2. Table of Content
SOLO
Sliding Mode Observers
2
Sliding Mode Observers
Sliding Mode Observer for a Linear Time Invariant (LTI) System
Generic Observer for a Linear Time Invariant (LTI) System
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
Sliding Mode Observers of Target Acceleration
References
3. SOLO
Sliding Mode Observers
In most of the Linear and Nonlinear Unknown Input Observers proposed so far,
the necessary and sufficient conditions for the construction of Observers is that the
Invariant Zeros of the System must lie in the open Left Half Complex Plane, and the
transfer Function Matrix between Unknown Inputs and Measurable Outputs satisfies
the Observer Matching Condition.
Observers are Dynamical Systems that are used to Estimate the State of a Plant using
its Input-Output Measurements; they were first proposed by Luenberger.
David G. Luenberger
Professor
Management Science
and Engineering
Stanford University
In some cases, the inputs of the System are unknown or partially
unknown, which led to the development of the so-called Unknown
Input Observers (UIO), first for Linear Systems. Motivated by the
development of Sliding-Mode Controllers, Sliding Mode UIOs
have been developed.
The main advantage of using Sliding-Mode Observer over their
Linear counterparts is that while in Sliding, they are Insensitive to
the Unknown Inputs and, moreover, they can be used to
Reconstruct Unknown Inputs which can be a combination of
System Disturbances, Faults or Nonlinearities.
3
Sliding Mode Observers
4. SOLO
4
Diagram of the LTI System
and a Sliding Mode Observer
pxnnxmnxmnxn
pmmn
CBBA
yuux
xCy
uBuBxAx
RRRR
RRRR
∈∈∈∈
∈∈∈∈
=
++=
,,,
,,,
21
21
21
21
2211
Assumptions:
( ) 222
222 ,,
mBCrankBrank
pmpCrankmBrank
==
≤==1
2
3
tu ∀≤ ρ2
Invariant Zero of the triple (A,B2,C)
are in the Open Left-Hand Complex
Plane, or equivalently
( ) 0Real
0
2
2
2
2
≥∀+=
−
smn
C
BAIs
rank
pxmpxn
nxmnxnn
Sliding Mode Observer for a Linear Time Invariant (LTI) System
Observer:
( ) ( )
2
,
ˆˆ
,ˆ,ˆˆˆ 211
mnxp
EL
xCy
yyEBuByyLxAx
RR ∈∈
=
−+−+= η
will stabilize the Observer on the Sliding Surface
nxp
L R∈ ( ) 0ˆ =− yyF
( )
( )
( )
( )
( )
xpm
F
yyFif
yyFif
yyF
yyF
yyE
2
,0
0ˆ0
0ˆ
ˆ
ˆ
,ˆ,
RR ∈∈<
=−
≠−
−
−
=
η
η
η
Reaching the Sliding Surface
using:
( ) 0ˆ =− yyF
Sliding Mode Observers
LTI System:
5. L is obtain by choosing a S.P.D. matrix Q and solving for the S.P.D. matrix P defined
by the Lyapunov Algebraic Equation (A.R.E.)
SOLO
5
pxnnxmnxmnxn
pmmn
CBBA
yuux
xCy
uBuBxAx
RRRR
RRRR
∈∈∈∈
∈∈∈∈
=
++=
,,,
,,,
21
21
21
21
2211
Sliding Mode Observer for a Linear Time Invariant (LTI) System
Observer:
( ) ( )
2
,
ˆˆ
,ˆ,ˆˆˆ 211
mnxp
EL
xCy
yyEBuByyLxAx
RR ∈∈
=
−+−+= η
will stabilize the Observer on the Sliding Surface
nxp
L R∈ ( ) 0ˆ =− yyF
( )
( )
( )
( )
( )
xpm
F
yyFif
yyFif
yyF
yyF
yyE 2
,0
0ˆ0
0ˆ
ˆ
ˆ
,ˆ, RR ∈∈<
=−
≠−
−
−
= η
η
η
Reaching the Sliding Surface
using:
( ) 0ˆ =− yyF
On the Sliding Surface ( ) 11
ˆˆˆ uByLxCLAx ++−=
( ) ( ) 02 <−=−+− QCLAPPCLA
T
F is obtain by solving
Sliding Mode Observers
( ) 1
22
−
=⇒= TTTT
CCCPBFPBCF
Diagram of the LTI System
and a Sliding Mode Observer
L.T.I. System:
6. SOLO
Generic Observer for a Linear Time Invariant (LTI) System
pxmpxnnxmnxn
pmn
DCBA
yux
uDxCy
uBxAx
RRRR
RRR
∈∈∈∈
∈∈∈
+=
+=
,,,
,,
Observer
sxpsxmsxq
qxpqxmqxq
sxnsq
RSM
GJF
Lwz
xLw
yRuSzMw
yGuJzFz
RRR
RRR
RRR
∈∈∈
∈∈∈
∈∈∈
→
++=
++=
,,
,,
,,
A Necessary Condition for obtaining an Observer is that (A,C) is Observable.
