Machine Learning aided Phase-Field Method in Constitutive Modeling of Concrete Structures

Resumo

Phase-field modeling has emerged as a promising approach for modelling crack propagation. Different from Griffiths theory, which deals with discrete cracks, phase-field modeling transforms cracks into diffusive entities that propagate within a defined region, regulated by a length scale parameter. This methodology introduces a phase-field variable as a novel nodal degree of freedom, governed by an additional equation integrated into the model. This variable quantifies the damage of the material at each point, with zero representing intact material and unity indicating complete damage.
However, the practical implementation of phase-field modeling presents significant computational challenges. Its requirement of very refined meshes makes the process computationally expensive. In this context, machine learning techniques can be used to simulate the constitutive model and ensure that the process enhance efficiency. The article aims to use the machine learning technique to train a neural network capable of simulating the constitutive behavior of a phase-field model. The validation of this approach will be carried out by comparing its results with those obtained through finite element numerical analysis. All simulations will be conducted using the INSANE software.