BP neural network for non-invasive IGBT junction temperature online detection
Li Liu, Qianlei Peng, Huaping Jiang, Li Ran, Yang Wang, Changhong Du, Jian Chen, Hongbo Zhou, Yang Chen, Zhiyuan Peng
Abstract
A number of studies have focused on Insulated Gate Bipolar Transistor (IGBT) junction temperature prediction methods. Some methods introduce extra circuits or sensors for an invasive prediction. However, the most of current methods hardly consider the situation that the junction temperature prediction may be affected by the IGBT degradation after long term operation. In this paper, the neural network using DC link voltage, switching frequency, load current and Negative Thermal Coefficient thermistor (NTC) temperature as inputs is proposed for junction temperature online prediction. The neural network training is completed through the IGBT module tests. The accelerated aging tests are then performed to validate the neural network results over lifetime. The static and transient working conditions are respectively validated at different aging degrees. Finally, a neural network with good accuracy could be achieved by a series of experiments considering IGBT device degradation.