Hi friends, Is there any specific method to determine the grid size for CFD problems? Any type of hint or approach being used for knowing the initial grid for a reasonable solution will be welcome. Thanks. Ram Dayal

10.67

tri-balance

computational resources (computer memory, solution speed) numerical consistency (numerical to analytical) numerical accuracy (physical to numerical)

no hard and fast rules of thumb. Just keep all 3 in mind. Strive for a solution that helps balance all three of these aspects.

regards

Q

Hi Ben, I am sorry I didn't got it ''10.67 ????" Bye Ram Dayal

Kinda like buddhism be at one with your mesh and your solution will be at one with you.....

Hi, Thanks, but if I have to start with some initial assumption that so many number of volume elements will be enough for a solution of reasonable accurecy. On what parameters that assumption should be? Bye Ram Dayal

You can't just put an arbitrary number on cell size or cell count, you build your grid depending on what you are looking to model, what you are trying to resolve, what type of elements you are using. It's impossible to say, go away and attend a CFD course then come back and ask a sensible question because this one is pretty much the king of stupid questions in CFD world.

rule number one: Model surfaces (and volumes) have to be discretized with a mesh fine enough to resolve all significant model geometry features.

rule number two (usually more demanding than no. 1): The flow domain has to be discretized with a mesh fine enough to locally capture all significant flow scales. For steady state, those are spatial scales. For unsteady computation, there is the additional requirement that the grid needs to be fine enough to describe the temporal flow variations on the time scales of interest.

Consistency is rather a requirement on the numerical algorithm (i.e. the discretized equations) not on the grid. If the above requirements are satisfied, a consistent and stable algorithm will give you an accurate solution (on the scales of interest).

Of course, to state the rule is always easier than to satisfy it. I would say the majority of CFD calculations is done on a few hundred to dozens of millions of grid cells (orders of 10^2 to 10^6 (toward 10^7)), depending on dimensionality and complexity of geometry and flow. I am sure you realize that there is no simple formula to let you know where your problem fits into that range. If you have any particular problem in mind, we can help you estimate the requirements. That initial estimate can then be optimized by a trial and error process called "grid convergence study".

Thanks Mani

I read thru the answers quickly, so forgive me if I am repeating, but you can also do a mesh-size check. Check your answers for different size meshes. Do a coarse mesh and a fine mesh and see how much your final answer differs. This is more prudent for computational/computer time.

I thought it was 42.

I was told you need to divide by the square root of the Blasius shear stress (estimation) and then multiply by the density difference using Menardo's formula

P.

For steady-state simulation, you need to check the grid error. For transient simulation, you need to check both the grid and time step errors. Read the Roache's book for start if you are new to this subject.

Roache, P.J., "Verification and validation in computational science and engineering", Hermosa Publishers, New Mexico, USA, 1998.

I think you will find that it is the cubed route of the day you were born on multiplied by the average height of a bolivian lama

A simple solution is to look for papers that are modeling similar problems and use their recommendations as a starting pt

Just wanted to say I enjoy your sense of humor, Ben. CFD can be a dry topic.

30/Re_inf

seems to be OK.

If your candidate mesh size is already large for your computing resources, you might want to use two schemes with different order of accuracy and view the relative error field.

30/Re_inf seems to be OK for turbulent flows.

~Characteristic length / 500 seems to be OK. for inviscid flows.

~Characteristic length / 1000 seems to be OK. for laminar flows.

Best regards.

how to do the error analysis ?