Modern Control Engineering

(Chris Devlin) #1
806 Chapter 10 / Control Systems Design in State Space

Concluding Comments on Optimal Regulator Systems


1.Given any initial state x(t 0 ),the optimal regulator problem is to find an allowable


control vector u(t)that transfers the state to the desired region of the state space


and for which the performance index is minimized. For the existence of an optimal


control vector u(t),the system must be completely state controllable.


2.The system that minimizes (or maximizes, as the case may be) the selected


performance index is, by definition, optimal. Although the controller may have


nothing to do with “optimality” in many practical applications, the important point


is that the design based on the quadratic performance index yields a stable control


system.


3.The characteristic of an optimal control law based on a quadratic performance


index is that it is a linear function of the state variables, which implies that we need


to feed back all state variables. This requires that all such variables be available for


feedback. If not all state variables are available for feedback, then we need to


employ a state observer to estimate unmeasurable state variables and use the es-


timated values to generate optimal control signals.


Note that the closed-loop poles of the system designed by the use of the


quadratic optimal regulator approach can be found from


Since these closed-loop poles correspond to the desired closed-loop poles in the


pole-placement approach, the transfer functions of the observer controllers can


be obtained from either Equation (10–74) if the observer is of full-order type or


Equation (10–108) if the observer is of minimum-order type.


4.When the optimal control system is designed in the time domain, it is desirable to


investigate the frequency-response characteristics to compensate for noise effects.


The system frequency-response characteristics must be such that the system at-


tenuates highly in the frequency range where noise and resonance of components


are expected. (To compensate for noise effects, we must in some cases either modify


the optimal configuration and accept suboptimal performance or modify the


performance index.)


5.If the upper limit of integration in the performance index Jgiven by Equation


(10–114) is finite, then it can be shown that the optimal control vector is still a


linear function of the state variables, but with time-varying coefficients. (Therefore,


the determination of the optimal control vector involves that of optimal time-


varying matrices.)


10–9 Robust Control Systems


Suppose that given a control object (i.e., a system with a flexible arm) we wish to de-


sign a control system. The first step in the design of a control system is to obtain a


mathematical model of the control object based on the physical law. Quite often the


model may be nonlinear and possibly with distributed parameters. Such a model may


be difficult to analyze. It is desirable to approximate it by a linear constant-coefficient


system that will approximate the actual object fairly well. Note that even though the


∑s I-A+BK∑= 0


Openmirrors.com

Free download pdf