Advancing material simulations: Physics-Informed Neural Networks and Object-Oriented Crystal Plasticity Finite Element Methods

Abstract

An innovative method for predicting the behavior of crystalline materials is presented by integrating Physics-Informed Neural Networks (PINNs) with an object-oriented Crystal Plasticity Finite Element (CPFE) code within a large deformation framework. The CPFE platform is utilized to generate reference data for training the PINNs, ensuring precise and fast predictions of material responses. The object-oriented design of the CPFE system facilitates the coherent incorporation of complex constitutive models and numerical methods, enhancing simulation flexibility and scalability. To demonstrate the adaptability of this approach, two problems are addressed: a fundamental power-law and a complex dislocation density-based constitutive models for predicting the behavior of ${\text{Ni}}_3\text{Al}$-based alloys. Both models are implemented within an object-oriented CPFE system powered by its flexible plug-in architecture. The resulting PINN model accurately captures intricate deformation mechanisms in crystalline materials, as validated through comparisons with CPFE simulations and experimental data. This work offers a promising alternative for efficient and accurate material behavior prediction, paving the way for advanced simulations in materials science.