The Observer will achieve if and only ifxLw →
=+
=+
−=
=−
−
0DRS
LTGCR
DGBTJ
CGTFAT
valueseigenstablehasF
where Tnxn is a Transformation Matrix.
L.T.I. System
Sliding Mode Observers
8. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
nxqpxnnxmnxn
qpmn
ECBA
dyux
xCy
dEuBxAx
RRRR
RRRR
∈∈∈∈
∈∈∈∈
=
++=
,,,
,,,
Observer
DesignedbetoMatricesHKTF
xz
yHzx
yKuBTzFz
nxpnxpnxnnxn
nn
RRRR
RR
∈∈∈∈
∈∈
+=
++=
,,,
ˆ,
ˆ
( ) ( ) ( ) ( ) ( ) dECHIuBCHIxACHIyKxCKuBTzFxCHIze nnn
KKK
n −−−−−−+++=−−=
=+
21
21
The Estimator Error ( ) xCHIzxxe n −−=−= ˆ:
An Unknown Input Observer for an
LTI System will derive its State Error
regardless of the unknown input (disturbance)
0ˆ:
asymptotic
xxe →−=
( )td
( ) ( )[ ] ( )[ ] ( )[ ] ( ) dECHIuBCHITyHCKACHAKzCKACHAFeCKACHAe nn −−−−+−−−+−−−+−−= 1211
eyHzx −+=Substitute in this equation:
L.T.I. System
Sliding Mode Observers
9. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
nxqpxnnxmnxn
qpmn
ECBA
dyux
xCy
dEuBxAx
RRRR
RRRR
∈∈∈∈
∈∈∈∈
=
++=
,,,
,,,
Observer
An Unknown Input Observer for an
LTI System will derive its State Error
regardless of the unknown input (disturbance)
0ˆ:
asymptotic
xxe →−=
( )td
( ) ( )[ ] ( )[ ] ( )[ ] ( ) dECHIuBCHITyHCKACHAKzCKACHAFeCKACHAe nn −−−−+−−−+−−−+−−= 1211
We can see that if we can make the following relations:
( )
HFK
CKACHAF
CHIT
ECHI
n
n
=
−−=
−=
=−
2
1
0
the State-Estimator Error will be: eFe =
We can see that the Observer Error will be zero asymptotically iff all the eigenvalues of
F are stable.
L.T.I. System
Sliding Mode Observers
10. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
Observer
Lemma 1: This Equation is solvable if:
( )
HFK
CKACHAF
CHIT
ECHI
n
n
=
−−=
−=
=−
2
1
0
but:
and a special solution for H is: ( ) ( )[ ] ( ) ( )†1
ECEECECECEH
TT
nxq
sp
nxp ==
−
Proof of Lemma 1:
nxqpxnnxpnxq ECHE = ( ) ( )nxqpxnnxq ECrankErank ≤⇒
( ) ( ) ( )[ ] ( )nxqnxqpxnnxqpxn ErankErankCrankECrank ≤≤ ,min
( ) ( ) qErankECrank nxqnxqpxn ==
Necesity
Sufficiency When rank (CE) = rank (E), (CE) is a full column rank matrix, because
E is assumed a full column rank matrix, and a left inverse of (CE) exists.
( ) ( ) ( )[ ] ( )TT
ECECECEC
1† −
=
and: ( )†
ECEH sp
nxp =
q.e.d.
L.T.I. System
( ) qErankECrank nxqnxqpxn == Observer Matching Condition
Sliding Mode Observers
11. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
Observer
Lemma 2: Let:
If s1 ϵ C is an unobservable mode of the pair (C1,A) then:
=
AC
C
C1
Then the Detectability for the pair (C,A) is equivalent to
that of (C1,A). (A pair (C,A) is Detectable when all the
unobservable modes of this pair are Stable).
Proof of Lemma 2:
n
AC
C
AIs
rank
C
AIs
rank
n
n
<
−
=
−
1
1
1
That means that exists a vector α ϵ Cn
such that:
n
C
AIs
rank
C
AIs
AC
C
AIs
nn
n
<
−
⇒=
−
⇒=
−
11
1
00 αα
s1 is also an unobservable mode of the pair (C,A).
If s2 ϵ C is an unobservable mode of the pair (C,A) then: n
C
AIs
rank n
<
−2
That means that always exists a vector β ϵ Cn
such that: 02
=
−
β
C
AIs n
s2 is also an unobservable mode of the pair (C1,A).
( )
00
0
0
1
1
1
22
2
=
−
=
−
⇒===⇒
=
=−
βββββ
β
β
C
AIs
AC
C
AIs
CssCAC
C
AIs n
n
n
q.e.d.
