Hello Paul
Light modifications of your script (please do not make a conclusion I am a specialist of Python... far from that, just a user)
2 remarks :
Light modifications of your script (please do not make a conclusion I am a specialist of Python... far from that, just a user)
- only odd odd modes
- log vertical axis
- passing the mouse over the graph shows the modes in the terminal (first time I test that! Thank you stackoverflow).
2 remarks :
- mode 1,3 (orange) and 3,3 (brown, 4th line ratio=1) lines become quickly close (risk of a too high level?)
- does ratio (vertical line) passing on mode line crossing be avoided? Is the minimum of crossing points a criteria?
I intend to use the python interface to Elmer when I get back to that again. Makes it much easier to do a series of trials and gather comparative data. I will post a pic of odd modes. As you know I am interested in irregular shapes to try to get an even distribution of modes and hopefully more modes at the bottom end. But also shapes with straight edges (eg with one long side slanted) and so on. Also, to find a complementary pair of aspect ratios that together give the smoothest response.
Great, I’ll have a look at the rollover feature. I think it’s worth including odd/even modes too.
I would say so. Two modes of the same frequency will give a higher peak. Plus, for every two modes that are degenerate (ie close together), you effectively have a gap somewhere else. So even distribution is a target for me.
Here is a graph of mean and SD for different aspect ratios. Frequency to 1000 Hz. Note the peaks at integer ratios. I don't know if this is significant though for a real build. The peaks are sharp, and any asymmetry in mounting, damping, exciter position etc may eliminate the peak. Also this graph uses all modes, not just odd-odd or odd-X modes. But if it's easy to avoid, why not?
The average SD (is that a thing?) is around 5.3 or so, seems quite high compared to the mean? I would hope to find it lower for asymmetrical, large-perimeter length or panels with concave (dispersive) edges.
My intuition about a two modes at the same frequency 'causing' a larger gap somewhere else is wrong. Zeros just occur where they occur.
The distribution is not gaussian, but more like exponential distribution, with decreasing counts for larger gaps. Here is the distribution of modes 0-1000Hz at aspect ratio 3.0. Lots of small gaps, but there is no related increase at the larger end.
In fact, the reverse may be true. Since the mean modal gap is relatively constant with changing aspect ratio, a FALL in the number of small gaps will be counterbalanced with a longer tail, and more larger gaps further out.
The distribution is not gaussian, but more like exponential distribution, with decreasing counts for larger gaps. Here is the distribution of modes 0-1000Hz at aspect ratio 3.0. Lots of small gaps, but there is no related increase at the larger end.
In fact, the reverse may be true. Since the mean modal gap is relatively constant with changing aspect ratio, a FALL in the number of small gaps will be counterbalanced with a longer tail, and more larger gaps further out.
Last edited:
Scrub that last sentence. I was thinking of a steep and narrow distribution that starts high, vs a flatter distribution that starts low, with a long tail. But that's not going to give the same mean!
What we know is that the number of modes less than a given frequency is quite constant, and the mean is quite constant. Looking for aspect ratios with the smallest number of near-equal modes will give you more further apart, but still less than the mean. Anyhow, we probably only need to consider near-equality for odd,odd modes as Christian said.
What we know is that the number of modes less than a given frequency is quite constant, and the mean is quite constant. Looking for aspect ratios with the smallest number of near-equal modes will give you more further apart, but still less than the mean. Anyhow, we probably only need to consider near-equality for odd,odd modes as Christian said.
Nice Paul, can't believe I missed your post for so long....so many posts in the monster thread lately I didn't even see your post here until last night.
Anyway, yes. Haha, your plots look familiar. See this post:
https://www.diyaudio.com/community/...s-as-a-full-range-speaker.272576/post-6129119
My observation of this is what got me interested in high aspect ratio panels to being with.
Eric
Good questions Christian,
On that topic, I can say that with my high aspect ratio panels I have often seen spikes in the FR corresponding to where the "second wave" of resonances start to hit. That is, where the (3,1), (3,5) (3.7) etc start to overlap with the first set. Sometimes (but not always) these peaks can be mitigated by offsetting the exciter close to the 0.33 position (along the short direction of the panel). That is, along the one of the two blue lines in the image below). Of course, this position is a nodal line for the entire "second wave" of modes, so it makes sense that placing the exciter along that line would reduce the contribution of the "second wave".
