An original instrument for thermo-mechanical polymer testing has been developed. This article describes the process of data acquisition, preprocessing and classification into 11 main polymer groups. The following polymer groups are used: polystyrene, acrylonitrile–butadiene–styrene, polycarbonate, low density polyethylene, polypropylene, high density polyethylene, polyamide 4.6, polyamide 6, polyamide 6/6, polybutylene terephthalate and polyethylene terephthalate. Three pattern recognition techniques of increasing complexities are applied in order to assess their suitability for the automated categorisation of polymer samples: k-nearest neighbours, various combinations of Q- and D-statistics (sometimes referred to as Soft Independent Modelling of Class Analogy, SIMCA) and Back-propagation Neural Networks. It is found that all the three methods categorise the materials into the correct polymer groups irrespective of their complexity. Methods based on the correlation structure in the data prove more beneficial than methods based on distance due to particular characteristics in the data. Best results are obtained using an adequate combination of two coefficients: one based on correlation and another based on distance.
