A Distance-based Health Indicator for RUL Prediction of Power Electronics

Power electronic (PE) converters are considered as essential components of electric vehicles and their reliability is receiving increasing attention by vehicle manufacturers. In this vein, remaining useful life (RUL) prediction of PE components is crucial for extending their lifetime and for avoiding unexpected downtime of the power-converter based electrical systems. In this paper, we developed a novel prognostic methodology, utilizing a distance-based health indicator and tracking it using an interacting multiple model (IMM), to estimate the RUL of a degraded system. The distance between the test vector and the cluster center of a self-organizing map (SOM) obtained from a trained healthy model is identified as the failure precursor. The time evolution of the precursor is estimated by an interacting multiple model (IMM) consisting of multiple particle filters or Kalman filters with a linear trend. Then the RUL estimate and the concomitant uncertainty of the estimate are evaluated by extrapolating the IMM state estimates until they cross a pre-defined failure threshold. The algorithms are tested on data from accelerated aging tests under various operating conditions. The results demonstrate that the proposed algorithms provide good RUL predictions for the degraded devices and the methodology is applicable to a variety of systems.