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

PID tuning based on loop-shaping H, control

W. Ta n
J.Liu
P.K.S.Ta m

Indexing terms: PID controllers. Tuning rules, Integrating processes, Stability robustness, Closed-loopperformance

controllers with a PID structure [8]; H , observations


Abstract: Loop-shaping H , controllers are weighted control might also result in controllers with a
designed for both stable and integrating PID structure [9].
processes. H , controllers for typical industrial The advantage of the first approach is that the PID
processes have a PID structure interpretation. controller can be tuned based on typical process con-
Further, if the precompensator is carefully stants instead of the explicit process model, while the
chosen, the design indicator will be independent advantage of the second approach is that modern con-
of plant constants. This fact leads to the trol theory can be utilised in controller design.
derivation of a direct relation between the In this paper the advantages of two approaches are
controller parameters and the plant constants, combined. A loop-shaping H , design approach [lo] is
and the results can be viewed as new PID tuning adopted to design H , controllers for some typical
rules for stable and integrating processes. The industrial processes. We find that for these processes
main contribution of the tuning rules is that the the resulting H , controllers can be reduced to a PID
PID parameters can be obtained using one design structure. Furthermore, if the precompensator design is
parameter which reflects the trade-off between carefully chosen, the design indicator can be made to
stability robustness and time domain performance be independent of plant constants, which makes it
of the closed-loop system, thus making it possible to obtain the PID-type H , controller
convenient to choose PID parameters dependent parameters as a direct function of plant constants. The
on the uncertainties of the processes. result can be interpreted as PID tuning based on H ,
optimal control.
The proposed tuning rule for stable processes is
equivalent to the IMC method, however, the derivation
1 Introduction is quite different. The advantage of the proposed set-
ting is that the trade-off between time domain perform-
It is known that controllers designed using modern ance and stability robustness can be directly reflected
control theory are usually very complex, thus even through the choice of design parameter. For integrating
though modern control theory has been developed sig- processes our rule predicts larger integral action than
nificantly in the past years, proportional integral deriv- existing tuning methods, and thus can reject the input
ative (PID) controllers are still widely used in the load disturbance more efficiently.
process industries due to their easy implementation.
Basically there are two different approaches to PID 2 Loop-shaping H, approach
controller design:
One is to first implement a controller using a PID In this Section, we first review the loop-shaping H ,
structure, and then try to find the 'best' PID parame- approach. Details can be found in [lo].
ters. This approach has been used for fifty years, e.g. Given a plant G, the loop-shaping H , design pro-
the Ziegler-Nichols (Z-N) method [11, the Cohen- ceeds as follows:
Coon (C-C) method [2], the two-frequency-data-based Loop shaping. Precompensator W , andlor postcom-
method [3], and the ISTE method [4]. These methods pensator W2are used to shape the singular values of G
work for stable processes. Integrating processes, how- so that the shaped plant G := W 2 GW ,has the desired
ever, appear to be more difficult to tune, thus there are open-loop shape.
few tuning methods reported in the literature [5-71. Robust stabilisation. For the shaped plant 6, solve
The other is to first design a controller using modern the following H , optimisation problem:
control theory and then try to reduce it to a PID struc-
,, ' &
, =
ture, e.g. internal model control (IMC) might result
0 IEE, 1998
IEE Proceedings online no. 19982407
Paper first received 22nd January 1998 and in revised form 21st July 1998 (1)
W. Tan and J. Liu are with the Department of Power Engineering, North
China Electric Power University, Baoding, Hebei Province, Peoples Here cmaxis a design indicator. A reasonable value (e.g.
Republic of China > 0.2) indicates that the loop shapes can be well
P.K.S. Tam is with the Department of Electronic Engineering, The Hong approximated and the closed-loop system will have
Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong good, robust stability.
IEE Proc-Control Theory Appl., Vol. 145, No. 6.November 1998 485
The final feedback controller is constructed as IY = controller given by eqn. 4 has the following structure:
WI K W2.
One advantage of the approach is that the H , opti-
misation problem (eqn. I) can be solved explicitly with-
out iteration. Suppose a minimal state-space realisation where K,, T,, T2 depend only on A. We note that the
for the shaped plant is controller cancels the pole of the shaped plant (in Pade
approximation form), thus Tl always equals 2. The
= AX + Bu
{ X
y = CX
terms gmax, K,, T2 as functions of A can be obtained
using a least-squares fit technique:
then we have = (1 + Am,,y( YX)-1’2,where X and Y 1
Emax =
are the unique semipositive stabilising solutions to the 0.572A 1.472 +
following algebraic Riccati equations: K , = 0.633X 1.154 +
ATX + X A - X B B T X + CTC = 0 (2) Ti = 2
AY + YAT YCTCY+ BBT = 0
- (3) T2 = 5.314X 0.951 (9) +
An optimal H , controller can be constructed using the Thus the final H , controller for the FOPDT model has
following generalised state-space description: the following structure:

