This paper focuses on the development of a diagnostic tool for detecting insulated gate bipolar transistor power electronic switch flaws caused by both open and short circuit faults in multi-level inverter time-frequency output voltage specifications. High-resolution laboratory virtual instrument engineering workbench software testing tool with a sample rate data collection system, as well as specialized signal processing and soft computing technologies, are used in this proposed method. On a single-phase cascaded H-bridge multilevel inverter, simulation and experimental investigations of both open and short issues of the insulated gate bipolar transistor components are performed out. In all conceivable switch issues, the output voltage signals are evaluated for different modulation index values. Fast fourier transform and discrete wavelet transform methods are used to investigate the frequency domain properties of output voltage signals. In the artificial neural network, the back propagation training technique was employed, and the generated neural parameter values were used in the Laboratory Virtual Instrumentation Engineering Workbench real-time fault diagnosis model.
Multilevel inverters (MLIs) are the most successful design in the field of high-power implementations, and they are utilized in a wide range of industrial applications that require lower operating costs and higher quality output, such as drive systems, electric cars, voltage profile compensators, voltage regulators, sustainable energy applications, and so on. The multilevel inverter is among the most common and reliable categories of MLI, wherein the stair cased pulse-width synthesized inverter voltage output is generated by a number of cells connected in series H-bridge voltage cells. The operation of insulated gate bipolar transistor (IGBT) power modules in cascaded h-bridge multilevel inverters (CHMIs) leads in substantial switching losses and severe overheating when switching frequency modulation is significant with high voltage and current. Thermal gradients and constant operation of switches can lead to serious issues including parameter drift and ageing, which can lead to semi-conductor system failure [
Short circuit (SC) and open circuit (OC) faults are the two types of failures which can occur in power switching devices in MLIs. Short circuit problems have the ability to damage the system right away. As a result, hardware approaches for short circuit protection, including as high potency fuses, de saturation methods, and di/dt feedback techniques, are frequently employed. Open-circuit faults in power electronic switches do not always result in the device shutting down, and they can go undetected for a long time. This could cause secondary defects in the inverter or other drive components, culminating in the entire system being shut down and expensive repairs. Short-circuit faults in power electronic switches, on the other hand, are extremely damaging and necessitate special precautions to automatically shut down the entire drive. These types of failures must be identified and repaired in a microsecond in order to safeguard analogous semiconductor devices from damage in the converter leg. On the other perspective, extended open circuit fault behavior of the power converters might cause the entire system to collapse. Expertise in fault behaviors, fault prediction, and fault diagnostics will be necessary to keep the multilevel inverter system functioning smoothly. The following are the two aspects of power electronic device fault diagnosis: one is fault information acquisition, which entails gathering data whenever a failure occurs using a specific fault detection approach; the other is fault identification and characterization, which rely upon that specific of failure modes to recognize the category and position of faults using only a specific fault detection method [
In this circumstance, faults should be recognized as soon as possible after they occur. If an inverter device is used continuously under abnormal settings, further issues will arise, resulting in severe consequences. Furthermore, since the multilevel inverter is made up of a lot of switching devices, the system is complicated and has a lot of nonlinear influences [
Numerous model-based techniques for transistor OC defect identification have been suggested for the system. To detect the voltages aberrations for diagnostics, an adaptable system is established [
The fundamental objectives of the proposed research effort are to develop a high-performance fault detection methodology for evaluating open and short circuit faults in MLI using enhanced signal processing and soft computing techniques. The fast fourier transform (FFT) and discrete wavelet transform (DWT) techniques in Laboratory Virtual Instrumentation Engineering Workbench (LabVIEW) software tool are used to evaluate the spectrum properties of output voltage wave forms produced using both modeling and experimental investigations at various fault situations. By using FFT technique and LabVIEW software, extract salient features such as total harmonic distortion (THD) and harmonic contents of output voltage signal at different fault cases. Extract important characteristics from the DWT multi resolution analysis using LabVIEW capabilities, such as energy content at various layers of decomposition of the output voltage signal for various fault occurrences. Designing an automated fault detection system based on artificial neural networks (ANNs) for detecting individual faulty switches, develop a LabVIEW based real-time defect detection system to leverage an off-line trained artificial neural network model. The performance characteristics of the FFT-ANN-LabVIEW model and the DWT-ANN-LabVIEW model-based fault detection approach for multilevel inverters can be compared to develop an effective fault diagnostic system.
The single stage five level cascaded H-bridged multilevel converter configuration included in this analysis, which includes two H-Bridges and eight IGBT switches, is depicted in
The voltage and current characteristics of a single cascaded H-Bridged multilevel inverter is shown in
The FFT method was used to extract the properties of output voltage patterns. In order to create an effective fault analysis tool, frequency domain assessments of the output voltage waveforms are necessary. The output voltage signal was subjected to the FFT technique to extract different properties. As a result, the signal processing approach is critical. Even though a competent characteristic extractor should offer enough crucial information about the neural network in the position taken, it was the maximum level of accuracy obtained inside the artificial neural convention.
