A New Algorithm for Real-Time Multiple Open- Circuit

A New Algorithm for Real-Time Multiple OpenCircuit Fault Diagnosis in Voltage-Fed PWM Motor Drives by the Reference Current Errors Jorge O. Estima, Member, IEEE, and A. J. Marques Cardoso, Senior Member, IEEE Φ Abstract—Three-phase inverters are currently utilized in an enormous variety of industrial applications, including variable speed ac drives. However, due to their complexity and exposure to several stresses, they are prone to suffer critical failures. Accordingly, this paper presents a novel diagnostic algorithm that allows the real-time detection and localization of multiple power switches open-circuit faults in inverter-fed ac motor drives. The proposed method is quite simple and just requires the measured motor phase currents and their corresponding reference signals, already available from the main control system, avoiding therefore the use of additional sensors and hardware. Several experimental results using a vector controlled permanent magnet synchronous motor drive are presented, showing the diagnostic algorithm effectiveness, its relatively fast detection time and its robustness against false alarms.

Index Terms — Fault diagnosis, fault detection, fault location, pulse width modulation inverters, variable speed drives, machine vector control, velocity control, ac motors, permanent magnet motors.

I. NOMENCLATURE en in i¤n dn Im ! an Dn An kf km kd hxi

Reference current errors Measured motor phase currents Motor phase reference currents Diagnostic variables Currents amplitude Stator currents frequency Auxiliary variables Diagnostic fault symptom variables Auxiliary fault symptom variables Single fault threshold Multiple faults threshold Same leg double fault threshold Average value of the variable x

Manuscript received November 9, 2011. Accepted for publication February 14, 2012. This work was supported by the Portuguese Government through the Foundation for Science and Technology (FCT) under Project No. SFRH/BD/40286/2007 and Project No. PTDC/EEA-ELC/105282/2008. Copyright (c) 2012 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to [email protected]. Jorge O. Estima is with the Dep. of Electrical and Computer Engineering, University of Coimbra, 3030-290 Coimbra, Portugal, and also with the Instituto de Telecomunicações (email: [email protected]). A. J. Marques Cardoso is with the Dep. of Electromechanical Engineering, University of Beira Interior, 6201-001 Covilhã, Portugal, and also with the Instituto de Telecomunicações (email: [email protected]).

II. INTRODUCTION

A

LTOUGH the widely used voltage source inverters (VSIs) have already achieved a certain level of maturity, they are prone to suffer critical failures due to their complexity and considering that they are often exposed to high stresses [1]. Therefore, in order to preclude this harmful influence as well as to improve the system reliability, the development of fault diagnostic methods has gain a lot of interest during the last years. This can be especially important for the development of fault-tolerant systems [2]-[5], which allow the system to continue operating under faulty conditions. The great susceptibility of these power converters is confirmed by several statistical studies that show that capacitors, semiconductors power switches and their corresponding gate drivers electronic circuits are the main responsible for critical failures [6]-[9]. In general, inverter power device failures can be broadly classified as open-circuit faults and short-circuit faults. Typically, short-circuit faults are very destructive, requiring special measures to shutdown the drive immediately. On the other hand, open-circuit faults do not necessarily cause the system shutdown and can remain undetected for an extended period of time. This may lead to secondary faults in the converter or in the remaining drive components, resulting in the total system shutdown and high repairing costs. The use of the Park’s Vector Approach as a fault diagnostic tool for voltage source inverter faults was successfully applied in [10]-[13]. However, this approach requires very complex pattern recognition algorithms which are not suitable for integration into the drive controller. The localization of the faulty switch can also be performed by the analysis of the current space vector trajectory diameter [14]-[15]. Nevertheless, this technique has serious drawbacks related to slow detection, tuning and problems under low current values. Fast detection times can be achieved using voltage-based techniques [16]-[18]. However, additional voltage sensors may be required, which increases the drive costs and complexity. Furthermore, in order to avoid false alarms, some time-delay values must be correctly defined which can be very difficult since they depend on several variables. A different technique, based on the average current Park’s Vector approach was proposed in [19]. This technique presents some problems regarding the tuning effort related to

the fact that it is load dependent, together with some false alarm issues. These problems were mitigated by the normalization of the diagnostic variables, as proposed in [20]. Despite of multiple faults diagnosis, this algorithm has some drawbacks such as a higher complexity and larger detection times. Some reviews and surveys can also be found in the literature regarding fault diagnostic methods for VSIs single power switch open-circuit faults [9], [21]-[22]. New algorithms with multiple faults diagnosis were proposed in [23]-[24]. They do not dependent on the machine load and speed conditions and show a very robust behavior regarding the issue of false alarms. An observer-based approach for single and simultaneous faults was addressed in [25]. The algorithm proves to be load independent and does not require additional sensors. However, as main drawbacks, it is computationally demanding and requires the knowledge of the machine parameters. A simple method based on the operating characteristics of brushless direct current motor drives was presented in [26] where the diagnosis is accomplished without using additional sensors or electrical devices. A new approach with a detection time lower than 10 µs was presented in [27]. However, the developed analog circuit requires some voltage measurements, which increases the system complexity, and the technique cannot be applied to all power devices. VSI open-circuit faults can also be effectively detected and localized by the analysis of the second-order harmonic component in the q-axis or by using the discrete wavelet transform [28]-[29]. A new approach for single power switch open-circuit fault diagnosis based on the reference current errors was also presented in [30], showing a fast detection time equivalent to 5% of the motor phase currents fundamental period. Fault diagnostic methods and fault-tolerant strategies applied to matrix converters are also gaining more and more interest by the scientific community [31]-[32]. The literature review shows that, despite of all the existing work done by several authors concerning the diagnostic methods for VSI open-circuit faults, there is a great lack of research regarding the development of algorithms that can detect single and multiple faults, having simultaneously other mandatory features such as the requirement of no extra sensors or electric devices, simple implementation, operating conditions independence and low computational requirements. In this paper a novel fault diagnostic algorithm for real-time open-circuit faults diagnosis is presented, which takes into account all these important features. In order to test and validate the proposed method, single and multiple open-circuit faults are introduced in the VSI of a vector controlled permanent magnet synchronous motor (PMSM) drive. Experimental results are presented, showing the algorithm performance regarding the detection and localization of different faulty power switches combinations

