Fastener Looseness Recognition Of Medium-low Speed Maglev Contact Rail Based On Instance Segmentation

Medium-low speed maglev can effectively alleviate the pressure of urban traffic,which has three main functions of suspension,guidance and traction.Based on the thorough functions,mechanical contact between train and track can be avoided to reduce the maintenance cost.Meanwhile,ensuring current-receiving quality of medium-low speed maglev train is the prerequisite to maintain the above functions.Generally,Medium-low speed maglev train collects current from collector shoe by contact with the contact rails.However,after a long-term interaction,fasteners of the contact rail are prone to loosening,which affects the normal flow between collector shoe and rails.Therefore,the state of the contact rail fastener needs to be effectively recognized.Regrettably,the serving recognition methods are mainly manual inspections and viewing images by technicians,which have problem on workload,efficiency and accuracy.So,automatically recognizing the looseness of the contact rail fasteners is highly imperative.With the images taken by the medium-low speed maglev contact rail imaging device in Changsha,in this thesis,we propose a method for fastener looseness recognition based on instance segmentation,and the design and development of corresponding recognition system is completed.The main contributions of this thesis is summarized as follows:Firstly,considering the overly small size of the contact rail fasteners,the areas in which they are distributed should be extracted from the figure.In this thesis,the existing detection algorithms are studied in depth,and the sample databases for various algorithms are established by using the typical contact rail images.According to the characteristics of the initial localization task,the regression-based detection network,YOLO v2,is selected to realize the localization of the fasteners distribution areas of the base and the connecting plate.Secondly,an instance segmentation network,Mask R-CNN,is used to segment the nuts and screw of the base bolt,the connecting plate and the head of the connecting plate screw in the two localized regions.Based on the results of segmentation,the corresponding recognition criteria is designed for different fasteners.For the base bolt,the relative position relationship between the mask of nut and the mask of screw is used as the basis of looseness recognition;for the connecting plate screw,the distance between the mask of screw head and the mask of connecting plate is used as the looseness criterion to realize the looseness recognition of fasteners.At last,based on PyQt5,Tensorflow and OpenCV,the algorithm and user interface of the recognition system of fasteners looseness for medium-low speed maglev are realized.The practicability of the algorithm and the reliability of the system are verified through the test on the images from the contact rail imaging device at different periods.