Machine-learning-based lightweight optimization for vibration isolation of dual acoustic black hole beams

This study proposes a data-driven design framework for the lightweight optimization of metamaterial beams with enhanced vibration-isolation performance. Dual acoustic black hole (ABH) beams with dual bandgap characteristics, namely a complete bandgap and an ABH bandgap, are adopted as unit cells, and topologically optimized hollow configurations are introduced in the root regions for mass reduction. A dataset containing 15,000 distinct hollow patterns and their corresponding bandgap distributions is constructed using parameterized finite-element simulations. A convolutional neural network (CNN) is then developed to establish the nonlinear mapping between geometric configurations and bandgap characteristics. Numerical results show that the proposed model achieves a prediction accuracy above 99.9% within 300 training epochs, with a mean squared error on the order of 10−6 and excellent robustness (Δaccuracy < 0.0004). For representative samples with η = 0.16 – 0.445, the model achieves 99.87% – 99.99% accuracy for the complete bandgap and 99.97% – 99.99% accuracy for the first ABH bandgap, while the latter exhibits higher numerical stability. The root excavation strategy enables 21.9% – 44.5% mass reduction while preserving the longitudinal, flexural, and ABH bandgap characteristics, demonstrating substantial lightweight potential. Compared with conventional design approaches, the proposed machine-learning-assisted framework significantly reduces the computational cost and design cycle, providing an efficient route for the lightweight design of ABH beam structures.