ABSTRACT
This chapter describes the shrinkage behavior of the poly(ethylene terephthalate) (PET) yarns in terms of structural parameters. A neural network for the relation between the structure and properties of PET yarns was trained in the way described. Learning algorithms for neural networks are usually methods that minimize the mean square of system error iteratively by adapting the weights. The use of artificial neural networks in analyzing large sets of data has been a new and successful approach to reveal the structure-property relations of PET yarns. To avoid the development of ad hoc relations, a large experiment, comprising 295 drawn PET yarns, was performed, covering a wide range of yarn structures and properties. The prediction set consisted of PET yarns manufactured in processes other than the process used to produce the yarns of the training set and the test set. The set was characterized by a great diversity of structures.
