Wind turbine power converter fault diagnosis using DC-link voltage time

Abstract

With the developments in power electronic devices, there is increasing use of the insulated gate bipolar transistor devices in power converters. Power converters are more and more gaining attention in the wind energy conversion system. In this article, a grid-connected wind energy conversion system is considered. Wind energy conversion system optimal operation requires a diagnostic method for back-to-back power converter to be addressed in detail in this article. Therefore, an open-circuit switching fault diagnosis method for a back-to-back power converter is developed to improve the system reliability and optimize the produced power and also achieve its real operational cost. In this work, we investigate only the DC-Link voltage signal and we use the time–frequency analysis to achieve the diagnosis purpose. The scheduled diagnosis approach is divided into two tasks: the first one is the fault detection task and the second one is the fault localization task. The main novelty of the proposed diagnosis approach is to localize the appropriate faulty power converter, that is, rotor side power converter or grid side power converter. Also, this approach is able to localize faults when there is a simultaneous fault on both power converters. Simulations are carried out to verify the robustness of the proposed diagnosis approach. The achieved simulation results would help in diagnosing of Back-to-back power converter that would expectedly cancel the traditional diagnosis method.

Keywords

Back-to-back converter DC-link voltage diagnostic discrete wavelet transform open-circuit semiconductor short time Fourier transform wind energy conversion systems

Introduction

Around the world, renewable energy has provided an important contribution to save the environment. Particularly, the Wind Energy Conversion System (WECS) is under the fast-rising type of renewable energy. For instance, recent statistics affirmed that global cumulative installed WECS capacity growth with a rate of 10% from 2016 to 2017 (Statistics, 2017). WECSs can be classified into two configurations: variable speed turbine generation and fixed speed turbine generation. One of the most widely used structures in currently installed WECS to realize the variable speed turbine generation configuration is based on the Doubly Fed Induction Generator (DFIG).The main advantages of the DFIG-based WECS over other variable speed turbine generation are the reduced mechanical loads, the efficiency of the energy conversion, and the reduced rate of the power converters subsystem (Yenduri and Sensarma, 2016).

With the increasing deployment of WECS, many challenges have appeared. In most instances, WECSs are installed in remote places. As a consequence, the maintenance mission is usually costly and it is time-consuming. Hence, the reliability of WECS is of paramount concern where it has gained much attention. On the other side, it has been reported in several WECS reliability survey that power converters are the most fragile subsystems in the WECS where they are responsible for more than 22% of the total downtimes of the WECS. This statistic shows the importance of the power converter reliability purpose (Kaidis et al., 2015).

The most common power converter topology used in DFIG-based WECS is the two-level topology. The simple structure of this topology is an advantage. Nevertheless, this topology is sensitive to its power semiconductor switches’ failures. Noteworthy, any failure in the power converter subsystem will decrease the system performance in term of power quality and energy efficiency; also the failure may propagate to other subsystems of the WECS. Moreover, it can cause the shutdown of the whole WECS. Therefore, an additional monitoring and diagnostic systems are required to prevent the damage of the whole WECS and improve the reliability and availability of this one (Amirat et al., 2009; Blaabjerg and Ma, 2013).

Typically, the major power semiconductor adopted in the power converter is the Insulated Gate Bipolar Transistor (IGBT) module. IGBT failures are divided into open-circuit (OC) and short-circuit (SC); both OC and SC failures are usually caused by a single overstress event. The OC fault is caused by different mechanisms such as gate driver failure, bond wire lift-off, and solder fatigue. The SC fault is caused by different mechanisms such as high voltage breakdown, static/dynamic latch-up, high temperature and impact ionization (Yang and Chai, 2016; Yang et al., 2011).

Furthermore, the major difference between the mentioned fault types is impaction. OC failure mainly causes the current distortion, where this distortion can propagate to the rest subsystems of the WECS, imposes a quality power supply problem and generates fluctuation which disturbs the electrical network. Noteworthy, the OC fault is not easily detected, then it is classified as a disturbing failure. Usually, SC faults cause a current larger than the rated current which leads to the destruction of other components, and then it is classified as a catastrophic failure. Therefore, in such a case, the system must stop in order to protect all the rest system components. Hence, crucial embedded protections have been employed to avoid the damage of SC fault such as fuses, and circuit breakers (Boudjellal and Benslimane, 2016).

Usually, the OC switching fault is less destructive than SC switching fault. Nevertheless, it is hard to detect and protect the WECS from the OC switching fault early enough (Choi et al., 2016). From the mentioned fault types, a power converter suffers mainly from the OC switching fault; therefore, the goal of this work is to build a diagnostic approach in order to diagnose defects due to the OC witching fault.

In this article, we propose an effective fault diagnosis method to detect and localize the OC switching fault as fast as possible in Back-to-Back Power Converter (BBPC), to increase the reliability of the WECS, and to reduce maintenance cost. This prevents further damage to WECS and reduces BBPC subsystem downtime. The first step in reacting to an OC switch fault in the power converter is to detect quickly the occurrence of fault event, also to localize the failure. The proposed fault diagnosis schemes for BBPC is discussed in the rest of this article.

