Physics-Informed Learning Accelerates Flow Prediction in Stirred Tanks

A new study from arXiv explores the use of physics-informed machine learning to predict steady flow fields in industrial-scale stirred vessels, aiming to reduce the computational cost of traditional CFD simulations. The research, published as arXiv:2605.07444, generates a dataset of steady flows using Reynolds Averaged Navier Stokes (RANS) simulations across realistic operating conditions, varying impeller speeds and liquid heights. The authors train implicit neural representations of flow fields, comparing purely data-driven models against those constrained by physical laws. They investigate the trade-off between training dataset size and predictive accuracy, evaluating model performance using global mean squared error. The work highlights the potential of machine learning as a surrogate for expensive fluid simulations, though it notes the need for large training datasets. The study does not specify a particular institution or author names, focusing instead on the methodological approach.