A Real-Time Method for the Diagnosis of Multiple Switch Faults in NPC Inverters Based on Output Currents Analysis

Abstract

This paper presents a new approach for fault diagnosis in three-level neutral point clamped inverters. The proposed method is based on the average values of the positive and negative parts of normalized output currents. This method is capable of detecting and locating multiple open-circuit faults in the controlled power switches of converters in half of a fundamental period of those currents. The implementation of this diagnostic approach only requires two output currents of the inverter. Therefore, no additional sensors are needed other than the ones already used by the control system of a drive based on this type of converter. Moreover, through the normalization of currents, the diagnosis is independent of the load level of the converter. The performance and effectiveness of the proposed diagnostic technique are validated by experimental results obtained under steady-state and transient conditions.

NOMENCLATURE

x ∈ {A,B,C} Inverter leg index

ix Inverter output current of leg x

ix.n ix normalized

Positive and negative parts of ix.n

Average values of

First and second threshold values for

First and second threshold values for

Ith+, Ith- Threshold values for ix.n

Diagnostic variables

Time of the occurrence of the first and second faults

First and second faulty pair number

First and second faulty switch number

 

I. INTRODUCTION

Multilevel converters are considered a very attractive solution for both rectifier and inverter stages of power converters used in medium-voltage and high-power drives [1], [2]. Higher voltage blocking capability, reduced output harmonic content, lower switching losses, higher power quality waveforms and lower voltage stresses are some of the advantages of multilevel converters when compared to conventional two-level converters [3], [4].

Three-level neutral point clamped (3LNPC) inverters belong in a category of multilevel converters that are increasingly being used in industry [5], [6]. Wind energy conversion [7], photovoltaic conversion [8] and multiphase AC drives [9] are some of the applications of 3LNPC inverters.

However, due to the complexity of the 3LNPC topology, and the requirement of a large number of switches and gate drivers, the probability of the occurrence of a fault is not negligible [10].

One common type of fault in the power switches of a converter is an IGBT open-circuit (OC) fault [11], [12]. This kind of fault can arise if the IGBT rated parameters, such as the rated collector-emitter voltage Vce or the rated collector current Ic, are exceeded, or when the gate driver unit fails [13].

In standard applications, after the occurrence of a fault in one or more of the power switches, the converter is stopped. However, in recent years, there has been an increase in the demand for nonstop service in many industrial applications. To achieve this goal, the power converter should be able to operate continuously, even during fault conditions, being that one of the main aims of fault tolerance [14].

To implement fault tolerant control strategies in a converter, the use of a reliable real-time diagnostic system is required. Furthermore, a diagnostic system with the ability to identify the faulty element has some advantages such as the ability to select the best fault tolerant control strategy, or the reduction of the inverter repair time as well as the associated repair cost.

This work gives special attention to the OC faults in IGBTs since in many cases this type of fault is not detected by the IGBT gate driver circuit or by converter protection systems. If an OC fault is not detected and the protection system of the converter does not actuate, it will continue to run and can be the source of other faults [15].

Several approaches were proposed for fault diagnosis in 3LNPC inverters. A method for detecting the OC faults in 3LNPC inverters by measuring the pole voltage is presented in [16].A method for OC fault diagnosis in grid-connected 3LNPC inverters based on the output line currents is proposed in [10]. In [17], a method for detection of the OC faults in the 3LNPC back-to-back converters used in permanent magnet synchronous generators (PMSG) is proposed.

A new diagnostic method based on the average current Park’s Vector (ACPV) is proposed in [18], [19]. This method does not need any extra sensors and is able to identify the single faulty IGBT in a 3LNPC inverter.

The state of the art in fault diagnosis in 3LNPC inverters indicates that all of the proposed approaches for fault diagnosis in 3LNPC converters only pay attention to the occurrence of a single fault and do not have the ability to identify simultaneous failures. Due to the high number of power switches, the probability of the occurrence of the multiple faults in 3LNPC inverters is not negligible. If more than one OC fault is present, in some cases the implementation of fault tolerant control strategies is impossible. In such cases, the diagnostic method should be able to diagnose multiple IGBT OC faults to avoid running the reconfiguration by the fault tolerant system. Otherwise, the results obtained by the fault diagnostic system lead to run the fault tolerant strategy in an inappropriate way. This can be the cause of further damage in the other healthy devices of the inverter.

