Sei sulla pagina 1di 6

An LPV Pole-placement approach to friction compensation as an FTC

problem
Lejun Chen, Ron Patton IEEE Fellow and Supat Klinkhieo
AbstractThe concept of combining robust fault estimation
within a controller system to achieve active Fault Tolerant
Control (FTC) has been the subject of considerable interest
in the recent literature. The current study is motivated by the
need to develop model-based FTC schemes for systems that
have no unique equilbria and are therefore difcult to linearise.
Linear Parameter Varying (LPV) strategies are well suited
to model-based control and fault estimation for such systems.
This contribution involves pole-placement within suitable LMI
regions, guaranteeing both stability and performance of a multi-
fault LPV estimator employed within an FTC structure. The
proposed design strategy is illustrated using a non-linear two-
link manipulator system with friction forces acting simulta-
neously at each joint. The friction forces, considered as a
special case of actuator faults, are estimated and their effect
compensated within a polytope controller system, yielding a
robust form of active FTC that is easy to apply to real robot
systems.
I. INTRODUCTION
Friction phenomena are widely encountered in mechan-
ical/mechatronic systems and the problem of control of
systems with friction presents interesting challenges. In
recent years there has been a substantial literature on this
subject focussed on modelling of friction dynamics [1]. From
a control point of view, friction compensation strategies
that require detailed models of friction characteristics have
limitations arising from the representation of non-smooth
non-linearity and the fact that friction modelling remains an
imprecise subject, thereby resulting in difcult robustness
issues.
Friction forces degrade the performance of a mechanical
system and the concept of compensating their effect in a
closed-loop system can be viewed as an FTC problem in
which the friction forces actually act on the system as com-
ponent faults [2]. In [2],the sliding mode observer theory is
combined with a sliding mode controller to achieve a robust
method of friction force estimation and FTC compensation
for an inverted pendulum system. In a similar way to [2]
a recent investigation [3] based on a two-link manipulator
study, has shown that the fault estimation can be achieved
using an LPV estimation approach, based on a combination
of LPV estimation and LPV control within an active or
direct FTC framework. In [3] simple perturbation faults
are considered, whilst the current paper revisits the friction
compensation problem for the same manipulator system.
The LPV modelling methodology has been widely adopted
in control system design especially related to vehicle and
L Chen and R J Patton (r.j.patton@hull.ac.uk) are with the Department
of Engineering, University of Hull, Cottingham, Hull, HU6 7RX, UK and
S Klinkhieo is with the Synchrotron Light Research Institute, Thailand
aerospace control [4], [5]. The essential idea is that a time-
varying system can be represented via an LPV model system,
for example using a polytopic structure with the control
for each vertex system computed via efcient interior-point
algorithms and LMIs. LPV modelling can also be used to
design residual-based fault detection and diagnosis (FDD)
estimators that provide robustness to modelling uncertainty
[6], [7]. FTC studies based on the LPV concept have also
focussed on the use of FDD residuals and system recong-
uration [5].
The current paper extends the theoretical approach in [8]
by (a) providing a proof for the LPV estimator stability,
via pole-placement design in an LMI region, (b) including
the use of an LPV stabilizing controller and (c) introducing
simultaneously acting friction forces based on the two-
link manipulator model. Section II outlines the theoretical
foundations of LPV estimator design and introduces the
LMI-based pole-placement design utilised to guarantee the
performance and the stability of the fault estimator is deter-
mined efciently via the solution of a set of LMIs. Section
III describes the two-link manipulator case study example.
The Stribeck friction model is outlined prior to constructing
the corresponding active LPV FTC scheme. The LMI-based
LPV pole-placement strategy is developed from the time-
varying joint angle measurements and the estimated joint
friction forces are used in an active FTC compensation
mechanism in each control signal, based on the concept of
fault effect factors, rst dened in [9]. Section IV gives a
brief conclusion to the work.
II. THEORETICAL FOUNDATION OF ROBUST LPV
ESTIMATOR
The LPV system with faults is described as follows:
x
p
(t) = A
p
()x
p
(t) +B
p
()u(t) +E
p
()d(t) +F
p
()f(t)
y
p
(t) = C
p
()x
p
(t) +D
p
()u(t) +G
p
()d(t) +H
p
()f(t)
(1)
where, x
p
(t) R
n
, u(t) R
r
, y
p
(t) R
m
and d(t)
R
q
are the states, control inputs, outputs, and disturbances.
f(t) R
g
is the fault vector. R
s
is a varying parameter
vector, and A
p
(), B
p
(), C
p
(), D
p
(), E
p
(), F
p
(),
G
p
() and H
p
() are matrices with appropriate dimensions.
The assumptions that apply to system (1) are [8] that the:
(A1) System (1) is stable.
(A2) The vector (t) varies in a polytope with vertices
2010 Conference on Control and Fault Tolerant Systems
Nice, France, October 6-8, 2010
WeA4.4
978-1-4244-8154-5/10/$26.00 2010 IEEE 100
w
udf
(t) e
P()
F()
_
u
y
p
_

