A Time-varying Sliding Surface for Fast and robust tracking control of second-order uncertain systems. Proposed surface is repeatedly, instantaneously moved: shifted or / and rotated. After each instantaneous movement of the surface, the system representative point is no longer on the surface.
A Time-varying Sliding Surface for Fast and robust tracking control of second-order uncertain systems. Proposed surface is repeatedly, instantaneously moved: shifted or / and rotated. After each instantaneous movement of the surface, the system representative point is no longer on the surface.
A Time-varying Sliding Surface for Fast and robust tracking control of second-order uncertain systems. Proposed surface is repeatedly, instantaneously moved: shifted or / and rotated. After each instantaneous movement of the surface, the system representative point is no longer on the surface.
Copyright @ 1995 Elsevier Science Ltd Printed in Great Britain. All rights resewed ooo5-1098/95 $9.50 + 0.00 Technical Communique A Comment on A Time-varying Sliding Surface for Fast and Robust Tracking Control of Second-order Uncertain Systems* ANDRZEJ BARTOSZEWICZt Key Words-Variable structure control; sliding surfaces. Abstract-In a recent paper, Seung-Bok Choi, Dong-Won Park and Suhada Jayasuriya (Automatica, 30,899-904, 1994) have presented a stepwise time-varying sliding surface for variable structure control of a class of second-order uncertain systems. As the surface does not truly guarantee insensitivity of the systems to parameter variations and external disturbances, in this comment we propose a continuously time-varying surface that allows faster tracking and really guarantees robust bebaviour of the systems. 1. I ntroduction Seung-Bok Choi et al. (1994) considered a second-order nonlinear uncertain dynamic system. They introduced a time-varying sliding surface that is adaptable to arbitrary initial conditions of the system. At the beginning, the surface passes through the representative point specified by any given or measured initial errors, and then the surface is moved step by step to its predetermined, desired position. Seung-Bok Choi et al. have shown that their variable structure control scheme with the proposed surface guarantees tracking time essentially shorter than conven- tional VSC with time-invariant sliding surfaces. However, the proposed scheme does not guarantee insensitivity to parameter variations or external disturbances, and therefore it does not allow one to enforce a desired dynamic behaviour of the system. In the proposed scheme the sliding surface is repeatedly, instantaneously moved: shifted or/and rotated. As a consequence, after each movement of the surface, the system representative point is no longer on the surface, and the system is, at least for some time, not insensitive to disturbances and parameter uncertainties. In other words, after each instantaneous movement of the sliding surface, the system is in the phase of reaching of a new surface-thus it is not robust. Whether or not it actually reaches the new surface before the surface is moved again depends on the relation between dwelling time Ar, constants AL and Afs (all three of them set arbityarily), current disturbance and parameter variations. In this comment we propose to substitute the stepwise time-varying sliding surface (line) with a continuously time-varying one. Then we introduce a modified variable structure control law that satisfies sufficient conditions for the existence of a sliding mode on our surface. Actually, to achieve the sliding mode, we add to the control signal of Seung-Bok Choi et a/.: a proportional term for rotated sliding *Received 10 October 1994; revised 22 March 1995; received in final form 26 July 1995. This paper was not presented at any IFAC meeting. This paper was recom- mended for publication in revised form by Peter Dorato. Corresponding author Andrzej Bartbszewicz. Tel. (48 42) 312554: Fax (4842) 312551: E-mail ~ndpb~rt@lodzl.p.lodz.pl. t Institute of Automatic Control I-13, Technical University of Lodz, 18/22 Stefanowskiego St.. 90-924 Lodz, Poland. lines and a constant term for shifted ones. By this means, the representative point of the considered system is forced to stay always on our sliding line. Therefore the system is truly insensitive to parameter variations and external disturbances; hence its dynamic behaviour can be precisely predetermined. We also avoid setting the constants AT, Ah, A& arbitrarily, and, if required, we can further decrease the tracking time of the system. Finally, we point out a minor ambiguity in the illustrative example presented by Seung-Bok Choi et al. 2. Continuously time-varying sliding surface Consider a second-order nonlinear uncertain dynamic system described by i,(t) =X2(t), J &(t) = 2 f;(x,(th x20), t) + 5 wtlgi(~Io), X20), 0 (1) r=l i=, + W,(t), x20), OuW + 4th with given initial conditions xl@,,), x2(to), where Gai(t)gi(x,(t), x2(t), I ) and d(t) represent plant uncertainty and external disturbance respectively. Let us denote the tracking error by e(r) = [e,(t) e2WlT = h(t) - xdlW 120) - h2WlT and define the sliding line by a(eW, t) =c(t)e,O) +e2(t) +a(t), (2) where c(t) =At +B and a(t) = Ct +D, with A, B, C and D constant. Taking into account the function V(c) = :02(e(t), t), (3) which is positive definite with respect to a, and provided that uncertainties and disturbances are bounded, ( i )mi n 5 hi 5 (ai )max t (4) YI 5 d(t) 5 ~2, (5) the control law u(t) = [-(k + ,$ Ikfdxt 1)l) W ta) - ,$ ftx9 t, - ,$ Ei ( X9 t, - cx2 + cxd2 + id2 - Ae, - C I/ b(x, t), (6) where Zitx~ t, = zigi(x, t). di = l [ (ai )r ni n + (ai)msxl7 &qx, t) = a,g& t), lij = (aJmax - iii, k>max(l~,l, Ir21) (7) 1893 1894 Technical - 1' I 0 0.1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 01 Fig. 1. Phase portrait for various sliding lines: (a) conventional fixed; (b) stepwise shifted and rotated; (c) continuously shifted and rotated. ri = ati = u[Ae,(t) +c(t)e2(t) + P*(r) + C] -(~&(X,f)/)sgn(o)+d(r)-ksgn(u)]<O (8) negative-definite. Thus V is a Lyapunov function for the system (1) with control (6), the sliding mode on the surface (2) actually occurs, and the system is truly insensitive to disturbances and parameter variations. Furthermore, since in our algorithm the relation a(e(r), r) =0 holds for any r >to when the parameters A, B, C and D are constant, the error of the system is described by e(r) = [ e(r,,) +sr,, +s - $1 exp I- [$A(? - r;) +B(r - to)]} C DC --r---+-_. c c c2 Equations (l)-(8) imply that if - 0. 