EE-612: Fault Diagnosis and Tolerance

Dr. Abdul Qayyum Khan

Room No. SE-302

Department of Electrical Engineering,
Pakistan Institute of Engineering and Applied Sciences,
P.O. Nilore Islamabad Pakistan

Email: aqkhan@pieas.edu.pk
http://www.pieas.edu.pk/aqayyum/

February 16, 2017

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 1 / 23
Outline

1 Modeling of Processes and Faults

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 2 / 23
Process Disturbance Modeling and Noises

Frequency domain

y (s ) = Gyu (s )u(s )

y Rm : output vector
u Rku : input vector

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 3 / 23
Process Disturbance Modeling and Noises

Frequency domain

y (s ) = Gyu (s )u(s ) + Gyd d (s ) + Gy (s )

y Rm : output vector
u Rku : input vector
d Rkd : deterministic disturbance vector
Rk : stochastic disturbance vector

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 3 / 23
Process Disturbance Modeling and Noises

Frequency domain

y (s ) = Gyu (s )u(s ) + Gyd d (s ) + Gy (s )

y Rm : output vector
u Rku : input vector
d Rkd : deterministic disturbance vector
Rk : stochastic disturbance vector
State space

x = Ax + Bu
y = Cx + Du

x Rn : is the state vector, d is deterministic process disturbances

and Rk , Rk stochastic disturbance vector

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 3 / 23
Process Disturbance Modeling and Noises

Frequency domain

y (s ) = Gyu (s )u(s ) + Gyd d (s ) + Gy (s )

y Rm : output vector
u Rku : input vector
d Rkd : deterministic disturbance vector
Rk : stochastic disturbance vector
State space

x = Ax + Bu + Ed d +
y = Cx + Du + Fd d +

x Rn : is the state vector, d is deterministic process disturbances

and Rk , Rk stochastic disturbance vector

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 3 / 23
Modeling uncertainties

sources of uncertainty
parameters are not exactly known
variation in operating conditions
limited resolution
high frequency behavior
model simplification

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 4 / 23
Modeling uncertainties

sources of uncertainty
parameters are not exactly known
variation in operating conditions
limited resolution
high frequency behavior
model simplification
Process with uncertainties

x + Bu
x = A +E
d d
+ Du
y = Cx +Fd d

= A + A , B
A = B + B ,

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 4 / 23
Uncertainties Description

Norm Bounded
" # " #
A B E E h i
= (t ) G H J
C D F F

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 5 / 23
Uncertainties Description

Norm Bounded
" # " #
A B E E h i
= (t ) G H J
C D F F

(t )

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 5 / 23
Uncertainties Description

Norm Bounded
" # " #
A B E E h i
= (t ) G H J
C D F F

(t )

Polytopic uncertainty
" #
A B E
C D F
(" # " # " #)
A1 B1 E1 A2 B2 E2 Al Bl El
= Co , , ,
C1 D1 F1 C2 D2 F2 Cl Dl Fl

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 5 / 23
Uncertainties Description

Norm Bounded
" # " #
A B E E h i
= (t ) G H J
C D F F

(t )

Polytopic uncertainty
" #
A B E
C D F
(" # " # " #)
A1 B1 E1 A2 B2 E2 Al Bl El
= Co , , ,
C1 D1 F1 C2 D2 F2 Cl Dl Fl

Ai , Bi , Ci , known matrices
Co {} Convex Set
A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 5 / 23
Polytopic Uncertainty

Convex Set: A set C is convex if the line joining any two points c1
and c2 in C remains entirely in C.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 6 / 23
Polytopic Uncertainty

Convex Set: A set C is convex if the line joining any two points c1
and c2 in C remains entirely in C.

c1

c2

convex set nonconvex set

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 6 / 23
Polytopic Uncertainty

Convex Set: A set C is convex if the line joining any two points c1
and c2 in C remains entirely in C.

c1

c2

convex set nonconvex set

Given a set C (not necessarily convex) its convex hull is the smallest
convex set which contains C.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 6 / 23
Polytopic Uncertainty

Convex Set: A set C is convex if the line joining any two points c1
and c2 in C remains entirely in C.

c1

c2

convex set nonconvex set

Given a set C (not necessarily convex) its convex hull is the smallest
convex set which contains C.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 6 / 23
Convex Hull and convex function

The convex hull of a given set C may be defined as

The (unique) minimal convex set containing C
The intersection of all convex sets containing C
The set of all convex combinations of points in C
The union of all simplices with vertices in C
A convex function is a real-valued function defined on an interval with
the property that its epigraph (the set of points on or above the graph
of the function) is a convex set.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 7 / 23
Polytopic Uncertainty

