Research On Key Techniques Of 3D Printing Based On Point Cloud Data

3D-Printing,the groundbreaking technology being greatly developed recently in engineering field,could prototype digital models in any shapes in an additive manufacturing style to fasten product developing cycle and lower cost price.The key focus in 3D printing is the digital models to be prototyped.Other than designing sophisticated models in CAD software,inversely reconstructing virtual models via scanning real objects or scenes provides a quick solution which is quite suitable for structure-complicated things.Benefits from fast-developing of 3D data acquiring devices in last few years,it becomes rather easy to gain three-dimensional information(e.g.point cloud data)of surrounding environments.This thesis is dedicated to investigate the 3D printing related technologies on point cloud data.Starts from the gathered three-dimensional point clouds of realistic scenes or objects,this dissertation explores the reconstructing algorithms of 3D models on multi-view point clouds and the slicing approaches of 3D printing to reproduce physical entities,further to establish an integrated inverse-engineering solution from data capturing to entity manufacturing.Restricted by data acquiring principles,it is required to register multi-frames of point cloud scanned from multi-viewpoints for complete surface representation of an object or environment.To generate printer processible models,it is indispensable to reconstruct topological triangular mesh model from discrete point cloud data.After that,the reconstructed triangular model is sent into 3D printer to determine suitable slicing parameters then fabricate solid entity with minimum volume deviation.This dissertation studies abovementioned three research contents,the main work and contributions are stated as follows:1.In point cloud registration,feature-based pairwise aligning is employed.Due to the inherent data noises and performance limitation of feature operators,there inevitably exists outlier correspondences in obtained feature matches which drags the estimation of rigid transformation.The third chapter proposes a Distance Disparity Matrix(DDM)approach for detecting and eliminating outlier matches upon the maintaining property of the Euclidean distance between feature points before and after rigid motion.Proposed approach prunes over 90% of the outlier correspondences while reserves 95% of inliers in remained correspondence set,and the portion of inliers in total rises from 17% to 70%.Based on which,a Least-Squares Backward method(LSBW)is introduced to estimate more accurate initial transformation matrix in fewer iterations to align two point clouds to a closer position which effectively improves the efficiency of ICP optimization procedure by 30%.The employed thresholds,such as radius of support region for feature operators,inlier/outlier decision criterion,are adaptively determined with respect to the resolution of input point cloud data.The only operation required from user is to control the precision through a scale factor,through which way,the inconvenience and inaccuracy of manually parameter setting are avoided.2.In point cloud mesh reconstruction,implicit surfaces approach produces abnormal surfaces in regions without data and projection-based region growing method results in local holes in generated mesh,which makes the reconstructed model difficult to apply in 3D printing.The fourth chapter of this dissertation presents an enhanced self-repairing surface triangulating approach in region growing principle.Presented approach starts from a seed triangle and then generates triangles in an incremental style according to the visible relationships of neighboring points to the reference point on its tangent plane.It summarizes the triangulating process to seven different cases with respect to the position and status of neighboring points,and details the triangulating steps for each case.Furthermore,it introduces the Single Edge Table(SET)structure for recording and updating the boundaries of triangular mesh during mesh growing.The depth traversing on SET after triangulation detects the local holes which are repaired later by the minimum edge angle prior strategy.Compared with available methods,proposed approach achieves more tight and normal triangular mesh which is applicable to 3D printing.3.In 3D printing,suitable parameters,such as slicing thickness and orientation,impact the volume precision and surface quality of fabricated model.The fifth chapter proposes the volumetric deviation model of 3D printing based on the analysis of staircase effect and its formation in layer-by-layer manufacturing.The relationship between slicing orientation and volumetric deviation is drawn from the model,and the optimal slicing orientation determining problem is mathematically modeled to the least absolute deviation linear regression problem.Originated from this conclusion,a strategy of PCA-approximating then local-optimizing for optimal slicing orientation is developed to achieve minimum fabrication deviation.Proposed volumetric deviation model gives more direct and accurate representation of manufacturing error than existing cusp height or surface roughness model.Presented slicing orientation determining algorithm improves the efficiency and precision by 60% and 2% respectively,and possesses more adaptability comparing to other orientation traverse-based algorithms.This dissertation summarizes the pipeline of point cloud based 3D printing to three main parts: point cloud registration,surface triangular reconstruction,model slicing,and declares the research content related to each part.Then,conducts targeted study to each part and initially establishes the inverse engineering solution from point cloud acquiring to solid fabrication.