i¤n

in

+ −

en

u (t )

hen i

Average Values

Average Values

× ÷

hjin ji

dn

Auxiliary Variables

Fault Detection and Localization

an

Faulty Switches

Fig.1 – Block diagram scheme of the proposed fault diagnostic method.

for distinct operating conditions. III. PROPOSED DIAGNOSTIC METHOD The proposed algorithm results from an enhanced version of the method firstly presented in [30]. A block diagram of its structure is shown in Fig. 1. It is desirable that the fault diagnostic method utilizes variables already used by the main control, avoiding the use of extra sensors and the subsequent increase of the system complexity and costs. Hence, this method has as inputs the motor phase currents and their corresponding reference signals, which can be obtained from the main control system. Considering a vector controlled drive with hysteresis current controllers in the abc reference frame, the motor reference currents are directly available from the control strategy. For the case where PI current controllers are used together with a SV-PWM or other modulation techniques, if the reference currents are not directly available, they can be easily obtained from the dq current components using the inverse Park transformation. The three reference current errors en are calculated from the measured motor phase currents in and from their corresponding reference signals i¤n by:

en = i¤n ¡ in

(1)

where n = a; b; c . The properties of the final diagnostic variables result from the average values of the three reference current errors hen i: hen i =

! 2¼

Z

2¼ !

en dt

(2)

0

In order to overcome the problems associated to the drive rated power and machine mechanical operating conditions dependency such as the speed and load levels, the average reference current errors are normalized using the average absolute values of the motor phase currents hjin ji. Therefore, the final diagnostic variables dn are given by:

dn =

hen i hjin ji

(3)

Under healthy operating conditions, and neglecting the low amplitude high frequency noise, the motor is supplied by a perfectly balanced three-phase sinusoidal current system that

can be given by: T1

8 > ¡ : ic = Im sin !t +

¢

( 0 0 < t · !¼ ia = Im sin (!t) !¼ < t · 2¼ !

(4)

da =

0

R 2¼ ! ! Im sin (!t) dt ¡ 2¼ Im sin (!t) dt ¼ ! 2¼ R ! ! 2¼ ¼ ¡Im sin (!t) dt

PMSM

(5)

(6)

Solving this equation by calculating the average values of the three signals, the value of the diagnostic variable da can be obtained by the expression:

da =

Im ¼

=1

T2

T4

(7)

According to this last equation and taking into account that this theory is also extended to the remaining cases, it can be proved that after the fault occurrence, the diagnostic variable corresponding to the inverter faulty leg will converge to a value of approximately +1 or -1, if the damaged switch is the top or the bottom one, respectively. Furthermore, the diagnostic variable associated to the affected phase will converge more rapidly to its final value. Therefore, a single open-circuit fault is detected by observing which is the first variable dn that crosses a defined threshold

T6

Fig. 2 – Diagram of a typical VSI feeding an ac motor. TABLE I SINGLE IGBT OPEN-CIRCUIT FAULT DETECTION AND LOCALIZATION [30] da db dc Faulty IGBT > kf T1 < ¡kf T2 > kf T3 < ¡kf T4 > kf T5 < ¡kf T6

value kf , whereas its signal gives information about the fault localization. This diagnostic can be easily accomplished taking into account Fig. 2 and using Table I [30]. Regarding the multiple open-circuit fault diagnosis, the formulated variables carry also information that allows to achieve this goal. However, in case of a double fault in the same inverter leg, the diagnostic variables defined by (3) become ill-conditioned since the corresponding value of hjin ji is close to zero. With the aim to handle with these specific faults, three additional auxiliary variables can be defined as: an =

!

¡ ¢ 0 ¡ ¡ I¼m

ib

ic

Regarding the variable jia j and taking into account that under these faulty operating conditions ia just assume negative values, it can be verified that jia j = ¡ia. The reference current signals are always sinusoidal, in accordance to (4). Following this and using (3), the phase a diagnostic variable can be calculated as:

R 2¼ !

T5 ia

2¼ 3 ¢ 2¼ 3

As the inverter is operating normally, the three motor phase currents follow their corresponding reference signals i¤n, which can be also given by (4). As a result, the diagnostic variables dn will be zero since hen i are also zero. In practice en are not exactly zero since they present the currents low amplitude high frequency ripple. Nevertheless, as their average values are calculated, hen i are zero under these conditions. When a fault is introduced, the motor currents cannot perfectly follow their reference signals. Taking as an example a fault in the phase a upper switch, the current in this phase is strongly affected since during its positive half-cycle the current is zero. Under these conditions, during one fundamental period ia can be given by:

! 2¼

T3

2 hjin ji hjil ji + hjim ji

(8)

where l; m; n 2 fa; b; cg, and l 6= m 6= n. Therefore, for a double fault in phase n, the variable an will decrease considerably to a value near to zero, indicating a lack of current flow in that phase. To accomplish multiple fault diagnosis, two more threshold values km and kd can be used and fault symptom variables can be formulated according to the following expressions: 8 > : P if dn ¸ km

