Professional Documents
Culture Documents
4, JULY 2015
1465
I. I NTRODUCTION
Manuscript received July 16, 2014; accepted November 14, 2014. Date
of publication January 8, 2015; date of current version June 12, 2015.
Manuscript received in final form November 24, 2014. This work was
supported by the Italian Government within the Fund for Investment in
Basic ResearchFuturo in Ricerca 2008 through the NECTAR Project under
Grant RBFR08QWUV. Recommended by Associate Editor X. Zhang.
F. Arrichiello is with the University of Cassino and Southern Lazio,
Cassino 03043, Italy (e-mail: f.arrichiello@unicas.it).
A. Marino is with the University of Salerno, Salerno 84084, Italy (e-mail:
almarino@unisa.it).
F. Pierri is with the University of Basilicata, Potenza 85100, Italy (e-mail:
francesco.pierri@unibas.it).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCST.2014.2377175
1063-6536 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1466
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015
(1)
(2)
(3)
where
vector,
vector,
dynamic uncertainty term.
It is supposed that each robot has access to a noisy
measure x i,m of its own state
x i,m = x i + i
(4)
where i
is the additive noise, assumed to be norm
bounded by a positive scalar
IRn
i
i = 1, 2, . . . , N.
(5)
(6)
1467
N
ai j , li j = ai j , i = j.
j =1, j =i
In
On}
ith node
i
j
y i y + i ym i y + i u
(7)
x = ko
j Ni
t
where ko > 0 is a scalar gain; ym = x m t0 u( )d ,
t
and i y = ix t0 i u(,
i x )d , where t0 is the initial time
i
i
instant; and u = u(t,
i x ) is the estimate of the collective
input elaborated by the i th robot on the basis of its estimate
of the collective state i x , and of the control law detailed
in Section II-C. The i th input ui in (1) is extracted from i u
thanks to the selection matrix i
ui = i i u.
(8)
1468
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015
(9)
2n
y = [1 yT , . . . , N yT ]T IR N
2n
2n
2n
y = 1 N y y IR N
t
with y = x t0 u( )d .
2n
(10)
N
1
xi = J 1 x
N
(i x ) =
i
i
(i x )
(i x )
=
1,d
1
2,d i 2 (i x )
i
(i x )
(11)
i=1
where
2 nN 2 n
y = ko L y + ko .
(17)
1469
j Ni
(21)
(22)
r k = diag{ k , k , . . . , k }r = k ( L y + )
= L k yk + k
(23)
where the vector yk = [1 yTk , . . . , N yTk ]T IR Nn collects the
estimation errors i yk of the observers and L k = L I n + k .
(24)
ko k ( ))d.
(26)
r k = i r k = i L k yk + i k = i hk + i f k
(27)
respectively.
Therefore, from (27)(29), the following can be argued.
1) The residuals i r k for all i = 1, 2, . . . , N (i.e., the
residuals referred to the faulty robot) are affected by
the fault k via the term if k in (29).
2) All the residuals i r l for all i = 1, 2, . . . , N, and for all
l = k (i.e., all the residuals referred to a robot different
to the faulty one) are insensitive to the fault on robot k.
This completes the proof.
Remark 3.1: It is worth remarking that since the residuals
are decoupled in such a way that the fault k affects only
the residuals i r k (i = 1, 2, . . . , N), the proposed scheme
is effective also in the presence of multiple faults affecting
different robots.
The following Lemma for the detection and isolation of the
fault k can be stated.
Lemma 3.2: The fault k , affecting the kth robot, can be
detected and isolated by the robot i if
t > t f : i r k (t) > i k (t)
(30)
l (1, 2, . . . , N), l = k t > 0, i r l (t) i l (t)
where t f > 0 is the instant at which the fault starts and
j (t) (i, j ) are suitable thresholds defined in the following.
1470
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015
(31)
e
0
i k + i L k eko L k t yk (0)
t
ko L k (t )
+ i Lk
e
ko + N d
0
i k yk (0)et
N
k o +
(1 et )
+ i Lk
(32)
(33)
with , > 0 has been exploited being ko L k Hurwitz [14].
Finally, since
i L i (L I n ) + i k
k
n i (L I n ) + i k
n d i + i k
(34)
+ i k = i k (t).
(35)
As far as the detectability condition is concerned, in the
presence of fault, based on (27), the following chain of
inequalities holds:
i r k = i f k + i hk i f k i hk i f k i k (t).
(36)
1471
Fig. 3.
Fig. 1.
Fig. 2. Simulation case study. Paths of the robots during the mission. Crosses
are the initial positions, diamonds the final ones, and circles intermediate
configurations.
Fig. 4. Simulation case study. Plots of the state estimation ( x ), centroid,
and formation task errors ( 1 and 2 ).
Fig. 5. Simulation case study. Solid line: norm of the residual components
relative to robot 1 (a nonfaulty robot) as computed by the different robots.
Dashed line: threshold used for the fault detection.
1472
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015
Fig. 8. Experimental case study. Picture of the four Khepera III robots used
for the experiments.
Fig. 6. Simulation case study. Solid line: norm of the residual components
relative to robot 2 (the faulty robot with null input term for t > 30) as
computed by the different robots. Dashed line: threshold used for the fault
detection. Fault occurred at t = 30 s.
Fig. 7. Simulation case study. Solid line: norm of the residual components
relative to robot 3 (the faulty robot with constant additive constant input term
for t > 20) as computed by the different robots. Dashed line: threshold used
for the fault detection. Fault occurred at t = 20 s.
