Prediction of Foam Rheology Models Parameters Utilizing Machine Learning Tools.

Area of Science:

• Rheology and materials science
• Computational modeling and simulation
• Chemical engineering

Background:

• Predicting foamed fluid viscosity traditionally relies on rheological models, but determining model constants experimentally under diverse conditions is complex.
• Foam rheology is influenced by numerous factors including shear rate, temperature, pressure, surfactant type, gas phase composition, and salinity.
• Accurate rheological characterization is crucial for optimizing processes involving foamed fluids.

Purpose of the Study:

• To investigate the impact of various parameters on foam rheology.
• To develop and evaluate machine learning models for predicting rheological model constants.
• To identify the most effective machine learning technique for this predictive task.

Main Methods:

• Experimental measurements of foam rheology using a high-pressure, high-temperature rheometer.
• Fitting experimental data to established rheological models (Power-law, Bingham plastic, Casson).
• Application and comparison of seven machine learning techniques (Decision Tree, Random Forest, XGBoost, etc.) to predict model constants.

Main Results:

• The Casson fluid model effectively described the foam's rheological behavior.
• The XGBoost (XGB) machine learning model demonstrated superior performance in predicting rheological constants, achieving 95% accuracy under optimal conditions.
• Pearson's correlation analysis indicated that XGBoost utilizes most features significantly for predictions, unlike other models.

Conclusions:

• Machine learning, particularly XGBoost, offers a robust and cost-effective method for predicting foamed fluid rheological constants.
• The developed methodology can significantly reduce the experimental effort required for rheological parameter determination.
• This approach provides a valuable tool for rapid assessment and optimization in applications involving foamed fluids.