ESISAR 5A
Department : AUTOMATIQUE Damien KOENIG
2018 : Fault Diagnosis
TP 2: P ARITY S PACE APPROACH
Objective: Build a bank of residuals in order to detect and isolate the fault sensor
1 S
ENSOR FAULT DETECTION AND ISOLATION FOR ABOILER SYSTEM EXCHANGER:
P
ARITY SPACE APPROACHTheoretical study
Consider the following boiler system exchanger
The variables are:
Tc : output water temperature at the boiler °C
Tp : output water temperature of the primary circuit of the exchanger °C
Ts : output water temperature of the second circuit of the exchanger °C
Qg : gas flow m3/h
Qp : water flow of the primary circuit of the exchanger l/h
Qs: water flow of the second circuit of the exchanger l/h
The state of the system is desbribed by the following relation :
k
Tc
k Tp k Ts k
Tx
the control inputs by :
k
Q
k2
Q k1
Q k5
Q k1
u g p p s
And the output by :
k T
c k T
p k T
s k T
y
TpO
Chaudière
Circuit primaire échangeur
Circuit secondaire échangeur TpO
Qg
Qp
Qs
TcO TcO
TpO
TsO
AC-514
2 The matrices A, B and C are given by :
33 31
22 21
12 11
0 0 0 a a
a a
a a
A ,
34 32
24 22
13 11
0 0
0 0
0 0
b b
b b
b b
B , C = I3
1.1 Parity space
Without increasing the time k, how many parity equations can be deduced?
Gives the parity equations with a minimum horizon of observation, to limit the detection delays.
Gives the fault location decision table, each sensor fault alarm with the logic decision.
2 S
TATIC PARITY SPACE WITH NON PERFECT DECOUPLING Consider the static output relation:
k Cx
k k Fd
ky
where xRn, yRm anddRp .
y(k) is the output measurement, x(k) the state variable, d(k) the fault vector to be detected and (k) the noise measurement. Matrices C and F are known with appropriate dimension.
Assumption: m > n for a redundancy information existence.
We will find a parity vector sensitive to p-1 defaults and insensitive to di which represent the ith component of d.
This leads to explain the measurement vector in the form:
y
k Cx
k k Fd
k Fd
kwhere d+ are d- are respectively the sensitive and insensitive faults. F + et F – are the associated matrices.
2.1 Numerical Application:
Consider the following system:
2 0 2
1 0 1
1 1 1
2 0 1
1 2 1
C ,
1 0
5 2
0 0
2 1
2 1
F ,
1 1 3 0 1 F
Find the kernel w with the optimization problem described bellow. Give wTF- and wTF+.
Check the products wTC, wTF- and wTF+ and conclude.
Solution:
The problem is reduced to find a matrices W such that
C F 0W (1)
where the parity vector is given by : p
k W
k WFd
kAC-514
3 The exact solution W of (1) holds if and if the line rank of (C F-) is deficient. If not the following optimization problem can be used:
2 2
0
F w
F min w
C w
T T w
T
remark: max(f(x)) = - min(-f(x)) example f(x) = x2-2
Explanation
Consider
2 1
C
C C where C1 is of full rank. The constraints wTC = 0 is true by using the following changement :
Pw2
w
where
I
C
P C T
T 1 2
1 .
Proof :
1 2 2 112 1 2
1
0
0
w w C C
C w C w C
w
T T T T T thus
02 1 1 1 2
2
C I C C C
wT .
With identification:
2 2 1 2
1 1 1 2
2 w w Pw
I C w C
I C C w
w T
T T
T
.
Using the new variable w2, the problem 2
2
0
F w
F min w
; C
w T
T w
T becomes :
2 22 22 22 2
2 2
2
2 w B~w
w A~ minw Pw F F P w
Pw F F P minw w F F w
w F F minw F
w F
min w T
T T w
T T
T T T T w
T T T T w
T
w
where A~ PTF
F TP and B~PTF
F TP.Since the small eigenvalue of the pair (A~,B~
) is the minimal of the criteria, the corresponding eigenvector w2 is the solution of the optimization problem. Find , such that
B~
w2 0A~
. From w2 and P deduce w and the parity relation:
d F d F w p
y w p
T T
That conclude the proof.