Intelligent Condition Monitoring of Multiple Thermal Degradation of IGBT Modules Based on Case Temperature Matrix

This article presents an intelligent approach for monitoring various thermal degradations in insulated gate bipolar transistor (IGBT) modules. Thermal degradation at the baseplate solder and thermal interface material (TIM) both induce changes in the case temperature distribution, offering insights into the corresponding degradation type and level. To capture this information, a strategically placed thermocouple array forms a temperature matrix on the case surface. The normalization of the case temperature matrix effectively mitigates the influence of working conditions, such as load current or heatsink temperature. Furthermore, a model augmented with a convolutional neural network is utilized to classify the degradation type and level based on temperature matrixes. To generate datasets under diverse operational conditions, a finite-element model of a conventional IGBT module is developed. The model includes various degradation locations (baseplate solder fatigue, TIM degradation, baseplate solder, and TIM degradation) and levels (health, early degradation, severe degradation, failure). Gaussian noise is introduced to the simulation data to enhance both training and prediction accuracy. Additionally, experimental data from a two-level inverter is incorporated into the training dataset to further refine accuracy. The results affirm that this approach excels in accurately distinguishing the degradation type and evaluating the degradation level.