( L if an · kd An = H if an > kd

(10)

Taking this into account and considering a typical motor drive system with a voltage source inverter supplying an ac motor (Fig. 2), the generated fault signatures allow to detect

TABLE II MULTIPLE IGBT OPEN-CIRCUIT FAULT DETECTION AND LOCALIZATION Db Dc Aa Ab Ac Faulty Switches Da T1 P 0 0 H H H T2 N 0 0 H H H T3 0 P 0 H H H T4 0 N 0 H H H T5 0 0 P H H H T6 0 0 N H H H T1, T2 − 0 0 L H H T3, T4 0 − 0 H L H T5, T6 0 0 − H H L T1, T4 P N 0 H H H T2, T3 N P 0 H H H T1, T6 P 0 N H H H T2, T5 N 0 P H H H T3, T6 0 P N H H H T4, T5 0 N P H H H T1, T3, [T6] P P N H H H T2, T4, [T5] N N P H H H T3, T5, [T2] N P P H H H T4, T6, [T1] P N N H H H T1, T5, [T4] P N P H H H T2, T6, [T3] N P N H H H T1, T2, (T3|T6) − P N L H H T1, T2, (T4|T5) − N P L H H T3, T4, (T1|T6) P − N H L H T3, T4, (T2|T5) N − P H L H T5, T6, (T1|T4) P N − H H L T5, T6, (T2|T3) N P − H H L Note: [ ] means that the switch may or may not be in open-circuit and ( | ) means that both or at least one switch is in open-circuit.

and localize 27 possible combinations of faulty IGBTs, as shown in Table II. All these combinations correspond exactly to the maximum number of faulty modes that can be distinguished from one another based on just the current analysis. According to Table II, all these open-circuit fault combinations can be separated into five groups as follows: single switch faults; single phase faults; crossed double faults in different legs, i.e., an upper and a bottom switch in different legs are damaged; double faults in upper (lower) switches, which are indistinguishable from triple faults that includes two upper (lower) switches and the bottom (upper) switch of the remaining leg; and triple faults involving a single phase fault, which are also indistinguishable from quadruple faults combining a single phase fault with an upper and a bottom faulty switch in each remaining leg. The threshold values kf , km and kd can be empirically established by simply analyzing the variables behavior for different faulty operating conditions and taking into account a tradeoff between fast detection and robustness against false alarms. A more detailed explanation regarding the selection of these thresholds is presented in Section V. IV. EXPERIMENTAL RESULTS The experimental setup basically comprises a PMSM coupled to a four-quadrant servomotor test system (Fig. 3), a three-phase diode bridge rectifier, a Semikron SKiiP three-

Fig. 3 – Detail of the PMSM coupled to the servo machine.

Fig. 4 – General view of the power and control stages.

Fig. 5 – Block diagram with the components of the experimental setup.

phase VSI, a dSPACE DS1103 digital controller and two precision digital power analysers (Fig. 4). The voltage and current signals as well as all the IGBTs gate commands are connected to the dSPACE controller trough interface and isolation boards (Fig. 5). Two digital power analyzers Yokogawa WT3000 were connected in series in the power circuit in order to measure important variables. The parameters of the PMSM used for the experimental tests are reported in Table III, included in Appendix. The realtime interface board library for the DS1103 controller is designed as a common Matlab/Simulink Blockset that provides blocks to implement the I/O capabilities in Simulink models. The dSPACE ControlDesk experiment software provides functions for real-time control and monitoring,

A.

900 rpm Operation Fig. 6 presents the time-domain waveforms of the motor phase currents together with the diagnostic and auxiliary variables for a fault in IGBT T1 and T5 under 25% of load. Under normal operating conditions, it can be seen that the PMSM phase currents are sinusoidal, resulting in a null value for the diagnostic variables since the actual motor currents follow their corresponding reference signals. At the instant t=0.4961 s a fault in IGBT T1 is introduced, and as a result, the motor phase currents are no longer sinusoidal, being the phase a current half-cycle eliminated. After this moment, the da diagnostic variable increases immediately converging to a value near to 1, while the remaining variables converge to approximately -0.31. Finally, 0.1 seconds after the first fault occurrence, another opencircuit fault is introduced for the IGBT T5. As a consequence, the fault signature changes since the da and dc will converge to approximately 1 whereas the variable db will converge to near -1. Regarding the auxiliary variables, despite that they are not relevant for this situation, all the three take values above 0.2, which indicates, as expected, current flow in all phases. Considering the defined threshold values kf and km and the information presented in Table II and Table III, both failures are correctly detected and localized. Furthermore, the inverter faulty operation, which corresponds to the fault in T1, is detected 700 µs after the power switch failure. In a different way, this is equivalent to a detection time of about 5.25% of the PMSM phase currents period. Fig. 7 presents the time-domain waveforms of the motor

10 Motor Phase Currents (A)

ia

ib

ic

5 0 −5 Normal

−10 0.48

Fault in T1

0.51

0.54

da

db

0.51

0.54

aa

ab

0.51

0.54

Fault in T1+T5

0.57 0.6 Time (s)

