Sliding Mode Control: An Introduction

Dr. Shyam Kamal

Kyushu Institute of Technology, Japan

kamal@ces.kyutech.ac.jp

February 23, 2015

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Outline

1 The Sliding Mode Control

2 Filippov Theory

3 Physical Interpretation of the Equivalent Control

4 Invariance in Sliding-modes

5 Sliding Mode Control Design Using Regular Form Transformation

Dr. Shyam Kamal Sliding Mode Control: An Introduction 2

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

What is the Sliding mode and how did its study start?

Figure : Primitive Examples - Electrical Figure : Primitive Examples-Mechanical

Dr. Shyam Kamal Sliding Mode Control: An Introduction 3

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

Variable Structure Control: Prof.Emelyanov

The first steps of sliding mode control theory originated in the early 1950 initiated
by S. V. Emelyanov.
Started as Variable Structure Control (VSC):-Varying system structure for
stabilization.

Figure : Mode I
x = 1 x Figure : Mode II x = 2 x, 0 < 2 < 1

Dr. Shyam Kamal Sliding Mode Control: An Introduction 4

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

Piecing together

Properties of VSC
Both constituent systems were oscillatory and were not asymptotically stable.
Combined system is asymptotically stable.
Property not present in any of the constituent system is obtained by VSC.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 5

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

Unstable Constituent Systems

Figure : x x x = 0 Figure : x x + x = 0

Dr. Shyam Kamal Sliding Mode Control: An Introduction 6

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

Analysis
Both systems are unstable
Only stable mode is one mode of system
s
2
x x x = 0, = +
2 4

If the following VSC is employed

(
I if xs 0
Mode = , s = cx + x, c =
II if xs > 0.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 7

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

Unstable Constituent Systems: More Analysis

Again, property not present in constituent systems is found in the combined
system.
A stable structure can be obtain by varying between two unstable structures.
However, a more interesting behavior can be observed if we use a different
switching logic.
(
I if xs 0
Mode = , s = cx + x, 0 < c (c = )
II if xs > 0.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 8

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

The Sliding Mode Control: Revisited

New trajectory that was not present in any of the two original systems

Sliding Mode:-Motion of the system trajectory along a chosen line/plane/surface of the

state space.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 9

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design

The Linear System

x = Ax + B(u + (t, x))

x Rn, u R

The sliding surface is designed as (x) = Cx = 0.

The sliding mode control is designed to reach the sliding surface in finite time.

Assumptions
The pair (A, B), is controllable.
The scalar CB 6= 0.
The disturbance/uncertainties are matched and bounded, i.e.,|(t, x)| DM .
The zero dynamics(reduced order dynamics) of the system with the output (x) is
stable.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 10

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design

The Controller Design

Since the function (x) is to be made zero, it helps to look at its derivative.

= C x (2.1)
= CAx + CB(u + (t, x)) (2.2)

Now, let us design the control as,

u = (CB)1 (CAx + ())

Substituting this control in (2.2) gives,

= () + CB(t, x) (2.3)

The function () must be selected such that, the equation (2.3) provides finite
time convergence of to zero, in presence of (t, x).
Once the sliding surface is reached, i.e., (x) 0, the system trajectory is
governed by the zero dynamics, also called reduced order dynamics.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 11

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design

Examples of ()
1 () = K sgn()
2 () = K sgn()
3 () = K || sgn()
However, the third choice can not reject the disturbances if (t, x) 6= 0 at x = 0.

The Chattering Problem

The control is discontinuous, resulting in high frequency switching by the actuator.

The Relative Degree Problem

= CAx + CB(u + (t, x)) (2.4)

If CB = 0, then control does not appear in the expression for .
Usually, C is designed surface matrix so CB = 0 can be avoided.
However, may be some output of the system and C is determined by system
dynamics. The objective is to make the output zero robustly, then with CB = 0, the
sliding mode control design is not possible.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 12

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

A Brief Review of Filippov Theory

Convexification

Figure : Filippov Solution

Physical Interpretation of Filippov Method Using Regularization Approach

Consider the system of the following form
n
x = f (x, u), x, f R , u(x) R (3.1)
(
u+ if s(x) > 0
u(x) = (3.2)
u if s(x) < 0

where the component of vector f , scalar functions u + , u and s(x) are continuous and smooth,
and u + (x) 6= u (x).

