Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Model-Based Auto-Tuning System Using Relay Feedback

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Model-based Auto-tuning System Using Relay Feedback

Hsioa-Ping Huang, Kuo-Yaun Luo


Pse, Department of Chemical Engineering, National Taiwan University,
Taipei Taiwan 10617, R.O.C.

Abstract: An on-line model based autotune system that employes conventional ATV test
is proposed. The ATV responses from normalized FOPDT and SOPDT processes are
grouped into two zones in a space, which has normalized amplitude and normalized period
as coordinates. In terms of the amplitude and the period of constant cycles of an ATV test,
model-based tuning rules are also prepared for two types of process, one for FOPDT in
one zone and one for under-damped SOPDT in the other zone. Thus, from an ATV test,
the amplitude and period of constant cycles are used in an identification step to select
tuning rules to be applied to a given process. Then, PID controller settings are computed
according to the selected tuning rules. The system can be inplemented as simple as the
conventional autotune system, and the resulting control performance is is compatible to
that from a model-based controller design.
Keywords: Relay Feedback, Inverse-based PID Controller, On-line Autotuning

1.INTRODUCTION
The PID auto-tuning has shorten the time to commission
control system, and facilitated control optimization through
regular retuning. The relay feedback auto-tuning method
proposed by Astrom and Hagglund was attractive owing to
its simplicity and robustness. The important ingredients in
the auto-tuning system are identification (parametric or nonparametric) and rules for controller tuning. Many researches
on these two aspects have been reported in recent years. For
examples, improvements on the accuracy and efficiency by
saturation relay feedback test (Yu, et al.,1995) or by reducing
high-order harmonic terms or using the Fourier analysis
(Lee, et al., 1995; Sung, et al., 1995; Wang, et al., 1997) have
been reported. On the other hand, extensive works on the
PID tuning formulae and refinement have been published.
Extensions of Auto-tuning system to other different cases,
such as: time varying delay (Leva,1993), cascade controllers
(Hang, et al.,1994), the Smith Preidictor (Plamor, et al.,1994)
and the FSA (finite spectrum assignment) controller (Wang,
et al.,1995) have also been found in literature. It used to
observe that use of model-based tuning rules results in
better control performance. But, the trade-off is lots of
efforts are required to identify an appropriate model to apply
these model-based tuning rules. As has been mentioned,
although methods of identification using relay or ATV test
has been extensively addressed in literature, the absence of
an model-based auto-tuning system as simple and robust as
the one of Astrom and Hagglund is a simple fact. In this
paper, we will reformulate the inverse-based tuning rules in
terms of ultimate gain and frequency so that can be easily

+
R -

e-

h
h

u-

Process G(s)

y-

Figure 1: Relay Feedback System


applied using those data obtained from a simple ATV test. On
the other hand, a simple method based on the same ultimate
gain and frequency obtained is proposed to select the tuning
rules which are formulated specifically for different simple
dynamics.

2.RELAY FEEDBACK IDENTIFICATION


2.1 Ideal Relay Test
In 1984, Astrom and Hagglund (Astrom et al., 1984) suggest
the relay feedback test to generate sustained oscillation as an
alternative to the conventional continuous cycling technique.
This relay feedback test was soon (Luyben, 1987) referred as
autotune variation (abbr. ATV) test. As shown in Fig.6 is the
ATV test. Controller tuning using ATV test is attractive, because it is operated under closed loop and no a priori knowledge of system is need. The test provides ultimate gain and ultimate period for applying Z-N rules to tune a PID controller.

y6

6
A

Pu

2.2 Relay Feedback Response Curves

Consider the FOPDT and SOPDT models for representing


general dynamics in chemical plants. The original and
normalized models are shown in the following.

u6

FOPDT Model
h
6
h ?

6
?

