Skip to the content.

Data-Driven Model Reduction, Scientific Frontiers, and Applications ()

Suparno Bhattacharyya, Institute of Data Science
Uncertainty Quantification with Hyper-Reduced Order Models

Abstract

Data-driven Reduced Order Models (ROMs) are crucial to Digital Twins, enabling efficient real-time simulations of complex systems. However, optimal performance requires a deep understanding of the associated uncertainties. Uncertainty quantification (UQ) is increasingly critical across industries for system analysis, predictive maintenance, and real-time decision-making. For example, Digital Twins in nuclear reactors must account for uncertainties in temperature and material properties, while aircraft engine Digital Twins assess uncertainties in sensor data and environmental conditions.

In this presentation, we introduce a specific class of reduced-order models: hyper-reduced order models (hyper-ROMs). Unlike traditional ROMs, hyper-ROMs are designed to maximize computational efficiency, especially in nonlinear scenarios, and have shown significant potential for integration with Digital Twins. UQ typically involves many-query computations, which can be computationally expensive, particularly with nonlinear systems. We show that hyper-ROMs make UQ more efficient and practical, for complex, high-dimensional problems.