数据驱动下的声学器件音质优化
Data-Driven Sound Quality Optimization of Acoustic Devices
许磊 1张维声 1朱宝 1郭旭1
作者信息
- 1. 大连理工大学 工程力学系,辽宁 大连 116024
- 折叠
摘要
音质是声学器件声音表现的重要衡量标准.但音质的优化过程需要对大量频点的响应进行协同优化,造成优化问题的可求解性较差.该文提出了一种数据驱动下的声学通道拓扑优化设计方法,可实现声-结构系统中的声频响快速预测,进而借助显式拓扑优化技术实现声学器件的音质优化.通过人工神经网络对结构几何参数、激励频率与声频响之间的非线性关系进行建模,以可移动变形组件(moving morphable components,MMC)法中的结构几何参数、激励频率为输入变量,以声压频响作为输出变量,通过训练多层前馈网络建立了声频响的人工神经网络模型.所得结果可以有效地将目标频带内的声压级范围差从 44.89 dB缩小至 6.49 dB,相较于传统优化方法,求解速度约为之前的 16.3 倍,表明了当前方法对音质优化问题的快速求解具有明显效果.
Abstract
Sound quality is an important measure of the sound performance of acoustic devices.However,the process of optimizing the sound quality requires a collaborative optimization of the responses at multiple fre-quency points,resulting in poor solvability of the optimization problem.A data-driven acoustic channel topology optimization design method was proposed to enable fast prediction of the acoustic frequency responses in the a-coustic-structural system and then optimize the sound quality of acoustic devices with explicit topology optimi-zation techniques.The non-linear relationship between structural geometry parameters,excitation frequencies and acoustic frequency responses was modelled with artificial neural networks.An artificial neural network model for acoustic frequency responses was developed by training a multilayer feedforward network with the structural geometrical parameters in the moving morphable components method and the excitation frequencies as input variables,and the acoustic pressure frequency responses as output variables.The obtained results can ef-fectively reduce the range difference of the sound pressure level(SPL)in the target frequency band from 44.89 dB to 6.49 dB.Compared with the traditional optimization method,the solution speed is about 16.3 times as be-fore,which shows that the current method is effective for the rapid solution of sound quality optimization prob-lems.
关键词
拓扑优化/声-结构系统/人工神经网络/可移动变形组件法/音质Key words
topology optimization/acoustic-structural system/artificial neural network/moving morphable components method/sound quality引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据