Вы находитесь на странице: 1из 2

Introduction to the Linear Algebraic Method for C;Onfroi dy .

jg
Chi,Tsong Chen
ABSTRACT: This tutorial article discusses some basic issues in the design of control systems. The concepts of well posedness and total stability are introduced to deal with noise and disturbance problems. The implementable transfer function is developed and is shown to solve completely pole-and- ero assignment and model matching problems. Two feedbac! configurations are introduced to reali e the implementable transfer functions" and feedbac! compensation is obtained by solving sets of linear algebraic e#uations. $ntroduction %hen the open-loop plant transfer function has been specified" there are basically two approaches to carry out design. $n the first approach" we choose a feedbac! con figuration and compensation with undetermined parameters and then ad&ust the parameters so that the resulting closedloop system will meet design specifications. The root-locus and fre#uency-domain methods are ways to evaluate the ad&ustments used in this approach. $n the second approach" we choose an overall closed-loop system to meet design specifications. %e then choose an appropriate feedbac! configuration and compute the re#uired compensation. The linear #uadratic optimal control method '$( and design through pole- ero pattern are e)amples of the second approach '*(. %e call the first approach the outward approach and the second the inward approach. $n the inward approach" the first step is to choose an overall closed-loop transfer f unction to meet a set of specif ications. Because of physical constraints" this choice is not entirely arbitrary. $n this paper" we introduce four constraints on the choice of the overall transfer function: namely" properness of compensators" well posedness" total stability" and no plant lea!age. An overall transfer function that can be implemented under these four constraints is called an itn\o/elnentable transfer function. The implementable transfer function is shown to solve pole-and- ero assignment and model matching problems. 4nce an implementable overall transfer function is chosen" the ne)t step is to choose a control configuration. %e show by e)ample that the most commonly used con5guration" unity feedbac!" cannot be used to achieve pole assignment and ero assignment simultaneously. 6owever" unity feedbac! can be used to achieve arbitrary pole assignment. %e then introduce two more sophisticated configurations" which can be used to achieve pole assignment and ero assignment simultaneously: namely" the two-parameter and the plant input7output 8$749 feedbac! configurations. The compensators are obtained by solving sets of linear algebraic e#uations.

Example and Issues


:irst" we will use an e)ample to illustrate the issues that may arise in the design of control systems. Consider the plant with the open-loop transfer function ;8s9. ;8s9 < 8s - $9$'s8s - *9( 8$9

Chi-Tsong Chen is with the +epartment of ,lectrical ,ngineering" State -niversity of .ew /or!" Stony Broo!" ./ $0123.

The problem is to design an overall closedloop system such that the plant output yet9 will trac! a reference input r8t9. As can be seen" the plant transfer function is unstable and has a nonminimum-phase ero. This is a difficult problem if the root-locus method or fre#uency-domain method is used to carry out the design. $f the inward approach is used" then the first step is to select an overall closed-loop transfer function. $t is clear that an overall transfer function of unity is the best possible system we can design. $ndeed" if an overall transfer function is unity" then the plant output is identical to any reference input= the position and velocity errors are ero= and the rise time" settling time" and overshoot are also all ero. Therefore" no other transfer function can perform better than a transfer function of unity. .ote that" for a unity transfer function" the power levels at the reference input and plant output are different= otherwise" the control system would be unnecessary. 4f course" a transfer function of unity usually cannot be implemented in practice because we must use pure differentiators as compensator the actual-

0272-1708,'-87:1000-0036

$01.00 ~@ 1987 IEEE

36

IEEE Contro !y"t#rn" $magazin e

ing control signal may get very large" causing the plant to saturate. Therefore" a more realistic overall transfer function must be chosen. The calculations to obtain a realistic transfer function may be carried out by computer simulation using e)isting computeraided design pac!ages. $f we use the #uadratic performance inde) >" with r8t9 a reference step input" yet9 output" and act9 control" then the closed-loop transfer function ;o8s9 can be calculated directly '$(" '?(.

(2) ;o8s9 < 8s @ $978s* @ *.5s @ $9


(3)

Although ,#. 8?9 is optimal for the performance inde) in ,#. 8*9" its implementation will encounter difficulties because the nonminimum-phase ero of the plant transfer function ;8s9 in ,#. 8$9 does not appear in the closedAloop transfer function in ,#. 8?9. Therefore" ,#. 8?9 must be modified. This e)ample shows that the selection of an overall transfer function is not entirely arbitrary. Bhysical constraints and implementations problems must be considered in the selection of overall transfer functions. Suppose" after considering physical constraints and implementations problems" the following overall transfer function ;o8s9 in ,#. 839 is found to be acceptable:
Go(s) = -(s - I)/(S2 + 2.6s + I)

(4)

The ne)t problem is the selection of a configuration. %e rule out open-loop configuration because of its poor sensitivity. The most often used configuration is the unityfeedbac! system shown in :ig. $. $f we use

n r + e u Y

:ig. $. ,)ample system with unity feedbac!.

Вам также может понравиться