Automated Welding Defect Detection Using YOLO: A Deep Learning Approach for Industrial Quality Assurance

Abstract

Welding is a critical manufacturing process in which defects such as cracks, porosity, spatter, and incomplete fusion can significantly compromise structural integrity and product reliability. Conventional inspection methods are labor-intensive, subjective, and difficult to scale for high-throughput industrial environments. This paper presents an automated welding defect detection system based on a lightweight YOLO11s object detection model designed for real-time industrial quality assur ance. The proposed system is trained and validated on a limited dataset of labeled weld images, reflecting realistic industrial data constraints. Experimental results demonstrate that the model achieves a mean Average Precision (mAP@0.5) of 46.23% with moderate recall, indicating its suitability as an automated screeningand inspection assistance tool. The study highlights the feasibility of deploying lightweight deep learning models for visual inspection tasks under limited data availability and computational constraints.