Neural Topology Optimization Via Active Learning for Efficient Channel Design in Turbulent Mass Transfer.

Area of Science:

• Chemical Engineering
• Computational Fluid Dynamics
• Machine Learning

Background:

• Optimizing fluid channel structures is crucial for enhancing mass transfer in chemical reactors and separators.
• Systematic design of channel topology for complex turbulent flows remains a significant challenge.

Purpose of the Study:

• To develop a machine learning framework for efficient topology optimization of channel structures for turbulent mass transfer.
• To simultaneously maximize mass transfer efficiency and minimize energy consumption.

Main Methods:

• A neural network (
• neural topology
• ) represents channel structures.
• Pre-trained neural operators with active data augmentation are used for optimization.
• The framework targets dual objectives: mass transfer efficiency and energy consumption.

Main Results:

• The machine learning framework achieves data-efficient training and superior computational efficiency compared to traditional methods.
• Optimized channels experimentally demonstrated a 37% improvement in concentration uniformity.
• The study shows how optimal structures vary with inlet velocity, providing design insights.

Conclusions:

• The developed machine learning framework offers an efficient and effective approach for topology optimization of turbulent mass transfer devices.
• The findings provide valuable insights for designing chemical process equipment under diverse operating conditions.