Fault Diagnosis Using Recursive Least Square Algorithm

for Permanent Magnet Synchronous Motor Drives

Byoung-Gun Park, Rae-Young Kim, and Dong-Seok Hyun
Hanyang University
Dept. of Electrical Engineering
17 Haengdang-dong Seongdong-gu, Seoul, 133-791
Korea
E-mail: redalarm@hanyang.ac.kr
Abstract This paper proposes a novel diagnosis scheme using
parameter estimation of stator resistance, especially, in subject
to the open-phase faults of PMSM drives. The stator resistances
of PMSM are estimated by recursive least square (RLS)
algorithm in real time. The fault diagnosis is achieved by
analyzing the estimated stator resistance of each phase
according to fault condition. The proposed diagnosis scheme is
implemented without any extra devices. Moreover, the
estimated parameter information can be used in order to
improve the control performance. The feasibility of the
proposed fault diagnosis scheme is proved by simulation and
experimental results.
I. INTRODUCTION
A permanent magnet synchronous motor (PMSM) has
been widely used in various applications including electrical
vehicles, appliances, aircraft, and industrial servo drives due
to its high power density and large torque to inertia ratio,
while improvements in the properties of permanent-magnet
materials have increased their viability. The reliability of the
PMSM drives is the one of critical factors in several
industries, in particularly, the areas where the precise
operation and/or high performance are required. In these
areas, a sudden drive failure results in serious damages and
economical losses [1].
Several kinds of the fault occur in the PMSM systems. The
most frequent one has been known as the electrical fault of the
PMSM, such as the open or short circuit of stator or rotor
windings and bearing faults [2]-[8]. Power electronics has
also been considered as weak equipment. The frequent failure
usually originates from the open or short circuit in either
power devices or wire connection.
In recent years, several papers for faults in motor drive
system have been published using the following schemes:
fault detection and identification methods [2]-[8] and fault
tolerant schemes [3],[4]. The phase voltage comparison
method [5] and lower switch voltage method [6] using the
measurement of the voltages could quickly diagnose the
open-circuit fault to reduce the time between the fault
occurrence and diagnosis. It needs additional voltage sensors
to measure phase voltages. The wavelet-fuzzy algorithm [7]
Fig. 1. The electrical equivalent circuit of PMSM drives
and wavelet-neural network method [8] use the wavelet
analysis to detect fault signatures. Wavelet transform is an
emerging DSP algorithm that has variable time and frequency
resolutions. However, these expert systems require a
relatively high computing process. However, most existing
fault detection and identification methods have problems
because the fault detection time takes at least one
fundamental period, the process for detecting the fault is
complex, and schemes to identify the fault are inadequate.
Moreover, this methods use additional sensor for fault
diagnosis. To solve these problems in the novel proposed
fault diagnosis, it is possible to detect and identify the fault
having a single algorithm.
This paper proposes a novel diagnosis scheme that is
presented by the combination of stator resistance variations
approach and recursive least square (RLS) algorithm,
especially, in subject to the open fault of the PMSM drives.
The stator resistances of PMSM are estimated by the RLS
algorithm in real time. If an open-phase fault occurs, the
stator resistance of a faulty phase estimated by the RLS
algorithm is rapidly changed. This characteristic of the stator
resistance offer a simple algorithm to detect open-phase fault.
The proposed diagnosis scheme is implemented without
any extra devices. Moreover, since it uses a simple algorithm
by analyzing only estimated stator resistances of each phase,
the detection speed becomes fast. The feasibility of the
proposed fault diagnosis scheme is proved by several
simulation and experimental results.
8th International Conference on Power Electronics - ECCE Asia
May 30-June 3, 2011, The Shilla Jeju, Korea
978-1-61284-957-7/11/$26.00 2011 IEEE
[WeP1-015]
II. MODELLING OF PMSM
Generally, the PMSM drive system can be modelled as an
electrical equivalent circuit that consists of a resistance, an
inductance, and back-EMF per phase. The electrical
equivalent circuit of PMSM drives is shown in Fig. 1.
Although the conventional d-q motor model obtained through
the transformation of phase voltage model is widely used to
analyze and control AC motor, it cannot be used under open
faults in switching devices since 3-phase balanced condition
is no longer hold under the open fault and it is not easy to
obtain motor input voltage in open phase from the
pole-voltage. Therefore, in the normal condition without
faulty, the dynamic model of 3-phase balanced PMSM is
written as
(
(
(

