JACS: Reconstruction of liquid-liquid phase diagrams and extraction process design using machine learning with physical constraints

1/ AI application method: physics-informed machine learning
2/ Application task: reconstruction of liquid-liquid phase diagrams and extraction process design
3/ Reported result: the phase diagram was reconstructed and applied to extraction process design
4/ This JACS paper combined site-specific molecular interaction learning and physics-informed learning to reconstruct the liquid-liquid phase diagram and apply it to extraction process design.
5/ In other words, it is a case that did not stop at describing the phase diagram, but restored phase-equilibrium information through a learning framework that reflects physical constraints and connected it to process design.
6/ The significance from a materials and chemical engineering perspective is that ML with physical information can serve as an auxiliary tool in phase-equilibrium problems where experimental data are limited or difficult to measure.
7/ Phase diagram reconstruction and extraction design are directly connected to solvent selection, process-condition search, and separation-process optimization, so the application range is broad.
8/ The limitations are also clear.
9/ Even if physical constraints are included, they do not always completely replace actual phase behavior, and extrapolation error can become large outside the composition range of the training data.
10/ Therefore, the reconstructed phase diagram and design proposals should be confirmed by independent experiments or additional thermodynamic validation. [1][2]