First Order Sliding Mode Control

Professor Aamer Iqbal Bhatti

Department of Electrical Engineering

Capital University of Science and Technology, Islamabad

April 26, 2019

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 1 / 24
 First Order Sliding Mode Sliding Mode Control

Outline

State-Feedback Sliding Surface Design

Regular Form
Eigenvalue Placement
Quadratic Minimization
Robustness of the LQR Sliding Surface Design
State-Feedback Unit-Vector Control

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 2 / 24
 First Order Sliding Mode Sliding Mode Control

State-Feedback Sliding Surface Design

Consider the nth-order linear time-invariant system with m inputs given by

ẋ(t) = Ax(t) + Bu(t) (1)

Where A ∈ Rnxn and B in Rnxm with 1 ≤ m ≤ n. Without loss of
generality it can be assumed that the input distribution matrix B has full
rank. Define a switching function σ : R −→ Rm to be

σ(x) = Sx(t) (2)

Where S ∈ Rnxn is of full rank and let S be the hyperplane defined by

S = {x ∈ Rn : Sx = 0} (3)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 3 / 24
 First Order Sliding Mode Sliding Mode Control

Sliding Surface Design

Figure: Evolution of shaft position

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 4 / 24
 First Order Sliding Mode Sliding Mode Control

Sliding Surface Design

Figure: Evolution of switching function

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 5 / 24
 First Order Sliding Mode Sliding Mode Control

Sliding Surface Design

Figure: Phase portrait

This implies the switching function σ(x) is a linear combination of the

states. Also from Eq. (1) it follows that the sliding motion is associated
with the null space of the matrix S. Also note that the number of rows of
the matrix S corresponds to the number of columns of the input
distribution matrix B and consequently the matrix SB is square

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 6 / 24
 First Order Sliding Mode Sliding Mode Control

Sliding Surface Design

Suppose u represents a sliding mode control law where the changes in

control strategy depend on the value of the switching function σ(x) It is
natural to explore the possibility of choosing the control action and
selecting the switching strategy so that an ideal sliding motion takes place
on the hyperplane, i.e., there exists a time tr such that

σ(x) = Sx(t) = 0∀t > tr (4)

S ẋ(t) = SAx(t) + SBu(t) = 0 (5)
Suppose the matrix S is designed so that the square matrix SB is
non-singular (in practice this is easily accomplished since B is full rank and
S is a free parameter)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 7 / 24
 First Order Sliding Mode Sliding Mode Control

Equivalent Control Design

The equivalent control, written as ueq , as argued above, is the unique

solution to the algebraic equation (5), namely
ueq (t) = −(SB)−1 SAx(t) (6)
ẋ(t) = (In − B(SB)−1 S)Ax(t)∀t ≥ tr and Sx(t) = 0 (7)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 8 / 24
 First Order Sliding Mode Sliding Mode Control

Equivalent Control Design with uncertainty

Now suppose the system is uncertain then we can write the equation
ẋ(t) = Ax(t) + Bu(t) + Bζ(t, x, u) (8)
n m
where ζ :R X R X R is unknown but bounded and encapsulates any
nonlinearities or uncertainties in the system. Uncertainty which acts in the
channel of the inputs is often referred to as matched uncertainty.
ueq (t) = −(SB)−1 SAx(t) − ζ(t, x, u) (9)
The closed-loop sliding motion is given by substituting Eq. (8) in Eq. (9)
and yields
ẋ(t) = (In − B(SB)−1 S)Ax(t) (10)
This motion is completely independent of the uncertainty. Although the
sliding motion is clearly dependent on the matrix S, how to select S to
achieve a specific design goal is not transparent.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 9 / 24
 First Order Sliding Mode Sliding Mode Control

Regular Form

.
A coordinate transformation is introduced to create a special
structure in the input distribution matrix.
By assumption rank(B)= m there exists an orthogonal matrix Tr ∈
Rnxn such that " #
0
Tr B = (11)
B2

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 10 / 24
 First Order Sliding Mode Sliding Mode Control

Eigenvalue Placement

 
0 1 0 0 ... 0
 0 1 ... 0 
A= (12)
 
 . . 0 0 ... 1 


−a1 −a2 −a3 −a4 ... −an

 
0
0
B=  (13)
 
.
1
where the scalars ai are the coefficients of the characteristic equation of
the A matrix:
λn + an λn−1 + ... + a2 λ + a1 = 0 (14)
Therefore the characteristic equation of the sliding motion is given by
λn + sn λn−1 + ... + s2 λ + s1 = 0 (15)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 11 / 24
 First Order Sliding Mode Sliding Mode Control

