Application of machine learning methods for predicting hazardous failures of railway track assets

The Aim of the paper is to reduce the number of hazardous events on railway tracks by developing a method of prediction of rare hazardous failures based on processing of large amounts of data on each kilometre of track obtained in real time from diagnostics systems. Hazardous failures are rare events; the set of variate values of the number of such events for an individual kilometre of track per year is: [0, 1]. However, for a railway network as a whole the yearly number of such events is in the dozens and efficient management requires the transition from the estimation of the probability of hazardous failure occurrence to the identification of the most probable location of failure. Methods. The problem of identification of rare, but hazardous possible events out of hundreds of thousands of records of non-critical railway track parameter divergences cannot be solved by conventional means of statistical processing. Hazardous events are predicted using the above statistics and artificial intelligence. Big Data and Data Science technology is used. Such technology includes methods of machine learning that enable item classification based on characteristics (features, predicates) and known cases of undesired event occurrence. The application of various algorithms of machine learning is demonstrated using the example of prediction of track superstructure failures using records collected between 2014 and 2019 on the Kuybyshevskaya Railway. Findings and conclusions. The result of facility ranking is the conclusion regarding the location of the most probable hazardous failure of railway track. That conclusion is based on the correspondence analysis between the actual characteristics of an item and conditions of its operation and the cases of adverse events and cases of their non-occurrence. The practical value of this paper consists in the fact that the proposed set of methods and means can be considered as an integral part of the track maintenance decision-making system. It can be easily adapted for online operation and integrated into the automated measurement system installed on a vehicle.

machine learning, railway track facility failure, decision trees

1. Sukonnikov G.V. [Application of the Internet of Things by JSC RZD]. www.rzd-expo.ru; 2017. Available at: URL: http://www.rzd-expo.ru/innovation/novosti/1.pdf. (in Russ.)

2. Bondarenko Yu.V., Kukso A.A., Markevich I.G. [Information technology in railway track diagnostics system management]. Proceedings of the IX international research and practice conference of students, post-graduate students and young scientists in 4 volumes. 2018;214-215. (in Russ.)

3. Nazarov D.G., Guda D.A. [On systems for automated track measurement]. Krasnodar: Kuban State Technological University: Scientific works of the Kuban State Technological University; 2019:135-146. (in Russ.)

4. Kurakina S.G., Shumakova E.G. [Automation of diagnostics and monitoring of railway catenary sections]. Sovremennye innovatsii. 2017;8(22):15-17. (in Russ.)

5. Zamyshliaev A.M. Premises of the creation of a digital traffic safety management system. Dependability. 2019;4(71):45-52.

6. Kuznetsova G.A., Krasheninnikov S.V., Kraysvitny V.P., etc. [Upgrading the GID Ural-VNIIZhT system]. Avtomatika, sviaz, informatika. 2016;11:15-19. (in Russ.)

7. Zamyshliaev A.M., Ignatov A.N., Kibzun A.I., Novozhilov E.O. Functional dependency between the number of wagons derailed due to wagon or track defects and the traffic factors. Dependability. 2018;18(1):53-60.

8. Liu X., Saat M., Barkan C. Analysis of causes of major train derailment and their effect on accident rates. Transp. Res. Rec. J. Transp. Res. Board. 2012;2289:154–163.

9. Zamyshliaev A.M., Ignatov A.N., Kibzun A.I., Novozhilov E.O. On traffic safety incidents caused by intrusion of derailed freight cars into the operational space of an adjacent track. Dependability. 2018;18(3):39–45.

10. Lasisi A., Attoh-Okine N. Principal components analysis and track quality index: a machine learning approach. Transp. Res. Part C. Emerg. Technol;91:230–248.

11. Zarembski A.M. Better railroading through Big Data. www.railwayage.com; 2018. Available at: https://www.railwayage.com/analytics/better-railroadingthrough-big-data.

12. Thaduri A., Galar D., Kumar U. Railway assets: a potential domain for big data analytics. Proc. Comput. Sci. 2015;53:457–467.

13. Li Q., Zhong Z., Liang Z. et al. Rail inspection meets big data: methods and trends. 18th International Conference on Network-Based Information Systems. 2015:302–308.

14. Flach P. Machine Learning: the art and science of algorithms that make sense of data. Moscow: BMK Press; 2015.

15. Goodfellow I., Bengio Y., Courville A. Deep Learning. Moscow: DMK Press; 2018.

16. Widodo A., Yang B.S. Support vector machine in machine condition monitoring and fault diagnosis. Mech. Syst. Signal Process. 2007;21:2560–2574.

17. Sun W., Chen J., Li J. Decision tree and PCAbased fault diagnosis of rotating machinery. Mech. Syst. Signal Process. 2007;21:1300–1317.

18. Cerrada M., Zurita G., Cabrera D. et al. Fault diagnosis in spur gears based on genetic algorithm and random forest. Mech. Syst. Signal Process. 2016;7071:87–103.

19. Santur Y., Karakose M., Akin E. Random forest based diagnosis approach for rail fault inspection in railways. National Conference on Electrical, Electronics and Biomedical Engineering. 2016:714–719.

20. Chistiakov S.P. [Random forests: an overview]. Transactions of Karelian Research Centre of the Russian Academy of Sciences. 2013;1:117-136.

21. Hosmer D., Lemeshov S., Sturdivant R.X. Applied Logistic Regression. New York: John Wiley & Sons; 2013.

22. Samworth R.J. Optimal weighted nearest neighbour classifiers. Ann. Statist. 2012;40(5):2733–2763.

23. Famurewa S.M., Zhang L., Asplund M. Maintenance analytics for railway infrastructure decision support. Journal Qual. Maint. Eng. 2017;23:310–325.

24. Hu C., Liu X. Modeling Track Geometry Degradation Using Support Vector Machine Technique. Joint Rail Conference; 2016.

25. Boyko P.Yu., Bikov E.M., Sokolov E.U., Yarotsky D.A. Application of Machine Learning to Incident Ranking at Moscow Railway. Journal of Information Technologies and Computing Systems. 2017;2:43-53. (in Russ.)

26. Jiang Y., Wang H., Tian G. et al. Fast classification for rail defect depths using a hybrid intelligent method. Optik (Stuttg). 2019;180:455–468.

27. Gibert X., Patel V.M., Chellappa R. Deep multitask learning for railway track inspection. IEEE Trans. Intell. Transp. Syst. 2017;18:153–164.

28. Reznitskiy M.A., Arshinskiy L.V. Software implementation of the upper structure of the railway track defects detection automated system based on the technology of the convolutional neural networks. Electronic Scientific Journal “Young Science of Siberia”. 2018;1. (in Russ.)

29. Lee J.S., Hwang S.H., Choi I.Y. et al. Prediction of track deterioration using maintenance data and machine learning schemes. J. Transp. Eng. Part A Syst. 2018;144:04018045-1:9.

30. Nakhaee M.C., Hiemstra D., Stoelinga M. et al. The Recent Applications of Machine Learning in Rail Track Maintenance: A Survey. Lecture Notes in Computer Science. 2019;91–105.

31. Fawcett T. An introduction to ROC analysis. Pattern Recognition Letters. 2006;27:861–874.