Beruflich Dokumente
Kultur Dokumente
Third Edition
By
Armando B. Corripio, Ph.D., P.E.
Chemical Engineering
Louisiana State University
and
Michael Newell
Automation Designer
Polaris Engineering
Notice
The information presented in this publication is for the general education of the reader. Because
neither the author nor the publisher has any control over the use of the information by the reader,
both the author and the publisher disclaim any and all liability of any kind arising out of such use.
The reader is expected to exercise sound professional judgment in using any of the information
presented in a particular application.
Additionally, neither the author nor the publisher has investigated or considered the effect of
any patents on the ability of the reader to use any of the information in a particular application.
The reader is responsible for reviewing any possible patents that may affect any particular use of
the information presented.
Any references to commercial products in the work are cited as examples only. Neither the
author nor the publisher endorses any referenced commercial product. Any trademarks or
tradenames referenced belong to the respective owner of the mark or name. Neither the author
nor the publisher makes any representation regarding the availability of any referenced
commercial product at any time. The manufacturers instructions on the use of any commercial
product must be followed at all times, even if in conflict with the information in this publication.
Preface to the
Third Edition
This third edition of Tuning of Industrial Control Systems has been significantly
simplified from the second edition with the goal of having the discussion
more in line with modern control systems and with language that is less academic and more in tune with the vocabulary of the technicians who do the
actual tuning of control systems in industry. For example, we have eliminated
any references to first- and second-order models since these terms are highly
mathematical and may discourage some from appreciating the usefulness of
the models. We have also eliminated the distinction between series and parallel PID controllers since most modern installations use the series version and
there is not much difference between the tuning of the two versions.
We have reduced the tuning strategies to just one; the quarter-decay-ratio
(QDR) formulas slightly modified by the Internal Model Control (IMC) rules
for certain process characteristics. All the tuning strategies are intended for
responses to disturbances with a discussion on how to modify these responses
to avoid sudden excessive changes of the controller output on set point
changes when such changes are undesirable.
Chapter 10 is new and deals with the auto-tuning feature that has become
standard on current process control systems. We have successfully used the
auto-tuning feature in our tuning work on oil refineries as a reference to guide
our selection of the final tuning parameters for the controllers.
We have kept the previous editions discussions on the problems of process
nonlinearities and reset windup, and how to compensate for them. All of the
tuning strategies are demonstrated with computer simulation examples.
ix
Contents
2
3
7
8
8
8
8
12
19
21
29
29
30
30
34
36
41
46
50
53
54
54
vi
Contents
vii
152
158
165
166
166
170
174
186
193
196
198
199
199
202
205
209
212
213
1
Introduction
Learning Objectives When you have completed this chapter, you should be
able to
A. Define the main goal of tuning a control system.
B.
Introduction
parameters are usually compromises selected to work in the range of operating conditions, and so their values are not precise.
Understanding this simplifies the task of tuning because it reduces the number of values of the tuning parameters to be tried. For example, it is a lot easier
to decide between gain values of 1.0 or 1.5 than to try to find out whether the
gain should be 1.276. In practice, all three of these values will work about the
same.
Armed with these heuristics and basic concepts, we are now ready to look at
the feedback control strategy.
It is important to realize that a feedback controller does not use a model of the
process to compute its output. It takes action by trial and error. Tuning the
controller is the procedure of adjusting the controller parameters to ensure
that the controller output converges quickly to its correct value.
In order to better understand the concept of feedback control, consider as an
example the process heater sketched in Figure 1-1. The process fluid flows
inside the tubes of the heater and is heated by steam condensing on the outside of the tubes. The objective is to control the outlet temperature T of the
process fluid in the presence of variations in process fluid flow (throughput or
load) F and in its inlet temperature Ti. This is accomplished by manipulating
or adjusting the steam flow to the heater Fs and with it the rate at which heat is
transferred into the process fluid, thus affecting its outlet temperature.
SP
Steam
OP
TC
Fs
PV
F
Ti
TT
Process
fluid
Steam
trap
Condensate
Introduction
deviation of the process variable from the set point SP and acts on this error
by changing the signal OP to the control valve; the control valve position
changes, causing a change in steam flow Fs to the heater; this in turn causes a
change in the outlet temperature T which then starts a new cycle of changes
around the loop.
Figure 1-2. Block Diagram of the Temperature Control Loop of the Process
Heater
Ti
F
-
SP
Controller
Fs +
OP+
Control Valve
Heater
PV
Sensor
Transmitter
The signs in Figure 1-2 represent the action of the various input signals on the
output signal; that is, a positive sign means that an increase in input causes an
increase in outputdirect actionwhile a negative sign means that an increase
in input causes a decrease in outputor reverse action. For example, the negative sign by the process flow into the heater means that an increase in flow
results in a decrease in outlet temperature. Notice that by following the signals around the loop, there is a net reverse action in the loop. This property is
known as negative feedback and is a required characteristic of a feedback loop
for the loop to be stable. In this example it means that an increase in heater
outlet temperature results in a decrease in controller output, which in turn
closes the control valve and reduces the steam flow. This results in a decrease
in outlet temperature, as desired.
To ensure this self-regulating effect the controller must act in the correct direction when the process variable changes. In this example the controller action is
reverse, that is, an increase in process variable results in a decrease in control-
Introduction
ler output. Other processes may require direct action, for example when a tank
level controller adjusts the flow out of the tank. In this case, an increase in liquid level in the tank requires that the exit control valve open to increase the
flow out of the tank and decrease the level. Consequently, the action (direct or
reverse) of the feedback controller is its most important characteristic.
the process when each controller output affects the process variables
controlled by the other controllers.
1-5. Summary
This first chapter has presented the goals of the tuning procedure and has
introduced the feedback control strategy. A brief description of other common
control strategies has also been presented.
References
1. ANSI/ISA-5.1-2009 - Instrumentation Symbols and Identification, International Society of Automation, Research Triangle Park, NC.
Review Questions
1-1. What is the main goal of controller tuning?
1-2. Which two process characteristics must be considered when tuning the
controller?
1-3. What are the three instrumentation components of a feedback control
loop?
1-4. What is the fourth element of the feedback loop?
1-5. What is the most important characteristic of a feedback control loop?
Introduction
1-6. A controller controls the temperature in an exothermic reactor by manipulating the flow of cooling water to the jacket around the reactor. What
should be the fail position of the cooling water control valve, open or
closed? What must be the action of the controller, direct or reverse?
1-7. A controller controls the level in a stirred tank reactor by manipulating
the flow of the reactants into the reactor. Recommend the fail position of
the reactants control valve, open or close, and the controller action, direct
or reverse.
1-8. A controller controls the composition of a caustic stream by manipulating the flow of the water that dilutes the concentrated caustic stream
entering a mixer. The control valve fails closed. What must be the controller action, direct or reverse?
2
The Feedback
Controller
The basic concept of feedback control was introduced in the preceding chapter. This chapter presents details of the feedback controller and one of the
methods proposed to tune it: the ultimate gain and period method.
11
12
Proportional Mode
The purpose of the proportional mode is to cause an instantaneous response
of the controller output to changes in the process variable. The adjustable
parameter for the proportional mode is the gainproportional gain or controller gainKc. Figure 2-1 illustrates how the proportional mode responds to
the process variable PV assuming that the controller is reverse acting and that
the loop is open, that is, that the controller output does not affect the process
variable. The figure shows that:
The controller output OP responds instantaneously to the process variable PV.
The response is proportional to the gain Kc.
The proportional mode does not eliminate the sustained deviation (offset) between the process variable PV and the set point SP.
If a controller only has proportional mode there will normally be an offset.
Since console operators prefer to see all the variables at their set points, not
many controllers are proportional only.
13
Figure 2-1. Response of the Proportional Mode with the Loop Open
PV
SP
OP
Kc = 1.0
Kc = 2.0
time
Integral or Reset Mode
The purpose of the integral or reset mode is to eliminate the deviation
between the process variable and the set point. The controller does this by
moving its output with time at a rate proportional to the magnitude of the
deviation. Thus, as long as there is a deviation, the integral mode will keep
moving the output. The adjustable tuning parameter for the integral mode is
the integral timeor reset timeTI, which is inversely proportional to the
rate at which the controller output changes. Figure 2-2 illustrates how an integral reverse-acting controller responds to a sustained deviation between the
PV and the SP with the loop open. The figure shows that:
The output does not change when the deviation is zero.
The output changes continuously as long as there is a deviation.
The response is not instantaneous; that is, the integral mode takes time
to act.
The rate of change is slower the higher the integral time.
14
Figure 2-2. Response of the Integral Mode to a Step Change in PV with the Loop
Open
PV
SP
Kc = 1.0
OP
TI = 6
TI = 12
time
The step in output shown in the figure is the instantaneous response of the
proportional mode. It takes the integral mode a period of time equal to TI to
duplicate the instantaneous response of the proportional mode.
The integral mode thus forces the process variable to the set point at the
expense of slower action than the proportional mode. This slow action introduces some instability into the response of the loop.
15
Figure 2-3. Response of the Derivative Mode to a Ramp in the PV with the Loop
Open
PV
SP
OP
TD = 6.0
TD = 3.0
0
time
16
To better illustrate the anticipation action of the derivative mode, the response
to a ramp in the process variable is shown in Figure 2-4 for a direct-acting controller having both proportional mode (with a gain of 1.0) and derivative
mode. The initial step in the output is caused by the derivative mode and the
continuous change is caused by the proportional mode. As a result, the output
leads the process variable by a period of time equal to the derivative time.
Notice that this does not mean the controller can predict the future, since the
output cannot change until the process variable starts changing.
Kc = 1.0
OP
10
PV
TD = 10
SP
time
Although the derivative mode increases the stability of the control loop, it has
two undesirable characteristics. One is that if the transmitter signal (PV) is
noisy, the derivative can amplify noise. To limit this amplification as the frequency of the noise increases, practical controllers have a built-in filter on the
derivative mode that limits the amplification factor. The other undesirable
characteristic is that the derivative mode can cause sudden changes in controller output with sudden changes in the process variable. This is usually not a
problem because very seldom will the process variable change suddenly in
practice. To prevent sudden changes in set point from causing sudden
changes in output, all practical controllers have the derivative mode work
only on the process variable, not on the deviation from the set point.
17
T, F
250
150
50
PV, %
OP, %
100
100
80
80
60
60
SP
40
40
20
20
18
Figure 2-5 also illustrates the concept of the controller proportional band (PB)
defined as the fraction of the transmitter output range that causes a 100%
change in the controller output OP. For the assumed gain of 5.0 the proportional band is 20%. In some older controllers the gain was specified as the proportional band, but that is no longer the practice.
19
the controller output does not change, and, optionally, the set point is set to
the current value of the process variable when the switch is performed.
When the console is properly authorized under password protection, the
instrument person or engineer can access the following features:
Proportional gain, integral time, and derivative time adjustments
Direct/reverse action switch
Having introduced the feedback controller in this section, the next section
presents the concept of loop stability, that is, the effect of the controller on the
process response.
20
As pointed out earlier, the oscillatory type of instability is caused by the controller having too high a gain, too short an integral time, or too long a derivative time, or a combination of these. This leads into the simplest method for
characterizing the process for the purpose of tuning the controller, that of
determining the ultimate gain and period of oscillation of the loop.
21
The first controller tuning method will now be introduced, one that depends
on measuring the characteristics of the control loop by determining the limit
of stability of the closed loop with a proportional controller.
22
Kc = 2
Kc = 0.5
Kc = 2
Kc = 0.5
3. Carefully increase the proportional gain in steps. After each increase, disturb the loop by introducing a small step change in set point and observe
the response of the controlled and manipulated variables, preferably on a
trend recorder. The variables should start oscillating as the gain is
increased, as in Figure 2-7.
4. When the amplitude of the oscillations remains constant (or approximately constant) from one oscillation to the next, the ultimate controller
gain has been reached. Record it as Kcu.
5. Measure the period of the oscillations from the trend recordings, as in Figure 2-8. For better accuracy, time several oscillations and calculate the
average period. In Figure 2-8, for example, the time required by five oscillations is measured and then divided by 5.
6. Stop the oscillations by reducing the gain to about half of the ultimate
gain.
23
Kcu = 3.42
5Tu
The procedure just outlined is simple and requires a minimum upset to the
process, just enough to be able to observe the oscillations. Nevertheless, the
prospect of taking a process control loop to the verge of instability is not an
attractive one from a process operation standpoint. However, it is not absolutely necessary in practice to obtain sustained oscillations (see the section on
Practical Ultimate Gain Tuning Tips). It is also important to realize that some
simple loops cannot be made to oscillate with constant amplitude with just a
proportional controller. Fortunately, these are usually the simplest loops to
control and tune.
24
Its characteristic is that each oscillation has an amplitude that is approximately one-fourth that of the previous oscillation. The formulas proposed by
Ziegler and Nichols1 for calculating the QDR tuning parameters of P, PI, and
PID controllers from the ultimate gain Kcu and period Tu are summarized in
Table 2-1.
Gain
Integral Time
Kc = 0.50 Kcu
PI
Kc = 0.45 Kcu
TI = Tu/1.2
PID
Kc = 0.60 Kcu
TI = Tu/2
Derivative Time
TD = Tu/8
A
A/4
SP
SP
A/4
Disturbance
A
25
It is intuitively obvious that for the proportional (P) controller the gain for a
QDR response should be one-half of the ultimate gain, as Table 2-1 shows. At
the ultimate gain, the maximum error in each direction causes an identical
maximum error in the opposite direction; at one-half the ultimate gain, the
maximum error in each direction is exactly one-half the preceding maximum
error in the opposite direction and one-fourth the previous maximum error in
the same direction. This is the quarter-decay-ratio response.
In Table 2-1 notice that the addition of integral mode results in a reduction of
10% in the QDR gain between the P and the PI controller tuning formulas.
This is due to the additional lag introduced by the integral mode. On the other
hand, the addition of the derivative mode allows increasing the controller
gain by 20% over the proportional controller. Therein lies the justification for
the derivative mode, that is, the increase in the controllability of the loop.
Finally, the derivative and integral times in the PID formulas are in the ratio of
1:4. This is a useful relationship to keep in mind when tuning PID controllers
by trial-and-error (i.e., in those cases when the ultimate gain and period cannot be determined).
26
Using the formulas of Table 2-1, the QDR tuning parameters are:
P controller:
PI controller:
PID controller:
Figure 2-10 shows the response of the controller output and the outlet process
temperature to an increase in process flow for the proportional controller with
a QDR gain of 1.7 and with a lower gain of 1.0. The figure shows that the
lower gain results in a larger initial deviation of the PV and a larger offset, but
the oscillations are smaller and the required variation in controller output is
less.
Kc = 1.7
Kc = 1.0
Kc = 1.7
Kc = 1.0
27
Figures 2-11 and 2-12 show the responses of the PI and PID controllers,
respectively. In each case, the smaller proportional gain results in less oscillatory behavior and less initial movement of the controller output, at the
expense of a larger initial deviation of the PV and a slower return to the set
point. This shows that the tuning parameters, particularly the gain, can be
varied from the values given by the tuning formulas to obtain the desired
response.
Figure 2-11. Proportional-Integral (PI) Controller Responses to a Change in
Process Flow
Kc = 1.5
Kc = 1.0
TI = 5.7 min
Kc = 1.5
Kc = 1.0
Notice the offset in Figure 2 10, and the significant improvement that the
derivative mode produces in the responses of Figure 2-12 over those of
Figure 2-11.
28
Kc = 2.0
Kc = 1.0
TI = 3.4 min
TD = 0.85 min
Kc = 2.0
Kc = 1.0
period gives good enough values of the integral and derivative times. The
proportional gain can then be adjusted to obtain an acceptable response.
For example, notice in Figure 2 7 that, for the case of a gain of 2, the period
of oscillation is 8.0 minutes, which is less than 20% away from the actual
ultimate period (6.8 min).
2. The performance of the feedback controller is not usually sensitive to the
tuning parameters. Thus, when adjusting the parameters from the values
given by the formulas one would be wasting time by changing them by
less than 50%.
3. The recommended parameter adjustment policy is to leave the integral
and derivative times fixed at the values calculated from the tuning formulas, and adjust the gain, up or down, to obtain the desired response.
The QDR tuning formulas allow the tuning of controllers for a specific
response when the ultimate gain and period of the loop can be determined.
29
The chapters that follow present alternative methods for characterizing the
dynamic response of the loop and for tuning feedback controllers. The following section brings up the need for such alternative methods.
2-5. Summary
This chapter has introduced the three modes of the proportional-integralderivative controller and one method to tune it based on the ultimate gain and
period of the closed control loop. The next chapter introduces an open loop
method for characterizing the dynamic response of the process in the loop; the
chapters that follow present tuning formulas based on the parameters of the
open-loop model.
30
References
1. Ziegler, J. G. and Nichols, N. B. Optimum Settings for Automatic Controllers, Transactions of the ASME, Vol. 64, Nov. 1942, p. 759.
Review Questions
2-1. A controller has a gain of 3. For each of the following cases determine by
how much the proportional mode causes the output of the controller to
change and in which direction increase or decrease:
a. The PV increases by 10% and the controller is reverse acting.
b. The PV decreases by 15C, the transmitter range is 0 to 150F, and the
controller is direct acting.
c. The PV increases by 250 kg/hr, the transmitter range is 0 to
50,000 kg/hr and the controller is reverse acting.
2-2.
2-3.
2-4.
A controller is switched to Automatic and its output starts rising immediately and does not stop until it reaches its upper limit. What do you
think is the cause?
2-5.
A controller is switched to Automatic and starts oscillating with increasing amplitude of the oscillations. What would you do to correct this
problem?
31
2-6.
Why do you think that the tuning formulas of Table 2-1 relate the integral and derivative times to the ultimate period of oscillation of the
loop?
2-7.
After tuning a controller using the formulas of Table 2-1 you find the
variation in the controller output when a disturbance upsets the system
is higher than you would like it to be. How would you adjust the tuning
to obtain a more reasonable behavior?
3
Open-loop
Characterization of
Process Dynamics
33
34
35
Steam
OP
TC
Fs
PV
F
Ti
TT
Process
fluid
Steam
trap
Condensate
36
Another reason that generally the process variable does not start changing
immediately is the presence of transportation lag in the loop. This is the lag
caused by the time it takes for the process fluid to move through the process.
However, for most loops real transportation lag, usually of the order of seconds, is negligible relative to the lags that are commonly of the order of minutes. (See Section 3-5 for further discussion of transportation lag.)
37
Gain
The gain determined from the step response is the product of the gains of the
control valve (or other final control element), the process, and the sensor/
transmitter. It must be expressed as percent change in transmitter output per
percent change in controller output, not in engineering units. Figure 3-3
shows the determination of the gain from the step response of Figure 3-2. As
in most control computer-generated plots, the PV is displayed in engineering
units, F, and the final steady change in the PV is (209.5 190) = 19.5F. To
convert to percent of the transmitter output, we need the range of the transmitter. Let us say it is 50 to 250F; the change in the PV is then:
19.5F
-------------------------------- 100% = 9.75%
( 250 50 )F
Since this change is caused by a 5% change in controller output, the gain is:
9.75%
--------------- = 1.95
5%
38
t2 = 15 5 = 10 min
190 + 0.632(19.5) =
202.3F
t1 = 10 5 = 5 min
39
t1 = 10 5 = 5 min
t2 = 15 5 = 10 min
Note that we subtracted 5 minutes from the times read in the response to
determine the times from the application of the step change in controller
output.
( 635.8 638.2 )F
-------------------------------------------- 100% = 0.8%
( 1200 0 )F
40
Figure 3-4. Determination of Loop Parameters from Step Test of a Hot Oil
Temperature Controller
638.5
638.2F
638.0
Temperature, F
637.5
637.0
t1 = 54 sec
636.5
t2 = 87 sec
635.8F
635.5
0
50
100
150
200
time, sec
Measurement
250
300
350
400
Calculated
From the figure, t1 = 54 seconds and t2 = 87 seconds. The data were manually
obtained from the trend recorder on the control system in a refinery and
entered into a spreadsheet. Interpolation was then used on the spreadsheet to
obtain the model parameters. The time constant and dead time are then:
Time constant:
Dead time:
The dashed line in Figure 3-4 is a plot of the response of a process with a gain
of 0.267, a transportation lag of 37 sec, and a time constant of 50 sec.
