One of the main causes of failure in Insulated Gate Bipolar Transistor (IGBT) modules used in high-power conversion applications is thermal-stress-induced degradation. In this paper, an experimental testing setup for thermal stress and real-time degradation monitoring, as well as a deep neural network (DNN)-based lifetime prediction of IGBT modules under thermo-electrically stressed inverter operation, is proposed. A two-level SVPWM inverter is implemented to create a hybrid power cycling test platform that imposes well-defined junction-temperature swings representative of real-world operation by combining controlled electrical loading and active induction heating with water cooling. Throughout the aging process, on-state voltage and module temperature are constantly monitored to identify degradation precursors associated with thermo-mechanical fatigue. A physics-based Coffin–Manson lifetime model is fitted using failure datasets to characterize temperature-dependent lifetime behavior. An offline deep neural network (DNN) is trained on degradation trajectories derived from on-state collector–emitter voltage (Vce,on) to predict remaining useful lifetime. This approach uses partial degradation histories for accurate early-life prediction. The proposed DNN model for competitive and computationally efficient lifetime prediction is validated experimentally on several IGBT modules under different thermal stresses, and its accuracy is compared with other prediction methods.
Okilly et al. (Thu,) studied this question.