[kyo022]

Hello,

I'm analyzing the turbulence cased by the pantograph of high-speed trains, and
one of the review of my paper told me that I should use K-omega SST model NOT the standard k-epsilon model.

I understand that SST is better than the standard K-epsilon model, for example, when the flow separation is the interest.

But, in my case, the flow speed is above 300km/h, which means a high Re number and as far as I know SST model is suitable for low Re number cases.

Moreover, the Y+ value should be maintained within <1 when using SST model to my knowledge, but I frankly don't understand why my y+ values still range from 20~300.

I would really appreciate if you could share your knowledge on this.

[Gert-Jan]

You are mixing up the different models.
SST is some kind of a hybrid model that can handle full range of Y+. From less than 1 and up to 300.
On one side of the spectrum it behaves like k-omega. On the otherside it behaves likes k-epsilon. Maybe you could consider it as a universal model.
So, with SST, don't let Y+ be larger than 300 and you're more or less safe.

[kyo022]

Thank you for such a clear explanation!

I tried to look up at various sources about this but no one explained easily like you.

I guess now I have to work on reducing Y+ values for few regions that overly exceed 300.

Again, thank you!

[Gert-Jan]

Maybe it is a bit too crude. If someone has a more detailed view, do not hesitate to post a comment..........

[ghorrocks]

Do a sensitivity study If the simulation is not sensitive to the boundary layer then it won't matter (yes, many types simulations don't care what near wall model you use).

So run a model with y+ being large and another with y+<300 and see if it makes a difference.

[fede32]

Can somebody provide some reference on the y+<300 limit? Thanks!

[ghorrocks]

There is no limit at y+=300. The upper limit of allowable y+ values is problem dependent and needs to be determined case by case with a sensitivity analysis.

[fede32]

Originally Posted by ghorrocks

There is no limit at y+=300. The upper limit of allowable y+ values is problem dependent and needs to be determined case by case with a sensitivity analysis.

Which kind of analysis should we do in order to determine the y+ upper limit?

[Gert-Jan]

In pinciple you have to perform multiple calculations with different grids and see how results (velocity profiles, pressures, forces, etc) change with grid density. Meanwhile, see what Y-plus you obtained in each calculation and how results are affected.
The results depends a lot on the turbulence model you apply and the geometry you have. If it is streamlined or blunt, etc.

[fede32]

Originally Posted by Gert-Jan

In pinciple you have to perform multiple calculations with different grids and see how results (velocity profiles, pressures, forces, etc) change with grid density. Meanwhile, see what Y-plus you obtained in each calculation and how results are affected.
The results depends a lot on the turbulence model you apply and the geometry you have. If it is streamlined or blunt, etc.

Great! I think this is what I usually do, trying to get convergence in results. For example, using k-epsilon, I usually get result convergence once the grid has y+ under 50/70. But I haven't found this limit changing with problems, for example, I haven't seen a problem in which the results don't converge until y+<10. So I thought that upper limits were more general.

[Duke711]

Originally Posted by kyo022

I understand that SST is better than the standard K-epsilon model, for example, when the flow separation is the interest.

No, that's not correct. Decide are the wall functions of the k-e model. k-e model with scalable or enhanced wall treatment and fine mesh is even better than k-o sst. k-o sst only good with high y+ (<300) and large meshgrid size without energy therms. When the flow separation is the interest, the k-o sst is very bad, because it is not working with a fine mesh (y+ < 11). Standard k-o or k-e with enhanced wall treatment and y+ < 4 recommend. Flow seperation results are prooly resolution with a y+ > 11.

[ari003]

Originally Posted by Duke711

No, that's not correct. Decide are the wall functions of the k-e model. k-e model with scalable or enhanced wall treatment and fine mesh is even better than k-o sst. k-o sst only good with high y+ (<300) and large meshgrid size without energy therms. When the flow separation is the interest, the k-o sst is very bad, because it is not working with a fine mesh (y+ < 11). Standard k-o or k-e with enhanced wall treatment and y+ < 4 recommend. Flow seperation results are prooly resolution with a y+ > 11.

You are mixing up the models. There are two possible ways to properly model with RANS. One is with y+ in viscous sublayer (y+<1) which k-omega standard model use or with y+ in logarithmic region (30<y+300) which k-epsilon uses. For the later one, wall functions are necessary so you wont find turbulent model using wall function to model when y+ is in viscous sublayer.
For SST k-omega it switches from k-omega near wall to k-epsillon in free-stream with some blend.
In my opinion, even when the y+ near wall is in log law region the SST k-omega uses k-epsilon term to model the flow which is kinda similar to standard k-epsilon.
Please let me know, if I stated anything wrong.