Track Geometry Prediction Using Three-Dimensional Recurrent Neural Network-Based Models Cross-Functionally...

Area of Science:

• Railway Engineering
• Artificial Intelligence
• Data Science

Background:

• Increasing rail transport demands necessitate optimized railway track maintenance for safety and efficiency.
• Current maintenance strategies may not align with escalating operational demands, leading to premature track deterioration.
• Predictive and preventative maintenance are crucial for optimizing railway system performance.

Purpose of the Study:

• To develop novel deep machine learning models for predicting railway track geometry parameters.
• To integrate these predictive models with a building information modeling (BIM) framework for enhanced maintenance planning.
• To demonstrate the efficacy of machine learning and BIM in advancing predictive maintenance in railways.

Main Methods:

• Development of three-dimensional recurrent neural network-based co-simulation models for track geometry prediction.
• Application of various recurrent neural network techniques for predictive modeling.
• Creation of a building information modeling (BIM) model for integrating track geometry data and predictive outputs.

Main Results:

• Machine learning models achieved an average R² of 0.95 and a mean absolute error of 0.56 mm in predicting track geometry.
• The developed BIM models facilitate information exchange for effective predictive maintenance.
• The study successfully co-simulated track geometry measurements with predictions.

Conclusions:

• Deep machine learning and BIM integration offer a powerful approach for predictive railway maintenance.
• This methodology can significantly enhance the safety and cost-effectiveness of railway infrastructure management.
• The findings represent a breakthrough in applying AI for optimizing railway operations and maintenance.