High-speed Railway Catenary Based On UAV Patrol Inspection Analysis And Research On The Condition Of Nut And Cotter Pin

In an electrified railway catenary system,the high-speed rail catenary is a device installed above the high-speed train and powered by a pantograph.In high-speed rail catenary,fasteners are required for the connection of equipment.Fasteners in the railway mainly include bolts,nuts,cotter pins,pins and U hoops and other strong and small connecting elements,which play the role of connecting and anchoring in the catenary.Due to various outdoor weather and other influencing factors,the high-speed rail catenary often has loose nuts and missing cotter pins,which seriously endangers the safety of high-speed railways.At this stage,the inspection of the high-speed rail catenary is mainly carried out through manual inspection and high-speed rail catenary power supply safety detection and monitoring(6C system).Among them,manual inspection is dangerous,and the inspection efficiency is low.After the high-speed rail catenary suspension state detection device(4C system)is completed,a large number of testing workshop personnel are required to screen the problem pictures,the task is cumbersome,and the angle of the 4C inspection vehicle camera is fixed during shooting,and there is a blind area of inspection,which is not conducive to discovering equipment defects.In this thesis,it is proposed to use UAV as the inspection carrier to make up for the blind spot of the high-speed railway catenary suspension state detection device,and use deep learning technology to detect defects in the upper bolt nut area of the hanging string and the additional system cotter pin area collected by the inspection to ensure the safe operation of the high-speed railway catenary.Firstly,the high-speed railway catenary is inspected by UAV to obtain the inspection data set,and this thesis proposes to carry out preprocessing experiments on the data set,mainly including image processing,brightness processing,small target replication processing and simulation processing under different weather conditions,so as to enhance the diversity of the dataset and improve the accuracy of model training.Then,the nut and cotter pin are positioned,and an improved Efficient Det model is proposed,which introduces the Fused-MBConv module into the feature extraction network of the original model to improve the running speed of the model,and then introduces the hybrid attention mechanism(CBAM)module to strengthen the feature extraction ability.In the feature fusion network,the weighted bidirectional feature pyramid network(Bi FPN)is added with cross-layer connection to strengthen the feature fusion ability,and the ablation experiment of the improved Efficient Det model and the comparative experiment of different object detection models are carried out.Experiments show that the mean accuracy(m AP)of the improved Efficient Det model is increased by 5.98% compared with the original model,and the average detection time(ADT)is reduced by 0.016 s compared with the original model.After the segmentation detection and state analysis of nuts and cotter pins,this paper proposes to improve the DeepLabv3+ model,improve on the original DeepLabv3+ model,first use the lightweight model Mobile Net V2 to replace the feature extraction network of the Xception model,and then introduce the effective channel attention(ECANet)module in the decoding part,and carry out ablation experiments and comparative experiments on different semantic segmentation networks for the improved DeepLabv3+.Experiments show that the number of parameters and calculation speed(FLOPs)of the model are reduced by 85% compared with the original model,and the uniformly intersected union ratio(MIo U)is increased by 3.91%compared with the original model,which improves the semantic segmentation accuracy of the model,improves the operation speed of the model,and reduces the number of parameters of the model.This paper introduces the principle of state analysis of nut and cotter pin,and measures the judgment results by missing detection rate and false detection rate,and experiments show that the missed detection rate and false detection rate of nut and cotter pin state analysis are less than 8%.Finally,the UAV high-speed rail catenary patrol and intelligent analysis system software is introduced,which is based on Python language,which can realize user data storage,UAV inspection sample extraction,key component target detection and key part semantic segmentation.To a certain extent,the research content of this thesis promotes the intelligence and specialization of UAV inspection high-speed railway catenary.