Professional Documents
Culture Documents
Lecture # 1
Introduction
– p. 1/1
Nonlinear State Model
ẋ1 = f1 (t, x1 , . . . , xn , u1 , . . . , up )
ẋ2 = f2 (t, x1 , . . . , xn , u1 , . . . , up )
.. ..
. .
ẋn = fn (t, x1 , . . . , xn , u1 , . . . , up )
– p. 2/1
x1 f1 (t, x, u)
u1
x2
f2 (t, x, u)
u
2
.. ..
x= . , u = , f (t, x, u) = .
.
..
.. ..
.
.
u p
xn fn (t, x, u)
ẋ = f (t, x, u)
– p. 3/1
ẋ = f (t, x, u)
y = h(t, x, u)
ẋ = A(t)x + B(t)u
y = C(t)x + D(t)u
ẋ = f (t, x)
ẋ = f (x)
Time-Invariant System:
ẋ = f (x, u)
y = h(x, u)
– p. 5/1
Existence and Uniqueness of Solutions
ẋ = f (t, x)
|f (y) − f (x)|
≤L
|y − x|
– p. 9/1
Example:
ẋ = −x2
f (x) = −x2 is locally Lipschitz for all x
1
x(0) = −1 ⇒ x(t) =
(t − 1)
x(t) → −∞ as t → 1
the solution has a finite escape time at t = 1
– p. 11/1
Lemma: Let f (t, x) be piecewise continuous in t and
globally Lipschitz in x for all t ∈ [t0 , t1 ]. Then, the state
equation ẋ = f (t, x), with x(t0 ) = x0 , has a unique
solution over [t0 , t1 ]
– p. 12/1
Lemma: Let f (t, x) be piecewise continuous in t and
locally Lipschitz in x for all t ≥ t0 and all x in a domain
D ⊂ Rn . Let W be a compact subset of D , and suppose
that every solution of
– p. 13/1
Example:
ẋ = −x3 = f (x)
f (x) is locally Lipschitz on R, but not globally Lipschitz
because f ′ (x) = −3x2 is not globally bounded
– p. 14/1
Equilibrium Points
x(t0 ) = x∗ ⇒ x(t) ≡ x∗ , ∀ t ≥ t0
f (x) = 0
– p. 15/1
A linear system ẋ = Ax can have an isolated equilibrium
point at x = 0 (if A is nonsingular) or a continuum of
equilibrium points in the null space of A (if A is singular)
Limit cycles
Chaos
– p. 18/1
Nonlinear Systems and Control
Lecture # 2
Examples of Nonlinear Systems
– p. 1/1
Pendulum Equation
l
θ
•
mg
x1 = θ, x2 = θ̇
– p. 2/1
ẋ1 = x2
g k
ẋ2 = − sin x1 − x2
l m
Equilibrium Points:
0 = x2
g k
0 = − sin x1 − x2
l m
(nπ, 0) for n = 0, ±1, ±2, . . .
Nontrivial equilibrium points at (0, 0) and (π, 0)
– p. 3/1
Pendulum without friction:
ẋ1 = x2
g
ẋ2 = − sin x1
l
Pendulum with torque input:
ẋ1 = x2
g k 1
ẋ2 = − sin x1 − x2 + T
l m ml2
– p. 4/1
Tunnel-Diode Circuit
iL L
s i,mA
+ vL
X
X
1
CC iC CC iR i=h(v)
R
P
P
P
P
P
P
P
P
0.5
P
P
+ +
vC C J
J
J
vR 0
E
−0.5
(a) 0 0.5 1 v,V
(b)
dvC diL
iC = C , vL = L
dt dt
x1 = vC , x2 = iL , u=E
– p. 5/1
iC + iR − iL = 0 ⇒ iC = −h(x1 ) + x2
1
ẋ1 = [−h(x1 ) + x2 ]
C
1
ẋ2 = [−x1 − Rx2 + u]
L
Equilibrium Points:
0 = −h(x1 ) + x2
0 = −x1 − Rx2 + u
– p. 6/1
E 1
h(x1 ) = − x1
R R
i
R
1.2
1
0.8
Q
Q 2
0.6 1
0.4
Q
0.2 3
0
0 0.5 1 v
R
– p. 7/1
Mass–Spring System
Ff
Fsp
B B B m F-
B B B p -y
mÿ + Ff + Fsp = F
Sources of nonlinearity:
Nonlinear spring restoring force Fsp = g(y)
v v
(a) (b)
F Ff
f
v v
(c) (d)
(a) Coulomb friction; (b) Coulomb plus linear viscous friction; (c) static, Coulomb, and linear
viscous friction; (d) static, Coulomb, and linear viscous friction—Stribeck effect
– p. 10/1
Negative-Resistance Oscillator
i
X
X
+ i = h(v)
C L
v Resistive
Element
CC
iC CC
iL v
(a)
(b)
dv 1
Z t
C + v(s) ds + h(v) = 0
dt L −∞
d2 v ′ dv
CL + v + Lh (v) =0
dt2 dt
√
τ = t/ CL
dv √ dv d2 v d2 v
= CL , = CL
dτ dt dτ 2 dt2
– p. 12/1
Denote the derivative of v with respect to τ by v̇
′
p
v̈ + εh (v)v̇ + v = 0, ε = L/C
h(v) = −v + 13 v 3
v̈ − ε(1 − v 2 )v̇ + v = 0
State model: x1 = v, x2 = v̇
ẋ1 = x2
ẋ2 = −x1 − εh′ (x1 )x2
– p. 13/1
Another State Model: z1 = iL , z2 = vC
1
ż1 = z2
ε
ż2 = −ε[z1 + h(z2 )]
x1 = v = z2
√
r
dv dv L
x2 = = CL = [−iL − h(vC )]
dτ dt C
= ε[−z1 − h(z2 )]
" # " #
−h(x1 ) − 1ε x2 −1 z2
T (x) = , T (z) =
x1 −εz1 − εh(z2 )
– p. 14/1
Adaptive Control
P lant : ẏp = ap yp + kp u
– p. 15/1
Adaptive Law (gradient algorithm):
θ̇1 = −γ(yp − ym )r
θ̇2 = −γ(yp − ym )yp , γ>0
– p. 17/1
Nonlinear Systems and Control
Lecture # 3
Second-Order Systems
– p. 1/?
ẋ1 = f1 (x1 , x2 ) = f1 (x)
ẋ2 = f2 (x1 , x2 ) = f2 (x)
– p. 2/?
Vector Field diagram
x + fq (x) = (3, 2)
f (x)
*
q
x = (1, 1)
x1
Repeat at every point in a grid covering the plane
– p. 3/?
6
2
2
0
x
−2
−4
−6
−5 0 5
x
1
ẋ = f (x), x(0) = x0
ẋ = −f (x), x(0) = x0
x(t) = M z(t)
ż = Jr z(t)
– p. 6/?
Case 1. Both eigenvalues are real: λ1 6= λ2 6= 0
M = [v1 , v2 ]
ż1 = λ1 z1 , ż2 = λ2 z2
– p. 7/?
λ2 < λ1 < 0
z1
Stable Node
λ2 > λ1 > 0
Reverse arrowheads
Reverse arrowheads =⇒ Unstable Node – p. 9/?
x2 v2 x2 v2
v1 v1
x1 x1
(a) (b)
– p. 10/?
λ2 < 0 < λ1
Saddle
– p. 11/?
z2 x2
v1
v2
z1 x1
(a)
(b)
– p. 12/?
Case 2. Complex eigenvalues: λ1,2 = α ± jβ
−1 z2
q
r = z12 + z22 , θ = tan
z1
r(t) = r0 eαt and θ(t) = θ0 + βt
α < 0 ⇒ r(t) → 0 as t → ∞
α > 0 ⇒ r(t) → ∞ as t → ∞
α = 0 ⇒ r(t) ≡ r0 ∀ t
– p. 13/?
z2 (a) z2 (b) z
2 (c)
z1 z1 z1
x2 x2 x2
(a) (b) (c)
x1 x x1
1
– p. 14/?
Effect of Perturbations
– p. 15/?
A center is not structurally stable
" #
µ 1
−1 µ
Eigenvalues = µ ± j
µ < 0 ⇒ Stable Focus
µ > 0 ⇒ Unstable Focus
– p. 16/?
Nonlinear Systems and Control
Lecture # 4
Qualitative Behavior Near
Equilibrium Points
&
Multiple Equilibria
– p. 1/?
The qualitative behavior of a nonlinear system near an
equilibrium point can take one of the patterns we have seen
with linear systems. Correspondingly the equilibrium points
are classified as stable node, unstable node, saddle, stable
focus, unstable focus, or center
– p. 2/?
Let p = (p1 , p2 ) be an equilibrium point of the system
– p. 3/?
f1 (p1 , p2 ) = f2 (p1 , p2 ) = 0
y1 = x1 − p1 y2 = x2 − p2
ẏ ≈ Ay
∂f ∂f1
1
a11 a12 ∂x1 ∂x2
∂f
A= = =
∂x x=p
∂f2 ∂f2
a21 a22 ∂x1 ∂x2
x=p
– p. 4/?
Eigenvalues of A Type of equilibrium point
of the nonlinear system
λ2 < λ1 < 0 Stable Node
λ2 > λ1 > 0 Unstable Node
λ2 < 0 < λ1 Saddle
α ± jβ, α < 0 Stable Focus
α ± jβ, α > 0 Unstable Focus
±jβ Linearization Fails
– p. 5/?
Example
ẋ1 = −x2 − µx1 (x21 + x22 )
ẋ2 = x1 − µx2 (x21 + x22 )
x = 0 is an equilibrium point
" #
∂f −µ(3x21 + x22 ) −(1 + 2µx1 x2 )
=
∂x (1 − 2µx1 x2 ) −µ(x21 + 3x22 )
" #
∂f 0 −1
A= =
∂x x=0
1 0
ẋ1 = 0.5[−h(x1 ) + x2 ]
ẋ2 = 0.2(−x1 − 1.5x2 + 1.2)
i
R
1
0.2 Q
3 Q3 = (0.884, 0.21)
0
0 0.5 1 v
R
– p. 8/?
" #
∂f −0.5h′ (x1 ) 0.5
=
∂x −0.2 −0.3
" #
−3.598 0.5
A1 = , Eigenvalues : − 3.57, −0.33
−0.2 −0.3
" #
1.82 0.5
A2 = , Eigenvalues : 1.77, −0.25
−0.2 −0.3
" #
−1.427 0.5
A3 = , Eigenvalues : − 1.33, −0.4
−0.2 −0.3
1.5
1
y
0.5
−0.5
−0.5 0 0.5 1 1.5
x
– p. 10/?
Hysteresis characteristics of the tunnel-diode circuit
u = E, y = vR
i y
1.2
R
F
1 1 C
0.8 0.8
Q Q 0.6
0.6 1 2
D
0.4 0.4
0.2 Q 0.2
3 B
E A
0
0 0 1 2 3
0 0.5 1 v
R u
– p. 11/?
Nonlinear Systems and Control
Lecture # 5
Limit Cycles
– p. 1/?
Oscillation: A system oscillates when it has a nontrivial
periodic solution
x(t + T ) = x(t), ∀ t ≥ 0
−1 z2 (0)
q
r0 = z12 (0) + z22 (0), θ0 = tan
z1 (0)
– p. 2/?
The linear oscillation is not practical because
It is not structurally stable. Infinitesimally small
perturbations may change the type of the equilibrium
point to a stable focus (decaying oscillation) or unstable
focus (growing oscillation)
– p. 3/?
1
Limit Cycles:
i
X
X
+ i = h(v)
C L
v Resistive
Element
CC
iC CC
iL v
(a)
(b)
– p. 4/?
ẋ1 = x2
ẋ2 = −x1 − εh′ (x1 )x2
λ2 + εh′ (0)λ + 1 = 0
h′ (0) < 0 ⇒ Unstable Focus or Unstable Node
– p. 5/?
Energy Analysis:
E = 12 CvC
2
+ 12 Li2L
1
vC = x1 and iL = −h(x1 ) − x2
ε
E = 12 C{x21 + [εh(x1 ) + x2 ]2 }
– p. 6/?
−a x
b 1
Ė = −εCx1 h(x1 )
– p. 7/?
Example: Van der Pol Oscillator
ẋ1 = x2
ẋ2 = −x1 + ε(1 − x21 )x2
4 x2 x
3 2
3
2 2
1 1
0 0
x x
1 −1 1
−1
−2 −2
−3
−2 0 2 4 −2 0 2 4
(a) (b)
ε = 0.2 ε=1
– p. 8/?
1
ż1 = z2
ε
ż2 = −ε(z1 − z2 + 13 z23 )
3
10 x z
2 2
2
5 1
0
0 z
−1 1
x
1
−5 −2
−3
−5 0 5 10 −2 0 2
(a) (b)
ε=5
– p. 9/?
x x
2 2
x x
1 1
(a) (b)
– p. 10/?
Example: Wien-Bridge Oscillator
s R1
B B B
C1
+ v1
#
BB BB BB
"!
+ +
v2 C2 R2 gv
P
P
( 2)
P
P
P
P
P
P
P
P
Equivalent Circuit
– p. 11/?
State variables x1 = v1 and x2 = v2
1
ẋ1 = [−x1 + x2 − g(x2 )]
C1 R1
1 1
ẋ2 = − [−x1 + x2 − g(x2 )] − x2
C2 R1 C2 R2
There is a unique equilibrium point at x = 0
Numerical data: C1 = C2 = R1 = R2 = 1
– p. 12/?
x ’ = − x + y − (3.234 y − 2.195 y3 + 0.666 y5)
y ’ = − ( − x + y − (3.234 y − 2.195 y3 + 0.666 y5)) − y
0.8
0.6
0.4
0.2
0
y
−0.2
−0.4
−0.6
−0.8
−1
– p. 13/?
x ’ = − x + y − (3.234 y − 2.195 y3 + 0.666 y5)
y ’ = − ( − x + y − (3.234 y − 2.195 y3 + 0.666 y5)) − y
0
y
−2
−4
−6
−6 −4 −2 0 2 4 6
x
– p. 14/?
Nonlinear Systems and Control
Lecture # 6
Bifurcation
– p. 1/?
Bifurcation is a change in the equilibrium points or periodic
orbits, or in their stability properties, as a parameter is
varied
Example
ẋ1 = µ − x21
ẋ2 = −x2
– p. 2/?
√
Linearization at ( µ, 0):
" √ #
−2 µ 0
0 −1
√
( µ, 0) is a stable node
√
Linearization at (− µ, 0):
" √ #
2 µ 0
0 −1
√
(− µ, 0) is a saddle
– p. 3/?
ẋ1 = µ − x21 , ẋ2 = −x2
x x x
2 2 2
x x x
1 1 1
– p. 4/?
µ is called the bifurcation parameter and µ = 0 is the
bifurcation point
Bifurcation Diagram
µ
(a) Saddle−node bifurcation
– p. 5/?
Example
ẋ1 = µx1 − x21 , ẋ2 = −x2
Two equilibrium points at (0, 0) and (µ, 0)
" #
µ 0
The Jacobian at (0, 0) is
0 −1
– p. 6/?
While the equilibrium points persist through the bifurcation
point µ = 0, (0, 0) changes from a stable node to a saddle
and (µ, 0) changes from a saddle to a stable node
µ µ
(a) Saddle−node bifurcation (b) Transcritical bifurcation
– p. 7/?
Example
ẋ1 = µx1 − x31 , ẋ2 = −x2
For µ < 0, there is a stable node at the origin
µ
(c) Supercritical pitchfork bifurcation
– p. 8/?
Example
ẋ1 = µx1 + x31 , ẋ2 = −x2
For µ < 0, there are three equilibrium points: a stable node
√
at (0, 0) and two saddles at (± −µ, 0)
µ
(d) Subcritical pitchfork bifurcation
– p. 9/?
Notice the difference between supercritical and subcritical
pitchfork bifurcations
µ µ
(c) Supercritical pitchfork bifurcation (d) Subcritical pitchfork bifurcation
– p. 10/?
Example: Tunnel diode Circuit
1
ẋ1 = [−h(x1 ) + x2 ]
C
1
ẋ2 = [−x1 − Rx2 + µ]
L
x2
x = h(x )
2 1
A B x1 A B µ
(a) (b)
– p. 11/?
Example
– p. 12/?
ṙ = µr − r 3 and θ̇ = 1
√
For µ > 0, there is a stable limit cycle at r = µ
x x
2 2
x x
1 1
µ<0 µ>0
– p. 13/?
Example
2 2 2 2 2
ẋ1 = x1 µ + (x1 + x2 ) − (x1 + x2 ) − x2
2 2 2 2 2
ẋ2 = x2 µ + (x1 + x2 ) − (x1 + x2 ) + x1
– p. 14/?
ṙ = µr + r 3 − r 5 and θ̇ = 1
Sketch of µr + r 3 − r 5 :
r r
µ<0 µ>0
– p. 15/?
As µ increases from negative to positive values, the stable
focus at the origin merges with the unstable limit cycle and
bifurcates into unstable focus
µ µ
(e) Supercritical Hopf bifurcation (f) Subcrtitical Hopf bifurcation
– p. 16/?
All six types of bifurcation occur in the vicinity of an
equilibrium point. They are called local bifurcations
ẋ1 = x2
ẋ2 = µx2 + x1 − x21 + x1 x2
– p. 17/?
x x
2 2
µ=−0.95 µ=−0.88
x x
1 1
x2 x2
µ=−0.8645 µ=−0.8
x x
1 1
– p. 18/?
Nonlinear Systems and Control
Lecture # 7
Stability of Equilibrium Points
Basic Concepts & Linearization
– p. 1/?
ẋ = f (x)
y = x − x̄
def
ẏ = ẋ = f (x) = f (y + x̄) = g(y), where g(0) = 0
– p. 2/?
Definition: The equilibrium point x = 0 of ẋ = f (x) is
stable if for each ε > 0 there is δ > 0 (dependent on ε)
such that
– p. 3/?
First-Order Systems (n = 1)
x x x
– p. 4/?
The origin is stable if and only if xf (x) ≤ 0 in some
neighborhood of the origin
x x x
– p. 5/?
The origin is asymptotically stable if and only if xf (x) < 0
in some neighborhood of the origin
f(x) f(x)
−a b x x
(a) (b)
– p. 6/?
Definition: Let the origin be an asymptotically stable
equilibrium point of the system ẋ = f (x), where f is a
locally Lipschitz function defined over a domain D ⊂ Rn
( 0 ∈ D)
The region of attraction (also called region of
asymptotic stability, domain of attraction, or basin) is the
set of all points x0 in D such that the solution of
ẋ = f (x), x(0) = x0
– p. 7/?
Second-Order Systems (n = 2)
– p. 8/?
Example: Tunnel Diode Circuit
x ’ = 0.5 ( − 17.76 x + 103.79 x2 − 229.62 x3 + 226.31 x4 − 83.72 x5 + y)
y ’ = 0.2 ( − x − 1.5 y + 1.2)
1.5
1
y
0.5
−0.5
−0.5 0 0.5 1 1.5
x
– p. 9/?
