Research On Ultrasonic Field Guided 3D Generation And Reconstruction Of Rail Defects

Ultrasonic non-destructive testing is one of the main methods in the field of industrial non-destructive testing and is widely used in scenarios such as rail flaw detection.Three-dimensional reconstruction of rail defects can visually restore the 3D size and spatial position of defects inside the rail,improve the efficiency of rail defect detection,and help to detect rail problems early,avoiding accidents and safety issues caused by rail defects.However,current three-dimensional reconstruction mainly relies on visual or laser-based methods.Ultrasound-based 3D reconstruction is only applied in laboratory situations and requires multiple probes or mechanical scanning platforms,making it impossible to achieve non-destructive testing inside the rail.In addition,the large amount of data required for three-dimensional reconstruction makes it difficult to acquire sufficient defect three-dimensional data using wave propagation or B-scan methods in long-distance rail ultrasonic non-destructive testing,lacking an effective rail defect three-dimensional reconstruction scheme.In Thesis,based on the ultrasonic B-scan data of rail defects collected by a multi-channel inspection wheel,a rail defect threedimensional reconstruction method is proposed.The main work is as follows:(1)Thesis proposes a sound field-guided CycleGAN defect cross-section generation model for predicting the cross-sectional morphology of defects corresponding to B-scan data.The method simulates the ultrasonic sound field distribution of the probe in the Bscan plane through simulation,simulates the B-scan process of the inspection wheel,inverts the potential region of the defect,and generates a defect prior probability mask to guide the generation of the defect cross-section.Compared with models trained by other image generation methods,the defect cross-section generation model guided by this mask improves the F-score of the defect cross-section generation results by 0.134-0.313 and reduces the defect aperture error rate by 11.8%-42.2% in testing with different types of defect B-scan data.(2)In order to extract the three-dimensional size of defects from two-dimensional B-scan data,thesis proposes a defect three-dimensional size calculation model based on ultrasonic physical echoes.The model fits the three-dimensional shape of defects as a cylinder,and calculates the size of the defect in the new dimension by extracting the defect’s two-dimensional information from the generated defect cross-section and the original defect B-scan data.We tested the feasibility and error of the model by simulating multiple sets of echo signals of different defects and collecting defect B-scan data in an actual laboratory using Abaqus simulation.(3)Thesis focuses on calculating the three-dimensional size and position information of defects in the railway track environment.Based on the characteristics of defects in different parts of the rail,a KNN classification is used to improve the accuracy of defect size calculation by establishing different ultrasonic echo physical models.In order to restore the actual three-dimensional position of the defect in the rail,a spatial mapping method is proposed to map the defect from the B-scan coordinate of probe wheel system to the rail coordinate system,and the ultrasonic three-dimensional software platform designed in thesis is used to visually restore the defect to the rail model.The threedimensional defects can be reconstructed from B-scan data in different channels of rail probe wheel and test piece through this method,compared with other point cloud-based 3D reconstruction methods,the three-dimensional size and position of the defect in the rail can be restored more accurately.