Predicting Bond Qualities of Fabric Composites after Wash and Dry Wash Based on Principal-BP Neural Network Model

Abstract

This article reports the construction of principal-BP neural network for predicting the bond quality of fabric composites after wash and dry wash. The parameters of fabrics and interlinings were analyzed by principal analysis and eight principal components were obtained through this method. A BP neural network with a single hidden layer was constructed including eight input nodes, six hidden nodes and four output nodes. During training the network with a back-propagation algorithm, the eight principal components were used as input parameters, while bond qualities were used as output parameters. The weight values were modified with momentum and learning rate self-adaptation to solve the two defects of the BP network. All original data were preprocessed and the learning process was successful in achieving a global error minimum. The bond qualities could be predicted with this training network and there was a good agreement between the predicted and tested values.