Implicit Euler Numerical Simulation of Sliding Mode Systems
Transcript of Implicit Euler Numerical Simulation of Sliding Mode Systems
![Page 1: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/1.jpg)
Implicit Euler Numerical Simulation of Sliding
Mode Systems
Vincent Acary and Bernard Brogliato,INRIA Grenoble Rhone-Alpes
Conference in honour of E. Hairer’s 60th birthday
Geneve, 17-20 June 2009
![Page 2: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/2.jpg)
Objectives
◮ Study the implicit Euler discretization of a class of differentialinclusions with sliding surfaces (⊂ Filippov’s systems)
◮ Show that this numerical method permits a smoothstabilization on the sliding surface, in a finite number of steps
◮ Show how this may be used in real-time implementations ofsliding mode control
![Page 3: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/3.jpg)
To start with we consider the simplest case:
x(t) ∈ −sgn(x(t)) =
1 if x(t) < 0−1 if x(t) > 0[-1,1] if x(t) = 0
, x(0) = x0 (1)
with x(t) ∈ IR. This system possesses a unique Lipschitzcontinuous solution for any x0. The backward Euler discretizationof (1) reads as:
xk+1 − xk = −hsk+1
sk+1 ∈ sgn(xk+1)(2)
![Page 4: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/4.jpg)
As is known the explicit Euler discretization of such discontinuoussystems yields spurious oscillations around the switching surface[Galias et al, IEEE TAC and CAS 2006, 2007, 2008].
this means that the derivative of the switching function whilesliding occurs, is very badly estimated.
Both the explicit and the implicit methods converge (theapproximated solution xN(·) tends to the Filippov’s solution ash → 0).However or the backward Euler method the following holds:
LemmaFor all h > 0 and x0 ∈ IR, there exists k0 such that xk0+n = 0 andxk0+n+1 − xk0+n
h= 0 for all n ≥ 1.
![Page 5: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/5.jpg)
On this simple case this has the following graphical interpretation,as the intersection of two graphs:
xi
−h
h
xk xk+1 xk+2xk−1
si
Figure: Iterations of the backward Euler method.
![Page 6: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/6.jpg)
An interesting property is that the smooth stabilization and thefinite-time convergence on the switching surface, hold (more orless) independently of the step h > 0:
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
t
x(t)−s(t)
(a) h = 0.2
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
t
x(t)−s(t)
(ii)
(b) h = 0.02
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
t
x(t)−s(t)
(c) h = 0.01
Figure: A simple example for x0 = 1.01 at t0 = 0.
![Page 7: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/7.jpg)
EXTENSIONS
We shall focus on inclusions of the form:
x(t) ∈ f (t, x(t)) − B Sgn(Cx(t) + D), a.e. on (0,T )
x(0) = x0
(3)
with
B ∈ IRn×m
Sgn(Cx(t) + D)∆= (sgn(C1x + D1), ..., sgn(Cmx + Dm))T ∈ IRm,
where sgn(·) is multivalued at 0.
![Page 8: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/8.jpg)
Well-posedness of the differential inclusions (3)
Proposition
Consider the differential inclusion in (3). Suppose that
◮ There exists L ≥ 0 such that for all t ∈ [0, T ], for all x1, x2 ∈ IRn,one has ||f (t, x1) − f (t, x2)|| ≤ L||x1 − x2||.
◮ There exists a function Φ(·) such that for all R ≥ 0:
Φ(R) = sup
{
‖ ∂f
∂t(·, v) ‖
L2((0,T );IRn) | ‖ v ‖
L2((0,T );IR n)≤ R
}
< +∞.
If there exists an n × n matrix P = PT > 0 such that
PB•i = CTi• (4)
for all 1 ≤ i ≤ m, then for any initial data the differential inclusion(3) has a unique solution x : (0,T ) → IRn that is Lipschitzcontinuous.
![Page 9: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/9.jpg)
Proof of Proposition 1: The proof uses the change of statevariables z = Rx where R = RT > 0 and R2 = P . After somemanipulations the system is rewritten as
z(t) ∈ Rf (t,R−1z(t)) −m
∑
i=1
∂fi (z(t)) (5)
where fi (z) = |Ci•R−1z + Di |. The multivalued mapping
z 7→ ∑
m
i=1 ∂fi (z(t)) is maximal monotone. We then use a result in[Bastien-Schatzman ESAIM M2AN 2002] to conclude.
