Vous êtes sur la page 1sur 10

708 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO.

2, FEBRUARY 2014
Speed Sensorless and Sensor-Fault Tolerant Optimal
PI Regulator for Networked DC Motor System With
Unknown Time-Delay and Packet Dropout
Ali-Akbar Ahmadi, Farzad R. Salmasi, Senior Member, IEEE, Mojtaba Noori-Manzar, and
Tooraj Abbasian Najafabadi
AbstractSensorless and sensor-fault-resilient control of a net-
worked dc motor system (NDCMS) with an optimal integral-
square-error proportionalintegral (PI) controller is considered,
while network-induced delays and packet dropouts are taken into
account. A sliding-mode observer is developed to estimate rotor
speed and unknown load torque for the networked system. Then,
a PI controller is designed such that the overall NDCMS with com-
plete or partial sensor failure is stabilized and a linear quadratic
cost function is sufciently minimized. Optimal controller pa-
rameters are determined by solving bilinear matrix inequalities.
The numerical and experimental tests are performed to evaluate
the feasibility and applicability of the networked sensorless or
sensor-fault-tolerant controller. The results show good perfor-
mance in both estimation and control objectives.
Index TermsFault tolerant, linear quadratic regulator (LQR),
network-induced delay, networked control systems (NCSs), packet
dropout, proportionalintegral (PI) controller, sliding-mode ob-
server, speed sensorless.
NOMENCLATURE
NDCMS Networked dc motor system.
LMI Linear matrix inequality.
BMI Bilinear matrix inequality.
LQR Linear quadratic regulator.
USB Universal serial bus.
PID Proportionalintegralderivative.
PWM Pulsewidth modulation.
I
n
(:, i) (i)th column of I
n
.
I
n
n n unitary matrix.
u(t) Heaviside step function.
i
a
Armature current of the dc motor.
Angular speed of the dc motor.
v Armature winding input voltage.
r
a
Armature winding resistance.
l
a
Armature winding inductance.
Manuscript received September 24, 2012; revised December 26, 2012;
accepted February 22, 2013. Date of publication March 15, 2013; date of
current version August 9, 2013.
A.-A. Ahmadi, F. R. Salmasi, and T. A. Najafabadi are with the School of
Electrical and Computer Engineering, Faculty of Engineering, University of
Tehran, Tehran 14395, Iran (e-mail: a.ahmadi@ut.ac.ir; farzad_rs@ieee.org;
najafabadi@ut.ac.ir).
M. Noori-Manzar is with the Department of Electrical Engineering,
K. N. Toosi University of Technology, Tehran 14317, Iran (e-mail:
m.noorimanzar@mail.kntu.ac.ir).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TIE.2013.2253073
k
b
Back electromotive force constant.
k Torque constant.
b Damping coefcient.
f Motor angular moment of inertia.
T
l
External disturbance or load torque.
I. INTRODUCTION
I
NNETWORKEDcontrol systems (NCSs), the control loops
are closed via a communication network, i.e., the data
exchange between different parts of the control loop is per-
formed through the network. NCSs have several benets in
comparison with conventional control systems such as reduced
system wiring, easy maintenance, and increase system agility.
For those reasons, NCSs have been used in various elds such
as dc motors [1], mobile sensor networks, remote surgery,
automated highway systems, and unmanned aerial vehicles
[2]. Nevertheless, insertion of the communication network in
control loop makes system analysis and design more com-
plicated. This is arisen from some network inherent sophis-
tications such as network-induced delays and packet losses.
For this reason, many reports have been proposed for mod-
eling, stability analysis, and controller design for NCSs (e.g.,
see [3][9]). Most of these approaches are very conservative,
and controller performance has not been seriously taken into
account. In [10], it is stated that the NCS performance de-
grades when the network quality-of-service changes. There-
fore, networked proportionalintegral (PI) controller gains are
adapted to compensate for these setbacks. Suboptimal mixed
H
2
/H