The negative sign of the gain means that an increase in controller output
results in a decrease in temperature because it causes an increase in the process flow through the coil. The controller must then have direct action.
41
This example illustrates how modern computer control systems allow the precise determination of the step response parameters using a small step change
in controller output that results in a very small change in the process variable
(less than 1%).
This step response method of characterizing the process dynamics applies
only when the process is self-regulating; that is, when it is one that reaches a
new steady state when driven by a sustained change in controller output.
There are two types of processes that are not self-regulating: imbalanced or
integrating processes, and open-loop unstable processes. A typical example of
an imbalanced process is the liquid level in a tank, and an example of an
unstable process is an exothermic chemical reactor. It is obviously impractical
to perform step tests on processes that are not self-regulating. Fortunately,
most processes are self-regulating.
42
Figure 3-5. Physical Systems with Simple Lag Dynamics: (a) Electrical Circuit; (b)
Liquid Storage Tank; (c) Gas Surge Tank; (d) Blending Tank
Inlet flow
ein
eout
Level
Outlet flow
(b)
(a)
Outlet flow
Inlet flow
Kv
F1 C1
F C
V
F2 C2
P V T
Kv
(c)
(d)
The time constant of a lag is defined as the ratio of its capacitance to its conductance or the product of the capacitance times the resistance (the resistance
is the reciprocal of the conductance):
Capacitance
= ---------------------------------- = Capacitance Resistance
Conductance
where is the time constant.
The concepts of capacitance, resistance, and conductance are best understood
by analyzing the physical systems of Figure 3-5. In each of them there is a
physical quantity which is conserved, a rate of flow of that quantity, and a
potential that drives the flow. The capacitance is defined by the amount of
quantity conserved per unit of potential:
Amount of quantity conserved
Capacitance = --------------------------------------------------------------------------------Potential
43
The conductance is the ratio of the flow to the potential that drives it:
Flow of quantity conserved
Conductance = ------------------------------------------------------------------------Potential
To better understand the physical meanings to the terms just presented, consider each of the physical systems of Figure 3-5.
Electrical Circuit (Figure 3-5a). For this system the quantity conserved is electric charge, the potential is electric voltage, and the flow is electric current. The
capacitance is provided by the ability of the capacitor to store electric charge
and the conductance is the reciprocal of the resistance of the electrical resistor.
The time constant is then given by:
= RC
where:
R =
C =
Kv =
(3-1)
44
The conductance of the valve depends on the valve size, and is usually known
in terms of flow per unit pressure drop. Note that the change in pressure drop
across the valve per unit change in level can be calculated by multiplying the
density of the liquid times the local acceleration of gravity.
Gas Surge Tank (Figure 3-5c). This system is analogous to the liquid storage
tank. The quantity conserved is the mass of gas, the potential that drives the
flow through the valve is the pressure in the tank, and the capacitance is provided by the ability of the tank to store gas as it is compressed. The capacitance can be calculated by the formula MV/zRT lb/psi, where V is the volume
of the tank (ft3), R is the ideal gas constant (10.73 psi ft3/lbmole R), z is the
compressibility factor of the gas, M is its molecular weight (lb/lbmole), and T
is its absolute temperature (R). The conductance of the valve is expressed in
change of mass flow per unit change in pressure drop across the valve. The
time constant of the tank can be estimated by the formula:
= (MV/zRT)/Kv
(3-2)
where:
Kv =
(3-3)
45
If there is a chemical reaction, the time constant for the concentration of reactants is decreased because the conductance is increased to the sum (F + kV)
where k is the reaction coefficient, defined here as the change in reaction rate
divided by the change in the reactant concentration. The conductances are
added because the processes of reaction and convection occur in parallel.
Similarly, if there is heat transfer to the surroundings or to a coil or jacket, the
time constant for temperature changes is reduced because the conductance is
increased to the sum (F + [UA/Cp]) where U is the coefficient of heat transfer
(Btu/min-ft2-F), A is the heat transfer area (ft2), is the density of the fluid
(lb/ft3), and Cp is the heat capacity of the fluid (Btu/lbF). In this case the conductances are additive because the processes of conduction and convection
occur in parallel.
For the preceding examples of simple lags the time constant may be estimated
from process parameters and thus a dynamic test on the process is not
needed. For more complex processes such as distillation columns and heat
exchangers, the time constant cannot be estimated because it is made up of
many resistance-capacitance combinations in series and in parallel. For these
systems the only recourse is to perform a dynamic test such as the one presented earlier in this chapter.
46
The conductance of the valve can be estimated from the formulas given by
valve manufacturers to size the valves. Since the pressure drop through the
valve is small compared to the pressure in the tank, the flow is subcritical
and the conductance is given by the following formula:
Kv = W (1 + Pv/P)/(2Pv)
= (100/60)[1 + 5/(30+14.7)]/[(2)(5)] = 0.1853 (lb/min)/psi
The time constant is then:
= 0.0443/0.1853 = 0.24 min (14.3 sec)
The conductance calculated for the valve is the change in gas flow per unit
change in tank pressure, P. It takes into account the variation in gas density
with pressure, and the variation in flow with the square root of the product
of density times the pressure drop across the valve, Pv. For critical flow,
when the pressure drop across the valve is more than one-half the upstream
absolute pressure, the conductance can be calculated by the formula
Kv = W/P.
(3-4)
47
Figure 3-6. Physical Occurrence of Dead Time (Transportation Lag or Time Delay)
and Response
L
C1
C2
C1
Dead time
L/v
C2
Time
Pressure and flow travel at the velocity of sound in the fluid; for example, 340 m/s or 1,100 ft/s for air at ambient temperature.
Temperature, composition, and other fluid properties travel at the
velocity of the fluid, up to about 5 m/s (15 ft/s) for liquids, and up to
about 60 m/s (200 ft/s) for gases.
Solid properties vary at the velocity of the solid (e.g., paper in a paper
machine or coal in a conveyor).
These numbers show that for the reasonable distances, which are typical of
process control systems, pure dead time is only significant for temperature,
composition, and other fluid and solid properties. The velocity of a fluid in a
pipe can be calculated by the following formula:
v = F/Ap
(3-5)
48
where
v
Ap =
Given that (as shall be seen shortly) the dead time makes a feedback loop less
controllable, most process control loops are designed to reduce the dead time
as much as possible. Dead time can be reduced by installing the sensor as
close to the equipment as possible or in the equipment itself.
Although pure dead time is usually not significant for most processes, the
process dead time estimated from the response to the step test arises from
phenomena which are not necessarily transportation lag, but consist of the
presence of two or more lags in series (e.g., the trays in a distillation column).
When these processes are modeled with a simple lag, the dead time is needed
to represent the delay caused by the multiple lags in series. Figure 3-7 shows
the response of composition in a blending train when it consists of one, three,
and five tanks in series, assuming that the total blending volume is the same.
For example, each of the five tanks has one-fifth the volume of the single tank.
As the figure shows, the higher the number of tanks in series, the longer it
takes for the process to start changing and the shorter the total response time.
This behavior makes the model dead-time-to-time-constant ratio higher, making the loop less controllable since it takes the feedback controller longer to
see the change in PV relative to the time it takes the PV to respond.
At the limit, an infinite number of infinitesimal tanks in series results in a pure
dead time equal to the time constant of the single tankthat is, the total volume divided by the volumetric flow.
Most real processes fall somewhere between the two extremes of single perfectly mixed processes, and transportation (unmixed) processes. The simplelag-plus-dead-time (SLPDT) model used to model such processes is the simplest model that can be used for characterizing them. It is the model commonly used to tune the controllers by practitioners in industry and by autotune software.
49
PV
V/3
V/3
V
V/3
/
V/5
V/5
V/5
PV
V/5
V/5
PV
Ft/V
50
51
Valve capacity
70%
60%
50%
40%
30%
20%
10%
0%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Valve position
the controller output, provided that the pressure drop across the valve
remains approximately constant.
Reset Windup. The other type of process nonlinearity is caused by saturation
of the controller output and of the final control element, not necessarily at the
same points. Saturation gives rise to various degrees of the problem known as
reset windup, which happens when the reset or integral mode drives the controller output against one of its limits. Reset windup is worse when the controller output limit is different from the corresponding limit of its destination
(e.g., the position of the control valve).
Reset windup is more common in batch processes and during the startup and
shutdown of continuous processes, but the possibility of windup must always
be kept in mind when tuning controllers. Some apparent tuning problems are
really caused by unexpected reset windup. Chapter 4 looks at reset windup in
more detail.
52
The following example illustrates the variation of the process gain in a process
heater. It takes advantage of the fact that for the heater, the gain can be calculated from a simple steady-state energy balance on the heater.
53
For the linear valve with constant pressure drop, the gain of the valve is
equal to its capacity, (2 kg/s)/100% = 0.02 kg/s/%, and the gain of the transmitter is (100%)/(150C 50C) = 1.0%/C. The dimensionless gain is then:
%
C 0.020kg s
K = 37.5 -------------- ---------------------------1.0 ------- = 0.75
C
%
kg s
Now, if the process were to be run at one-half its full capacity, 8 kg/s, the
gain at this capacity would be:
%
2250kJ kg 0.020kg s
K = ------------------------------------------- ---------------------------1.0 ------- = 1.50
%
C
3.75kJ
8kg
---------- --------------------
s kg * C
This doubling of the process gain could cause the loop to become unstable
if the controller was tuned at the design process flow.
One way to compensate for variable process gain is to use an equal-percentage control valve instead of a linear valve. As discussed above, the equalpercentage valve is designed to have its gain proportional to the valve position, so as the process flow decreases and the control valve closes, the gain
of the valve decreases proportional to the flow.
This example shows the variation of the process gain, indicating that the
process heater is nonlinear. As mentioned earlier, this decrease in process
gain with an increase in flow is characteristic of many process control systems, hence the popularity of equal-percentage control valves, which
exactly compensate for this gain variation.
3-6. Summary
This chapter showed how to perform and analyze a process step test to determine the parameters of a simple-lag-plus-dead-time (SLPDT) model of the
process. These parameters are the gain, the time constant, and the dead time.
It also presented the physical significance of these parameters and showed
how to estimate them from process design parameters for some simple process loops. The chapters to follow will use the estimated dynamic parameters
to design and tune feedback, feedforward, and multivariable controllers.
54
Regardless of the method used to measure the dynamic parameters of a process, it is important to realize that even a rough estimate of the process
dynamic parameters can be quite helpful in tuning and troubleshooting process control systems.
References
1. Smith, C. L., Digital Computer Process Control, Scranton: International Textbook Co., 1972.
2. Ziegler, J. G., and Nichols, N. B., Optimum Settings for Automatic Controllers, Transactions ASME, V. 64, Nov. 1942, p. 759.
Review Questions
3-1. Summarize the procedure for performing an open-loop step test on a
process.
3-2. What are the parameters of a single-lag-plus-dead-time (SLPDT) model
of the process? Give a brief description of each one.
3-3. Figure 3-9 shows the response of the composition out of a reactor to a
step change in the controller output at time 1.0 min. The composition
controller has a range of 0 to 1.5 lb/gal. Estimate the parameters of a simple-lag-plus-dead-time model of the response.
3-4. A passive low-pass filter can be built with a resistor and capacitor. For
use in printed circuit boards, the maximum magnitudes of these components are, respectively, 10 megohms (million ohms) and 100 microfarads
(millionth of farad). What would be the maximum time constant of a filter built with these components?
3-5. The liquid surge tank of Figure 3-5b has an area of 20 ft2 and the valve
has a conductance of 50 gpm/ft of level change (1 ft3 = 7.48 gallons). Estimate the time constant of the response of the level to a step change in
inlet flow.
3-6. The blending tank of Figure 3-5d has a volume of 2000 gallons. Calculate
the time constant of the composition response for product flows of (a) 50
gpm, (b) 500 gpm, and (c) 5000 gpm.
55
3-7. The blending tank of Figure 3-5d mixes 100 gpm of concentrated solution
at 20 lb/gallon with 400 gpm of dilute solution at 2 lb/gallon. Calculate
the steady-state product concentration in lb/gallon. How much would
the outlet concentration change if the concentrated solution rate were to
change to 110 gpm, with all other conditions remaining the same? Calculate the process gain.
3-8. Repeat the previous question assuming that the initial rates are 10 gpm
of concentrated solution and 40 gpm of dilute solution, and the concentrated solution is changed to 11 gpm to do the test. Also estimate the time
constant of the tank for both questions if the tank has a volume of
5,000 gal.
4
How to Tune Feedback
Controllers
This chapter presents formulas for tuning controllers based on the three
parameters obtained from the open-loop step test presented in the previous
chapter: gain, time constant, and dead time.
Learning Objectives When you have completed this chapter, you should be
able to:
A. Tune feedback controllers from estimates of the process gain, time
constant, and dead time.
B.
57
58
Gain
Integral Time
Kc = /Kto
PI
Kc = 0.9/Kto
TI = 3.33to
PID
Kc = 1.2/Kto
TI = 2.0 to
Derivative Time
TD = 0.5to
The formulas of Table 4-1 are very similar to those of Table 2-1. Notice, for
example, that in both sets of formulas the proportional gain of the PI controller is 10% lower and the PID gain is 20% higher than that of the P controller,
and that the derivative or rate time is one-fourth of the integral or reset time
for the PID controller. The ratio of the integral time of the PID controller to
that of the PI controller is 1.7, which is also the same as in Table 2-1; that is, the
derivative mode allows the integral mode to be 1.7 times faster.
The formulas of Table 4-1, however, provide important insight into how the
parameters of the process affect the tuning of the controller and thus the performance of the loop, in particular:
The controller gain is inversely proportional to the process gain K.
Since the process gain represents the product of the gains of all the elements in the loop other than the controller (control valve, process
equipment, and sensor/transmitter), this means that the loop response
depends on the loop gain; that is, it depends on the product of all the
elements in the loop. It also means that if the gain of any of the elements should change because of recalibration, resizing, or nonlinearity
(see Section 3-5), the response of the feedback loop will change unless
the controller gain is readjusted.
59
The controller gain must be reduced when the ratio of the loop dead
time to its time constant increases. This means that the controllability of
the loop decreases when the ratio of the process dead time to its time
constant increases, and leads us to define the ratio of dead time to time
constant as the uncontrollability parameter of the loop:
t
P u = ---0
where:
to = the loop dead time
= the loop time constant
Note that it is the ratio of the dead time to the time constant that determines the degree of uncontrollability of the loop. In other words, a process with a long dead time is not uncontrollable if its time constant is
much longer than its dead time.
The speed of response of the controller, which is determined by the
integral and derivative times, must match the speed of response of the
process. The QDR formulas match these response speeds by relating
the controller time parameters to the process dead time.
These three conclusions can be helpful in guiding the tuning of feedback controllers, even in cases when the tuning formulas cannot be used directly
because the process parameters cannot be accurately estimated. For example,
if the performance of a well-tuned controller should deteriorate during operation, look for a change in the process gain, its uncontrollability parameter, or
its speed of response. At other times the controller performance may be poor
because the integral time is much shorter than the process response time,
because in such a case the process cannot respond as fast as the controller
wants it to respond. The point here is that the speed of response of the process
must be considered when setting the integral time. This is what the tuning formulas of Tables 2-1 and 4-1 do.
The conclusions just drawn from the tuning formulas, coupled with the methods for estimating time constants and dead times given in Sections 3-4 and 35, can also guide the design of the process and its instrumentation. For exam-
60
ple, loop controllability can be improved by reducing the dead time between
the manipulated variable and the sensor, or by increasing the process time
constant. It is also possible to quantitatively estimate the effect of process, control valve, and sensor nonlinearities on the variability of the loop gain and
thus determine the need for readjusting the controller gain when process conditions change.
61
The QDR parameters for a PID controller, also from Table 4-1, are:
Kc = (1.2/1.95) (7.5/2.5) = 1.8
Figure 4-1 compares the PI and PID controller responses of the temperature
transmitter output PV and of the controller output OP using these tuning
parameters for a step increase in process flow to the heater. For this loop with
an uncontrollability parameter of 2.5/7.5 = 0.33, the advantage of adding the
derivative mode is obvious: it produces a smaller initial deviation and maintains the temperature closer to the set point for the entire response, with fewer
oscillations. In addition, the initial change in controller output is not much
greater for the PID response than it is for the PI response.
The next two examples address the question of whether the QDR tuning
parameters will always perform this well, regardless of the degree of loop
controllability.
62
PID
PI
PID
PI
Figure 4-2 shows the responses of the transmitter output PV and the controller output OP to a step change in the disturbance variable. As expected,
the shorter integral time recommended by the QDR formula in comparison
with the IMC rule results in a faster return to the set point with about the
same oscillatory behavior. This is at the expense of a slightly larger initial
change in the controller output.
63
Figure 4-2. Responses of PI Controllers for a Very Controllable Loop Tuned with
QDR and IMC Formulas
TI = 3 min
TI = W = 11.6 min
KKc = 11.6
TI = 3 min
TI = W = 11.6 min
64
Figure 4-3. Responses of PID Controllers For Low Controllability Loop Tuned
with QDR and IMC Formulas
TI = W = 3.5 min
TI = 9 min
KKc = 0.93
TD = 2.2 min
TI = W = 3.5 min
TI = 9 min
65
in set point can cause a large change in the controller output, as shown in Figure 4-4. The response shown in Figure 4-4 is to a 3F change in set point for the
process of Example 4-2. Since the loop gain is 11.6, the figure shows that the
controller output initially changes by over 30%. Note that this is for a process
gain K = 1; if the process gain is higher the change in output will be smaller,
but if the process gain is lower than usual the controller output will have a
larger change. This sudden large change in controller output is bound to cause
a disturbance to other loops. For example, if the controller manipulates the
steam flow as in Figure 3-1, there may be a drop in the steam header pressure
that will affect other systems on that steam line.
Figure 4-4. Responses of a Loop with High Proportional Gain to a Step Change in
Set Point (Set Point Shown as a Dashed Line)
SP
KKc = 11.6
TI = 3 min
66
There are several ways to prevent sudden large changes in controller output
when the set point is changed:
1. Decrease the controller gain so as to prevent the large change in controller
output. This, however, may reduce the tightness of control on disturbance
inputs.
2. Configure the controller so that the proportional mode acts on the process
variable and not on the deviation from the set point. When the set point is
changed, the output will ramp to the new value at a rate controlled by the
integral time.
3. Configure the controller so that the set point is always ramped at an
adjustable rate when the set point is changed.
4. Have the operator slowly change the set point in small steps.
Keep in mind that a large change in controller output happens only when the
controller gain is high, as in Example 4-2. This is not a problem with less controllable loops such as the one in Example 4-3.
A similar problem could occur with the derivative mode except that the
default configuration for the derivative mode is to act on the process variable
and not on the deviation from set point.
There are cases in which set point changes are common, such as with batch
processes and on-line optimization. One recent development in industrial
operations is the incorporation of on-line optimization programs that automatically change controller set points as the optimum conditions change.
Most of these programs contain limits on the size of the set point changes they
make. At any rate, as mentioned above, one sure way to prevent large changes
in controller output with set point changes is to have the proportional mode
act on the process variable instead of on the deviation from set point. As long
as there is integral mode, this option will not affect the performance of the
controller on the disturbance variables.
It is important to recognize that good controller performance on maintaining
the process variable at or near the set point must be balanced against too
much action on the controller output. The reason is that the controller output
usually causes disturbances to other control loops and in some cases manipu-
67
lates safety-sensitive variables. For example, in a furnace temperature controller, the controller output could be manipulating the fuel flow to the furnace. A
large drop in fuel flow could cause a loss of the flame in the firing box.
The examples in this section have illustrated the performance of the controller
when tuned with the parameters of the open-loop test. To summarize our
findings:
Derivative mode provides superior performance for processes with a
high dead-time-to-time-constant ratiothat is, those processes with
high uncontrollability parameter.
Except for controllers that must constantly respond to set point changes
(e.g., slaves in cascade loops; see Chapter 7), the controller should be
tuned for good performance on disturbance inputs, and sudden set
point changes should be limited in magnitude.
