Reverse design of cylindrical shell metasurface structures based on machine learning methods

Ruiqi Meng, Lingyun Yao, Siwei Chen, Jialin Wu,
Reverse design of cylindrical shell metasurface structures based on machine learning methods,
Composite Structures,
Volume 372,
2025,
119606,
ISSN 0263-8223,
https://doi.org/10.1016/j.compstruct.2025.119606.
(https://www.sciencedirect.com/science/article/pii/S0263822325007718)
Abstract: Metasurfaces were considered highly promising devices for elastic wave manipulation due to their wave-control capabilities. However, forward prediction approach was predominantly employed in current metasurface designs. This approach required extensive computational resources and time. For cylindrical shell metasurface with more complex three-dimensional structural parameters, forward prediction result require more complex calculations and design. Design complexity of cylindrical shell metasurface structures was required to be reduced for rapid design implementation, which remained a significant challenge. A machine learning-based reverse design method for cylindrical shell metasurface was proposed in this study. Advances in machine learning technology provided new insights for designing complex metasurface. A serrated cylindrical shell metasurface was initially designed. Multiple variable parameters of serrated cylindrical shell metasurface were input into machine learning network. Relationship between variable parameters and elastic wave propagation characteristics was established through machine learning techniques. Neural network model was subsequently utilized to reverse design cylindrical shell metasurface. Abnormal refraction and beam focusing effects of elastic waves were effectively achieved by reverse designed cylindrical shell metasurface. Finally, constructed neural network model was employed to enhance transmittance of cylindrical shell metasurface. Improved beam focusing performance was achieved through cylindrical shell metasurface with enhanced transmittance. Machine learning-based reverse design approach significantly reduced design time of cylindrical shell metasurface. The method proposed in this study was not only developed to provide a novel design strategy for complex three-dimensional metasurface structures but also established fundamental principles for their applications in elastic wave manipulation, beam focusing and so on.
Keywords: Cylindrical shells; Machine learning; Elastic wave metasurface; Elastic wave modulation