Leob, You are welcome.
Good work. Doesn't it feel great having a clue regarding what should actually happen to the modes when you move the constraints?
As far as what you should be looking for, well, that's the big question right?
Two notes:
- Regarding the last mode you showed (and below): That is indeed a valid mode but it is not a "transverse" mode. That is, its displacements are in the X and Y direction, and not the Z. You can see this if you use the animation tool. I suppose this mode might radiate from the sides a little bit, but I think you should basically ignore this mode and any others of this sort. These modes are often fairly conspicuous once you have seen them a few times, due to the relative scarcity of blue in them. I don't think they contribute anything meaningful to the sound.
- I still recommend that you determine real numbers for the Elastic Modulus and poisson ratio for your plate (assuming you have not already done so). Now that you have the LISA model, to do so is really a very simple exercise. All you need to do is to experimentally find the natural frequencies of the first two modes of your plate in the "free" (no constraints) condition. You can do this this "tap testing" using the RTA mode of REW and close mic-ing. Then, first adjust the value used in the model for Young's Modulus until the LISA prediction matches your tap test result for the mode on the left below. Then, adjust the value used in the model for poisson ratio until the LISA prediction matches your tap test result for the mode on the right below. When the model predictions match both tap test results, you have identified the real properties of your plate and can use those to model any other size and shape of plate with any other constraints.
https://www.diyaudio.com/community/...r-dml-design-and-analysis.383567/post-7210940
Eric
Last edited:
Looking at Lisa and Mecway à bit more, I admit to being puzzled as to where these frequency ranges are coming from. They seem not to be fundamental or harmonics, but random values, and some of the vibration movements are truly weird. Shimmy shaking in all sorts of directions... Have I been hitting the grog too much?
And can you select a target frequency range?...Adding more modes just tracks more higher frequency modes,, again seemingly random
Eucy
And can you select a target frequency range?...Adding more modes just tracks more higher frequency modes,, again seemingly random
Eucy
Eucy! Wow, I never heard of Mecway, I'm going to have to check that out!
I'm not quite sure what you mean by "where these frequency ranges come from". Can you explain? Do you mean where the modal frequencies come from? The answer to that is that they come from the physics of the plate: determined by the plate's size, properties, aspect ratio and boundary conditions.
Regarding harmonics: Yes, that is correct. Unlike a tensioned string (guitar, etc) the natural frequencies of plates are not "harmonics" of the fundamental. For a tensioned string, all the overtones (natural frequencies) are harmonics, that is, their frequencies are at integer multiples of the frequency of the fundamental. But the same thing is absolutely not true for plates. The "overtones" of a plate are not integer multiples of the fundamental, but rather occur at (seemingly) random intervals. If all the natural frequencies were harmonics, there would be very little need or value to modeling, as a single measurement of the fundamental would tell us everything we needed to know.
Eric
Hello Eric,
Always nice to see by testing what is coming from the theory.
In addition I am always glad to see the good use to do of your impedance measurement jig.
I guess from the my few tests like your that the frequencies shown are not high. Below 1kHz?
In an other hand, the DML is said directive above the coincidence frequency which is not expected below 1kHz...
So is it the panel which beaming or an effect of some standing waves in the room. I imagine here the tests are not far from the Schroeder frequency of the room...
If you run new sweeps, could you share a 0° (axis) one and a one between 60 and 80° (REW native file) to see if we can detect the coincidence frequency. I am in the opinion that it might be possible with the wavelet spectrogram with the FR at 0s (REW offers a function to make a 2D slice of the 3D spectrogram)
Christian
Indeed! I had no idea it was possible to get decent simulation without more work analysing the material and learning advanced or expensive software.
Thinking a bit more about how to take the simulation results into account for placement of the constraints, something that struck me is that pushing the first mode down below the first mode for an unconstrained plate is probably not very useful. At least in the plate with constraints on the long sides, while the amplitude around 80-90Hz was had a bump, it was still too low to be useful.
Perhaps pushing down the lowest mode is likely to result in the kind of response I was seeing, with the first few modes being bumpy and getting overall lower mid? And consequently good constraints have a first mode close to the unconstrained plate?
Looking at the sims I have for free plate, the first mode I have is at 100Hz and looks like a valid mode:
For the plate with constraints on the long side, first mode is at 100Hz as well, however it seems transverse?
So the first actual mode for the long side supported plate would be 131Hz?
For the plate with centered contraints, again first mode is transverse?
So the first "real" mode is at 144Hz:
And in Erics original simulation of the plate with the bad response, first mode was a valid mode at 96Hz:
So the plate with the bad response has first mode closest to the free plate, but also the lowest mode and even a bit lower than for the free plate. It seems natural that constraining the plate can move modes upward, a bit like a drum skin that will become higher in pitch the tighter it is, but that pushing modes down is due to interference caused by the constraints. Pure speculations though.
Anyway it seems like moving the suspension points will not have the effect I hoped for to make the response go down deeper might not work out since I did not consider the first mode being transverse. It could lower first mode to 131 from 144, but not to 100Hz.
For the plate with constraints on the long side, first mode is at 100Hz as well, however it seems transverse?
So the first actual mode for the long side supported plate would be 131Hz?
For the plate with centered contraints, again first mode is transverse?
So the first "real" mode is at 144Hz:
And in Erics original simulation of the plate with the bad response, first mode was a valid mode at 96Hz:
So the plate with the bad response has first mode closest to the free plate, but also the lowest mode and even a bit lower than for the free plate. It seems natural that constraining the plate can move modes upward, a bit like a drum skin that will become higher in pitch the tighter it is, but that pushing modes down is due to interference caused by the constraints. Pure speculations though.
