QSPR Modeling Applied to High Molecular Weight Polymers: Ductility Characterization from Elongation at Break

MARÍA JIMENA MARTÍNEZ; FIORELLA CRAVERO; IGNACIO PONZONI; MÓNICA. F. DIAZ

Aveiros

Conferencia; VIII International Symposium on Materials (Materias2017); 2017

New polymeric materials with specific requirements are designed to satisfy a demanding market and the study of the mechanical profile of a polymer helps to define its application range. During last years, we are proposing virtual testing tools based on Quantitative Structure-Property Relationships (QSPR) to assist the polymer designers prior to the synthesis, obtaining savings in times and costs. QSPR models estimate a target property of chemical compounds from variables that describe their molecular structure. In this work, we present classification QSPR models for ductility characterization of polymeric materials, using information provided by tensile test experiments in order to classify polymer ductility based on elongation at break. The models are computed from a dataset of 77 linear amorphous thermoplastic polymers with high molecular weight. The first step detects alternative subsets of relevant molecular descriptors related to the target property using a feature selection method. These subsets are contrasted by an expert via a visual analytics software tool. From this analysis, we conclude that an experimental parameter of the tensile test, cross head speed, plays a central role in all models. Finally, QSPR models are inferred by different machine learning methods. The output of these models allows classifying a new virtual material under design as ductile, fragile or undefined. The model with the highest performance correctly classifies 88.46% (%CC) of polymers and has a receiver operating characteristic (ROC) curve equal to 0.97. As conclusion, this QSPR model can predict if a material will be ductile or not in early phases of polymer design, previous synthesis, with high confidence.