University Consortium for Applied Hypersonics Seminar Series: Efficient Aero-Database Construction with Weighted Active Learning

Abstract:

Constructing a well-characterized aero-database is essential for understanding the flight characteristics of a vehicle. High-fidelity CFD simulations can be time-consuming to generate. Generating samples in an efficient manner can greatly reduce the total simulation time. In practice, standard approaches often consider the whole space to be equally important, but there are many situations where this assumption does not hold. For example, the surrogate model is often queried non-uniformly and this information can be utilized to better refine the database. We develop an efficient approach to weighted active-learning for Gaussian process and multi-fidelity cokriging models. We apply this approach to multiple examples to demonstrate potential use cases and show its effectiveness in practice. 

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.