Polymer chemistry informed neural networks (PCINNs) for data-driven modelling of polymerization processes

Abstract

Although the use of neural networks is now widespread in many practical applications, their use as predictive models in scientific work is often challenging due to the high amounts of data required to train the models and the unreliable predictive performance when extrapolating outside of the training dataset. In this work, we demonstrate a method by which our knowledge of polymerization processes in the form of kinetic models can be incorporated into the training process in order to overcome both of these problems in the modelling of polymerization reactions. This allows for the generation of accurate, data-driven predictive models of polymerization processes using datasets as small as a single sample. This approach is demonstrated for an example solution polymerization process where it is shown to significantly outperform purely inductive learning systems, such as conventional neural networks, but can also improve predictions of existing first principles kinetic models.