Simulating and Comparing CO₂/CH₄ Separation Performance of Membrane-Zeolite Contactors by Cascade Neural Networks

Area of Science:

• Materials Science
• Chemical Engineering
• Artificial Intelligence

Background:

• Greenhouse gas emissions necessitate efficient carbon dioxide (CO2) capture technologies.
• Membrane technology, particularly mixed matrix membranes (MMMs) with SAPO-34 zeolite, shows promise for CO2 separation.
• Limited research exists on modeling CO2 capture performance in MMMs.

Purpose of the Study:

• To develop and validate a machine learning model for predicting CO2/CH4 selectivity in SAPO-34 MMMs.
• To investigate the application of cascade neural networks (CNN) for simulating MMM performance.
• To establish a predictive tool for optimizing MMMs for CO2 capture.

Main Methods:

• Synthesized mixed matrix membranes (MMMs) incorporating SAPO-34 zeolite.
• Employed cascade neural networks (CNN) for data modeling and prediction.
• Optimized CNN topology (4-11-1) through trial-and-error and statistical monitoring.
• Validated the model against experimental data for CO2/CH4 selectivity.

Main Results:

• The optimized 4-11-1 CNN topology achieved high accuracy in predicting CO2/CH4 selectivity.
• The model accurately simulated selectivity for seven different MMMs under varying conditions.
• The CNN model predicted 118 experimental measurements with an average absolute relative deviation (AARD) of 2.92%.

Conclusions:

• Cascade neural networks provide a highly accurate and reliable method for modeling CO2/CH4 selectivity in SAPO-34 MMMs.
• The developed CNN model can precisely predict the performance of diverse MMMs, aiding in material selection and process optimization.
• This research highlights the potential of AI in advancing membrane-based carbon capture technologies.