To me this whole idea seems rather difficult, if not impossible, to implement. We are talking about interactions happening at the atomic level of matter here.
It's not really the atomic level that we have to implement. You can actually treat the material as continuous, and sampled at discrete points so that the wavelengths are much, much larger than the distance between points, and the steps in time are small compared to the size of the wavelength/(wave speed). In the limit as the difference between spatial samples and temporal samples goes to 0, you get something that represents the real physics of the problem. The real question is just what's good enough?
And it would be far more interesting to sample real cymatic figures in the physical world and reuse them for synthesis or whatever.
I really liked the video link you posted for the figures. The rectangular plate has some cool symmetries, that would be better than a circular membrane. Plus, it's easier to do simulations of partial differential equations in rectangular coords rather than polar coords.