A Physics-Informed Machine Learning Model for Mounting Optimization in Printed Circuit Boards

The increased use of lead-free solder and miniaturization in surface mount technology present challenges in achieving accurate assembly alignment. In modern electronics manufacturing, self-alignment is crucial for improving assembly quality by aligning misaligned components. Previous research explored self-alignment using physics-based models or machine learning approaches. However, these methods require significant computational costs or extensive datasets. To address this, the study introduces a hybrid model that integrates data from inspection machines and features representing component displacement caused by self-alignment, derived from a physics-based model utilizing surface tension energy, Young's equation, and a simplified solder fillet profile. The proposed model is compared with other machine learning algorithms to demonstrate its effectiveness in predicting component locations. Side overhang and end overlap metrics are employed to evaluate its performance in assembly quality. The study aims to optimize the mounting process and enhance assembly quality by leveraging the accurate predictive capabilities of the model.