Optimization of Deep Learning Model Parameters in Classification of Solder Paste Defects

Abstract

Mass production processes of printed circuit boards (PCBs) are interrupted due to problems caused by soldering defects during the assembly of surface-mounted semiconductor electronics components to PCBs. This situation causes both an increase in production processes and costs and a decrease in production quality. Increasing production processes due to solder paste defects on PCBs, which can generally be detected at the final stage of the mass production process, cause the test processes of especially strategic projects to be disrupted. In this study, a deep learning model whose model parameters are estimated with population-based optimization algorithm that mimics atomic motion is proposed in order to detect the solder paste defects on PCBs at the early phase of the mass production process. AlexNet, one of the architectures with the least model complexity, is chosen for the convolutional neural network (CNN) model. The proposed optimization algorithm plays an important role in improving the performance of the model. In the study, six types classes are used, consisting of correct soldering, incorrect soldering, missing soldering, excess soldering, short circuit and undefined object. The performance of the proposed model has been experimentally tested and compared with the particle swarm optimization (PSO) based model approach. The results obtained confirm that the proposed model is satisfactorily successful in detecting solder paste defects on the PCB.