Abstract
Recent research has focused on sustainable development and renewable energy resources, thus motivating nonconventional cutting-edge technology development. Multilevel inverters are cost-efficient devices with IGBT switches that can be used in ac power applications with reduced harmonics. They are widely used in the power electronics industry. However, under extreme stress, the IGBT switches can experience a fault, which can lead to undesirable operation. There is a need for a reliable system for detecting switch faults. This paper proposes a signal processing method to detect open-circuit problems in IGBT switches. Relative wavelet energy has been used as a feature for a machine learning algorithm to diagnose and classify the faulted switches. The switching sequence can be altered to restore a healthy output voltage. Inverter faults have been diagnosed by using support vector machine (SVM) and decision tree (DT), and an ensemble model based on decision tree (DT) and XG boost algorithm was developed, which yielded 92%, 88%, and 94.12% accuracy, respectively.