Smart-Inspection System on Assembly Process of Pin-Through Components Using Machine Learning

Abstract

This paper proposes using machine learning techniques to implement a failure mode classifier for automatic fail classification in pin-through hole (PTH) connector terminals in printed circuit boards (PCB). The Support Vector Machine (SVM), K-nearest neighbor (KNN), and Decision Tree (DT) algorithms were used. It was evaluated using a dataset of real images from manufacturing multimedia centers for the algorithm training phase. Subsequently, it thoroughly evaluated the results of the metrics obtained from each trained model. The main objective is to select the model with the best precision in predicting two failure modes to be implemented at the automotive factory and improve the inspection phase to reduce the defect and rework rates. The failure mode classifier trained with the SVM algorithm obtains the best precision, with an accuracy of 99% in predicting the dataset of tested images. KNN and DT achieved 78% and 79% accuracy, respectively, but DT was unstable. The final decision was to implement the SVM algorithm that obtained the best accuracy in decision-making for the failure modes evaluated in the research.