46 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO.

1, JANUARY/FEBRUARY 2005

Fault Detection and Fault-Tolerant Control of Interior

Permanent-Magnet Motor Drive System
for Electric Vehicle
Yu-seok Jeong, Student Member, IEEE, Seung-Ki Sul, Fellow, IEEE, Steven E. Schulz, Senior Member, IEEE,
and Nitin R. Patel, Member, IEEE

Abstract—This paper presents a control strategy that provides

fault tolerance to the major sensor faults which may occur in an
interior-permanent-magnet-motor (IPMM)-based electric vehicle
propulsion drive system. Failures of a position sensor, a dc-link
voltage sensor, and current sensors are all included in the study
assuming no multiple faults. For each possible sensor fault, a cor-
responding method of detection or diagnosis is provided. Addition-
ally, once the fault is detected, the control scheme is automatically
reconfigured to provide post-fault operational capability. A state
observer is used to provide missing current information in the case
of current sensor faults. Experimental results demonstrate the ef-
fectiveness of both the fault detection algorithm and the reconfig-
urable control scheme. The resulting IPMM drive system proves to
Fig. 1. IPMM drive system and sensors.
be resilient to sensor failures while providing smooth transition to
the post-fault operational mode.
Index Terms—Fault tolerance, reconfigurable control, sensor
fault, state observer.

I. INTRODUCTION

R ECENTLY, the interior permanent-magnet motor (IPMM)

is receiving strong interest for electric/hybrid vehicle trac-
tion due to its capability of field-weakening control, high effi-
ciency, and high power density [1], [2]. For the vehicle trac-
tion control, fault detection and fault tolerance are important
issues not only for the reliability of the drive system but also
for the proper operation of the vehicle following a fault. There Fig. 2. Control block diagram of IPMM drive.
are numerous study results about fault detection and fault-tol-
erant control [3]–[6], but most of them focused on the faults of
power semiconductors of an inverter and stator windings of a a vehicle gives gracefully degraded performance according to
motor. In [6], sensor faults of an induction motor drive system the faults. In this paper a method to detect sensor faults and an
are addressed and a method to reconfigure the control system algorithm to reconfigure the control system for IPMM drive are
according to the specific fault of a sensor is discussed. In the described. The main focus of the study has been on the detec-
paper, the control system tolerates the faults by changing the tion of current sensor faults, the analysis of a current observer,
control algorithm from a high-performance indirect vector con- and the resilient control of the drive system while retaining the
trol to a simple V/f control. In this way the traction system of same basic vector control strategy.

Paper IPCSD-04-060, presented at the 2003 Industry Applications Society

II. IPMM DRIVE SYSTEM AND SENSORS
Annual Meeting, Salt Lake City, UT, October 12–16, and approved for publica- The drive system including motor and sensors is shown in
tion in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial
Drives Committee of the IEEE Industry Applications Society. Manuscript sub- Fig. 1, where the rotor position is measured by an absolute en-
mitted for review July 1, 2003 and released for publication August 12, 2004. coder or a resolver, the stator currents, namely, and phase by
This work was supported in part by General Motors. two Hall-effect sensors, and the dc-link voltage by an isolated
Y. Jeong and S.-K. Sul are with the School of Electrical Engineering and
Computer Sciences, Seoul National University, Seoul 151-744, Korea (e-mail: transducer. It is assumed that the drive system is under a posi-
yu-seok@eepel.snu.ac.kr; sulsk@plaza.snu.ac.kr). tion-sensor-based operation with maximum-torque-per- ampere
S. E. Schulz and N. R. Patel are with the Advanced Technology Center, (MTPA) control proposed in [7]. The control block diagram is
General Motors, Torrance, CA 90505 USA (e-mail: steven.schulz@gm.com;
nitinkumar.patel@gm.com). shown in Fig. 2, where the - and -axes current references in
Digital Object Identifier 10.1109/TIA.2004.840947 the rotor reference frame are generated by a lookup table based
0093-9994/$20.00 © 2005 IEEE
JEONG et al.: FAULT DETECTION AND FAULT-TOLERANT CONTROL OF IPMM DRIVE SYSTEM FOR ELECTRIC VEHICLE 47

on the torque reference and the maximum flux and are com-
pared with the measured current to be regulated by a propor-
tional and integral (PI) controller with back-electromotive-force
(EMF) decoupling term. The voltage generated by space-vector
pulsewidth modulation (PWM) is applied to the IPMM.