controller design for NCSs is considered in [11],


in which a quadratic performance index and a disturbance
rejection constraint are minimized simultaneously. In [12],
NCS is modeled as a discrete-time switched system, and then,
optimal stabilizing controller gain is selected using a heuristic
search approach, so-called estimation of distribution algorithm
(EDA). The results are implemented on an NDCMS. Robust
static output feedback (SOF) controller design problem for
discrete-time NCSs is considered in [13]. It is assumed that
the system matrices are unknown but lie in a convex polytope.
The controller gain is obtained by solving a set of LMIs. Also,
using augmentation method, a remote PID controller design
problem is formulated for designing the SOF controller. The
proposed approach is implemented on an NDCMS. In [28],
measurement and actuation delays in NCSs are modeled by two
0278-0046 2013 IEEE
AHMADI et al.: SPEED SENSORLESS AND SENSOR-FAULT TOLERANT OPTIMAL PI REGULATOR FOR NDCM 709
mutually independent random variables with Bernoulli binary
distribution. Then, a less conservative observer-based controller
for networked systems is developed by using full information of
the delays in the design process.
On the other hand, electric motor drives are the cornerstones
of industrial systems. If these drives are controlled through
remote stations, then the inherent problems with networked sys-
tems such as time-varying delays and packet dropout degrade or
even destabilize the overall system, if conventional controllers
are utilized. In a networked motor drive system, the controller
or regulator is assumed to reside on a remote station, but the
power inverter is located close to the electric motor. There are
several reports on NDCMS (e.g., see [14], [15], [29], [30]).
In [14], a sufcient condition for asymptotic stability and a
necessary condition for zero steady-state tracking error for an
NDCMS are obtained. It is assumed that full-state measurement
is available and no disturbance exists in the system. Gain-
scheduling-based state feedback integral control for NCS is
investigated in [15] and implemented on an NDCMS. Integral
action is employed for disturbance rejection. Output feedback
case is also introduced when the full-state measurement of the
system is not available. Moreover, mean square error, percent
overshoot, and settling time of the rotor angular speed response
are improved using EDA.
It should be noted that the design problem suffers from more
complexity if the speed sensor failure is taken into account.
During the recent decades, speed or position sensorless electric
motor drives have gained most of the attentions and research
works among motor drive scholars [16], [17]. Partial speed
sensor failure has been under investigation too [18]. Edwards
and Tan developed sliding-mode observers for sensor-fault-
tolerant control of conventional systems [19]. A sliding-mode
observer for linear systems with unknown time-varying delay
and external disturbance is discussed in [20]. Furthermore,
a reliable observer-based control against sensor failures for
systems with time delays in both state and input is considered
in [21], which combines the results of observer and stabilizing
controller design. In [22], reliable control for NCSs against both
probabilistic sensor and actuator faults is tackled. Liu and Xia
used likelihood ratios for fault detection and compensation in
networked predictive control systems with random network-
induced time delays [23].
In this paper, a speed sensorless or sensor-fault-resilient
optimal controller is proposed for NDCMS with unknown time
delay, packet dropout, and unknown load torque. Moreover, PI
controller parameters are obtained by solving BMIs, not only
to stabilize the overall closed-loop system but also to minimize
a linear quadratic or integral-square-error cost function. Thus,
two critical problems in industrial networked dc motor drive
systems are tackled: speed sensor fault/failure, or exclusion
from the drive system for cost reduction, and controller per-
formance improvement, besides overall stability.
The rest of this paper is organized as follows. The model-
ing of the NDCMS and some preliminaries are presented in
Section II. In Section III, a sliding-mode observer with expo-
nential convergence to zero estimation error is presented to esti-
mate the networked motor speed and load torque. In Section IV,
an observer-based optimal PI regulator is designed to stabilize
the overall system while minimizing a linear quadratic cost
function. In Section V, numerical simulation and experimental
results on the networked dc motor are provided to show the
validity of the proposed method. Finally, the conclusion is given
in Section VI.
II. MODELING OF NDSMS
The state equations of an NDCMS is given by
di
a
(t)
dt
=
r
a
l
a
i
a
(t)
k
b
l
a
(t) +
1
l
a
v(i
k
h)
d(t)
dt
=
k
j
i
a
(t)
b
j
(t)
1
j
T
l
(1)
where t [i
k
h +
k
, i
k+1
h +
k+1
), k N. Overall, network-
induced delay at the kth step is denoted by
k
. Both sensor-to-
controller and controller-to-actuator delays are incorporated in
inducing the overall delay, which is assumed to be an unknown
variable. In this conguration, the remote speed controller
sends the packet at i
k
h instant. It takes
k
seconds for the
local dc-motor chopper to receive the packets. The motor states
vary based on the transmitted packets until the next packets are
received by the controller at (i
k+1
h +
k+1
)th second. It should
be noted that packet dropouts are embedded in this model if
> h. Since the overall network delay includes both induced
delays, we can assume that the dedicated sensors measure the
armature current and speed at (t)th second.
With x
a
def
= [i
a
]
T
and assuming that the output of the
speed sensor is given by (t) +
F
(t), where
F
(t) repre-
sents speed sensor fault, the dc-motor state and output equations
are given by
x
a
(t) =A
a
x
a
(t) +B
a
v(i
k
h) +E
a
T
l
y
a
(t) =C
a
x
a
(t) +D
F
(t) (2)
where
A
a
=
_

r
a
l
a

k
b
l
a
k
j

b
j
_
B
a
=
1
l
a
I
2
(:, 1) C
a
= I
2
E
a
=
1
j
D D = I
2
(:, 2).
Remark 1: If the drive system is speed sensorless or, equiv-
alently, the sensor suffers from complete failure, then
F
(t) =
(t), which corresponds to a speed sensor with zero speed
sensor output.
Similar to [24] and with the assumption that the load torque
is an extra state variable with very slow variations compared to
motor time constants and x
def
= [x
T
a
T
l
]
T
, we have
x(t) =Ax(t) +Bv(i
k
h)
y(t) =Cx(t) +D
F
(t) (3)
where
A =
_
A
a
E
a
0 0 0
_
B =
_
B
a
0
_
C =
_
C
a

0
0
_
. (4)
710 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 2, FEBRUARY 2014
III. SPEED ESTIMATION AND RECONSTRUCTION
FOR AN NDCMS
In this section, a sliding-mode observer is designed to esti-
mate the rotor speed and load torque.
The observer structure is given by

x(t) =A x(t) +Bv (i


k
h e

(t)) +G(y(t) y(t)) +B(t)


y(t) =C x(t)
(t) =D
T
y(t) (5)
where the hat-signed variables stand for the estimated ones and
e

(t) denotes the estimation error of network-induced delay.


