sliding mode control: basic theory, advances and applications · discrete-time sliding mode control...
TRANSCRIPT
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 1
Dept of Electrical and Electronic Eng.Dept of Electrical and Electronic Eng.University of CagliariUniversity of Cagliari
Summer School onODEs with Discontinuous Right-Hand Side: Theory and Applications
Dobbiaco (BZ) – Italy
Sliding Mode Control: Basic Theory, Advances and Applications
Elio USAI
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 2
• Approximate Sliding Modes• Discrete Time implementation• Effect of Parasitic Dynamics• Effect of Measurement Noise• Chattering Attenuation
➢ Control magnitude adaptation➢ Parameter tuning in 2-SMC➢ System dynamics shaping
Lecture 3
Implementation Issues of
Sliding Mode Control Systems
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 3
L3 – Approximate SM
Ideal Sliding Modes in Control Systems can be established if infinite frequency switching in the closed loop dynamics appears
Real devices has low-pass characteristics and therefore cannot perform infinite frequency switching
Switching delay appears
The system state is no more constrained on the sliding surface
WHAT is the EFFECT of Switching Delays?
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 4
L3 – Approximate SM
Assume that the sliding surface is an attractive set of the closed-loop dynamics
At certain time instant t0 the system state is within an vicinity of the sliding
surface and the system dynamics is represented by the input-output and internal dynamics
( ) ( ) ( ) ( )( )( ) ( ) ( )( ) ( )
∈∈∈==
= − qqnqσtttt
tttttR,R,R
,,,,,
uwxywyw
uwyyψϕ
( ) ( )( ) ( ) ( ) ( )( )tttttttf ,,,,, uwyuxxσ ϕ=⋅
∂∂
( ) qnqn −∈∈∈Φ=
R,R,R wyxxwy
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 5
L3 – Approximate SM
General treatment of the analysis of the system behaviour nearby the sliding surface is quite complex and could need a Poicaré analysis
qnqqnq −− ∈∈∈∈⋅=
⋅
=
R,R,RRˆ21
2
1
22
1211 wyxxxCxx
C0CC
wy
Complete results can be quite easily obtained in the linear case for the classic first order sliding mode control systems
( ) ( ) ( ) ( ) ( )( ) ( )tt
tttttxCy
uBxAx⋅=
⋅+⋅=nqqqn <∈∈∈ ,R,R,R yux
( ) ( ) ( ) ( ) ( ) ( ) ( )( ) ( ) ( ) ( ) ( )
⋅+⋅=⋅⋅+⋅+⋅=
tttttttttttt
wΨyΨwuBCwΦyΦy
21
21
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 6
L3 – Approximate SM
Assuming that the control is designed taking into account for the nominal dynamics
( ) ( )[ ] ( ) ( ) ( ) ( ) ( )( )( )tUtttttt ywΦyΦBCu sgn211 +⋅+⋅⋅−= −
( ) ( ) ( ) ( ) ( ) ( )( )( ) ( ) ( ) ( ) ( )
⋅+⋅=−⋅+⋅=
ttttttUttttt
wΨyΨwywΦyΦy
21
21 sgn~~
If the system were in ideal sliding mode the system dynamics will be characterized by its zero dynamics
( ) ( ) ( ) ( )( )( ) ( ) ( ) ( )( ) ( ) ( )∫
+
−→ ⋅ →
⋅=−⋅= ε
εεττ
ε
t
t
ttdUttt
tUtttwΦy
wΨwywΦy
202
2 ~sgn2
sgn~
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 7
L3 – Approximate SM
The dynamics of the error between the ideal and real sliding behaviour
( ) ( ) ( ) ( ) ( ) ( )( )( ) ( ) ( ) ( ) ( )
⋅+⋅=−⋅+⋅=
ttttttUttttt
wΨyΨwywΦyΦy
~~~sgn~~~~
21
21
( ) ( )( ) ( ) ( )ttt
ttwww
yy−=
=~~
Assume that T is the switching delay and that the sliding dynamics can be upper bounded by a constant D
