Machine Learning-Assisted Pool Boiling Parameter Prediction

Abstract

In this study, the pool boiling behavior of n-pentane fluid on different metallic surfaces was investigated using experimental and machine learning-based approaches. Copper, aluminum, and stainless-steel substrates were used in the experimental studies; for each surface, the uncoated condition and three different nickel coating concentrations were evaluated. Surface characterization included measurements of average surface roughness and contact angle; boiling performance was analyzed using surface superheat, heat flux, heat transfer coefficient (h), onset of nucleate boiling (ONB), and critical heat flux (CHF) parameters. Experimental results showed that nickel-coated surfaces generally reduced the onset of nucleate boiling, allowed the same heat flux to be achieved at lower surface superheat, and improved the CHF behavior, especially on aluminum and copper surfaces. Changes in surface roughness and contact angle were found to have significant effects on boiling performance. Random Forest, Gradient Boosting, and Support Vector Regression (SVR) models were developed to estimate heat flux and heat transfer coefficient using an experimental dataset. In the revised model structure, ΔT, material type, surface type, nickel concentration, Ra, and contact angle were used as machine learning inputs, while pressure and saturation temperature were treated only as fixed experimental conditions. For heat transfer coefficient prediction, the primary model was reconstructed without q″ as an input variable to reduce direct information leakage from the defining relation h = q″/ΔT. The q″-included h model was retained only as a secondary consistency case. Model performance was evaluated using R², RMSE, MAE, and MAPE metrics. The results showed that the SVR model provided the most successful performance in estimating h, while the Gradient Boosting model provided higher accuracy in estimating heat flux. Furthermore, the risk of overly optimistic performance, which might arise from including experimental points of the same surface in both training and test sets, was reduced by using a group-based cross-validation approach. This study evaluates the coating-material-boiling performance relationship in n-pentane pool boiling within an experimental and AI-supported framework by integrating surface characterization parameters with data-based modeling.