Академический Документы
Профессиональный Документы
Культура Документы
In order for the reader to fully understand the content and as regulators, having a set point that remains unchanged for
concepts of this section, it is advisable to first become familiar days and even months at a time.
with some basic topics. These include gains, time lags and reac- Examples of variables held at constant set points are
tion curves (Section 2.22), the PID control modes (Section 2.3), drum-level and steam temperature of a boiler, most pressure
feedback and feedforward control (Section 2.9), and relative and level variables, pH of process and effluent streams, most
gain calculations (Section 2.25). product-quality variables, and most temperature loops. Set-
Controllers are designed to eliminate the need for con- point response is of no importance to these loops, but they
tinuous operator attention when controlling a process. In the must contend with load upsets minute by minute. In fact, the
automatic mode, the goal is to keep the controlled variable only loops in a continuous plant that must follow set-point
(or process variable) on set point. The controller tuning changes are flow loops.
parameters determine how well the controller achieves this Batch plants have frequent transitions between steady
goal when in automatic mode. states, some of which require rapid response to set-point
changes with minimal overshoot. However, some of these
changes are large enough to saturate the controller, particu-
larly at startup. This can cause integral windup, which
DISTURBANCES requires special means of prevention to overcome.
The purpose of a controller is to keep the controlled variable The Load Only pure-batch processes — where no flow into
as close as possible to its set point at all times. How well it or out of the process takes place — operate at constant load,
achieves this objective depends on the responsiveness of the and that load is zero. All other processes can expect to encoun-
process, its control modes and their tuning, and the size of ter variations in load, which are principally changing flow rates
the disturbances and their frequency distribution. entering and leaving vessels. A liquid-level controller, for
example, manipulates the flow of one liquid stream, while
Sources other streams represent the load. Feedwater flow to a boiler is
manipulated to control drum level and must balance the com-
Disturbances arise from three different sources: set point, bined flows of steam and blowdown leaving to keep level at
load, and noise. Noise is defined as a random disturbance set point. The load changes frequently and often unpredictably,
whose frequency distribution exceeds the bandwidth of the but the set point may never change.
control loop. As such, the controller has no impact on it, In a typical temperature control loop, the load is the flow
other than possibly amplifying it and passing it on to the of heat required to keep temperature constant. Liquid entering
final actuator, which can cause excessive wear and ultimate a heat exchanger will require a certain flow of steam to reach
failure. a controlled exit temperature. Variations in liquid flow and
Set point and load changes affect the behavior of the inlet temperature will change the demand for steam flow
control loop quite differently, owing to the dynamics in their manipulated to keep exit temperature at set point.
path. A controller tuned to follow set point changes tends to
respond sluggishly to load variations, and conversely a con-
troller tuned to correct disturbances tends to overshoot when Dynamics
its set point is changed.
The term process dynamics can refer to capacitance, inertia,
resistance, time constant, dead time or their combinations.
Set Point The set point is the desired value of the controlled There is no dynamics involved with changing the set point,
variable and is subject to adjustment by the operator. In a unless intentionally placed there for the purpose of filtering
continuous process plant, most of the control loops operate the set point.
414
Step Responses
Controlled variable
manual mode and stepping the output by the desired amount Load
in the desired direction, and immediately (before a deviation SP filter tuning
Load
develops) transferring back to the automatic mode.
This procedure can be followed for all but the fastest
loops, such as flow loops. For them, a step in set point is
acceptable, both because flow loops must follow set-point
changes, and because for them, set-point tuning gives accept-
able load response. 0 1 2 3 4
Time, t/St
Comparing Set-Point and Load Responses
FIG. 2.35c
The steady-state process gain of a flow loop is typically Set-point tuning slows load recovery for lag-dominant processes.
between 1 and 2, as indicated by the controller output being
between 50 and 100% when the flow measurement is at full
scale. The proportional gain of a typical flow controller is in
and longer integral time than is optimum for load regulation,
the range of 0.3 to 1.0, with the higher number associated
or a filter must be applied to the set point.
with the process that has the lower steady-state gain. There-
Figure 2.35c compares responses to steps in set point and
fore, the proportional loop gain for a typical flow loop is in
load for a process with distributed lag such as a dashed
range of 0.6 to 1.0. As a result, a step change in set point
exchanger, distillation column, or stirred tank. The time scale
will move the controller output approximately the correct
is normalized to Στ , which is the time required for the dis-
amount to produce the same change in flow, by proportional
tributed lag to reach 63.2% of the full response to a step input
action alone, that gives excellent set-point response.
in the open loop. It is also the residence time of liquid in a
This is not the case for other loops. Level has the opposite
stirred tank.
behavior. To maintain a constant level, the controller must
If the PID settings are adjusted to minimize the Integrated
match the vessel’s inflow and outflow precisely. Changing
Absolute Error (IAE) to the set point change, the dashed
the set point will cause the controller to change the manip-
response curve is produced (SP tuning). Note that following
ulated flow, but only temporarily — when the level reaches
a step change in load, the return to set point is sluggish. This
the new set point, the manipulated flow must return to its
is commonly observed with lag-dominant processes. The PID
original steady-state value.
settings that produce the minimum-IAE load response, shown
Therefore, no steady-state change in output is required
in black (no filter), result in a large set-point overshoot,
for a level controller to respond to a set-point change. The
however.
