Machine Learning for Two-Phase Flow Separation in a Liquid-Liquid Interface Manipulation Separator.

Area of Science:

• Chemical Engineering
• Separation Science
• Machine Learning Applications

Background:

• Two-phase flow separation is crucial for downstream purification processes, requiring effective methods to prevent contamination.
• Controllable flow behavior in separators is essential for reliable and efficient separation.
• Existing methods may lack the precision needed for complex water-solvent systems.

Purpose of the Study:

• To develop and evaluate a core-annular separator for biphasic flow separation.
• To implement machine learning (ML) models for predicting separation success and optimizing flow rates.
• To address industrial needs for process automation and cost reduction in separation development.

Main Methods:

• Development of a core-annular separator for biphasic liquid-liquid separation.
• Application of process prediction for automation using ML-based classifiers.
• Utilizing a multilayer perceptron (MLP) network to predict maximum input flow rates.
• Training ML models with limited experimental data for unknown water-solvent systems.

Main Results:

• The core-annular separator demonstrated complete two-phase water-solvent separation at a maximum total input flow rate of 4000 μL min⁻¹.
• ML-based classification achieved 92.2% accuracy in predicting successful separation.
• The MLP network exhibited superior performance in predicting the maximum flow rate.

Conclusions:

• The developed core-annular separator effectively achieves complete biphasic separation.
• ML models provide accurate predictions for separation success and optimal flow rates.
• This integrated approach of advanced separator design and ML can significantly reduce process development costs.