The principal contributions of this work can be summarized as follows:

The signature signal number is reduced to only one signal.

A practical diagnosis approach is presented using the time–frequency analysis to analyze the signature signal.

Ability to detect and localize not only single OC switching fault, but also simultaneous OC switching fault in the BBPC.

Provide a fast fault detection process.

This article is organized as follows. The second section describes the operation of the WECS under an OC switching fault. The third section presents a brief review of the OC fault diagnosis method. The proposed diagnosis approach and theoretical backgrounds are presented in the fourth section. These simulation results are presented and discussed in the fifth section. The final section concludes this article.

Operation of the grid-connected WECS under open-circuit fault condition

In this section, the operation of the grid-connected WECS under the OC faults, considering the OC faults in phase A, is analyzed. Two different kinds of OC faults within the two-level converter are considered: the first is when the OC fault occurs in one of the power semiconductors switches in the Grid Side Power Converter (GSPC), and the second is when the OC fault occurs in one of the power semiconductor switches in the Rotor Side Power Converter (RSPC). As shown in Figure 1, the OC faulty operating condition starts at time 2 seconds.

Figure 1.

The effect of an OC fault at the (a) GSPC and (b) RSPC.

Effect of open-circuit switching fault in the grid side power converter

The effects of OC faults can be classified into two types: local effects and global effects. Figure 1(a) shows waves under the healthy operating condition and after the occurrence of the fault. The local effect is the distortion of the current line that goes through the faulty converter. The line current path is formed through the power semiconductor components. Therefore, once the OC switch fault occurs, the current of the faulty leg loses its healthy waveform. If the OC fault occurs in the upper arm, the current loses its positive half-cycle amplitude. If the OC fault occurs in the lower arm, the current loses its negative half-cycle amplitude.

According to the supposition of balanced current, the neutral current IN is equal to zero. IN = I_a + I_b + I_c. To achieve the aforementioned assumption, the other healthy phase current creates a DC offset in each phase current as shown in Figure 1.

On the other hand, the global effects lead the significant variables of the DFIG and the other WECS components. In this section, only the DC-Link voltage, the stator current, and the electromagnetic torque, will be discussed.

Typically, the two-level Neural Point Clamped (NPC) converter consists of a capacitor. The capacitor is characterized by its electric potential difference named the DC-Link voltage V_dc, considering that the failure is in the GSPC. According to Figure 1, an OC fault in this converter leads to the electromagnetic torque and DC-Link voltage oscillations. Figure 1(a) shows the effect of an OC switching fault at phase A of the GSPC.

Effect of open-circuit switching fault in the rotor side power converter

The following part is the detail of the damage caused by the OC switching fault in the RSPC. Figure 1(b) shows the effect of an OC switching fault at phase A of the RSPC. If an OC switching fault occurs in the RSPC, the mean of local effect is presented in the distortion of the rotot side current that goes through the faulty converter. In regards to the global effects, the normal operation of the GSPC will be affected. Also, the stability of the whole WECS can be broken. An OC switching fault in the RSPC can cause severe damages such as oscillations in the electromagnetic torque and DFIG rotational speed. Then, it can cause large pulsations of active and reactive power in the stator windings. Also, it reduces the lifetime of electrolytic capacitors, where it causes fluctuations of the DC-Link voltage. In addition, there is a significant oscillation appearing on the stator current. Besides, an OC switching fault in RSPC leads to an over current, this current stress to other power electronic devices when the DFIG is operating around the synchronous speed.

Compared with OC switching fault in the GSPC, the OC switching fault in the RSPC usually leads to more and several damages to the DFIG, electrical network, and wind turbine.

