![](bg2.png)
or thermal stress acting on the stator winding. This
type of fault is responsible for one quarter of all faults.
This fault can rapidly propagate to other stator turns
because it creates a large circulating current in the
shorted path. Since the magnitude of the second-
order harmonics in the q-axis current is proportional
to the number of shorted turns and operating speed,
the fault becomes considerably large as the number
of shorted turns and the magnitude of speed increases
(Kim, 2011). Consequently, other faults such as ec-
centricity faults, open circuit faults, and demagneti-
zation faults are generated by the inter-turn faults. It
is necessary to detect inter-turn faults immediately be-
cause of these dangers. The early detection of stator
winding failures is important in order to avoid greater
risk to the drive.
2.2 Motor Current Signature Analysis
(MCSA)
Motor current signature analysis (MCSA) analyzes
the amplitudes of the harmonics of the stator cur-
rent. When the amplitude is over a threshold value,
which is denoted by the standard amplitude of a nor-
mal current, an inter-turn fault is detected. MCSA
has proven to be an efficient technique for fault de-
tection and is the most popular technique (Gandhi et
al, 2011). In order to detect inter-turn faults, it is im-
portant to determine their frequencies. As shown in
(Sottile, 2001), the faulty harmonics are located at
f
f ault
= 3 × f
f und
where f
f ault
is the frequency component associated
with inter-coil shorts within the stator winding, and
f
f und
is the stator fundamental frequency. This
means that the frequency component associated with
an inter-turn fault depends on the fundamental fre-
quency. As the speed of a vehicle increases or de-
creases, the vehicle is in a transient state. In a tran-
sient state, the fundamental frequency should vary
with the vehicle speed. Thus, the frequency of the
faults should change proportionally. Hence, an ex-
traction algorithm under transient conditions through
MCSA is required.
2.3 Description of Overall System
The block diagram of the overall detection system is
shown in Fig. 1. In an electric vehicle, the battery
is used to power the motor, which is an IPMSM. The
battery has a voltage of approximately 15V. Because
the output of the battery is DC, an inverter should
be used to convert DC to AC. The output of the in-
verter consequently becomes AC, which powers the
Battery
DC-AC
Inverter
IPMSM
Motor
Controller
Low Pass
Filter
Fault
Detection
Algorithm
Fault
Classification
Fault
Indicator
I(t)
Figure 1: Block diagram of motor drive and fault detection
strategy.
IPMSM. To control the speed of the IPMSM, the mo-
tor controller receives the angle and angular velocity
of the IPMSM and the angle of the rotor from the re-
solver. The input current of the IPMSM is the target
signal that is extracted to detect faults. The input cur-
rent is passed through a low pass filter and then used
in the fault detection algorithm. The algorithm ex-
tracts the input signal and classifies each harmonic. It
is impossible to classify the harmonics from the entire
signal that is received from the current sensor. Using
fault equations (Barendse and Pillay, 2006), the fre-
quencies of the faults are determined. Subsequently,
the fault indicator indicates the faulty signal to the ve-
hicle control unit. The significant harmonic compo-
nents are identified by the fault detection algorithm.
3 PROPOSED ALGORITHM
3.1 Theoretical Background
In this section, the adaptive algorithm is introduced
along with a description of how the algorithm is
adapted in a conventional algorithm to extract the in-
put current signal. The adaptive algorithm shows re-
markable qualities in tracking and extracting the non-
stationary sinusoid, while minimizing the square er-
ror. Let i(t) denote a stator current signal
i(t) = i
c
(t) + i
n
(t)
where i
c
(t) is the pure current signal, and i
n
(t) is the
noise component. i
c
(t) can be represented in detail as
follows;
i
c
(t) = I
c
sin(
Z
ω(τ)dτ + δ)
where I
c
is the amplitude of the current, ω is the vary-
ing angular velocity dependent on time, and δ is the
angle shift. Let i
out
denote the output current esti-
mated by the adaptive algorithm.
i
out
= i
f und
+ i
1
+ i
2
+ .. (1)
ICINCO 2012 - 9th International Conference on Informatics in Control, Automation and Robotics
428