Minimum number of points in wall normal

canopus · November 12, 2015, 17:33

Hello
I am trying to do RANS for a plane diffuser geometry.

To use High Reynolds Model (like k epsilon) I want to keep average y+ > 30 but this will mean putting fewer points in the wall normal direction.

Is there a minimum number of points that should be across the channel ?
(i think I read 10 somewhere cant remember now)

Thanks in advance

LuckyTran · November 13, 2015, 00:18

For boundary layers ~8-10. So maybe 16-20 across the channel is appropriate. But what's important is the size of the structures that need to be resolved. In a uniform flow for example, there is nothing to resolve and 1 grid point is sufficient to represent a uniform flow. And they're more like guidelines than hard rules (i.e. no minimum).

canopus · November 13, 2015, 06:03

Thanks for the reply.

Its a developing flow and then separates.

The question is then how to ensure y+ > 30?

One can have coarse mesh at wall and fine near center or is there any other way out?

skewness abyss · November 15, 2015, 15:47

If you look at the theory guide in the FLUENT manual in the turbulence section, they say it is better to have more cells in the BL than to achieve a certain y+ value. I'm not sure if you are using FLUENT, but I believe this statement still holds if you are using another solver. If you are placing 10 cells in the BL but achieving a y+<30, you might need to switch to a k-epsilon model that integrates all the way down to the laminar sublayer (y+<1). That way, you get the correct BL resolution and you meet the required y+ value at the same time. In FLUENT, they call this option "Enhanced wall treatment".

canopus · November 15, 2015, 16:56

Thanks for the reply. I agree that the more points to resolve the BL the better. But I want to study the performance of High Reynolds Models i.e y+ > 30.

So if I switch to EWT or Low Reynolds Model it gives better results surely but doesn't serve my purpose.

skewness abyss · November 15, 2015, 20:13

In that case, I would try two things

First, I would use the adaption feature in your solver (if it has one) and adapt based off of the y+ value. I would tell the program that I want a minimum y+ of 30 and it will coarsen those cells near the wall to achieve that y+ value. You just have to be careful with using the adaption feature as sometimes it makes the quality of the cells worse.

The second thing I'd try is to go back to my mesher and raise the first cell height value near the walls and run the simulation again. Then I would check to see if most of my cells near the wall meet the y+>30 requirement. You can view contours of y+ on your walls to visually inspect the y+ values. If too many cells do not meet the y+ requirement, then you need to keep raising the first cell height value in your mesher until they meet the criteria.

As LuckyTran says, there is no hard rule as to how many cells should be across the channel. It should be a number such that your discretization error is a low enough number for a variable of interest. Say you can tolerate a 1% error in velocity across the channel. Then keep increasing the number of cells across the channel while keeping the first cell height fixed until your calculated grid convergence index value for velocity is less than 1 %.

November 12, 2015, 17:33	Minimum number of points in wall normal	#1
canopus Member SM Join Date: Dec 2010 Posts: 97 Rep Power: 15	Hello I am trying to do RANS for a plane diffuser geometry. To use High Reynolds Model (like k epsilon) I want to keep average y+ > 30 but this will mean putting fewer points in the wall normal direction. Is there a minimum number of points that should be across the channel ? (i think I read 10 somewhere cant remember now) Thanks in advance

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[snappyHexMesh] snappyHexMesh sticking point	natty_king	OpenFOAM Meshing & Mesh Conversion	11	February 20, 2024 09:12
Natural convection in a closed domain STILL NEEDING help!	Yr0gErG	FLUENT	4	December 2, 2019 00:04
Wrong flow in ratating domain problem	Sanyo	CFX	17	August 15, 2015 06:20
foam-extend_3.1 decompose and pyfoam warning	shipman	OpenFOAM	3	July 24, 2014 08:14
decomposePar pointfield	flying	OpenFOAM Running, Solving & CFD	28	December 30, 2013 15:05

November 13, 2015, 00:18		#2
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,674 Rep Power: 65	For boundary layers ~8-10. So maybe 16-20 across the channel is appropriate. But what's important is the size of the structures that need to be resolved. In a uniform flow for example, there is nothing to resolve and 1 grid point is sufficient to represent a uniform flow. And they're more like guidelines than hard rules (i.e. no minimum).

November 13, 2015, 06:03		#3
canopus Member SM Join Date: Dec 2010 Posts: 97 Rep Power: 15	Thanks for the reply. Its a developing flow and then separates. The question is then how to ensure y+ > 30? One can have coarse mesh at wall and fine near center or is there any other way out?

November 15, 2015, 15:47		#4
skewness abyss New Member Danny Join Date: Feb 2015 Posts: 13 Rep Power: 11	If you look at the theory guide in the FLUENT manual in the turbulence section, they say it is better to have more cells in the BL than to achieve a certain y+ value. I'm not sure if you are using FLUENT, but I believe this statement still holds if you are using another solver. If you are placing 10 cells in the BL but achieving a y+<30, you might need to switch to a k-epsilon model that integrates all the way down to the laminar sublayer (y+<1). That way, you get the correct BL resolution and you meet the required y+ value at the same time. In FLUENT, they call this option "Enhanced wall treatment".

November 15, 2015, 16:56		#5
canopus Member SM Join Date: Dec 2010 Posts: 97 Rep Power: 15	Thanks for the reply. I agree that the more points to resolve the BL the better. But I want to study the performance of High Reynolds Models i.e y+ > 30. So if I switch to EWT or Low Reynolds Model it gives better results surely but doesn't serve my purpose.

November 15, 2015, 20:13		#6
skewness abyss New Member Danny Join Date: Feb 2015 Posts: 13 Rep Power: 11	In that case, I would try two things First, I would use the adaption feature in your solver (if it has one) and adapt based off of the y+ value. I would tell the program that I want a minimum y+ of 30 and it will coarsen those cells near the wall to achieve that y+ value. You just have to be careful with using the adaption feature as sometimes it makes the quality of the cells worse. The second thing I'd try is to go back to my mesher and raise the first cell height value near the walls and run the simulation again. Then I would check to see if most of my cells near the wall meet the y+>30 requirement. You can view contours of y+ on your walls to visually inspect the y+ values. If too many cells do not meet the y+ requirement, then you need to keep raising the first cell height value in your mesher until they meet the criteria. As LuckyTran says, there is no hard rule as to how many cells should be across the channel. It should be a number such that your discretization error is a low enough number for a variable of interest. Say you can tolerate a 1% error in velocity across the channel. Then keep increasing the number of cells across the channel while keeping the first cell height fixed until your calculated grid convergence index value for velocity is less than 1 %.