Furthermore, G and (t) are observer gain matrix and auxil-
iary sliding-mode input, respectively, and will be designed to
guarantee asymptotic stability of the observer. Let e
F
= x x
be the state estimation error. Thus, the error dynamic equations
are determined to be
e
F
(t) = (AGC)e
F
(t) +B(v(i
k
h) v (i
k
h e

(t)))
GD
F
(t) B(t). (6)
The variations of armature voltage controlled by PWM signal
is limited during a time period, or mathematically, the control
signal v(t) is Lipschitz, i.e.,
|v(i
k
h) v (i
k
h e

(t))| |e

(t)| (7)
where is known positive scalar. The error system in (6) is
asymptotically stable if
(AGC)
T
+ (AGC) + < 0
GD = [ 0 0 0 ]
T
(8)
and (t) is selected to be
(t) =
e
i
(t)
|e
i
(t)|
|e

(t)| + (9)
where e
i
(t) = i
a
(t)

i
a
(t) is the armature current estimation
error, is a positive-denite matrix, and denotes an arbitrary
positive scalar. This can be veried by choosing the following
Lyapunov function:
V
o
(t) = e
T
F
(t)e
F
(t). (10)
Therefore, the derivative of the aforementioned function along
with the error system trajectory is as follows:

V
o
(t) =e
T
F
(t)
_
(AGC)
T
+ (AGC)

e
F
(t)
+ 2e
T
F
(t)B(v(i
k
h) v (i
k
h e

(t)))
2e
T
F
(t)GD
F
(t) 2e
T
F
(t)B(t). (11)
Since e
T
F
(t)B = (1/l
a
)e
i
(t) and from (7) and (9) and the
assumptions in (8)

V
o
(t) e
T
F
(t)e
F
(t) +
2
l
a
|e
i
(t)e

(t)|
2
l
a
e
i
(t)(t)
e
T
F
(t)e
F
(t)
2
l
a
|e
i
(t)|
e
T
F
(t)e
F
(t) <

e
F
(t)
2
(12)
where

=
min
(). Consequently, lim
t
[

T
l
] =
[ T
l
].
IV. SPEED SENSORLESS OPTIMAL PI CONTROLLER
DESIGN FOR NETWORKED DC MOTOR
In this section, an optimal sensorless PI controller will be
designed such that the overall networked dc motor is stable and
(t) tracks a step reference input r
s
.
From (1), it is clear that, when (t) tracks the reference
input r
s
in steady state, then the steady-state armature current
is given by i
ass
= (b/k)r
s
+ (1/k)T
l
. With x
ass
def
= [(b/k)r
s
+
(1/k)T
l
r
s
]
T
and e
R
def
= x
a
x
ass
, it follows that
e
R
(t) = A
a
(e
R
(t) +x
ass
) +B
a
v(i
k
h) +E
a
T
l
.
At steady state, e
R
(t) 0 and v(i
k
h) v
ss
, which yields to
v
ss
=
br
a
+kk
b
k
r
s
+
r
a
k
T
l
.
With v(i
k
h)
def
= v
ss
+v
a
(i
k
h), we have
e
R
(t) = A
a
e
R
(t) +B
a
v
a
(i
k
h). (13)
Remark 2: Stabilizing the aforementioned error system via
v
a
(i
k
h) is equivalent to determining controller output v(i
k
h)
such that (t) asymptotically tracks r
s
.
In the following, we aim to design an optimal PI controller
for (13), as follows:
v
a
(t) =k
Po
e

(t) +k
Io
I
e

(t)
=D
T
k
Po
e
R
(t) +k
Io
I
e

(t)
or, equivalently,
v
a
(t (t)) = D
T
k
Po
e
R
(i
k
h) +k
Io
I
e

(i
k
h) (14)
in which (t) = t i
k
h, e

(t)
def
= (t) r
s
, and e
R
will be
introduced later. Moreover, k
Po
and k
Io
are PI controller gains,
respectively. Furthermore, the integral term is updated at each
sample as follows:
I
e

(i
k
h) =
i
k
h
_
0
e

()d
=
i
k
h
_
0
()d (i
k
h)r
s
=I

(i
k
h) (i
k
h)r
s
. (15)
AHMADI et al.: SPEED SENSORLESS AND SENSOR-FAULT TOLERANT OPTIMAL PI REGULATOR FOR NDCM 711
Remark 3: The proportional part of the controller is applied
only on the speed tracking error, unlike the conventional state
feedback PI controller in [15], in which the proportional gain
vector is applied to the vector of speed and current tracking
errors.
Remark 4: The commanded armature voltage is given by
v(i
k
h) = v
a
(i
k
h) +
_
br
a
+kk
b
k
r
s
+
r
a
k

T
l
(i
k
h)
_
in which the bracket term plays the rule of a feedforward
controller to improve the transient response.
Without loss of generality and for simplicity of the design
process, we assume that the reference speed and load torque are
zero. Therefore, if the estimation error vector e
F
is decomposed
as e
F
= [e
T
E
e
T
L
]
T
, where e
E
and e
T
l
denote the current-
speed estimation error vector and load torque estimation error,
respectively, then the estimated tracking error vector is given by
e
R
(t) = e
R
e
E
.
By substituting (14) in (13), we have
e
R
(t) =A
a
e
R
(t) +k
Po
B
a
D
T
e
R
(i
k
h) +k
Io
B
a
I
e
w
(i
k
h)
=A
a
e
R
(t) +k
Po
B
a
D
T
e
R
(i
k
h) k
Po
B
a
D
T
e
E
(i
k
h)
+k
Io
B
a
I
e

(i
k
h),
t [i
k
h +
k
, i
k+1
h +
k+1
), k N. (16)
Let E
R
(t) = [e
T
R
(t) I
e

(t)]
T
be the state vector of the aug-
mented system. Since

I
e

(t) = e

(t) = D
T
e
R
, thus, from
(16), the dynamic of the augmented system is obtained as
follows:

E
R
(t) =

AE
R
(t) +BKE
R
(i
k
h) W
E
(t, i
k
h),
t [i
k
h +
k
, i
k+1
h +
k+1
), k N (17)
where

A =
_
A
a
0
D
T
0
_
B =
_
B
a
0
_
K =[ D
T
k
po
k
Io
]
W =
_
B
a
D
T
k
Po
0
0 D
T
_

E
(t, i
k
h) = [ e
T
E
(i
k
h) e
T
E
(t) ]
T
. (18)
We consider the following quadratic cost function associated
with (17):
J =

_
0
C
F
(t)dt
C
F
(t) =E
T
R
(t)QE
R
(t) +v
T
a
(t)Rv
a
(t) (19)
where Q and R are positive-denite matrices with appropriate
dimensions. The following theorem gives the procedure for
nding matrix K such that it makes (17) asymptotically stable
and suboptimally minimizes the cost function in (19).
Theorem 1: For given positive scalar and positive-denite
matrices Q and R, if positive-denite matrices P and and
matrices M
i
and N
i
(i = 1, 2, 3) with appropriate dimensions
exist and minimize any positive scalar C
0
as follows:
min
P>0,>0,M
i
,N
i
,K
C
0
st. < 0 (20)
_
C
0
E
T
R
(0)(P + )
(P + )E
R
(0) (P + )
_
< 0 (21)
where is given in the Appendix, then controller gain matrix
K guarantees overall asymptotical stability and minimizes the
cost function in (19).
Proof: Inspired by [8] and [21], the following Lyapunov
functional for (17) is considered:
V
T
(t) = V
c
(t) +g
o
V
o
(t) (22)
where
V
c
(t) = E
T
R
(t)PE
R
(t) +
t
_
t
t
_
s

E
T
R
(v)

E
R
(v)dvds (23)
where P and are positive-denite matrices and g
o
is positive
scalar. V
o
is dened in (10). Differentiating (23) along the
trajectories of (17) and performing the same tasks as [8] yields

V
c
(t) E
R
T
(t)
1
E
R
(t) +
0
_
_
E
R
(t)
_
_

E
(t) (24)
where E
R
(t) = [E
T
R
(t) E
T
R
(i
k
h)

E
T
R
(t)]
T
. The other vari-
ables are dened in the Appendix. From (12) and (24), we have

V
T
(t) =

V
c
(t) +g
o

V
o
(t)
E
R
T
(t)
1
E
R
(t) +
0
_
_
E
R
(t)
_
_

E
(t)
g
o

e
F
(t)
2
. (25)
Adding and subtracting C
F
(t) in the right-hand side of (25)
follow that

V
T
(t) E
R
T
(t)E
R
(t) +
0
_
_
E
R
(t)
_
_

E
(t)
g
o

e
F
(t)
2
+
1

E
(t)
2
C
F
(t) (26)
where and
1
are also given in the Appendix.
Note that the vector
E
() converges to zero exponentially;
therefore, one can nd an appropriate positive scalar
2
such
that
E
(t)
2
e
F
(t). Thus, (26) becomes

V
T
(t) E
R
T
(t)E
R
(t) +
0

2
_
_
E
R
(t)
_
_
e
F
(t)
(g
o

2
) e
F
(t)
2
C
F
(t).
If < 0, we have

V
T
(t)

_
_
E
R
(t)
_
_
2
+
0

2
_
_
E
R
(t)
_
_
e
F
(t)
(g
o

2
) e
F
(t)
2
C
F
(t)
712 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 2, FEBRUARY 2014
where

=
min
(). By the selection of g
o
such that
(g
o

2
)

(
0

2
)
2
we have

V
T
(t)

_
_
E
R
(t)
_
_
2
+
_
(g
o

2
)

_
_
E
R
(t)
_
_
e
F
(t)
(g
o

2
) e
F
(t)
2
C
F
(t)

2
_
_
E
R
(t)
_
_
2

(g
o

2
)
2
e
F
(t)
2
C
F
(t) < 0. (27)
Thus, (17) is asymptotically stable. Using Schur complements
[25], < 0 is equivalent to (20).
From (27), we have

V
T
(t) < C
F
(t). (28)
Integrating both sides of (28) from zero to t
f
yields
t
f
_
0
C
F
(t)dt < V
T
(0) V
T
(t
f
).
Since (17) is asymptotically stable, lim
t
f

