Towards industrial large eddy simulation using the FR/CPR method

Abstract

NASA’s 2030 CFD Vision calls for the development of accurate and efficient scale-resolving simulations for turbulent flow, such as large eddy simulation (LES) and direct numerical simulation (DNS). This is primarily because the Reynolds-averaged Navier-Stokes (RANS) approach has failed to predict vortex-dominated flow involving large flow separations, e.g., flow through a jet engine or over aircraft near the edge of the flight envelope, i.e., during take-off and landing at high angles of attack. Although the DNS approach resolves all turbulence scales, it is too expensive in the foreseeable future for real world flow problems because of the disparate length and time scales in the flow. LES resolves the energetic large scales while modeling the smaller scales, so it provides a good compromise between accuracy and cost. As a result, LES is widely considered to be the method of choice for next generation CFD design tool. The major obstacle for LES is its considerable computational cost since unsteady 3D simulations need to be performed to obtain the mean flow quantities such as the drag and lift coefficients. In order to resolve the dominant scales in a turbulent flow, numerical methods used for LES should have low dissipation and dispersion errors. This means standard second order finite-volume methods are usually not accurate or efficient enough for LES applications. High-order methods (order of accuracy 2) have demonstrated their potential for LES and DNS in the past decade because of their low embedded numerical dissipation and dispersion errors. In the present study, we develop and demonstrate a recently developed high-order method, called flux reconstruction (FR) or correction procedure via reconstruction (CPR), for industrial LES. A major advantage of the FR/CPR method is its capability to handle unstructured mixed meshes, and its compactness and scalability, which is particularly desired on modern super-computers. We therefore address the following major pacing items in industrial LES in the present study: High-order methods Geometric flexibility Efficient time integration Efficient implementation on modern super computers Demonstration for real world applications