward compensator.
controller
+ +
r e + y
PID Plant Cascade Control
+ +
−
Section 2.6 discusses cascade control in detail; therefore,
only an introduction is given here. In cascade control two
controllers are used to control a single process variable. For
example, consider the composition control of the product
FIG. 2.1ff from a continuously stirred reactor in which an exothermic
Block diagram representation of a combination of a feedback PID chemical reaction takes place. If the compensation is directly
control loop with a feedforward compensator. related to the temperature within the reactor and if the com-
position itself is difficult to measure online, then the reactor
temperature is often chosen as the controlled variable. To
control that temperature, the cooling water flow to the jacket
Now consider a control system that includes a 100%
is usually manipulated by the master (outer) temperature
effective feedforward controller as shown in Figure 2.1ff. The
controller, which is adjusting the set point of the slave (inner)
bottom part of Figure 2.1gg shows the response of a perfect
flow controller. Together these two controllers form a cas-
feedforward compensator to an upset that has occurred at
caded control loop, as shown in Figure 2.1hh.
t = 5. The middle section of the figure shows the feedback
The control loop for the cooling water flow is called the
PID output and the top portion shows the controlled variable,
inner loop. The outer loop is the controller for the reactor
which remains undisturbed.
temperature. The inner loop must have a response time that
Note that the compensator does not respond until the
is faster than that of the outer loop. The outer loop controller
disturbance occurs at time t = 5 seconds. With the feedfor-
provides the set point for the inner loop. The inner loop must
ward compensation that is added to the PID controller output
be faster than the outer loop, so that the set point will not
change too fast for the inner loop dynamics to follow; if it
did, stability problems would result.
1
Filters are needed to eliminate noise from such signals as the
0.5
controlled variable. If the noise in the measurement signal is
0 not reduced, it can pass through the controller and cause
0 1 2 3 4 5 6 7 8 9 10
cycling and eventual damage to the control valve or other
Controller output with disturbance final control element.
2 Noise tends to be of a high frequency and changes more
Amplitude
1.5 rapidly in time than does the controlled process variable. The
1 task of a filter is to block the rapidly changing component
0.5 of the signal but pass the slowly changing component, so that
0 1 2 3 4 5 6 7 8 9 10 the filter output will have less noise than does the raw signal.
Feedforward compensator output with disturbance Filtering may have little or no effect on control perfor-
1 mance if the closed-loop system response time is slow com-
pared to the response time of the filter. In other words, if the
0.5
0
0 1 2 3 4 5 6 7 8 9 10
r e + Flow y
Time in seconds Reactor
PID PID
+ system
− −
FIG. 2.1gg
The response of the control loop shown in Figure 2.1ff to a process
upset that occurs at t = 5. In the bottom part, the output of the
feedforward compensator is shown. When this is added to the output
of the feedback PID controller, the upset is fully corrected and as FIG. 2.1hh
shown in the top curve, the controlled variable is not upset at all. Block diagram of a cascade control loop.
filter reaction time is assumed to be instantaneous compared Frerichs, D., “Integrating Diagnostic Expert System with Statistical Process
to the speed of response of the closed-loop system, then there Control in a Modern Distributed Control System,” paper presented at
ISA-90, Instrument Society of America, 1990.
is no dynamic effect on the closed-loop system. On the other Grant, E., and Leavenworth, R., Statistical Quality Control, New York:
hand, if the closed-loop is fast relative to the speed of response McGraw-Hill, 1988.
of the filter, then the dynamic effects of the filter will have an Holmes, D. S., “Time Series Analysis Overcomes SPC Problems,” Control,
effect. February 1991, pp. 36–38.
Therefore, filtering can complicate the controller and can Hoo, K. A., Piovoso, M. J., Schnelle, P. D., and Rowan, D. A., “Process and
Controller Performance Monitoring: Overview with Industrial Appli-
limit the response of a PID controller. One way to compensate
cations,” IEEE Journal of Adaptive Control & Signal Processing, Vol.
for this is to tune the controller with the filter inserted into the 17, No. 7–9, 2003, pp. 635–662.
feedback loop. Although the PID controller may give better Hougen, J. O., Measurements and Control Applications for Practicing
performance without the filter, the effectiveness of the filter Engineers, Boston: Cahners Books, 1972.
depends on the nature and amount of the noise and the desired Johnson, C. D., Process Control Instrumentation Technology, New York:
Wiley & Sons, 1988.
closed-loop dynamics.
Jones, B. E., Instrumentation, Measurement and Feedback, New York:
McGraw-Hill, 1977.
Joseph, B., et al., “Knowledge Acquisition and Knowledge Representation
ADVANCED CONTROLS in an Expert System for Control of a Batch Process,” paper presented
at AIChE National Meeting, Houston, 1991.
The scope of the discussion in this first section was limited Kecman, V., State-Space Models of Lumped and Distributed Systems,
Heidelberg: Springer-Verlag, 1988.
to basics of process control, and therefore such subjects as Kelly, K. P., “Should You Be Training Your Operators with Simulation
multivariable loops, interaction and decoupling, optimiza- Software,” Control, June 1992, pp. 40–43.
tion, artificial intelligence, statistical process control, relative Koenig, D. M., Control and Analysis of Noisy Processes, Englewood Cliffs,
gain calculations, neural networks, model-based and model- NJ: Prentice Hall, 1991.
free controls, adaptive control, fuzzy logic and statistical Korn, G., Interactive System Simulation, New York: McGraw-Hill, 1989.
