Lifetime prediction of power MOSFET based on LSTM with successive variational mode decomposition and error compensation

Abstract

Accurate prediction of the remaining useful life (RUL) of metal oxide semiconductor field effect transistors (MOSFETs) is the key to safe and reliable operation of power electronics. In this paper, we combine long short-term memory (LSTM) networks with successive variational mode decomposition (SVMD) and use error compensation methods to build a lifetime prediction model, which improves the performance of the prediction model by reducing the interaction between different sequences and using error sequence compensation. The results show that, compared with the Bayesian optimized LSTM, the method has the advantages of high prediction accuracy and low prediction uncertainty.

Monitoring method of solder layer void damage of IGBT module based on transfer function

“…There are two existing methods for monitoring the status of solder layer: model-based method and datadriven method [5,16]. Based on the model method, the parametric failure physical model is used to determine the material characteristics and the aging of packaging factors [17]. The data-driven method uses the measured data information to monitor the aging status of the solder layer, which needs the relevant theories such as pattern recognition and machine learning [18].…”