Mechanical property prediction and inverse structural design of corrugated cardboard based on FEM and machine learning

Abstract

This study aims to develop machine learning models based on simulation data for predicting the mechanical properties of corrugated cardboard and performing inverse optimization of its structural parameters, thereby promoting the transition of corrugated cardboard design from empirical practices to an intelligent, data-driven paradigm. A constitutive model for the corrugated cardboard base paper is constructed, and edge crush test (ECT) and flat crush test (FCT) are performed in ABAQUS, with maximum errors of 12.78% for ECT and 11.41% for FCT compared to experimental results. The BPNN, GA-BPNN, and PSO-BPNN models are established to achieve forward prediction of the ECT and FCT. Their accuracy and stability are evaluated using statistical metrics and cross-validation. The BPNN model combined with GA is employed for the inverse design of structural parameters (flute width, flute height, liner paper thickness, flute paper thickness), and the accuracy of the inverse design is validated with FEM. The results show that the GA-BPNN model exhibits optimal predictive performance and stability, with R² values of 0.980 and 0.982, and RMSE values of 0.219 and 0.00938, respectively, on the ECT and FCT test sets. The comparison between the inverse design results and the simulations shows that the errors in ECT and FCT are within 5% and 10%, respectively. This study provides more accurate and customized solutions for the packaging industry, promoting the intelligent development of related industries.