Development of Explainable AI (XAI) Framework for Fault Diagnosis in Power Electronics Systems

Abstract

This study has sought to develop an explainable artificial intelligence (XAI) system used to diagnose power electronics system faults. The research deals with the necessity to introduce the transparent and explainable fault detection measures in complex power electronic devices such as converters, inverters, and power motors. It utilized a rigorous procedure that covered an extensive gathering of data that comes about because of different operating environments, simulation of a severity of faults and the execution of modern machine learning strategies. The framework provides explainability methods like LIME, SHAP and attention mechanisms to yield the transparent decision-making processes. Various learning strategies comprising deep neural networks, support vector machines, and multi-algorithm strategies were tested to determine their accuracy and computing speed in the diagnosis process. The final support structure was cross checked utilizing approaches of cross-materials and also tested in experimental testbeds (simulated and realistic). The accuracy, precision, recall, F1-score, explainability measures were used to evaluate the measure of performance, which showed better results than existing methods of fault diagnosis. The obtained outcomes suggest that the offered XAI framework is highly accurate at the diagnostic level, as well as gives interpretable knowledge about the mechanisms of faults; thus, can be successfully implemented in the industrial sector of power electronics, where the aspects of reliability and transparency are vital.