A Generalized Deep Entropy Scaling Architecture Framework to Predict Viscosities

A challenge in the development of new sustainable processes and materials is the scarcity of physical property data. Data driven models are often not able to handle thermodynamic constraints adequately. Integrating advanced machine learning methods with physically-based modeling techniques allows to combine advantages of machine learning methodologies and thermodynamic models, leading to better performance than the approaches on their own. We have previously proposed a neural network architecture incorporating entropy scaling to predict shear viscosities over a large range of thermodynamic state points and different chemical species. In this work, we extend the previous model to a generalized approach, avoiding the dependency on a particular entropy scaling framework. The resulting model demonstrates good prediction accuracy even for complex molecules with various functional groups, is easy to train and can be used in a variety of ways.