Optimization of Mesoscopic Model Structure Parameters of <scp>CB</scp> / <scp>NRs</scp> Based on <scp>FE</scp> Simulation With Comparison Machine Learning Algorithms

ABSTRACT In this study, SEM observation showed that in four different CB/NRs, as the amount of filled CB increased, CB agglomeration became more serious, and the CB shape wasn't a simple sphere. Combining FEA with machine learning, the study optimized CB particle structure parameters, shape, and the aspect ratio of the ellipse. Using these optimized parameters, it established a 3D stochastic model without aggregates and a 3D model with aggregates to study the uniaxial tensile mechanical behavior of CB/NRs, adopting the Mooney–Rivlin phenomenological model. The results indicated that the ellipsoidal particle model was slightly better than the spherical one in predicting the mechanical behavior of CB/NRs. Specifically, the Random Forest algorithm fitting, cross‐validation, and grid search for hyperparameters to obtain the minimum RMSE had high prediction accuracy and fitting effectiveness. The optimal aspect ratio range was determined to be 2.2–2.4. Moreover, compared with experimental results, the RVE model with aggregates described the constitutive behavior of CB/NRs more accurately, better addressed the large deviations between FE simulation and experimental curves at high CB volume fractions in CB/NRs, and provided modeling solutions for CB/NRs.