A model comparison method in digital inspection of railway track wear

Abstract

Railway track condition has its importance in relation to the driving safety and transport capacity of railways, and its inspection is a crucial task for railway maintenance. Rail wear directly influences wheel-rail contact and the life of rail tracks, so precise and effective inspection of railway wear is a continuous demand. In this paper, a comparison method between point clouds from a structured lightscanner and CAD models of the rail is proposed for railway track wear measurement. With the segmentation algorithm based on Euclidean clustering and random sample consensus (RANSAC), the wheel-rail and non-wheel-rail contact profile are extracted for the alignment. Comparing the measured data and CAD model, random points sampling from CAD models is conducted to generate enough points for the data alignment. In the coarse registration, the curvature of the rail profile is utilized for the point feature histogram (PFH) generation. In the fine registration, the points on non-wheel-rail contact profiles are utilized in the ICP algorithm. To verify the effectiveness of the model comparison, the analysis from software Geomagic Qualify is referenced. It proves that the model comparison method can provide the fundamental support for railway track wear evaluation.

Rail wear inspection based on computer-aided design model and point cloud data

“…Rail profile was obtained by a 3D scanner in Zhou et al, 11 but the rail profile alignment was implemented on a 2D cross section of the rail. To inspect continuous rail profile, stripe structured light scanners were introduced in Chen and colleagues, 12,13 and the comparisons between the point cloud data and 3D computer-aided design (CAD) model were conducted. The 3D point cloud was also introduced to reconstruct the surface of the worn parts for remanufacturing.…”

“…The convergence of this algorithm is ensured by the input states of the two groups of data. 16 In the comparisons of Chen and colleagues's 12,13 studies, the point clouds from 3D scanner were aligned to those from CAD models in coarse and fine registrations. For good initial pose estimation, sample consensus initial alignment (SAC-IA) was used in the coarse registration.…”