[oj.bulmer]

I work in industry and we use CFD studies to assess the pressure drop characteristics of our products. But lately there is something that is bothering me.

Often clients are interested in the guarantee of the pressure drop values we provide. Now we have done experimental testing of few of our products and typically our CFD results are within 20% of experimental pressure drop values. But then it is not possible to benchmark every product and every size of of that product, and which is the reason CFD is used to fill this gap. But then, I was wondering, if a product of say 6" size is bench-marked with CFD, how to assess whether the same accuracy can be expected in say 30" size of the same product. Or, for that matter any other type of the product.

I have never worked in consultancy, but I am curious to know, if CFD community typically guarantees the results they give for all sizes of the product, based on representative bench-marking done for a single size of the product. Or if the commitment about the accuracy is expected, how it is handled.

[FMDenaro]

Originally Posted by oj.bulmer

I work in industry and we use CFD studies to assess the pressure drop characteristics of our products. But lately there is something that is bothering me.

Often clients are interested in the guarantee of the pressure drop values we provide. Now we have done experimental testing of few of our products and typically our CFD results are within 20% of experimental pressure drop values. But then it is not possible to benchmark every product and every size of of that product, and which is the reason CFD is used to fill this gap. But then, I was wondering, if a product of say 6" size is bench-marked with CFD, how to assess whether the same accuracy can be expected in say 30" size of the same product. Or, for that matter any other type of the product.

I have never worked in consultancy, but I am curious to know, if CFD community typically guarantees the results they give for all sizes of the product, based on representative bench-marking done for a single size of the product. Or if the commitment about the accuracy is expected, how it is handled.

Size is not the only relevant parameter... let me do an example for a flow around a cylinder of radius R. The flow is characterized by the non-dimensional Reynolds number Re=U*L/ni, so you can fix L=2R as reference lenght. Now, if the group U/ni can be varied in such a way that Re=constant for different R sizes, you have the same flow features and the same quality in the CFD solution. But if the Re increases, the quality of the CFD solution can be much more difficult to be ensured, at present Direct Numerical Simulation for complex geometries and flows is possible at moderate Re number. Using turbulence model you have more parameters that introduce problems in guarantee the quality of the solution

[oj.bulmer]

It is probably easy to assess what conditions would imply similar non dimensional results for simple geometries, but for an industrial product, like say a filtration equipment, it is difficult to judge these criteria.

DNS (if done correctly) would be obviously more accurate than typical RANS models given that the closure assumptions are not involved, but then, one needs a humongous computational power to do this, along with patience and time to find the correct lengthscales/timescales. This is not possible in industry where there is a budget for everything and it needs to be justified. And most importantly, results are needed yesterday!

The trickiest part is, when management thinks.. "Oh well, we have invested in expensive CFD packages, man-hours of CFD engineers and experimental testing. The CFD guys take 2-3 days for every result. So we should get results with 0% inaccuracy, otherwise what is the point in spending such money!"

They do not understand the uncertainties involved, nor do they posses the CFD-vocabulary/patience to understand the explanation given. So I was curious to know how others handle this

[FMDenaro]

Originally Posted by oj.bulmer

It is probably easy to assess what conditions would imply similar non dimensional results for simple geometries, but for an industrial product, like say a filtration equipment, it is difficult to judge these criteria.

DNS (if done correctly) would be obviously more accurate than typical RANS models given that the closure assumptions are not involved, but then, one needs a humongous computational power to do this, along with patience and time to find the correct lengthscales/timescales. This is not possible in industry where there is a budget for everything and it needs to be justified. And most importantly, results are needed yesterday!

The trickiest part is, when management thinks.. "Oh well, we have invested in expensive CFD packages, man-hours of CFD engineers and experimental testing. The CFD guys take 2-3 days for every result. So we should get results with 0% inaccuracy, otherwise what is the point in spending such money!"

They do not understand the uncertainties involved, nor do they posses the CFD-vocabulary/patience to understand the explanation given. So I was curious to know how others handle this

Yes! This is the real problem... speaking to people that have no experience in CFD applications!!
Unfortunately, RANS in industry is still the standard and you cannot reach 0% inaccuracy even if you produce grid-refinement study...

[sbaffini]

In my experience, even if i was never involved in actual consultancy, this is not the usual practice of serious people. But simply because it's impossible to give such ensurance. Notice also that the actual developers of turbulence models also wouldn't.

What should enter the CFD practice is the Uncertainty Quantification, but that usually means running several tens, or hundreds, simulations, to only understand how a single turbulence model affects a single flow case. No way it is gonna be everyday practice but, with that, you could give confidence intervals.

However, besides the turbulence modeling effects (which are huge in practical CFD), this is not different from the experimental practice. No wind-tunnelist would ever give you ensurance if your Re number is different from the one tested. Without direct experimentation he can, at most, give you an advice based on experience. That's all.

Also, notice that most consultancy firms use black-box commercial CFD tools. They cannot give ensurance over something which is not produced by them.

In practice this might also be applicable to the users of OpenSource tools, which i never found to be more prepared on CFD or Fluid Dynamics than generally experienced people not using such codes. It is actually more probable that they are more programmers than actual CFD experts (otherwise they should already have their own code instead of putting efforts in recompiling/repacking OpenFoam in all the possible flavors).

[SB123]

there's also the grid convergence index and richardson extrapolation, but that will only estimate the error in your grid resolution, not the error inherint in your models ie turbulence and what not. and this also requires you to do a grid study, which i'm sure becomes difficult in industry

[garrison]

Originally Posted by oj.bulmer

I work in industry and we use CFD studies to assess the pressure drop characteristics of our products. But lately there is something that is bothering me.

Often clients are interested in the guarantee of the pressure drop values we provide. Now we have done experimental testing of few of our products and typically our CFD results are within 20% of experimental pressure drop values. But then it is not possible to benchmark every product and every size of of that product, and which is the reason CFD is used to fill this gap. But then, I was wondering, if a product of say 6" size is bench-marked with CFD, how to assess whether the same accuracy can be expected in say 30" size of the same product. Or, for that matter any other type of the product.

I have never worked in consultancy, but I am curious to know, if CFD community typically guarantees the results they give for all sizes of the product, based on representative bench-marking done for a single size of the product. Or if the commitment about the accuracy is expected, how it is handled.

In many cases I've seen, people just enjoy those colorful pictures plotted by commercial packages... They don't even realize results can be wrong.. Or, engineers need a number say pressure drop and they can't calculate one without CFD, even CFD will give them a wrong number.

[mehulkumar]

Hi All,

You can read relative answer at 1.4.4 Potential Problems in the book "Computational Fluid Dynamics in Industrial Combustion".

Regards,
Mehulkumar

[FMDenaro]

Some paragraph about industrial application with LES is in the Book of Sagaut