MODELING FABRIC HAND OF A TEXTILE PROCESS USING A MULTILAYER PERCEPTRON PRUNING ALGORITHM

摘要

As the textile materials and processes have inherent variability, estimation of their properties by mathematical models often yields a very high prediction error. Artificial Neural Network (ANN) systems present the potential solutions for the modeling and optimization of predicting. This paper presents a new multilayer perceptron (MLP) pruning algorithm for predicting denim fabric hand from stonewash treatment parameters. To optimize the MLP structure two techniques are used: (ⅰ) variance sensitivity analysis to prune hidden neurons (pruning does not concern inputs that corresponds to stonewash parameters) and (ⅱ) k-fold Cross-Validation. The stop criteria are based on a performance evaluation of the network results from both learning and validation datasets. The obtained results show that neural network models could predict the desired fabric hand with reasonable accuracy.