[Fole]

Hello all,

Which model is more precise for a transitional flow in convective heat transfer: laminar unsteady, laminar steady or turbulence model? The figure sought is the total heat from a wall to the surrounding air.

I am working in the solar field, and from the literature some authors rely on turbulence models for natural convection. In these problems it is easy to find transitional regimes, i.e. Grashof number in to .

I've got suspicions that all models hold relatively well for heat transfer modeling given that the needed precision is not as high as for the case when lift&drag coefficients are sought.

What is your experience?

Thanks for reading,
Fole.

[FMDenaro]

laminar (steady or unsteady) is not a model...it is a solution regime of the NS equations without any additional modelling, acceptable in case the flow is at low Reynolds number.

If you can, solve the equations for turbulent regime using DNS, otherwise you can think to use LES. If you want a statistically averaged unsteady solution you can use URANS formulation that is less expensive.

The answer depends on your computational resources and on the physical parameters of your problem

[Fole]

Thank you for pointing that, I mistakenly packaged all that in the same bag (Not sure if I should change it or keep it for the sake of thread consistency)

A single time-averaged figure is sought and the computer resources are very limited in time and memory (I have a tuned personal computer but of last generation). However a RANS or URANS simulation of 1 million cells runs pretty fast (1 hour).

My concern with LES is that it is a new class of models and I found that some of the subgrid models yield wrong viscosity levels for the viscosity sublayer. They have to be developed further.

The concern with unsteady DNS is that if the Kolmogorov scales are not reached, then DNS would become implicit-LES depending on the numerical implementation. Here the subgrid is not controlled, so from a theoretical point of view it is wrong basically.

That's what I understand until now.

Thanks,
Fole.

[FMDenaro]

Originally Posted by Fole

Thank you for pointing that, I mistakenly packaged all that in the same bag.

A single time-averaged figure is sought and the computer resources are very limited in time and memory (I have a tuned personal computer but of last generation). However a RANS or URANS simulation of 1 million cells runs pretty fast (1 hour).

My concern with LES is that it is a new class of models and I found that some of the subgrid models yield wrong viscosity levels for the viscosity sublayer. They have to be developed further.

The concern with unsteady DNS is that if the Kolmogorov scales are not reached, then DNS would become implicit-LES depending on the numerical implementation. Here the subgrid is not controled, so from a theoretically point of view it is wrong basically.

That's what I understand until now.

Thanks,
Fole.

LES is a quite old formulation having almost 50 years of history... While it is true that the old models were based on the eddy viscosity assumption (e.g. Smagorisnky), the state of the art of the SGS models is now very wide. For example dynamic mixed models are well suited but you can find a lot of details about many other modern models in the book of Sagaut.

When you use DNS, as you stated, you have to resolve until the Kolmogorov lenght scale, I don't know if you can do that on a PC for your problem. An unresolved DNS (also called LES no-model) could give not so bad results in some cases. That mainly depends on the numerical discretization.

[Fole]

@FMDenaro

I wasn't really aware of Sagaut's book. I read from Wilcox, Moukalled,Mangani&Darwish books, Internet and this forum.

I'm afraid of not reaching Kolmogorov length. However, I performed some simulations and they conform with previous qualitative knowledge (e.g. "this design is better that the other in 3 or 2 times"). They are unsteady DNS and the grid is resolving the viscosity sublayer and detachment regions. The other regions have little cells and turn the flow stable because that (I suspect that the residuals usually present a low bound that corresponds to the approximation/discretization error).

Thank you, I'll read more about that models.

I find LES a quite interesting and promising field,
Fole.

[FMDenaro]

if you are interested in turbulence with heat transfer, you could find useful this LES reading and the references where many papers are cited.

https://www.researchgate.net/publica...ied_turbulence

[Fole]

@FMDenaro

Thank you!

After a fast reading I have found it very interesting. It is even shown a case where the noSGS matches well with DNS, according to what you pointed out here.

It makes sense that FVM has good properties, since it returns back to the conservation on small volumes of the original physics involved.

Nice field to be involved in

Regards,
Fole.