Predictive Maintenance of SiC Power Converters Using AI for Ultra-Fast Electric Vehicle Charging Systems

Abstract

Widespread acceptance of EVs has led to greater development of the ultra-fast EV charging infrastructure, thereby raising demands on the power converter’s reliability and efficiency. The SiC-based power converter has emerged as an ideal solution to the need of efficient and fast EV charging due to its superior switching characteristics and low power loss. However, continued operation under such high stress conditions can contribute to the acceleration of degradation processes that influence the performance and lifetime of the converter. The current periodic approach to maintenance of power converters fails to detect the early stages of degradation; additionally, the available techniques that rely on artificial intelligence (AI) cannot predict converter degradation in real time. This work presents an AI-based predictive maintenance approach for SiC-based converters used for ultra-fast EV charging. The presented methodology utilizes multi-sensor health data collected via voltage, current waveforms, junction temperature, switching frequency, and harmonic distortion measurements. The developed hybrid CNN-BiLSTM-attention model is designed to automatically detect degradation characteristics, learn aging patterns, perform fault diagnosis, and calculate remaining useful life (RUL). Evaluation experiments conducted based on varying conditions of the converter reveal that the presented approach outperforms the traditional methods for machine learning and deep learning. It provides fault classification accuracy of 97.3%, lowers detection latency down to 45 milliseconds, and delivers higher degradation prediction capacity via attention-based analysis of health features. These outcomes prove that the AI approach presented here allows for accurate fault detection, early degradation prediction, and maintenance planning for future ultra-fast EV charging systems.