Weld Defect Detection with YOLOv10

Abstract

Welding is one of the important processes used in various industries with various applications. The change of weld defects has the feature of continuous critical monitoring of safety, quality control and cost-effectiveness in industrial production ranges. Although traditional high accuracy offers time-consuming, it depends on the product and operator experience. This study implements three-class detection of Bad Weld, Good Weld and defect with YOLOv10 object detection for automatic detection of weld defects. In the relevant data set, the model provides 0.939 Precision-Confidence and 0.91 Recall-Confidence values. The obtained results show that the model can detect defects. This study aims to reveal the potential of deep learning in the detection of weld defects, providing a faster, cost-effective and reliable solution.