![Page 10: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/10.jpg)
◮ The existence of a positive definite P such that PB = CT issatisfied in many instances of sliding-mode control:observer-based sliding-mode control, Lyapunov-baseddiscontinuous robust control.
◮ This is an “input-output” constraint on the system,constraining the relative degree of the triple (A,B ,C ).
◮ It is satisfied when (A,B ,C ) is positive real (dissipative).
![Page 11: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/11.jpg)
Time-discretization of (3)
The differential inclusion in (3) is therefore discretized as follows:
{ xk+1 − xk
h∈ f (tk , xk) − BSgn(Cxk+1 + D), a.e. on (0,T )
x(0) = x0
(6)From [Bastien-Schatzman ESAIM M2AN 2002] we have that:
Proposition
Under Proposition 1 conditions, there exists η such that for allh > 0 one has
For all t ∈ [0,T ], ||x(t) − xN(t)|| ≤ η√
h (7)
Moreoverlimh→0+ maxt∈[0,T ] ||x(t) − xN(t)||2 +
∫
t
0 ||x(s) − xN(s)||2ds = 0.
![Page 12: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/12.jpg)
However we have more: the discrete state reaches the slidingsurface (when it exists) in a finite number of steps, and stabilizeson it in a smooth way.
Let y(t)∆= Cx(t) + D.
LemmaLet us assume that a sliding mode occurs for the indexα ⊂ {1 . . . m}, that is yα(t) = 0, t > t∗. Let C and B be such that(4) holds and Cα•B•α > 0. Then there exists hc > 0 such that∀h < hc , there exists k0 ∈ IN such that yk0+n = Cxk0+n+1 + D = 0for all integers n ≥ 1.
Such algorithms are similar to proximal algorithms which possess
finite-time stabilization properties [Baji and Cabot, Set-Valued Analysis
2006].
![Page 13: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/13.jpg)
Remarks
◮ Contrarily to other methods that reduce (not suppress...)chattering, the discrete-time sliding surface is equal to thecontinuous-time sliding surface.
◮ At each step one has to solve a generalized equation withunknown xk+1 that takes the form of a mixed linearcomplementarity system (MLCP).
◮ Specific MLCP solvers are needed to implement the method.
![Page 14: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/14.jpg)
NUMERICAL EXPERIMENTS
Let us consider the following two examples:
x =
[
0 10 −c1
]
x −[
0α
]
sgn([
c1 1]
x). (8)
(codimension one sliding surface)
B =
[
1 22 −1
]
, C =
[
1 22 −1
]
, D = 0, f (x(t), t) = 0
(9)(codimension two sliding surface)
![Page 15: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/15.jpg)
-1
-0.5
0
0.5
1
1.5
2
2.5
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
x1(t)
x 2(t
)
(a) h = 0.3.Explicit Euler
-1
-0.5
0
0.5
1
1.5
2
2.5
-0.2 0 0.2 0.4 0.6 0.8 1
x1(t)
x 2(t
)
(b) h = 0.1.Explicit Euler
-1
-0.5
0
0.5
1
1.5
2
2.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
x1(t)
x 2(t
)
(c) h = 1. Im-plicit Euler
-1
-0.5
0
0.5
1
1.5
2
2.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
x1(t)
x 2(t
)
(d) h = 0.3.Implicit Euler
-1
-0.5
0
0.5
1
1.5
2
2.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x1(t)
x 2(t
)
s
(e) h = 0.1.Implicit Euler
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
0 0.2 0.4 0.6 0.8 1 1.2
x1(t)
x 2(t
)
(f) h = 0.05.Implicit Euler
Figure: Equivalent control based SMC, c1 = 1, α = 1 andx0 = [0, 2.21]T . State x1(t) versus x2(t).
![Page 16: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/16.jpg)
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
time t
x1(t)x2(t)
stat
ex 1
(t)
and
x 2(t
)
(a) state x1(t) and x2(t)versus time
-1
-0.8
-0.6
-0.4
-0.2
0
0 0.2 0.4 0.6 0.8 1
x1(t)
x 2(t
)
(b) phase portrait x2(t)versus x1(t)
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
time t
s1(t)s2(t)
sva
lues
(c) sgn function s1(t) ands2(t)
Figure: Multiple Sliding surface. h = 0.02, x(0) = [1.0,−1.0]T
The system reaches firstly the sliding surface 2x2 + x1 = 0 without any
chattering, The system then slides on the surface up to reaching the
second sliding surface 2x1 − x2 = 0 and comes to rest at the origin.
