Machine learning characterization of structural defects in amorphous packings of dimers and ellipses

Abstract

Structural defects within amorphous packings of symmetric particles can be characterized using a machine learning approach that incorporates structure functions of radial distances and angular arrangement. This yields a scalar field, softness, that correlates with the probability that a particle is about to rearrange. However, when particle shapes are elongated, as in the case of dimers and ellipses, we find the standard structure functions produce imprecise softness measurements. Moreover, ellipses exhibit deformation profiles in stark contrast to circular particles. In order to account for effects of orientation and alignment, we introduce new structure functions to recover predictive performance of softness, as well as provide physical insight to local and extended dynamics. We study a model disordered solid, a bidisperse two-dimensional granular pillar, driven by uniaxial compression and composed entirely of monomers, dimers, or ellipses. We demonstrate how the computation of softness via support vector machine extends to dimers and ellipses with the introduction of new orientational structure functions. Then, we highlight the spatial extent of rearrangements and defects, as well as their cross-correlation, for each particle shape. Finally, we demonstrate how an additional machine learning algorithm, recursive feature elimination, provides an avenue to better understand how softness arises from particular structural aspects. We identify the most crucial structure functions in determining softness and discuss their physical implications. arXiv:1811.03690v2 [cond-mat.soft]

Predicting the propensity for thermally activated β events in metallic glasses via interpretable machine learning

“…ii. Our ML model is able to link structure with both local favored and unfavored structural motifs, rather than only identifying the latter as in previous ML literature [17][18][19][20][21] . This is aided by the explicit and sufficient ART perturbation tests around each atom, and the Gaussian-like distribution of thermal activation energetics that gives sufficient resolution to both the soft and hard ends.…”

“…Over the past several decades, many efforts have been devoted to addressing this critical question. Recently, the emerging machine learning (ML) technique, based on wellcrafted representations of the atomic environment, has been proven to be promising for establishing atomic-level structureproperty relationships in liquids and glasses [17][18][19][20][21][22][23][24] . For example, Schoenholz et al 17 studied L-J model liquids and utilized ML to derive a structural parameter called "softness", which was found to correlate well with the particle's propensity for hopping, reflecting its susceptibility to β relaxation of liquids 10 .…”

Characterising soft matter using machine learning

“…89 Subsequent studies have applied the learning of ''softness'' to simulations of thin polymer films and pillars and to the analysis of granular experiments using spheres, dimers and ellipsoids. [90][91][92] In the former case, Sussman and coworkers found that the enhanced dynamics close to the surface of a polymer thin film is uncorrelated with the ''softness'' parameter. The SVM approach worked as before for predicting which sites would be likely to move, it just failed to identify any changes close to the free surface (or to the substrate).…”

“…This gave an excellent ability to predict rearrangements for ellipses and reasonable performance for dimers. 92 Inspired by the success of SVMs, the ''softness'' approach has been generalized via the use of graph neural networks that are able to predict the location of structural rearrangements. 80 Graph neural networks are being envisioned as a flexible machine learning methodology in which the role of the algorithm in shaping the character of the solution can be productively employed.…”

Predicting and following T1 events in dry foams from geometric features

“…Due to the crucial role of the yield events in the deformation of foams and other amorphous materials, the ability to predict where and when the yield events occur is critical for anticipating the behavior of foams under mechanical load. Solving complex problems, such as finding structural features that indicate yielding from amorphous materials, is well suited for machine learning tools, which continue to gain popularity in science [21][22][23].…”