A brief review of diagnosis approaches

Recently, the power converter failure diagnosis methods are widely explored in the literature reviews. Many approaches for the diagnostic of power converter failure in WECSs have been developed. Plenty of methods have been published based on the current signal as a signature; some of them are mentioned in this part. Where Zidani et al. (2008) uses the output inverter current as a signature to diagnose the inverter. They transformed the current shape into the Park’s plan. The authors investigated the fuzzy logic technique in order to diagnose the appearance of the OC switching fault in the power converter. It should be noticed that the online implementation of this approach is not feasible. Also, Sae-Kok et al. (2010) used the rotor current and the stator current in order to detect the OC faults in the BBPC. In this study, current is transformed into Park’s plan. Therefore, the current Park’s shape is used in the detection purpose. The principal disadvantage of this approach is that the detection time is large where this approach needs a period to detect the OC switching fault. Besides, Duan et al. (2010) brought to bear the rotor side current in order to diagnose the rotor side converter. They computed the moving average value of the rotor side current and its absolute value in order to generate the diagnosis variable used in the detection of the OC switching fault. This fault diagnosis method can detect at most two OC switch faults. This approach is too slow due to the additional system used to improve the efficiency of the diagnosis method when the DFIG operate within the synchronous speed. In addition, Freire et al. (2012) proposed an OC switching fault detection method. This method uses current Parks’ vector, current polarity and errors of the current average. These signals act as a signature of faults that help in the diagnosis purpose. The current phase is transformed into Parks’ vector to characterize the failure. This approach can detect single and double faults in power converters. However, it suffers from a complex calculation. Furthermore, Jlassi et al. (2014) introduced an OC fault detection method for both GSPC and RSPC. This method adopts the analysis of the three-phase current. This method investigates the current Luenberger observer model and the Current Form Factors in the diagnosis purpose. This approach can detect single and double faults. Besides, the problem of this method is the cost and the calculation complexity. Moreover, Trabelsi et al. (2017) presented a real-time OC fault diagnosis method for the inverter. They used two-phase current and transform it into the Park’s frame. In this method, the normalized average value is computed to detect and localize the OC switching fault. The main drawback of this method is the inability to avoid the system from the false alarm appears when the DFIG cross the synchronous speed. What is more, Zhao and Cheng (2017, 2018) presented an OC power switch fault diagnosis method in the BBPC. The authors used six phases of current to diagnose the whole BBPC. They transform the current signals into the absolute normalized Park’s vector to achieve the diagnosis purpose. This method can detect single and double OC faults in the converter. Also, it guarantees the detection even when the DFIG operates around the synchronous speed. The most important drawback is the time consuming due to the number of signature signals, the complexity of implementation, and the computational time. Also, Ge et al. (2018) proposed an OC fault diagnosis approach based on the grid current through the computation of the current residual. This method is able to detect the OC faults and identifies the faulty switch. However, this method suffers from the computing complexity. Also, Joseph et al. (2018) proposed an OC power switch fault diagnosis method for inverters. The authors use three-phase current and DC link voltage signals to achieve the diagnosis purpose, where the phase current is transformed into Park’s vector. This method suffers from a high number of signature signals. Hang et al. (2018) presented an OC fault diagnosis for three-phase inverter using three-phase current. This method is based on the computation of the distance between the phase current and the normalized Mannharden distance phase current. This method suffers from high complex computation.

However, a few methods have been published based on the three-phase of voltage signals. Karimi et al. (2008) proposed a real-time failure diagnosis method to detect OC and SC power switch faults. The authors used the pole voltage signal as a signature signal in the diagnosis purpose. They used the rotor side pole voltage to diagnose the RSPC and the grid side pole voltage to diagnose the GSPC. The experimental implementation is achieved using an integrated circuit named Field Programmable Gate Array (FPGA) hardware; this method is characterized by a fast detection in less than 10 μs. Nevertheless, the main drawbacks of this method are the need for an extra voltage sensor and the complexity of calculation. In addition, Caseiro and Mendes (2015) proposed a diagnosis algorithm for rectifiers. This algorithm is based on the computation of the difference between the measured pole voltage signal and its expected value.

A brief theoretical background

This section presents a brief theoretical background and the relevant notation and terminology adopted to achieve the OC fault diagnosis approach for BBPC. First, the Sort Time Fourier Transform (STFT) is described; then, the Discrete Wavelet Transform (DWT) is explicated. And finally, the proposed fault diagnosis method is described.

It should be noticed that the STFT and the DWT are a time–frequency analysis. Usually, these analysis tools are used as features to characterize faults for diagnostic and prognostic (Vachtsevanos et al., 2006). In this article, the STFT is used as a spectral analysis to detect the appearance of an OC switching fault. The spectral analysis is a powerful tool for the diagnostic purpose (Saidi et al., 2016).

In this article, we consider using the DC-Link part where the DC-Link voltage signal operates as a fault signature in the proposed method.

Short time Fourier transform

The STFT is a non-stationary signal analysis, which transforms the signal from one-dimensional into a two-dimensional plane. Therefore, it is able to provide a time–frequency representation. This analysis is used in order to extract useful information for diagnosis. STFT can be expressed as follows (Niu, 2017)

S T F T (n, f) = \int_{n - N / 2}^{n + N / 2} x (N) ω^{*} (N - n) e^{- j 2 π f N} d N

(1)

where x(N) denotes time signal being analyzed, n is the analysis sample, f is frequency, N is time variable, * represents a complex conjugate presentation, N is the number of samples per window, and w(N) is the windowing function.

The STFT implementation remains window length constant during the analysis. Therefore, the analysis results are in a constant time–frequency resolution in the whole time–frequency plane. It can be noticed that a large window length is used for low-frequency estimates, whereas the short window length is used for high-frequency estimation.

In our work to improve the time resolution of the Transform and to provide a good time and frequency resolution, we use a sliding window with overlapping. The major advantage of the STFT algorithm is its simple and fast implementation.

Wavelet transform

The wavelet transform is a signal processing tool; it is selected as a powerful fault diagnosis tool. It is suitable for both stationary and non-stationary signals (Vachtsevanos et al., 2006).The benefit of the wavelet transform technique is its ability to describe signals at multi-localization levels in time and frequency domains. There are two basic types of wavelet transform: the Continuous Wavelet Transform (CWT) and the Discrete Wavelet Transform (DWT). Note that usually the DWT is used in fault diagnosis. Hence, in this part we discuss DWT.

