Implementation of Metallic Material Constitutive Models Based on Artificial Neural Networks in Explicit Finite Element Analysis

Machine learning techniques have emerged as a powerful tool for constitutive modeling and prediction of material response, offering superior accuracy and computational efficiency compared to conventional numerical approaches. This study presents a strategy for embedding artificial neural network (ANN) models into the finite element (FE) framework to enable automated and high-accuracy constitutive modeling of complex metallic materials. A FORTRAN-based automated code generator was developed to convert trained ANN models into the VUHARD user subroutine format, allowing direct integration into the Abaqus/Explicit solver without manual coding. The CoCrFeNiMn high-entropy alloy was chosen as a case study to validate the proposed method. A modified Johnson-Cook model was established as a benchmark for comparison with ANN-based simulations. Results show that the ANN-driven approach achieves superior predictive accuracy, with relative errors in flow stress consistently below 1%, while significantly improving computational efficiency. In addition, by leveraging the inherent differentiability of ANNs, the framework can automatically compute the derivatives required for the radial return algorithm, thereby ensuring robust convergence in nonlinear constitutive calculations. The findings confirm that the data-driven ANN-VUHARD integration provides a highly accurate and efficient alternative to conventional constitutive models. This approach eliminates the need fortime-consuming manual subroutine development when built-in models in commercial FE software are insufficient to describe complex thermomechanical responses. More broadly, the study establishes a generalized and scalable pathway for incorporating machine learning models into FE simulations of advanced materials, with considerable potential for application to high-entropy alloys and other systems with intricate deformation mechanisms. The proposed methodology demonstrates clear advantages for accelerating both academic research and industrial design processes.