Process performance maps for membrane-based CO2 separation using artificial neural networks

Membrane-based gas separation processes are currently being implemented at different scales for several industrial applications. The optimal design of such processes, which is of key importance for their large-scale commercial deployment, has been extensively studied through parametric analyses and optimisation procedures. Nevertheless, the applicability of such design methodologies is generally limited by the large computational time and effort they require. In this work, surrogate models based on artificial neural networks are developed to circumvent the lengthy optimisation of a one-stage and two-stage cascade membrane-based gas separation process. In 200 ms, the surrogate model generates a Pareto front that describes the optimal trade-off between the process specific electricity consumption and productivity based on given input data, i.e., membrane material properties, feed composition and separation target. Whereas the surrogate model is applicable to any binary gas mixture, here its features are illustrated by creating process performance maps for post-combustion CO2 capture. Such maps provide valuable insights on: (i) attainable gas separation regions in term of CO2 recovery and CO2 purity, and (ii) the impact of membrane material, feed composition and separation target on the Pareto fronts and the optimal operating conditions.

Automated summary

Surrogate models based on artificial neural networks are developed to optimize membrane-based gas separation processes. These models can quickly generate Pareto fronts that describe the optimal trade-off between process-specific electricity consumption and productivity. The models are applicable to any binary gas mixture, and in this study, they are used to create process performance maps for post-combustion CO2 capture.