A machine learning-based data-driven approach for modelling anisotropic and tension-compression asymmetry behavior of elastoplastic materials using limited experiment data

Machine learning (ML)-based data-driven methods offer efficient approaches to characterize the anisotropic behavior of elastoplastic materials. This paper proposes a ML-based data-driven approach to model anisotropic and tension-compression asymmetry (TCA) behavior using ML techniques and limited anisotropic experimental data. To improve the predictive capability under unknown large deformations, two separate artificial neural network (ANN) models are employed within this modelling framework. Firstly, an enhanced data-driven constitutive model constructed using an ANN model, is utilized to characterize the mechanical behavior under tension loading along the rolled direction. This characterization forms the foundation for calculating the anisotropic factor. Secondly, another ANN model is used to learn the relationship between this anisotropic factor and stress components, enabling modeling of elastoplastic anisotropic and TCA behavior. Additionally, to address the challenge posed by limited available anisotropic experimental data, Hermite interpolation method is used to enrich the training dataset. For numerical implementation, a stress-updating algorithm based on the isotropic hardening assumption is introduced. Finally, to validate the proposed anisotropic data-driven approach, it is applied to predict the anisotropic and TCA behavior of the 2024-T351 aluminum alloy. The predictions are compared with those obtained from the Barlat model and anisotropic experiments. The investigation results show that, the proposed anisotropic data-driven approach effectively captures the evolution of anisotropic and TCA characteristics during the plastic deformation processes.