Prediction of tensile modulus for linear polymers applied to new materials development (5 pag)

PALOMBA, DAMIAN; CRAVERO, FIORELLA; VAZQUEZ, GUSTAVO, E.; DÍAZ, MÓNICA F.

Santa Fe

Congreso; Congreso Internacional de Metalurgia y Materiales SAM-CONAMET / IBEROMAT 2014; 2014

SAM - Asociación Argentina de Materiales

Any engineering activity is dependent on a careful selection between materials in order to satisfy requirements of performance and/or cost. In this regard, polymers have been modified to improve their utility and consequently synthetic polymers were developed. While the typical approach in the design of new materials has been empirical, at present the ability to predict material properties previous synthesis has been included and it has led to tremendous savings in time and cost. Nevertheless, it is not easy to achieve these predictions since the variables involved are very complex. Mechanical properties of polymers delineate their application profile. In this paper we present results about prediction of tensile modulus for a group of amorphous synthetic and linear polymers by means the QSPR (Quantitative Structure-Property Relationship) technique. To the best of our knowledge, this is one of the first attempts to investigate the possibility of predicting tensile properties for polymers by using QSPR technique. The dataset was generated for this work. Simplified molecular models were designed in order to depict polymers and also new descriptors were proposed. A new QSPR model is introduced providing a tool for the design of materials with specific application profile. A combined variable selection method was applied to the whole set of descriptors with the aim of reducing the set of descriptors using a variable selection method and a physicochemical-motivated strategy. This last was done manually by domain experts so that important and interpretable features are considered and redundancy is kept minimal. The final set consisted of 5 descriptors (4 classic and 1 ad hoc) with very good statistical performance for regard to R2 (squared correlation coefficient) and other classical statistical parameters; since the lack of related works in the literature performance could not be compared.