Hello. I have a problem with calculation of the pressure drop along pipe. It's very simple case. Pipe with 4 bends has internal diameter of ~0.36m and length of ~42m. Wall rougness is 400micron. Fluid is a particle laden gas flow, but let's assume just gas phase here. Initial task is connected with bends shape but the problem is the distributed pressure drop.
I started with CFX and there are no problems with CFX variants for now, but I'd like to have some proof from other software for results reliabiliry, so I used Saturne 7.0.2.
I tried different settings but results look strange... First, I used SST model and mesh with inflation layers targeted to Y+=1 (meshes are identical with CFX). Resulting pressure drop is apparently very low. Then I tried k-epsilon (with other mesh, Y+ is around 120), pressure drop is, in opposite, very high. Let's look at numbers (all for clean gas without particles).
CFX, SST, roughness 400um: 590 Pa
CFX, SST, smooth wall: 410 Pa
CFX, k-epsilon, roughness 400um: around 590...600 Pa (oscillates)
Fluent: cannot stabilize with lots off different standard tricks, no results.
Saturne, SST, roughness 400um: 371 Pa (!)
Saturne, SST, smooth wall: 317 Pa (!)
Saturne, k-epsilon, roughness 400um: 1153 Pa (!)
Saturne, k-epsilon, smooth wall: 639 Pa
Semi-empirical correlations, rough wall 400um: 744 Pa
As you can see, we should have, roughly, something like 600...700 Pa. According to empirical formulae, 409 Pa is from distributed pressure drop and remaining resistance is due to bends (semi-empirical method accuonts for any Re / roughness range). But with SST I get just 371 Pa with 400um roughness and 317 Pa with smooth wall, while, from empirical formulae, difference must be 141 Pa, not 54 Pa (there is some level of error, but I don't think it's 2.6 times). With k-epsilon resistance is, in opposite, too high: 1153 Pa with roughness and 639 Pa with smooth wall, even more than empirical method that tends to higher values. Difference smooth/rough is 514 Pa (3.6 times more than with empirical method). In CFX, smooth/rough difference is 180 Pa that is comparable with 141 Pa in empirical method.
Fluent behavior was very strange. It can't even produce a solution, pressure oscillates and the case doesn't converge with any relaxation (down to 0.01), discretization, numeric scheme, roughness, density and viscosity. Completely disappointed with it, Saturne looks much better in this task (we use Fluent for furnaces due to it's powerful reaction mechanism support, but with just one long pipe it won't work).
My questions are:
1. Why there is so huge difference between SST and k-epsilon cases in Saturne?
2. What model and wall function in Saturne is more appropriate for pipe pressure drop calculations?
Here are some case features.
Gas velocity: 20 m/s
Gas density: 0.8 kg/m3
Gas dynamic viscosity: 2.2E-5 Pa*s
Wall function: none in SST smooth wall case, 1-scale or 2-scale model in other cases (SST, k-e with rough wall 400um)
Target CFL: 1...2
Discretization: SOLU with blend ~ 0.8 for velocity, Upwind for others.
Velocity/pressure limits: yes, but only on early iterations
Velocity pressure relaxation: relaxv=0.1...0.3
Mesh: tetra with max size of 25 mm, 10mm at the wall for k-e, 50micron at the wall (inflation first layer) for SST
Convergence control: maximum/minimum pressure stabilization + "standard" minimum iteration number
Case XML file and listing are attached for SST variant with roughness.
Pressure drop in long pipe
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
Pressure drop in long pipe
- Attachments
-
- run_solver.log
- Listing for SST case
- (499.76 KiB) Downloaded 158 times
-
- L-G000-R000-S001R400.xml
- Case file (SST)
- (8.92 KiB) Downloaded 151 times
Last edited by Antech on Thu Oct 27, 2022 8:36 am, edited 1 time in total.
-
- Posts: 284
- Joined: Fri Dec 04, 2015 1:42 pm
Re: Pressure drop in long pipe
Hello,
Did you use the same mesh in CFX and CS?
Regards,
Luciano
Did you use the same mesh in CFX and CS?
Regards,
Luciano
Re: Pressure drop in long pipe
Hello.
Yes, meshes are absolutely identical (exported from Ansys Meshing to CGNS) because they are small: 6 millions for SST and 12 millions for k-epsilon (because of wall sizing of 10mm).
I checked Y+ in CFD-Post (CFX postprocessor) and it's OK for both meshes, although, for some model reason, it significantly higher (up to 10) with roughness (with the same mesh with layers).
In Saturne, Y+ with SST/layers is OK (around 1.0) when the wall is smooth, but becomes ~0.06 with wall roughness of 400um. With k-epsilon/no-layers Y+ is ~20...60 when the wall is smooth and becomes ~5 with 400 um roughness.
According to empirical model, for 400um roughness "sand grain size", flow/roughness regime is in range where both roughness and Reynolds number affect resistance coefficient (friction factor Lambda).
