Professional Documents
Culture Documents
University of Cagliari
Summer School on
ODEs with Discontinuous Right-Hand Side: Theory and Applications
Dobbiaco (BZ) Italy
eusai@diee.unica.it
x = f ( x, u, t ) x G Rn u Rq
G = {x R n : ( x, t ) = 0} : R n R + R q
When the differential inclusion defining the closed loop dynamics belongs to the
tangential space of the surface G, a 2nd Order Sliding Mode (2-Sliding Mode)
appears
x = f ( x, u, t ) x G2 R n u R q
G2 = {x R n : ( x, t ) = ( x, t ) = 0} : R n R + R q
The 2nd Order Sliding Mode is characterized by a n-2q reduced order dynamics
x1 ( x, t ) = 0
x3
( x, t ) = ( x, t ) = 0
The constraint ( x, t ) = 0 depends on the system dynamics as ( x, t ) = f ( x, u, t ) +
x t
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 4
L2 HOSM in VSS
Example
x1 = x2
y 1 1 0 x1
x2 = x3 y = 0 1 1 x y = x3 + f ( y , y , w, u, t )
x3 = f ( x, u, t ) 2 w = w + y
w 1 0 0 x3
y = x1 + x2
x2
x2 + x3 = 0
x1 + x2 = 0
x1
x3
x = f ( x, u, t ) x Gr R n u Rq
n d ( x, t )
k
Gr = x R : k
= 0, k = 0,1,, r 1 : R n R + R q
dt
The overall number of constraints must be less than the system order
n>rq
Definition
Let the r-sliding set G be non-empty and assume that it is locally an integral
r
system state belonging to the set G is called r-Sliding Mode (r-SM) with
r
when the control u is able to constrain the system state onto the set G
r
x = f ( x , u, t )
x
x 0
1 0
x0 =
0 + Iq u
u
f ( x , u, x0 ) x + f ( x , u, x0 ) f ( x , u, x0 )
x x x0
u
x = f ( x ) + g ( x ) u
h
h( x ) : R R
n
h( x ) = Gradient vector
x
h1 ( x )
h( x ) : R n R q R p h( x ) = = J h Jacobian matrix
h1 ( x )
h( x ) : R n R Lg h = h ( x ) g ( x )
Lkg h = ( Lkg 1h ) g ( x ) i = 1,2, L0g h = h ( x )
Lie derivatives
g( x ) : R n R n
h( x ) : R R
n p Lg h = h( x ) g ( x ) = J h g ( x )
g( x ) : R n R n Lkg h = ( Lkg 1h ) g ( x ) i = 1,2, L0g h = h( x )
x Rn u Rq
x = f ( x ) + g ( x ) u
f : R n Rn g : R n R n Rq
n d ( x )
k
Gr = x R : k
= 0, k = 0,1,, r 1 : R n R + R q u Rq
dt
Lg Lkf = 0 i = 0,2,, r 2 The control does not appear in the first r-1
derivatives of the system output (x)
, Terminal Zhihong, Yu
Coefficients ci are chosen so that the polynomial P(p) has all roots with
negative real part r 2
P( p ) = p r 1
+ ci p i
i= 0
The control u(t) affects the time derivative of s(x), and if it is designed so that
s(x) is stabilized to zero, each output variable s , and its r-1 time derivatives, i
are stabilized to zero asymptotically
Chose the sliding variable set (x) such that the internal dynamics corresponding
, , ( r 1) is BIBS stable.
to the output variables , ,
1 T
Lyapunov function V ( s) = s s
2
V = T
r 2
( x ) + i = 0 i i+ 1
c ( x ) +
r 2 r
s
= s Lf + ci Lf
T r i+ 1 r 1
Lg Lf
i= 0 m s 2
sT Lg Lrf 1 s s 2 < V < 0
ms2
Once the condition s=0 is fulfilled, each sliding variable motion is characterized
by a asymptotically stable linear dynamics.
Dynamic Higher Order Sliding Modes are an extension of the classic 1-SM
Perfect knowledge of the time derivatives of the sliding variable up to the r-1
order are needed
+ c0 = 0
Coefficients p and m (p>m) are odd numbers so that the motion on the
hyper-surface s=0 in the phase space is characterized by the finite time
reaching of the origin
The control u(t) affects the time derivative of s(x), and if it is designed so that
s(x) is stabilized to zero, each output variable s , and its r-1 time derivatives,
i
are stabilized to zero asymptotically
Chose the sliding variable set (x) such that the internal dynamics corresponding
to the output variables , is BIBS stable.
