PID Control

Control System
System
Analysis and Design
PROBLEMS, REMEDIES, AND FUTURE DIRECTIONS

By YUN LI, KIAM HEONG ANG, and GREGORY C.Y. CHONG

© IMAGESTATE

W
ith its three-term functionality offering treatment of both transient and steady-state responses,
proportional-integral-derivative (PID) control provides a generic and efficient solution to real-
world control problems [1]–[4]. The wide application of PID control has stimulated and sus-
tained research and development to “get the best out of PID’’ [5], and “the search is on to find
the next key technology or methodology for PID tuning” [6].
This article presents remedies for problems involving the integral and derivative terms. PID design objec-
tives, methods, and future directions are discussed. Subsequently, a computerized, simulation-based approach
is presented, together with illustrative design results for first-order, higher order, and nonlinear plants. Finally,
we discuss differences between academic research and industrial practice, so as to motivate new research
directions in PID control.

STANDARD STRUCTURES OF PID CONTROLLERS

Parallel Structure and Three-Term Functionality

The transfer function of a PID controller is often expressed in the ideal form

32 IEEE CONTROL SYSTEMS MAGAZINE » FEBRUARY 2006 1066-033X/06/$20.00©2006IEEE

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.


U(s) 1 where GPD (s) and GPI (s) are the factored PD and PI parts of
GPID (s) = = KP 1 + + TD s , (1)
E(s) TI s the PID controller, respectively, and

√
where U(s) is the control signal acting on the error signal E(s), 1± 1 − 4TD /TI
α= > 0.
KP is the proportional gain, TI is the integral time constant, TD 2
is the derivative time constant, and s is the argument of the
Laplace transform. The control signal can also be expressed in THE INTEGRAL TERM
three terms as
Destabilizing Effect of the Integral Term
1 Referring to (1) for TI = 0 and TD = 0, it can be seen that
U(s) = KP E(s) + KI E(s) + KD sE(s)
s adding an integral term to a pure proportional term increases
= UP (s) + UI (s) + UD (s), (2)
the gain by a factor of

where KI = KP /TI is the integral gain and KD = KP TD is the   

 1  1
derivative gain. The three-term functionalities include: 1 + = 1 + 2 2 > 1, for all ω, (5)
 jωTI  ω TI
1) The proportional term provides an overall control action
proportional to the error signal through the allpass gain
factor. and simultaneously increases the phase-lag since
2) The integral term reduces steady-state errors through



low-frequency compensation. 1 −1
 1+ = tan−1 < 0, for all ω. (6)
3) The derivative term improves transient response through jωTI ωTI
high-frequency compensation.
A PID controller can be considered as an extreme form of a Hence, both gain margin (GM) and phase margin (PM) are
phase lead-lag compensator with one pole at the origin and reduced, and the closed-loop system becomes more oscillatory
the other at infinity. Similarly, its cousins, the PI and the PD and potentially unstable.
controllers, can also be regarded as extreme forms of phase-
lag and phase-lead compensators, respectively. However, the Integrator Windup and Remedies
message that the derivative term improves transient response If the actuator that realizes the control action has saturated
and stability is often wrongly expounded. Practitioners have range limits, and the saturations are neglected in a linear con-
found that the derivative term can degrade stability when trol design, the integrator may suffer from windup; this caus-
there exists a transport delay [4], [7]. Frustration in tuning KD es low-frequency oscillations and leads to instability. The
has thus made many practitioners switch off the derivative windup is due to the controller states becoming inconsistent
term. This matter has now reached a point that requires clarifi- with the saturated control signal, and future correction is
cation, as discussed in this article. For optimum performance, ignored until the actuator desaturates.
KP, KI (or TI ), and KD (or TD) must be tuned jointly, although
the individual effects of these three parameters on the closed- Automatic Reset
loop performance of stable plants are summarized in Table 1. If TI ≥ 4TD so that the series form (3) exists, antiwindup can be
achieved implicitly through automatic reset. The factored PI
The Series Structure part of (3) is thus implemented as shown in Figure 1 [8], [9].
If TI ≥ 4TD, the PID controller can also be realized in a series
form [7] Explicit Antiwindup


1 In nearly all commercial PID software packages and hardware
GPID (s) = (α + TD s) KP 1 + (3)
αTI s modules, however, antiwindup is implemented explicitly
= GPD (s)GPI (s), (4) through internal negative feedback, reducing UI (s) to [8]–[10]

TABLE 1 Effects of independent P, I, and D tuning on closed-loop response.

For example, while K I and K D are fixed, increasing K P alone can decrease rise time,
increase overshoot, slightly increase settling time, decrease the steady-state error, and decrease stability margins.

Rise Time Overshoot Settling Time Steady-State Error Stability

Increasing K P Decrease Increase Small Increase Decrease Degrade
Increasing K I Small Decrease Increase Increase Large Decrease Degrade
Increasing K D Small Decrease Decrease Decrease Minor Change Improve

FEBRUARY 2006 « IEEE CONTROL SYSTEMS MAGAZINE 33

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 which tends to increase the PM. In the meantime, however,
Actuator Model the gain increases by a factor of
UPD(s) U(s) U(s)
KP +
+   
1 + jωTD = 1 + ω2 T2 > 1, for all ω, (9)
D

1 and hence the overall stability may be improved or degraded.

