This document presents an extension to a previous method for simultaneously generating second-order sensitivity functions for linear systems. The previous method was limited to systems with one direct path where the weighting functions depended only on plant transfer functions. The proposed extension removes this restriction, allowing the weighting functions to depend on any system parameters. An example demonstrating the extension is provided.
This document presents an extension to a previous method for simultaneously generating second-order sensitivity functions for linear systems. The previous method was limited to systems with one direct path where the weighting functions depended only on plant transfer functions. The proposed extension removes this restriction, allowing the weighting functions to depend on any system parameters. An example demonstrating the extension is provided.
This document presents an extension to a previous method for simultaneously generating second-order sensitivity functions for linear systems. The previous method was limited to systems with one direct path where the weighting functions depended only on plant transfer functions. The proposed extension removes this restriction, allowing the weighting functions to depend on any system parameters. An example demonstrating the extension is provided.
This document presents an extension to a previous method for simultaneously generating second-order sensitivity functions for linear systems. The previous method was limited to systems with one direct path where the weighting functions depended only on plant transfer functions. The proposed extension removes this restriction, allowing the weighting functions to depend on any system parameters. An example demonstrating the extension is provided.
form 6 =T +h ( U , (6) where 9; is a transport lead function cor- responding to the transport lag function ~ i . For example, in the case of a linearly varying delay T(1) Ut f b (7) where Q and b are constants, i t follows that T + b 1--a 1--a t =- 9 +( T) =-. (8) Note that for the particular case of a con- stant delay (a =O), (8) becomes t =T +- b, +(T) =b (9) which coincides with the form obtained by Ragg. However, this is practically the only case in which his work leads to an exact result, thus invalidating the main purpose of his otherwise excellent paper. With the proper change of variable speci- fied by (j ), Raggs necessary conditions (21) to (24) should now be relvritten as t o I t < - ~ ( ~ + k ) f (I la) Fa,: =0, tr - T ( ~ ~ : ) J I t I t j (I lb) for k=l , 2, e . . , Y. tween formulations, consider the system I n order to illustrate the difference be- k(1) - X ( t ) +2l(t - T) (12) where the transport l a g is given by ~ ( t ) =0.951 (13) and the performance functional 10 P =s, tU*(t)dt. (14) I t is desired to drive the system under a continuous control from the initial state x(0) =0 to the terminal state s(10) =1 with a minimum cost. The necessary conditions are found from the function F where, i n this example, F ( X , x,, i, ZI, x,, A) = ~ 2 r ~ + X( i + z - 2 c 5 ) . (15) Since F is not a function of x;, (loa) and (lob) become d - F; =F,, 0 I f 5 10 (16a) dt or i ( t ) =h(t), 0 5 t 5 10. (16b) Equati ons (l l a) and (l l b) become Fu =0, 0.5 5 t 5 10. (l i b) The superscript s has been purposely used in (17a) to point out the similarity in the form of the equations arrived at by both methods, and indicates the time at which these expressions are to be evaluated. Using Raggs solution, s =t +~ ( f ) =1.93. (184 Thus, (17a) and (17b) become .z~(f ) - - =0, 0 5 t <0.5 (18b) A(1.93) 1 - 0.95 z ~ ( t ) =0, 0.5 5 t 5 10 (18c) which leads to the optimal control and a minimum cost P=O.133. iXe propose instead s =t ++(t) =206 (204 to find that (17a) and (l i b) become .zr(t) - ___ = A (201) 1 - 0.95 0, 0 5 6 <0.5 (20b) u( t ) =0, 0.5 5 t 5 10 (204 and the optimal control with a minimum cost P =0.050, which is con- siderably smaller. I t can be noticed that since (2) requires a solution in the form of (6) instead of (4), the solution of problems of optimal control of systems with time-varying transport lags generally becomes very complicated. Fur- thermore, it can be expected that the prob- lem would become even more complex if the time lags are of the general form given in (I ). I n such a case, additional difficulties would arise in the solution (6) of (2), which leads us to doubt that the proposed general prob- lem could be easily solved for any but the most trivial cases, such as that of the above example in which the T? are simple functions of time only. J AMES F. RaN.