Bubble Size Determination in a Half-Scale Curved Water Model Mold for Various Casting Conditions Using Imaging and Machine Learning

Parametric studies were performed in a 1:2 scaled, curved water model using shadowgraphy to estimate bubble sizes for different casting parameters such as gas flowrate, liquid flowrate and mold width. Bubble diameter calculations were based on a machine learning algorithm using ImageJ software. Bubble diameters were correlated with input parameters using a deep-learning algorithm. The model performance was determined based on the coefficient of determination (R2) and root means squared error (RMSE). The model showed significant promise with bootstrapping aggregation, validated with five-fold cross-validation and improved accuracy.