Research On Surface Defect Detection Of High-speed Train Gearbox Based On Binocular Stereo Vision

As the core component of the train transmission system,the gearbox of high-speed train has the function of transmitting power to drive the train forward.Through regular inspection,it is found that the high-speed train gearbox is prone to surface defects,which will seriously affect the driving safety of high-speed trains.In order to adapt to the vigorous development of high-speed railway and ensure the safety of high-speed trains,intelligent and efficient detection technology is urgently needed to replace the traditional manual visual detection method.In this thesis,the high-speed train gearbox with defects is taken as the research object.A set of surface defect detection system for high-speed train gearbox based on binocular stereo vision is designed around the detection and recognition of local defects,surface cracks and surface pit defects on the gearbox surface,spatial positioning,and acquisition of three-dimensional information.The research focuses on the spatial coordinate positioning of defects and the acquisition of three-dimensional information of defects.The specific research work is as follows :(1)Aiming at identifying and locating the surface defects of high-speed train gearbox housing,binocular stereo vision technology is used to perform threedimensional imaging of the housing and obtain its point cloud data to complete threedimensional reconstruction.The three-dimensional imaging principle of binocular camera is studied.The transformation relationship between plane coordinate system and space coordinate system under camera imaging is deduced and analyzed.The internal and external parameters and distortion parameters of the camera required to obtain the spatial coordinates of the defect are clarified,and the two-dimensional checkerboard is used as the calibration plate to calibrate the system to solve the relevant parameters.After the parameters are obtained,the image stereo correction is performed to ensure a good mapping between the plane and the spatial dimension.Finally,based on the semi-global stereo matching technology(SGM),the three-dimensional reconstruction is completed to obtain the spatial coordinates of the defects,and the spatial detection and location of all defects on one side of the box are completed at one time.Experiments show that the method has a good effect on the detection and positioning of local defects and pit defects on the surface of the high-speed train gearbox.(2)In order to avoid the impact of noise points and the accuracy of post-processing results,it is necessary to pre-process the point cloud of high-speed train gearbox imaged by the detection system.Therefore,a method combining statistical filtering and bilateral filtering is proposed to remove redundant noise points.The experiment shows that this method not only simplifies the point cloud data of the high-speed train gearbox,but also does not lose a large number of target point clouds,achieving good point cloud denoising and smoothing effect,and also lays a foundation for the extraction and analysis of the surface defect point cloud of the high-speed train gearbox.(3)Aiming at the specific three-dimensional information acquisition of the surface defects of the gearbox of the high-speed train,the point cloud slicing method is used to cut the box point cloud in the depth direction,and the depth value of the surface defects of the box is obtained by the shape change of the slice spacing and the defect section.According to the difference of RGB data in the depth direction of the point cloud,the difference between the defect and the non-defect area is effectively observed through the experiment.The ROI region of interest extraction processing and the point cloud detection software are used to effectively detect the geometric dimensions of the horizontal and vertical directions on the two-dimensional plane of the defect.The experimental results show that the two methods are effective in obtaining the threedimensional data information and crack length of rib defects and pit defects on the surface of the box,and the minimum error is 1.21 %.