Large Eddy Simulation of Junction Vortex Flows

The aim of this work is to learn more about the complex turbulent flow around a surface mounted slender object, commonly referred to as junction flows. This type of flows are common in many engineering applications such as turbine blade-hub assemblies, bridge support pillars, aircraft wing fuselage junctions, and submarine appendage-hull junctions, and is thus of great interest. In addition, we are interested in comparing how different Computational Fluid Dynamic (CFD) methods, such as Reynolds Averaged Navier Stokes (RANS), Detached Eddy Simulation (DES) and Large Eddy Simulation (LES) models manage to predict this flow when compared with experimental data. The flow physics is very complicated, with between two and six vortices interacting in the region between the upstream separation line and the body depending on the boundary layer and Re number. In addition, the flow is highly strained, anisotropic, and strongly affected by (unsteady) small-scale transport, and hence poses a great challenge to all CFD models of today. Here, we compare predictions from RANS, DES and LES with the experimental data of Pierce & Shin. RANS predicts somewhat too late separation, and a resulting horseshoe vortex system with extremely weak legs downstream of the body. In addition, RANS cannot provide information about the unsteadiness of the flow. DES typically predicts too early separation, resulting in a horseshow vortex located too far away from the body, but with horseshow vortex system that prevails throughout the computational domain. Wall modeled LES appears to capture the location of the separation well, and thus also the location of the horseshoe vortex system that prevails throughout the computational domain