In order to overcome this limitation, this paper presents a new fault diagnostic method based on the average values of the positive and negative parts of the normalized motor line current waveforms. The proposed diagnostic approach is able to identify multiple OC faults in the controlled power switches of 3LNPC inverters.

The remaining part of this paper is structured as follows. Section II presents the 3LNPC operation under normal and IGBT OC fault conditions. Section III presents the proposed technique for the diagnosis of OC faults. Details of the experimental setup and discussion of the experimental results are presented in Section IV, which includes three subsections with results for single faults, multiple faults and load variations. Finally, Section V presents the main conclusions of this work.

 

II. 3LNPC DRIVE UNDER HEALTHY AND FAULTY CONDITIONS

The topology of a 3LNPC inverter is shown in Fig. 1 [20]. It consists of three legs, with four active switches per leg, labeled Sx1, Sx2, Sx3, Sx4, four freewheeling diodes labeled Dx1, Dx2, Dx3, Dx4 and two clamping diodes labeled Dx5 and Dx6 (x means leg A, B or C). Each leg includes 2 IGBT pairs, Px1 and Px2, as shown in Fig. 1. Moreover, in this work, Sx1 and Sx4 are the outer switches while Sx2 and Sx3 are the inner switches. The active switches are IGBTs and the DC-bus has 2 capacitors, which provide the middle point “O”.

Fig. 1.Schematic representation of a 3LNPC inverter.

The IGBTs of each inverter leg provide three different switching states, designated as 1, 0, and -1, according to the information listed in TABLE I. More information about the operating modes of inverter legs can be found in [21].

TABLE ISWITCHING STATES OF LEG X

Fig. 2 shows the 6 available current paths for each leg according to the current direction and the corresponding state [16]. The unavailable switching state(s) and current path(s) in leg x after the occurrence of an OC fault are listed in TABLE II for 10 different faulty switch combinations (conditions 1 to 4 for a single fault, and conditions 5 to 10 for double faults).

Fig. 2.Current paths according to the current direction and leg state.

TABLE IIINVERTER LEG SITUATION AFTER AN OC FAULT IN LEG X

When an inverter leg is in switching state 1 and the output current is positive (according to Fig. 2), if a fault occurs in Sx1, the output terminal of the affected leg is disconnected from the positive terminal of the DC bus and is connected to terminal O. Hence, current path 1 is changed to current path 2. As a result, ix has a smaller amplitude when compared with the normal operation of that leg. When a fault occurs in Sx2, no positive current ix flows to the load (motor). Thus, the behavior of the positive values of ix depends on whether the fault occurs in an outer IGBT during switching state 1 (small positive current values available) or in an inner IGBT with states 1 or 0 (no positive current values are available).

 

III. PROPOSED DIAGNOSTIC TECHNIQUE

A new method for the diagnoses of multiple OC faults in a 3LNPC inverter, based on the average values of the positive and negative parts of the line currents, is proposed in this paper.

To obtain a diagnostic method independent of the load condition, the motor line currents iA[k], iB[k] and iC[k] are normalized and ix.n[k] is obtained according to (2), using the magnitude of the motor line current Park’s Vector |is[k]| obtained by (1). Then, the normalized line current ix.n[k] is split into two parts, one with positive values and the other with negative values which are determined by (3) and (4), respectively. Thus, the average values of are calculated by (5) and (6), respectively, where N is the number of samples in one period of currents, k is the index of a sample, and x is the inverter leg index.

Under healthy conditions, signals are equivalent to a half-wave rectified signal. Therefore,

After the occurrence of an OC fault in one of the IGBTs located in Px1 (Sx1 or Sx2), decreases and changes a little but is still far away from the threshold. In the presence of an OC fault in one of the IGBTs located in Px2 (Sx3 or Sx4), increases and changes a little but is still far away from the threshold. The evolution of under an OC fault can be used for fault detection and the identification of the faulty IGBT pair (Fpair).