0.63

0.66

0.69

0.63

0.66

0.69

0.63

0.66

0.69

Diagnostic Variables

2 1 0 −1 −2 0.48 2 Auxiliary Variables

allowing simultaneously to capture data files that can be plotted using the Matlab. A rotor field oriented control strategy employing hysteresis current controllers was used to control the motor speed. This technique together with the developed diagnostic algorithm were implemented in the DS1103 digital controller board, using a sampling time of 25 µs. The rotor position is obtained by an incremental encoder with 1024 pulses per revolution. Inverter power switch open-circuit faults are controlled by the user using the dSPACE ControlDesk software. These are accomplished by removing the gate command signals of the required IGBTs. Experimental results were obtained for two different mechanical operating speeds, namely 900 rpm and 1500 rpm. Considering both cases, several inverter faulty combinations are analyzed for different load levels equivalent to 25% and 50% of the PMSM rated torque. By doing this, it is possible to evaluate the algorithm performance for distinct mechanical operating conditions and its multiple failures diagnostic capability. The threshold values kf , km and kd were set to be equal to 0.08, 0.5 and 0.2, respectively. The exact procedure on how these values were chosen is explained in a more detailed way in Section V.

dc 0.57 0.6 Time (s) ac

1.5 1 0.5 0 0.48

0.57 0.6 Time (s)

Fig. 6 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T1 and T5 under 25% of rated torque and for a reference speed of 900 rpm.

phase currents together with the diagnostic and auxiliary variables for a simultaneous fault in IGBT T2 and T5 under 25% of load torque, and considering a reference speed of 900 rpm. Like in the previous case, for the inverter healthy operation the PMSM is supplied by a balanced and sinusoidal current system, where the motor measured currents follow their reference signals. At t=0.3274 s, a simultaneous fault in power switches T2 and T5 is introduced. As a consequence, the diagnostic variables of phase a and phase c will converge to -1 and 1, respectively, while db assumes a value near to zero. The information provided by the auxiliary variables an is also not relevant in this case. Nevertheless, under normal and faulty operating conditions, their values converge to approximately 1. The fault signature generated by equations 4 and 5 also allow to effectively detect and localize this inverter faulty combination. Regarding the detection speed, for this specific situation, the inverter abnormal behavior is diagnosed by the identification of transistor T1 in 1.149 ms, which corresponds to approximately 8.62% of the motor currents period. Considering now a double failure in the same inverter leg, Fig. 8 presents the time-domain waveforms of the motor phase currents together with the diagnostic and auxiliary variables for a fault in IGBT T1 and T2, with a load equivalent to 25% of the motor rated torque.

10

10 ib

ic

5 0 −5 Normal

−10 0.3

Fault in T2+T5

0.33

0.36

ia

Motor Phase Currents (A)

Motor Phase Currents (A)

ia

0.39 0.42 Time (s)

0.45

0.48

0 −5 Normal

2 Diagnostic Variables

Diagnostic Variables

0.51

0.54

Fault in T1+T2

0.57 0.6 Time (s)

0.63

0.66

0.69

0.63

0.66

0.69

0.63

0.66

0.69

dc

0 −1

0.33

0.36

0.39 0.42 Time (s)

1.5

0.45

aa

0.48

ab

1 0.75

0.36

0.39 0.42 Time (s)

0.45

0.48

0 −1

2

ac

1.25

0.33

1

−2 0.48

0.51

Auxiliary Variables

Auxiliary Variables

db

1

0.5 0.3

Fault in T1

2 da

−2 0.3

ic

5

−10 0.48

0.51

ib

db 0.54

aa

ab

0.51

0.54

dc 0.57 0.6 Time (s) ac

1.5 1 0.5 0 0.48

0.51

da 0.51

0.57 0.6 Time (s)

Fig. 7 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T2 and T5 under 25% of rated torque and for a reference speed of 900 rpm.

Fig. 8 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T1 and T2 under 25% of rated torque and for a reference speed of 900 rpm.

When a fault in IGBT T1 is introduced at the instant t=0.4960 s, the behavior is virtually the same to the one shown in Fig. 5, being this fault correctly detected and localized in a time interval equivalent to 5.44% of the currents fundamental period. However, immediately after the fault occurrence in T2 at the instant t=0.5960 s, it can be seen that the phase a current is zero, and the PMSM is only supplied by the remaining two healthy phases. As explained before, for this kind of situations, the diagnostic variables become ill-conditioned. This is clearly visible by the high-amplitude oscillations taken by the variable da. In order to handle with these specific faults, the auxiliary variables an assume now an important role. As a result, after the fault in T2, the phase a auxiliary variable decreases considerably to a near zero value, whereas the remaining ones converge to approximately 2. Considering all this, the generated fault signature also allows to correctly detect and localize the two faulty power switches in the inverter phase a. Finally, Fig. 9 presents the time-domain waveforms of the motor phase currents together with the diagnostic and auxiliary variables for a fault in IGBT T1 and T4 under 50% of load torque. These results allow to verify that when a fault in transistor T1 is introduced at t=0.2293 s, the behavior of the motor phase currents and all the considered variables is virtually equal to the previous cases, despite of the higher operating load level.

An equivalent result is also obtained when another fault in IGBT T4 is introduced at t=0.3293 s (double fault in T1 and T4), where it can be observed a similar behavior to the results shown in Fig. 6, where da and db converge to 1 and -1, respectively, and dc assume a near zero value. For this case, the inverter abnormal operation is effectively detected and localized in a time interval of about 848 µs, which corresponds to 6.37% of the motor phase currents fundamental period. B.