Integrated Error (IE) sustained by a controller following a
disturbance varies directly with the change in output between
Set-Point Filtering
its initial and final steady states. In response to a set-point
change, the level controller has the same initial and final If optimum load rejection is desired, without the large set-
steady-state output values and hence sustains zero integrated point overshoot that it produces on lag-dominant processes,
error. the proportional response to set-point changes needs to be
As a result, the error that is integrated while the level is reduced.
approaching the new set point will be matched exactly by an Some PID controllers have the option to eliminate pro-
equal area of overshoot. In other words, set-point overshoot portional action on the set point altogether. This tends to
is unavoidable in a level loop unless set-point filtering is produce a set-point undershoot, which can significantly dashed
provided. the controller response, and it should never be used in the
Most other processes, such as temperature, pressure, and secondary controller of a cascade system. (Incidentally,
composition, have steady-state gains higher than those of a derivative action should never be applied to the set point, as
flow process. But more importantly, they are also dominated this always produces overshoot.)
by lags, which allows the use of a higher controller propor- Some controllers can reduce the controller’s proportional
tional gain for tight load regulation. When this high propor- gain when it acts on set point changes, either through the use
tional gain is multiplied by the process steady-state gain, the of a lead-lag filter, or by the use of a specially structured
resulting loop gain can be as high as 5 to 10 or more. algorithm. This adjustment allows separate optimization of
A set-point step then moves the controller output far more set-point response, after the PID settings have been tuned to
than required to drive the controlled variable to the new set optimize the controller’s load response. The filter used for
point, producing a large overshoot. To minimize set-point the loop whose response is shown in Figure 2.35c has applied
overshoot, the controller must be detuned, with lower gain only half the controller’s proportional gain to the set point.
130
Process Variable
120
23 s
40 80
50
48
Controller Output
46
44
42
40
40 80
Time (sec)
FIG. 2.35e
Process reaction curve in response of a change in controller output (CO). The process variable (PV) range is 0 to 300 degC and the CO
range is 0 to 100%.
170 τ = 183 − 33 s
160
Process variable
150
140
130
120
400 800
50
Controller output
48
46
44
42
40
400 800
Time (sec)
FIG. 2.35g
Maximum slope curve, Fit 1.
(manipulated variable) to a process. If a different-size step The first-order lag time constants are given by:
change in controller output was applied, the value of τd would
not change significantly.
Fit1: τ F1 = K / Rr = 1.5/0.1%/s = 150 s
However, the value of Rr is essentially directly propor-
tional to the magnitude of the change in controller output. (same slope as previous section) 2.35(3)
Therefore, if a two-unit change in output was used instead Fit 2: τ F1 = t63.2% − t0 = 155 s – 33 s = 122 s 2.35(4)
of a one-unit change, the value of Rr would be approximately
twice as large. For this reason the value of Rr used in Equation
In Equation 2.35(4), t63.2% is the time necessary to reach
2.35(2) or any other must be the value that would be obtained
63.2 % of the final value, and t0 is the time elapsed between
for a one-unit change in controller output.
the CO change and the beginning of the PV change. Note that
In addition to the dead time and reaction rate, the value
the parameters for Fit 1 are based on a single point on the
of the process gain K must also be determined as follows:
response curve, which is the point of maximum rate of ascent.
However, the parameters obtained with Fit 2 are based on two
final steady-state change in controlled vaariable (%) separate points.
K= 3
change in controller ouput (control unit) Studies indicate that the open-loop response based on
2.35(2) Fit 2 always provides an approximation that is as good or
better than the Fit 1 approximation. A typical curve resulting
There is a second method to determine the pure time delay from the above procedure is shown in Figure 2.35i.
plus the first-order lag approximation. In order to distinguish The response shown in Figure 2.35i resulted from a ten-
between these two methods, they will be called Fit 1 (described unit change in controller output. For different step changes,
in Figure 2.35g) and Fit 2 (described in Figure 2.35h). K and Rr must be adjusted accordingly. From a curve such
The only difference between these two is in how the first- as in Figure 2.35i, a number of parameters can be determined.
order time constant is obtained. In case of Fit 2, the time The controller settings are calculated from the equations in
constant of the process is determined as the difference Table 2.23j:
between the time when the dead time ends and the time when Table 2.35k compares the results obtained in terms of pro-
the controlled variable has covered 63.2% of the distance cess gain (K), time constant (τ (s)), dead time (td (s)), and the
between the pre-test steady state and the new one. The dead resulting controller gain (Kc) and integral time setting (Ti (min))
time determination by both fits is the same and was already of a PI controller.
described. In the following discussion, the process model deter-
Another method to determine the dead time is to measure mined by the second fit will be used to compare against a
the time when the PV moves by 2% of the total change. variety of tuning criteria.
