Machine learning models for vapor-liquid equilibrium of binary mixtures: State of the art and future opportunities

Machine Learning (ML) models, especially, Artificial Neural Networks (ANNs) are widely used in chemical processes modeling and also have been used for vapor-liquid equilibrium (VLE) determination. Despite, a comprehensive review for this topic was never written. In this review article we intend to present to the interested reader a review regarding the technical details of ANN modeling of VLE for binary mixtures, such as: direct or indirect VLE estimation, ANN type, inputs and outputs, training algorithm, activation functions, objective function, target mixtures, number of mixtures, data division, best structure found and main results. Based on the compilation of results obtained from selected articles, an evolution of research in the application of ML for modeling VLE of binary mixtures was provided. Within this context, we could show that most of the studies considered mixtures with one component remaining fixed, containing 8–10 mixtures on average. Also, that the best results were obtained by using linear activation function in the output layer and one hidden layer. Finally, with the analysis of the technical details, this work also presented the limitations in the field and opportunities for future research.