V
T
(t
f
) = 0.
Therefore,

_
0
_
E
T
R
(t)QE
R
(t) +v
T
a
(t)Rv
a
(t)
_
dt < V
T
(0). (29)
Thus, K should be chosen such that V
T
(0) becomes sufciently
small. Since the observer is designed independently, V
o
(0) is
not taken into account for the minimization of V
T
(0). The rst
term of V
c
(0) is E
T
R
(0)PE
R
(0), and the second term of V
c
(t) at
t = 0 can be approximated by E
T
R
(0)E
R
(0) +O(E
T
R
E
R
(t)),
where O() represents the higher order terms. The stability
of the closed-loop system forces the second term to tend
to zero fast enough. Therefore, it would sufce to minimize
E
T
R
(0)(P + )E
R
(0). Note that using Schur complements (21)
is equivalent to E
T
R
(0)(P + )E
R
(0) < C
0
. Thus, the mini-
mization of C
0
in (21) yields to optimal PI controller gains.
This completes the proof.
Remark 5: Note that (20) is a BMI. There are some powerful
tools for solving BMIs. In this paper, YALMIP parser [26] and
PENBMI solver of the TOMLAB optimization environment in
MATLAB [27] are used for this end.
Remark 6: With the assumption that the observer initial
conditions are zero, E
R
(0) contains the tracking error at t = 0,
i.e., the difference between the armature current and angular
speed of the motor and their corresponding steady-state values
just before the next set point is commanded.
Remark 7: As dened in the Appendix, the positive scalar
that appears in the matrix is an upper bound on the
unknown network-induced delay and data packet dropout in
data transmission through the network. Note that Theorem 1
gives the sufcient conditions for the existence of an suboptimal
TABLE I
PARAMETERS OF THE DC MOTOR
observer-based PI controller for an NDCMS for a given . This
means that the designed controller for a given may tolerate
the bigger network-induced delay and/or data packet dropout.
In the computer simulation and experimental tests, the effect
of increasing on the controller performance is evaluated as
reported in the next section.
V. SIMULATION AND EXPERIMENTAL RESULTS
In this section, numerical simulations and experimental
tests are discussed to evaluate the feasibility of the proposed
approach.
A. Numerical Simulation
Consider a dc motor with the parameters given in Table I.
With these parameters, the augmented system matrices in (4)
are given as follows:
A =
_
_
1436.46 42.54 0
4666.67 22.22 2777.78
0 0 0
_
_
B =
_
_
552.49
0
0
_
_
C =
_
1 0 0
0 1 0
_
.
In the simulation runs, h is chosen to be 0.02 sec, and the
network-induced delays are random numbers between [0 3h].
Moreover, data loss rate between the sensor and actuator is 5%,
and is selected to be 0.08. The stochastic behavior of the
mentioned network-induced delays and data packet dropouts
will be described in the experimental test subsection. First,
we impose the following sinusoidal fault on the angular speed
sensor:
f(t) = 10u(t 16.5) cos(0.45t)
where the load torque is T
l
= 0.14u(t 38). Moreover, the
reference input r
s
= 85u(t) 10u(t 22).
Speed sensor-fault-tolerant optimal PI controller is designed
based on the proposed approach. It is assumed that no
AHMADI et al.: SPEED SENSORLESS AND SENSOR-FAULT TOLERANT OPTIMAL PI REGULATOR FOR NDCM 713
Fig. 1. Reference, faulty, actual, and estimated angular speeds of the net-
worked dc motor in the presence of sensor fault and disturbance with reliable
observer-based PI controller. (Dashed line) Reference. (Solid line) Faulty.
(Dashdot line) Actual. (Dotted line) Estimated.
estimation of network-induced delays are available; thus, we
choose (t) = 0. With = 0.05, the sliding-mode observer
gain is chosen to be = 3.05. Moreover, the observer gain
matrix is determined as
G =
_
2 1 3
0 0 0
_
T
.
With Q = 4I and R = 2.7 and solving the optimization
problem which is given in Theorem 1 (see also Remark
5), the optimal controller gains are determined to be k
Po
=
0.0144 10
3
and k
Io
= 0.0496. The actual, estimated,
and faulty angular speeds of the motor are shown in Fig. 1. The
proposed observer converges to the real value of the angular
speed in the presence of the sensor fault. The optimal observer-
based PI controller rejects unknown time delay, accommodates
the sensor fault, where an unknown load torque is applied to the
motor, and tracks the set points.
Speed sensor fault and its reconstruction are depicted in
Fig. 2. It is clear that sensor fault is reconstructed with good
precision.
Now, the performance of the designed sensor-fault-resilient
LQR controller is compared to a nonoptimal stabilizing net-
worked state feedback controller one. Note that if the cost func-
tion in (19) is not taken into account, a BMI should be solved
that gives the nonoptimal controller gains such as the ones in
[8]. To perform a fair comparison, conventional state feedback
integral version of the proposed controller in [8] is considered.
In this case, the obtained controller gains are determined to
be k
P
= [0.8753 0.1026] 10
3
and k
Io
= 0.0131.
Note that, in this case, there is no restriction on the proportional
part of the controller gain, unlike the proposed approach in
this paper. The simulation results in Fig. 3 demonstrate the
improved performance of the proposed optimal controller in
comparison with the former approach.
In the next simulation run, the networked sensorless dc motor
is studied, in which trapezoidal reference speed is commanded
and a load disturbance is applied at t = 60 sec. Reference,
faulty, actual, and estimated speeds of the dc motor are shown
Fig. 2. Speed sensor fault and its estimation. (Dashed line) Actual. (Solid line)
Reconstructed.
Fig. 3. Reference and actual angular speeds of the dc motor with LQR and
conventional controller. (Dashed line) Reference. (Solid line) Actual speed
with LQR controller. (Dashdot line) Actual speed with conventional stabilizing
networked state feedback controller.
in Fig. 4. It is clear that the networked sensorless motor speed
tracks the reference prole.
To investigate the robustness of the designed controller
against network-induced delay and data packet dropout, two
network scenarios are considered. First, network-induced de-
lays are assumed to be random numbers between [0 4h] and
data loss rate is 10% (case 1). Then, network-induced delays are
assumed to be random numbers between [0 5h], in addition
to 15% packet dropout rate. The previous designed controller
gains are used for simulation in these two cases. The results are
depicted in Fig. 5. It is clear that the designed controller could
keep its healthy operation in steady state; however, the transient
