Open switch fault diagnosis of cascade H-bridge multi-level inverter in distributed power generators by machine learning algorithms

Abstract

In recent years, multi-level inverters have had remarkable applications in renewable energy sources, high voltage, and other high-power applications. The multi-level inverter has advantages like minimum harmonic distortion and can operate on several voltage levels. A multi-level inverter is being utilized for multipurpose applications such as transportation, communication, industrial manufacturing, aerospace active power filter, Static Var Compensator, and machine drive. Power electronics equipment reliability is very important, and to ensure a multi-level inverter system’s stable operation; it is important to detect and locate faults as quickly as possible. It is difficult to diagnose a fault in a multi-level inverter using a mathematical model because it consists of many switching devices, in this context and to improve fault diagnosis accuracy and efficiency of a cascaded multi-level inverter (CHMLI), a fault diagnosis strategy based on the probability principal component analysis (PPCA) might be utilized. Different machine learning algorithms are used to classify and diagnose the faults under different conditions in a cascaded H-bridge multi-level inverter (CHMLI). This paper presents the comparison of two different machine learning algorithms, such as support vector machine (SVM) and k-Nearest neighbors algorithm (k-NN), based on probabilistic principal component analysis (PPCA) for the effective open switch fault diagnosis in CHMLI employed in distributed generator units. PPCA is a useful technique used for optimizing and data processing without changing the input data’s original properties and characteristics. Using the phase shift pulse width modulation technique, the output voltage signals under different switching fault conditions if the CHMLI are taken as fault features. Both algorithms are used to identify and locate the fault under different modes in CHMLI of distributed generator units. The proposed fault diagnosis methods are compared using simulations and experimental results employing a field-programmable gate array (FPGA) controller. The developed system’s simulation and experimental results perform satisfactorily to detect the fault type, fault location, and reconfiguration. The fault diagnosis time using PPCA-SVM as a fault diagnosis tool for the simulation and experimental case is 0.065 ms and 2.12 ms, respectively. On the other hand, 347 ms and 415 ms fault diagnosis time for simulation and experimental case, respectively, are recorded for the PPCA-kNN based fault diagnosis technique. Therefore, the SVM-based fault diagnosis method is much more efficient and accurate than the k-NN based fault diagnosis method. Moreover, the proposed SVM-based fault diagnosis guaranteed high reliability for CHMLI.