[Akshay]

Hey Foamers!

I wanted to know if any of you have worked on handling bad cells at the solver level. Something like ... ignore the bad cells during the solving phase..interpolate the values around the bad cell and move on.
Or has someone worked on any other ideas ?

**Bruno ... point me in the right direction!

Thanks
Akshay

[wyldckat]

Greetings Akshay,

This question has already been asked a few times, although I don't have the time to go search for it now.

It really depends on what kinds of bad cells you're dealing with. If it's an orthogonality issue, you can try using a value greater than 0 for the "nonOrtho*" option in the "SIMPLE", "PISO" and "PIMPLE" solvers in "system/fvSolution".

If the cells are reaaaaaaly bad, it's usually better to just re-do the mesh, because the only other solution is to fix it manually, either:

• by using setSet +subsetMesh: http://openfoamwiki.net/index.php/SetSet#Usage_example
• or by using modifyMesh: http://www.cfd-online.com/Forums/ope...tml#post345332 post #3

Best regards,
Bruno

PS: Next time you want to bump the thread up, you could take the time to list some of the information you've already found and for what you've looked for.

[Akshay]

Hi Bruno

Thanks for the reply.
All the correctors...schemes...limiters... and modifying mesh options..all that is there (been there ... done that )
But when I have a mesh count of say 5 million and I have say 100 skewed cells and 100 non-ortho cells, I'd like to ignore these and carry on with my simulation. If this can be done at the solver phase, it would be very handy. I don't really have to worry about stability and accuracy...unless the cells are in areas of high gradients, which let us assume is not in my case.

So I wanted to know if anyone has implemented anything like this in the solver. And it would help a great deal, wouldn't it?

Akshay

[Tushar@cfd]

Can you explain your case?

But when I have a mesh count of say 5 million and I have say 100 skewed cells and 100 non-ortho cells, I'd like to ignore these and carry on with my simulation. If this can be done at the solver phase, it would be very handy. I don't really have to worry about stability and accuracy...unless the cells are in areas of high gradients, which let us assume is not in my case.

Are you sure about this? (highlighted point)

If your cells are away from high gradient region, then you can use coarse mesh in that region and avoid non-ortho cells. It is easy to handle non-orthogonality of coarse mesh and you can do smoothing of such cells. As per your case, why you require non-ortho cells in those regions (low gradient regions)?

It's just a suggestion

-
Best Luck!

[Akshay]

Hi Tushar

It is hard to make sure that I will not end up(in the first attempt) with a bad cell in some region considering we use snappyhexmesh...and I DO NOT want to mesh it again and ensure that I don't have those cells in my domain.
But what I said(my case) is just an assumption...I'm sure Fluent and other commercial solvers have this implemented (apart from their limiters and other robust numerical implementations).
And again we are deviating from my original point...I feel this kind of an implementation will help the solver a great deal. Comments?

Akshay