1 T
Lyapunov function V ( s) = s s
2
V =
1
p T
m
{ p
}
s diag i2 m ( L2f + Lg Lf u( t ) )
=
1
p T
m
s diag i {
2 p m 2
} m 2 p m
Lf
+ ( 2 + ) sgn( s )
p
p T p m 1
s < 0 0
m
sgn( ) = 0
Once the condition s=0 is fulfilled, each sliding variable motion is characterized
by a finite time stable linear dynamics.
Perfect knowledge of 1
-3
-6 -4 -2 0 2 4 6
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 21
L2 Single Input 2-SMC
Both dynamic HOSM and terminal 2-SM are implementations of the classical
sliding mode control approach but with a peculiar sliding manifold
Several algorithms that implement a 2nd Order Sliding Mode directly have been
presented in the literature
Because of the lack of Lyapunov like methods for proving their finite time stability,
they are usually presented with reference to Single Input systems
x = f ( x ) + g ( x ) u x G Rn u R
G = {x R n : ( x, t ) = 0} : R n R + R
d /d t
-5
( 0;1)
= 07
: modulation parameter
: anticipation parameter 0
d /d t
U: gain parameter
-5
u = ( t )U sgn ( M ) 1
1 ( M ) M 0
(t) = 0 .5
( M ) M < 0 0 .5 *y 1 M y1M
( 0;1) 0
-0 .5
s lid in g v a r ia b le d e r iv a t iv e
-1
increasing VM
-1 .5
anticipation
-2
modulation 1/
-2 .5
reference
-3
-3 -2 -1 0 1 2 3
s lid in g v a r ia b le
Chose the sliding variable set (x) such that the internal dynamics corresponding
to the output variables , is BIBS stable.
U>
1
Dominance condition Mk+ 1 Mk
m 0
-1
Convergence condition -2
+ MU mU
2 + M U ( 1 )
[1;+ )
-3
;+
mU ( 1 + ) -4
-5
-6
+ mU
Mk+ 1 Mk , ( 0,1) -7
M k
-8
MU
M k 2(1 )( M U + ) Mk , -9
-1 0
-8 -6 -4 -2 0 2 4 6 8 10
The values M,k can be ideally evaluated by looking at the past, or by checking
the time instant at which the time derivative of the sliding variable vanishes
The reaching phase can be characterized by twisting around the origin of the ,
plane, or by monotonic convergence of the sliding variable
+ (1 ) M U
Condition for monotonic convergence [1;+ ) ;+
mU
= 0
Condition for Twisting algorithm 2 + M U
>
mU
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 29
L2 Single Input 2-SMC Sub Optimal
The parameters of the Generalized Sub-Optimal 2nd Order Sliding Mode Control
algorithm can be adjusted to guarantee the global convergence property
M k 2(1 )( M U + ) Mk
Very small values of will cause very slow convergence to the 2-SM
m + M 2(1 ) M ,1 1
0+U
mU + MU + [ (1 ) M m ]U
1
MU
L2f ( , )
u = U M k sign k ( )
0 < m Lg Lf M ( , )
Mk
1
k = max ;1
2 (
2
2 k + M kU M k )
UMk >
1
Mk
( k + 13 2
)
k = ( Mk , Mk )
Mk = M ( Mk , Mk )
A prediction of the steepest transient is performed to evaluate the right control
gain and anticipation in order to limit the magnitude of the time derivative of
3 10
S T D s u b -o p t im a l a lg o rit h m
2 .5 5
2 0
1 .5 -5
p la n t c o n t ro l
o u tp u t
G lo b a l s u b -o p t im a l a lg o rit h m
1 -1 0
G lo b a l s u b -o p t im a l a lg o rit h m
0 .5 -1 5
0 -2 0
S T D s u b -o p t im a l a lg o rit h m
-0 .5 -2 5
0 0 .5 1 1 .5 0 0 .5 1 1 .5
t im e [s ] t im e [s ]
The red line shows the case of a more restrictive condition on the time derivative
of the sliding variable
2
: upper bound of the drift term
1
m, M: positive lower and upper
0
bounds of the gain term
-1
-3
only -6
-3 -2 -1 0 1 2
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 33
L2 Single Input 2-SMC Supertwisting
Theorem.