1 + α TIs To demonstrate that adding a differentiator can destabilize
some systems, consider the typical first-order delayed plant
FIGURE 1 The PI part of a PID controller in series form for automatic K −Ls
reset. The PI part of a PD-PI factored PID transfer function can be G(s) = e , (10)
1 + Ts
configured to counter actuator saturation without the need for sepa-
rate antiwindup action. Here, UPD (s) is the control signal from the where K is the process gain, T is the time constant, and L is the
preceding PD section. When U(s) does not saturate, the feedfor-
dead time or transport delay. Suppose that this plant is con-
ward-path gain is unity and the overall transfer function from UPD (s)
to Ū(s) is thus 1 + 1/(αTI s), the same as the last factor of (3). trolled by a proportional controller with gain KP and that a
derivative term is added. The resulting PD controller


1 U(s) − Ū(s)
ŨI (s) = KP E(s) − , (7) GPD (s) = KP (1+TD s) (11)
TI s γ

leads to an open-loop feedforward path transfer function with

where U(s) is the theoretically computed control signal, Ū(s) frequency response
is the actual control signal capped by the actuator limits, and
1 + jTD ω − jLω
γ is a correcting factor. A value of γ in the range [0.1, 1.0] usu- G( jω)GPD ( jω) = KKP e . (12)
1 + jTω
ally results in satisfactory performance when the PID coeffi-
cients are reasonably tuned [7]. For all ω, the gain satisfies


Accounting for Windup in Design Simulations 1 + TD2 ω2 TD
KKP
KK P min 1, , (13)
Another solution to antiwindup is to reduce the possibilities 1 + T2 ω 2 T
for saturation by reducing the control signal, as in linear
1/2
quadratic optimal control schemes that minimize the tracking where inequality (13) holds since ((1 + TD2 ω2)/(1 + T2 ω2 )) is
error and control signal through a weighted objective func- monotonic in ω.
tion. However, preemptive minimization of the control signal Hence, if KP > 1/K and TD > T/K KP , then, for all ω,
can impede performance due to minimal control amplitude.
 
Therefore, during the design evaluation and optimization G( jω)GPD ( jω) > 1. (14)
process, the control signal should not be minimized but
rather capped at the actuator limits when the input to the Inequality (14) implies that the 0-dB gain crossover frequency is
plant saturates since a simulation-based optimization process at infinity. Furthermore, due to the transport delay, the phase is
can automatically account for windup that might occur.

G( jω)GPD ( jω) = tan−1 ωT
1
D − tan−1 Tω − Lω.
1
THE DERIVATIVE TERM
Therefore, when ω approaches infinity,
Stabilizing and Destabilizing Effect of the Derivative Term
Derivative action is useful for providing phase lead, which

G( jω)GPD ( jω) < −180◦ . (15)
offsets phase lag caused by integration. This action is also
helpful in hastening loop recovery from disturbances. Deriva-
tive action can have a more dramatic effect on second-order Hence, if TD > T/KKP and KP > 1/K, then by the Nyquist crite-
plants than first-order plants [9]. rion, the closed-loop system is unstable. This analysis also con-
However, the derivative term is often misunderstood and firms that some PID mapping formulas, such as the
misused. For example, it has been widely perceived in the Ziegler-Nichol (Z-N) formula obtained from the step-response
control community that the derivative term improves tran- method, in which KP = (1.2(T/L)) (1/K) and TD is proportional
sient performance and stability. But this perception is not to L, are valid for only a limited range of values of the T/L ratio.
always valid. To see this, note that adding a derivative term to As an example, consider plant (10) with K = 10, T = 1 s, and
a pure proportional term reduces the phase lag by L = 0.1 s [7]. Control by means of a PI controller with KP =
0.644 > 1/K and TI = 1.03 s yields reasonable stability margins
  ωTD and time-domain performance, as seen in Figures 2 and 3 (Set 1,

1 + jωTD = tan−1 ∈ [0, π/2] for all ω, (8)
1 red curves). However, when a differentiator is added, gradually