Kuo, B. C., Automatic Control, Englewood Cliffs, NJ.: Prentice Hall, 1982.
process control have not been covered. These topics and
Luyben, W. L., Process Modeling, Simulation and Control for Chemical
many others are discussed in the sections that follow. Engineers, 2nd ed., New York: McGraw-Hill, 1990.
MacFarlane, A. G. J., and Hung, Y. S., “Quasi-Classical Approach to Mul-
tivariable Feedback Systems Design,” 2nd IFAC Symposium, Purdue
References University, September 1982, pp. 39–48.
McAvoy, T. J., Interaction Analysis, Research Triangle Park, NC: Instrument
Society of America, 1983.
1. Eckman, D. P., Automatic Process Control, New York: John Wiley
McColl, M., and Collins, S., “Deadtime Compensation Algorithm,” ISA/93
& Sons, 1958.
Technical Conference, Chicago, 1993.
2. Shinskey, F. G., Process Control Systems, 4th ed., New York: McGraw-
McMillan, G., Tuning and Control Loop Performance, ISA, 2nd ed., 1990,
Hill, 1996.
pp. 45–55.
Morely, R., “Faster Than Real Time,” Manufacturing Systems, May 1992,
p. 58.
Bibliography Nachtigal, C. N., Instrumentation and Control, New York: Wiley, 1986.
Nise, N. S., Control System Engineering, New York: John Wiley & Sons,
Arkun, Y., and Downs, J., “A General Method to Calculate Input–Output 2000.
Gains and the Relative Gain Array for Integrating Processes,” Computers Nyquist, H., “Regeneration Theory,” Bell System Technical Journal, Vol. 13,
& Chemical Engineering, 14, Vol. 10, 1990, pp. 1101–1110. 1952, p. 1.
Balakrishman, R., “Dead Time Controller,” ISA/94 Technical Conference, Oakland, J., Statistical Process Control, New York: Wiley, 1986.
Anaheim, 1994. Ogata, K., Modern Control Engineering, Englewood Cliffs, NJ.: Prentice Hall,
Balchen, J. G., and Mummè, K. I., Process Control Structures and Appli- 1970.
cations, New York: Van Nostrand Reinhold, 1988. Perron, S., “The New Generation Language for Controls Design,” Control
Butterworths, W. R., Advanced Process Control, New York: McGraw-Hill, Engineering, February 1992.
1981. Phillips, C. L., and Harbor, R. D., Feedback Control Systems, Englewood
Chang, J.-W., and Yu, C.-C., “The Relative Gain for Non-Square Multi- Cliffs, NJ: Prentice Hall, 1991, p. 136.
variable Systems,” Chemical Engineering Science 45, Vol. 5, 1990, Rich, S., et al., “Development of a Diagnostic Expert System for a Whipped
pp. 1309–1323. Toppings Process,” Journal of Loss Prevention in Process Industry,
Cutler, C. R., and Ramaker, B. L., “Dynamic Matrix Control—A Computer Vol. 2, July 1989.
Control Algorithm,” ACC Conference Proceedings, San Francisco, Rijnsdorp, J. E., MacGregor, J. F., Tyreus, B. D., and Takamatsu, T., Dynamics
California, 1980. and Control of Chemical Reactors, Distillation Columns and Batch
D’Azzo, J. J., and Houpis, C. H., Linear Control System Design, Conven- Processes, Oxford: Pergamon Press, 1990.
tional and Modern, New York: McGraw-Hill, 1988. Roffel, B., and Chin, P., Computer Control in the Process Industries,
Doherty, J., “Detecting Problems with SPC,” Control, November 1990, Chelsea, MI: Lewis Publishers, 1989.
pp. 70–73. Rowan, D., “On-Line Expert Systems in Process Industries,” AI/Expert,
Dorsey, J., Continuous and Discrete Control Systems: Modeling, Identifica- August 1989.
tion, Design, and Implementation, New York: McGraw-Hill, 2002. Ryskamp, C. J., “New Control Strategy Improves Dual Composition Con-
Franklin, G. F., Powell, J. D., and Emami-Naeini, A., Feedback Control of trol,” Hydrocarbon Processing, May 1981.
Dynamic Systems, 2nd ed., Reading, MA: Addison-Wesley, 1991. Seborg, D., et al., Process Dynamics and Control, New York: Wiley, 1989.
Shinskey, F. G., Chemical Process Control II: Proceedings, T. J. McAvoy Stephanopoulos, G., Chemical Process Control: An Introduction to Theory
et al., Eds., Amsterdam: Elsevier, 1986. and Practice, Englewood Cliffs, NJ: Prentice Hall, 1984, p. 310.
Shinskey, F. G., Distillation Control, 2nd ed., New York: McGraw-Hill, 1984. Vasbinder, E. M., and Hoo, K. A., “A Decision-Based Approach to Plantwide
Shinskey, F. G., Process Control Systems, New York: McGraw-Hill, 1996. Control Structure Synthesis,” Industrial & Engineering Chemistry
Skogestad, S., Lundstrom, P., and Jacobsen, E. W., “Selecting the Best Research, Vol. 42, No. 20, 2003, pp. 4586–4598.
Distillation Control Configuration,” AIChE Journal 36, May 1990, Vervalin, C. H., “Training by Simulation,” Hydrocarbon Processing, December
pp. 753–764. 1984, pp. 42–49.