Prism NoPyramids

8/10/2019 Prism NoPyramids

1/21

ICEM CFD Tetra/Prism For CFD++ or Fluent

2007 ANSYS, Inc. All rights reserved. 1 ANSYS, Inc. Proprietary

Simon PereiraANSYS ICEM CFDProduct Manager


2/21

Large model Strategy

If your models are large and generating the mesh takestime.

Break out a chunk (most difficult chunk) for optimizing setup andsettings

With only a few million cells, your turn around time will be quicker Apply lessons learned to the larger models


,apply that to other models within that class.


3/21

Geometry Setup

Prism starts with geometry For geometry patch independence, delete unnecessary

curves and points. Over constrained patches (too many curves) will make it

difficult for the smoother to give you a smooth mesh.


an a so set s zes on curves or ens ty reg ons Finer mesh in difficult areas gives more freedom to the smoother

Some users set width parameter on surfaces to keep the mesh sizefrom coarsening for a number of layers beyond the surface. This puts

more nodes in the volume and makes it easier for prism to connect up. Dont set width to more than 2 or 3 on surfaces your plan to prism. Yes, it will increase your Octree tetra count It only helps with octree tetra mesh (use density regions for a more

general solution)


4/21

Prism Preparation

It is difficult to smooth prisms, so we want to start with the best possible quality. Preparation isthe key.

Start with good tetra or tri-surface mesh Check aspect ratios / quality

Check and fix all diagnostics Single edges, Non-manifold vertices, or Duplicate elements will crash the prism mesher

Laplace smooth the surface tris It may be a good idea to use one of the bottom up tetra methods (Advancing front, Tgrid, or

Delaunay) to replace your octree tetras.


You could reduce the tetra element count by 30 to 50% and you will get better mesh transitions. Be careful

Laplace smooth the surface mesh before filling Make sure you have no single edges, non-manifold verts, etc. which could cause trouble for the bottom up tetra

methods (and prism for similar reasons). If you set surface width for volume refinement, it is not directly respected by the bottom up methods

Visually scan the surface mesh Look for kinks or sharp tent-like structures in the mesh

Diagnostics may not reveal all surface discrepancies

Make sure part associations are correct Look for a few elements in one part scattered among another part

This can sometimes happen and then when prism is generated it is tough to organize the transitions to

the surrounding tetras between scattered prism columns.


5/21

Mesh Smoothing

Use the Laplacian based surface smoother

pLaplacian smootherrp= Avg (r i)

1

2i


Some users like to float the tets and smooth the tris, then turn offLaplace and freeze the tris and smooth the tets. Other users just delete the Octree tets, smooth the tris and then fill

with a bottom up tetra method.

Laplace moves the point towards an ideal location defined as theaverage of the surrounding nodes This procedure significantly improves the mesh Has the effect of averaging the surface mesh transitions and

triangles become more equilateral


6/21

Tetra Fill

Octree tetra has sharp transitions and uses about 50%more nodes on these models than Delaunay Tetra

Advancing Front, TGRID Delaunay and TGRID AFT are alsoavailable Advancing front methods give smoother mesh, but that

means more elements.


AFT is an advancing front option that can be set for theTGRID mesher My tests have shown it is roughly 6 times faster than the other

advancing front algorithm

12.1 has a Delaunay with TGLib-AFT option (best) We recommend doing the fill before Prism

Tetras enable collision detection

Bottom up methods may be more difficult after prism


7/21

ICEM Tetra Meshing

Octree Delaunay


Expansion Ratio=1.2

Advancing Front Mix and Match

with Hexa Core or

Prism


8/21

Tetra Fill

Under global settins, you can change the growth ratioand adjust other properties for the fill.

OCTREE Mesh


Delaunay Mesh


9/21

Prism Requirements

The settings are determined by yourrequirements.

In this case Fluent or CFD++ You require All Prisms (Few or no

Pyramids)


Can handle sharp high angles andrelatively poor quality to achieve this You require smooth transitions

Cell to cell and between prisms andtetras

You need to set the initial height. Y+


10/21


11/21

Pyramids due to intersection

If the height is floating, you must refinethe mesh to control the prism height andprevent intersection from being resolvedwith pyramids Before refinement, prismlayers intersected and

pyramids were used

after refinement there is lots


Auto reduction (discussedlater) can also preventcollision without pyramids


12/21

Prism Settings

Min Prism quality = 0.000001 Default is 0.01; If you set this closer

to zero, prism will permit lowerquality prisms without resorting to

r mids.