For very controllable processes, the tuning formulas call for high loop
gains.
Very uncontrollable processes require low loop gains that result in
large initial deviations of the process variable from its set point. Better
performance is possible with techniques such as dead time compensation (Section 6-4) or feedforward control (Chapter 8).
68
69
When the formulas result in higher gains and shorter integral times than
seem reasonable, let your judgment override the formulas.
6. Tuning very uncontrollable processes.
For processes with uncontrollability parameters of 1 and higher, it is
important to recognize that even an optimally tuned feedback controller
will display poor performance; that is, show large initial deviations on disturbance inputs and slow return to set point. In such cases, improved performance can be achieved through feedforward control (see Chapter 8) or
dead time compensation in the feedback controller (see Section 6-4).
7. Beware of problems that are not related to tuning.
The following problems interfere with the normal operation of a controller
and although they may appear to be tuning problems, they are not:
Reset windup, caused by saturation of the controller output (see Section 4-3).
Interaction between loops (see Chapter 9).
Processes with inverse or overshoot response, caused by parallel effects
of opposite direction between a process input and the controlled variable (see Section 4-4).
Changes in process parameters because of nonlinearities, which must
be handled by adaptive control methods.
Control valve hysteresisthat is, the valve stops at a different position
than the one desired, and the difference changes directions depending
on the direction of motion of the valve. Hysteresis is due to dry friction
on the valve packing. It causes the controller output to oscillate around
the desired position of the valve.
Limit cycles due to nonlinear behavior.
All of these problems cause poor feedback controller performance that must
be handled by means other than controller tuning; for example decoupling,
feedforward control, adaptive control, or the use of valve positioners. The
chapters that follow present these techniques.
70
71
Valve opened
ple, 50% or if known, to a value near the required output (20% in this
example) before switching the controller to Automatic.
The problem of reset windup can also occur during normal operation when
there is a gap between the limits of the controller output and the operating
limits of the valve position or other manipulated variable. For example, if in
case of Figure 4-5 the controller output were limited to -10% to 110% while the
valve operates between 0 and 100%, windup would occur when a large disturbance causes the controller output to enter the gap. This is because the controller has no effect on the process variable while its output is in the gap and
the integral mode keeps the controller output in the gap until the process variable crosses the set point, resulting in an overshoot.
72
a step change. A typical example of a process with inverse response is an exothermic reactor with the feed colder than the reactor. An increase in the feed
rate to the reactor causes the temperature to initially drop due to the larger
rate in cold feed. However, the increase in the inlet flow of the reactants eventually increases the rate of the reaction, and with it the rate of heat generation
by the reaction. This causes the temperature in the reactor to end up higher
than it was initially.
Another typical example of inverse response is the level in the steam drum of
a water tube boiler when the steam demand changes. The inverse response is
caused by the phenomena of swell and shrink of the steam bubbles in the
boiler tubes.
Figure 4-6 shows the response of the temperature control of an exothermic
reactor to a step increase in reactant flow followed by a step increase in the
inlet temperature. The temperature is controlled by a PID controller tuned by
the QDR formulas. Since the reactants enter at a lower temperature than the
reactor, the reactor temperature initially decreases, but as the reactant concentration increases, the reaction rate and corresponding rate of heat generation
increase, causing the reactor temperature to increase. The initial drop in temperature fools the controller into increasing its output to decrease the coolant
flow (the control valve fails open), but eventually a lower controller output
higher coolant flowis required because of the higher rate of heat generation.
As the figure shows, the result is a very oscillatory response, particularly in
the controller output. By comparison, the step decrease in inlet temperature
results in lower amplitude of the oscillations because there is no inverse
response to the inlet temperature. Note that the temperature control loop is
very controllable because the reactor has a high capacitance and hardly any
dead time; this is why the deviations in the temperature are so small.
Feedforward control (Chapter 8) can compensate for the effect of inverse
response to improve the performance of the feedback controller when necessary. The feedforward model considers the long-term effect of the disturbance
and takes action that cancels out the initial change in the feedback controller
output in the wrong direction. This is basically how the swell and shrink
problems in boiler levels are handled by two- and three-element boiler level
control systems.
73
Step increase in
Step decrease in
reactants flow
inlet temperature
When the inverse response is to a change in the controller output, the loop
becomes very uncontrollable. For example, if in the reactor of Figure 4-6 the
temperature were controlled by manipulating the reactant flow, every action
by the controller would be followed by an immediate change in the temperature in the wrong direction. This would be worse than if the process had a
dead time equal to the duration of the inverse response. Fortunately, such a
situation is extremely rare, but it should be kept in mind when troubleshooting difficult tuning problems.
74
4-6. Summary
This chapter presented a simple set of tuning formulas for feedback
controllers based on the parameters of the open-loop test: the gain, the time
constant, and the dead time. The set of formulas originally proposed by
Ziegler and Nichols3 for quarter-decay-ratio response was proposed with a
limitation on the integral time when the process is very uncontrollable. The
limit on the integral time is the time constant, which is the value proposed by
the IMC formulas.
The effectiveness of the tuning formulas was demonstrated for normal processes as well as for very controllable and very uncontrollable processes. The
advantage of the PID over the PI controller was also demonstrated.
Helpful tuning hints were presented for those instances when the open-loop
test cannot be performed.
Several process characteristics that reduce the performance of the controller
were presented, namely reset windup, inverse response, and process nonlin-
75
Figure 4-7. The Response Is More Oscillatory at Half Production Rate because the
Loop Gain is Twice as High than at Full Rate
earity. The problem of large initial changes in the controller output when the
set point is changed and when the controller gain is high was also discussed.
The next chapter presents the selection of controller modes and tuning for a
number of common control loops.
References
1. Martin, J. Jr., Corripio, A. B. and Smith, C. L. How to Select Controller
Modes and Tuning Parameters from Simple Process Models, ISA Transactions, V. 15 (Apr. 1976), pp. 314-319.
2. Rivera, D. E., Morari, M. and Skogestad, S. Internal Model Control, 4. PID
Controller Design, Industrial and Engineering Chemistry Process Design and
Development, V. 25 (1986), p. 252.
76
3. Ziegler, J. G. and Nichols, N. B. Optimum Settings for Automatic Controllers, Transactions ASME, V. 64, (Nov. 1942), p. 759.
Review Questions
4-1. Based on the tuning formulas given in this chapter, how must you
change the controller gain if, after the controller is tuned, the process
gain were to double because of the nonlinear behavior of the process?
4-2. How is the uncontrollability of a feedback loop measured?
4-3. Assuming that the quarter-decay-ratio formulas of Table 4-1 give the
same tuning parameters as those of Table 2-1, what relationship can be
established between the controller ultimate gain and the gain with the
uncontrollability parameter of the process in the loop? What is the relationship between the ultimate period and the process dead time?
4-4. Compare the following processes in regards to sensitivity, speed of
response, and controllability:
Process A
Process B
Process C
Gain
0.5
2.0
4.0
2.0
30
5.0
0.2
3.0
3.0
4-5. Estimate the tuning parameters of a PID controller for the three processes of question 4-4.
4-6. Why would one want to configure the controller so that the proportional
mode acts on the process variable and not the deviation from set point?
What would the response of the controller output be when the controller
is configured as such and the set point is changed?
4-7. What is the typical symptom of reset windup? What causes it? How can
it be prevented?
4-8. What is known as inverse response? What effect does it have on the performance of a feedback controller and why?
5
Mode Selection and
Tuning of Common
Feedback Loops
The preceding chapters dealt with the tuning of feedback controllers for general processes that can be represented by a single-lag-plus-deadtime
(SLPDT) model. This chapter presents tuning guidelines for the most typical
process control loops, specifically flow, level, pressure, temperature, and composition control loops.
C. Design and tune simple feedback controllers for flow, level, pressure,
temperature, and composition.
77
78
79
gas surge tanks, because in these cases the purpose of the tank is to attenuate
variations in process flow.
SP
FC
FT
80
However, when the flow controller is the slave in a cascade control scheme
(see Chapter 7), it is important for the flow to respond quickly to set point
changes. This requires a proportional-integral controller with a gain near
unity, which to maintain stability may require an increase in the integral time
from the few seconds normally used in flow controllers. The IMC2 tuning
rules (see Section 4-1) suggest that the integral time be set equal to the time
constant of the loop, usually that of the control valve actuator. In cascade situations, tight flow control is indicated.
The proportional gain should also be increased when hysteresis of the control
valve causes variations in the flow around its set point. As mentioned in Section 4-2, hysteresis is caused by static friction in the valve packing that creates
a difference between the actual valve position and the corresponding controller output. The error changes direction according to the direction in which the
valve stem must move, and this causes a dead band around the desired valve
position; that is, a band within which the valve does not respond to changes in
the controller output. Increasing the flow controller gain reduces the amplitude of the flow variations caused by hysteresis. A valve positioner also
reduces hysteresis and speeds up the valve response, but positioners are usually difficult to cost-justify for flow control loops.
81
Figure 5-2. Oscillations on a Flow Control Loop with Valve Hysteresis are
Reduced in Amplitude with a Higher Controller Gain
Kc = 0.25
TI = 0.2 min
Kc = 1.5
TI = 0.2 min
Tight Control
One example of tight liquid level control and one example of tight pressure
control are shown in Figure 5-3. The control of level in natural-circulation
evaporators and reboilers is important because too low a level causes deposits
on the bare hot tubes and overheating of the tubes at the top. Conversely, too
high a level causes elevation of the boiling point, reducing the heat transfer
rate and preventing the formation of bubbles, which enhances heat transfer by
promoting turbulence. The example of tight pressure control or pressure regu-
82
83
Figure 5-3. Examples of Tight Control: (a) Evaporator Level; (b) Supply Header
Pressure
Vapors
Feed
Steam
LC
LT
Condensate
(a)
Product
PC
PT
(b)
Loads
Supply
84
Figure 5-4. Examples of Averaging Level Control: (a) Surge Tank; (b) Condenser
Accumulator
Feeds
SP
Surge
tank
LT
LC
Outlet flow
(a)
Vapors
Condenser
Accumulator
Column
LT
LC
FC
FT
(b)
Reflux
Distillate
85
at the set point, a term that is only important when the controller has no integral mode. This will cause the outlet valve to be fully open when the level is at
100% of range and fully closed when the level is at 0% of range, thus using the
full capacity of the valve and of the tank. A proportional gain higher than
unity would reduce the effective capacity of the tank for smoothing variations
in flow, while a gain lower than unity would reduce the effective capacity of
the control valve and create the possibility of the tank overflowing or running
dry. With the proposed design the tank behaves as a low-pass filter to flow
variations; a low-pass filter allows low-frequency input through while it
attenuates high frequency variations. The time constant of such a filter is:
A ( h max h min )
= ----------------------------------------K c F max
(5-1)
where:
A
hmin and hmax = the low and high points of the range of the level
transmitter, respectively, ft
Fmax = the maximum flow through the control valve when opened fully
(100% controller output), ft3/min
Kc
The controller gain is assumed to be 1.0 in this design. When the level controller is cascaded to a flow controller, Fmax is the upper limit of the range of the
flow transmitter in the flow control loop. Note that a proportional gain greater
than unity results in a reduction of the filter time constant and therefore less
smoothing of the variations in flow. A good way to see it is to note that doubling the gain would be equivalent to reducing either the tank area or the
transmitter range by a factor of two, thus reducing the effective capacity of the
tank. On the other hand, reducing the controller gain to half would be equivalent to reducing the capacity of the valve by half, thus increasing the possibility of the tank overflowing.
Although averaging level control can be accomplished by a simple proportional controller, most level control applications use PI controllers. This is
because control room operators have an aversion to variables that are not at
86
their set points. The process in a level control loop is unlike most other loops
in that it does not self-regulate; that is, the level tends to continuously rise or
fall when the feedback controller is not in Automatic. This usually means that
for level control loops, a time constant cannot be determined by an open-loop
test. Even when there is some degree of self-regulation, the process time constant is very long, on the order of one hour or longer. Because of this, PI controllers in level control loops have the following characteristics:
The level, and the flow that is manipulated to control it, oscillate for a
long period. Sometimes the period is so long that the oscillation is
imperceptible, unless it is trended over a very long time.
The shorter the integral time, the shorter the period of oscillation.
The level control loop is unstable when the integral time is equal to or
shorter than the time constant of the control valve.
Unlike most other loops, there is a range of controller gains over which
the oscillations increase as the controller gain is decreased.
This leads to the following general rules for tuning PI controllers for averaging level control:
Set the integral time to 60 minutes or longer.
Set the proportional gain to at least 1.0.
Averaging pressure control is not as common as averaging level control
because in the case of gas systems, a simple fixed resistance on the outlet of
the surge tank is usually all that is required to smooth out variations in flow.
87
Cp
(5-2)
88
Figure 5-5. Responses of Averaging Level Control (Continuous Lines) and Tight
Level Control (Dashed Lines) on a Surge Tank
Level
Kc = 1
Kc = 20
TI = 20 min
Outlet flow
Kc = 20
Kc = 1
When these units are used, the time constant is calculated in seconds.
Temperature is the variable most often controlled in chemical reactors, furnaces, and heat exchangers. When the temperature controller manipulates the
flow of steam (Figure 3-1) or fuel to a heater or furnace (Figure 5-6), the rate of
heat supplied to the process fluid is proportional to the flow of steam or fuel.
This is because the heat of condensation of the steam and the heating value of
the fuel remain approximately constant with load. However, when the manipulated variable is cooling water or hot oil, the rate of heat removed or supplied to the process fluid is very nonlinear with water or oil flow because the
heat transfer rate requires that the outlet utility temperature moves closer to
its inlet temperature as the heat transfer rate increases. This means that it
requires higher increments in flow for equal increments in heat rate as the
load increases. To reduce the nonlinear nature of the loop, the temperature
controller TC is sometimes cascaded to a heat rate controller (QC), as in Figure
5-7. The process variable for the heat rate controller is the rate of heat transfer
89
in the exchanger, which is proportional to the flow and to the change in temperature of the hot oil:
Q = FoilCp(Toin - Toout)
where:
Q
Foil
Cp
90
SP
Process
flow
TC
TT
Air
Fuel
91
Hot oil
Toin
TT
Foil
FT
SP
TC
SP
QC
Process
in
TT
Toout
TT
Process
out
92
the order of several process time constants, theories such as IMC2 and controller synthesis1 call for a pure integral controller. This is because the process
responds fast relative to the time frame in which the analysis is done. Chapter
6 discusses the tuning of controllers that make use of sampled, rather than
continuous, measurements.
5-6. Summary
This chapter presented some guidelines for selecting and tuning feedback
controllers for several common process variables. While flow control calls for
fast PI controllers with low gains, level and pressure control can be achieved
with simple proportional controllers with high or low gains, depending on
whether the objective is tight control or smoothing of flow disturbances.
When PI controllers are used for level control, the integral time should be
long, on the order of one hour or longer. PID controllers are commonly used
for temperature and analyzer control.
References
1. Martin, J. Jr., Corripio, A. B. and Smith, C. L. How to Select Controller
Modes and Tuning Parameters from Simple Process Models, ISA Transactions, V. 15 (Apr. 1976), pp. 314-319.
2. Rivera, D. E., Morari, M. and Skogestad, S. Internal Model Control, 4. PID
Controller Design, Industrial and Engineering Chemistry Process Design and
Development, V. 25 (1986), p. 252.
Review Questions
5-1. Briefly state the difference between tight level control and averaging
level control. In which of the two is it important to maintain the level at
the set point? Give an example of each.
5-2. What type of controller is recommended for flow control loops? Indicate
typical values for the gain and integral times.
5-3. What type of controller is indicated for tight level control? Indicate typical gains for the controller.
93
5-4. What type of controller is indicated for averaging level control? Indicate
typical gains for the controller.
5-5. When a PI controller is used for averaging level control, what should the
integral time be? Would an increase in gain increase oscillations or
decrease oscillations?
5-6. Estimate the time constant of a temperature sensor weighing 0.03 kg,
with a specific heat of 23 kJ/kg-C. The thermowell has a contact area of
0.012 m2 and the heat transfer coefficient is 0.6 kW/m2-C.
5-7. Why are PID controllers commonly used for controlling temperature?
5-8. What is the major problem with the control of composition?
6
Tuning Sampled-Data
Control Loops
This chapter deals with tuning methods for loopssuch as analyzer control
loopsin which the process variable cannot be measured continuously. In
such loops the process variable must be sampled at discrete intervals of time,
at which the control calculations are carried out and the controller output is
updated, to be kept constant until the next update.
Learning ObjectivesWhen you have completed this chapter you should be able
to:
A. Understand the effect of sampling on control loop performance.
B.
95
96
97
derivative mode work on the process variable PVk instead of on the deviation
from set point. The PD calculation also contains a filter with time constant
TD, which is intended to limit the magnitude of pulses in the controller output upon sudden changes in the process variable. It is seldom desirable for the
derivative mode of the controller to respond to set point changes, because on a
set point change there would be a large change in the controller output lasting
for just one samplethat is, a large undesirable output pulse known as a
derivative kick. Such pulses are completely avoided by the algorithm of
Table 6-1 since the derivative mode, acting on the process variable, does not
see changes in set point.
Table 6-1. Discrete PID Control Algorithm
Proportional-Derivative (P+D) unit:
T D
( + 1 )T D
T
Y k = -------------------- Y k 1 + -------------------- PV k + ------------------------- ( PV k PV k 1 )
T + T D
T + T D
T + T D
Deviation from set point:
Ek = SPk - Yk
Increment in controller output:
T
M k = K c E k E k 1 + ------ E k
T
I
Controller output:
Mk = Mk-1 + Mk
where:
SPk = set point
PVk = process variable (measurement)
Mk = controller output
Ek = error or set point deviation
= derivative filter parameter
T = sampling interval, min
Kc = proportional gain
TI = integral time
TD = derivative time
The deviation from the set point or error (E) in Table 6-1 is for a reverse acting
controller. For a direct-acting controller the terms Yk and SPk are reversed in
the formula or, alternatively, the controller gain is set to a negative value.
When either the derivative time TD is set to zero (PI controller) or the process
98
variable reaches a steady value, Yk = PVk, the algorithm still drives the process variable to the set point.
The filter parameter of Table 6-1 has a special meaning; its reciprocal, 1/, is
the amplification factor on the change of the PV at each sampling instant and
is also called the dynamic gain limit. Note that, if were set to zero, the
amplification factor on the change in PV would have no limit. For example, if
the sampling interval is one second (1/60 min) and the derivative time is one
minute, the change in PV at each sample with = 0 would be multiplied by a
factor of 60 (TD/T = 60). By setting the nonadjustable parameter to a
reasonable value, say 0.1, the change in PV cannot be amplified by a factor
greater than 10, independent of the sampling interval and the derivative time.
The dynamic limit permits setting the derivative time to any desired value
without the danger of introducing large undesirable pulses in the controller
output.
0.
1.
2.
3.
4.
5.
10.
20.
40.
PVk
0.
1.
2.
3.
4.
5.
10.
20.
40.
Yk
0.
8.5
15.2
20.3
24.5
27.9
38.3
49.9
70.0
Ideal
30.
31.0
32.0
33.0
34.0
35.0
40.0
50.0
70.0
99
For example, for the value of Y1, at the first sample Y0 = 0, PV1 = 1%,
PV0 = 0%, TD = 0.5(60) = 30 s, so:
( 0.1 + 1 )30
1
0.1 ( 30 )
Y 1 = ---------------------------- 0 + ---------------------------- 1 + ---------------------------- ( 1 0 ) = 0 + 0.25 + 8.25 = 8.50
1 + 0.1 ( 30 )
1 + 0.1 ( 30 )
1 + 0.1 ( 30 )
Notice that the unfiltered (ideal) derivative unit jumps to 30 at time 0 with
increments of 1 at each sample. Both of these responses are shown graphically in Figure 6-1. The unfiltered derivative unit is leading the PV by one
derivative time (30 s), while the derivative unit with the filter, after a brief
lag, also leads the PV by one derivative time. In practice, the lag is too small
to significantly affect the performance of the controller.