%s A 4 ~ ~ ~ 6 G. VACROUX Dept. of Elec. Engrg. Illinois I nstitute of Technology Chicago, Ill. Authors Reply2 I wish to thank Banas and Vacroux for their comments on my paper and for point- ing out an error, Correction of this error, when the lags are functions of time only, makes quantitative changes to the limits on (21) and (23) of the paper. However, it does not invalidate the approach to the problem or the general nature of the results. Since publication of the paper, I have become aware that the derivation needs modification when the lags are functions of the states and derivatives of the states, and some further attention has been given to this. I agree that problems with time-varying transport lags are difficult to solve, but knowing some nec- 2 Manuscript received December 28, 1967. essary conditions for optimality is an im- portant step towards a greater understand- ing of such systems. In addition, on page 739, (16) should read 0; 5 Ti 5 &+I, j =0, . . . ~ Q - 1. (16) E. C. ~ G G Dept. of Control Engrg. The University of Sew South Wales Kensington, New South \t:ales, Australia A Practical Chart for an Optimal Control of a Second-Order Under- damped System Recentll- i t has become common practice to use an optimal control in a second-order control system. Simple alternative methods and charts are given in the literature. I n this correspondence, the emphasis is on using a new chart when a minimum-time optimal control for a second-order control system with respect to a step input is required. Let us consider a second-order control system described by the equation 2 +2Cw,k f wn?x =w,l u (1) for O<T <l , where 3c is the system output and u is an optimal control signal defined by the relation I m for 0 <f <t1 -m ior t l <d <bl +t n (2) z f ( f ) I 1 for t >61 +t2. The control signal u brings the system out- put from 0 to 1. The methods used to determine the times tl and t n as a function of amplitude ?I? and a damping ratio j- are well known from the theory of optimal control systems.? There- fore, this calculation is not given here. I t is rather difficult to present the results in an- alytical form and therefore, i n this paper, a charted representation, shown in Fig. 1, is preferred. The settling time versus ampli- tude w 7 when r=const (see Fig. 2) can be deduced from the chart in Fig. 1. When a second-order system is given, the charts of Figs. 1 and 2 might be used i n the following cases: 1) I n choosing the optimal control signal -the amplitude wz and the times tl and bz- when a settling time is specified. I n this case, the value of the amplitude PI? is determined from Fig. 2, and the times tl and t 2 are de- termined from Fig. l by intersecting the curve of II? =const by the curve of r=const. 2) I n establishing the settling time and the times tl and tn for a given control signal ( 2) , even if only its amplitude PZ is given. For values of ~ l t and {given, the times t~ and t n are determined from Fig. 1 and the settling time from Fig. 2. ber 21, 1967. Manuscript received J une 7, 1967; revised Decem- 1 P. I. Cinaev. Samonastraivaurxiesia sistemi.? hloszhiz. hfoscoG, 1963. Yorork: McGran-HiU. 1958. 2 0. J . M. Smith, Feedback Corrfiol Systems. Kew 210 IEEE TRANTSACTIONS ON AUTOMATIC CONTROL, APRI L 1968 Fig. 1. A chart of times t~ and t ? as a function of the amplitude m and the damping ratio r. Fig. 2. The settling time versus the amplitude 7n when r is constant. CORRESPONDENCE 21 1 3) I n choosing the optimal control signal (2) when the ratio of the values tl and t? is specified, i.e., t : =kh. I n this case, the straight line w,tn =kwJ1 intersects the curves of [=const shown in Fig. 1, and thus the amplitude m is deter- mined. Practically, it is sometimes much easier to deal with the case in which tl=t?. L. LUPAS I nstitute for &Automation Bucharest, R.S. Romania An Extension to "Simultaneous Generation of the Second-Order Sensitivity Functions" INTRODUCTION Bi ng~l ac[~I proposed a method for the simultaneous generation of second-order sensitivity functions for a class of linear sys- tems. The purpose of this correspondence is to extend the method to a more general case and to provide an example. The second-order sensitivity functions are required when we formulate performance criteria so as to minimize ~ensi ti vi ty:[~] where p =weighting factor .& __ = ay ) sensitivity of output y a In qm to parameter qm and e(t , 41, - - . , q.v) =.(t, 41, . * . , q,v) - yo(t) (2) where y o ( t ) is the desired response. Minimization of J by a choice of qt using a gradient procedure requires generation of the components Bingulac's methodl11 was proposed for linear multiloop control systems with one direct path (Fig. 1): with the restriction W ; ( S , Si ) =qi G;(~). (9 This restriction is removed in the following derivation. METHOD Consider the transfer function of the system in Fig. 1. P( s ) is the product of transfer functions in the fonvard path, and l ( 5 ) is the determinant of the system graph. From Kokotoui ~,[~] + + 4 Fig. 1. System considered. "13 Fig. 2. Results for m =l. (a) Extension. (b) Bingulac's result. Manuscript received October 17. 1967.