III. FAULT DETECTION IN SENSORS AND RECONFIGURATION

OF CONTROL LOOP

A. Encoder/Resolver Fault
The faults of a rotor position sensor may result in immediate
overcurrent trip of the drive system in a heavy load or high-speed
operating condition due to the decoupling failure of the - and
-axes current controllers. In a light-load and low-speed condi-
tion the torque linearity would not be maintained and the faults
eventually lead to overcurrent trip. To detect a fault of the po-
sition sensor the sensorless control algorithm in [8] could be
used for the rotor angle estimation, and if the difference be-
tween the measured angle and the estimated one is larger than a
threshold value, then the control algorithm should be reconfig-
ured to the sensorless control. A fault from the slippage of the
Fig. 3. Position sensorless restart of IPMM in the case of an encoder/resolver
position sensor is harder to detect than that from the breakdown fault while the machine is running.
[9]. In the case of low speed operation including starting, where
the back EMF is too small to give the accurate rotor position, TABLE I
high-frequency injection methods in [10], [11] may be used. The IPMM PARAMETERS
performance of a sensorless IPMM drive is comparable to that of
a sensored one due to the inherent rotor saliency of the IPMM. In
either low- or high-speed operation of the IPMM, the seamless
transition from a sensored control to a sensorless control can be
achieved with a proper transition algorithm. In high speed oper-
ation with a large inertia, which is usually the case in a vehicle
drive, the seamless transition based on the estimated back-EMF
information is possible. After the inverter is tripped due to the
position sensor faults and all phase currents fall off, the current small. If the power balance exceeds a certain threshold, then
references in the rotor reference frame can be set to zero and the measured dc-link voltage is replaced by the nominal value
the back EMF can be estimated from the current observer if the in the current controller. The error in the voltage information
IPMM is still running due to the system inertia. Based on the results in mistuned gains of the PI controller and back-EMF de-
estimated back EMF, a sensorless control algorithm can start coupling term in the current controller. Hence, for successful
without stopping the IPMM. The initial torque variation during operation without voltage sensor it is crucial that the system sta-
the zero-current regulation period depends on the bandwidth of bility should be guaranteed with at least 50% variation of con-
the current regulator and the motor speed. The experimental re- troller gains and back-EMF decoupling terms due to the voltage
sult is depicted in Fig. 3. The test IPMM parameters are shown variation of a battery.
in Table I.
C. Current Sensors
B. DC-Link Voltage Sensor
Sudden severe faults of a current sensor [9] result in the over-
The faults of a dc-link voltage sensor are more easily detected current malfunction of the system, and if there is no proper pro-
than those of other sensors since the nominal value of the dc-link tection scheme in the gate-drive circuit, it leads to irrecoverable
voltage is far from zero. The scale/offset error can be detected faults of power semiconductors in the inverter. The minor faults
by the following power balance: such as gain drift and nonzero offset would result in torque pul-
sations synchronized with the inverter output frequency [12].
(1) The larger the offset and scaling error, the worse the perfor-
mance of torque regulation. Ultimately, if the offset and gain
where the first term stands for the estimated electrical input drift are above a certain level, this results in overcurrent trip at
power, the second term is the mechanical output power, and high speed and in heavy load conditions. The faults including
denotes the estimated power loss which includes copper loss, the offset and gain drift can be easily detected when the ma-
iron loss, etc. chine is not running.
It can be easily seen that the fault detection is difficult at low With the system shown in Fig. 4, at first the gating signals to
speed and in light-load condition because the power itself is too -phase semiconductor switches, i.e., and are blocked,
48 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 1, JANUARY/FEBRUARY 2005

Fig. 4. Inverter circuit and IPMM windings.

Fig. 6. Measured current for c-phase blocked operation.

