Research on X-ray nondestructive defect detection method of tire based on dynamic Snake Convolution YOLO model

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Vision

Background:

• Tire X-ray nondestructive testing is vital for driving safety but challenged by complex structures and diverse defects.
• Traditional inspection and current machine learning methods lack sufficient accuracy and efficiency for tire defect detection.

Purpose of the Study:

• To develop an innovative and highly accurate tire X-ray nondestructive testing technique.
• To enhance the detection performance of tire defects using an improved YOLOv5 model.

Main Methods:

• Proposed an enhanced YOLOv5 model incorporating Dynamic Snake Convolution (DSConv) for slender features.
• Introduced a DSConv-based C3 module for specific defects like cord-overlap and cord-cracking.
• Redesigned the neck network with Scale sequence feature fusion (SSFF) and Triple feature encoding (TFE) for multi-scale information integration.
• Integrated Convolution Block Attention Module and employed Soft-NMS for improved defect recognition and bounding box selection.

Main Results:

• The proposed algorithm achieved a 5.9% increase in mAP$_{0.5}$ and a 5.7% increase in mAP$_{0.5:0.95}$ compared to the benchmark YOLOv5 model.
• Demonstrated superior detection accuracy over mainstream object detection algorithms.
• Successfully completed the nondestructive testing task for tire defects.

Conclusions:

• The developed YOLOv5-based technique significantly improves tire X-ray nondestructive testing accuracy and efficiency.
• The integration of DSConv, SSFF, TFE, and attention mechanisms effectively addresses challenges in detecting diverse tire defects.
• This method offers a robust solution for ensuring tire safety through advanced defect identification.