Quadratic Form

Consider the problem of minimizing the quadratic performance index

Z ∞
J= x(t)T Qx(t)dt (16)
ts
where Q is both symmetric and positive definite, and ts is the time at
which the sliding motion commences. The objective is to minimize Eq.(11)
subject to the system equation under the assumption that sliding takes
place. Notice this is quite different from the classical LQR problem
formulation which includes a penalty weighting on the control effort. Here
no penalty cost on the control is imposed, and this represents a so-called
cost-free control problem. It is assumed that the state of the system at
time ts, given by x(t)is a known initial condition, and the closed-loop
system is stable such that x(t) → ∞.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 12 / 24
 First Order Sliding Mode Sliding Mode Control

Robustness of the LQR Sliding Surface Design

An advantage of this approach compared to pole placement is that the

LQR optimization method inherits robustness.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 13 / 24
 First Order Sliding Mode Sliding Mode Control

State-Feedback Unit-Vector Control

ẋ(t) = Ax(t) + Bu(t) + f (t, x, u) (17)

n m m
where the function f : R X R X R → R , which represents the
uncertainties or nonlinearities, satisfies the so-called matching condition,
i.e.,
f (t, x, u) = Bζ(t, x, u) (18)
n m m
where ζ : R X R X R → R , Rm and is unknown but satisfies
kBζ(t, x, u)k ≤ k1 kuk + α(t, x) (19)
where 1 ≥ k1 0 is a known constant and α(.) is a known function.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 14 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Matched Uncertainty I

The proposed control law comprises two components: a linear component

to stabilize the nominal linear system and a discontinuous component.
Specifically

u(t) = ul (t) + un (t) (20)

where the linear component is given by

ul (t) = −Λ−1 (SA − ΦS)x(t) (21)

where Φ ∈ Rm is any stable design matrix and Λ = SB. The nonlinear
component is defined to be

P 2σ(t)
un (t) = −ρ(t, x)Λ−1 (22)
kP 2σ(t)k

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 15 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Matched Uncertainty II

where P2 ∈ Rmxm is a symmetric positive definite matrix satisfying the
Lyapunov equation

P2 Φ + ΦT P2 = −I (23)
and the scalar function ρ(t, x), which depends only on the magnitude of
the uncertainty, is any function satisfying

kΛk(k1 kul k + α(t, x)) + γ

ρ(t, x) ≥ (24)
1 − k1 kΛkkΛk−1
where γ ≥ 0 is a design parameter. In this equation it is assumed that the
scaling parameter S2 has been chosen so that Λ = SB has the property
that

1 − k1 kΛkkΛk−1 ≤ 1 (25)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 16 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Matched Uncertainty III

A necessary condition for the above equation to hold is that 1 − k1 ≤ 1
because kΛkkΛk−1 ≥ 1 for all choices of S.
Before demonstrating that the above controller induces a sliding motion, it
will first be established that any scalar modulation function satisfying
equation 24 bounds the uncertain term ζ(t, x, u). On rearranging the
equation 24 gives

ρ(t, x) ≥ kΛk(k1 kul k + α(t, x)) + γ + k1 kΛkkΛk−1 ρ(t, x) (26)

−1
ρ(t, x) ≥ kΛk(k1 kΛk ρ(t, x) + k(k1 kkul k + α(t, x)) + γ (27)
ρ(t, x) ≥ kΛk(k1 kkul k + α(t, x)) + γ (28)
ρ(t, x) ≥ kΛkζ(t, x, u) + γ (29)
In obtaining the third inequality the fact that
u = ul − ρ(t, x)kΛk−1 kPP22 σk
σ
⇒ kuk ≤ kul k + ρ(t, x)kΛk−1

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 17 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Matched Uncertainty IV

Inequality(29) demonstrates ρ(t, x) is greater in magnitude than the
matched uncertainty. The sliding surface dynamics can be written as :
σ̇ = SAx(t) + Λu + Λζ(t, x, u) (30)

P2 σ
σ̇ = Φσ − ρ(t, x) + Λζ(t, x, u) (31)
kP2 σk
It can be shown that V (σ) = σ T P2 σ guarantees quadratic stability for the
switching states σ, and in particular

V̇ = σ T (P2 Φ + ΦT P2 )σ − 2ρkP2 σk + 2ρT P2 Λζ (32)

