Noncontact Reflow Oven Thermal Profile Prediction Based on Artificial Neural Network

Abstract

This research proposes a thermal profile prediction model in the soldering reflow process (SRP). The thermal profile is the temperature of the solder paste during SRP and is affected by multiple factors [i.e., temperature setting for the oven chamber (thermal recipe) and conveyor belt speed]. The quality of the printed circuit board (PCB) is determined mainly by the thermal profile, because the components are attached to the board by the solder joint, which is generated during SRP. The thermal profile shape decides the performance of the solder joint, such as its strength. Solder paste manufacturers usually provide a target profile based on the physical properties of the solder paste composition. Comparing the experimental profile and target profile is the most simple, direct, and effective way to judge the PCB quality. In general, the thermal profile is collected from the thermocouples attached to the solder paste. To overcome drawbacks such as low testing efficiency, a noncontact machine learning-based thermal profile prediction model is developed in this research to monitor thermal performance. The noncontact prediction is of great significance to the PCB production line and meets the requirements of Industry 4.0 with a high degree of automation. Specifically, the prediction model is developed based on an artificial neural network (ANN). Several environmental factors are utilized as input data to forecast the board temperature. Then, a pooling model is devised to address the thermal profile. The proposed model is mainly for the passive components, such as 1005. The prediction model shows the promising ability to predict the thermal profile by showing a 97.2% accuracy with the testing data.