Fault Detection Method For Bolts In Center Sheath Of Train Bottom Based On 2D-3D Image Information Fusion

The rapid development of urban rail transit in my country has brought great convenience to people’s travel,but at the same time it also brings challenges to the safety of train operation.With the continuous increase of train speed and the continuous increase of load capacity,the maintenance of trains has become more and more important.During the long-term operation of the train,due to problems such as collision and aging,surface cracks,abrasion,looseness,missing and other faults will appear at the bolts of the center sheath of the train bottom,which may easily cause major traffic accidents,causing casualties and property losses.Therefore,the sheath was studied.The bolt failure detection method and technology can detect the failure of the sheath bolt in time through online detection and avoid the occurrence of train accidents.Its research has great application value.The current two-dimensional image-based detection methods for key components of the train bottom cannot measure the longitudinal displacement of the loose bolts because they do not contain depth information.Failures such as loose bolts cannot be detected in time.Compared with a simple two-dimensional image,the point cloud data method based on visual inspection can obtain more detection target parameters,and can better reflect the three-dimensional characteristics of the object.It has become an effective method for bolt fault online detection.For the detection of the center sheath bolt failure,the bolt loss failure can be detected from the two-dimensional image,but the bolt looseness failure can only be judged through the point cloud data.Based on this,this paper proposes a method for bolt fault detection based on 2D-3D image information fusion,and conducts research.The main work of this paper includes:1.The focus is on the target positioning method under the complex background of2 D images.Two target detections based on HOG+SVM and deep learning SSD are used to realize the positioning of the bolt area of the center sheath under the vehicle.The sample making of the two methods,the construction of the environment and the training of the model are studied.Finally,the test data set was used to locate and detect the bolt area of the center sheath of the vehicle,and the performance of the two detection models was compared and analyzed.The final result showed that the deep learning method has higher recognition accuracy and faster running speed.The detection time for a picture is about 0.2s.2.The preprocessing method of 3D data is studied.Firstly,Kd-tree is used to establish the topological structure of discrete point cloud data,and then the bolt point cloud is down-sampled by the voxel grid method.Finally,the statistical method is used to calculate the variance Separate the group points from the mean rejection part.Finally,after preprocessing,the data volume of the point cloud is greatly reduced while keeping the original geometric features unchanged,paving the way for the following processing.3.The focus is on the recognition method of bolt point cloud target.First,the preprocessed point cloud needs to be segmented.The traditional European clustering segmentation has very poor segmentation effect when there are adhesion points.This article uses the region growing segmentation method to add The constraint conditions of normal and curvature complete the segmentation of bolt point cloud.The shape factor of the segmented point cloud is calculated,the non-target point cloud is eliminated,and the bolt identification and positioning are completed by calculating the geometric characteristics of the bolt.The proposed method for detecting the failure of the center sheath bolt on the undercarriage can provide a basis for the automation of the detection of the undercarriage of the train,and the algorithm is encapsulated in a dynamic link library in the 360-degree detection system of a metro vehicle for testing.The test results show that: for metro vehicles The failure identification of the loss and loosening of the bolts of the center sheath of the vehicle bottom reaches expectations and can be used for actual measurement.