( ) DTt +≤ 0~~ yy
Assume also that the matrices in the error dynamics can be upper bound by proper constants during the switching delay
( ) ( ) PtQt << 21 ΨΨ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 8
L3 – Approximate SM
( ) ( )
( )
+≤
+++≤ ∫+
DTt
dPQDTTQtTt
t
0
221
00
~~
~~~0
0
yy
wyww ττ
Taking into account the BIBS assumption for the internal dynamics and that T is the switching delay
( ) [ ]TtttTt +∈∀=∆∆≤ 00,~ υx
The system trajectory remains confined within a O(T) vicinity of the ideal sliding trajectory
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 9
L3 – Approximate SM
TheoremConsider system
Assume that conditions for conver-gence and stability of a r-SM arefulfilled by a homogeneous r–ordersliding mode controller
Then, if the switching device has a switching delay T, the real r-SM has the following finite time accuracy
( ) ( ) RR: →+= nu σxgxfx
Γ≤≤Γ<−==
Φ≤
−M
rm
k
r
σLLrkσLL
σL
102,,1,0 ,0
fg
fg
f
( )( )( )1,1 ,,,sgn −
−−= rrru σσσα Φ
( ) 1,,1,0 −=≤ − rkT krk
k υσ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 10
L3 – Approximate SM
ProofUnder the considered assumption the system trajectories are infinitely extendible in time for any Lebesgue-measurable bounded feedback control and at the ideal switching time the sliding variable and its time derivatives are bounded
( )
[ ] [ ]( ) krpTK
dtd pk
Ttttk
Ttttp
kp
−−=∀≤ −
+∈+∈
1,,1,0sup,,*
σσ
Applying the Lagrange theorem
( ) rkkk ,,1,0 =Σ≤σ
Integrating σ (k) k times and taking into account above inequalities
( ) ( ) ∞−
+ ∈−=Σ≤ Kkkr
kkk rkT υυσ 1,,1,01
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 11
L3 – Approximate SM
Real actuation devices cannot implement infinite switching and therefore the system trajectory cannot be constrained on the sliding surface
The real sliding is a motion confined into a vicinity of the sliding surface
The thickness of the real sliding vicinity depends on the the switching delay T and on the control magnitude (Σ
r)
The real sliding accuracy can be improved by means of HOSM, if the switching delay is T<1
The accuracy can be also marginally improved by avoiding unnecessary large magnitude controls
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 12
L3 – Approximate SM
Example ( ) ( ) ( ) ( )cyy
tuyykkyybbytm+=
−=++++
σπsin2
3121
0 1 2 3 4 5 6- 8
- 6
- 4
- 2
0
2
4
6
8x 1 0
- 4
T i m e [ s ]
τ= 1 e - 5
0 1 2 3 4 5 6- 8
- 6
- 4
- 2
0
2
4
6
8x 1 0
- 3
T e m p o [ s ]
τ = 1 e - 4
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 13
L3 – Discrete Time Implementation
The most common cause of delay switching is the digital implementation of the controller
Discrete-time sliding mode control
[ ] [ ] [ ][ ] [ ] [ ]( ) [ ][ ] [ ]( )kk
kkkkk
kkk
dd
dd
σu
uΓσxΦσ
σΒxAx
sgn
,,1
1
α−=
+=+
+=+
Discrete-time sliding mode control can appear also in systems with continuous right-hand-side (e.g., deadbeat control)
Discrete time sliding mode control is sensible only in the presence of uncertainties or disturbances
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 14
L3 – Discrete Time Implementation
[ ] ,2,1,01 ==+ kk 0σ
What is a discrete time sliding mode?
[ ] [ ] ,2,1,01 ==−+ kkk 0σσ
The second is not convincing and does not imply the first
?
