|
[Sponsors] |
May 10, 2006, 21:30 |
Is this possible
|
#1 |
Guest
Posts: n/a
|
CFD++ customers are not afraid of using unstructured grids for applications that demand a high level of accuracy, robustness and speed, even when they are "solving to the wall" with mesh aspect ratios that exceed 100000.(http://www.metacomptech.com/)
|
|
May 11, 2006, 09:15 |
Re: Is this possible
|
#2 |
Guest
Posts: n/a
|
In a word, yes.
|
|
May 11, 2006, 11:08 |
Re: Is this possible
|
#3 |
Guest
Posts: n/a
|
Doesnt the solution degrade because of the wide change in the element size.What about the time stepping used for the calculations.
|
|
May 11, 2006, 11:44 |
Re: Is this possible
|
#4 |
Guest
Posts: n/a
|
If you have a 100,000 aspect ratio against the wall and then a large cell immediately next to it, then obviously that impacts boundary layer prediction.
Change in element size is a function of your grid generator. If you have a grid generator that creates a sufficient boundary layer extrusion and transitions to the inviscid grid smoothly, there shouldn't be an issue. Time stepping can be a problem with largely disparate cell sizes adjacent to each other, but again that's going to depend on how you make your grid. I believe there are a few different meshers out there that support CFD++. |
|
May 11, 2006, 17:44 |
Re: Is this possible
|
#5 |
Guest
Posts: n/a
|
Aspect ratios of 100,000 or higher? For one-dimensional flow that's not going to be a problem at all.
With most (if not all) numerical methods, the cell aspect ratio itself (not just cell-to-cell variations) are a concern regarding both stability and accuracy. Accuracy is not a problem as long as the flow is still sufficiently resolved even along the longer side of the cell. However, even if the flow warrants the use of such cells, I would always be skeptical... With a finite volume method, it's hard to imagine getting a reasonable flux balance when some cell edges are simply neglible in size compared to others. You may as well use a 1-D cell. With a finite element method, it's hard to imagine shape functions that reasonably interpolate the solution based on awkwardly placed control points (cell corners/edges). All these cells will give you is 1-D flow, regardless of the real physics, so you better use them wisely, if you have to use them at all. A trivial example I can think of is fully developed flow in a straight pipe, where gradients in flow direction are really zero. So, is it possible? Yes, I suppose you can come up with examples where locally the flow really is one-dimensional. Then you should be ok provided that your solver can handle such disparity in resolution and you don't get convergence or stability problems as your flux balance approaches the size of numerical round-off errors. |
|
Thread Tools | Search this Thread |
Display Modes | |
|
|