f
Fig. 1. Polytopic LPV Estimation System Structure

1
,
2
, . . . ,
j
(j = 2
s
), i.e,:
(t) : = Co{
1
,
2
, . . . ,
j
}
= {
j

i=1
a
i

i
: a
i
0,
j

i=1
= 1}
(2)
(A3) As the state-space matrices depend afnely on (t),
system (1) is assumed to be polytopic, i.e.
_
A
p
() B
p
() E
p
() F
p
()
C
p
() D
p
() G
p
() H
p
()
_

Co
_

_
_
A
p
() B
p
() E
p
() F
p
()
C
p
() D
p
() G
p
() H
p
()
_
i = 1, . . . , j
_

_
(3)
(A4) C
p
(), D
p
(), G
p
(), and H
p
() are parameter inde-
pendent, i.e.
C
p
() = C
p
, D
p
() = D
p
, G
p
() = G
p
, H
p
() = H
p
(4)
The design of a polytopic estimator can be written as
x
f
(t) = A
f
()x
f
(t) +B
f
()
_
u(t)
y
p
(t)
_

f(t) = C
f
()x
f
(t) +D
f
()
_
u(t)
y
p
(t)
_ (5)
Therefore, the estimated error vector e
f
(t) =

f(t) f(t)
R
g
, is minimised. Here u(t) and y
p
(t) are dened in (1), and
x
f
(t) R
n
is the state vector of the estimator,

f(t) is the
estimation of the fault f(t). A
f
(), B
f
(), C
f
() and D
f
()
are matrices with appropriate dimensions, to be designed.
The LPV estimator (5) can be rewritten as:
F():=
_
A
f
() B
f
()
C
f
() D
f
()
_
Co
_

_
_
A
f
(
i
) B
f
(
i
)
C
f
(
i
) D
f
(
i
)
_
i = 1, . . . , j
_

_
(6)
The system structure is shown in Figure 1.
Dene x
pf
(t) and w
udf
(t) to be [x
p
x
f
]
T
and
[u(t) d(t) f(t)]
T
respectively. Rewrite (1) as
x
p
= Ax
p
+B
1
w
udf
+B
2

f (7)
e = C
1
x
p
+D
11
w
udf
+D
12

f (8)
y
p
= C
2
x
p
+D
12
w
udf
+D
22

f (9)
Above system in Figure 1 can then be expressed as
x
pf
= A()x
pf
(t) +B()w
udf
(t)
e
f
= C()x
pf
(t) +D()w
udf
(t)
(10)
where,
A() = A
0
+BF()C B() = B
0
+BF()D
21
C() = D
12
F()C D() = D
11
+D
12
F()D
21
A
0
=
_
A 0
0 0
_
B
0
=
_
B
1
0
_
B =
_
0 B
2
I 0
_
C =
_
0 I
C
2
0
_
D
12
=
_
0 D
12