Furthermore the total sum of the measured values of each phase

current should be around zero without transient characteristics,
Fig. 5. Equivalent circuit for c-phase blocked operation. otherwise there might be significant offset error in case of the
current sensor faults or other faults in inverter power circuits or
and the line-to-line voltage is synthesized by a PWM in- IPMM windings.
verter as follows: Next, another test voltage of the same type in (2) is applied
between - and -phase terminals of the motor and -phase ter-
(2) minal is shorted with phase as shown in Fig. 7(a). The - and
-phase currents are measured and stored in a memory compo-
where stands for the magnitude of the test voltage, for
nent of the digital control system. Then the test voltage is ap-
the angular frequency, and for the initial phase angle. The
plied between and phases as shown in Fig. 7(b), and lastly
system in Fig. 4 can be expressed as an equivalent circuit shown
as in Fig. 7(c). The sum of the stored phase current at each cor-
in Fig. 5. The current flowing in the circuit as a result of the test
responding time in Fig. 7 should be time-point-wise zero if the
voltage in (2) can be derived as follows:
motor windings are well balanced.
If the faulted sensor is detected in the above-mentioned
method, the measured value from it can be disregarded. The -
(3) and -axes currents in the rotor reference frame are regulated
where . based on the estimated currents which are observed by an
in (3) stands for the sum of the winding resistance of open-loop observer in case that neither of the current sensors
the IPMM and the conduction resistance of the power semicon- is available. The stator voltage model of IPMM in the rotor
ductor, and denotes the inductance between - and -phase reference frame can be expressed as follows:
terminals of the motor, and is a function of the rotor position.
From (3) it can be seen that the transient term can be suppressed
by adjusting the voltage phase according to power factor of the
circuit. Fig. 6 shows the traces of the measured - and -phase (4)
currents when the reference voltage in (2) is applied.
The inductance is several hundreds of microhenrys varying The transfer function from the actual current to the estimated
with the rotor position and the resistance is around 10 m in- current of the open-loop observer takes the following form as-
cluding the resistance of power semiconductor, and the time suming that the mechanical dynamics is much slower than the
constant of the circuit is several tens of msec. However, with the electrical dynamics:
proper setting of the initial phase angle of the reference voltage
there is no transient in the current trace. The frequency of the
test voltage is 200 Hz and the duration is 5 cycles. Hence, the
test takes only 25 ms. If the winding, inverter and the current
sensors have no problem, the measured - and -phase currents (5)
should be the same in magnitude but opposite in sign as shown
in Fig. 6. Also the root mean square (RMS) value of the cur- where the circumflexed parameters denote estimated values. It
rent should be around the value of the steady-state term in (3). should be noted that the system has no disturbance input and
JEONG et al.: FAULT DETECTION AND FAULT-TOLERANT CONTROL OF IPMM DRIVE SYSTEM FOR ELECTRIC VEHICLE 49

Fig. 7. Test circuits for fault detection of current sensors. (a) Test voltage between a and bphase. (b) Test voltage between b and cphase. (c) Test voltage between
c and a phase.

Fig. 9. Closed-loop current observer with a sensor gain matrix.

gain matrix which has time-varying and even cross-coupling el-

ements as follows:

Fig. 8. Trajectory of the current error vector in the rotor reference frame.

the estimation error depends only on the parameter variation.

The stator resistance and the rotor magnet flux linkage can be (8)
changed with the operating temperature and the stator induc- Thus, the observer depicted in Fig. 9 can be analyzed in the
tances decrease when the IPMM operates in high load condi- unified form. In normal operation, both of the current sensors
tion. are utilized, and the sensor gain matrix is nothing but the identity
If either of the current sensors is available, a closed-loop ob- matrix. If either current sensor has failed, it is replaced with the
server can be used to estimate the stator currents. The current singular matrix defined in (8). If the other sensor also
error vector with single phase feedback in the stationary refer- breaks down, it becomes the zero matrix.
ence frame can be defined as follows if the rest of phase current The experimental results show the dynamic characteristics of
errors are assumed to be negative one-half of the sensed one: the current control in the rotor reference frame when a pulse
torque of 10 N m is applied at a speed of 1500 r/min. In Fig. 10,
(6) the -phase current sensor has failed at 100 msec and the -axis
current in the stationary reference frame was replaced with the
where is the measured phase current, is the estimated cur- estimated one from the observer in Fig. 9. The - and -axes
rent of the corresponding phase, and is zero with sensing stator inductances in the current observer were intentionally
phase, one with phase, and two with phase, hereafter. This mistuned to show the feasibility of the proposed reconfigura-
yields the following in the rotor reference frame: tion algorithm. It can be seen that the current ripple at the dou-
bled synchronous frequency decreases gradually in steady state.
It should be noted that the observer gains are high enough to
(7) make the closed-loop poles with both sensors higher than the
synchronous frequency, otherwise it can cause resonance and
This error vector in Fig. 8 has an additional term which has the make the time-varying system unstable.
same magnitude but rotates in the opposite direction at a speed Under the same condition, the transition characteristic from
of . This phenomenon is a result of insufficient current infor- the closed-loop to the open-loop observer is depicted in Fig. 11.
mation feedback due to the fault of either current sensor. From The transient after the other sensor failure at 100 ms is based on
the viewpoint of the state feedback, this single phase error can the loss of the correction term which compensated for the error
be interpreted as a product of the current error and a sensor resulting from the parameter variation.
50 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 41, NO. 1, JANUARY/FEBRUARY 2005