responses have been degraded. Performance degradations are
due to increased network delay and data loss rate.
B. Experimental Tests
To show the effectiveness of the proposed method, a net-
worked dc-motor setup is developed, as shown in Fig. 6. The
714 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 2, FEBRUARY 2014
Fig. 4. Reference, faulty, actual, and estimated angular speeds of the dc motor
in the presence of sensor failure and disturbance with reliable observer-based
PI controller for trapezoidal reference speed. (Dashed line) Reference. (Solid
line) Faulty. (Dashdot line) Actual. (Dotted line) Estimated.
Fig. 5. Reference and actual angular speed of the dc motor in presence
of sensor failure and disturbance with reliable observer-based proportional-
integral controller for trapezoidal reference speed. (Dashed line) Reference.
(Solid line) Case 1. (Dashdot line) Case 2.
Fig. 6. Conguration of the networked dc motor.
setup includes a dc motor (with the given parameters in Table I),
microcontroller-based board, power board, simulated network,
remote controller, and a personal computer (PC). A 360-pulse
encoder and Hall-effect current sensor ACS712 are used for
speed and armature current measurements.
The three main tasks of the microcontroller-based board are
as follows: 1) Preprocessing and sending the measured data
to the PC; 2) receiving the control input from the PC and
generating PWM signal; and 3) sending the PWM signal to
the power board. A USB module is used for data exchanging
between the microcontroller board and the PC. The power
Fig. 7. Histogram of RTT.
Fig. 8. Data packet dropout incidents.
board receives the PWM signal from the microcontroller board
and then drives the dc motor. The simulated network and remote
controller are implemented in MATLABenvironment on the PC
running windows 7.
Since using PC and serial to USB module imposes unknown
delays on the experimental tests, an extra analysis should be
performed for the overall round-trip-time (RTT) characteristics.
RTT is the duration between the sending and receiving of
a packet to/from the PC by the microcontroller-based board.
For this purpose, a set of uniformly random numbers between
[0 1.8h] is generated in MATLAB and considered to be pri-
mary network-induced delays. RTT is the sum of those primary
delays and time delays imposed by PC and serial to USB mod-
ule. Based on this analysis, a histogram of the RTT for 20 000
data packets is given in Fig. 7. This information is used to set
the upper bound parameter . The obtained distribution has also
been utilized in numerical simulation for RTT generation. Note
that, in numerical simulation, the values of generated RTT are
halved and then used for sensor-to-controller and controller-
to-actuator delays. The packet dropouts are also simulated in
MATLAB. Typical 5% data packet dropouts for 1000 packets
are depicted in Fig. 8. The bars indicate incidents of the data
packet dropouts.
AHMADI et al.: SPEED SENSORLESS AND SENSOR-FAULT TOLERANT OPTIMAL PI REGULATOR FOR NDCM 715
Fig. 9. Reference, faulty, actual, and estimated angular speeds of the dc motor
in the presence of sensor fault and disturbance with reliable observer-based PI
controller. (Dashed line) Reference. (Solid line) Faulty. (Dashdot line) Actual.
(Dotted line) Estimated.
Fig. 10. Speed sensor fault and its estimation. (Dashed line) Actual. (Solid
line) Reconstructed.
Determined observer and controller gains in the last subsec-
tion along with the same load torque are used for experimental
evaluations. In the rst experiment, robustness against sensor
fault is evaluated. As shown in Fig. 9, motor speed tracks the
reference input in the presence of the imposed sensor fault.
Estimated sensor fault is given in Fig. 10, as compared to the
imposed one.
In the next experimental test, speed sensor is removed, and
the networked sensorless motor drive is studied. The proposed
observer reconstructs the actual motor speed, as shown in
Fig. 11. Then, in Fig. 12, the performance of the proposed
sensorless optimal PI controller is compared with the conven-
tional networked PI controller. As shown in Fig. 12, the optimal
controller benets from superior transient response.
The next experiment is devoted to the networked sensorless
tracking of a trapezoidal reference speed. As shown in Fig. 13,
good tracking performance is achieved with the proposed
scheme.
Fig. 11. Reference, faulty, actual, and estimated angular speeds of the dc
motor in the presence of sensor failure and disturbance with reliable observer-
based PI controller. (Dashed line) Reference. (Solid line) Faulty. (Dashdot line)
Actual. (Dotted line) Estimated.
Fig. 12. Reference and actual angular speeds of the dc motor with LQR
and conventional controller in sensor failure case. (Dashed line) Reference.
(Solid line) Actual speed with LQR controller. (Dashdot line) Actual speed
with conventional controller.
Fig. 13. Reference, faulty, actual, and estimated angular speeds of the dc
motor in the presence of sensor failure and disturbance with reliable observer-
based PI controller for trapezoidal reference speed. (Dashed line) Reference.
(Solid line) Faulty. (Dashdot line) Actual. (Dotted line) Estimated.
716 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 2, FEBRUARY 2014
Fig. 14. Reference and actual angular speeds of the dc motor in the presence
of sensor failure and disturbance with reliable observer-based PI controller
for trapezoidal reference speed. (Dashed line) Reference. (Solid line) Case 1.
(Dashdot line) Case 2.
In the last experiment, the robustness of the proposed con-
troller against the elevated network delay and packet dropouts
is evaluated, as described by case 1 and case 2 in the last
subsection. From Fig. 14, it is concluded that, although the
performance degrades with new scenarios, the designed con-
troller demonstrates reasonable performance under unforeseen
network conditions.
VI. CONCLUSION
A speed sensor-fault-resilient networked dc motor in the
presence of network-induced delays, packet dropouts, and un-
known load torque has been developed. Complete failure of
speed sensor or, in other words, speed sensorless scheme is also
covered with the proposed approach. Furthermore, an optimal
PI regulator is introduced for the networked system. Simu-
lation and experimental verications show that the proposed
observer/controller is capable of dumping the adverse effects
of unknown and time-varying delays and packet dropouts.
APPENDIX
Variables in proof of Theorem 1 are as follows:
=
_