In order to implement the diagnostic algorithm in a 3LNPC digital controller and to have the ability to detect and locate the faulty IGBT, it is necessary to use the diagnostic variables

According to (7) and (8), the variables are obtained from the comparison of with the corresponding threshold values According to (9), the variable fx[k] is obtained from the comparison of ix.n[k] with the threshold values Ith+ and Ith-.

According to (7) and (8), if the inverter operates with no faults, the diagnostic variables present null values. In addition, under normal conditions, the variable fx[k] changes between -1, 0 and 1, as shown in (9).

Details of the threshold values used in this paper are listed in TABLE III. These threshold values have been chosen by a trial and error procedure for several faulty conditions of the 3LNPC. Later they were validated by simulation and experimental diagnostic results. Since ix.n[k], are normalized values, the employed threshold values are independent of the operating conditions of the inverter and do not need to be adjusted for each specific operating condition.

TABLE IIIDETAILS OF THE USED THRESHOLDS VALUES

For fault detection and identification Fpair, are compared with two threshold values respectively. These threshold values represent 31% of the values of under healthy conditions. Therefore, there is a sufficient safety margin to avoid false alarms due to more severe or unusual working conditions such as during transient operation or unbalanced load conditions.

For discrimination between the outer and inner IGBT faults in an IGBT pair, ix.n[k] is compared with two threshold values Ith+ = 0.1 and Ith- = -0.1 (10% of the maximum value of ix.n under healthy conditions), while are compared with two threshold values (3.1% of the values under healthy conditions).

To address the detection and identification of Fpair, the values of are used, as presented in TABLE IV. If present a value that is higher than 0, a fault is detected and both Fpair and the faulty leg (Fleg) are immediately identified. The symbol “--“ in TABLE IV means that the variable value is irrelevant for the diagnostic process.

TABLE IVLOOKUP TABLE FOR FAULTY PAIR IDENTIFICATION

After Fpair identification, Fpair together with the values of are necessary to identify Fswitch, as shown in TABLE V. For example, if a fault occurs in the IGBT pair PA1, based on TABLE V, the values of fA[k] and are necessary to identify Fswitch. If fA[k] takes the value 1, then Fswitch=SA1, and if takes the value 2, Fswitch=SA2.

TABLE VLOOKUP TABLE FOR IDENTIFICATION OF THE FAULTY IGBT(S)

The block diagram and the flowchart of the proposed OC fault diagnostic procedure are presented in Fig. 3 and Fig. 4, respectively.

Fig. 3.Block diagram of the proposed diagnostic algorithm.

Fig. 4.Flowchart of the proposed diagnostic algorithm.

 

IV. VALIDATION OF THE DIAGNOSTIC APPROACH

A. System under test

To test and validate the proposed diagnostic approach, a 3LNPC motor drive system, comprising an induction motor fed by a 3LNPC converter and controlled by a rotor field oriented control (RFOC) strategy was used.

A simulation model of this system was built in the Matlab/Simulink environment. In this model, the space vector modulation (SVM) technique with even-order harmonic elimination was used [20], [22].

In addition to the simulation model, an experimental setup (Fig. 5) was also built. A 3LNPC inverter prototype was developed, where the power switches are Semix IGBT modules, with SKYPER 32 PRO drivers that constitute an interface between the IGBT modules and the controller. OC faults can be introduced in all of the power switches of the inverter in a controlled way. In the experiments, OC faults were introduced by sending the zero control state to specific IGBTs.

Fig. 5.The experimental setup.

This 3LNPC inverter feeds a WEG W22, 4 kW, 400 V, 50 Hz, 8.12 A, 1435 rpm, three-phase induction motor. The motor and inverter control systems, as well as the diagnostic system, are implemented in a digital controller based on a dSPACE ds1103 platform.

The 3LNPC inverter is supplied by a constant DC voltage of 200 V. The mechanical load of the induction motor is a WEG 4 kW permanent magnet synchronous generator feeding a variable load resistor.

