Estimation of Buckling Load in Sandwich Beams with Novel Honeycomb Core: A Non-Destructive Approach Enhanced by Machine Learning and Genetic Algorithm Optimization

This study presents an advanced, non-destructive method for evaluating the buckling behavior of sandwich beams by integrating the vibration correlation technique (VCT) with machine learning (ML) and genetic algorithms (GAs). The investigation focuses on a sandwich beam featuring an innovative honeycomb core, designed from a unit cell characterized by five adjustable parameters. Three ML models — [Formula: see text]-Nearest Neighbor, Single-Layer Perceptron, and Multi-layer Perceptron — are developed to accurately relate these design parameters to two key performance metrics: the critical buckling load-to-weight ratio and the first natural frequency-to-weight ratio. The continuous predictive model derived from these algorithms is incorporated into a GA framework to simultaneously maximize both performance indices, thereby identifying an optimal set of design parameters. The optimized configuration is then realized through 3D printing of the sandwich beam. To determine the equivalent mechanical properties of the honeycomb core, a homogenization approach combined with finite element analysis is employed. Additionally, the buckling and free vibration responses of the structure are analyzed using equations derived from Timoshenko beam theory and solved via a meshless numerical method. Comparison of the VCT-based predictions with experimental measurements confirms the accuracy and robustness of the integrated approach. The results demonstrate that the synergy of ML, evolutionary optimization, and advanced structural modeling provides a powerful tool for the design and analysis of complex structural systems.