U = BTXX This can be transformed into the following practical


PID structure:
where Q = (1 ~ E ~ ; , ) I+ YX.

3 PID tuning for stable processes


Most stable processes encountered in industry car1 be
described as first order plus deadtime (FOPDT)
models:
where

Kp=
AK,(TiT + T )
krT1
-
0.265X + 0.307
k
(T +
- 0.5
)
An interesting observation is that when we choose a PI
precompensator:

(12)
(where A is a design parameter), the optimum E,,?,, in
Note that the integral time and the derivative time are
eqn. 1 depends only on A and is independent of z. This
independent of h. We can view eqn. 12 as a PID tuning
property guarantees that direct relations between the
rule based on H , optimal control for a FOPDT plant.
parameters of the H , controller and the deadtime zcan
It has the following properties:
be obtained, and a parametrisation of the H , optimal
controller for this type of process can be found. (i) The response of the closed-loop system is independ-
Now the transfer function of the shaped plant is (A/ ent of plant time constant T, and rise time and settling
zs)e-rs. To make use of the state-space solution, the time are proportional to z.
delay is approximated by (ii) The shape of the response depends only on A.
1 - T2 S
1 5s +
thus a minimal state-space realisation for the shaped
plant is:

It is easy to verify that now the unique stabilising solu-


tions to eqns. 2 and 3 are

respectively. Thus 01 t I I
0 5 10 15 20 25 30
YX= [ + +
1 2X X 2 / 2
+
X/2 X2/4 -X2 /4 Fig. 1
... /1 = 1 ,
time
Process output for different PID settings: FOPDTprocess
~ A= i s - - a = 2
is independent of z, and so will be E,,,.
Since the state-space realisation of the shaped plant To illustrate the responses of the PID settings given
is of order 2, it is easy to verify that its H , optimal above, suppose all parameters of the process are scaled
486 IEE ProcControl Theory Appl., Vol. 145. No. 6, November 1998
to 1. Suppose a setpoint change of magnitude 1 is Example 1: Consider the plant
introduced at t = 1, a step input load disturbance of e-50~
magnitude 0.5 enters at t = IO and a step output load
disturbance of magnitude 0.5 enters at t = 20. The 1 + 10s
process output responses for three typical PID settings This example is studied in [ 3 ] to illustrate their tuning
are shown at Fig. 1 and the controller output at Fig. 2. method. The ratio of deadtime to time constant is 5,
We find that as A becomes larger, the response time for thus some tuning methods (e.g. the Z-N method) can-
the closed-loop system becomes shorter. However, the not be used. Table 1 gives the PID controller parame-
overshoot becomes larger, and the robust margin ters predicted by our method (with A = 1.5), and by the
becomes smaller, which means that the robust stability C-C method, the ISTE method and the method in [3].
of the closed-loop system becomes worse. To obtain a To compare these PID controllers, suppose a set-
fast setpoint response A can be chosen larger (A = 2 point change of magnitude 1 is introduced at t = 1 and
with E, = 0.383), while to be robust against a greater a step load disturbance of magnitude 0.5 enters at t =
variation of plant parameters, A can be chosen smaller 500, the responses for different PID settings are shown
( A = 1 with = 0.490). Usually we can choose A = in Fig. 3. In order to compare fairly, we simulate the
1.5 with = 0.423. controllers as ideal PIDs. Filters are used to ensure
smooth responses and enough robustness against high
frequency uncertainties in practice. In this example,
they do not affect the process output significantly so
they are omitted. it is observed that our method pre-
dicts suitable proportional gain and integral time, and
the closed-loop system has a faster and steady setpoint
response.