Anyway it seems like moving the suspension points will not have the effect I hoped for to make the response go down deeper might not work out since I did not consider the first mode being transverse. It could lower first mode to 131 from 144, but not to 100Hz.
Attachments
Christian,
I'm thinking now that the variation in SPL that I observed was indeed due to room effects rather than lobing, as you suggested. I put the speaker on a turntable and rotated it slowly from -90 to +90 degrees and didn't sense anywhere near the variation in SPL that I had observed with the speaker stationary and just moving my head around to different places.
I did some sweeps for you at various angles at the same time. I'm on a trip now for the week but will share the files with you when I get back home next week. How do you think the CF might be revealed in the Spectrogram slice at 0 seconds?
Eric
@Leob
I wanted to clarify my earlier comment about how the FEA may display modes that you should probably ignore. The modes I'm talking about in that case are ones for which the displacements are all in the XY plane, and not the Z direction. These are modes that, if you use the "view deformed" tool and use "XY View" they will not appear as rectangles. The reason I suggest to ignore such modes is that surface that would be "pumping" the air in this case would just the plate edges, which have only a very small area in comparison to the face of the plate. Especially for plates that are only a few mm thick. But even for relatively thick plates like 25 mm EPS, the edge surface area is quite small compared to the face surface area.
One example is this mode, which appears as mode 21 in the example I did in the LISA guide.
And I think I spoke too soon when I suggested that the mode shown below from your model was one of these modes. In retrospect, I think I was wrong about that particular mode. I think now that it's a normal "transverse" mode and not an "in plane" mode.
Eric
I wanted to clarify my earlier comment about how the FEA may display modes that you should probably ignore. The modes I'm talking about in that case are ones for which the displacements are all in the XY plane, and not the Z direction. These are modes that, if you use the "view deformed" tool and use "XY View" they will not appear as rectangles. The reason I suggest to ignore such modes is that surface that would be "pumping" the air in this case would just the plate edges, which have only a very small area in comparison to the face of the plate. Especially for plates that are only a few mm thick. But even for relatively thick plates like 25 mm EPS, the edge surface area is quite small compared to the face surface area.
One example is this mode, which appears as mode 21 in the example I did in the LISA guide.
And I think I spoke too soon when I suggested that the mode shown below from your model was one of these modes. In retrospect, I think I was wrong about that particular mode. I think now that it's a normal "transverse" mode and not an "in plane" mode.
Eric
Leob,
I was trying to recreate the results you shared in that post, but couldn't quite come up with a match. Can you share more detail of that result? What was the size and location of the suspension points you used? And what were the plate dimensions and properties? It's not that I suspect you did anything wrong, or anything like that. I'm just curious why I can't reproduce it. I can get results looking really similar to the 2nd to 9 modes, but not the 1st and 10th.
Eric
Eric
I've been mulling over forced vibration simulation.
If a plate was meshed into 25 mm squares and a transient load modelled to make a square wave approximation of a sinusoid, then applied to one of the 25x25 faces for a few seconds, then a transient modal analysis carried out, would it be worth the effort. ?
The load would have to be edited for each frequency change.
Thoughts?
Eucy
I've been mulling over forced vibration simulation.
If a plate was meshed into 25 mm squares and a transient load modelled to make a square wave approximation of a sinusoid, then applied to one of the 25x25 faces for a few seconds, then a transient modal analysis carried out, would it be worth the effort. ?
The load would have to be edited for each frequency change.
Thoughts?
Eucy
Hello Eric,
The lobbing of the DML is expected by the theory.
There is a trace of it in the Tectonic papers. Those measurements are from an anechoic chamber.
When we get a measurement in a room with REW with a standard FFT, the observation time is defined by a window. If the window it "too long", some reflections will enter into it which will mask the CF. If the window is set to reject the reflection, the frequency resolution is then reduced like by a smoothing.
The principle of wavelets is a bit different. The IR is analyzed thanks to a specific signal. For all the frequencies, this shape of this signal is the same. It is the duration which changes. The general shape is defined by a so called "mother wavelet" which is scaled (compressed in time or dilated).
The consequence of that is observation time decreases when the frequency increases. For a wavelet of 1/6 of octave (a wavelet is like a bandpass filter), the observation window is maybe 5 or 6 periods (to be checked). For a 1/3 of octave, it will be less. For a 1/9, more.
So by nature, the wavelet is a good way to reduce the observation time when the frequency increases.
As the CF is expected at some kHz, my hypothesis is it is high enough for the reflections to be rejected.
In the the time to frequency transformation, it is known that it is not possible to be precise in time and in frequency. The wavelets offer an excellent performance in this time/frequency compromise.
Christian
Thickness 0.025, Young's 30 MPa, Poisson's 0.3 and density 25
Perhaps I'm doing the displacement in a bad way, but since I cannot have that many nodes in the free version, I have only fixed the edge nodes and not an area. It is a very rough approximation, but due to the low resolution, if I would select a couple of complete quads the area would be too big instead:
@Veleric :
Hello Eric
I made a test on measurements yesterday... I was not able to find the CF... If the long window FR of a panel is not too bad, the short time FR seem messy. Let me know from your tests.
Christian
Just stumbled across this paper on DML and bending wave, I'm not sure if it's been linked here already. https://publications.lib.chalmers.se/records/fulltext/154618.pdf