Evaluating the Impact of Constitutive Models on Xgboost Performance for Material Parameter Identification
Please login to view abstract download link
This study investigates the impact of 3D constitutive models on the predictive performance of machine learning, focusing on XGBoost for material parameter identification. While the Hill'48 yield criterion showed strong predictive results [1], the more advanced CPB2006 criterion [2] significantly degraded model performance. Experiments involved training XGBoost on datasets from biaxial tensile tests using Hill'48, then switching to CPB2006 to assess its effect. Efforts to improve performance with CPB2006, including PCA, hyperparameter tuning, and ensemble methods, were only partially effective, highlighting the need to better address the interaction between constitutive models and machine learning.
