Discovery of Physics and Characterization of Microstructure from Data with Bayesian Hidden Physics Models

Abstract

There has been a surge in the interest of using machine learning techniques to assist in the scientific process of formulating knowledge to explain observational data. We demonstrate the use of Bayesian Hidden Physics Models to first uncover the physics governing the propagation of acoustic impulses in metallic specimens using data obtained from a pristine sample. We then use the learned physics to characterize the microstructure of a separate specimen with a surface-breaking crack flaw. Remarkably, we find that the physics learned from the first specimen allows us to understand the backscattering observed in the latter sample, a qualitative feature that is wholly absent from the specimen from which the physics were inferred. The backscattering is explained through inhomogeneities of a latent spatial field that can be recognized as the speed of sound in the media.

7, pages, 6 figures. To appear, AAAI 2021 Spring Symposium on Combining Artificial Intelligence with Physical Sciences

References in corpus (4)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• The NumPy array: a structure for efficient numerical computation
• Implicit Neural Representations with Periodic Activation Functions
• Bayesian Hidden Physics Models: Uncertainty Quantification for Discovery of Nonlinear Partial Differential Operators from Data