combination of saliency-based and model-based position esti-

mation algorithms are utilized to estimate the rotor position at all
operating speeds. For the case of dc-link voltage sensor failure,
the principle of power balance is adopted to detect sensor mal-
function and the effect to the stability of current control is dis-
cussed. Lastly, in the case of current sensor failure, a sequence of
simple tests is devised to detect the sensor failure. Missing cur-
rent information is obtained using a state observer to estimate
the unknown quantities and the analysis is made introducing a
sensor gain matrix.

REFERENCES
[1] Y. Honda, T. Nakamura, T. Higaki, and Y. Takeda, “Motor design con-
siderations and test results of an interior permanent magnet synchronous
motor for electric vehicle,” in Conf. Rec. IEEE-IAS Annu. Meeting, 1997,
pp. 75–82.
[2] J. Park, D. Koo, J. Kim, and H. Kim, “Improvement of control character-
istics of interior permanent-magnet synchronous motor for electric ve-
hicle,” IEEE Trans. Ind. Appl., vol. 37, no. 6, pp. 1754–1760, Nov./Dec.
2001.
[3] A. El-Antably, L. Xiaogang, and R. Martin, “System simulation of fault
conditions in the components of the electric drive system of an electric
vehicle or an industrial drive,” in Conf. Rec. IEEE IECON’93, 1993, pp.
Fig. 10. Transition from both-sensored to single-sensored operation. 1146–1150.
[4] B. A. Welchko, T. M. Jahns, and S. Hiti, “IPM synchronous machine
drive response to a single-phase open circuit fault,” IEEE Trans. Power
Electron., vol. 17, no. 5, pp. 764–771, Sep. 2002.
[5] N. Retiere, D. Roye, and P. Mannevy, “Vector-based investigation of in-
duction motor drive under inverter fault operations,” in Conf. Rec. IEEE
PESC’97, 1997, pp. 1288–1294.
[6] R. B. Sepe Jr., B. Fahimi, C. Morrison, and J. M. Miller, “Fault-tolerant
operation of induction motor drives with automatic controller reconfig-
uration,” in Conf. Rec. IEEE IEMDC’01, 2001, pp. 156–162.
[7] T. M. Jahns, G. B. Kliman, and T. W. Neumann, “Interior permanent-
magnet synchronous motors for adjustable-speed drives,” IEEE Trans.
Ind. Appl., vol. 22, no. 4, pp. 738–747, Jul./Aug. 1986.
[8] S. Morimoto, K. Kawamoto, M. Sanada, and Y. Takeda, “Sensorless con-
trol strategy for salient-pole PMSM based on extended EMF in rotating
reference frame,” IEEE Trans. Ind. Appl., vol. 38, no. 4, pp. 1054–1061,
Jul./Aug. 2002.
[9] C. Thybo, “Fault-Tolerant Control of Inverter Fed Induction Motor
Drives,” Ph.D. dissertation, Dept. Control Eng., Aalborg Univ., Grasten,
Denmark, 1999.
[10] J. Ha, K. Ide, T. Sawa, and S. Sul, “Sensorless rotor position estimation
of an interior permanent-magnet motor from initial states,” IEEE Trans.
Ind. Appl., vol. 39, no. 3, pp. 761–767, May/Jun. 2003.
[11] H. Kim, M. C. Harke, and R. D. Lorenz, “Sensorless control of interior
permanent-magnet machine drives with zero-phase lag position estima-
tion,” IEEE Trans. Ind. Appl., vol. 39, no. 6, pp. 1726–1733, Nov./Dec.
2003.
[12] D. Chung and S. Sul, “Analysis and compensation of current measure-
ment error in vector-controlled AC motor drives,” IEEE Trans. Ind.
Appl., vol. 34, no. 2, pp. 340–345, Mar./Apr. 1998.

