l1-based sparsification of reduced order models of high Reynolds number turbulent flows
Abstract
In high Reynolds number flows, dynamics are characterized by a large number of temporal and spatial scales. This multi-scale nature poses a significant challenge for model order reduction, because a large number of energy-containing scales needs to be retained in the model. Hence, the computational cost associated to propagating the model forward in time can be high. However, although the total number of scales is large, the interaction between them is rather sparse. In fact, dynamics of motions at a given length scale are determined by those of commensurate lengths, and not very much by significantly larger or smaller scales. In this project, we aim to develop a model sparsification strategy, whereby this physical feature is incorporated into well-established model order reduction techniques and exploited to develop sparse reduced order models that have the same overall prediction accuracy but a significantly lower computational cost. The algorithmic key of this work will be a set of sparsity-promoting statistical and machine learningtechniques that have gained significant interest in the fluid mechanics community in the last years. These will be used to unravel and identify the sparsity features in dense reduced order models of the turbulent flow past a representative delta wing geometry.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2017
- Source ID
- FA95501710324
Entities
People
- Andrea Da Ronch
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Southampton