Pollution Detection on Rail Surface for Adhesion Evaluation Using Multispectral Images

Abstract

There is a continuous rise of the number of trains' passengers to transport, therefore a densification of traffic is studied. For safety reasons, it becomes necessary to know the wheel-rail contact condition, in traffic, as it affects driving variables as adherence. The latest determines passenger's safety and the proper functioning of train equipment but it is often deteriorated due to recurrent pollution of the rail. From the analysis of the rolling surface, one can know if the rail is polluted or not and extract the kind of pollution and its influence on the train driving. The aim of this work is to contribute to guarantee and improve passengers' safety. In this paper, we propose a method for pollution detection and clustering, as a first step to the rail surface characterization. Methods using multispectral image analysis, specific data calibration, and hierarchical clustering based on Non Negative Matrix Factorization (NMF) are used.