Projection-based Model Order Reduction and Hyperreduction of Turbulent Flow Models
Speaker: Sebastian Grimberg, Stanford University
Presentation
Abstract
Despite significant advances in the simulation-based engineering sciences in recent decades, time-critical applications struggle to take advantage of high-fidelity, partial differential equation-based computer simulation. This is due to the large processing time and storage requirements associated with large-scale computational models. Projection-based model order reduction (PMOR) methods offer the ability to dramatically reduce this computational cost by generating compact, low-dimensional models for which solutions can be obtained in near real-time while still retaining the accuracy of an associated high-fidelity, high-dimensional model for the time and parameter domain of interest. PMOR is thus an essential technology for the application of model-based control, probabilistic analysis, or design optimization to problems involving increasingly complex engineered systems and physical phenomena. Unfortunately, nonlinear problems, in particular turbulent computational fluid dynamics (CFD) applications, continue to present a number of challenges for constructing stable, accurate, and computationally efficient projection-based reduced order models (PROMs).
This talk will address some of these challenges, demonstrating in the end PROMs for a number of large-scale, unsteady, turbulent flow applications which showcase the potential of PMOR for nonlinear CFD models with multiscale physics. These results leverage advancements in hyperreduction, a technique for the treatment of nonlinearities in the underlying computational model, to obtain CPU time and wall-clock time speedup factors of several orders of magnitude. This talk will also present an investigation of the role of projection in the numerical stability of PMOR for convection-dominated flow problems, ultimately disproving an often-stated claim in the literature with the support of several numerical examples.
Biography
Sebastian Grimberg is a Ph.D. candidate in the Department of Aeronautics and Astronautics at Stanford University and is advised by Charbel Farhat. His current research focuses on projection-based model order reduction methods for computational fluid dynamics and computational mechanics problems. Prior to Stanford, Sebastian graduated from Princeton University in 2015 with a B.S.E. in mechanical and aerospace engineering, completing the program in engineering physics.