ReflowNet: ConvLSTM-based direct reflow oven recipe optimization framework

This paper presents ReflowNet, a domain-adaptive convolutional long short-term memory (ConvLSTM) neural network-based oven recipe optimization framework for reflow soldering, a critical step in printed circuit board (PCB) assembly using Surface Mount Technology (SMT). The proposed framework simultaneously utilizes synthetic datasets generated from physics-based computational fluid dynamics (CFD) simulations and real-world experimental trial data to predict the optimal oven recipe based on process-specific spatiotemporal information. ReflowNet addresses the limitations of previous methods by (I) considering deviations between simulation and experimental results, (II) directly predicting the oven recipe instead of the solder temperature profile, and (III) explicitly incorporating spatiotemporal information related to the reflow soldering process. Experimental results demonstrate that the proposed model accurately estimates the oven recipe and provides lower estimation variance across different recipe settings. By leveraging the power of domain adaptation (DA) and ConvLSTM network, ReflowNet offers a novel and effective solution for optimizing reflow soldering oven recipes in PCB assembly.