1' 1 0 2 4 6 6 10 Ti me [ s] Fig. 2. Position tracking error for various sliding lines: (a) conventional fixed; (b) stepwise shifted and rotated; (c) continuously shifted and rotated. Zommuniques - 1' 0 2 4 6 a Ti me [ S] J Fig. 3. Velocity tracking error for various sliding lines: (a) conventional fixed; (b) stepwise shifted and rotated; (c) continuously shifted and rotated. (i) the parameters A, B, C and D are not actually constant but rather piecewise-constant, (i) the system is controlled according to (6), and (iii) the system state belongs to the sliding line (2) at some moment T, then the condition for the state to belong to the line for all subsequent times r >T is that at any moment when the surface parameters A, B, C and D change, they do so in such a way that the representative point of the system stays on the line. An example of the surface with appropriate piecewise-constant parameters is presented in the sequel. To compare our algorithm with that presented by Seung-Bok Choi et al. (1994), we considered the nonlinear spring-damper system introduced in their illustrative example. We used the same numerical values of all signals and parameters that they did, and we simulated the system controlled with the following continuously time-varying, shifted and then rotated, sliding line: c(r) =c, =5, C =2, D =-3.5 for r <r,, where r, is the moment when ez becomes smaller than zero; A =5, B =-5r,, a(r) =0 for r, <r <r2 (i.e. after e2 becomes smaller than zero until the system representative point (RP) reaches the predetermined sliding line) and c(r) =cp =7, a(r) =0 for r >r2 (i.e. after the system RP reaches the predetermined sliding line). The time 7- 6- 5- 4- Q 3- 2- l - O- 1 -I! J 0 1 z ^ J A 4 c J Time [s] Fig. 4. Variations of c(r) for various sliding lines: (a) stepwise shifted and rotated; (c) continuously shifted and rotated. Technical Communiques 1895 - 4. - 4 - 5. - 5 - 6. - 6 _, - 6. 6026 , - 6. 9026 0 2 4 6 6 10 _, 0 2 4 Time [s] 6 a 10 Ti me [ s] Fig. 5. Control signal of the algorithm with conventional fixed sliding line. Fig. 7. Control signal of the algorithm with continuously shifted and rotated sliding line. t,=--+lnfi=O.25055s has been determined by differentiating (9) and equating it to zero, while t2 = cp/A +t, =1.65055 s. Then we compared our simulations with the results of Seung-Bok Choi et al. Figures l-7 show that, with the same maximum value of input control signal, our scheme guarantees faster error convergence and truly robust behaviour of the system. Although it can be seen on Figs 5-7 that none of these algorithms are free of chatter, this issue can be addressed without difficulty, using for example a boundary-layer controller. There is no trade-off between robustness and covergence rate in our algorithm. k=O. 301 - - - - - _ k&5 3. A remark on the illustrative example The example presented by Seung-Bok Choi et al. (1994) might suggest that their control law with dwelling time A7=0.@2s makes the system insensitive to external disturbances--see their Fig. 6. Actually, the reason why the disturbances are not visible on the figure is that in the example the control gain k =0.5 is use-d instead of k just greater than 0.3, which would be satisfactory according to their equation (7). Our Fig. 8 shows the phase portrait of the system controlled according to Seung-Bok Choi et al., with k =0.301 and k =0.5. It can be seen that k =0.301 does not guarantee system insensitivity. This leads to the conclusion that the system robustness was achieved in the example partly by increasing the magnitude of the discontinuous control gain and therefore also by increasing undesirable chattering. 2 0.4- 0.2- -0.4 0 0. 1 0.2 0.3 0.4 0.5 0.6 0.7 e1 Fig. 8. Phase portrait for stepwise time-varying sliding lines with k =0.301 and k =0.5 (AT= 0.02, AL=O.O05, AL= 0.004). 4. Conclusions -4. -5 In this comment we have substituted the stepwise time-varying sliding surface proposed by Seung-Bok Choi et al. (1994) with a continuously time-varying one, and we have proved the existence of a sliding mode on our surface. By this means, we truly guaranteed insensitivity to disturbances and/or parameter variations and we further decreased the tracking time for a class of second-order nonlinear uncertain dynamic systems. In principle, our surface is similar to that proposed in the original paper. The difference between the two surfaces lies in the way they are moved: our surface moves in a continuous manner, as opposed to the surface proposed by Seung-Bok Choi et al., which moves instantaneously. In our algorithm the trade-off between robustness and tracking time is eliminated. -6 _7 -6.9026 0 2 4 6 6 10 Time (sl Fig. 6. Control signal of the algorithm with stepwise shifted and rotated sliding line. Reference Seung-Bok Choi, Dong-Won Park and Suhada Jayasuriya (1994). A time-varvinp. slidine surface for fast and robust tracking control -of - second-order uncertain systems. Automatica, 30,89!%304.