Inverse Design Of Acoustic Metamaterials Based On Machine Learning

Due to the strong penetration,slow attenuation and concealment of low-frequency noise,long-term exposure to it will do great harm to people's physiology and psychology,and therefore the control of low-frequency noise is attracting more and more attention.The traditional sound insulation materials based on the mass theorem and sound absorption materials based on porous fiber have poor noise reduction effect in low frequency band,poor ventilation and large volume,which cannot be applied in low-frequency noise reduction of small space with ventilation.In recent years,the appearance of acoustic metamaterials provides a new idea for the design of lowfrequency noise reduction devices due to the characteristic that large-wavelength sound field is manipulated with subwavelength size.Meanwhile,how to quickly and efficiently design structural parameters of various complex metamaterials for desired acoustic performance is becoming the core issue of metamaterial design.Traditional forward design method based on physical mechanism is difficult to achieve a new breakthrough due to unclear mechanism of complex structures or too many degrees of freedom of theoretical models.This paper proposes a sub-wavelength ventilated acoustic structure for absorbing broadband low-frequency noise,and by focusing on the inverse design of acoustic metamaterials based on machine learning,proposes an inverse design method of acoustic metamaterials based on genetic algorithm and an inverse design method of acoustic metamaterials based on Gauss-Bayesian model,providing new ideas for the parameter design of acoustic metamaterials.In Chapter ?,the introduction mainly introduces the research background of this paper,research status of low-frequency sound absorbing / insulating metamaterials at home and abroad,and parameter design methods of acoustic metamaterials.In Chapter ?,aiming at low-frequency sound absorption with ventilation in small space,a sub-wavelength ventilated low-frequency sound absorption absorber suitable for compact space is proposed.Its basic unit is nested by micro-perforated cavities and connected to a hollow pipe.Using nested design to increase the sound path can shift the resonance frequency to low frequency,and the small quality factor of micro-perforated plate can make the resonant absorption peak wider.Therefore,by designing the structural parameters of different units,we realize the low-frequency broadband sound absorption through the combination of units.In this chapter,the two-layer nested structure is studied by theoretical modeling,numerical simulation and experimental measurement,and the three-layer nested structure is studied by theoretical modeling,both of which achieve the low-frequency broadband sound absorption at subwavelength scale.In Chapter ?,an inverse design method of acoustic metamaterials based on genetic algorithm is proposed to optimize the high-dimensional structural parameters by taking the expected acoustic response as the objective function.The method is firstly applied to the sound absorber composed of cascaded Helmholtz resonators with the assistance of the theoretical model,and the 8 dimensional parameters are designed to achieve high efficiency(up to 95%)and broadband sound absorption in low frequency band;Secondly,a deep subwavelength sound absorber is constructed by combining the nested square open tubes with a straight tube.The unit is explored by theoretical model and numerical simulation,and then the 20 dimensional structural parameters are inversely designed using the method,achieving over 1 3 relative bandwidth in the low frequency range with 50% cross-section opening ratio.In Chapter ?,aiming at the efficient inverse design of parameters independent of analytical models,an inverse design method for acoustic metamaterials based on Gauss-Bayesian model is proposed.Through the cyclic iteration of training and prediction,the efficient inverse design of structural parameters independent of the physical models is realized according to the expected acoustic responses.Here,the adaptive convergence strategy is introduced into the acquisition function for the first time,and the method is utilized to inversely design the high-dimensional parameters of the typical ventilated acoustic metamaterial for low-frequency high-efficiency sound absorption.The sound absorption at 400Hz(up to 90%)is achieved with 37 iterations in the 10 dimensional parameter space much more efficiently than other methods.The experimental measurement and numerical simulation results verify the effectiveness of the inverse design method.In Chapter ?,we summarize the main content of this paper and look forward to the future work.