Example: Pendulum Without Friction
x’=y
y ’ = − 10 sin(x)
0
y
−2
−4
−6
−8
−4 −3 −2 −1 0 1 2 3 4
x
– p. 10/?
Example: Pendulum With Friction
x’=y
y ’ = − 10 sin(x) − y
0
y
−2
−4
−6
−8
−4 −3 −2 −1 0 1 2 3 4
x
– p. 11/?
Linear Time-Invariant Systems
ẋ = Ax
x(t) = exp(At)x(0)
P −1 AP = J = block diag[J1 , J2 , . . . , Jr ]
λi 1 0 ... ... 0
0 λ 1 0 . . . 0
i
. ... ..
.. .
Ji = .
. . . .
. 0
..
. . . . 1
0 . . . . . . . . . 0 λi m×m
– p. 12/?
r X
X mi
exp(At) = P exp(Jt)P −1 = tk−1 exp(λi t)Rik
i=1 k=1
– p. 13/?
Theorem: The equilibrium point x = 0 of ẋ = Ax is stable if
and only if all eigenvalues of A satisfy Re[λi ] ≤ 0 and for
every eigenvalue with Re[λi ] = 0 and algebraic multiplicity
qi ≥ 2, rank(A − λi I) = n − qi , where n is the dimension
of x. The equilibrium point x = 0 is globally asymptotically
stable if and only if all eigenvalues of A satisfy Re[λi ] < 0
k ≥ 1, λ > 0
– p. 14/?
Exponential Stability
– p. 15/?
Example
ẋ = −x3
The origin is asymptotically stable
x(0)
x(t) = p
1 + 2tx2 (0)
e2λt
|x(t)| ≤ ke−λt |x(0)| ⇒ ≤ k2
1 + 2tx2 (0)
e2λt
Impossible because lim =∞
t→∞ 1+ 2tx2 (0)
– p. 16/?
Linearization
ẋ = f (x), f (0) = 0
f is continuously differentiable over D = {kxk < r}
∂f
J(x) = (x)
∂x
h(σ) = f (σx) for 0 ≤ σ ≤ 1
h′ (σ) = J(σx)x
Z 1
h(1) − h(0) = h′ (σ) dσ, h(0) = f (0) = 0
0
Z 1
f (x) = J(σx) dσ x
0
– p. 17/?
Z 1
f (x) = J(σx) dσ x
0
G(x) → 0 as x → 0
This suggests that in a small neighborhood of the origin we
can approximate the nonlinear system ẋ = f (x) by its
linearization about the origin ẋ = Ax
– p. 18/?
Theorem:
The origin is exponentially stable if and only if
Re[λi ] < 0 for all eigenvalues of A
– p. 19/?
Nonlinear Systems and Control
Lecture # 8
Lyapunov Stability
– p. 1/1
Let V (x) be a continuously differentiable function defined in
a domain D ⊂ Rn ; 0 ∈ D . The derivative of V along the
trajectories of ẋ = f (x) is
n n
X ∂V X ∂V
V̇ (x) = ẋi = fi (x)
i=1
∂xi i=1
∂xi
f1 (x)
i f2 (x)
h
∂V ∂V ∂V
= ∂x1
, ∂x2
, ... , ∂xn
..
.
fn (x)
∂V
= f (x)
∂x
– p. 2/1
If φ(t; x) is the solution of ẋ = f (x) that starts at initial
state x at time t = 0, then
d
V̇ (x) = V (φ(t; x))
dt t=0
– p. 3/1
Lyapunov’s Theorem:
If there is V (x) such that
V̇ (x) ≤ 0, ∀x∈D
then the origin is a stable
Moreover, if
– p. 4/1
Furthermore, if V (x) > 0, ∀ x 6= 0,
kxk → ∞ ⇒ V (x) → ∞
– p. 5/1
Solutions starting in Ωβ stay in Ωβ because V̇ (x) ≤ 0 in Ωβ
lim V (x(t)) = c ≥ 0
t→∞
c 1 <c 2 <c 3
– p. 9/1
Why do we need the radial unboundedness condition to
show global asymptotic stability?
It ensures that Ωc = {x ∈ Rn | V (x) ≤ c} is bounded for
every c > 0
Without it Ωc might not bounded for large c
Example
x21 2
V (x) = + x 2
1 + x21
x2 c
c
c
c
c
c
hhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh
c
c x1
c
c
c
c
c
– p. 10/1
Nonlinear Systems and Control
Lecture # 9
Lyapunov Stability
– p. 1/1
Quadratic Forms
n X
X n
V (x) = xT P x = pij xi xj , P = PT
i=1 j=1
P A + AT P = −Q
Matlab: P = lyap(A′ , Q)
– p. 3/1
Theorem A matrix A is Hurwitz if and only if for any
Q = QT > 0 there is P = P T > 0 that satisfies the
Lyapunov equation
P A + AT P = −Q
– p. 4/1
Linearization
ẋ = f (x) = [A + G(x)]x
G(x) → 0 as x → 0
Suppose A is Hurwitz. Choose Q = QT > 0 and solve the
Lyapunov equation P A + AT P = −Q for P . Use
V (x) = xT P x as a Lyapunov function candidate for
ẋ = f (x)
– p. 6/1
We can use V (x) = xT P x to estimate the region of
attraction
– p. 7/1
Example
ẋ1 = −x2
ẋ2 = x1 + (x21 − 1)x2
" #
∂f
0 −1
A= =
∂x x=0
1 −1
√
has eigenvalues (−1 ± j 3)/2. Hence the origin is
asymptotically stable
" #
T 1.5 −0.5
Take Q = I, P A + A P = −I ⇒ P =
−0.5 1
λmin (P ) = 0.691
– p. 8/1
V (x) = xT P x = 1.5x21 − x1 x2 + x22
2 2 def 2
V̇ (x) < 0 for 0 < kxk < √ = r
5
2 2
Take c = λmin (P )r = 0.691 × √ = 0.618
5
{V (x) < c} is an estimate of the region of attraction
– p. 9/1
Example:
ẋ = −g(x)
g(0) = 0; xg(x) > 0, ∀ x 6= 0 and x ∈ (−a, a)
Z x
V (x) = g(y) dy
0
∂V
V̇ (x) = [−g(x)] = −g 2 (x) < 0, ∀ x ∈ (−a, a), x 6= 0
∂x
The origin is asymptotically stable
– p. 11/1
Example: Pendulum equation without friction
ẋ1 = x2
ẋ2 = − a sin x1
– p. 12/1
Example: Pendulum equation with friction
ẋ1 = x2
ẋ2 = − a sin x1 − bx2
1
V (x) = a(1 − cos x1 ) + x22
2
V̇ (x) = aẋ1 sin x1 + x2 ẋ2 = − bx22
The origin is stable
– p. 13/1
The conditions of Lyapunov’s theorem are only sufficient.
Failure of a Lyapunov function candidate to satisfy the
conditions for stability or asymptotic stability does not mean
that the equilibrium point is not stable or asymptotically
stable. It only means that such stability property cannot be
established by using this Lyapunov function candidate
Try
1 T
V (x) = 2x P x +" a(1 − cos#x"1 ) #
1 p11 p12 x1
= 2 [x1 x2 ] + a(1 − cos x1 )
p12 p22 x2
– p. 14/1
V̇ (x) = (p11 x1 + p12 x2 + a sin x1 ) x2
+ (p12 x1 + p22 x2 ) (−a sin x1 − bx2 )
= a(1 − p22 )x2 sin x1 − ap12 x1 sin x1
+ (p11 − p12 b) x1 x2 + (p12 − p22 b) x22
D = {x ∈ R2 | |x1 | < π}
V (x) is positive definite and V̇ (x) is negative definite over D
The origin is asymptotically stable
– p. 15/1
Nonlinear Systems and Control
Lecture # 10
The Invariance Principle
– p. 1/1
Example: Pendulum equation with friction
ẋ1 = x2
ẋ2 = − a sin x1 − bx2
1
V (x) = a(1 − cos x1 ) + x22
2
V̇ (x) = aẋ1 sin x1 + x2 ẋ2 = − bx22
The origin is stable. V̇ (x) is not negative definite because
V̇ (x) = 0 for x2 = 0 irrespective of the value of x1
– p. 2/1
Definitions: Let x(t) be a solution of ẋ = f (x)
– p. 3/1
A set M is an invariant set with respect to ẋ = f (x) if
x(0) ∈ M ⇒ x(t) ∈ M, ∀ t ∈ R
Examples:
Equilibrium points
Limit Cycles
x(0) ∈ M ⇒ x(t) ∈ M, ∀ t ≥ 0
– p. 4/1
The distance from a point p to a set M is defined by
dist(p, M ) = inf kp − xk
x∈M
– p. 5/1
Lemma: If a solution x(t) of ẋ = f (x) is bounded and
belongs to D for t ≥ 0, then its positive limit set L+ is a
nonempty, compact, invariant set. Moreover, x(t)
approaches L+ as t → ∞
– p. 6/1
Proof:
⇒ lim V (x(t)) = a
t→∞
– p. 7/1
V (x) = a on L+ and L+ invariant ⇒ V̇ (x) = 0, ∀ x ∈ L+
L+ ⊂ M ⊂ E ⊂ Ω
x(t) approaches L+ ⇒ x(t) approaches M (as t → ∞)
– p. 8/1
Theorem: Let f (x) be a locally Lipschitz function defined
over a domain D ⊂ Rn ; 0 ∈ D . Let V (x) be a continuously
differentiable positive definite function defined over D such
that V̇ (x) ≤ 0 in D . Let S = {x ∈ D | V̇ (x) = 0}
If no solution can stay identically in S , other than the
trivial solution x(t) ≡ 0, then the origin is asymptotically
stable
– p. 9/1
Example:
ẋ1 = x2
ẋ2 = −h1 (x1 ) − h2 (x2 )
hi (0) = 0, yhi (y) > 0, for 0 < |y| < a
Z x1
V (x) = h1 (y) dy + 21 x22
0
V̇ (x) = h1 (x1 )x2 +x2 [−h1 (x1 )−h2 (x2 )] = −x2 h2 (x2 ) ≤ 0
V̇ (x) = 0 ⇒ x2 h2 (x2 ) = 0 ⇒ x2 = 0
S = {x ∈ D | x2 = 0}
– p. 10/1
ẋ1 = x2 , ẋ2 = −h1 (x1 ) − h2 (x2 )
H Load
H
q
2
r
q
r 1
– p. 12/1
M (q)q̈ + C(q, q̇)q̇ + D q̇ + g(q) = u
u = g(q) − Kp (q − q ∗ ) − Kd q̇
e = q − q∗ , ė = q̇
M ë = M q̈
= −C q̇ − D q̇ − g(q) + u
= −C q̇ − D q̇ − Kp (q − q ∗ ) − Kd q̇
= −C ė − D ė − Kp e − Kd ė
– p. 14/1
M ë = −C ė − D ė − Kp e − Kd ė
V = 21 ėT M (q)ė + 21 eT Kp e
V̇ = ėT M ë + 12 ėT Ṁ ė + eT Kp ė
1 T
= 2 ė (Ṁ − 2C)ė − ėT (Kd + D)ė
– p. 15/1
(Kd + D) is positive definite
M ë = −C ė − D ė − Kp e − Kd ė
ė(t) ≡ 0 ⇒ ë(t) ≡ 0 ⇒ Kp e(t) ≡ 0 ⇒ e(t) ≡ 0
By LaSalle’s theorem the origin (e = 0, ė = 0) is globally
asymptotically stable
– p. 16/1
Nonlinear Systems and Control
Lecture # 11
Exponential Stability
&
Region of Attraction
– p. 1/1
Exponential Stability:
The origin of ẋ = f (x) is exponentially stable if and only if
the linearization of f (x) at the origin is Hurwitz
{k1 kxka ≤ c} ⊂ D
V (x) ≤ c ⇒ k1 kxka ≤ c
Ωc = {V (x) ≤ c} ⊂ { k1 kxka ≤ c} ⊂ D
Ωc is compact and positively invariant; ∀ x(0) ∈ Ωc
a k3
V̇ ≤ −k3 kxk ≤− V
k2
dV k3
≤− dt
V k2
V (x(t)) ≤ V (x(0))e−(k3 /k2 )t
– p. 3/1
1/a
V (x(t))
kx(t)k ≤
k1
" #1/a
V (x(0))e−(k3 /k2 )t
≤
k1
" #1/a
k2 kx(0)ka e−(k3 /k2 )t
≤
k1
1/a
k2
= e−γt kx(0)k, γ = k3 /(k2 a)
k1
– p. 4/1
Example
ẋ1 = x2
ẋ2 = −h(x1 ) − x2
– p. 6/1
Example
ẋ1 = −x2
ẋ2 = x1 + (x21 − 1)x2
4
x
2
2
0
x
1
−2
−4
−4 −2 0 2 4
– p. 7/1
Example
ẋ1 = x2
ẋ2 = −x1 + 13 x31 − x2
4
x
2
2
0
x
1
−2
−4
−4 −2 0 2 4
– p. 8/1
Estimates of the Region of Attraction: Find a subset of the
region of attraction
NO
Why?
– p. 9/1
Example: Reconsider
ẋ1 = x2
ẋ2 = −x1 + 13 x31 − x2
" #
1 T 1 1 R x1
V (x) = 2
x x+2 0 (y − 13 y 3 ) dy
1 2
3 2
= x
2 1
− 16 x41 + x1 x2 + x22
– p. 10/1
The simplest estimate is the bounded component of
{V (x) < c}, where c = minx∈∂D V (x)
r2
For D = {|bT x| < r}, min xT P x =
|bT x|=r bT P −1 b
– p. 11/1
Example (Revisited)
ẋ1 = −x2
ẋ2 = x1 + (x21 − 1)x2
2 0.691
Take c = λmin (P )r = = 0.803
0.861
Alternatively, choose c as the largest constant such that
{xT P x < c} is a subset of {V̇ (x) < 0}
– p. 13/1
2 x 3 x
2 2
2
1
1
0 x 0 x
1 1
−1
−1
−2
−2 −3
−2 −1 0 1 2 −2 0 2
(a) (b)
– p. 14/1
If D is a domain where V (x) is positive definite and V̇ (x)
is negative definite (or V̇ (x) is negative semidefinite and no
solution can stay identically in the set V̇ (x) = 0 other than
x = 0), then according to LaSalle’s theorem any compact
positively invariant subset of D is a subset of the region of
attraction
Example
ẋ1 = x2
ẋ2 = −4(x1 + x2 ) − h(x1 + x2 )
– p. 15/1
" #
T T 2 1
V (x) = x P x = x x = 2x21 + 2x1 x2 + x22
1 1
T T 1
b = [1 1], c= min x Px = =1
|x1 +x2 |=1 bT P −1 b
– p. 16/1
σ = x1 + x2
d
σ 2 = 2σx2 − 8σ 2 − 2σh(σ) ≤ 2σx2 − 8σ 2 , ∀ |σ| ≤ 1
dt
d
On σ = 1, σ 2 ≤ 2x2 − 8 ≤ 0, ∀ x2 ≤ 4
dt
d
On σ = −1, σ 2 ≤ −2x2 − 8 ≤ 0, ∀ x2 ≥ −4
dt
c1 = V (x)|x1 =−3,x2 =4 = 10, c2 = V (x)|x1 =3,x2 =−4 = 10
V(x) = 10
0 x1
V(x) = 1
(3,−4)
−5
−5 0 5
– p. 18/1
Nonlinear Systems and Control
Lecture # 12
Converse Lyapunov Functions
&
Time Varying Systems
– p. 1/1
Converse Lyapunov Theorem–Exponential Stability
– p. 2/1
c1 kxk2 ≤ V (x) ≤ c2 kxk2
∂V
f (x) ≤ −c3 kxk2
∂x
∂V
≤ c4 kxk
∂x
– p. 3/1
Idea of the proof: Let ψ(t; x) be the solution of
ẏ = f (y), y(0) = x
Take
Z δ
V (x) = ψ T (t; x) ψ(t; x) dt, δ>0
0
– p. 4/1
Example: Consider the system ẋ = f (x) where f is
continuously differentiable in the neighborhood of the origin
and f (0) = 0. Show that the origin is exponentially stable
only if A = [∂f /∂x](0) is Hurwitz
– p. 5/1
∂V ∂V ∂V
Ax = f (x) − G(x)x
∂x ∂x ∂x
≤ −c3 kxk2 + c4 Lkxk2
= −(c3 − c4 L)kxk2
def
Take L < c3 /c4 , γ = (c3 − c4 L) > 0 ⇒
∂V
Ax ≤ −γkxk2 , ∀ kxk < min{r0 , r1 }
∂x
The origin of ẋ = Ax is exponentially stable
– p. 6/1
Converse Lyapunov Theorem–Asymptotic Stability
V (x) → ∞ as x → ∂RA
∂V
f (x) ≤ −W (x), ∀ x ∈ RA
∂x
and for any c > 0, {V (x) ≤ c} is a compact subset of RA
When RA = Rn , V (x) is radially unbounded
– p. 7/1
Time-varying Systems
ẋ = f (t, x)
– p. 9/1
Example
α(r) = tan−1 (r) is strictly increasing since
α′ (r) = 1/(1 + r 2 ) > 0. It belongs to class K, but not
to class K∞ since limr→∞ α(r) = π/2 < ∞
– p. 10/1
β(r, s) = r/(ksr + 1), for any positive real number k,
is strictly increasing in r since
∂β 1
= >0
∂r (ksr + 1)2
∂β −kr 2
= <0
∂s (ksr + 1)2
– p. 12/1
exponentially stable if there exist positive constants c,
k, and λ such that
– p. 13/1
Theorem: Let the origin x = 0 be an equilibrium point for
ẋ = f (t, x) and D ⊂ Rn be a domain containing x = 0.
Suppose f (t, x) is piecewise continuous in t and locally
Lipschitz in x for all t ≥ 0 and x ∈ D . Let V (t, x) be a
continuously differentiable function such that
∂V ∂V
(2) + f (t, x) ≤ 0
∂t ∂x
for all t ≥ 0 and x ∈ D , where W1 (x) and W2 (x) are
continuous positive definite functions on D . Then, the origin
is uniformly stable
– p. 14/1
Theorem: Suppose the assumptions of the previous
theorem are satisfied with
∂V ∂V
+ f (t, x) ≤ −W3 (x)
∂t ∂x
for all t ≥ 0 and x ∈ D , where W3 (x) is a continuous
positive definite function on D . Then, the origin is uniformly
asymptotically stable. Moreover, if r and c are chosen such
that Br = {kxk ≤ r} ⊂ D and c < minkxk=r W1 (x), then
every trajectory starting in {x ∈ Br | W2 (x) ≤ c} satisfies
kx(t)k ≤ β(kx(t0 )k, t − t0 ), ∀ t ≥ t0 ≥ 0
∂V ∂V
+ f (t, x) ≤ −k3 kxka
∂t ∂x
for all t ≥ 0 and x ∈ D , where k1 , k2 , k3 , and a are
positive constants. Then, the origin is exponentially stable.
