Research On 3D Printing Surface Defect Detection Based On Point Cloud Data And Deep Learning

With the rapid development of 3D printing technology,surface defect detection,as one of the important quality assurance methods,has attracted more and more attention.In the process of 3D printing,detecting the defects of the printing layer on the surface of the part can not only find the topography quality problems in the printing in time,but also avoid the subsequent serious metallurgical defects and quality problems.This research is to detect defects on the surface of 3D printed parts based on the surface point cloud data of printed parts obtained by 3D vision and deep learning technology.According to the 3D printing scene and the features of the parts’ surface,the method of obtaining and processing the point cloud data on the parts’ surface is designed.To solve the problems of noise existence and large amount of data in the originally collected point cloud data,the original point cloud data is denoised and simplified by using statistical filtering algorithm and voxel mesh down-sampling method.In order to segment the point cloud data of the current printing layer from the point cloud data containing the substrate and other redundant information,a point cloud plane segmentation algorithm based on the improved RANSAC is proposed,which is improved in two aspects of sampling point selection and segmentation optimization.The experimental results show that compared with the traditional RANSAC algorithm,the improved RANSAC point cloud plane segmentation algorithm has a great improvement in both time efficiency and accuracy.A nonlinear mapping transformation is established to transform 3D point cloud data into 2D depth images,and an improved defect segmentation algorithm of U-Net is designed.The residual network structure is used to replace the original convolution,the attention mechanism module combining channel and space was embedded,and the weighted crossentropy loss function based on defect proportion and Adam optimization algorithm are used to improve and optimize U-Net.Comparative experiments show that the improved U-Net defect segmentation algorithm can effectively segment the defect area on the twodimensional depth map,and classify the defects,so as to realize the detection and recognition of defects on the surface of 3D printed parts.The improved U-Net defect segmentation algorithm has a better comprehensive performance than the original U-Net and other similar semantic segmentation algorithms.