Unveiling the potential of machine learning for heat transfer prediction and flow regime classification in falling film evaporation over plain horizontal tubes

Accurate characterization of the Heat Transfer Coefficient (HTC) and reliable identification of the prevailing flow regime constitute fundamental prerequisites for the rational design and optimization of falling film evaporators. Machine Learning (ML) methodologies offer a valuable complement to conventional semi-empirical correlations, providing enhanced capability to resolve the intricate, high-dimensional functional dependencies governing the falling film evaporation process. In the present study, a suite of ML models is developed to predict the HTC during falling film evaporation over a plain horizontal tube across both the Full Wetting (FW) and Partial Dryout (PD) regimes, alongside dedicated classification models for flow regime determination. The regression database for the FW regime comprises 1,481 data points consolidated from 21 independent sources, while that for the PD regime encompasses 183 data points drawn from 5 sources. The developed models demonstrate remarkable predictive fidelity, achieving mean accuracy errors below 2% and 3% for the FW and PD regimes, respectively, while maintaining generalization errors of approximately 25% and 23%. The classification models identify the FW regime with misclassification rates below 0.5% and 10% in terms of accuracy and generalization, respectively. The generalization capability for PD regime classification, however, remains comparatively limited, a limitation fundamentally rooted in the scarcity and parametric narrowness of the available experimental data for this regime.