Lambda=0.11*(Delta/d+68/Re)^0.25
DeltaP=Lambda*L/d*0.5*Rho*w^2
Where Delta is roughness [m], d and L are pipe internal diameter and length [m], Rho is fluid density [kg/m3]. For 400um roughness Lambda=0.0212.
For smooth wall (Delta=0), empirical formula for the friction factor is:
Lambda=0.3164/Re^0.25
It gives Lambda=0.0139.
Typical Lambda value for pipes is 0.02 (it's used for preliminary estimation or if roughness is not known).
Maybe there is some problem with accomodating different flow/roughness regimes in wall function? With smooth wall and SST pressure drop is OK, with roughness it's not on any turbulence model (maybe I set something wrong or I need some user parameters tweak, I don't know), with smooth wall and k-epsilon pressure difference is also too high (it should be about 400 Pa according to CFX [SST/k-e] and 317 Pa according to Saturne with SST, 639 Pa on k-e looks too high compared with 317 Pa with SST in Saturne).
I also corrected previous post, semi-empirical value there is for rough wall (Delta=400um). All those pressure differences in previous post are for distributed + local resistances (formulae above are for distributed pressure drop only).
Yes, meshes are absolutely identical (exported from Ansys Meshing to CGNS) because they are small: 6 millions for SST and 12 millions for k-epsilon (because of wall sizing of 10mm).
I checked Y+ in CFD-Post (CFX postprocessor) and it's OK for both meshes, although, for some model reason, it significantly higher (up to 10) with roughness (with the same mesh with layers).
In Saturne, Y+ with SST/layers is OK (around 1.0) when the wall is smooth, but becomes ~0.06 with wall roughness of 400um. With k-epsilon/no-layers Y+ is ~20...60 when the wall is smooth and becomes ~5 with 400 um roughness.
According to empirical model, for 400um roughness "sand grain size", flow/roughness regime is in range where both roughness and Reynolds number affect resistance coefficient (friction factor Lambda).
Lambda=0.11*(Delta/d+68/Re)^0.25
DeltaP=Lambda*L/d*0.5*Rho*w^2
Where Delta is roughness [m], d and L are pipe internal diameter and length [m], Rho is fluid density [kg/m3]. For 400um roughness Lambda=0.0212.
For smooth wall (Delta=0), empirical formula for the friction factor is:
Lambda=0.3164/Re^0.25
It gives Lambda=0.0139.
Typical Lambda value for pipes is 0.02 (it's used for preliminary estimation or if roughness is not known).
Maybe there is some problem with accomodating different flow/roughness regimes in wall function? With smooth wall and SST pressure drop is OK, with roughness it's not on any turbulence model (maybe I set something wrong or I need some user parameters tweak, I don't know), with smooth wall and k-epsilon pressure difference is also too high (it should be about 400 Pa according to CFX [SST/k-e] and 317 Pa according to Saturne with SST, 639 Pa on k-e looks too high compared with 317 Pa with SST in Saturne).
I also corrected previous post, semi-empirical value there is for rough wall (Delta=400um). All those pressure differences in previous post are for distributed + local resistances (formulae above are for distributed pressure drop only).
Re: Pressure drop in long pipe
Hello. Would anybody, please, clarify my question with distributed pressure drop with roughness?
I think you have validated the software in this aspect, maybe you can share some small test case? I will calculate with current Saturne version and other methods and present you with results.
I think you have validated the software in this aspect, maybe you can share some small test case? I will calculate with current Saturne version and other methods and present you with results.
-
- Posts: 4153
- Joined: Mon Feb 20, 2012 3:25 pm
Re: Pressure drop in long pipe
Hello,
I mentioned your post to colleagues, but have not had an answer yet. I'll send them a reminder.
If I understand correctly, you do have a prism layer at the wall ? This is quite important quality-wise.
i think the k-omega will the "all-y+" wall model could be interesting to try here, but am not a turbulence specialist.
Regarding roughness, I am not sure. There has been recent work in the code towards unifying the roughness and smooth models, but as of today, in v7.0, I am not sure a rough wall with very small roughness would provide exactly the same result as a smooth wall...
Best regards,
Yvan
I mentioned your post to colleagues, but have not had an answer yet. I'll send them a reminder.
If I understand correctly, you do have a prism layer at the wall ? This is quite important quality-wise.
i think the k-omega will the "all-y+" wall model could be interesting to try here, but am not a turbulence specialist.
Regarding roughness, I am not sure. There has been recent work in the code towards unifying the roughness and smooth models, but as of today, in v7.0, I am not sure a rough wall with very small roughness would provide exactly the same result as a smooth wall...
Best regards,
Yvan
Re: Pressure drop in long pipe
Thanks for your attention!
Saturne version is 7.0.2.
I use 2 meshes: with prism layers (Y+ approx. 1.0) for SST model and just tetra without layers for k-epsilon model. So every turbulence model should have an appropriate mesh. Meshes are exactly the same as for CFX, only converted to CGNS format. If I use SST + mesh with layers, the resulting pressure drop with rough wall is quite low (lower than just friction/distributed pressure drop calculated with empirical formulae), with k-epsilon model, in opposite, pressure drop is very (approx. 2x times) high, with the same roughness. The same is for smooth wall (please find numbers in my first post).
