Data-Driven Investigation of Flow Boiling Heat Transfer Characteristics in Micro-Channels under Variable Gravity Environment

ObjectiveTo address the demand for high-efficiency heat dissipation in railway transportation and aerospace equipment under variable gravity environments， we systematically investigated the flow boiling heat transfer characteristics of water-glycol mixtures in microchannels and developed data-driven predictive models. Although microchannel flow boiling offers a compact cooling solution， its characteristics under variable gravity are not well understood， and traditional empirical correlations lack prediction accuracy. In this study， we aimed to fill this gap and provide a theoretical basis for optimizing cooling systems for both railway and aerospace applications.MethodsBoth experimental and machine-learning approaches were employed to evaluate flow boiling heat transfer performance. A variable gravity experimental platform based on a centrifugal rotating table was established， capable of simulating gravity environments from 1.00<italic>g</italic> to 3.16<italic>g</italic>. The experimental system featured closed-loop circulation with a 200 mm-long， 2 mm-inner-diameter copper microchannel test section. Experiments were conducted across mass fluxes of 50-500 kg/（m²·s）， heat fluxes of 100-800 kW/m²， system pressures of 0.1-0.3 MPa， and inlet subcoolings of 10-30 ℃. T-type thermocouples with ±0.1 ℃ accuracy were used for temperature measurements， while pressure transducers and differential pressure sensors monitored system pressures. Three machine learning algorithms—Random Forest （RF）， Support Vector Machine （SVM）， and Extreme Gradient Boosting （XGBoost）—were developed using 80% training and 20% testing data splits with 5-fold cross-validation for hyperparameter optimization.Results and DiscussionsThe results demonstrate significant gravity-induced enhancement of flow boiling heat transfer. As gravity increased from 1.00<italic>g</italic> to 3.16<italic>g</italic>， the heat transfer coefficient （HTC） improved by 60%-80%， while the critical heat flux （CHF） increased by 20%-35%. In the region of low vapor quality （<italic>x</italic>&lt;0.3）， gravity enhancement reduced bubble departure diameter according to the relationship <italic>D_b</italic> ∝ g^-0.5， leading to increased departure frequency and enhanced microlayer evaporation. In the medium quality region （0.3&lt;<italic>x</italic>&lt;0.7）， gravity intensification resulted in thinner and more uniform liquid films， with peak HTC values reaching 23 000 W/（m²·K） at 3.16<italic>g</italic> compared to 14 100 W/（m²·K） at 1.00<italic>g</italic>. In the high-quality region （<italic>x</italic>&gt;0.7）， hypergravity delayed the onset from <italic>x</italic>=0.75 to <italic>x</italic>=0.8. A comparison with ten classical correlations showed that traditional models exhibit large prediction errors under variable gravity， with the best-performing Fang model achieving only a 9.6% mean absolute error （MAE）. In contrast， the XGBoost model achieves an MAE of 3.1% across all gravity conditions， with particularly superior performance at 3.16<italic>g</italic> （MAE=3.35%） compared to the Fang model （MAE=18.61%）.ConclusionsThis study confirms that gravity is a critical factor in flow boiling heat transfer， significantly enhancing both HTC and CHF through mechanisms such as bubble dynamic optimization and liquid film redistribution. The XGBoost machine-learning model demonstrates superior accuracy in predicting heat transfer performance under variable gravity compared to traditional methods. These findings provide a crucial theoretical basis for the optimal design of cooling systems for railway-vehicle-mounted aerospace airborne equipment that operate in complex gravitational environments.