| Under the general trend of the railway industry developing in the direction of high speed and heavy load,the importance of railway transportation in people’s lives has increased.Safety issues in the operation of high-speed railways have become a top priority.However,high-speed railways’ inspection and maintenance face many difficulties and restrictions,such as short skylight time,harsh operating environment,low intelligence of detecting equipment,and low operating efficiency.Therefore,this study relies on multi-functional track inspection equipment to research efficient and intelligent track disease detection methods.The main research contents are as follows:(1)This study proposes a deep learning method using B-scan image recognition of rail defects with an improved YOLO(You Only Look Once)V3 algorithm.Precisely,the developed model can recognize detail fractures,broken base,rail surface defect,bolt hole break,section,joint,bolt hole,weld and automatically position corresponding boxes in B-scan images.By improving the YOLO V3 network structure,data enhancement,parameter adjustment,etc.,the detection accuracy and efficiency of the YOLO V3 algorithm are improved compared with the original model.The final m AP can reach 92.3%.(2)This study proposes a multi-source data fusion algorithm for rail surface defect detection in both camera-based rail inspection images and ultrasound B-scan images.This study builds a camera and ultrasound data fusion(CUFuse)model for rail surface defect detection,including two main networks:multi-source data feature extraction and multi-scale feature fusion networks.The CUFuse model can detect the rail surface defect dataset,and output five rail surface state types,including Light,Moderate,Severe,Normal,and Joint.The results show that the accuracy of the CUFuse model is 96.97%,which can accomplish the task of rail surface defect detection on railway sites.(3)This study proposes a semi-supervised algorithm for detecting foreign objects in ballastless beds based on the improved deep SVDD(Support Vector Data Description)algorithm.By improving the FPN and RPN of Mask R-CNN,the model’s ability to extract the rail and fastener areas in ballastless beds is improved.And by deepening the backbone network of Deep SVDD and blurring the rail and fastener areas,the detection effect of Deep SVDD on foreign objects in ballastless beds is further improved.The results show that the AUC of our improved deep SVDD algorithm is 89.23% and improves the AUC compared to that of the original model by 11.09%,which can complete the task of detecting foreign objects on a ballastless trackbed well. |