A polytope is the convex hull of a finite set of points

P = convex hull{p1 , p2 , , pN }
A vertex is a point that cannot be generated as the convex combination of
two distinct points

p2
p5
p4

p1 p3
p6

p6

p7

(i) Polytope generated by (j) Two 3-D polytopes

vertices p1 , p2 , p5 , p6 , p7

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 8 / 23
Polytopic Uncertainty

Every point p in a polytope P can be generated as the convex combination

of the vertices of P
N
X N
X
p= pi i , i = 1, i 0
i =1 i =1

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 9 / 23
Polytopic Uncertainty

(" # " # " #)

A1 B1 E1 A2 B2 E2 A B E
Co , , ,
C1 D1 F1 C2 D2 F2 C D F
" #
X Ai Bi Ei
= i ,
Ci Di Fi
i =1

X
i = 1, i 0, i = 1, ,
i =1

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 10 / 23
Stochastic Uncertainty

" # " #
A B E X Ai Bi Ei
= pi (k )
C D F Ci Di Fi
i =1

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 11 / 23
Stochastic Uncertainty

" # " #
A B E X Ai Bi Ei
= pi (k )
C D F Ci Di Fi
i =1

p1 (k )


.
p (k ) = ..

p (k )

p (k ) is stochastic process
E [p (k )] = 0, E [p (k )p T (k )] = diag (1 , , )

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 11 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Types of faults based on their source

Actuator faults
Sensor faults
Component faults

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Actuator Faults
More consumption of energy or total loss of control
Example: Stuck up of control valve, faults in pumps, motors, etc.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Actuator Faults

(k) floating around (l) lock in place

trim

(m) hard-over failure (n) loss of

effectiveness
A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Sensor Faults
Sensor measurements are used in feedback, result into change in
operating point
Examples: drift, bias, freezing of sensor, calibration error

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Sensor Faults

(o) bias (p) drift

(q) loss of (r) freezing (s) calibrating

accuracy error

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Component Faults
Alter the physical parameter of plant
Due to wear and tear
Examples: leakage, cracks, change in friction,
Dynamic properties are changed, instability
Most important to be detected

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Types of faults based on time-domain behavior

(t) Abrupt fault (u) Incipient fault

(v) Intermittent fault

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Types of Faults

Types of faults based on their source

Types of faults based on their time-domain behavior
Types of faults with regards to process models

Types of faults with regards to process models

Additive faults
Multiplicative faults

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 12 / 23
Modeling of Faults

Frequency domain

y (s ) = Gyu (s )u(s )+Gyd d (s ) + Gy (s )+Gyf f (s )

f Rkd vector of faults

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 13 / 23
Modeling of Faults

Frequency domain

y (s ) = Gyu (s )u(s )+Gyd d (s ) + Gy (s )+Gyf f (s )

f Rkd vector of faults

State space

x = Ax + Bu+Ed d +Ef f
y = Cx + Du+Fd d +Ff f

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 13 / 23
Modeling of Faults

Frequency domain

y (s ) = Gyu (s )u(s )+Gyd d (s ) + Gy (s )+Gyf f (s )

f Rkd vector of faults

State space

x = Ax + Bu+Ed d +Ef f
y = Cx + Du+Fd d +Ff f

Ef and Ff depend upon the location of faults

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 13 / 23
Modeling of Faults Contd.

Sensor faults, Ff = I

y = Cx + du + Fd d + fs

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 14 / 23
Modeling of Faults Contd.

Sensor faults, Ff = I

y = Cx + du + Fd d + fs

Actuator fault, Ef = B , Ff = D

x = Ax + B (u + fA ) + Ed d
y = Cx + D (u + fA ) + Fd d

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 14 / 23
Modeling of Faults Contd.

Changes in signals are modeled additive faults

Changes in parameters are modeled as multiplicative faults
Representation as multiplicative faults

+ AF ) x + ( B
x = (A + BF )u + Ed d
y = (C + CF )x + (D + DF )u + Fd d

A
X B
X
AF = Ai Ai , BF = Bi Bi
i =1 i =1
C
X D
X
CF = Ci Ci , DF = Di Di ,
i =1 i =1

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 15 / 23
Modeling of Faults Contd.

Can be transformed into additive faults

qM = GF x + HF u,
fM = F (t )qM

Inn 0


. .
.. ..



I 0
GF = nn

, HF =
0 Iku ku
. .
.. ..

0 Iku ku

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 16 / 23
Modeling of Faults Contd.

A1 Inn
..

.

AA Inn
F (t ) =
B1 Iku ku
..

