A Machine-Learning-Based Approach for Railway Track Monitoring Using Acceleration Measured on an In-Service Train

Area of Science:

• Civil Engineering
• Railway Engineering
• Structural Health Monitoring

Background:

• Railway track degradation, particularly loss of stiffness in sub-layers, can compromise safety and operational efficiency.
• Traditional track inspection methods are often labor-intensive, costly, and may not detect subsurface issues effectively.
• Developing non-disruptive monitoring techniques is crucial for proactive maintenance and ensuring railway infrastructure integrity.

Purpose of the Study:

• To propose and validate a novel approach for detecting stiffness loss in railway track sub-layers using in-service train measurements.
• To develop an Artificial Neural Network (ANN) model capable of identifying track damage based on acceleration response energies.
• To introduce a Damage Indicator (DI) for visualizing and quantifying track degradation.

Main Methods:

• Utilized acceleration responses from an in-service train, processed into energy values per 15m track slice.
• Developed and trained an ANN model using simulated and real-world train acceleration data for healthy track conditions.
• Simulated track damage by reducing sub-ballast layer stiffness, representing hanging sleepers.
• Introduced a DI based on prediction errors between simulated and predicted energies to assess damage levels.
• Conducted a sensitivity analysis on the DI's performance considering signal noise, slice size, and multiple damage locations.

Main Results:

• The ANN model successfully learned healthy track conditions from acceleration signal energies.
• The proposed Damage Indicator (DI) effectively visualized and differentiated various levels of stiffness loss in the track sub-layers.
• Sensitivity analysis indicated the DI's robustness against moderate signal noise and variations in slice size.
• The method demonstrated potential in identifying multiple damaged locations, though performance varied.

Conclusions:

• The developed ANN-based approach offers a promising, non-disruptive method for monitoring railway track sub-layer stiffness.
• The proposed Damage Indicator provides a valuable tool for visualizing and assessing track degradation, aiding proactive maintenance.
• Further research is recommended to refine the DI for complex scenarios, including varying train speeds and diverse damage types.