34 IEEE CONTROL SYSTEMS MAGAZINE » FEBRUARY 2006

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 increasing TD from zero improves both GM and PM. The GM Linear Lowpass Filter
peaks when TD approaches 0.03 s; this value of TD maximizes The filtering remedy most commonly adopted is to cascade
the speed of the transient response without oscillation. Howev- the differentiator with a first-order, lowpass filter, a technique
er, if TD is increased further to 0.1 s, the GM deteriorates and often used in preprocessing for data acquisition. Hence, the
the transient exhibits oscillation. In fact, the closed-loop system derivative term becomes
can be destabilized if TD increases to 0.2 with T/KKP = 0.155.
Hence, care needs to be taken to tune and use the derivative TD s
G̃D (s) = KP , (16)
term properly when the plant is subject to delay. 1+ TD
β
s
This destabilizing phenomenon can contribute to difficul-
ties in designing PID controllers. These difficulties help explain where β is a constant factor. Most industrial PID hardware
why 80% of PID controllers in use have the derivative part provides a value ranging from 1 to 33, with the majority
switched off or omitted completely [5]. Thus, the functionality falling between 8 and 16 [12]. A second-order Butterworth
and potential of a PID controller is not fully exploited, while filter is recommended in [13] if further attenuation of high-
proper use of a derivative term can increase stability and help frequency gains is required. Sometimes, the lowpass filter is
maximize the integral gain for better performance [11]. cascaded to the entire GPID in internal-model-control (IMC)-
based design, which therefore leads to more sluggish tran-
Remedies for Derivative Action sients.
Differentiation increases the high-frequency gain, as shown in
(9) and demonstrated by the four sets of frequency responses Velocity Feedback
in Figure 2. A pure differentiator is not proper or causal. Because a lowpass filter does not completely remove, but
When a step change of the setpoint or disturbance occurs, dif- rather averages, impulse derivative signals caused by sud-
ferentiation results in a theoretically infinite control signal. To den changes of the setpoint or disturbance, modifications of
prevent this impulse control signal, most PID software pack- the unity negative feedback PID structure are of interest [8].
ages and hardware modules add a filter to the differentiator. To block the effect of sudden changes of the setpoint, we
Filtering is particularly useful in a noisy environment. consider a variant of the standard feedback. This variant uses

PIDeasy(TM) II - Nichols Chart

Nichols Chart

Nichols Chart Set 1:

Set 1 Gain Margin: 7.75757
40 PIDeasy Phase Margin: 53.37157
Set 2
PIDeasy:
Set 3
30 Gain Margin: 9.16494
Phase Margin: 64.72558
Set 3: TD = 0.2
20 Set 2:
Gain (dB)

Gain Margin: 3.41323

Set 2: TD = 0.1 Phase Margin: 86.39067
10
Set 3:
Gain Margin: −2.26496
0 Phase Margin: – ∞

−10
PIDeasy: TD = 0.0303

−20
−3.5 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0
Phase (deg.)
Set 1: TD = 0 ×102

FIGURE 2 Destabilizing effect of the derivative term, measured in the frequency domain by GM and PM. Adding a derivative term increases
both the GM and PM, although raising the derivative gain further tends to reverse the GM and destabilize the closed-loop system. For
example, if the derivative gain is increased to 20% of the proportional gain (TD = 0.2 s), the overall open-loop gain becomes greater than
2.2 dB for all ω. At ω = 30 rad/s, the phase decreases to −π while the gain remains above 2.2 dB. Hence, by the Nyquist criterion, the
closed-loop system is unstable. It is interesting to note that MATLAB does not compute the frequency response as shown here, since MAT-
LAB handles the transport delay factor e−jωL in state space through a Padé approximation.

FEBRUARY 2006 « IEEE CONTROL SYSTEMS MAGAZINE 35

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 the process variable instead of the error signal for the deriva- tively using standard techniques of stability and robust-
tive action [14], as in ness analysis.

Structure (17) is referred to as Type B PID (or PI-D) control,
t
d structure (18) is known as Type C PID (or I-PD) control, and
u(t) = KP e(t) + KI e(τ ) dτ − KD y(t), (17)
0 dt structures (1)–(3) constitute Type A PID control. Types B and C
introduce more structures, and the need for preselection of, or
where y(t) is the process variable, e(t) = r(t) − y(t) is the error switching between, suitable structures can pose a design chal-
signal, and r(t) is the setpoint or reference signal. The last term lenge. To meet this need, PID hardware vendors have developed
of (17) forms velocity feedback and, hence, an extra loop that artificial intelligence techniques to suppress overshoots [15], [16].
is not directly affected by a sudden change in the setpoint. Nevertheless, the ideal, parallel, series, and modified PID con-
However, sudden changes in disturbance or noise at the plant troller structures can be found in many software packages and
output can cause the differentiator to produce a theoretically hardware modules. Techmation’s Applications Manual [12] docu-
infinite control signal. ments the structures employed in many industrial PID con-
trollers. Since vendors often recommend their own controller
Setpoint Filter structures, tuning rules for a specific structure do not necessarily
To further reduce sensitivity to setpoint changes and avoid perform well with other structures. Readers may refer to [17] and
overshoot, a setpoint filter may be adopted. To calculate the [18] for detailed discussions on the use of various PID structures.
proportional action, the setpoint signal is weighted by a factor
b < 1, as in [8] and [14] Prefilter
For setpoint tracking applications, an alternative to using a Type
  t d B or C structure is to cascade the setpoint with a prefilter that
u(t) = KP b r(t) − y(t) + KI 0 e(τ ) dτ − KD y(t). (18)
dt has critically damped dynamics. When a step change in the set-
point occurs, continuous output of the prefilter helps achieve
This modification results in a bumpless control signal and soft start and bumpless control [8], [19]. However, a prefilter
improved transients if the value of b is carefully chosen [2]. does not solve the problem caused by sudden changes in the
However, modification (18) is difficult to analyze quantita- disturbance since it is not embedded in the feedback loop.

