Variable Structure Control (VSC)

SUMMARY

Addressed the design of a variable structure controller for

the output regulation of a linear system

Evaluation of the performance of the controller on a

numerical example, including robustness with respect to
parametric uncertainty
SUMMARY
 SUMMARY
System S:

Key ingredients in the design of the controller:

the switching function = with = 1 / and
= 1 2 1 1
 SUMMARY
System S:

Key ingredients in the design of the controller:

the switching function = with = 1 / and
= 1 2 1 1 , which defines:
the sliding surface = 0
the characteristic polynomial of the (n-1)-dimensional
system S* governing the state evolution over the sliding
surface:
= 1 + 1 1 + + 1
the value for the system output at the equilibrium over
the sliding surface: =
 SUMMARY
System S:

Key ingredients in the design of the controller:

the switching function = with = 1 / and
= 1 2 1 1
the control input
= + + r g
with > 0, 0, > 0, = , = A, where
A is the matrix of S in controllable form
 SUMMARY
System S:

Key ingredients in the design of the controller:

the switching function = with = 1 / and
= 1 2 1 1
the control input
= + + r g
with > 0, 0, > 0, = , = A, where
A is the matrix of S in controllable form, which makes
system S reach the sliding surface in finite time
0 /
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

Suppose that S is subject to some load disturbance ()

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

Suppose that S is subject to some load disturbance ()

Statement:
If () < , , then:
the state will reach the sliding surface = 0 in finite time
and will evolve according to the dynamics of S*.
Only the time to convergence is affected by
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

Sub-system

has input + and output

=
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

=
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

=
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

=
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

System :
- externally + behaves like an integrator

- has hidden dynamics with characteristic polynomial

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

If = 0

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

If = 0

The (pseudo)-equilibrium = is reached in finite time.

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

If = 0

The sliding surface = = = 0 is reached in finite time.

If () < , , then, the sign of the derivative of is preserved,

so that = is still reached in finite time.
The time to reach = will however depend on .
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

More formally
1 2
Consider = , where = . Then,
2

= = = +

since
> 0
and
w (< ) implies that
we get

= = = + ( )||

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

More formally
1 2
Consider = , where = . Then,
2

||

where (> 0) .
 ROBUSTNESS W.R.T. LOAD DISTURBANCE

More formally
1 2
Consider = , where = . Then,
2

||

where (> 0) . Hence,

1
() 0 t
0 2 0 2 2
which leads to
1 1
(()) t + ((0))
2 2 2

and the time required to reach s = 0 is upper bounded by

((0))

 ROBUSTNESS W.R.T. LOAD DISTURBANCE

When = = = 0, then,
the system evolves on the sliding surface according to the
dynamics of S*;
the output tends to
STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

400( + 5)
= 2
( +5 + 20)( + 10)( 1)
400( + 5)
= 4
+ 14 3 + 55 2 + 130 200
STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

400( + 5)
= 4
+ 14 3 + 55 2 + 130 200
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

Recall that the coefficients of the switching function [n=4]

appear in the characteristic polynomial of the linear system S*

of order 3 governing the dynamics of S when restricted to the
sliding surface () = 0:
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

Recall that the coefficients of the switching function [n=4]

appear in the characteristic polynomial of the linear system S*

of order 3 governing the dynamics of S when restricted to the
sliding surface () = 0:

Set the eigenvalues of S* equal to those of S that are stable:

then,

So that
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

400( + 5)
= 4
+ 14 3 + 55 2 + 130 200

Since the state is not directly measurable, we resort to the

asymptotic state observer (Luenberger observer) and use in
place of in the sliding mode control law:

u = + +
 -s

u = + +
ASYMPTOTIC OBSERVER

+
+

+
+
ASYMPTOTIC OBSERVER

+
+

+ +
+
ASYMPTOTIC OBSERVER

observer gain

+ +
+
DYNAMICS OF THE STATE ESTIMATION ERROR
DYNAMICS OF THE STATE ESTIMATION ERROR
 ASYMPTOTIC OBSERVER

If (A,C) is observable, then, L can be designed so that

A-LC has arbitrarily chosen eigenvalues and
the estimation error converges exponentially to zero with
rate 0 2 (0, mini |Re{i(A-LC)}|)

Im
eigenvalues of A-LC
o

o o
Re

o
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE

If we choose the eigenvalues -10, -8, -6, -5 for the observer

dynamics, we get:
= [0.0025 0.0250 0.0175 0.2550]

and 0<0 <5

 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE
y = 5 2 , = 1, = 0 [/2 = 0.02, /2 = 1]
0 = 0 = [0 0 0 0], 0 = 0 = 0

y u
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE
y = 5 2 , = 1, = 0 [/2 = 0.02, /2 = 1]
0 = 0 = [0 0 0 0], 0 = 0 = 0

y u

Same behavior as without the observer since the estimation error

is zero at time 0 and then keeps being zero.
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE
y = 5 2 , = 1, = 0 [/2 = 0.02, /2 = 1]
0 = 0 0 0 0 , 0 = 0; 0 = 0.012 0 0 0 , 0 = 24

y u
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE
y = 5 2 , = 1, = 0 [/2 = 0.02, /2 = 1]
0 = 0 0 0 0 , 0 = 0; 0 = 0.012 0 0 0 , 0 = 24

y u

estimation error goes to 0

s
in 1 time unit ( = 5).

