SPT-NRTL: A physics-guided machine learning model to predict thermodynamically consistent activity coefficients

The availability of property data is one of the major bottlenecks in the development of chemical processes and products, often requiring time-consuming and expensive experiments or limiting the chemical space to a small number of known molecules. This bottleneck has been the motivation behind the continuing development of predictive property models. For the property prediction of novel molecules, group contribution methods have been groundbreaking. In recent times, machine learning has joined the more established property prediction models. However, even with recent successes, the integration of physical constraints into machine learning models remains challenging. Physical constraints are vital to many thermodynamic properties, such as the Gibbs-Dunham relation, introducing an additional layer of complexity into the prediction. Here, we introduce SPT-NRTL, a natural language processing model to predict thermodynamically consistent activity coefficients and provide NRTL parameters for easy use in process simulations.SPT-NRTL can be classed as an advanced group contribution model that uses characters of the SMILES code as atom groups and then dynamically constructs higher-order groups. The results show that SPT-NRTL achieves higher accuracy than UNIFAC in the prediction of activity coefficients across all functional groups and is able to predict many vapor-liquid-equilibria with near experimental accuracy, as illustrated for multiple exemplary mixtures. To ease the application of SPT-NRTL, NRTL-parameters of 100000000 mixtures are calculated with SPT-NRTL and provided online.

Automated summary

The availability of property data is a major obstacle in chemical development. Predictive models, such as group contribution methods and machine learning, have been developed to overcome this bottleneck. However, integrating physical constraints into machine learning models remains a challenge.