Soldering Defect Detection with kNN and SVM for Quality Enhancement in Industrial

Abstract

Research into defect detection in industrial fields based on image processing techniques has gained significant momentum in recent years. The primary objective of this research has been to enhance productivity and reduce errors in industrial processes. In this context, the present paper introduces a novel approach for the classification of visual inspections in the domain of printed circuit board (PCB) soldering issues. The method leverages two machine learning algorithms, specifically the K-nearest neighbor (kNN) and the support vector machine (SVM). The study involved the collection of extensive data from a production line, culminating in a substantial dataset of 1 GB in size. This dataset was thoughtfully divided, with 75% of the data allocated for training the model and the remaining 25% reserved for rigorous testing to evaluate its performance. The research findings reveal that the support vector machine (SVM) algorithm outperforms the K-nearest neighbor (KNN) algorithm in terms of accuracy. Specifically, the SVM algorithm achieves an impressive accuracy rate of 97%, while the KNN algorithm attains a still commendable 92% accuracy rate. This outcome underscores the efficacy of the proposed classification method in addressing soldering problems within PCBs and underscores the potential for improving quality control and defect detection in industrial settings.