Loading...

Modeling the Impact of Climatic Conditions on Geometric Variations of Railway Tracks: A Deep Learning Approach

Sedigh, Saeedeh | 2025

14 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58778 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Shafahi, Yousef
  7. Abstract:
  8. Climate conditions are one of the key factors influencing railway infrastructure, and future climate changes and the occurrence of extreme weather events increase the risks to the sustainability and performance of these infrastructures. In this study, the relationship between climate conditions and geometric degradation of railway tracks in Iran is investigated using deep learning models. Geometric data of railway tracks from the EM120 track recording machine over a 14-year period were combined with meteorological data including temperature, humidity, precipitation, and wind speed for all 19 railway regions in Iran (covering approximately 14,000 kilometers of track and 180 GB of data). Subsequently, LSTM and GRU deep learning models were developed to analyze and predict geometric degradation patterns. To evaluate the role of meteorological data in improving prediction accuracy, an ablation study was conducted. In this study, the models were trained once using only geometric data, and then again using a combination of geometric and meteorological data. The results showed that incorporating meteorological data significantly improved the accuracy of degradation predictions, and the LSTM model outperformed the GRU model. This study can enhance the understanding of the impact of climatic variables on railway infrastructure and enable more accurate prediction of track lifespan and optimized maintenance planning. Furthermore, the findings provide a foundation for developing more climate-resilient railway infrastructure
  9. Keywords:
  10. Degradation Trend Prediction ; Deep Learning ; Track Car Recorder (EM120) ; Meteorological Data ; Ablation Study

 Digital Object List

 Bookmark

No TOC