Then, same behavior as

when x(0) known
 STATE NOT AVAILABLE: A NUMERICAL EXAMPLE
y = 5 2 , = 1, = 0 [/2 = 0.02, /2 = 1]
0 = 0 0 0 0 , 0 = 0; 0 = 0.012 0 0 0 , 0 = 24

y u

y u
ISSUE OF THE HIGH FREQUENCY INPUT SWITCHING

The high frequency switching of the control input in the sliding

mode phase can be undesirable and even unacceptable.
ISSUE OF THE HIGH FREQUENCY INPUT SWITCHING

Possible solution:
filtering the high frequency components of the control input by
introducing an auxiliary control variable whose integral is the
actual control variable
 A NUMERICAL EXAMPLE

controllable and observable system (no pole-zero cancellations)

A NUMERICAL EXAMPLE

enlarged controlled system

A NUMERICAL EXAMPLE

enlarged controlled system

not in controllable form

A NUMERICAL EXAMPLE
 A NUMERICAL EXAMPLE

Dynamics within the sliding surface:

In the original state variables:

 A NUMERICAL EXAMPLE
y = 5 1 , = 10 = 0[/2 = 0.02, /2 = 1]
 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2

Goal: design a feedback controller that globally asymptotically

stabilizes the origin
 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2

Goal: design a feedback controller that globally asymptotically

stabilizes the origin

Difficult to stabilize even locally. Indeed, linearization in the

vicinity of the origin gives a non-controllable system:
1 =
2 =
3 = 0
If 1 and 2 are too close to zero, 3 cannot be steered to zero
 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2

Goal: design a feedback controller that globally asymptotically

stabilizes the origin

We adopt a sliding mode approach

 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2

Idea: make 3 converge faster to zero than 1 and 2

Sliding mode control:
= 1 + 2 3
= 2 1 3
[sliding surface 3 =0]
 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2

Idea: make 3 converge faster to zero than 1 and 2

Sliding mode control:
= 1 + 2 3
= 2 1 3
[sliding surface 3 =0]

We first show that there exists a set of initial conditions such

that trajectories starting there converge to the origin.
 STABILIZATION OF DOUBLE INTEGRATOR

System S:
1 =
2 =
3 = 1 2
Sliding mode control:
= 1 + 2 3
= 2 1 3

Control system:
1 = 1 + 2 3
2 = 2 1 3
3 = (1 2 + 1 2 ) 3
 STABILIZATION OF DOUBLE INTEGRATOR

Control system:
1 = 1 + 2 3
2 = 2 1 3
3 = (1 2 + 1 2 ) 3

Lyapunov function for the (1 , 2 ) space

1
1 , 2 = (1 2 + 2 2 )
2

= 1 1 + 2 2 = 1 2 + 2 2 = 2V

= 0 2 , 0

1 , 2 tend to zero asymptotically

 STABILIZATION OF DOUBLE INTEGRATOR

Control system:
1 = 1 + 2 3
2 = 2 1 3
3 = (1 2 + 1 2 ) 3

Lyapunov function for the (1 , 2 ) space

1
1 , 2 = (1 2 + 2 2 )
2
1 , 2 tend to zero asymptotically

As for 3 :
3 = (1 2 + 1 2 ) 3 = 2 3
 STABILIZATION OF DOUBLE INTEGRATOR

As for 3 :
3 = (1 2 + 1 2 ) 3 = 2 3
3
= 3 3 = 2


3 () 3 0 = 2
0

3 () = 3 0 2
0

If 3 0 <2 0
, then 3 tends to zero in finite time

If 3 0 =2 0
, then 3 tends to zero in infinite time
 STABILIZATION OF DOUBLE INTEGRATOR

= 0 2 , 0
1
Then, 2 0
= 0 = (1 2 (0) + 2 2 (0))
2

Hence: 3 0 2 0

becomes
1
3 0 (1 2 (0) + 2 2 (0))
2
 STABILIZATION OF DOUBLE INTEGRATOR

If initial state satisfies

1
3 0 1 2 0 + 2 2 0 ()
2
then the sliding mode controller leads the state to the origin
If the initial state doesnt satisfy that condition, then, apply
constant control to drive it in that region:
1 =
2 =
3 = 1 2

1 = + 1 0
2 = + 2 (0)
3 = 1 0 2 0 + 3 0
Then equation () will be satisfied at some finite time instant t.
 STABILIZATION OF DOUBLE INTEGRATOR

If initial state satisfies

1
3 0 (1 2 (0) + 2 2 (0))
2
then the sliding mode controller leads the state to the origin

If the initial state satisfies

1
3 0 > (1 2 (0) + 2 2 (0))
2
then, apply constant control to drive it in the complementary
region

Hybrid control scheme