When
Where
Student: Criston Hyett, Program in Applied Mathematics
Title: Applications of Scientific Machine Learning in Fluid Dynamics
Advisor: Misha Chertkov, Program in Applied Mathematics
Location: Math building, Room 401 and zoom https://arizona.zoom.us/j/86262993084
Abstract: Machine learning (ML) has ushered in a new age of mathematical modeling. Fluids, a ubiquitous component of engineering and fundamental research with high simulation cost, are a natural focus for efficient modeling using ML. This dissertation examines the application of structured ML from a physics-first perspective.
We present two differing applications of ML in fluids:
(1) The construction of a physics-informed, autoregressive ML model to close the equations of the Lagrangian velocity gradient tensor in isotropic turbulence. Besides improving the state-of-the-art, we interpret the model to uncover an interesting connection between the evolution of the pressure Hessian and the history of the strain-rate tensor.
(2) Using differentiable programming to perform PDE-constrained, (physics-adherent) optimization for contingency planning in natural gas networks. We empower network operators to utilize optimization with this proof-of-concept approach to unifying the gas flow simulator and optimizer.
Taken together, these applications show the potential of ML - appropriately applied - to advance understanding, increase fidelity, and enable engineering robustness. This work was advised by Dr. Misha Chertkov.