+
(
(
(

(
(
(

+
(
(
(

(
(
(

=
(
(
(

c
b
a
c
b
a
c
b
a
c
b
a
c
b
a
c
b
a
e
e
e
i
i
i
L
L
L
dt
d
i
i
i
R
R
R
v
v
v
0 0
0 0
0 0
0 0
0 0
0 0
(1)
where i
a
, i
b
, and i
c
are phases currents. v
a
, v
b
, and v
c
are
three-phase terminal voltages. e
a
, e
b
, and e
c
are phase
back-EMFs. R and L are resistance and inductance of a phase
windings.
The phase back-EMFs (e
a
, e
b
, and e
c
) can be approximately
expressed as:
) / cos( e
) / cos( e
cos e
e f e c
e f e b
e f e a
3 2
3 2
t u e
t u e
u e
+ =
=
=
(2)
where
f
,
e
, and
e
represent flux linkage of permanent
magnet, electrical rotor speed, and electrical rotor position.
Equation (3) is derived by (1). This mathematical equation
of a PMSM model is applied to RLS algorithm.
(
(
(

(
(
(
(
(
(
(

+
(
(
(

(
(
(
(
(
(
(

=
(
(
(

c c
b b
a a
c
b
a
c
b
a
c
c
b
b
a
a
c
b
a
e v
e v
e v
L
L
L
i
i
i
L
R
L
R
L
R
i
i
i
dt
d
1
1
1
0 0
0 0
0 0
(3)
III. RECURSIVE LEAST SQUARE ALGORITHM
For digital implementation of the RLS algorithm, the
discrete dynamic model is required. The discrete phase
current is given by
)) k ( e ) k ( v (
T R L
T
) k ( i
T R L
L
) k ( i
a
samp
samp
a
samp
a

+
+
+
= 1
(4)
Fig. 2. Space vector Modulation.
) ( Z ) Y(
T
k k
T
O =
|
|
.
|

\
|
+

=
) ( ) 1 ( ) (
) 1 ( ) ( ) ( ) 1 (
- ) 1 (
1
) (
k Z k P k Z
k P k Z k Z k P
k P k P
T
T

) ( ) 1 ( ) (
) ( ) 1 (
) (
k Z k P k Z
k Z k P
k K
T
+

)) 1 ( ) ( Z - ) )(Y( ( ) 1 ( ) (
T
O + O = O k k k k K k k
Fig. 3. Fault Diagnosis using RLS algorithm.
The recursive least squares (RLS) algorithm mainly has
been used to estimate parameters of the motor due to simple
implementation. The general RLS algorithm is formulated as
follows [9]:
) ( Z ) Y(
T
k

k
T
O = (5)
|
|
.
|

\
|
+

=
) ( ) 1 ( ) (
) 1 ( ) ( ) ( ) 1 (
- ) 1 (
1
) (
k Z k P k Z
k P k Z k Z k P
k P k P
T
T

(6)
) ( ) 1 ( ) (
) ( ) 1 (
) (
k Z k P k Z
k Z k P
k K
T
+

(7)
)) 1 ( ) ( Z - ) )(Y( ( ) 1 ( ) (
T
O + O = O k

k k k K k

(8)
where, Y(k) is the output, is the parameter vector, O

is the
estimated parameter vector(^ denotes the estimated value).
Z(k) is the signal vector, P(k) is the covariance matrix, and
is the forgetting factor given by 0< <1. The forgetting factor
is related to the sensitivity of the parameter variation.
The relation between recursive equation coefficients and
physical parameters can be expressed as follows;
(a) (b)
Fig. 4. Effective voltage vector (a) V1(100) (b) V2(110).
TABLE I
RESISTANCE VARIATION FOR SECTOR 1-6
samp
samp
samp
T R L
T
b ,
T R L
L
a
+
=
+
=
(9)
b
T R b T
a
T R a
L
samp samp samp

=

=
1
(10)
b
T a
L ,
b
a
R
samp

=
1
(11)
From (4), the formulation of the RLS algorithm is obtained
by:
) ( ) Y( k i k
a
= (12)
T
a a a
k e k v k i k ] ) ( ) ( ) 1 ( [ ) Z( = (13)
T
k k a k

