Application of Machine Learning to Recognize Wire Bond Lift-Off in Power Electronics Manufacturing

Abstract

In this paper, machine learning is used to teach a software to detect wire bond lift-offs in power electronic modules. To show the feasibility of this method, a DCB with four aluminum wires is analyzed. Using SolidWorks, different failure states of the device under test with varying wire bond lengths and heights are created. The models are then running through a magnetostatic simulation in Ansys Maxwell. Using the simulation results, sixteen magnetic field values are extracted based on their placements in an existing printed circuit board. The values of some simulation results are used to train a machine learning algorithm based on supervised learning, while the rest are used to verify the algorithm. For this work, the support vector machine and decision tree algorithms are tested and compared to each other. The results show that both methods work well and can give good results for one wire failure using a data set of only 100.