Nonlinear Systems and Control

Lecture # 38
Observers
High-Gain Observers
Motivating Example

– p. 1/1

γ(ˆ x)) – p. x ˜2 = x2 − x ˆ2 x ˜˙ 2 = −h2 x ˜1 + δ(x. γ(x)) Observer: x ˆ˙ 1 = x ˆ2 + h1 (y − x ˆ1 ). x ˆ˙ 2 = φ0 (ˆ x. γ(ˆ x)) − φ0 (ˆ x. x˙ 2 = φ(x. u) x ˜1 = x1 − x ˆ1 . u) + h2 (y − x ˆ1 ) φ0 (x. x ˜˙ 1 = −h1 x ˜1 + x ˜2 . y = x1 Let u = γ(x) stabilize the origin of x˙ 1 = x2 . u) is a nominal model φ(x. x ˜) δ(x. x˙ 2 = φ(x. 2/1 . x ˜) = φ(x.x˙ 1 = x2 . u).

ε (εs)2 + α1 εs + α2 ε>0 " ε εs + α1 # – p.Design H = " h1 h2 # such that Ao = " −h1 1 −h2 0 # is Hurwitz Transfer function from δ to x ˜: Go (s) = 1 s2 + h1 s + h2 " 1 s + h1 # Design H to make supω∈R kGo (jω)k as small as possible h1 = Go (s) = α1 ε . 3/1 . h2 = α2 ε2 .

4/1 .Go (s) = ε (εs)2 + α1 εs + α2 " ε εs + α1 # Observer eigenvalues are (λ1 /ε) and (λ2 /ε) where λ1 and λ2 are the roots of λ2 + α1 λ + α2 = 0 sup kGo (jω)k = O(ε) ω∈R – p.

η1 = x ˜1 εη˙ 1 = −α1 η1 + η2 . a>0 Peaking Phenomenon: x1 (0) 6= x ˆ1 (0) ⇒ η1 (0) = O(1/ε) The solution contains a term of the form 1 ε 1 ε e−at/ε e−at/ε approaches an impulse function as ε → 0 – p. x ˜) Ultimate bound of η is O(ε) η decays faster than an exponential mode e−at/ε . η2 = x ˜2 εη˙ 2 = −α2 η1 + εδ(x. ε . 5/1 .

x˙ 2 = x32 + u.Example x˙ 1 = x2 . 6/1 . y = x1 State feedback control: u = −x32 − x1 − x2 Output feedback control: u = −ˆ x32 − x ˆ1 − x ˆ2 x ˆ˙ 1 = x ˆ2 + (2/ε)(y − x ˆ1 ) x ˆ˙ 2 = (1/ε2 )(y − x ˆ1 ) – p.

03 0.01 0.5 −2 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 0 0.5 0 SFB OFB ε = 0.5 −1 −1.06 0.1 1 0 x 2 −1 −2 −3 0 u −100 −200 −300 −400 – p.02 0.08 0.07 0.09 0.04 0.05 t 0.1 OFB ε = 0.005 x 1 −0.01 OFB ε = 0. 7/1 .0.

ε = 0.08 0 0.01 0.04 0. 8/1 .2 1 0 x −0.07 0.03 0.6 0 0.02 0.04 t 0.4 −0.03 0.06 0.02 0.07 0.08 0 x 2 −200 −400 −600 2000 u 1000 0 −1000 – p.08 0 0.04 0.01 0.05 0.06 0.2 −0.03 0.06 0.05 0.05 0.07 0.01 0.004 0.02 0.

05 x2 0 −0.05 t 0.1 0.1 u 0 −0.05 −0.03 0.001 0.u = sat(−ˆ x32 − x ˆ1 − x ˆ2 ) SFB OFB ε = 0.07 0.01 0.04 0.1 OFB ε = 0.01 OFB ε = 0. 9/1 .08 0.05 0 −0.05 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 0 0.1 0.02 0.15 x1 0.5 −1 – p.09 0.06 0.

Region of attraction under state feedback: 2 x2 1 0 −1 −2 −3 −2 −1 0 x1 1 2 3 – p. 10/1 .

5 0 x1 0.5 ε = 0.5 0 −0.1 (dashed) and ε = 0.Region of attraction under outputfeedback: 1 x2 0.5 −1 −0.5 −1 −1. 11/1 .05 (dash-dot) – p.5 1 1.

γ(x − x ˜)) εη˙ 2 = −α2 η1 + εδ(x.Analysis of the closed-loop system: x˙ 1 = x2 εη˙ 1 = −α1 η1 + η2 x˙ 2 = φ(x. x ˜) η6 q D D D D D O(1/ε) q D D D D D D D O(ε) D D W W  Ωb  Ωc - x – p. 12/1 .

13/1 . kn = sup |v(t)| < ∞ t≥0 – p.What is the effect of measurement noise? The high-gain observer is an approximate differentiator Transfer function from y to x ˆ (with φ0 = 0): " # " # α2 1 + (εα1 /α2 )s 1 → as ε → 0 2 s s (εs) + α1 εs + α2 Differentiation amplifies the effect of measurement noise y = x1 + v.

3 0.04 0.2 0.7 0. 0.06 0.1 0.1 1 0.5 0.14 0. 14/1 .9 the error bound 0. t≥0 0.08 0.16 kn = sup |v(t)| t≥0 – p.02 ! .6 0.1.8 0.12 the high−gain parameter epsilon kd = sup |¨ x1 (t)|.4 0.1 εopt = O s 0 kn kd 0.