InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.

org
41
ActiveFaulttolerantControlofUnmanned
UnderwaterVehicles
EssamNabil
*1
,AbdelAzemSobaih
1
andBelalAbouZalam
1

1
IndustrialElectronicsandControlEngineeringDept.,FacultyofElectronicEngineering,MinoufiyaUniversity
Minouf,Minoufiya,Egypt

*1
essam.abdelaziz@eleng.menofia.edu.eg

Abstract
In this paper, a hierarchical systematic design methodology
of active fault tolerant control system is developed to
accommodate partial actuator and/or sensor faults of
Unmanned Underwater Vehicles (UUVs). An integrated
sensitiveFaultDetectionandDiagnosis(FDD)mechanismis
developed with acceptable time period and a computation
load reduction for large number of actuators and sensors in
the presence of system disturbances and random noise. An
optimized robust reconfigurable controller design is
presented based on singular value decomposition (SVD)
principle and eigenstructure assignment technique. The
effectiveness of the proposed scheme has justified by
simulationresultonthedecoupledsteeringsubsystemofthe
Naval Postgraduate School (NPS) Unmanned Underwater
Vehicles(UUVs).
Keywords
Unmanned Underwater Vehicles (UUVs); Modelbased Fault
Detection and Diagnosis (FDD); FaultTolerant Control System
(FTCS); Controller reconfiguration; Singular Value
Decomposition (SVD) Principle; Eigenstructure Assignment
Technique(EA)
I nt r oduc t i on
Nowadays, control systems are everywhere in our life
butunfortunately,theycantworkperfectlyatalltime.
The need to design controllers that guarantee both
system stability and acceptable performance upon the
occurrence of faults has been a hot topic of research.
Faults being dynamic in nature, the faulttolerant
control system (FTCS) should be capable of
accommodating them quickly, especially for complex
systems e.g. unmanned underwater vehicles,
aircraft/helicopter, spacecraft, hazardous chemical
plants, and control of nuclear reactors. The fault
tolerant control system (FTCS) is defined as a control
system that possesses the ability to accommodate
systemcomponentfailuresautomatically(Y.M.Zhang,
2008) (J. Jiang, 2005). A system is Fault Tolerant (FT)
abletorecoveritsoriginaltaskafterafaultoccures,

with the same or degraded acceptable performances.

Faulttolerantcontrol(FTC)approachesarecategorized
into two types (Y. M. Zhang, 2008): passive andactive
approaches.PassiveFTC(PFTCS)cantolerateaclassof
predefined set of faults considered at the controller
design stage and needs neither fault detection and
diagnosis (FDD) schemes nor controller
reconfiguration. Active FTCS (AFTCS) on the other
hand,employsFaultDetectionandIdentification(FDI)
scheme that monitors system performance to detect
and isolate faults (S., Katipamula, PartI, 2005) (S.,
(Katipamula,PartII,2005).Accordingly,thecontrollaw
is reconfigured online. Figure 1 demonstrates the
typical components of AFTCS that implies a sensitive
FDD scheme to provide precise and the most upto
date information about the system as soon as possible
after the fault occurrence and controller
reconfiguration mechanism that organizes the online
reconfigurable controller to recover the prefault
systemperformance.

FIG.1ACTIVEFAULTTOLERANTCONTROLSYSTEM(FTCS)
COMPONENTS
Since Unmanned Underwater Vehicles (UUVs) are
widely used in commercial, scientific, and military
www.ijape.orgInternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013
42
missionsforvariouspurposes,itisnecessarytoembed
FTCparadigmsintoUUVstoincreasethereliabilityof
the vehicles and enable them to execute and finalize
complex missions (A. Alessandri, , M. Caccia, G.
Veruggio, 1997) (A. Alessandri, M. Caccia, and G.
Veruggio, 1999) (M. Caccia, & G. Veruggio, 2000)
(Podder T.K. & Sarkar N., 2001) (A. Healey, and D.
Lienard, 1993). Earlier investigations of literature
employed parameter estimation and stateestimation
methods for fault detection and diagnosis in UUVs.
The stateestimation method monitors the system
status based on an analytical model, whereas the
parameter estimation method determines the system
status based on system identification, where the
residuals refer to plant parameters, rather than state
variables. The parameterestimation method does not
require the analytical model in advance. This is
especially helpful in those areas where analytical
models are not easy to develop. In this paper, a
systematicactivefaulttolerantcontrolsystem(AFTCS)
design that integrates fault detection, diagnosis and
accommodation is developed. The design procedures
imply: (1) an integrated design of a modelbased fault
detectionandisolation(FDI)techniqueusingabankof
constrained Kalman filters estimators, (2) an obvious
fault estimation algorithm is derived that can estimate
the effectiveness factor of a detected faulty sensor or
actuator in the presence of the simulated system
disturbance and measurement noise, and (3) a flexible
approach to an online controller reconfiguration
design for the postfault system to compensate for
detected failures leads to performance degradation or
damagestothesystem.Areconfigurablestatefeedback
controller with reference input is designed to recover
the prefault system both transient and steady state
performance based on eigenstructure assignment
technique and singular value decomposition (SVD)
principle.Thispaperisorganizedasfollows:Section2
providessystemanalysis.Section3,faultdetectionand
diagnosis. Section 4 shows the reconfigurable
controller design. Section 5 presents simulation results
for the steering subsystem of the Naval Postgraduate
School (NPS) UUV with simulated actuator (rudder)
failures. Finally, concluding remarks are made in
section6followedbythelistofreferences.
Syst em Anal ysi s
UnmannedUnderwaterVehicleKinematics
The kinematic relationships for the unmanned
underwater vehicles (UUV) modeling discusses the
motion in six degrees of freedom (DOF). Figure 2
shows the schematic diagram of UUV bodys inertial
coordinate axes definition and six motion components
i.e.surge,sway,heave,roll,pitchandyaw.Formarine
vehicles, it is common to use the Society of Naval
Architects and Marine Engineers (SNAME) notation
summarizedinTable1(SNAME,1950).

