Learning hidden physics from reduced-order modeling of bubble dynamics in boiling heat transfer

Boiling heat transfer is a highly effective but stochastic process where the working fluid undergoes vigorous liquid to vapor transition. Physics-based modeling of bubble dynamics during boiling is challenging due to the drastic changes in system parameters, such as nucleation, bubble morphology, temperature, pressure, and velocity. Principal component analysis (PCA), an unsupervised machine learning (ML) approach, is used to extract hidden physics from boiling heat transfer processes using in-house pool boiling experimental images. PCA relies purely on reduced order representations of bubble images and can effectively extract the dominant physical dynamics with a significant reduction in data size. The behaviors of the dominant frequency and its associated amplitude of the PCA versus the heat flux can be linked to the bubble nucleation site densities, bubble departure, coalescence, and morphology. The current approach automatically detects the boiling regimes, without data labeling and human errors.