IGBT modules fault prediction based on particle filter with an improved nonlinear characteristics representation of state-space model

Insulated Gate Bipolar Transistor (IGBT) modules, being widely applied in many fields, are prone to aging and even fail under high voltage or temperature operation, so it is necessary to conduct IGBT modules fault pre-diction to avoid critical failures. Particle filter (PF) has strong applicability in IGBT modules fault prediction due to the fact that hidden state of IGBT modules can be evaluated from observed measurements containing noises. Nevertheless, the analytic form of state transition equation for PF is insufficient to represent nonlinear charac-teristics of complex systems, and the degradation process of IGBT modules in practical engineering does not conform to Markov property. Hence, a novel fault prediction method for IGBT modules based on improved particle filter is proposed in this paper, which has an improved nonlinear characteristics representation of state -space model. Specifically, long short-term memory (LSTM) model and curve fitting function are utilized to construct the state transition equation and the measurement equation, respectively. The IGBT accelerated aging test data published by NASA PCoE research center are used to verify the proposed fault prediction method, and the comparison studies show effectiveness and superiority of the proposed method in the IGBT modules fault prediction.