Junction Temperature Monitoring of Power Devices Using Convolutional Neural Networks

The temperature-sensitive electrical parameter (TSEP) -based method enables accurate junction temperature monitoring (JTM) of power devices. However, the majority of TSEPs are susceptible to errors due to the effects of load currents and device aging, reducing the accuracy of JTM. To address this, a JTM method based on a convolutional neural network (CNN) model is proposed to deal with the unfavorable effects of two factors. In this method, the turn-on collector current (I_C) is selected as the TSEP, and the temperature characteristics of the turn-on I_C are thoroughly analyzed by a mathematical model. Moreover, the parameter dependence of the turn-on I_C is fully investigated with extensive double-pulse tests. Then, considering the significant effect and the frequent variations of load current in practice, the adverse effects of load current are mitigated based on the CNN. Finally, experimental verification is given to prove the effectiveness and accuracy of the proposed model based on a single-phase rectifier.