The Optimal Solution of Reflow Oven Recipe based on Physics-guided Machine Learning Model

Abstract

This paper presents a physics-guided machine learning model to provide the optimal reflow recipe for a 7-zone oven. The numerical method based on the computational fluid dynamics (CFD) simulation was used to predict reflow profiles of a BGA package. After validating the CFD model with the measurement results, an automated system was programmed to collect profiles subjected to 81 sets of boundary conditions (reflow recipe). A machine learning model trained by 81 sets of input data was employed to predict profiles subjected to 148,176 sets boundary conditions rapidly. The peak temperature and time above liquidous of output profiles were extracted to quantify the performance of the corresponding boundary conditions. The boundary condition with the best reflow performance was regarded as the optimal recipe.