Data-Driven Kinetic Reaction Networks for Separation Chemistry

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Data Science

Background:

• Understanding complex chemical reactions is crucial for process design.
• Separating rare-earth and actinide elements requires improved chemical insights.
• Liquid-liquid extraction is a key process for element separation.

Purpose of the Study:

• To develop kinetic reaction networks for uranium extraction using AI and machine learning.
• To compare purely data-driven models with chemistry-informed models.
• To enhance the interpretability and accuracy of chemical process modeling.

Main Methods:

• Leveraging artificial intelligence and machine learning for kinetic reaction networks.
• Developing purely data-driven models with L1 regression.
• Creating chemistry-informed models using quantum mechanical calculations for reaction energies.
• Comparing model performance based on experimental data.

Main Results:

• Purely data-driven models are accurate but lack interpretability.
• Chemistry-informed models show improved interpretability and consistency.
• Ensemble averaging in chemistry-informed models enhances accuracy.
• The dominant extracted species is UO2(NO3)2(DEHiBA)2, consistent with experimental data.

Conclusions:

• AI and machine learning can effectively model complex chemical reactions.
• Chemistry-informed models provide valuable insights into separation mechanisms.
• This approach offers accurate predictions and chemical understanding at low computational cost.
• The study advances the design and optimization of chemical separation processes.