Integrating dimensionless transport features with machine learning for evaporation rate prediction under mixed and forced convection

Masalha, Ismail and Abu-Zienah, Hamza and Alawi, Omer A. and Badran, Omar and Poós, Tibor (2026) Integrating dimensionless transport features with machine learning for evaporation rate prediction under mixed and forced convection. RESULTS IN ENGINEERING, 32. no.-113183. ISSN 2590-1230 (In Press)

Preview

Text Vegleges_1-s2.0-S2590123026041940-main.pdf - Published Version Available under License Creative Commons Attribution. Download (7MB) | Preview

Abstract

Accurately predicting evaporation rate under mixed and forced convection remains challenging because of the interaction among heat, mass transfer, and flow dynamics. Conventional correlations are limited to specific conditions, while purely data-driven models often lack Physically Consistent Trends. This study proposes a Dimensionless Transport Feature-Based Machine Learning (DTF-ML) framework that integrates dimensionless parameters such as Reynolds, Richardson, and Schmidt numbers with data-driven models. The two utilized datasets are a limited dataset of 32 observations and a multi-source database containing 487 observations for model development, of which 462 remained after outliers were eliminated. An additional 32 observations were reserved exclusively for independent external testing. While, for the limited 32-point dataset, the Hybrid–Linear Regression model achieved R2=0.977, MAE = 0.037, and RMSE = 0.043 using Leave-One-Out Cross-Validation. On the internal held-out test set of the extended database, CatBoost achieved R2≈0.98, MAE≈0.02, and RMSE≈0.06. On the independent 32-point external dataset, the Physics–CatBoost configuration achieved R2=0.983, MAE=0.006, and RMSE=0.021. The PINN-like model shows competitive performance, while maintaining Physically Consistent Trends. The results demonstrate that incorporating dimensionless transport features improves prediction accuracy and model stability across different operating conditions. The proposed framework provides a data-supported approach to evaporation-rate prediction within the investigated operating-condition ranges.

Actions (login required)

View Item