170
Process variable
(50− 40)%
150 120 degC + 0.632 * = 4.5 degC = 100*(4.5/300)%
45 degC = 148 degC % %
140
= 1.5
130
120 155 s
33 s
400 800
50
Controller output
48
46
44
42
40
400 800
Time (sec)
FIG. 2.35h
Bump test, Fit 2 curve.
Comparing the Tuning Methods One of the earliest methods empirical equations to be used to predict controller settings to
for using the process reaction curve was proposed by Ziegler obtain a decay ratio of 1/4 are given in Table 2.35j in terms of
and Nichols. When using their process reaction curve method, K, td, and τ.
which was described in connection with Figure 2.35e, only Rr In developing their equations, Ziegler and Nichols con-
and td or t0 must be determined. Using these parameters, the sidered processes that were not “self-regulating.” To illustrate,
170
160
Process variable
150
140
130
120
400 800
50
48
Controller output
46
44
42
40
400 800
Time (sec)
FIG. 2.35i
A typical reaction curve using the dead time and time constant obtained by a bump test.
the level in the tank will rise to some new position but will
TABLE 2.35j not increase indefinitely, and therefore system will be self-
Ziegler–Nichols’ Recommendations to Obtain the Tuning 4
regulating. To account for self-regulation, Cohen and Coon
Parameters for an Interacting Controller Based on the Readings
introduced an index of self-regulation µ defined as:
Calculated from a Model
Type of Controller P (gain) I (minutes/repeat) D (minutes) µ = Rr Lr /K 2.35(5)
τ Note that this term can also be determined from the process
P — —
K ∗ td reaction curve. For processes originally considered by Ziegler
and Nichols, µ equals zero and therefore there is no self-
τ regulation. To account for variations in µ, Cohen and Coon
PI 3.33 td —
0.9 *
K ∗ td suggested the equations given in Table 2.35l in terms of td and τ.
In case of a proportional control, the requirement that the
decay ratio be 1/4 is sufficient to ensure a unique solution, but
PID τ 2 td 0.5 td
1.2 for the case of proportional-plus-reset control, this restraint is
K ∗ td
not sufficient.
Another constraint in addition to the 1/4 decay ratio can
consider the level control of a tank with a constant rate of be placed on the response to determine the unique values of
liquid outflow. Assume that the tank is initially operating at Kc and Ti . This second constraint can be to require that the
constant level. If a step change is made in the inlet liquid flow, control area of the response be at its minimum, meaning that
the level in the tank will rise until it overflows. This process the area between the response curve and the set point be the
is not “self-regulating.” smallest. This area is called the error integral or the integral
On the other hand, if the outlet valve opening and outlet of the error with respect to time.
backpressure are constant, the rate of liquid removal With the proportional-plus-reset-plus-rate controller
increases as the liquid level increases. Hence, in this case, (PID), the same problem of not having a unique solution
exists even when the 1/4 decay ratio and the minimum error
TABLE 2.35k integral constraints are applied. Therefore, a third constraint
The Tuning Setting Recommendations for a PI Controller must be chosen to obtain a unique solution. Based on the
Resulting from the Three Methods of Testing Described work of Cohen and Coon, it has been suggested that this new
constraint could have a value of 0.5 for the dimensionless
τ (s) td (s) Kc Ti (min)
Testing Method Used K
group Kc Ktd /τ. The tuning relations that will result from
Reaction curve 3.91 1.28 applying these three constraints are given in Table 2.35l. This
5–7
Process model Fit1, ZN 1.5 150 23 3.91 1.28 method has been referred to as the 3C method.
Greg Shinskey suggested a variation to the above, where
Process model Fit 2, ZN 1.5 122 23 3.18 1.28
the proportional gain and the integral time are increased.