Fig. 11. Transition from single-sensored to sensorless operation.

Yu-seok Jeong (S’01) was born in Daegu, Korea,

IV. CONCLUSION in 1971. He received the B.S. and M.S. degrees in
electrical engineering in 1993 and 1995, respectively,
The issues of IPMM drive response to sensor faults have been from Seoul National University, Seoul, Korea, where
investigated. As a result, a fault tolerant IPMM control scheme he is currently working toward the Ph.D. degree,
has been proposed. For each potential sensor failure, a fault de- pursuing fault-tolerant control and robust adaptive
control of interior permanent-magnet synchronous
tection algorithm and post-fault control scheme has been devel- machine drives in collaboration with General Mo-
oped. Transition from fault mode to post-fault operating mode tors.
is seen to be smooth. The resultant drive provides slightly de- He joined the Kia Motors Technical Center, Seoul,
Korea, as a Research Engineer in 1995. At Kia,
graded performance in the case of some faults (dc voltage sensor he worked to develop ac motor drive systems for electric/hybrid vehicles.
and current sensor), but still allows crucial limp home capability In 1999, he joined the Korea Electrical Engineering and Science Research
in those cases. Institute, Seoul, Korea, where he worked on single-converter–multi-inverter
control design for crane applications. He spent a year as a Special Student at
For the case of a position sensor fault, a sensorless rotor posi- the University of Wisconsin, Madison. His interests include digital control of
tion estimation scheme is used to provide post-fault control. A power electronics and energy conversion.
JEONG et al.: FAULT DETECTION AND FAULT-TOLERANT CONTROL OF IPMM DRIVE SYSTEM FOR ELECTRIC VEHICLE 51

Seung-Ki Sul (S’78–M’80–SM’98–F’00) was born Nitin R. Patel (S’96–M’97) was born in Gujarat,
in Korea in 1958. He received the B.S., M.S., and India. He received the B.S. degree in instrumentation
Ph.D. degrees in electrical engineering from Seoul and control from the University of Poona, Poona,
National University, Seoul, Korea, in 1980, 1983, and India, in 1991, and the M.S. degree in electrical
1986, respectively. engineering from the University of Tennessee,
From 1986 to 1988, he was an Associate Re- Knoxville, in 1996.
searcher with the Department of Electrical and He was a Research Engineer with Dover Elevators
Computer Engineering, University of Wisconsin, following graduation. In 1997, he joined the General
Madison. From 1988 to 1990, he was a Principal Motors Advanced Technology Center, Torrance, CA,
Research Engineer with Gold-Star Industrial Sys- as a Control Engineer, where he is involved in the
tems Company. Since 1991, he has been a member development of ac drives of propulsion systems for
of the faculty of the School of Electrical Engineering, Seoul National Uni- FCEV/HEV applications. His research interests include fuzzy logic and neural
versity, where he is currently a Professor. His current research interests are network applications to power electronics, motor drives, and sensorless machine
power-electronic control of electric machines, electric/hybrid vehicle drives, control. He is the holder of nine U.S. patents with 12 pending.
and power-converter circuits. Mr. Patel is an active member of the Industrial Drives Committee of the IEEE
Industry Applications Society. He has authored several papers presented at IEEE
conferences. In 2003, he was awarded the General Motors Vice President’s
Award (Charles L. McCuen Award) for his contribution to wheel hub motor
Steven E. Schulz (S’88–M’90–SM’02) received the technology.
B.S. degree (summa cum laude) from North Carolina
State University, Releigh, in 1988, and the M.S.
degree from Virginia Polytechnic Institute and State
University, Blacksburg, in 1991, both in electrical
engineering.
In 1991, he joined Hughes Aircraft Space and
Communications Group, where he worked on space-
craft power supply design and modeling. Since 1992,
he has been with the General Motors Advanced
Technology Center, Torrance, CA. From 1992 to
1997, he designed and developed inductively coupled battery chargers for
electric vehicles. He is currently working in the area of variable-speed motor
drives as applied to electric and hybrid vehicles. His interests include power
electronics, ac motor drives, and digital controls.