G(s)

Kp es
s+1

G(s)

s
Kp e
s
+1

where s = s,
=
SOPDT Model

Figure 2: Autotune with ATV Test


The ultimate gain and ultimate frequency are extimated from
the ATV test as:
kcu

4h
A
2
Pu

(1)
(2)

By making use of the data obtained, PID controller parameters


can be computed using Z-N rules. The control performance
thus obtained is in general crude compared with those tuned
with model-based rules, such as IMC. This fact is most obvious when the process has underdamped second order dynamics. But, with one simple ATV test, it is usually not sufficient
for identifying parametric model of processes other than one
that has exactly FOPDT dynamics. This makes spaces for research on developing the parametric models from ATV tests.
There are quite a few researchers (e.g. Luyben, 1987; Li, et
al.,1991, Ching et al., 1992; Lee and Sung, 1993, Shen et al.,
1996, Sung and Lee, 1997, Wang, et al., 1997, Huang, et al.,
2000) worked on this problem. Nevertheless, methods from
those published works have different extents of sophisicatcy
and complexities, and is not convenient for apllication to autotune systems, where a simple and robust method is desirable.
As a result, there remains space for developing effective parametric autotune methods to enhance the controller tuning and
achieve better performance.
In model-based methods, tuning formulas are derived for
each specific type of transfer function model. But, as far as
control performance is concerned, to differentiate FOPDT
dynamics from the overdamped dynamics seems not so crucial. In general, the FOPDT-based formula works quite well
for over-damped or over-damped-like processes, but not for
under-damped ones. Tuning rules for underdampled SOPDT
process are indeed required. In this sense, the identification
in a autotune system needs only to select models from two
types, that is the FOPDT one or an under-damped SOPDT
one. In the following, we shall illustrate this identification
using one ATV test.

Kp es
+ 2 s + 1

Gp (s)

2 s2

Gp (
s)

2 s2

Kp es
+ 2
s + 1

where s = s,
= .
For these two types of processes, extensive ATV tests have
been conducted over wide ranges of normalized parameters.
For example, for SOPDT process, ranges from 0.1 to 100
and from 0.1 to 100. The resulting normalized ultimate
gains and ultimate period are plotted and is as shown in
Figure(3).

= 1.0

102
101

10

100
100
= 0

1
A 10
kp h
102

0.1
1.0

103

10

104

100

105
1

10

100
Pu

Figure 3: Generalized ATV Responses


2.3Identification

In Figure(3), it is thus possible to divide the ATV results into


different zones. As has been mentioned, for PID controler
tuning, overdamped processes and some of underdamped processes that has damping factors close to one can be approximated by FOPDT model. To differentiate the FOPDT and
overdamped SOPDT dynamics from those of underdamped
SOPDT, the curve having of 0.7 is selected as a criterion.
This is because that, in Bodes plot, an SOPDT process does
not have resonance peak when its is greater than or equal
0.707. Thus, as shown in Figure (4), there are two zones (Zone
I and Zone II) separated by two curves. Among the curves,
one is on the edge of zone I that represents FOPDT dynamics.
The normalized time constant (i.e. ) of this FOPDT process
ranges from 0.1 to 100. The other is on the edges of the zoneI
and Zone II. Zone II represents processes of underdamped dynamics. With Figure (4), identifying an unknown system using ATV test can thus be conducted using the resulting A and
Pu . By locating the experimental results on the fingure, parameters can be estimated, provide that kp and is known,
using the following equations. In case of both values of kp
and are not available, estimation method will be introduced
later in the section of tuning procedures.

where, Ku = kp kcu and = u

3. TUNING RULES
An inverse-based design is used to synthesize PID controller
for autotune. The controller is synthesized so as to have a
loop transfer function (abbr. LTF) of two standard forms of
the following. That is:
Gloop (s) =

0.65 (1 + 0.4s es )

s(1 + f s)

(3)

0.4 es
s(1 + f s)

(4)

or,
Gloop (s) =

These loop transfer function can provide the system reasonable stability robustness and control performance. According
to these standard forms, the PID controller is sysnthesized for
FOPDT dynamic process and is given as follows:


Gc (s) = Kc (1 +

1 + D s

R s 1 + f s
)

(5)

where,

10

(Kc ) =

0.65

10

A/hKp

Zone II

(R ) =

(D ) =

0.4

SOPDT =0.7

Zone I

Gc (s) = Kc (1 +

FOPDT
2
0

10

10

Kp

On the other hand, the PID controller for SOPDT type dynamic process is:

10

10

10

Pu/d

Kc

Figure 4: Simplified ATV Responses


1. For point located in FOPDT zone:


Ku 2 1
Pu
=
=
Ku 2 1
u
2

tan1 Ku 2 1
=
u

(6)

2
0.4
Kp
2
2
2

Since, all controller parameters are in terms of the dynamic


parameters as idtentified inthe previous section, these controller parameters can be re-formulated to be in terms of ultimate gain and ultimate period. In other words, the tuning parameters can be directly computed from the results of a ATV
test. As an example, for FOPDT process, the PID parameters
can be re-written as:


where, Ku = kp kcu
2. For point located in underdamped SOPDT zone:

tan2 ()
2 u = Y = Ku
1 + tan2 ()