V̇ ≤ σ T (P2 Φ + ΦT P2 )σ − 2ρkP2 σk + 2kP2 ρT kkΛkkζk (33)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 18 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Matched Uncertainty V

V̇ ≤ −σ T σ − 2kP2 σk(ρ(t, x) − kΛkkζk) (34)

V̇ ≤ −σ T σ − 2γkP2 σk (35)
Assuming that the closed-loop system has no finite-escape time during the
reaching phase, then this control law guarantees that the switching surface
is reached in finite time despite the disturbance or uncertainty. Once the
sliding motion is attained, it is completely independent of the uncertainty

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 19 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Unmatched Uncertainty I

If the uncertainty does not meet the matching requirements, after

transformation into regular form, in the âĂIJzâĂİ coordinates, a system of
the form

Z˙1 (t) = A11 Z1 (t) + A12 Z2 (t) + fu (t, Z1 , Z2 ) (36)

Z˙2 (t) = A21 Z1 (t) + A12 Z2 (t) + B2 u(t) + fm (t, Z1 , Z2 ) (37)

Where fm (t, z1, z2) and fu (t, z1, z2) represent the matched and
unmatched components of the uncertainty, respectively. As argued in the
earlier sections, the effects of the matched uncertainty fm.t; z1; z2/ can be
canceled. This section considers the null-space dynamics. If a sliding
motion can be induced on S, then Z2 = −M Z1 and the reduced-order
motion is governed by

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 20 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Unmatched Uncertainty II

Z˙1 (t) = (A11 − A12 )Z1 (t) + +fu (t, Z1 , Z2 ) (38)

Because of the presence of the term fu (t, Z1 , −M Z2 ) stability of the
system in Eq. (38) is not guaranteed. However, if the linear component is
dominant, then bounds on fu (t, Z1 , −M Z2 ) can be obtained to guarantee
that stability is maintained. Many different approaches and assumptions
can be made: here a Lyapunov approach will be adopted. Specifically it
will be assumed that

kfu (t, Z1 , Z2 )k ≤ µkZk (39)

where µ is a positive scalar. Since, by choice of the sliding surface, the
matrix (A11 − A12 M is stable, there exists a symmetric positive definite
matrix P1 such that

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 21 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Unmatched Uncertainty III

P1 (A11 − A12 M ) + (A11 − A12 M )T P1 = −In−m (40)

It can be shown that if q 1
µ 1 + kM k2 (41)
2kP 1k
then Eq. (38) is stable while sliding. To establish this, first the constraint
in Eq. (39) will be written in terms of kZ1 k. Since during the sliding
motion Z2 =−M Z2 it follows
 
Z 
kZ1 k = k  1  k2 ≤ kZ1 k2 + kZ2 k2 ≤ kZ1 k2 + kM Z1 k2 ≤ (kM k2 + 1)Z12
2  
Z2 
(42)

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 22 / 24
 First Order Sliding Mode Sliding Mode Control

Design in the Presence of Unmatched Uncertainty IV

p
and consequently kZkk 1 + kM k2 kkZ1 k Create from the symmetric
positive definite matrix P1 a candidate Lyapunov function
V (Z1 ) = Z1T P1 Z11 .
It follows
V̇ =Z1T P1 Z˙1 + Z˙1T P1 Z1
V̇ =Z1T (P1 (A11 −A12 M )+(A11 −A12 M )T P1 )Z1 +2Z1T P1 fu (t, Z1 , −M Z2 )
≤ −Z1T Z1 + 2Z1 kZ1 kkP1 kkfu (t, Z1 , −M Z2 )k
≤ −Z1T Z1 + 2Z1 kZ1 kkP1 kµkZ1 )k
q
≤ kZ1 k2 + 2(µkP1 k( 1 + kM k2 − 1) (43)
If the inequality in Eq.(41) holds then
q
V̇ ≤ kz1 k2 + 2µkP1 k( 1 + kM k2 − 1) < 0 (44)
and so the reduced-order motion is stable.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 23 / 24
 First Order Sliding Mode Sliding Mode Control

For Further Reading I

Utkin, V., Guldner J., Shi J.

Sliding Mode Control in Electro-Mechanical Systems, 2nd Edition.
CRC Press, 2009.
Shtessel, Y., Edwards, C., Fridman, L., Levant, A.
Sliding Mode Control and Observation.
Springer, 2014.

c Dr. A. I. Bhatti, CASPR SMC Workshop April 26, 2019 24 / 24