The integrated behavior of the fluid flow, encompassing variation of some specific response, has received more attention than mere examination of velocity and pressure distributions. A machine learning framework is introduced for the first time to elucidate and predict the complex fluid dynamic responses across varying time scales. For the same class of fluid processes, the dynamic response curves with varying time durations can be defined in terms of time scales and curve shapes. The former is predicted by a single-objective regression model, and the curves are normalized and recovered using time scaling and uniform interpolation. The latter are downscaled by Principal Component Analysis (PCA) and then predicted by a multi-objective regression model. As a novel application, this framework is employed to analyze small bubbles in a Newtonian fluid as they impact a horizontal wall, predicting the velocity, aspect ratio, and position dynamic response of the bubbles. Both Random Forest and Gaussian Process models, based on PCA-driven data processing, exhibit remarkable accuracy in describing and predicting complex dynamics of rising bubbles. Notably, PCA's principal components demonstrate independence and are set through rigorous cross-validation of the training set, rather than dependence on the Cumulative Variance Explained (CVE) values. Furthermore, our framework adeptly extends to predict the complex behavior of multiple bubble bounces by concatenating individual bounce predictions with high efficiency and precision. The proposed framework proves to be a catalyst for comprehending the dynamic response of fluids.
Read full abstract