[lnk]

Hi,

May I ask for a full answer to this basic question? Why we use turbulence modeling?

Thanks!

Best,
lnk

[yonchong]

Main reason is that we use meshes which does not have fine enough resolutions to capture all the fluid motion. If you have a fine enough mesh then you don't have to use any turbulence modelling and it is call Direct Numerical Simulation (DNS). Anything short on that you need some kind of modelling to count for the missing resolution.

[FMDenaro]

as already stated, using turbulence modelling has an historical motivation, when computers were so poor that the computational grids were unable to solve for the whole range of scales...

Now, DNS can be a chance in some problems, at least of small/average scales. However, for industrial applications is very often prefered using turbulence modelling in CFD prediction for very rapid evaluations ... for them, using DNS is a tool to expensive computationally, what is more even if they use DNS very often the consequent database is too large to be managed.

Turbulence modelling will continue survive for long time ... but we are slowly swithing from RANS/URANS to LES formulations...

[flotus1]

For almost every flow with practical relevance, it is simply impossible (and will be for the next few decades) to perform a DNS, e.g. to resolve all turbulent scales without modeling assumptions.
The limitation is the computational cost, which scales with Re^(9/4).
Additionally, engineers often only require Reynolds averaged values of the flow field, thus DNS or LES methods would simply be an overkill, even if applicable.

[leflix]

Within less than 30 years, turbulence modeling will be completely useless since computers will be powerful enough to handle all range of scales until kolmogorov scale with DNS.
You will be able to solve the flow over a F17 at mach 3 on your play station during your breakfast
All turbulence modelers have only jobs for 30 years left , so do not start in this field if you don't want to turn jobless soon

[FMDenaro]

Originally Posted by leflix

Within less than 30 years, turbulence modeling will be completely useless since computers will be powerful enough to handle all range of scales until kolmogorov scale with DNS.
You will be able to solve the flow over a F17 at mach 3 on your play station during your breakfast
All turbulence modelers have only jobs for 30 years left

That's enough for me!!

[leflix]

At present time PS4 is available, you just have to wait PS9 or Iphone13 and it will be done

[cfdnewbie]

Originally Posted by leflix

Within less than 30 years, turbulence modeling will be completely useless since computers will be powerful enough to handle all range of scales until kolmogorov scale with DNS.
You will be able to solve the flow over a F17 at mach 3 on your play station during your breakfast
All turbulence modelers have only jobs for 30 years left , so do not start in this field if you don't want to turn jobless soon

I am pretty sure that wont be the case, sadly enough... just increase Re by an order of magnitude, and you will blow up the systems but of course, the Re number we will be able to resolve will indeed increase drastically

[yonchong]

Originally Posted by leflix

Within less than 30 years, turbulence modeling will be completely useless since computers will be powerful enough to handle all range of scales until kolmogorov scale with DNS.
You will be able to solve the flow over a F17 at mach 3 on your play station during your breakfast
All turbulence modelers have only jobs for 30 years left , so do not start in this field if you don't want to turn jobless soon

I should have retired by then so I don't care.

[leflix]

Originally Posted by cfdnewbie

I am pretty sure that wont be the case, sadly enough...

bouhh you are defeatist cfdnewbie ! stay positive!!!
We shall continue this discussion in 30 years and you will see that I was right !

In 1950 the computers of that time needed 70 hours to compute the first 2000 digits of pi number. 50 years later a japanese guy Shigeru Kondo computed 135 millions of digits in 2h45mn on a simple pentium III...
I guess if we had said in 1950 to Von Neumann that we could compute 135 milllions of digits of pi in 50 years later in less than 3h ,while he was struggling with only 2000 digits he would have required that one send you in a psychiatric hospital.

The record by the same guy is 5000 billions of digits and it required 90 days of computations. If we had said this to Von Neuman he would have lost its mind immediatly
so you see...just be patient...

[leflix]

Originally Posted by yonchong

I should have retired by then so I don't care.

so you are safe !!!you can go on

[Ford Prefect]

While I enjoy this discussion I am a bit more cautious. It seems to me that the time to solve a CFD problem has stayed more or less constant over time. With more computational power we just tend to add more cells and physics. Also, the information in DNS might not even be useful for some industrial applications, just think about how to get all boundary data for complex systems. I'd use a RANS model any day over DNS if it produced tolerable answers, well at least until the computers are so powerful that I won't even notice the difference in calculation time. Less is more, right?

[leflix]

Originally Posted by Ford Prefect

It seems to me that the time to solve a CFD problem has stayed more or less constant over time. With more computational power we just tend to add more cells and physics.

That is so true!!!! We are never satisfied and we always want more when more is available! that's completely human!!!
But a time will come when our more secrets will, could be technically satisfied...Also over a limit it won't bring anything to keep increasing the number of nodes.

[FMDenaro]

several years ago I ran simulations of 2d flows on a RISC-based computer and waited for days, now I run simulation of 3d flows on I7-based computer and ... wait for days!

[leflix]

Originally Posted by FMDenaro

several years ago I ran simulations of 2d flows on a RISC-based computer and waited for days, now I run simulation of 3d flows on I7-based computer and ... wait for days!

yes I know this problem!! It's exactly what was saying Fordperfect
But in several decades you could perform ten times bigger cases on a massive computer with 10.000 I7 that you should have bought for few dollars in your super market while making your shopping for dinner

[FMDenaro]

Originally Posted by leflix

yes I know this problem!! It's exactly what was saying Fordperfect
But in several decades you could perform ten times bigger cases on a massive computer with 10.000 I7 that you should have bought for few dollars in your super market while making your shopping for dinner

At that time I will mount them on my wheelchair ...

[leflix]

Originally Posted by FMDenaro

At that time I will mount them on my wheelchair ...

No because at that time wheelchairs will be old-fashioned Everyone could have an exoskeleton !!

[truffaldino]

Originally Posted by leflix

No because at that time wheelchairs will be old-fashioned Everyone could have an exoskeleton !!

... and chip enchanced brain, so everyone will do DNS in his/her own imagination

[cfdnewbie]

you guys having a party here? smoking some herbs?

[FMDenaro]

Originally Posted by truffaldino

... and chip enchanced brain, so everyone will do DNS in his/her own imagination

... but ...this already happens now !