Sei sulla pagina 1di 25

30 June 2012 1

ACTIVE FAULT ANALYSIS OF


PARAMETRIC FAULTS IN
DYNAMIC SYSTEMS
xxxxxxxxxxxxxxxxxxxxx
Presented by xxxxxxxxxxxx
Roll xxxxxx
Regd. No :- xxxxxxxxx
Branch :- Electrical &Electronics
Engineering

30 June 2012 2





CONTENTS
1. Importance of fault tolerance
2. Structure of fault tolerant control
system
3. Fault Detection and Isolation(FDI)
4. Robust design of FDI
5. Fault Diagnosis of Nonlinear Dynamic
System
6. System reconfiguration

30 June 2012 3
1. Importance of fault
tolerance
The growing complexity of control system
increases the possibility of component and
system failure.
System failure can cause mission abortion,
material damage and human fatality.
The improvement of system reliability can
not been totally dependent upon the
improvement of component reliability,
because of the restriction of technological
level and cost consideration
30 June 2012 4
A fault tolerant system is designed with redundancy
capacity. When its some components or subsystems fail,
it can reconfigure its remaining components and
information- processing capability to continue operation
properly.
Hardware redundancy and Analytical (functional)
redundancy.

Example:
Aircraft control system has several channels of
amplifiers. When one of them is failed, the system
can switch to other channel and continue
operation ( e.g. landing) instead of disaster with
hundreds of lives lost.

30 June 2012 5
Fault tolerance can greatly reduce the system
fault rate and it can use low reliable
components to achieve high reliable system.
Single channel system fault rate
Triple channel redundant system
More researches need to be done.

Challenges to Control(A collective view), IEEE Trans. AC-32
No.4, April 1987.
Aircraft Control SystemA projection to the year 2000, IEEE
Contr. Syst. Mag, pp11-13, Feb. 1985.

=

10
3
h
s

10
7
h
30 June 2012 6
2. Structure of fault
tolerant control systems
Actuators
Plant
Dynamics
Sensors
Residual
Generator
Decision
Making
Reconfiguration
Algorithm
Controller
Output
reference input
30 June 2012 7
Residual Generation
Observer Approach
Kalman Filter Approach
Parity Space Approach
Decision Making
Fault Detection: Whether there is a fault or not?
Fault Isolation: Where has the fault happened?
Fault Estimation: When did the fault happen?
How serious is the fault?

30 June 2012 8
Approaches to Decision Making
Statistical Hypotheses Test
Fuzzy Logic Inference
Neural Network Classifier
Fault Diagnosis Expert Systems

30 June 2012 9
3. Fault Detection and
Isolation
(1) Observer scheme
System model with faults:
(1)
(2)
Actuator fault
Sensor fault
d(t)Unknown Input (Modeling error
and/or disturbances)

( ) ( ) ( ) ( ) ( ) x t Ax t Bu t R f t Ed t = + + +
1 1
y t Cx t Du t R f t ( ) ( ) ( ) ( ) = + +
2 2
f t
1
( )
f t
2
( )
30 June 2012 10
Observer equation
(3)
(4)

Weighted output residual
(5)

Consider sensor faults ( )
(6)

(7)

(8)
(9)

( ) ( )

( ) ( ) ( ) ( ) x t A KC x t B KD u t Ky t = + +

( )

( ) ( ) y t Cx t Du t = +
r t W y t y t ( ) ( ( )

( )) =
R I, f 0
2 1
= =
r s H s f s H s d s
f d
( ) ( ) ( ) ( ) ( ) = +
2
H s W I sI A K
f c
( ) [ ( ) ] =
1
H s WC sI A E
d c
( ) ( ) =
1
A A KC
c
=
30 June 2012 11
Fault detection rule(Assuming d(t)=0)
no fault (10)
with fault (11)
threshold
Unknown Input d(t) will affect the correctness of fault
detection.
It is needed to seek for more robust methods.
Fault isolation
We use m observers where the ith observer uses

i.e., all sensor outputs but the ith sensor output .
Therefore, the ith observer is free of

|| ( )|| r t T
D
<
|| ( )|| r t T
D
>
T
D
y y y y y
i i m 1 2 1 1
, , , , , ,
+
y
i
30 June 2012 12
When only the ith sensor output has failed, the ith observer will still
give all correct estimates: , however, other observers will
give wrong estimates. Therefore, the ith sensor fault can be
isolated.