FIG.2THESCHEMATICDIAGRAMOFUUVBODYAND
INERTIALCOORDINATEAXESDEFINITION
TABLE1SNAMENOTATIONUSEDFORMARINEVEHICLES
6motioncomponents
Forces
and
moments
Linearand
angular
velocities
Positions
AndEuler
Angles
surge
(motionsinthex
direction)
X u x
sway
(motionsinthey
direction)
Y y
heave
(motionsinthez
direction)
Z z
roll
(rotationaboutthex
axis)
K p
pitch
(rotationaboutthey
axis)
M q
yaw
(rotationaboutthez
axis)
N r
There are two coordinate reference frames denoted as
the bodyfixed reference frame B (Xo Yo Zo), and the
inertiafixed frame I(XYZ)as indicated in figure A.1.
The motion of the bodyfixed frame is described
relativetotheinertialreferenceframeandtheoriginO
of frame B is usually chosen at the center of gravity
(CG) when the CG is in the principal plane of
symmetry.ThebodyfixedaxesXo,Yo,Zocoincidewith
the principal axes of inertia with the longitudinal axis
Xo pointing from aft to fore, the transverse axis Yo to
InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.org
43
starboard, and the normal axis Zo from top to bottom.
Thepositionandorientationofthevehicle(x,y,z,,,
) are described relative to frame I, while expressing
thelinearandangularvelocitiesofthevehicle(u,,,
p,q,r)inthebodyfixedcoordinatesystem.

FIG.3BODYREFERENCEFRAMESFORUNDERWATER
VEHICLE
Thecoordinatesaregroupedintotwovectorsas:
T T T T
z y x ] [ ] [
2 1
u | q q q = =
(1)
T T T T
r q p u v v v ] [ ] [
2 1
e u = =
(2)
where represents the vector of position and
orientation of the vehicle in the inertial frame while
represents the vector of linear and angular velocity of
thevehicleinthebodyfixedframe.Thesixdegreesof
freedom(DOF)equationsthatareusedtodescribethe
kinematictransformationbetweenframeIandframeB
aregivenas(T.I.Fossen,1994)(T.I.Fossen,2002):
v q q ) ( J = (3)
whereJ()isanonlineartransformationmatrixandcan
bewrittenas:
(

=
) ( 0
0 ) (
) (
2 2
1 1
q
q
q
J
J
J
(4)
UnmannedUnderwaterVehicleDynamics
The6DOFnonlineardynamicequationsofmotionare
expressedinacompactformas(T.I.Fossen,1994)(T.I.
Fossen,2002):
t q v v v v v = + + + ) ( ) ( ) ( g D C M (5)
whereMistheinertiamatrixofthevehicle,C()isthe
matrix of Coriolis and centrifugal terms, D() is the
matrixofhydrodynamicdampingterms,gisthevector
ofgravityandbuoyantforcesandisthecontrolinput
vectordescribingtheforcesandmomentsactingonthe
vehicle in the bodyfixed frame. The bodys inertia
tensor corresponding to the bodyfixed coordinate
systemXo,YoandZowithoriginOcanbedefinedas:
(
(
(