L.T.I. System
Sliding Mode Observers
12. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
=
++=
xCy
dEuBxAx
U.I.O. Observer
+=
++=
yHzx
yKuBTzFz
ˆ
Lemma 3: Necessary and Sufficient Conditions to have an
U.I.O. Observer for the L.T.I. System are:
(1)rank (C E) = rank (E)
(2)(C, A1) is a Detectable pair, where ( ) ( ) ( )[ ] ( ) ACECECECEAACECEAA
TT 1†
1 :
−
−=−=
The condition (2) is equivalent to the condition that the Invariant Zeros for the
Unknown Input, i.e., of the triplet (A,E,C) must be stable:
qpnsqn
C
EAIs
rank
pxqpxn
nxqnxnn
≥≥∈∀+=
−
−C
0
Proof of Lemma 3 (Sufficiency):
According to Lemma 1 if rank (C E)= rank (E) exists a solution for H:
( ) ( )[ ] ( ) ( )†1
ECEECECECEH
TT
nxq
sp
nxp ==
−
and: ( ) ( )[ ]( ) CKACKACECECECEACKACHAF
TT
pxnnxnpxn
sp
nxpnxnnxn nxp 1111 −=−−=−−=
We can see that F may be Stabilized by choosing a proper K1, only if the pair (C, A1)
is Detectable.
L.T.I. System
Sliding Mode Observers
13. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
=
++=
xCy
dEuBxAxL.T.I. System
U.I.O. Observer
+=
++=
yHzx
yKuBTzFz
ˆ
Lemma 3: Necessary and Sufficient Conditions to have an
U.I.O. Observer for the L.T.I. System are:
(1)rank (C E) = rank (E)
(2)(C, A1) is a Detectable pair, where ( ) ( ) ( )[ ] ( ) ACECECECEAACECEAA
TT 1†
1 :
−
−=−=
Proof of Lemma 3 (Necessity): ( )
HFK
CKACHAF
CHIT
ECHI
n
n
=
−−=
−=
=−
2
1
0
A General Solution for H
is
( ) ( )( )[ ]†
0
†
ECECIHECEH mnxp −+=
where is an arbitrary matrix andnxm
H R∈0
( ) ( ) ( )[ ] ( )TT
ECECECEC
1† −
=
Since the Observer is a U.I.O. Observer for the L.T.I. System
we can solve for H, T, K1, F and K2
( )[ ] [ ]
( )[ ] [ ] 111
1
011011
1
1
CKA
AC
C
HKA
ACECECI
C
HKACECEICKACHAF
C
K
T
m
T
n −=
−=
−
−−=−−=
Since the Matrix F is Stable the pair is Detectable, therefore the pair (C, A1)
is also detectable, according to Lemma 2.
( )11, AC
q.e.d.
Sliding Mode Observers
14. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
=
++=
xCy
dEuBxAxL.T.I. System
U.I.O. Observer
+=
++=
yHzx
yKuBTzFz
ˆ
Lemma 3: Necessary and Sufficient Conditions to have an
U.I.O. Observer for the L.T.I. System are:
(1)rank (C E) = rank (E)
(2)(C, A1) is a Detectable pair, where ( ) ( ) ( )[ ] ( ) ACECECECEAACECEAA
TT 1†
1 :
−
−=−=
Proof of Lemma 3 (Necessity) (continue):
q.e.d.
