Automated Detection of IQI Wires and Weld Defects in Ship Pipe Radiographic Images Using Deep Object Detection Model

Nondestructive Testing(NDT) is used to ensure the reliability of ship pipe welds, and RT is the most widely used method. However, RT interpretation relies on experimented technicians. Thus IQI interpretation and defect detection automatic systems are in demand. Accordingly, we analyzed and evaluated the performance of object detection models for IQI and weld defects. The dataset was constructed using YOLOv8x and Faster R-CNN models for IQI wire detection, and YOLOv8s for defect detection. The result of experiment showed that YOLOv8x had high robustness in detection IQI wires in RT images and demonstrated excellent performance overall. Faster R-CNN showed high detection rates in high-quality images but had some over detection. In defect detection, the YOLOv8s model showed high precision but low recall due to missing small defects. The results of this study demonstrate that object detection models can effectively be applied to the automation of RT image-based pipe welding defect inspection. In the future work, we include expanding the dataset to enhance model performance.