Y
u = X = 1 +
tan()

1
+ D s)(1 + f s)
R s

Kc

(Kc )ZN 0.54

(R )ZN F1

(D )ZN F2

where,
F1 =


2 K2 1
Kcu
p

(7)


F2 =

1.6
1

tan1

2 K2 1
Kcu
p

+
R -

A comparison of this inverse-based tuning with that of Z N


tuning is as shown in Figure(5).
10

ue =
h u ?Process G(s)

y-

Figure 6: ATV Test with External Excitation

ZN
IB
kc

e-

0
0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

10

ZN
IB

y6

0
0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.5
ZN
IB

0.45
0.4
0.35
0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Figure 5: Comparisons of Inverse-based And Z-N Controller


Settings

6
A
A1
?
?
...............................................................................
?
6 - y
6
A2 6
t
A
?
?
Pu

u6
h
6
h ?

6
?

4. AUTOTUNE PROCEDURES
As has been methioned, when the values of kp and the apparent deadtime, , are available, the autotune procedure proceeds just like the conventional one, except that the selection
of tuning rules by locating (A/(kp h), Pu /) on Figure (4) is
required as a proceding step, and the tuning rules itself are
different.
In case of where kp and are not available, the ATV test has
to make a slightly modification. The excitation to the relay
feedback system is introduced through an external input at
the output of relay. As shown in Figure (6). The external input has to be small like up to 10% of h, the level of relay output. Nomenclatures for computing amplitudes and period are
given in Fighue (??). Then, kp is estimated by the following
equation:
 t+P/2
y( )d
kp = tt+P/2
(8)
u(
)d
t
On the other hand, the apparent deadtime for applying the
identification zones in Figure (4) are estimated as an average
value of true deadtime detected at the very begining of the test
and q quater of period later at constant cycling, as shown in
Figure.(??). This is because the deadtime interms of SOPDT
dynamics is always lying between te two values aforementioned.
To calculte the amplitudes for identification, the centerline of
the cycling has to be updated to to the bias form the input to
a height of y = 0.5(A1 |A2 |).

Fig 7: External Excitation ATV Test


Besides the preceding steps just mentioned, all other remaining procedures are the same as the previously described ones.
Thus, by estimating the normalized amplitude and the nornmalized period, and by making use of Figure (4), tuning rules
can be selected and controller parameters can be computed.
..
y ...
- L
..
..
6A
...
..
..
....................................................................................................
D
?
...
..
..
..
...
..
Pu

..
.
Time
Fig 8: Estimation for Apparent Deadtime

5. SIMULATION RESULTS

Table 4: PID parameters and IAE values

The following are the results of applying this autotune


procedure to a few example processes.
Ex.1
1
es
(s + 1)5

Process

h
1.000

Kcu
1.9758

Ku
1.9758

a
0.6444

u
0.5689

0.5
proposed
ZN
0

Pu
11.0444

L
2.8337

D
1.500

proposed
0.2094
4.6926
5.3974
21.749

1.5

Table1: Simulation Results


Kp
1.0000

Z-N
0.7141
17.3900
4.3475
38.325

kc
I
D
IAE

50

100

150

d
2.0206

3
2.5

Table2: PID Parameters and IAE values


Z-N
1.1623
5.5222
1.3805
8.518

kc
I
D
IAE

1.5
1
0.5

proposed
0.3348
2.2904
1.4776
6.685

proposed
ZN
0

50

100

150

Figure 8: Ex.2 Tuning result


Ex.3

1.5

Process

1
es
(s2 + 0.6s + 1)

0.5

Table5: Simulation Results

proposed
ZN
0

10

15

20

25

30

35

40

45

50

t
6

Kp
1.0000

a
1.727

h
1.000

Pu
5.4806

L
1.3609

Kcu
0.7373

Ku
0.7373

D
1.000

d
1.1805

u
1.1464

4
3
2
1
0

proposed
ZN
0

10

15

20

25

30

35

40

45

50

Table6: PID parameters and IAE values

Figure 7: Ex.1 Tuning result

kc
I
D
IAE

Ex.2
Process

1.08
e10s
(s + 1)2 (2s + 1)3

Z-N
0.4337
2.7403
0.6851
7.3250

proposed
0.2128
0.6280
1.4044
4.0262

5. CONCLUSION

Table3: Simulation Results


Kp
1.0826

a
1.0487
Pu
34.78

h
1.000
L
11.2275

Kcu
1.2141
D
11

Ku
1.3144
d
11.1138

u
0.1807

In this paper, we have presented a autotune system which


make uses of model-based tuning rules to enhance control
performance. The presented system works in a similiar way
to the one of Astrom and Hugglund in the sense it uses a
conventional AYV test and the resulting amplitude and period
of constant cycles. Because of using parametric models,
addtional steps it needs are the estimation of kp and . The
tuning rules are derived from an inverse-based approach,

