Computational Modeling Sciences Department

Why Is Hex Meshing So Hard?

What is the Cubit Team Doing About it?

Cubit User’s Meeting

June 27, 2007

Matt Staten
 Tet Meshing Vs. Hex Meshing
Computational Modeling Sciences Department

Tet Meshing Hex Meshing

1. Fully Automated 1. Partially Automated, some Manual
2. Generate millions of elements in 2. Generate millions of elements in
minutes/seconds days/weeks/months
3. Adequate for some analysis 3. Preferred by most analysts for
4. Inaccurate for other Analysis solution quality
5. “The only people that use tets are
those that don’t care about their
answers.” – Sandia Analyst

Matt Staten
 Design Through Analysis
Analysts Man-Hours Spent (Not CPU Time)
Computational Modeling Sciences Department

8% 32%
21%

6% 6%
14%

4% 4% 5%

73%
10 Step of time is spent
Design in
through
either
Analysis meshing
Process or geometry
1%
manipulation
Iteration loops notfor meshing.
shown for clarity
Matt Staten
 Why is Hex Meshing Harder
Than Tet Meshing?
Computational Modeling Sciences Department

• More complex element connectivity

• Must consider warp

• Larger minimum angles required

• Propagation

Matt Staten
 More Complex Element Shape
Computational Modeling Sciences Department

Hex Element Tet Element

8 nodes 4 nodes
12 edges 6 edges
6 faces 4 faces

Matt Staten
 More Complex Element Shape
Computational Modeling Sciences Department

• Most Meshing performed with a bottoms-up

approach
– 1st mesh all vertices with nodes
– 2nd mesh all curves with mesh edges
– 3rd mesh all surfaces with surface elements
– 4th mesh volumes

Matt Staten
 Advancing Front Element Creation
Computational Modeling Sciences Department

Only a single node is required to finish the tet element

Matt Staten
 Advancing Front Element Creation
Computational Modeling Sciences Department

To Complete the hex, 4 well positioned nodes are required,

which may not be readily available.
Matt Staten
 Advancing Front Element Creation
Computational Modeling Sciences Department

To Complete the hex, 4 well positioned nodes are required,

which may not be readily available.
Matt Staten
Hex Elements Must Consider Warp
Computational Modeling Sciences Department

Matt Staten
 Hexahedra Require Larger
Minimum Angles
Computational Modeling Sciences Department

Shape: 0.41 Shape: 0.28

Cond#: 3.13 Cond#: 5.5

Small angles can

be filled with high
quality tets much
easier than high
quality hexahedra.

Matt Staten
 Hex Mesh Propagation
Computational Modeling Sciences Department

• Local Changes to tri/tet meshes can easily be

performed:
– Insertion of a single node
– Removal of a single node
– Swap a single face

Matt Staten
 Hex Mesh Propagation
Computational Modeling Sciences Department

• Local Changes to quad/hex meshes are difficult

to keep local without introducing a hybrid mesh

• Changes Propagate in
order to maintain
all/quad hex mesh.

Matt Staten
 Hex Mesh Propagation
Computational Modeling Sciences Department

• Local Changes to quad/hex meshes are difficult

to keep local without introducing a hybrid mesh

• Changes Propagate in
order to maintain
all/quad hex mesh.

Matt Staten
Quad/Hex Mesh Dual
Computational Modeling Sciences Department

Matt Staten
 Quad/Hex Mesh Dual
Computational Modeling Sciences Department

Chords define
continual chain
of connected
quadrilaterals
which force
propagation

Matt Staten
 Quad/Hex Mesh Dual
Computational Modeling Sciences Department

Chords define
continual chain
of connected
quadrilaterals
which force
propagation

Matt Staten
 Quad/Hex Mesh Dual
Computational Modeling Sciences Department

Chords define
continual chain
of connected
quadrilaterals
which force
propagation

Matt Staten
 Quad/Hex Mesh Dual
Computational Modeling Sciences Department

In 3D, hexahedra combine to form Dual Sheets, the 3D analog

of Quad Chords. A local change to a hex mesh must be
propagated through the sheets in order to avoid introducing
hybrid element types.

Matt Staten
 Current Hex-Meshing State of the Art
Pave-and-Sweep
Computational Modeling Sciences Department

Sweeping a source quad mesh along a topologically constant cross section

Source
Sweep Direction

Target

Also handles multiple sources and targets. No restriction on planar/perpendicular

source/targets or linear sweep path. Limited to 2½D Topologies
More Complex solids meshed with Partition-Pave-and-Sweep

Decomposing general geometries into sweepable partitions is time and

expertise intensive … an “art” form, requiring tedium,Slide
creativity and Mattexperience.
2 of 11 – Matt Staten
Staten
 What is Cubit Doing about it?
Computational Modeling Sciences Department

• Two-Approaches being pursued:

– Improved Geometry tools to make Pave-And-Sweep
Easier
• Virtual operators
• ITEM wizard
• Power Tools

• Continued research on completely automated

Unstructured Hex Mesh Generation
– CSRF funded projects
– DSP participants research
– Graduate Student Research

Matt Staten
 CSRF Funded Projects
Computational Modeling Sciences Department

• CSRF has funded research in the Cubit team for

the past decade.

• Unconstrained Plastering is current research

focus. Some simple to medium complex models
can be meshed automatically.

Matt Staten
 CSRF – Unconstrained Plastering Research
Computational Modeling Sciences Department

Plastering – 3D
(1992)
•Initial quad
boundary defined
•Advancing Front
Paving - 2D (1990) by discrete
•Initial boundary intervals defined elements
•Advancing front by discrete elements •Element-by-
•Element-by-element placement element placement
•Interior voids easily resolved •Cannot resolve
•Robustly quad meshes nearly any surface interior voids
•Creates skewed elements in thin surfaces

Unconstrained Plastering – 3D (2005)

Unconstrained Paving – 2D (2005)
•No initial boundary intervals •No initial boundary quad mesh
•Advancing front by discrete geometric layers •Advancing front by discrete geometric layers
•Layer-by-layer placement •Sheet-by-sheet placement
•Under active research and development •Under active research and development
Matt Staten
Slide 4 of 11 – Matt Staten
 DSP Participant Research
Computational Modeling Sciences Department

• Jason Shepherd – University of Utah (2004-2007)

– Theory of Sheet Based Hex mesh generation
• Primary Sheets
• Secondary Sheets

– Manual tools allowing for unstructured hex mesh

generation on non-sweepable models with little (if any)
geometry modification

– Automation of methods into Cubit currently being

considered

Matt Staten
 DSP Participant Research
Computational Modeling Sciences Department

• Jason Shepherd – University of Utah (2004-2007)

– Example meshes

Matt Staten
 DSP Participant Research
Computational Modeling Sciences Department

• Matt Staten – Carnegie Mellon University (2007-2010)

– Theory and Application of Sheet Based Hex mesh generation
• Continued Unconstrained Plastering
• Advanced Pillowing Techniques
• Mesh Matching Techniques

• Mike Borden – University of Texas at Austin (2007-2010)

– IsoGeometric Analysis as an alternative to traditional Hex-
based FEA analysis.

Matt Staten
 Graduate Student Research
Computational Modeling Sciences Department

• Brigham Young University (1994-Present)

– Recent Successes
• Improved Local Hexahedral Refinement
• Localized Hexahedral Coarsening
• Improved node placement in Pave-and-Sweep

• Carnegie Mellon University (2002-Present)

– Recent Successes
• Geometry based sizing functions for tet meshing

Matt Staten