Mechanical Property Prediction of Multitopology Composite Corrugated Structures Using Machine Learning Surrogate

This study investigates the equivalent mechanical behavior of composite corrugated structures with five cross-sectional topologies: sinusoidal, triangular, rectangular, trapezoidal, and reentrant. These structures offer high strength-to-weight ratios and tailorable orthotropic stiffness, making them suitable for adaptive aerospace applications. All designs were constrained within a fixed [Formula: see text] control volume, and equivalent properties—the longitudinal modulus [Formula: see text], transverse modulus [Formula: see text], and shear modulus [Formula: see text]—were obtained using a validated analytical homogenization model. The parametric space includes laminate thickness ([Formula: see text]), corrugation count ([Formula: see text]), and web angle ([Formula: see text] for trapezoidal and reentrant geometries). To reduce computational cost, a random forest regressor surrogate was trained across all topologies using a unified feature representation, with characteristic angle placeholders for angle-independent cases. The model achieved strong predictive accuracy ([Formula: see text] for [Formula: see text], 0.99 for [Formula: see text], and 0.96 for [Formula: see text]), demonstrating that a single model can capture diverse corrugation responses. Feature importance analysis identifies corrugation count and topology as dominant predictors, consistent with homogenization physics.