Self Organising Maps for distinguishing polymergroups using thermal response curves obtained by dynamic mechanical analysis

Abstract

Self Organising Maps are described including the U-Matrix, component planes, hit histograms, quality indicators as mean quantisation error and topological error. Software was written in Matlab and several new approaches for visualising multiclass maps are employed. The method is applied to a dataset consisting of the Dynamic Mechanical Analysis of 293 polymers, involving heating the polymers over a temperature range of −51 °C to 270 °C. These can be characterised in three different ways (a) amorphous or semi-crystalline (b) as 9 groups (c) as 30 grades.