D
21
=
_
0
D
21
_
The denition of the LMI region and the existence theorem
for assignable poles inside this region are given as follows:
Denition (LMI Region): A subset D of the complex plane
is called an LMI region if there exist a symmetric matrix
= [
kl
] and a matrix = [
kl
] such that
D = {z C : f
D
< 0} (11)
with
f
D
:= +z +z
T
= [
kl
+
kl
z +
lk
z]
l<k,l<m
(12)
Theorem 0.1: [10] The matrix A is D-stable if and only
if there exists a symmetric matrix X such that
M
D
(A, X) < 0, X > 0 (13)
where
M
D
(A, X) = X + (AX) +
T
(AX)
T
= [
kl
X +
kl
AX +
lk
XA
T
]
l<k,l<m
(14)
Here, Problem 1 is dened to solve the estimation problem
with pole placement.
Problem 1 For the LPV system (1) with assumptions (A1)-
(A4), design a polytopic LPV estimator (5), such that the L
2
-
induced norm of the operator mapping w
udf
(t) into e
f
(t) is
bounded by a scalar number for all parameter trajectories
(t) in the polytope . Also, all poles of the LPV system
(10) lie in the given LMI region D.
Considering the structure of (10) and according to the
assumptions (A2)-(A4), it can be veried that the system (10)
is polytopic, and the Lemma 1 can be used as an adaptation
of the results from [8].
Lemma 1 For LPV system (10), the following statements
are equivalent:
(1) L
2
-induced norm of the operator mapping w
udf
(t) into
e
f
(t) is bounded by a scalar number for all parameter
trajectories (t) in the polytope .
(2) Parameter trajectories (t) in the polytope . There
exists X = X
T
> 0 satisfying the system of LMIs:
_

_
XA(
i
) +A
T
(
i
)X XB(
i
) C
T
(
i
)
B
T
(
i
)X I D
T
(
i
)
C(
i
) D(
i
) I
_

_
< 0 (15)
101
where,i = 1, . . . , j. The main result of this section is stated
in Theorem 0.2, which provides the solution of the Problem
1.
Theorem 0.2: Consider the LPV system in (1) with as-
sumption (A1)-(A3). Let N
R
= [N
T
1
N
T
2
]
T
and N
S
=
[V
T
1
V
T
2
]
T
denote bases of the null spaces of [B
T
2
D
T
12
]
and [C
2
D
21
] respectively. There exists a polytopic LPV
estimator that can determine the solution of Problem 1 if
matrices 0 < R = R
T
R
nn
, 0 < S = S
T
R
nn
can
be found that:

N
R
0
0 I

A
p
(
i
)R +RA
T
p
(
i
) 0 B
1
(
i
)
0 I D
11
B
T
1
(
i
) D
T
11
I

N
R
0
0 I

<0
(16)

N
S
0
0 I

SA
p
(
i
) +A
T
p
(
i
)S SB
1
(
i
) 0
B
T
1
(
i
)S I D
T
11
0 D
11
I

N
S
0
0 I

<0
(17)
U
T
1
(
kl
S +
kl
A
p
(
i
)S +
lk
SA
T
p
(
i
))U
1
< 0 (18)
V
T
1
(
kl
R +
kl
RA
p
(
i
) +
lk
A
T
p
(
i
)R)V
1
< 0 (19)
_
R I
I S
_
0 i = 1, . . . , j (20)
Proof: Based upon Theorem 0.1, the poles of the LPV
system (10) lie in LMI region D if and only if there exists
X, such that,
M
D
(A(), X) = [
kl
X +
kl
A()X +
lk
XA
T
()] < 0
(21)
By Lemma 1, Theorem 0.1 and considering the notations
in (5-14), there exists a polytopic LPV fault estimator (5)
which solves the Problem 1 if:
(
i
) +U
T
x
F(
i
)V +V
T
F
T
(
i
)U
x
< 0 (22)
(
i
) +P
T
F(
i
)Q

+Q

T
F
T
(
i
)P < 0 (23)
where,
(
i
) =
_

_
XA
0
(
i
) +A
T
0
(
i
)X XB(
i
) 0
B
T
(
i
)X I D
T
11
0 D
11
I
_

_
(24)
(
i
) =
kl
X +
kl
A
0
X +
lk
XA
T
0
(25)
U
x
= [B
T
X 0 D
T
12
] (26)
V = [C D
21
0] (27)
P =
lk
B
T
, Q

= CX (28)
Based on the projection lemma, the LMIs of (22) and (23)
hold for some F(
i
) if and only if:
W
T
U
x
(
i
)W
U
x
< 0 (29)
W
T
V
(
i
)W
V
< 0 (30)
W
T
P
(
i
)W
P
< 0 (31)
W
T
Q

