[flow_CH]

Hello. I hope you are great.


I have 2 basic questions about validation of model. I will be grateful if you help me.


1- Assume I simulate a model with a turbulence model like standard - model (the order of turbulence model used are not important in this question). I validate the model with experimental (physical) model. Then I simulate the same model with two another turbulence models like RSM and RNG. Should I again validate two latter models with experimental model?


2- Assume I simulate a channel with water flow for 30 degrees slope and I validate the model with experimental (physical) model. Then I have an idea and I make a decision to simulate the same model for 30 degrees slope but in this case there is no experimental data to validate. Can I simulate it?



Thanks

[PositronCascade]

Hello. Here my opinion:

When you are dealing with RANS, you use one of the turbulence models. After selecting the model, you can complete the numerical simulations and you can tune the coefficients of the model by using an experimental study. If you change the model, you need to validate your results once more.

If you use the same-type problem and same turbulence model but you just change the slope, in that case, you may validate your results with an experimental study that is available in the literature and then you can change the slope.

[flow_CH]

Originally Posted by PositronCascade

Hello. Here my opinion:

If you use the same-type problem and same turbulence model but you just change the slope, in that case, you may validate your results with an experimental study that is available in the literature and then you can change the slope.

Thanks for your reply.

What do you mean "available in the literature"? There is only one experimental study for this channel. No other information for this particular channel in literature.


Actually, I want to see that for every small change in geometry or flow condition or fluid properties, we must validate the numerical model?

Let me explain the channel simulation simply,

SIMULATION 1: Slope "a" degree. flow rate "a" m3/s.
Be validated by experimental.

SIMULATION 2: Slope "b" degree. flow rate "a" m3/s.
Without any change for other parameters like turbulence.

SIMULATION 3: Slope "a" degree. flow rate "b" m3/s.
Without any change for other parameters like turbulence.

Can I simulate number 2 and 3 without validation? You can see for every simulation, I only change one of the parameters.

[LuckyTran]

Regardless, if there's no data, then it's not even possible to validate it.

[sbaffini]

First let me friendly complain about a misuse of terminology that I see frequently and might be a cause of confusion in your case too.

In physical simulations we do not typically refer to the whole numerical simulation as "the model". Taking CFD as example, you first have:

1) Reynolds transport theorem + Newton Law
2) Viscou stress model + Stokes Hypothesis
3) Thermodynamic model

We kind of give them for granted. On top of these, you might have:

4) Turbulence model
5) Combustion model
6) Multiphase model

etc.

To which, instead, we typically refer as models.

I'm not saying that there is an absolute, correct terminology, but just that, as engineers, as opposed to physicists, the level of accuracy we should typically care of is not at ground zero. Defining the whole numerical simulation as "model" gives more often than not a wrong impression about what you have control on.

Once you have 1-3 granted, you can manage to separate your numerical aspects from the modeling aspects in 4-... That is, how you discretize your problem is a purely mathematical task whose results can be verified (as opposed to validated). There is no modeling here, just not necessary assumptions that should be controllable by the analyst.

On top of these come the models, with which we refer to physical processes that we already know that we can't correctly represent in our simulation, either because of limited resources or simply because we don't know how. Hence we use a model for them.

It is this model level that we usually need to validate (as opposed to verify).

Seen from a different perspective, when we choose 1+2+3 we kind of pick items that for our application are well validated and don't need further validation, especially if we simply change the geometry.

Going to your example, you have CFD performed on a set of cases with different geometry and different turbulence models. An engineer gives for granted 1+2+3 (even if 2+3 might need a careful choice among different models) and is the only one in charge for the verifiable part (grid convergence, correct numerical schemes, etc.).

This leaves out the suitability of a certain turbulence model for a given geometry. You are not supposed to be a turbulence expert, so you either work in a field where there are already best practices for the choice of the turbulence model for a given kind of flow, or you need to come up with something.

If you need to validate a turbulence model you need experimental data for a case as close as possible. And you can't infer nothing from a model to the other.

In practical terms, this means:

1) Pick up the closest experimental data set for each case you need to work on. This also involves being able to identify flow patterns and mechanisms that might be represented differently by different turbulence models. Every pattern/mechanism would ideally need a different experimental reference. Of course, we can only do our best with what we have.

2) Test each turbulence model.

3) Pick up the one that performs best according to some measure which is up to you to define.

It is apparent, I hope, that you can't then change turbulence model afterward, because the turbulence model is what you are actually validating here. You can change the geometry, but as long as you are confident that the case still falls into the validated range of parameters. Rough example: airfoil at 0 and 5 degrees are probably in the same room, same airfoil at 15 degrees is probably not and needs a separate validation.