Pipeline Weld Radiographs Defect Detection Based on Improved YOLO V8

Abstract

Abstract Pipeline transportation is a crucial method for transporting oil and gas, and with the growing energy demand, the scale of oil and gas pipelines is expanding rapidly. Most of the oil and gas pipelines are joined together by welding. Weld defects often lead to significant safety accidents, making it essential to test the quality of the welds. Radiographic Testing (RT) is a widely used Non-Destructive Testing (NDT) technology. The manual inspection of weld negatives obtained by X-ray inspection is a common method for detecting weld defects, but it involves a heavy judgment workload, making it prone to misjudgment and omission. An enhanced algorithm based on YOLO V8 for identifying and localizing weld defects was proposed in this paper to address this issue. The Receptive Field Block (RFB) mechanism was introduced to enhance the network’s feature extraction ability by using different sizes of receptive fields. This approach aims to simulate the human visual system and reduce misleading image backgrounds for small object detection. The EIOU loss function, which considers the overlap area between the detection frame and the real frame, the distance from the center point, and the actual difference between the length and width of the sides, was employed to address the issue of the fuzzy definition of the aspect ratio. Meanwhile, a dataset of six types of defects, including round defects, strip defects, cracks, slag inclusion, incomplete fusion, and incomplete penetration, was established. The experimental results show that the mean average precision (mAP) of the improved YOLO V8 reaches 89.3% for six types of defects. Compared to advanced object detection algorithms like Faster R-CNN, YOLO V5, and the original YOLO V8, the improvements are 44.92%, 22.88%, and 1.4%. The method can accurately and quickly identify a wide range of pipe weld defects.