[dpalko]

Dear All,

I'm challenging a task of performing DNS simulation of strongly heated single-phase flow in a pipe. I'm wondering if anyone has an experience with DNS simulation in a pipe geometry with heat transfer using OpenFoam?
Any kind on help is highly appreciated.

Regards,

David

[LijieNPIC]

Originally Posted by dpalko

Dear All,

I'm challenging a task of performing DNS simulation of strongly heated single-phase flow in a pipe. I'm wondering if anyone has an experience with DNS simulation in a pipe geometry with heat transfer using OpenFoam?
Any kind on help is highly appreciated.

Regards,

David

Hi, David. Are you still working on this field?

[alexj]

Hi,

did anybody follow up on this and performed DNS simulations of heated single phase pipe flow in OpenFOAM?

Would love to hear about that.

Regards,
Alex

[calf.Z]

I am doing DNS of heated pipe flow. But I do not know how to recognize the current state of simulation. I am not sure if the simulation is going in the right way and when it will end. It costs a lot of time. Any suggestion?

[Santiago]

Originally Posted by calf.Z

I am doing DNS of heated pipe flow. But I do not know how to recognize the current state of simulation. I am not sure if the simulation is going in the right way and when it will end. It costs a lot of time. Any suggestion?

the same way you check any numerical experiment. You do some structural (statistically speaking) analysis of your results, and compare them with experiments. you have azimuthal homogeneity in pipe flows. if theres no stratification, a fully developed pipe flow behaves similar to a channel flow, for first order staristics. you can start verifying a "baseline" case with no buoyancy. Hint: Stable startification "relaminarizes" the flow.

[Santiago]

incidental to this discussion but... Have in mind PISO is not an algorithm suitable for DNS. it is over-dissipative in the inertial range of turbulence, even for time scales below kolmogorov's.

... Not to say you cannot use it for such purpose though.

[calf.Z]

Originally Posted by Santiago

incidental to this discussion but... Have in mind PISO is not an algorithm suitable for DNS. it is over-dissipative in the inertial range of turbulence, even for time scales below kolmogorov's.

... Not to say you cannot use it for such purpose though.

Thank you for your suggestions.

Is there any paper or book that verifies PISO is not an algorithm suitable for DNS? Except for PISO, are there other suitable algorithms for DNS? PIMPLE?
Thank you.

[Santiago]

Hi,

The first order accuracy reduction of PISO for low (<1) CFL was acknowledged by ISSA, the proposer of PISO. You can easily check this issue by running a classic channel flow, driven by a CONSTANT pressure gradient (not by adjusting the bodyforce to maintain a mean velocity), with a higher-than-180 Re_tau. You'll see spurious wiggles in the pressure RMS profiles near the walls.


Articles? First, the one of Issa (1986?). Then you can check the one by Tukovic & Jasak:
"Consistent second-order time-accurate non-iterative PISO-algorithm" That skims through the inconsistency problem of PISO. Then, you can read the following:
"Direct numerical simulation and wall-resolved Large Eddy Simulation in nuclear thermal hydraulics"
Where they do some DNS and LES using OpenFOAM.

The problem of over-dissipation in numerical codes can be tackled by running with time scales of the order of Kolmogorov's time scales, preventing such dissipation to be produced on the first place. But then there's the inconsistency problem (1st order accuracy) of PISO...

As I mentioned, I'm not saying you **should not** use PISO for DNS; instead, you should prove that the solutions obtained therein are not "contaminated" by the aforementioned issues.

One can simply avoid this problems either by using a code that linearizes the advection term in a consistent way for implicit codes, or project the term in time explicitely, or just use one of the many spectral/DG/FR codes that exist in the wild.


(SIDE NOTE/RANT: Validating any solver/turbulence model using a channel flow with Re_tau = 180 is "cheating"; The inertial range of turbulence is almost absent, hence the flow is almost purely dissipative which, in general, masks any (unexpected) dissipative errors that the code may have. Furthermore, the turbulence models are just there for the free ride, specially for low-order (or FV) codes.)