基于BP人工神经网络的轴流风扇气动噪声预测  

Aerodynamic noise prediction of tube-axial fan based on BP artificial neural network

作　　者：姚景瑜 孟海洋 杨京[1,2] 梁彬 程建春[1,2] Yao Jingyu;Meng Haiyang;Yang Jing;Liang Bin;Cheng Jianchun(Institute of Acoustics,Nanjing University,Nanjing,210093,China;Collaborative Innovation Center of Advanced Microstructures,Nanjing,210093,China)

机构地区：[1]南京大学声学研究所,南京210093 [2]人工微结构科学与技术协同创新中心,南京210093

出　　处：《南京大学学报（自然科学版）》2020年第6期900-908,共9页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(11374157)。

摘　　要：轴流风扇被广泛应用于如刀片服务器等机械设备的通风散热中,在工作时风扇旋转形成的湍流与自身以及固壁发生耦合作用,进而诱发气动噪声.湍流尺度极小,稳定性差,所以对气动噪声模型的数值计算需要耗费大量的计算资源.将人工神经网络与传统的数值模拟方法相结合,首先基于k-ε两方程模型求解纳维-斯托克斯方程得到轴流风扇的流场解,再通过FW-H声类比理论得到其声场解,然后对轴流风扇搭建人工神经网络模型,同时在声远场选取气动噪声声压级样本点,通过BP神经网络算法进行学习,最后利用训练好的人工神经网络模型对轴流风扇工作时的噪声场进行预测,实现了良好的预报效果.该方法可以有效节省计算资源,在工业上有一定的应用价值.Tube-axial fan is widely used in ventilation and heat dissipation of mechanical equipment such as blade server,etc.Turbulence caused by fan couples with itself and solid boundaries,so aerodynamic noise comes into being.Turbulence is small-scaled and unstable.As a result,the numerical simulation of aerodynamic noise consumes a lot of computing resources.This paper integrates artificial neural network with numerical simulations.First,turbulent field is solved by using N-S equations and acoustic field is calculated with help of FW-H analogy theory.Then,an artificial neural network of tube-axial fan is built and fed with sample data of sound pressure level chosen in far acoustic field.At last,the training-completed neural network is used to predict the aerodynamic noise of tube-axial fan and the result shows high accuracy.This method can save computing resources efficiently and has certain application value in industry.

关 键 词：轴流风扇 气动噪声 声类比理论 BP 神经网络 声场预测 

分 类 号：O429[理学—声学]