PIDeasy(TM) II - Step and Control Signal Response Dialog

View Both Responses Sampling Rate: 0.00014 sec.

Step Response Set 1:

Kp: 0.6439
Step Response Plot Ti: 1.0278
Td: 0
1.4 Set 1
Set 1: TD = 0 ITAE: 228.93
PIDeasy
1.2 Set 2 PIDeasy:
Set 3 Kp: 0.6439
1.0 Ti: 1.02781
Td: 0.03025
ITAE: 104.86
Output

0.8
Set 2:
0.6 Kp: 0.6439
Set 2: TD = 0.1 Ti: 1.0278
0.4 Td: 0.1
ITAE: 247.69
0.2 Set 3:
Kp: 0.6439
0.0 Ti: 1.0278
Td: 0.2
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 ITAE: 52,547.83
Time (Sec.) Set 3: TD = 0.2
PIDeasy: TD = 0.0303

FIGURE 3 Destabilizing effect of the derivative term, confirmed in the time domain by the closed-loop step response. Although increasing the
derivative gain initially decreases the oscillation, this trend soon reverses and the oscillation grows into instability.

36 IEEE CONTROL SYSTEMS MAGAZINE » FEBRUARY 2006

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 Nonlinear Median Filter model and a targeted closed-loop transfer function with an
Another method for smoothing the derivative action is to use indirect performance objective, such as pole placement, IMC,
a median filter [7], which is nonlinear and widely applied in or lambda tuning. To derive a rational, closed-loop transfer
image processing. Such a filter compares several data points function, this method requires that transport delays be
around the current point and selects their median for the con- replaced by Padé approximations.
trol action. Consequently, unusual or unwanted spikes result-
ing from a step command, noise, or disturbance are removed
completely. Median filters are easily realized, as illustrated in derivative = (error -previous_error) / sampling_period;
Figure 4, since almost all PID controllers are now implemented if (derivative > max_d)
in a digital processor. Another benefit of this method is that new_derivative = max_d; // median found
extra parameters are not needed to devise the filter. A median else if (derivative < min_d)
filter outperforms a prefilter as the median filter is embedded new_derivative = min_d; // median found
else
in the feedback loop and, hence, can deal with sudden new_derivative=derivative; // median found
changes in both the setpoint and the disturbance; a median fil-
if (derivative > previous_derivative) { // for next cycle
ter may, however, overly smooth underdamped processes. max_d=derivative;
min_d=previous_derivative;
Design Objectives and Methods } else {
max_d=previous_derivative;
Design Objectives and Existing Methods min_d=derivative;
Excellent summaries of PID design and tuning methods can be }
previous_derivative = derivative; // for next cycle
found in [4], [8], [20], and [21]. While matters concerning com-
missioning and maintenance (such as pre- and postprocessing as FIGURE 4 Pseudocode for a three-point median filter to illustrate the
well as fault tolerance) also need to be considered in a complete mechanism of complete removal of impulse spikes. Median filters
PID design, controller parameters are usually tuned so that the are widely adopted in image processing but not yet in control sys-
closed-loop system meets the following five objectives: tem design. This nonlinear filter completely removes extraordinary
derivative values resulting from sudden changes in the error signal,
1) stability and stability robustness, usually measured in the
unlike a lowpass filter, which averages past values.
frequency domain
2) transient response, including rise time, overshoot, and set-
tling time TABLE 2 ABB’s Easy-Tune PID formulas mapping the three
3) steady-state accuracy parameters K, T, and L of the first-order delayed plant (10)
4) disturbance attenuation and robustness against environ- to coefficients of P, PI, PID, and PD controllers, respectively
mental uncertainty, often at steady state [23]. The formulas are obtained by minimizing the
integral of time-weighted error index, except the PD formula
5) robustness against plant modeling uncertainty, usually
for which empirical estimates are used. Usually expressed
measured in the frequency domain. in percentage, PB = (Umax − Umin )/K P is the proportional
Most methods target one objective or a weighted compos- band, where Umax and Umin are the upper and lower
ite of the objectives listed above. With a given objective, saturation levels of the control signal, respectively, and
design methods can be grouped according to their underlying |Umax − Umin | is usually normalized to one.
nature listed below [7], [8].
Mode Action Value

1.084
L
Heuristic Methods P PB 2.04K
T
Heuristic methods evolve from empirical tuning (such as the Z-
0.977
L
N tuning rule), often with a tradeoff among design objectives. PI PB 1.164K
T
Heuristic search now involves expert systems, fuzzy logic,
0.68
T L
neural networks, and evolutionary computation [19], [22]. TI
40.44 T

