EXPLAINS: Explainable Anomaly Prediction for SMT Solder Joints Using SPI Data

This research proposes an explainable anomaly prediction framework, called EXPLAINS, for solder joint quality prediction through printing quality data. Early anomaly detection is vital for preventing solder joints with anomalies, which lowers the rework costs. Meanwhile, if the reason behind the anomalous pattern can be found, it can help manipulators identify the cause of bad solder joints. Motivated by several inspiring novel observations on the correlation between the quality indicators of printed solder paste and corresponding solder joint defects, this research aims to accurately predict the abnormality probability of solder joints’ height after the Solder Paste Printing process and provide explainable and instructive prediction outcomes for manipulators. It bridges the gap between explainable machine learning and predictive anomaly detection for solder joints while considering the non-negligible noise from production lines. The proposed approach has been evaluated with chip resistor R0603M and SAC305 solder paste. The experimental results indicate that EXPLAINS can provide manipulators serving SMT product lines with valuable guidance.