] ) b( ) ( [ ) ( = O (14)
IV. PROPOSED FAULT DIAGNOSIS SCHEME
As shown in Fig. 2, the operation region of the PMSM
drive system is divided into the six triangle domains, denoted
from Sector 1 to Sector 6, in the space vector pulse width
modulation (SVPWM) hexagon. The timing flowchart of the
proposed fault detection/ identification scheme is
Fig. 5. Block diagram of the proposed RLS-based fault diagnosis
illustrated in Fig. 3, where stator resistances of PMSM are
estimated by RLS algorithm in real time.
A. Fault Detection
Generally the stator resistance of PMSM is changed by the
variation of temperature. A variation of temperature can
cause a significant variation in stator resistance. However, the
stator resistance under general operation is determined by the
temperature dependence, which is given by
) 1 (
0
T R R A + = o (15)
where R
0
is resistance at the reference temperature C T
$
25 = ,
o
is resistance temperature coefficient, and T A is temperature
increase.
If the open-phase fault occurs, the stator resistance of a
faulty phase estimated by RLS algorithm is rapidly changed.
The stator resistance varied by temperature is distinguished
clearly from that of open-phase fault by its change rate.
Therefore, the threshold value to detect open-phase fault must
be larger than variation of stator resistance for temperature
effect. The algorithm for the fault detection is given by
) (
0
1
_
_
k Sector at
R R if F
R R if F
RLS j th D
RLS j th D

> =
< =
(16)

where R
th
is the threshold value of stator resistance to detect
the open-phase fault continuously.
If the stator resistance (R
j_RLS
) estimated by RLS algorithm
continuously is larger than the threshold value (R
th
), the fault
detection signal (F
D
) changes from low to high.
B. Fault Identification
After the detection, it is possible to identify the position of
faulty phase during a minimum two Sector. Similar
phenomena are observed in other Sector shown in Fig. 2. The
identification of open-phase fault is obtained by the fault
identification signal (F
I
) when the Sector(k) is converted to
(a)
(b)
(c)
(d)
Fig. 6. Simulation results when switch S
1
fault occurs. (a) Current waveform
[A] (b) Estimated stator resistances [] (c) Extended stator resistances []
(d) Fault detection and identification signals
Sector(k+2). The algorithm for the fault identification is
given by
) 2 (
0
1
_
_
+