=
z zy zx
yz y yx
xz xy x
o
I I I
I I I
I I I
I
(6)
whereIx,IyandIzarethemomentsofinertiaaboutthe
Xo,YoandZoaxes,respectively.Ixy=Iyx,Ixz=Izx,andIyz=
Izyareproductsofinertia.Fossen(T.I.Fossen,1994)(T.
I.Fossen,2002)simplifiestheequationsofmotionas:
X q pr z r pq y r q x q r u m
G G G
= + + + + )] ( ) ( ) ( [
2 2
e u (7)
Y r qp x p qr z p r y ur p m
G G G
= + + + + )] ( ) ( ) ( [
2 2
e u
(8)
Z p rq y q rp x q p z p q u m
G G G
= + + + + )] ( ) ( ) ( [
2 2
u e
(9)
K ur p z p q u y m qr I I p I
G G y z x
= + + + + )] ( ) ( [ ) ( e u u e
(10)
M p q u x q r u z m rp I I q I
G G z x y
= + + + + )] ( ) ( [ ) ( u e e u
(11)
N q r u y ur p x m pq I I r I
G G x y z
= + + + + )] ( ) ( [ ) ( e u e u
(12)
wheremisthemassofthebody,xG,yGandzGarethe
coordinate components of rG. Due to the symmetric
shape of UUVs, the center of gravity (CG) is usually
taken as the origin O in such cases rG = [0,0,0]. When
thebodyaxescoincidewiththeprincipalaxesofinertia,
it implies that the inertial tensor about the bodys
centerofgravityisdiagonal,i.e.
) , , (
C C C
z y x C
I I I diag I = (13)
Subsequently, a simple representation will be yielded
as(T.I.Fossen,1994):
X q r u m = + ) ( e u (14)
Y ur p m = + ) ( e u (15)
Z p q u m = + ) ( u e (16)
K qr I I p I
C C C
y z x
= + ) ( (17)
M rp I I q I
C C C
z x y
= + ) ( (18)
N pq I I r I
C C C
x y z
= + ) ( (19)
DecoupledSubsystemsofUnmannedUnderwater
Vehicle
Equation of motion introduced in the previous section
is not practical for controller or observer design.
Therefore,acontrolstrategyforslenderandsymmetric
vehicles is possible to separate the sixDOF equations
of motion into noninteracting (or lightly interacting)
subsystems, and design one controller for each
www.ijape.orgInternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013
44
subsystem. For decoupled control system design, we
grouptogetherrelatedequationsofmotionforseparate
functions of steering, diving, and forward speed (A.
Healey,andD.Lienard,1993)(T.I.Fossen,1994)(T.I.
Fossen, 2002) (B. Jalving, 1994). The three subsystems
andtheircontrolandstatevariablesaresummarizedin
table A.2. Each subsystem has multiple states and a
singlecontrolelement.Thesteeringsubsystemcontrols
heading errors, while the diving subsystem controls
depth and pitch errors aswell as the speed subsystem
i.e.thesurgesubsystemcontrolsthepropellerdcmotor.
The steering subsystem dynamics is the most
challenging one of the three almost decoupled sub
systems.
TABLE2CONTROLANDSTATEVARIABLESOFUUVSUBSYSTEMS
Subsystem StateVariables Controlinputs
Speed
Steering
Diving
u(t)
(t),r(t),(t)
(t),q(t),(t),z(t)
n(t)
r(t)
s(t)
SteeringSubsystemofNPSUUV
Under the assumption of nearly constant speed u u0,
the vehicle dynamics in sway and yaw can be
simplifiedto:
Y r u m m = +
0
u
(20)
N r I
z
= (21)
where
r r r
Y r Y Y r Y Y Y o u u
o u u
+ + + + =

(22)
r r r
N r N N r N N N o u u
o u u
+ + + + =

(23)
For small roll and pitch angles we can assume that
r ~ . Based on the above assumptions, the linear
equations of motion for the steering subsystem are
expressedas(T.I.Fossen,1994)(B.Jalving,1994):
r G r v
r v
r
z
v
G
r
G
v
N
Y
r

u mx N u N
mu Y u Y
r

- N I - N mx
Y x m -Y m
. .
.
o

u
o
o
(
(
(

+
(
(
(

(
(
(

=
(
(
(

(
(
(

0 0 1 0
0
0

1 0 0
0
0
0 0
0 0
.

(24)
From(24)wecanwrite:
(

(
(
(

+
(
(
(

(
(
(

=
(
(
(

(
(
(

rs
rb
G r v
r v
r
z
v
G
r
G
v
Y u
r u mx N u N
u m Y u Y
r

- N I - N mx
Y x m -Y m
. .
.
o
o

u
o
0 0
0.377L - 0.283L
1 1
5 . 0
0 1 0
0 ) (
0 ) (


1 0 0
0
0
2
0
0 0
0 0
.

(25)
Naval Postgraduate Schools work determines the
valuesofconstantsandcoefficientsin(25).Substituting
and finally, rearranging it into a statespace form, we
have:

0 0
0.302 0.1924
0.2650 - .1708 0

0 1 0
0 4659 . 0 0591 . 0
0 4439 . 1 6383 . 0
(

(
(
(

+
(
(
(

(
(
(

=
(
(
(

rs
rb
r r
o
o

(26)

1 0 0
0 1 0
(
(
(

u
r Y
(27)
The equivalent discrete linear state space model
representationofthegivensystemcanbedescribedas:
k k k 1 k
w u B A x x + + =
+
(28)

k k k
v x C y + = (29)
where xk is the state vector R
nx1
, A is the system
matrix R
nxn
, B is the control matrix R
nxp
, uk is the
input control vector R
px1
, wk is the system
disturbances, yk is the measured output R
qx1
, C is the
output matrix R
qxn
, vk is the measurement noise with
zero mean. Using zero order hold with a sampling
periodTs=0.1s,thediscretemodelparameterswillbe
givenas:
(
(
(
(

