Comparative Analysis of Machine Learning Approaches for Boiling ONB Prediction

Abstract

This study investigates the use of Machine Learning models for predicting both wall temperature and heat flux at the Onset of Nucleate Boiling (ONB). The dataset used in this work was obtained from an experimental test bench using Joule heating for boiling generation. Furthermore, five models, including Artificial Neural Networks (ANN), XGBoost, Support Vector Regression, AdaBoost, and Random Forest, were trained and evaluated. Results reveal that AdaBoost performed the worst in both wall temperature and heat flux predictions, indicating limitations in its ability to accurately forecast the ONB parameters. Conversely, the Random Forest model showed signs of overfitting in both predictions, suggesting that it may struggle to generalize to unseen data. In contrast, ANN demonstrated superior performance in predicting wall temperature (with mean square errors of 3.79 °C² and 3.84 °C² for training and testing), while XGBoost outperformed other models in heat flux prediction. Both models successfully captured the complex relationships between inputs (bulk temperature, pressure, channel inclination and velocity) and ONB parameters, leading to accurate predictions.