Loading...

Application of Reinforcement Learning in Railway Track Maintenance Management and Renewal

Amiri, Hanieh | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57614 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Shafahi, Yusof
  7. Abstract:
  8. The significance of the rail transportation system in the economic, political, and social realms of countries, along with the high costs of managing and maintaining its infrastructure, are among the most significant reasons for the importance of a railway track maintenance management system. Such a system is crucial to maintaining service levels, ensuring reliability and safety of the tracks to an acceptable level, and minimizing maintenance costs. This research focuses is on identifying the optimal maintenance policy for Iranian railway tracks, considering the budget constraints in the maintenance planning. For this purpose, the output of the EM120 machine, which was obtained from the survey of tracks from 2009 to 2020, was utilized. This dataset, encompassing all 19 railway regions of Iran, covers about 14,000 kilometers of tracks and and includes around 140 gigabytes of data. Following necessary preprocessing of the database, the Iranian railway tracks were classified into 14 categories based on geographical location, axle load, and traffic volume. The tracks were then segmented into 200-meter sections. The study employed composite indicators such as the Track Geometry Index (TGI) and the single track quality indicator (SD) to assess track quality. Subsequently, models for predicting track deterioration were developed using machine learning and deep learning methods. Among deep learning approaches, the Long Short-Term Memory (LSTM) model was evaluated, while the Multilayer Perceptron (MLP) algorithm was tested within machine learning methods. The predictive results of these models were presented and compared. To model the problem and identify an optimal maintenance policy that minimizes maintenance costs while adhering to safety standards and ensuring adequate service levels, a reinforcement learning approach and Q-learning algorithm were employed. The developed model, utilizing specific railway track data, reports an optimized maintenance plan for the next five years for various points within the railway network. The results from the optimized maintenance planning using reinforcement learning, in addition to confirming previous research findings, provided valuable information for future planning periods, thereby enhancing the effectiveness of maintenance strategies
  9. Keywords:
  10. Maintenance ; Reinforcement Learning ; Deep Learning ; Machine Learning ; Railway Network ; Railway ; Maintenance Management ; Maintenance Cost ; Maintenance Scheduling

 Digital Object List

 Bookmark

No TOC