Hybrid YOLOv8 and auto-encoder based welding defect detection model for industrial inspection

Welding serves as an essential process that determines the strength and safety of industrial manufacturing welds. The detection of material defects through cracks and porosity and improper fusion leads to material failure during early stages of the process. Traditional inspection methods require substantial time and high costs while depending on human expertise which creates a high risk of mistakes. The research presents a hybrid deep learning approach which combines YOLOv8 for object detection with an auto-encoder based anomaly detection system. The YOLOv8 model detects known defects, while the auto-encoder identifies deviations from normal weld patterns. The hybrid system, which consists of both supervised and unsupervised learning methods, helps to enhance system reliability. The experimental results demonstrate that the model achieves an accuracy of 70% and a precision of 62% together with a 100% recall rate which guarantees detection of all defective welds. The proposed system is suitable for industrial applications where safety is a priority.