(
i
)W
Q

< 0 (32)
where, W
U
x
, W
V
, W
P
and W
Q

denote any bases of the


null spaces of U
x
, V , P and Q

, respectively.
Observing that:
U
x
=
_
B
T
0 D
T
12

_
X 0 0
0 I 0
0 0 I
_

_ (33)
U =
_
B
T
0 D
T
12

(34)
Q = C (35)
The basis for the null space of U
x
and Q

is given by:
W
U
x
=
_

_
X
1
0 0
0 I 0
0 0 I
_

_W
U
(36)
W
Q

= X
1
W
Q
(37)
where, W
U
and W
Q
denote any basis of the null space of U
and Q. Hence, the inequalities (29) and (32) can be rewritten
as:
W
T
U
(
i
)W
U
< 0 (38)
W
T
Q
(
i
)W
Q
< 0 (39)
with
(
i
) =
_

_
A
0
(
i
)X
1
+X
1
A
T
0
(
i
) B
0
(
i
) 0
B
T
0
(
i
) I D
T
11
0 D
11
I
_

_
(40)
(
i
) = (X
1
)
T

kl
+ (X
1
)
T

kl
A
0
+
lk
A
T
0
X
1
(41)
X and X
1
can be partitioned as:
X =
_
S N
N
T

_
X
1
=
_
R M
M
T

_
(42)
where S, R, M, N R
nn
and S, R, M > 0, and stands
for the matrix entries which are not of interest.
Then, X and X
1
are substituted into (24), (25), (40) and
(41), yielding (
i
), (
i
), (
i
) and (
i
).
(
i
) =

SA
p
(
i
) +A
T
p
(
i
)S A
T
p
(
i
)N SB
1
(
i
) 0
N
T
A
p
(
i
) 0 N
T
B
1
(
i
) 0
B
T
1
(
i
) B
T
1
(
i
)N I D
T
11
0 0 D
11
I

(43)
(
i
) =

A
p
(
i
)R +RA
T
p
(
i
) A
p
(
i
)M B
1
(
i
) 0
M
T
A
T
p
(
i
) 0 0 0
B
T
1
(
i
)S 0 I D
T
11
0 0 D
11
I

(44)
(
i
) =
kl

S N
N
T

+
kl

A
p
(
i
)S A
p
(
i
)N
0 0

+
lk

SA
T
p
(
i
) 0
N
T
A
T
p
(
i
) 0

(45)
102
(
i
) =
kl

R M
M
T

+
kl

RA
p
(
i
) 0
M
T
A
p
(
i
) 0

+
lk

A
T
p
(
i
)R A
T
p
(
i
)M
0 0

(46)
Let N
R
=
_
U
1
U
2
_
and N
S
=
_
V
1
V
2
_
denote bases of the
null spaces of [B
T
2
D
T
12
] and [C
2
D
21
] respectively, where,
U
1
and V
1
span the null spaces of B
T
2
and C
2
. Then, the
bases of the null spaces of U, V , P and Q are given by:
W
U
=
_

_
U
1
0
0 0
0 I
U
2
0
_

_
, W
V
=
_

_
V
1
0
0 0
0 I
V
2
0
_

_
(47)
W
P
=
_
U
1
0
0 0
_
, W
Q
=
_
V
1
0
0 0
_
(48)
Substituting (47), (48), (43), (44), (45) and (46) into (30),
(31), (38) and (39), to yield

N
R
0
0 I

A
p
(
i
)R +RA
T
p
(
i
) 0 B
1
(
i
)
0 I D
11
B
T
1
(
i
) D
T
11
I

N
R
0
0 I

< 0
(49)

N
S
0
0 I

SA
p
(
i
) +A
T
p
(
i
)S SB
1
(
i
) 0
B
T
1
(
i
)S I D
T
11
0 D
11
I

N
S
0
0 I

< 0
(50)
U
T
1
(
kl
S +
kl
A
p
(
i
)S +
lk
SA
T
p
(
i
))U
1
< 0 (51)
V
T
1
(
kl
R +
kl
RA
p
(
i
) +
lk
A
T
p
(
i
)R)V
1
< 0 (52)
Based on the matrix completion result, the condition X >
0 is equivalent to:
_
R I
I S
_
0 (53)
which completes the proof of Theorem 0.2.
Once the matrices R and S are obtained, the LPV estimator
can be constructed as following Algorithm 1:
Step1. Computing the full rank matrices M, N using SVD
such that:
MN
T
= I RS (54)
Step2. Computing X as the unique solution of the linear
matrix equation:
X
_
I R
0 M
T
_
=
_
S I
N
T
0
_
T
(55)
Step3.Compute F(
i
) by solving (22) and (23).
Step4.Solve the polytopic LPV estimator:
F() =
j

