Neural network representation of a phase‐field model for brittle fracture

Abstract

Phase‐field models constitute a powerful tool in fracture mechanics. However, the main issues of these types of methods are the selection of the phase‐field model parameters, the numerical requirement of very finely meshed structures, and extremely small integration time steps. The subject of this work is the construction of an artificial neural network representing a finite element model coupled with a phenomenological phase‐field approach for brittle fracture. The approach is demonstrated on a notched plate in 2D. The neural network results and computation time is compared to established phase‐field models. After an initial training process, the neural network is able to predict the response of the complex system in a fraction of the time compared to the non‐local continuum mechanical model. (© 2017 Wiley‐VCH Verlag GmbH & Co. KGaA, Weinheim)