Coupled Fluid-Structure Eigen Analysis

[kevinarden]

For the HelmHoltz Structure case
https://github.com/ElmerCSC/elmerfem/tr ... zStructure

I test a beam in fluid, 2D, the frequency should be 81 Hz in air

with fluid properties of air the frequency is 84.9 Hz
with fluid properties of water the frequency is 84.9 Hz, it should be lower, but it is exactly the same.

The pressure is calculated in the fluid, and the pressure could be used in a load, but an eigenvalue uses no load.

Therefore the hydrodynamic added mass of the fluid is not added to the eigenvalue solver. The fluid-structure coupling appears to be pressure only.

wetcase.sif

(2.88 KiB) Downloaded 5 times

2d.zip

(176.75 KiB) Downloaded 5 times

[kevinarden]

I tested the beam case in transient by tweaking the top of the beam the letting it vibrate freely. I used the FSI obstacle tutorial as a guide.
The magnitude of the motion in water is much different than in air, but the frequency is unchanged. So FSI means loading only not mass.

vibes.png

(79.14 KiB) Not downloaded yet

[eschenk]

Thanks Kevin,

Looking at the case you did, this is a weakly coupled solution, yes?

I ask because, with the strongly coupled solution on my guitar test case, I do see a shift in frequencies in the structure by shifting the density of the fluid. Without the fluid coupling, the top of the guitar's first mode is around 155 Hz in my current model. With the coupled air, it shifts up to about 199Hz. (Since I am scanning, it's hard to be more precise about the exact frequencies here without scanning a smaller time steps)

This is roughly what I wold expect, although I think I don't have the overall mass/stiffness of the model correct yet.

This does make me think an interesting test case might run the transient system with an impulse to the bridge of the guitar and see what it looks like.

Can you share the .sif file that tweaked the beam? I think I roughly undertand what you did, but I'm unsure how to apply the transient force.
Also, How did you generate that vibration graph?

Cheers,

Eric

[kevinarden]

wetcase2.sif

(3.91 KiB) Downloaded 5 times

That final plot was actually from spreadsheet. To get to spreedsheet I go into paraview.
You can actually plot nodal data as a function of step directly in paraview. Using interactive select node and then plot selected data over time.
But if I want to have better control or plot results from multiple analyses. I save the paraview plot data to a csv fil.

You can select which areas you want it to dump to.

[eschenk]

Thanks. I see how you set this up.

I've started a transient run with a blow to the top of the guitar saddle in my model, which interestingly is exactly how one recovers the modes with microphone and an FFT analysis on a real instrument.

I'm running a simulation at time steps of 0.0001, or 10k Hz, which should give me enough bandwidth / resolution to see what is happening in the 0-500 Hz range or so.

I'll also see if I can drop a probe onto a node or boundary somewhere and recover a signal that I can run through an FFT, which might give me an alternative way to recover the joint eigen frequencies that I care about.

Also, there is some promise that might be able to use this techqique to recover signal faster than a scanning run. A single thread run of 1000 steps looks like it will take about 12 hrs, which might be enough to get some intersting numbers out to 400 Hz or so.

If I run this in parallel, I can probably take that time down considerably. I was already running the scan with 8 paralell runs for subsections of the scan space. Probably I'm memory limited, rather than CPU limited, as I only have about 32 GB avaiable under WSL for this run right now, but I have 32 cores I could be running with, so I can likely get a few more runs in parallel going.

I should also take a look and see if using the prior time step solution to precondition the next step might speed things up a bit.

Cheers,

Eric

[kevinarden]

There is a contributed case index on the board, your case would make a great contribution.

[eschenk]

I'll be happy to contribute a case once I get this working well enough.

I've been trying transient simulations and ran into a few problems with the model set up I was using.
I found that only having air inside the box is leading the transient simulation to "inflate" the box at the start of the simulation.
I've tried to build now a model with the guitar structure embedded at the center of a sphere of air, but this greatly increased the size of the simulation. I'm starting now to get some runs completed and successfully extract some signal with an FFT, but I'm not convinced now this is any faster than the frequency space scanning I was doing previously. That said, I was probably always going to have to do this in order to generate the acoustic response curve I also want, so that is progress.

Right now I'm trying to learn how to do better spatial divisions. Ideally I'd model the air with a uniform fixed cell size, and use a projection onto the structure at the fluid structure/boundary, but I have only been able to get forces to transfer sucesfully when I have joint mesh boundaries between the structure and the air. Are there any good examples of independent geometry densities between structure and fluid I can look at?

In any case, more updates as I make (slow) progress in my spare time.

Cheers,

Eric

[kevinarden]

"independent geometry densities between structure and fluid I can look at"

In Elmer this requires the use of mortar boundary conditions. With mortar you have a master and slave so the mesh does not have to conform.

Elmer Solver Manual Section 9.2

https://www.nic.funet.fi/pub/sci/physic ... Manual.pdf

example of mortar

https://github.com/ElmerCSC/elmerfem/tr ... ngBeamFlow

[raback]

Hi

Interesting discussions here. I am travelling so cannot spent time on this now. However, I can confirm that some real-world benchmarks have been solved a few years ago. These test cases are reflections of that work which itself was not made public.

There should be clear effect of density. So that is odd.

I thing I didn't immediately see is that I guess the system is open. The air should be able to get out. The results are very different if the system is closed. For a guitar maybe give some impedance BCs at the hole. Ideally one should extend the acoustic field outside the hole since certainly the outside waves are not well developed immediately.

-Peter

[kevinarden]

I went to the test cases
HelmholtStructure
HelmholtStructure2
HelmholtStructure3

and change the density from 1 to 1000, and the sound speed from 300 to 1443.

In each one there was no difference in the structural result, but the pressure calculation changed.