Prediction of yarn crimp in PES multifilament woven barrier fabrics using artificial neural network

This research was aimed to develop artificial neural network (ANN) models to predict yarn crimp in woven barrier fabrics. For ANN training, 52 polyester (PES) multifilament barrier fabrics were produced by varying weft yarn and filament fineness, yarn type, weft density, weave type, and loom parameters. The supervised training of neural network was performed using Matlab (R) ANN toolbox function trainbr' which is the incorporation of Levenberg-Marquardt (LM) optimization and automated Bayesian regularization into backpropagation. From modeling outcomes, it was observed that both warp and weft yarn crimp models have generalized well with excellent coefficient of determination and trivial mean absolute error when tested on novel data. Moreover, input rank analysis of optimized network provided important information about model stability with respect to input variables, and trend analysis elucidated the input-crimp behavior using different input levels.