NEURAL NETWORKS FOR AIR PERMEABILITY PREDICTION

The fabric porosity is leading factor for explaining air permeability of fabrics. Due to differences between ideal and real geometry and random variation of fabric structure there are no linear dependences between air permeability and predicted fabric porosity. Lack of theoretical model for this situation leads to the utilization of nonparametric regression techniques as neural networks. Introduction to the rational basis function neural networks in the context of nonparametric multivariate regression modeling is done. The advantages of RBF in comparison with classical neural networks are shown. The strategy of optimal RBF numbers selection is described. The main aim of this work is prediction of fabric air permeability from predicted surface and volume porosity (as combination of yarn diameters, weft and warp sett, planar weight and thickness of fabric). The 27- wool/PET plain weaves with constant sett of warp and varying sett of weft and varying yarn fineness will be used for predictive model building. The predictive ability will be characterized by the cross validation principle. The prediction of air permeability from basic geometrical characteristics is realized as well. .