Detection of stencil printing direction in printed circuit boards assembly using solder paste inspection data

This study addresses the challenge of accurately detecting the printing direction of solder paste in printed circuit boards (PCB) assembly, a crucial factor for optimizing process parameters aligned with Industry 4.0 standards. Traditional methods heavily rely on computer vision algorithms for pattern recognition and face significant limitations due to the diversity of pad shapes in PCB assembly. These methods are computationally expensive for online streaming monitoring and process optimization. Therefore, two alternative feature extraction algorithms were developed utilizing solder paste inspection data to address these challenges. The first approach automatically extracts solder paste measurements from the PCB’s edges to train a random forest (RF) model. The second utilizes a moving window technique with a genetic algorithm (GA) for optimized region extraction. These methods significantly reduce the dependency on manual settings and the limitations of image-based algorithms. Upon testing two different types of PCBs, a detection accuracy of 94% was achieved in the evaluation at the worst case when setting the classification threshold to 85%. The edge extraction method, in particular, showed superior performance, especially on PCBs with consistent and clear edges. However, challenges were observed with PCBs printed using different machines, suggesting the impact of varying printing conditions on the detection accuracy. This highlights the importance of considering machine-specific patterns and the adaptability of threshold settings in classification.