Regarding roughness, in my case (400 um) it's indeed quite high (standard roughness is 80um). Sorry, but, maybe, during wall function unification some problem appeared that was not caught by the validation. Distributed pressure drop is rarely in user's interest, that's why the problem may not be noticed by others. In my practice during years, there are just 2-3 cases where distributed pressure drop is important.
Saturne version is 7.0.2.
I use 2 meshes: with prism layers (Y+ approx. 1.0) for SST model and just tetra without layers for k-epsilon model. So every turbulence model should have an appropriate mesh. Meshes are exactly the same as for CFX, only converted to CGNS format. If I use SST + mesh with layers, the resulting pressure drop with rough wall is quite low (lower than just friction/distributed pressure drop calculated with empirical formulae), with k-epsilon model, in opposite, pressure drop is very (approx. 2x times) high, with the same roughness. The same is for smooth wall (please find numbers in my first post).
Regarding roughness, in my case (400 um) it's indeed quite high (standard roughness is 80um). Sorry, but, maybe, during wall function unification some problem appeared that was not caught by the validation. Distributed pressure drop is rarely in user's interest, that's why the problem may not be noticed by others. In my practice during years, there are just 2-3 cases where distributed pressure drop is important.
-
- Posts: 4153
- Joined: Mon Feb 20, 2012 3:25 pm
Re: Pressure drop in long pipe
Hello,
I'm not too sure which discretization CFX uses (some form of dual mesh maybe ?), but using code_saturne's discretiation, even for k-epsilon, using prism layers at wall is highly recommended, sot it would e interesting to run k-epsilon on your k-omega mesh. I suspect the mesh has a huge impact.
Regarding the porosity model, I cannot say. In many cases ,it is used for the atmospheric model, where the rougness can be used to represent vegetation (su much more than 400 um). It is probably also used for more classical configurations, but I am not sure what other test cases we may have.
Best regards,
Yvan
I'm not too sure which discretization CFX uses (some form of dual mesh maybe ?), but using code_saturne's discretiation, even for k-epsilon, using prism layers at wall is highly recommended, sot it would e interesting to run k-epsilon on your k-omega mesh. I suspect the mesh has a huge impact.
Regarding the porosity model, I cannot say. In many cases ,it is used for the atmospheric model, where the rougness can be used to represent vegetation (su much more than 400 um). It is probably also used for more classical configurations, but I am not sure what other test cases we may have.
Best regards,
Yvan
Re: Pressure drop in long pipe
CFX uses "automatic" or "scalable" wall function (both for SST and k-epsilon models) that, I believe, means that it automatically chooses wall function suitable for any mesh. It can't turn simple tetra into inflated mesh but, if the mesh is more or less appropriate for the turbulence model used, it automatically selects suitable wall function.
I'll try k-epsilon with inflation layers when I'll have a time (Saturne is used now for other case with air-cooled heat exchanger on RSM and it's doing well, very stable with limitation and relaxation). What Y+ do you recommend for k-epsilon and rough wall in Saturne?
I'll try k-epsilon with inflation layers when I'll have a time (Saturne is used now for other case with air-cooled heat exchanger on RSM and it's doing well, very stable with limitation and relaxation). What Y+ do you recommend for k-epsilon and rough wall in Saturne?
Re: Pressure drop in long pipe
Hello.
I tried k-epsilon with prism layers as you suggested. The mesh was for Y+ ~ 1.0. Result is not good, pressure difference is very high, 1320 Pa after 185 iterations and rising (usually 150...300 iterations is enough for "clear aerodynamics", so the pressure will not be lower with iterations). Almost all wall-adjacent cells are in viscous sublayer. Case XML is attached.
Maybe I need some particular Y+ ? This calculation is on layers providing Y+ at unity level in average. Previous k-epsilon run (results in first post) was on mesh with Y+ ~ 120 that is in recommended region 30...300 or so. What Y+ can you suggest for k-epsilon in Saturne?
Thanks for attention...
I tried k-epsilon with prism layers as you suggested. The mesh was for Y+ ~ 1.0. Result is not good, pressure difference is very high, 1320 Pa after 185 iterations and rising (usually 150...300 iterations is enough for "clear aerodynamics", so the pressure will not be lower with iterations). Almost all wall-adjacent cells are in viscous sublayer. Case XML is attached.
Maybe I need some particular Y+ ? This calculation is on layers providing Y+ at unity level in average. Previous k-epsilon run (results in first post) was on mesh with Y+ ~ 120 that is in recommended region 30...300 or so. What Y+ can you suggest for k-epsilon in Saturne?
Thanks for attention...
- Attachments
-
- L-G000-R000-S001R0KE+Layers.xml
- (8.98 KiB) Downloaded 149 times