A Real-Time Sensor Fault Detection, Isolation and
Reconfiguration Method for Vector Controlled
Induction Motors Based on Extended Kalman Filter
Feng Wu, Jin Zhao
∗
, Yang Liu, Wenming Cao
School of Automation,Huazhong University of Science and Technology
Wuhan, China
E-mail: wf199010202051@163.com, jinzhao617@163.com, yangliu30@gmail.com
Abstract—Vector controlled motor drives are widely used in
industry application areas, usually they contain two current
sensors and a speed sensor. A fault diagnosis and reconfigu-
ration structure is proposed in this paper including current
sensor measurement errors and sensors open-circuit fault. Sliding
windows and special features are designed to real-time detect
the measurement errors, compensations are made according
to detected offset and scaling values. When open-circuit faults
occur, sensor outputs are constant-zero, the residuals between
the Extended Kalman Filter (EKF) outputs and the sensors
outputs are larger than pre-defined close-to-zero thresholds,
under healthy condition, the residuals are equal to zero, as a
result, the residuals can be used for open circuit fault detection.
In this situation, the feedback signals immediately switch to EKF
outputs to realize reconfiguration. Fair robustness are evaluated
under disturbance such as load torque changes and variable
speed. Simulation results show the effectiveness and merits of
the proposed methods in this paper.
I. INTRODUCTION
Induction motors are widely used in industrial application
areas due to their simple mechanical structure and easy main-
tenance. However, it is known that the control of induction
motors is relatively difficult compared with other kinds of
motors. To solve this problem and achieve high performance,
vector controlled schemes have been developed, in which
torque and flux can be decoupled to separately control. In ad-
dition, growing concerns about robust drives at lowest possible
cost, high reliability becomes a challenge in major application
fields. During last few years, motor drives reliability was a
very active research field, for example, in [1]–[3]. Generally,
vector controlled motor drives need a speed sensor and at least
two current sensors, in most cases. When one of the sensors
is faulty, abnormal performance or even instability might be
caused. This is very harmful for system operation and should
be avoided.
Generally, Except speed open-circuit fault, current sensor
faults can be divided into two kinds: open-circuit fault and
measurement errors, current measurement error can be classi-
fied as offset error and scaling error [4].
Measurement errors in current sensors will cause torque
and speed ripples [4], [5], which reduce the performance of
the drive systems and should be avoided. The offset error
causes the torque of a motor to oscillate at the stator electrical
frequency, f
e
; the scaling error at the frequency, 2f
e
, respec-
tively [4]. Fourier transform is used for current and speed
compensation in this method. However, complex calculation
makes it hard to implement. Parity space approach based on
temporal redundancies is presented in [6], however, it is only
available for a large class if electrical system. The measure-
ment errors and their compensation are investigated in [7], [8].
Works presented in [8]–[10] focus on sensor fault detection
and isolation (FDI) and control reconfiguration. Relative low-
cost senseless control schemes are presented in [11]–[13],
which can be applied to fault tolerant control reconfiguration.
These studies show clearly that robust operation relies strongly
on the quality of the sensor measurements and healthy of the
sensors. If the sensor open-circuit and measurement errors are
not detected and handled quickly, their effect can reduce the
system performance, such as the torque and speed ripples, or
even lead to catastrophe of the electrical system such as single-
or three-phase power converters, dc or ac electrical drives.
On open-circuit fault, in [14], [15], the effect of sensors
open-circuit fault on a doubly fed induction machine is studied.
Although the issue of the sensors fault detection, isolation,
tolerance and measurement error for IM drives has already
been investigated by some researchers, many of them focus
on only one type of faults. For sensor open-circuit fault,
there are totally three kinds FDI techniques: model-based
technique, signal-processing technique and intelligent tech-
nique [16]. Neural network [17], fuzzy logic [18] and wavelet
[19] are proposed as FDI for sensors and applied to fault
tolerance as a senseless control scheme, these methods avoid
the knowledge of the model of the system, they perform well
with variable parameters and are independent to disturbance,
however, complex calculation and parallel structure make these
methods hard to implement. Signal-processing technique in
frequency-domain has been proposed as FDI for sensors, for
open-circuit is transient and signal is distorted by generating
a high-frequency transient. A proposed sensor open-circuit
fault detection method with variables resulted from the average
absolute value of the sum of the of the normalized three-phase
currents is presented in [20], this method preforms well in
permanent magnet synchronous motor (PMSM), but control
reconfiguration is not discussed. In [15], residuals between
measured currents and the predicted ones are used for sensor
978-1-5090-2067-6/16/$31.00 ©2016 IEEE
2016
International Symposium on Power Electronics,
Electrical Drives, Automation and Motion
617