IGBT Junction Temperature Estimation Based on External Parameters of Multiple Drive Devices in Practical Industrial Scenario

The insulated gate bipolar transistor (IGBT) is one of the most important power semiconductor devices in power electronics and is also prone to failure. High junction temperature and junction temperature fluctuation of IGBT are the main causes of IGBT module aging failure. The high-precision monitoring of the junction temperature of the IGBT module is a prerequisite for IGBT life prediction, which is crucial for reducing maintenance costs and improving equipment reliability. Therefore, an IGBT junction temperature estimation method based on long short-term memory (LSTM) neural network and sliding window estimation model is proposed and applied in practical industrial scenarios. This method uses the operating data of the motor drive device in the actual industrial application scenario as the training and test data set and uses the external operating parameters of the IGBT module to estimate the junction temperature of the IGBT module. Compared with the internal operating parameters of the IGBT module based on switching transient, the external operating parameters are easier to collect and process, and more suitable for practical application scenes. A sliding window estimation model is proposed to estimate the junction temperature of the IGBT module. Compared with the point-to-point estimation method, the sliding window estimation method can capture the influence of historical operation data better and has a higher capability of time series data estimation. The IGBT junction temperature estimation of sliding windows is realized by the LSTM neural network, which is more suitable for time series estimation in real industrial scenarios. The experimental results show that the estimation accuracy of the sliding window estimation method is better than that of the point-to-point estimation method, and the accuracy of the sliding window estimation method based on LSTM is better than that of the sliding window estimation method based on other machine learning models. It proves that the proposed method can better capture the dynamic process of the system and has higher estimation accuracy.