Documente Academic
Documente Profesional
Documente Cultură
www.elsevier.com/locate/jprocont
Abstract
In this paper, we present an approach using relay feedback for automatic identication of the PID conguration and parameters
in existing and operational closed-loop systems. During the duration of the same experiment, the control performance may be
evaluated, providing an assessment of whether the current control parameter set is adequate and if re-tuning is necessary. Systematic and on-line automatic PID tuning rules are further developed and provided for common congurations of PID control.
Simulation examples and a real-time experiment are provided to illustrate the practical appeal of the proposed method. # 2001
Elsevier Science Ltd. All rights reserved.
Keywords: PID control; PID structure; Relay feedback; Performance assessment; PID autotuning
1. Introduction
The PID controller has remained, by far, as the most
commonly used controller in practically all industrial
control applications. Estimated production gure currently stands at a heftyseveral hundred thousands per
year and this gure is still healthily on the uptrend [1,2].
PID controllers come in various structures, depending
on the manufacturer. Among them, the two largest
classes of PID structure are the parallel (non-interacting) and the series (interacting) conguration [2]. The
main reason for the non-standard structure is due to the
historical transition of the controller from pneumaticbased analog implementation to micro-processor-based
digital implementation of the present age, and the corresponding change in the structural requirements.
Although functionally, the parallel conguration of PID
controller is the more general version of the two, the
series conguration has been relatively more rmly
rooted in the pneumatic analog version of PID control
as it requires fewer ampliers which were very expensive
just a few decades ago. Many manufacturers also retain
their conventional and trademark structures, in spite of
the widespread proliferation of digital controllers, due
to the large base of users already familiar with them.
The dierent structures of PID controller have direct
implications on controller tuning and the resultant
* Corresponding author. Tel.: +65-874-211; fax: +65-779-1103.
E-mail address: eletankk@nus.edu.sg (K.K. Tan).
0959-1524/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PII: S0959-1524(00)00012-3
484
1
det
ut Kcp et
etdt Tdp
:
Tip
dt
In many literature, this is also known as the ideal
type. For the sake of simplicity, the paper will only
focus on this basic structure without considering other
minor structural modications for derivative ltering,
set-point weighting, etc.
The parallel type can be shown diagrammatically as in
Fig. 1. It is the one normally described in text books and
the more general one functionally, compared to the series type of PID controller. The main characteristic for
the parallel type is that the P, I and D parts are separated and connected in parallel. This type is less commonly seen in industrial process controllers until more
recent years. The reason for this is that this conguration was dicult to design using pneumatic controllers
due to the requirement for more analog ampliers
which were very expensive at that time. Traditionally,
this type has not been adopted, even though it is just as
easy to implement as the other designs in the modernmicrocomputer based controllers. However, this type is
becoming more and more commonly used in modern
control systems.
2.2. Series type
The series type is still the most common type in
industrial process controllers. It is given by the
equations:
e1 t et Tds
ut Kcs
det
dt
1
e1 t
Tis
e1 tdt :
Gcp s Kcp
1
Tdp s ;
1
Tip s
1
Gcs s Kcs 1
1 Tds s:
Tis s
485
2
3
This conguration has been applied in the development of a robust on-line relay automatic tuning method
for general control applications. The developed on-line
tuning method has been eective against many constraints of the basic relay autotuning method (as
explained in [11]), and it retains the simplicity of the
original method. A rich set of examples, where the online tuning method is more eective than the o-line
one, is documented in [11] (including the performance
under dierent kinds of disturbances). The parallel type
PID controller considered in [11] may be designed based
on a direct or an indirect approach. In the direct
approach, the control gains are tuned directly based on
frequency response information obtained from the
experiment so that the actual closed-loop behaviour
approaches the desired one. In the indirect approach,
the process is approximated by a model and the parameters of the model are obtained from the signals
resulting from the proposed conguration. The controller is subsequently designed based on the process
model. In Section 5 of this paper, the corresponding
tuning rules for the series type will be provided for both
approaches.
In this section, the application of the conguration of
Fig. 3 for PID control structure and parameter identication will be elaborated. To this end, we consider the
following two cases.
3.2. Case I: PID structure unknown; PID parameters
known
where Gc s denotes the PID controller during the tuning process, and depending on its structure, may be
described by the following transfer functions (2) or (3).
If the system settles in a steady-state limit cycle oscillation at frequency !u , Gc j!u can be obtained from
amplitude and phase response considerations in the
usual manner. It then follows that multiple points of the
486
u2 te
jk!u t
e2 te
jk!u dt
Gc jk!u 0Tu
0
dt
; k 1; 3; 5; . . . ;
where Tu !2u is the period of the stationary oscillations. If there are 2n samples, where n is an integer, the
Fourier transforms can be eciently computed using
the fast Fourier transform (FFT) algorithm [17] as
Gc jk!u
FFTu2
:
FFTe2
1
jTdp k!u ;
jTip k!u
k 1; 3; 5; . . . ; 2n
1;
Kcp
;
Tip k!u
7
G~ c j!:
! Gc j!
!
..
.
1
!
k!
..
.
2n 1!
2
1
k!
..
.
..
.
1
2n 1!
7
7
7"
#
7 Kcp Tdp
7
7
Kcp
7
Tip
7
7
5
3
ImGc j!
7
6
..
7
6
.
