Inverse design of cylindrical curved shell metasurface for elastic wave modulation based on PSO-BP model

Abstract

Nowadays, there is a lack of systematic and comprehensive theory regarding the propagation of elastic wave on cylindrical curved shells metasurface, largely due to the complex three-dimensional structural parameters on cylindrical curved shells. As a new powerful tool for predictive modeling, machine learning method can be learned and updated constantly, and is being used to design complex metasurface structures for controlling elastic wave propagation. In this work, an inverse design method was proposed to deal with the design of cylindrical curved shell metasurface with different curvatures. Firstly, machine learning technique was utilized to construct a mapping between structural parameters and properties of elastic wave propagation. Transmittance and phase shift properties under multiple outputs are initially predicted using multiple variable geometric parameters as inputs to machine learning network. Subsequently, particle swarm optimization (PSO) algorithm is utilized on original network to improve the quality of the predictions. Finally, the constructed models are used to design the unitary metasurface with high transmittance and special phase shifts, which are combined to realize the various modulation functions of elastic wave. Experimental results disclose that the network trained based on a large data set has high prediction accuracy. It is demonstrated the proposed method can not only be used to explore more elastic wave propagation laws in cylindrical curved shells, but also lays foundation for future functions of vibration isolation, energy focusing related to these structures.