Finding nat freq mode using experiment.

[APH]

I need to get rid of noise "ringing" on aluminum casting from outside vibration at 2kHz. I could do this 2 ways, raise the natural frequency of the casting above 2 kHz (adding mass, stiffening ribs, etc) or place isolation material like rubber/foam sheet inside the casting to absorb the energy. The second approach isn't practical from manufacturing stand point. I'm left with the first approach.

My question is, how do I find out that the mode of frequency at 2kHz in FEA study is actually the 2 kHz mode frequecny that I'm seing during testing?. I'm using frequency analysis in COSMOS. (no dynamic response analysis)

I try to add stiffning braces locally in the area that is needed by looking at specific mode that gives me the 2kHz freq, but I'm not sure which mode to use.

I would appreciate your respond.

Thank you
APH

 

[GregLocock]

You need to correlate your model properly using an experimental modal analysis of the real casting. Then you'll know which mode shape is the problem.

However, in this specific instance I'd guess you are running 4-6 mm of aluminum, in a curved structure. You may be lucky, and be able to move the frequency away from a narrow band excitation, enough to get rid of the problem in service. What is more likely is that the exciation is of variable frequency and you'll just move it to a different speed range.

That frequency is typical of aluminium castings, due to the coincidence frequency (speed of bending waves in aluminium = speed of sound in air, giving 3 dB gain).

In order to attenuate across a broad band then generally thickening the casting is more effective than ribs, in terms of dB/kg.

So, is your excitation at a single frequency? How many Hz do you have to shift the mode shape?



Cheers

Greg Locock

 

[APH]

Greg,

The thickness of the aluminum casting is .125". The radius in the corner is very small, .060-.125" max.

The casting have 650 hz, and 950 hz mode before it hits another mode at 2Khz. We succesfully attenuated the 650 hz and 950 hz by putting rubber mount on the casting. We tried different rubber mount, still the 2kHz showed up. I then tried to fix the problem by adding thickness/ribs locally to gain stifness, thus higher natural freq to excite. This made it worse, now I got other mode at different frequency (1200 kHz). I guess in order to shift the natural frequency of the same mode, you have to increase thickness uniformly?. This would creat weight issue to me.

I need to pass 3kHz in order for the electronic inside the casting to work.

Back to your first paragraph. How do you correlate your model properly using an experimental modal analysis of the real casting in the vibe table?

Thanks
APH

 

[GregLocock]

2 kHz to 3 kHz is a fair ask. What's the lower limit of the excitation (I assume it is below 2 kHz)?

I hadn't realised it wasn't an acoustic problem, so you can ignore everything I wrote about coincidence frequency.

Correlating FEA models is a black art.

If you have a good modal test result then you can try and do a modal assurance criterion (least squares fit) between the experimental and analytical modes. That's a bit optimistic. More usually people just use a descrition of each mode, and if the first few are each within 20% of the experimental result then you've got a good model. If that doesn't sound very scientific, well, such is life.

Things you might try - isolating the electronics, building a locally heavy and stiff raft for them to sit on, changing the casting material. Magnesium has more damping.

I don't think there is any intrinsic reason why you can't isoalte for 2 kHz, you need to be careful with the high frequency characteristics of the rubber - high damping rubbers tend to get worse at high frequency.



Cheers

Greg Locock

 

[APH]

One think that I know will fix the problem is putting hot melt(potting it) around the electronic components (basically similar to what you have said). This methods is expensive (laborious). I have also tried putting a rubber strip on the side of the walls, this also fixed the issue. I haven't tried to use magnesium, which is a possibility. Ideally we want to change the casting, this way it doesn't introduce more parts. I've squeking this issues for 3 months already to save $2.50 on cost...hmm..I wonder...

The low limit of excitation is 500Hz. If you don't mind, what is modal assurance criterion.

Thanks!
APH

 

[GregLocock]

MAC is doing a least squares fit for the mode shapes. So you have your analytical eigenvectors, each normalised to 1 at the maximum (or whatever) and your experimental mode shapes normalised in the same way, for the same subset of measurement points.

Then do a least squares fit. If you are very lucky or very good or both, one analytical mode will have a good fit to one experimental mode, for all your modes.

Haven't seen it myself...

google for the precise method.


Cheers

Greg Locock

 

[SrinivasAluri]

Formula for MAC from DJ Ewins Modal Testing

{phiX} = Experimental mode shape (vector)

{phiA} = Theoretically predicted or analytical mode shape (vector)

{phiX}T is transpose of {phiX}

MAC = |{phiX}T*{phiA}|^2 / [{phiX}T*{phiX})*({phiA}T*{phiA})]

Hope this formula is useful.

 

[ibrahimucak]

Excite the system with an impulse hammer and get the results. From the transfer functions you can plot the mode shapes and operational deflection shapes(ODS). If you can excite the system with a bandwidth of more than 2 kHz, The Operational deflection shapes directly show how the system moves at the specified frequency. If the bandwidth of excitation is not that large, then you can validate the FE model by the first several natural frequencies and mode shapes you gathered from the experiments.

Hope this helps.

Ibrahim

 

[GregLocock]

It might help but it is (at best) pretty confusing.

ODS are obtained by using self excitation of the running system - in essence they are an animation of the vibration at each frequency while it is in use.

Some software vendors call an animation of the FRF at each frequency an ODS, I think they could have come up with a better, unique, term for that.

You will find it is at least twice as difficult to correlate an FE model with (true) ODS data as with a proper modal analysis.



Cheers

Greg Locock