Machine learning assisted prediction and analysis of in-plane elastic modulus of hybrid hierarchical square honeycombs: Fid-Bau Portal

Machine learning assisted prediction and analysis of in-plane elastic modulus of hybrid hierarchical square honeycombs

Yang, Jian / Yang, Dingkun / Tao, Yong / Shi, Jun

Highlights • Hybrid hierarchical square honeycombs (HHSHs) with different configurations were fabricated and tested for evaluating their in-plane elastic modulus. • 11 machine learning (ML) models are trained using a database with 234 results from experimentally validated FE models. • A modified theory for the in-plane elastic modulus of HHSH is developed based on the XGBoost model and existing theory. • The upper and lower bounds of the in-plane elastic modulus of HHSH are determined by the modified theory. • Based on the SHAP method, a quantitative comparison of the importance of hierarchical parameters is conducted.

Abstract In this study, experimental, finite element (FE) simulation, machine learning (ML), and theoretical techniques are employed to investigate the in-plane elastic modulus $(E_{HHSH})$ of hybrid hierarchical square honeycombs (HHSHs). First, HHSHs with different configurations were fabricated using a 3D printer, and in-plane quasi-static compression tests were conducted on them. Then, 234 FE models are simulated to determine the $E_{HHSH}$ of HHSHs with various configurations, and the results are used to train 11 ML models. Comparative analysis demonstrates that the Extreme Gradient Boosting (XGBoost) model has the best predictive capability. Moreover, a modified theory for $E_{HHSH}$ is established based on the XGBoost model and existing theory, and its exceptional predictive capability is verified by comparing with experimental, FE, and existing theoretical results. Finally, the upper and lower bounds of $E_{HHSH}$ are determined by the modified theory, and the Shapley Additive Explanation (SHAP) method is used to identify the importance of different geometric parameters on tailoring $E_{HHSH}$ . The combination of theoretical and ML techniques provides a promising approach for developing a robust prediction model of material properties.