The DWT is a powerful technique, which provides an appropriate description of a time domain signal (Yang and Wang, 2015), where it provides a multi-scale representation of a time domain signal. The DWT is obtained by discretizing the scaling and shifting parameters. The DWT is efficiently realized through a filter bank. The principle of the filter bank is based on the decomposition of the signal into two sub-bands using the pair of low-pass and high-pass filters. These filters are reconstructed from the selected wavelet and its corresponding scaling function. Through these filters, the signal is decomposed into two components: the low-frequency and the high-frequency components. Usually, this decomposition is followed by a downsampling step. It should be noted that the low-frequency component is named the approximation coefficient and the high-frequency component is named the detail coefficient. The DWT is implemented as an iterating form of the pair filters to achieve the multi-scale presentation. To summarize, the DWT is the result of applying recursively the low-pass and the high-pass filters on the approximation coefficient signal (Yang and Wang, 2015). This procedure is expressed in Figure 2

{\begin{matrix} D_{j} (n) = \sum_{k} g (2 n - k) A_{j - 1} (k) \\ A_{j} (n) = \sum_{k} h (2 n - k) A_{j - 1} (k); j \geq 1 \\ A_{0} (n) = x (n) \end{matrix}

(2)

where x(n) denotes time signal being analyzed, n = 1,2,..N, N is the length of time signal being analyzed, h(n) is the low-pass filter, g(n) is the high-pass filter, D_j denotes the detail coefficient in the decomposition level j, and A_j denotes the approximation coefficient in the decomposition level j.

Figure 2.

Discrete wavelet transform filter bank.

The wavelet transform provides a great compromise between localization and frequency resolution, compared with STFT. Also, it is one of the most accessible and simplest tools.

In real industrial application, the wavelet feature extraction process is always used to transform the information acquired from the DWT output signals into a set of features. Usually, these features are used in order to provide the amount of information related to the system operating conditions, that is, healthy conditions and faulty conditions. The great useful features of the DWT are the Wavelet Shannon Entropy (WSE) feature and the Wavelet Energy Shannon Entropy (WECSE) feature (Ismail et al., 2019). These features are used to provide useful information in the proposed BBPC diagnosis method.

In practical applications, the WSE feature is commonly used for signal processing. In this work, we use the WSE in order to extract the characteristic of the OC switching fault and allow carrying out diagnostics. Here, the WSE is introduced (Yang and Wang, 2015).

WSE feature is a powerful tool for characterizing specific phenomena. Usually, it is used to determine a specific property of the OC switching failure (Yang and Wang, 2015). It is given by

W S E_{i} = - \sum_{j = 1}^{N} D_{i j} \ln (D_{i j})

(3)

where i denotes the scales, and j denotes shifts.

The WECSE feature is a robust tool usually used to extract useful information and characteristic from the DWT detail signals of the signal under study (Yang and Wang, 2015). It is defined as

W E S E_{i} = \sum_{j = 1}^{N} {| D_{i j} |}^{2}

(4)

The WSE and the WECSE are statistical features given to provide suitable information for detecting and characterizing specific phenomena in the scale-shift plane.

The proposed fault diagnosis approach

The newly proposed fault diagnosis method is built on the analysis of the DC-Link voltage signal. The analysis of the DC-Link voltage signal is achieved through the time–frequency analysis. Typically, the time–frequency analysis transforms the DC-Link voltage signal into the frequency domain with consideration of the time domain. Then, the frequencies are used as the OC switching fault characterization principle. The main advantage of the proposed approach over other methods explored in the literature is the reduced number of the signature signal; therefore, there is no need for extra sensors, where the methods presented in the literature demand more signals to realize the diagnosis purpose and need more sensors.

The algorithm of the proposed OC fault diagnosis method is summarized by the flowchart presented in Figure 3.

Figure 3.

Flowchart of the proposed fault diagnosis method.

Principally, the proposed fault diagnosis method has two main tasks: the fault detection task and the fault classification task. As revealed in Figure 3, the first task in the fault diagnosis approach is the fault detection task; to achieve this step, the STFT is used in order to detect the appearance of oscillation in the DC-Link voltage signal. In this context the STFT is used in order to detect the fault as fast as possible. It is considered as a time–frequency analysis. In this work, the STFT is used to extract the frequency of the DC-Link voltage signal. After the computation of the STFT, an analysis of the appeared frequency is required in order to detect the fault; in the case of no large frequency appeared, there is no failure detected and therefore the algorithm returns to the first task and requires a new data. Whenever a large frequency appears, the failure is detected and the failure alarm triggers. Therefore, the fault localization task is planned. In this task, the failure can be localized by computing the DWT and then the detail coefficients are used to compute energy and entropy features, where these features are used to generate the fault localization criterion. This criterion is a flag used to localize the analyzed OC switching fault, that is, OC fault in the RSPC or OC fault in the GSPC or OC fault in both converters.

The proposed OC fault diagnosis method requires three basic techniques, which are summarized as follows:

STFT computation,

DWT computation,

Detail coefficients Energy and Entropy computation.

Simulation results

The universal schematic of WECS with detail is depicted in Figure 9. In the WECS, the nacelle is mainly composed of the aerodynamic rotor blade, gearbox transmission, DFIG generator, and BBPC. The BBPC subsystem consists of three parts: the RSPC part, the GSPC part, and the DC-Link component. The wind captured by the rotor blade is converted to electricity at the nacelle. The wind power runs the rotor with low speed while in gearbox transmission the low speed is converted to high speed. The issue of this work is to diagnose the BBPC subsystem through the DC-Link voltage signal. It can be deduced that the diagnosed BBPC subsystem includes diagnosing the GSPC and the RSPC.