7
6
7
6
7;
6
Im
G
jk!
c
7
6
7
6
..
7
6
.
5
4
ImGc j2n 1!
k 1; 3; 5; . . . ; 2n
1:
11
P
2
k2 !u
P
k
k2 !2u
P
k
1
k2 !2u
1
k!u ImGc jk!u n
k 1; 3; 5; . . . ; 2n
1:
P
k
12
and
where
P
Tdp
1
k2 !2u
P
k
P
k!u ImGc jk!u n k!1 u ImGc jk!u
k
;
P 2 2P 1
2
k !u k2 !2 n
Kcp
k
k 1; 3; 5; . . . ; 2n
1:
k 1; 3; 5; . . . ; 2n
14
Kcs
;
Tis k!u
15
q
22 41
2
Kcs
Tis P
k
k2 !2u
P
k
16
P
k
P
2
k2 !u
1
k2 !2u
n2
1
k!u ImGc jk!u n
k 1; 3; 5; . . . ; 2n
1:
17
X
X
k2 !2u ImGc jk!u
k
X
n k!u ImGc jk!u
k
X
13
1
Kcs
487
1
k2 !2u
X
k
P
k
1
k!u
P
P
ImGc jk!u =
k2 !2u k21!2
k
k 1; 3; 5; . . . ; 2n
P
2
2
n2 ;
1:
ReGc jk!u
n
k 1; 3; 5; . . . ; 2n
1:
20
21
and
P
Tds
1
k2 !2u
P
k
P
k!u ImGc jk!u n k!1 u ImGc jk!u
k
;
P 2 2P 1
k !u k2 !2 n2
Kcs
k
k 1; 3; 5; . . . ; 2n
1:
18
Kc
ReGc jk!u
n
k 1; 3; 5; . . . ; 2n
1:
22
488
Kc
P
Ti
1
k!u
P
k
1
k2 !2u
ImGc jk!u
k 1; 3; 5; . . . ; 2n
1:
23
k!u
Kc
ImGc jk!u
Ti k!u
P 2 2
;
Kc k !u
k
k 1; 3; 5; . . . ; 2n
1:
depend on the type of disturbance present. The robustness of the controller is also related to the demand for
speed. Robustness means the sensitivity to variations in
the process dynamics. Controllers which are tuned to
give fast control are usually more sensitive to variations
in the process than controllers which are more conservative in their settings.
In some cases, it is important to tune the controller
such that there is no overshoot; in some cases, a slow
and smooth response is desired; some cases warrant fast
response and signicant oscillations are no problem.
The point is, the denition of good control depends on
specic requirements.
4.2. Benchmark against frequency response prototype
If a desired frequency response prototype [i.e. G yr j!]
is available, assessment of control performance in terms
of speed and stability can be made with reference to this
specication.
A suitable stability measure is
Gyr j!u
:
25
G yr j!u
24
Gyr j!u
Gc j!u 1 Gyr j!u
27
G yr j!u
Gp j!u 1 G yr j!u
28
29
2
arg
G
2
Tip
tan
2
!u
30
31
and
Kcp
Tip !u G c j!u
0:25T2ip !2u 1
32
489
q
2 argG c j!u
6:25!2u 4!2u tan2
2
Tis
2 argG
c j!u
2
!u tan
2
2:5!u
33
and
Tis !u G c j!u
Kcs qq :
T2is !2u 1 0:0625T2is !2u 1
34
kp
e
1 s
sL
35
;
36
1
:
ku 1
37
arctan !u r
!u L 0;
39
490
s
!2u 2 1
:
ku 1Kcp kp !u Ti
2
!u 21 !2u 22
where 1 , 2
Tip
2
Tip 4Tip Tdp
2Tip Tdp
40
1
2
arctan !1u arctan !2u
arctan !u
!u
42
kp ttTu
t
ytdt
rtdt
43
sL
44
arctan !u
!u L 0
45
s
!2u 2 1
ku 1Kcs kp !u Tis
!2u T2is 1 !2u T2ds 1
46
L2
47
arctan !u
48
6. Simulation
In this section, the practical appeal and other useful
characteristics of the identication method will be
G p s
1
e
s 1 2
49
Assume also the following parallel type PID controller has already been commissioned and is in force:
1
0:9167s :
50
Gc s 1:2 1
2:4s
Assume further the PID parameters are known, but
the structure of this controller is to be determined.
From the on-line relay experiment, the input and output
signals of the controller are obtained and shown in Fig.
4. If the parallel type is assumed, (9)(12) yield the PID
parameters as Kcp 1:2151, Tip 2:4273 and
Tdp 0:9056. If the series type is assumed, (15)(17)
yield Kcs 1:67, Tis 3:33 and Tds 0:649. Thus, the
parallel structure is correctly identied.
6.2. Example 2: identication under noisy measurements
Consider the same process:
G p s
1
e
s 1 2
51
Gp s
1:2e 10s
;
5s 12:5s 1
53
491
1
;
15:45s
Fig. 5. On-line identication experiment: unknown structure only and noisy measurements.
55
492
1
e s:
s 1 2
Fig. 6. On-line identication experiment unknown PID structure and parameters; Td disabled.
Fig. 7. On-line identication experiment unknown PID structure and parameters; Td enabled.
56
57
493
494
1
e
1 1:778s
1:5007s
59
1
e
2s 12
1:00s
60
1
0:519s :
2:074s
62
495
496