70
60
Yk
50
D=0
40
30
PVk
D = 0.1
20
10
TD = 30 T
0
0
10
15
20
25
t/T
30
35
40
45
100
101
(6-1)
where:
KL
102
80%
OP
60%
40%
20%
0%
0%
20%
40%
60%
80%
100%
PV
Prevention of the tanks overflowing or running dry requires that the valve
be fully open when the level is at 100% of the range and fully closed when
the level is at 0%. Since the set point is 50%, either requirement takes effect
when the magnitude of the error is 50%. With the output bias of 50%,
using the upper limit requirement in Equation 6-1:
100% = 50% + Kc(100% - 50%)
= 50% + 0.25[1 + KNL(50%)](50%)
KNL = [(100% - 50%)/(0.25)(50%) - 1]/50% = 0.06
The controller gain at each end of the range is:
Kc = 0.25[1 + 0.06(50%)] = 1.0
103
Recall from Equation 5-1 that the time constant of the tank is inversely proportional to the controller gain, thus the effective capacity of the tank, as
used for smoothing flow variations, can be increased from its real value at
full and zero flow to four times that value at one-half of full flow.
Section Summary
This section presented the computer- and microprocessor-based PID control
algorithm and the options that are made available by its configurable nature.
The next section concerns the tuning of sampled-data controllers.
104
3, and such a step will always take place at a sampling instant and will remain
constant after that.
Moore and his coworkers3 developed a simple correction of the controller tuning parameters for the effect of sampling. They point out that when a continuous signal is sampled at regular intervals of time and then reconstructed by
holding the sampled values constant for each sampling period, the reconstructed signal is effectively delayed by approximately one-half the sampling
interval, as shown by the dashed line in Figure 6-3. In sampled-data systems
the controller output is held constant between updates, thus adding one-half
the sampling interval to the dead time of the rest of the loop components. The
correction for sampling is then simply to add one-half the sampling interval to
the dead time obtained from the step response. The uncontrollability parameter is then given by:
T
t 0 + --2
P u = --------------
(6-2)
where:
Pu
t0
105
Figure 6-3. Effective Delay of Sampling and Holding a Signal is One Half the
Sampling Interval
Let
t
N = -----oT
a = e
T
--
Proportional gain:
( 1 q )a
K c = ------------------------------------------------------------------------------K( 1 a)[1 + N(1 q)]
Integral time:
aT
T I = ---------------1a
106
exact fit to the process response, the value of q would be the fraction of the
error at any one sample that would remain after one dead time plus one sampling interval. For example, setting q = 0 specifies the process variable to
match the set point after N + 1 samples, where N is the number of samples of
dead timethat is, the sample time divided by the sampling interval. This
would result in the highest gain, and therefore in the tightest control. However, for any value of q, the tightness of the closed-loop response depends on
the ratio of the sampling interval to the time constant, T/. A more fundamental adjustable parameter is the closed-loop time constant c, which can be
related to the time parameters of the loopshort for fast processes and long
for slow processes. If c is specified, the value of q can be computed by:
q = e
T --c
Setting q = 0 results in an upper limit for the controller gain. This value can be
used as a guide for the initial tuning of the controller. As is the case with the
tuning formulas of Chapter 4, the upper limit of the controller gain decreases
with increasing the process dead time, parameter N, in number of samples.
The formulas of Table 6-3 are intended to tune only a PI controller. Two time
constants plus a dead time would be required to tune a PID controller by this
procedure, but it is difficult to accurately determine more than one time constant and a dead time from a simple open-loop step test.
As mentioned earlier, the formulas of Table 6-3 are applicable to any value of
the loop parameters and the sampling interval; moreover, the controller gain
can be adjusted to obtain fast response with reasonable variation of the controller output. They are highly recommended because they relate the integral
and derivative times to the process time constants, thus reducing the tuning
procedure to the adjustment of the controller gain.
107
The process parameters for the heater were determined in Section 3-1:
K = 1.95
= 7.5 min
t0 = 2.5 min
The tuning parameters, for a sampling interval of 0.5 min and q = 0, are:
N = 2.5 0.5 = 5
a = e
0.5
------7.5
( 1 0 )0.936
K c = ------------------------------------------------------------------------------------------ = 1.2
1.95 ( 1 0.936 ) ( 1 0 ) [ 1 + 5 ( 1 0 ) ]
= 0.936
0.5 ( 0.936 )
T I = -------------------------- = 7.3 min
1 0.936
TD = 0
For the other sampling times the tuning parameters are given in Table 6-4.
1/60
1/6
0.5
1.0
2.0
4.0
Dead time, N
150
15
Maximum Kc (q = 0)
1.5
1.4
1.2
1.2
0.84
0.72
7.5
7.4
7.3
7.0
6.5
5.7
Notice that the gain is lower and the integral time is shorter as the sampling
interval is increased. This means that the loop is less controllable at the longer
sampling interval. Also notice that there is a very small change in the tuning
parameters when the sampling frequency is increased from 6 times per minute to 60 times per minute. Since most control systems sample at the rate of
more than once per second, the effect of sampling is negligible for most control loops.
Figure 6-4 shows the responses of the heater temperature controller to a step
change in disturbance with the tuning parameters given in Table 6-4 and sampling intervals of 0.5, 1.0, and 2.0 min. The figure shows that there is little difference in controller performance for sampling interval times of 0.5 and 1 min,
but the performance deteriorates with a sampling interval of 2 min.
108
T = 1 min
T = 0.5 min
T = 2 min
(6-3)
109
Notice that for the case N = 0 and q = 0, the controller gain is the reciprocal of
the process gain. This result makes sense, since a loop gain of 1.0 is what is
needed to reduce the error to zero in one sampling interval if the process
reaches steady state during that interval. An interesting application of this
case is a chromatographic analyzer sampling a fast process. The nature of
such an analyzer is that the composition is not available to the controller until
the end of the analysis cycle, because it takes a full analyzer cycle to separate
the mixture and analyze it. This means that the process dead time is approximately one sampling interval, or N = 1. For q = 0, Equation 6-3 gives a gain of
KI = 1/K(1 + 1) = 1/2K, or one-half the reciprocal of the process gain. This also
makes sense, because when action is taken by the controller, it takes two sampling intervals for the controller to see the result of that action, so the formula
says to spread the corrective action equally over two samples.
10
20
Dead time, N
Maximum gain (q = 0)
0.54
0.18
0.038
5.2
3.6
1.5
0.51
0.51
0.51
110
ters are those shown in Table 6-5 except that the gain was reduced by half (q
= 0.5) with the sampling interval of 5 min, and to one-fourth (q = 0.75) with
the sampling intervals of 10 and 20 min. As expected, controller performance is a lot slower than it is in the responses of Figure 6-4, but still
acceptable when the sampling interval is forced to be that long. This shows
that the tuning formulas of Table 6-2 can be applied to a wide range of sampling-interval-to-time-constant ratios.
T = 5 min
q = 0.5
T = 10 min
q = 0.75
T = 20 min
q = 0.75
To summarize, the formulas presented in Table 6-2 can be used with the openloop test model, resulting in a PI controller. They are applicable over a wide
range of sampling intervals and dead-time-to-time-constant ratios. Although
formulas used to tune PID controllers have not been developed by this
method, this does not present a problem because the derivative mode should
not be effective when the process variable is sampled slowly.
111
112
The increment in output contributed by the proportional and derivative modes is very small because the process variable changes very little from one sample to the next.
The increment in output contributed by the integral mode is very small
because the ratio of the sampling interval to the integral time, T/TI, is
very small.
For example, the pressure control loop of an ammonia synthesis loop, where
the pressure is controlled with the purge flow, has a time constant of about 40
minutes (2,400 s). If the controller is updated 10 times per second and the integral time is of the order of magnitude of the time constant, the term T/TI in
Table 6-1 is 0.1/2,400 = 0.00004. This makes the controller output, when the
error is 1% and the output 50%, equal to 50.00004%, so that there is an imperceptible change on the position of the valve after each update of the controller
output. The reason this is not commonly a problem is that the controller output is usually computed with double precision (twice as many digits as normal in the representation of the output result) and the output can be
accurately computed so that after many samples (about 10,000 in this case),
the small increments are not lost.
In our experience, the control loop performance is acceptable when the controller update frequency is as high as one-tenth the loop time constant. At one
time, Fisher Controls marketed a very successful computer control package
for ammonia plants that used a sampling interval of 5 minutes to control the
synthesis loop pressure with the purge flow.
113
114
ler, so that it ends up with too many adjustable parameters: the model parameters plus the controller tuning parameters. Because there are so many
parameters to adjust, there is no convenient way to adjust the closed-loop
response when the model does not properly fit the process. Given the nonlinear nature of process dynamics, any technique that depends heavily on exact
process modeling is doomed to fail.
Figure 6-6. Block Diagram of the Smith Predictor for Dead Time Compensation
Disturbances
SP
Controller
Process
with Dead
Time
OP
PV
+
Process
Model
Corrected model output
Model
Dead Time
+
+
Model error
(6-4)
115
where Mk can be computed by the control algorithm of Table 6-1. The last
term in the calculation of the output provides the dead time compensation.
Note that the term vanishes when there is no dead time, N = 0. The actual controller is tuned with the formulas of Table 6-2 except for the controller gain,
which is given by:
( 1 q )a
K c = --------------------K(1 a)
(6-5)
116
Figure 6-7 compares the steam heater responses of the controllers to a step
increase in process flow to the heater. The dead time compensation controller
results in a smaller deviation from set point and less oscillation than the regular PI controller. The improvement in performance is not spectacular, probably because this is a relatively controllable process. The value of q = 0.67 was
selected to prevent excessive oscillation in the controller output. The proportional gains are then one-third, (1 0.67) = 0.33, of the maximum gains.
More sophisticated dynamic compensation controllers have been proposed in
the last few years, for example the Vogel-Edgar controller6 and Internal
Model Control2. These controllers can incorporate a more precise compensator than the Dahlin Controller, provided that a precise model of the process is
available. Nevertheless, the Dahlin Controller has been applied successfully to
the control of paper machines and other processes with high dead-time-totime-constant ratios.
117
Figure 6-7. Response of Heater Temperature Control with and without Dead
Time Compensation
PI
controller
with dead time
compensation
Standard PI
controller
6-5. Summary
This chapter presented various sampled-data feedback controllers, how to
tune them, and how to select the sampling interval for them. The control algorithm of Table 6-1 and the tuning formulas of Table 6-2 are strongly recommended, in addition to being the most commonly used in sampled-data
control applications. For processes with high dead-time-to-time-constant
ratios, the Dahlin Controller, Equation 6-4, is commonly used in industry and
is also recommended here.
118
References
1. Dahlin, E. B. Designing and Tuning Digital Controllers, Instruments and
Control Systems, V. 41, June 1968, p. 77.
2. Garcia, C. E. and Morari, M. Internal Model Control, 1. A Unifying
Review and Some Results, Industrial and Engineering Chemistry Process
Design and Development, V. 21, 1982, pp. 308-323.
3. Moore, C. F., Smith, C. L. and Murrill, P. W. Simplifying Digital Control
Dynamics for Controller Tuning and Hardware Lag Effects, Instrument
Practice, V. 23, Jan. 1969, p. 45.
4. Smith, C. A. and Corripio, A. B. Principles and Practice of Automatic Process
Control 2nd ed., New York: Wiley, 1997, Chapter 15.
5. Smith, O. J. M. Closer Control of Loops with Dead Time, Chemical Engineering Progress, V. 53, May 1957, pp. 217-219.
6. Vogel, E. F. and Edgar, T. F. A New Dead Time Compensator for Digital
Control, Proceedings ISA/80, Research Triangle Park: ISA, 1980.
Review Questions
6-1. How are computer- and microprocessor-based controllers different from
analog controllers?
6-2. What is derivative kick? How is it prevented? Why is a dynamic gain
limit needed in the derivative term of the PID controller?
6-3. How and why would you eliminate proportional kick on set point
changes? Will the process variable approach its set point faster or slower
when proportional kick is avoided? When must proportional kick be
allowed?
6-4. What is the advantage of a nonlinear proportional gain in averaging
level control situations? In such a case, what must the nonlinear gain be
for the gain to be 0.2 at zero error and still have the controller output
reach its limits when the level reaches its limits (0 and 100%)? Assume a
level set point of 50% and an output bias of 50%.
119
6-5. A process has a gain of 1.6%, a time constant of 20 min and a dead time
of 5 min. Calculate the tuning parameters for a discrete controller if the
sampling interval is (a) 4 s, (b) 1 min, (c) 5 min and (d) 50 min.
6-6. Repeat question 6-5, but for a controller with dead time compensation.
Specify also how many samples of dead time compensation, N, must be
used in each case.
6-7. What is the basic idea behind the Smith Predictor? What is its major disadvantage? How does the Dahlin Controller with dead time compensation overcome the disadvantage of the Smith Predictor?
7
Tuning Cascade Control
Systems
Cascade control is a common strategy for improving the performance of process control loops. In its simplest form it consists of closing a feedback loop
inside the primary control loop by measuring an intermediate process variable. This chapter presents an overview of cascade control and the tuning of
cascade control systems.
Select the control modes and tune the controllers in a cascade control
system.
121
122
123
SP
TC
1
Reactants
TC
2
TT
2
TT
1
Water
out
Coolant
Products
Steam
124
SP
Master
Controller
TC-1
+
-
+
PV2
PV1
Slave
Controller
TC-2
TJ
Jacket
TR
Reactor
Jacket
TT
2
Reactor
TT
1
Disturbances into the inner loop will not be eliminated fast enough to
avoid their affecting the primary control variable.
Speeding up of the inner loop would result in a decrease in the controllability of the overall loop because its dead-time-to-time-constant ratio
would increase.
Nonlinearities would become a part of the slower and possibly less
controllable inner loop, thus affecting the stability of the control
system.
Besides the inner loop having to be faster to respond than the outer loop, the
success of cascade control also requires that the sensor of the inner loop be fast
and reliable. One would not consider, for example, cascading a temperature
controller to a chromatographic analyzer controller. On the other hand, the
sensor for the inner loop does not have to be accurate, only repeatable,
because the integral mode in the primary controller compensates for errors in
the measurement of the secondary variable. In other words, it is acceptable for
the inner loop sensor to be wrong as long as it is consistently wrong, and to
the same degree.
Finally, cascade control would not be able to improve the performance of
loops that are already very controllable; for example, this occurs with liquid
125
level and gas pressure control loops, or when the controlled variable does not
have to be maintained tightly around its set point, as in averaging level control. When a level controller is cascaded to a flow controller, it is usually justified by the greater flexibility in the operation of the process, not by improved
control performance.
The above sections have examined the reasons and requirements for using
cascade control. The following sections will look at how to select the controller
modes for cascade control systems and how to tune them.
126
element than when a single feedback loop is used. This amplification results
in a faster response of the primary loop.
127
128
After each inner loop has been tuned in succession, the primary loop can be
tuned to follow any desired performance criteria by any of the methods of
Chapters 2, 4, 5 and 6. Given that what is special in cascade systems is the tuning of the secondary loop(s), some typical secondary loops, namely flow, temperature, and pressure, are briefly discussed next. Keep in mind, however,
that any variableincluding compositioncan be used as a secondary variable provided it can be measured quickly and reliably (e.g., when a simple
continuous thermal conductivity detector is used to measure the hydrogen
composition in the ammonia synthesis process - see Example 7-2).
129
Vapors to
condenser
SP
TC
TT
Column
SP
FC
FT
Reflux
The reactor temperature control scheme of Figure 7-1 is a typical example of a
secondary temperature controller. In this application, temperature has the
advantage over coolant flow as a secondary variable in that it compensates for
changes in both coolant header pressure and inlet temperature, while coolant
flow only compensates for variations in coolant header pressure. The
temperature controller also closes a loop around the jacket, reducing its
effective time constant and thus making the reactor temperature control loop
more controllable.
130
system. The pressure in the steam chest in the reboiler directly determines the
heat transfer rate because it controls the steam condensing temperature and
therefore the difference in temperature across the heat transfer area.
Like temperature, pressure presents the difficulty of reset windup, discussed
in the next section. Another difficulty with pressure as a secondary variable is
that it can move out of the transmitter range and thus get out of control. For
example, in the scheme of Figure 7-4, if at a low production rate the reboiler
temperature drops below 100C (212F), the pressure in the steam chest will
drop below atmospheric pressure, getting out of the transmitter range, unless
the pressure transmitter is calibrated to read negative pressures (vacuum).
Reboiler
131
at a higher frequency than the outer loop, so that the secondary controller has
time to respond to a set point change from the primary controller before the
next set point change takes place. Recall that the inner loop should respond
faster than the outer loop. If the sampling frequency is low and the same for
both the primary and secondary loops, the secondary loop must be processed
after the primary loop; otherwise, the change in set point will be delayed by
one sample before the secondary loop can take action.
One important consideration when digital feedback algorithms are cascaded
is bumpless transfer from Manual to Automatic. This is done by initializing
the output of the primary controller to the process variable of the secondary
controller when the loops are switched to automatic control, making for a
smooth transition to Automatic.
= 7.5 min
t0 = 1.5 min.
= 1.5 min
t0 = 0 min.
132
Although an increase in coolant flow results in a decrease in both the reactor and jacket temperatures, the signs on the process gains are positive
because, for safety, the coolant valve fails to open. As a result, an increase in
controller output results in a decrease in coolant flow and consequently an
increase in the temperatures, hence the positive gains.
Use the Ziegler-Nichols QDR tuning formulas of Table 4-1 to tune the single reactor temperature PID controller:
Kc = 1.2(7.5/1.5)/2.2 = 2.8
TI = 2.0(1.5) = 3.0 min
The parameters from the response of the jacket temperature are used to
tune jacket temperature controller TC 2 in the cascade scheme. Since the
dead time is zero, a PI controller is indicated, the IMC rule of Section 4-1 is
used for the integral time and the gain can be as high as desired. To keep
the overshoot to set point changes reasonable:
Kc = 1.5
TI = = 1.5 min
TD = 0.
Once jacket temperature controller TC 2 is tuned, it is switched to Automatic and a step test in its set point is applied with the reactor temperature
TC 1 set in Manual. The response of the reactor temperature is recorded
with the following results:
K = 1.1
= 4.5 min
t0 = 0.75 min.
Comparison with the results of the response to the step in coolant flow
shows that the reactor temperature loop has both a shorter time constant
and a shorter dead time when the jacket temperature controller is used.
Recall, however, that these parameters depend on the tuning of the jacket
temperature controller. For example, if a higher gain were used for TC 2, the
time parameters would be shorter still.
133
Figure 7-5 compares the responses of the single reactor temperature controller
and the cascade control scheme to a 5C step increase in coolant inlet temperature followed by a step decrease of 20% in the flow of the reactants. Since the
response of the outlet coolant temperature to the change in inlet coolant temperature is immediately detected and corrected for by the secondary controller, the reactor temperature in the cascade scheme hardly deviates from its set
point. The cascade scheme immediately increases the coolant flow to compensate for the increase in inlet coolant temperature.
The figure shows that the cascade control scheme also improves the response
of the reactor temperature to a step decrease in reactant feed to the reactor.
However, the improvement in performance is not as dramatic because the
feed flow has a direct effect on the reactor temperature and cannot be corrected in time by the jacket temperature controller. The improvement in control is due to the faster response of the reactor temperature to controller
output in the cascade scheme. Another reason that the performance improvement is not as dramatic for the feed flow disturbance is the inverse response of
the temperature to the feed flow. This is because the reactants are colder than
the reactor and the decrease in the flow of reactants causes an immediate rise
in temperature, but the decrease in flow of reactants also causes a decrease in
reactant concentration that eventually results in a decrease in reaction rate
and consequently in temperature.
The following is an example of a successful industrial application of cascade
control. It is an example of composition-to-composition cascade, which is not
very common. It also shows a three-level cascade control system, with the
flow controller being the lowest level.
134
135
Figure 7-6. Cascade Control of Reactor Inlet Composition and Synthesis Loop
Pressure in the Ammonia Process
SP
SP
AC
1
RC
2
Vent
SP
Air
Compressor
FC
2
FT
AT
1
SP
AC
2
Synthesis
Gas
Compressor
Air
AT
2
Ammonia
Synthesis
Reactor
SP
FC
1
FT
Reforming
Process
Natural
Gas
SP
PC
4
PT
SP
RC
3
FC
3
FT
Steam
SP
Ammonia
Product
SP
CO2
FC
4
FT
Purge
The objective is to control the hydrogen to nitrogen ratio (H/N) of the mixture
entering the synthesis reactor at its optimum value (about 2.85 for a slight
excess of nitrogen). The primary controller (AC 1) receives the measurement
of the composition at the reactor inlet from a very accurate analyzer (AT 1).
