Pattern recognition and feature selection for the discrimination between grades of commercial plastics

Abstract

A new method of polymer classification is described involving thermal analysis of polymer properties as temperature is changed. It is based on the analysis of the damping factor (tan δ) as a function of temperature. In this study four polymer groups, Polypropylene, Low Density Polyethylene, Polystyrene and Acrylonitrile–Butadiene–Styrene, each characterised by different grades, were studied using Dynamic Mechanical Analysis. The aim is to distinguish polymer (plastics) grades. The first steps are to group polymers into semi-crystalline and amorphous and remove outliers. Prior to employing classification algorithms, feature reduction was performed: two methods are compared, namely principal components analysis and parameterization by means of curve fitting using Gaussian end Exponential functions. Three classification methods are then compared, namely k-nearest neighbours method and two methods for discriminant analysis, based on class distances using both Euclidean and Mahalanobis measures. Both methods of feature reduction gave good results; using Nearest Neighbours on PCA scores 97% of samples were correctly classified, whereas parameterisation combined with Euclidean distance gave 94% correct classification.