.


BB Iku ku

h i
EP = A1 AA B1 BB ,
h i
FP = C1 CC D1 DD

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 17 / 23
Modeling of Faults Contd.

x + Bu
x = A + Ed d + EP fM
+ Du
y = Cx + Fd d + FP fM

Effects stability, additive representation does not give this information

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 18 / 23
Modeling of Faults Contd.

Sensor faults, actuator faults, component faults

fA B D
f = fP , Ef = EP , Ff = FP

fs I

0

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 19 / 23
Examples
Example (Three Tanks System)

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Mass balance equations

accumulation = input output

Fluid flow through a sharp-edged hole

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Ac h 1 = Q1 Q13
Ac h 2 = Q2 + Q32 Q20
Ac h 3 = Q13 Q32

p
Q13 =a1 s13 sgn(h1 h3 ) 2g |h1 h3 |
p
Q32 =a3 s23 sgn(h3 h2 ) 2g |h3 h2 |
p
Q20 =a2 s0 2gh2

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Linearized Model

x =Ax + Bu
y =Cx

Operating point h1 = 45cm, h2 = 30cm, h3 = 15cm

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Linearized Model

x =Ax + Bu
y =Cx

Operating point h1 = 45cm, h2 = 30cm, h3 = 15cm

h1 " #
Q
x = y = h2 , u = 1

Q2
h3


0.0085 0 0.0085
0.0065 0

A = 0 0.00195 0.0084 , B = 0 0.0065 C = I3


0.0085 0.0084 0.0169 0 0

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Model Uncertainty caused by linearization

x =(A + A )x + Bu
y =Cx

A = (t )H

0.0085 0 0.0085

H = 0 0.00195 0.0084


0.0085 0.0084 0.0169
1 (t )

0 0

(t ) = 0 2 ( t ) 0

3 ( t )

0 0

(t )) 1.3620

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

Component faults,
Leaks in tanks p p p
A1 2gh1 , A2 2gh2 , A3 2gh3
Pluggings between two tanks
p
A4 a1 s13 sgn(h1 h3 ) 2g |h1 h3 |
p
A5 a3 s23 sgn(h3 h2 ) 2g |h3 h2 |
p
A6 a2 s0 2gh2
Sensor Faults, three additive faults in three sensors, f1 , f2 , f3
Actuator Faults, faults in pumps, f4 , f5

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

x =(A + AF )x + Bu + Ef f
y =Cx + Ff f

6
X h iT
AF = Ai Ai , f = f1 f6
i =1

0.0214 0 0 0 0 0

A1 = 0 0 0 , A2 = 0 0.0371 0 ,


0 0 0 0 0 0

0 0 0 0.0082 0 0.0085
A3 = 0 0 0 , A4 = 0 0 0


0 0 0.0262 0.0085 0 0.0085
A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Examples
Example (Three Tanks System)

0 0 0
0 0 0
A5 = 0 0.0111 0 , A6 = 0 0.0084 0.0084 ,


0 0 0 0 0.0084 0.0084
h i h i
Ef = 0 B , Ff = I33 0

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 20 / 23
Assignment

Assignment
1 Obtain the mathematical modeling and fault representation for the
Coupled Liquid Tanks System in DEE
2 Excercise 3.4 in M. Blanke, M. Kinnaert, J. Lunze, and
M. Staroswiecki, Diagnosis and fault-tolerant control. Springer, 2006

To be submitted on February 24, 2017

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 21 / 23
Further Readings

A. Q. Khan, Observer-based fault detection in nonlinear systems.

PhD thesis, University of Duisburg-Essen, Duisburg, Germany,
December 2010, Chapter 3
S. X. Ding, Model-based fault diagnosis techniques - design
schemes, algorithms and tools. Springer, 2008, Chapter 3
R. Isermann, Fault-diagnosis systems: an introduction form fault
detection to fault tolerance. Springer, 2006, Chapter 5

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 22 / 23
References I

M. Blanke, M. Kinnaert, J. Lunze, and M. Staroswiecki, Diagnosis and

fault-tolerant control.
Springer, 2006.
A. Q. Khan, Observer-based fault detection in nonlinear systems.
PhD thesis, University of Duisburg-Essen, Duisburg, Germany,
December 2010.
S. X. Ding, Model-based fault diagnosis techniques - design schemes,
algorithms and tools.
Springer, 2008.
R. Isermann, Fault-diagnosis systems: an introduction form fault
detection to fault tolerance.
Springer, 2006.

A. Q. Khan (DEE,PIEAS) EE-612: Fault Diagnosis and Tolerance February 16, 2017 23 / 23