1.0

2
01
0
1
5 10 15 20 25
I

30
I i t 0.8

time
Fig.2 Controller output for d@erent PID settings: FOPDTprocess
..... /z = 1, -/z = 1.5, _ - - a = 2

Our method is found to be equivalent to the IMC


PID tuning method [8]. Both predict the same integral
time and derivative time. The only difference lies in the
proportional gain and the filter time constant. How- time
ever, we can always choose a design parameter to Fig.3 Process output for exumple 1
match the IMC method. For example the recom- -this paper, ..... [3], . . - ISTE, -
- - - - C-C

mended situation with the IMC method (filter time


constant = 10) is close to our method with A = 2 . Note
that the IMC tuning method is based on robust control
with H2 performance and multiplicative uncertainty,
while our method is based on robust control for uncer-
tainty in the sense of a gap metric; thus robust control-
lers for a FOPDT for these two different situations
coincide in structure and have close parameters. One
advantage of our PID tuning setting is that we can
reflect the trade-off between time domain performance
and stability robustness directly through the choice of
a.
Although our method is derived from the FOPDT
model, it applies to a wide range of stable processes
such as higher-order models. Simulation also shows O t
that it works well for processes with a large delay. I , I
0 200 LOO 600 800 1000
time
Table 1: PID controller settings for example 1 Fig.4 Controller output for exumple I

Method Kp T Td T'
This paper 0.493 35.0 7.143 5.604 We observe that although the filter does not affect
the process output, it does affect the controller output.
c-c 0.517 58.49 9.42 -
Since other methods do not provide guidelines for
ISTE 0.508 33.61 13.28 - selecting the filter time constant, the initial controller
131 0.275 24.85 3.72 - outputs are large due to the derivative action of an
IEE Proc.-Conirol Theory Appl., Vol. 145, No. 6, November 1998 487
iueai riu. When a filter is added, the output of the
PID controller tuned with our method becomes
smooth, as can be seen in Fig. 4.
Using a least-squares fit technique, we can also obtain
4 PID tuning for integrating processes the terms E,,,, K,, Tl, T, as functions of k
1
For integrating process Emax =
1.449X 3.085 +
e-Ts K , = 1.503X 2.922 +
d S )= s ( T s 1) + Ti = 0.238X 0.123 +
if we choose a PD precompensator T2 = 5.750X +
0.590 (19)
X
w1= -(1kr +Ts) Thus the final H , controller for the integrating process
has the following structure:
then the shaped plant is the same as for the FOPDT
plant, thus we can use the results obtained in the previ-
ous Section. The final controller has the following
structure: It is also in the PID interacting form, and can be trans-
KC
K ( s ) = -(1+ T s )
1 rs/T1 + formed into the noninteracting structure (eqn. 11) with
kr 1 TS/T~ + (13)
K p = 0‘463X 0.277 ((0.238X + 0.123)T + r )
+

This can be interpreted as a PD plus a lead-lag con- kr2


I
troller. It has the same properties as mentioned above. Ti=T+
However, a drawback of this controller is that it can- + 0.123
0.238X
not reduce any input disturbance. We have to choose a Tr
Td =
PI precompensator and make sure that the design indi- (0.238X + 0.123)T + r
cator does not depend on plant constants. T
The precompensator we choose is a PI controller Tf =
5.750X + 0.590
w1=
X(1 + Ts) (21)
k r 2s This can be viewed as a PID tuning rule for integrating
Now the shaped plant is processes. It has the same properties as listed in the
previous Section.
r2s2 1.5r

