[Gkchpa]

Hello everyone, I am working on a project related to acoustic stimulation of rotating machines- Improvement of Fan acoustics ( sound reduction) compared with previously available data. and I am thinking about 2 way of approach

Making CFD simulation of the fan and taking the Acoustic pressure and want to convert into sound ( I have no idea about this )

or using directly some direct acoustic predicted software which is open

I need some suggestions regarding the conversion of data or any information about this topic, Anyone here who have a similar background or experience?
Thank you In advance

[Feecry]

Hi,
As far I am concerned, Fluent cannot give direct result on data that you can convert it to sound. You can do it indirectly by flow analysis thru turbulent. Most of the time, firstly you must basically perform below step:
1.The usual way is to perform flow simulation with Turbulent mode activate.
2. From the result, identify the region that have high turbulent.
3. The region that has high turbulent is associated with noise.
By the way, this is the simplest method to extract the noise due to flow. More assumption and input needed. You can give more attention to turbulent study.

[Feecry]

But still, if you interested in obtaining sound pressure, you can you Boundary Element Method. So far, this is mostly used for acoustic problem.

[sbaffini]

Actually, Fluent also has the Ffowcs Williams Hawkings Model (and most major commercial vendors should as well).

I don't have much experience on that, so you might start giving a look here https://www.afs.enea.it/project/nept...th/node230.htm

[LuckyTran]

The direct approach is to do a LES/DNS and get the instantaneous pressure fields. As we all know, LES/DNS is very computationally expensive and you must collect a lot of data to do the phase-averaging. How expensive and how much of course depends on the complexity of the problem.

How you want to show that the sound decreases will strongly influence the post processing. The entire spatially and temporally acoustic pressure can be obtained by phase-averaging. But most people never need this level of detail. You can get away with a simple DFT, still get the amplitude, and show that the spectrum has decreased for one case.


Then there is the aeroacoustic approach. You generally run some steady or unsteady RANS to get a background flow field and then feed this into an acoustic propagating wave model. I don't have knowledge about any free/Open-source aeroacoustic solvers but they are popular in commercial software (Fluent, Star-CCM, Wave6, COMSOL, and more). The folks over at Tu Delft have spent a long time developing an acoustic solver in OpenFOAM.

[Gkchpa]

Hello Lucky Tran, Thank you for your Reply. From the 2nd approach, you stated ( acoustic ), Is it like I go with CFD simulation using RANS model and then transfer data to some acoustic prediction or propagation software. or else can I Use FWHwilliam Hawkings ) analogy using ANSYS .

Thanks you
Best regards

[wangsen992]

Hi there,

I don't have any experience in actually modeling sound, but here is my take on it.

Noise for human, or sound, in general, ranges between frequency of 20 to 20000 Hz. What this means is the pressure fluctuation due to turbulence is within that same range. So assuming you have the pressure fluctuation time series from your simulation, you can take DFT to obtain the power spectrum for the noise amplitude.

In terms of using CFD to get pressure fluctuations, as you see you need to have a time step of 2pi/20 to 2pi/20000, which then sets constraints on your case set up. With this in mind, for practical purposes, running DNS is not an option. I would suggest using LES and obtain small scale fluctuation values from the subgrid model.

Above are some of the thoughts on the theoretical aspects for your issue and I hope it helps!

Cheers!
Sen

[Blanco]

Hi,


I'd like to add that the CFL<=1 constraint is still there and typically is the most stringent in most aeroacustic problems, so at the end you'll have a time-step which could be much lower than 1/20k... The local cell size in your mesh, in particular, should be chosen appropriately to resolve the flowfield up to the highest frequency of your interest (for both sound generation and sound propagation), and the cell size + the local velocity field will set your highest acceptable time-step. Mesh constraints in the boundary layer regions are typically so demanding that sometimes hybrid RANS+LES approach are suggested.

[Gkchpa]

Thank You, Blanco, for the suggestions, do you have any Idea about Using any acoustic analogy in Hybrid methods, for example, FW-H, Should I go for any Matlab functions? any Idea, How acoustic Analogy is used to convert this pressure fluctuation into Frequency

regards

[LuckyTran]

The acoustic analogy is the part that turns a flow-field into noise sources for the aeroacoustic equation which itself is a PDE. How this PDE gets solved is another implementation problem.


In Fluent, you would run RANS/URANS. Choose a surface to record the flow pressure in time, this becomes the noise source for the FW-H model in Fluent. Then you'd click some buttons to switch over to the FW-H model and then hit another run button. Specifically within Fluent, the FW-H approach spits out the time-series pressure on a farfield target surface. Fluent has builtin tools to take FFT's for you and give you a spectrum. You could also export the pressure into matlab and take the FFT yourself, among other things.

[Gkchpa]

Thank you, Lucky Tran, for explaining. I will look after this in Detail, Can I came back to you if I have any questions regarding this topic ?

Bests