TABLE 2.35l
Comparison of Equations Recommended by Ziegler–Nichols, Shinskey, Cohen–Coon, and
3C for the Determination of the Tuning Settings for PID Controllers
Ziegler–Nichols Shinskey Cohen–Coon 3C
P KKc = (td /τ ) −1.0 (td /τ ) −1.0 (td /τ ) −1.0 + 0.333 1.208(td /τ ) −0.956
P KKc = 0.9(td /τ ) −1.0 0.95(td /τ ) −1.0 0.9(td /τ ) −1.0 + 0.082 0.928(td /τ ) −0.946
I
Ti
= 3.33(td /τ ) 4.0(td /τ ) 3.33(td /τ )[1 + (td /τ )/11] 0.928(td /τ )−0.583
τ 1 + 2.2 + (td /τ )
P KKc = 1.2(td /τ ) −1.0 0.855(td /τ ) −1.0 1.35(td /τ ) −1.0 + 0.270 1.370(td /τ ) −0.950
Td 0.37(td /τ )
D = 0.5(td /τ ) 0.6(td /τ ) 0.365(td /τ )0.950
τ 1 + 0.2(td /τ )
8
Integral Criteria Tuning
TABLE 2.35m
Table 2.35m provides the controller settings that minimize Tuning Settings for Load and Set Point Disturbances
the respective integral criteria to the ratio td/τ. The settings Load Change Set Point Change
differ if tuning is based on load (disturbance) changes as
A B A B
opposed to set point changes. Settings based on load changes
IAE P 0.902 –0.985
will generally be much tighter than those based on set point
changes. When loops tuned to load changes are subjected to P 0.984 –0.986 0.758 –0.861
a set point change, a more oscillatory response is observed. I 0.608 –0.707 1.020 –0.323
P 1.435 –0.921 1.086 –0.869
I 0.878 –0.749 0.740 –0.130
Which Disturbance to Tune for D 0.482 1.137 0.348 0.914
With tuning parameters calculated for load rejection, the ITAE P 0.490 –1.084
integral time (Ti) and derivative time (Td) will depend mostly P 0.859 –0.977 0.586 –0.916
on the dead time (td) of the process. I 0.674 –0.680 1.030 –0.165
In contrast, if the tuning parameters are calculated for a P 1.357 –0.947 0.965 –0.855
set-point change, the integral time will be longer and the I 0.842 –0.738 0.796 –0.147
derivative time will be shorter, and they will depend mostly D 0.381 0.995 0.308 0.929
on the time constant of the process. ISE P 1.411 –0.917
The relationship between the controller settings based on P 1.305 –0.959
integral criteria and the ratio t0 /τ is expressed by the tuning I 0.492 –0.739
relationship given in Equation 2.35(6).
P 1.495 –0.945
I 1.101 –0.771
B
t D 0.560 1.006
Y = A 0 2.35(6)
τ ZN P 1.000 –1.000
P 0.900 –1.000
I 0.333 –1.000
where Y = KKc for proportional mode, τ /Ti for reset mode,
P 1.200 –1.000
Td/τ for rate mode; A, B = constant for given controller and
mode; t0, τ = pure delay time and first-order lag time constant. I 0.500 –1.000
Hence, using these equations, D 0.500 1.000
CCC P 1.208 –0.956
B P 0.928 –0.946
A t0
Kc = 2.35(7) I 1.078 –0.583
K τ P 1.370 –0.950
B I 1.351 –0.738
1 A t0
= 2.35(8) D 0.365 0.950
Ti τ τ Shinskey P 1.000 –1.000
B P 0.952 –1.000
t
Td = τ ∗ A 0 2.35(9) I 0.250 –1.000
τ P 0.855 –1.000
I 0.625 –1.000
Lambda Tuning D 0.600 1.000
4 to 1 decay P 1.235 –0.924
Lambda tuning originated from Dahlin in 1968; it is based Critical damping P 0.300 –1.000 0.300 –1.000
4,5
on the same IMC theory as MPC, is model-based, and uses (no overshoot, P 0.600 –1.000 0.350 –1.000
a model inverse and pole-zero cancellation to achieve the maximum speed) I 0.250 –1.000 Ti =1.16τ
desired closed-loop performance.
P 0.950 –1.000 0.600 –1.000
Lambda tuning is a method to tune loops based on pole
I 0.420 –1.000 Ti = τ
placement. This method ensures a defined response after a
set-point change but is generally too sluggish to properly D 0.420 1.000 0.500 1.000
reject disturbances. Promoters for this method often claim
that all loops should be tuned on the basis of Lambda tuning.
Doing so, the controllers are almost in “idling mode” and
where λ = closed loop time constant; it is recommended to Adjusting Robustness To remove oscillations in a control
use λ = 3τ. loop, hence to increase the robustness, it is necessary to give
165
IAE SP
155
145 ExperTune SP
135
ExperTune Load
125 IAE
ZN
115 CCC Shinskey
200 400 600 800 1000
45
Controller output
35
25
15
5
200 400 600 800 1000
Time (sec)
FIG. 2.35o
Load responses of the different tuning techniques (example).
