The classical Iterative Closest Point(ICP)registration algorithm is commonly applied to rail profile wear detection sys-tems.However,during the operation of track inspection vehicles,the profile data collected by the rail profile wear detection system based on linear structured light is often disrupted by noise from various outliers,leading to significant geometric variations in the rail pro-file calculation results.Given the dynamic detection accuracy requirements for rail profile wear using track inspection vehicles of±0.5 mm for lateral wear and±1.0 mm for vertical wear in practical applications,it is crucial that the design of wear detection algorithms do not compromise the detection accuracy due to outlier-induced noise in the profile data.To address this,this paper presented an improved two-stage rail profile wear detection algorithm.In the first stage,this algorithm enabled rapid initial registration based on feature point pairs extracted from the rail profile,providing an improved initial pose for the two point clouds.The second stage utilized an enhanced robust ICP algorithm for precise registration,followed by calculating geometric parameters related to rail profile wear.A laboratory ex-perimental platform was set up to evaluate the developed rail profile wear detection system.This setup simulated typical disruptions from outliers in the measured data of rail web and rail base segments,in comparison with the measurements obtained using a manual contact-type wear tester for benchmarking.Moreover,the accuracy errors and effectiveness of wear detection were analyzed by comparing the classical ICP algorithm and articles with the proposed improved algorithm.Furthermore,a comparative evaluation was conducted to ex-amine the detection speed of the algorithms,and the accuracy in repeatability measurements using the improved algorithm was verified.Finally,validation was conducted on a metro line using a track inspection vehicle,and the results highlighted the enhanced accuracy and speed of rail profile wear detection based on the improved algorithm in scenarios with disruptions from outliers,demonstrating the practi-cal engineering value of the proposed algorithm.
维普
万方数据