基于人工神经网络的复合材料层合板隔声性能预测  

Prediction of Sound Insulation Performance of Composite Laminated Plates Based on Artificial Neural Network

作　　者：董静捷 郑辉[1] 李富才[1] DONG Jingjie;ZHENG Hui;LI Fucai(State Key Laboratory of Mechanical System and Vibration,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械系统与振动全国重点实验室,上海200240

出　　处：《噪声与振动控制》2024年第3期22-27,108,共7页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(51975352)。

摘　　要：面向纤维增强复合材料层合板隔声设计需求,应用正交试验方法设计18组对称铺设复合材料层合板结构,并以基于切比雪夫多项式展开的半解析法对复合材料层合板结构在10~1500 Hz频率范围内的传声损失进行预测。以单层厚度、纤维铺设角度作为人工神经网络的输入,复合材料层合板结构1/3倍频程传声损失作为输出,分别建立前馈(Back Propagation,BP)神经网络、径向基函数(Radial Basis Function,RBF)神经网络和广义回归(General Regression,GR)神经网络预测模型。结果表明,RBF神经网络的预测效果最好,均方根误差仅为1.0937,GR神经网络和BP神经网络的预测效果逊于RBF神经网络,均方根误差分别为2.6499和2.9697。最后,基于分析结果构建具有良好局部预测性能的神经网络模型以用于复合材料层合板结构的隔声性能预测。Facing the demand of sound insulation design of fiber reinforced composite plates,18 groups of symmetrically laid composite laminates are designed by orthogonal test method,and their sound transmission losses in the frequency range of 10-1500 Hz are respectively predicted by a semi-analytical method based on Chebyshev polynomial expansion.Aiming at a fast and accurate prediction and optimization of sound transmission loss of the composite laminates,The prediction models of Back Propagation(BP)neural network,Radial Basis Function(RBF)neural network and General Regression(GR)neural network are respectively established by taking the lamina thickness and the fiber angles as the input of artificial neural networks,and 1/3 octave sound transmission loss of composite laminates as the output.The results show that RBF neural network has the best predictive performance,and its root mean square error is only 1.0937.Both of BP and GR neural networks have lower prediction effects than RBF neural network,and their root mean square errors are respectively 2.9697 and 2.6499.Finally,based on the analysis results,a RBF neural network model with good local prediction performance is constructed for the prediction of sound insulation performance of composite laminates particularly in low-frequency range.

关 键 词：声学 隔声 人工神经网络 复合材料 层合板 性能预测 

分 类 号：TB535[理学—物理]