Multi-scale design of MOF-based membrane separation for CO2/CH4 mixture via integration of molecular simulation, machine learning and process modeling and simulation

Metal-organic framework (MOF) membranes have demonstrated high efficiency for CO2 capture due to their wide range of pore sizes, high surface area, high porosity, and open metal sites. In this work, we propose a multi -scale design framework of MOF-based membrane separation for CO2/CH4 mixture via integration of molecular simulation, machine learning, and process modelling and simulation. The GCMC and MD molecular simulation is first used to evaluate adsorption isotherms, isosteric adsorption heat, self-diffusivity, activation energy, permeability, and selectivity of a MOF-based membrane (e.g., IRMOF-1) for CO2/CH4 separation at different operating conditions. It is found that the simulated isotherms at 298 K are consistent with experimental results. CO2 permeability can range from 4.090 x 104 to 3.818 x 105 barrer with variation in operating conditions. The same phenomenon is also reflected in the selectivity. We then establish prediction models of adsorption capacity and self-diffusivity using machine learning methods such as artificial neural network (ANN), which are used to calculate the permeability. The mean squared error is 0.0086, and the coefficient of determination is 0.9822. The proposed ANN models are integrated with the tanks-in-series model of a hollow fiber membrane separation process, which is simulated using the finite volume method. Three case studies illustrate the feasibility and superiority of the proposed integrated framework.

Automated summary

In this work, a multi-scale design framework of MOF-based membrane separation for CO2/CH4 mixture is proposed, which integrates molecular simulation, machine learning, and process modeling and simulation. The adsorption isotherms, permeability, and selectivity of a MOF-based membrane (IRMOF-1) for CO2/CH4 separation are evaluated using molecular simulation. Prediction models of adsorption capacity and self-diffusivity are established using machine learning methods, and then integrated with membrane separation process modeling. Case studies demonstrate the feasibility and superiority of the proposed integrated framework.