Process variable
ExperTune Load
150
140 )
IAE (SPLoad Lambda
xp er Tu ne
130 E
120
200 400 600 800 1000
80
Controller output
70
60
50
40
30
200 400 600 800 1000
Time (sec)
FIG. 2.35p
Set-point responses of the different tuning techniques (example).
160
CCC
IAE (L)
150
Process variable
ZN
Shinskey
140
IAE (SP)
130 Lambda
120
56
52
48
44
40
200 400 600 800 1000
Time (sec)
FIG. 2.35q
Set-point responses of the different tuning techniques (example) with a controller where the P is applied only on PV changes.
previously may be used, considering that the dead time used Output
is the sum of the true process dead time and one-half of the
sampling time, as expressed by Equation 2.35(12):
t0′ = t0 + T /2 2.35(12)
Curve A
where T is the sampling time. t′0 is used in the tuning rela- Curve B
tionships instead of t0. (Section 2.38 deals with the subject Curve C
of controller tuning by computer.)
Process variable
be used to illustrate the “ultimate” closed-loop tuning 145
method. The aim of this tuning process is to determine the
Pu
controller gain or proportional band that would cause sus-
125
tained, undampened oscillation (Ku) and to measure the cor-
responding period of oscillation, called the ultimate period 300 600
(Pu). The steps in this tuning sequence are as follows:
Controller output
55
1. Set all controller dynamics to zero. In other words, set
the integral to infinite (or maximum) minutes per
35
repeat or zero (or minimum) repeats per minute and
set derivative to zero (or minimum) minutes. 300 600
2. Set the gain or proportional band to some arbitrary Time (sec)
value near the expected setting (if known) or at Kc = 1
(PB = 100%) if no better information is available. FIG. 2.35t
3. Let the process stabilize. Once the PV is stable, intro- Ultimate cycling response of the same process that was tested in
duce an upset. The simplest way to do that is to move Figure 2.35e.
the set point up or down by a safe amount (for exam-
ple, move it by 2% for half a minute) and then return In order to use the ultimate gain and the ultimate period
it to its original value. to obtain the controller settings for proportional controllers,
Ziegler and Nichols correlated the decay ratio vs. gain
The result will be an upset in the PV resembling the expressed as a fraction of the ultimate gain for several sys-
characteristics of curve A, B, or C in Figure 2.35r. If the tems. From the results they concluded that if the controller
response is undampened (curve A), the gain (or proportional) gain is set to equal one-half of the ultimate gain, it will often
setting is too high (proportional narrow); inversely, if the give a decay ratio of 1/4, i.e.,
response is damped (curve C), the gain (or proportional)
Kc = 0.5Ku (PB = 2PBu) 2.35(13)
setting is too low (proportional wide). Therefore, if the
response resembles curve A, the controller gain is increased; By analogous reasoning and testing, the equations in
if it resembles curve C, the gain is reduced, and the test is Table 2.35s were found to also give reasonably good settings
repeated until curve B is obtained. for noninteracting two- and three-mode controllers. Again it
After one or more trials, the state of sustained, undamp- should be noted that these equations are empirical and excep-
ened oscillation will be obtained (curve B), and at that point tions abound. For the same example as before, Figure 2.35t
the test is finished. (Make sure that the oscillation is a sinu- illustrates the ultimate cycling response.
soidal and not a limit cycle.) Next, read the proportional gain The ultimate gain and period obtained from Figure 2.35t
that caused the sustained oscillation. This is called the “ulti- are: Ku = 7.75 and Pu = 87 s.
mate gain” (Ku), and the corresponding period is the ultimate
Hence the recommended tuning settings for the process
period of oscillation (Pu).
that was used in the example are: Kp = 3.49 and Ti = 1.21
Once the values of Ku and Pu are known, one might use minutes.
the recommendations of Ziegler–Nichols (Table 2.35s) or the
There are a few exceptions to the tuning procedure
recommendations that were described earlier, which also con-
described here because in some cases, decreasing the gain
sider the dead time-to-time constant ratio. No one tuning is
makes the process more unstable. In these cases, the “ultimate”
perfect, and experienced process control engineers do come
method will not give good settings. Usually in cases of this
up with their own “fudge factors” based on experience.
type, the system is stable at high and low values of gain but
unstable at intermediate values. Thus, the ultimate gain for
systems of this type has a different meaning. To use the ulti-
mate method for these cases, the lower value of the ultimate
TABLE 2.35s gain is sought.
Tuning Parameters Based on the Measurement of Ku and Pu
Recommended by Ziegler–Nichols for a Noninteracting
Advantages and Disadvantages The main advantage of the
Controller
closed-loop tuning method is that it considers the dynamics
Type of Controller P (gain) I (minutes/repeat) D (minutes) of all system components and therefore gives accurate results
P 0.5 Ku — —
at the load where the test is performed. Another advantage
is that the readings of Ku and Pu are easy to read and the
PI 0.45 Ku Pu /1.2 —
period of oscillation can be accurately read even if the mea-
PID 0.6 Ku Pu /2 Pu /8 surement is noisy.
