Turbine stage Y+

spl · January 22, 2015, 15:33

Hi guys,

I'm modelling flow through a radial turbine stage. In order to calculate my initial cell height for a Y+=1 and the overall boundary layer mesh height I need an estimate of the Reynolds number for the stage. Does anyone know which reference length and velocity would make a good starting point for the BL mesh? For example using the blade tip speed and tip diameter
(Re=(rho*U(tip)*dia(tip))/mu) would give the equivalent Reynolds number for the rotor but would this be sufficient for the whole stage or is there a better way of defining the value?

Thanks

student666 · February 10, 2015, 14:50

Hi,

have you found a good answer to this question?
I think main problem is that you can't have a constant value for y+ as radial velocity increase according to radius, so the viscous sub layer shrink as well and viscous Reynolds number too; you should decrease progressively your first node height, if you want to resolve totally the boundary layer.

So one hint I can tell you is: it strongly depends on your turbulence model.

I normally use k-epsilon (not the best), as I'm interested on mean values of the flow on discharge and suction air-ducts, and I don't want to spend much time on mesh resolution.

I do as follow: I run few time steps and I check the y+ value (I use openfoam) and then I choose if I need to refine or not, the mesh.

At the end, it's up to you:

-to spend weeks to adapt mesh resolution to your specific case
-to average values and choose to run cases with different mesh resolution and evaluate grid convergence.

But any other suggestions are welcome.

Bye.

spl · February 10, 2015, 16:44

Hi,

Thanks for the reply. No I haven't really got a good answer yet. I came to the same conclusion as you, that progressively reducing the initial cell height would be the best method. However this seems like a very complex approach, is this the method you would usually use? Also the local velocity changes over an aerofoil but a progressive change in initial cell height is not usual practice as far as I'm aware. Is this because the velocity variation over an aerofoil is much smaller than that through a turbine stage?

Would calculating the initial cell height based on the maximum velocity in the volute, therefore ensuring the y+ is always less that 1, be a reasonable approach? Or would this cause issues as in some regions the Y+ would be very small? I would then have a separate boundary layer mesh for the blade.

I am using the SST model and I intend to do a separate boundary layer mesh independence study. I'm just trying to get a reasonable starting point.

Thanks

student666 · February 10, 2015, 23:30

Hi,

Quote:

Thanks for the reply. No I haven't really got a good answer yet. I came to the same conclusion as you, that progressively reducing the initial cell height would be the best method. However this seems like a very complex approach, is this the method you would usually use?

no, this is not the method I would follow,because I think it's an usefulness usage of your own energy!

Quote:

Also the local velocity changes over an aerofoil but a progressive change in initial cell height is not usual practice as far as I'm aware. Is this because the velocity variation over an aerofoil is much smaller than that through a turbine stage?

No I don't think this is the reason, I think you can have high velocity gradient on an aerofoil as well, but in any case, all over my search on this topic, I haven't found (yet?) this way of proceeding.

Quote:

Would calculating the initial cell height based on the maximum velocity in the volute, therefore ensuring the y+ is always less that 1, be a reasonable approach? Or would this cause issues as in some regions the Y+ would be very small? I would then have a separate boundary layer mesh for the blade.

I am using the SST model and I intend to do a separate boundary layer mesh independence study. I'm just trying to get a reasonable starting point.

I have to do mainly with small backward impellers (fan), some people I asked about told me to take account of the velocity at the inlet side of the fan, but I think the solution is: it's a matter of experience and it strongly depends on each case and each turbulence model you use; so as said before I would go for run small iterations (steady analysis) to evaluate the y+ and remesh the surfaces.

Anyway, first question I would ask my self is: what kind of accuracy of flow I want to get? If I want to study separation fo boundary layer, then SST model, or others Low Reynolds models, make good job, but you're up to make a good layer addiction on your surface. In my case some times I can't add more then 3 layers as for small clearance between rotating frame and stationary one; if I try, my meshing SW will surely fail by merging the nearest cells.
That's one of the reason I switched to k-epsilon and on these cells I don't go for any addiction of layers, because I take the (wrong?) assumption that in my small gaps, velocity is quite near to zero (air flows somewhere else), and I think that's not going to improve or worsen my results drastically.

I think one other solution is to use scalable wall function (in a wall function point of view); I don't know anything about that, but the name suggests me that adapt the wall function behavior to your locally y+ value.

It's hard to say, but I think there's not the RIGHT way to proceed, but the one that makes not too much error!
For time dependent initial value problem, even a small change in your initial condition, boundary conditions, or even material properties will surely lead to different solution.

these are my thoughts about turbulence modelling over volumetric machines, but I'm not an expert and I won't lead you to wrong conclusion.