i=1
a
i
p
F(
i
) (56)
where a
i
p
is any solution of the convex decomposition
problem:
=
j

i=1
a
i
p

i
(57)
III. TWO-LINK MANIPULATOR CASE STUDY
A. Robust LPV Fault Estimator for Two-Link Manipulator
A two-link manipulator demonstrator system with actuator
faults is illustrated. The actuator faults are considered as the
friction forces acting at the joints of system, changing in
terms of the angular velocity of the joints. Four types of
dynamic torques that arise from the motion of the manipula-
tor: Inertial, Centripetal, Coriolis and Friction torques. The
model is shown in Fig.2. The position of the two arms are
dened by the joint angles = [
1

2
]
T
. The system inputs
are u = [u
1
u
2
]
T
, which are the torques actinging onto the
joints.
Load
m g 2
m g 1
l
c
2
lc
1
l1
fric2
u2
fric1
u1
g
1
2
.
.
Fig. 2. Two-link Manipulator Structure
The dynamic of the manipulator [11], [12] is shown as
()

+O(,

)

+g() = u fric(

) (58)
where, () R
22
is the manipulator inertia tensor matrix,
O(,

)

R
2
is the function containing the Centripetal and
Coriolis torques. g() R
2
and fric() the gravitational
and friction torques,respectively. The equations of motion
and the physical system parameters are described in [3].
The friction forces acting on the joints is described by the
discontinuous Stribeck friction model [13].
f
fric
= g(x
p
)Sign(x
p
) (59)
where, Sign(x
p
) =
_

_
1 if x
p
< 0
[1, 1] if x
p
= 0
1 if x
p
> 0
and g(x
p
) = F
c
+(F
s
F
c
)exp(|x
p
|/v
s
)

is the Stribeck
friction function with F
c
and F
s
are the Coulomb and static
friction levels, respectively and v
s
, > 0 are the Stribeck
velocity and shaping parameters, respectively. In the friction
simulation, the following parameter values are used: F
c
=
5N, v
s
= 0.15ms
1
, F
s
= 2.5N and
s
= 1.22. Here, the
quadratic terms O(,

) are not considered since they are
not bounded. (58) can be simplied as shown in (60).
()

+g() = u fric(

) (60)
where,
() =
_
m
1
lc
2
1
+m
2
lc
2
1
+I
1
m
2
l
1
lc
2
cos(
1

2
)
m
2
l
1
lc
2
cos(
1

2
) m
2
lc
2
2
+I
2
_
103
g() =
_
(m
1
lc
1
+m
2
l
1
)g sin(
1
)
m
2
glc
2
sin(
2
)
_
The nonlinear term in () is clearly a bounded function

1
() = cos(
1

2
) [1 1] (61)
Hence, () can be represented by a polytope whose
vertices are dened by:
() Co{
1

2
} (62)
To construct a state-space formulation, the vector eld
gg() with R
2
can be arranged in the form of G
g
()
and function
2
() can now be dened which is bounded.
sin(
1
) = (
sin(
1
)

1
)
1
=
2
()
1
(63)
where, 0.2
2
1.
From the bound of function
2
() in terms of the angle
, G
g
() is considered as a polytope as follows:
G
g
() Co{G
g
1
, G
g
2
, G
g
3
, G
g
4
} (64)
To dene the state space representation of the two-link
system, let: x(t) = [
1
(t)
2
(t)

1
(t)

2
(t)]

.
The LMI constraints with nonlinear system dynamics are
then given by the following descriptor system:
_
I 0
0 ()
_
x(t) =
_
0 I
G
g
() 0
_
x(t)+W
b
u(t) (65)
The state space equation can be expressed as follows:
x(t) = A()x(t) +B()u(t) (66)
where, A() =
1
_
0 I
G
g
() 0
_
and B() =
1
W
b
.
With the faults,
x(t) = A()x(t) +B()u(t) +F
a
f
a
(t)
= A
ij
x(t) +B
i
u(t) +F
a
f
a
(t)
(67)
where, F
a
is fault distribution matrix and f
a
represents
actuator faults which represent the friction forces acting on
each joint. The actuator fault estimate

