[kveki]

I apologize in advance for my poor English. I use a translator.

I have a general question.

Do we need structured grids in CFDs now? Is it necessary to mess with the block topology in ICEM CFD (or another program) now, or is this direction dead and it is enough to use, for example, unstructured polyhedral grids with corresponding boundary layers.

15-20 years ago, when unstructured solvers were not very developed, structured grids provided some advantages associated with greater accuracy. Currently, the accuracy of the results obtained on tetrahedral and hexahedral grids is practically the same for most problems. It turns out that if the geometry is quite complex, then it is pointless to spend time creating a structured hex grid, since the results will not be better and at the same time you will spend many times more time.

Is there any good reason to use structured grids today?

[aerosayan]

It's complicated ...


Unstructured grids are useful because they're generic and work on any domain shapes.


That's why commercial or open source codes use them. They want to develop for a large group of users and they want to make it flexible for any kind of problem.


Unstructured grid codes aren't exactly slow, they can be optimized to be very fast.



Structured grids still have their uses. I mainly use them because the solvers are easy to develop for structured grids.



With overset meshing, you can technically use structured grids for everything ...


But overset meshing introduces little bit of communication overhead.

[LuckyTran]

If the question is about structured vs unstructured grids, then most ppl have access to unstructured solvers now and few people still CFD exclusively use structured grid solvers. I personally have not used a structured CFD solver in my lifetime.


But if the question is about whether a good hexahedral grid is still good, then yes block meshes such as that generated in ICEM are still used when they are appropriate. When I do acoustic calculations for example, these are always on hexahedral grids. What is a waste of your time and the cluster's time and everyone's time is you doing a time resolved simulation of a propagating wave and your grid dissipates the entire wave. What a waste!

[CFDfan]

I would rephrase the question as "is there any good reason NOT to use structured grids". In mine experience the structured grids were the only option when dealing with very complex, "not clean" 3D geometries. One could spent countless of hours there to "clean" these and still not be able to mesh them with unstructured grids. Finally, the structured grids solvers are more stable.

[andy_]

Originally Posted by kveki

Is there any good reason to use structured grids today?

Structured grids require less computational resources for a given level of accuracy but at the cost of a less flexible grid distribution. Large expensive DNS and LES simulations that require most of the available computational nodes and run for days are examples of CFD simulations that generally ought to be using structured grids. Low order, steady simulations of the flow around complicated shapes with a lot of strong local gradients would generally benefit from using unstructured grids.

[FMDenaro]

There are also theoretical reasons to take into account.

Anyone knows what is the sense of requiring that the local truncation error vanishes for h->0 on structured grid. But when we talk about the consistence of a scheme for unstructured grid this has to be extended in a more general way, consequently the "accuracy order" has to be carefully considered.

Still from a theoretical point of view, what about the spectral analysis of a scheme on unstructured grid? That is, it is possible to analyse the spectral resolution in an extended "modified wavenumber" analysis?


My opinion is that unstructured grid can be associated to FEM analysis and the theory for linear problems exists. Non linear problems, time-dependent problems have a theory to be extended properly.


Generally, unstructured grids could be a well suited ground for FVM that exploit the available FEM theory. But I still haven't seen a clear framework for specific formulations like LES. That is an issue I've seen since the first papers on unstructured grid I read about 30 years ago.

[CFDfan]

a similar, general question I'd like to ask is:

Are the transient analyses (with the right, small enough time step) perceived to be more accurate than the steady state ones?