Then a minimal state-space realisation for the shaped


plant is
0 1
A = [ 0, 0 - 2 / r :] B=[
-X(r
X(T
+l)/r
]
C = [ ’7 1. 1.
T 71

(14)
It can be shown that now the unique stabilising solu-
tions to eqns. 2 and 3 have the following structure:
1 0 1/r 0
o 1/r2 O
O 1 -0.5 , :
J
1

o o 1/r3

Y = [0
1 0 0
r 0 1 Yx
0 0 r2
[; :]
0
0
r2
0 r3
(15)

respectively, where X d and Y, depend only on A and To illustrate the responses of the PID settings given
are independent of 2. Thus above, suppose all parameters of the process are scaled
to 1. Suppose a setpoint change of magnitude 1 is
Xma, (YX) introduced at t = 1, a step input load disturbance of
(16)
=L ” Y A X A ) magnitude 0.5 enters at t = 50 and a step output load
This is independent of z and so will be E,,,. disturbance of magnitude 0.5 enters at t = 100; the
Now the state-space realisation of the shaped plant is process output responses for three typical PID settings
of order 3, so that its H , optimal controller, given by are shown at Fig. 5 and the controller output at Fig. 6.
eqn. 4, takes the following structure: It is well known that the setpoint step response for
integrating processes has a large overshoot, and usually
+
- Kc(r2s2 zjrs 2 2 )
K=
+ a setpoint filter is added to reduce it. By trial and sim-
+
r 2 s 2 p1rs +pa (17)
ulation, we find that if the filter time is adopted as 0.8
where K d , zl, z2, p l , p 2 depend only on A. This can be times the integral time the closed-loop system response
approximated by a lead-lag controller: is a good compromise between overshoot and rise time.
488 IEE Proc.-Control Theory Appl., Vol. 145, No. 6, November 1998
'I
Here the second row is computed from the method
proposed in [5] with Tf = 0.2. The third row is com-
puted from Table 1 in [6]. Note that the PID control-
1.5 lers given in [5] and [6] are of the interacting form, we
have transformed them to the noninteracting form. The
fourth row is computed with p = 1 and 5 = 0.707 using
the method proposed in [7].
Suppose a setpoint change of magnitude 1 is intro-
duced at t = 10 and a step input load disturbance of
magnitude 0.5 enters at t = 40; the responses for differ-
ent PID settings are shown in Fig. 7. To have a fair
comparison setpoint filters are not used here and the
controllers are simulated as ideal PID. Our method
predicts suitable proportional gain and integral time
-0.51 I I I and thus has a fast setpoint response and the best load
0 50 100 150 disturbance rejection.
time
Fig.6 Controller output for different PID settings: integratingprocess 5 PID tuning for integrator/deadtime processes
..... A2=0.5,- A = 0.25, . - . - a. = 0.1
~

Another type of process frequently encountered is the


We observe that as A becomes larger, the response integrator plus deadtime model
time for the closed-loop system becomes shorter and
the ability to reject load disturbance is improved, how-
S
ever, the robust margin &ax becomes smaller and the
controller output becomes larger. So there is a trade-off For this type of process, if we choose the precompensa-
between control energy required and good performance tor as hlk$s, then the shaped plant is the same as for
desired. Simulation results show that to obtain good integrating processes, thus we will have the same H ,
performance, A should be chosen larger (A = 0.5 with optimal controller. However, the final H , controller
E,,, = 0.260), while to maintain robustness and take has the following structure:
less control energy, A can be chosen smaller (A = 0.1
with = 0.319). Usually we can choose A = 0.25
with = 0.289.
Example 2: To compare our rule with other rules,
consider the plant
e-o.2s
(22)
s(1 s ) + This is in the form of a PI controller plus a filter,
which is studied in [SI. Table 2 gives the PID parame- where
ters predicted by our method (with A. = 0.25) and those XK, - 0.463Xf0.277
by [5-71. Kp=--
krT2 ICT
-
yn--=
7 r
Table 2: PID controller settings for example 2
- Ti 0.238X 0.123 +
Method Kp T, Td T' T f = I = I

This paper 3.744 2.099 0.524 0.0987 T2 5.750.A + 0.590


[51 1.9997 3.029 0.670 0.2 and has the same properties as discussed above.
161 1.526 9.40 0.894 0.2 We note that the filter is used only to provide suffi-
[71 1.157 3.122 0.544 - cient robustness against high frequency uncertainties.
When the plant model is accurate it should not be used
since it can reduce the performance of the closed-loop
system.