If the assumptions hold globally, the origin will be globally
exponentially stable.
– p. 16/1
Example:
V (x) = 12 x2
Example:
ẋ1 = −x1 − g(t)x2
ẋ2 = x1 − x2
0 ≤ g(t) ≤ k and ġ(t) ≤ g(t), ∀ t ≥ 0
– p. 17/1
V (t, x) = x21 + [1 + g(t)]x22
– p. 18/1
Nonlinear Systems and Control
Lecture # 13
Perturbed Systems
– p. 1/?
Nominal System:
ẋ = f (x), f (0) = 0
Perturbed System:
∂V
2
V̇ (t, x) ≤ −c3 kxk +
∂x
kg(t, x)k
– p. 3/?
c3
γ<
c4
– p. 4/?
Example
ẋ1 = x2
ẋ2 = −4x1 − 2x2 + βx32 , β≥0
ẋ = Ax + g(x)
" # " #
0 1 0
A= , g(x) =
−4 −2 βx32
√
The eigenvalues of A are −1 ± j 3
3 1
2 8
P A + AT P = −I ⇒ P =
1 5
8 16
– p. 5/?
T ∂V
V (x) = x P x, Ax = −xT x
∂x
c3 = 1, c4 = 2 kP k = 2λmax (P ) = 2 × 1.513 = 3.026
kg(x)k = β|x2 |3
g(x) satisfies the bound kg(x)k ≤ γkxk over compact sets
of x. Consider the compact set
– p. 6/?
Fact:
√
max kLxk = c kLP −1/2 k
xT P x≤c
Proof
T 1 T 1
x Px ≤ c ⇔ x Px ≤ 1 ⇔ xT P 1/2 P 1/2 x ≤ 1
c c
1
y = √ P 1/2 x
c
√ −1/2 √
max kLxk = max kL c P yk = c kLP −1/2 k
xT P x≤c y T y≤1
– p. 7/?
√ −1/2 √
k2 = max |[0 1]x| = c k[0 1]P k = 1.8194 c
xT P x≤c
– p. 8/?
Alternative Bound on β
– p. 10/?
Case 2: The origin of the nominal system is asymptotically
stable
∂V ∂V
∂V
V̇ (t, x) = f (x)+ g(t, x) ≤ −W3 (x)+
g(t, x)
∂x ∂x ∂x
– p. 11/?
V̇ (t, x) ≤ −c3 φ2 (x) + c4 φ(x)kg(t, x)k
c3
If kg(t, x)k ≤ γφ(x), with γ <
c4
V̇ (t, x) ≤ −(c3 − c4 γ)φ2 (x)
– p. 12/?
Example
ẋ = −x3 + g(t, x)
V (x) = x4 is a quadratic-type Lyapunov function for the
nominal system ẋ = −x3
∂V ∂V
3 6 = 4|x|3
(−x ) = −4x ,
∂x ∂x
φ(x) = |x|3 , c3 = 4, c4 = 4
Suppose |g(t, x)| ≤ γ|x|3 , ∀ x, with γ < 1
V̇ (t, x) ≤ −4(1 − γ)φ2 (x)
Hence, the origin is a globally uniformly asymptotically
stable
– p. 13/?
Remark: A nominal system with asymptotically, but not
exponentially, stable origin is not robust to smooth
perturbations with arbitrarily small linear growth bounds
Example
ẋ = −x3 + γx
The origin is unstable for any γ > 0
– p. 14/?
Nonlinear Systems and Control
Lecture # 14
Passivity
Memoryless Functions
&
State Models
– p. 1/1
Memoryless Functions
+ y -
y
PPP
u P P
P
u
(a)
(b)
power inflow = uy
Resistor is passive if uy ≥ 0
– p. 2/1
y y y
u u u
lossless if uT y = 0
– p. 4/1
Sector Nonlinearity: h belongs to the sector [α, β]
(h ∈ [α, β]) if
αu2 ≤ uh(t, u) ≤ βu2
y=βu
y=β u
y y
y=αu
u u
y=αu
[K1 , ∞] if uT [h(t, u) − K1 u] ≥ 0
h ∈ [K1 , K2 ]
" #
β1 − α1 0
K = K2 − K1 =
0 β2 − α2
– p. 7/1
Example
kh(u) − Luk ≤ γkuk
K1 = L − γI, K2 = L + γI
K = K2 − K1 = 2γI
– p. 8/1
A function in the sector [K1 , K2 ] can be transformed into a
function in the sector [0, ∞] by input feedforward followed
by output feedback
+ +
- - - y = h(t, u) - -
K −1
+6 −6
- K1
Feedforward K −1 Feedback
[K1 , K2 ] [0, K] [0, I] [0, ∞]
−→ −→ −→
– p. 9/1
State Models
v2 = h2(i2) L
y
B B B
i2 + v2 iL
- - BB BB BB -
+ + + +
u v1 i1 = h1(v1) vC C v3 i3 = h3(v3)
P
P P
P
P
P
PP PP
P
P
PP PP
P
P
P P
i1 ?
i3
?
Lẋ1 = u − h2 (x1 ) − x2
C ẋ2 = x1 − h3 (x2 )
y = x1 + h1 (u)
– p. 10/1
V (x) = 12 Lx21 + 12 Cx22
Z t
u(s)y(s) ds ≥ V (x(t)) − V (x(0))
0
– p. 11/1
uy = V̇ + uh1 (u) + x1 h2 (x1 ) + x2 h3 (x2 )
uy ≥ V̇ + uh1 (u)
uy ≥ V̇ + x1 h2 (x1 ) + x2 h3 (x2 )
T ∂V
u y ≥ V̇ = f (x, u), ∀ (x, u)
∂x
Moreover, it is said to be
lossless if uT y = V̇
– p. 14/1
output strictly passive if uT y ≥ V̇ + y T ρ(y) for some
function ρ such that y T ρ(y) > 0, ∀ y = 6 0
Example
ẋ = u, y=x
V (x) = 12 x2 ⇒ uy = V̇ ⇒ Lossless
– p. 15/1
Example
ẋ = u, y = x + h(u), h ∈ [0, ∞]
ẋ = −h(x) + u, y = x, h ∈ [0, ∞]
– p. 16/1
Example
ẋ = u, y = h(x), h ∈ [0, ∞]
Z x
V (x) = h(σ) dσ ⇒ V̇ = h(x)ẋ = yu ⇒ Lossless
0
Example
yu = V̇ + xh(x) ⇒ Passive
h ∈ (0, ∞] ⇒ Strictly passive
– p. 17/1
Nonlinear Systems and Control
Lecture # 15
Positive Real Transfer Functions
&
Connection with Lyapunov Stability
– p. 1/?
Definition: A p × p proper rational transfer function matrix
G(s) is positive real if
poles of all elements of G(s) are in Re[s] ≤ 0
for all real ω for which jω is not a pole of any element of
G(s), the matrix G(jω) + GT (−jω) is positive
semidefinite
any pure imaginary pole jω of any element of G(s) is a
simple pole and the residue matrix
lims→jω (s − jω)G(s) is positive semidefinite Hermitian
G(s) is called strictly positive real if G(s − ε) is positive real
for some ε > 0
– p. 2/?
Scalar Case (p = 1):
Re[G(jω)] ≥ 0, ∀ ω ∈ [0, ∞)
– p. 3/?
Lemma: A p × p proper rational transfer function matrix
G(s) is strictly positive real if and only if
G(s) is Hurwitz
– p. 4/?
Scalar Case (p = 1): G(s) is strictly positive real if and only
if
G(s) is Hurwitz
Re[G(jω)] > 0, ∀ ω ∈ [0, ∞)
G(∞) > 0 or
– p. 5/?
Example:
1
G(s) =
s
has a simple pole at s = 0 whose residue is 1
1
Re[G(jω)] = Re = 0, ∀ ω 6= 0
jω
– p. 6/?
Example:
1
G(s) = , a > 0, is Hurwitz
s+a
a
Re[G(jω)] = > 0, ∀ ω ∈ [0, ∞)
ω2 + a2
ω2a
lim ω 2 Re[G(jω)] = lim = a > 0 ⇒ G is SPR
ω→∞ ω→∞ ω 2 + a2
Example:
1 1 − ω2
G(s) = , Re[G(jω)] =
s2 +s+1 (1 − ω 2 )2 + ω 2
G is not PR
– p. 7/?
Example:
s+2 1
s+1 s+2
G(s) = is Hurwitz
−1 2
s+2 s+1
2(2+ω 2 ) −2jω
1+ω 2 4+ω 2
G(jω) + GT (−jω) = > 0, ∀ ω ∈ R
2jω 4
4+ω 2 1+ω 2
" #
T 2 0
G(∞) + G (∞) = , q=1
0 0
– p. 8/?
Positive Real Lemma: Let
P A + AT P = −LT L
P B = C T − LT W
W T W = D + DT
– p. 9/?
Kalman–Yakubovich–Popov Lemma: Let
P A + AT P = −LT L − εP
P B = C T − LT W
W T W = D + DT
– p. 10/?
Lemma: The linear time-invariant minimal realization
ẋ = Ax + Bu
y = Cx + Du
with
G(s) = C(sI − A)−1 B + D
is
passive if G(s) is positive real
– p. 11/?
T ∂V
u y− (Ax + Bu)
∂x
= uT (Cx + Du) − xT P (Ax + Bu)
= uT Cx + 12 uT (D + D T )u
− 12 xT (P A + AT P )x − xT P Bu
= uT (B T P + W T L)x + 12 uT W T W u
+ 12 xT LT Lx + 12 εxT P x − xT P Bu
1
= 2 (Lx + W u)T (Lx + W u) + 12 εxT P x ≥ 12 εxT P x
– p. 12/?
Connection with Lyapunov Stability
Proof:
T ∂V ∂V
u y≥ f (x, u) ⇒ f (x, 0) ≤ 0
∂x ∂x
– p. 13/?
Lemma: If the system
T ∂V ∂V
u y≥ f (x, u) + ψ(x) ⇒ f (x, 0) ≤ −ψ(x)
∂x ∂x
Why is V (x) positive definite? Let φ(t; x) be the solution
of ż = f (z, 0), z(0) = x
– p. 14/?
V̇ ≤ −ψ(x)
Z τ
V (φ(τ, x)) − V (x) ≤ − ψ(φ(t; x)) dt, ∀ τ ∈ [0, δ]
0
Z τ
V (φ(τ, x)) ≥ 0 ⇒ V (x) ≥ ψ(φ(t; x)) dt
0
Z τ
V (x̄) = 0 ⇒ ψ(φ(t; x̄)) dt = 0, ∀ τ ∈ [0, δ]
0
– p. 15/?
Definition: The system
Linear Systems
ẋ = Ax, y = Cx
– p. 16/?
Lemma: If the system
T ∂V T ∂V
u y≥ f (x, u) + y ρ(y) ⇒ f (x, 0) ≤ −y T ρ(y)
∂x ∂x
– p. 18/?
Nonlinear Systems and Control
Lecture # 16
– p. 1/2
e1 -
u1- y1 -
+
H1
−6
y2 e + u
?
2 + 2
H2
yi = hi (t, ei )
– p. 2/2
Passivity Theorems
V̇ = 0 ⇒ x1 = 0 and y2 = 0
y2 (t) ≡ 0 ⇒ e1 (t) ≡ 0 ( & x1 (t) ≡ 0) ⇒ y1 (t) ≡ 0
y1 (t) ≡ 0 ⇒ e2 (t) ≡ 0
By zero-state observability of H2 : y2 (t) ≡ 0 ⇒ x2 (t) ≡ 0
Apply the invariance principle
– p. 5/2
Example
ẋ1 = x2 ẋ3 = x4
ẋ2 = 3
−ax1 − kx2 + e1 ẋ4 = −bx3 − x34 + e2
y1 = x2 + e1 y = x
2 4
| {z } | {z }
H
H
1 2
a, b, k > 0
V1 = 14 ax41 + 21 x22
– p. 6/2
V2 = 12 bx23 + 12 x24
– p. 7/2
Loop Transformations
+ +
- - - y = h(t, u) - -
K −1
+6 −6
- K
1
– p. 8/2
+h
- -H -
1
−6
H2
H1 is a dynamical system
H2 is a memoryless function in the sector [K1 , K2 ]
– p. 9/2
+h
- +h -
- -
H1
−6 −6
K1
+
h H2
−
6
K1
– p. 10/2
+h
- +h -
- - K -
H1
−6 −6
K1
+
h H2 K −1
−
6
K1
– p. 11/2
H̃1
+h
- +h -
- - K ?
+
- h -
H1
−6 −6 +
K1
+
+
h H2 K −1
h
−
6 +6
K1
H̃2
– p. 12/2
Example
ẋ1 = x2
ẋ2 = −h(x1 ) + bx2 + e1 y2 = σ(e2 )
| {z }
y1 = x2
H2
| {z }
H1
– p. 14/2
V̇ = −2p12 x1 h(x1 ) − (ka − 2p12 )x22
+ kx2 ẽ1 + 2p12 x1 ẽ1
= −2p12 x1 h(x1 ) − (ka − 2p12 )x22
+ (kx2 + ẽ1 )ẽ1 − ẽ21 + 2p12 x1 ẽ1
ỹ1 ẽ1 = V̇ + 2p12 x1 h(x1 ) + (ka − 2p12 )x22
+ (ẽ1 − p12 x1 )2 − p212 x21
≥ V̇ + p12 (2α1 − p12 )x21 + (ka − 2p12 )x22
ak k
Take 0 < p12 < min , 2α1 ⇒ p212 < 2p12 = p11 p22
2 2
H̃1 is strictly passive. By Theorem 6.4 the origin is globally
asymptotically stable (when u = 0)
– p. 15/2
+
- - H1 -
−6
H2
– p. 16/2
H̃1
+
- - H1 - W −1 (s) -
−6
H2 W (s)
H̃2
– p. 17/2
Example
H1 :
ẋ = Ax + Be1 , y1 = Cx
" # " #
0 1 0 h i
A= , B= , C= 1 0
−1 −1 1
H2 : y2 = h(e2 ), h ∈ [0, ∞]
−1 1
C(sI − A) B= Not PR
(s2 + s + 1)
1 (as + 1)
W (s) = ⇒
as + 1 (s2 + s + 1)
h i
H̃1 : ẋ = Ax + Be1 , ỹ1 = C̃x = 1 a x
– p. 18/2
(as + 1)
(s2 + s + 1)
1 + jωa 1 + (a − 1)ω 2
Re =
1− ω2
+ jω (1 − ω 2 )2 + ω 2
2 1 + jωa
lim ω Re 2
=a−1
ω→∞ 1 − ω + jω
(as + 1)
a>1 ⇒ is SPR
(s2 + s + 1)
V1 = 12 xT P x, P A + AT P = −LT L − εP, P B = C̃ T
– p. 19/2
H̃2 : aė2 = −e2 + ẽ2 , y2 = h(e2 ), h ∈ [0, ∞]
R e2
H̃2 is strictly passive with V2 = a 0 h(s) ds. Use
Z e2
V = V1 + V2 = 21 xT P x + a h(s) ds
0
– p. 20/2
1 T 1 T
V̇ = 2 x P ẋ + 2 ẋ P x + ah(e2 )ė2
1 T
= 2 x P [Ax − Bh(e 2 )] + 1
2 [Ax − Bh(e2 )]T P x
+ ah(e2 )C[Ax − Bh(e2 )]
= − 21 xT LT Lx − (ε/2)xT P x − xT C̃ T h(e2 )
+ ah(e2 )CAx
= − 12 xT LT Lx − (ε/2)xT P x
− xT [C + aCA]T h(e2 ) + ah(e2 )CAx
= − 12 xT LT Lx − (ε/2)xT P x − eT2 h(e2 )
≤ −(ε/2)xT P x
– p. 1/2
Absolute Stability
r + u
- -
y -
G(s)
−6
ψ(·)
– p. 2/2
Circle Criterion
Suppose G(s) = C(sI − A)−1 B + D is SPR and
ψ ∈ [0, ∞]
ẋ = Ax + Bu
y = Cx + Du
u = −ψ(y)
P A + AT P = −LT L − εP
P B = C T − LT W
W T W = D + DT
V (x) = 12 xT P x
– p. 3/2
1 T 1 T
V̇ = 2
x P ẋ + 2
ẋ P x
1 T T T
= 2 x (P A + A P )x + x P Bu
= − 12 xT LT Lx − 12 εxT P x + xT (C T − LT W )u
= − 12 xT LT Lx − 12 εxT P x + (Cx + Du)T u
T T T
− u Du − x L W u
uT Du = 21 uT (D + D T )u = 12 uT W T W u
V̇ = − 12 εxT P x − 1
2
(Lx + W u)T (Lx + W u) − y T ψ(y)
y T ψ(y) ≥ 0 ⇒ V̇ ≤ − 12 εxT P x
The origin is globally exponentially stable
– p. 4/2
What if ψ ∈ [K1 , ∞]?
-
+f - G(s) - -
+f -+f -
G(s) -
− − −
6 6 6
K1
ψ(·) ψ(·)
+
f
−
6
K1 ψ̃(·)
– p. 5/2
What if ψ ∈ [K1 , K2 ]?
-
+f - G(s) - -
+f -+f -
G(s) - K -?+ -
f
− − − +
6 6 6
K1
ψ(·) ψ(·) −1
+
f +
f
− K +
6 6
K1 ψ̃(·)
– p. 6/2
I + KG(s)[I + K1 G(s)]−1 = [I + K2 G(s)][I + K1 G(s)]−1
1 + βG(jω)
Re > 0, ∀ ω ∈ [0, ∞]
1 + αG(jω)
– p. 7/2
Case 1: α > 0
By the Nyquist criterion
1 + βG(s) 1 βG(s)
= +
1 + αG(s) 1 + αG(s) 1 + αG(s)
– p. 8/2
"1 #
β + G(jω)
Re 1 > 0, ∀ ω ∈ [0, ∞]
α + G(jω)
D(α,β) q
θ θ
2 1
−1/α −1/β
1 + βG(s)
– p. 10/2
Case 3: α < 0 < β
"1 #
1 + βG(jω)
β + G(jω)
Re > 0 ⇔ Re 1 <0
1 + αG(jω) α + G(jω)
The Nyquist plot of G(jω) must lie inside the disk D(α, β).
The Nyquist plot cannot encircle the point −(1/α) + j0.
