I keep hearing that "CFD isn't good for acoustics" because the implicit solvers used "dampen acoustic waves out". Is this true? Does anyone have real experience of (for example) CFD-predicted turbocharger whistle or whine?

- Steve

It is true that most production-oriented CFD solvers do not handle acoustics well. There are several reasons for this. First, unless numerical viscosity can be strictly limited it tends to kill off or grossly alter the very modes that you are interested in. Second, the inclusion of any turbulence model will also have a pronounced impact on possible acoustic behavior, unless the turbulence model is designed properly. Third, boundary conditions need to be carefully implemented in that sloppy BCs can have a dramatic impact on what would otherwise be a good algorithm. Having said that, it is not impossible to use CFD for acoustic problems. But you need to be very careful in your validation process to ensure that what you get is a true reflection of the physics.

Yes you are right. The problem is just the turbulence model. Infact if you use a 2-equation model, some frequencies cannot be captured. Instead, I strongly suggest LES.

Davide

Computational Aero Acoustic is a very difficult subject.

I ve been working on it for several years (almost 10) for 2 different major companies. One in the automotive industry, the other in the aerospace industry. We tested different solutions (commercial and/or academic) and always came to the conclusion the art is not mature yet.

let me give you just a few hints :

-lets assume the typical pressures for your application are about 1atm : 1e5Pa

-and assume you are looking for an acoustic prediction that is about 15dB accurate (not very good), then the corresponding pressure fluctuation you are looking for is about 1.12e-4 Pa.

Compared with the 'average' pressure , the ratio is about 9e8 !!!

in other words, the minimum accuracy you are looking for is 1e-7 % !!!!

No code I know is capable of this, (specially) not even those who claim they can ! Obviously the code has to be compiled using double precision but that is not enough.

A (very) low troncation error is also required which means high order numerical schemes. This is not to say any high order numerical scheme can be used.

Also needed is an accurate turbulence model. LES will probably provide the best candidates, but I dont think a Smagorinsky model is enough (overdissipative for small scales). I would rather bet on a mixed scale , or dynamic mixed scale model. (Be aware that troncation error and subgrid scale modelling are actually deeply linked : some models can give corrections that are lower than the troncation error, some others can even be rewritten as a troncation error, and you can even find people using no explicit subgrid scale model but only the truncation error as a turbulence model)

Also needed is an accurate control of the boundary conditions. (even experimentally, it is not straighfoward)

Anyway, best universities in the world are still working very hard on the subject. I dont think any commercial code already have included what these universities dont even have developped.

The 'best' I could see, needed weeks of simulations on the biggest computers in the world and was not even valid over a few hundred hertz, which in term of acoustic is very low.

We generally obtained over-estimation of the main frequency (f) of interest of the problem we were adressing plus a collection of harmonics (2f 3f 4f ....) (a/ NS is a quadratic equation b/ the numerical scheme has to be carrefully defined) (e.g. the Von Karman frequency (see also Strouhal number) and its harmonics or the blade passing frequency and its harmonics) All other frequencies were either under-estimated (possibly due to an inaccurate turbulence model) either overestimated (possibly due to high truncation error).

I hope I m not too discouraging but at least this should help you understand why CAA is so difficult.

[emreg]

Originally Posted by Eric
;31158

Computational Aero Acoustic is a very difficult subject.

I ve been working on it for several years (almost 10) for 2 different major companies. One in the automotive industry, the other in the aerospace industry. We tested different solutions (commercial and/or academic) and always came to the conclusion the art is not mature yet.

let me give you just a few hints :

-lets assume the typical pressures for your application are about 1atm : 1e5Pa

-and assume you are looking for an acoustic prediction that is about 15dB accurate (not very good), then the corresponding pressure fluctuation you are looking for is about 1.12e-4 Pa.

Compared with the 'average' pressure , the ratio is about 9e8 !!!

in other words, the minimum accuracy you are looking for is 1e-7 % !!!!

No code I know is capable of this, (specially) not even those who claim they can ! Obviously the code has to be compiled using double precision but that is not enough.

A (very) low troncation error is also required which means high order numerical schemes. This is not to say any high order numerical scheme can be used.

Also needed is an accurate turbulence model. LES will probably provide the best candidates, but I dont think a Smagorinsky model is enough (overdissipative for small scales). I would rather bet on a mixed scale , or dynamic mixed scale model. (Be aware that troncation error and subgrid scale modelling are actually deeply linked : some models can give corrections that are lower than the troncation error, some others can even be rewritten as a troncation error, and you can even find people using no explicit subgrid scale model but only the truncation error as a turbulence model)

Also needed is an accurate control of the boundary conditions. (even experimentally, it is not straighfoward)

Anyway, best universities in the world are still working very hard on the subject. I dont think any commercial code already have included what these universities dont even have developped.

The 'best' I could see, needed weeks of simulations on the biggest computers in the world and was not even valid over a few hundred hertz, which in term of acoustic is very low.

We generally obtained over-estimation of the main frequency (f) of interest of the problem we were adressing plus a collection of harmonics (2f 3f 4f ....) (a/ NS is a quadratic equation b/ the numerical scheme has to be carrefully defined) (e.g. the Von Karman frequency (see also Strouhal number) and its harmonics or the blade passing frequency and its harmonics) All other frequencies were either under-estimated (possibly due to an inaccurate turbulence model) either overestimated (possibly due to high truncation error).

I hope I m not too discouraging but at least this should help you understand why CAA is so difficult.

Discouraging enough however much time is passed after you wrote the message. How about the situation for FEM and BEM. Is it possible to perform CFD calculations and upload the results to the FEM/BEM software to investigate acoustic calculations?