Prognostic Analysis of IGBT Health: Real-Time On-State Voltage Prediction through Machine Learning

Abstract:
This study demonstrates a machine learning method to estimate the on-state voltage of semiconductors in a converter prototype exposed to dynamically changing mission profiles of electric vehicles (EVs) and photovoltaic (PV) systems under different ambient temperatures. The real-time monitoring of on-state voltage is an indicative measure of the junction temperature for the predictive maintenance of power converters. The method offers novel insights into uncertainties and feature importance for the on-state voltage predictions using available parameters of a converter. The approach is industrial-ready and applicable to various converter systems regardless of their specifications, achieving a remarkable absolute prediction error of 0.75%.