From the Nyquist criterion, G(s) must be Hurwitz
– p. 11/2
Example
4
G(s) =
(s + 1)( 12 s + 1)( 13 s + 1)
6
Im G
4
0
Re G
−2
−4
−5 0 5
– p. 12/2
Apply Case 3 with center (0, 0) and radius = 4
– p. 13/2
Example
4
G(s) =
(s − 1)( 12 s + 1)( 13 s + 1)
0.4 Im G
0.2
G is not Hurwitz
0
Re G
−0.2
Apply Case 1
−0.4
−4 −2 0
Center = (−3.2, 0), Radius = 0.168 ⇒ [0.2969, 0.3298]
– p. 14/2
Popov Criterion
-
+f -
G(s) -
−
6
ψ(·)
ẋ = Ax + Bu
y = Cx
ui = −ψi (yi ), 1 ≤ i ≤ p
ψi ∈ [0, ki ], 1 ≤ i ≤ p, (0 < ki ≤ ∞)
G(s) = C(sI − A)−1 B
Γ = diag(γ1 , . . . , γp ), M = diag(1/k1 , · · · , 1/kp )
– p. 15/2
- M
H̃1
- ?
+ +
- i - G(s) - (I + sΓ) i -
−6 +
+
ψ(·) (I + sΓ)−1
i
+6
H̃2
M
P A + AT P = −LT L − εP
P B = (C + ΓCA)T − LT W
W T W = 2M + ΓCB + B T C T Γ
– p. 17/2
H̃2 consists of p decoupled components:
1
γi żi = −zi + ψi (zi ) + ẽ2i , ỹ2i = ψi (zi )
ki
Z zi
V2i = γi ψi (σ) dσ
0
h i
1
V̇2i = γi ψi (zi )żi = ψi (zi ) −zi + ki ψi (zi ) + ẽ2i
1
= y2i e2i + ki ψi (zi ) [ψi (zi ) − ki z i ]
– p. 18/2
p
X Z yi
Use V = 21 xT P x + γi ψi (σ) dσ
i=1 0
1 T 1 T T (y)Γẏ
V̇ = 2
x P ẋ + 2
ẋ P x + ψ
1 T T P )x + xT P Bu
= 2
x (P A + A
+ ψ T (y)ΓC(Ax + Bu)
= − 12 xT LT Lx − 12 εxT P x
+ xT (C T + AT C T Γ − LT W )u
+ ψ T (y)ΓCAx + ψ T (y)ΓCBu
– p. 19/2
V̇ = − 21 εxT P x − 12 (Lx + W u)T (Lx + W u)
− ψ(y)T [y − M ψ(y)]
≤ − 12 εxT P x − ψ(y)T [y − M ψ(y)]
– p. 20/2
Scalar case
1
+ (1 + sγ)G(s)
k
is SPR if G(s) is Hurwitz and
1
+ Re[G(jω)] − γωIm[G(jω)] > 0, ∀ ω ∈ [0, ∞)
k
If
1
lim + Re[G(jω)] − γωIm[G(jω)] =0
ω→∞ k
we also need
2 1
lim ω + Re[G(jω)] − γωIm[G(jω)] > 0
ω→∞ k
– p. 21/2
1
+ Re[G(jω)] − γωIm[G(jω)] > 0, ∀ ω ∈ [0, ∞)
k
−1/k Re[G(jω)]
Popov Plot
– p. 22/2
Example
– p. 23/2
The system is absolutely stable for ψ ∈ [0, ∞] (h ∈ [α, ∞])
ω Im G
0.2 slope=1
0
Re G
−0.2
−0.4
−0.6
−0.8
−1
−0.5 0 0.5 1
1 α − ω2 √
+ > 0, ∀ ω ∈ [0, ∞], for k < 1 + 2 α
k (α − ω 2 )2 + ω 2
– p. 24/2
Nonlinear Systems and Control
Lecture # 18
Boundedness
&
Ultimate Boundedness
– p. 1/1
Definition: The solutions of ẋ = f (t, x) are
uniformly bounded if ∃ c > 0 and for every
0 < a < c, ∃ β = β(a) > 0 such that
kx(t0 )k ≤ a ⇒ kx(t)k ≤ β, ∀ t ≥ t0 ≥ 0
kx(t0 )k ≤ a ⇒ kx(t)k ≤ b, ∀ t ≥ t0 + T
Ωε
Ωc
– p. 3/1
Suppose
∂V
V̇ (t, x) = f (t, x) ≤ −W3 (x), ∀ x ∈ Λ, ∀ t ≥ 0
∂x
W3 (x) is continuous and positive definite
V̇ (t, x) ≤ −k, ∀ x ∈ Λ, ∀ t ≥ t0 ≥ 0
V (x(t)) ≤ V (x(t0 )) − k(t − t0 ) ≤ c − k(t − t0 )
x(t) enters the set Ωε within the interval [t0 , t0 + (c − ε)/k]
– p. 4/1
Suppose
Bµ
Ωε
Ωc
Br
– p. 5/1
Let α1 and α2 be class K functions such that
c = α1 (r) ⇒ Ωc ⊂ Br
kxk ≤ µ ⇒ V (x) ≤ α2 (µ)
ε = α2 (µ) ⇒ Bµ ⊂ Ωε
What is the ultimate bound?
– p. 6/1
Theorem (special case of Thm 4.18): Suppose
∂V
f (t, x) ≤ −W3 (x), ∀ kxk ≥ µ > 0
∂x
∀ t ≥ 0 and kxk ≤ r , where α1 , α2 ∈ K, W3 (x) is
continuous & positive definite, and µ < α−12 (α1 (r)). Then,
for every initial state x(t0 ) ∈ {kxk ≤ α−1
2 (α1 (r))}, there is
T ≥ 0 (dependent on x(t0 ) and µ) such that
kx(t)k ≤ α−1
1 (α2 (µ)), ∀ t ≥ t0 + T
– p. 8/1
Example
– p. 9/1
λmin (P )kxk2 ≤ V (x) ≤ λmax (P )kxk2 + 12 kxk4
Perturbed Systems
&
Input-to-State Stability
– p. 1/?
Perturbed Systems: Nonvanishing Perturbation
Nominal System:
ẋ = f (x), f (0) = 0
Perturbed System:
∂V ∂V
2
f (x) ≤ −c3 kxk ,
≤ c4 kxk
∂x
∂x
∀ x ∈ Br = {kxk ≤ r}
– p. 2/?
Use V (x) to investigate ultimate boundedness of the
perturbed system
∂V ∂V
V̇ (t, x) = f (x) + g(t, x)
∂x ∂x
Assume
kg(t, x)k ≤ δ, ∀ t ≥ 0, x ∈ Br
V̇ (t, x) ≤ −c3 kxk2 +
∂V kg(t, x)k
∂x
≤ −c3 kxk2 + c4 δkxk
= −(1 − θ)c3 kxk2 − θc3 kxk2 + c4 δkxk
0<θ<1
def
≤ −(1 − θ)c3 kxk2 , ∀ kxk ≥ δc4 /(θc3 ) = µ
– p. 3/?
Apply Theorem 4.18
c1
r
kx(t0 )k ≤ α−1
2 (α1 (r)) ⇔ kx(t0 )k ≤ r
c2
δc4 c1 c3 c1
r r
µ< α−1
2 (α1 (r)) ⇔ <r ⇔ δ< θr
θc3 c2 c4 c2
c2 δc4 c2
r r
−1
b = α1 (α2 (µ)) ⇔ b = µ ⇔ b=
c1 θc3 c1
p
For all kx(t0 )k ≤ r c1 /c2 , the solutions of the perturbed
system are ultimately bounded by b
– p. 4/?
Example
β ≥ 0, |d(t)| ≤ δ, ∀ t ≥ 0
3 1
2 8
T T
V (x) = x P x = x x (Lecture 13)
1 5
8 16
2 1 5 2
2βx22
V̇ (t, x) = −kxk + 8 x1 x2 + 16 x2
1 5
+ 2d(t) x +
8 1
x
16 2
√ √
2 29 29δ
≤ −kxk + βk22 kxk2 + kxk
8 8
– p. 5/?
√
k2 = max |x2 | = 1.8194 c
xT P x≤c
√
Suppose β ≤ 8(1 − ζ)/( 29k22 ) (0 < ζ < 1)
√
29δ
V̇ (t, x) ≤ −ζkxk2 + 8
kxk
√
29δ def
≤ −(1 − θ)ζkxk2 , ∀ kxk ≥ 8ζθ
= µ
(0 < θ < 1)
If µ2 λmax (P ) < c, then all solutions of the perturbed
system, starting in Ωc , are uniformly ultimately bounded by
√ s
29δ λmax (P )
b=
8ζθ λmin (P )
– p. 6/?
Case 2: The origin of ẋ = f (x) is asymptotically stable
∂V
∂V
f (x) ≤ −α3 (kxk),
≤k
∂x
∂x
∀ x ∈ Br = {kxk ≤ r}, αi ∈ K, i = 1, 2, 3
V̇ (t, x) ≤ −α3 (kxk) +
∂V
∂x
kg(t, x)k
≤ −α3 (kxk) + δk
≤ −(1 − θ)α3 (kxk) − θα3 (kxk) + δk
0<θ<1
−1 δk def
≤ −(1 − θ)α3 (kxk), ∀ kxk ≥ α3 θ
= µ
– p. 7/?
Apply Theorem 4.18
δk
µ< α−1
2 (α1 (r)) ⇔ α−1
3 < α−1
2 (α1 (r))
θ
θα3 (α−1
2 (α1 (r)))
⇔ δ<
k
−1 δk
−1
µ < α2 (α1 (r)) ⇔ α3 < α−12 (α1 (r))
θ
−1
θα3 (α2 (α1 (r))) c3 c1
r
⇔ δ< Compare with δ < θr
k c4 c2
Example
x
ẋ = −
1 + x2
∂V x 4x4
V (x) = x4 ⇒ − 2
=− 2
– p. 8/?
The origin is globally asymptotically stable
θα3 (α−1
2 (α1 (r))) θα3 (r) rθ
= =
k k 1 + r2
rθ
→ 0 as r → ∞
1+ r2
x
ẋ = − + δ, δ>0
1+ x2
1
δ> 2 ⇒ lim x(t) = ∞
t→∞
– p. 9/?
Input-to-State Stability (ISS)
– p. 10/?
Theorem (Special case of Thm 4.19): Let V (x) be a
continuously differentiable function such that
∂V
f (x, u) ≤ −W3 (x), ∀ kxk ≥ ρ(kuk) > 0
∂x
∀ x ∈ Rn , u ∈ Rm , where α1 , α2 ∈ K∞ , ρ ∈ K, and
W3 (x) is a continuous positive definite function. Then, the
system ẋ = f (x, u) is ISS with γ = α−11 ◦ α2 ◦ ρ
∂V
f (x, u) ≤ −W3 (x), ∀ kxk ≥ µ
∂x
– p. 11/?
Choose ε and c such that
∂V
f (x, u) ≤ −W3 (x), ∀ x ∈ Λ = {ε ≤ V (x) ≤ c}
∂x
Suppose x(t0 ) ∈ Λ and x(t) reaches Ωε at t = t0 + T . For
t0 ≤ t ≤ t0 + T , V satisfies the conditions for the uniform
asymptotic stability. Therefore, the trajectory behaves as if
the origin was uniformly asymptotically stable and satisfies
For t ≥ t0 + T ,
kx(t)k ≤ α−1
1 (α2 (µ))
– p. 12/?
kx(t)k ≤ β(kx(t0 )k, t − t0 ) + α−1
1 (α2 (µ)), ∀ t ≥ t0
!
kx(t)k ≤ β(kx(t0 )k, t − t0 ) + γ sup ku(τ )k , ∀ t ≥ t0
τ ≥t0
– p. 13/?
Example
ẋ = −x3 + u
The origin of ẋ = −x3 is globally asymptotically stable
V = 12 x2
V̇ = −x4 + xu
= −(1 − θ)x4 − θx4 + xu
1/3
≤ −(1 − θ)x4 , ∀ |x| ≥ |u|
θ
0<θ<1
The system is ISS with
γ(r) = (r/θ)1/3
– p. 14/?
Example
ẋ = −x − 2x3 + (1 + x2 )u2
The origin of ẋ = −x − 2x3 is globally exponentially stable
V = 12 x2
– p. 15/?
Example
– p. 16/?
− 12 θ(x21 + x42 ) + |x2 |3 |u| ≤ 0
2
2|u| 2|u| 2|u|
if |x2 | ≥ or |x2 | ≤ and |x1 | ≥
θ θ θ
s
2
2|u| 2|u|
if kxk ≥ 1+
θ θ
s
2
2r 2r
ρ(r) = 1+
θ θ
– p. 17/?
Find γ
V (x) = 12 x21 + 14 x42
1
For |x2 | ≤ |x1 |, 4
(x2
1 + x22 ) ≤ 14 x21 + 14 x21 = 12 x21 ≤ V (x)
1 1
For |x2 | ≥ |x1 |, 16 (x21 +x22 )2 ≤ 16 (x22 +x22 )2 = 14 x42 ≤ V (x)
1
α1 (r) = 4 min r , 4 r , α2 (r) = 12 r 2 + 14 r 4
2 1 4
γ = α−1
1 ◦ α2 ◦ ρ
( 1
α−1 (s) = 2(s) 4 , if s ≤ 1
1 √
2 s, if s ≥ 1
– p. 18/?
Nonlinear Systems and Control
Lecture # 20
Input-Output Stability
– p. 1/1
Input-Output Models
y = Hu
The
R∞ T
space of square-integrable functions:
0 u (t)u(t) dt < ∞
sZ
∞
L2 ; kukL2 = uT (t)u(t) dt < ∞
0
Z ∞ 1/p
Lp ; kukLp = ku(t)kp dt < ∞, 1 ≤ p < ∞
0
Notation Lm
p : p is the type of p-norm used to define the
space and m is the dimension of u
– p. 3/1
Extended Space: Le = {u | uτ ∈ L, ( ∀ τ ∈ [0, ∞)}
u(t), 0 ≤ t ≤ τ
uτ is a truncation of u: uτ (t) =
0, t>τ
/ L∞ but uτ ∈ L∞e
u∈
Causality: A mapping H : Lm e → L q is causal if the value
e
of the output (Hu)(t) at any time t depends only on the
values of the input up to time t
(Hu)τ = (Huτ )τ
– p. 4/1
Definition: A mapping H : Lm
e → L q is L stable if ∃ α ∈ K
e
β ≥ 0 such that
k(Hu)τ kL ≤ α (kuτ kL ) + β, ∀ u ∈ Lm
e and τ ∈ [0, ∞)
k(Hu)τ kL ≤ γkuτ kL + β, ∀ u ∈ Lm
e and τ ∈ [0, ∞)
– p. 5/1
Example: Memoryless function y = h(u)
ecu − e−cu
h(u) = a + b tanh cu = a + b , a, b, c > 0
ecu + e−cu
′ 4bc
h (u) = 2 ≤ bc ⇒ |h(u)| ≤ a+bc|u|, ∀ u ∈ R
(ecu + e−cu )
Finite-gain L∞ stable with β = a and γ = bc
h(u) = b tanh cu, |h(u)| ≤ bc|u|, ∀ u ∈ R
Z ∞ Z ∞
|h(u(t))|p dt ≤ (bc)p |u(t)|p dt, for p ∈ [1, ∞)
0 0
– p. 6/1
h(u) = u2
!2
sup |h(u(t))| ≤ sup |u(t)|
t≥0 t≥0
– p. 7/1
Example: SISO causal convolution operator
Z t
y(t) = h(t − σ)u(σ) dσ, h(t) = 0 for t < 0
0
Z ∞
Suppose h ∈ L1 ⇔ khkL1 = |h(σ)| dσ < ∞
0
Rt
|y(t)| ≤ 0 |h(t − σ)| |u(σ)| dσ
Rt
≤ 0 |h(t − σ)| dσ sup0≤σ≤τ |u(σ)|
Rt
= 0 |h(s)| ds sup0≤σ≤τ |u(σ)|
kyτ kL∞ ≤ khkL1 kuτ kL∞ , ∀ τ ∈ [0, ∞)
Finite-gain L∞ stable
Also, finite-gain Lp stable for p ∈ [1, ∞) (see textbook)
– p. 8/1
L Stability of State Models
∂V
∂V
2
f (x, 0) ≤ −c3 kxk ,
≤ c4 kxk
∂x
∂x
kf (x, u)−f (x, 0)k ≤ Lkuk, kh(x, u)k ≤ η1 kxk+η2 kuk
∀ kxk ≤ r and kuk ≤ ru
– p. 9/1
V̇ = ∂V
∂x f (x, 0) + ∂V
∂x [f (x, u) − f (x, 0)] √
≤ −c3 kxk2 + c4 Lkxk kuk ≤ − cc32 V + c4 L
√
c1
kuk V
c3 c4 L
p
W (t) = V (x(t)) ⇒ Ẇ ≤ − W + √ ku(t)k
2c2 2 c1
tc c4 L
Z t (t−τ )c 3
− 2c3 −
W (t) ≤ e 2 W (0) + √ e 2c 2
ku(τ )k dτ
2 c1 0
r
c2 tc c4 L
Z t (t−τ )c 3
− 2c3 −
kx(t)k ≤ kx(0)ke 2 + e 2c 2
ku(τ )k dτ
c1 2c1 0
Z t
ky(t)k ≤ k0 kx(0)ke−at +k2 e−a(t−τ ) ku(τ )k dτ +k3 ku(t)k
0
– p. 10/1
p
Theorem 5.1: For each x(0) with kx(0)k ≤ r c1 /c2 , the
system is small-signal finite-gain Lp stable for each
p ∈ [1, ∞]
Example
ẋ = −x − x3 + u, y = tanh x + u
V = 12 x2 ⇒ x(−x − x3 ) ≤ −x2
c1 = c2 = 12 , c3 = c4 = 1, L = η1 = η2 = 1
Finite-gain Lp stable for each x(0) ∈ R and each p ∈ [1, ∞]
– p. 11/1
Case 2: The origin of ẋ = f (x, 0) is asymptotically stable
is L∞ stable
– p. 12/1
Proof
!