(2). Kalman filter scheme
Residual generation

(12)
is the estimate of K.F.
is the estimate of a Shadow Filter
(predictor)
, (13)

( ) x k
r k x k x k ( ) ( ) ( ) = '
( ) ' x k

( ) ( , )

( ) ' = ' x k k k x k u 1 1

( )

( ) ' = x x 0 0
, , y y
m 1

30 June 2012 13
Binary hypotheses test

(14)

(15)
According to Generalized Likelihood Ratio Test Theory, We
construct the following Decision function of Fault Detection(DFD)
(16)
W(k): Covariance matrix of r(k).
(17)
, where n (dimension of r(k)) is the degree of
freedom of distribution

H E r k no fault
0
0 : [ ( )] =
H E r k with fault
1
0 : [ ( )] =
r k N W ( ) ~ ( , ) 0
_ ( ) ~ ( ) k n
2
_
2
( ) ( ) ( ) ( ) k r k W k r k = '
1
30 June 2012 14
Fault detection rule:
no fault (18)
with fault (19)
threshold which can be determined by using
Distribution table given (False detection ratio)
Disadvantage of using Shadow Filter (S.F.)
If system model is not accurate, the estimate of S.F.
will gradually deviate from true value of state vector x
and r(k) will include the effect of modeling error.

( ) k T
D
<
( ) k T
D
>
T
D
30 June 2012 15
Innovation test


(20)

White noise: no fault

Zhang, H.G. and H. Y. Zhang, Fault tolerant scheme for
multisensor navigation systems, Proc. of 18th Congress of the
International of Aeronautical Sciences,
Beijing, China, Sept. 1992.

r k y k H k x k k ( ) ( ) ( )

( | ) = 1
30 June 2012 16
(3). Parity space approach
Static case
(21)

zero mean Gaussian noise vector,
f-sensor fault vector
If m>n, we can find a matrix V

VH=0
(22)

y Hx f = + + c
y R
m
e
c
x R
n
e
p Vy V Vf = = + c
30 June 2012 17
Decision function of Fault Detection(DFD)
(23)

(24)

Fault detection rule:
no fault (25)
with fault (26)

= '

p W p
1
W E V V VE V = ' = ' ' [( )( ) ] [ ] c c cc
< T
D
> T
D
30 June 2012 18
4. Robust design of FDI
It is difficult to distinguish between effects of
fault f(t) and modeling uncertainty d(t).

Eigenstructure assignment:
Design gain matrix K and weight matrix W to
make
then residual r(s) is decoupled from uncertainty
(unknown input) d(t)
H s WC sI A KC E
d
( ) [ ( )] = =
1
0
30 June 2012 19
5. Fault Diagnosis of
Nonlinear Dynamic System
(1)Introduction
Linearized model method may not give
satisfactory result due to mismatch between
linear model and nonlinear behavior.
Analytical solution for FDI of general
nonlinear systems is difficult.
Two ways to overcome the difficulties:


30 June 2012 20
> Restricting the class of nonlinear
systems
> Neuro-Fuzzy approach
(2) Bilinear Systems:

) ( ) ( ) (
) ( ) ( ) ( ) ( ) (
2
1
1
t d E t Cx t y
t d E t x t u B t Ax t x
r
i
i i
+ =
+ + =

=

30 June 2012 21
where
respectively the state,
output and unknown
disturbance of the
system.

Observer:

exists if and only if



(

+ =
(

+
(

+ =
(

0
) ( dim
0
0
) (
0 0
2
1
2
1
2
1
2
2
1 2
E
E E
rank x
E C
E E A sI
rank
E
E E
rank E rank
E
E C CE E
rank
) ( ) ( ) (
) ( ) ( ) ( ) ( ) (
1
t Ny t H t x
t y t u L t Gy t F t
r
i
i i
+ =
+ + =

=
,
, ,

d y x , ,
30 June 2012 22
Fuzzy Logic Approach for
FDI of Nonlinear System
() Takagi-Sugeno fuzzy model
The nonlinear system behavior is
described by
a fuzzy fusion of the outputs of all linear
models which are linearized at different
operating points of the nonlinear system.


30 June 2012 23
6. System reconfiguratio
(1) Sensor reconfiguration
l Hardware redundancy case:
Switch off faulty sensor(s) and switch on redundant
sensor(s)

Analytical redundancy case:
Use observer to estimate the measurement of
faulty sensor(s)


30 June 2012 24
(2) Actuator reconfiguration
Switch off faulty actuator(s) and switch on
redundant actuator(s)

(3) Control signal reconfiguration

For modeled faults
B has changed to because of some faults of
control devices (e.g. surface damage of rudder,
elevator, flap etc).


( ) ( ) ( ) x t Ax t Bu t = +
B
n
30 June 2012 25

THANKYOU
ANY QUERIES????

Potrebbero piacerti anche