Perspective: Machine Learning of Thermophysical Properties

In this first contribution to Fluid Phase Equilibria’s Perspective Series, we discuss the role of machine learning (ML) in research on thermophysical properties. Following the idea behind the new series, this is no classical review aiming at a comprehensive coverage of previous work on the field. Instead, we provide an overview of the developments and point out promising new directions in the field, linking the perspectives of chemical engineers and computer scientists. The topics we cover include the role of data in research on thermophysical properties; the long history of ML methods in this field, which, however, stemmed so far almost exclusively from supervised learning; other ML methods of interest; as well as the important subject of how to merge physical modeling with ML to create hybrid approaches, which we expect to play a central role in the future. The discussion is illustrated by examples of the application of matrix completion methods from ML for the prediction of mixture properties, which we have recently introduced.