Accelerating computational fluid dynamics simulation of post-combustion carbon capture modeling with MeshGraphNets.

Area of Science:

• Chemical Engineering
• Environmental Science
• Computational Science

Background:

• Packed columns are crucial for post-combustion carbon dioxide (CO2) capture, enhancing solvent-gas contact.
• Computational Fluid Dynamics (CFD) models liquid-gas hydrodynamics for CO2 capture efficiency but is computationally expensive.
• Optimizing carbon capture systems (CCSs) is hindered by the extensive design space and high cost of CFD.

Purpose of the Study:

• To develop a data-driven surrogate model for efficient simulation of packed columns in CO2 capture.
• To utilize MeshGraphNets (MGN), a graph neural network, for accurate fluid dynamics modeling.
• To enable rapid design optimization of solvent-based post-combustion carbon capture systems.

Main Methods:

• Employed MeshGraphNets (MGN), a graph neural network framework, for data-driven modeling.
• Simulated a random packed column with over 160,000 graph nodes.
• Varied key input parameters: solvent surface tension, inlet velocity, and contact angle.

Main Results:

• MGN models accurately predicted complex fluid dynamics across a wide parameter range.
• Achieved simulation speeds over 1700 times faster than traditional CFD.
• Demonstrated adaptability to varying system parameters for CO2 capture analysis.

Conclusions:

• MGN offers a computationally efficient and accurate alternative to CFD for packed column modeling.
• The data-driven approach is practical for large-scale design optimization of carbon capture systems.
• MGN shows robustness and versatility for complex fluid dynamics in CCS applications.