Effective approach is constituted by continuous time design and subsequent discretization analysis
The system behaviour within a sampling period is almost unpredictable, apart from the maximum deviation from the sliding surface
In some conditions chaotic behavior within the boundary layer has been recognized
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 15
L3 – Discrete Time Implementation
The usual implementation of the control law has two parts * the nominal part * the discontinuous part to cope with uncertainties
This allows for implementing learning and adaptive methods that can improve the accuracy by one order, i.e., O (T) → O (T2)
( ) ( ) ( ) ( ) ( ) ( ) ( )xxxxxxx Mm
n
ii btbbFtfutbtfx
nixx≤≤<≤
+=−== + ,0,,
,,1,2,11
( ) ( )
( )( )( ) ( )
( )( ) ( )( ) ( ]TttttTtb
TtxcTtFtu
xcutbtf
xcx
kkkkm
n
ikiik
n
iii
n
iiin
+∈+
++++−=
++=
+=
∑
∑
∑
−
=+
−
=+
−
=
,sgn
,,
21
11
1
11
1
1
σκ
ηκκ
σ
σ
x
cx
xx
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 16
L3 – Discrete Time Implementation
The switching delay due to sampling causes an approximate sliding motion in a O (T) boundary layer of the ideal sliding
σ
t
u
tk-1
t
tk
tk+1
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 17
L3 – Discrete Time Implementation
( ) ( )τ
κκτκ TTtutu eqav 321 ++≤−
In the ideal case the equivalent control can be estimated by a low-pass filter
Since the sliding variable in constrained in a O (T) boundary layer of the ideal sliding and the equivalent control remains bounded
( ) ( ) ( )tututu avav =+τ
The estimation error can be minimized and the actual value of the average control computed exactly at each sampling time
( ) ( ) TtutuT eqav 43
1 κκκτ ≤−⇒=
[ ] [ ] [ ]kuekuekuT
av
T
av
−+=+ −− ττ 11
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 18
L3 – Discrete Time Implementation
The control input can be implemented as a combination of two components
Adapting step by step the time constant of the filter and the magnitude of the discontinuous control
( ) ,2,122
215
21
3
1
=>∀≤⇒
=
= −
−
−
−
−
−
jTKtTO
TU
Tj
j
j j
j
j
j
j
j
σ
κ
κκ
τ
[ ] [ ]( )( ) ( )
( )( ) ( )( ) ( ]TttttTtb
TtxcTtFTkuku kkk
km
n
ikiik
av +∈+
++++−=
∑−
=+
,sgn
21
11
5 σκ
ηκκκ
x
cx
[ ]
→ →2
3
1TOk Kkσ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 19
L3 – Discrete Time Implementation
Example of discrete time sliding mode control with recursive estimation and adaptation of the control
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 20
L3 – Effect of the parasitic dynamics
( ) ( )( )
( )( )σ
σσµ
switchus
uz
n
m
n
=→=
→×=→+=
RR:RRR:,
RR:,m
n1
xhzhz
gfxgxfx
If the switching control is applied to the plant by means of a dynamic actuator the relative degree between the sliding variable and the switching control increases and the ideal sliding cannot be achieved
( ) ( ) ( )( )( )
( )( )σ
σσµ
σσσ
switchussss
usss
LLLzsssLLsssLs
rnrr
mm
rnrnrr
rnrrrrrrrr
=→×=
→×=→×=
→×⋅+=
−−
−−−
−−−−−
RRR:,,,,RRR:,
RRR:,,,,RRR:,,,,,,,,,
1
1
11
111
whzhzψwψw
ww fgffgf
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 21
L3 – Effect of the parasitic dynamics
If the parameter µ is sufficiently small the actuator dynamics is a singular perturbation of the nominal dynamics
a) Poincaré analysis of the fast dynamics, freezing the slow dynamics
b) Phase trajectory analysis considering differential inclusions with switching delays
c) Homogeneity of the differential inclusion
( ) [ ] [ ] 1,, zs Mmrrr ΓΓ+ΛΛ−∈
Method a) is very much involved and hard to implement for nonlinear uncertain systems;
Method b) is relative simple only for relative degree 2 sliding dynamics
Method c) is general but require the homogeneity of the controller
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 22
