Sim2Joint: Dynamic hybrid model for solder joint prediction across Sim2Real

The objective of this research is to predict the solder joint's fillet profile before its formation. Solder joints are crucial for the structural and operational reliability of electronic assemblies, yet their integrity can be compromised by defects such as cold joints, voids, or insufficient solder. Traditional physics-based simulations attempt to model these phenomena but often fall short due to simplifications that fail to capture real-world variability. Conversely, data-driven approaches leverage historical data from Surface Mount Technology (SMT) lines to predict joint quality, though their effectiveness can be hampered by data noise and imbalance. Addressing these limitations, this research introduces a hybrid modeling framework named Sim2Joint, which combines physics knowledge-based simulations with the adaptability of data-driven methods. By introducing Sim2Real in the joint simulation domain, Sim2Joint bridges the gap between simulation and real-world situations by integrating dynamic weights for printing and placing factors with real-world data, enhancing prediction accuracy and reliability. The framework also includes uncertainty quantification to provide more reliable solder joint fillet profile predictions, thereby enabling better decision-making and optimization in SMT processes. Sim2Joint is validated against various baselines, showcasing its capability to adapt to real-time changes and improve the predictive performance of solder joint quality assessments.