P Kc1/4 — —
Advantages and Disadvantages In general, there are two
major disadvantages to the ultimate and damped oscillation
PID adjusted P/1.5 P/6
methods. First, both are essentially trial-and-error methods,
since several values of gain must be tested before the ultimate
The disadvantages of the closed-loop tuning method are gain or the gain to give a 1/4 decay ratio are to be determined.
that when tuning unknown processes, the amplitudes of To make one test, especially at values near the desired gain,
undampened oscillations can become excessive (unsafe) and it is often necessary to wait for the completion of several
the test can take a long time to perform. One can see that when oscillations before it can be determined whether the trial
tuning a slow process (period of oscillation of over an hour), value of gain is the desired one.
it can take a long time before a state of sustained, undampened Second, while one loop is being tested in this manner, its
oscillation is achieved through this trial-and-error technique. output may affect several other loops, thus possibly upsetting
For these reasons, other tuning techniques have also been an entire unit. While all tuning methods require that some
developed and some of them are described below. changes be made in the control loop, other techniques require
only one and not several tests, unlike the closed-loop methods.
Damped Oscillation Method Also, if the tuning parameters are too aggressive, the
expected response can be obtained by increasing the propor-
Harriott has proposed a slight modification of the previous tional band (or decreasing the proportional gain). The integral
procedure. For some processes, it is not feasible to allow and derivative settings probably need to be modified. The
sustained oscillations and therefore, the ultimate method can- proportional gain has to be reduced to 3.5 to have a quarter-
not be used. In this modification of the ultimate method, the of-amplitude decay.
gain (proportional control only) is adjusted, using steps anal-
ogous to those used in the ultimate method, until a response
curve with 1/4 of the decay ratio is obtained. However, with
COMPARISON OF CLOSED AND OPEN LOOP
this tuning method, it is necessary to note only the period P
of the response.
Table 2.35w provides a comparison of open-loop and closed-
Again it should be noted that the equations in Table 2.35u
loop results for the process example used in Figure 2.35e.
are empirical and exceptions abound.
150
1. FRM require only one process bump to identify the pro-
140
cess. The bump can be a change in automatic or manual,
130
120
300 600
TABLE 2.35w
80
Controller output
and be either a pulse, step, or other type of bump. A set- of the output when aligned with the crest of the input is
point change provides excellent data from FRM. generally thought to be 180 degrees out of phase or 180
2. FRM do not require any prior knowledge of the pro- degrees of phase lag.
cess dead time or time constant. With the other time By applying a variety of sine wave inputs to a process,
response methods, one often needs a dead time esti- one can obtain a table of amplitude ratios and phase lags
mate and a time constant estimate. dependent on the sine wave frequency. If one plots these, the
3. FRM do not require any prior knowledge of the pro- result will be the frequency response of the process.
cess structure. Time response methods often require Using Fourier analysis computer software programs one
the user to have such model structure knowledge, i.e., can calculate the process frequency response from a bump,
whether it is first or second order or whether it is an pulse, or any other signal that applies sufficient excitation to
integrator. For FRM-based tuning none of this is the controller output. In the evaluation both the CO and PV
required; only the process data are needed. trends are used. The data provided for these programs should
start from a settled state, experience a quick change, and end
Obtaining the Frequency Response settled.
Any one of the responses in Figures 2.35e, g, h, i, o, p, q,
The process frequency response is a graph of amplitude ratio
and v would provide adequate data for frequency response–
and phase vs. oscillation or sine wave frequency. If one injects
based testing. Figure 2.35x shows the typical frequency
a sine wave into a linear process at the controller output, the
response arrived at by the use of computer software.
PV will also display a sine wave. The output (PV) sine wave
will probably be of smaller height relative to the input and
will be shifted in time. PID Tuning Based on Frequency Response
The ratio of the heights is the amplitude ratio at the
frequency of the input sine wave. The shift in time is the In most processes, both the amplitude ratio and the phase
phase shift or phase lag. A time shift resulting in the trough angle will decrease with increasing frequencies. Assuming
Amplitude ratio (dB)
−1.25
−23.75
0
Phase angle (deg)
−180
−360
FIG. 2.35x
Typical model (solid line) and actual (dash line) process frequency response.
Deviation
frequency response has an amplitude ratio that is larger than
one when the phase lag is 180 degrees. lopt × 1.5
To provide proper tuning, a margin of safety in the gain 0
and phase is desired. Tuning constants are therefore adjusted lopt
to result in the highest gain at all frequencies and yet achieve lopt ÷ 1.5
a certain margin of safety or stability. This is best accom-
plished using computer software. 0 1 2 3 4
Time, t/St
FIG. 2.35z
FINE TUNING The integral setting primarily affects the location of the second peak.