=
1 0.09769 0.0002848
0 0.9541 0.005592
0 0.1366 - 0.9378
A
(
(
(
(

=
0.001484 0.0009488
0.02943 0.01884
0.02777 - 0.01521
B

and
(
(

=
1 0 0
0 1 0
C

InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.org
45
ModelingFaults
The system model with actuator and/or sensor faults
canbeconsideredas(G.Liu,D.Wang,andY.Li,2004)
(Y.M.ZhangandJ.Jiang,2002):
k k f k 1 k
w u B A x x + + =
+
(30)
k k f k
v x C y + = (31)
wheretheactuatorfaultsaremodeledby:
) ..., , , diag( ), - (I B B
ap a2 a1 a a f
= = (32)
whereai,i=1,2,,prepresentthecontroleffectiveness
factorsthatindicatetheactuatormultiplicativefaults.If
ai=0thenthereisnofault.If0<ai<1thenthereisa
partialfaultorlossincontrolaction.Ifai=1thenthere
is a complete actuator failure. Similarly the sensor
faultsaremodeledby:
) ..., , , diag( C, ) - (I C
sq s2 s1 c c f
= = (33)
wheresi,i=1,2,,q,representtheeffectivenessfactors
thatindicatethesensorfaults.Ifsi=0thenthereisno
fault.If0<si<1thenthereisapartialsensorfault.Ifsi=
1thenthereisacompletesensorfailure.
Faul t Det ec t i on and Di agnosi s
FaultDetectionandIsolation
AmodelbasedFDItechniqueisdevelopedbyusinga
setofconstrainedKalmanfiltersforfaultdetectionand
isolation (G. Liu, D. Wang, and Y. Li, 2004). Fault
isolation for a set of components and/or a single
component is achieved by generating structured
residual signals sensitive to certain faults and
insensitive to others by partitioning the input control
matrix B to two input matrices Bincl and Bexcl
consequently the input control vector uk is partitioned
to uincl and uexcl respectively for individual actuator
fault isolation. Similarly, the output matrix C is
partitioned to two matrices Cincl and Cexcl and so the
output vector yk is partitioned to yincl and yexcl
respectively for individual sensor fault isolation. The
system state space model of equations (28) and (29)
becomes:
k excl excl incl incl k 1 k
w (k) u B (k) u B A x x + + + =
+
(34)
k k
excl
incl
excl
incl
k
v x
C
C

(k) y
(k) y
y +
(

=
(

=
(35)
whereBinclandBexclforthei
th
actuatoraregivenas:
(
(
(
(

=
0
0
0

0
0
0

b 0 0
b 0 0
b 0 0
B
ni
2i
1i
incl

,
(
(
(
(
(

=
np
2p
1p
1) - n(i
1) - 2(i
1) - 1(i
1) - n(i n1
1) - 2(i 21
1) - 1(i 11
excl
b
b
b

b
b
b

0 b b
0 b b
0 b b
B

SimilarlyCinclandCexclforthei
th
sensoraregivenas:
(
(
(
(
(
(
(
(
(

=
(
(
(
(
(
(
(
(
(

=
+ + +
qn q2 q1
1)n (i 1)2 (i 1)1 (i
1)n - (i 1)2 - (i 1)1 - (i
1n 12 11
excl in i2 i1 incl
c c c
c c c
0 0 0
c c c
c c c
C ,
0 0 0
0 0 0
c c c
0 0 0
0 0 0
C

ThediscreteKalmanfilteralgorithmequationsfallinto
two groups: time update equations (prediction) and
measurement update equations (correction)
representedas:
k k/k 1/k k
u B x A x + =
+
(36)
] x C - [y g x A x
1/k k 1 k 1/k k 1 1/k k + + + + +
+ = (37)
by suitable selection of The innovation updating gain
matrixgprovidedecayedresponseforestimationerror
ekwhere:
k/k k k
x - x e = (38)
and

1 1/k k 1 k 1 k
x - x e
+ + + +
= (39)
from(28)and(37)wecanwrite:
k k k 1 k
v g - w C) g - I ( e A C) g - I ( e + =
+
(40)
squaringbothsidesandtakingstochasticmeanof(40)
gives:
T
vv
T
ww
T T
ee ee
g g C) g - I ( C) g - I (
C) g - I ( A (k) A C) g - I ( 1) (k
+
+ = +
(41)
where ee is the state estimation error covariance, ww
is the process disturbance covariance, vv is the
measurementnoisecovariance.Thepredictorcorrector
equationsfor(34)and(35)canbewrittenas:
(k) u B x A x
incl incl k/k 1/k k
+ =
+
(42)
] x C - 1) (k [y g x A x
1/k k incl incl 1/k k 1 1/k k + + + +
+ + = (43)
Theerrorcovarianceof(41)hasbecome:
www.ijape.orgInternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013
46
T
vv
T
incl ww incl
T
incl
T
ee incl ee
g g ) C g - (I ) C g - (I
) C g - (I A (k) A ) C g - (I 1) (k
+
+ = +
(44)
taking the 1
st
order variance of (44) and then solving
for ee and g. Fault detection here is reliant on
determiningekthatwillbeclosetozerodependingon
availablenoiseandsystemdisturbancesbutafterfault
occurrence it will be considerably high. For each
included set of actuators or sensors and for each
includedsingleactuatororsensor,aconstraintKalman
filterestimatorisdesigned(E.Larson,B.Parker,andB.
Clark,2002).Faultisolationhasperformedbydetecting
thefaultyset,thenisolatingthefaultysingleindividual
componentwithinthefaultydetectedset.Theresidual
signals generated by the state estimators are affected
by system disturbances and noise, so a suitable
threshold value can be used for fault detection level
dependingontherealsystem.
FaultEstimation
Afterthefaulthasbeendetectedandisolatedinasetof
component(orinasinglecomponent)thenextstepfor
faultdiagnosisistoestimatethevalueofthepostfault
system,sensoreffectivenessfactorinequation(33)and
thecontroleffectivenessfactorinequation(32).Forthe
faultfree system, the sensor measurement matrix will
beaffectedaftersensorfaultfrom:

q 1,2,.., = i , ] c ,..., c ,..., [c C

T
q i 1
=
(45)
to:
q 1,2,.., = i , ] c ,..., )c - (1 ,..., [c C
T
q i si 1 f
=
(46)
where si represents the estimated effectiveness factor
in case of a faultin the i
th
sensor. The discrete Kalman
filter algorithm equations (36), (37) considering i
th

sensorfaultswillbe:
] x C v x [C g x x
1/k k k 1 k f 1/k k 1 1/k k + + + + +
+ + =
(47)
Thenwecanhave:
k 1 k f 1/k k 1 1/k k
v g x C g x C) g - (I x + + =
+ + + +
(48)
usingequations(28)and(36)wecanwrite:
k 1 k f k
1 k k/k k 1 1/k k
v g x C g w C) g - (I
x C) g - (I ) x C)A(x g - (I x
+ +
+ =
+
+ + +
(49)
andequation(40)willbe:
k k 1 k
f
k
f
1 k
v g w C) g - (I x C g e C)A g - (I e + + =
+ +
(50)
where
T
i si f
0] ,..., c [0,..., C C C = =

Theerrorcovarianceof(44)canbewrittenas:
T
vv
T
ww
T T
xx
T T f
ee
f
ee
g g C) g - (I C) g - (I
g C C g C) g - (I A (k) A C) g - (I 1) (k
+ +
+ = +
(51)
where
f
ee is the state estimation error covariance
consideringi
th
sensorfault,andxxisgivenas:

} x E{x
T
1 k 1 k xx + +
=
(52)
from(41)and(51)wehave:
T T
xx
T T
ee
f
ee
g C C g
C) g - (I A A C) g - (I -

+ = =
(53)
Let
C
s si
=
(54)
and
T T
xx s
g C C g M =
(55)
then
s
2
si
T T
s xx s
2
si s
) g g ( M = =
(56)
where
T T
s xx s s
g g =
(57)
The effectiveness factor can be estimated as a mean
value by rearranging nonzero elements of Ms and the
correspondingnonzeroelementsofOsintoarraysMsk
andOskrespectively,sowecanwrite:
)

M
(
Ns
1

Ns
1 k sk
sk
si
=
=
(58)
where Ns is the total number of nonzero elements in
Ms(orOs)andk=1,2,,Ns.Similarly,forthefaultfree
system,theinputcontrolmatrixthatgivenas:
p 1,2,.., = i , ] b ,..., b ,..., [b B
p i 1
= (59)
willbeaffectedafteractuatorfaultto:
p 1,2,.., = i ], b ,..., )b - (1 ,..., [b B
p i ai 1 f
= (60)
where ai represents the loss in control effectiveness
factor of the i
th
actuator. The discrete Kalman filter
algorithmequationsconsideringactuatorfaultswillbe:
k k
k k k k/k 1 1/k k
v g w C g
u B C g e A C g u B x A x
+
+ + + + =
+ +
(61)
with 0] ,..., b [0,..., B B B
i ai f
= = ,equation(39)willbe:
k k k k
f
1 k
v g - w C) g - (I u B C) g - (I e A C) g - (I e + + =
+
(62)
alsoerrorcovarianceof(62)canbewrittenas:
T
vv
T
ww
T T
uu
T T f
ee
f
ee
g g C) g - (I C) g - (I
C) g - (I B B C) g - (I
C) g - (I A (k) A C) g - (I 1) (k
+
+
+ = +
(63)
where
f
ee is the state estimation error covariance
considering i
th
actuator fault, and uu is expected
InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.org
47
covariance of the control input. From (41) and (63) we
have:
T T
uu
T T
ee
f
ee
C) g - (I B B C) g - (I
C) g - (I A A C) g - (I -

+ = =
(64)
Let
B
a ai
=
(65)
and
T T
uu a
C) g - (I B B C) g - (I M = (66)
then
a
2
ai
T T
a uu a
2
ai a
) C) g - (I C) g - ((I M = = (67)
where
T T
a uu a a
C) g - (I C) g - (I = (68)
The effectiveness factor can be estimated as a mean
value by rearranging nonzero elements of Ma and the
corresponding nonzero elements of Oainto arrays Mak
andOakrespectively,sowecanwrite:
)