Dr. Shyam Kamal Sliding Mode Control: An Introduction 13

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

A Brief Review of Filippov Theory

Physical Interpretation of Filippov Method Using Regularization Approach

We assume that sliding mode occurs on the sliding surface s(x) = 0.
Try to derive the motion equations using the regularization method.
Let the discontinuous control be implemented with some imperfections of
unexpected nature, control is assumed to take one of the two extreme values, u +
or u , and the discontinuity points are isolated in time.
Since discontinuity points are isolated in time, the solution exists in the
conventional sense beyond the sliding surface.
Further assume that the state velocity vector f + = f1 = f (x, u + ) and
f = f2 = f (x, u ) to be constant for some point x on the surface s(x) = 0 within
a short interval [t, t + t].
Let the time interval t consists of two sets of intervals t1 and t2 such that
t = t1 + t2 , where t1 and t2 is the amount of that time when control of
magnitude u + and u is active.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 14

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Physical Interpretation of Filippov Method Using Regularization

Approach

Mathematically increment of the state vector in this interval t is given by

x = f1 t1 + f2 t2 (3.3)

and the average state velocity is given as

x f1 t1 + f2 t2 t1 t2
xaverage = = = f1 + f2
t t1 + t2 t1 + t2 t1 + t2
= f1 + (1 )f2 (3.4)

t
where = t1 is relative time for control to take value u + and (1 ) to take value u
and also 0 < 1.

To get the velocity vector x along the sliding surface we have to take limit t 0.
Hence sliding motion is represented as

x = f1 + (1 )f2 (3.5)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 15

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Physical Interpretation of Filippov Method Using Regularization

Approach

Remark:-One can also interpret Eqn.(3.5) as the velocity vector in the vicinity of a point
on a discontinuous surface which is complemented by a minimal convex set, and the
state velocity vector of the sliding motion belongs to this set.

Because the state trajectories during sliding mode are in the sliding surface s = 0, the
parameter should be selected such that the state velocity vector of the system (3.5)
is in the tangential plane to the sliding surface.
Mathematically one can write

s = [s(x)].x = [s(x)](f1 + (1 )f2 ) = 0 (3.6)

h i
s s
where [s(x)] = x1
xn
. The solution of above equation is given by

(s).f2
= (3.7)
(s).(f2 f1 )

Substituting the from Eqn.(3.8) to (3.5), one can get motion in sliding mode as

(s).f2 (s).f1
x = fsliding = f1 f2 (3.8)
(s).(f2 f1 ) (s).(f2 f1 )
Dr. Shyam Kamal Sliding Mode Control: An Introduction 16
 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Equivalent Control Method

Remark:-
Sliding mode occurs in the surface s(x) = 0, therefore, the function s and s have
different signs in the vicinity of the surface and s+ = (s).f1 < 0, s = (s).f2 > 0.
Also one can easily check that s = (s)fsliding = 0 for the trajectories of system (3.8)
and show that they are confined to the switching surface s(x) = 0.

Equivalent Control Method by the Geometrical Point of View

In sliding mode control, our main aim is to design a control law so that the state
trajectories are confined to a sliding manifold in finite time.
From a geometrical point of view, the equivalent control method does the same
job. It replace the discontinuous control on the intersection of the switching
surface by a continuous one such that, the state velocity vector lies in the
tangential manifold.

Mathematically, consider the system

x = f (x) + B(x)u x, f (x) R n , B(x) R, u R n (3.9)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 17

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Equivalent Control Method

u(x) is defined as

(
u+ if s(x) > 0
u(x) = (3.10)
u if s(x) < 0

So,
s
s = x = G(x)f (x) + G(x)B(x)uequivalent = 0 (3.11)
x
s
where G = x . Assuming the matrix GB is nonsingular for any x, find the equivalent
control uequivalent as the solution of the Eqn.(3.11)

uequivalent = G(x)B(x)1 G(x)f (x) (3.12)

and substituting uequivalent into (3.9) to yield the sliding mode equation s = 0 as

x = f (x) G(x)B(x)1 B(x)G(x)f (x) (3.13)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 18

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Equivalent Control Method

Physical Interpretation of the Equivalent Control

For the occurrence of the ideal sliding mode it was assumed that the control
changes at high (theoretically infinite) frequency such that the state vector is
oriented precisely along the intersection of discontinuity surfaces.
In reality however, various imperfections make the state oscillate in some vicinity
of the intersection and control components are switched at finite frequency
alternatively taking the positive and negative values.
These oscillations have high frequency as well as slow components.
All most all plants under control act as a low pass filter.
Due to this low pass filter characteristic high frequency component is filtered out,
and its motion in sliding mode is determined by the slow component.
Practically it is reasonable to assume that the equivalent control is close to the
slow component of the real control, which can be derived by filtering out the
high-frequency components using a low-pass filter.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 19

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Physical Interpretation of the Equivalent Control

Mathematically, the output of a low-pass filter

z + z = u (4.1)

tends to the equivalent control

lim z = uequivalent (4.2)

0,

0

where , are the time constant of low pass filter and width of the manifold
respectively.

