CargoCBM – Feature Generation and Classification for a Condition Monitoring System for Freight Wagons

Despite the fact that rail freight transport is one of the most environmentally friendly matters of transport, its growth has been far behind the growth of freight transport in general. Studies showed that a competitive disadvantage is caused by a low availability of rolling stock, especially freight wagons. Changing from a time based to a condition based maintenance strategy is believed to decrease down times by at least one third. To make condition based maintenance for freight wagons possible the TU Berlin and five industry partners started the research project CargoCBM. One task in this project is to develop algorithms for the automatic on-board diagnosis of wheel flats. The focus of the work is on the process of feature generation and feature selection as well as the application of different classifiers to automatically evaluate the data. Based on the results of measured data, features were selected and tested with different classifiers. Thought advanced classifiers such as neural networks have been analysed in accordance to their classification accuracy. It can be shown that with carefully constructed and selected features comparatively simple classifiers can lead to excellent results.