Defect Identification Of Locomotive Bearing Caps Based On 3D Point Cloud

There are many small parts in the key parts of the locomotive,for example,bolts on the bearing cover of locomotive bottom.If only rely on manual detection of defects,then the workload of workers will be very large,people will easily become fatigued,leading to missed detection,false detection.It is easy to cause serious consequences when the inspection is missed.In recent years,in railway applications,automatic defect detection systems based on two-dimensional images have matured,but there are still some insurmountable problems.Compared with the three-dimensional image,the two-dimensional image cannot directly obtain the depth information of the object.The three-dimensional data can not only record the shape of the surface of the object and geometric features,but also obtain the spatial coordinates and other information,but there is no natural topological connection between the three-dimensional point clouds.It is impossible to accurately depict the outline of the target object.Then,it is important that how to extract the features of the three-dimensional point cloud data,form a more actual contour of the composite body,and realize the defect recognition of the three-dimensional image.Therefore,it is of great practical significance to detect and identify the fault of the locomotive under the vehicle more accurately and reliably.In order to achieve the above questions,this thesis makes the following attempts:1.This thesis collects 3D point cloud experimental data by laser line sweeping method.This method can quickly obtain 3D point cloud image with good effect.However,due to many external interference information in the experimental scene,the actual data collected in this thesis contains many noise points and the presence of noise points will affect the feature extraction and post-processing of the point cloud.Therefore,the combination of statistical filtering and bilateral filtering is used in the thesis to achieve better filtering effect.2.The thesis selects three commonly used 3D feature point extraction algorithms for experimental comparison,compares the running time and extraction effect,extracts the optimal extraction algorithm,and combines it with clustering analysis to extract obtaining the line of 3D point cloud data,this method has certain improvement and optimization in time and effect compared with the traditional feature line extraction algorithm;3.For the common defects of locomotive underbody bolts,the corresponding solutions are proposed respectively.The feature descriptors are used for solving the bolt missing problem to achieve the matching effect and the missing position is determined.The surface defects of the bolts need to pass.The contour of the three-dimensional image of the surface of the bolt is extracted for detection purposes,but due to various interferences in the actual data,such as external light,the integrity of the collected data is affected,and there is a problem of missing at the corner of the surface of the object.Therefore,this thesis will repair the corner.The method is combined with the current method of feature line extraction,so that the experimental results can be more suitable for the surface contour of the actual object;the bolts on the bottom of the locomotive will also have problems of convexity and bending.For such problems,the point cloud data extracted from the surface of the bolt and the point cloud data collected on the surface of the bearing cover are plane-fitted,so that the two planes exist in the same coordinate system.After fitting,the detection is realized by judging the relationship between the two planes.