Neuro-Fuzzy Approaches to Estimating Thermal Overstress Behavior of IGBTs

Abstract

The Thermal overstress behavior of power semiconductor components is a determining factor to evaluate the reliability and performance of power electronic devices. Many theoretical and empirical methods have been presented to address the thermal effects of power electronics components on the quality of power systems. However, analyzing temperature brings to us a large number of uncertainties and nonlinearities affecting the accuracy of modeling. This paper proposes three neuro-fuzzy based techniques to estimate the temperature of Insulated Gate Bipolar Transistors (IGBTs). These techniques include grid partitioning clustering, Fuzzy C-Means (FCM) clustering, and subtractive clustering. An experimental dataset containing over 1.5 million data points is used to develop and train the proposed neuro-fuzzy approaches. This dataset is obtained during a comprehensive investigation on IGBTs and thermal effects by scientists at Ames Research Center of NASA. Preliminary results have demonstrated that the applied approaches are superior to estimating the thermal overstress behavior of IGBTs.