[ANSYS Meshing] How to: Grid independence study for structured mesh

Dronzer · March 30, 2018, 00:46

Hi everyone,

I am trying to predict cd and cl for NACA 0012 which is a test case from Drag Prediction Workshop 6. I have constructed a structured mesh (minimum quality=0.9 and minimum angle=72) as shown in Figures (attached).
Using the above mesh I completed the simulation in Fluent with 7% and 2% errors in cd and cl.
I want to improve the above results and planning to do a grid independence study. Is there any standard way of doing it for structured mesh in ICEM?
If yes, please tell me.

NablaDyn · March 30, 2018, 05:04

A simple method is: Find a reasonable solution on a coarse grid. Refine globally, e.g. by 2x2 cell splitting (in 2D). Rerun your simulation and compare the results.
Refine further until you reach satisfying convergence.
BUT: Keep in mind which discretisation schemes you are using. Second order accuracy formally leads to 'faster' convergence when reducing discrete step sizes than first order schemes.

Dronzer · March 30, 2018, 05:14

Hi NablaDyn,

Thank you for your reply.
Yes, I do have an initial mesh and solution now.
Can you tell me how to refine the grid globally for a structured mesh in ICEM ?

NablaDyn · March 30, 2018, 05:26

I'm not familiar with ICEM. But I'm perfectly sure it offers such refinement tool(s) . Searching the GUI or ICEM's help pages would be the best next step.

Dronzer · March 30, 2018, 05:54

Thanks for the suggestions

I could see global sizing parameter for unstructured meshing. But I could not find such options for structured meshing.
I am quite new to ICEM so I may be wrong.

NablaDyn · March 30, 2018, 05:59

Can you apply that refinement tool or is it not available for your mesh? If so, is the resulting mesh still of quad type? If not, do you have access to the mesh generation data, i.e. where the point distributions on the lines and the surface meshes have been defined?

Dronzer · March 30, 2018, 09:09

Quote:

Originally Posted by NablaDyn

Can you apply that refinement tool or is it not available for your mesh?

I believe that global refinement tools are only for unstructured mesh in ICEM(as far as I know

).

Quote:

Originally Posted by NablaDyn

If so, is the resulting mesh still of quad type? If not, do you have access to the mesh generation data, i.e. where the point distributions on the lines and the surface meshes have been defined?

Yes, I do have access to the number of elements, distribution etc of edges of the blocks (as in fig).

NablaDyn · March 30, 2018, 09:42

Well, that's a good starting point. I assume the 'nodes' entry in the bottom left mask is what you are looking for. Try to adapt the number of nodes uniformly at each mesh-defining edge. As your mesh looks pretty dense I would suggest to first coarsen it, say half number of points, before going too fine.

Shawnnnn · April 1, 2018, 00:37

If you use blocks to generate your mesh, you can start with corse mesh, e.g. 1 million nodes. Then you can increase the nodes to larger numbers, like 2, 4, 8, 16 million.

For 2D problem, you can double the nodes for 1 direction each time.

Compare the results to check convergence. For this case, you can compare global parameters like drag force. Or you can plot out the velocity profiles, or other variables your project is interested in, at certain location to compare. When the change is not much, the mesh can be thought as converged.

Above is my understanding for grid convergence check. Hope it help.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
What is meant by Grid Independence Study?	Khan	FLUENT	10	July 2, 2015 23:40
Moving mesh	Niklas Wikstrom (Wikstrom)	OpenFOAM Running, Solving & CFD	122	June 15, 2014 07:20
grid study, important question	hamid1	FLUENT	1	August 4, 2013 01:14
[ANSYS Meshing] grid study, important questions	hamid1	ANSYS Meshing & Geometry	2	February 10, 2012 14:28
A doubt on grid independence study	G.Balakrishnan	FLUENT	4	November 21, 2000 12:05

March 30, 2018, 05:04		#2
NablaDyn Senior Member Join Date: Oct 2015 Location: Germany Posts: 100 Rep Power: 11	A simple method is: Find a reasonable solution on a coarse grid. Refine globally, e.g. by 2x2 cell splitting (in 2D). Rerun your simulation and compare the results. Refine further until you reach satisfying convergence. BUT: Keep in mind which discretisation schemes you are using. Second order accuracy formally leads to 'faster' convergence when reducing discrete step sizes than first order schemes.

March 30, 2018, 05:14		#3
Dronzer Member Join Date: Apr 2016 Posts: 53 Rep Power: 10	Hi NablaDyn, Thank you for your reply. Yes, I do have an initial mesh and solution now. Can you tell me how to refine the grid globally for a structured mesh in ICEM ?

March 30, 2018, 05:26		#4
NablaDyn Senior Member Join Date: Oct 2015 Location: Germany Posts: 100 Rep Power: 11	I'm not familiar with ICEM. But I'm perfectly sure it offers such refinement tool(s) . Searching the GUI or ICEM's help pages would be the best next step.

March 30, 2018, 05:54		#5
Dronzer Member Join Date: Apr 2016 Posts: 53 Rep Power: 10	Thanks for the suggestions I could see global sizing parameter for unstructured meshing. But I could not find such options for structured meshing. I am quite new to ICEM so I may be wrong.

March 30, 2018, 05:59		#6
NablaDyn Senior Member Join Date: Oct 2015 Location: Germany Posts: 100 Rep Power: 11	Can you apply that refinement tool or is it not available for your mesh? If so, is the resulting mesh still of quad type? If not, do you have access to the mesh generation data, i.e. where the point distributions on the lines and the surface meshes have been defined?

March 30, 2018, 09:42		#8
NablaDyn Senior Member Join Date: Oct 2015 Location: Germany Posts: 100 Rep Power: 11	Well, that's a good starting point. I assume the 'nodes' entry in the bottom left mask is what you are looking for. Try to adapt the number of nodes uniformly at each mesh-defining edge. As your mesh looks pretty dense I would suggest to first coarsen it, say half number of points, before going too fine.

April 1, 2018, 00:37		#9
Shawnnnn New Member Join Date: Jun 2017 Location: USA Posts: 4 Rep Power: 9	If you use blocks to generate your mesh, you can start with corse mesh, e.g. 1 million nodes. Then you can increase the nodes to larger numbers, like 2, 4, 8, 16 million. For 2D problem, you can double the nodes for 1 direction each time. Compare the results to check convergence. For this case, you can compare global parameters like drag force. Or you can plot out the velocity profiles, or other variables your project is interested in, at certain location to compare. When the change is not much, the mesh can be thought as converged. Above is my understanding for grid convergence check. Hope it help.