Seminar

On Tuesday, March 3, 2026 at 12:15 p.m., in the Prof. W.R. Gundlach Auditorium, room 230 at 217/221/ Wolczanska Street, the seminar on Reduced order modeling of CFD solutions with Neural ODEs will be held. The meeting will be hosted by: 

Matthew Schulz 

Masters student in the Department of Aerospace Engineering at Texas A&M 

A reduced-order model (ROM) was used to parametrize CFD solutions of single- and two-phase flows. This ROM consists of a set of proper orthogonal decomposition (POD) modes and a neural ordinary differential equation. Given a set of full-order model (FOM) flow solutions, the POD provides spatial basis functions which minimize projection error. Instead of using the Galerkin projection, where the governing equations are projected onto the reduced basis, a NODE was trained using the FOM data to predict the evolution of flows in the low-dimensional POD space over time. Given a sufficient dataset, this reduced-order modeling approach can be applied to any parametrized flow problem. Once the model was developed, flows with new parameters were predicted with computational speeds over four orders of magnitude faster than the full-order model.

We invite everyone interested in the topic of the presented work to participate in the seminar.