Learning the Effective Adhesive Properties of Heterogeneous Substrates

TL;DR

This paper uses machine learning to predict effective adhesive properties in peeling from heterogeneous substrates, capturing complex phenomena like pinning and depinning with a novel neural network approach.

Contribution

It introduces a neural architecture that accurately predicts critical peel force and the singular peel force relationship from heterogeneous patterns, extending multiscale modeling techniques.

Findings

Neural network accurately predicts critical peel force.

Captures the singular peel force vs. rate relationship.

Applicable to other free boundary problems.

Abstract

Adhesion is a fundamental phenomenon that plays a role in many engineering and biological applications. This paper concerns the use of machine learning to characterize the effective adhesive properties when a thin film is peeled from a heterogeneous substrate. There has been recent interest in the use of machine learning in multiscale modeling where macroscale constitutive relations are learnt from data gathered from repeated solution of the microscale problem. We extend this approach to peeling; this is challenging because peeling from heterogenous substrates is characterized by pinning where the peel front gets stuck at a heterogeneity followed by an abrupt depinning. This results in a heterogeneity dependent critical force and a singular peel force vs. overall peel rate relationship. We propose a neural architecture that is able to accurately predict both the critical peel force and the singular nature of the peel force vs. overall peel rate relationship from the heterogeneous adhesive pattern. Similar issues arise in other free boundary and free discontinuity problems, and the methods we develop are applicable in those contexts.