The output of the primary controller adjusts the set point on the secondary
controller (AC 2). The secondary controller receives the measurement of the
composition of the fresh feed from a fast and inexpensive analyzer (AT 2),
usually a simple thermal conductivity detector, and its output adjusts the
ratio of air to natural gas through ratio controller (RC 2). The ratio controller,
in turn, adjusts the set point of the process air flow controller (FC 2). To pre-
136
vent from throttling the air compressor, flow controller (FC 2) adjusts a valve
on a vent on the compressor discharge.
This example illustrates the point made earlier about the secondary measurement not having to be accurate but having to be fast and consistent. Inaccuracy in the secondary measurement is corrected by the integral mode of the
primary controller. On the other hand, the measurement of the primary controller can be slow, but it must be accurate. Disturbances in the reforming process are handled quickly by the secondary controller, before they have a
chance to affect the primary controlled variable.
Figure 7-6 also shows a pressure-to-flow cascade loop for the control of the
pressure in the synthesis loop. In this cascade the primary controller is the
pressure controller (PC) and the secondary controller is the purge flow controller (FC 4). The purge is a small stream removed from the loop to avoid the
accumulation of inert gases (e.g., argon and methane) and the excess nitrogen.
Although both cascade control loops of Figure 7-6 could be carried out with
analog controllers, computer control offers an unexpected virtue to this
scheme: patience. For example, in one installation where the pressure control
scheme was carried out with analog controllers, the primary controller was
operated on Manual because it was swinging the purge flow all over its range.
This was because the time constant of this loop is about one hour. A digital
controller with a sampling interval of 5 minutes and an integral time of 45
minutes was able to maintain the pressure at its optimum set point on the
same installation.
137
cooling water valve closed and the steam valve manually opened to bring the
reactor up to the operating temperature, 104C (see Figure 7-7). The jacket
temperature transmitter, TT 2, has a range of 40 to 115C, and the steam condenses at 110C, which is the value of the jacket temperature when the steam
valve is closed and the cascade control system is initialized and switched to
Automatic. This is done before the reactor temperature reaches its 104C set
point, say when it reaches 100C.
Following the bumpless transfer procedure of the control system, the output
of the primary controller is initialized to the measured temperature of the secondary controller, 110C. At this time the jacket temperature begins to drop
because the steam has been turned off and the reactor is at the lower temperature of 100C, while the reactor temperature is rising because of the heat of the
reaction. For the time that the reactor temperature is between 100 and 104C
(its set point), the control situation is as follows:
The secondary controller sees a jacket temperature below its set point
(110C) and calls for the cooling water valve to remain closed.
The primary controller also sees its temperature below set point and
calls for an increase in the jacket temperature set point above the current 110C value.
Most computer and DCS controllers detect that the secondary controller output is limited or clamped at the closed position and prevent the primary
controller from increasing its output since this would only call for the closing
of the coolant valve, which is already closed. Does this logic prevent the cascade control system from winding up? Let us see what happens next.
Notice that a gap has been created between the set point of the secondary controller, clamped at 110C, and its measured temperature, the jacket temperature that drops to the reactor temperature as soon as the steam is turned off.
As the reactor temperature crosses its set point of 104C, the primary controller starts decreasing the set point of the secondary controller to bring the temperature down, but the coolant valve will not open until the set point of the
secondary controller drops below its measured temperature; that is, until the
gap mentioned earlier is overcome. Since the set point of the secondary controller will change at a rate controlled by the integral time of the primary controller, it takes a long time for the coolant valve to start to open and the reactor
temperature overshoots its set point badly, the common symptom of reset
windup. The situation continues as the coolant valve is driven from closed to
open and back again, as the oscillations of the dashed lines in Figure 7-7 show.
As you can see, the saturation or clamp limit detection system could not
avoid reset windup in this case.
Figure 7-7. Oscillatory Behavior Caused by Reset Windup in the Cascade Control
of an Exothermic Reactor Startup (Dashed Lines) and Solution Using Reset
Feedback on the Master Controller (Continuous Lines)
Reset Feedback
An elegant and effective way to protect against cascade reset windup is the
use of the reset feedback feature on the primary controller. In the cascade
139
FBk
140
7-5. Summary
This chapter presented the cascade control scheme; that is, the cascading of a
primary controller to a secondary controller to improve control performance.
The discussion included the reasons for using cascade control, the selection of
modes for the secondary controller, and the procedure for tuning cascade control systems. It also looked at cascade reset windup and ways to protect
against it.
Cascade control has proliferated in computer control installations because
there is essentially no cost for the additional secondary controllers. The only
additional cost in a computer control system is the cost of one transmitter and
one multiplexer input channel for each secondary loop.
Review Questions
7-1. What are the three major advantages of cascade control?
7-2. What is the main requirement for a cascade control system to result in
improved control performance? What is required of the sensor for the
secondary loop?
7-3. Are the tuning and selection of modes different for the primary controller in a cascade control system than for the controller in a simple feedback control loop? Explain.
7-4. What is different about the secondary controller in a cascade control system? When should it not have integral mode? If the secondary is to have
derivative mode, should it operate on the process variable or on the error
(deviation)?
7-5. In what order must the controllers in a cascade control system be tuned?
Why?
7-6. What are the two major difficulties with using temperature as the process variable of the secondary controller in a cascade control system?
How can they be handled?
7-7. Why is pressure a good variable to use as the secondary variable in cascade control? What are the two major difficulties with using pressure as
the secondary variable?
141
7-8. What is the relationship between the processing frequencies of the primary and secondary controllers in a computer cascade control system?
7-9. How can reset windup occur in a cascade control system? How can it be
avoided?
8
Feedforward and Ratio
Control
This chapter presents the design and tuning methods of feedforward and ratio
control strategies. Along with cascade control, these strategies can be classified as multiple input, single output (MISO) because they require more than
one process measurement but only one final control element (usually a control
valve) because there is only one control objective.
143
144
145
U
G2
SP
Feedback
Controller
OP
+
G1
Sensor
- PV
146
U
G2/G1
G2
-
SP
1/G1
OP
PV
G1
Feedforward-Feedback Control
It is seldom practical to measure all the disturbances that affect the process
variable. A more reasonable approach is to measure only those disturbances
that are expected to cause the greatest deviations in the process variable and
handle the so-called minor disturbances by adding feedback trim to the
feedforward controller. Figure 8-3 shows a block diagram for a feedforwardfeedback control system. Note that the feedback controller takes the place of
147
the set point element of Figure 8-2, and only the feedforward element is necessary in the combined control scheme. A feedforward element is required for
each disturbance measured.
U
G2/G1
G2
Feedback
Controller
SP
OP
PV
G1
+
-
Sensor
When the outputs of the feedforward and feedback controllers are summed,
as in Figure 8-3, the presence of the feedforward controller does not affect the
response of the loop to inputs other than the measured disturbance, thus the
feedback controller tuning does not have to be adjusted because of the installation of the feedforward controller.
Economics dictates that only those disturbances that are frequent enough and
important enoughin regards to their effect on product quality or safety, or
for similar considerationsshould be measured and compensated for with a
feedforward controller.
The advantages of the feedforward-feedback scheme are:
The feedback controller takes care of those disturbances that are not
important enough to be measured and compensated for.
148
The feedforward controller does not have to compensate exactly for the
measured disturbances since any minor errors in the model are
trimmed off by the feedback loop, hence the term feedback trim.
Because of these advantages, feedback trim is a part of almost every feedforward control scheme.
Ratio Control
The simplest form of feedforward control is ratio control. It simply consists of
the establishment of a ratio between two flows. Figure 8-4 shows an example
of ratio control between the steam and process flows of a steam heater. In this
example, the process flow is the disturbance or wild flow, and the steam is
the manipulated flow. The steam flow controller takes care of both variations
in the pressure drop across the control valve and its nonlinearity. By maintaining a constant ratio when the process flow is changed by the operator or
by another controller, the outlet process temperature is kept constant as long
as the steam latent heat and process inlet temperature remain constant; in
other words, the ratio controller compensates only for variations in process
flow to the heater. The temperature feedback controller in the figure, TC,
adjusts the ratio to provide the feedback trim in this example to compensate
for variations in process inlet temperature and steam heat of condensation.
The diagram of the control of the ammonia process in Figure 7-6 shows two
ratio controllers to compensate for variations in the natural gas flow to the
process.
Some control engineers prefer to calculate the ratio by dividing the manipulated flow by the wild flow and then controlling the ratio with a feedback controller, as in RC in Figure 8-5. This alternative has the disadvantage of creating
a very nonlinear feedback control loop; note that the gain of the feedback loop
in Figure 8-5 is inversely proportional to the wild flow, which is the major disturbance. The ratio controllers in some computer and distributed control systems display the calculated ratio, but do not use it for control. Instead, the
output is calculated by multiplying the input or wild flow by the ratio set
point, as in Figure 8-4.
149
RC
SP
Steam
FT
SP
FC
TC
OP
Fs
PV
FT
TT
F
Process
fluid
Steam
trap
Condensate
150
Wild flow
FT
(B/A)set
B/A
B
Manipulated
flow
FY
%
SP
RC
FT
(8-1)
This is the design equation for the feedforward controller having set point SP
and disturbance U as inputs and output variable OP as output. Equation 8-1
provides the design formulas for both the set point and feedforward elements
of Figure 8-2:
Set point element:
1
G s = ------G1
Feedforward element:
G
G F = ------2G1
151
When feedback trim is used, as in Figure 8-3, only the feedforward element is
needed, since the feedback controller takes the place of the set point element.
(8-2)
with:
K
Gain = ------2K1
Lead of
Lead Lag = --------------------------1Lag of 2
Dead Time Compensator = t02 - t01
(8-3)
where:
K1
K2
152
The dead time compensator of Equation 8-3 can only be realized when the
dead time between the disturbance and the process variable is longer than the
dead time between the manipulated variable and the process variable. Otherwise, it would call for the feedforward correction to start before the disturbance takes place, which is obviously not possible.
The dead time compensator requires the memory of digital devices (computers and microprocessors) for its implementation. The dead time compensator
can often be left out because the lead-lag unit can be tuned to provide all of
the required dynamic compensation, thus simplifying the tuning task. In general, the dead time compensator should only be used when the lead-lag unit
cannot do the job by itself.
Gain Adjustment
The adjustment of the feedforward gain can be carried out with the feedback
controller in Manual or Automatic. If it is done with the feedback controller in
Manual, when the gain is not correct, the process variable will deviate from its
set point after a sustained disturbance input. The gain can then be adjusted
until the process variable is at the set point again. Because of process nonlinearities, the required feedforward gain may change with operating conditions,
thus exact compensation may not be possible with a simple linear controller.
If the feedforward gain is adjusted with the feedback controller in Automatic,
the variable to observe is the output of the feedback controller. If the feedback
controller has integral mode, the process variable will always return to its set
point after a disturbance, but if the feedforward gain is incorrect, the output of
the feedback controller will be changed to compensate for the error in the
feedforward controller. The feedforward gain must then be adjusted until the
feedback controller output returns to its initial value. As before, process nonlinearities will prevent a single value of the gain from working for all process
conditions.
153
The one thing to remember when tuning the feedforward gain is that it is necessary to wait until the system reaches steady state before making the next
adjustment.
154
Output
Lead = 2xLag
2.5
Input
Output
Lead = 0.5xLag
1.5
0.5
0
0
10
15
20
25
30
35
40
45
2. Select the ratio of the lead to the lag based on how much to amplify or
attenuate sudden changes in the disturbance inputs. For example, suppose
it is desired to lead the disturbance by one minute; a lead of 1.1 minutes
and a lag of 0.1 minutes give an amplification factor of 1.1/0.1 = 11, while
a lead of 3 minutes and a lag of 2 minutes give an amplification factor of
only 3/2 = 1.5. If the disturbance is noisy; for example, in the case of a
flow, the second choice is preferred since it results in less amplification of
the noise.
Although it is possible to have a lag with zero lead, it is not possible to have a
lead without a lag. The ratio of the lead to the lag should not be greater than
10. When a net lag is required, the lead can usually be set to zero, simplifying
the tuning task.
155
100
Net lead
Output
Lead > Lag
80
Net lag
60
Input
40
Output
Lead < Lag
20
0
0
10
15
20
25
30
35
40
45
LG
where:
Xk
Yk
(8-4)
156
157
where N is the number of samples of dead time and unity gain is assumed.
Figure 8-8 shows a plot of the responses of dead time compensation to an
input signal. Notice that the dead time compensator does not start responding
until one dead time after the change in the input; the output then reproduces
the input exactly.
The dead time compensator is easy to tune, since it only has one dynamic
parameter, the number of samples of dead time N.
Before applying dead time compensation, it is important to ensure that the
dead time does not delay the action in a feedback control loop. Recall that
dead time always makes a feedback control loop less controllable. The reason
it can be used in feedforward control is that the corrective action always goes
forward; that is, no loop is involved.
Input
0
0
10
15
20
25
30
35
40
45
158
159
as simple as possible. Solve for the output variable so that it can be calculated from the measured disturbances and the set point of the process
variable. The resulting formula or formulas constitute the design equation(s) to be programmed into the computer for on-line execution. Caution: the formula must use the set point of the process variable and not its
measured value.
5. Re-evaluate the list of measured disturbances. The effect of the expected
variation of each disturbance on the process variable can be calculated
from the basic design formulas; if the effect of a disturbance is small the
disturbance need not be measured. On the other hand, there may be a disturbance that was not on the original list which may be found from the
formulas to have a significant effect on the process variable. The decision
to measure or not to measure must weigh the effect of the disturbance, its
expected magnitude, speed and frequency of variation, and the cost of
measuring it. Unmeasured disturbances are treated as constants in the
design equation, equal to their design or average expected values. Alternatively, if they are difficult to measure but are still expected to vary, they
may be adjusted by feedback trim.
6. Introduce the feedback trim, if any, into the design equation. This is done
by grouping unknown terms and unmeasured disturbances as much as
possible and letting the output of the feedback controller adjust the group
of terms that is expected to vary the most. A simple and effective approach
is to have the output of the feedback controller adjust the set point of the
feedforward controller.
7. Decide whether dynamic compensation is needed and how it is to be
introduced into the design. Simple lead-lag or dead time compensators are
commonly used. A separate dynamic compensator should be installed on
each measured disturbance. It is not good practice to install the dynamic
compensator in such a way that it becomes part of the feedback trim loop,
especially if it contains dead-time compensation.
8. Draw the instrumentation diagram for the feedforward controller. This is
a diagram showing the various computations and relationships between
the signals. It is good practice to draw it so that all the input signals enter
from the top (or left) and the output signals exit at the bottom (or right). It
is at this point that implementation details, largely dependent on the
equipment used, must be decided upon. A good design should be able to
160
(8-5)
2. Measured disturbances:
W, the flow through the exchanger, kg/h
Ti, the inlet temperature, C
3. Manipulated variable:
F, steam flow controller set point, kg/h
4. A steady-state energy balance on the exchanger yields the equation for
the static feedforward controller:
FHv = WCp(To - Ti) + QL
where:
Cp
Hv
QL
(8-6)
161
SP
Feedforward
Controller
Fset
Toset
SP
Steam
FT
SP
FC
TC
Ti
TT
OP
PV
FT
TT
Process
fluid
To
Steam
trap
Condensate
5. At this point it is possible to evaluate the quantitative effect of the possible disturbances on the outlet temperature; such analysis may determine that the heat loss rate is as important as the two measured
disturbances but is difficult to measure and is thus a candidate for feedback trim adjustment. Conversely, the inlet temperature may not have
enough effect to merit the cost of measuring it, in which case the feedforward controller becomes a simple steam-to-process-flow ratio controller.
6. The need for feedback trim is determined by considering how much the
unknown terms in the design formula are expected to vary. The three
unknown terms are the physical properties, Cp and Hv, and the heat loss
rate, QL. The three can be lumped together by assuming that the heat
loss rate is proportional to the heat transfer rate:
QL = (1 - )FHv
(8-7)
162
set
Cp
set
= ----------( T o T i )W
Hv
(8-8)
Notice that the outlet temperature in the formula has been replaced by
its set point; that is, the control objective, Equation 8-5, has been
substituted into the design formula to ensure that it is enforced by the
feedforward controller. In modern computer control systems it is
possible to retrieve the set point from the feedback controller to use in
the feedforward calculation, so that only one set point has to be entered
by the operator. This is an important design requirement.
All the unknowns of the model have been lumped into a single
coefficient, Cp/Hv, and it would seem natural for the feedback trim
controller to adjust this coefficient to correct for variations in the
specific heat Cp, the steam latent heat of condensation Hv, and the
heater efficiency . However, these parameters are not expected to vary
much, thus it would be undesirable for the feedback trim controller to
control by adjusting a term that is not expected to vary. A better control
system structure results if the feedback controller output is made to
adjust the set point of the feedforward controller or equivalently, the
product of the unknown coefficient and the set point, as follows:
set
Cp
= OP -----------T
W
H v i
where:
OP = CpToset/Hv = output of feedback controller
The coefficient Cp/Hv becomes the tunable gain of the inlet
temperature correction. This term can be calculated from measured
values of the temperatures and flows, averaged over long enough
periods of time. From Equation 8-8:
163
Cp
F
= ---------------------------------------Hv
W ( To Ti )
7. The feedforward formula is derived from an energy balance on the
heater at steady state. Dynamic compensation will probably be required
because changes in steam flow, which is the output variable, are
delayed by the lags of the control valve and steam chest, while the process flow will have a faster effect on the outlet temperature. On the
other hand, the effect of changes in inlet temperature will be delayed by
the transportation lag in the heater. To compensate for these dynamic
imbalances, lead-lag units can be applied to the two measured disturbances before they are used in the computation.
8. Figure 8-10 shows the instrumentation diagram for the feedforward controller. In some computer control systems, the multiplier may be carried
out as a ratio controller, with the ratio being set by the adder, which
combines the feedback controller output and the inlet temperature correction.
164
TT
Toset
TT
Ti
To
Lead-Lag
1
FT
SP
TC
Lead-Lag
2
+
-
Adder
Toset - Ti
Multiplier
Fset
lead-lag unit reduces the deviation of the outlet temperature to about onethird the deviation of the simple feedback controller and about two-thirds
that of the static feedforward controller.
In terms of the response of the outlet temperature to a 10C increase in inlet
temperature, the action of the static feedforward controller is hindered by
the lag of the inlet temperature sensor, causing the correction to be too slow
to prevent the outlet temperature from deviating as much as if feedforward
compensation were not used. Once again this can be corrected by inserting
a net lead to dynamically correct for the disturbance. Curve (c) of Figure 811 shows the response with a lead of 2.1 min and a lag of 0.4 min for a net
lead of 1.7 min, considerably reducing the initial deviation in outlet temperature through both simple feedback control and static compensation.
165
(a)
(b)
(c)
Decrease in
process flow
Increase in process
inlet temperature
To better demonstrate the performance of feedforward control, the feedforward responses of Figure 8-11 do not include feedback trim. Although in this
case the feedforward model is accurate, feedback trim is almost always
needed in practice to correct for inaccuracies in the feedforward model.
The preceding example has a characteristic typical of many successful feedforward control applications: the formulas used in the compensation are simple
steady-state relationships. If dynamic compensation is needed, lead-lag and
dead time compensation may be added to the nonlinear steady-state compensator. The moral is: keep your design super simple.
8-5. Summary
In summary, ratio and feedforward control complement feedback control,
reducing the magnitude of the deviations of the process variable caused by
disturbances. The feedforward controller is free of stability concerns but its
166
application requires a model of the process. The best approach is a combination of feedforward and feedback control. Ratio control is the simplest form of
feedforward control; it establishes a simple proportionality between two
flows.
References
1. Luyben, W. L. Process Modeling, Simulation, and Control for Chemical Engineers, 2nd ed. New York: McGraw-Hill, 1990, Sections 8.7 and 11.2.
