All-Hex Meshing using Singularity-Restricted Field Yufei Li1, Yang Liu2, Weiwei Xu2, Wenping Wang1, Baining Guo2 1.
Download ReportTranscript All-Hex Meshing using Singularity-Restricted Field Yufei Li1, Yang Liu2, Weiwei Xu2, Wenping Wang1, Baining Guo2 1.
All-Hex Meshing using Singularity-Restricted Field Yufei Li1, Yang Liu2, Weiwei Xu2, Wenping Wang1, Baining Guo2 1. The University of Hong Kong 2. Microsoft Research Asia 1/28 Motivation • All-hex mesh – A 3D volume tessellated entirely by hexahedron elements. All-hex mesh Tetrahedral mesh • Why alll-hex mesh? – Reduced number of elements. – Improved speed and accuracy of physical simulations [Shimada 2006; Shepherd and Johnson 2008]. 2/28 Motivation • Issues – Highly constrained connectivity. – Require much user interaction. Semi-automatic User interaction • Industrial practice – Multiple sweeping [Shepherd et al. 2000]; – Paving and plastering [Staten et al. 2005]; – … ANSYS software 3/28 Motivation • Quality criteria for all-hex mesh – Boundary conformity – Feature alignment – Low distortion Feature Alignment Boundary Conformity Low Distortion All-hex mesh Goal: automatically generate all-hex meshes with high-quality 4/28 Existing methods: all-hex meshing based on volume parameterization guided by 3D frame field Input volume (tetrahedral mesh) 3D frame field (inside the volume) Volume parameterization (guided by 3D frame field) All-hex mesh 5/28 Existing methods: all-hex meshing based on volume parameterization guided by 3D frame field Input volume (tetrahedral mesh) 3D frame field (inside the volume) [Huang et al. 2011] Volume parameterization (guided by 3D frame field) All-hex mesh Hex-dominant mesh 6/28 Existing methods: all-hex meshing based on volume parameterization guided by 3D frame field Input volume (tetrahedral mesh) 3D frame field (inside the volume) Volume parameterization (guided by 3D frame field) All-hex mesh Manually designed meta-mesh CubeCover [Nieser et al. 2011] 7/28 Our approach: all-hex meshing framework based on singularity-restricted field (SRF). Key condition Input volume (tetrahedral mesh) 3D frame field SRF 3D frame field (singularity-restricted field) Volume parameterization (guided by 3D frame field) All-hex mesh Major contribution Automatic SRF conversion SRF Hex-dominant mesh 8/28 Basics of 3D frame field • Discrete setting [Nieser et al. 2011] – 3D frame: 24 permutations. Chiral Cubical Symmetry Group (24 matrices) – Discrete 3D frame field for input tet mesh: a constant 3D frame for each tet. 9/28 Basics of 3D frame field • A pair of arbitrary frames – Difference: a general rotation. – Matching: the permutation that best matches the two frames (24 choices). Fs Matching Ft 10/28 Basics of 3D frame field • An interior edge – How the frames rotate around it? • • Identity matrix: regular edge. Non-identity matrix: singular edge (23 types). Proposition: Any singular edge does not end inside the volume. Singular graph 11/28 Singularity-restricted field (SRF) • Definition of SRF – A 3D frame field is an SRF if all of its edge types fall into the following subset of rotations: 3D frame field (24 edge types) SRF (10 edge types) – Ru, Rv, Rw represent the 90 degree rotations around u-, v-, wcoordinate axes, respectively. SRF is necessary for inducing a valid all-hex structure [Nieser et al. 2011] 12/28 Converting general 3D frame field to singularity-restricted field (SRF) 3D frame field (24 edge types ) SRF (10 edge types Eliminate the improper singular edges (14 types) • Operations for SRF conversion: – Matching adjustment: tentatively adjust the matching for any triangular face, and check if improper singular edges could be eliminated. Geometric – Improper singular edge collapse. operation Necessary for all-hex meshing ) 13/28 Converting general 3D frame field to singularity-restricted field (SRF) • Improper singular edge collapse (topological operation) – Collapse improper singular edges without introducing new ones; Key – Preserve the validity of mesh topology during the collapsing process. s1 s2 Our algorithm could eliminate all the improper singular edges, except two extreme t e cases that do not happen in practice. (See proof in the paper) Collapse improper singular edge e 14/28 SRF Conversion Input frame field Matching adjustment could also smooth the singular graph. Output SRF Improper singular edges (in red) are collapsed. 