Predicting Rail Corrugation Based on Convolutional Neural Networks Using Vehicle's Acceleration Measurements.

Area of Science:

• Railway Engineering
• Machine Learning
• Predictive Maintenance

Background:

• Rail corrugation is a significant issue affecting railway infrastructure and operations.
• Accurate prediction of rail corrugation is crucial for efficient maintenance strategies.

Purpose of the Study:

• To develop and validate a deep learning model for predicting rail corrugation.
• To assess the model's performance using on-board sensor data (vertical acceleration and forward velocity).

Main Methods:

• Implementation of a One-Dimensional Convolutional Neural Network (CNN-1D).
• Utilizing on-board rolling-stock vertical acceleration and forward velocity measurements.
• Employing Gradient-weighted Class Activation Mapping (Grad-CAM) for model interpretability.

Main Results:

• The CNN-1D model achieved mean absolute percentage errors below 5% in predicting corrugation profiles.
• The model demonstrated the ability to reproduce corrugation based on real-time sensor data.
• Grad-CAM analysis confirmed the model's capability to identify different corrugation regions.

Conclusions:

• Data-driven approaches like CNN-1D show significant potential for real-time rail corrugation prediction.
• This predictive capability can enhance the reliability and efficiency of railway maintenance.
• Online monitoring of rolling-stock dynamics offers a viable pathway for proactive infrastructure management.