Remark:-

To eliminate the high-frequency component of the control in sliding mode, the

frequency should be much higher than 1 , or 1f << , hence, << .
Finally, the time constant of the low-pass filter should be made to tend to zero
because the filter should not distort the slow component of the control.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 20

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Invariance in Sliding-modes

B. Drazenovic The invariance conditions in variable structure systems Automatica,

v.5, No.3, Pergamon Press, 1969.

Figure : Prof.B. Drazenovic

Mathematical Treatment of Matched Uncertainty

Mathematical meaning of equivalent control is that, when the trajectories are on the
sliding surface then and , both must be zero. But this assumption is valid only when
the control input appears linearly for a given state. Now consider the following system

x = f (x) + g(x)u + d (5.1)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 21

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Invariance in Sliding-modes

Consequently, one can write

= x = f (x) + g(x)u + d (5.2)
x x x x

For calculating equivalent value of control substitute = 0 in Eqn.(5.2), the only

condition is that
x
g(x) 6= 0, we can write

 1  

uequivalent = g(x) f (x) + d (5.3)
x x x

The closed loop dynamics during sliding is obtained by substituting (5.3) into (5.1), one
can write

 1  

x = f (x) g(x) g(x) f (x) + d +d
x x x
 1 !  1 !

= I g(x) g(x) f (x) + I g(x) g(x) d (5.4)
x x x x

Dr. Shyam Kamal Sliding Mode Control: An Introduction 22

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Invariance in Sliding-modes

Now suppose that disturbance d is entering through control channel. Therefore one
can write d = g(x)(t), where (t) is unknown signal, but with a known bound. This
kind of disturbances are known as matched disturbances.
After substituting the value of d, into (5.4), one can write
 1 !  1 !

x = I g(x) g(x) f (x) + I g(x) g(x) g(x)(t)
x x x x
 1 !

= I g(x) g(x) f (x) + g(x)(t) g(x)(t)
x x
 1 !

= I g(x) g(x) f (x) (5.5)
x x

Hence the closed loop dynamics which is given in (5.5), is completely independent of
any matched disturbances, when the system is in the sliding mode. This property is
known as the invariance property with respect to disturbance.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 23

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Consider the linear time invariant system of the following form

x(t) = (A + A(t))x(t) + Bu + Df (t) (6.1)

where x R n1 ,
u A R m1 ,and B R nn R nm
are the state vector, control
input, system matrix and input matrix respectively. A(t) and f (t) represent parameter
variations and disturbance vector, where D is the disturbance input matrix.

Assumptions:-
The system is assumed to be controllable i.e the controllability matrix
[B AB A2 B An1 B] has full rank.
A range(B) and D range(B), or there exist constant or time varying
matrices 1 and 2 , such that

A = B1 , D = B2 (6.2)

Using Eqn.(6.1) and assumption (6.2), one can write

x(t) = Ax(t) + B(u + 1 x(t) + 2 f (t)) = Ax(t) + B(u + 0 ), 0 = 1 x(t) + 2 f (t)

| {z }
(6.3)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 24

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Regular Form Transformation:

Input matrix B in Eqn.(6.1) may be partitioned (after reordering the state vector
components)as  
B1
B= (6.4)
B2

where B1 R (nm)m , B2 R mm with detB2 6= 0.

The nonsingular coordinate transformation T , which is able to convert system (6.1) into
regular form is given by
" #
Inm B1 B21
T = (6.5)
0 B21

and change of coordinate is

 
z1
Tx = (6.6)
z2

Dr. Shyam Kamal Sliding Mode Control: An Introduction 25

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Transformed system (6.1), after applying above change of coordinate

z1 (t) = A11 z1 (t) + A12 z2 (t)

z2 (t) = A21 z1 (t) + A22 z2 (t) + u + 0 (6.7)

where z1 R nm , z2 R m and
   
A11 A12 0
TAT 1 = , TB = (6.8)
A21 A22 I

Design of Sliding Surface:-

Now our next aim is to design a specified switching manifold such that after reaching
the manifold, the system is free from its dynamics. The designed manifold can be linear
or nonlinear, depends on our specified goal. The simplest manifold is a linear one. So,
in this chapter we focus towards linear manifolds only, unless otherwise specified.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 26

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

The linear switching manifold (sliding surface) is given as follow

S = C1 z1 + z2 (6.9)
where S R m1 and C1 R m(nm) > 0.

