Research On Rapid Detection Techniques For Rail Fastener Tension State Based On 3D Machine Vision

Rail fasteners serve as essential connecting components between rail and sleepers or other underlying foundations.Their primary role is to secure rail,preventing rail displacement and rail tilting,among other issues.The failure of these components presents potential threats to the safe operation of railways.Traditional methods of addressing this are manual track inspections during specific maintenance windows,which are marked by low efficiency and a risk of omissions.The adoption of sensors for rapid rail fastener state assessment is an emerging trend,with machine vision-based inspection technologies demonstrating considerable potential and advantages.Current research primarily focuses on 2D machine vision;however,due to the absence of height information,this approach is unable to detect the fastening force state of fasteners,leading to critical inspection clip gap.Therefore,this paper design a rail fastener tension state detection method based on 3D machine vision,aiming to further enrich and improve the rapid machine vision inspection capabilities for rail fastener conditions.The primary research work includes:Firstly,a 3D imaging system utilizing a line laser sensor is designed.In order to meet practical engineering requirements,the Gocator 2450 line laser sensor is chosen,and a mounting platform is designed to acquire dual-modal data consisting of rail fastener contour point clouds and RGB-D images.Experiments are designed to analyze the sampling data under diverse environmental conditions,demonstrating that the imaging system is capable of stably and effectively acquiring the required data.Secondly,a complete fastener positioning method based on dual-modal data is proposed.After data cleaning and point cloud denoising,two different rail fastener region segmentation schemes are designed for contour point cloud data based on point cloud and dual-modal data.Experimental results show that using dual-modal data can locate most fastener point clouds,with a detection rate of over 98%,meeting subsequent detection requirements and significantly reducing program runtime.Thirdly,a clip gap detection method based on the complete fastener point cloud is proposed,and the rail fastener tension state is determined according to the size of the gap value.Initially,the elastic strip point cloud is extracted using Euclidean clustering and height information.Subsequently,various rapid extraction methods for the elastic strip skeleton,based on binary image mapping,are designed,and the clip gap values are calculated using the elastic strip skeleton.A comparison of the clip gap errors obtained from different methods reveals that the most accurate approach involves fitting multiple points to the center of each cross-section of the elastic strip.Finally,in order to validate the accuracy of the proposed method for identifying the rail fastener tension state,several experiments were designed to assess its performance.Initially,different sampling intervals were tested,with a distance of1.67 mm found to yield the best overall results.Next,fasteners with varying gap intervals were examined,and the obtained results demonstrated an accuracy rate exceeding 99% within the permissible error range.Lastly,a total of 6,600 data files for WJ-2,WJ-8,and WJ-7 fasteners were tested,revealing detection rates of 98.883%,99.009%,and 99.075% for the rail fastener tension state detection,respectively.The experimental results indicate that the method presented in this paper can achieve rapid and accurate detection of rail fastener tension states.