Predicting contact-without-connection defects on printed circuit boards employing ball grid array package types: a data analytics case study in the smart manufacturing environment

Abstract

This research presents an exploratory data analytics case study in defect prediction on printed circuit boards (PCB) employing ball grid array (BGA) package types during assembly. BGA package types are of interest because defects are difficult to identify and costly to rework. While much of the existing research is dedicated to techniques to identify and diagnose BGA defects, this research attempts to preempt them by using parametric data measured by solder paste inspection (SPI) machines as input data to applied machine learning models. Two modeling approaches are explored: one approach to analyze individual solder paste deposits and the other approach to holistically analyze all solder paste deposits on a single PCB location. The latter approach employs feature generation to extract a broad set of features from the arrays of SPI data and feature selection techniques for dimensionality reduction. Models trained on the reduced feature sets provide encouraging initial results, with precision, recall, and f1 score metrics exceeding 0.82, 0.50, and 0.62 respectively for each of two datasets analyzed.