Data driven prognostics for predicting remaining useful life of IGBT

Power electronic devices such IGBT (Integrated Gate Bipolar Transistor) are used in wide range of applications such as automotive, aerospace and telecommunications. The ability to predict degradation of power electronic components can minimise the risk of their failure while in operation. Research in this area aims to develop prognostics strategies for predicting degradation behaviour, failure modes and mechanisms, and remaining useful life of these electronic components. In this paper, data driven prognostics approaches based on Neural Network (NN) and Adaptive Neuro Fuzzy Inference System (ANFIS) models are developed and used to predict the degradation of an IGBT device. IGBT life data under thermal overstress load condition with square signal gate voltage bias, available from NASA prognostics data repository, is used to demonstrate the proposed data-driven prognostics strategy. The monitored collector-emitter voltage is used to identify the pattern and duration of different phases in the applied voltage load. The NN and ANFIS models are trained with a subset of the test data to predict remaining useful life (RUL) of the IGBT device under varying load test profiles. The predictive capability and performance of the models is observed and analysed