0.947
L
Frequency Response Methods PB 0.7369K
T
Frequency-domain constraints, such as GM, PM, and sensi-
0.738
T L
tivities, are used to synthesize PID controllers offline [2], PID TI
51.02 T
[3]. For real-time applications, frequency-domain measure-
0.995
T L
ments require time-frequency, localization-based methods TD
157.5 T
such as wavelets.
0.947
L
PD PB 0.5438K
T
Analytical Methods
0.995
T L
Because of the simplicity of PID control, parameters can be TD
157.5 T
derived analytically using algebraic relations between a plant

FEBRUARY 2006 « IEEE CONTROL SYSTEMS MAGAZINE 37

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 Numerical Optimization Methods time and frequency domains [19], [22].
Optimization-based methods can be regarded as a special type Some design methods can be computerized, so that designs
of optimal control. PID parameters are obtained by numerical are automatically performed online once the plant is identified;
optimization for a weighted objective in the time domain. hence, these designs are suitable for adaptive tuning. While
Alternatively, a self-learning evolutionary algorithm (EA) can PID design has progressed from analysis-based methods to
be used to search for both the parameters and their associated numerical optimization-based methods, there are few tech-
structure or to meet multiple design objectives in both the niques that are as widely applicable as Z-N tuning [2], [3]. The
most widely adopted initial tuning methods are based
on the Z-N empirical formulas and their extensions, such
15 as those shown in Table 2 [23]. These formulas offer a
Gain Margin (dB)

direct mapping from plant parameters to controller coef-

ficients.
10
Over the past half century, researchers have sought the
next key technology for PID tuning and modular realiza-
5 tion [6]. With simulation packages widely available and
10−3 10−2 10−1 100 101 102 103 heavily adopted, computerizing simulation-based designs
80 is gaining momentum, enabling simulations to be carried
Phase Margin (°)

out automatically so as to search for the best possible PID

70
settings for the application at hand [22]. By using a com-
puterized approach, multiple design methods can be com-
60
bined within a single software or firmware package to
50 support various plant types and PID structures.
10−3 10−2 10−1 100 101 102 103
L/T A Computerized Simulation Approach
PIDeasy [7] is a software package that uses automatic sim-
FIGURE 5 Gain and phase margins resulting from PIDeasy designs for first- ulations to search globally for controllers that meet all five
order delayed plants with various L/T ratios. While requirements of fast
transient response, no overshoot, and zero steady-state error are accom-
design objectives in both the time and frequency domains.
modated by time-domain criteria, multiobjective design goals provide The search is initially performed offline in a batch mode
frequency-domain margins in the range of 9–11 dB and 65–66◦ . [19] using artificial evolution techniques that evolve both

TABLE 3 Multioptimal PID settings for normalized typical high-order plants. Since PIDeasy’s search priorities are time-domain
tracking and regulation, the corresponding gain and phase margins are given to assess frequency-domain properties.

PID Coefficients Resultant Margins

Plants Kp Ti (s) Td (s) GM (dB) PM (◦ )

α =1 92.1 1.0 0.0022 ∞ 102

α =2 1.95 1.61 0.14 ∞ 62.4
1
G1 (s) = α =3 1.12 2.13 0.28 26.8 60.7
(s + 1)α
α =4 0.83 2.61 0.43 13.9 61
α =8 0.50 4.31 1.01 9.05 58.9

1 α = 0.1 5.53 1.03 0.04 52.8 68.7

G2 (s) =
(s + 1)(1 + αs)(1 + α 2 s)(1 + α 3 s) α = 0.2 2.87 1.08 0.07 38.6 66.3
α = 0.5 1.19 1.36 0.17 19.1 62.6
α = 0.1 1.03 2.15 0.31 19.4 61.2
α = 0.2 0.96 2.18 0.33 16.6 61.6
1 − αs α = 0.5 0.79 2.23 0.39 13 62.4
G3 (s) =
(s + 1)3 α = 1.0 0.63 2.30 0.47 7.52 50.9
α = 2.0 0.48 2.39 0.57 7.45 58.6
α = 5.0 0.36 2.58 0.72 2.69 40.4

α = 0.1 0.23 0.43 0.12 10.4 66

α = 0.2 0.30 0.59 0.17 10.4 65.8
1
G4 (s) = e−s α = 0.5 0.49 1.07 0.26 10.5 65.6
(1 + sα)2 α = 2.0 1.04 3.49 0.49 15 62.4
α = 5.0 1.42 8.32 0.92 24.2 62.1
α = 10 1.65 16.35 1.59 32.8 62.1

38 IEEE CONTROL SYSTEMS MAGAZINE » FEBRUARY 2006

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 controller parameters and their associated structures. For practi- ear projection linking the starting and ending points of the
cal simplicity and reliability, the standard PID structure is main- operating envelope, as illustrated by node 2 in Figure 6 [22].
tained as much as possible, while allowing augmentation with Similarly, two more controllers can be added at nodes or
either lowpass or median filtering for the differentiator and with setpoints 1 and 3, forming a pseudolinear controller net-
explicit antiwindup for the integrator. The resulting designs are work comprised of three PIDs to be interweighted by sched-
then embedded in the PIDeasy package. Further specific tuning uling functions S1 (y), S2 (y), and S3 (y), examples of which are
can be continued by local, fast numerical optimization if the shown in Figure 7.
plant differs from its model or data used in the initial design.

