Machine learning enhanced exploration of bubble dynamics beneath a horizontal wall

Abstract

A better understanding of the dynamic behavior of a single bubble rising and impacting on a horizontal substrate is crucial, yet it has traditionally been constrained by the necessity for demanding experiments and extensive time-consuming simulations. This work develops a machine learning-enhanced model to determine characteristic quantities for the dynamic behavior of rising bubbles beneath the horizontal wall based on Computational Fluid Dynamics (CFD) simulation. We establish an unprecedentedly extensive dataset from accurate CFD modeling, covering a vast range of values for input parameters: Galilei number (1∼100), Bond number (0.001∼20), and rising distance (0.1R∼40R). Three popular regression models (Support Vector Regression, Back Propagation Neural Network, and Random Forest) are trained to capture key features for the first bounce of a bubble after collision, exhibiting promising predictive capacity, generalization, and interpretability. Our current approach facilitates the generation of dense datasets, enabling finer interpolation and detection of the physical properties of the unknown fluids. This pioneering data-driven methodology paves the way for advancing our understanding and prediction of bubble dynamics, marking a significant contribution to fluid dynamics research.