kx(t)k ≤ β(kx(0)k, t)+γ sup ku(t)k , β ∈ KL, γ ∈ K
0≤t≤τ
ky(t)k ≤ α1 β(kx(0)k, t) + γ sup0≤t≤τ ku(t)k
+ α2 (ku(t)k) + η
α1 (a + b) ≤ α1 (2a) + α1 (2b)
ky(t)k ≤ α1 (2β(kx(0)k, t)) + α1 2γ sup0≤t≤τ ku(t)k
+ α2 (ku(t)k) + η
kyτ kL∞ ≤ γ0 (kuτ kL∞ ) + β0
γ0 = α1 ◦ 2γ + α2 and β0 = α1 (2β(kx(0)k, 0)) + η
– p. 13/1
Theorem (Rephrasing of Thm 5.2): Suppose f is locally
Lipschitz and h is continuous in some neighborhood of
(x = 0, u = 0). If the origin of ẋ = f (x, 0) is
asymptotically stable, then there is a constant k1 > 0 such
that for each x(0) with kx(0)k < k1 , the system
is small-signal L∞ stable
– p. 14/1
Example
V̇ ≤ −kxk4 + 2kxk|u|
= −(1 − θ)kxk4 − θkxk4 + 2kxk|u|, 0 < θ < 1
1/3
≤ −(1 − θ)kxk4 , ∀ kxk ≥ 2|u|
θ
ISS
L∞ stable
– p. 15/1
Nonlinear Systems and Control
Lecture # 21
L2 Gain
&
– p. 1/1
Theorem 5.4: Consider the linear time-invariant system
ẋ = Ax + Bu, y = Cx + Du
– p. 2/1
Lemma: Consider the time-invariant system
– p. 4/1
Proof
Z τ Z τ
V (x(τ ))−V (x(0)) ≤ aγ 2 ku(t)k2 dt− a ky(t)k2 dt
0 0
V (x) ≥ 0
Z τ Z τ
V (x(0))
2 2 2
ky(t)k dt ≤ γ ku(t)k dt +
0 0 a
– p. 4/1
Proof
Z τ Z τ
V (x(τ ))−V (x(0)) ≤ aγ 2 ku(t)k2 dt− a ky(t)k2 dt
0 0
V (x) ≥ 0
Z τ Z τ
V (x(0))
2 2 2
ky(t)k dt ≤ γ ku(t)k dt +
0 0 a
s
V (x(0))
kyτ kL2 ≤ γkuτ kL2 +
a
– p. 4/1
Lemma 6.5: If the system
uT y ≥ V̇ + δy T y, δ>0
– p. 5/1
Lemma 6.5: If the system
uT y ≥ V̇ + δy T y, δ>0
V̇ ≤ uT y − δy T y
1
= − 2δ (u − δy)T (u − δy) + 1 T
2δ u u − 2δ y T y
δ 1 T T
≤ 2 δ2 u u − y y
– p. 5/1
Example
– p. 6/1
Example
– p. 6/1
Example
– p. 6/1
Example
– p. 6/1
Theorem 5.5: Consider the time-invariant system
f (0) = 0, h(0) = 0
where f and G are locally Lipschitz and h is continuous
over Rn . Suppose ∃ γ > 0 and a continuously
differentiable, positive semidefinite function V (x) that
satisfies the Hamilton–Jacobi inequality
T
∂V 1 ∂V ∂V 1
T
f (x)+ G(x)G (x) + hT (x)h(x) ≤ 0
∂x 2γ 2 ∂x ∂x 2
– p. 8/1
Proof
∂V ∂V
f (x) + G(x)u =
∂x ∂x
2
T
1 2
1 ∂V ∂V
T
− γ
u − 2 G (x)
+ f (x)
2
γ ∂x
∂x
T
1 ∂V ∂V 1 2
T 2
+ G(x)G (x) + γ kuk
2γ 2 ∂x ∂x 2
1 2 2 1
V̇ ≤ γ kuk − kyk2
2 2
– p. 8/1
Example
ẋ = Ax + Bu, y = Cx
– p. 9/1
Example
ẋ = Ax + Bu, y = Cx
Suppose there is P = P T ≥ 0 that satisfies the Riccati
equation
T 1 T T
PA + A P + 2
P BB P + C C=0
γ
– p. 9/1
Example
ẋ = Ax + Bu, y = Cx
Suppose there is P = P T ≥ 0 that satisfies the Riccati
equation
T 1 T T
PA + A P + 2
P BB P + C C=0
γ
– p. 9/1
Example
ẋ = Ax + Bu, y = Cx
Suppose there is P = P T ≥ 0 that satisfies the Riccati
equation
T 1 T T
PA + A P + 2
P BB P + C C=0
γ
– p. 9/1
The Small-Gain Theorem
u1-e1- y1
H1 -
+−
6
y2 e + u
?
H2 2 + 2
– p. 10/1
The Small-Gain Theorem
u1-e1- y1
H1 -
+−
6
y2 e + u
?
H2 2 + 2
ky1τ kL ≤ γ1 ke1τ kL + β1 , ∀ e1 ∈ Lm
e , ∀ τ ∈ [0, ∞)
– p. 10/1
The Small-Gain Theorem
u1-e1- y1
H1 -
+−
6
y2 e + u
?
H2 2 + 2
ky1τ kL ≤ γ1 ke1τ kL + β1 , ∀ e1 ∈ Lm
e , ∀ τ ∈ [0, ∞)
– p. 10/1
" # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2
– p. 11/1
" # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2
– p. 11/1
" # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2
Proof
– p. 11/1
" # " # " #
u1 y1 e1
u= , y= , e=
u2 y2 e2
Proof
– p. 11/1
ke1τ kL ≤ ku1τ kL + γ2 (ku2τ kL + γ1 ke1τ kL + β1 ) + β2
= γ1 γ2 ke1τ kL
+ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
– p. 12/1
ke1τ kL ≤ ku1τ kL + γ2 (ku2τ kL + γ1 ke1τ kL + β1 ) + β2
= γ1 γ2 ke1τ kL
+ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1
ke1τ kL ≤ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1 − γ1 γ2
– p. 12/1
ke1τ kL ≤ ku1τ kL + γ2 (ku2τ kL + γ1 ke1τ kL + β1 ) + β2
= γ1 γ2 ke1τ kL
+ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1
ke1τ kL ≤ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1 − γ1 γ2
1
ke2τ kL ≤ (ku2τ kL + γ1 ku1τ kL + β1 + γ1 β2 )
1 − γ1 γ2
– p. 12/1
ke1τ kL ≤ ku1τ kL + γ2 (ku2τ kL + γ1 ke1τ kL + β1 ) + β2
= γ1 γ2 ke1τ kL
+ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1
ke1τ kL ≤ (ku1τ kL + γ2 ku2τ kL + β2 + γ2 β1 )
1 − γ1 γ2
1
ke2τ kL ≤ (ku2τ kL + γ1 ku1τ kL + β1 + γ1 β2 )
1 − γ1 γ2
keτ kL ≤ ke1τ kL + ke2τ kL
– p. 12/1
Nonlinear Systems and Control
Lecture # 22
Normal Form
– p. 1/1
Relative Degree
– p. 2/1
∂(Lf h)
Lg Lf h(x) = g(x)
∂x
∂(Lf h)
L2f h(x) = Lf Lf h(x) = f (x)
∂x
∂(Lk−1
f h)
Lkf h(x) = Lf Lk−1
f h(x) = f (x)
∂x
L0f h(x) = h(x)
ẏ = Lf h(x) + Lg h(x) u
Lg h(x) = 0 ⇒ ẏ = Lf h(x)
(2) ∂(Lf h)
y = [f (x) + g(x)u] = L2f h(x) + Lg Lf h(x) u
∂x
– p. 3/1
Lg Lf h(x) = 0 ⇒ y (2) = L2f h(x)
– p. 4/1
Example
ẏ = ẋ1 = x2
ÿ = ẋ2 = −x1 + ε(1 − x21 )x2 + u
Relative degree = 2 over R2
Example
Change of variables:
φ1 (x)
..
.
φn−ρ (x) φ(x) η
def def
z = T (x) = − − − = −−− = −−−
h(x) ψ(x) ξ
..
.
ρ−1
Lf h(x)
– p. 7/1
∂φ
η̇ = [f (x) + g(x)u] = f0 (η, ξ) + g0 (η, ξ)u
∂x
ξ̇i = ξi+1 , 1≤i≤ρ−1
ξ̇ρ = Lρf h(x) + Lg Lρ−1
f h(x) u
y = ξ1
– p. 8/1
Theorem 13.1: Suppose the system
ξ̇ = Ac ξ + Bc γ(x)[u − α(x)]
If x∗ is an open-loop equilibrium point at which y = 0; i.e.,
f (x∗ ) = 0 and h(x∗ ) = 0, then ψ(x∗) = 0. Take
φ(x∗ ) = 0 so that z = 0 is an open-loop equilibrium point.
– p. 11/1
Zero Dynamics
η̇ = f0 (η, ξ)
ξ̇ = Ac ξ + Bc γ(x)[u − α(x)]
y = Cc ξ
– p. 12/1
ρ−1
Z ∗ = {x ∈ D0 | h(x) = Lf h(x) = · · · = Lf h(x) = 0}
y(t) ≡ 0 ⇒ x(t) ∈ Z ∗
def
⇒ u = u∗ (x) = α(x)|x∈Z ∗
The restricted motion of the system is described by
def
ẋ = f ∗ (x) = [f (x) + g(x)α(x)]x∈Z ∗
– p. 13/1
Example
– p. 14/1
Example
2 + x23
ẋ1 = −x1 + u, ẋ2 = x3 , ẋ3 = x1 x3 + u, y = x2
1+ x23
ẏ = ẋ2 = x3
ÿ = ẋ3 = x1 x3 + u ⇒ ρ = 2
L2f h(x)
γ = Lg Lf h(x) = 1, α = − = −x1 x3
Lg Lf h(x)
Z ∗ = {x2 = x3 = 0}
u = u∗ (x) = 0 ⇒ ẋ1 = −x1
Minimum phase
– p. 15/1
Find φ(x) such that
2+x23
∂φ h i 1+x23
∂φ ∂φ ∂φ
φ(0) = 0, g(x) = , , 0 =0
∂x ∂x1 ∂x2 ∂x3
1
and h iT
T (x) = φ(x) x2 x3
is a diffeomorphism
∂φ 2 + x23 ∂φ
· + =0
∂x1 1+ x23 ∂x3
is a global diffeomorphism
η = −x1 + x3 + tan−1 x3 , ξ1 = x2 , ξ2 = x3
!
2 + ξ22
−η + ξ2 + tan−1 ξ2
η̇ = 1+ 2
ξ2
1+ ξ2
ξ̇1 = ξ2
−1
ξ̇2 = −η + ξ2 + tan ξ2 ξ2 + u
y = ξ1
– p. 17/1
Nonlinear Systems and Control
Lecture # 23
Controller Form
– p. 1/1
Definition: A nonlinear system is in the controller form if
ẋ = Ax + Bγ(x)[u − α(x)]
u = α(x) + γ −1 (x)v ⇒ ẋ = Ax + Bv
ẋ = f (x) + g(x)u
M AM −1 = Ac + Bc λT , M B = Bc
def
z = M ζ = M S(x) = T (x)
ż = Ac z + Bc γ(·)[u − α(·)]
h(x) = T1 (x)
– p. 4/1
In summary, the n-dimensional SI system
ẋ = f (x) + g(x)u
Lg Li−1
f h(x) = 0, i = 1, 2, . . . , n−1, and L Ln−1
g f h(x) 6= 0
h i
T (x) = h(x), Lf h(x), ··· Ln−1
f h(x)
– p. 5/1
The Lie Bracket: For two vector fields f and g , the Lie
bracket [f, g] is a third vector field defined by
∂g ∂f
[f, g](x) = f (x) − g(x)
∂x ∂x
Notation:
– p. 7/1
" # " #
x2 −x1
f = , adf g =
− sin x1 − x2 x1 + x2
– p. 8/1
Distribution: For vector fields f1 , f2 , . . . , fk on D ⊂ Rn , let
∆ = span{f1 , f2 , . . . , fk }
g1 ∈ ∆ and g2 ∈ ∆ ⇒ [g1 , g2 ] ∈ ∆
– p. 9/1
Lemma: If ∆ is a nonsingular distribution, generated by
f1 , . . . , fk , then it is involutive if and only if
[fi , fj ] ∈ ∆, ∀ 1 ≤ i, j ≤ k
Example: D = R3 ; ∆ = span{f1 , f2 }
2x2 1
f1 = 1 , f2 = 0 , dim(∆(x)) = 2, ∀ x ∈ D
0 x2
0
∂f2 ∂f1
[f1 , f2 ] = f1 − f2 = 0
∂x ∂x
1
– p. 10/1
rank [f1 (x), f2 (x), [f1 , f2 ](x)] =
2x2 1 0
rank 1 0 0 = 3, ∀ x ∈ D
0 x2 1
∆ is not involutive
– p. 11/1
Example: D = {x ∈ R3 | x21 + x23 6= 0}; ∆ = span{f1 , f2 }
2x3 −x1
f1 = −1 , f2 = −2x2 , dim(∆(x)) = 2, ∀ x ∈ D
0 x3
−4x3
∂f2 ∂f1
[f1 , f2 ] = f1 − f2 = 2
∂x ∂x
0
2x3 −x1 −4x3
rank −1 −2x2 2 = 2, ∀ x ∈ D
0 x3 0
∆ is involutive
– p. 12/1
Theorem: The n-dimensional SI system
ẋ = f (x) + g(x)u
and
– p. 13/1
Example
" # " #
a sin x2 0
ẋ = + u
−x21 1
" #
∂f −a cos x2
adf g = [f, g] = − g=
∂x 0
" #
0 −a cos x2
[g(x), adf g(x)] =
1 0
– p. 14/1
∂h ∂h
g= = 0 ⇒ h is independent of x2
∂x ∂x2
∂h
Lf h(x) = a sin x2
∂x1
∂(Lf h) ∂(Lf h) ∂h
Lg Lf h(x) = g= = a cos x2
∂x ∂x2 ∂x1
∂h
Lg Lf h(x) 6= 0 in D0 = {x ∈ R2 | cos x2 6= 0} if ∂x1 6= 0
" # " #
h x1
Take h(x) = x1 ⇒ T (x) = =
Lf h a sin x2
– p. 15/1
Example (Field-Controlled DC Motor)
−ax1 1
ẋ = −bx2 + k − cx1 x3 + 0 u
θx1 x2 0
a a2
adf g = cx3 ; ad2f g = (a + b)cx3
−θx2 (b − a)θx2 − θk
1 a a2
2
[g(x), adf g(x), adf g(x)] = 0 cx3 (a + b)cx3
0 −θx2 (b − a)θx2 − θk
– p. 16/1
det[·] = cθ(−k + 2bx2 )x3
3 k
D0 = {x ∈ R | x2 > and x3 > 0}
2b
Find h such that Lg h(x) = Lg Lf h(x) = 0; Lg L2f h(x) 6= 0
– p. 17/1
x∗ = [0, k/b, ω0 ]T , h(x∗ ) = 0
∂h ∂h
g= = 0 ⇒ h is independent of x1
∂x ∂x1
∂h ∂h
Lf h(x) = [−bx2 + k − cx1 x3 ] + θx1 x2
∂x2 ∂x3
∂h ∂h
[∂(Lf h)/∂x]g = 0 ⇒ cx3 = θx2
∂x2 ∂x3
h = c1 [θx22 + cx23 ] + c2 , Lg L2f h(x) = −2c1 cθ(k − 2bx2 )x3
c1 = 1, c2 = −θ(k/b)2 − cω02
– p. 18/1
Nonlinear Systems and Control
Lecture # 24
– p. 1/1
Definition: A nonlinear system is in the observer form if
ẋ = Ax + γ(y, u), y = Cx
Observer:
x̃ = x − x̂
x̃˙ = (A − HC)x̃
Design H such that (A − HC) is Hurwitz
– p. 2/1
Theorem: An n-dimensional single-output (SO) system
and
j
[g, adf τ ] = 0, 0≤j ≤n−2
The change of variables z = T (x) is given by
∂T h i
τ1 , τ2 , ··· τn =I
∂x
where
τi = (−1)i−1 adi−1
f τ, 1≤i≤n
– p. 4/1
Example
" # " #
β1 (x1 ) + x2 0
ẋ = + u, y = x1
f2 (x) 1
" # " #
h(x) x1
φ(x) = =
Lf h(x) β1 (x1 ) + x2
" #
∂φ ∂φ
1 0
= ∂β1 ; rank (x) = 2, ∀ x
∂x ∂x1 1 ∂x
" # " #
∂φ 0 0
τ = ⇒ τ =
∂x 1 1
– p. 5/1
" #" # " #
∂f ∗ 1 0 1
adf τ = [f, τ ] = − τ =− ∂f2 =− ∂f2
∂x ∗ ∂x2 1 ∂x2
" #" #
∂(adf τ ) 0 0 0
[τ, adf τ ] = τ =− ∂ 2 f2 ∂ 2 f2
∂x ∂x1 ∂x2 ∂x22 1
∂ 2 f2
[τ, adf τ ] = 0 ⇔ = 0 ⇔ f2 (x) = β2 (x1 )+x2 β3 (x1 )
∂x22
[g, τ ] = 0 (g and τ are constant vector fields)
All the conditions are satisfied
– p. 6/1
" #
0 0
τ1 = (−1) ad0f τ =τ =
1
" #
1 1
τ2 = (−1) ad1f τ = −adf τ =
β3 (x1 )
∂T h i
τ1 , τ2 =I
∂x
" #" # " #
∂T1 ∂T1
∂x1 ∂x2 0 1 1 0
∂T2 ∂T2 =
∂x1 ∂x2
1 β3 (x1 ) 0 1
∂T1 ∂T1 ∂T1
∂x2
= 1, ∂x1
+ β3 (x1 ) ∂x 2
=0
– p. 7/1
∂T1 ∂T1
∂x2
=1 ⇒ ∂x1
+ β3 (x1 ) = 0
R x1
T1 (x) = x2 − 0 β3 (σ) dσ
∂T2 ∂T2
∂x2
=0 ⇒ ∂x1
= 1, T2 (x) = x1
Z x1
z1 = x2 − β3 (σ) dσ, z2 = x1
0
y = z2
" # " #
0 0 β2 (y) − β1 (y)β3 (y) + u
ż = z+ Ry
1 0 0 β3 (σ) dσ + β1 (y)
h i
y = 0 1 z
– p. 8/1
Definition: A nonlinear system is in the output feedback
form if
ẋ1 = x2 + γ1 (y)
ẋ2 = x3 + γ2 (y)
..
.
ẋρ−1 = xρ + γρ−1 (y)
ẋρ = xρ+1 + γρ (y) + bm u, bm > 0
..
.
ẋn−1 = xn + γn−1 (y) + b1 u
ẋn = γn (y) + b0 u
y = x1
– p. 9/1
Show that
The output feedback form is a special case of the
observer form
bm sm + · · · + b1 s + b0
is Hurwitz
– p. 10/1
Definition: A nonlinear system is in the strict feedback form
if
ẋ = f0 (x) + g0 (x)z1
ż1 = f1 (x, z1 ) + g1 (x, z1 )z2
ż2 = f2 (x, z1 , z2 ) + g2 (x, z1 , z2 )z3
..
.
żk−1 = fk−1 (x, z1 , . . . , zk−1 ) + gk−1 (x, z1 , . . . , zk−1 )zk
żk = fk (x, z1 , . . . , zk ) + gk (x, z1 , . . . , zk )u
x ∈ Rn , z1 to zk are scalars
gi (x, z1 , . . . , zi ) 6= 0 for 1 ≤ i ≤ k
– p. 11/1
Find the relative degree if y = z1
fi (x, 0) = 0, ∀1≤i≤k
– p. 12/1
Nonlinear Systems and Control
Lecture # 25
Stabilization
– p. 1/?
