Machine Learning-Driven Optimization of Multiphase Chemical Processes for Energy-Efficient and Sustainable Manufacturing

Abstract

There are many industrial chemical manufacturing processes that involve multi-phase chemistry, such as gas–liquid, liquid–liquid, gas–solid, solid–liquid and three-phase systems, and these are among the most energy-intensive and complex unit processes to optimise, due to the complex, nonlinear and often unmeasurable interfacial phenomena. The hydrodynamic, thermal and mass transfer coupling involved in these systems are difficult to model with traditional first-principles or empirical optimization approaches and make it challenging to progress towards industrially relevant scales and energy efficient, sustainable manufacturing targets. This narrative review aims to collect and condense the surge of literature on the application of machine learning (ML) algorithms for optimizing multiphase chemical processes, which spans from classic statistical process control (SPC) to advanced data-driven, hybrid and physics-informed learning algorithms. We study the use of supervised learning, deep learning, reinforcement learning, Bayesian optimization and evolutionary algorithms in important multiphase unit operations for surrogate modelling, soft sensing, real-time control and multi-objective optimization. The literature reviewed indicates that the combined results of the different strategies with ML can produce energy efficiency gains ranging from 8% - 28%, improvements in yield and product quality and predictive accuracy can be identified, and computational costs can be reduced when compared to other conventional approaches such as computational fluid dynamics (CFD) and first-principles (FP). The challenges highlighted include issues such as limited data, interpretability of models, scalability, and real-time deployment. The future scope for this research is physics-informed and hybrid models, digital twins, transfer and federated learning, and benchmarking.