Machine Learning for Turbulent Multiphase Flows

We develop machine learning algorithms to predict spatiotemporal behaviors of turbulent, non-reacting and reacting multiphase flows to facilitate design space explorations of practical engineering devices. Our algorithms are not limited to predicting specific global quantities, but we predict all the details including interfacial instabilities, turbulent statistics, and spatiotemporal evolution of all relevant quantities.

Please right click on the video, choose show controls, and click the play button to play the videos.
Comparison of diesel jet inejction and atomization predicted by machine learning (right) with DNS-based computations (left). Comparison of flow dynamics when air flows over a circular cylinder predicted by machine learning (right) with DNS-based computations (left).
Relevant Publications
  1. H. Ganti & P. Khare "Data-Driven Surrogate Modeling of Multiphase Flows using Machine Learning Techniques"" [PDF]
    Computers & Fluids, 211, 2020, 104626. DOI: https://doi.org/10.1016/j.compfluid.2020.104626.
  2. H. Ganti, M. Kamin & P. Khare "Design Space Exploration of Turbulent Multiphase Flows Using Machine Learning-Based Surrogate Model" [PDF]
    Energies, 13(17), 2020, 4565. DOI: https://doi.org/10.3390/en13174565.