Fig. 9.
1473
Fig. 10. Experimental case study. Norm of the state estimation error x
(thick line) and of its individual components (thin lines).
Fig. 12. Experimental case study. Solid line: norm of the residual components
i r (i = 1, 2, . . . , N ) as computed by the different robots relative to a
1
nonfaulty robot (robot 1 in this case). Dashed line: threshold used for the
fault detection.
Fig. 13. Experimental case study. Solid line: norm of the residual components
i r (i = 1, 2, . . . , N ) as computed by the different robots relative to the faulty
4
robot (robot 4). Dashed line: threshold used for the fault detection.
1474
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015
for
and thus by collecting the differences u(t) i u(t)
i = 1, 2, . . . , N, it is
!
N
1x
k
k x
k=1
!
N
2
kx
x
k
. (42)
u = 1 N u u = kc
k=1
..
!
N
N
k
x
k x
k=1
) x .
1 N u u = kc (I N 2 n
) = N 2 n rank(
) = N 2 n Nn.
rank(I N 2 n
x = ko L y + u .
(38)
(39)
with u = 1 N u u = [1 u T 2 u T , . . . , N u T ]T IR N n and
d = ko L y + . Provided that the conditions in
Theorem 2.1 and in (2) and (5) hold, the above system can be
seen as a dynamical systems with d as a bounded disturbance
term.
As a consequence, the error x in (39) exponentially reaches
the origin provided the system
2
x = u (t, x ) + d
(40)
is exponentially stable. It is, thus, necessary to find an expression of u as a function of x in the case of the control law
in (14). At this end, it holds
u = u(t) i u(t)
k k u(t)
i u(t)
k=1
= kc
N
J x
i
N
k=1
k J
(44)
(43)
J x
k
(41)
(45)
(47)
i
(49)
x x =
t0
t0
and since i (u i u ) = i (
it holds
t
t0
ud
i (x i x ) = 0
t
t0
i ud
)
= 0, i ,
(50)
1475
[20] N. Meskin and K. Khorasani, Actuator fault detection and isolation for
a network of unmanned vehicles, IEEE Trans. Autom. Control, vol. 54,
no. 4, pp. 835840, Apr. 2009.
[21] N. Meskin, K. Khorasani, and C. A. Rabbath, A hybrid fault detection
and isolation strategy for a network of unmanned vehicles in presence of
large environmental disturbances, IEEE Trans. Control Syst. Technol.,
vol. 18, no. 6, pp. 14221429, Nov. 2010.
[22] R. Olfati-Saber, J. A. Fax, and R. M. Murray, Consensus and cooperation in networked multi-agent systems, Proc. IEEE, vol. 95, no. 1,
pp. 215233, Jan. 2007.
[23] R. Olfati-Saber and R. M. Murray, Consensus problems in networks of
agents with switching topology and time-delays, IEEE Trans. Autom.
Control, vol. 49, no. 9, pp. 15201533, Sep. 2004.
[24] P. Panagi and M. M. Polycarpou, Decentralized fault tolerant control
of a class of interconnected nonlinear systems, IEEE Trans. Autom.
Control, vol. 56, no. 1, pp. 178184, Jan. 2011.
[25] W. Ren and R. W. Beard, Distributed Consensus in Multi-vehicle Cooperative Control (Communications and Control Engineering). Berlin,
Germany: Springer-Verlag, 2008.
[26] W. Ren, R. W. Beard, and E. M. Atkins, Information consensus in
multivehicle cooperative control, IEEE Control Syst. Mag., vol. 27,
no. 2, pp. 7182, Apr. 2007.
[27] I. Shames, A. M. H. Teixeira, H. Sandberg, and K. H. Johansson,
Distributed fault detection for interconnected second-order systems,
Automatica, vol. 47, no. 12, pp. 27572764, 2011.
[28] R. S. Smith and F. Y. Hadaegh, Closed-loop dynamics of cooperative
vehicle formations with parallel estimators and communication, IEEE
Trans. Autom. Control, vol. 52, no. 8, pp. 14041414, Aug. 2007.
[29] D. P. Spanos, R. Olfati-Saber, and R. M. Murray, Dynamic consensus
on mobile networks, in Proc. IFAC World Congr., 2005.
[30] S. Stankovic, N. Ilic, Z. Djurovic, M. Stankovic, and
K. H. Johansson, Consensus based overlapping decentralized fault
detection and isolation, in Proc. Conf. Control Fault-Tolerant Syst.,
Oct. 2010, pp. 570575.
[31] Y. Wang, H. Ye, S. X. Ding, Y. Cheng, P. Zhang, and G. Wang, Fault
detection of networked control systems with limited communication,
Int. J. Control, vol. 82, no. 7, pp. 13441356, 2009.
[32] P. Yang, R. A. Freeman, and K. M. Lynch, Multi-agent coordination
by decentralized estimation and control, IEEE Trans. Autom. Control,
vol. 53, no. 11, pp. 24802496, Dec. 2008.
[33] X. Zhang, Decentralized fault detection for a class of large-scale
nonlinear uncertain systems, in Proc. Amer. Control Conf., Baltimore,
MD, USA, Jun./Jul. 2010, pp. 56505655.
[34] X. Zhang and Q. Zhang, Distributed fault diagnosis in a class of
interconnected nonlinear uncertain systems, Int. J. Control, vol. 85,
no. 11, pp. 16441662, 2012.
1476
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 23, NO. 4, JULY 2015