Dryout-aware empirical correlation for saturated flow-boiling heat transfer in smooth circular micro/mini tubes

Accurate prediction of saturated flow-boiling heat transfer coefficients remains challenging because wall nucleation, two-phase convection, confinement, vapor shear, liquid-film depletion, and dryout-region deterioration interact across fluids, tube sizes, and operating conditions. This study develops a dryout-aware empirical correlation for saturated flow boiling in smooth circular micro/mini tubes using a curated literature database of 18,833 heat transfer coefficient measurements from 78 sources. The correlation is formulated in closed form using dimensionless variables informed by a preceding machine-learning feature-importance study. Separate branches are developed for micro pre-dryout, mini pre-dryout, micro dryout-region, and mini dryout-region data, allowing the model to distinguish confinement-sensitive pre-dryout behavior from vapor-convection and residual-liquid effects after dryout incipience. The pre-dryout form combines heat-flux/thermodynamic forcing, inertial two-phase convection, and confinement/film-structure terms, while the dryout-region form combines vapor convection with a residual liquid/film/droplet contribution. Gravity/orientation and thermophysical-descriptor factors are included through continuous dimensionless groups, and a targeted micro-pre-dryout correction is introduced to capture confined high-shear behavior. The correlation achieved an overall mean absolute percentage error of 17.81%, median absolute percentage error of 13.35%, and root mean square error of 2.04 kWm −2 K −1 across the full curated database, with 82.58% and 95.25% of the data predicted within strict ±30% and ±50% error bands, respectively. Regime-resolved MAPE values were 17.61%, 17.99%, 16.86% and 17.47% for micro pre-dryout, mini pre-dryout, micro dryout-region, and mini dryout-region data, respectively. For pre-dryout data, the current model reduced MAPE relative to Liu–Winterton, Kim–Mudawar, Bertsch, and Sun–Mishima correlations evaluated on the same curated dataset. For the full evaluated dataset, the current model achieved 17.81% compared with 31.19% for the Bertsch composite correlation used as a full-quality-range stress test. The results demonstrate that a physics-informed, dryout-aware empirical structure can improve accuracy over conventional operating-variable correlations while retaining interpretability and requiring no wall-temperature input.