In this paper, the authors introduced a comprehensive KDD methodology to build models to predict high impact wheel events before they disrupt operation. The proposed methodology innovates in key data-mining areas such as data representation, automatic labeling, model fusion, and model evaluation. Most of the steps are generic and can be reused in other circumstances such as in medical and environmental applications. They evaluated the feasibility of the methodology through a large-scale experiment in which they build a set of the heterogeneous base-level models and a set of meta-models to implement model fusion. After combining four base-level models, they have successfully constructed a multiple classifier system (MCS) capable of predicting 97 % of wheel failures while maintaining a reasonable false alert rate (8 %). Such results clearly show the great potential of this innovative application of data mining in the railway industry.
Learning to predict train wheel failures
Lernprozeß bei der Vorhersage von Ausfällen bei Eisenbahnrädern
2005
10 Seiten, 5 Bilder, 7 Tabellen, 27 Quellen
Conference paper
English
Engineering Index Backfile | 1930
|Engineering Index Backfile | 1950
Engineering Index Backfile | 1935
Engineering Index Backfile | 1935