Research On Intelligent Inspection Trolley For Track Defects Based On Machine Vision

Railway safety is the basic guarantee for national economic development,and regular track inspection has also become an indispensable part of railway work.In addition to the regular inspection of large-scale rail inspection vehicles in my country’s railway inspection work,various railway bureaus also use different rail inspection equipment in daily inspections to replace manual inspections.Large-scale rail inspection vehicles have high detection efficiency,but are inconvenient to schedule and expensive.Testing equipment such as rail detectors is inefficient and not suitable for long-term testing.In view of the above problems,this paper will design a track defect intelligent inspection trolley based on machine vision to complete the daily track inspection tasks of the railway,and use the deep learning target detection algorithm to replace the traditional detection algorithm to improve the speed and accuracy of rail defect detection.First of all,the general design scheme of the track inspection trolley is proposed.After understanding the industry background and the operation principle of the railway inspection car,the basic performance requirements and scheme of the inspection car in this paper are determined.According to the composition of the inspection trolley,it is divided into three parts: image acquisition system,defect detection algorithm and inspection trolley processing test.Secondly,the orbit image acquisition system is designed according to the orbit characteristics.The high-speed line scan camera is used as the core device for image acquisition,and the photoelectric encoder is used to control the camera for image acquisition.Considering that the external light has a great influence on the image acquisition,the strip light source is selected as the auxiliary light source to ensure the image acquisition quality,and the lithium battery is used for the image acquisition.The device is powered and the acquired images are stored with a disk array.In the software part,the acquisition and display interface of the image-based rail inspection system is designed to display the image acquisition results and use it for data comparison in the later rail defect detection.It mainly includes three parts: parameter setting module,continuous display and saving module,file naming and image extraction.After the image acquisition software and hardware design is completed,an image acquisition platform is built in the laboratory to test the acquisition results,and the motion state of the inspection trolley is simulated by a conveyor belt.The test results are good,and the high-speed acquisition of rail images is realized.In addition,use the deep learning target detection algorithm to locate and identify the track defects.After data analysis,it is found that the amount of sample data of rail surface defects is small,and the necessary premise of the high accuracy of the deep learning image detection method is that the model training samples are sufficient.DCGAN)rail sample augmentation method to solve the problem of insufficient data.After the samples are sufficient,the deep learning model for rail surface and fastener defect detection is trained.After analyzing the current mainstream deep learning target detection algorithms,combined with the actual object,a candidate region-based deep convolutional neural network(Faster Region-Convolutional Nrural Networks,Faster R-CNN)for defect detection,and test the generated rail samples to verify their usability.Finally,the track inspection trolley is designed to be equipped with an image acquisition device to obtain track defect images and verify the accuracy of the defect detection algorithm.A portable track image acquisition trolley was built to acquire images on the rail test line,and the 3D modeling of the track inspection trolley was improved after improving the problems in the on-site acquisition process.According to the load and speed requirements of the trolley,the matching motor is calculated and controlled,and then the trolley is processed by statics simulation of the frame part to verify the rationality of the structure.Image acquisition After the electrical equipment and motors are installed,the test results in the laboratory are good and the on-site rail images are obtained.After classifying the collected samples with defects,a defect data set of the track surface and fasteners is made to test the defect detection results.The experimental results show that the image collection efficiency of the inspection trolley is high,and the collected rail images are clear,complete and undistorted.Compared with manual inspection,the defect detection speed is fast and the accuracy is high.Compared with manual inspection,it reduces the workload and improves the detection efficiency,which meets the design requirements.