The FFT is widely utilized in signal processing applications, including power systems, power electronics, communications, broadcasting, and entertainment. The frequency response description of any periodic or non-periodic signal is provided by the FFT. The Fourier Transform of a signal or function g(t) is described as following
THD is evaluated for a voltage signal using the below
The new fault diagnosis algorithms work by identifying important characteristics in output voltage data, which are then aggressively removed from the basic data to create diagnostic data. Simultaneous temporal and frequency analysis of voltage signals is required to build an effective fault diagnosis system.
The decomposition scale is p, the number of sampling points is denoted as q, and the translation coefficient is n. The purpose of multi resolution analysis is to accomplish two essential qualities, the first of which is the temporal mapping ability, which is useful for identifying changes in features. At the time of the disruption, this will appear as high coefficients. The signal energy is partitioned into different frequency bands in the second step. This gives an indication of the distorted signal’s frequency quality. The Daubechies 4 wavelet is well-known for its capacity to identify signal power transitions, but it has previously been employed as an origin wavelet. As a result, the Daubechies 4 wavelet was employed to analyze the data. The frequency spectrum of comprehensive elements for output voltage signals was divided into 9 levels, as indicated in
Comprehensive features of DWT (D1 to D9) | Operating range of frequency (kHz) |
---|---|
1 | 5–10 |
2 | 2.5–5 |
3 | 1.25–2.5 |
4 | 0.625–1.25 |
5 | 0.3125–0.625 |
6 | 0.15625–0.3125 |
7 | 0.078125–0.15625 |
8 | 0.0390625–0.078125 |
9 | 0.01953125–0.0390625 |
During normal conditions,
The DWT MRA different band levels of the voltage amplitude are shown in
For practical real-time applications, laboratory tests must be used to validate the results of simulation studies. A photo of a laboratory experimental setup for obtaining multilevel inverter output voltage signals during various switch operation condition is shown in
Open and short circuit faults are generated on each switch, and the output voltage waveform is analyzed. A voltage sensor mounted to the USB senses the output voltage and captures the data of output signals. The corresponding FFT and DWT characteristics generated from output voltage waveforms confirmed with experimental investigations under various fault situations and at numerous opposing modulation index values of 0.85 and 0.9 are displayed in
The LabVIEW front panel for spectrum analysis of output voltage impulses using the FFT approach is shown in
The cascaded H-bridged multilevel inverter faulty switch analysis module shown in
The failure detection of multilevel inverters was automated using an ANN in these investigations. The ANN was used to solve the difficulty of detecting the faulty switch in a cascaded multilevel inverter. The multi-layer feed forward network with back propagation learning technique has been considered as one of the many ANN designs available in the literature due to its easy approach and high predictive capability.
The neural network requires longer learning and following the convergence criteria as the number of hidden layer neurons increases beyond 18. In order to reach an ideal value for the number of epochs, the network’s mean square error was computed by retaining the step size at 0.1 with 18 hidden layer neurons. The mean square error values derived from different amounts of hidden layer neurons are shown in
The fault diagnostic framework is then implemented in the LabVIEW graphical user interface module using the extracted features as an input to ANN. The FFT-ANN-LabVIEW approach and the DWT-ANN-LabVIEW approach are two separate fault diagnostic methodologies proposed in this work.
The THD, Vrms, and Harmonic/Fundamental ratios up to 11th order harmonics are displayed using the FFT approach, which provides critical information regarding the defective switch in the multilevel inverter. For each failure condition, the energy content of the output voltage shows various patterns at different degrees of decomposition of the DWT MRA technique. Front panels based on LabVIEW are being developed for real-time applications, and they are extremely useful and user-friendly for the processing of industrial applications. The fault diagnostic efficiency of the DWT-ANN-LabVIEW technique is significantly better than the FFT-ANN-LabVIEW approach for all fault situations, from standard to multiple open and short-circuit switch fault states. Unlike the FFT-ANN method, the DWT-ANN method solely uses DWT characteristics to train and assess.
The DWT-ANN technique outperforms the FFT-ANN approach without any external inputs such as rms voltage values. The DWT-ANN-LabVIEW technique provides a nearly 100% identification rate. Unlike the FFT-ANN technique, the DWT-ANN approach solely trains and evaluates using DWT features. This method does not require any intermediate calculations, such as rms voltage values. As opposed to the FFT-ANN approach, the DWT-ANN approach performs better without any external inputs such as rms voltage values. For high-power applications, the number of levels in multilevel inverters is expanding by the day; the recommended device may be evaluated for 7 and 9 level inverter systems. For 7 level and 9 level inverters, the transients in current signal recorded during open or short circuit faults may be examined, and a diagnostic process based on this information can be devised.
The authors would like to thank Sona College of Technology, Salem, TN, India and also, we like to thank Anonymous reviewers for their so-called insights.