M
(
Na
1

Na
1 k ak
ak
ai
=
=
(69)
where Na is the total number of nonzero elements in
Ma(orOa)andk=1,2,,Na
Rec onf i gur abl e Cont r ol l er Desi gn
The activation of the reconfiguration process is
constrained by the following condition (Y. M. Zhang
andJ.Jiang,2002)(Y.M.ZhangandJ.Jiang,2001):
p 1,..., i ,
i i
1 - k
i
k
i
k
= < =
(70)
where o
i
is the threshold level involved as a design
parameter and
i
k represents the errors in the
consecutivecontroleffectivenessfactorestimation.The
reconfigurable fault tolerant control system design is
required to recover the prefault system both transient
and steady state performance based on the accurate
FDD,takingintoaccounttheperformancedegradation
intheoccurrenceofactuatorfaults(Y.M.ZhangandJ.
Jiang, 2003). The system representation in (28), (29) is
assumed to be both controllable, observable and input
as well as output matrices are of full rank, then the
statefeedbackcontrolleroftheform:
k forward k k
r K Kx u + = (71)
can be designed to find the gain matrix K R
pxq
using
singular value decomposition (SVD) principle for
predefined eigenvalues {sj ; j = 1,2,...,m} where rk
represents the reference input and Kforward denotes the
feedforward control gain to achieve steadystate
tracking of the reference input (T. Didier, J. Cdric, Y.
M.Zhang,2008)i.e.
r y lim
k
=
+
(72)
By using a linear state transformation of the state
vectorxktotheinputclosedstatespaceas:
kT
x T x
k
= (73)
Theclosedloopsystemmodelcanbewrittenas:
k forward T kT T T 1)T (k
r K B x K) B (A x + =
+
(74)
where
(
(
(

=
p n n,
p
0
I
B T
andIpisidentitymatrix R
pxp
,
B T B T, A T A
-1
T
-1
T
= =
The closed loop eigenvalues denoted by sj and the j
th

rightassociatedeigenvectorsdenotedbyvjrepresented
as:
m 2,..., 1, j , v s v K) B (A
j j j T T
= = (75)
rearranging(75)wecanwrite:
0
Kv
v
h
Kv
v

0
I
A - I s
j
j
Tj
j
j
p n n,
p
T j
=
(
(
(

=
(
(
(

(
(
(

(76)
Applying singular value decomposition (SVD)
principletohTjgives:

T
jn jn jn Tj
P L h =
(77)
whereLjnandP
T
jnaretheleftandrightsingularvectors
respectively and ojn represents the set of singular
values of hTj matrix. Through that principle, the
achievable eigenvectors are located in the subspace
with dimension determined by the number of inputs
(rank (B)) and orientation determined by the matrices
(AT, BT) and the desired eigenvalues sj. As all column
vectors {Pjk ,k=p+1,,n+p}of the right singular vectors
Pjn generated by using singular value decomposition
(SVD)proceduressatisfythefollowingcondition:

0 P h
jk Tj
=
(78)
from(76)wecanwrite:
(

=
(

=
(

=
j
j
j
j
j
j
jk
Q K
Q
K v
v
w
Q
P
(79)
Where Q R
nxn
, W R
pxn
, then using singular value
decomposition (SVD) based solution for the desired
eigenvaluestoconstructthegainmatrixKwehave:

Q K W =
(80)
andgainmatrixKisgivenas:

-1
Q W K =
(81)
The reconfigurable control law for the postfault
systemcanbewrittenas:

k
f
forward k f k
r K x -K u + = (82)
www.ijape.orgInternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013
48
where Kf, K
f
forward are the control gain matrices of the
postfault system. To achieve the same steadystate
trackingofthereferenceinputforthepostfaultsystem
(G.Liu,D.Wang,andY.Li,2004):
K B K B
f
forward f forward
= (83)
Bythedefinitiononanotherlinearstatetransformation
ofthestatevectorxktotheinputclosedstatespace,the
transformedsystemmatricesaregivenas:
(
(
(

=
p n n,
p
f f
0
I
B T
andIpisidentitymatrix R
pxp
,
f
-1
f T f
-1
f fT
T A T A , B T B = = (84)
The control reconfiguration using eigenstructure
assignmentisusedheretoguaranteethestabilityofthe
reconfigured system by recovering the system
eigenvalues and to recover the prefault system
performance to the maximum extent by placing the
eigenvectorsasclosetothoseoftheoriginalsystemas
possible (D. Krokavec, , 2005) (A. Esna Ashari, A.
KhakiSedigh,M.J.Yazdanpanah,2005)i.e.
n 1,2,..., i , ) K B (A K) B (A
i f fT T T T
f
i
= = = = (85)
Both(75),(76)canberewrittenas:
m 2,..., 1, j , v s v ) K B (A
fj j fj f fT T
= = (86)
0
v K
v
h
v K
v

