Data driven modelling as a new route to design PLA based materials with improved barrier properties

Abstract

In this study, two Machine Learning models (ML), namely Random Forest and Support Vector Machine, have been trained to predict the relative permeability of PLA based materials containing an impermeable phase. A large range of impermeable phases have been investigated, going from spherical, tubular, lamellar fillers to spherulites. Input parameters have been chosen wisely from the material characteristics and statistical correlations have been established to keep only the relevant ones. By this way, additional samples could have been added to the dataset. Good accuracy of the predicted relative permeability has been obtained for both models. The corresponding models were therefore used for proposing the material designs meeting targeted relative permeability values. For the first time, ML models are used to predict the permeability of polymer containing an impermeable phase.