We want to stabilize the system
ẋ = f (x, u)
0 = f (xss , uss )
xδ = x − xss , uδ = u − uss
def
ẋδ = f (xss + xδ , uss + uδ ) = fδ (xδ , uδ )
fδ (0, 0) = 0
uδ = γ(xδ ) ⇒ u = uss + γ(x − xss )
– p. 2/?
State Feedback Stabilization: Given
ẋ = f (x, u) [f (0, 0) = 0]
find
u = γ(x) [γ(0) = 0]
s.t. the origin is an asymptotically stable equilibrium point of
ẋ = f (x, γ(x))
– p. 3/?
Linear Systems
ẋ = Ax + Bu
(A, B) is stabilizable (controllable or every uncontrollable
eigenvalue has a negative real part)
u = −Kx
Typical methods:
Eigenvalue Placement
Eigenvalue-Eigenvector Placement
LQR
– p. 4/?
Linearization
ẋ = f (x, u)
f (0, 0) = 0 and f is continuously differentiable in a domain
Dx × Du that contains the origin (x = 0, u = 0)
(Dx ⊂ Rn , Du ⊂ Rp )
ẋ = Ax + Bu
∂f ∂f
A= (x, u) ; B= (x, u)
∂x x=0,u=0 ∂u x=0,u=0
u = −Kx
– p. 5/?
Closed-loop system:
ẋ = f (x, −Kx)
∂f ∂f
ẋ = (x, −Kx) + (x, −Kx) (−K) x
∂x ∂u x=0
= (A − BK)x
– p. 6/?
Example (Pendulum Equation):
θ̈ = −a sin θ − bθ̇ + cT
0 = −a sin δ + cTss
x1 = θ − δ, x2 = θ̇, u = T − Tss
ẋ1 = x2
ẋ2 = −a[sin(x1 + δ) − sin δ] − bx2 + cu
" # " #
0 1 0 1
A= =
−a cos(x1 + δ) −b −a cos δ −b
x1 =0
– p. 7/?
" # " #
0 1 0
A= ; B=
−a cos δ −b c
h i
K= k1 k2
" #
0 1
A − BK =
−(a cos δ + ck1 ) −(b + ck2 )
a cos δ b
k1 > − , k2 > −
c c
a sin δ a sin δ
T = − Kx = − k1 (θ − δ) − k2 θ̇
c c
– p. 8/?
Notions of Stabilization
– p. 9/?
Semiglobal Stabilization: The origin of ẋ = f (x, γ(x)) is
asymptotically stable and γ(x) can be designed such that
any given compact set (no matter how large) can be
included in the region of attraction (Typically u = γp (x) is
dependent on a parameter p such that for any compact set
G, p can be chosen to ensure that G is a subset of the
region of attraction )
– p. 10/?
Example
ẋ = x2 + u
Linearization:
ẋ = u, u = −kx, k > 0
Closed-loop system:
ẋ = −kx + x2
The control
u = −x2 − kx
achieves global stabilization because it yields the linear
closed-loop system ẋ = −kx whose origin is globally
exponentially stable
– p. 12/?
Practical Stabilization
ẋ = f (x, u) + g(x, u, t)
f (0, 0) = 0, g(0, 0, t) 6= 0
kg(x, u, t)k ≤ δ, ∀ x ∈ Dx , u ∈ Du , t ≥ 0
There is no control u = γ(x), with γ(0) = 0, that can make
the origin of
– p. 13/?
Definition: The system
ẋ = f (x, u) + g(x, u, t)
kx(t)k ≤ ε, ∀t≥T
– p. 14/?
With practical stabilization, we may have
local practical stabilization
regional practical stabilization
global practical stabilization, or
semiglobal practical stabilization
depending on the region of initial states
– p. 15/?
Example
ẋ = x2 + u + d(t), |d(t)| ≤ δ, ∀ t ≥ 0
V = 12 x2 ⇒ V̇ = x3 − kx2 + xd(t)
V̇ ≤ − k3 x2 − x2 k3 − |x| − |x| k3 |x| − δ
V̇ ≤ − k3 x2 , for 3δ
k
≤ |x| ≤ k
3
3δ 3δ
Take k <ε ⇔ k≥ ε
By choosing k large enough we can achieve semiglobal
practical stabilization
– p. 16/?
ẋ = x2 + u + d(t)
V = 21 x2 ⇒ V̇ = −kx2 + xd(t)
V̇ ≤ − k2 x2 − |x| k2 |x| − δ
– p. 17/?
Nonlinear Systems and Control
Lecture # 25
Stabilization
Feedback Lineaization
– p. 1/?
Consider the nonlinear system
ẋ = f (x) + G(x)u
f (0) = 0, x ∈ Rn , u ∈ Rm
Suppose there is a change of variables z = T (x), defined
for all x ∈ D ⊂ Rn , that transforms the system into the
controller form
ż = Az + Bγ(x)[u − α(x)]
u = α(x) + γ −1 (x)v ⇒ ż = Az + Bv
– p. 2/?
v = −Kz
ż = (A − BK)z
– p. 3/?
What can we say about the stability of x = 0 as an
equilibrium point of
−1
ẋ = f (x) + G(x) α(x) − γ (x)KT (x)
– p. 4/?
What information do we need to implement the control
Closed-loop system:
ż = (A − BK)z + Bδ(z)
δ = γ[α̂ − α + γ −1 KT − γ̂ −1 K T̂ ]
– p. 5/?
ż = (A − BK)z + Bδ(z) (∗)
– p. 6/?
Example (Pendulum Equation):
θ̈ = −a sin θ − bθ̇ + cT
a
x1 = θ − δ, x2 = θ̇, u = T − Tss = T − sin δ
c
ẋ1 = x2
ẋ2 = −a[sin(x1 + δ) − sin δ] − bx2 + cu
1
u= {a[sin(x1 + δ) − sin δ] − k1 x1 − k2 x2 }
c
" #
0 1
A − BK = is Hurwitz
−k1 −(k2 + b)
– p. 7/?
a 1
T =u+ sin δ = [a sin(x1 + δ) − k1 x1 − k2 x2 ]
c c
Let â and ĉ be nominal models of a and c
1
T = [â sin(x1 + δ) − k1 x1 − k2 x2 ]
ĉ
ẋ = (A − BK)x + Bδ(x)
âc − aĉ c − ĉ
δ(x) = sin(x1 + δ1 ) − (k1 x1 + k2 x2 )
ĉ ĉ
– p. 8/?
âc − aĉ c − ĉ
δ(x) = sin(x1 + δ1 ) − (k1 x1 + k2 x2 )
ĉ ĉ
|δ(x)| ≤ kkxk + ε
âc − aĉ c − ĉ âc − aĉ
q
k = + 2 2
k +k , ε= | sin δ1 |
1 2
ĉ ĉ ĉ
" # " #
p11 p12 p12
P = , PB =
p12 p22 p22
1
k< q
2 p212 + p222
sin δ1 = 0 ⇒ ε = 0
– p. 9/?
Is feedback linearization a good idea?
Example
ẋ = ax − bx3 + u, a, b > 0
u = −(k + a)x + bx3 , k > 0, ⇒ ẋ = −kx
−bx3 is a damping term. Why cancel it?
u = −(k + a)x, k > 0, ⇒ ẋ = −kx − bx3
Which design is better?
– p. 10/?
Example
ẋ1 = x2
ẋ2 = −h(x1 ) + u
h(0) = 0 and x1 h(x1 ) > 0, ∀ x1 6= 0
Feedback Linearization:
u = h(x1 ) − (k1 x1 + k2 x2 )
V̇ = −x2 σ(x2 )
– p. 12/?
Nonlinear Systems and Control
Lecture # 27
Stabilization
– p. 1/1
Consider the nonlinear system
η̇ = f0 (η, ξ)
ξ̇ = Aξ + Bγ(x)[u − α(x)]
– p. 2/1
u = α(x) + γ −1 (x)v
η̇ = f0 (η, ξ), ξ̇ = Aξ + Bv
Suppose the origin of η̇ = f0 (η, 0) is asymptotically stable
– p. 4/1
Example
η̇ = − 12 (1 + ξ2 )η 3 , ξ̇1 = ξ2 , ξ̇2 = v
e(A−BK)t =
−k2 te−kt (1 − kt)e−kt
– p. 5/1
Peaking Phenomenon:
2 −kt k
max{k te }= → ∞ as k → ∞
t e
η02
η 2 (t) =
1 + η02 [t + (1 + kt)e−kt − 1]
If η02 > 1, the system will have a finite escape time if k is
chosen large enough
– p. 6/1
Lemma 13.2: The origin of
Proof: Use
Lemma 4.7: If ẋ1 = f1 (x1 , x2 ) is ISS and the origin of
ẋ2 = f2 (x2 ) is globally asymptotically stable, then the
origin of
ẋ1 = f1 (x1 , x2 ), ẋ2 = f2 (x2 )
is globally asymptotically stable
– p. 7/1
u = α(x) − γ −1 (x)KT2 (x)
– p. 8/1
Lemma 13.4
If kδ(z)k ≤ ε for all z and η̇ = f0 (η, ξ) is input-to-state
stable, then the state z is globally ultimately bounded by
a class K function of ε
– p. 9/1
Proof–First Part: As in Lemma 13.3
kξ(t)k≤ cε, ∀ t ≥ t0
∂V1
f0 (η, 0) ≤ −c3 kηk2
∂η
∂V1
≤ c4 kηk
∂η
– p. 10/1
V (z) = bV1 (η) + ξT P ξ
" #T " #
kηk kηk
V̇ ≤ − Q
kξk kξk
" #
bc3 −(kkP Bk + bc4 L/2)
Q=
−(kkP Bk + bc4 L/2) 1 − 2kkP Bk
b=k
Q is positive definite for sufficiently small k
– p. 11/1
Nonlinear Systems and Control
Lecture # 28
Stabilization
Backstepping
– p. 1/?
η̇ = f (η) + g(η)ξ
ξ̇ = u, η ∈ Rn , ξ, u ∈ R
η̇ = f (η) + g(η)ξ
η̇ = f (η) + g(η)φ(η)
∂V
[f (η) + g(η)φ(η)] ≤ −W (η), ∀η∈D
∂η
– p. 2/?
z = ξ − φ(η)
∂φ
u= [f (η) + g(η)ξ] + v
∂η
– p. 3/?
Vc (η, ξ) = V (η) + 12 z 2
∂V ∂V
V̇c = [f (η) + g(η)φ(η)] + g(η)z + zv
∂η ∂η
∂V
≤ −W (η) + g(η)z + zv
∂η
∂V
v=− g(η) − kz, k > 0
∂η
V̇c ≤ −W (η) − kz 2
– p. 4/?
Example
z2 = x2 − φ(x1 ) = x2 + x1 + x21
– p. 5/?
Vc (x) = 12 x21 + 21 z22
∂V ∂V
V̇c = (x21 − x31 + x2 ) + (z3 + φ)
∂x1 ∂x2
∂φ 2 ∂φ
3
+ z3 u − (x1 − x1 + x2 ) − (z3 + φ)
∂x1 ∂x2
∂V ∂φ ∂φ
u=− + (x21 − x31 + x2 ) + (z3 + φ) − z3
∂x2 ∂x1 ∂x2
– p. 8/?
η̇ = f (η) + g(η)ξ
ξ̇ = fa (η, ξ) + ga (η, ξ)u, ga (η, ξ) 6= 0
1
u= [v − fa (η, ξ)]
ga (η, ξ)
η̇ = f (η) + g(η)ξ
ξ̇ = v
– p. 9/?
Strict-Feedback Form
ẋ = f0 (x) + g0 (x)z1
ż1 = f1 (x, z1 ) + g1 (x, z1 )z2
ż2 = f2 (x, z1 , z2 ) + g2 (x, z1 , z2 )z3
..
.
żk−1 = fk−1 (x, z1 , . . . , zk−1 ) + gk−1 (x, z1 , . . . , zk−1 )zk
żk = fk (x, z1 , . . . , zk ) + gk (x, z1 , . . . , zk )u
gi (x, z1 , . . . , zi ) 6= 0 for 1 ≤ i ≤ k
– p. 10/?
Example
η̇ = −η + η 2 ξ, ξ̇ = u
η̇ = −η + η 2 ξ
ξ = 0 ⇒ η̇ = −η
V0 = 12 η 2 ⇒ V̇0 = −η 2 , ∀ η ∈ R
V = 12 (η 2 + ξ2 )
V̇ = η(−η + η 2 ξ) + ξu = −η 2 + ξ(η 3 + u)
u = −η 3 − kξ, k>0
V̇ = −η 2 − kξ2 Global stabilization
– p. 11/?
Nonlinear Systems and Control
Lecture # 29
Stabilization
Passivity-Based Control
– p. 1/?
ẋ = f (x, u), y = h(x)
f (0, 0) = 0
T ∂V
u y ≥ V̇ = f (x, u)
∂x
Theorem 14.4: If the system is
(1) passive with a radially unbounded positive definite
storage function and
(2) zero-state observable,
then the origin can be globally stabilized by
– p. 2/?
Proof:
∂V
V̇ = f (x, −φ(y)) ≤ −y T φ(y) ≤ 0
∂x
– p. 3/?
Choice of Output
∂V
ẋ = f (x) + G(x)u, f (x) ≤ 0, ∀x
∂x
No output is defined. Choose the output as
T
∂V
def
y = h(x) = G(x)
∂x
∂V ∂V
V̇ = f (x) + G(x)u ≤ y T u
∂x ∂x
Check zero-state observability
– p. 4/?
Example
ẋ1 = x2 , ẋ2 = −x31 + u
– p. 5/?
Feedback Passivation
u = α(x) + β(x)v
such that
is passive
– p. 6/?
Theorem [31]: The system
– p. 7/?
Stabilize the system at q = qr
e = q − qr , ė = q̇
– p. 9/?
How does passivity-based control compare with feedback
linearization?
Example 13.20
u = h(x1 ) − (k1 x1 + k2 x2 )
" #
0 1
ẋ = x
−k1 −k2
– p. 10/?
Passivity-based control:
Z x1
V = h(z) dz + 12 x22
0
V̇ = x2 h(x1 ) − x2 h(x1 ) + x2 u = x2 u
Take y = x2
With u = 0, y(t) ≡ 0 ⇒ h(x1 (t)) ≡ 0 ⇒ x1 (t) ≡ 0
u = −σ(x2 ), [σ(0) = 0, yσ(y) > 0 ∀ y 6= 0]
ẋ1 = x2 , ẋ2 = −h(x1 ) − σ(x2 )
– p. 11/?
Linearization:
" #
0 1
, k = σ ′ (0)
−h′ (0) −k
s2 + ks + h′ (0) = 0
Sketch the root locus as k varies from zero to infinity
One of the p
two roots cannot be moved to the left of
Re[s] = − h′ (0)
– p. 12/?
Cascade Connection:
– p. 13/?
T
∂W
u=− F (z, y) + v ⇒ U̇ ≤ y T v
∂z
The system
– p. 14/?
Nonlinear Systems and Control
Lecture # 30
Stabilization
– p. 1/1
ẋ = f (x) + g(x)u, f (0) = 0, x ∈ Rn , u ∈ R
ẋ = f (x) + g(x)ψ(x)
is asymptotically stable
∂V
[f (x) + g(x)u] < 0
∂x
Small Control Property
– p. 3/1
Definition: A continuously differentiable positive definite
function V (x) is a Control Lyapunov Function (CLF) for the
system ẋ = f (x) + g(x)u if
∂V ∂V
g(x) = 0 for x ∈ D, x 6= 0 ⇒ f (x) < 0 (∗)
∂x ∂x
it satisfies the small control property
It is a Global Control Lyapunov Function if it is radially
unbounded and (∗) holds with D = Rn
Sontag’s Formula
– p. 5/1
Proof: For properties of ψ , see Section 9.4 of [88]
∂V
[f (x) + g(x)ψ(x)]
∂x
∂V ∂V
If g(x) = 0, V̇ = f (x) < 0 for x 6= 0
∂x ∂x
∂V
If g(x) 6= 0
∂x
" r #
2 4
∂V ∂V ∂V ∂V
V̇ = ∂x f − ∂x f + ∂x f + ∂x g
r 2 4
∂V ∂V
= − ∂x f + ∂x g < 0 for x 6= 0
– p. 6/1
How can we find a CLF?