1500 rpm Operation Fig. 10 presents the time-domain waveforms of the motor phase currents together with the diagnostic and auxiliary variables for a fault in IGBT T1 and T6 under 25% of load torque, and considering a reference speed of 1500 rpm. Similarly to the previous cases, for healthy conditions, the motor currents follow their reference values and the diagnostic variables assume zero values. When a fault in IGBT T6 is introduced at t=0.6278 s, the value of the phase c diagnostic variable dc decreases immediately, converging to a value near to -1, while the remaining two assume a value of approximately 0.31. Accordingly, using the defined threshold values, the faulty power switch is correctly detect and localized in 475 µs, which is equivalent to approximately 5.94% of the PMSM phase currents period. 50 ms after the

10

16 ib

ic

8 0 −8 Normal

−16 0.21

Fault in T1

0.24

0.27

Fault in T1+T4

0.3

0.33 Time (s)

0.36

ia

Motor Phase Currents (A)

Motor Phase Currents (A)

ia

0.39

−5 Normal

Diagnostic Variables

Diagnostic Variables

1 0 −1 da

db

dc

0.24

0.27

0.3

0.33 Time (s)

0.36

aa

0.39

0.62

0.64

da

db

0.62

0.64

aa

ab

0.62

0.64

0.66 0.68 Time (s)

Fault in T6+T1

0.7

0.72

0.74

0.7

0.72

0.74

ab

1.25 1 0.75

0.24

0.27

0.3

0.33 Time (s)

0.36

0.39

0.7

0.72

0.74

0 −1

1.5

ac

0.42

dc

1

−2 0.6

0.42

Auxiliary Variables

Auxiliary Variables

1.5

0.5 0.21

Fault in T6

2

2

−2 0.21

ic

0

−10 0.6

0.42

ib

5

0.66 0.68 Time (s) ac

1.25 1 0.75 0.5 0.6

0.66 0.68 Time (s)

Fig. 9 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T1 and T4 under 50% of rated torque and for a reference speed of 900 rpm.

Fig. 10 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T6 and T1 under 25% of rated torque and for a reference speed of 1500 rpm.

first fault, another failure in transistor T1 is introduced. As a result, the fault signature changes and the variable da increases to a value near to 1 whereas db decreases to approximately a zero value. The information provided by the auxiliary variables is also not relevant for this case, where all variables are greater than 0.2, generating the same signature. Taking all this into account, and by using Table III, this inverter fault combination is also effectively detected and localized. Considering now the motor operation for a higher load level, Fig. 11 presents the time-domain waveforms of the PMSM phase currents together with the diagnostic and auxiliary variables for a fault in IGBT T1 and T4, with a load equivalent to 50% of the motor rated torque. These faults are introduced at the instants t=0.4576 s and t=0.5076 s, respectively. Despite of the motor higher operating speed, comparing these results to the ones shown in Fig. 9, it can be verified that the behavior of the diagnostic and auxiliary variables is virtually the same. Therefore, the algorithm can also effectively identify the inverter faulty switches. Regarding the detection time, for this case, the inverter abnormal operation is detected and localized 500 µs after the first fault occurrence, which is equivalent to 6.25% of the PMSM phase currents period.

phase currents together with the diagnostic variables for a load transient from rated load torque to no-load conditions, and considering a motor reference speed of 1200 rpm. As the information of the auxiliary variables is not important for this analysis, their behavior is not shown. At the instant t=0.57 s, an instantaneous load torque variation from rated load to no-load conditions is introduced. As a result of this fast and strong transient, the average values of the reference current errors slightly increase, leading to the arising of variations in the diagnostic variables. Under no-load operating conditions, the diagnostic variables present more appreciable variations around zero than for the motor operation at its rated load. This is a consequence of the larger current ripple presented in the PMSM phase currents, which contributes to the increase of the reference current errors amplitude, generating therefore larger variations in the diagnostic variables. Nonetheless, these experimental results also prove that even with this fast and strong transient, the absolute values of the diagnostic variables are still well below from the considered threshold value of 0.08 for kf . Therefore, it can be confirmed that the chosen threshold value allows the proposed diagnostic algorithm to handle with these extreme situations, and obviously, also handling with less significant load variations, without emitting false alarms. Comparing with the majority of the existing methods, they cannot typically handle with such

C.

Load Variation Behavior Fig. 12 presents the time-domain waveforms of the motor

16 Motor Phase Currents (A)

ia

ib

still present values near to zero and, consequently, no false alarms are produced.

ic

8 0

V. OTHER REMARKS −8 Normal

−16 0.44

Fault in T1

Fault in T1+T4

0.46

0.48

0.5

0.52 Time (s)

da

db

dc

0.46

0.48

0.5

A.

0.54

0.56

0.58

0.54

0.56

0.58

Diagnostic Variables

2 1 0 −1 −2 0.44

0.52 Time (s)

Auxiliary Variables

1.5

aa

ab

ac

1.25 1 0.75 0.5 0.44

0.46

0.48

0.5

0.52 Time (s)

0.54

0.56

0.58

Motor Phase Currents (A)

Fig. 11 – Time-domain waveforms of the PMSM phase currents, the diagnostic variables and auxiliary variables for a fault in T1 and T4 under 50% of rated torque and for a reference speed of 1500 rpm.

