Detecting and classifying multiple track defects using clustering algorithm

Abhilash Vishwakarma

doi:10.1016/j.engstruct.2025.120197

Detecting and classifying multiple track defects using clustering algorithm

2025-04-01

0

PRE

AI

A

Abhilash Vishwakarma *

A

Anirvan DasGupta

DOI：10.1016/j.engstruct.2025.120197

Original

Original request for help

Save

Abstract

En 中文

This work proposes an algorithm to detect and classify railway track defects by analysing the axle-box acceleration in the vertical direction. The axle-box acceleration is generated by using an experimentally validated vehicle track model. The vehicle-track model is built in the commercial multibody dynamics software SIMPACK. When the vehicle moves over the track defects, each defect excites a specific band of frequencies. To separate the track defects based on the frequency content the axle-box acceleration is passed through a filter bank to decompose it into its constituents. Mean and standard deviation are calculated for each decomposed signal to form a feature matrix. On the feature matrix, principal component analysis is performed to extract the orthogonal features and select the dominant features. On these dominant features, density-based spatial clustering of application with noise (DBSCAN) is applied to group the data points into defective and perfect clusters. The effect of vehicle speed, axle load, defect features, and epsilon value is analysed for the defect detection algorithm. The track defects are classified by applying the DBSCAN algorithm on selected decomposed signals. The proposed algorithm is found to accurately detect and locate the track defects without being affected by the external parameters.

Keywords:

Track defects

Filter bank

Clustering algorithm

Detecting and classifying multiple track defects using clustering algorithm

Abstract

Journal

Organization

Cited Papers