2. Shinskey, F. G. Feedforward Control Applied, ISA Journal, Nov. 1963, p.
61.
3. Smith, C. A. and Corripio, A. B. Principles and Practice of Automatic Process
Control, 3rd ed. New York: Wiley, 2006, Chapter 11.
Review Questions
8-1. Why isnt it possible to have perfect control; that is, the process variable
always being equal to the set point, with feedback control alone? Is perfect control possible with feedforward control?
8-2. What are the main requirements of feedforward control? What are the
advantages of feedforward control with feedback trim over pure feedforward control?
8-3. What is ratio control? What is the control objective of the air-to-natural
gas ratio controller in the control system sketched in Figure 7-6 for the
ammonia process? Which are the measured disturbance and the manipulated variable for that ratio controller?
8-4. What is a lead-lag compensator? How is it used in a feedforward control
scheme? Describe the step and ramp responses of a lead-lag unit.
8-5. It is desired to lead a disturbance in a feedforward controller by 1.5 minutes. If the amplification factor for the noise in the disturbance measurement must not exceed 2, what must the lead and lag be?
8-6. What is dead time compensation in a feedforward controller? When can
it be used? When should it be used?
167
Flue gas
W
Ti
FT TT
Process
Stream
Fset
SP
F
FC
FT
Main Fuel
To
TT
Fs
FT
Air
Auxiliary
Fuel
9
Multivariable Control
Systems
F.
169
170
SP
AC
OP1
SP
FC
x
F1x1
F
OP2
AT FT
F2x2
171
SP2 PV2, to produce the two controller outputs. Signals SP1 and SP2 represent the set points of the loops. In the diagram of Figure 9-2 each of the four
process blocks, G11, G12, G21 and G22, includes the gains and dynamics of the
final control elements (valves), the process and the sensor/transmitters. For
simplicity the disturbances are not shown.
SP1
+
Controller
1
OP1
PV1
G11
+
G12
G21
SP2
+
Controller
2
OP2
+
G22
PV2
To look at the effect of interaction, assume that the gains of all four process
blocks are positive; that is, an increase in each controller output results in an
increase in each of the process variables. Let us follow the sequence of events
plotted in Figure 9-3:
1. Suppose that at a point in time a step change in controller output OP1
takes place with both loops in Manual (opened). Figure 9-3 shows the
responses of both process variables, PV1 and PV2, where the time of the
step change is marked as point a.
2. Now suppose that at time b control loop 2 is closed (switched to Automatic) and that it has integral or reset mode. Controller output OP2 will
decrease until process variable PV2 comes back down to its original value,
assumed to be its set point.
172
3. The decrease in OP2 also causes, through interaction block G12, a decrease
in process variable PV1, so that the net change in PV1 is smaller than the
initial change. Note that this initial change is the only change that would
take place if there were no interaction, or if controller 2 were kept in
Manual.
The difference between the initial change and the net change in PV1 is the
effect of interaction. It depends on the effect that OP1 has on PV2 (G21), the
effect that OP2 has on PV2 (G22)which determines the necessary corrective
action on OP2and the effect that OP2 has on PV1 (G12). Note also that the
steady-state effect of interaction depends only on the process gains, not on the
controller tuning, provided that controller 2 has integral mode.
Figure 9-3. Effect of Interaction for a 2x2 Control System with all Four Gains
Positive
OP1
b
SP1
PV1
a
OP2
PV2
SP2
a
Time
The authors invite you to verify that a step in OP2, followed by closing control
loop 1, has the same effect on PV2at least qualitativelyas the effect just
observed on PV1. It will be shown shortly that the relative effect of interaction
is quantitatively the same for control loop 2 as it is for control loop 1.
173
In the case just analyzed, all four process gains were assumed positive (direct
action). The effect of interaction was in the opposite direction as the direct (initial) effect of the step change, resulting in a net change that was smaller than
the initial change. This type of situation, in which the two loops fight each
other, is known as negative interaction. Note that it is possible for the
effect of interaction to be greater than the initial effect, in which case the net
change will be in the opposite direction as the initial change. Here we could
say that the wrong loop wins the fight, a situation that results from incorrect
pairing of the loops, as shall be shown shortly. You can easily verify that if any
two of the process gains were positive, and the other two were negative, the
interaction would also be negative.
If one of the four process gains has a sign opposite to that of the other three,
the effect of interaction would be in the same direction as the direct action and
the net change would be larger than the initial change, as you can also verify.
This is the case of positive interaction, in which the two loops help each
other.
Positive interaction is usually easier to handle than negative interaction,
because the possibility of inverse response (i.e., the process variable moving in
the wrong direction right after a change) or of open-loop overshoot exists only
in the case of negative interaction.
It is evident that both positive and negative interaction can be detrimental to
the performance of the control system. This is because the response of each
loop is affected when the other loop is switched into and out of Automatic, or
when its output saturates. In summary, the following are characteristics of
loop interaction:
For interaction to affect the performance of the control system, it must
work both ways; that is, each controller output must affect both process
variables through the process. Notice that if either G12 or G21 is absent
from the diagram of Figure 9-2, there is no interaction effect.
Because of interaction, a set point change to either loop produces at
least a transitory change in both process variables.
The interaction effect on one loop can be eliminated by interrupting the
other loop. That is, if one of the two controllers is switched to Manual
the remaining loop is no longer affected by interaction.
174
The following sections look at two ways to approach the problem of loop
interaction:
Pair the process variables and controller outputs so as to minimize the
effect of interaction between the loops.
Combine the controller output signals through decouplers to eliminate
the interaction between the loops.
More advanced multivariable control design techniques will be discussed in a
later section.
Open-loop Gains
Consider the 2x2 system of Figure 9-2. If a change is applied to controller
output OP1, while keeping the other controller output constant, and the
175
changes in process variables PV1 and PV2 are measured, the open-loop gains
can be calculated:
Change in PV
K 11 = ---------------------------------------1Change in OP 1
Change in PV
K 21 = ---------------------------------------2Change in OP 1
(9-1)
Similarly, when a change is applied to OP2, keeping OP1 constant, the other
two open-loop gains can be calculated:
Change in PV
K 12 = ---------------------------------------1Change in OP 2
Change in PV
K 22 = ---------------------------------------2Change in OP 2
(9-2)
The open-loop gains can be determined from the steady-state equations or the
computer simulation programs used to design the plant.
There is a natural tendency to try to use the open-loop gains in the pairing of
the variables. However, it is immediately apparent that PV1 and PV2, and OP1
and OP2 do not necessarily have the same dimensions. Thus, attempting to
compare open-loop gains would be like trying to decide between buying a
new sofa or a new house. To overcome this problem, Bristol1 proposes to compute relative gains that are independent of dimensions.
Closed-loop Gains
Because of interaction, the effect of OP1 on PV1 is different when the other
loop is closed than when it is opened, as discussed in the previous section.
This requires the definition of the closed-loop gains K11', K21', K12' and K22'.
These are defined exactly in Equations 9-1 and 9-2, but with the changes in
PV1 determined with PV2 kept constant, and the changes in PV2 determined
with PV1 kept constant. For example, to determine K11', a change is made in
OP1 and the change in PV1 is measured while a feedback controller with integral mode controls PV2 by adjusting OP2.
176
However, closed-loop tests are not needed because the closed-loop gains can
be computed from the open-loop gains previously defined. For example,
when both OP1 and OP2 change, the total change in PV1 can be estimated by
the sum of the two changes:
Change in PV1 = K11(change in OP1) + K12(change in OP2)
and similarly for the total change in PV2. Now, if PV2 is kept constant, its
change is zero:
Change in PV2 = K21(change in OP1) + K22(change in OP2) = 0.
Solving for the change in OP2 required for PV2 to remain constant:
K 21
Change in OP 2 = -------- ( Change in OP 1 )
K 22
Substitute to obtain the total change in PV1:
K 12 K 21
The bracketed expression is then the closed-loop gain K11'. The closed-loop
gains for each of the other three pairings can be similarly derived.
177
The formulas of Equations 9-3 can be used to compute the relative gains for
any 2x2 system:
K 11 K 22
11 = 22 = ------------------------------------------K 11 K 22 K 12 K 21
K 12 K 21
12 = 21 = ------------------------------------------K 12 K 21 K 11 K 22
(9-3)
It makes sense that the interaction measure for the PV1-OP1 pair is the same as
for the PV2-OP2 pair since they represent one option in the 2x2 system, the
other option being PV1-OP2 and PV2-OP1.
The relative gains are dimensionless and can therefore be compared to one
another. To minimize the effect of interaction, the process variables and controller outputs are paired so that the relative gain for the pair is closest to
unity. This results in the least change in gain when the other loop of the pair is
closed. Note that for the case of no interaction, the open-loop gain is equal to
the closed-loop gain, and the relative gains are 1.0 for one pairing and 0.0 for
the other.
178
179
180
Unlike the specific numerical solution that was developed in Example 9-1, a
general solution for the blender will be developed here. To do this, the conservation of mass and the conservation of solute are used to develop formulas for the open-loop gains.
Conservation of mass: F = F1 + F2
F1 x1 + F2 x2
Conservation of solute: x = ----------------------------F1 + F2
Using differential calculus, the steady-state gains are:
KF1 = Kv1
F2 ( x1 x2 )
K x1 = ---------------------------K
2 v1
( F1 + F2 )
KF2 = Kv2
F1 ( x2 x1 )
K x2 = ---------------------------K
2 v2
( F1 + F2 )
where Kv1 and Kv2 are the valve gains in (kg/h)/fraction valve position.
Next substitute the open-loop gains into the formulas for the relative gains,
Equations 9-3. A little algebraic manipulation produces the following general expressions for the relative gains:
F1
F1 = x2 = ----------------F1 + F2
F2
F2 = x1 = ----------------F1 + F2
In words, the pairing that minimizes interaction has the flow controller
adjusting the larger of the two flows and the composition controller adjusting the smaller of the two flows. If a ratio controller were to be used, the
smaller flow would be ratioed to the larger flow, with the flow controller
adjusting the larger flow and the composition controller adjusting the ratio.
It could easily be shown that the ratio controller decouples the two loops so
that a change in the product stream flow does not affect the composition.
Note that the valve gains Kv1 and Kv2 do not affect the relative gains. This is
why they were not considered in Example 9-1.
181
For most processes, the relative gains tell all that needs to be known about
interaction. They are determined from the open-loop steady-state gains,
which are easy to determine by either on-line or off-line methods. However,
in systems with negative interaction, the pairing recommended by relative
gain analysis may not result in the best control performance because it does
not consider the dynamic response. This is illustrated in the following
example.
182
Condenser
SP
PC
SP
PT
LC
1
LT
TT
TC
1
SP
Feed
Reflux
Distillate
SP
TT
SP
LC
2
TC
2
LT
Steam
Reboiler
Bottoms
183
Condenser
SP
PC
SP
PT
LC
1
LT
TT
TC
1
SP
Feed
Reflux
l
Distillate
SP
TT
SP
LC
2
TC
2
LT
Steam
Reboiler
Bottoms
TC 1
TC 2
Reflux
-2.85
-0.438
Steam
1.16
2.53
TC 1
TC 2
Reflux
1.08
-0.08
Steam
-0.08
1.08
The recommended pairing is the same as the intuitive one: control the overhead composition with the reflux flow and the bottoms composition with the
184
bottoms flow. The interaction is negative, with the gain of each loop decreasing by 8% when the other loop is closed.
Condenser
SP
PC
SP
PT
LC
1
LT
TT
TC
1
SP
Feed
TT
SP
LC
2
Reflux
TC
2
l
Distillate
SP
LT
Steam
Reboiler
Bottoms
The sensitivity study on the simulated column gives the following open-loop
gains:
TC 1
TC 2
Reflux
-0.35
0.07
Bottoms
-1.05
-1.93
185
TC 1
TC 2
Reflux
0.90
0.10
Bottoms
0.1
0.90
The pairing for this scheme is also the obvious one: top temperature with
reflux and bottom temperature with bottoms product flow, and the relative
gains show about 10% positive interaction; that is, the two loops help each
other, indicated by the relative gains being positive and less than unity.
It would appear, then, from steady-state relative gain analysis that the Direct
Material Balance Control results in positive interaction, which is preferred to
the negative interaction resulting from Energy Balance Control. Unfortunately, the Energy Balance Control scheme was found by simulation to perform better in this particular example than the Direct Material Balance Control
scheme. The reason is dynamic interaction, which goes undetected by the relative gain matrix. For the first scheme the open-loop responses are monotonic;
that is, the temperature stays between its initial value and its final value during the entire response. On the other hand, for the second scheme the openloop responses exhibit inverse response; that is, the temperature moves in one
direction at the beginning of the response and then moves back to a final
value on the opposite side of its initial value. This causes the feedback controller to initially take action in the wrong direction, degrading the performance
of the control system.
Although in this particular example relative gain analysis fails to properly
predict which of the two control schemes performs better, it is still useful in
verifying that the intuitive pairing is the correct one for each scheme. It would
also have evaluated the interaction for each scheme correctly had all the
responses been monotonic. This example also shows that the arrangement of
the level controllers affects the interaction between the other loops in the
column.
186
SP1
+
Controller OP1
1
G12
G21
D2
Controller
2
+
D1
SP2
PV1
G11
OP2
+
G22
PV2
Each of the two decoupler terms, D1 and D2, can be considered to be feedforward controllers for which the disturbances are the controller output signals OP1 and OP2. The design of the decouplers is therefore identical to the
design of a feedforward controller presented in Chapter 8.
187
(9-4)
(9-5)
Decoupling, like feedforward, can be designed to varied degrees of complexity. The simplest is given by linear static compensation (i.e., forfeiting the
dynamic compensation), which can be accomplished in practice by a simple
summer (adder) with adjustable gains. The next degree of complexity is to
add dynamic compensation in the form of lead-lag units (see Chapter 8). Ultimately, nonlinear models of the process could be used to design nonlinear
decouplers, following the procedure outlined in Chapter 8. Equations 9-4 and
9-5 assume linear models.
188
(9-6)
(9-7)
Half Decoupling
As discussed earlier, the interaction effect depends on both controller outputs
affecting both process variables. Thus, interaction can be eliminated by decou-
189
pling one loop and letting the other loop be affected; this can be achieved by
implementing either D1 or D2 but not both. This is referred to as half decoupling. In deciding which decoupler to select, the first consideration may be
which of the process variables is more important to keep at its set point. A secondary consideration may be the ease with which the dynamic terms of the
decouplers can be implemented.
F 2 K v1
- ( OP 1 OP 10 )
M 2 = OP 2 --------------F K
1
v2
190
stream compositions and the product composition set point. If any of these
compositions were to vary, the gain of the decoupler would have to be readjusted. There is, however, another way to design the decoupler which does
not require readjustment of the parameters when process conditions
change. It consists of using simple process models to set up the structure of
the control system, as discussed next.
(9-8)
This formula requires the measurement of the smaller flow and flow control
of the larger flow.
The conservation of solute mass shows that the product composition depends
on the ratio of the flows rather than on any one of the inlet flows. It is then
assumed that the output of the composition controller is the ratio of the
smaller flow to the larger flow, and the smaller flow is calculated as follows:
F2set = OP2F1
(9-9)
This formula requires that the smaller flow also be controlled. Figure 9-8
shows the diagram of the resulting control system. In this scheme the ratio
controller keeps the product composition from changing when the total flow
is changed, and the adder keeps the total flow from changing when the composition controller takes action. The multivariable control system is therefore
fully decoupled.
The last two design formulas, Equations (9-8) and (9-9), do not show the scale
factors that may be necessary to convert the flow signals into percent of the
191
Adder
SP
OP2 = (F2/F1)set
F1set
FC
SP
AC
SP
FC
F1
SP
RC
FT
F2set
F1x1
SP
FC
F
AT FT
F2
FT
F2x2
scales of the flow controllers. The scale factors depend on the spans of the two
flow transmitters rather than on the sizes of the control valves. The flow controllers allow the signals to be linear with flow, and they also take care of
changes in pressure drop across the control valves.
192
TI = 6 min
TI = 0.1 min
Figure 9-9 shows the responses of the product composition and flow, as
well as the inlet flows, for a step increase of 10 kg/hr in product flow set
point followed by a 2% increase in product composition set point. The continuous curves plot the responses for simple feedback controllers and the
dashed lines are for the decoupled system. Notice that for the change in
product flow the decoupler immediately changes both flows, keeping the
ratio between the two flows constant and the product composition constant,
while the un-decoupled system must correct the dilute flow to bring the
composition back to set point. Similarly, for the change in product composition, the decoupler keeps the total flow constant while the un-decoupled
system requires a small delay in correcting the flows, causing the total flow
to dip temporarily.
The reason that the improvement in control afforded by the decoupler is
not more dramatic is that the blender is a highly controllable system that
allows the composition controller to be very tightly tuned. Nevertheless,
this example shows that the decoupler can fulfill its objective of maintaining each process variable constant when the other one is changed.
193
Figure 9-9. Product Flow and Composition Control of Blender with Decoupler
(Dashed Lines) and with Simple Feedback Controllers (Solid Lines)
194
195
If all the loops are of equal importance and speed of response, each must be
tuned while the other loops are in Manual. The gain of each loop must then be
adjusted by multiplying the gain obtained when all other loops were opened
by the relative gain for the loop:
Kcij' = Kcijij
(9-10)
where:
Kcij' = the adjusted controller gain
Kcij
= the controller gain tuned with all the other loops opened
ij
This adjustment accounts for the change in steady-state gain when the other
loops are closed, but it does not account for dynamic effects. If some of the
loops are slower than the others or can be detuned, the relative gains for the
remaining loops must be recalculated as if those were the only interacting
loops; that is, as if the slower or detuned loops were always opened.
The gain adjustment suggested by Equation 9-10 should be sufficient for those
loops with positive interaction since their response remains monotonic when
the other loops are closed. However, the loops with negative interaction may
require retuning after the other loops are closed because the other loops will
cause either inverse or overshoot responses that normally require lower gains
and slower integral than monotonic loops. Note that the formula results in a
gain reduction for the loops with positive interaction and a gain increase for
the loops with negative interaction, assuming that the pairing with the positive relative gain is always used, as it should be.
When decouplers are used, they must be tuned first and kept active while the
feedback controllers are tuned. Recall that perfect decoupling has the same
effect on a loop as if the other loops were very tightly tuned. For example, in
the blender control system of Figure 9-8, the ratio and mass balance controllers must be tuned first and then kept active while the flow and composition
controllers are tuned.
196
197
Figure 9-10. Control of Product Composition with the Product Flow Controller in
Automatic (Solid Lines) and in Manual (Dashed Lines)
cusses Model Reference Control and presents an example. For a simple introduction to the mathematics of the DMC scheme see Smith and Corripio3.
A model reference controller uses an on-line process model and feedback
from a process measurement to correct for unmeasured disturbances and
model error. The more successful schemes are not restricted to specific model
structures, such as the single-lagplus-dead-time model of Chapter 3. Instead,
the models are developed from process data. For example, DMC models consist of unit step responses of each process variable PV to each controller output OP and measured disturbance D.
One common characteristic of all the successful model reference controllers is
that they use the models to predict the future response of the process or dependent variables. Then, by comparing current process measurements with the values predicted by the model for the current time, the predicted values are
198
corrected. The corrected predicted values from the model are then used to
determine the changes in the manipulated or independent variables that minimize the deviations of the dependent variables from their set points.
Because the different process variables have different units of measure (e.g.,
temperatures, flows, compositions, etc.), their deviations must be weighted in
the function to be minimized. One way this is done is by defining an equalconcern error for each variable. For example, equal-concern errors in a given
application may be 5C, 200 kg/h, 2 weight%, etc. Weighing the deviations by
the reciprocals of the equal-concern errors normalizes them into deviations of
equivalent magnitude.
Another common characteristic of model reference controllers is the presence
of penalties in the function to be minimized for excessive movements in the
controller outputs. In fact, the penalty factors for the controller output moves,
known as move-suppression parameters, are some of the tuning parameters
of the multivariable control system.
