Development and evaluation of data-driven modeling for bubble size in turbulent air-water bubbly flows using artificial multi-layer neural networks

•This first data-driven modeling of bubble size for a closure model with validation.•Bubble sizes predicted using the model are much more accurate than previous models.•A regime-adaptive data-driven model for bubble size & lift for successful validation.•The devised model works for both wall and core peaking regimes using a single model.•Sensitivity analyses along with PCA and random forest method for important parameters. In the present study, we consider a new reliable model of the bubble size based on multi-layer artificial neural networks (ANN). A multi-layer ANN is used to establish a function for the bubble size without any assumption on the form. In the training procedure, the proposed ANN is trained using data sets collected from open literature and experiments performed in the present study. An excellent agreement was obtained between the trained ANN and experimental data in the bubble size. Also, sensitivity analyses along with principal component analysis and random forest method provide important physical parameters for the bubble size. Next, in order to rigorously evaluate the prediction capability of the present model, flow simulations were conducted for turbulent bubbly flows, for which experimental data are available. The present validation results show that a regime-adaptive data-driven model for the bubble size achieves successful estimation for both wall and core peaking regimes.