Develops machine learning models for predicting mechanical properties in corrugated cardboard, indicating a shift towards data-driven design.
Key Points
The aim is to create machine learning models using simulation data to predict mechanical properties of corrugated cardboard and optimize its structure.
Developed a constitutive model for corrugated cardboard base paper.
Conducted edge crush tests (ECT) and flat crush tests (FCT) in ABAQUS.
Established BPNN, GA-BPNN, and PSO-BPNN models for ECT and FCT predictions.
Evaluated prediction accuracy and stability using statistical metrics and cross-validation.
Used GA-BPNN for inverse design of structural parameters validated with finite element method.
GA-BPNN model achieved R2 values of 0.980 for ECT and 0.982 for FCT, showing high predictive performance.
Maximum errors for ECT and FCT predictions were 12.78% and 11.41%, respectively, compared to experimental results.
Inverse design results showed errors within 5% for ECT and 10% for FCT, demonstrating accuracy.