Optimal Design Of Phononic Crystals Based On Data-Driven Methods

Noise and vibration are the most important factors that cause device damage and property damage in acoustic and collision problems,and have become a research hotspot in academia and industry.Researchers have proposed a variety of methods to noise control and vibration,to reduce damage caused by elastic waves.Among them,energy absorption theory based on material damage and reaction momentum theory based on elastic wave are the most widely used.Although some results have been achieved in the energy absorption theory based on material damage,there are still some limitations.Therefore,there is an urgent need to explore new materials with high efficiency to attenuate elastic waves.Phononic crystals(PCs)are the hotspots of new acoustic metamaterial.PCs,that consist of periodic distribution of scatterers embedded in a matrix,convey a few extraordinary physical properties beyond those materials found in natural or chemically synthesized substances.One of remarkable trait is the band gap,in which elastic wave cannot propagate due to the action of its internal structure,while transmit energy in other frequency ranges.As a result of this feature,the characterize and optimize the properties of PCs has fundamental significance on noise control and vibration for acoustic metamaterial structure.In this work,the data-driven approach is formulated to predict the band gap and optimize the band structure of PCs.And the main work of the study includes:(1)The neural network(NN)is formulated to get the mapping between the geometric parameters and the band gap parameters of PCs.Compared with the traditional finite element method(FEM)that takes a large amount of time to compute the band gap,the NN model is very efficient and accurate to predict the band gap of different types of PCs.(2)Traditional algorithms regular performance time-consuming,laborious and low precision to find the optimal solution of complex neural network model.To this end,Marine Predators Algorithm(MPA)has been deeply explored in this study to tun of energy gap of PCs.(3)The results on multiple data sets,including porous,solid/solid,fluid/solid and solid/fluids PCs,show that the numerical tools with a combination of machine learning and MPA have great robustness,high computational efficiency and accuracy.And the data-driven approach is very proposing to design PCs with complex shapes and large design parameters.