Research On Key Detection Technologies Of Rail Fastener Assembly Robot

Freight railway plays an important role in the transfer of goods at ship ports and terminals,and under the background of the rapid development of China’s railways,the requirements for intelligent railway construction equipment are becoming higher and higher.At present,there are professional equipment operations for the construction of steel rails,sleepers and other parts.However,the assembly of rail fasteners is still manually operated during the construction and maintenance of the railway.Railway construction companies urgently need a rail fastener assembly robot to improve efficiency of the work and realize the standardization of fasteners construction and one of the key technologies to realize the rail fastener assembly robots is to realize the positioning of rail fasteners.In this paper,the detection of rail fasteners as the entry point,the multi-sensor fusion scheme is adopted to realize the detection and positioning function of the rail fastener assembly robot on the whole,and the corresponding solutions for the detection of rail bolts and rail fasteners are proposed,which lays the foundation for realizing the automatic and standardized assembly of the rail fastener assembly robot.In view of the hopping of the rail bolt data collected by the laser sensor and the various feedback signal waveforms,this paper proposes a rail bolt detection model based on a one-dimensional convolutional neural network.The optimal structure of rail bolt detection model was determined by comparing the parameters of different convolution layers,convolution kernel size and number of convolution kernel.Based on the collected laser sensor data set,the one-dimensional convolutional neural network detection model has a better effect than the two-dimensional convolutional neural network and the traditional shallow classification model,and the recognition accuracy of the rail bolt is99.39%.Aiming at the identification and location of rail fastener and nut in complex construction environment,this paper proposes a rail fastener identification and location method based on improved YOLOV3.The main improvements include adding SPP,adjusting YOLOV3 network prediction structure,and re-clustering of prior box through improved K-means ++.Based on the self-established VOC format rail fastener data set,the detection and evaluation indexes of Faster R-CNN,SSD,YOLOV3 and the improved YOLOV3 detection model were compared and analyzed.The improved YOLOV3 model has improved the recognition accuracy and speed to a certain degree,with an average detection accuracy of 98.67%.The results meet the requirements of rail fastener assembly robot for the whole rail fastener and rail nut detection and positioning.By analyzing the construction requirements of the rail assembly robot of track fasteners,the system software of the assembly robot of track fasteners is designed and developed.In order to verify the multi-sensor fusion detection scheme proposed in this paper,the positioning test platform of the track coupler assembly robot was built,and the positioning test method of the track coupler proposed in this paper was verified by the test platform and the system software of the track coupler assembly robot.