L3 – Effect of the parasitic dynamics
All methods confirmed that the accuracy of the sliding mode depends on the singular parameter µ only, no matter the relative degree m of the parasitic dynamics is
( ) ( ) 1,,1,0 −== − rkkrk µσ O
In general information about the system behavior within the boundary layer are not available apart for linear system
Approximate method Exact methods
Describing Function Tzipkin locus
LPRS
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 23
L3 – Effect of the parasitic dynamics
f’(u) G (jω)+∆r’(t)=0 ∆u(t) ∆m(t) ∆w(t)
_
This methods refer to linear systems with nonlinear static feedback
Give only necessary conditions for the stability of limit cycles because they consider the steady state behavior only, and are based on the harmonic balance of the feedback loop
∑∞
=
−=1k
tjkjk
jk
tj eeMeGUe kk ωϑϕω
If the linear system has low-pass characteristics the Describing Function method can be applied
( ) ( ) 0,1 =+ ωω UNjG ( ) ( )111, jabU
UN +=ω
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 24
L3 – Effect of the parasitic dynamics
Example
( ) ( ) ( ) etaa
t
kkBJjRLjjkjG
+++=
ωωωω
N y q u i s t a n d D F p l o t s
R e a l A x i s
Imag
inar
y A
xis
- 2 - 1 . 5 - 1 - 0 . 5 0 0 . 5 1- 2 0
- 1 8
- 1 6
- 1 4
- 1 2
- 1 0
- 8
- 6
- 4
- 2
0
ω
U
( )( ) ( )( ) rad 0.17594M-Urad/s 056.360
=ℜ⋅==+===ℑ crp
a
etacrjW
jWJL
kkBRp
ωπ
ωωω
R=0.4; % rotor resistance
L=0.001; % rotor inductance
ke=0.3; % voltage feedback constant
kt=0.3; % torque constant
Jm=0.01; % motor inertia
Jl=0.09 % load inertia
Bm=0.05; % motor friction coefficient
Bl=0.05; %load friction coefficient
J=Jm+Jl;
B=Bm+Bl;
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 25
L3 – Effect of the parasitic dynamics
Example
( ) ( ) ( ) etaa
t
kkBJjRLjjkjG
+++=
ωωωω
( )( ) ( )( ) rad 0.17594M-Urad/s 056.360
=ℜ⋅==+===ℑ crp
a
etacrjW
jWJL
kkBRp
ωπ
ωωω
R=0.4; % rotor resistance
L=0.001; % rotor inductance
ke=0.3; % voltage feedback constant
kt=0.3; % torque constant
Jm=0.01; % motor inertia
Jl=0.09 % load inertia
Bm=0.05; % motor friction coefficient
Bl=0.05; %load friction coefficient
J=Jm+Jl;
B=Bm+Bl;
N y q u i s t a n d D F p l o t s
R e a l A x i sIm
agin
ary
Axi
s- 2 0 - 1 5 - 1 0 - 5 0
x 1 0- 3
- 5
- 4
- 3
- 2
- 1
0
1
2
3
4
5
x 1 0- 4
S y s t e m : W p R e a l : - 0 . 0 0 5 6 8
I m a g : - 8 . 7 3 e - 0 0 7 F r e q ( r a d / s e c ) : 3 6 . 8
ω
U
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 26
L3 – Effect of the parasitic dynamics
0 1 2 3 4 5 6 7 8 9 1 0-8 0
-6 0
-4 0
-2 0
0
2 0
4 0
6 0
8 0
T im e
a n g les p e e dc u r r e n t
The system presents a periodic steady-state oscillation
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 27
L3 – Effect of the Measurement noise
A measurement noise super-imposed on the ideal sliding variable
( ) ( ) ( ) ( ) δσσ ≤+= tntntt ,ˆ
1-SMC ( ) ( )δσ O=t
2-SMC( ) ( )( ) ( )δσ
δσOO
==
tt
Possibly not convergent
r-SMC Possibly not convergent( )
,2,1,0
,
=
=−
i
O riri δσ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 28
L3 – Effect of the Measurement noise
Robust sliding mode differentiators can make the HOSM convergent even in the presence of noise
For the generalized sub-optimal a peculiar adaptation of the anticipation parameter β allows for implementing a noise robust 2-SMC
( ) ( )( ) ( )
( )[ )
( ) ( ) 0ˆ ˆlast theis ˆ1;0
0ˆˆˆ 10ˆˆˆ 1
,ˆˆsgn
*
=∋∈
<−>≥−
=
−−=
exexex
exex
exex
ex
tt
ifsif
t
sUtuσσσ
βσβσσασβσ
α
βσα
( )( )
+
−+∈
>
1,*2UGG
UGGFGFU
mM
mM
m
αβ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 29
L3 – Effect of the Measurement noise
Is it possible to estimate the sequence of the extremal values
by inspection of the measured values of σ in a proper time window if the measurements are noisy?