10 ia

ib

ic

5 0 −5 −10 0.55

0.57

0.59

da

db

0.61 0.63 Time (s)

0.65

0.67

0.69

0.65

0.67

0.69

Diagnostic Variables

0.08 dc

0.04 0 −0.04 −0.08 0.55

Rated torque

No−load conditions

0.57

0.59

0.61 0.63 Time (s)

Fig. 12 – Time-domain waveforms of the PMSM phase currents and the diagnostic variables for a load step variation from rated load torque to no-load and for a reference speed of 1200 rpm.

extreme scenarios, leading to the generation of erroneous diagnostics. Other experimental tests were carried out in order to evaluate the algorithm behavior under speed variations. The obtained results also showed that, even when imposing very high acceleration/deceleration rates, the diagnostic variables

Threshold Values Selection The correct selection of the used threshold values is critically important for the correct algorithm diagnostic performance. In a similar way to the majority of the existing open-circuit fault diagnostic techniques, the definition of the threshold values is accomplished by analyzing the diagnostic variables behavior for the healthy case and all faulty modes. Then, the values are set taking into account a tradeoff between a fast diagnosis and the robustness against the issue false alarms. According to this, the choice of kf is based on the fact that if more than two power switches fail at exactly the same instant, the initial behavior is always equivalent to a single fault in one of them. Therefore, Table I can be used to rapidly diagnose one of the faulty devices using kf . Hence, for the selection of its value, the behavior of dn under normal operating conditions and for a single fault is considered. For these two cases, the absolute values of these variables are between 0 and 1. This means that, theoretically, kf can have a value within this range, where a low value means fast detection speed but worse robustness against false diagnostics. On the contrary, a large value greatly increases the algorithm robustness but decreases its diagnostic speed. The recommended procedure that was followed in this paper consists in analyzing first the diagnostic variables behavior for the worst case scenario, presented in Fig. 12. It is known that the majority of the existing methods fail (by generating false alarms) for these extreme situations since there are very fast variations and the currents become very small, which lead to incorrect average values calculation. All these problematic issues are presented for the transient shown in Fig. 12, including the worst possible case i.e. practically zero currents (between t=0.58 s and t=0.59 s). From these results it can be verified that after the transient, the dn oscillations increase but their amplitude are always below 0.025. This means that this value could be used for kf , assuring a good diagnostic behavior for the other less extreme operating conditions. However, it is always better to consider a safety margin and therefore, the considered value of 0.08 can guarantee a much more robust diagnostic. On the other hand, the maximum value of kf should be lower than 1, since in practice the diagnostic variables may not reach this value. Considering all this and according to the obtained results, kf can be chosen within 0.025 and 0.9. The selection of km is done by analyzing the behavior of the variables dn for single faults and multiple faults (excluding double faults in the same inverter leg). For a single fault, the values of jdn j (corresponding to the faulty leg) will converge

to almost 1 while the other two converge to approximately 0.32. For multiple faults, they will converge to about 1 and 0. Therefore, and taking also into account a safety margin, km can be chosen to be between 0.4 and 0.9. It must be noticed that in order to maintain the diagnostic coherence, the condition kf · km must be verified. The threshold kd is the less important one since it is only used for same inverter leg double faults. Excluding these specific faults, it can be verified from the presented results that the values taken by an are always greater than 0.5. When this fault type occurs, one of the variables will fall to practically zero. Hence, and assuring also a safety margin, kd can assume a value between 0.4 and 0.05. It is worth noting that if a fast diagnosis is not a major requirement (since the lower limit of km is higher than for kf ), the threshold kf and Table I can be neglected and only Table II is used. It must be noticed that as the diagnostic variables are normalized, these thresholds can be considered universal since they do not depend on the motor rated power, its operating load level and mechanical speed. Moreover, the adopted normalization strategy also makes the algorithm much more insensitive to the characteristics of the control scheme since it is based on the vector control basic principle, where the motor phase currents are always controlled in order to follow their reference signals. The used modulation strategy will directly affect the motor phase currents ripple amplitude. Accordingly, if a high switching frequency is used, in this specific case this is accomplished by decreasing the hysteresis band, the current ripple is reduced, and consequently, also the amplitude of en. On the contrary, by increasing the hysteresis band, the current ripple also increases, leading to higher amplitude oscillations in the variables en. Nevertheless, as the average values of these quantities are used, this will act as a low-pass filter. Thus, for both cases, this effect is strongly attenuated, which also makes the algorithm quite insensitive to these issues. B.

Algorithm Detection Speed The algorithm detection speed depends on the time instant of the fault occurrence. If a fault occurs during a positive current half-cycle and in the top IGBT of that phase, the effect can be clearly seen since the current tends immediately to zero. An equivalent result is also verified during a negative current half-cycle for a fault in the respective bottom transistor of that phase. For these cases, the detection and localization is, typically, relatively fast. However, if the fault occurs during a current half-cycle where the faulty IGBT does not immediately affect it, the fault effects will just be noticed at the next current half-cycle. Under these conditions, the fault can remain undetected for a period of time that can reach more than onehalf of the motor phase currents fundamental period. Considering all this, additional experimental tests allow to verify that the diagnostic detection speed can be as fast as 5%

and as slow as 67% of the motor phase currents fundamental period. This can be considered quite fast since, comparing with the majority of the existing methods, the diagnostic is typically accomplish in practically one fundamental period. Finally, as this algorithm requires the reference current signals obtained from the main control system of a vector controlled drive, it cannot therefore be applied to open-loop control strategies such as V/f control, since these signals are not available. VI. CONCLUSIONS A novel algorithm for real-time diagnostic of multiple power switches open-circuit faults in voltage-fed PWM motor drives has been proposed in this paper. The method just uses, as inputs, variables that are already available from the main control system. This means that comparing with other techniques, it avoids the use of extra sensors or electric devices and the subsequent increase of the system complexity and costs. The obtained results allow to conclude that in opposition to the majority of the existing techniques, the presented method can handles with extreme and fast transients, without emitting false diagnostics. Furthermore, thanks to the use of normalized quantities, the algorithm behavior does not depend neither on the motor rated power, and its load level, nor on its mechanical speed. Accordingly, universal threshold values can be defined, independently of these issues. Regarding the diagnostic method detection speed, the obtained results show that, comparing with similar diagnostic methods that usually need almost one fundamental period, the presented algorithm is much faster, allowing to perform a diagnostic in a time interval as short as 5% of the motor phase currents period. Comparing with the existing methods, the developed diagnostic algorithm is also much more simple since it just requires a few and basic mathematical operations. This makes it not computationally demanding and therefore, it can be easily integrated into the main control system without great effort. APPENDIX TABLE III PARAMETERS OF THE USED PMSM Power