Lee, J., S.W. Sung (1993), Comparison of two identification methods for pid controller tuning, AIChE J.,39,695697

1.5

0.5

0
proposed
ZN
0.5

20

40

60

80

100

120

140

160

180

200

Lee, T.H., Q.G. Wang and K.K. Tan (1995), A modified


relay-based technique for improved critical point in process control, IEEE Trans. Control System Tech., 3,,330337

Leva, A. (1993), PID auto-tuning algorithm based on relay feedback, IEE Proc. Pt. D., 140,328-338

1
proposed
ZN
0

20

40

60

80

100

120

140

160

180

200

Li, W., E. Eskinat and W.L. Luyben (1991), An improved autotune identification method, Ind. Eng. Chem.
Res.,20,1023-1042
Luyben, W.L. (1987),Derivation of transfer functions for
highly nonlinear distillation columns, Ind. Eng. Chem.
Res., 26,2490-2495

Figure 9: Ex.3 Tuning result


1

10

Ex.1

Plamor, Z.J. and M. Blau (1994), An auto-tuner fot


Smith dead time compesator, Int. J. Control, 60,117-135

Ex.3

Schei, T.S. (1992), A method for closed loop automatic


tuning of PID controllers, Automatica, 28,587-591

10

A/hKp

Ex.2

Shen, S.H., J.S. Wu and C.C. Yu (1996), Use of biased relay feedback for system identification, AIChE
J.,42,1174-1180

10

10

10

10

10

Pu/d

Figure 10: Identification Results


which is similar to the IMC method but simpler. From the
simulation results, we can observe and conclude that this
autotune system is efficient and self-contained.

6. REFERENCE

Shen, S.H., H.D. Yu and C.C. Yu(1996), Use of


saturation-relay feedback method for autotune identification, Chem. Eng. Sci. 51,1187-1198
Shen, S.H., J.S. Wu and C.C. Yu (1996), Autotune identification under load disturbance, Ind. Eng. Chem.
Res.,35,1642-1651
Sung, S.W. and I.B. Lee (1997), Enhance relay feedback
method, Ind. Eng. Chem. Res.,36,375-381
Sung, S.W., J.H. Park and I. Lee (1995), Modified relay
feedback method, Ind. Eng. Chem. Res., 34,4133-4135

Astrom and K.J., T. Hagglund (1984),Automatic tuning


of simple regulators with specifications on phase and amplitude margins, Automatica, 20,645-651

Wang, Q.G., C.C. Huang and B. Zou (1997), Low-order


modeling from relay feedback, Industrial and Engineering Chemistry Research, 36,375-381

Ching, R.C., S.H. Shen and C.C. Yu (1992),Derivation


of transfer function from relay feedback systems, Ind.
Eng. Chem. Res.,31,855-860

Wang, Q.G., C.C. Huang and B. Zou (1997), A frequency response approach to auto-tuning of multivariable controllers, Chemical Engineering Research and
Design, 75,797-806

Huang C.C., A.P. Loh and V.U. Vasnani (1994), Relay feedback auto-tuning of cascade controllers, IEEE
Trans. Control System Tech., 2,42-45
Huang, H.P., M.W. Lee and I.L. Chien (2000),Identification of transfer function models from the relay feedback
test, Chem. Eng. Comm., 180,231-253
Huang, H.P., M.W. Lee and Y.T. Tang (2000), Identification of Wiener model using relay feedback test, J. of Ch.
E. Japan,31,604-612

Wang, Q.G., T.H. Lee and K.K. Tan (1995), Automatic


tuning of finite spectrum assignment controllers for delay system, Automatica, 28,477-482
Wang, Q.G. and Y. Zhang (2001),Robust identification of continuous systems with deadtime from step responses, Automatica, 37,377-390

You might also like