Learning Forum

[AlexC]

Hello, 

I was having trouble obtaining reasonable solutions of pressure drop from Fluent. I created a very simple case of water flow in a circular pipe (files attached). 

Pipe diameter: 1mm
Pipe length: 100 mm
Re=9950
Inlet velocity: 10 m/s
Water density 998.2 kg/m3; viscosity 0.001 Pa.s
Model: k-epsilon standard

It's a pretty straight forward case, and the pressure drop calculated by many empirical models would be about 160 kPa. I obtained similar numbers from CFX simulation and COMSOL simulation. Yet using Fluent, the result is about twice as big, i.e., ~300 kPa. 

I tried several different cases and always got over-predicted pressure using Fluent. I also tried different mesh, yet didn't make it work. 

This is a pretty straight forward case. I don't know where I might have problem. I guess I might have some very stupid rookie mistake somewhere. 

Please advise. 

Many thanks. 

[Kalyan Goparaju]

Hello, 

How are you running the simulation? Axisymmetric or full 3D? Any reasons for not using the default k-w SST? 

Thanks, 

Kalyan

[AlexC]

It's full 3D, steady state. I thought this is a most simple case, and standard k-epsilon model should be able to handle it. Yet I wasn't able to get a reasonable pressure drop. 

BTW, per your another message, the Darcy friction coefficient is about 0.032 based on f=0.3164*Re^-0.25. Pressure drop calculated by Darcy–Weisbach equation is about 160 kPa. Yet I got about  300 kPa in Fluent. 

 

[Kalyan Goparaju]

Hello, 

I did avery quick run based on your setup and got a value of 162kpa which is very close to the theoretical value. I ran 2D axisymmetric with SST k-w. The maximum y+ was around 10. I am wondering if there is some issue with your setup. Can you share more details? Can you perhaps try a 2D axisymmetric run?

Thanks,

Kalyan

[AlexC]

Hey Kalyan,

Thank you for your reply. 

I also ran a 2D axisymmetric with sst k-w, and got similar pressure drop as you (~160 kPa). 

Yet when I shift to standard k-epsilon model in this 2D axisymmetric case(without changing anything else), the pressure drop doubled (~320 kPa). 

I also did the same thing using 3D simulation. sst k-w gives ~160 kPa, and k-epsilon model gives ~320 kPa. 

I tried two different mesh cases. The maximum y+ is 4 in one case, and 1.5 in the other case. It has no effect on the solution. 

So it seems my problem is in the k-epsilon case. But I don't know where is the problem. Not sure if I provided enough info. Let me know if you need anything else. 

Alex

 

[Kalyan Goparaju]

Hello Alex, 

I believe I understand what may be going on. What was the wall-function you used when solving with standard k-epsilon? The standard wall-functions deteriorate for y+ less than 30. In your case, you should be using wall functions that are tailored for low y+ values. SST k-w is y+ insensitive and hence is able to correctly predict the solution. Can you try k-epsilon using Enhanced Wall Treatment (EWT) and check the pressure drop?

Thanks, 

Kalyan

[AlexC]

Hey Kalyan,

Very impressive. I think you pointed the key problem. I tried k-epsilon with enhanced wall treatment, and got a pressure drop of ~180 kPa. Still about 10% deviation from k-w SST model, but not too bad. 

 

BTW, I also did the same simulation in Ansys CFX and COMSOL. Both generated reasonable pressure drop (~160 kPa) with k-epsilon model. I guess these different softwares define k-epsilon in the same way. Probably you know why CFX works well with k-epsilon but Fluent doesn't for the same case? BTW, in CFX, it's the wall function for k-epsilon is "scalable".

It's very interesting since the pipe flow is such a simple case. Many people told me k-epsilon should be OK for most of the problems in simple geometries and normal Reynolds number range. Apparently I should be more careful about using it. 

 

 

[Kalyan Goparaju]

Hello Alex, 

Fluent offers 3 sets of wall treatments with the standard k-epsilon model - Standard and Scalable wall functions, Enhanced wall treatment. In addition to being a y+ insensitive approach, EWT is also a viscous sublayer resolving approach. This means that when it is used with a mesh that is fine enough to capture the flow behavior in the viscous sublayer, more accurate results would be expected with EWT than with scalable wall functions. In a case where the first grid point is always in the log layer, there should not be much difference between EWT and scalable wall functions.

K-epsilon is not necessarily a great model when it comes to wall bounded flows. However, for a simple pipe flow, it should not show any issues, as long as appropriate methods are used in resolving the boundary layer. The issue you were facing was not necessarily a problem of k-epsilon but that of choosing the correct wall modeling approach. 

Thanks,

Kalyan

 

 

[AlexC]

Dear Kalyan,

Thank you very much. It's very helpful! Appreciate your help. 

Sincerely,

Alex