0
I
A - I s
fj f
fj
fTj
fj f
fj
p n n,
p
T j
=
(
(
(

=
(
(
(

(
(
(

(87)
then applying singular value decomposition (SVD)
principle procedures as previous steps to obtain the
reconfiguredgainmatrixKf.
Si mul at i on Resul t s
The following simulation results for the decoupled
steering subsystem of NPS UUVs confirmed the
effectiveness of the developed integrated AFTCS. The
fault scenario is simulated as a partial actuator faults
with effectiveness factor of 0.75 introduced at the
sampling instants of 50 for the second actuator only.
TheproposedschemeusessixKalmanfilterestimators,
one for actuator group, one for sensor group, one for
each actuator, and one for each sensor. Computer
simulationsareconductedusinginitialstatevectorx(0)
=[0 0 0]
T
. Due to the pure integration in the yaw
channel of the steering subsystem, there is an
eigenvalue of one for the equivalent discrete model.
The desired eigenvalues of the steering subsystem are
chosen as Sj= {0.9158, 0.9094, 1}. Applying the
proposed controller design algorithm to the given
faulty conditions provides the achievable K and Kf
respectivelyas:
K=
(

0.1473 0.1285 0.2971 -

0.0240 0.0953 - 0.1866 -

Kf=
(

0.3567 0.1555 0.1405 -

0.0827 0.0030 0.2136 -

To achieve the same steadystate tracking of the

reference input for the postfault system, simulation
resultsareperformedusing:

(
(
(

= =
0.0005 0.0003
0.0101 0.0064
0.0095 - 0.0052
K B K B
f
forward f forward

In path tracking, we try to keep a constant heading
angle20degreeasareferencesetpointfortheheading
angle of the steering subsystem. In simulating path
tracking, the yaw rate of turn for the steering path is
made to follow a reference input shown in figure 4.
Figure 5 demonstrates the system output responses to
faultfree system, the faulty system without controller
reconfiguration case that declares the loss of control
effect due to the simulated partial actuator fault, and
finally, the postfault system response with controller
reconfigurationtoshowtheeffectivenessoftheapplied
reconfiguration algorithm that recovers the prefault
system both transient and steady state performances.
The constrained Kalman Filter estimator for the
actuators group has detected a fault at the sampling
instants of 50 by tracking the actual faulty response of
the system but the estimator for sensors group hasnt
detectedanyfaultsandthatisemphasizedthroughthe
generated residual signals for actuators group and
sensors group shown in figure 6 and figure 7
respectively. The residual signals of sensor group are
close to zero depending on available measurement
noise and system disturbances. To isolate the faulty
actuator, the generated residual signals for the first
actuatorandthesecondactuatoraredeclaredinfigure
8 and figure 9 respectively from which it is noted that
thesecondactuatorisresponsibleforthedetectedfault
and that can be modeled as loss in control with
effectiveness factor a2. Through the introduced fault
estimation technique, the last estimate of the loss of
control effectiveness factor i.e. a2 is 0.74995 that is
close to the simulated applied partial actuator fault.
Figure 10 shows the rudder control signals provided
through the designed reconfigurable state feedback
controller with reference input in which it can be seen
InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.org
49
thatthereconfigurablemechanismisactivatedafterthe
fault has been detected, isolated and estimated using
the integrated fault detection and diagnosis (FDD)
scheme.

FIG.4REFERENCEINPUTFORTHEYAWRATEOFTURNFOR
THESTEERINGPATH

A. THESYSTEMOUTPUTRESPONSE(Y1)

B. THESYSTEMOUTPUTRESPONSE(Y2)
FIG.5THESYSTEMOUTPUTRESPONSESOFFAULTFREEAND
FAULTYSYSTEMWITHANDWITHOUTRECONFIGURATION

A. THERESIDUALFORTHEOUTPUT(Y1)

B. THERESIDUALFORTHEOUTPUT(Y2)
FIG.6THERESIDUALSIGNALSFORACTUATORGROUP

A. THERESIDUALFORTHEOUTPUT(Y1)

B. THERESIDUALFORTHEOUTPUT(Y2)
FIG.7THERESIDUALSIGNALSFORSENSORGROUP

A. THERESIDUALFORTHEOUTPUT(Y1)

B. THERESIDUALFORTHEOUTPUT(Y2)
FIG.8THERESIDUALSIGNALSFORTHEFIRSTACTUATOR
www.ijape.orgInternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013
50