f
a
(t) in system (67)
can be implemented by using Algorithm1.
The open-loop two-link manipulator system is unstable.
Therefore, a stable closed-loop system has to be congured
to satisfy (A1). In this case, a common constant gain matrix
K is build to stabilise the fault-free open-loop system on
each vertex. Let L
c
= KS
c
, then the quadratic stability
conditions for 8-vertex system can be written as:
A
ij
S
c
+B
i
L
c
+S
c
A
T
ij
+L
T
c
B
T
i
0 (68)
The gain matrix can be calculated based upon solving
above LMIs.
The desired LMI region D is forced to lie in the half-plane
x < 0.7. Also, choosing to be 2.7550, after 15 iterations,
the estimator F
i
() is calculated, corresponding to each ver-
tex. The LPV estimator is then built through the combination
of F
i
(), i = 1, 2, . . . , 8. Fig.3 shows the results of the
friction estimation, with a Gaussian random disturbance d(t)
of zero-mean and variance 0.02. Simulation results show that
robust LPV fault estimator provides a good online estimation
performance for both friction forces simultaneously. The blue
curve represents the fault signal and the red curve depicts
the estimated fault signal. However, the estimation errors
are existed since the two faults affect the system dynamics
simultaneously. The estimation performance for each fault is
affected by the existence of another fault.
10 15 20 25 30 35 40
10
0
10
Time [s]
F
r
i
c
t
i
o
n

F
o
r
c
e

1

[
N
]
10 15 20 25 30 35 40
10
0
10
Time [s]
F
r
i
c
t
i
o
n

F
o
r
c
e

2

[
N
]
Fig. 3. Fault Estimation provided by the Polytopic LPV Estimator
B. Active LPV FTC Scheme
The dynamic system of in (67) includes an additive term
of the actuator faults. However, the faults can have a multi-
plicative effect in the system representation. A multiplicative
actuator fault representation can be dened as:
x(t) = A
ij
x(t) +B
i
[I
r

a
(t)]u(t) (69)
where,
a
is the so-called fault-effect factor, and
a
=
diag[
a
1
,
a
2
, . . . ,
a
r
], and 0
a
i
< 1 represents a fault in
the i
th
actuator and
a
i
= 0 means that i
th
actuator operates
normally (fault-free), whilst
a
i
> 0 means that some degree
of fault effect occurs in the actuator [9]. The distribution
matrix F
a
is equal to the matrix B in an actuator fault case.
The estimation of fault-effect factor
a
(t) is determined from
the fault estimation

f(t) provided by the LPV fault estimator.
In Section III-A, a constant controller is developed to achieve
P()
K
FTC
(
a
, )
LPV Estimator
u
FTC
(t)

a
Fig. 4. Active LPV Fault-tolerant Control Scheme
the stability of the 8-vertex system to satisfy assumption
104
(A1) that the framework of the LPV estimator design can
be implemented. The LPV controller can be expressed by:
K
lpv
= K +K() (70)
where, K is the developed constant controller. The structure
of the active LPV FTC system is shown in Fig.4, wherein,
K
FTC
(
a
, ) is the adaptive LPV controller for the FTC
mechanism, depending on the on-line estimation
a
and
measurement .
Theorem 0.3: From a design consideration consider the
system in (69) with i = 1, . . . , r actuator faults (
a
= 0)
acting independently within the control system with LPV
gain matrix K
lpv
. Dene = [I
a
(t)]. The new control
action(assuming non-zero fault effects) is given as:
u
FTC
(t) = (

+(I

)Z)K
lpv
x(t) = K
FTC
(
a
, )x(t)
(71)
where is required to be full row rank. represents the
Pseudo-Inverse. Z is the free matrix. Degree of freedom of
designing the LPV controller can be fully utilised through
choosing various Z [14].
Proof: Based upon the system state equation with
faults (69) and the new control input u
FTC
(t), the fault
compensation system is given by
x(t) = Ax(t) +Bu
FTC
(t)
= Ax(t) +B(

+ (I

)Z)K
lpv
x(t)
= Ax(t) +Bu
nom
(t)
(72)
It can be seen that the term (I
a
) acting on the system of
(69) can be removed through replacing u(t) with u
FTC
(t),
which completes the proof.
Fig.5 shows the system outputs after the fault being
compensated. The nominal performance of system outputs
generated by the LPV controller can be effectively recovered
after the fault compensation being activated at 40s. This
demonstrates very well the fault-tolerance of the active LPV
FTC system.
0 20 40 60 80
3.5
3
2.5
2
1.5
1
0.5
0
0.5
Time[s]
R
o
t
a
t
i
o
n