Backstepping
– p. 7/1
Example:
ẋ = ax − bx3 + u, a, b > 0
Feedback Linearization:
ẋ = −kx
V (x) = 12 x2 is a CLF
∂V ∂V
g = x, f = x(ax − bx3 )
∂x ∂x
– p. 8/1
r 2 4
∂V ∂V ∂V
∂x f + ∂x f + ∂x g
−
∂V
∂x g
p
x(ax − bx3 ) + x2 (ax − bx3 )2 + x4
= −
x
q
= −ax + bx3 − x (a − bx2 )2 + 1
q
ψ(x) = −ax + bx3 − x (a − bx2 )2 + 1
Compare with
−ax + bx3 − kx
– p. 9/1
Method Expression
FL-u −ax + bx3 − kx
FL-CLS ẋ = −kx
p
CLF-u 3
−ax + bx − x (a − bx2 )2 + 1
p
CLF-CLS −x (a − bx2 )2 + 1
– p. 10/1
Lemma: Let V (x) be a CLF for ẋ = f (x) + g(x)u and
suppose ∂V ∂x (0) = 0. Then, Sontag’s formula has a gain
margin [ 12 , ∞); that is, u = kψ(x) is stabilizing for all k ≥ 1
2
Proof: Let
s
2 4
∂V ∂V ∂V
1
q(x) = 2
− f+ f + g
∂x ∂x ∂x
∂V
q(0) = 0, g 6= 0 ⇒ q > 0
∂x
∂V ∂V
g=0 ⇒ q=− f > 0 for x 6= 0
∂x ∂x
q(x) is positive definite
– p. 11/1
u = kψ(x) ⇒ ẋ = f (x) + g(x)kψ(x)
∂V ∂V
V̇ = f+ gkψ
∂x ∂x
∂V ∂V
g = 0 ⇒ V̇ = f < 0 for x 6= 0
∂x ∂x
∂V ∂V ∂V
g 6= 0, V̇ = −q + q + f+ gkψ
∂x ∂x ∂x
∂V ∂V
q+ f+
gkψ
∂x ∂x
s
2 4
1 ∂V ∂V ∂V
= − k−2 f+ f + g
∂x ∂x ∂x
– p. 12/1
Nonlinear Systems and Control
Lecture # 31
Stabilization
Output Feedback
– p. 1/1
In general, output feedback stabilization requires the use of
observers. In this lecture we deal with three simple cases
where an observer is not needed
Minimum Phase Relative Degree One Systems
Passive systems
– p. 2/1
Minimum Phase Relative Degree One Systems
– p. 3/1
Assumptions:
The origin of η̇ = f0 (η, 0) is exponentially stable
∂V1
f0 (η, 0) ≤ −c3 kηk2
∂η
∂V1
≤ c4 kηk
∂η
– p. 4/1
High-Gain Feedback:
u = −ky, k>0
– p. 5/1
V̇ ≤ −c3 kηk2 + c4 L1 kηk |y| − kγ0 y 2 + L2 kηk |y| + L3 y 2
" #T " #" #
kηk c3 − 21 (L2 + c4 L1 ) kηk
V̇ ≤ −
|y| − 21 (L2 + c4 L1 ) (kγ0 − L3 ) |y|
| {z }
Q
– p. 6/1
Passive Systems
V̇ ≤ uT y = −y T ψ(y) ≤ 0
V̇ = 0 ⇒ y = 0 ⇒ u = 0
y(t) ≡ 0 ⇒ x(t) ≡ 0
– p. 7/1
Systems with Passive Maps from u to ẏ
V̇ ≤ uT ẏ
– p. 8/1
+ h u- y
- Plant -
−6
z
ψ(·) bs
s+a
+ h u- y ẏ
- Plant - s -
−6
z
ψ(·) b
s+a
– p. 9/1
For 1 ≤ i ≤ m
bi s
zi is the output of driven by yi
s + ai
ui = −ψi (zi )
ai , bi > 0, ψi (0) = 0, zi ψi (zi ) > 0 ∀ zi 6= 0
żi = −ai zi + bi ẏi
Use
m
X Z zi
1
Vc (x, z) = V (x) + bi ψi (σ) dσ
i=1 0
– p. 10/1
Example: Stabilize the pendulum
mℓθ̈ + mg sin θ = u
x1 = θ − δ1 , x2 = θ̇
1
ẋ1 = x2 , ẋ2 = mℓ [−mg sin θ + u]
u = mg sin θ − kp x1 + v, kp > 0
kp 1
ẋ1 = x2 , ẋ2 = − mℓ x1 + mℓ v
y = x1 , ẏ = x2
– p. 11/1
V = 12 kp x21 + 12 mℓx22
h i
kp 1
V̇ = kp x1 x2 + mℓx2 − mℓ x1 + mℓ
v
V̇ = x2 v = ẏv
With v = 0, x2 (t) ≡ 0 ⇒ x1 (t) ≡ 0
y z
- bs -
s+a
ξ̇ = −aξ + y, z = b(−aξ + y)
v = −kd z, kd > 0
u = mg sin θ − kp (θ − δ1 ) − kd z
– p. 12/1
Nonlinear Systems and Control
Lecture # 32
Robust Stabilization
– p. 1/1
Example
s = a1 x1 + x2 = 0
– p. 2/1
ṡ = a1 ẋ1 + ẋ2 = a1 x2 + h(x) + g(x)u
Suppose
a1 x2 + h(x)
≤ ̺(x)
g(x)
V = 12 s2
V̇ ≤ g(x)|s|̺(x) − g(x)β(x)|s|
– p. 3/1
s < 0, u = β(x)
u = −β(x) sgn(s)
– p. 4/1
√
V̇ ≤ −g0 β0 2V
dV √
√ ≤ −g0 β0 2 dt
V
√ V (s(t)) √
2 V ≤ −g0 β0 2 t
V (s(0))
p p 1
V (s(t)) ≤ V (s(0)) − g0 β0 √ t
2
|s(t)| ≤ |s(0)| − g0 β0 t
s(t) reaches zero in finite time
Once on the surface s = 0, the trajectory cannot leave it
– p. 5/1
s=0
– p. 6/1
ẋ1 = x2 ẋ2 = h(x) − g(x)β(x)sgn(s)
u = −k sgn(s)
– p. 8/1
Chattering
B Sliding manifold
B
B
B
B
B
H
HH B
s < 0 BH Y s>0
B
B
B
H
HH B a
B HH
B YH
H
– p. 9/1
Reduce the amplitude of the signum function
ṡ = a1 x2 + h(x) + g(x)u
[a1 x2 + ĥ(x)]
u=− +v
ĝ(x)
ṡ = δ(x) + g(x)v
g(x) g(x)
δ(x) = a1 1 − x2 + h(x) − ĥ(x)
ĝ(x) ĝ(x)
δ(x)
≤ ̺(x), β(x) ≥ ̺(x) + β0
g(x)
v = −β(x) sgn(s)
– p. 10/1
Replace the signum function by a high-slope saturation
function
s
u = −β(x) sat
ε
(
y, if |y| ≤ 1
sat(y) =
sgn(y), if |y| > 1
y
sgn(y) 6 sat ε 6
1 1
- -
y ε y
−1 −1
– p. 11/1
How can we analyze the system?
With c ≥ ε
n o
c
Ω = |x1 | ≤ a1
, |s| ≤ c is positively invariant
– p. 12/1
Inside the boundary layer:
s
ẋ1 = −a1 x1 + s ṡ = a1 x2 + h(x) − g(x)β(x)
ε
in finite time
– p. 13/1
What happens inside Ωε ?
h(0) = 0 ⇒ x̄1 = 0
– p. 14/1
z1 = x1 − x̄1 , z2 = s − a1 x̄1
s
ż2 = a1 x2 + h(x) − g(x)β(x)
ε
z2 + a1 x̄1
= a1 (z2 − a1 z1 ) + h(x) − g(x)β(x)
ε
z2
ż2 = ℓ(z) − g(x)β(x)
ε
h(x) x̄1
ℓ(z) = a1 (z2 − a1 z1 ) + a1 g(x)β(x) −
a1 g(x)β(x) ε
– p. 15/1
z2
ż1 = −a1 z1 + z2 , ż2 = ℓ(z) − g(x)β(x)
ε
ℓ(0) = 0, |ℓ(z)| ≤ ℓ1 |z1 | + ℓ2 |z2 |
g(x)β(x) ≥ g0 β0
V = 21 z12 + 12 z22
z2
V̇ = z1 (−a1 z1 + z2 ) + z2 ℓ(z) − g(x)β(x)
ε
g0 β0
V̇ ≤ −a1 z12 + (1 + ℓ1 )|z1 | |z2 | + ℓ2 z22 − z22
ε
– p. 16/1
g0 β0
V̇ ≤ −a1 z12 + (1 + ℓ1 )|z1 | |z2 | + ℓ2 z22
− z22
" #T " # ε
" #
1
|z1 | a1 − 2 (1 + ℓ1
) |z1 |
V̇ ≤ −
g0 β0
|z2 | − 12 (1 + ℓ1 ) ε − ℓ2 |z2 |
| {z }
Q
g0 β0
det(Q) = a1 − ℓ2 − 41 (1 + ℓ1 )2
ε
h(0) = 0 ⇒ lim x(t) = 0
"t→∞ #
x̄1
h(0) =
6 0 ⇒ lim x(t) =
t→∞ 0
Robust Stabilization
– p. 1/1
Regular Form:
η̇ = fa (η, ξ)
ξ̇ = fb (η, ξ) + g(η, ξ)u + δ(t, η, ξ, u)
η ∈ Rn−1 , ξ ∈ R, u ∈ R
fa (0, 0) = 0, fb (0, 0) = 0, g(η, ξ) ≥ g0 > 0
Sliding Manifold:
s = ξ − φ(η) = 0, φ(0) = 0
– p. 2/1
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + g(η, ξ)u + δ(t, η, ξ, u)
∂η
1 ˆ ∂φ ˆ
u=− fb − fa + v or u = v
ĝ ∂η
∂φ ˆ 1
ˆ
u = −L fb − fa + v, L = or L = 0
∂η ĝ
ṡ = g(η, ξ)v + ∆(t, η, ξ, v)
∂φ ∂φ ˆ
∆ = fb − ˆ
fa + δ − gL fb − fa
∂η ∂η
∆(t, η, ξ, v)
g(η, ξ) ≤ ̺(η, ξ) + κ0 |v|
– p. 3/1
∆(t, η, ξ, v)
g(η, ξ) ≤ ̺(η, ξ) + κ0 |v|
– p. 4/1
sṡ ≤ −g(η, ξ)(1 − κ0 )β0 |s| ≤ −g0 β0 (1 − κ0 )|s|
s
v = −β(x) sat , ε>0
ε
sṡ ≤ −g0 β0 (1 − κ0 )|s|, for |s| ≥ ε
The trajectory reaches the boundary layer {|s| ≤ ε} in finite
time and remains inside thereafter
η̇ = fa (η, φ(η) + s)
– p. 5/1
α1 (kηk) ≤ V (η) ≤ α2 (kηk)
∂V
fa (η, φ(η) + s) ≤ −α3 (kηk), ∀ kηk ≥ γ(|s|)
∂η
|s| ≤ c ⇒ V̇ ≤ −α3 (kηk), for kηk ≥ γ(c)
α(r) = α2 (γ(r))
– p. 6/1
V
c α(.)
0
α(c)
α(ε)
ε c |s|
– p. 7/1
Theorem 14.1: Suppose all the assumptions hold over Ω.
Then, for all (η(0), ξ(0)) ∈ Ω, the trajectory (η(t), ξ(t)) is
bounded for all t ≥ 0 and reaches the positively invariant
set Ωε in finite time. If the assumptions hold globally and
V (η) is radially unbounded, the foregoing conclusion holds
for any initial state
Theorem 14.2: Suppose all the assumptions hold over Ω
̺(0) = 0, κ0 = 0
The origin of η̇ = fa (η, φ(η)) is exponentially stale
Then there exits ε∗ > 0 such that for all 0 < ε < ε∗ , the
origin of the closed-loop system is exponentially stable and
Ω is a subset of its region of attraction. If the assumptions
hold globally, the origin will be globally uniformly
asymptotically stable
– p. 8/1
Example
|θ1 | ≤ a, |θ2 | ≤ b
x2 = −kx1 ⇒ ẋ1 = −kx1 + θ1 x1 sin x2
V1 = 12 x21 ⇒ x1 ẋ1 ≤ −kx21 + ax21
s = x2 + kx1 , k>a
ṡ = θ2 x22 + x1 + u + k(x2 + θ1 x1 sin x2 )
u = −x1 − kx2 + v ⇒ ṡ = v + ∆(x)
∆(x) = θ2 x22 + kθ1 x1 sin x2
– p. 9/1
∆(x) = θ2 x22 + kθ1 x1 sin x2
– p. 10/1
Example: Normal Form
η̇ = f0 (η, ξ)
ξ̇i = ξi+1 , 1≤i≤ρ−1
ρ ρ−1
ξ̇ρ = Lf h(x) + Lg Lf h(x) u
y = ξ1
η̇ = f0 (η, ξ1 , · · · , ξρ−1 , ξρ )
ξ̇i = ξi+1 , 1≤i≤ρ−2
ξ̇ρ−1 = ξρ
s = ξρ + k1 ξ1 + · · · + kρ−1 ξρ−1
– p. 12/1
Multi-Input Systems
η̇ = fa (η, ξ)
ξ̇ = fb (η, ξ) + G(η, ξ)E(η, ξ)u + δ(t, η, ξ, u)
η ∈ Rn−p , ξ ∈ Rp , u ∈ Rp
fa (0, 0) = 0, fb (0, 0) = 0, det(G) 6= 0, det(E) 6= 0
G = diag[g1 , g2 , · · · , gm ], gi (η, ξ) ≥ g0 > 0
Design φ s.t. the origin of η̇ = fa (η, φ(η)) is asymp. stable
s = ξ − φ(η)
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + G(η, ξ)E(η, ξ)u + δ(t, η, ξ, u)
∂η
– p. 13/1
∂φ
ṡ = fb (η, ξ) − fa (η, ξ) + G(η, ξ)E(η, ξ)u + δ(t, η, ξ, u)
∂η
∂φ ˆ
u=E −1 ˆ
−L fb − fa + v , L = Ĝ−1 or L = 0
∂η
ṡi = gi (η, ξ)vi + ∆i (t, η, ξ, v), 1 ≤ i ≤ p
∆i (t, η, ξ, v)
≤ ̺(η, ξ) + κ0 max |vi |, ∀ 1 ≤ i ≤ p
g (η, ξ) 1≤i≤p
i
– p. 14/1
si ṡi = si gi vi + si ∆i ≤ gi {si vi + |si |[̺ + κ0 max |vi |]}
1≤i≤p
vi = −β sgn(si ), 1≤i≤p
Now use
si
vi = −β sat , 1≤i≤p
ε
Read Theorem 14.1 and 14.2 in the textbook
– p. 15/1
Nonlinear Systems and Control
Lecture # 34
Robust Stabilization
– p. 1/?
Lyapunov Redesign (Min-max control)
– p. 2/?
T ∂V
w = G
∂x
V̇ ≤ −W (x) + w T v + w T δ
w T v +w T δ ≤ w T v +kwk kδk ≤ w T v +kwk[ρ(x)+κ0 kvk]
w w
v = −η(x) = sgn(w) for p = 1
kwk kwk
ρ(x)
η(x) ≥ ⇒ w T v + w T δ ≤ 0 ⇒ V̇ ≤ −W (x)
(1 − κ0 )
– p. 3/?
−η(x) w , if η(x)kwk ≥ ε
kwk
v=
−η 2 (x) w , if η(x)kwk < ε
ε
η(x)kwk ≥ ε ⇒ V̇ ≤ −W (x)
For η(x)kwk < ε
T 2 w
V̇ ≤ −W (x) + w −η · +δ
ε
η2
≤ −W (x) − kwk2 + ρkwk + κ0 kwkkvk
ε
η2 κ0 η 2
= −W (x) − kwk2 + ρkwk + kwk2
ε ε
– p. 4/?
!
η2
V̇ ≤ −W (x) + (1 − κ0 ) − kwk2 + ηkwk
ε
y2 ε
− +y ≤ , for y ≥ 0
ε 4
(1 − κ0 )
V̇ ≤ −W (x) + ε , ∀x∈D
4
Theorem 14.3: x(t) is uniformly ultimately bounded by a
class K function of ε. If the assumptions hold globally and
V is radially unbounded, then x(t) globally uniformly
ultimately bounded
Corollary 14.1: If ρ(0) = 0 and η(x) ≥ η0 > 0 we can
recover uniform asymptotic stability
– p. 5/?
Example: Pendulum with horizontal acceleration of
suspension point
h i
m ℓθ̈ + A(t) cos θ = T /ℓ − mg sin θ
g 1 A(t)
x1 = θ − π, x2 = θ̇, a = , c= , h(t) =
ℓ mℓ2 ℓ
ẋ1 = x2 , ẋ2 = a sin x1 + cu + h(t) cos x1
Nominal Model: ẋ1 = x2 , ẋ2 = â sin x1 + ĉu
â 1
ψ(x) = − sin x1 − (k1 x1 + k2 x2 )
ĉ ĉ
– p. 6/?
" #
0 1
, V (x) = xT P x
−k1 −k2
| {z }
Hurwitz
1 aĉ − âc
δ = sin x1 + h(t) cos x1
ĉ ĉ
c − ĉ c − ĉ
− (k1 x1 + k2 x2 ) + v
ĉ ĉ
q
c − ĉ âc − aĉ c − ĉ
≤ κ0 , + k2 + k2 ≤ k, |h(t)| ≤ H
ĉ 1 2
ĉ ĉ
(kkxk + H) def
|δ| ≤ + κ0 |v| = ρ(x) + κ0 |v|, (κ0 < 1)
ĉ
– p. 7/?
ρ(x) H
η(x) = , η(x) ≥
(1 − κ0 ) ĉ(1 − κ0 )
" #
∂V T 0
w= G = 2x P = 2(p12 x1 + p22 x2 )
∂x 1
−η(x)sgn(w), if η(x)|w| ≥ ε
v=
−η 2 (x) w , if η(x)|w| < ε
ε
â 1
u=− sin x1 − (k1 x1 + k2 x2 ) + v
ĉ ĉ
Will this control stabilize the origin x = 0?
– p. 8/?
Backstepping
gi 6= 0, 1≤i≤k
– p. 9/?
ż1 = f1 + g1 z2 + δ1 (z)
ż2 = f2 + g2 z3 + δ2 (z)
..
.
żk−1 = fk−1 + gk−1 zk + δk−1 (z)
żk = fk + gk u + δk (z)
|θ1 | ≤ a, |θ2 | ≤ b
δ1 = θ1 x1 sin x2 , δ2 = θ2 x22
ẋ1 = x2 + θ1 x1 sin x2 , |θ1 x1 sin x2 | ≤ a|x1 |
x2 = −k1 x1
V1 = 12 x21 , V̇1 ≤ −(k1 − a)x21 ; Take k1 = 1 + a
z2 = x2 + (1 + a)x1
– p. 11/?
ẋ1 = −(1 + a)x1 + θ1 x1 sin x2 + z2
ż2 = ψ1 (x) + ψ2 (x, θ) + u
Vc = 21 x21 + 12 z22
– p. 12/?
ψ2 = (1 + a)θ1 x1 sin x2 + θ2 x22
x2 = z2 − (1 + a)x1
|ψ2 | ≤ (1 + a)(a + bρ)|x1 | + bρ|z2 |
– p. 13/?
Second Approach (Example 14.14):
Tracking
– p. 1/1
SISO relative-degree ρ system:
f (0) = 0, h(0) = 0
ρ−1
Lg Li−1
f h(x) = 0, for 1 ≤ i ≤ ρ − 1, Lg Lf h(x) 6= 0
Normal form:
η̇ = f0 (η, ξ)
ξ̇i = ξi+1 , 1≤i≤ρ−1
ξ̇ρ = Lρf h(x) + Lg Lρ−1
f h(x)u
y = ξ1
f0 (0, 0) = 0
– p. 2/1
Reference signal r(t)
r(t) and its derivatives up to r (ρ) (t) are bounded for all
t ≥ 0 and the ρth derivative r (ρ) (t) is a piecewise
continuous function of t;
the signals r ,. . . ,r (ρ) are available on-line.
r ξ1 − r
R= ..
e= ..
. , . =ξ−R
r (ρ−1) ξρ − r (ρ−1)
– p. 3/1
η̇ = f0 (η, e + R)
h i
ρ ρ−1
ė = Ac e + Bc Lf h(x) + Lg Lf h(x)u − r (ρ)
0 1 0 ... 0 0
0 0 1 ... 0
0
.. ... .. ..
Ac = . . , Bc =
.
..
. 0
1 0
0 ... ... 0 0 1
1 h
ρ
i
u= ρ−1 −Lf h(x) + r (ρ) + v
Lg Lf h(x)
ė = Ac e + Bc v
– p. 4/1
v = −Ke ⇒ ė = (Ac − Bc K) e
| {z }
H urwitz
Example 13.21
e1 = x1 − r, e2 = x2 − ṙ
ė1 = e2 , ė2 = −a sin x1 − bx2 + cu − r̈
1
u= [a sin x1 + bx2 + r̈ − k1 e1 − k2 e2 ]
c
ė1 = e2 , ė2 = −k1 e1 − k2 e2
See simulation in the textbook
– p. 6/1
Sliding Mode Control
η̇ = f0 (η, ξ)
ξ̇1 = ξ2
.. ..
. .