1
First-Order Delayed Plants
An example of PIDeasy for a first-order delayed plant is 0.9
shown in Figures 2 and 3. To assess the robustness of design 0.8 0.74 3
using PIDeasy, GMs and PMs resulting from designs for 0.7

Output (mol/l )
plants with various L/T ratios are shown in Figure 5 [19]. 0.6
While requirements of fast transient response, no overshoot, 0.49 2
0.5
and zero steady-state error are accommodated by time- ∆dmax
0.4
domain criteria, PIDeasy’s multiobjective goals provide fre- 0.31
0.3
quency-domain margins in the range of 9–11 dB and 65–66◦ .
1
0.2
Higher Order Plants 0.1
For higher order plants, we obtain multioptimal designs for 0
the 20 benchmark plants [24] 0 1 2 3 4 5 6
Input (l/h)

1
G1 (s) = , α = 1, 2, 3, 4, 8, (19) FIGURE 6 Operating trajectory (bold curve) of the nonlinear chemical
(s + 1)α process (23) for setpoints ranging from 0 to 1 mol/, as given by
1 (24). A PID controller is first placed at y = 0.49 (node 2) by using
G2 (s) = ,
(s + 1)(1 + αs)(1 + α 2 s)(1 + α 3 s) the maximum distance from the nonlinear trajectory to the linear
α = 0.1, 0.2, 0.5, (20) projection (thin dotted line) linking the starting and ending points of
1 − αs the operating envelope. Similarly, two more controllers can be
G3 (s) = , α = 0.1, 0.2, 0.5, 1, 2, 5, (21) added at nodes 1 and 3, forming a pseudo-linear controller network
(s + 1)3
comprised of three PIDs. Without the need for linearization, these
1
G4 (s) = e−s, α = 0.1, 0.2, 0.5, 2, 5, 10. (22) PID controllers can be obtained individually by PIDeasy or other PID
(1 + sα)2 software directly through step-response data, or obtained jointly by
using an evolutionary algorithm [22].

The resulting designs and their corresponding gain and PMs

are summarized in Table 3.

Setpoint-Scheduled PID Network S1 S2 S3

1
Consider the constant-temperature reaction process

dy(t) 1
 0.8
= −Ky2 (t) + d − y(t) u(t), (23)
dt V
Weighting

0.6
where
y(t) = concentration in the outlet stream (mol/), 0.4
u(t) = flow rate of the feed stream (/h),
K = rate of reaction (/mol-h), 0.2
V = reactor volume ( ),
0
d = concentration in the inlet stream (mol/ ). 0.31 0.49 0.74
The setpoint, equilibrium, or steady-state operating trajectory
of the plant is governed by 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Scheduling Variable y (mol/l)
1 d
Ky2 + u y − u = 0. (24)
V V FIGURE 7 Fuzzy logic membership-like scheduling functions S1 (y),
S2 (y), and S3 (y) for individual PID controllers contributing to the
For setpoints ranging from 0 to 1 mol/, an initial PID
PID network at nodes 1, 2, and 3, respectively. Due to nonlinearity,
controller can be placed effectively at y = 0.49 by using the these functions are often asymmetric. Similar to gain scheduling, lin-
maximum distance from the nonlinear trajectory to the lin- ear interpolation suffices for setpoint scheduling.

FEBRUARY 2006 « IEEE CONTROL SYSTEMS MAGAZINE 39

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 The PID controller centered at node 2 can be obtained by point [22]. The remaining PIDs can be obtained similarly. For
PIDeasy or other PID software directly through step-response simplicity, we obtain the controller centered at node x, where
data without the need for linearization at the current operating x = 1, 2, and 3, by using step-response data from the starting
point to node x. The resulting PID network is given by



0.8 9.82 1.22 0.0376 1
Output (mol l−1)

u(t) = [S1 S2 S3 ] 15.6 0.784 0.0241 p−1 e(t),

0.6 0.53
28.6 0.481 0.0137 p
0.4 (25)

0.2
where p denotes the derivative operator.
0 To validate tracking performance using a setpoint that is
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
not originally used in the design process, the setpoint r = 0.53
Time (h)
6 mol/
is used to test the control system. The response is
Control Signal (l h−1)

shown in Figure 8, where a 10% disturbance occurs during [3,

4 3.5] h, confirming load disturbance rejection at steady state.
Figure 9 shows the performance of the network at multiple
2 operating levels not originally encountered in the design. If a
more sophisticated PID network is desirable, the number of
0 nodes, controller parameters for each node, and scheduling
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (h) functions can be optimized globally in a single design process
by using an EA [22].
FIGURE 8 Performance of the pseudolinear PID network applied to It is known that gain scheduling provides advantages over
the nonlinear process example (23). To validate tracking perfor- continuous adaptation in most situations [8]. The setpoint-
mance using a setpoint that is not originally used in the design scheduled network utilizes these advantages of gain schedul-
process, the setpoint r = 0.53 mol/
is used to test the control sys- ing. Furthermore, by bumpless scheduling, the network does
tem. The controller network tracks this setpoint change accurately
not require discontinuous switching between various con-
without oscillation and rejects a 10% load disturbance occurring
during [3, 3.5] h. troller structures.

