Predictive Monitoring of Wirebond Ultrasonic Signal on Electrical Test Result Using Machine Learning

This study presents the development of a cost-effective predictive monitoring system interfaced to an existing wirebonding machine to determine its corresponding electrical test result. The system was tested on one production lot with the data gathered subjected to a machine learning process to predict if the gathered signal will be rejected and will become a possible failure at later electrical testing. Three classification algorithms: Logistic Regression, Decision Tree, and Support Vector Machine were used to evaluate which prediction algorithm provides the expected electrical results. The study shows that 98% to 99 % accuracy was achieved in order to predict whether the lot will produce high production yield or not. Its embedded design approach lend itself well to real-time operation. More importantly this study would provide a way for legacy manufacturing equipment to be upgraded and thus be integrated into other system that are designed for “Industry 4. 0” implementation.