ANN and SVR Modeling to predict Bio-surfactant Enhanced Pool Boiling Heat Transfer

This study focuses on predicting saturated pool boiling heat transfer enhanced with a nonionic biosurfactant (coco glucoside) using machine learning models. Pool boiling experiments were conducted on a copper surface using methanol, ethanol, and acetone as working fluids. Each fluid was mixed with different concentrations of coco glucoside (0.5, 1.0, and 1.5 milliliters per 100 milliliters) and tested under heat fluxes ranging from 13,636 to 60,000 watts per square meter. The heat transfer coefficients obtained from these experiments served as target outputs for model training. Two supervised machine learning models, Artificial Neural Network (ANN) and Support Vector Regression (SVR), were developed to capture the complex relationships among thermophysical properties, biosurfactant concentration, and heat flux. Both models demonstrated excellent predictive performance, with SVR achieving an R² of 0.99, a mean absolute percentage error (MAPE) of 2.8%, and a root mean square error (RMSE) of 117.8 W/m ² K, while ANN achieved an R² of 0.98, a MAPE of 3.4%, and an RMSE of 147.6 W/m ² K. Parity plots confirmed strong agreement between predicted and experimental heat transfer coefficient values. This study highlights the effectiveness of integrating biosurfactant-enhanced boiling with machine learning models to optimize thermal systems and provide a reliable predictive framework for different fluids and operating conditions.