Loading...

Quantitative comparison of the analytical computer-vision baseline, XGBoost and multimodal deep-learning fusion models for prediction of nozzle–wheel gap and optical-flow-based flow proxy. Performance metrics are reported using group-wise cross-validation by casting run using mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R2). Both machine-learning models substantially improved gap prediction compared with the analytical baseline, while prediction of the optical-flow-based flow proxy remained considerably more challenging for all methods, indicating that foil flow behaviour depends on process dynamics beyond the image-derived descriptors available in the present study. - Figshare | NLM Dataset Catalog