Band gap analysis and prediction for phononic metamaterials with different spiral shapes based on transfer learning

Abstract

This study presents a comprehensive computational investigation of band gap characteristics in spiral-based phononic metamaterials, including Archimedean, Octagon, Hexagon, and Square spiral configurations. It offers a quantitative understanding of the similarities in Bloch wave properties across these spiral types and demonstrates the feasibility of using data from known spiral patterns to facilitate the property prediction of new types. Based on the spiral datasets that vary in the number of turns, cutting width, and inner radius, we observed strong correlations in band gap counts among patterns (e.g., Rotated Octagon and Octagon, Archimedean and Rotated Octagon), indicating similar behaviors in band gap occurrence across different geometries. It was also found that the rotation of geometric shapes had a minor impact on band gap counts. However, we observed that the distribution of band gap width varies significantly across different types of spirals, with weak correlations. Furthermore, we demonstrate that transfer learning (TL) enhances prediction accuracy for new spiral types compared to traditional neural network approaches. TL model demonstrated superior performance, effectively capturing complex band gap details and improving overall prediction accuracy, without requiring extensive training data.