Deep learning identifies transversely isotropic material properties using kinematics fields

Abstract

Determining the stress-strain relationship in materials that exhibit complex behaviors, such as anisotropy, is pivotal for applications in structural engineering and materials science, as the behavior of materials under stress directly impacts safety and performance. This study introduces an innovative approach that leverages Artificial Intelligence (AI) through deep learning (DL) techniques to accurately predict transversely isotropic material properties using kinematic fields. These kinematic fields are derived from Finite Element Method (FEM) computations, which can realistically be obtained through advanced image correlation techniques, ensuring high precision and applicability in real-world scenarios. The objective of this research is to precisely characterize the behavioral parameters governing transversely isotropic materials. This methodology can also be applied to other constitutive laws, extending its utility across different material models. The proposed methodology, which utilizes a multi-scale encapsulated AI architecture, not only provides nearly instantaneous analytical solutions but also achieves remarkable accuracy, with average errors in parameter identification remaining below 3 % across all parameters. This sophisticated AI model plays a crucial role in accurately ascertaining the mechanical properties of transversely isotropic materials. By offering a method that is significantly faster and more precise than traditional experimental techniques, this research advances the current understanding of transversely isotropic materials' behavior. Such improvements in analysis speed and accuracy facilitate quicker iterations in material design and testing, potentially accelerating advancements in materials science and engineering applications.