IGBT aging monitoring and remaining lifetime prediction based on long short-term memory (LSTM) networks

摘要

Reliable remaining useful lifetime (RUL) prediction of power semiconductors is challenging, despite knowledge of valid aging precursors. In this paper, an aging platform for IGBTs is built to collect the real-time Vce-on data and its thermal circuit is carefully designed to control the temperature range. A decent error and uncertainty of the measurement system have been identified. A novel machine learning technique that is effective to deal with the time-sequence data, i.e. recurrent neural networks (RNN) using long short-term memory (LSTM) units, is introduced to predict the RUL of IGBTs and compared with two conventional methods. The proposed method is found able to deliver a proper prediction at an early stage and update the results during the aging process.