This study explores the application of empirical and machine learning techniques to assess the impact of surfactants and time on the stability of oil-water emulsions and the characteristics of droplets. It utilizes a novel machine learning approach to forecast cumulative mass percentages by considering parameters such as drop size and time. The actual data was at 1st, 30, and 60 minutes after emulsion preparation and were forecasted up to 180 minutes with a Long-Short Term Memory (LSTM) machine learning model. The model demonstrates promising results in capturing the intricate relationships characterized by achieving an R-Squared (R2) score of 0.898 and Mean Squared Error (MSE) 0.00466. Under similar conditions and analysis, the results predicted for all three surfactants Gum Arabic (GA), Tween-20 (T20), and Poly Vinyl Alcohol (PVA) demonstrated similar behavior. Overall change in cumulative mass is lower confirming emulsion stability; however, at time stamps coalescence occurs, that can be neglected due to little impact. The results also show that interfacial tension is directly related to emulsion stability. Gum Arabic having highest interfacial tension (16mN/m) resulted in the most stable emulsion as compared to lowest interfacial tension surfactant Tween-20 (4mN/m). It is important to acknowledge certain limitations such as variations in surfactant concentration, temperature fluctuations, and shear forces, which may impact the experimental results and model performance. In conclusion, the current finding indicates that predictive modeling with LSTM in understanding emulsion dynamics is providing a foundation for future developments aimed at improving product performance and stability in a variety of industrial sectors like oil/gas, food and pharmaceutical.
Read full abstract