Flow boiling of HFE-7100 in a single microchannel

Abstract

The global drive towards electrification and carbon reduction has created an urgent requirement for compact, high-performance thermal management systems capable of dissipating high heat fluxes while maintaining component reliability. Flow boiling in microchannels has emerged as a promising solution due to its high heat transfer capability arising from latent heat effects and large surface area to volume ratios. However, industrial implementation remains limited by incomplete understanding of microscale flow behaviour, heat transfer, pressure drop, and critical heat flux (CHF). Existing predictive methods are often empirical, fluid-specific, and lack general applicability. This thesis addresses these limitations through an experimental investigation of flow boiling of HFE-7100 in a single rectangular microchannel. An experimental facility was designed and constructed to obtain accurate measurements of wall temperature, pressure drop, and high-speed flow visualisation. Two oxygen-free copper test sections with rectangular channels and hydraulic diameters of 0.75 mm were fabricated, with heated lengths of 30 mm and 75 mm to investigate length effects. Experiments were conducted over mass fluxes from 100–1000 kg/m²s, wall heat fluxes up to CHF, and inlet pressures of 1–2 bar. HFE-7100 was selected due to its dielectric properties and lack of prior single-channel flow boiling data. Single-phase validation and uncertainty analysis confirmed the reliability of the methodology. Flow visualisation identified bubbly, slug, churn, and annular flow regimes, with transitions strongly dependent on mass flux, heat flux, inlet pressure, and heated length. Stratified flow was not observed, confirming the confined nature of microscale boiling. Nucleation consistently initiated at channel corners due to geometric confinement. Increasing mass flux and pressure altered bubble coalescence and stability, shifting transition boundaries accordingly. Shorter heated lengths promoted earlier slug formation due to accelerated bubble growth at higher local heat fluxes. Transition maps demonstrated that existing predictive models only partially capture the combined effects of confinement, evaporation momentum, interfacial shear, and geometry. Heat transfer measurements showed that boiling incipience occurred at lower wall superheats with increasing inlet pressure. Heat flux significantly enhanced heat transfer coefficients through increased nucleation activity, while mass flux primarily extended the subcooled boiling region. Higher inlet pressures further improved heat transfer by stabilising nucleation. The 30 mm channel produced higher average heat transfer coefficients than the 75 mm channel at equivalent outlet vapour qualities due to intensified nucleation from concentrated heat input. Nucleate boiling persisted throughout the vapour quality range, limiting the applicability of convective boiling models. Evaluation of twelve existing correlations showed that the best-performing models were those capable of accounting for nucleate boiling dominance. Two-phase pressure drop increased with heat flux due to increased vapour generation and interfacial shear. Increasing pressure generally reduced pressure drop by delaying flow regime transitions through changes in fluid properties. Shorter heated lengths produced steeper pressure gradients due to intensified bubble activity and frictional losses. Existing pressure drop correlations showed considerable discrepancies, although the homogeneous flow model provided reasonable agreement. CHF investigations in the 30 mm channel showed that dryout occurred primarily through liquid film depletion in annular flow. Higher mass flux and pressure increased CHF, with mass flux having the strongest influence due to improved liquid replenishment and bubble removal. Existing CHF correlations performed reasonably well overall but did not accurately capture the weak pressure dependence observed experimentally. Finally, a novel neural network framework for predicting microscale flow pattern transitions was developed. A multi-stage feed-forward neural network architecture was trained using a large experimental database with robust preprocessing, cross-validation, and hyperparameter optimisation. The model achieved an overall classification accuracy of 95.5%, outperforming conventional empirical approaches and demonstrating the potential of data-driven methods for predicting complex microscale two-phase flow behaviour.