Discussion and Conclusions

PID is a generally applicable control technique that
1
0.9 derives its success from simple and easy-to-under-
Output mol (l−1)

stand operation. However, because of limited infor-

0.6 mation exchange and problem analysis, there
0.5 remain misunderstandings between academia and
0.3 industry concerning PID control. For example, the
message that increasing the derivative gain leads to
improved transient response and stability is often
0
0 1 2 3 4 5 6 7 8 9 wrongly expounded. These misconceptions may
Time (h) explain why the argument exists that academically
10 proposed PID tuning rules sometimes do not work
Control Signal (l h−1)

8 well on industrial controllers. In practice, therefore,

switching between different structures and func-
6
tional modes is used to optimize transient response
4 and meet multiple objectives.
Difficulties in setting optimal derivative action
2
can be eased by complete understanding and careful
0 tuning of the D term. Median filtering, which is
0 1 2 3 4 5 6 7 8 9
Time (h) widely adopted for preprocessing in image process-
ing but yet to be adopted in controller design, is a
FIGURE 9 Performance of the pseudolinear PID network applied to the nonlin- convenient tool for solving problems that the PI-D
ear chemical process (23) at multiple operating levels that are not originally and I-PD structures are intended to address. A medi-
used in the design process. The network tracks these setpoint changes accu-
an filter outperforms a lowpass filter in removing
rately without oscillation. It can be seen that the control effort increases dispro-
portionally to the setpoint change along the nonlinear trajectory, compensating impulse spikes of derivative action resulting from a
for the decreasing gain of the plant when the operating level is raised. sudden change of setpoint or disturbance. Embedded

