"Large Eddy Simulation for Incompressible Flows: An Introduction" --Pierre Sagaut

"Turbulent Flows" -- Pope

Thanks for the ref. Any special ref for pratical usage for industrial field ?

http://www.wolfdynamics.com/tutorials.html?id=120 training material "at this link" - download the slides (it's Joel Guerrero behind that) They relate to turbulence but have some LES comments Shared LES cases are also there if you're brave enough to DL the 4GB folder 😄 if I remember correctly, some mesh / fv* guidelines are given and for sanity check, look at the energy spectrum to be sure you're resolving what you think you are resolving

nice !!

I am wondering if it's possible to get (blind) reasonable results with medium LES mesh and wall function 🙂

I bet slopezcastano has tips regarding this 😄 Maybe you'll find guidelines in FluidsMechanics101 videos regarding LES?

yes. I am going to explore the LES world. tired of SST RANS

oh I see mister is bragging about his computational power

XD. Only 128 cores. This is very small in comparison to Academic power. In fact maybe slopezCastano can answer, but I got a very interesting discussion with a CFD researcher. He was telling me that we can't use RANS for unsteady problem ! An that's true that I read in the past something like that (If I am correct in the lesson of McDonought, 2007)

So I was wondering if any URANS simulation is "wrong" by essence

I think there was this discussion just a few days ago here, and I guess URANS is disputed yes, see this from 2003 https://www.cfd-online.com/Forums/main/6519-rans-urans.html

thanks I will read that 🙂

Algebraic wall functions are still the achilles heel of wall-modelled LES (read Ugo Piomelli's work). You might want to start with (ID-)DES, given the resemblance with RAS

Criticism of time-dependent RAS comes from the incorrect assumption that the ergodicity theorem holds for time-dependent simulations. Basically it means that one cannot assume averaging along one direction (space) will not have an effect in the other direction (time). (U)RANS may be improved by using Scale-Adaptive Terms (Rotta-like terms) in order to model the time-scales that are lost in the Reynolds decomposition. Problem of turbulence modelling solve by this? Of course not. All of this, however, are criticisms from a fundamental point of view that shouldn't bear much weight in everyday's computations. These are "philosophies" of turbulence modelling, and sometimes the nuisance of modelling has to be confronted with the practical aspects of engineering. LES may be a better choice, yes, if robust algorithms such as PIMPLE/PISO wouldn't generate so much artificial viscosity by themselves. ILES would be perfect, if there was any a-priori metric that could be used to verify its validity for certain archetype flows. The achilles heel of LES is wall-models... And so on and so forth...

Thanks 🙂 so much work in front of me !

is it possible that geometry and boundary conditions interfere with URANS modelling, like in my pump, vortices and rotating blades work on the same scales as turbulent phenomena?

if I increase Co, the case acts like steady state and timesteps don't converge but if I lower it, I see huge oscillations in pressure and torque

in my understanding, you can't run RANS on time scales comparable to turbulent scale (which? IDK) because there's nothing to be averaged... but I don't know how to check that 🙁

are those physical oscillations? And is the average the same as with bigger Co? My guess: Pump physics are linked to your rotation per timestep more than Co, which is recommended to be 0.5 to 3deg per timestep, whatever Co is. If you use a very fine mesh, you maxCo will be 10x bigger for example for a given timestep, but the physics didn't change because of that. I'd say fix your timesteps and close your eyes (or experiment with it and let us know... :D) I don't know about "time scale" of turbulence and if it applies to URANS

yes, that's exactly what I did. I have 700 timesteps for 7-blade impeller and had to increase residualControl to 1e-3 for pressure and now I've also been keeping my eyes closed. Looks like OF knows whem I'm watching because at the moment it seems to work fine

Well, I'm still not at peace with this RANS/URANS saga. Why don't steady-state cases converge if they aren't steady in real-life? RANS models turbulence by time-averaging velocity fluctuations, but turbulent wakes and whatnot are never averaged enough to converge a simpleFoam case? Also, a coarser grid might converge but finer will not, why's that? Where is the limit between what is averaged and what will keep fluctuating even in steady-state solvers and why this sounds like LES

Malalasekera doesn't give sufficient answers... or should I read it again 😕 of course I should

Over the time step, we assume that fluctuations are zero on average, in RANS. If the time step is fine, you will see unsteady behaviour over time steps... However the solution will suffer from absence of first order terms of some part of the momentum. In LES these first order terms interact with the zero order fluctuation terms (the mean). This leads to a different transient evolution of the equation. You won't be able to get a single steady solution with simpleFoam if your steady behaviour is actually two states (eg clockwise or counterclockwise rotating field)...pimplrFoam solution will show you a field oscillating between such extreme configurations. It will also 'converge' in the transient sense, while simpleFoam will not since it's steadystate is of a more absolute sense. Coarser grids have a more diffused structure so perhaps oscillations , which disturb these alternate states, do not emerge.

"Over the time step, we assume that fluctuations are zero on average" - suppose we're studying a wake behind a cylinder; we can average the pressure field to a single value (drag coefficient) either from a transient experiment, LES, URANS, but not from steady-state RANS. The fluctuations (perpendicular to flow) ARE zero on average and RANS is the only one defined as a 'proper' average, yet it is the only model that can't handle this. That's what confuses me most

Can you explain what you mean when you say we can't average the pressure field in steady RANS? What do we see...?

we can but I'd expect the solver to do it...

Numerical solutions are just extension of analytical solutions. So let us say for a very simple problem the cfd process is trying to solve the equation: x^2 = 1, What does the solver try to do? What solution will we expect at the end of any solution process? x= +-1, But will simplFoam be able to achieve it? At the same time pimpleFoam may achieve one of them, and possibly oscillate between the two answers... Oscillations may be initiated by small numerics, turbulence etc. For simple foam, what will cause it's iteration to iteration answer to not vary extremely between the two branches? What the simplefoam solver actually do? What is the bigger picture if we look at oscillations in solution + residuals?

You can understand wakes as a mean flow + oscillating flow . The idea of fnding eigenvalues, or principle components, or support vectores are related to the core solution characteristics .. the numerical methods in standard cfd get us to the final solutions state, but it is not decomposed into these modes. Depending on the algorithm, a steady or transient solver may or may not be able to capture the properties of the equation, (the left or right handed components which are both equally probably, but their average = zero has no stability)

"Why don't steady-state cases converge if they aren't steady in real-life? " What would you expect for a wake behind cylinder case.?