MULTI-FIDELITY PARAMETRIC SENSITIVITY FOR LARGE EDDY SIMULATION

Abstract

Designing engineering systems involving fluid flow under uncertainty or for optimality often requires performing many computational fluid dynamics (CFD) calculations. For low-fidelity turbulence modeling simulations such as Reynolds-averaged Navier-Stokes (RANS), such a framework has been established and is in use. However, for high-fidelity turbulence-resolving simulations such as large eddy simulations (LES), the relatively high computational cost of even a single calculation hinders the development of such a framework. The overarching goal of this work is to aid LES in becoming a usable engineering design tool. In this thesis, a computationally affordable approach to estimate parametric sensitivities of engineering relevant quantities of interest in an LES is explored. The method is based on defining a RANS problem that is constrained to reproduce the LES mean flow field. The proposed method is described and assessed for a shock/boundary layer interaction problem, where the shock angle and wall temperature are considered variable or uncertain. In the current work, a proof-of-concept of the proposed method is demonstrated. The method offers qualitative improvements to the sensitivity prediction of certain flow features as compared to standalone RANS simulations, while using a fraction of the LES cost. Different cost functions to infer auxiliary RANS variables are also examined and their influence on the sensitivity estimation is assessed. Overall, the results serve as an important proof-of-concept of the method and suggests the most promising path for future developments.