Predicting Cargo Train Failures: A Machine Learning Approach for a Lightweight Prototype
- 1.2k Downloads
In cargo transportation, reliability is a crucial issue. In the case of railway traffic, the consequences of locomotive failure are not limited to the affected machine, but are propagated through the railway network and may affect public transport as well. Therefore it is desirable to predict and avoid failures. In order to do this, constant monitoring of the trains’ systems and measurement of the relevant variables is required, but often not implemented. In this paper we leverage the existing technology of the 185 locomotive series and build a layered model for power converter failure prediction that can be applied without additional technology. We train instance anomaly detectors based on the pattern structure of the locomotives’ diagnostic messages from historical data records. For this purpose we selected rule and decision tree learning because they can be easily implemented in the existing software, whereas more complex classifiers would require costly software adaptations. In order to predict a time series of instances, we construct a meta classification layer. We then evaluate our model on the data of 180 locomotive tours by leave one out classification. The results show that the meta classifier improves classification accuracy, which will allow us to use this technology in a fielded prototype installation without disturbing daily operations.
KeywordsFalse Alarm Window Size Instance Classifier Positive Instance Instance Level
This work has been co-funded by the DB Schenker Rail project “TechLok” and by the LOEWE initiative (Hessen, Germany) within the NICER project [III L 5-518/81.004].
- 1.Cohen, W.W.: Fast effective rule induction. In: Proceedings of the Twelfth International Conference on Machine Learning, pp. 115–123 (1995)Google Scholar
- 2.Kauschke, S., Schweizer, I., Janssen, F.: On the challenges of real world data in predictive maintenance scenarios: a railway application. In: Görg, S., Müller, G., Bergmann, R. (eds.) Proceedings of the LWA 2015 Workshops: KDML, FGWM, IR, and FGDB, pp. 121–132. CEUR Workshop Proceedings, October 2015Google Scholar
- 5.Martinez-Rego, D., Fontenla-Romero, O., Alonso-Betanzos, A.: Power wind mill fault detection via one-class v-svm vibration signal analysis. In: Proceedings of International Joint Conference on Neural Networks (2011)Google Scholar
- 7.Ross Quinlan, J.: C4. 5: Programs for Machine Learning. Morgan Kaufman Publishers, Inc., San Francisco (1993)Google Scholar
- 8.Sipos, R., Fradkin, D., Moerchen, F., Wang, Z.: Log-based predictive maintenance. In: Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1867–1876. ACM (2014)Google Scholar
- 9.Vaarandi, R., et al.: A data clustering algorithm for mining patterns from event logs. In: Proceedings of the 2003 IEEE Workshop on IP Operations and Management (IPOM), pp. 119–126 (2003)Google Scholar