B. Experimental results

In order to assess the proposed diagnostic technique, several IGBT OC faults were introduced into the 3LNPC inverter. Each IGBT OC fault test was conducted for different values of motor load torque and several reference speeds, under both steady state and transient conditions.

In this subsection, results for different inverter faulty conditions, with single and double IGBT OC faults are presented.

1) Results for a single IGBT OC fault

In this subsection, results for three different single IGBT OC faults are analyzed. The employed reference speed profile for the first and second tests is shown in Fig. 6. In both tests, at t=3 s (in the middle of the acceleration ramp), an OC fault is introduced in the inverter. In the first test, an OC fault is introduced in IGBT SA1. In the second test, the fault is introduced in IGBT SA2.

Fig. 6.The employed profile of the reference speed.

The motor mechanical speed and load torque evolutions for these faulty conditions are shown in Fig. 7. In these figures, the instant when the fault was introduced is marked with a vertical line labeled FT. As can be noticed, before the fault occurrence, the motor speed follows the reference and the motor electromagnetic torque is stable with 5 Nm. After t=3 s, both of the results show that the motor is unable to follow the speed reference and the motor torque has oscillations. The figures also show that the oscillations of the motor speed and torque due to SA2 fault are higher than those with SA1 fault. The motor line current waveforms with an OC fault in the IGBTs SA1 and SA2 are represented in Fig. 8(a) and Fig. 8(b), respectively.

Fig. 7.Transient of motor speed and load torque.

Fig. 8.Transient of motor line currents.

The diagnosis of these faults is developed in two steps. The first step involves the OC fault detection and the faulty IGBT pair (Fpair) identification. For this purpose, TABLE IV is used. After the identification of Fpair, it is necessary to check the diagnostic variable values presented in TABLE V, in order to identify the faulty IGBT.

For the case of an OC in SA1, the results are presented in Fig. 9. The zoomed motor line current waveforms around t=3 s are presented in Fig. 9(a). Considering an OC fault at t=3 s in the inverter leg A, Fig. 9(b) shows the normalized current of leg A (iA.n[k]) together with the average values of both the positive part and the negative part As can be noticed, before the fault occurrence, iA.n[k] has a sinusoidal shape with amplitude of 1 pu and therefore After the appearance of the fault, almost all of the positive values of iA.n[k] are eliminated and the value of decreases. When is below the threshold value (Fig. 9(b)), takes the value 1 (at t=3.0115 s) as shown in Fig. 9(c). Considering these results, TABLE IV allows the identification of Fleg=1 (leg A) and Fpair=1 (PA1).

Fig. 9.Results for the inverter operation with OC fault in SA1.

Since Fpair=1, there are two possible faulty switches, SA1 and SA2. Thus, based on TABLE V, the value of and fA[k] are necessary to identify the faulty switch. The first time that fA[k] takes the value of 1 after the faulty pair 1 identification, TABLE V allows for identifying SA1 as the faulty IGBT. As can be seen in the diagnostic timeline represented in Fig. 9(d), the IGBT pair number 1 (PA1) and the faulty switch number 1 (SA1) are identified by the algorithm.

For the second test with an OC in SA2, a zoom of Fig. 8(b) around t=3 s, is presented in Fig. 10(a). The results for iA.n[k], are shown in Fig. 10(b).

Fig. 10.Results for the inverter operation with OC fault in SA2.

After the fault occurrence at t=3 s, the positive part of iA.n[k] is completely eliminated. Therefore, decreases to zero. As observed in Fig. 10(b), decreases to a value below t=3.0115 s. At this point (Fig. 10(c)) and the fault is detected. Using these diagnostic variable values as entries in TABLE IV, the identification of the faulty IGBT pair as number 1 (PA1) is achieved (Fig. 10(d)).

After fault detection, taking into account that Fpair=PA1, identifying the faulty IGBT is necessary to check the values of and fA[k]. Fig. 10(c) shows that after fault occurrence, fA[k] changes between -1 and 0, and assumes the value 2 at t=3.0175 s. Considering these values, TABLE V allows for the identification of SA2 as the faulty IGBT.

