Pioneering oil-water interfacial tension using surfactant properties and cutting-edge algorithms

This study addresses the challenge of accurately predicting oil–water interfacial tension through the integration of surfactant physicochemical descriptors and machine learning algorithms. The main objective was to establish quantitative relationships between surfactant structure, concentration, and oil characteristics to determine their collective effect on interfacial energy minimization. A dataset of 260 experimentally reported points was compiled from peer-reviewed studies, encompassing seven inputs (Surfactant MW, Charge, HLB value, CMC, Concentration, oil ZPC, and Oil API) against measured IFT as output. After ensuring dataset uniformity through leverage-based outlier detection, six algorithms (DT, AdaBoost, RF, KNN, CNN, and MLP-ANN) and one hybrid framework were trained and evaluated using 5-fold cross-validation with performance indices R 2 , MSE, and AARE%. The Ensemble Learning model achieved the highest accuracy (R 2 test = 0.986, MSE test = 4.09), demonstrating superior generalization compared with single learners. SHAP analysis confirmed surfactant concentration as the dominant factor with a strong negative association to IFT, followed by Oil API and ZPC, consistent with Gibbs adsorption theory. The results emphasize that interfacial behavior is mainly dictated by surfactant molecular architecture and concentration rather than oil composition. This unified data-driven approach provides a reproducible framework for evaluating and optimizing surfactant formulations to minimize IFT effectively in industrial applications.

Ҳали таржима қилинмаган