Another characteristic is provision for optimization of the set points. In a linear scheme like DMC, a Linear Program is used to do the optimization, which
means that the system is driven to its constraints, since linear systems cannot
have optimums inside the range of operating conditions. Since there are constraints in both the set points and the controller outputs, and the number of
degrees of freedom is equal to the number of manipulated (independent) variables, the optimum operating conditions occur when the sum of the number
of variables constrained is equal to the number of controller outputs.
Finally, model reference control systems are designed to handle constraints in
both the dependent and independent variables. The main concern addressed
by these techniques is that when one or more variables are driven against a
constraint, the optimum values of the remaining variables are not the same as
when all the variables can be set to their optimum values.
9-6. Summary
This chapter dealt with multivariable control systems and their tuning. It
showed the effect that loop interaction has on the response of feedback control
systems and presented two methods to deal with it: Bristols relative gains for
quantitatively determining the amount of interaction and for selecting the
199
References
1. Bristol, E. H. On a Measure of Interaction for Multivariable Process Control, IEEE Transactions on Automatic Control, V. AC-11, Jan. 1966, pp.133134.
2. Cutler, C. R. and Ramaker, B. L. DMC - A Computer Control Algorithm,
AIChE 1979 Houston Meeting, Paper #516; New York: AIChE, 1979.
3. Smith, C. A. and Corripio, A. B. Principles and Practice of Automatic Process
Control, 2nd ed., New York: Wiley, 1997.
Review Questions
9-1. Under what conditions does loop interaction take place? What are its
effects? What two things can be done about it?
9-2. For any given loop in a multivariable (interacting) system, define the
open-loop gain, the closed-loop gain, and the relative gain (interaction
measure).
9-3. How are the relative gains used to pair process variables and controller
outputs in an interacting control system? What makes it easy to determine the relative gains? What is the major shortcoming of the relative
gain approach?
9-4. In a 2x2 control system the four relative gains are 0.5. Is there a best way
to pair the variables to minimize the effect of interaction? By how much
does the gain of a loop change when the other loop is closed? Is the interaction positive or negative?
9-5. Define positive and negative interaction. What is the range of values of
the relative gain for each type of interaction?
200
9-6. The open-loop gains for the top and bottom compositions of a distillation
column are the following:
Distillate Composition
Bottoms Composition
Reflux
0.05
-0.02
Steam
-0.02
0.05
Calculate the relative gains and pair the compositions of the distillate
and bottoms to the reflux and steam rates so that the effect of interaction
is minimized.
9-7. The automated showers in the house of the future will adjust the hot and
cold water flows to maintain constant water temperature and flow. In a
typical design the system is to deliver 2.5 gallons per minute (gpm) of
water at 110F by mixing water at 170F with water at 80F. Determine
the open-loop gains, the relative gains and the preferred pairing for the
two control loops. Hint: the solution to this problem is identical to that of
Example 9-2.
9-8. Design a decoupler to maintain the temperature constant when the flow
is changed in the shower control system of the preceding exercise.
Dynamic effects can be neglected.
10
The Auto-tuner
Application
This chapter will describe the functionality and operation of the auto-tuner
application that is built into most process control software packages on the
market at this time. Knowing how to properly set up and use the application
is essential to obtaining acceptable results from the auto-tuner.
201
202
10-1. Operation
The auto-tuner application is an advanced software application that automatically handles all the conventional functions of tuning for the engineer or technician. Everything that you have learned in this book thus far, the auto-tuner
can do for you. Does this mean that you can forget all of the preceding chapters? Definitely not! You must have a working knowledge of the principles
that underlie those tuning procedures to be able to use the auto-tuner reliably
and effectively.
The auto-tuner is simply a tool that allows the engineer or technician to perform the tuning procedures discussed in this book in a timely, more efficient
manner. However, the application must be set up for the task at hand. This
requires knowing how to configure the application for each loop that needs to
be tuned.
Once the application has been configured properly, it will perform satisfactorily. The auto-tuner will perform the step test, make the necessary measurements and solve the calculations depending on the selected algorithm. There
are many algorithms to choose from, including Modified Ziegler-Nichols,
Internal Model Control, and Lambda tuning among others. Once the autotune test has been completed, the user can also switch between different algorithms instantaneously to compare the different outcomes based on the different algorithms. The auto-tuner can, in most cases, perform the step tests, make
the measurements and solve the calculations in less time than that which is
required to manually perform a step test, thereby saving a considerable
amount of time during the start-up of a process or during the general tuning
of process loops.
Local Override
In use, the auto-tuner will take control of the loop being tuned and put the
loop into LOCAL OVERRIDE. In Figure 10-1, LOCAL OVERRIDE is displayed on the faceplate as LO. This will let the board operator know that
another program is controlling the output. During this period, the auto-tuner
has control of the control valve (or another final control element) and it will
make adjustments according to how it has been configured. This means that
the auto-tuner will move the final control element as it sees fit for the purpose
of measuring the process parameters to obtain the desired tuning results,
instead of maintaining process control during normal operations.
203
Figure 10-1. Faceplate Showing the Control Loop in LOCAL OVERRIDE (LO)
204
Step Size
The step size can be selected depending on several variables of the loop being
tuned, the most important of which is process gain. Be aware that by performing a step test, you are causing a disturbance in a running process, and the
goal is to keep the disturbance to a minimum. For loops with a high process
gain, a small step size should generally be used. A step size of 3% of scale or
less can be chosen if there is very little process noise in the loop. On some
loops, a step size as small as 0.5% of scale can produce good results. For noisier loops or loops with a low process gain, a larger step size should be
selected. If you are unsure of the process gain, start with a smaller step size
and retest with a larger step size if necessary. Figure 10-2 shows the autotuner application tuning page. You can see that the step size parameter (in the
lower left corner) is selectable by dropping down the step size selection box.
One method that will aid in selecting an initial step size is to examine the
plant historian. Look at the historical data that was recorded during recent
process upsets and see whether small output responses were needed to rectify
a large process upset. If so, you can assume that the process gain is large and
you should start with a small step size. On the other hand, if you notice that
205
10-2. Applications
The first requirement in tuning any process is that the loop must be stable. If
the loop is in automatic mode and the process variable is stable, you should
have no problem with auto-tuning the process while it is in automatic mode.
If the historical trend shows that the process variable is stable overall but has
some significant process noise, you may still be able to tune the loop using a
larger step size.
If stability cannot be achieved in automatic mode, you should switch the loop
to manual mode. Remember to monitor the process variable any time that the
loop is not under automatic control. If after switching the loop to manual
mode you still observe variability in the loop, then there is another loop acting
on the loop under test. You must find the loop that is interacting with the loop
under test and stabilize that loop first to obtain a steady process variable in
preparation for the tuning procedures. Keep in mind that other process problems, such as feed composition changes or equipment malfunctions, may also
cause interactions with the loop under test. Interacting loops may be easy to
locate if they are cyclic or recurring, but they may be harder to find if they are
intermittent and interact with the loop under test less frequently. The facility
process engineer may be able to offer helpful advice when it comes to tracking
down interacting loops.
206
Not all loops can or should be tuned using the auto-tuner. In some cases, the
auto-tuner may not be able to achieve a sufficient response from the process
variable to provide accurate results. On some slower loops, the auto-tuner
may seem to hang up and not move the final control element as required to
generate oscillations in the process. When this happens, you have the option
to either wait and see whether the auto-tuner will switch states or to abort the
procedure and tune the loop using an alternate method.
In all cases, you should closely monitor the reaction of any loop in which the
tuning parameters have been modified. If the response of the loop is not
acceptable, you can retune the loop, or if necessary, revert to the original tuning values until the process becomes stable again. We will discuss the application of the auto-tuner later in this chapter.
The auto-tuner is just one tool that the engineer or technician has at his or her
disposal to help maintain the most efficient control of the process. A control
loop with a tight control response (when required) can save the facility unnecessary expense compared to a control loop that has high variability. For
instance, if you take the example of a process heater in which the fuel gas flow
is controlled by the heater outlet temperature, you can see where a steady control loop would cost less in expensive fuel gas when compared to a poorly
tuned loop in which the outlet temperature falls a few degrees and the control
valve opens a significant amount allowing a surge of fuel gas into the heater
to recover the few degrees of lost heat. Such a loop can be tuned using any
method, but if the time it takes to complete the tuning can be shortened by
using the auto-tuner, then the technician or engineer can move on to the next
loop sooner, thereby increasing profitability for the company.
207
The auto-tuner has special settings for slow time constant loops. These slow
time constant loops will usually take longer to auto-tune than standard loops.
They should be watched closely and the tuning should be aborted if the process does not respond within a time period that will not interfere with normal
process operations.
Loops that are classified as dead-time-dominant present a special challenge to
the auto-tuner. Some auto-tuners have a dead-time-dominant setting that can
be used to help with these loops. Be sure to take advantage of these advanced
settings when working with more challenging loops such as deadtime-dominant and other more challenging loops.
The auto-tuner works well with fast loops. A typical flow loop usually has a
small time constant. Look back at Figure 10-2 and notice how quickly the step
test takes place. When this is the case, the auto-tuner can almost always finish
the testing and report the results faster than the process control engineer can
manually perform the step test. A loop that has a small time constant will also
settle back to normal operating conditions very quickly after the step test disturbances have been injected. Most pressure loops are very fast by nature and
are easily tunable using the auto-tuner application. Some fast-acting level
loops are also good candidates for auto-tuning. Tight level control is easily
achievable using the auto-tuner on fast level loops. Loops with a small time
constant provide the most benefit from using the auto-tuner application
because of the relatively short amount of time it takes to complete the tuning
process using the auto-tuner.
There may be some loops in the process that would respond better if they
were not tightly tuned. For instance, the liquid level in a large tower may be
tuned less tightly to allow the level to fluctuate within reasonable limits if this
fluctuation does not cause disturbances in other loops (see Figure 10-3). If the
level control loop in the tower has a control valve that controls the level, and
the product leaving the tower is the feed to another unit, it may be more desirable to have the feed flow rate to the downstream unit remain more uniform
while allowing the level to fluctuate a bit more. In this case, the technician
could use the auto-tuner to achieve tight level control and then modify the
suggested tuning parameters to deaden the process response time a small
amount, or simply choose a more conservative tuning method, such as
Lambda tuning.
208
Loops with a very small process gain should be monitored closely during the
auto-tune process. See Figure 10-4 for an example of a loop with a small process gain. Notice that a step size of 12 percent of span was chosen for tuning
this particular loop. Also notice on the graph that the process variable only
moved a very small amount given the very large step size that was chosen.
209
Process Noise
Loops with a high level of process noise should also be monitored closely
during the tuning procedure. At times, the auto-tuner may actually mistake
process noise for a movement of the process variable PV in an incorrect
direction. In these cases, the auto-tuner may abort the tuning procedure,
forcing you to start over again. If possible, PV filtering may be used to limit
the amount of process noise seen by the auto-tuner. Remember that the autotuner cannot distinguish between process noise and an actual process reading.
Use the minimum amount of filtering required to limit the process noise and
dont forget to set the filter back to normal (if required) after the tuning
process is completed.
In all tuning procedures the loop (or loops) should be evaluated closely after
any change in tuning parameters is made. Watch for the loops response to
load changes, upsets, and set point adjustments. If the loop does not perform
satisfactorily, you may try retuning the loop, or you may want to reinstall the
original tuning parameters. The auto-tuner has a button that will allow you to
revert to the original tuning parameters easily, should it be required.
Desired Response
Most auto-tuner software has a feature that will allow the user to choose certain options in the tuning such as Normal, Fast, or Slow to obtain a particular
desired response from the tuning. The Fast option allows the user to try a
210
more aggressive approach for loops that need to return to set point quickly,
with overshoot being of a minimal concern. The Slow option reduces the
amount of overshoot experienced at the cost of PV being away from set point
longer. The Normal option is a trade-off between a quick return to set point
and a minimal amount of overshoot. Figure 10-5 illustrates the DESIRED
RESPONSE parameter in the highlighted drop-down box. As you can see, in
this case a NORMAL response has been selected.
Tuning Methods
Some expanded features of the software will allow you to choose which tuning method is employed to enhance the performance of the loop. You may
choose Ziegler-Nichols, Lambda, or Internal Model Control for more desirable
tuning results. One of the best things about the auto-tuner is that once you
have performed the step test and recorded the process dynamics of the loop,
you do not have to retest each time you wish to switch between tuning methods. By simply selecting the new tuning method, you will instantly have the
new calculation results and will be able to quickly compare them to one
another with just a few clicks of the mouse.
211
As you can see, the recommended setting is represented by a dot with RECOMMENDED beside it on the Tuning for Robustness graph. The second dot
depicts the simulated tuning value. It can be placed anywhere in the Tuning
for Robustness graph by simply clicking the mouse in the shaded area. The
auto-tuner will instantly recalculate the predicted response to the simulated
tuning values and plot the results in graphical form in the Simulation
Response area of the auto-tuners Simulate tab.
212
10-4. Summary
The auto-tuner application does not replace the knowledge necessary to perform tuning procedures. The auto-tuner can drastically reduce the amount of
time required for manual tuning step testing and the required mathematical
calculations, but it must be properly configured before beginning any tuning
procedures.
Proper setup is crucial to obtaining accurate results. By doing the necessary
preparation work before running the auto-tuner application, the engineer or
technician can ensure a better outcome.
The application will take control of the final control element and cause it to
move to generate a measurable response in the process variable. Constant
communication with the process operators is a must during the period that
the control loop is in LOCAL OVERRIDE.
213
The user can select a certain desired response before testing. Once the process
dynamics have been measured, the user can switch between several of the
most common tuning methods available. The auto-tuner application also has
the ability to project expected responses to process disturbances based on
input from the user using the simulation feature that is included with the
auto-tuner.
Review Questions
10-1. Briefly describe LOCAL OVERRIDE and explain its effects on the control loop.
10-2. While selecting a step size for auto-tuning a control loop, you notice that
the process historian shows that the last upset took large movements of
the final control element to bring the process variable back to set point.
What can you assume about this loop and what step size should you
begin with?
10-3. What is the first requirement in tuning any process?
10-4. What types of loops does the auto-tuner work best on?
10-5. Explain why the user must maintain constant contact with the board
operator while performing tuning procedures.
10-6. Describe the Simulate feature and how it outputs the results of simulated tuning.
Appendix A
Suggested Reading and
Study Materials
Blevins, T. L., McMillan, G. K., Wojsznis, W. K., and Brown, M. W., Advanced
Control Unleashed: Plant Performance Management for Optimum Benefit, ISA,
Research Triangle Park, NC.
McMillan, G. K. and Cameron, R. A., Advanced pH Measurement and Control,
3rd edition, ISA, Research Triangle Park, NC.
McMillan, G. K., and Toarmina, C. M., Advanced Temperature Control, 2nd edition, ISA, Research Triangle Park, NC.
McMillan, G. K., Good Tuning, A Pocket Guide, 3rd edition, ISA, Research Triangle Park, NC
Murrill, P. W., Fundamentals of Process Control Theory, 3rd edition, ISA,
Research Triangle Park, NC.
Shinskey, F. G., Process Control Systems, 3rd ed., McGraw-Hill, New York, NY,
1989.
Smith, C. A., and Corripio, A. B., Principles and Practice of Automatic Process
Control, 3rd edition, Wiley, New York, NY, 2006. (Note: The material on
Dynamic Matrix Control is in the 2nd edition, 1997).
215
Appendix B
Answers to Study
Questions
Chapter 1
1-1.
1-2.
The two process characteristics to be considered when tuning the controller are the process sensitivity or gain and its rate of response
(Section 1-1).
1-3.
1-4.
The fourth element of the feedback loop is the process (Section 1-2).
1-5.
1-6.
The fail position of the cooling water valve must be open so that coolant
is not lost on loss of power; the controller signal then closes the valve, so
the controller action must be reverse. In other words, when the tempera-
217
218
ture increases the controller output decreases, opening the valve to supply a higher-cooling water flow.
1-7.
The reactants control valve must fail closed so that the reactor does not
overflow on loss of power. The controller action must then be reverse;
that is, increasing level decreases the controller output to close the valve
and reduce the flow of reactants into the reactor.
1-8.
The controller must have direct action; that is, increasing caustic composition must increase the controller output to increase the flow of dilution
water and decrease the composition of the caustic product.
Chapter 2
2-1.
Proportional gain of 3.
a. The controller output decreases by 3(10%) = 30%.
b. The change in PV is a decrease of 15C/(150C 0)100% = 10%.
Controller output decreases by 3(10%) = 30%.
c. The PV increases by:
(250 kg/hr)/(50,000 kg/hr 0)100% = 0.5%.
Controller output decreases by 3(0.5%) = 1.5%.
2-2.
2-3.
2-4.
The cause is that the controller has the incorrect action. To correct,
change the controller action.
2-5.
The problem is that the controller gain is too high (or the integral time is
too low). To correct, decrease the controller gain by at least half. If this
does not work, adjust the integral time using the period of the oscillations as a guide; that is, set the integral time to be of the order of magnitude of the period of the oscillations.
219
2-6.
The integral and derivative times in Table 2-1 are related to the period of
oscillation of the loop because the period is an indication of the speed of
response of the loop.
2-7.
Chapter 3
3-1.
The procedure for the open-loop step test is as follows (Section 3-1):
a. Switch the controller to manual output.
b. Apply a small step change, 1 to 3%, in the controller output.
c. Record on the same trend the controller output and the process
variable until the PV reaches a new steady value.
d. Analyze the PV response to obtain the gain, dead time, and time
constant.
3-2.
The parameters of the SLPDT model are the gain, the dead time, and the
time constant (Section 3-2). The gain is an indication of the sensitivity of
the process variable to the controller output; the dead time is a measure
of how long it takes the PV to start responding to the change in controller output; and the time constant is a measure of how long it takes the
process to respond to the action of the controller.
3-3.
220
3-5.
3-6.
3-7.
3-8.
221
Chapter 4
4-1.
4-2.
The uncontrollability is the ratio of the dead time to the time constant.
The actual magnitude of the time constant and the dead time determine
how fast the loop can respond, but not its controllability (Section 4-1).
4-3.
The proportional gain from the ultimate gain is Kcu/2 and from the
uncontrollability parameter is 1/KPu. For them to be the same, Kcu = 2/
KPu. The PID integral time from the ultimate period is Tu/2 and from
the dead time is 2t0, so for them to be the same, Tu = 4t0 (Section 4-1).
4-4.
The sensitivity is the measure of how much the process variable changes
when the controller output changes; that is, the gain, so Process A is the
least sensitive and Process C the most. The speed of response is deter-
222
mined by the process time constant, so Process B is the slowest and Process A the fastest. The uncontrollability is the ratio of the dead time to
the time constant, so Process C is the least controllable and Processes A
and B are equally controllable (Section 4-1).
4-5.
Using the tuning strategy proposed in the chapter, the PID tuning
parameters are (Section 4-1):
Process A:
Process B:
Process C:
Kc = (1.2/0.5)(2.0/0.2) = 24
TI = 2(0.2) = 0.4 min
TD = 0.2/2 = 0.1 min
Kc = (1.2/2.0)(30/3.0) = 6.0
TI = 2(3.0) = 6.0 min
TD = 3.0/2 = 1.5 min
Kc = (1.2/4.0)(5.0/3.0) = 0.5
TI = 5.0 min (use IMC rule, 2(3) > 5)
TD = 3.0/2 = 1.5 min
4-6.
The controller would be configured to act on the process variable to prevent large changes in controller output when the set point is changed
and the controller gain is high. When configured in this manner, the
output ramps to the new required value at the rate controlled by the
integral time when the set point is changed (Section 4-1).
4-7.
4-8.
Inverse response is the case where the process variable initially moves
in the opposite direction of its eventual direction of change. It is detrimental to the performance of a feedback controller because it causes the
controller to initially move the output in the wrong direction
(Section 4-4).
223
Chapter 5
5-1.
Tight level control is indicated when the level has significant effect on
the process operation, as in a natural-circulation evaporator or reboiler.
Averaging level control is to be used when it is necessary to smooth out
sudden variations in flow, such as in a surge tank receiving discharge
from batch operations to feed a continuous process. Tight level control
is the one that requires the level to be kept at or very near its set point
(Section 5-3).
5-2.
For flow control loops a proportional-integral (PI) controller is recommended with a gain near but less than 1.0. The integral time is usually
small, on the order of 0.05 to 0.1 minutes (Section 5-2).
