A Siamese Network-Based Non-Contact Measurement Method for Railway Catenary Uplift Trained in a Free Vibration Test.

Area of Science:

• Mechanical Engineering
• Electrical Engineering
• Artificial Intelligence

Background:

• Pantograph-catenary interaction significantly impacts railway operation and safety.
• Monitoring catenary dynamic uplift is essential for evaluating contact performance and interaction mechanisms.

Purpose of the Study:

• To propose a non-contact, deep learning-based method for real-time monitoring of catenary vibration.
• To validate the method's effectiveness through field tests under various conditions.

Main Methods:

• Utilized a fully convolutional Siamese neural network for contact wire tracking.
• Implemented a class-agnostic binary segmentation mask to mitigate recognition errors.
• Incorporated a novel down-sampling block to reduce feature loss under variable lighting.

Main Results:

• The proposed deep learning method demonstrated robust performance in identifying catenary vibration during field tests.
• Extracted uplift data correlated well with numerical simulations.
• The method proved effective under diverse lighting conditions and excitation levels.

Conclusions:

• The developed non-contact measurement method accurately monitors catenary vibration and uplift.
• This technique offers a valuable tool for railway inspection and understanding pantograph-catenary dynamics.
• The findings contribute to improved wave propagation and damping analysis in catenary systems.