In this work, the data were generated in the MATLAB/Simulink software. The simulated WECS structure comprised a 1.5 MW induction generator. The technical specifications of the WECS structure under investigation are given in Table 1.

Table 1.

Specifications of the WECS.

Generator DFIG
Nominal power	Pn	1.5	MW
Roto voltage	Vr	1975	V
Stator voltage	Vs	575	V
Stator resistance	Rs	0.023	pu
Stator inductance	Ls	0.18	pu
Rotor resistance	Rr	0.016	pu
Rotor inductance	Lr	0.16	pu
Magnetizing inductance	Lm	2.9	pu
Pole numbers	p	3	/
Frequency	f	60	Hz
Power converters
DC-Link voltage	Vdc	1150	V
DC-Link capacitor	Cdc	10,000	µF
Switching frequency GSPC	fsg	2700	Hz
Switching frequency RSPC	fsr	1620	Hz

WECS: wind energy conversion system; DFIG: doubly fed induction generator; DC: direct current; GSPC: grid side power converter; RSPC: rotor side power converter.

The power semiconductor OC switching fault in the BBPC leads to the infusion of the current phase in the DC-Link component. Therefore, this phenomenon explains the oscillations of the DC-Link voltage signal under an OC switching fault; the oscillations of the DC-Link voltage follow the current phase of the faulty power converter in the order of frequency.

Figure 4 displays DC-Link voltage signals recorded from the simulation realized at the MATLAB/Simulink software. The three signals were extracted from the same WECS model. In the first case, the RSPC was subjected to an OC switching fault at time 1 second. In the second case, the OC switching fault occurs in the GSPC at time 1 second. Finally, in the third case, there are two OC switching faults occurring at the same time, one in the RSPC and the other in the GSPC. Figure 4 shows in the three cases a very clear presence of oscillation after the occurrence of the OC fault event.

Figure 4.

Simulation of the DC-Link voltage under (a) an OC fault in the RSPC, (b) an OC fault in the GSPC, and (c) simultaneous OC faults in both RSPC and GSPC.

Figure 5 presents the analysis of the nature of the DC-Link voltage signal recorded from the healthy operating conditions and shows the low-order harmonics of the DC-Link voltage signal where it is virtually invisible in the STFT spectrum where it is signed with blue color. However, the recorded data from the faulty operating conditions showed the high-order harmonics of the DC-Link voltage that appears clearly in the STFT spectrum where it is highlighted with yellow color.

Figure 5.

STFT spectrum of the DC-Link voltage under (a) an OC fault in the RSPC, (b) an OC fault in the GSPC, and (c) simultaneous OC faults in both RSPC and GSPC.

The introduction of the oscillation after the occurrence of the OC fault event depends on the faulty power converter (RSPC or GSPC). In general, there is a distinction regarding the frequency and the amplitude of the DC-Link voltage ripple. For the first case when the OC fault appears in the RSPC, the DC-Link voltage oscillation keeps the operating rotor side current frequency f_r. For the second case when the OC failure appears in the GSPC, the DC-Link voltage ripple keeps the operating fundamental system frequency f_s. For the third case a simultaneous OC failure occurs; it can be seen that the ripple of the DC-Link voltage keeps both operating rotor side current frequency and operating fundamental system frequency (Ismail et al., 2018, 2019).

The information collected from STFT spectrum presented in Figure 5 is summarized in the Table 2. It can be deduced from the Table 2 that after the occurrence of the OC switching fault in the RSPC, the frequency of the DC-Link voltage is a multiple of the generator rotational frequency f_r. In the other case when the OC switching fault occurs in the GSPC, the frequency of the DC-Link voltage is a multiple of the power system frequency f_s. In the final case, after the occurrence of a simultaneous OC switching fault, the frequencies appear are a multiple of power system frequency f_s and generator rotational frequency f_r. It should be mentioned that usually, the frequency of the grid side current is the same as the system frequency. In our study, the system frequency is fixed at 60 Hz, as illustrated in Table 1. Noteworthy, the generator rotational frequency cannot be fixed due to the operation under the variable speed of the proposed WECS.

Table 2.

The characteristic harmonics of the DC-Link voltage signal.

Location of faults	DC-Link voltage signal harmonics
RSPC	k.fr
GSPC	k.fs
RSPC and GSPC	k.fr and k.fs

DC: direct current; GSPC: grid side power converter; RSPC: rotor side power converter.

In practical applications, fault detection task and fault location task are two important issues. The first is to find whether there is an OC switching fault in the BBPC and the latter is to determine fault location; this task is the prerequisite for maintenance and correction.

The OC switching faults usually manifest themselves in the frequency domain. Therefore, in this work, we proposed a method to detect the OC failure based on the frequency analysis discussed above. Because this work deals with the detection of a damaged semiconductor, the frequencies of interest are the rotational frequency of the generator f_r and the rotational frequency of the generator f_r. Therefore, we use the STFT to extract the frequencies of interest.

