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

TL;DR

This paper uses AI and neighbor analysis to detect and predict local yielding events in dry foams, revealing the importance of vertices and the temporal asymmetry in predictability.

Contribution

It introduces a novel AI-based method for detecting and predicting T1 events in foam flow using single-frame images and neighbor analysis.

Findings

AI achieves high accuracy with single-frame images.

T1 event predictability is linked to local neighborhood statistics.

Predictability development is asymmetric over time.

Abstract

The yielding of dry foams is enabled by small elementary yield events on the bubble scale, "T1"s. We study the large scale detection of these in an expanding 2D flow geometry using artificial intelligence (AI) and nearest neighbour analysis. A good level of accuracy is reached by the AI approach using only a single frame, with the maximum score for vertex centered images highlighting the important role the vertices play in the local yielding of foams. We study the predictability of T1s ahead of time and show that this is possible on a timescale related to the waiting time statistics of T1s in local neighborhoods. The local T1 event predictability development is asymmetric in time, and measures the variation of the local property to yielding and similarly the existence of a relaxation timescale post local yielding.