( ) ( )
( ) ( )
−∈∀+=
−
−
−
=
−
Csqn
C
EAIs
ECEsCECE
I
ECEsCECEI
rank
C
EAIs
rank
pxqpxn
nxqnxnn
p
n
pxqpxn
nxqnxnn
0
0
0
††
††
( )
( ) ( )
( ) −∈∀+
−
−
=
−
+−
=
−
CsErank
CAECE
C
AIs
rank
ECAECE
C
ACECEAIs
rank
C
EAIs
rank
q
n
nn
pxqpxn
nxqnxnn
nxn
†
1
†
†
0
0
0
The condition that the pair (C, A1) is detectable, is equivalent to
therefore equivalent to the Invariant Zeros of the triplet (A,E,C) being stable
−∈∀=
−
Csn
C
AIs
rank nxnn 1
The Condition that the Invariant Zeros of the triplet (A,E,C) are stable is:
Sliding Mode Observers
15. SOLO
Unknown Input Observer (UIO) for a Linear Time Invariant (LTI) System
=
++=
xCy
dEuBxAxL.T.I. System
U.I.O. Observer
Design Procedure
+=
++=
yHzx
yKuBTzFz
ˆ
1 Check if rank (CE)=rank(E). If rank (CE)≠rank(E) go to .10
2 Compute: ( ) ( )[ ] ( ) ATACHTECECECEH pxnnxpnxn
TT
nxq
sp
nxp ===
−
1
1
,,
3 If (C1, A) Observable a U.I.O. exists, and K1 can be computed using Pole
Placements or any other Method. Go to 9
T
n
T
pp 1
,,1
4 Construct a Transformation Matrix P by choosing n1=rank (WO) (where
WO=[C, CA1,…,CA1
n-1
]) row vectors , together othe n-n1 row vectors
to construct the nonsingular
T
n
T
n pp ,,11
+ [ ]T
n
T
n
T
n
T
ppppP ,,11 11
+=
5 Perform [ ]0*
0 1
2221
111
1 CPC
AA
A
PAP =
= −−
6 Check Detectability of (C,A1). If one of eigenvalues of A22 is unstable, a
U.I.O. doesn’t exist and go to
10
7 Select n1 eigenvalues and assign them to using Pole Placement.*1
11 CKA p−
8 Compute where is any (n-n1)xn matrix.( ) ( )[ ]TT
p
T
pp KKPKPK 2111
1 −−
== 2
pK
9 Compute HFKKKKCKAF +=+=−= 12111 ,
10 Stop
Sliding Mode Observers
16. 16
Sliding Mode Observers of Target Acceleration
Kinematics: ( )
→→
⋅−⋅+Λ−=Λ tataRR
td
d
MT 11
We want to Observe (Estimate) the Unknown Target Acceleration Component:
→
⋅ taT 1
Define: 0:1_ vtaestAt
Est
T =
⋅=
→
( ) mEstEst AvRz
td
d
−+Λ−= 00
The Differential Equation of the Observer will be a copy of the kinematics:
mM Ata =
⋅
→
:1
Define the Observer Error: EstEstO Rz Λ−=
0:σ
Define the Sliding Mode Observers that must drive σO→0:
( )
( )
( )
22
11
1232
201
2/1
0121
1
3/2
10
1.1
5.1
2
vz
vz
vzsigntv
zvzsignvztv
zsigntv OO
=
=
−⋅⋅=
+−−⋅⋅=
+⋅⋅⋅=
σσ
Missile Command Acceleration
( ) ( )
( )
estAtdRsignRRsignRRNa
EstEstRSM
EstEstEstEstEstEstEstEstEstEstC _'
2
3/1
2
2/1
1 +ΛΛ+ΛΛ+Λ−=
Λ
∫
µαα
t1, t2, t3 are Design Parameters
Observer 4: Variation of 1
( )
( )
( )
22
11
122
201
2/1
01
2/1
1
1
3/23/1
0
1.1
5.1
2
vz
vz
vzsignLv
zvzsignvzLv
zsignLv OO
=
=
−⋅⋅=
+−−⋅⋅=
+⋅⋅⋅=
σσ
Observer 1:
( )
( )
02
11
3/1
21
1
2/1
10
=
=
⋅⋅−=
+⋅⋅−=
z
vz
signv
zsignv
OOO
OOO
σσα
σσα ( )
( )
02
11
21
1
2/1
10
=
=
⋅−=
+⋅⋅−=
z
vz
signv
zsignv
OO
OOO
σρ
σσρ
L is a Design Parameter are Design Parameters21, OO αα
are Design Parameters21, OO ρρ
Observer 2: Observer 3:
Sliding Mode Observers
23. References
SOLO
O’Reilly, J., “Observers for Linear Systems”, Academic Press, 1983
23
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”,
Kluwer Academic Publishers, 1999
Sliding Mode Observers
24. 24
SOLO
Technion
Israeli Institute of Technology
1964 – 1968 BSc EE
1968 – 1971 MSc EE
Israeli Air Force
1970 – 1974
RAFAEL
Israeli Armament Development Authority
1974 –2013
Stanford University
1983 – 1986 PhD AA
Notes de l'éditeur
Zak, S.H., Hui, S., “Output Feedback Variable Structure Controllers and Stste Estimators for Uncertain Dynamic Systems”, TR-EE 91-46, November 1991
Floquet,T., Edwards, C., Spurgeon, S.,K., “On Sliding Mode Observers for Systems with Unknown Inputs”, Int. J. Adapt. Control Signal Process, vol. 21, pp. 638-656, 2007
Kalsi, K., Lian, J., Hui, S., Zak, S.,H., “Sliding-Mode Observers for Systems With Unknown Inputs”, August 22, 2008, Draft
Zak, S.,H., Hui, S., “Output Feedback Variable Structure Controllers and State Estimators for Uncertain Dynamic Systems”, TR-EE 91-46, November 1991
Kalsi, K., Lian, J., Hui, S., Zak, S.,H., “Sliding-Mode Observers for Systems with Unknown Inputs”
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999
Chen, J., Patton, R., J., “Robust Model-Based Fault Diagnosis for Dynamic Systems”, Kluwer Academic Publishers, 1999