Soldering reflow process optimization based on simulation

This research proposes a reflow thermal recipe (RTR) optimization model based on the after-reflow defect inspection simulation results. RTR, which refers to the temperature setting of each zone in a forced convection reflow oven, is the most important parameter during the soldering reflow process (SRP). The components are attached to the printed circuit board (PCB) by the solder joint generated during SRP. RTR can affect the solder joint quality, so it becomes the main factor to determine PCB quality. Therefore, a proper RTR is very important for SRP. An optimization model is proposed in this research to reduce the after-reflow solder joint defect level by maintaining the optimal RTR setting. In particular, a customized defect metric (DM) is proposed based on automatic optical inspection (AOI) results and is defined as the objective function (OF) of the optimization model. Then, a simulation model is applied to improve the optimization efficiency. Simulation is a commonly used tool to generate the data based on the history data characteristics or mimicking the actual production line environment. Simulation is widely applied in industry-related research because it significantly saves the experimental costs (i.e., material and time). A simulation model can be developed based on environmental factors or data characteristics. In this research, the random forest regression (RFR) model is selected and trained with practical experimental data to estimate the OF value of the optimization model. To solve the optimization problem with high efficiency, evolution strategy (ES) with a self-adaptive search area is applied. The proposed model has important significance for SRP. There are several advantages of the proposed model. First, the application of the machine learning (ML) method increases the calculation ability and optimization efficiency. Also, the proposed model combines the reflow oven and AOI machines, which increases the machine-to-machine communication efficiency and automation degree. The proposed optimization framework is also possible to apply to other PCB assembly lines for quality improvement and RTR setup. The experimental results prove the effectiveness of the proposed model. To be specific, the identified RTR reduces the DM by 74% and increases the pass rate by 2.3% compared to the original RTR and is consistent with the practical experimental results.