A Thermal-Mechanical Deformation Based Fault Diagnosis Method Utilizing Support Vector Machine Algorithm for IGBT Bonding Wire Crack Detection

Abstract

Bonding wire cracking is one of the most common failure modes in welded encapsulated insulated gate bipolar transistor (IGBT) modules. This paper presents a thermal-mechanical deformation based fault diagnosis method utilizing support vector machine (SVM) algorithm for IGBT bonding wire crack detection. Firstly, an experimental platform is set up to obtain the thermal-mechanical deformation signals of chip in the IGBT modules. Then in order to facilitate the algorithm for fault identification, wavelet transform is carried out to denoise the origin signals, based on which the deformation distribution maps are reconstructed for intact bonding wires and defective bonding wires, respectively. Finally, an SVM algorithm classification model which utilizes obtained deformation distribution maps to find the optimal separation hyperplane with the largest interval defined in feature space is proposed to diagnose crack fault. The result shows that the model can classify whether the bonding wire cracked with high accuracy up to 98%. The proposed fault diagnosis method provides an effective and reliable technical means for the detection of bonding wire faults in power electronic devices.