Вы находитесь на странице: 1из 9

- 1 -

1 Sliding Mode Control



Consider the problem of doing set point control for a system of the form

( )
( ) ( )
n
x f x b x u = +
Where u x, . Further, we seek a solution that is robust to uncertainties in ( ) f x and
( ) b x .

Note that the system can be re-written in vector form. It is then equivalent to

( ) ( )
2 1
1 n n
n
x x
x x
f x b x u x

| | | |
| |
| |
=
| |
| |
|
+
\ . \ .
`
. .
`
`
, with
1
1 n
n
x
x
x
x

| |
|
|
=
|
|
\ .
`
.
`
`


The system is a diagonal nonlinear system.

Define
d
x (the setpoint) and x (the error signal) the difference between x and
d
x .

( ) , , ,
d d d d
x x x x = ` `` .
d
x x x =

We take a 2 step approach to designing the controller:
1. Define the sliding mode. This is a surface that is invariant of the controlled
dynamics, where the controlled dynamics are exponentially stable, and where the
system tracks the desired set-point
2. Define the control that drives the state to the sliding mode in finite time

Define the sliding mode ( ) S t as follows:

( ) ( ) { }
, 0 S t x s x t = =

where ( ) , s x t is defined by

( ) ( )
1
,
n
d
s x t x t
dt

| |
= +
|
\ .
, 0 >

Note that on the surface ( ) S t , the error dynamics are governed by the equation
( )
1
0
n
d
x t
dt

| |
+ =
|
\ .


On this surface the error will converge to 0 exponentially.
- 2 -

This implies that if there exists a control input ( ) u t such that ( ) x t is in ( ) S t it follows that
( ) x T is in ( ) S T for all T t > and the error will converge exponentially to 0 for this
control input.


Remark 1.1: The choice of ( ) , s x t is somewhat arbitrary. You may choose any error
dynamic which leads to exponentially stable behavior

( ) ,
d
s x t p x
dt
| |
=
|
\ .


Where ( ) p s is a polynomial with all zeros in the left half plane, then the error dynamics
will converge exponentially. For example we could have

( )
2
1 p s s s = + +

1
d
p x x x
dt
| |
= + +
|
\ .
`` `


The strategy to converge to the sliding mode is that we add something to the ( ) u t which
will drive us to the sliding mode in finite time.

More formally, assume that for all x, 0 ) ( x b . To prove that the sliding mode exists, we
derive a control such that 0 = s` . With s(x,t) defined as above:
x x s
n n ~ ~ 1 ) 1 (
+ = .
Differentiating:
x x u x b x f
x x x
x x s
n n
d
n n
d
n
n n
`
.
`
.
`
. `
~
) ( ) (
~
~ ~
1 ) (
1 ) ( ) (
1 ) (

+ + + =
+ + =
+ + =


Defining u as:
( ) x x x f
x b
u
n n
d
`
.
~
) (
) (
1
1 ) ( *
+ =
Then for
*
u u = it is clear that for all time 0 = s` . In order to complete the second control
design step, we consider the following control:
) sgn(
) (
*
x
x b
k
u u =
Where:
0
0
0
1
0
1
) sgn(
>
=
<

=
x
x
x
x

- 3 -
To prove convergence to the sliding mode, we show that with this control, s will converge to
0 in finite time. Consider the development of
2
2
1
s over time. Differentiating gives:
( )
s k
s s k
s s k
s s
s
t
=
=
=
=

). sgn(
)). sgn( (
.
2
1
2
`

Thus k s = ` , so we get convergence to the sliding mode in finite time.

In summary, the motion consists of a reaching phase during in which trajectories starting off
the manifold 0 s = move toward it and reach it in finite time, followed by a sliding phase
during which the motion is confined to the manifold 0 s = and the dynamics of the system
are represented by a reduced-order model with exponentially stable error dynamics. The
manifold 0 s = is called the sliding mode, and the control law ( ) ( ) sgn u x s = is called
sliding control mode.

We now present 2 examples demonstrating how to achieve robustness with this approach.


