Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
66 views

Benchmark Systems For PID Control

Uploaded by

Rodrigo Schneider
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
66 views

Benchmark Systems For PID Control

Uploaded by

Rodrigo Schneider
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Copyright CS> IFAC Digital Control: Past, Present and

Future of PlO Control, Terrassa, Spain, 2000

BENCHMARK SYSTEMS FOR PID CONTROL

K. J. Astrom and T. Hiigglund


Department of Automatic Control
Lund University, Lund, Sweden

Abstract This paper describes a collection of systems that are suitable for
testing PID controllers. The systems are collected from a wide range of sources.
Copyright «J2000 [FAC
Keywords: PID control, benchmark problems.

INTRODUCTION 2. FOURTH ORDER SYSTEM


When evaluating PID controllers it is of interest Transfer function
to have a batch of test examples so that differ-
ent schemes can be evaluated. Over the years 1
G( s) - -,-----:-:-~--:-:-:--....,,----,-~____;:~
we have collected a large number of test exam- - (s + 1)(1 + as)(1 + a 2 s)(1 + a 3 s) (2)
ples which we have used for research and for a = 0.1,0.2,0.5,1.0
evaluation of commercial systems. This paper
describes a collection of systems that we have This system has four poles whose spacing is
found useful. The material is based on many determined by parameter a. For small values
different sources. The systems 1-5 are standard there are drastic improvements when going
systems that are well suited to parametric stud- from PI to PID control. For a = 1 the system is
ies. Their properties can easily be changed by identical to the system (1) for n=4.
varying a parameter. The systems 6-10 are more
specialized. They illustrate systems with vari-
ous difficulties of control. PID control is not well 3. RIGHT HALF PLANE ZERO
suited for all of them. Parameters for PID con-
Transfer function
trollers for many of these systems are found
in (Astrom and Hagglund (1995)), (Astrom G(s) _ 1- as
et al. (1998)) (Panagopoulos et al. (1999)). - (s + 1)3 (3)
a = 0.1,0.2,0.5,1,2,5
1. MULTIPLE EQUAL POLES This system has three equal poles and a right
Transfer function half plane zero. The achievable performance is
determined by parameter a. The difficulty of
control increases with increasing a.
G (s) _ -,-----1---,--- n= 1,2,3,4,8 (1)
- (s+1)n
4. TIME DELAY AND LAG
These systems are very common. For n = 1 Transfer function
and 2 anything can be achieved by PI or PID
control respectively. For large values of n the G(s) = _1_ e- s
system behaves like systems with long dead 1+sT (4)
times. The systems have been used by controller T = 0,0.1,0.2,0.5,2,5,10
manufacturers as test cases for a long time. We
have obtained this information from Eurotherm This is the classic system which has been used
and Foxboro. in many investigations of PID control. The

165
system reduces to a pure time delay for T = 0 9. OSCILLATORY SYSTEM
and represents lag dominated systems for large
Transfer function
T. Many of the early tuning rules were derived
based on this model. A drawback with the model OJ2
is that it has slow roll-off at high frequencies. G(8) = 0 2 '
(s + 1)(8 + 2( OJoS + OJ o )
2
(9)
( = 0.1, OJo = 1,2,5,10.
5. TIME DELAY AND DOUBLE LAG Systems of this type with small damping (
Transfer function are not good candidates for PID control. The
system is easy to control if OJo is large. The
performance can often be improved drastically
G(8) = (1 +18T )2 e- s (5) by more general controller structures.
T = 0,0.1,0.2,0.5,2,5,10
10. UNSTABLE POLE
This system is similar to (4) but it has more
high frequency roll off. The system reduces to a Transfer function
pure time delay for T = o. 1
G(8) = --Y--1
8 -
(10)

6. HEAT CONDUCTION This is a simple model of an inverted pendulum.


An unstable batch reactors is an example from
Transfer function
industry. Notice that particular care must be
taken with saturating actuators in this case.
(6)

This system represents the dynamics of one di- 11. SYSTEMS WITH INTEGRAL ACTION
mensional heat conduction. The ultimate gain is
k u = elC • Analog implementations of this system It is very useful to also have systems with
has been used by Eurotherm to test temperature integral action. A good collection is obtaind by
controllers. Other common industrial examples adding an integrator to the systems 1-5.
are heat exchangers.
ACKNOWLEDGEMENTS
7. FAST AND SLOW MODES We are greatful to many collegues who have
Transfer function been willing to share their experiences with
us particularly Dr M. Sommerville from Eu-
rotherm, Dr. E. Bristol from Foxboro, Dr S. Ya-
G( ) 100 ( 1 0.5) (7)
s = (8 + 10)2 8 + 1 + 8 + 0.05 mamoto from Yokogawa, Dr Shigemasa from
Thshiba, Prof. D. Atherthon University of Sus-
The essential dynamics of this system has a sex, Prof N. Munro of UMIST and Prof C. C.
fast mode with time constant 1 and moderate Hang of Singapore National University.
gain (1) and a slow pole with time constant
20 and a large gain (10). Simple tuning rules
based on the step response do normally not give REFERENCES
good tuning for systems of this type because it Astrom, K J. and T. Hagglund (1995): PIn Con-
is difficult to get a good estimate of the gain and trollers: Theory, Design, and Tuning. Instru-
the time constant. ment Society of America, Research Triangle
Park, North Carolina.

8. CONDITIONALLY STABLE SYSTEM Astrom, K J., H. Panagopoulos, and T. Hag-


glund (1998): "Design of PI controllers based
Transfer function on non-convex optimization." Automatica,
35:5.
G s _ (8 + 6)2 Panagopoulos, H., K J. Astrom, and T. Hag-
(8)
()- 8(8+1)2(8+36)
glund (1999): "Design of PID controllers
based on constrained optimization." In 1999
This system is conditionally stable. The stability American Control Conference. San Diego,
region under PI control consists of two disjoint California.
sets.

166

You might also like