40 IEEE CONTROL SYSTEMS MAGAZINE » FEBRUARY 2006

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.
 in the feedback loop, a median filter also outperforms a pre- neering Asia Pte. Ltd., Singapore. Since 2005, within the same
filter in dealing with disturbances. company, he has been working on process industry stand-
Over the past half century, researchers have sought the ardization and new technology development. His research
next key technology for PID tuning and modular realization. interests include evolutionary, multiobjective learning, com-
Many design methods can be computerized and, with simula- putational intelligence, control systems, and engineering
tion packages widely used, the trend of computerizing simu- design optimization.
lation-based designs is gaining momentum. Computerizing Gregory Chong received a First-Class Honors B.Eng. degree
enables simulations to be carried out automatically, which in electronics and electrical engineering from the University of
facilitates the search for the best possible PID settings for the Glasgow, United Kingdom, in 1999. He is completing his
application at hand. A simulation-based approach requires no Ph.D. at the same university in evolutionary, multiobjective
artificial minimization of the control amplitude and helps modeling, and control for nonlinear systems.
improve sluggish transient response without windup.
In tackling PID problems, it is desirable to use standard
PID structures for a reasonable range of plant types and oper- REFERENCES
[1] J.G. Ziegler and N.B. Nichols, “Optimum settings for automatic con-
ations. Modularization around standard PID structures trollers,” Trans. ASME, vol. 64, no. 8, pp. 759–768, 1942.
should also help improve the cost effectiveness of PID control [2] W.S. Levine, Ed., The Control Handbook. Piscataway, NJ: CRC Press/IEEE
and maintenance. This way, robustly optimal design methods Press, 1996.
[3] L. Wang, T.J.D. Barnes, and W.R. Cluett, “New frequency-domain design
such as PIDeasy can be developed. By including system iden- method for PID controllers,” Proc. Inst. Elec. Eng., pt. D, vol. 142, no. 4, pp.
tification techniques, the entire PID design and tuning process 265–271, 1995.
can be automated, and modular code blocks can be made [4] J. Quevedo and T. Escobet, Eds., “Digital control: Past, present and future
of PID control,” in Proc. IFAC Workshop, Terrassa, Spain, Apr. 5, 2000.
available for timely application and real-time adaptation. [5] I.E.E. Digest, “Getting the best out of PID in machine control,” in Digest
IEE PG16 Colloquium (96/287), London, UK, Oct. 24, 1996.
ACKNOWLEDGMENTS [6] P. Marsh, “Turn on, tune in—Where can the PID controller go next,” New
Electron., vol. 31, no. 4, pp. 31–32, 1998.
This article is based on [25]. Kiam Heong Ang and Gregory [7] Y. Li, W. Feng, K.C. Tan, X.K. Zhu, X. Guan, and K.H. Ang, “PIDeasy and
Chong are grateful to the University of Glasgow for a post- automated generation of optimal PID controllers,” in Proc. 3rd Asia-Pacific
graduate research scholarship and to Universities UK for an Conf. Control and Measurement, Dunhuang, P.R. China, 1998, pp. 29–33.
[8] K.J. Åström and T. Hägglund, PID Controllers: Theory, Design, and Tuning.
Overseas Research Students Award. The authors thank Prof. Research Triangle Park, NC: Instrum. Soc. Amer., 1995.
Hiroshi Kashiwagi of Kumamoto University and Mitsubishi [9] F.G. Shinskey, Feedback Controllers for the Process Industries. New York:
Chemical Corp., Japan, for the nonlinear reaction process McGraw-Hill, 1994.
[10] C. Bohn and D.P. Atherton, “An analysis package comparing PID anti-
model and data. windup strategies,” IEEE Contr. Syst. Mag., vol. 15, no. 2, pp. 34–40, Apr.
1995.
AUTHOR INFORMATION [11] K.J. Åström and T. Hägglund, “The future of PID control,” Contr. Eng.
Pract., vol. 9, no. 11, pp. 1163–1175, 2001.
Yun Li (Y.Li@elec.gla.ac.uk) is a senior lecturer at the Universi- [12] Techmation Inc., Techmation [Online], May 2004. Available: http://pro-
ty of Glasgow, United Kingdom, where he has taught and tuner.com
conducted research in evolutionary computation and control [13] J.P. Gerry and F.G. Shinskey, “PID controller specification,” white paper
[Online]. May 2004. Available: http://www.expertune.com/PIDspec.htm
engineering since 1991. He worked in the U.K. National Engi- [14] BESTune, PID controller tuning [Online]. May 2004. Available:
neering Laboratory and Industrial Systems and Control Ltd, http://bestune.50megs.com
Glasgow, in 1989 and 1990. In 1998, he established the IEEE [15] Honeywell International Inc. [Online]. May 2004. Available: http://
www.Acs.Honeywell.Com/Ichome/Rooms/DisplayPages/LayoutInitial
CACSD Evolutionary Computation Working Group and the [16] Y. Li, K.H. Ang, and G. Chong, “Patents, software, and hardware for
European Network of Excellence in Evolutionary Computing PID control,” IEEE Contr. Syst. Mag., vol. 26, no. 1, pp. 42–54, 2006.
(EvoNet) Workgroup on Systems, Control, and Drives. He [17] J.P. Gerry, “A comparison of PID control algorithms,” Contr. Eng., vol.
34, no. 3, pp. 102–105, Mar. 1987.
was a visiting professor at Kumamoto University, Japan. He is [18] A. Kaya and T.J. Scheib, “Tuning of PID controls of different structures,”
currently a visiting professor at the University of Electronic Contr. Eng., vol. 35, no. 7, pp. 62–65, July 1988.
Science and Technology of China. His research interests are in [19] W. Feng and Y. Li, “Performance indices in evolutionary CACSD
automation with application to batch PID generation,” in Proc. 10th IEEE Int.
parallel processing, design automation, and discovery of engi- Symp. Computer Aided Control System, Hawaii, Aug. 1999, pp. 486–491.
neering systems using evolutionary learning and intelligent [20] R. Gorez, “A survey of PID auto-tuning methods,” Journal A, vol. 38, no. 1,
search techniques. He has advised 12 Ph.D. students and has pp. 3–10, 1997.
[21] A. O’Dwyer, Handbook of PI and PID Controller Tuning Rules. London:
140 publications. He can be contacted at the Department of Imperial College Press, 2003.
Electronics and Electrical Engineering, University of Glasgow, [22] Y. Li, K.H. Ang, G. Chong, W. Feng, K.C. Tan, and H. Kashiwagi,
Glasgow G12 8LT, U.K. “CAutoCSD—Evolutionary search and optimisation enabled computer-
automated control system design,” Int. J. Automat. Comput., vol. 1, no. 1,
Kiam Heong Ang received a First-Class Honors B.Eng. and pp. 76–88, 2004.
a Ph.D. degree in electronics and electrical engineering from [23] Specification Data File of Commander 355, ABB, SS/C355, Issue 3, 2001.
the University of Glasgow, United Kingdom, in 1996 and [24] K.J. Åström and T. Hägglund, “Benchmark systems for PID control,” in
Proc. IFAC Workshop, Terrassa, Spain, 2000, pp. 165–166.
2005, respectively. From 1997–2000, he was a software engi- [25] K.H. Ang, G. Chong, and Y. Li, “PID control system analysis, design, and
neer with Advanced Process Control Group, Yokogawa Engi- technology,” IEEE Trans. Contr. Syst. Tech., vol. 13, no. 4, pp. 559–576, 2005.

FEBRUARY 2006 « IEEE CONTROL SYSTEMS MAGAZINE 41

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on February 01,2024 at 05:47:01 UTC from IEEE Xplore. Restrictions apply.