5-3.
For tight level control a proportional controller with a high gain, usually
greater than 10, should be used. When the lag of the control valve is significant, a proportional-derivative controller could be used. If a proportional-integral controller is used, the integral time should be long, on the
order of one hour or longer (Section 5-3).
5-4.
5-5.
5-6.
5-7.
PID controllers are commonly used for temperature control so that the
derivative mode compensates for the lag of the temperature sensor,
which is usually significant (Section 5-4).
5-8.
The major difficulty with the control of composition is the dead time
introduced by sampling and by the analysis (Section 5-5).
224
Chapter 6
6-1.
6-2.
The derivative kick is a pulse on the controller output that takes place
at the next sample after the set point is changed and lasts for one sample. It can be prevented by having the derivative term act on the process
variable instead of on the error (deviation) (Section 6-1).
The derivative filter or dynamic gain limit is needed to prevent large
amplification of changes in the process variable when the derivative
time is much longer than the algorithm sampling interval.
6-3.
6-4.
6-5.
Using the formulas of Table 6-2, with q = 0 (for maximum gain) and the
following parameters (Section 6-2):
K = 1.6
6-6.
225
t = 20 min
0.067
50
a = exp(T/t)
.9934
0.905
0.368
0.0067
N = t0/T
75
Gain
2.0
1.7
0.90
0.046
20.0
19.5
17.6
4.47
If the algorithm has dead time compensation, the gain can be higher
because it does not have to be adjusted for dead time. This does not
affect case (d) because the dead time is less than one sample, and therefore, no dead time compensation is necessary. From Equation 6-5 and
Table 6-3 (Section 6-4):
Sample time, min
6-7.
t0 = 5 min
0.067
N = t0/T
75
Gain
153
5.9
1.8
20
19.5
17.6
The basic idea of the Smith Predictor is to bypass the process dead time
to make the loop more controllable. This is accomplished with an internal model of the process responding to the manipulated variable in parallel with the process. The basic disadvantage is that a complete process
model is required but it is not used to tune the controller, creating too
many adjustable parameters (Section 6-4).
The Dahlin Algorithm produces the same dead time compensation as
the Smith Predictor but it uses the model to tune the controller, reducing
the number of adjustable parameters to one: q.
Chapter 7
7-1.
Cascade control (1) takes care of disturbances into the secondary loop
reducing their effect on the controlled variable, (2) makes the primary
loop more controllable by speeding up the secondary loop and (3) handles the nonlinearities in the inner loop, where they have less effect on
controllability (Section 7-1).
226
7-2.
7-3.
The primary controller in a cascade control system has the same requirements as the controller in a simple feedback control loop, thus the tuning and mode selection of the primary controller are no different from
those for a single controller (Section 7-2).
7-4.
7-5.
The controllers in a cascade system must be tuned from the inside out,
because each secondary controller forms part of the process controlled
by the primary around it (Section 7-3).
7-6.
7-7.
7-8.
7-9.
Integral windup can occur in cascade control when the operating range
of the secondary variable is narrower than the transmitter range. To pre-
227
vent it, the secondary measurement can be passed to the integral feedback of the primary; in such a scheme the primary always takes action
based on the current measurement, not on its set point (Section 7-4).
Chapter 8
8-1.
8-2.
To be used by itself, feedforward control requires that all the disturbances be measured and accurate models be developed of how the disturbances and the manipulated variable affect the controlled variable
(Section 8-1).
Feedforward with feedback trim has the advantage that only the major
disturbances have to be measured and compensation does not have to
be exact, because the integral action of the feedback controller takes care
of the minor disturbances and the model inaccuracies.
8-3.
8-4.
228
8-6.
Dead time compensation consists of storing the feedforward compensation and playing it back some time later. The time delay is the adjustable
dead time parameter (Section 8-3).
Dead time compensation can be used only when the feedforward action
is to be delayed and a computer or microprocessor device is available to
implement it. It should be used only when a simple lag cannot
effectively delay the feedforward compensation.
8-7.
229
5. Numerical values are needed to evaluate the importance of each disturbance. The change in each disturbance required to cause a given
change in main fuel flow would be calculated (Section 8-4).
6. Feedback trim can be added as in Example 8-1:
Feedback output: OP = Toset
Design formula:
Fset = (Cp/Hm)[OP - Ti]W - (Hs/Hm)Fs
7. Lead-lag units must be installed on the process flow and inlet temperatures, but not on the supplementary fuel gas flow, because its
dynamic effect should match that of the main fuel gas flow
(Section 8-4).
8. Instrumentation diagram (Figure B-8-1):
Figure B-8-1.
TT
Toset
TT
Ti
SP
To
Lead-Lag
1
Lead-Lag
2
TC
OP
FT
FT
Fs
Adder
Toset - Ti
Multiplier
+
Adder
Fset
230
Chapter 9
9-1.
Loop interaction takes place when the controller output of each loop
affects the process variable of the other loop. The effect is that the gain
and the dynamic response of each loop change when the auto/manual
state or tuning of the other loops changes (Section 9-1).
When loop interaction is present, we can (1) pair the loops in the way
that minimizes the effect of interaction and (2) design a control scheme
that decouples the loops.
9-2.
9-3.
To minimize interaction for a loop, the relative gain for that loop must
be as close to unity as possible. Thus, the loops must be paired to keep
the relative gains close to unity, which in a system with more than two
control objectives may require ranking the objectives (Section 9-2).
The relative gains are easy to determine because they involve only a
steady-state model of the process, which is usually available at design
time.
The main shortcoming of the relative gain is that it does not take into
account the dynamic response of the loops.
9-4.
When all four relative gains are 0.5, the effect of interaction is the same
for both pairing options. The gain of each loop will double when the
other loop is switched to Automatic. The interaction is positive; that is,
the loops help each other (Section 9-2).
9-5.
When the effect of interaction with other loops is in the same direction
as the direct effect for that loop, the interaction is positive; if the interaction and direct effects are in opposite direction, the interaction is negative. For positive interaction, the relative gain is positive and less than
231
unity, while for negative interaction the relative gain is either negative
or greater than unity (Section 9-2).
9-6.
| Reflux
Yd | 1.19
Xb | 0.19
Steam
-0.19
1.19
The top composition must be paired to the reflux and the bottoms
composition to the steam to minimize the effect of interaction.
9-7.
Let H be the flow of hot water in gpm, let C be the flow of cold water in
gpm, let F be the total flow in gpm and let T be the shower temperature
in F. The mass and energy balances on the shower, neglecting variations in density and specific heat, give the following formulas:
F=H+C
T = (170H + 80C)/(H + C)
These are the same formulas as for the blender of Example 9-2. The
relative gains are therefore:
| Hot
F | H/F
T | C/F
Cold
C/F
H/F
232
9-8.
Chapter 10
10-1. LOCAL OVERRIDE is when another program takes control of the output and moves it to measure process parameters to obtain tuning
results.
10-2. The loop has a small process gain and you should start with a larger
step size.
10-3. The process must be stable.
10-4. Loops with a fast (that is, small) time constant.
10-5. If the process starts to become unstable, the board operator will need to
retake control of the loop immediately.
10-6. The Simulate feature allows the user to simulate loop response for different tuning settings. The simulated response is presented in the form
of a chart that displays the projected amount of overshoot versus the
time required to reach set point again.
INDEX
Index Terms
Links
A
action (direct or reverse)
adaptive control
69
algorithm
12
9698
100
103
115
131
139
145
151
155156
202
224225
112
128
134
96
136
153
ammonia synthesis
analog
224
analysis cycle
91
109
analyzer control
92
95
191192
196
arrow
124
automatic output
auto-tuner application
auto-tuning
34
202
8
205207
213
82
8487
101
125
223224
batch process
51
66
87
bias
82
101102
118
44
49
224
blending tank
42
86
Index Terms
block diagram
Links
56
113114
122
124
127
144147
149
C
capacitance
cascade control
cascade dontrol
4145
72
78
18
80
100
121125
127129
131
133
135138
140
143
225226
130
135
characteristics
control loop
control valve
21
5051
derivative mode
16
feedback control
144
19
feedforward control
144
loop interaction
173
PI controllers
process
86
12
29
74
217
clamp limit
138
clamped
137
closed-loop gain
closed-loop time constant
coarse tuning
compensation for dead time
composition control
175178
199
41
230
106
67
113
156
54
77
190
193196
180
141
conductance
4146
conductance, valve
4344
46
54
This page has been reformatted by Knovel to provide easier navigation.
50
Index Terms
Links
control objective
7778
122
143144
158
160
162
169170
178179
192194
227228
230
56
3435
37
5053
58
60
6970
72
74
7980
82
8586
122123
128
136
143
148
163
179
186
189
191
202
206207
217218
222223
21
25
5961
6364
76
113
122
124
221
control valve
controllability
225
controllable process
64
116
controlled variable
34
6870
78
96
113
125
136
206
225
227
18
27
80
128
223
11
18
2829
87
91
101
224
217218
controller
proportional-integral
proportional-integral-derivative
96
proportional-only
82
single-mode
18
three-mode
18
two-mode
18
controller action
Index Terms
controller gain
controller output
controller synthesis
Links
12
2122
29
5860
64
66
7576
8081
8586
97
102
104
106
109
115
127
195
218
221222
224
3
79
92
104
80
113
3334
3841
4650
53
5760
6264
68
7274
78
91
103106
111
119
127
132
144
156157
196
206
219223
67
69
95
113117
119
156157
165166
225
228
113
151152
157159
decoupler
78
169
174
186194
200
114
current-to-pressure transducer
D
Dahlin controller
dead band
113117
78
128
dead time
decoupling
dependent variables
derivative
filter
197
12
16
126127
226
9798
224
98
Index Terms
Links
derivative (Cont.)
kick
97
100
118
1416
21
25
27
58
61
6667
7778
82
97
110
112
126
128
140
218
223
226
1520
24
2830
5860
68
82
9799
106
127128
224
mode
time
224
unit
98100
derivative-on-PV
127128
100
desired response
209210
digital controller
136
direct action
67
40
194
218
disturbance
173
184185
45
48
129130
170
181184
199
18
96
148
160
12
19
21
2324
31
34
6267
6972
78
82
92
100
107
112113
122
124
126
133
136
144147
149154
156
161
164165
171
186
203205
207
Index Terms
Links
disturbance (Cont.)
213
222
227
229
116
146
151152
156
159
163
165
187189
98
118
dynamic interaction
185
199
162
228
dynamic compensation
225
224
E
efficiency
electrical circuit
43
electronic circuits
96
estimation of time
182
185
50
198
50
57
20
25
80
97
101102
106
109
112113
118
124
140
148
152
197
205
224
41
45
106
108109
34
78
1112
78
96
123
144
150
164166
198
1920
23
29
34
49
F
fast process
feedback control
Index Terms
Links
feedback controller
79
103
111
125
140
144
148
157
169
193
217
226
45
1112
19
23
2829
3637
48
67
69
72
76
86
113114
125
136
145148
151152
159
162164
175
185
222
144
146149
151
159
161162
165166
227
229
67
69
72
144146
148
151
157
160161
163
165166
192
227
feedback trim
feedforward control
227
feedforward element
146
feedforward-feedback control
146
fieldbus
filter parameter
fine tuning
flow control
frequency
9798
155156
68
7981
85
90
92
123
190
196
223
16
78
85
112
131
159
Index Terms
Links
G
gain
1112
16
18
20
22
2426
30
34
3637
3940
50
53
5758
6061
64
74
76
80
82
87
92
101
105
107
110
113
126
128
132
148
151153
156157
162
177
179
181
184185
190
194196
199
217
219
223
225
230
closed-loop
175178
199
230
nonlinear
101102
118
224
open-loop
174178
180
183184
199200
230
175180
183
185186
189
193196
198200
178
180181
113
137
44
79
relative
230231
steady-state
158
195
variation
53
gap
7071
139
42
160
Index Terms
Links
H
half decoupling
188189
heat exchanger
45
61
88
4445
50
81
87
89
93
130
161
36
19
21
25
3436
52
6162
74
88
9091
107
115
149
160161
163165
144
heat rate controller (QC)
heat transfer
heater
88
206
efficiency
162
feedforward controller
163
temperature
hydrogen/nitrogen ratio
hysteresis
106110
117
134
69
8081
128
60
6264
74
80
92
132
92
108109
1314
21
25
51
58
66
7071
7879
82
85
112
124
126
136
139140
206
I
IMC
222
independent variables
integral controller
198
7980
223
integral mode
Index Terms
Links
integral time
integrating process
interaction
interaction measure
145
152
172
175
188
218
222
226
230
13
17
1921
2425
30
5860
6264
66
6869
74
7980
82
8687
92
97
100
105
107
109
112
127128
132
136137
196
218
221225
41
7
69
112113
158
169
171173
175
178181
184185
189
193194
196
199200
205
174
176177
199
230
intermediate level control
86
60
6264
74
80
92
132
7174
76
133
173
185
188
194
222
222
intuitive
174
inverse response
J
jacketed reactor
136
Index Terms
Links
L
lag
41
153
lagniappe
123
lead
153
lead-lag compensation
163
lead-lag unit
144
151153
155156
163164
166
187
72
8182
8487
92
96
118
125
182
207
227228
level control
223224
linear feedforward controllers
liquid storage tank
149
4244
local override
202203
18
loop interaction
152
205
212213
170
173174
186
198199
230
22
60
68
7071
78
88
149
151152
160
162
166
174
206
217
225
227228
M
manipulated variable
manual output
master controller
material balance control
measured disturbance
34
7
138
184185
144
147148
158161
163
166
197
227228
Index Terms
microprocessor
microprocessor-based controllers
Links
96
130
156
228
35
100101
111
113
118
224
96
205
103
minor disturbances
146
mode
12
152
automatic
205
derivative
1416
21
25
27
30
58
61
6667
78
82
87
97
110
126128
218
223
226
1314
21
25
51
58
7071
7879
82
112
124
126
136
139140
145
152
171172
175
188
218
222
226
12
14
16
30
66
76
82
100
125
integral
230
manual
205
proportional
218
rate
14
reset
13
171
selection
77
226
196
move-suppression parameters
198
143
multiplexer
140
This page has been reformatted by Knovel to provide easier navigation.
Index Terms
multivariable control
Links
169170
174
182
190
192194
196
173
179
181
185186
188
194195
199
231
198
N
negative feedback
negative interaction
noise
209
158
nonlinear gain
nonlinearity
101102
118
224
58
148
221
174
181
12
2627
78
126
139
226
78
90
174178
180
199200
230
3336
58
6163
67
74
86
103
110
O
off-line
offset
on-off controllers
open-loop gain
open-loop test
112
oscillations
203
output
184
47
1214
16
1820
2527
30
3437
3941
5051
54
6163
6576
80
82
85
9598
100104
106
108
111114
Index Terms
Links
output (Cont.)
overshoot
116
118
122
125
128
131
133
135137
139
144
146
148150
152153
155159
162163
169171
173174
176179
181
186
188
190
194
197199
202
204
218219
221222
224
229230
63
6971
100
125
127
132
136137
139
173
188
194195
210211
222
226
232
173180
183
185186
188189
195
199200
69
194
P
pairing
230
parallel
45
225
parallel paths
149
PD controller
78
17
perfect control
performance
188
144
166
227
2829
50
5759
64
6669
72
74
76
9596
99100
107
110113
116
121
123125
127
133
140
144
151
Index Terms
Links
performance (Cont.)
158
165
169
173174
181
185
188
210
221222
2526
30
58
6163
74
79
8586
92
97
105106
109110
115116
132
192
223
226
pH control
101
PI controller
PID
PID algorithm
96
PID controller
1112
17
21
2427
30
58
6164
72
76
9293
106
110
114
132
223
173
179
185
195
230
8081
86
92
112
125
136
pneumatic
96
positive interaction
pressure control
primary controller
122
problematic loops
206
37
48
59
76
106
109
111
113
225
33
50
5253
55
5759
65
68
109
122
204
208
220221
process gain
232
process noise
209
Index Terms
Links
process nonlinearity
33
51
74
37
41
60
86
91
106
108
111
222
13
58
1219
25
3436
41
46
64
6667
71
76
78
82
8889
9192
9598
100
102
106
110
112
115
118
121
125128
131
136
139140
144145
149
151152
158159
166
169174
176
178179
181
187188
192193
197199
205206
208209
212213
217
219
221222
224
226227
process variable
230
processing frequency
programmable logic controllers (PLC)
111
96
proportional band
1718
78
82
proportional controller
2123
2526
29
78
8586
90
92
102
223
12
17
19
22
2728
58
65
68
78
80
82
8586
97
101
118
125126
133
218
221
226
proportional gain
Index Terms
proportional kick
proportional mode
proportional-integral controller
proportional-integral-derivative controller
Links
100
118
224
1214
16
66
76
82
100
125
218
18
27
128
223
11
18
2829
87
91
101
224
9698
100
224
1217
2527
30
34
3739
48
6164
9699
102
145146
149150
170172
175177
187188
209210
218219
23
25
58
74
133
24
26
6163
132
80
96
proportional-only controller
proportional-on-PV
pulse
PV
82
100
Q
QDR response
QDR tuning
28
58
R
rate time
ratio control
reactor
14
58
78
143144
148149
166
179
227
41
54
7073
122123
126
Index Terms
Links
reactor (Cont.)
129
131133
135
137
139
218
175180
183
185186
189
193196
198200
174
178
185
138139
226
reset time
13
58
reset windup
51
57
6971
74
76
121
128
130
136
138
140141
222
4243
45
86
19
217
relative gain
230231
matrix
reset feedback
resistance
resistance temperature device (RTD)
89
reverse
reverse action
running away
19
S
sample time
106
225
sampling frequency
103
107
111
131
sampling period
saturation
secondary controller
self-regulating
91
104
5051
6970
138
188
136
122
6
35
41
48
60
78
82
87
108
123124
127128
145
158
164
206
223
226
205
sensor
Index Terms
sensor time constant
sensor/transmitter
Links
87
89
128
3435
37
58
79
91
45
4849
98
127128
226
57
1219
2224
27
30
6171
7576
7880
82
86
89
92
9698
100102
106
113
116
118
122123
125128
131133
135137
139
144148
150152
158159
162
166
170171
173
181
186
189190
192193
198
203
209210
213
222224
226227
232
80
100
171
series
set point
146
shrink
72
simple lags
41
211
simulation response
211
single-mode controller
18
slave
67
slave controller
18
slow sampling
108109
Smith Predictor
113114
119
225
11
14
16
19
21
78
80
124126
stability
Index Terms
Links
stability (Cont.)
165
188
192
205
static compensation
static compensator
163164
187
152
static friction
80
steady-state
52
55
71
78
158
160
165
172
174175
177178
228
158
178
180181
148
160
steady-state gain
195
steam heater
115116
step size
204
step test
33
3536
3941
48
5354
57
68
103
106
131132
202
204
207
210
212
19
25
35
70
72
87
9091
108110
117
123124
129
139
160
165
167
223
219
stiction
206
subcritical
46
swell
72
T
temperature control
128
18
Index Terms
tight control
time constant
time delay
Links
68
81
83
87
92
206
3334
3746
48
50
5355
5760
6264
68
74
76
7880
82
8586
89
91
93
97
103106
108
111113
115
119
128129
132
135136
144
151
153
156
189
191
194
207
219223
227228
232
4647
91
145
40
4648
228
transducer
transfer function
113
transportation lag
36
163
tuning
211
tuning methods
210
tuning parameter
23
1214
17
2429
6163
6768
76
104
107
109
114
116
119
126
193
198
206207
209
212
222
two-mode controller
two-point method
18
181
Index Terms
Links
U
ultimate gain
ultimate period
uncontrollability
uncontrollable process
21
2425
27
29
34
58
76
103
21
23
25
27
31
76
103
221
5962
6769
76
91
104
111
144
221222
64
67
69
1921
41
53
86
188
203
74
unstable
232
V
vacuum
130
226
valve
characteristics
50
conductance
50
gain
50
hysteresis
69
112
velocity
139
8081
W
windup
51
57
6971
74
76
121
128
130
136
138141
222
226
Index Terms
Links
Z
Ziegler and Nichols
2324
38
74
58