User login

You are here

Is machine learning a research priority now in mechanics?

Hi all,

I  know that the study and application of machine learning/artificial intelligence are somehow popular nowadays, in a general academia. I happen to be interested in that. I found people use this tool to predict crystal structure in material science, to characterize molecule structure in chemistry. The first impression is that the machanince learning helps to deal with the raw data, to constitute empirical models etc. However, there are not as many published works as I imagine, on the application of machine learning in mechanics.

Maybe what I know about mechanics, especially statistic mechanics, is limited. But I am wondering that whether is the some kind of research like "on the constitutive modeling for elastomer with machine learning" interesting and encouragable?




Machine learning includes a large set of techniques that can be summarized as curve fitting in high dimensional spaces.  Mechanicians have used these techniques, such as neural networks or genetic algorithms, extensively over the years without calling them "machine learning".  The recent resurgence in interest in machine learning comes form the development for new algorithms (in particular, significant improvements in neural net algorithms) and the availablity of large amounts of data which make the curve fits much more accurate.  The usefulness of the new  techniques should not be underestimated.

One application domain that comes to mind in improving the speed of detailed mechanics computations (say FEA) by fitting nets instead of redoing the same calculations over and over at each Gauss point.  For other approaches see

-- Biswajit

Thanks, Biswajit. I am agree with you. My first impression of mechanicians using machine learning is its applications in empirical modeling. However, the part of FEA quite interests me and I'm going to search for some examples.

A very late comment, but still, just because something struck me only this late... May as well share it....

I think that, as Biswajit points out, it's a question of matching a technique to an area where it is likely to be "good enough" of a fit.

I mean to say, consider fluid dynamics, and contrast it to QM.

In (C)FD, the nonlinearity present in the advective term is a major headache. As far as I can gather, the nonlinearity has all but been "proved" as the basic cause behind the phenomenon of turbulence. If so, using machine learning in CFD would be, by simple-minded "analysis", a hopeless endeavour---the very idea of using a potential presupposes differential linearity.

But then, consider the role of the BCs and the ICs in any simulation. It is true that if you don't handle nonlinearities right, then as the simulation time progresses, errors are soon enough going to multiply (sort of), and lead to a blowup---or at least a dramatic departure from a realistic simulation.

But then, also notice that there still is some small but nonzero period of time before a sufficiently bad amplification of the errors actually begins to occur. Now what if a new "IC" gets imposed right within that time-period (the one showing "good enough" an accuracy)? In this case, you can expect the simulation to remain "sufficiently" realistic-looking!

Something like that seems to have been the line of thought implicit in the results reported by this paper: [(.PDF) ^]. 

Machine learning seems to work even in CFD, because in an interactive session, a new "IC" is every now and then is manually being introduced by none other than the end-user himself!

It's somewhat like an electron rushing through a cloud chamber. By the uncertainty principle, the electron path sure begins to get hazy immediately after it is "measured" (i.e. absorbed and re-emitted) by a water molecule at a definite point in space. The uncertainty in the position grows quite rapidly. However, what actually happens in a cloud chamber is that, before this cone of haziness becomes too big, comes along another water molecule, and "zaps" i.e. "measures" the electron back on to a classical track. ... The end result is a very, very classical-looking (line-like) path as if the electron were only a particle, not a wave.

Conclusion? Be realistic about how smart "dumb curve-fitting" can at all get. At the same time, also remain open to all the application areas where it can at all be made it work---even in those areas where, "intuitively", you wouldn't expect it to have any chance to work!




Subscribe to Comments for "Is machine learning a research priority now in mechanics?"

Recent comments

More comments


Subscribe to Syndicate