Font Size: a A A

Design Of Rail Surface Defect Detection System Based On Neural Network

Posted on:2021-09-17Degree:MasterType:Thesis
Country:ChinaCandidate:Q Y BaiFull Text:PDF
GTID:2492306332970269Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
Aiming at the problems of slow detection speed,high detection cost,low detection accuracy and poor system adaptability in current rail surface inspection,this paper proposes a multi model rail detection algorithm based on parallel structure,and constructs a complete rail surface defect recognition and analysis system to realize the digital management of rail defects.The main work of this paper is as follows.(1)The model of rail surface defect pre detection based on WGAN is constructed.The model can be used to quickly pre-screen the suspected rail surface faults and effectively solve the problem of low detection accuracy caused by the imbalance of positive and negative samples in the training data set.(2)In this paper,a parallel algorithm architecture is established.MGMM model can be used for the coarse segmentation of rail defect image,and CASAE semantic segmentation model can be used for fine segmentation.The parallel processing structure can be used to fuse the two models reasonably,and the segmentation results of different scales are comprehensively utilized,which greatly improves the accuracy and robustness of the defect segmentation algorithm.(3)An intelligent rail surface defect detection system is constructed.The overall hardware framework of the system is constructed by using locomotive operation monitoring device(LKJ),high-resolution linear scanning camera and large computing power workstation.At the same time,parallel multi-mode detection algorithm,database processing technology and UI interface design technology are comprehensively used to realize automatic detection,visual operation and digital operation of rail surface defects Management.The experimental results show that:the detection accuracy of the system reaches 96.98%,the average detection speed reaches 0.0897s/frame,and the detection performance is obviously better than other detection technologies.Moreover,the system constructed in this paper can be adapted to different lights,acquisition angle,background and acquisition equipment,and has strong environmental robustness.At the same time,the system is easy to operate and easy to use,which can replace China to a certain extent The manual detection method is mainly used to realize the automatic detection and intelligent management of rail defects.
Keywords/Search Tags:Rail surface defect, Neural network, Automatic detection, Parallel architecture
PDF Full Text Request
Related items