Download as pdf
Download as pdf
You are on page 1of 2
10.2. LYAPUNOV REDESIGN 207 The stabilization result of Theorem 10.4 is dependent on the choice of (to satisfy lz) M > 0. Under this condition, the feedback control law (10.26) acts in the region lw] < ye as a high-gain feedback controller v = —w/p. Such high-gain feedback can stabilize the origin when (10.30) is satisfied. It can be shown that if 8 does not satisfy the condition (x) > So > 0, the feedback control may fail to stabilize the origin.” Example 10.5 Reconsider the pendulum equation of Example 9.4. ym, y= —sin(xy +6) — boxe tow where x; = 6~6;, 22 = 9, and the uncertain parameters by and c satisfy 0 < by < 0.2 and 0.5 < c <2. We want to stabilize the pendulum at @ = 6; by stabilizing the origin 2 = 0. This system is feedback linearizable and can be written in the form 2 + Bl sin(x1 + 61) ~ bore + cu] where the pair (A, B) represents a chain of two integrators. With nominal param- ctert by = 0 and é, a nominal stsbilizng feedback contol ean be taken a8 (2) = (2) [sin(e1 +6) — kre — bozo] where A= [ky ka] is chosen such that A— BK is Hurwitz, With w= 6(2) +» the uncertain term 3 is given by = ($$) tints = 6) has — a Assuming Ky <1, we take (2) as 20 + ex|zi| + ealzal Ble) > Bo + fg > 0 ‘The nominal closed-loop system has a Lyapunov function V(z) = 27 Px, where P is the solution of the Lyapunov equation P(A BK) +(A—BK)"P = —I. Therefore, w= 227 BYP = 2pr2x1 ~ parza) and the control w= o(x) — B(x) sat (B(x)w/'n) Tce Exercise 1420 of (74) 268 CHAPTER 10, ROBUST STATE FEEDBACK STABILIZATION achieves global ultimate boundedness with ultimate bound proportional to yl. If sin dy = 0, we take 05 = 0 and the origin of the closed-loop system will be globally exponentially stable. In Example 9.4, we analyzed the same system under the control u = 6(2). Comparing the results of the two examples shows the contribution of the additional control component v. In Example 9.4, we had to restrict the uncertainty such that oa L 4 fis Yeve] < | [ v | 2h This restriction has now boon removed, Wen sind, #0, we showed there thatthe ultimate hound is proportional to |sin(:)(e~ @)/e. With the current control, the ultimate bound is proportional to 7; henee it eam be made arbitrarily small by choosing j¢ small enough. Consider a case where the pendulum i to be stabilized a 6 \(e-a)/al, take & ~ (24 05)/2 ~ 1.25, which allows us to take ka With K = [1 2], A~ BK has multiple eigenvalues at —1 and the solution of the Lyapunov equation yields w = 21 +22. With 6 = 0.3, 6(2) is given by 5/2. To minimize B(x) = 24ers] +2.8]z2| +15 and w= 0.8(c0821 ~ 21 ~ zy) ~ (2) sat (9(z)w/a) ‘The parameter jis chosen small enough to meet the requirement on the ultimate bound. For example, suppose it is required that 2(t) be ultimately inside the set {lzi| $0.01, |z2| < 0.01}. We usc the ultimate bound provided by Theorem 10.3 with @ = 0.9 to choose x. The Lyapunov function V(x) ~ 2" Pz satisfies (10.28) with au(r) = Awin(P)r®, aa(r) = Amas(P)r?, and as(r) = r2. Hence, b(us) = a (a3(a5*(0.2540(1 — n0)/9))) OSV Choosing ff < 0.01/0.8 ensures that the ball {|x|| < 6} is inside the set {|zi| < 0.01, [zz £0.01}. A less conservative estimate of 1 can be obtained by examining the trajectories inside {|[rl| < 6}. It can be shown that the trajectories inside this set converge to an equilibrium point (2) © y(0.8e ~ 1)/(2.28c),22 = 0). With € [05,2], |a(0.8c~ 1)/(2.25e)| < 0.53. Hence, it is sufficient to choose #1 <0.01/0.58. Simulation results with by = 0.01, ¢= 0.5, and p= 0.01 are shown in Figure 10.9. Figure 10.9(a) shows the response of @ when 0(0) = 6(0) Figure 10.9(b) shows that the phase portrait resembles sliding mode control where tuajectories roach the surface w — 0 in finite time, then slide on it towards the origin. Even though the controller was not designed to make wii negative, it can be shown that when Blu) > y1, wir < —w? 4 "Recall fom Example 94 that for the sense numbers, (c~ 2)/é| has tobe lees thats 0.399 2Pamalysis inside {| 2] <0) ue the fact thet, in thio example, wi < —w" whenever | = 4 ond Sip almost constant. Therefore, the tajctorie enter the set {|

You might also like