Now taking the first time derivative of the sliding surface (6.9) and substituting the value
from the Eqn.(6.7), one can get

S = C1 z1 + z2
= (C1 A11 + A21 )z1 (t) + (C1 A12 + A22 )z2 (t) + u + 0 (6.10)

Now expressing the system (6.7) into z1 and S coordinate, using the differential
Eqn.(6.9) and (6.10)

z1 (t) = A11 z1 (t) + A12 (S C1 z1 (t))

S = (C1 A11 + A21 )z1 (t) + (C1 A12 + A22 )(S C1 z1 (t)) + u + 0
= (C1 A11 + A21 C1 A12 C1 A22 C1 )z1 (t) + (C1 A12 + A22 )S + u + 0 (6.11)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 27

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Existence Condition of Sliding Mode:-The main objective here is to design a control

u such that, the sliding motion occurs in finite time. For this, the control is selected
according to the following theorem

Theorem:-
The control input u which is defined as

u = (C1 A11 + A21 C1 A12 C1 A22 C1 )z1 (t) (C1 A12 + A22 )S (6.12)

where,
= K1 S K2 sign(S)) (6.13)
with K1 > 0, K2 > k0 k leads to the establishment S equal to zero in finite time.

Proof:-
Substituting u from Eqn.(6.12) to (6.11), one can write

z1 (t) = A11 z1 (t) + A12 (S C1 z1 (t))

S = K1 S K2 sign(S)) + 0 (6.14)

Dr. Shyam Kamal Sliding Mode Control: An Introduction 28

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Choosing the Lyapunov function as V = 21 S T S, and calculating the time derivative of

the Lyapunov function along (6.14), one can get

V = S T S = S T (K1 S K2 sign(S)) + 0 )
= K1 V K2 kSk + S T 0 (6.15)

1 1
Using Cauchy-Schwartz inequality S T 0 kSkkk and (2V ) 2 = (S T S) 2 = kSk, we
get

V K1 V (K2 kk)kSk
1 1
K1 V 2 2 (K2 kk)V 2
1
K1 V V 2 0 (6.16)

Therefore, one can ensure the asymptotic stability of the trajectory to the sliding
1
manifold where = 2 2 (K2 kk).

Dr. Shyam Kamal Sliding Mode Control: An Introduction 29

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Now, we have to prove that convergence takes place in finite time.

Time calculation:-
dV 1
V = k1 V V 2
dt
Z 0 Z 0
dV dV
T 1
= 1
 1

k1 V V
V (0) V (0) V
2 2 k1 V 2 +
2   1
 
= ln k1 V (0) 2 + ln() (6.17)
k1

Therefore, the above claim is justified since the time T is always finite.

Equivalent Control
It is clear from Eqn.(6.12), during the sliding mode S = 0 and average value of = 0,
therefore the remaining control is only
uequivalent = (C1 A11 + A21 C1 A12 C1 A22 C1 )z1 (t), which is theoretically
interpreted as equivalent control, which is the control required to maintain a sliding
motion on the manifold S.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 30

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

Analysis of Reduced Order Dynamics

Therefore, after time T , dynamics of S has been collapsed and the reduced order
dynamics of system (6.14) after substituting S = 0, is given as

z1 (t) = (A11 A12 C1 )z1 (t) (6.18)

The stability and performance of the above reduced order system depends on
(A11 , A12 ). Therefore, the design of C1 depends on the controllability of the pair
(A11 , A12 ).

Lemma
The matrix pair (A11 , A12 ) is controllable, if and only if the pair (A, B) is controllable.

Proof:-

 
sI A11 A12 0
rank[sI A B] = rank
A21 sI A22 I
 
= rank sI A11 A12 + m for all s C

Dr. Shyam Kamal Sliding Mode Control: An Introduction 31

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Sliding Mode Control Design Using Regular Form Transformation

This implies

rank[sI A B] = n rank [sI A11 A12 ] = n m

and from the Popov-Belevitch-Hautus (PBH) rank test, it follows that (A, B) is
controllable if and only if the pair (A11 , A12 ) is controllable.2

Now one can design C1 using any robust linear state feedback method, such as
quadratic minimization, direct or robust eigenvalue assignment, using LMI etc.

Dr. Shyam Kamal Sliding Mode Control: An Introduction 32

 The Sliding Mode Control Filippov Theory Physical Interpretation of the Equivalent Control Invariance in Sliding-modes Sliding Mode Control Design Usin

Thank You!

Dr. Shyam Kamal Sliding Mode Control: An Introduction 33