In the fault detection task, it is required to acquire a small section of the DC-Link voltage signal data where the window length of the analyzed data in the fault detection step is equal to 800 samples to ensure an optimal detection. Then, the STFT is used to release a Time–frequency analysis and detect the incipience of large frequencies. To enforce the robustness of the detection task and provide a suitable detection task, the analyzed the window should be sliding with overlapping between every two successive windows. It should be noticed that the fault detection task is an online task. The chief vantage of the proposed implementation of the STFT technique is its fastness and its robustness detection. Where after the occurrence of the OC fault the proposed STFT is able to detect the fault in less than one switching cycle, one switching cycle is fast enough comparing the other studies achieved in this topic.

After detecting the incipience of the OC switching fault, the fault location is an important task for follow-up the fault diagnosis process. The OC switching Faults occurring in power converter are often related to an abrupt event that usually leads to a nonlinear effect on the DC-Link component. Therefore, the DC-Link voltage signal is subjected to the visualization by the time–frequency–domain analysis to extract information that is impossible to see it in a regular time domain analysis. The most simple and accessible time–frequency–domain analysis is the shift-scale analysis, as its experimental implementation is simple, does not need additional setups, is performed directly on the experimental equipment, and finally, it is not time-consuming. Noteworthy, the DC-link voltage signal is classified as is a noisy signal. The presence of noise in the DC-Link voltage signal complicates the localization task. Then, it is easier said than done to localize the OC failure using STFT. Hence, in the localization task, we investigate the DWT. First, we investigate the DWT to characterize the DC-Link voltage signal. Then, we use time-domain features to characterize the output of the DWT technique.

The implementation of the DWT requires first the fixation of the suitable number of scales which highlights the OC switching fault. It should be noted that the maximum number of scales is fixed using the following formula

V D C_{l} = 2^{n s}

(5)

where VDC_l is the analyzed length of DC-Link voltage signal, and ns denote the maximum number of scales.

In practice, to visualize the effect of the OC switching fault on the analyzed DC-Link voltage signal, we need a signal length equal to 1024 samples. Therefore, the maximum number of scales is fixed at 10 scales.

Figure 6 illustrates the detail coefficients of the DWT in 10 scales. The first case is under healthy operating conditions, the second case is under OC switching fault on the GSPC, the third case is under OC switching fault on the RSPC, and the fourth case is under a simultaneous OC switching fault. As revealed in Figure 6, under healthy operating condition all the detail coefficient signals are noisy and there are no signified waves. Then, the detail coefficient signals under OC switching fault on the GSPC are periodic and the detail coefficient signal of the 7th scale tracks the feature of a sinusoidal function. Besides, the detail coefficient signals under OC switching fault on the RSPC are periodic and the detail coefficient signal of the 9th scale keeps slightly the feature of the sinusoidal function. Then, the major detail coefficient signals under a simultaneous OC switching fault are periodic and the detail coefficient signal of the 7th and 9th scale follow the feature of the sinusoidal function.

Figure 6.

Readings of the operational and defective converter to the characteristic scale: (a) healthy operating conditions, (b) an OC fault in the GSPC, (c) an OC fault in the RSPC, (d) an OC fault in both converters.

The main goal of this work is to detect and localize faulty operating conditions due to the existence of OC switching faults based on the DWT. For this purpose, the transformation of the V_dc signal from the time domain into the shift-scale-domain is necessary. Once the computation of the DWT is achieved, the next step is to extract the statistical features mentioned in the previous section, that is, WSE and WECSE, and generate the fault localization criterion.

In this article, we perform a comparative study across the effect of different OC switching faults in the power converters. This comparison is based on WSE feature and WECSE feature. Once these features are computed, we can pass to compute the fault localization criterion. It should be noted that each OC switching fault case has its own and specific fault flag wave.

In this work, the simulation results analysis has been treated on three DC-Link signals under healthy conditions, three DC-Link signals under OC switching fault on the GSPC, three DC-Link signals under OC switching fault on the RSPC, and three DC-Link signals under simultaneous OC switching fault.

Figure 7 presents the mean of the WECSE features and the mean of WSE features for each scale. It shows the reading of the DC-Link voltage signal within healthy conditions and within OC switching fault conditions on the scale of range. Within healthy conditions, the WECSE feature and WSE feature waveforms are balanced. Besides, under an OC switching fault condition, the WECSE and WSE waveforms are transitory and reach peak at a specific scale. Noteworthy, each localization case fault is characterized by its specific WECSE and WSE waveforms. Figure 7 clearly proves this assumption.

Figure 7.

WSE and WECSE features for healthy and the three studied cases of faulty conditions.