We set this low enough to get nopyramids and CFD++ can still handlethe worst prisms.


13/21

Prism Settings

On complex models, prism orthogonality and thequality of the inner tri surface are at odds. Set this below 0.5 to favor tetra quality over

orthogonality.

Ortho wei ht = 0.1 Ortho weight = 0.5 Ortho weight = 0.9


Poor quality tets

Directional smoothing may also

reduce orthogonality as wasnecessary in this example


14/21

Prism Settings

Fillet ratio works together with Ortho weight toimprove tetra quality in convex regions. Larger fillet (number is a ratio to total prism height)

means more inner surface area for the tops of theprism columns and their adjacent tetras


Fillet Ratio = 0.1

Fillet Ratio = 0.5

Fillet Ratio = 1.0


15/21

Prism Settings

Max Prism Angle keeps the prism layers connectedaround tight bends. Set this at 180 to prevent pyramids


Max Prism Angle = 140 Max Prism Angle = 180


16/21

Prism Settings

Smoothing Options Set surface smoothing steps to 0; you should

smooth before running prism Volume/directional Smoothing is the most time

consuming part of Prism. If you turn these off, Prismgenerates very quickly


weight to be less than 1. For complex models withlots of tight spots, smoothing steps are important

Without Directional

smoothing steps, prisms arevery orthogonal, but theywill still bend for fillet ratio.

Directional collisions resultin pyramids


17/21

Prism Settings

Advanced prism params, Auto Reduction Proximity can cause prism to stop and give

pyramids instead. This option causes the prisms to squeeze down This is fixed with redistribute prisms (automatic at 12.1)


yram s

Reduced (Squished) prismsRedistributed andNo Pyramids


18/21

Redistribute Prism

Edit Mesh, Redistribute Prism Reset the initial prism height. Solves the squished mesh from auto reduction Allows the user to adjust for the desired y+ Roughly maintains the smooth volume transition

Auto reduced Prisms(Squished)

Before


Issues If the initial height is large, the min ratio will be 1

(uniform) distribution.

11.0 could not redistribute layers containingpyramids. This is fixed in 12.0

12.0 changed the way Prisms were calculated to givethe most precise column by column initial height

12.0 could not redistribute hexas (inflated quads), orimported mesh, these have both been fixed for 12.1

.

Prisms fixed Initial height set

After


19/21

Smooth Prism

Prism is difficult to smooth which is why we try to startwith as good a surface mesh as possible.

Once the mesh is done, we recommend smoothingwith the prisms (Penta_6) frozen. This will improve thetetra mesh. Do this until you dont see anyimprovements.


also, but use judiciously or you may add many newelements.

If you want to smooth some more turn on the smoothfor the Penta_6 elements, set the up to Value very

low (< 0.05) so that most of the prism mesh is leftalone. I also recommend saving before smoothing prisms

in case you dont like the result.

From all accounts CFD++ is very robust and seems tobe able to handle even the worst prisms.


20/21

One Layer/Multi Layer

The one layer, multi-layer argument has gone back and forth for a long time The one layer advantage is speed since it only needs to smooth the volume

mesh out of the way one time and doesnt do any directionalsmoothing. Smoothing is the most time consuming part. The downside is thatyour prisms are very simplistic and not good at capturing complexgeometry If that is all you wanted, then you may as well run prism with allthe layers, but turn the smoothing steps way down (maybe to 0).


e a vantage o mu t p e ayers s t at t e smoot ng can e p en t eprisms around difficult areas, plus you can take advantage of a number of otherintelligent features that the developers worked hard on (floating height, filletratio, etc.). The down side is that it is slow.

I usually try a middle solution. If you want 21 layers in the end, then maybe do

three layers or seven layers instead. Use the max height over base to controlthe growth and then split and redistribute. We put a fair amount ofdevelopment resources on split and redistribute prisms over the 12.0 and 12.1releases.


21/21

Closing

These settings should minimize the number of pyramids you produce,while allowing for a better Prism/tetra transition and control of the first cellheight.

With more discussion, we can adjust these recommendations as needed Contact us if you have any trouble Tetra/Prism is a quick way to get a decent estimate. For more accuracy


,methods.

Documents

Prism NoPyramids