1
The integral time will need correcting regardless of the
2 controller used—it must be multiplied by a factor repre-
0 sented by the top curve for a PID controller and the middle
curve for a PI controller. If the controller is a noninteracting
(ideal, parallel) PID, then its proportional band should be
0 1 2 3 4 multiplied by the correction factor, which is indicated by the
Time, t/Σt dashed line. PI and interacting (series) PID controllers do not
require a correction to the proportional setting, and no cor-
FIG. 2.35aa rection to the derivative setting is required for any controller.
Load dynamics affect the shape of the response curve.
Symbols, Abbreviations
passes through different dynamics than in the controller-output
path, the resulting response curve is no longer minimum-IAE. CO Controller output
As might be expected, faster dynamics in the load path FRM Frequency response method
produce a faster rising deviation, which therefore peaks at a Gp Process gain
higher value before the controller can overcome it. The larger IAE Integral of absolute error
peak is then followed by a large overshoot. By contrast, IE Integral error
slower dynamics in the load path reduce the peak deviation, ISE Integral of squared error
followed by an undershoot. The IAE for the curve where ITAE Integral of absolute error × time
Στq /Στm = 0.5 is actually only 6% higher than the minimum, ITSE Integral of squared error × time
but for the case where Στq /Στ m = 2, the IAE is actually 53% Kp Proportional gain for a PID controller
above the minimum, indicating substantial need for correc- Ku Ultimate controller gain
tion. The appearance of the exaggerated overshoot and the PV Process variable or measurement
undershoot indicate that those responses are no longer min- Rr Reaction rate, slope
imum-IAE, and that the integral time needs adjusting. Retun- RRT Relative response time, the time to remove most
ing the controllers for minimum-IAE response in the pres- of a disturbance
ence of different load dynamics resulted in the development SP Set point
of the correction factors plotted in Figure 2.3bb. (Tau) Process time constant (seconds)
10
PID
Integral corr.
PI
1 Prop. corr.
Nonint. PID
0.1
0.1 1 10
Σtq /Σtm
FIG. 2.35bb
To obtain minimum-IAE response, integral time and possibly the proportional band may require correction for load dynamics.
t0(td) Process dead time (seconds) Rovira, A.A., Murrill, P.W., and Smith, C.L., Tuning Controllers for Set-Point
Ti Integral time (s) -for a PID controller Changes, Instruments and Control Systems, December 1969.
Ruel, M., Loop Optimization: before You Tune, Control Magazine, Vol. 12,
Td Derivative time (s) for a PID controller No. 03 (March 1999), pp. 63–67.
F PV filter time constant Ruel, M., Loop Optimization: Troubleshooting, Control Magazine, Vol. 12,
tu Ultimate period No. 04, (April 1999), pp. 64–69.
Ruel, M., Loop Optimization: How to Tune a Loop, Control Magazine, Vol. 12,
No. 05 (May 1999), pp. 83–86.
Ruel, M., Instrument Engineers' Handbook, chap. 5.9, Plantwide Control
Bibliography Loop Optimization, 3rd ed., Bela G. Liptak, Ed., Boca Raton, FL:
CRC Press, 2002, 40 pp.
Astrom, K.J., and Hagglund, T., Automatic Tuning of PID Controllers,
Shinskey, F.G., Process Control Systems, 4th ed., New York: McGraw-Hill,
Research Triangle Park, NC: ISA Press, 1988.
1996.
Bjornsen, B.G., A Fluid Amplifier Pneumatic Controller, Control Engineering,
Shinskey, F.G., Process Control Systems, 4th ed., New York: McGraw-Hill,
June 1965.
1996.
Box, G.E.P., and Jenkins, G.M., Time Series Forecasting and Control, San
Shinskey, F.G., Interaction between Control Loops, Instruments and Control
Francisco: Holden-Day, 1970.
Systems, May and June 1976.
Caldwell, W.I., Coon, G.A., and Zoss, L.M., Frequency Response for Process
Salmon, D.M., and Kokotovic, P.V., Design of Feedback Controllers for Non-
Control, New York: McGraw-Hill, 1959.
linear Plants, IEEE Trans. on Automatic Control, Vol. AC 14, No. 3,
Campbell, D.P., Process Dynamics, New York: John Wiley & Sons, 1958. 1969,
Chen, C.T., Introduction to Linear System Theory, New York: Holt, Rinehart Sastry, S., and Bodson, M., Adaptive Control: Stability, Convergence and
and Winston, 1970. Robustness, Englewood Cliffs, NJ: Prentice Hall, 1989.
