Fractional Long-Range Dependence Multi-Mode for Remaining Useful Life Prognosis of IGBT Under DC Voltage Control

Since the degradation process of IGBTs is a slowly varying nonstationary stochastic process with random switching among three modes (active region, saturation region, and cut-off region), remaining useful life (RUL) prediction of IGBTs remains an urgent research issue. This paper investigates IGBT RUL prediction under DC voltage-controlled operating conditions. First, a health feature parameter (HFP) is constructed via the weighted linear combination of gate voltage, collector voltage, and temperature for fault feature extraction of degradation processes. Second, the Markov state transition matrix describing the three operating modes of IGBTs is established. Furthermore, a long-range dependent fractional Weibull process (fWp) prognostic model with three-mode characteristics is developed. Finally, validation is carried out on the public NASA IGBT thermal overload aging dataset. Error analysis demonstrates that the proposed method achieves superior prediction accuracy compared with four prognostic approaches: APE, OSGP, TDNN-GP, and SA-NN.