Professional Documents
Culture Documents
Abstract
In this paper a new PID controller design scheme that uses optimisation in the frequency domain is proposed for industrial process
control. An optimal-tuning PID controller is designed to satisfy a set of frequency-domain performance requirements: gain margin,
phase margin, crossover frequency and steady-state error. Using an estimated process frequency response, the method can provide
optimal PID parameters even in cases where the process dynamics are time variant. This scheme is demonstrated through its
application to a rotary hydraulic system and its performance is compared with six alternative PID tuning rules. 1999 Elsevier
Science Ltd. All rights reserved.
0967-0661/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 6 7 - 0 6 6 1 ( 9 9 ) 0 0 0 4 7 - 7
822 G.P. Liu, S. Daley / Control Engineering Practice 7 (1999) 821}830
In order to address this problem a PID controller & Landou, 1995). A summary of the six PID tuning rules
design based on direct optimisation in the frequency- are given in Table 1.
domain is developed in this paper. A set of frequency- In Table 1, K and ¹ are the inverse of the system gain
S S
domain performance functions are considered, which are and frequency at which the phase is !1803. K and
the gain-margin, phase-margin, crossover-frequency and u correspond to the gain of the system and frequency
steady-state error. The PID controller is then designed to at which the system phase is !1353. B3[1, 2] is an
directly satisfy the requirements set for these functions. acceleration factor. It can be seen from the table that the
Using a non-parametric model represented by the system six PID design methods do not require a parametric
frequency response, a method is derived that can pro- transfer function model of the process and only need
vide optimal PID parameters for di!erent working con- either one or two frequency response measurements of
ditions. This controller is demonstrated through its the process.
application to a rotary hydraulic system and its perfor-
mance evaluated through comparison with six other PID
tuning rules. 3. Optimal-tuning PID control
K e\QO
G(s)" N , (1) 3.1. Frequency response estimation
1#s¹
To estimate the system frequency response, the design
where K is the gain, ¹ the time constant and q the dead
N of the excitation signal is very important. The excitation
time. It is assumed that the ideal transfer function of
signal utilised here is the commonly used multi-sine
a PID controller is given by
which has the following form:
1 ,
K(s)"K 1# #¹ s , (2) x(t)" A sin(2n f t#u ) ,
A ¹ s B I G I
(3)
G
I
where K , ¹ and ¹ are the PID parameters. where A , u and f are the amplitude, phase and fre-
A G B I I I
Based on the FOPDT model, there are a number of quency of the signal components. The phase chosen after
PID tuning formula available. Six PID tuning rules are the desired power distribution is determined has an im-
introduced and assessed in later sections. These tuning portant in#uence on the time-domain signal shape. For
rules are Ziegler}Nichols (NZ) (Ziegler & Nichols, 1942), example, the maximal peak can be signi"cantly com-
integral of absolute error (IAE) (Pessen, 1994), some- pressed, enabling greater energy to be injected for the
overshoot rule (SOR) (Seborg et al., 1989), no-overshoot given input range of the measurement device, and the
rule (NOR) (Seborg et al., 1989), integral of squared time system is kept in the linear working region.
weighted error (ISTWE) (Zhuang & Atherton, 1993), During the identi"cation, the closed-loop system with
symmetric optimum rule (SO) (Kessler, 1958; Voda the PID controller is assumed to be of the structure
shown in Fig. 1. The multi-sine signal is added to the
Table 1 reference r(t). For this system, the unbiased estimate of
Six PID tuning rules the frequency-response of the process using spectral anal-
ysis is given by (Wellstand, 1981; McCormack & God-
PID rule PID parameters
frey, 1998)
Ziegler}Nichols K "0.60K , ¹ "0.5¹ , ¹ "0.125¹
A S G S B S PK ( ju)
Integral of absolute K "0.70K , ¹ "0.4¹ , ¹ "0.150¹
A S G S B S
GK ( ju)" WP , (4)
PK ( ju)
error SP
Some-overshoot rule K "0.33K , ¹ "0.5¹ , ¹ "0.330¹
A S G S B S
No-overshoot rule K "0.20K , ¹ "0.5¹ , ¹ "0.330¹
A S G S B S
Integral of squared K "0.509K , ¹ "0.051(3.302K K #1)¹
A S G N S S
time weighted error
n ¹ "0.125¹
B S
Symmetric optimum 4#B B 4#B B
K" , ¹" , ¹"
A 8(2 K G Bu B (4#B)u
Fig. 1. Standard PID control system.