> =
< =
k Sector at
R R if F
R R if F
RLS j th I
RLS j th I
(17)
The fault diagnosis is achieved by the fault signal (F
D
, F
I
)
at Sector (k+2). For example, assuming that the switch S
1
in
Fig. 1 fails in a open condition and the PMSM operates in
Sector 1 shown in Fig. 2, two effective vectors of the V
1
(100)
and V
2
(110) is generated from inverter output. Consider the
switch configurations under the V
1
(100) and V
2
(110) vectors,
illustrated in the Fig. 4. Before open fault of the switch S
1
, the
V
1
(100) is configured that the phase A is connected to the
positive bus, and the phase B and phase C are tied to the
negative bus. The V
2
(110) is configured with the phase A and
phase B to the positive bus, and the phase C to the negative
bus. After the open fault, however, the V
1
(100) has zero
currents in each phase, and accordingly, all stator resistances
of each phase increase. For the V
2
(110), current on the phase
A becomes zero, and the phase B and C increase slightly for
compensating q-axis current reduced by fault phase as
compared to general operation. Consequently, the open fault
of the S
1
at the Sector 1 can be detected by observing the
variation of stator resistances, such as (i) the estimated stator
resistance R
a_RLS
(of the phase A) continuously increase for
both V
1
(100) and V
2
(110), (ii) the estimated stator resistance
R
b_RLS
(of the phase B) increase for the V
1
(100), but decrease
for the V
2
(110), and (iii) the resistance R
c_RLS
(of the phase C)
increase for the V
1
(100), and remain without any variation
during the V
2
(110). By applying this principle, the variation
characteristics of stator resistance to detect open-phase fault
is analyzed definitely in Sector 1-6.
TABLE I shows how fault diagnosis algorithm under
open-phase fault is achieved in Sector 1-6 and the stator
resistance change for the relationship between switching state
of Sector1-6. The block diagram for overall structure of the
proposed fault diagnosis system is shown in Fig. 5.
V. SIMULATION AND EXPERIMENTAL RESULTS
In order to verify the proposed fault diagnosis algorithm,
the simulation and experiment were performed in the same
simulation conditions. Fig. 6(a)-(d) show simulation results
of a novel proposed method by RLS algorithm when the
open-phase fault of the switch S
1
happens. The open-phase
fault of phase A occurs at 0.394[s]. Fig. 6(a) shows current
waveform under the open fault of switch S
1
. The current of
phase A becomes zero within about 11[ms] due to the open
fault of phase A. As shown in Fig. 6(b) and (c), the stator
resistance estimated by RLS algorithm of fault phase A
suddenly increase. After the fault of switch S
1
, the stator
resistances of each phase differently appear. Because of open
fault of switch S
1
, stator resistance of phase A in comparison
with other stator resistance rapidly increase. Fig. 6(d) shows
the fault detection signal of phase A at 395.8[ms]. In
simulation results, the fault detection signal is achieved by
applying threshold value that is set to the 200% of nominal
stator resistance.
The designed laboratory prototype for the experiment is
shown in Fig. 7. The main controller was configured by using
DSP TMS320VC33 and the sampling time in the control
algorithm was 100s. The inverter used in the experiment was
implemented with IPM PM20CJ060 devices. A 250 W
PMSM was coupled with laboratory in order to test proposed
fault diagnosis algorithm.
In this work, open-circuit fault of switch (S
1
) was described
by the enforced off-signal of gate drive at the fault occurrence.
Fig. 8(a) shows experimental results for estimated stator
resistances R
a_RLS
, R
b_RLS
and R
c_RLS
and fault signal in switch
S
1
fault. After fault occurrence, R
a_RLS
of fault phase suddenly
increases. If stator resistance value R
a_RLS
is larger than the
given threshold value (th), the fault detection signal happens
Fig. 7. Experimental setup for the proposed fault diagnosis.
(a)
(b)
Fig. 8. Experimental results when switch S1 fault occurs. (a) Estimated stator
resistances[] and fault signal. (b) R
a_RLS
, Fault detection signal, Fault
identification signal and Theta.
and then the fault identification signal occurs within two
Sector about 7[ms] as shown in Fig. 8(b).
Fig. 9 shows the current waveforms for three phases under
fault tolerant control. Under the open fault of switch S
1
, the
current of phase A decreased to zero at the positive period.
After fault identification, the fault reconfiguration was
achieved by fault tolerant control. And then three phase
currents can flow through the motor.
Fig. 9. Experimental results for reconfiguration after fault detection.
VI. CONCLUSION
In this paper, the fault diagnosis using RLS algorithm has
been proposed to detect the open-phase fault of inverter
switches in PMSM drives. The proposed scheme has been
achieved by analyzing the variation of stator resistances,
which are estimated by RLS algorithm in real time. The
proposed fault diagnosis scheme can be implemented without
any extra devices. Moreover, since it uses a simple algorithm
by analyzing only estimated stator resistances of each phase,
it has the fast fault detection time. The feasibility of the
proposed fault diagnosis scheme has been proved by
simulation and experimental results.
REFERENCE
[1] P. Pillay and P. Freere, Literature survey of permanent magnet Ac
motors and drives, in Conf. Rec. IEEE-IAS Annu. Meeting. pp. 74-84,
1989.
[2] W. le Roux, R.G. Harley, and T.G. Habetler, Detecting rotor faults in
low power permanent magnet synchronous machines, IEEE Trans. on
Power Electronics, vol. 22, no 1, pp. 332-328, Jan. 2007
[3] R. Peuget, S. Courtine, and J. P. Rognon, Fault Detection and Isolation
on a PWM Inverter by Knowledge-Based Model, IEEE Trans. on Ind.
Applicat., vol. 34, no. 6, pp. 1318-1326, Nov./Dec. 1998.
[4] S. Bolognani, M. Ziglitto and M. Zordan, Innovative Remedial
Strategies for Inverter Faults in IPM Synchronous Motor Drives, IEEE
Trans. Energy Conversion, vol. 18, no. 2, pp. 306-312, June. 2003.
[5] R. L. A. Ribeiro, C. B. Jacobina and E. R. C. da Silva, Fault-Tolerant
Voltage-Fed PWM Inverter Ac Motor Drive Systems, IEEE Trans. Ind.
Applicat., vol. 51, no. 2, pp. 439-446, Apr. 2004.
[6] S. M. Cruz and A. Cardoso, Stator fault diagnosis in three-phase
synchronous and asynchronous motors, IEEE Trans. Ind. Applicat., vol.
37, no. 5, pp. 1227-1233, Sep. 2001.
[7] B. M. Ebrahimi, J. Faiz, and M. J. Roshtkhari, Static-, Dynamic-, and
Mixed-Eccentricity Fault Diagnoses in Permanent-Magnet Synchronous
Motor, IEEE Trans. on Ind. Electron., vol. 56, no. 11, pp. 4727-4739,
Nov. 2009.
[8] M. S. Nejad, B. N. Mobarakeh, S. Pierfederici, and F. M. Tabar, Fault
tolerant and minimum loss control of double-star synchronous machines
under open phase conditions, IEEE Trans. Ind. Electron., vol. 55, no. 5,
pp. 10151020, May 2008.
[9] S. Morimoto, M. Sanada, and Y. Takeda,Mechanical Sensorless Drives
of IPMSM with Online Parameter Identification, IEEE Trans. Ind.
Applicat., vol. 42, no. 5, pp. 1241-1248, Sep./Oct.. 2006.