A
n
g
l
e
1

[
r
a
d
]
0 20 40 60 80
0.5
0
0.5
1
1.5
2
2.5
3
3.5
Time[s]
R
o
t
a
t
i
o
n

A
n
g
l
e
2

[
r
a
d
]

2
LPV Controller
being activated
at 40s
LPV Controller
being activated
at 40s
Fig. 5. Nonlinear System Output Responses with Active FTC
IV. CONCLUSIONS
This study concerns the development of an LPV fault
estimation scheme within an active or direct FTC for a tradi-
tional non-linear robot manipulator problem with challenging
control requirements. Friction forces at the manipulator joints
are considered as actuator faults acting in direct competition
with the control system. As a part of the structure of an LPV
fault compensation scheme, the principle of LPV robust fault
estimation with pole placement has been introduced through
via a set of LMIs using efcient interior point algorithms. It
is shown that not only can the robustness of the estimation
error be improved, corresponding to the system control
inputs, disturbances and the faults, but also the structure of
the robust LPV estimator can be modied online through
the measurement of the varying parameters. The friction
forces acting in the manipulator joints are generated using
the Stribeck friction model. The simulation results show that
the simultaneously acting friction forces can be estimated
effectively using the polytopic estimator and the actuator
faults can be compensated through the developed active LPV
FTC scheme.
V. ACKNOWLEDGMENTS
The authors wish to thank the European Commission
for research funding in the contract FP7-233815, Advanced
Fault Diagnosis for Safer Flight Guidance and Control
(ADDSAFE).
REFERENCES
[1] K.J.

Astr om and C. D. Wit, Revisiting the LuGre model stick-slip


motion and rate dependence, IEEE Control Systems Magazine, vol.
28(6), pp. 101114, 2008.
[2] R.J.Patton, D.Putra, and S.Klinkhieo, Friction compensation as a
fault-tolerant control problem, special issue on fault diagnosis and
fault tolerant control, Int. J. of Sys. Sci., vol. 41(8), pp. 987 1001,
2010.
[3] R.J.Patton and S.Klinkhieo, LPV fault estimation and FTC of a
two-link manipulator, special session on application of LPV control
methods, in 2010 American Control Conference, Baltimore, June
30July 2 2010.
[4] F.Wu, A generalised LPV system analysis and control synthesis
framework, Int. J. of Control, vol. 74, pp. 745749, 2001.
[5] S.Ganguli, A.Marcos, and G.J.Balas, Recongurable LPV control
design for b-747-100/200 longitudinal axis, in American Control
Conference, Anchorage, 2002.
[6] J.Bokor and G.Balas, Detection lter design for LPV systems: a
geometric approach, Automatica, vol. 40(3), pp. 511518, 2004.
[7] A. Casavola, D. Famularo, G. Franze, and M. Sorbara, A fault
detection lter design method for linear parameter-varying systems,
Proc. IMechE Part I: J. Sys. & Control Eng, vol. 221(6), pp. 865874,
2007.
[8] P. Apkarian, P.Gahinet, and G.Becker, Self-scheduled H

control of
linear parameter-varying systems: a design example, Automatica, vol.
31(9), pp. 12511261, 1995.
[9] J.Chen, R. Patton, and Z.Chen, Active fault-tolerant ight control
systems design using the linear matrix inequality method, Trans.
Institute of Measurement & Control, vol. 21(2-3), pp. 7784, 1999.
[10] M.Chilali and P.Gahinet, H

design with pole placement constraints:


An LMI approach, IEEE Transactions on Automatic Control, vol.
41(3), pp. 358367, 1996.
[11] P.J.McKerrow, Introduction to Robotics. Addison-Wesley Publishing
Company, Inc, 1991.
[12] S.Hassen, F.Crusca, and H.Abachi, Modelling system for control
studies - an overview, in ISCA 15th International conference on
Computer and their Application, Orlando, March 11-13 2000.
[13] D.Putra, L.Moreau, and H.Nijmeijer, Observer-based compensation
of discontinuous friction, in Proceedings of the 43rd IEEE Conference
on Decision and Control, Bahamas, 2004, pp. 49404945.
[14] L.Chen, Multirate eigenstructure assignment using lifting, Ph.D.
dissertation, University of York, 2009.
105

Potrebbero piacerti anche