Panel Discussion Notes

---

October 13, 1999

Format:

Five (5) panelists, listed below, gave summaries of the Birds of a Feather (BOF) discussions held on Monday, which they lead. They also gave their own views on their subject area. The moderator summarized these views, then opened the floor for questions and discussion. The discussion and questions portion was lively, without much need for the moderator to "coerce" participation from the audience.

Panelists:

• Mesh quality: Lori Freitag, Argonne National Lab
• Geometry: Cecil Armstrong, Queen,s University Belfast
• Tri/tet algorithms: David Marcum, Mississippi State Univ.
• Structured meshing: John Steinbrenner, Pointwise Inc.
• Hex meshing: David White, Sandia National Laboratories

Moderator: Tim Tautges, Sandia National Laboratories

Panelist summaries:

TRI/TET ALGORITHMS

(David Marcum, Mississippi State U.)

• Is tet meshing solved? Yes & no
• generally know how to generate a tet mesh
• work is needed on application-specific capabilities
• Higher order elements
• Need to account for actual geometry during mesh generation
• What is "ideal" element shape? Similar problem with h-refinement

Geometry/surface mesh problems

• nearby surfs & small included angles can produce a "poor" volume mesh even for good surface meshes
• thin sections / shell meshes have similar problems

Dynamic meshes

• moving components
• small scale ok; large scale requires tighter integration between solvers & meshers

Geometry definition

• may be needed for adaptation, dynamic, high-order
• generic cad API needed

STRUCTURED GRID GENERATION ISSUES

(John Steinbrenner, Pointwise)

• bad geometry
• bad geometry navigable, but still a problem
• grid quality
• pde-based & algebraic methods work well
• adaptation sometimes difficult
• block topology construction
• templates from library (effective in niche markets)
• semi-automatic (schematic, extrusion/sweep)
  • requires user to invent baseline topology
  • requires robust grid generation tools for difficult cases
• fully automatic block construction
• currently non-existent?
• Use current unstructured algorithms to form blocks
• New tools required
• [Problems there with number of blocks, currently]
• continued work on automatic block construction
• unstructured techniques may help
• ** - suggestion: new BOF session on hybrid grids

HEX MESHING

(David White, SNL)

• Technology advances: who will "win"?
• traditional approaches
  • pave/sweep, automation, multi-sweep, grafting
• Fully automatic approaches
  • Medials, whisker weaving
  • Mid-point subdivision
  • Inside-out
  • Feature-based decomposition
• Who will win? User, with combination of tools
• Incrementally getting closer

What to look for?

• better analysis methods developed
• can we use tets/mixed elements (e.g. 8-node tet?)
• improved contact/mpc?
  • Dohrmann/Key/Heinstein IJNME paper
  • Example: MC4217 much easier with non-conformance

In summary:

• traditional approaches becoming more automated
• learning from automatic approaches
• global problem
• methods to improve traditional approaches
• mixed approaches promising
• Will we ever satisfy the user?
• Do they want unstructured mesh, or structured with few irregularities?
• forum for analysis community to come together and tell us about quality requirements?
• ** - invite analysis community to next meshing roundtable?

DISCUSSION ISSUES

Mesh quality: need a forum for quality metrics from analysis community

• should invite analysis community to next IMR
• shameless plug: VERDE mesh quality evaluation tool from Sandia National Laboratories

Adaptive remeshing may not be possible because of size, or because initial geometry wasn,t captured well enough

Also, haven't thought about modeling errors, e.g. fidelity to original geometry

What are "acceptable" quality metric values?

• need a small list of shape metrics on which to standardize
• need a collection point for info on how shape metrics affect analysis results

Question: can mesh generators help geometry modelers?

EDITORIAL SUMMARY

• Mesh quality clearly a great concern
• Need information from analysis community on what,s acceptable
• Need small, standard set of metrics which all pre-processing codes implement
• There is potential for VERDE to serve a role here
• ACTION: invite analysis community to next roundtable (possible keynote speech?)
• ACTION: need collection point for quality information
• Making progress on both tet and hex meshing, but plenty of work still to be done
• Do users want unstructured hexes, or structured with few irregularities?
• Can unstructured hex meshing techniques help with block definition?
• ACTION: need BOF session on mixed element meshing