An analysis of voltage source inverter switches fault classification using short time Fourier transform

The dependability of power electronics systems, such as three-phase inverters, is critical in a variety of applications. Different types of failures that occur in an inverter circuit might affect system operation and raise the entire cost of the manufacturing process. As a result, detecting and identifying inverter problems for such devices is critical in industry. This study presents the short-time Fourier transform (STFT) for fault classification and identification in three-phase type, voltage source inverter (VSI) switches. Time-frequency representation (TFR) represents the signal analysis of STFT, which includes total harmonic distortion, instantaneous RMS current, RMS fundamental current, total non harmonic distortion, total waveform distortion and average current. The features of the faults are used with a rule-based classifier based on the signal parameters to categorise and detect the switch faults. The suggested method's performance is evaluated using 60 signals containing short and open circuit faults with varying characteristics for each switch in VSI. The classification results demonstrate the proposed technique is good to be implemented for VSI switches faults classification, with an accuracy classification rate of 98.3%.


INTRODUCTION
Voltage source inverter (VSI) is commonly used to convert a DC voltage into an AC voltage with varying magnitude and frequency [1]- [6]. As a single failure in the converter components will result in a fault in the overall system, the system reliability of these converters is critical [7]- [15]. As a result, in several crucial procedures, the VSI must be able to operate continuously even under malfunctioning situations [16]- [19]. Semiconductor switches and aluminium capacitors are the two most critical components of a voltage source inverter. Soldering joint and semiconductor failures account for more than 34% of all malfunctions and failures [20]- [24]. Figure 1 shows the flow chart of the proposed method. The fault signals are modelled using MATLAB software and represented in both frequency, and time domain namely the STFT. From the TFR, the significant signal parameters such as instantaneous of total waveform distortion, RMS current, total nonharmonic distortion, average current, total harmonic distortion and RMS fundamental current are estimated. The switches fault are then classified and identified using a rule-based classifier based on the signal characteristics.

Modeling of VSI switches fault
Despite the fact that overcurrent or short circuit protection against power switches has become increasingly popular on industrial drives, open circuit failures are still disregarded in the industrial context [89]. Overheating from thermic cycling, for example, can cause connections to break, which can be caused by bonding wire lifting, device driver problems, or connection breakage itself [90]. Furthermore, the OC has the ability to develop secondary faults and cause several device problems in the system if used in such an unusual manner over an extended period of time [91]. The topology of open and short circuit faults of VSI are shown in Figure 2(a) and 2(b), recpectively. Figure 3(a) and 3(b) depict a VSI switches fault model using MATLAB. In addition, the designed circuit has a DC voltage input of 50V, a sampling time of 100s, and a fundamental frequency of 60 Hz.

Short time Fourier transform
STFT is a significant tool for indicating a signal in both the frequency and time [84], [92]- [94]. The signal spectral characteristic in time domain can be recognised using the TFR. As a result, STFTs are an appropriate method for analysing switch fault signals with non-stationary and multi-frequency components. where ω is the observation window.

Signal characteristic
The STFT is used to calculate signal characteristics in order to offer timely data about the signal. It is distinguished by the following parameters: total harmonic distortion mean (THDmean), fault duration (Td,fault), total non-harmonic distortion mean (TnHDmean), average RMS current mean (Irms,mean), average current mean (Iave,mean), and total waveform distortion mean (TWDmean). Using signal parameter information, it is then feasible to identify the faulty switch location. Furthermore, the sample frequency for the recommended approach is 10 kHz.

Ruled-based classifier
A deterministic classification, which is rule-based classifier, is widely utilised in a variety of applications due to its simplicity and ease of implementation [95]. The efficiency of the classification is strongly depending on competent threshold, and rules preferences. To classify and identify failures in the switches, a rule-based classifier based on signal characteristics is used. This study will use 60 different signals with varied characteristics of different switch fault signal in order to establish the optimum potential threshold setup parameters for the suggested technique. In addition, the total waveform distortion threshold (TWDthres,fault) is used to calculate the threshold values of the proposed method.

RESULTS AND DISCUSSION
This section discusses the findings of the switch faults analysis using STFT. The open switch fault take place at S31 of phase c upper, and the short switch fault appears at S12 of phase. Figure 4 Following that, a short-circuit fault of the VSI switch occurs at S12 and. As demonstrated in 60 Hz frequency appears along the time axis in Figure 5(a). The DC component (0 Hz) does exist, however, during the fault period, which ranges from 195ms to 255 ms. Figure 5(b) shows the signal magnitude abruptly drops from 1.6 A to 0.05 A for 60 ms between 195 ms and 255 ms. Figure 6 depicts a graphical representation of the average current, fundamental current, and RMS current of open circuit switch faults. Starting at 195 ms and lasting 60 ms, the RMS current is reduced from 1.17 A to 0.9 A. The basic current exhibits a similar behaviour, with the signal's current suddenly dropping for 60 ms. The current decline from the nominal value ranges from 1.17 to 0.7 A. Similarly, once the fault signals are recognised, the average current signal is reduced to -0.7 A. As illustrated in Figure 7(a), the RMS current for a short circuit increases abruptly to 1.35 A from the nominal value which is 1.17 A, whereas for a period of 60 ms the RMS fundamental current declines to 0.75 A. Comparable to average current, the current drops from 0 A to -1.1 A and exists at the negative cycle. present the total waveform distortion and total nonharmonic distortion magnitude for a period 60 ms that increases from 2% to 55% and 47%, respectively. As shown in Figure 8(c), the magnitude of total harmonic distortion increases by 30% for open circuit switches fault. According to the investigation of open circuit switch faults, total nonharmonic distortion has a greater percentage value than total harmonic distortion. Total waveform distortion is the combination of total nonharmonic distortion, and total harmonic distortion.  Figure 9 (a) depicts the amount of total waveform distortion or short circuit switches fault, which is identical to total nonharmonic distortion since total harmonic distortion and total nonharmonic distortion are added together. Meanwhile, as seen in Figure 9, the amount of total harmonic distortion remains low (3%). As a result of this, Figure 9 (c) depicts the total nonharmonic distortion magnitude, which increases from 4% to 55% at 180 ms for a period of 60 ms.
The fault signal was derived through the analysis of 60 signals with varied properties for each type of switch (open and short circuit for S11, S12, S21, S22, S31, and S32). The best threshold value for a rule-based classifier is found to be 0.05 or 5%. The pseudo code as shown in Figure 10 describes a rule-based classifier for classifying and identifying switch faults based on signal characteristics. The fault signals of switches are evaluated and classified using STFT and a rule-based method. In addition, 60 signals with various fault signal characteristics are generated to determine the system's performance. According to Table 1, the proposed technique provides 98.3% accuracy of classification.

CONCLUSION
The STFT approach, in conjunction with a rule-based classifier, was used to create a classification and identification system for switch fault signals. The system's performance is then validated by categorising 60 actual signals with varying characteristics for each sort of switch fault signal. The suggested approach performs admirably, with 98.3 percent of faults correctly classified. As a result, it shows that the system is well-suited for usage as a switch fault monitoring system. Other time-frequency domain methods, such as the Gabor transform and S-transform, should be investigated in future study to improve classification accuracy.