G.P. Liu, S. Daley / Control Engineering Practice 7 (1999) 821}830 823
"K( ju)G( ju)"
input and the output, respectively. If the averaging of
(K , ¹ , ¹ )" ,
A G B G
every quantity is done in a recursive way, then, for +
k"1, 2,2
PK I( ju)
GK ( ju)" WP , (7)
LK( ju)G( ju)"!1803 ,
(10)
PK I( ju)
SP
180#LK( ju)G( ju)
(K , ¹ , ¹ )" 2! ,
where A G B P
+
k!1 1
PK I( ju)"
WP k
PK I\( ju)# RH( ju)> ( ju),
WP k I I
(8) "K( ju)G( ju)""1 ,
(11)
k!1 1
PK I( ju)" PK I\( ju)# RH( ju); ( ju) (9) u
SP k SP k I I
(K , ¹ , ¹ )" , "K( ju) G( ju)""1 , (12)
A G B 2n f
B
with PK ( ju)"0 and PK ( ju)"0. For the open-loop
WP SP
identi"cation case, the multi-sine signal is directly added 1
(K , ¹ , ¹ )" , u"1 , (13)
to the control input of the process. The estimation of the A G B e "K( ju)G( ju)"
QQ
frequency response of the process is still the same as the
above with R ( ju)"1, ∀k. Generally speaking, the accu- where
(K , ¹ , ¹ ), for i"1, 2, 3, 4, are the normalised
I G A G B
racy of the open-loop estimation is better than the gain-margin, phase-margin, crossover-frequency and the
closed-loop case. But, in practice, the latter is more con- steady-state error functions with the desired values
G , P , f and e , respectively. Thus, the following
venient for implementation than the former. + + B QQ
performance criteria should be satis"ed:
3.2. Optimal PID controller design
(K , ¹ , ¹ ))1, i"1, 2, 3, 4. (14)
G A G B
Although the six PID design methods give simple If the above inequalities are met, then the problem is
tuning rules for the controller parameters using either solved. Clearly, the design problem is to "nd a PID
one or two measurement points of the system frequency controller to make Eq. (14) hold. There are a number of
response, their control performance may not satisfy the methods to solve the performance criteria problem (14).
desired requirements. To overcome this disadvantage, an Here, two methods are brie#y introduced: the minimax
optimal PID controller design is proposed in the fre- optimisation method and the method of inequalities.
quency domain. Using the minimax optimisation method (Gill, Murray
In the frequency domain, there are two quantities used & Wright, 1981), the performance criteria (14) can be
to measure the stability margin of the system. One is the satis"ed if
gain margin, which is the factor by which the gain is less
than the neutral stability value. The other is the phase min max +
(K , ¹ , ¹ ),)1. (15)
G A G B
margin, which is the amount by which the phase of the )A 2G 2B G
system exceeds !1803 when the system gain is unity. Clearly, the above minimises the worst case values of the
The gain and phase margins are also related to the performance functions.
damping of a system. In addition to the stability of The method of inequalities (Zakian & Al-Naib, 1973;
a design, the system is also expected to meet a speed-of- Liu, 1992; Whidborne & Liu, 1993) uses optimisation
response speci"cation like bandwidth. The crossover fre- algorithms (e.g., moving boundaries algorithm) to "nd
quency, which is the frequency at which the gain is unity, the admissible or feasible set of parameter vectors, for
would be a good measurement in the frequency domain which all the performance inequalities hold. The admiss-
for the system's speed of time response. Also, the larger ible set is de"ned as
the value of the magnitude on the low-frequency asym-
ptote, the lower the steady-state errors will be for the X"X 5X 5X 5X , (16)
824 G.P. Liu, S. Daley / Control Engineering Practice 7 (1999) 821}830
where X is the set of parameter vectors for which the ith The operating procedure of the optimal-tuning PID
G
functional inequality is satis"ed, that is control is as follows. When the system's operating-point
or dynamics change, the new process frequency response
X "+(K , ¹ , ¹ ),
(K , ¹ , ¹ ))1,. (17)
G A G B G A G B is re-estimated by switching on the excitation signal.
Similar algorithms for solving the problem de"ned in (14) Then, using this updated frequency-response, the tuning
now exist in standard libraries of optimisation software, mechanism searches for the optimal parameters for the
for example, the optimisation toolbox for use with MAT- PID controller to satisfy the desired system speci"ca-
LAB (Grace, 1994). tions. Finally, the PID controller is set to the obtained
optimal parameters. In this way, the PID controller may
3.3. Optimal-tuning PID control scheme cope with all operating-points of the system and the
closed-loop system will have similar optimal-control
When a system has di!erent operating points with performance. But, compared with a "xed parameter con-
widely di!ering dynamic properties, it is not always pos- trol, the disadvantage of this strategy is that it needs
sible to exercise control with a "xed parameter control- slightly more computation to search for the optimal
ler, even if this is a highly robust controller. For this case, parameters.
the optimal-tuning PID control scheme shown in Fig. 2
is proposed. It mainly consists of four parts: frequency
response estimation, desired system speci"cations, 4. Application to a rotary hydraulic system
optimal-tuning mechanism and PID controller. The fre-
quency response estimated using frequency-domain iden- The optimal-tuning PID control scheme is applied to
ti"cation methods provides a non-parametric model for a rotary hydraulic test rig (Daley, 1987), which is repre-
the process. The desired system speci"cations includes sentative of many industrial systems that utilise #uid
a set of requirements in the frequency domain: gain power. This is a particularly opposite application of the
margin, phase margin, crossover frequency and steady- method since hydraulic systems are often very conserva-
state error. The optimal-tuning mechanism uses the pro- tively tuned, due to the fact that the cost of getting the
cess frequency response to "nd optimal parameters for tuning wrong can be highly destructive and costly. To
the PID controller so that the desired system speci"ca- e!ectively assess the performance of the proposed tuning
tions are satis"ed. method the other six tuning rules which were introduced
in Section 2 are also applied to the rig.
