Machine learning-inspired study of dynamical parameters of single vapor bubble under nucleate flow boiling regime

Heat transfer performance of nucleate boiling is interlinked with the departure dynamics of vapor bubbles, which also serves as the basis for heat transfer partitioning schemes and bubble growth models. Given the fast transients and multiple bubble ebullition cycles, boiling experiments lead to voluminous data that are to be analysed for determining the bubble dynamic parameters, and subsequently, the boiling heat transfer rates. Conventional approach of manual handling of such temporal and spatially-resolved experimental data is inherently time-consuming and error-prone. In the backdrop of these limitations, importance of machine learning algorithms gets highlighted in making the entire process automated and accurate as they minimize the need of any human intervention. The present work explores the potential of machine learning techniques towards quantifying the bubble departure characteristics and heat transfer partitioning during nucleate flow boiling. Water-based experiments have been conducted in a vertical channel under varying subcooling levels (ΔT_sub = 2, 5, and 8 K) and flow rates (Re = 2400, and 3600). Mask R-CNN machine learning (ML) model is employed to evaluate the temporal variations of dynamical parameters of vapor bubbles and departure characteristics. The bubble departure characteristics as well as the corresponding evaporative and convective heat transfer rates, as retrieved through ML-generated masks, showed a strong dependence on the level of bulk fluid subcooling and flow rates. Heat transfer partitioning analysis revealed a competing interplay between the evaporative and condensation components of overall boiling heat transfer rates. These findings demonstrate the potential of ML techniques towards optimizing the thermal performance of boiling phenomena that have significant implications in a range of critical engineering applications.