Corripio, A.B., Tuning of Industrial Control Systems, Research Triangle Seborg, D., Edgar, E., and Mellichamp, D., Process Control, New York:
Park, NC: ISA Press, 1990. Wiley, 1990.
Gerry, J.P., How to Control Processes with Large Dead Times, Control Seborg, D.E., Edgar, T.F., and Shah, S.L., Adaptive Control Strategies for
Engineering, March 1998. Process Control, AI Chem. E. Journal, 32, 1986, pp. 881–913.
Gerry, J.P., PID settings sometimes fail to make the leap, Intech, November Smith, C.A., and Corripio, A.B., Principles and Practice of Automatic Process
1999. Control, New York: John Wiley & Sons, 1985.
Gerry, J.P., Tune Loops for Load Upsets VS Setpoint Changes, CONTROL Smith, C.L., Controller Tuning That Works, Instruments and Control Systems,
Magazine, September 1991. September 1976.
Higham, J.D., Single-Term Control of First- and Second-Order Processes with Smith, C.L., and Murrill, P.W., A More Precise Method for the Tuning of
Dead Time, Control, February 1968. Controllers, ISA Journal, May 1996.
Hougen, J.O., Experiences and Experiments with Process Dynamics, Chemical Smith, C.L., Controllers—Set Them Right, Hydrocarbon Processing and
Engineering Process Monograph Series, Vol. 60, No. 4, 1964. Petroleum Refiner, February 1966.
Howell, J., Rost, D., and Gerry, J.P., PC Software Tunes Plant Startup, InTech, Smith, O.J.M., Close Centrol of Loops with Dead Time, Chemical Engineering
June 1988. Progress, May 1957.
Kraus, T.W., and Myron, T.J., Self-Tuning PID Controller Uses Pattern Sood, M., Tuning Proportional-Integral Controllers for Random Load
Recognition Approach, Control Engineering, June 1984. Changes, M.S. thesis, Department of Chemical Engineering, University
Larson, R.E., et al., State Estimation in Power Systems, Part I and Part II, of Mississippi, University, MS, 1972.
IEEE Trans. on Power Apparatus and Systems, Vol. PAS 89, No. 3, Sood, M., and Huddleston, H.T., Optimal Control Settings for Random Dis-
March 1970. turbances, Instrumentation Technology, March 1973.
Lelic, M.A., and Zarrop, M.B., Generalized Pole-Placement Self-Tuning Stephanopoulos, G., Chemical Process Control: An Introduction to Theory
Controller, Int. J. Control, 46, 1987, pp. 569–607. and Practice, Englewood Cliffs, NJ: Prentice Hall, 1984.
Lipták, B.G., How to Set Process Controllers, Chemical Engineering, Suchanti, N., Tuning Controllers for Interacting Processes, Instruments and
November 1964. Control Systems, April 1973.
Lopez, A.M., et al., Controller Tuning Relationships Based on Instrumentation Wade, H.L., High-Capability Single-Station Controllers: A Survey, InTech,
Technology, Control Engineering, November 1967. September 1988.
Luyben, W.L., Process Modeling, Simulation and Control for Chemical Wells, C.H., Application of Modern Estimation and Identification Techniques
Engineers, 2nd edition, New York: McGraw-Hill, 1990. to Chemical Processes, AIChE Journal, 1971.
Mehra, R.K., Identification of Stochastic Linear Dynamic Systems Using Wells, C.H., and Mehra, R.K., Dynamic Modeling and Estimation of Carbon
Kalman Filter Representation, AIAA Journal, January 1971. in a Basic Oxygen Furnace, paper presented at 3rd Int. IFAC/IFIPS
Mehra, R.K., On-Line Identification of Linear Dynamic Systems with Applica- Conf. on the Use of Digital Computers in Process Control, Helsinki,
tions to Kalman Filtering, IEEE Trans. on Automatic Control, Vol. AC 16, Finland, June 1971.
No. 1, February 1971. Wellstead, P.E., and Zarrop, M.B., Self-Tuning Systems, New York: Wiley,
Miller, J.A., et al., A Comparison of Open-Loop Techniques for Tuning Con- 1991.
trollers, Control Engineering, December 1967. Widrow, B., and Stearns, S.D., Adaptive Signal Processing, Englewood
Murrill, P.W., Automatic Control of Processes, Scranton, PA: International Cliffs, NJ: Prentice Hall, 1985.
Textbook Co., 1967. Ziegler, J.G., and Nichols, N.B., Optimal Settings for Automatic Controllers,
Raven, F., Automatic Control Engineering, New York: McGraw-Hill, 1961. ASME Trans., November 1942.
Rovira, A.A., Ph.D. dissertation, Louisiana State University, Baton Rouge, Ziegler, J.G., and Nichols, N.B., Optimal Settings for Controllers, ISA Journal,
LA, 1969. August 1964.