11

12

13

14

22

23

24

33

34

44
_

11
=N
1
+N
T
1
M
1
AA
T
M
T
1
+Q

12
=N
T
2
N
1
A
T
M
T
2
M
1
BK

13
=N
T
3
A
T
M
T
3
+M
1
+P
14
= N
1

22
= N
2
N
T
2
M
2
BK K
T
B
T
M
T
2
+ 2K
T
RK

23
= N
T
3
+M
2
K
T
B
T
M
T
3

24
=N
2

33
= M
3
+M
T
3
+

34
=N
3

44
=
where is an upper bound for the network-induced delay and
data packet dropout, i.e., (i
k+1
i
k
)h +
k+1

0
=2MW, M =
_
M
T
1
M
T
2
M
T
3

1
=
_
_

11
Q
12

13

22
2K
T
RK
23

33
_
_
+N
1
N
T
N =
_
N
T
1
N
T
2
N
T
3

T
=
_
_

11

12

13

22

23

33
_
_
+N
1
N
T

1
=2
max
(H
T
RH). H = [D
T
k
Po
0].
REFERENCES
[1] S. Chai, G. P. Liu, D. Rees, and Y. Xia, Design and practical implemen-
tation of Internet-based predictive control of a servo system, IEEE Trans.
Control Syst. Technol., vol. 16, no. 1, pp. 158168, Jan. 2008.
[2] J. P. Hespanha, P. Naghshtabrizi, and Y. Xu, A survey of recent results
in networked control systems, Proc. IEEE, vol. 95, no. 1, pp. 138162,
Jan. 2007.
[3] W. Zhang, M. Branicky, and S. Phillips, Stability of networked control
systems, IEEE Control Syst. Mag., vol. 21, no. 1, pp. 8499, Feb. 2001.
[4] J. Nilsson, Real-time control systems with delays, Ph.D. dissertation,
Dept. Autom. Control, Lund Inst. Technol., Lund, Sweden, Jan. 1998.
[5] A. Hassibi, S. P. Boyd, and J. P. How, Control of asynchronous dynamical
systems with rate constraints on events, in Proc. 38th IEEE Conf. Decis.
Control, Phoenix, AZ, USA, Dec. 1999, pp. 13451351.
[6] S. Hu and Q. Zhu, Stochastic optimal control and analysis of stability of
networked control systems with long delay, Automatica, vol. 39, no. 11,
pp. 18771884, Nov. 2003.
[7] P. Seiler and R. Sengupta, An H

approach to networked control, IEEE


Trans. Autom. Control, vol. 50, no. 3, pp. 356364, Mar. 2005.
[8] D. Yue, Q.-L. Han, and C. Peng, State feedback controller design of
networked control systems, IEEE Trans. Circuits Syst. II, Exp. Briefs,
vol. 51, no. 11, pp. 640644, Nov. 2004.
[9] Y. Tipsuwan and M.-Y. Chow, Control methodologies in networked
control systems, Control Eng. Pract., vol. 11, no. 10, pp. 10991111,
Jun. 2003.
[10] M.-Y. Chow and Y. Tipsuwan, Gain adaptation of networked DC motor
controllers based on QoS variations, IEEE Trans. Ind. Electron., vol. 50,
no. 5, pp. 936943, Oct. 2003.
[11] P. Millan, L. Orihuela, G. Bejarano, C. Vivas, T. Alamo, and F. R. Rubio,
Design and application of suboptimal mixed H
2
/H

