Construção de modelos de aprendizado de máquina para predição de tensão de ruptura de nanocompositos baseados em grafenos e derivados

Abstract

The development of nanocomposite materials through experimental studies for mechanical properties analysis is a costly activity, requiring considerable time and effort. Concurrently, advanced Machine Learning (ML) techniques have emerged as a promising alternative for efficiently generating information and predictions, also being used for predicting the mechanical properties of polymer nanocomposites based on graphene. Thus, the objective of this work was to develop an ML model for predicting the rupture stress of polymeric nanocomposites using graphene and derivatives as reinforcing materials. In this study, an independent database was created from academic literature, addressing studies on the rupture stress of polymeric nanocomposites using graphene and derivatives as reinforcing materials, and in parallel, chemoinformatics was employed, based on the unsupervised ML technique, Mol2vec, to generate information based on polymer structure. Consequently, a unique database was formed, and Auto-Sklearn was used to predict the rupture stress gain from the addition of graphene to polymeric nanocomposites. Auto-Sklearn, utilizing a pre-trained Mol2vec model with a dimension of 300, identified the Gaussian Process based regression model as the best-performing one, achieving an R² value of 0.769. This application demonstrated that unsupervised ML models efficiently provide information using molecular structures in conjunction with nanocomposites, extracting complementary information for the prediction of mechanical properties through ML models. This implementation emerges as an alternative to reduce the need for empirical data generation in mechanical property analysis, accelerating the evaluation process for nanocomposite compounds.