基于人工神经网络的超材料声吸收体参量预测及逆向设计  

Parameter prediction and inverse design of metamaterial acoustic absorber based on artificial neural network

作　　者：朱远舟 龚纯 顾页 龙厚友 程营[1] 刘晓峻[1] ZHU Yuanzhou;GONG Chun;GU Ye;LONG Houyou;CHENG Ying;LIU Xiaojun(Institute of Acoustics,Nanjing University,Nanjing 210008;School of Naval Architecture,Ocean and Civil Engineering,Shanghai Jiao Tong University,Shanghai 200240;Midea Corporate Research Center,Foshan 528300)

机构地区：[1]南京大学声学研究所,南京210008 [2]上海交通大学船舶海洋与建筑工程学院,上海200240 [3]美的集团中央研究院,佛山528300

出　　处：《声学学报》2025年第2期257-267,共11页Acta Acustica

基　　金：国家重点研发计划项目(2022YFA1404400);国家自然科学基金项目(12074183,12004176)资助。

摘　　要：超材料声吸收体由于深度亚波长特性广受关注,但目前其设计及性能优化主要基于参数化扫描方法,该方法依赖人工物理直觉及设计因子,在多参数优化过程中存在计算资源占用量大、优化耗时且无法保证得到全局最优解等问题。为此,基于人工神经网络算法,提出了一种便捷高效的超材料声吸收体设计和优化方案,包括根据基元几何参量正向预测声吸收体的整体声学特性,以及根据目标声学频谱逆向设计所需的基元几何结构。同时,通过耦合由该神经网络优化得到的吸收元胞,设计了一种深度亚波长的宽带声吸收体,可在300~400 Hz范围内实现92.6%的平均声吸收,其结构总厚度仅为36.2 mm。所提优化方案可多参量同步优化,计算速度快,适用度广。Metamaterial sound absorbers have received widespread attention due to their deep subwavelength characteristics,while the design and optimization of metamaterial acoustic absorbers mainly relies on parameterized scanning method at present.However,this method largely relies on artificial physical intuition and design experience,which poses challenges in the process of multiparameter optimization,such as consumption of massive computing resources and much time,and difficulties in obtaining global optimal design.For this reason,this article proposes a convenient and efficient design and optimization scheme for metamaterial sound absorbers based on artificial neural network algorithms,including forward prediction of the overall acoustic characteristics for the sound absorber based on primitive geometric parameters,as well as inverse design of the required primitive geometric structure based on the target acoustic spectra.Moreover,a deep sub-wavelength broadband sound absorber with an average sound absorption at 92.6%in the range of 300−400 Hz is also achieved with this method,which is described a compact thickness at 36.2 mm.The optimization scheme proposed in this article displays the advantages of multi-parameter synchronous optimization,fast calculation speed and wide applicability.

关 键 词：声学超材料 声吸收体 人工神经网络 逆向设计 

分 类 号：TB34[一般工业技术—材料科学与工程] TP183[自动化与计算机技术—控制理论与控制工程]