- 0 . 2 0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2- 2
- 1 . 5
- 1
- 0 . 5
0
0 . 5
S l i d i n g v a r i a b l e σ
Slid
ing
varia
ble
deriv
ativ
e d
σ /dt
0 0 . 5 1 1 . 5 2 2 . 5 3-0 . 2
0
0 . 2
0 . 4
0 . 6
0 . 8
1
1 . 2
T im e
Mea
sure
d sl
idin
g va
riabl
e
?
!
e x t r e m a l v a lu e s
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 30
L3 – Effect of the Measurement noise
Since the main problem is the detection of the flex points of the sliding variable σ(t), the control switchings needed as local minima are reached could be postponed by a fixed ratio of the distance between two subsequent extremal values, i.e., a local maximum and a local minimum
1,ˆˆˆˆ >−≤− N
NmM
mσσσσ
This choice guarantees the reduction of the estimation error of the flex points up to δ as the approximate sliding mode is reached
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 31
L3 – Effect of the Measurement noise
N affects the magnitude of the loop, and therefore the ultimate accuracy
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 32
L3 – Effect of the Measurement noise
NmM
mσσσσˆˆˆˆ −≥−
Mσβσ ˆˆ ≤
σσ ˆ:ˆ =m
σσ ˆ:ˆ =MN
mMM
σσσσˆˆˆˆ −≥−
mσβσ ˆˆ ≥
σσ ˆ:ˆ =M
σσ ˆ:ˆ =m
0ˆ <σ
0ˆ >σ
( )MM
Uuσσσ ˆ,ˆmaxˆ =
−=1( )MM
Uuσσσ ˆ,ˆmaxˆ =
−= 4
( )mm
Uuσσσ ˆ,ˆminˆ =
+=2
( )mm
Uuσσσ ˆ,ˆminˆ =
+= 3
Implementation of the switching logic require an automaton and modified stability conditions
n
son
Nβ
ββ
−=
+=
11
,2
1
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 33
L3 – Effect of the Measurement noise
With exact measurements and ideal switching the 2-Sliding set is reached in a finite time T
∞
( ) ( )( ) ( ) 2
102
210
,,
,,
τδσ
τδσ
FkFkk
FkFk
FU
FU
Tt
FU
FU
Tt
MM
MM
′+′′→
′+′→
∞
∞
→
→
With noisy measurements and switching delays only a boundary layer of the 2-Sliding set can be reached in a finite time T
0,0∞∞ →→
→→TtTt
σσ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 34
L3 – Effect of the Measurement noiseExample
( )( ) dd yyty
nnyyutyyy−==
≤+=+++=σ,5sin2
2.0,ˆ,3sin22
0 0 .2 0 .4 0 .6 0 .8 1 1 .2 1 .4- 4
- 3
- 2
- 1
0
1
2
3
4S u b -O p t im a l 2 - S M c o n t ro l l e r , τ = 0 . 0 1
T im e [ s e c ]
Tra
ckin
g er
ror
e x t r e m a l v a lu e s
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 35
L3 – Chattering Attenuation
Chattering cannot be eliminated but only attenuated!
Many definitions can be found:
• Discontinuous control
• Not precise attainment of the sliding
• Oscillation of the system state due to unmodelled dynamics
Continuous control can produce large and unpredictable state oscillationsNot precise attainment of the sliding can be due to design errors of the sliding surfaceUnmodelled dynamics is always present
What is Chattering ?
