Abstract: In order to predict the sound insulation of multilayer composite spunbonded nonwovens, a prediction method based on rough set theory and artificial neural network was introduced.Using attribute reduction, a reduction set composed of thickness, fiber diameter and porosity was extracted from the fiber web structural parameter set of multilayer composite spunbonded nonwovens which comprises 10 structural parameters. The values of sound transmission loss corresponding to 24 frequencies for 25 multilayer composite spunbonded nonwoven samples were predicted based on the 120 BP neural network models by taking the above three parameters as the inputs and changing the number of neurons in the hidden layer. The experimental results show that the overall average of mean absolute percentage error between the predicted and measured values of sound transmission loss for all samples is only 3.47%. Besides, those models with eight neurons in hidden layer have the highest prediction accuracy. This indicates that the sound insulation of multilayer composite spunbonded nonwovens can be accurately predicted using thickness, fiber diameter and porosity, which proves the rationality of the reduction result based on rough set.

Key words: spunbonded nonwovens, multilayer composite, fiber web structure, attribute reduction, artificial neural network, sound transmission loss

摘要： 为预测复合纺粘非织造布的隔音性能,提出基于粗糙集理论和人工神经网络的预测方法。运用属性约简方法对含有10个参数的复合纺粘非织造纤网结构参数集进行降维,得到含厚度、纤维直径和孔隙率的约简集。将上述3个参数作为输入并通过改变隐含层神经元个数建立120个BP神经网络模型,对25个复合纺粘非织造布样本的所有24个频率所对应的透射损失数值进行预测。实验结果显示所有样本透射损失预测值与实测值之间的平均绝对百分比误差的总平均值仅为3.47%,其中以隐含层神经元个数为8的模型的预测准确度最高。研究结果表明基于厚度、纤维直径和孔隙率能够对复合纺粘非织造布的隔音性能进行准确的预测,印证了粗糙集约简结果的合理性。

关键词: 纺粘非织造布, 多层复合, 纤网结构, 属性约简, 人工神经网络, 透射损失