The timeline of the diagnostic process is shown in Fig. 10(d).

In the third test, the capability of the proposed method under unbalanced load conditions (that can be the result of a motor fault for example) is evaluated. To emulate a fault in the motor, an extra resistor is connected in series with phase A. As a consequence of this procedure, the motor line currents became very unbalanced (equivalent to a motor fault with a high severity). Then, under this unbalanced operating condition, an OC fault is introduced in IGBT SA1.

The motor speeds and load torque evolution are shown in Fig. 11(a), where the time of occurrence of the unbalanced load and the IGBT OC fault are marked with two vertical lines labeled t1 and t2, respectively. The vertical line labeled t3 indicates the instant in time when the OC fault in SA1 is eliminated. As shown in this figure, after the occurrence of the motor fault (1 s

Fig. 11.Results for an OC in SA1 under unbalanced operation.

After the occurrence of the motor fault (1 s3.2 s), the value of increase to a value higher than

Zooms of Fig. 11(b) and Fig. 11(c) around t=3 s are presented in Fig. 12(a) and Fig. 12(b), respectively. As can be observed in Fig. 12(a), before the occurrence of the IGBT fault (t<3 s), the motor line currents are unbalanced (due to the motor fault), while and (Fig. 12(b)). After the appearance of the fault, the value of decreases and becomes smaller than the threshold Therefore, changes to 1 (at t=3.0136 s), as shown in Fig. 12(c). Based on this value, and considering TABLE IV, Fpair is identified as pair 1 and SA1 is identified as the faulty switch as soon as fA[k] changes to 1 for the first time (at t=3.0167 s).

Fig. 12.Results for the inverter operation with an OC fault in SA1 and with an unbalanced load.

The final diagnostic results are shown in Fig. 12(d). These results, which were obtained under an enormous unbalance in the motor line currents, shows that the proposed method does not give any false alarms under these conditions.

2) Results for multiple OC faults

Several tests were also conducted for different combinations of double OC faults. Results for three different double IGBT OC faults under transient operating condition are presented. The employed reference speed profile for all of the results represented in this subsection is similar to the one shown in Fig. 6.

In the first case, results for two OC faults occurring in two IGBTs located in the same inverter leg (SB4 and SB2) are presented. The first OC is introduced in SB4 at t=8 s and the second one is applied to SB2 at t=8.015 s.

The motor speeds and load torque evolution are shown in Fig. 13(a). The motor line current waveforms are shown in Fig. 13(b). A zoomed version of these current signals, centered at t=8 s, is presented in Fig. 14(a). The two vertical lines labeled indicate the instants in time when the faults in SB4 and SB2, are introduced. Since both OC faults are located in inverter leg B, iB.n[k], are the variables needed for the diagnostic process, and are presented in Fig. 14(b). As observed in this figure, after the occurrence of the first fault in SB4 at t=8 s, some of the negative values of iB.n[k] are eliminated and, therefore, rises. Once is larger than assumes the value of 1, and the first fault is detected by the diagnostic system at t=8.011 s. According to TABLE IV, these results allow for the identification of Fpair=4 (PB2). At this point, the behavior of fB[k] shown in Fig. 14(c), permits the identification of SB4 as the first faulty IGBT at t=8.0125 s.

Fig. 13.Results for two OC faults in SB4 and SB2 under transient operation.

Fig. 14.Results for the inverter operation with two OC faults in SB4 and SB2.

The second OC fault is applied in SB2 at t=8.015 s. As can be observed in Fig. 14(b), after this moment, decreases to zero. When is lower than takes the value of 1 at t=8.028 s as shown in Fig. 14(d). This allows for the identification of PB1 as the faulty pair.

To identify the second OC fault, it is necessary to analyze the values of and fB[k], presented in Fig. 14(c). When takes the value of 2 (at t=8.0335 s), the identification of SB2 as the second faulty IGBT is achieved.

At this point, a different combination of double OC faults is analyzed for two different inverter legs. The first OC fault is applied to IGBT SA4 at t=8 s, and the second one is applied to IGBT SC1 at t=8.01 s. The motor line current waveforms are shown in Fig. 15(a).