P

2.2 kW

Torque

T

12 Nm

Voltage

V

316 V

Current

I

5.3 A

Number of pole pairs

p

5

Armature resistance

Rs

1.72 Ω

Magnet flux linkage

ÃP M

0.244 Wb

d-axis inductance

Ld

20.5 mH

q-axis inductance

Lq

20.5 mH

Moment of inertia

J

0.007 Kg.m2

Finally, as a main drawback it can be pointed out its nonapplicability to open-loop control strategies such as V/f scalar control since the reference current signals are not available.

REFERENCES [1] V. Smet, F. Forest, J. Huselstein, F. Richardeau, Z. Khatir, S. Lefebvre and M. Berkani, “Ageing and failure modes of IGBT modules in hightemperature power cycling”, IEEE Transactions on Industrial Electronics, vol. 58, no. 10, pp. 4931-4941, October 2011. [2] B. A. Welchko, T. A. Lipo, T. M. Jahns and , S. E. Schulz, “Fault tolerant three-phase ac motor drive topologies; a comparison of features, cost, and limitations”, IEEE Transactions on Power Electronics, vol. 19, no. 4, pp. 1108- 1116, July 2004. [3] R. Errabelli and P. Mutschler, “Fault tolerant voltage source inverter for permanent magnet drives”, IEEE Transactions on Power Electronics, vol. 27, no. 2, February 2012. [4] F. Aghili, “Fault-tolerant torque control of BLDC motors”, IEEE Transactions on Power Electronics, vol. 26, no. 2, pp. 355-363, February 2011. [5] S. Dwari and L. Parsa, “Fault-tolerant control of five-phase permanentmagnet motors with trapezoidal back EMF”, IEEE Transactions on Industrial Electronics, vol. 58, no. 2, pp. 476-485, February 2011. [6] H. B. A. Sethom and M. A. Ghedamsi, “Intermittent misfiring default detection and localisation on a PWM inverter using wavelet decomposition”, Journal of Electrical Systems, vol. 4, no. 2, pp. 222-234, 2008. [7] S. Yang, A. Bryant, P. Mawby, D. Xiang, L. Ran and P. Tavner, “An industry-based survey of reliability in power electronic converters”, IEEE Transaction on Industry Applications, vol. 47, no. 3, pp. 1441-1451, May/June 2011. [8] S. Yang, D. Xiang, A. Bryant, P. Mawby, L. Ran, and P. Tavner, “Condition monitoring for device reliability in power electronic converters - a review”, IEEE Transactions on Power Electronics, vol. 25, no. 11, November 2010. [9] F. W. Fuchs, “Some diagnosis methods for voltage source inverters in variable speed drives with induction machines - a survey”, 29th Annual Conference of the IEEE Industrial Electronics Society, vol. 2, pp. 13781385, 2-6 November, 2003. [10] A. M. S. Mendes, A. J. M. Cardoso and E. S. Saraiva, “Voltage source inverter fault diagnosis in variable speed ac drives, by Park's vector approach”, Seventh International Conference on Power Electronics and Variable Speed Drives, Conf. Publ. No. 456, pp. 538-543, 1998. [11] C. Kral and K. Kafka, “Power electronics monitoring for a controlled voltage source inverter drive with induction machines”, IEEE 31st Annual Power Electronics Specialists Conference, vol. 1, pp. 213-217, 2000. [12] D. Diallo, M. E. H. Benbouzid, D. Hamad, and X. Pierre, “Fault detection and diagnosis in an induction machine drive: a pattern recognition approach based on concordia stator mean current vector”, IEEE Transactions on Energy Conversion, vol. 20, no. 3, pp. 512-519, September 2005. [13] F. Zidani, D. Diallo and M. Benbouzid and R. Nait-Said, “A fuzzy-based approach for the diagnosis of fault modes in a voltage-fed PWM inverter induction motor drive”, IEEE Transactions on Industrial Electronics, vol. 55, no. 2, pp. 586-593, February, 2008. [14] R. Peuget, S. Courtine and J. P. Rognon, “Fault detection and isolation on a PWM inverter by knowledge-based model,” IEEE Transactions on Industry Applications, vol. 34, no. 6, pp. 1318–1326, Nov./Dec., 1998. [15] M. Trabelsi, M. Boussak and M. Gossa, “Multiple IGBTs open circuit faults diagnosis in voltage source inverter fed induction motor using modified slope method”, XIX International Conference on Electrical Machines, 6 pp., 6-8 September 2010. [16] R. L. A. Ribeiro, C. B. Jacobina, E. R. C. Silva and A. M. N. Lima, “Fault detection of open-switch damage in voltage-fed PWM motor drive systems,” IEEE Transactions on Power Electronics, vol. 18, no. 2, pp. 587–593, March, 2003.

