Modeling constituent-property relationship of polyvinylchloride composites by neural networks

Abstract

The purpose of this study is to develop an artificial neural network (ANN) model to predict and analyze the relationship between properties and process parameters of polyvinyl chloride (PVC) composites. The tensile strength, ductility, and density of PVC are modeled as a function of virgin PVC, recycled PVC, CaCO3, di-2-ethylhexyl phthalate, chlorinated paraffin wax, and CaCO3 particle size. The ANN model is trained using the backpropagation algorithm. The developed model was validated with a set of unseen test data. The correlation coefficient adj. R-2 values for test data were 0.95, 0.83, and 0.90 for tensile strength, density, and ductility, respectively. The relationship between constituents and properties of PVC composites were analyzed by sensitivity analysis, index of relative importance, and quantitative estimation. The study concluded that ANN modeling was a dependable tool for the optimization of constituents for the desired properties of PVCs.