| Intelligent railway is an important component in the construction of new infrastructure construction.Qinghai-Tibet Railway has put forward higher requirements for unmanned management and decision-making for the cold and high altitute operating conditions.The TFDS（trouble of moving freight car detection system）dynamically collects data such as real-time status images of the moving vehicles and predicts potential safety hazards.Studies on the TFDS system has great application potentials on improving the monitoring ability of Qinghai-Tibet railway freight car safe operation and reducing the waste of man-time for maintenance.This research work starts from the system and application of TFDS,analyzes its potential development direction in the future,and realizes the rapid fault detection with the intelligent image detection algorithm.The main research work of the full text mainly includes the following sections,Firstly,this paper analyzes the basic composition,technology and function of TFDS railway wagon operation fault dynamic image detection system,expounds the operation mode and operation organization,operation flow and standard of TFDS centralized operating platform,as well as the current process of TFDS fault detection,confirmation and escalation.Secondly,based on the historical big data of the TFDS of the Qinghai-Tibet Railway,the main fault types,the temporal and spatial frequency of faults and the detection rate are analyzed.The main types of failures include bending of hook joists,damage to steel floor,loss of relief valve levers,foreign objects on vehicles,and deformation of human brake pedals.The temporal and spatial frequency and detection rate of truck faults are counted,and the trial situation of the TFDS truck fault image intelligent recognition system developed by related manufacturers is analyzed.Finally,based on the YOLOX algorithm for intelligent image recognition of vehicle faults and foreign objects,the relevant research basis of the YOLOX algorithm is first introduced,and then the model data set is constructed based on Labelme.Based on RTX3090 model training,for the fault of coupler joist bending as an example,the m AP value of the training result is 0.65;for foreign objects mounted on the vehicle,the m AP value of the training result is 0.52.The intelligent identification method could realize fault or abnormal detection,and the detection speed could reach 2 images per second,and the average error rate is 5.61%.Anomaly detection solutions based on cameras and intelligent algorithms have low cost and could be deployed on a large scale.In addition,this chapter also discusses the intelligent railway perception system framework based on cloud-edge-device collaboration,which provides a feasible solution for balancing the cost of information transmission and computing efficiency.This research "Qinghai-Tibet Railway Freight Car Fault Diagnosis System and Intelligent Image Detection Method" carries out research work from the aspects of system,application and key technology,and proposes the further improvement direction of TFDS system.good application value. |