Fig. 15.Results for the inverter operation with two OC faults in SA4 and SC1.

For the OC fault in leg A, iA.n[k], are represented in Fig. 15(b). Following the procedures previously mentioned, at t=8.0115 s which allows for the identification of the faulty pair PA2. Fig. 15(c) also shows that after fault detection, fA[k] takes the value of -1 at t=8.012 s. With these collected values for Fpair and fA[k], the identification of SA4 is accomplished.

For the OC fault in leg C, iC,n[k], are represented in Fig. 15(d). Fig. 15(e), shows that at t=8.017 s takes the value of 1, and TABLE IV allows for the identification of Fpair=PC1. Since at the moment of the fault detection fC[k]=1, SC1 is identified as the second faulty IGBT.

For the third case, another combination of double OC faults is analyzed. The first OC fault is applied to IGBT SA4 at t=8 s (similar to the previous example) and the second one is applied to IGBT SC2 at t=8.01 s. The motor line current waveforms are represented in Fig. 16(a).

Fig. 16.Results for the inverter operation with two OC faults in SA4 and SC2.

The procedure to identify the first faulty IGBT is similar to that in the previous example.

To analysis the second OC fault, iC,n[k], are represented in Fig. 16(d). This figure shows that after the fault in SC2, decreases to below at t=8.017 s. Therefore, takes the value of 1, and TABLE IV allows for the identification of PC1 as the second faulty pair.

To identify the IGBT with the OC fault, and fC[k] are shown in Fig. 16(e), at the moment takes the value of 2 (t=8.023 s), and independently of fC[k] value, TABLE V give the results of SC2 as the second faulty IGBT. Fig. 16(f) represents the summary of the diagnostic results for this double OC fault.

If a double OC fault occurs in the IGBTs of any other inverter leg, the procedure to identify the faulty IGBT is similar to the one previously explained. TABLE VI summarizes the time taken by the diagnostic algorithm to detect and identify the faulty IGBTs in the experimental tests.

TABLE VIDIAGNOSTIC TIMES OF THE PROPOSED APPROACH

As can be seen, for all of the inverter fault operations, the diagnostic is performed within half a period of the fundamental motor line currents.

3) Results for load variation

The proposed diagnostic method was also tested for a load variation. The motor runs at 500 rpm with no load. At t=0.48 s, the load changes to 5 Nm. The results are shown in Fig. 17, where the load changing instant is marked with a vertical line labeled Tv. The torque and speed variations are shown in Fig. 17(a). As can be seen in Fig. 17(b), there is a large increase in the motor line currents. It can also be seen that there are large oscillations.

Fig. 17.Results for the inverter operation under a torque variation between no load and 5 Nm, without fault.

As already explained (see equation (2)), the motor line current waveforms are normalized by dividing to the magnitude of the motor line currents Park’s Vector, presented in Fig. 17(c). As an example, iA,n[k], are represented in Fig. 17(d), which shows that their values are slightly affected by the load variation. However, despite this transient condition, are far from the corresponding threshold values of Therefore, there is a sufficient safety margin to avoid any false alarms due to load variation conditions.

 

V. CONCLUSIONS

This paper proposes a new approach for fault diagnosis in 3LNPC induction motor drives based on the average values of the positive and negative parts of the normalized motor line currents. This method is able to diagnose multiple OC faults in all of the controlled power switches of the inverter.

Its implementation only requires two motor line currents, which are already used by the AC drive controller. Therefore, it does not require the installation of any additional hardware or sensors. Thus, guaranteeing a low implementation cost and ease of integration into motor drive control systems.

Typically, the faulty IGBT is isolated within half of a fundamental inverter output period. The experimental results obtained with the proposed diagnostic approach under transient conditions show that this approach works well under all conditions without any incorrect results. Furthermore, the capability of the method is verified under unbalanced load operating conditions.

The proposed method has several advantages such as its ability to detect and identify multiple OC faults in all of the controlled power switches without the need for extra sensors. This diagnostic method also possess independence from the working conditions, easy implementation and reliable performance under transient working conditions.