Example 1.1: Consider the second order system

( ) , x f x x u = + `` `

where x is a scalar. Let ( ) , f x x ` be an uncertain function, where only and estimate of the
true state equation ( )

, f x x ` is known, and that the error may be bounded



( ) ( ) ( )

, , , f x x f x x F x x ` ` ` , , x x `

Define the sliding mode:

1 n
d
s x x
dt

| |
= + = +
|
\ .
`
, 2 n =

then

( ) ,
d d
s x x x f x x u x x = + = + +
` `
` `` `` ` ``

Choose

( )

,
d
u f x x x x = +
`
` `` ,

f f = and 0 s = `

and define the control

( )
sgn u u k s =
- 4 -

We consider ( ) , s x t to be a measure of how far we are from the sliding mode. In order to
force the system to stay on the sliding mode, we choose ( ) u t such that 0 = s` .

To prove convergence to the sliding mode, consider the derivative of the distance of the
point from the sliding mode
2
1
2
s .

( ) ( ) ( )
( )
( ) ( )
( )

2
0
1
2

, , sgn

, ,
F
d
s s s
dt
f x x f x x k s s
f x x f x x s k s
<
<
=
=
=
`
` `
` `
_


and for ( ) ( ) , , k x x F x x = + ` ` we have

2
1
2
d
s s
dt


With the control ( )
sgn u u k s = we achieve convergence to the sliding mode. In order to
cope with uncertainty, we choose a higher gain k in the control.



- 5 -
Example 1.2: Consider the second order system


( ) ( )
1 2
2
x x
x h x g x u
=
= +
`
`


where h and g are unknown nonlinear functions and ( )
0
0 g x g > for all x . We want to
design a state feedback control law to stabilize the origin. Suppose we can design a control
law that constrains the motion of the system to the manifold (or surface)
1 1 2
0 s a x x = + = .
On this manifold, the motion is governed by
1 1 1
x a x = ` . Choosing
1
0 a > guarantees that
( ) x t tends to zero as t tends to infinity and the rate of convergence can be controlled by
choice of
1
a . The motion on the manifold 0 s = is independent of h and g .

The variable s satisfies the equation

( ) ( )
1 1 2 1 2
s a x x a x h x g x u = + = + + ` ` `

Suppose h and g satisfy the inequality


( )
( )
( )
1 2
a x h x
x
g x

+
,
2
x

for some known function ( ) x . With ( )
2
1 2 V s = as a Lyapunov function candidate for
( ) ( )
1 2
s a x h x g x u = + + ` , we have

( ) ( ) ( ) ( ) ( )
1 2
V ss s a x h x g x su g x s x g x su ( = = + + +

`
`

Taking

( ) ( ) sgn u x s =

where ( ) ( )
0
x x + ,
0
0 > , and

( )
1, 0
sgn 0, 0
1, 0
s
s s
s
>

= =

<



yields

( ) ( ) ( ) ( ) ( ) ( )
0 0 0 0
sgn V g x s x g x x s s g x s g s ( + =

`


Therefore, the trajectory reaches the manifold 0 s = in finite time, and once on the manifold,
it cannot leave it, as seen from the inequality
0 0
V g s
`
.
- 6 -






The next figure represents a typical phase portrait under sliding mode control




We have to note that the controller is discontinuous at 0 s = . Due to the effects of sampling,
switching and delays in the devices used to implement the controller, respectively in the
simulation engines used when modelling the controlled system, sliding mode control suffers
from chattering

The next figure shows how delays can cause chattering. It depicts a trajectory in the region
0 s > heading toward the sliding manifold 0 s = . It first hits the manifold at a point a . In
ideal sliding mode control, the trajectory should start sliding on the manifold from a point
a . In reality, there will be a delay between the time the sign of s changes and the time the
control switches. During this delay period, the trajectory crosses the manifold into the region
0 s < .




Chattering results in low control accuracy, high heat losses in electrical power circuits and
high wear of moving mechanical parts. It may also excite unmodeled high frequency
dynamics, which degrades the performance of the system and may even lead to instability.

- 7 -
There are many strategies used to avoid chattering, e.g. you can introduce a boundary layer.
Here, the sgn function is made continuous by using a piecewise linear approximation





Within the boundary layer you have exponentially convergence to the sliding mode. You
rely on continuity arguments to show that the system will still converge.
- 9 -

Вам также может понравиться