Consider the system dynamics x = f ( x ) + g ( x ) u
Chose the sliding variable set (x) such that the internal dynamics corresponding
to the output variables , is BIBS stable.
Convergence condition
1
L2f + ( Lf Lg + Lg Lf ) u + u 2 L2g M k + 1
0 .5
0 < m Lg M 0
+ M
> , 2 > 2
m m - 0 .5
-1
M k
M k + 1 M k , ( 0,1) - 1 .5
- 0 .2 - 0 .1 0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6
The reaching phase are characterized by a smooth twisting around the origin
of the , plane
The 2_SM feature allows for easy implementation of an exact finite time
convergent first order differentiator
The main point is to design a control law with the less measurement demand as
possible
Find a finite-time stabilizer that depends only on the sign of the phase variables
u = Usign( )
? u = U1sign( ) U 2 sign()
U > /m m [U1 + U 2 ] > M [U1 U 2 ] +
Anosov unstable
AU u = U 0 sign( )
The appearance of a sequence of
zero crossings is ensured. s(t)
t zi i = 1,2,...,
: ( t zi ) = 0
( t zi ), ( t zi ) as i
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 39
L2 Single Input 3-SMC
What is the system behaviour under the separate action of the two
considered structures ?
1 1
u = (U1 + U 2 ) sign( s) (U1 U 2 ) sign( s)
Generalized Twisting
G-TW 2 2
and both converge to zero in
finite time
1 1
Generalized Twisting
u = (U1 + U 2 ) sign( s) (U1 U 2 ) sign( s)
G-TW 2 2
Chose the sliding variable set (x) such that the internal dynamics corresponding
to the output variables , , is BIBS stable.
MU 0 + MU 0 +
1= 2=
Define the parameters mU1 mU 2
i
U U +
3= m 1 4= M 2
MU 2 + mU 2
1) Set U0 >
m
1 ( 3 + 2 1 ) + 2 2 (1 + 2 1 ) 2 <
3
e ,i + 1 e ,i i = 0,1,2,
A typical time-evolution of
the sliding quantity
The Hybrid 3rd Order Sliding Mode Control algorithm is based on switching
between an unstable controller and a stable controller that in general force the
closed loop system onto an equilibrium point with 0
Tuning is quite easy if the step of the procedure are carefully performed
The Arbitrary Order Sliding Mode Control laws are based on a recursive
algorithm deriving form the prescribed law of variation 2-SMC
A r-SMC algorithm needs the knowledge of the sliding variable, of its r-2 time
derivatives and the sign of its r-1 time derivative
As for the Dynamical Sliding Mode Control it is assumed that the time
derivatives of the sliding variable are available
Lg Lkf = 0 i = 1,2,, r 2
Define a suitable output (x)
Lg Lrf 1 0
0 ,r =
Define the hyper-surfaces
i ,r = ( i ) + i N i ,r sgn(i 1,r ) i = 1,2,, r 1
i ,r = 0 i 1,r t r1 i 0 i = 1,2,, r 1
Example r=3
m = 2 3 = 6
2
N1,r = 3
N 2 ,r = ( 6
3
+
6
2
) 1
6
0 ,r =
sgn( )
2
1,r = + 1 3
2 ,r = + 2 ( 6
3
+
6
2
) 1
6
(
sgn + 1
2
3
sgn( ))
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 49
L2 Arbitrary Order SMC
Lrf
If the sliding dynamics is bounded r 1
0 < m Lg Lf M
( r)
[ ,+ ] + [ m , M ] u
The coefficients appearing on each i,r are such that the closed loop
dynamics has homogeneity properties
F( x ) = q
d 1F( d x ) > 0
The homogeneity property implies the finite time stability once the
contraction is proved
S.V. Emelyanov, S.K. Korovin, L.V. Levantovsky, Higher order sliding regimes in the
binary control systems, Soviet Physics, Doklady, 31, pp. 291293, 1986.
A. Levant, Sliding order and sliding accuracy in sliding mode control, Int. J. Control,
58, pp. 1247-1263, 1993.