A NEURAL NETWORK SYSTEM FOR PREDICTION OF THERMAL RESISTANCE OF KNIT FABRICS

Authors:

Abstract
An artificial neural network (ANN) was developed to predict the thermal resistance of knit fabrics. Thickness, porosity, air permeability, weight per unit area, and fiber conductivity were taken as input variables of the ANN. Data on thermal resistance were measured on experiments carried out on jersey knitted structures. An original (virtual leave-one-out) approach dealing with the overfitting phenomenon and allowing the selection of the optimal neural network architecture was used. The optimal ANN model contained three hidden neurons in the hidden layer. This model was validated by testing data, and the confidence intervals on the predictions were evaluated. It shows good performance in prediction with better accuracy. The developed neural model is expected to be used for a wide industrial context.