A. THERESIDUALFORTHEOUTPUT(Y1)

B. THERESIDUALFORTHEOUTPUT(Y2)
FIG.9THERESIDUALSIGNALSFORTHESECONDACTUATOR

A. BOWRUDDERSCONTROLSIGNAL

B. STERNRUDDERSCONTROLSIGNAL
FIG.10THERUDDERSCONTROLSIGNALS
Conc l usi on
This paper presents an active faulttolerant control
system(AFTC)toaccommodatepartialactuatorand/or
sensor faults of Unmanned Underwater Vehicles
(UUVs). The induced systematic approach provides
sensitive and robust fault detection and diagnosis
(FDD)system,areconfigurablecontroldesigninaway
that will preserve the stability of the system,
combination of control reconfiguration and FDD
system,ashortandacceptabletimeperiodasbothfault
detection and isolation processes are achieved
simultaneously as well as a computation load
reductionforlargenumberofactuatorsandsensorsin
thepresenceofsystemdisturbancesandrandomnoise.
The developed flexible approach based on Singular
ValueDecomposition(SVD)solutionforeigenstructure
assignment (EA) technique offers a powerful way to
select robust reconfigurable control based on known
statespace models of a dynamic system. The
performances of the control reconfiguration approach
are emphasized by simulation results of the faultfree
case, the faulty case without reconfiguration and fault
accommodationwithcontrollerreconfiguration.
REFERENCES
A.Alessandri,,M.Caccia,G.Veruggio,Modelbasedfault
detection for unmanned underwater vehicles.
Safeprocess+97,Hull,UK,Vol.2,pp.615620,1997.
A. Alessandri, M. Caccia, and G. Veruggio, Fault detection
of actuator faults in unmanned underwater vehicles,
IFACControlEngineeringPractice,vol.7,no.3,pp.357
368,1999.
A. Esna Ashari, A. Khaki Sedigh, M. J. Yazdanpanah,
Reconfigurable control system design using
eigenstructure assignment: static,dynamic, and robust
approaches,Int.J.ofControl,78,pp.10051016,2005.
A. Healey, and D. Lienard, Multivariable Sliding Mode
Control for Autonomous Diving and Steering of
Unmanned Underwater Vehicles, IEEE Journal of
OceanicEngineering,Vol.18,No.3,pp.327339,1993.
B. Jalving, The NDREAUV Flight Control System, IEEE
Journal of Oceanic Engineering, Vol. 19, No. 4, pp. 497
501,October1994.
D. Krokavec, Eigenstructure assignment in reconfigurable
control system design, In Proceedings of the
International Conference Cybernetics and Informatics
2005,DolnKubn,SlovakRepublic,pp.8491,2005.
E.Larson,B.Parker,andB.Clark,Modelbasedsensorand
actuatorfaultdetectionandisolation,InProceedingsof
the American Control Conference, Anchorage, AK, pp.
42154219,2002.
G. Liu, D. Wang, and Y. Li, Active fault tolerant control
with actuation reconfiguration, IEEE Trans. Aerospace
and Electronic Systems, vol. 40, No. 3, pp. 11101117,
2004.
J. Jiang, Faulttolerant control systemsAn introductory
overview, Automatica SINCA, vol. 31, No.1, pp. 161
174,2005.
InternationalJournalofAutomationandPowerEngineering(IJAPE)Volume2Issue3,March2013www.ijape.org
51
M. Caccia, & G. Veruggio, Guidance and control of a
reconfigurable unmanned underwater vehicle, Control
EngineeringPractice,8(1),2137,2000.
Podder T.K. & Sarkar N., FaultTolerant Control of an
Autonomous Underwater Vehicle Under Thruster
Redundancy, Journal of Robotics and Autonomous
System,Vol.34,pp.3952,2001.
S., Katipamula, and M. R. Brambley,Methods for Fault
Detection, Diagnostics and Prognostics for Building
SystemsAReviewPartI,HVAC&RResearch,Vol.11,
No.1,pp.325,2005.
S., Katipamula, and M. R. Brambley, Methods for Fault
Detection, Diagnostics and Prognostics for Building
Systems A Review Part II, HVAC & R Research, Vol.
11,No.2,pp.169187,2005.
SNAME, The society of naval architects and marine
engineers, Nomenclature for treating the motion of a
submergedbodythroughafluid,TechnicalandResearch
Bulletin,Vol.15,1950.
T. Didier, J. Cdric, Y. M. Zhang, Actuator fault tolerant
control design based on a reconfigurable reference
input,Int.J.Appl.Math.Comput.Sci.,vol.18,No.4,pp.
553560,2008.DOI:10.2478/v1000600800481

T.I.Fossen,GuidanceandControlofOceanVehicles,John
WileyandSonsLtd.NewYork,1994.
T.I.Fossen,MarineControlSystems.Guidance,Navigation,
and Control of Ships, Righs and Underwater Vehicles,
Trondheim,Norway:MarineCybernetics,2002.
Y. M. Zhang and J. Jiang, An active fault tolerant control
systemagainstpartialactuatorfailures,IEEProceedings
Control Theory and Applications, vol. 149, No. 1, pp.
95104,2002.
Y. M. Zhang and J. Jiang, Integrated Design of
Reconfigurable FaultTolerant Control Systems, Journal
of Guidance, Control, and Dynamics, vol. 24, No.1, pp.
133136,2001.
Y. M. Zhang and J. Jiang, Fault Tolerant Control System
Design with Explicit Consideration of Performance
Degradation, IEEE Trans. on Aerospace and Electronic
Systems,vol.39,No.3,pp.838848,2003.
Y. M. Zhang and J. Jiang, Bibliographical review on
reconfigurable fault tolerant control systems, Annual
ReviewsinControl.32,pp.229252,2008.