Loading...

Application of Machine Learning Algorithms on Railway Track Deterioration Modeling

Sedighi, Alireza | 2020

456 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 52633 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Shafahi, Yosef
  7. Abstract:
  8. The vital and influential role of railways in the socio-economic growth and development of countries, along with the high costs of rehabilitation and maintenance of its infrastructure, has made the existence of a comprehensive maintenance management system a necessity to not only elevate the level of service, reliability, and safety but also to minimize such repair costs. Accurate modeling of track deterioration is essential as one of the components of such systems.In this study, a large database from outputs of EM120, track recording machine, was provided for the years 2009 to 2018 and for all 18 railway zones of Iranian Railways (approximately 13,000 km railway track and 80 GB of data). The necessary pre-processing was done, and in addition to identifying existing errors, methods for resolving it were presented. In this regard, it is necessary to first express the quality status of track segments in the form of a quantitative indicator. In order to compare the performance of the indices in expressing the quality, the new TTQI index was calculated for the equal segments of Railway tracks in comparison with the famous TGI index and the proposed standard EN indices (SD and CoSD). Also, to evaluate the effect of segment length, the above indices were calculated for segments of once 100 m long, and 200 m again.Track segments were categorized into 14 classes based on railroad identification information such as axial load, traffic volume, and geographical location. For each track segment, time series of the length17 (six-month interval) was created plus class attribute and zone number for use in modeling. Finally, by considering two prediction and classification approaches, as well as neural network algorithms, support vector machine, and decision tree, the best model in each case was presented and compared
  9. Keywords:
  10. Railway Track Deterioration ; Machine Learning ; Track Quality Index ; Track Car Recorder (EM120)

 Digital Object List

 Bookmark

No TOC