Artificial intelligence is currently transforming thermodynamics. Hybrid models that combine machine learning (ML) with physical modeling enable predictions of thermophysical properties with unprecedented scope and accuracy. Focusing on the thermophysical properties of fluids, recent advances in this field are highlighted, covering two hybridization techniques: (i) embedding ML into physical models and (ii) incorporating physical knowledge into ML models. The discussion covers different types of thermodynamic models: (i) excess Gibbs energy models, (ii) equations of state, and (iii) force field models. The new hybrid models combine the soundness of physical models with the flexibility of ML models and give the best results when trained on large data sets, which are, however, not always available. The new hybrid models often significantly outperform widely used classical physical thermodynamic benchmark models. We have only begun to explore the new routes opened up by hybrid thermodynamic modeling; this review provides a starting point for future work in this field.
Hasse et al. (2026) studied this question.