Different results from turbulence models switching from CS 2

[andrea-dyn]

Hi all,
in the past I've used CS 2.0.7 to analyze the pressure and velocity fields in complex geometries (oleodynamic valves) but recently I've switched to CS 3.2 from Ubuntu repositories (same kind of geometries) in order to gain advantages from the added new features.

Now I'm quite surprised seeing that the same mesh with the same set of calculation parameters returns different results in terms of pressure and velocity fields.

The test case that I've used can be summarized as following:
Geometry: oleodynamic control valve;
Reynold: very high (turbulence condition)
BC: normal velocity of the flow at the inlet face (21 m/s @ operating conditions)
Wall: smooth

For this geometry, some experimental data are also available.

The study has been conducted using the steady flow algorithm, because the goal is to evaluate the pressure and velocity fields @ steady operating condition.
The operating condition at the inlet surface is reached by increasing the fluid velocity from 1 m/s to 21 m/s in 100 calculation steps (linearly with the calculation steps), after that the velocity is kept constant for the next 200 steps, in order to allow the calculation convergence.
The turbulence parameters have been set to the lower and higher typical limits, in order to investigate also the sensitivity of the turbulence models depending from the parameters themselves.

The attached figures show, in terms of delta-pressure across the valve, the differences between the results obtained using CS 2.x and CS 3.x; the following turbulence models have been considered:
- Mixing length: Length Scale limits: 3% of Hydraulic Diameter; 18% of Hydraulic Diameter (0,285 mm and 1,4 mm respectively)
- K-epsilon: intensity value limits: 2%; 20% (the hydraulic diameter has been set to the real one: 7,5 mm)
- K-omega SST: intensity value limits:: 2%; 20% (the hydraulic diameter has been set to the real one: 7,5 mm)

As you can see, the CS 3.x seem to underestimate the DP in all the tested condition, while the CS 2.x seems to be more precise with the same parameters.
Moreover, the sensitivity of the parameters seems to be higher in CS 2.x; this is true especially for the ML model, of which the result @ 18% LS has been not reported in the graphic because it reaches the value around 95 bar, which overestimates a lot the experimental value.
Conversely, the turbulence intensity has a very small sensitivity in all the cases, with both the CS 2.x than with the CS 3.x.

Last but not least, the computation time has dramatically increased (using the same HW) in CS 3.x; the factor with respect to CS 2.x is about 2.5 times.

Any feedback about this test will be appreciated, we consider CS a very useful tool for the CFD analysis but we also wish to understand the possible reasons in this different behavior, in order to decide if permanently switch to CS 3.x or wait for the 4.x series.

Thanks in advance...

[Yvan Fournier]

Hello,

Which version x of 3.2.x are you using ? 3.2 is not a fully validated version, so there might be a bug, but differences might also be due to an incorrect setup.

If in doubt, you should try 3.0, which is a fully validated version (the next fully validated version will be 4.0, with a first beta possibly next week, but a validated version within a few months).

The performance factor of 3 is not normal. Performance between 2.0 and 3.0 should be pretty similar in most cases. This might be due to your setup or a packaging issue (I recently saw the package on Ubuntu 14.10 is broken, as it can't compile user subroutines; we'll have to file a bug report).

In any case, without additional info (as in the recommendations for this forum), I can't guess at much more.

Regards,

Yvan

[andrea-dyn]

Hi Yvan,
thanks for the quick reply.

I'm using the version 3.2.1 (from UBUNTU repository).

Unfortunately I cannot post further informations (mesh, listing and so on) because are protected by industrial IP.

My question was just to investigate if il the registered behavior was already observed in other cases.

Regards,
Andrea