Performance of neural network algorithms during the prediction of yarn breaking elongation

Abstract

Yarn breaking elongation is one of the most important yarn quality characteristics, since it affects the manufacture and usability of woven and knitted fabrics. One of the methods used to predict the breaking elongation of ring spun yarn is artificial neural network (NN). The design of an NN involves the choice of several parameters which include the network architecture, number of hidden layers, number of neurons in the hidden layers, training, learning and transfer functions. This paper endeavors to study the performance of NN as the design factors are varied during the prediction of cotton ring spun yarn breaking elongation. A study of the relative importance of the input parameters was also undertaken. The results indicated that there is a significant difference in the types of transfer and training functions used. Of the two transfer functions used, purelin performed far much better than logsig function. Among the five training functions, the best training functions in terms of performance was Levenberg-Marquardt. The study of the relative importance of input factors revealed that yarn twist, yarn count, fiber elongation, length, length uniformity and spindle speed, were the six most influential factors.