Research On Detection And Judgment Of Rail Damage Based On Intelligent Analysis Of Time And Space Detection Data

At present,China Railway uses rail inspection vehicles to quickly detect the internal damage of in-service rails,and the collected spatio-temporal detection data is used to identify and discriminate rail damage.On the one hand,the detection effect of the internal damage of the rail is affected by the detection speed of the rail inspection vehicle,the performance of the probe alignment,the coupling water saturation and other related factors.There is a problem that the detection performance of the inspection vehicle needs to be improved and the detection effect is not good.On the other hand,the rail inspection vehicle has the disadvantage of high false positive rate in the recognition of the damage.So the manual two-stage screen-by-screen playback mode has to be adopted.The data analysis workload is large and the discriminating efficiency is low.According to the above background,the detection and identification of rail damage based on intelligent analysis of spatio-temporal data were studied.The study is aiming at improving the intelligence and automation level of "inspection,identification and judgment" of rail damage,and further improving rail damage detection.The accuracy of detection and identification provides a more reliable guarantee for railway transportation safety.The main research contents are as follows:(1)From the aspects of rail inspection vehicle detection method,rail damage identification and rail damage discrimination of rail inspection vehicles,the research status of domestic and foreign research is studied.Problems and difficulties in rail damage detection and identification in China are analyzed.Research goals for the main issues of "inspection,identification,and judgment" were put forward and research technology routes were formulated.(2)In order to further improve the detection efficiency and damage detection effect of the rail inspection vehicle,the characteristics of the inspection system of the rail inspection vehicle were analyzed,and the influence of the detection speed of the inspection vehicle on the detection efficiency was studied.The ultrasonic reflector detection was analyzed.Based on the principle,the detection model of common reflector and rail bottom is proposed.The coupling degree of the probe wheels,including the performance of the probe wheels alignment and the coupling water saturation,is analyzed.The static test in the laboratory explores the probe alignment.The attenuation law of the ultrasonic signal is discovered.The attenuation model of the deviation of the probe wheel to the ultrasonic signal is proposed.The attenuation law of the coupled water saturation on the ultrasonic signal at different speeds is explored on the flat straight line.The attenuation model of the signal is proposed.Using the standard ultrasonic reflector as a reference,the influence of the detection speed on the deviation of the probe wheels and the coupling water saturation is studied.The prediction model of the ultrasonic reflector detection is proposed.The dynamic step size adjustment algorithm is studied.An inspection vehicle detection method based on detection prediction model and dynamic step size adjustment algorithm is proposed.(3)In order to solve the problem of timeliness of on-line intelligent identification of typical rail damage,the generation principle and data format of spatio-temporal detection data(ultrasound B-display data)collected by rail inspection vehicle are analyzed,and typical rail damage are analyzed.The rail typical damage images are studied,and the pre-processing of B-display data reorganization and standardization is studied.The DBSCAN method is proposed to spatially segment the ultrasonic B-display data.The characteristics of the AlexNet network are analyzed,and the segmented ultrasonic B-display images are input to the AlexNet network.Feature extraction and damage classification are done.The XGBoost method for neuron cutting of two fully connected layers is proposed.An online intelligent identification method based on DBSCAN-AlexNet model is proposed,which can quickly and automatically preprocess ultrasonic B-display data,spatially segment and extract features.(4)In order to solve the problem of comprehensively identifying all types of rail damages,the extra-long characteristics of ultrasonic B-display data are analyzed.The sliding window recommendation network technology is used to preprocess the ultrasonic B-display data.The characteristics of the Faster RCNN network architecture are analyzed.A fine recognition model of SW-Faster RCNN model for ultrasonic B-display data is proposed.To ensure that the damage is not missed and false positive,a cropping network and a relabel model are proposed.Based on this,a fine identification method based on SW-Faster RCNN model is proposed,which can detect and identify the type and position of rail damages in B-display data.The comprehensiveness and accuracy of rail damage identification is improved.(5)In order to solve the problem of low efficiency of manual damage detection by manual twice playback,based on the segmentation of the ultrasonic detection data and the fine identification of the rail damage,the segment ultrasonic reflectors are arranged in order.The relative distance attribute between the ultrasonic reflectors is added to generate segment detection data.According to the historical detection period segment detection data and the line infrastructure information database data,segment standard detection data is generated.The segment detection data and the segment standard detection data are regarded as two long strings,and an automatic contrast analysis method based on the Edit Distance algorithm is proposed.The segment detection data is automatically compared with the segment standard detection data to discriminate the rail damage.The ultrasonic reflectors of different sections and different grades are graded to reduce the computational complexity and time complexity of the automatic comparison.After the comparison is finished,the result of the rail damage determination is output.The review result is dynamically updated to the segment standard detection data for the identification of the rail damage in the next detection cycle.The discrimination report is automatically formed.Thereby,the efficiency of the rail damage discrimination is improved.The research contents of rail inspection vehicle detection method,the intelligent identification of the rail damage and the automatic discrimination of the rail damage were tested respectively,and the test data of the three stages of inspection,identification and judgment were analyzed respectively.As a result,the effectiveness of the proposed methods was verified.The detection effectiveness,recognition speed and discrimination efficiency of the rail damage were improved.