Prognosis for insulated gate bipolar transistor based on Gaussian Process Regression

Abstract

The failure issues of power insulated gate bipolar transistor (IGBT) modules are mainly related to thermal and thermo-mechanical aging mechanism. This aging causes degradation of the device performance and faults which can lead to the failure bringing huge loss and catastrophic influence. To avoid those failures, the monitoring of the device operation and the detection of an aging state remain as a priority. This paper, at first, describes the failure mechanism of power cycling by analyzing of the structure of lead-based solder and joint failure due to solder fatigue. Secondly, Gaussian Process Regression (GPR) model supporting uncertainty representation is used to realize the prognosis for the junction temperature of the IGBT. Furthermore, the comparison of GPR prediction with the Neural Network algorithm has been achieved, and the dynamic model is introduced to improve the prediction accuracy for the IGBT health assessment. Meantime, GPR owns more simple computational complexity and less time consuming.