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 36
L3 – Chattering Attenuation
νk is an increasing function of Σ
k+1 , and in particular, by a chain rule, of
Σr , that depends on the discontinuous control magnitude
Chattering depends on the ultimate accuracy of the sliding motion
( ) ( ) ∞−
+ ∈−=Σ≤ Kkkr
kkk rkT υυσ 1,,1,01
The shape of νk depends on the overall closed loop dynamics
T can be either:• the switching delay of a relay device• the sampling time in a discrete time implementation• the equivalent time constant of a dynamic actuator
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 37
L3 – Chattering AttenuationChattering can be attenuated by means of:
Smooth approximations of the discontinuous functionIt is effective only if the matching uncertainty vanishes on the sliding surface
Implementation of HOSMIt will require the knowledge of a number of time derivatives of the sliding variable, apart from the Super-Twisting and Sub-Optimal algorithms that require Single-Input systems
Using much higher sampling frequencyIt can be sensible to high frequency noise (Aliasing phenomenon)
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 38
L3 – Chattering AttenuationChattering can be attenuated by means of:
Adaptation of the switching control magnitudeIt will require more complicated control schemes
Tuning the parameter of a 2-Order Sliding Mode ControllerIt is quite easy but some drawback on the reaching phase will follow, and it is quite simple for Single-Input linear systems only
Shaping the dynamics of the systemIt requires an additional filter it is quite simple for Single-Input linear systems only
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 39
Adaptation of the switching control magnitude
Most of the adaptation algorithms resort to the estimation of the equivalent control, and define the system input as a proper combination of the averaged control and a switching control
SystemVSCref. + _
σ u y+ +
FF Equivalent control
LPF2
LPF1
x++
uav
usw
uc
k
L3 – Control Magnitude Adaptation
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 40
L3 – Control Magnitude Adaptation
Adaptation of the switching control magnitude
If the parameter are properly chosen the system input asymptotically approaches the ideal equivalent control and the accuracy is improved
( ) ( ) ( ) ( ) ( )( ) ( )( )
( )( )( ) ( )
( ) ( ) ( )( ) ( ) ασκτ
τσ
γϕ
σ
=+=+
−=
−=
++=
tktktututu
Utut
tttu
tutktututu
ad
swavavav
sw
c
swavc
sgn
,,
xx
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 41
L3 – Control Magnitude Adaptation
Adaptation of the switching control magnitude
( ) ( ) ( )( )
( ) ( )1122
22112
21
2122112
21
12
1sinsin12
xzcxzzzzz
zzutaDtDxxxx
xx
−+−=
−+−==
++++−+−==
σµ
ωµ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 42
L3 – Control Magnitude Adaptation
Another approach for adapting the switching control magnitude is not based on the on the estimation of the continuous equivalent control but on avoiding too large control authority
In real-sliding mode the derivatives of the sliding variable must be zero in media and the discontinuous control switches at very high, but finite, frequency
A real r-order sliding mode (r-sliding) in which the rth derivative of the sliding variable σ is always separated from zero is characterised by
( )
( )21
111
10
0,,,
RRkkk
rr
rrr
≤≤<≤≤≤ −
−−
στστστσ
τ is the switching delay
∈
− ττ1,
2 1
1
rsw k
Rf
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 43
Adaptation of the switching control magnitude
Lemma: Let σ be the sliding variable of a variable structure control system, and let the discontinuous control be always separated from zero. Assume that σ(i) (i=0,1,...,r-1) are continuous functions. Consider a time interval T* of length Nτ, N∈R+, and let Nsw be the number of zero crossings of σ during the time T*. If Nsw ≥ r then, over the time interval T*, the system trajectories are confined within the domain
( ) ( )
−=≤∈= − 1,...,1 ,0 ,
! :D i2 riN
iR iir τσ xRx n
The time interval between two subsequent zero-crossing of the sliding variable in the steady state varies over [Tm , TM] that depend on the specific VSC algorithm, its parameters and τ
A proper choice for N in order to detect the real sliding could be related to the maximum "cycle-time" TM
N τ = r TM
L3 – Control Magnitude Adaptation
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 44
Adaptation of the switching control magnitudeFirst order sliding modes
( ) ( )utt ,, 11 xx γϕσ +=( )
( ) 211
1
,0,
Γ≤≤Γ<Φ≤
tt
xx
γϕ
u(t)=-UM sign (σ(ti)) ti ≤ t < ti+1 ( ti+1 - ti = τ) i=0,1,2, ….