controllers for
networked control systems, IEEE Trans. Control Syst. Technol., vol. 20,
no. 4, pp. 10571065, Jul. 2012.
[12] H. Li, M.-Y. Chow, and Z. Sun, Optimal stabilizing gain selection for
networked control systems with time delays and packet losses, IEEE
Trans. Control Syst. Technol., vol. 17, no. 5, pp. 11541162, Sep. 2009.
[13] H. Zhang, Y. Shi, and A. S. Mehr, Robust static output feedback control
and remote PID design for networked motor systems, IEEE Trans. Ind.
Electron., vol. 58, no. 12, pp. 53965405, Dec. 2011.
[14] H. Li, M.-Y. Chow, and Z. Sun, EDA-based speed control of a networked
DC motor system with time delays and packet losses, IEEE Trans. Ind.
Electron., vol. 56, no. 5, pp. 17271735, May 2009.
[15] H. Li, Z. Sun, M.-Y. Chow, and F. Sun, Gain-scheduling-based state
feedback integral control for networked control systems, IEEE Trans.
Ind. Electron., vol. 58, no. 6, pp. 24652472, Jun. 2011.
[16] P. Vaclavek and P. Blaha, Lyapunov-function-based ux and speed ob-
server for AC induction motor sensorless control and parameters estima-
tion, IEEE Trans. Ind. Electron., vol. 53, no. 1, pp. 138145, Feb. 2006.
[17] K. Rajajshekara, A. Kawamura, and k. Matsuse, Sensorless Control of AC
Motor Drives. Piscataway, NJ, USA: IEEE Press, 1996.
[18] T. A. Najafabadi, F. R. Salmasi, and P. J. Maralani, Detection and iso-
lation of speed, DC-link voltage, and current sensor faults based on an
adaptive observer in induction motor drives, IEEE Trans. Ind. Electron.,
vol. 58, no. 5, pp. 16621672, May 2011.
[19] C. Edwards and C. P. Tan, Sensor fault tolerant control using sliding mode
observers, Control. Eng. Pract., vol. 14, no. 8, pp. 897908, Aug. 2006.
AHMADI et al.: SPEED SENSORLESS AND SENSOR-FAULT TOLERANT OPTIMAL PI REGULATOR FOR NDCM 717
[20] A. Seuret, T. Floquet, J.-P. Richard, and S. K. Spurgeon, A sliding mode
observer for linear systems with unknown time varying delay, in Proc.
Amer. Control. Conf., New York, NY, USA, Jul. 2007, pp. 45584563.
[21] Z. Gao, T. Breikin, and H. Wang, Reliable observer-based control against
sensor failures for systems with time delays in both state and input, IEEE
Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 38, no. 5, pp. 1018
1029, Sep. 2008.
[22] E. Tian, D. Yue, and C. Peng, Reliable control for networked control sys-
tems with probabilistic sensors and actuators faults, IET Control Theory
Appl., vol. 4, no. 8, pp. 14781488, Aug. 2010.
[23] B. Liu and Y. Xia, Fault detection and compensation for linear systems
over networks with random delays and clock asynchronism, IEEE Trans.
Ind. Electron, vol. 58, no. 9, pp. 43964406, Sep. 2011.
[24] G. S. Buja, R. Menis, and M. I. Valla, Disturbance torque estimation in a
sensorless DC drive, IEEE Trans. Ind. Electron., vol. 42, no. 4, pp. 351
357, Aug. 1995.
[25] S. Boyd, L. E. Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix In-
equalities in System and Control Theory. Philadelphia, PA, USA: SIAM,
1994.
[26] J. Lfberg, YALMIP: A toolbox for modeling and optimization
in MATLAB, in Proc. CACSD. Conf., Taipei, Taiwan, Sep. 2004,
pp. 284289.
[27] K. Holmstrm, Practical optimization with the TOMLAB environment
in MATLAB, in Proc. 42nd SIMS. Conf., Porsgrunn, Norway, Sep. 2001,
pp. 89108.
[28] X. Luan, P. Shi, and F. Liu, Stabilization of networked control systems
with random delays, IEEE Trans. Ind. Electron, vol. 58, no. 9, pp. 4323
4330, Sep. 2011.
[29] A. Onat, T. Naskali, E. Parlakay, and O. Mutluer, Control over imper-
fect networks: Model-based predictive networked control systems, IEEE
Trans. Ind. Electron., vol. 58, no. 3, pp. 905913, Mar. 2011.
[30] W. Zeng and M.-Y. Chow, Optimal tradeoff between performance and se-
curity in networked control systems based on coevolutionary algorithm,
IEEE Trans. Ind. Electron., vol. 59, no. 7, pp. 30163025, Jul. 2012.
Ali-Akbar Ahmadi received the B.S. degree in con-
trol engineering from Amirkabir University of Tech-
nology (Tehran Polytechnic), Tehran, Iran, in 2005
and the M.S. degree in control engineering from
Tarbiat Modares University, Tehran, in 2007. He is
currently working toward the Ph.D. degree in control
engineering in the School of Electrical and Computer
Engineering, Faculty of Engineering, University of
Tehran, Tehran.
His research interests include networked control
systems, fault detection and diagnosis, fault-tolerant
control, and robust adaptive observers and controllers.
Farzad R. Salmasi (S99M02SM09) received
the B.S. degree in electrical engineering from Sharif
University of Technology, Tehran, Iran, in 1994, the
M.Sc. degree in electrical engineering from Amirk-
abir University of Technology, Tehran, in 1997, and
the Ph.D. degree in electrical engineering from Texas
A&MUniversity, College Station, TX, USA, in 2002.
From 1999 to 2002, he was a Research Assistant
with the Advanced Motor Drives and Hybrid Vehi-
cles Laboratory, Department of Electrical engineer-
ing, Texas A&M University. From 2002 to 2004, he
was with the Electro Standards Laboratories, Cranston, RI, USA, as a Research
Scientist. Since September 2004, he has been with the School of Electrical and
Computer Engineering, Faculty of Engineering, University of Tehran, Tehran,
where he is currently the Head of the Control Engineering Department. His
main research areas include design and control of advanced motor drives, power
electronic systems, hybrid electric vehicles, networked control systems, and
smart grids.
Dr. Salmasi is a Technical Committee Member of IASTED on power and
energy systems.
Mojtaba Noori-Manzar received the B.S. degree in
control engineering from the University of Science
and Technology, Tehran, Iran, in 2009 and the M.S.
degree in control engineering from the University of
Tehran, Tehran, in 2012. He is currently working
toward the Ph.D. degree in control engineering at
K. N. Toosi University of Technology, Tehran.
His research interests include networked control
systems, time-delay systems, and electric motors.
Tooraj Abbasian Najafabadi received the B.S. de-
gree in electrical engineering from Esfahan Univer-
sity of Technology, Esfahan, Iran, in 2003 and the
M.S. and Ph.D. degrees in electrical engineering
from the University of Tehran, Tehran, Iran, in 2005
and 2009, respectively.
He is currently an Assistant Professor with the
School of Electrical and Computer Engineering, Uni-
versity of Tehran. His main research areas include
nonlinear and adaptive control, design and control
of advanced motor drives, power electronic systems,
hybrid electric vehicles, and mechatronics.

Vous aimerez peut-être aussi