A Defect Detection Method for Rail Surface and Fasteners Based on Deep Convolutional Neural Network.

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Railway track components like rails and fasteners are susceptible to defects due to operational stress and environmental exposure.
• These defects pose significant risks to train operation safety.
• Existing inspection methods may be limited in scope or efficiency.

Purpose of the Study:

• To propose a novel multiobject detection method for nondestructive inspection of railway track defects.
• To enhance the safety and reliability of railway operations through advanced defect detection.

Main Methods:

• Utilized an improved YOLOv5 framework for localizing rails and fasteners in track images.
• Employed Mask R-CNN for detecting and segmenting rail surface defects.
• Applied a ResNet-based model for classifying fastener conditions.
• Validated the method on diverse railway track images from a high-speed line.

Main Results:

• The proposed deep convolutional neural network method demonstrated superior performance in detecting rail surface and fastener defects.
• Achieved high accuracy in localization, detection, segmentation, and classification stages compared to other deep learning algorithms.
• Verified robustness and effectiveness across ballast and ballastless track types.

Conclusions:

• The developed deep learning approach provides an effective and robust solution for nondestructive detection of railway track defects.
• This method significantly contributes to improving railway safety and maintenance efficiency.