Machine learning and predicting the time-dependent dynamics of local yielding in dry foams

Leevi Viitanen,Jonatan R Mac Intyre,Antti Puisto,Juha Koivisto,Mikko Alava

doi:10.1103/physrevresearch.2.023338

Abstract

The authors use convolutional neural networks to extract information in neighbor swap events in sheared foams

Highlights

Dry foams are assemblies of gas pockets separated by thin films of liquid, forming a connected polygonal film structure [1] at a configuration globally minimizing the surface energy [2]
We study the convolutional neural network (CNN) for five different combinations of locations and grayscale or skeletonized frames for the full range of region of interest (ROI)
We show that typically T1 events initiate from the unstable vertices that appear to violate Plateau’s rules while the bubble shape is a less relevant quantity (Fig. 5)

Summary

Introduction

Dry foams are assemblies of gas pockets separated by thin films of liquid, forming a connected polygonal film structure [1] at a configuration globally minimizing the surface energy [2]. The viscoplastic flow of dry foams is enabled by small elementary topological yield events referred to as T1’s and T2’s, analogous, for instance, to shear transformation zones (STZs) in amorphous solids [6,7]. The T2 events involve the disappearance of three-sided bubbles, while the T1 events refer to a neighbor swap between four bubbles. Both events enable the system to jump from one metastable surface energy minimum to another, including a local relaxation of the stored elastic energy of the foam [8,9]. The T1 events can be either reversible or nonreversible depending on the geometric configuration and the stress direction [15,16]

Methods

Results

Conclusion