It's funny, though, that in a way this is opposite of the NXT way. That is, I'm suggesting to position the exciter in a place that avoids exciting particular modes, rather than in the place that excites as many modes as possible.
Eric
Paul,
Very interesting, I never noticed that before! I know you have suspected it, and I was skeptical. But you are correct. I had to go back to my own plots to see what's happening. Now I see that indeed at the integer ratios there is a higher probability of two modes at exactly the same frequency (see yellow highlights below).
One thing to note, however, is that these calculations all assume an isotropic panel. But few materials that we typically use for panels are isotropic. For an anisotropic panel, I suspect there will be similarly poor aspect ratios (i.e with higher standard deviation), but they probably won't fall at integer aspect ratios anymore.
Eric
You should Eric! I know there are many things to do, to learn in life! Excellent tool with many libraries for engineers and scientists. Large support. Free, multi-plateforne. Interfaces with FEM tools (not experimented). No need of deep knowledge in software. It runs almost out of the box. As you see I am a supporter of Python and Linux but the second is an other story (=choice).
By the way, Eric, could you remind me if there are other exact formulas for other boundary conditions (I am currently testing a Python script based on FDM).
Sound smart Eric.
I start scripting on the FDM solver to get proposals of a driving location. Your criteria is interesting.
Christian
Sorry Christian I am not aware of any other exact formulas for different boundary conditions.
Eric
Paul,
Yes that’s the equation. I thought there would be one for free panel too, but there is no closed form solution for the free panel.
Yes I intend to stick with EPS which I assume is close to isotopic. Presumably the results can be scaled for non-isotopic with the elastic constants, or one could do a series of fem runs overnight to plot a similar graph. I wonder if the peaks would show up with fem. I guess you would need a high resolution grid.
Paul,
Yes, EPS is probably pretty close to isotropic. I didn't care for the one EPS panel I tried, and haven't seen other convenient sources, so it's kind of off my radar. Eric
Well I now have a bunch of it so for the time being, I'll use that. Carbon fibre looks great but there is a lot of expense with the tooling and materials, quite a bit of skill to apply it, and a host of questions if the result is not as hoped. Did I use enough epoxy? Too much? Was the core strong enough? Did it delaminate? etc. I also feel there is a lot of unexplored territory with the cheaper materials, including shape (as I keep saying!), sculpting in the thickness dimension, applying skins of eg alfoil or mylar, a central insert of acrylic for hf, constrained layer damping, doublet or triplet panels with interleaved modes, applying a curve or pre-stressing, etc
Now that we've broached the topic of Python... Something I'd like to try with is to start to build a tool for experimentation. REW is great, but I find that poking about on a panel with either a scan or RTA, it's very difficult to discern what is happening. REW is not optimised for looking at changing modal frequencies, peak heights or widths.
There are several audio libraries for Python, and Im thinking to scan and record the peaks, then for a subsequent scan try to identify how the peaks have shifted relative to that baseline, and plot the differences. (Maybe f vs delta-f, with colour for the peak height).
There are several audio libraries for Python, and Im thinking to scan and record the peaks, then for a subsequent scan try to identify how the peaks have shifted relative to that baseline, and plot the differences. (Maybe f vs delta-f, with colour for the peak height).
Strangest thing happened on the other forum. After looking at a few papers of the illustrious Mr Karavashkin, and the Special Laboratory for Fundamental Elaborations, I called bull-****. Immediately chdsl seemed to back me up, but in strangely broken English. I made the accusation that he was a sock-puppet, we went back and forth, and very soon after all relevant posts were deleted by an administrator with a warning. It’s my firm belief that Alvipet is here to disrupt that forum with pseudo-science, and that he is in cahoots with chdsl, that they are both Russian, and I suspect that at least one admin of the site may be involved also. I don’t know why anyone would bother but suspect Russian manufacturers of DML speakers. I fully expect this post may disappear, which would prove it beyond any doubt. Sounds unlikely I know, Russians under the bed and all that, but something very strange is going on.
Last edited: