DeepBuckle: Extracting physical behavior directly from empirical observation for a material agnostic approach to analyze and predict buckling

Buckling is a long studied mechanical process that has been tackled from a variety of theoretical and numerical methods over the past two and a half centuries. Despite this, predicting buckling behavior of materials with complex structure and components, such as notched beams of non-homogeneous architected composites, remains non-trivial. Inspired by recent advancements in applying artificial intelligence methods to model physical relationships directly from observa-tional data, here we use a Variational Autoencoder in concert with a Long Short-Term Memory network to model the buckling behavior of notched beams. Our model, DeepBuckle, qualitatively and quantitatively learns buckling behavior of homogeneous polymer beams, and has the ca-pacity to predict novel designs that yield certain buckling behaviors with creative, out-of-the-box implementation. Importantly, we subsequently demonstrate that our approach directly general-izes to beams comprised of a far more complex composite foam material, without the increased computational resources or ancillary knowledge of material characteristics required by a more traditional finite element or other continuum approaches. Notably, the method reported here uses a simple table top experimental setup and can easily be transferred to other applications, for use in fundamental studies of mechanical phenomena, or in educational settings.

Automated summary

This study utilizes artificial intelligence methods and deep learning models to model the buckling behavior of notched beams. The model, called DeepBuckle, can qualitatively and quantitatively learn the buckling behavior of homogeneous polymer beams and predict new designs that produce specific buckling behaviors. Importantly, this approach can generalize to beams made of complex composite foam materials without requiring additional computational resources or knowledge of material characteristics. The method reported here can be easily transferred to other applications and is suitable for fundamental research and educational settings.