Truly Predictive Reduced Order Modeling for Complex Multi-scale, Multi-physics Problems

Join Professor Karthik Duraisamy from the University of Michigan in a seminar presentation exploring predictive reduced order modeling for complex multi-scale and multi-physics problems. Delve into existing data-driven modeling approaches and understand what makes models 'truly predictive.' Using rocket engine combustion dynamics as a case study, examine the intricate coupling between chemical reactions, heat release, hydrodynamics, and acoustics. Learn about innovative approaches to improve model robustness through structure-preserving transformations and discretely consistent least squares formulation. Discover how adaptive formulation and non-local procedures enable predictive capabilities in future state and parametric problems with minimal offline training. Explore the development of multi-fidelity frameworks where component-level ROMs are trained on small domains and integrated for full-system predictions, demonstrating enhanced predictive capabilities and the ability to capture emergent phenomena.