Any way around mesh study using mesh stats?
 

I am by no means trying to say mesh sensitivity studies are unnecessary. However, my boss says they aren't (he is not a CFD guy), and is demanding answers faster and faster. Is there any way that using mesh stats like y+, aspect ratio, mesh exp factor, etc... to determine if a mesh is giving decent results? I know this doesn't promise the same mesh independence of mesh refinement study, but it would hopefully mean that there is more numerical stability.

In fact most of the recent research is focused on getting results which are less sensitive to computational factors.

For example now we have automatic and scalable wall functions which are less sensitive to Y+.

To answer your question, I would say,

1a. Keep yplus above 30 if you not interested in separation, boundary layer etc. Use scalable wall function
1b. keep it between 1-10 with 10-15 nodes in boundary layer if you want to predict the seperation. Use hybrid wall functions aka automatic wall treatment.

2. Use SST if you are solving external flows specially aerodynamics .

3. Aspect ratio should be 1000 for single precision and can go up-to 10000 in boundary layer for the double precision. However I have used up-to 8000 aspect ratio for compressor simulation with single precision and results were good (less than 2% error).

4. Mesh expansion fator : At max 1.25

For axial compressor simulation, I know I can get good results with 0.2 million if I am not interested in minor flow features and may need up to 1 million mesh if want to resolve the boundary layer, tip clearance, shock boundary layer interaction, wake etc.

PS. Above comments can not be generalised. So don't use these advices unless you verify it for your case. However for quick iteration you can use it.

Rule No. 1. Boss is always right. :D
Rule No. 2. If boss is wrong then please refer to rule no. 1 ;)

Thanks Far. Rule 1 and 2 are definitely right! haha. I use SST with 15 layers and 1.2 exp rate. My y+ ave is about 1, and I know max value should be 1, but its only at very few areas that it goes past 1 and its been a victim of trying to cut mesh down. Do you have any general mesh tips for the rest of the body? I have a list of what I shoot for as far as overall domain ratios and factors for my rotating machinery apps.

for initial iteration, don't go for yplus 1. If it useless to get the yplus 1 if other nodes outside the viscous sub layer.

Moreover, in latest wall treatments yplus 10 should work fine as yplus 1 would do.

You're a big help Far. I have wall function on (I'm using CFX, probably should have said that before). Thanks again. I certainly want to meet the bosses schedule, but I don't want angry designers and test engineers coming back to me in a month or two asking how I could be so far off!

Run one case with fine mesh (within your computer limits) and one meduim and compare. Also run one mesh with Yplus 10 with automatic wall treatment and one with Y+ 30 or so and wall function and observe the effects.

Put in this way. If you dont have any separation then you will probably get the similar results with automatic wall treatment and wall function approach. Even you may get better solution and convergence due to low distortion in mesh, which is due to excessive stretching in boundary layer. You use saved mesh in the core flow region to resolve it properly