15/28 A high-quality all-hex meshing framework based on singularity-restricted field (SRF). Improvement Adaptive rounding Input volume (tetrahedral mesh) SRF (singularity-restricted field) Volume parameterization (guided by SRF) Improved CubeCover [Nieser et al. 2011] to solve this mixed-integer problem. Gradient field Given SRF Input domain Parameter domain All-hex mesh 16/28 Obstacle 1: degenerate element Input domain Parameter domain Zero volume Element Degeneration Fail to trace iso-lines Missing hex elements! Singular edge combination on triangular face. b a Degenerate elements (in red) Why degenerate? c 17/28 Handling degenerate elements See the paper Input volume (tetrahedral mesh) SRF (singularity-restricted field) All degeneration cases for any triangular face. Topological operations Preprocessing Volume parameterization (guided by SRF) All-hex mesh All the degenerate elements (in red) are removed 18/28 Obstacle 2: flipped element Erroneous topology of iso-curve network Fix the topology Restore a complete all-hex mesh Negative volume Input domain Parameter domain Flipped Elements 19/28 Comparison with [Huang et al. 2011] Free of improper singular edges. Red edges are improper edges. More smooth SRF by our method Frame field by [Huang et al. 2011] 20/28 Optimized SRF with different frame field initializations SRF All-hex mesh Singular structure 21/28 Comparison with CubeCover [Nieser et al. 2011] Cube-like elements Distorted elements Our method: J_min = 0.609 J_min CubeCover: J_min = 0.073 [-1,1]: the minimal scaled Jacobian of hexes, bigger is better. 22/28 Comparison with PolyCube [Gregson et al. 2011] Distortion Boundary conformity Our method: J_min = 0.351 Poor quality due to PolyCube nature PolyCube: J_min = 0.196 23/28 More results by our method J_min = 0.185 Feature alignment Cube-like elements J_min = 0.729 J_min = 0.599 24/28 A high-quality all-hex meshing framework based on singularity-restricted field (SRF). Input volume (tetrahedral mesh) SRF (singularity-restricted field) Volume parameterization (guided by SRF) All-hex mesh Contributions Effective smoothness of 3D frame fields Effective operations for SRF conversion Key ingredient Improved volume parameterization by handling degenerate& flipped elements 25/28 Limitations & Future work • No theoretical guarantee that SRF always leads to a valid all-hex structure. Necessary condition SRF Singularity-restricted field All-hex structure “Almost” but NOT sufficient Open problem: what is the sufficient condition for all-hex structures? 26/28 Limitations & Future work • No theoretical guarantee that SRF always leads to a valid all-hex structure. • Singularity mis-alignment: no global control of singularities. Singularity mis-alignment 27/28 Limitations & Future work • No theoretical guarantee that SRF always leads to a valid all-hex structure. • Singularity mis-alignment: no global control of singularities. • CANNOT guarantee a degeneracy-free or flip-free volume parameterization. Shortcoming shared by CubeCover [Nieser et al. 2011]. 28/28 Acknowledgements • • • Reviewers for constructive comments. Ulrich Reitebuch, Jin Huang for providing comparison data. Funding agencies: The National Basic Research Program of China (2011CB302400), the Research Grant Council of Hong Kong (718209, 718010, and 718311) Backup slides Boundary-aligned 3D frame field generation • Difference of two frames Fs and Ft • Smoothness: closeness from to (24 types of permutations) • Optimization – Solved by the L-BFGS method [Liu and Nocedal 1989] Frame field initialization • Boundary tets – Smooth boundary cross field + surface normals • Interior tets – Propagation from boundary tets. – Assigned to be the same as the one of its nearest boundary tet. Frame field guiding User intention Small features, not enough tets Robustness of SRF conversion • Test on a random frame field – Initialization: principal-dominant cross-field on the boundary + random frames inside. – Without optimization. • • • • 32320 tets 6825 vertices 775 proper singular edges 61 improper singular edges SRF conversion • • • • 31930 tets 6766 vertices 753 proper singular edges 0 improper singular edges SRF-Guided Volume Parametrization • Definition – The integer grids in induce a hex tessellation of the input volume • Computation Gradient field Integer variables: – Boundary faces – Vertices on the singular graph – Adjacent face gaps Given SRF . CANNOT guarantee degenerationfree volume parameterization • • The triangle has three regular edges (does not belong to the degeneration case in our analysis). Vertices a, b and c are on the singular graph. b a c The triangle might still degenerate due to the integer rounding on vertices a, b and c SRF is not sufficient What is the sufficient condition for all-hex structures? Triangular face The topology of the singular graph prohibits the existence of all-hex structures. CANNOT guarantee flip-free volume parameterization! The singular graph consists of two spiral and close curves inside a torus volume. Fail to retrieve an all-hex mesh. The tets mapped to negative volumes in the parameterization are rendered. Statistics