ECSL
State-feedback control systems
University of Ulsan
Prof. KIM
c
°2023-2024 All rights reserved
ECSL
6 Sliding mode control
• Sliding mode control (SMC) is a nonlinear control method to
regulate the behavior of dynamic systems subject to disturbances or
uncertainties.
• The core idea behind SMC is to create a sliding surface in the state
space of the system, such that
“Once the system reaches this sliding surface, it ideally remains there
regardless of disturbances or uncertainties.”
ECSL
• In other words, this control law is designed such that it drives the
system toward the sliding surface and then maintains it there.
• One of prominent features is that this control involves the use of a
discontinuous control signal which switches based on the system’s
position relative to the sliding surface.
• Sliding mode control has applications in various fields including
aerospace, automotive, robotics, and power systems.
ECSL
Let us consider the following linear systems with uncertainties and
external disturbances:
ẋ(t) = (A + ∆A)x(t) + Bu(t) + N w(t).
1
(44)
z(t) = Cx(t) + N2 w(t)
Assumption: In general, the matrix B is of full column rank.
In this study, we employ the following switching function s(t):
Z t
s(t) = Gx(t) − G(A + BK)x(τ )dτ (45)
0
where G ∈ Rm×n is chosen so that GB ∈ Rm×m is invertible and K
denote the linear control gain to be designed later.
ECSL
From (45), it follows that
ṡ(t) = Gẋ(t) − G(A + BK)x(t). (46)
Additionally, according to (44), it is also available that
ṡ(t) = GBu(t) + G∆Ax(t) + GN1 w(t) − GBKx(t). (47)
ECSL
Theorem 5. The trajectories of system (44) can be driven onto the
surface s(t) = 0 if the control input is given as
u(t) = Kx(t) + (GB)−1 ψ(t) (48)
where
ψ(t) = −sgn(s(t)) · (α + ²(x(t))) (49)
²(x(t)) = ²A · ||G|| · ||x(t)|| + ²w · ||GN1 ||.
Remark 2. The decision variables K and G will be designed from
the stability condition of closed-loop system given that ṡ(t) = 0.
ECSL
Proof: Consider the Lyapunov function:
1 T
Vs (t) = s (t)s(t).
2
Then, we have
V̇s (t) = sT (t)ṡ(t)
³ ´
= sT (t) GBu(t) + G∆Ax(t) + GN1 w(t) − GBKx(t) .
Using (48), it is given that
³ ´
V̇s (t) = sT (t) ψ(t) + G∆Ax(t) + GN1 w(t)
≤ sT (t)ψ(t) + ||s(t)|| · ||G∆Ax(t)|| + ||s(t)|| · ||GN1 w(t)||.
ECSL
Noting that
||G∆Ax(t)|| + ||GN1 w(t)||
≤ ||G|| · ||∆A|| · ||x(t)|| + ||GN1 || · ||w(t)||
≤ ²A · ||G|| · ||x(t)|| + ²w · ||GN1 ||
=:²(x(t))
we have
V̇s (t) ≤ sT (t)ψ(t) + ||s(t)||²(x(t)).
Finally, using (49) implies
V̇s (t) ≤ −α||s(t)|| < 0, ∀s(t) 6= 0
which means that
lim s(t) → 0.
t→∞
ECSL
According to sliding mode control theory, when the sliding motion
takes place, it follows that s(t) = 0 and ṡ(t) = 0.
That is, from (47), the equivalent control be given as
−1
¡ ¢
ueq (t) = Kx(t) − (GB) G ∆Ax(t) + N1 w(t) .
Thus, the sliding motion is described as
ẋ(t) = Ax(t) + Bueq (t) + ∆Ax(t) + N1 w(t)
³ ´ ³ ´³ ´
= A + BK x(t) + I − B(GB)−1 G ∆Ax(t) + N1 w(t)
³ ´
= Ā + Ḡ∆A x(t) + ḠN1 w(t) (50)
where
Ā = A + BK, Ḡ = I − B(GB)−1 G.
ECSL
Theorem 6. System (50) is stable under the condition of s(t) = 0 if
there exits matrices P̄ = P̄ T ∈ Rn×n , K̄ ∈ Rm×n and scalars ² > 0,
γ > 0, such that
P̄ > 0
−I C P̄ N2 0
(∗) He{AP̄ + B K̄} + ²ḠEE ḠT T
ḠN1 P̄ H T
< 0.
(∗) (∗) −γ 2 I 0
0 (∗) 0 −²I
Furthermore, the control gain can be reconstructed as follows:
K = K̄ P̄ −1 .
ECSL
Proof: Let us choose
V (t) = xT (t)P x(t).
Then we have
T
x(t) x(t)
V̇ (t) + z (t)z(t) − γ w (t)w(t) =
T 2 T Ψ
w(t) w(t)
where
He{P Ā + P Ḡ∆A} P ḠN1 CT h i
Ψ= + C N2 .
(∗) 2
−γ I N2T
ECSL
By the Schur complement, the robust H∞ stability condition Ψ < 0
is transformed into
−I C N2
(∗) He{P Ā + P Ḡ∆A} P ḠN < 0. (51)
1
(∗) (∗) −γ 2 I
Furthermore, pre- and post-multiplying (51) by diag(I, P̄ := P −1 , I)
and its transpose yields
−I C P̄ N2
(∗) He{ĀP̄ + Ḡ∆AP̄ } ḠN < 0. (52)
1
(∗) (∗) −γ 2 I
ECSL
Noting that
He{Ḡ∆AP̄ } = He{ḠEΥH P̄ } ≤ ²ḠEE T ḠT + ²−1 P̄ H T H P̄
and by apply the Schur complement, we can see that the following
condition implies (52):
−I C P̄ N2 0
(∗) He{AP̄ + B K̄} + ²ḠEE T ḠT ḠN T
P̄ H
1
<0
(∗) (∗) −γ I2
0
0 (∗) 0 −²I
where K̄ = K P̄ .