I suggest to make several cases in order to validate your model, grid sensitivity and whaever, and keep monitoring some typical value for your case. In my cases, I take as solution convergence, when I get less than 0.001 m^3/s difference between inlet and outlet.

Bye

spl · February 11, 2015, 04:12

In my case accuracy is quite important as I want to investigate losses and also extend the model for thermal losses through the wall. Therefore the boundary layer resolution is important.

I think your suggestion to test a number of cases is the best way forward. This way I can have a number of meshes with different initial cell heights and measure the effect on accuracy. Rather than starting with a calculated initial cell height I will make a rough estimate and go from there.

As for your issue with the mesh in the gap between blade and shroud. I don't know what meshing program you use but in ICEM if you float the prim layers first and then set initial cell height the layers will fit the gap much better.

Thanks for your help on the subject.

January 22, 2015, 15:33	Turbine stage Y+	#1
spl Member spl Join Date: Nov 2014 Posts: 33 Rep Power: 11	Hi guys, I'm modelling flow through a radial turbine stage. In order to calculate my initial cell height for a Y+=1 and the overall boundary layer mesh height I need an estimate of the Reynolds number for the stage. Does anyone know which reference length and velocity would make a good starting point for the BL mesh? For example using the blade tip speed and tip diameter (Re=(rhoU(tip)dia(tip))/mu) would give the equivalent Reynolds number for the rotor but would this be sufficient for the whole stage or is there a better way of defining the value? Thanks

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Two stage axial turbine in CFX	sherifkadry	CFX	16	June 8, 2020 07:58
Turbine stage	Bartek840	FLUENT	0	January 25, 2014 05:49
2 stage axial turbine model convergence issues	sherifkadry	CFX	2	September 7, 2009 20:51
Turbine stage mixing plane calculation	Knut	FLUENT	0	December 4, 2007 12:46
two stage steam turbine	sebastian	FLUENT	6	June 7, 2006 02:15

February 10, 2015, 14:50		#2
student666 Senior Member M. C. Join Date: May 2013 Location: Italy Posts: 286 Blog Entries: 6 Rep Power: 16	Hi, have you found a good answer to this question? I think main problem is that you can't have a constant value for y+ as radial velocity increase according to radius, so the viscous sub layer shrink as well and viscous Reynolds number too; you should decrease progressively your first node height, if you want to resolve totally the boundary layer. So one hint I can tell you is: it strongly depends on your turbulence model. I normally use k-epsilon (not the best), as I'm interested on mean values of the flow on discharge and suction air-ducts, and I don't want to spend much time on mesh resolution. I do as follow: I run few time steps and I check the y+ value (I use openfoam) and then I choose if I need to refine or not, the mesh. At the end, it's up to you: -to spend weeks to adapt mesh resolution to your specific case -to average values and choose to run cases with different mesh resolution and evaluate grid convergence. But any other suggestions are welcome. Bye.

February 10, 2015, 16:44		#3
spl Member spl Join Date: Nov 2014 Posts: 33 Rep Power: 11	Hi, Thanks for the reply. No I haven't really got a good answer yet. I came to the same conclusion as you, that progressively reducing the initial cell height would be the best method. However this seems like a very complex approach, is this the method you would usually use? Also the local velocity changes over an aerofoil but a progressive change in initial cell height is not usual practice as far as I'm aware. Is this because the velocity variation over an aerofoil is much smaller than that through a turbine stage? Would calculating the initial cell height based on the maximum velocity in the volute, therefore ensuring the y+ is always less that 1, be a reasonable approach? Or would this cause issues as in some regions the Y+ would be very small? I would then have a separate boundary layer mesh for the blade. I am using the SST model and I intend to do a separate boundary layer mesh independence study. I'm just trying to get a reasonable starting point. Thanks

February 11, 2015, 04:12		#5
spl Member spl Join Date: Nov 2014 Posts: 33 Rep Power: 11	In my case accuracy is quite important as I want to investigate losses and also extend the model for thermal losses through the wall. Therefore the boundary layer resolution is important. I think your suggestion to test a number of cases is the best way forward. This way I can have a number of meshes with different initial cell heights and measure the effect on accuracy. Rather than starting with a calculated initial cell height I will make a rough estimate and go from there. As for your issue with the mesh in the gap between blade and shroud. I don't know what meshing program you use but in ICEM if you float the prim layers first and then set initial cell height the layers will fit the gap much better. Thanks for your help on the subject.