Fig. 7. The speed of the hydraulic motor using NZ, ISTWE, IAE and SO controllers (Case I).
Fig. 8. The speed of the hydraulic motor using SOR, NOR and ODR controllers (Case I).
estimation, a de"nition of desired system speci"cations, domain identi"cation methods provides a non-paramet-
an optimal-tuning mechanism and a PID controller. ric model for the process. The desired system speci"ca-
The frequency response estimated using frequency- tions includes a set of requirements in the frequency
828 G.P. Liu, S. Daley / Control Engineering Practice 7 (1999) 821}830
Fig. 9. The speed of the hydraulic motor without load using SOR, NOR, SO and ODR controllers for Case I.
Fig. 11. The speed of the hydraulic motor using NZ, ISTWE, IAE and SO controllers (Case II).
Fig. 12. The speed of the hydraulic motor using SOR, NOR and ODR controllers (Case II).
Acknowledgements References
The authors are grateful to the management of the Astrom, K. J., & Hagglund, T. (1984). Automatic tuning simple regu-
ALSTOM Energy Technology Centre for giving per- lators with speci"cations on phase and amplitude margins. Auto-
mission to publish this work. matica, 20(5), 645}651.
830 G.P. Liu, S. Daley / Control Engineering Practice 7 (1999) 821}830
Daley, S. (1987). Application of a fast self-tuning control algorithm to McCormack, A. S., & Godfrey, K. (1998). Rule-based autotuning based
a hydraulic test rig. Proceedings of the Institution of Mechanical on frequency domain identi"cation. IEEE Transactions on Control
Engineers, 201, 285}295. Systems Technology, 6(1), 43}61.
Daley, S., & Liu, G. P. (1998). Optimal PID tuning using direct search Pessen, D. W. (1994). A new look at PID-controller tuning. Transactions
algorithm. Tuning-in to increase proxt-developments in PID tuning, of the American Society of Mechanical Engineers, Journal of Dynamic
IMechE Seminar, London. Systems, Measurement and Control, 116, 553}557.
Hang, C. C., Astrom, K. J., & Ho, W. K. (1991). Re"nements of the Radke, F., & Isermann, R. (1987). A parameter-adaptive PID controller
Ziegler}Nichols tuning formula. IEE Proceedings-D, 138(2), 111}118. with stepwise parameter optimisation. Automatica, 23, 449}457.
Gill, P. E., Murray, W., & Wright, M. H. (1981). Practical optimisation. Seborg, D. E., Edgar, T. F., & Mellichamp, D. A. (1989). Process
New York: Academic Press. dynamics and control. New York: Wiley.
Grace, A. (1994). Optimisation toolbox for use with MATLAB. The Math- Voda, A., & Landau, I. D. (1995). A method for the auto-calibration of
Works Inc. PID controllers. Automatica, 31(1), 41}53.
Kessler, C. (1958). Das symmetrische optimum. Regelungstetechnik, Wellstead, W. E. (1981). Nonparametric methods of systems identi"ca-
6(11), 395}400. tion. Automatica, 17(1), 55}69.
Kraus, T. W., & Mayron, T. J. (1984). Self-tuning PID controllers based Whidborne, J. F., & Liu, G. P. (1993). Critical control systems: Theory,
on a pattern recognition approach. Control Engineering, 106}111. design and applications. New York: Research Studies Press Ltd and
Liu, G. P. (1992). Theory and design of critical control systems. Ph.D. Wiley.
thesis, Control Systems Centre, University of Manchester Institute Zhuang, M., & Atherton, D. P. (1993). Automatic tuning of optimum
of Science and Technology, U.K. PID controllers. IEE Proceedings-D, 140(3), 216}224.
Liu, G. P., Dixon, R., & Daley, S. (1998). Multiobjective optimal-tuning Ziegler, J. G., & Nichols, N. B. (1942). Optimum settings for automatic
PI controller design for a gasi"er. The MEC benchmark challenge on controllers. Transactions of ASME, 64, 759}768.
gasixer control, IMechE Seminar, Coventry. Zakian, V., & Al-Naib, U. (1973). Design of dynamical and control
Mantz, R. J., & Tacconi, E. J. (1989). Complementary rules to Ziegler systems by the method of inequalities. IEE Proceedings, 120(11),
and Nichols' rules for a regulating and tracking controller. Interna- 1421}1427.
tional Journal of Control, 49(5), 1465}1471.