Comparative analysis of regression with ARIMA and PI-RNN models for predicting IGBT remaining useful life using switch health index embedding feature reconstruction approach

Abstract Switched reluctance motors (SRMs) are increasingly used in electric vehicles (EVs) due to their robust construction, cost-effectiveness, and high efficiency. Power converters typically employ insulated gate bipolar transistors (IGBTs) as key switching devices to control SRMs. However, these power semiconductor components are prone to degradation over time, making it essential to develop accurate prognostic models for predicting their remaining useful life. This study proposes a novel methodology called switch health index embedding (SHIE) for IGBT devices. The methodology combines unnecessary features’ elimination, feature reduction and reconstruction using a correlation dimension estimator, principal component analysis (PCA) and a sparse denoising autoencoder (SDAE). This is followed by predictive modeling using physics-informed recurrent neural networks, and then regression with an Autoregressive Integrated Moving Average (ARIMA) model. To assess the effectiveness of our approach, we utilized the accelerated aging database for IGBTs from NASA Ames Laboratory’s Prognostics Center of Excellence. The analysis identifies collector-emitter voltage ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mrow> <mml:msub> <mml:mi>V</mml:mi> <mml:mrow> <mml:mrow> <mml:mtext>CE</mml:mtext> </mml:mrow> </mml:mrow> </mml:msub> </mml:mrow> </mml:mrow> </mml:math> ) as the most suitable precursor signal for IGBT degradation monitoring. Comparative results indicate that the regression with the ARIMA (43, 1, 1) model outperforms other methods, achieving superior accuracy (99.395%) and the lowest error rates, i.e. symmetric mean absolute percentage error (SMAPE: 0.794%). The findings highlight the potential of regression with the ARIMA model to monitor the health in the real-time of power semiconductor devices, aiding in predictive maintenance and operational efficiency.