Research On Detection And Recognition Method Of Rail Bolt Fault

In recent years,with the opening of China-Europe Express and other long-distance crossborder passenger and freight trains,railway transportation has played an increasingly important role in the field of land long-distance transportation.Railway transportation has made important contributions to cross-regional economic integration,social communication,and humanistic interaction.How to ensure the safe operation of railway transportation has attracted wide attention.Various failures of rail joint bolts are one of the factors that cause railway transportation accidents.The daily maintenance and maintenance of rail joints account for 60%-70% of the track maintenance workload.Therefore,It is of great significance to study the rapid detection and identification of the failure of rail joint bolt components.The main research work is as follows:First,the rail bolt failure,the application of machine vision in rail transit detection,and the current research status of target detection algorithms at home and abroad are summarized,and then all possible failures of the rail bolt assembly are analyzed,and the causes of various failures are explained.Design the image acquisition method of rail bolt assembly.Secondly,a special data set of rail bolt component failures was established,and several algorithms commonly used in the field of target detection were selected for comparative experiments.The experiment proved that YOLOv4 is excellent in evaluation indicators such as detection speed,average accuracy,and average average accuracy,and it is suitable for use.Target detection and recognition algorithm used in engineering,and YOLOv4 is optimized according to the characteristics of rail bolts.According to the characteristics of the sample size of the track bolt assembly data set,a priori box suitable for the size of the bolt assembly is generated through clustering,which also provides a more reliable regression basis for coordinate prediction.In the improvement of the YOLOv4 optimizer,an adaptive optimization algorithm is introduced,combined with the default stochastic gradient descent algorithm of the YOLOv4 target detection model,to form an optimization algorithm of adaptive and stochastic gradient descent fusion,and set the switching time of the two algorithms.The fused algorithm exhibits a rapid decline in the loss function in the early stage of the training phase,and has good convergence in the later stage.Then the target detection and recognition algorithm model is pruned and reduced in volume.Finally,the lightweight algorithm was transplanted to the test platform for experiments and optimization,and the accuracy of the identification of the failed bolt components was tested.The test results show that the rail bolt component fault detection and identification method based on deep learning can effectively identify rail bolts such as missing nuts,broken nuts,loose nuts,missing gaskets,missing bolts,and broken bolts,with an average accuracy rate of 84.8 % Above,FPS can be kept stable at about 20.12 fps,which can quickly and accurately complete the detection task of bolt failure at the rail connection.