As presented in Figure 7, the corresponding WECSE is large, and the energy concentration is high. In other word, an appropriate wavelet can extract the maximum amount of WECSE while minimizing the WSE of the corresponding wavelet detail coefficients. In this work, we propose a Fault Localization Criterion (FLC); this criterion is computed by combining the WSE and WECSE features, it is defined as a WSE and WECSE mean, as denoted in equation (6). This criterion is used to localize the OC switching fault in the BBPC

F L C = \frac{1}{2} (W E S E + | W S E |)

(6)

In this work, the performed study fixes the scales in a range of 1 to 10 scales. Figure 8 illustrates the OC switching fault localization criterion waveforms. We can deduct from Figure 8 that in the first case when the occurrence of an OC switching fault is in the GSPC, the fault localization criterion is increasing and reach a peak at the scale number 7. In the second case, when the OC switching fault appears in the GSPC, the fault localization criterion is increasing and realizes the peak at the scale number 9. Finally, in the case when a simultaneous OC switching faults (i.e. fault in the RSPC and fault in the GSPC), the fault localization criterion is increasing and attain two peaks: the first is at scale number 7 and the second is at the scale number 9. This comparative research proves the necessity of scales of the DWT for diagnosis. Table 3 summarizes the above-mentioned assumption with a focus on the characteristic scales.

Figure 8.

Fault localization criterion under (a) an OC fault in the GSPC, (b) an OC fault in the RSPC, (c) and an OC fault in both converters.

Figure 9.

A typical WECS structure based on DFIG.

Table 3.

Fault converters in the criterion scales.

Location of faults	The scale of the fault flag criterion
RSPC	9
GSPC	7
RSPC and GSPC	9 and 7

DC: direct current; GSPC: grid side power converter; RSPC: rotor side power converter.

The benefit and attractiveness of the proposed OC fault diagnosis method reside on the capability to diagnose the whole BBPC, further using only the V_dc signal as a signature. Also, the proposed fault diagnosis method is able to diagnose the power converters under single, double, and simultaneous fault.

In order to better evaluate the effectiveness of the proposed OC fault diagnosis method, Table 4 gives a comparison results with some of the existing methods in the literature. Using the proposed fault detection method, the OC switching fault can be detected faster than the methods presented in the literature. Also, in the proposed fault diagnosis method, only one signature signal is required to diagnose the BBPC; however, the other methods require six signals to diagnose it. This is while the proposed method is not able to localize the faulty IGBT in the faulty power converter. Therefore, the proposed fault diagnosis method seems to be providing a good compromise between rapidity, performance, and cost.

Table 4.

Comparison of the proposed method with existing methods.

Methods	Signature	Detection time	Ability to detect multi-fault	Ability to localize the faulty IGBT
Proposed method	DC-Link voltage	9 ms	Yes	No
(Zhao and Cheng, 2017)	Three-phase current	40 ms	Yes	Yes
(Duan et al., 2010)	Three-phase current	55 ms	Yes	Yes

DC: direct current; IGBT: insulated gate bipolar transistor.

The results presented in this work demonstrate that the DWT is a powerful tool to diagnose the BBPC and to characterize the OC switching fault precisely. The major advantages of the proposed method are the minimization of the cumulative data using only one signal as a signature and reducing the calculation load (time) with using the DWT technique. The simulation results prove the validity and feasibility of the proposed fault diagnosis approach.

Conclusion

This article presented the results of a feasibility study aimed to assess whether the occurrence of OC switching fault within a BBPC can be diagnosed based on only the DC-Link voltage signal measurement. Our scientific research comprised four different measurement setups; these measurement setups are collected from the same WECS structure. The first is operated with a healthy BBPC. The second is a BBPC with one OC fault in the RSPC. The third is operated with one OC fault in the GSPC. Finally, a BBPC with two OC faults, one in the RSPC and the other in the GSPC. The main difficulty in diagnosis purpose is to minimize the number of sensors; therefore, minimize the cumulative data. To look for signatures of OC switching fault in the BBPC, the analysis of the DC-Link voltage signal was based on the output from a combination of STFT and DWT. The STFT was used to detect the appearance of the OC switching fault in the BBPC and the DWT was used to localize the fault. In the fault localization task, the DWT was followed by the computation of WSE, WECSE, and the fault localization criterion; these tools should be used in conjunction, where they complement each other and provide clear evidence of the localization of the OC switching fault. The results of the proposed fault diagnosis method prove great performance and feasibility. In future work, our proposed method could be further improved by using an artificial intelligence method, which is the automated version of the classification techniques; this makes our method more efficient.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Amirat

Benbouzid

MEH

Al-Ahmar

et al . (2009) A brief status on condition monitoring and fault diagnosis in wind energy conversion systems. Renewable and Sustainable Energy Reviews 13(9): 2629–2636.

Blaabjerg

(2013) Future on power electronics for wind turbine systems. IEEE Journal of Emerging and Selected Topics in Power Electronics 1(3): 139–152.

Boudjellal

Benslimane

(2016) Open-switch fault-tolerant control of power converters in a grid-connected photovoltaic system. International Journal of Power Electronics and Drive Systems 7(4): 1294–1308.

Caseiro

Mendes

(2015) Real-time IGBT open-circuit fault diagnosis in three-level neutral-point-clamped voltage-source rectifiers based on instant voltage error. IEEE Transactions on Industrial Electronics 62(3): 1669–1678.

Choi

Lee

Blaabjerg

et al . (2016) Open-circuit fault diagnosis and fault-tolerant control for a grid-connected NPC inverter. IEEE Transactions on Power Electronics 31(10): 7234–7247.