σ τ
A
B
C
D E t
R2τ
TM
Tm
τ
τ
+=
+=
1
2
2
1
1
1
RRT
RRT
M
m
σ τ
A
B
C
D E t
R2τ
TM
Tm
τ
τ
+=
+=
1
2
2
1
1
1
RRT
RRT
M
m
Φ=
−Γ+Γ=
MURR
η
ηη
11
1
2
2
1
L3 – Control Magnitude Adaptation
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 45
L3 – Chattering Attenuation
Adaptation of the switching control magnitude
η represents the “dominance factor”.
The larger the η the larger the boundary layer size and the highest the switching frequency
As far as the actual value of η guarantees the stability of the sliding mode, the switching frequency is higher than a certain minimum value, as well as the number of changes of sign of σ with a certain time interval
UM can be adapted on-line to maintain a prescribed, desired, “dominance factor” η*
+<Λ++≥Λ−
=+ 1 ,,1 ),0,(max
sw
sw1 rNNU
rNNUU
Mi
MiMi τ
τ
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 46
L3 – Chattering Attenuation
Adaptation of the switching control magnitude
ut ++= )5sin(23σ
1 2 3 4 5-0,005
0,000
0,005
1-SMC without adaptation. The Sliding Variable.
Time [sec]
1 2 3 4 5
-0,004
-0,002
0,000
0,002
0,004
1-SMC with adaptationThe sliding variable
Time [sec]
UM=20 20=ΛN=10
τ =10-4
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 47
L3 – Parameters Tuning in 2-SMC
A feedback Second-Order Sliding Mode System Scheme is simply represented
PLANT
2-VSC
σ - Sliding variable
( ) ( ) ( )( ) ( )
RRR
∈∈∈
=+=
σσ
utt
tuttnx
CxBAxx
The controller is the Generalized Sub-Optimal
( ) ( )( ) ( )
( )[ )1;0
0ˆˆˆ 10ˆˆˆ 1
,ˆˆsgn
*
∈
<−>≥−
=
−−=
βσβσσασβσσ
α
σβσα
exex
exex
ex
ifif
t
Utu
If the plant is stable linear system with low-pass properties
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 48
L3 – Parameters Tuning in 2-SMC
The steady state analysis of the system can be approximately carried out by means of the Describing Function approach
( ) ( ) ( ) ( )[ ]{ }1*1*11*2 2 ++−+−+= αβαβαπ
jyUyN p
M
MpM
( ) ( ) BAIC 1−−= ωω jjG
( ) ( ) ( ) ( )[ ]( ) ββα
αβαβαπω−++
++−+−+−=11*
1*1*11*4 2
2 jUyjG
M
pM
The Describing Function of the Generalized Sub-Optimal controller
The harmonic response of the system
The Harmonic Balance equation
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 49
L3 – Parameters Tuning in 2-SMC
The steady state analysis of the system can be approximately carried out by means of the Describing Function approach
Re
Im
( )pMyN
1−
( )ωjG
A
O
ψ
( ) ( )( ) 211*
1*1*arctanβα
αβαψ−+
++−=
( ) ββαπ
−++=
11*22 2pM
M
yUAO
Parameters α* and β can be tuned in
order to minimize ypM
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 50
L3 – System Dynamics Shaping
The Describing Function approach can be used to properly shaping the Nyquist plot of the linear system by introducing a proper linear filter
With reference to the Generalized Sub-Optimal controller only parameter β can
be considered since optimization is not much sensitive with respect to α*
PLANT
FILTER
σ - Sliding variable
2-VSC
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 51
L3 – System Dynamics Shaping
The introduction of a low-pass filter “compensates for” the actuator dynamics in the range of frequency of the steady state oscillations
Nominal plant
Nominal plant plus parasitic dynamics
B o d e D i a g r a m
F r e q u e n c y ( r a d / s e c )
Pha
se (d
eg)
Mag
nitu
de (d
B)
1 00
1 01
1 02
1 03
1 04
- 3 6 0
- 2 7 0
- 1 8 0
- 9 0
0- 2 0 0
- 1 5 0
- 1 0 0
- 5 0
0
5 0
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 52
L3 – System Dynamics Shaping
The introduction of a low-pass filter “compensates for” the actuator dynamics in the range of frequency of the steady state oscillations
Nominal plant
Nominal plant plus parasitic dynamics
Overall Harmonic Response with the shaping low-pass filter τ=0.1s
B o d e D i a g r a m
F r e q u e n c y ( r a d / s e c )
Pha
se (d
eg)
Mag
nitu
de (d
B)
1 00
1 01
1 02
1 03
1 04
- 4 5 0
- 3 6 0
- 2 7 0
- 1 8 0
- 9 0
0- 3 0 0
- 2 0 0
- 1 0 0
0
1 0 0
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 53
L3 – References
• V.I. Utkin, Sliding Modes In Control And Optimization, Springer Verlag, Berlin, 1992.