[17] Q.-T. An, L.-Z. Sun, and T. M. Jahns, “Low-cost diagnostic method for open-switch faults in inverters”, Electronics Letters, vol. 46, no. 14, pp. 1021-1022, July 2010. [18] Q.-T. An, L.-Z. Sun, K. Zhao and L. Sun, “Switching function model based fast-diagnostic method of open-switch faults in inverters without sensors”, IEEE Transactions on Power Electronics, vol. 26, no. 1, pp. 119-126, January 2011. [19] A. M. S. Mendes and A. J. M. Cardoso, “Voltage source inverter fault diagnosis in variable speed ac drives, by the average current Park’s Vector approach”, IEEE International Electric Machines and Drives Conference, Seatle, Washington, USA, pp. 704-706, May 9-12, 1999. [20] W. Sleszynski, J. Nieznanski and A. Cichowski, “Open-transistor fault diagnostics in voltage-source inverters by analyzing the load currents”, IEEE Transactions on Industrial Electronics, vol. 56, no. 11, pp. 46814688, November, 2009. [21] K. Rothenhagen and F. W. Fuchs, “Performance of diagnosis methods for IGBT open circuit faults in three phase voltage source inverters for ac variable speed drives”, European Conference on Power Electronics and Applications, 10 pp., 2005. [22] B. Lu and S. K. Sharma, “A literature review of IGBT fault diagnostic and protection methods for power inverters”, IEEE Transactions on Industry Applications, vol. 45, no. 5, pp. 1770-1777, Sept./Oct., 2009. [23] N. M. A. Freire, J. O. Estima and A. J. M. Cardoso, “Multiple opencircuit fault diagnosis in voltage-fed PWM motor drives using the current Park’s Vector phase and the currents polarity”, 8th IEEE International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives, Bologna, Italy, pp. 397-404, 5-8 September 2011. [24] J. O. Estima and A. J. M. Cardoso, “A new approach for real-time multiple open-circuit fault diagnosis in voltage source inverters”, IEEE Transactions on Industry Applications, vol. 47, no. 6, 8 pp., November/December 2011. [25] D. U. Campos-Delgado and D. R. Espinoza-Trejo, “An observer-based diagnosis scheme for single and simultaneous open-switch faults in induction motor drives”, IEEE Transactions on Industrial Electronics, vol. 58, no. 2, pp. 671-679, February 2011. [26] B.-G. Park, K.-J. Lee, R.-Y. Kim, T.-S. Kim, J.-S. Ryu and D.-S. Hyun, “Simple fault diagnosis based on operating characteristic of brushless direct-current motor drives”, IEEE Transactions on Industrial Electronics, vol. 58, no. 5, pp. 1586-1593, May 2011. [27] M. A. Rodríguez-Blanco, A. Claudio-Sánchez, D. Theilliol, L. G. VelaValdés, P. Sibaja-Terán, L. Hernández-González, J. Aguayo-Alquicira, “A failure-detection strategy for IGBT based on gate-voltage behavior applied to a motor drive system”, IEEE Transactions on Industrial Electronics, vol. 58, no. 5, pp. 1625-1633, May 2011. [28] K.-H. Kim, B.-G. Gu and I.-S. Jung, “Online fault-detecting scheme of an inverter-fed permanent magnet synchronous motor under stator winding shorted turn and inverter switch open”, IET Electric Power Applications, vol. 5, no. 6, pp. 529-539, July 2011. [29] M. Aktas and V. Turkmenoglu, “Wavelet-based switching faults detection in direct torque control induction motor drives”, IET Science, Measurement & Technology, vol. 4, no. 6, pp. 303-310, November 2010. [30] J. O. Estima and A. J. M. Cardoso, “Single power switch open-circuit fault diagnosis in voltage-fed PWM motor drives by the reference current errors”, 8th IEEE International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives, Bologna, Italy, pp. 364-371, 58 September, 2011. [31] K. Nguyen-Duy, L. Tian-Hua, C. Der-Fa and J. Y. Hung, “Improvement of matrix converter drive reliability by online fault detection and a faulttolerant switching strategy”, IEEE Transactions on Industrial Electronics, vol. 59, no. 1, pp. 244-256, January 2012. [32] S. Khwan-on, L. de Lillo, L. Empringham and P. Wheeler, “Fault-tolerant matrix converter motor drives with fault detection of open switch faults”, IEEE Transactions on Industrial Electronics, vol. 59, no. 1, pp. 257-268, January 2012.

Jorge O. Estima (S’08, M’10) was born in Aveiro, Portugal in 1984. He received the Diploma degree in electrical engineering from the University of Coimbra, Portugal, in 2007, where he is currently working toward the Ph.D. degree with the Department of Electrical and Computer Engineering, Faculty of Sciences and Technology. Since 2008 he has also been with the Instituto de Telecomunicações. His research interests are focused on condition monitoring and diagnostics of power electronics and AC motor drives, fault-tolerant variable speed AC drives, traction motor drives applied to electric/hybrid vehicles and wind energy conversion systems. António J. Marques Cardoso (S’89, A’95, SM’99) was born in Coimbra, Portugal, in 1962. He received the Electrical Engineering diploma, the Dr. Eng. degree and the Habilitation degree, all from the University of Coimbra, Coimbra, Portugal, in 1985, 1995 and 2008, respectively. From 1985 until 2011 he was with the University of Coimbra, Coimbra, Portugal, where he was Director of the Electrical Machines Laboratory. Since 2011, he has been with the University of Beira Interior (UBI), Covilhã, Portugal, where he is a Full Professor at the Department of Electromechanical Engineering. His teaching interests cover electrical rotating machines, transformers, and maintenance of electromechatronic systems and his research interests are focused on condition monitoring and diagnostics of electrical machines and drives. He is the author of a book entitled Fault Diagnosis in Three-Phase Induction Motors (Coimbra, Portugal: Coimbra Editora, 1991), (in Portuguese) and about 300 papers published in technical journals and conference proceedings.