基于神经网络的织物折皱回复性能预测  

Creasing properties prediction for woven fabrics based on neural network

作　　者：裴华强[1] 郑德均[2] 

机构地区：[1]中原工学院,河南郑州450007 [2]香港理工大学电子计算学系,香港00852

出　　处：《纺织科技进展》2008年第6期56-59,共4页Progress in Textile Science & Technology

摘　　要：探讨了将神经网络理论用于织物折皱回复性能的直接预测。根据研究对象特征,通过选取织物原料组成、经纬密度、抗弯长度、织物厚度及重量等重要影响因子作为神经输入元,将折皱回复角值作为输出目标,比较了径向基函数RBF、BP和广义回归GRNN 3种神经网络对织物折皱回复角的预测结果。试验结果表明,通过神经网络方法对织物的折皱性能预测具有较好的满意结果,且在预测织物的整体折皱回复性能时,BP模型与GRNN网络模型的预测值与实测值之间的相对误差要比RBF的小。利用输入神经元织物参数进行折皱回复性能的预测更有利于工艺与织物结构设计优化与质量控制,通过对输入神经元试验与优化有望达到满意的预测结果。The creasing properties of woven fabrics are predicted by using the method of artificial neural network. According to the eharacteristics of the investigated objects, some important factors, such as material composition, warp or weft density, bending-reslstant length, fabric thickness and weight, are considered to be input elements of neural network, while the wrinkle recovery angle is regarded as output result. Three different neural networks, i.e. RBF, BP, and GRNN neural network, are used to predict the output element. The results of the three methods are also compared with each other. From the experiments, it can be concluded that a satisfied agreement could be achieved by using the proposed neural network methods. Generally, the accuracy of BP and GRNN is relatively higher than RBF neural network. By experimental selection and optimization of influential ele- ments corresponding to the focus result, proper factors taken as neural network inputting elements can predict the output creasing properties of woven fabrics to provide guidance of technological processing, fabric design and quality control.

关 键 词：织物 神经网络 折皱回复性能 回复角 

分 类 号：TS190[轻工技术与工程—纺织化学与染整工程]