• K.D. Young, V.I. Utkin, U. Ozguner“, A control engineers guide to sliding mode control”, IEEE Trans. Control Sys. Technology, 7, pp. 328–342, 1999.
• A. Levant, “Sliding order and sliding accuracy in sliding mode control”, Int. J. Control, 58, pp. 1247-1263, 1993.
• A. Levant, “Homogeneity approach to high–order sliding mode design”, Automatica, 41, pp. 823–830, 2005.
• C. Milosavljevic, “General conditions for the existence of a quasisliding mode on the switching hyperplane in discrete variable structure systems”, Automation Remote Control, 46, pp. 307–314, 1985.
• Bartolini G., A. Pisano, E. Usai, “An improved Second-Order Sliding Mode Control Scheme Robust Against the Measurement Noise”, IEEE Trans. Automatic Control, 49, pp. 1731-1736, 2004
• K. Furuta, “Sliding mode control of a discrete system”, System and Control Letters, 14, pp. 145-152, 1990
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 54
L3 – References
• W.-C.Wu, S. V. Drakunov, and U. Ozguner“, An O(T 2) boundary layer in sliding mode for sampled-data systems”, IEEE Trans. Automatic Control, 45, pp. 482-484, 2000.
• G. Bartolini, A. Pisano, E. Usai, “Digital second-order sliding mode control for uncertain nonlinear systems”, Automatica, 37, pp. 1371–1377, 2001.
• H. Lee, V.I. Utkin, “Chattering suppression methods in sliding mode control systems”, Annuals Reviews in Control, 31, pp. 178-188, 2007.
• L. Fridman, “Chattering analysis in sliding mode systems with inertial sensors”, Int. J. Control, 76, pp. 906-912, 2003.
• I. Boiko,L. Fridman, A. Pisano, E. Usai, “Analysis of Chattering in Systems with Second-Order Sliding Modes”, IEEE Trans. Automatic Control, 52, pp. 2085–2102, 2007.
• J.-X. Xu, Y.-J. Pan, T.-H. Lee, “Sliding Mode Control With Closed-Loop Filtering Architecture for a Class of Nonlinear Systems”, IEEE Trans on Circuit and Systems—II: Express Brief, 51, 2004.
Dobbiaco Summer School - Dobbiaco (BZ) 22-26 June 2009 55
L3 – References
• Levant A., Bartolini G., Pisano A., Usai E., "A Real-Sliding Criterion for Control Adaptation", Advances in Variable Structure Systems – Theory and Applications. Proceedings of the 7th IEEE Int. Workshop on Variable Structure Sysstems (VSS 2002), Sarajevo, YUG, July 2002, A. Sabanovic ed., pp. 205-213, ETF Sarajevo, 2002
• Y.B. Shtessel, Y.-J. Lee,“New approach to chattering analysis in systems with sliding modes”, Proceedings of the 35th IEEE CDC, Kobe, Japan, 4014–4019.
• I. Boiko, Discontinuous Control Systems: Frequency-domain Analysis And Design, Birkhauser, Boston, 2008
• I. Boiko, L. Fridman, R. Iriarte, A. Pisano, E. Usai, “Parameter tuning of second-order sliding mode controllers for linear plants with dynamic actuators”, Automatica, 42, pp. 833–839, 2006.