Thermodynamically consistent machine-learned internal state variable approach for data-driven modeling of path-dependent materials

Characterization and modeling of path-dependent behaviors complex materials by phenomenological models remains challenging due to difficulties in formulating mathematical expressions internal state variables (ISVs) governing behaviors. Data-driven machine learning models, such as deep neural networks recurrent (RNNs), have become viable alternatives. However, pure black-box data-driven mapping inputs outputs without considering the underlying physics suffer from unstable inaccurate generalization performance. This study proposes a machine-learned physics-informed constitutive approach for based on measurable material states. The proposed model is designed with consideration universal thermodynamics principles, where ISVs essential path-dependency are inferred automatically hidden RNNs. RNN describing evolution follows second law. To enhance robustness accuracy stochasticity introduced training. effects number history steps, dimension, complexity, strain increment performances been investigated. effectiveness method evaluated soil under cyclic shear loading using experimental stress-strain data.