Polymer color properties: Neural network modeling

Abstract

The prediction and optimization of polymer color properties is a complex problem with no easy method to predict color properties directly and accurately. The problem is especially complicated with the formulation of polymers to achieve the physical properties. By considering the formulation, the neural network has been implemented for prediction of color properties consisting of sigmoid hidden units and a linear output unit arranged in a feed forward back propagation architecture. An optimal design is accomplished for 15, 16, 17, 18, 19, and 20 hidden neurons with four different algorithms including gradient descent with momentum (GDM), resilient backpropagation (resilient backpropagation), scaled conjugate gradient (SCG), and Levenberg-Marquardt (LM) algorithm. The best result in terms of statistics is presented by the LM algorithm with 16 neurons in the designed artificial neural network (ANN) model. The degree of accuracy of the ANN model is proven acceptable in all statistical analysis and shown in results. However, it was concluded that ANN provides a possible method for the prediction of specific color tristimulus values.