Application of data mining techniques for the investigation of track geometry and stiffness variation. (May 2020)

• Record Type: Journal Article
• Authors: Mehrali, Mohammad; Esmaeili, Morteza; Mohammadzadeh, Saeed
• Is Part Of: Proceedings of the Institution of Mechanical Engineers. Volume 234:Number 5(2020)
• Page Start: 439
• Page End: 453
• Publication Date: 2020-05
• Subjects: Track geometry -- track stiffness -- data mining -- classification -- decision tree -- clustering -- dominant wavelength filtering; Railroads -- Periodicals; Personal rapid transit -- Periodicals; 625.1
• DOI: 10.1177/0954409719844885
• Languages: English
• ISSNs: 0954-4097

Abstract

Railway tracks are one of the most important national assets of many countries. The major part of the annual budget of railway companies concerns repairing, improving, and maintaining railway tracks, which is a challenge for railway managers. The logical method of repair and maintenance should take into account all the economic and technical aspects of the problem and proper management of track maintenance—without knowing the factors and parameters responsible for the track failure—quality control methods, and finally, the choice of the appropriate repair methods. Railway track geometry is the main factor that identifies the track behavior and condition. It is based on measuring the geometric parameters of the track determined by the track quality indices. The existing track quality indices mostly represent the geometrical condition of the railway track superstructure. In the past years, the effects of track bed stiffness on the track condition have been investigated. This paper investigates the railway track condition based on the railway track geometry parameters as well as the vertical track stiffness. A method for continuous measurement of track stiffness along a railway line is described and demonstrated. By measuring the track geometry parameters and stiffness, the superstructure and the substructure condition of the railway track are assessed. In addition, the relation between these data is investigated by using data mining techniques such as classification, decision tree, clustering, and dominant wavelength filtering. It is shown that filtering the data based on the dominant wavelength provides the best correlation between the track geometry in the vertical direction and stiffness.