1.6 In Table 3: PID controller settings for example 3

Method KP T,
~~

24.8602
Td

-
This paper 1.6742
#.-.- [Ill 1.6041 52.50 -
/.
[71 1.8029 24.0696 1.3235
a

^. Example 3: Consider the plant


I 1
\ I 0.0506e-6,
S

0 10 20 30 LO 50 60 70 80
which is studied in [l 11 and [5]. The PI or PID parame-
time ters predicted by our method (with 1 = 0.5) and those
Fig.7 Process output for exumple 2 by [Ill and [7] (with p = 2, 5 = 0.707) are given in
~ this paper, ..... [5],
- .- . ~ [6], - - [7]
~
Table 3. The process output is shown in Fig. 8. The
IEE Proc-Control Theory Appl., Vol. 145,No. 6, November 1998 489
proposed PI works better that that tuned by [l 13 but portional gain and integral action and thus produces a
has a larger overshoot than the PID tuned by [7]. So fast response and good load disturbance rejection. In
the derivative action can be used to obtain a better per- our tuning rules the PID parameters can be obtained
formance but, of course, it will cause larger controller using one design parameter which reflects the trade-off
output. Our method for integratoddeadtime processes between stability robustness and time domain perform-
can be used only to tune PI not PID. ance of the closed-loop system, thus making it conven-
ient to choose PID parameters in the face of
uncertainties in the processes.

7 References

1 ZIEGLER, J.G., and NICHOLS, N.B.: ‘Optimum settings for


automatic controllers’, Trans. A S M E , 1942, 62, pp. 759-768
2 COHEN, G.H., and COON, G.A.: ‘Theoretical consideration of
retarded control’, Trans. A S M E , 1953, 75, pp. 827-834
3 WANG, L., BARNES, T.J.D., and CLUETT, W.R.: ‘New fre-
quency-domain method for PID controllers’, IEE Proc. Control
Theory Appl., 1995, 142, pp. 265-271
4 ZHUANG, M., and ATHERTON, D.P.: ‘Automatic tuning of
optimum PID controllers’, IEE Proc. D, 1993, 140, pp. 216-224
5 POULIN, E., and POMERLEAU, A.: ‘PID tuning for integrat-
ing and unstable processes’, IEE Proc. Control Theory Appl.,
1996, 143, pp. 429434
6 POULIN, E., POMERLEAU, A., DESBIENS, A., and
0 I J
HODOUIN, D.: ‘Development and evaluation of an auto-tuning
0 50 100 150 200 250 300 and adaptive PID controller’, Automatica, 1996, 32, pp. 71-82
time 7 WANG, L., and CLUETT, W.R.: ‘Tuning PID controllers for
Fig.8 Process output for example 3 integrating processes’, IEE Proc. Control Theory Appl., 1997, 144,
~ this paper, - . - . - [7], ..... [l I ] pp. 385-392
8 RIVERA, D.E., MORARI, M., and SKOGESTAD, S.: ‘Inter-
nal model control. 4: PID controller design’, Ind. Eng. Chem.
6 Conclusions Process Des. Dev., 1986, 25, pp. 252-265
9 GRIMBLE, M.J.: ‘H, controllers with a PID structure’, Trans.
A S M E J. Dvnam. Svst.. Meas. Control. 1990. 112. DD. 325-336
This paper presents simple tuning rules for both stiable 10 MCFARLANE, D:C., and GLOVER, K.:’ ‘Rod& controller
and integrating processes based on loop-shaping H , design using normalized coprime factor plant descriptions’
optimal control. For FOPDT processes, our tuning (Springer-Verlag, 1990)
11 TYREUS, B.D., and LUYBEN, W.L.: ‘Tuning PI controllers for
rule turns out to be equivalent to the IMC rule, while integratoddead time processes’, Ind. Eng. Chem. Res., 1992, 31,
for integrating processes, our rule predicts suitable pro- pp. 2625-2628

490 IEE Proc-Control Theory Appl., Vol. 145,No. 6 , November 1998

You might also like