Parametric analysis of railway infrastructure for improved performance and lower life-cycle costs using machine learning techniques

Jose A. Sainz-Aja, Diego Ferreño, Joao Pombo, Isidro A. Carrascal, Jose Casado, Soraya Diego, Jorge Castro

Research output: Contribution to journalArticlepeer-review

Abstract

Rigorous and efficient management of the railway infrastructure is crucial to avoid accidents and reduce operation and maintenance costs. This requires in-depth knowledge of the assets, the interaction among them and the effect that each track parameter has on the overall infrastructure performance. In this study, a large set of studies are carried out, on a previously calibrated finite element slab track model, where the relevant track parameters are varied within their usual ranges. The results are then used to train and validate a series of predictive models based on Machine Learning algorithms. This methodology provides greater understanding and enhanced prediction of the behaviour of tracks, which are composed of multiple variables such as the soil/subgrade, supporting layers, sleepers, pads and rails. The study also considers train axle loads and service speeds, which are other key elements that influence the track performance. The results show that the parameters that have greatest influence on the railway infrastructure are the properties of the soil, characteristics of the rail pads and the axle loads. This work can support the implementation of predictive maintenance procedures for railway tracks and the development of innovative technological solutions, providing responses to the industrial needs of reducing costs and contributing to improve the competitiveness of railway transport.

Original languageEnglish
Article number103357
Number of pages16
JournalAdvances in Engineering Software
Volume175
Early online date16 Nov 2022
DOIs
Publication statusPublished - 1 Jan 2023

Fingerprint

Dive into the research topics of 'Parametric analysis of railway infrastructure for improved performance and lower life-cycle costs using machine learning techniques'. Together they form a unique fingerprint.

Cite this