![Page 17: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/17.jpg)
The Filippov’s example with switches accumulation
B =
[
1 −22 1
]
, C =
[
1 00 1
]
, D = 0, f (x(t), t) = 0.
(10)
The trajectories may slide on the codimension 2 surface given byCx = 0. The origin is attained after an infinite number of switchesin finite time.
![Page 18: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/18.jpg)
-1.5
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2time t
x1(t)
x2(t)
stat
e
(a) state x1(t) and
x2(t) versus time
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1x1(t)
x 2(t
)
(b) phase portrait
x2(t) versus x1(t)
-1
-0.5
0
0.5
1
0 0.5 1 1.5 2
s va
lues
time t
s1(t)
s2(t)
(c) sgn function s1(t)
and s2(t)
Figure: Multiple Sliding surface. Filippov Example. h = 0.002,x(0) = [1.0,−1.0]T
The results show that the system reaches the origin without anychattering.
![Page 19: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/19.jpg)
The case of Zero Holding discretization (ZOH) for controlimplementation
The ZOH discretization of linear time invariant systemsx(t) = Fx(t) + Gu(t) with an ECB-SMC controller,u(x) = −(CG )−1(CFx + αSgn(Cx)), α > 0 results in adiscrete-time system of the form:
xk+1 = Φxk − Γsk for all t ∈ [kh, (k + 1)h) (11)
where h > 0 is the sampling period, and
Φ = exp(Fh) −∫
h
0exp(F τ)dτG (CG )−1CF (12)
Γ =
∫
h
0exp(F τ)G (CG )−1dτ (13)
with G ∈ IRn×m, C ∈ IRm×n, when an explicit Eulerimplementation of the control is performed.
![Page 20: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/20.jpg)
Similarly to the explicit Euler method, the explicit ZOHdiscretization may yield spurious oscillations as shown in[Galias-Yu, IEEE CAS 2008].
For an implicit Euler implementation, let us set
{
uk = −(CG )−1(CFxk + sk+1)
sk+1 = Sgn(Cxk+1),(14)
which corresponds to the implicit discrete time version of theECB-SMC controller. We therefore get on each sampling period:
xk+1 = Φxk − Γsk+1 for all t ∈ [kh, (k + 1)h) (15)
![Page 21: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/21.jpg)
At each time–step, one has to solve
xk+1 = Φxk − Γsk+1
yk+1 = Cxk+1 + D
sk+1 ∈ Sgn(yk+1)
. (16)
Inserting the first line of (16) into the second line we obtain thefollowing one–step system (MLCP)
{
yk+1 = CΦxk + D − CΓsk+1
sk+1 ∈ Sgn(yk+1). (17)
![Page 22: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/22.jpg)
The control scheme looks like:
MLCP solver
−
−
CFxk (CG)−1
uk xk
sk+1
Discrete-time Plant
Figure: Control system scheme with implicit Euler implementation.
![Page 23: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/23.jpg)
A numerical example
The LTI system with an ECB-SMC controller is defined by thefollowing data,
F =
[
0 1−a1 −a2
]
, G =
[
01
]
, C =[
c1 1]
. (18)
Starting from the initial data, x0 = [0.55, 0, 55]T , [Galias-Yu, IEEECAS 2008] have shown that the Explicit ZOH discretization of thesystem with a1 = −2, a2 = 2, c1 = 1 and h = 0.3 exhibits aperiod–2 orbit.
![Page 24: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/24.jpg)
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
x1(t)
x 2(t
)
(a) h = 0.3. Explicit ZOH
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
x1(t)
x 2(t
)
(b) h = 0.3. Implicit ZOH
Figure: Equivalent control based SMC, a1 = −2, a2 = 2, c1 = 1 andh = 0.3. x0 = [0.55, 0, 55]T State x1(t) versus x2(t).
the chattering on the switching surface is suppressed.
![Page 25: Implicit Euler Numerical Simulation of Sliding Mode Systems](https://reader031.fdocuments.in/reader031/viewer/2022011905/61d6d689d064fc5f462a54e5/html5/thumbnails/25.jpg)
CONCLUSIONS
The implicit Euler method allows one to nicely simulate the mainfeatures of sliding-mode systems:
◮ Finite-time stabilization on the switching surface (ofcodimension ≥ 1)
◮ Smooth stabilization on the switching surface
It extends to the discrete-time implementation with ZOHdiscretization: looks like a promising solution for discrete-timesliding modes.