ξ̇ρ−1 = ξρ
ρ ρ−1
ξ̇ρ = Lf h(x) + Lg Lf h(x)[u + δ(t, x, u)]
y = ξ1
e=ξ−R
– p. 7/1
η̇ = f0 (η, ξ)
ė1 = e2
.. ..
. .
ėρ−1 = eρ
ėρ = Lρf h(x) + Lg Lρ−1
f h(x)[u + δ(t, x, u)] − r (ρ)
(t)
Sliding surface:
– p. 8/1
η̇ = f0 (η, ξ)
ė1 = e2
.. ..
. .
ėρ−1 = −(k1 e1 + · · · + kρ−1 eρ−1 )
– p. 9/1
ρ−1
X
s = (k1 e1 + · · · + kρ−1 eρ−1 ) + eρ = ki ei + eρ
i=1
ρ−1
X ρ ρ−1
ṡ = ki ei+1 +Lf h(x)+Lg Lf h(x)[u+δ(t, x, u)]−r (ρ) (t)
i=1
"ρ−1 #
1 X
u=− ρ−1 ki ei+1 + Lρf h(x) − r (ρ) (t) + v
Lg Lf h(x) i=1
ṡ = Lg Lρ−1
f h(x)v + ∆(t, x, v)
∆(t, x, v)
ρ−1 ≤ ̺(x) + κ0 |v|, 0 ≤ κ0 < 1
Lg L h(x)
f
– p. 10/1
s
v = −β(x) sat , ε>0
ε
̺(x)
β(x) ≥ + β0 , β0 >
(1 − κ0 )
What properties can we prove for this control?
– p. 11/1
Nonlinear Systems and Control
Lecture # 36
Tracking
Equilibrium-to-Equilibrium
Transition
– p. 1/1
η̇ = f0 (η, ξ)
ξ̇i = ξi+1 , 1≤i≤ρ−1
ρ ρ−1
ξ̇ρ = Lf h(x) + Lg Lf h(x) u
| {z } | {z }
fb (η,ξ) gb (η,ξ)
y = ξ1
Equilibrium point:
0 = f0 (η̄, ξ̄)
0 = ξ̄i+1 , 1≤i≤ρ−1
0 = fb (η̄, ξ̄) + gb (η̄, ξ̄)ū
ȳ = ξ̄1
– p. 2/1
ξ̄1 = ȳ, ξ̄i = 0 for 2 ≤ i ≤ ρ − 1
fb (η̄, ȳ, 0, · · · , 0)
0 = f0 (η̄, ȳ, 0, · · · , 0), ū = −
gb (η̄, ȳ, 0, · · · , 0)
Assume f0 (η̄, ȳ, 0, · · · , 0) has a unique solution η̄ in the
domain of interest
η̄ = φη (ȳ), ū = φu (ȳ)
φη (0) = 0, φu (0) = 0
– p. 3/1
Goal: Move the system from equilibrium at y = 0 to
equilibrium at y = ȳ , either asymptotically or over a finite
time period
Is this allowed ?
r (i) = 0 for i ≥ 2
– p. 4/1
η(0) = 0, e1 (0) = −ȳ, ei (0) = 0 for i ≥ 2
Feedback Linearization:
– p. 6/1
The derivatives of r are generated by the pre-filter
1
. ..
z + ∗
ż = y
1
−aρ −a1 aρ
h i
r = 1 z
r = z1 , ṙ = z2 , . . . . . . r (ρ−1) = zρ
ρ
X
r (ρ) = − aρ−i+1 zi + aρ y ∗
i=1
Constraint : |u(t)| ≤ 2
(
∗ 0, for t < 0
y = π
2 for t ≥ 0
Pre-Filter:
1
(τ s + 1)2
– p. 8/1
" # " #
0 1 0
ż = −1 −2 z+ 1 y∗
τ2 τ τ2
h i
r = 1 0 z
1 ∗ 2
r = z1 , ṙ = z2 , r̈ = (y − r) − z2
τ2 τ
h t i
r(t) = π
2
1 − e− τ 1 + τt
u = 0.1(10 sin x1 + x2 + r̈ − k1 e1 − k2 e2 )
– p. 9/1
τ = 0.05 τ = 0.25
2 2
1.5 1.5
Output
Output
1 1
output output
reference reference
0.5 0.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
τ = 0.05 τ = 0.25
2 2
1 1
Control
0 Control 0
−1 −1
−2 −2
– p. 10/1
Third Approach: Plan a trajectory (r(t), ṙ(t), . . . , r (ρ) (t))
to move from (0, 0, . . . , 0) to (ȳ, 0, . . . , 0) in finite time T
Example: ρ = 2
a
r(2)
0 T
T/2
−a
r(1)
at a(T−t)
aT2/4
r
−aT2/4+aTt−at2/2
at2/2
– p. 11/1
at2
2
for 0 ≤ t ≤ T2
aT 2 at2
r(t) = − 4 + aT t − 2 for T2 ≤ t ≤ T
aT 2
4 for t ≥ T
4ȳ
a= ⇒ r(t) = ȳ for t ≥ T
T2
– p. 12/1
Nonlinear Systems and Control
Lecture # 37
Observers
Linearization
and
Extended Kalman Filter (EKF)
– p. 1/1
Linear Observer via Linearization
(A − HC) is Hurwitz
It will work locally for sufficiently small kxδ (0)k, kx̂δ (0)k,
and kuδ (t)k
– p. 2/1
Feedback Control:
x̃ = x − x̂
x̃˙ = (A − HC)x̃
– p. 4/1
Extended Kalman Filter (EKF)
ẋ = f (x, u) + w, y = h(x, u) + v
∂f ∂h
A(t) = (x̂(t), u(t)), C(t) = (x̂(t), u(t))
∂x ∂x
η(0, t) = 0, ξ(t) = w(t) − H(t)v(t)
– p. 5/1
Assuming that x(t), u(t), w(t), v(t), and H(t) are
bounded and f and h are twice continuously differentiable,
show that
Hint:
∂f
f (x, u) − f (x̂, u) − (x̂, u)x̃
∂x
Z 1
∂f ∂f
= (σ x̃ + x̂, u) dσ x̃ − (x̂, u)x̃ (Exercise 3.23)
0 ∂x ∂x
Z 1
∂f ∂f
= (σ x̃ + x̂, u) − (x̂, u) dσ x̃
0 ∂x ∂x
– p. 6/1
Kalman Filter Design: Let Q(t) and R(t) be symmetric
positive definite matrices that satisfy
Compute H(t)
– p. 8/1
Lemma: The origin of
Proof:
V = x̃T P −1 x̃
−1 ˙ ˙ T −1 T d
T
V̇ = x̃ P x̃ + x̃ P x̃ + x̃ P −1 x̃
dt
– p. 9/1
d
P −1 = −P −1 Ṗ P −1
dt
V̇ = x̃T P −1 (A − P C T R−1 C)x̃
+ x̃T (AT − C T R−1 CP )P −1 x̃
− x̃T P −1 Ṗ P −1 x̃ + 2x̃T P −1 (η + ξ)
– p. 10/1
Stochastic Interpretation: When w(t) and v(t) are
zero-mean, white noise stochastic processes,
uncorrelated, i.e., E{w(t)v T (τ )} = 0, ∀t, τ , and
E{w(t)w T (τ )} = Q(t)δ(t − τ )
E{v(t)v T (τ )} = R(t)δ(t − τ )
then x̂(t) is an approximation of the minimum variance
estimate that minimizes
n o
E [y(t) − h(x̂(t), u(t))]T [y(t) − h(x̂(t), u(t))]
– p. 11/1
Feedback Control: What can you say about the closed-loop
system when x̂ is used in feedback control?
– p. 12/1
Nonlinear Systems and Control
Lecture # 38
Observers
Exact Observers
– p. 1/1
Observer with Linear Error Dynamics
Observer Form:
ẋ = Ax + γ(y, u), y = Cx
τi = (−1)i−1 adi−1
f τ, 1≤i≤n
∂T h i
τ1 , τ2 , ··· τn =I
∂x
z = T (x)
– p. 3/1
ẋ = Ax + γ(y, u), y = Cx
– p. 4/1
How would you analyze the closed-loop system?
We know that
the origin of ẋ = Ax + γ(Cx, ψ(x)) is globally
asymptotically stable
the origin of x̃˙ = (A − HC)x̃ is globally exponentially
stable
What additional assumptions do we need to show that the
origin of the closed-loop system is globally asymptotically
stable?
– p. 5/1
Circle Criterion Design
x̃ = x − x̂
x̃˙ = (A − HC)x̃ − L[β(M x) − β(M x̂ − N (y − C x̂))]
def
G(s) = (M + N C)[sI − (A − HC)]−1 L
0 +
- n -
z -
G(s)
−6
ψ(·)
h iT
ψ(t, z) = ψ1 (t, z1 ), . . . , ψℓ (t, zℓ )
– p. 7/1
Main Assumption: βi (·) is a nondecreasing function
– p. 8/1
By the circle criterion (Theorem 7.1) the origin of
−1 s + N + h1
G(s) = (M + N C)[sI − (A − HC)] L=
s2 + h1 s + h2
– p. 10/1
From Exercise 6.7, G(s) is SPR if and only if
– p. 11/1
What about feedback control?
Let u = φ(x) be a globally stabilizing state feedback control
Closed-loop system under output feedback:
Observers
High-Gain Observers
Motivating Example
– p. 1/1
ẋ1 = x2 , ẋ2 = φ(x, u), y = x1
Observer:
– p. 2/1
" # " #
h1 −h1 1
Design H = such that Ao = is Hurwitz
h2 −h2 0
Transfer function from δ to x̃:
" #
1 1
Go (s) =
s2 + h1 s + h2 s + h1
Observer eigenvalues are (λ1 /ε) and (λ2 /ε) where λ1 and
λ2 are the roots of
λ2 + α1 λ + α2 = 0
– p. 4/1
x̃1
η1 = , η2 = x̃2
ε
εη̇1 = −α1 η1 + η2 , εη̇2 = −α2 η1 + εδ(x, x̃)
Ultimate bound of η is O(ε)
η decays faster than an exponential mode e−at/ε , a>0
Peaking Phenomenon:
1
The solution contains a term of the form e−at/ε
ε
1
e−at/ε approaches an impulse function as ε → 0
ε
– p. 5/1
Example
u = −x32 − x1 − x2
– p. 6/1
0.5
0 SFB
−0.5 OFB ε = 0.1
1
OFB ε = 0.01
x
−1
−1.5
OFB ε = 0.005
−2
0 1 2 3 4 5 6 7 8 9 10
−1
2
x
−2
−3
0 1 2 3 4 5 6 7 8 9 10
−100
u
−200
−300
−400
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
t
– p. 7/1
ε = 0.004
0.2
0
1
−0.2
x
−0.4
−0.6
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
−200
2
x
−400
−600
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
2000
1000
u
−1000
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
t
– p. 8/1
u = sat(−x̂32 − x̂1 − x̂2 )
SFB
0.15
OFB ε = 0.1
0.1 OFB ε = 0.01
OFB ε = 0.001
x1
0.05
−0.05
0 1 2 3 4 5 6 7 8 9 10
0.05
0
x2
−0.05
−0.1
0 1 2 3 4 5 6 7 8 9 10
0
u
−0.5
−1
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
t
– p. 9/1
Region of attraction under state feedback:
2
x2
0
−1
−2
−3 −2 −1 0 1 2 3
x1
– p. 10/1
Region of attraction under outputfeedback:
0.5
x2
−0.5
−1
– p. 12/1
What is the effect of measurement noise?
– p. 13/1
1.1
0.9
0.8
0.6
0.5
0.4
0.3
0.2
0.1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
the high−gain parameter epsilon
s !
kn
εopt = O , kd = sup |ẍ1 (t)|, kn = sup |v(t)|
kd t≥0 t≥0
– p. 14/1
Nonlinear Systems and Control
Lecture # 40
Observers
High-Gain Observers
Stabilization
– p. 1/1
ẋ = Ax + Bφ(x, z, u)
ż = ψ(x, z, u)
y = Cx
ζ = q(x, z)
u ∈ Rp , y ∈ Rm , ζ ∈ Rs , x ∈ Rρ , z ∈ Rℓ
A, B, C are block diagonal matrices
0 1 ··· ··· 0 0
0 0 1 ··· 0 0
. . .
. . ..
Ai =
. .
, B i =
0 ··· ··· 0 1 0
0 · · · · · · · · · 0 ρ ×ρ 1 ρi ×1
i i
– p. 2/1
h i m
X
Ci = 1 0 ··· ··· 0 , ρ= ρi
1×ρi
i=1
Normal form
ẋ1 = x2
k L0 ax23
ẋ2 = g − x2 −
m 2m(a + x1 )2
1 L0 ax2 x3
ẋ3 = −Rx3 + 2
+u
L(x1 ) (a + x1 )
– p. 3/1
Stabilizing (partial) state feedback controller:
u = γ(x, ζ)
Ẋ = f (X ), X = (x, z, ϑ)
Observer:
– p. 4/1
H is block diagonal
αi1 /ε
αi2 /ε2
..
Hi =
.
i
αρi −1 /ερi −1
αiρi /ερi ρi ×1
– p. 5/1
Theorem 14.6 (Nonlinear Separation Principle:
Suppose the origin of Ẋ = f (X ) is asymptotically stable
and R is its region of attraction. Let S be any compact set
in the interior of R and Q be any compact subset of Rρ .
Then,
∃ ε∗1 > 0 such that, for every 0 < ε ≤ ε∗1 , the solutions
(X (t), x̂(t)) of the closed-loop system, starting in
S × Q, are bounded for all t ≥ 0
– p. 6/1
given any µ > 0, ∃ ε∗3 > 0, dependent on µ, such that,
for every 0 < ε ≤ ε∗3 , the solutions of the closed-loop
system, starting in S × Q, satisfy
– p. 7/1
Key ideas of the proof:
Representation of the closed-loop system as a
singularly perturbed one with X as the slow and η
(scaled estimation error) as the fast
– p. 8/1
Example 14.19:
or
a1 (θ − π) + ω̂
u = −k sat
µ
– p. 10/1
(a) (b)
3.5 2
3
1
2.5
ω
θ
2 0
1.5 SFB
OFB ε = 0.05 −1
1 OFB ε = 0.01
0.5 −2
0 2 4 6 8 10 0 2 4 6 8 10
(c) (d)
3.5 3.5
3 3
2.5
2.5
2
θ
θ
2
1.5
1.5
1
1 0.5
0.5 0
0 2 4 6 8 10 0 2 4 6 8 10
Time Time
– p. 11/1
Nonlinear Systems and Control
Lecture # 41
Integral Control
– p. 1/1
ẋ = f (x, u, w)
y = h(x, w)
ym = hm (x, w)
– p. 2/1
Assumption: e can be measured
0 = f (xss , uss , w)
0 = h(xss , w) − r
xδ = x − xss , uδ = u − uss ?
– p. 3/1
Integral Action:
σ̇ = e
Augmented System:
ẋ = f (x, u, w)
σ̇ = h(x, w) − r
r - l - R −σ
- Stabilizing u- Plant y-
+ Controller
−6
6 6
Measured
Signals
– p. 4/1
Integral Control via Linearization
State Feedback:
u = −K1 x − K2 σ − K3 e
Closed-loop system:
Equilibrium points:
0 = f (x̄, ū, w)
0 = h(x̄, w) − r
ū = −K1 x̄ − K2 σ̄
ξ̇δ = (A − BK)ξδ
" #
A 0 ∂f ∂h
A= , A= (x, u, w) , C =
(x, w)
C 0 ∂x eq ∂x eq
" #
B ∂f
B= , B= (x, u, w)
0 ∂u eq
h i
K = K1 + K3 C K2
– p. 6/1
(A, B) is controllable if and only if (A, B) is controllable
and " #
A B
rank =n+p
C 0
Task: Design K, independent of v , such that (A − BK) is
Hurwitz for all v
e(t) → 0 as t → ∞
– p. 7/1
Pendulum Example:
θ̈ = −a sin θ − bθ̇ + cT
Regulate θ to δ
x1 = θ − δ, x2 = θ̇, u=T
ẋ1 = x2
ẋ2 = −a sin(x1 + δ) − bx2 + cu
" #
0 a
xss = , uss = sin δ
0 c
σ̇ = x1
– p. 8/1
0 1 0 0
A = −a cos δ −b 0 , B = c
1 0 0 0
K1 = [k1 k2 ], K 2 = k3 , K3 = 0
(A − BK) will be Hurwitz if
a 1
Suppose ≤ ρ1 , ≤ ρ2
c c
k3
k2 > 0, k3 > 0, k1 > ρ1 + ρ2
k2
– p. 9/1
Output Feedback: We only measure e and ym
σ̇ = e = y − r
ż = F z + G1 σ + G2 ym
u = Lz + M1 σ + M2 ym + M3 e
– p. 10/1
Integral Control via Sliding Mode Design
η̇ = f0 (η, ξ, w)
ξ̇1 = ξ2
.. ..
. .
ξ̇ρ−1 = ξρ
ξ̇ρ = b(η, ξ, u, w) + a(η, ξ, w)u
y = ξ1
a(η, ξ, w) ≥ a0 > 0
Goal:
y(t) → r as t → ∞
ξss = [r, 0, . . . , 0]T
– p. 11/1
Steady-state condition: There is a unique pair (ηss , uss )
that satisfies the equations
0 = f0 (ηss , ξss , w)
0 = b(ηss , ξss , uss , w) + a(ηss , ξss , w)uss
ė0 = y − r
e1 ξ1 − r
e2 ξ2
z = η − ηss , e= .. =
..
. .
eρ ξρ
– p. 12/1
def
ż = f0 (η, ξ, w) = f˜0 (z, e, w, r)
ė0 = e1
ė1 = e2
.. ..
. .
ėρ−1 = eρ
ėρ = b(η, ξ, u, w) + a(η, ξ, w)u
s = k0 e0 + k1 e1 + · · · + kρ−1 eρ−1 + eρ
s
u = −β(e) sat
µ
̺(e)
β(e) ≥ + β0 , β0 > 0
(1 − κ0 )
For |s| ≥ µ, sṡ ≤ −a0 (1 − κ0 )β0
What about the other state variables?
– p. 14/1
ż = f˜0 (z, e, w, r)
ζ̇ = Aζ + Bs (A is Hurwitz)
s
ṡ = −a(·)β(e) sat + ∆(·)
µ
ζ = [e0 , . . . , eρ−1 ]T
– p. 15/1
Ω = {|s| ≤ c} ∩ {V2 ≤ c2 ρ1 } ∩ {V1 ≤ c0 }
– p. 16/1
Output Feedback: Only e1 is measured
High-gain Observer:
ė0 = e1
k0 e0 + k1 e1 + k2 ê2 + · · · + êρ
u = −β sat
µ
˙êi = êi+1 + αi (e1 − ê1 ),
1≤i≤ρ−1
εi
αρ
˙êρ = (e1 − ê1 )
ερ
– p. 17/1