Duan

Xie

Zhang

et al . (2010) Open-switch fault diagnosis and system reconfiguration of doubly fed wind power converter used in a microgrid. IEEE Transactions on Power Electronics 26(3): 816–821.

Freire

Estima

Cardoso

AJM

(2012) Open-circuit fault diagnosis in PMSG drives for wind turbine applications. IEEE Transactions on Industrial Electronics 60(9): 3957–3967.

Gou

et al . (2018) An open-circuit fault diagnosis approach for single-phase three-level neutral-point-clamped converters. IEEE Transactions on Power Electronics 33(3): 2559–2570.

Hang

Ying

Shu

(2018) Transistor open-circuit fault diagnosis in two-level three-phase inverter based on similarity measurement. Microelectronics Reliability 91: 291–297.

10.

Ismail

Saidi

Sayadi

(2018) A novel back-to-back converter fault detection method using DC-link voltage signature: Application to WES. In: Proceedings of the 1st Euro-Mediterranean conferences on mathematical reliability (ECMR), Djerba, Tunisia, 4–6 July, pp. 117–124.

11.

Ismail

Saidi

Sayadi

(2019) An open circuit switching fault diagnosis approach for back-to-back converter using wavelet analysis. In: Proceedings of the 10th international renewable energy congress (IREC), Sousse, Tunisia, 26–28 March.

12.

Jlassi

Estima

El Khil

et al . (2014) Multiple open-circuit faults diagnosis in back-to-back converters of PMSG drives for wind turbine systems. IEEE Transactions on Power Electronics 30(5): 2689–2702.

13.

Joseph

Chelliah

Lee

(2018) Multi-channel VSI fed large variable speed asynchronous hydro-condenser: Fault analysis, fault diagnosis and fault tolerant control. IET Renewable Power Generation 13(3): 438–450.

14.

Kaidis

Uzunoglu

Amoiralis

(2015) Wind turbine reliability estimation for different assemblies and failure severity categories. IET Renewable Power Generation 9(8): 892–899.

15.

Karimi

Gaillard

Poure

et al . (2008) FPGA-based real-time power converter failure diagnosis for wind energy conversion systems. IEEE Transactions on Industrial Electronics 55(12): 4299–4308.

16.

Niu

(2017) Data-driven Technology for Engineering Systems Health Management: Design Approach, Feature Construction, Fault Diagnosis, Prognosis, Fusion and Decisions. Beijing, China: Springer.

17.

Sae-Kok

Grant

Williams

(2010) System reconfiguration under open-switch faults in a doubly fed induction machine. IET Renewable Power Generation 4(5): 458–470.

18.

Saidi

Ali

Benbouzid

et al . (2016) The use of SESK as a trend parameter for localized bearing fault diagnosis in induction machines. ISA Transactions 63: 436–447.

19.

Statistics

(2017) Global wind energy council. Washington, DC, USA.

20.

Trabelsi

Boussak

Benbouzid

(2017) Multiple criteria for high-performance real-time diagnostic of single and multiple open-switch faults in ac-motor drives: Application to IGBT-based voltage source inverter. Electric Power Systems Research 144: 136–149.

21.

Vachtsevanos

Lewis

Hess

et al . (2006) Intelligent Fault Diagnosis and Prognosis for Engineering Systems, Vol. 456. Hoboken, NJ: Wiley.

22.

Yang

Wang

(2015) Multi-level wavelet Shannon entropy-based method for single-sensor fault location. Entropy 17(10): 7101–7117.

23.

Yang

Bryant

Mawby

et al . (2011) An industry-based survey of reliability in power electronic converters. IEEE Transactions on Industry Applications 47(3): 1441–1451.

24.

Yang

Chai

(2016) A survey of fault diagnosis for onshore grid-connected converter in wind energy conversion systems. Renewable and Sustainable Energy Reviews 66: 345–359.

25.

Yenduri

Sensarma

(2016) Maximum power point tracking of variable speed wind turbines with flexible shaft. IEEE Transactions on Sustainable Energy 7(3): 956–965.

26.

Zhao

Cheng

(2017) Open-circuit faults diagnosis in back-to-back converters of DF wind turbine. IET Renewable Power Generation 11(4): 417–424.

27.

Zhao

Cheng

(2018) Open-switch fault-diagnostic method for back-to-back converters of a doubly fed wind power generation system. IEEE Transactions on Power Electronics 33(4): 3452–3461.

28.

Zidani

Diallo

Benbouzid

MEH

et al . (2008) A fuzzy-based approach for the diagnosis of fault modes in a voltage-fed PWM inverter induction motor drive. IEEE Transactions on Industrial Electronics 55(2): 586–593.

Wind turbine power converter fault diagnosis using DC-link voltage time–frequency analysis

Abstract

Keywords

Introduction

Operation of the grid-connected WECS under open-circuit fault condition

Effect of open-circuit switching fault in the grid side power converter

Effect of open-circuit switching fault in the rotor side power converter

A brief review of diagnosis approaches

A brief theoretical background

Short time Fourier transform

Wavelet transform

The proposed fault diagnosis approach

Simulation results

Conclusion

Footnotes

Declaration of conflicting interests

Funding

References