Remaining Useful Life Prediction of SiC MOSFETs Using the Autoformer-RELM Model

The reliability of silicon carbide (SiC) MOSFETs plays a critical role in ensuring the stability and safety of power electronic converters, highlighting the need for a precise and data-driven approach to predict their remaining useful life (RUL). This study focuses on developing an accurate RUL prediction model by identifying the on-state drain-source voltage (V ${}_{\mathrm {ds\_on}}$ ) as the critical failure parameter in power cycling experiments for SiC MOSFETs. Based on experimental data, a hybrid prediction model, Autoformer-regularized extreme learning machine (RELM), is proposed and validated for RUL prediction. The experimental results demonstrate that the Autoformer-RELM model reduces the mean absolute percentage error (MAPE) by 26.84% compared to Autoformer while achieving 55.56% and 57.45% improvements over Informer and long short-term memory (LSTM), respectively. The Autoformer-RELM model outperforms traditional approaches by effectively handling nonlinearity and long-term dependencies, resulting in significantly improved prediction accuracy. In addition to advancing RUL prediction methodologies, this research provides meaningful insights into the reliability analysis and practical applications of SiC MOSFETs and other power devices.