Machine Learning- and Finite Element-Based Temperature- and Rate-Dependent Plasticity Model: Application to the Tensile Behavior
Bo Zhang,Yang Yang,Hao Wu,Yida Zhang,Quanyi Wang,Hong Zhang,Yongjie Liu,Qingyuan Wang
DOI: https://doi.org/10.1007/s11665-024-10167-5
IF: 2.3
2024-10-10
Journal of Materials Engineering and Performance
Abstract:This paper aims to investigate the in-depth relationship between plastic hardening models based on J2 plasticity theory with temperature and strain rate using machine learning methods. A data set related to temperature and strain rate is constructed by randomly generating equivalent plastic strain increments. To replace the conventional process of solving equivalent plastic strain increments using Newton's iterative method under isotropic hardening and associated flow rules, artificial neural network (ANN) and support vector regression (SVR) models are developed to predict equivalent plastic strain increments efficiently. The hyperparameters of these models are systematically optimized to identify the most suitable configurations. After determining the optimal hyperparameters, the model is trained. The plastic hardening models based on ANN and SVR are then implemented within the Abaqus User MATerial (UMAT) subroutine, and their performance is validated. For this purpose, a single element and tensile test simulation were applied to evaluate the accuracy of ANN and SVR. The results indicate that the ANN model's performance initially improves with increasing temperature but subsequently deteriorates. In contrast, the strain rate appears to have minimal impact on the ANN model's performance. On the other hand, the SVR model's performance remains unaffected by variations in both temperature and strain rate. Furthermore, the SVR model demonstrates higher accuracy, requires fewer hyperparameters and is more suitable for this application. Nonetheless, the ANN model's results remain within acceptable limits.
materials science, multidisciplinary