Key factors in the use of DDES for the flow around a simplified car

• Systematic sensitivity study to investigate accuracy of DDES with varying grid, numerics and underlying RANS model.
• Demonstrated the potential for a loss of accuracy when using DDES with insufficient mesh refinement or too dissipative numerical scheme.
• Highlighted that common user-controlled parameters can yield differences on the same scale as using different approaches.
• Indentified ‘grey-area’ behaviour of DDES at point of separation which leads to erroneous prediction of flow.
• Demonstrated that this can be overcome via the injection of synthetic fluctuations upstream of the onset of separation.

The Ahmed car body represents a generic vehicle exhibiting key aspects of the 3D flow arising due to standard automobile designs. It is recognised to be a challenging test case for the turbulence modelling community; combining strong separation with a pair of counter-rotating vortices, which interact to produce a downstream recirculation region. In recent years this case has been extensively studied using a range of methods, with varying success. In general, conclusions have been made on the basis of the standard form of each model, while in the present work we focus on variants of the common Delayed Detached-Eddy Simulation (DDES) approach, in order to demonstrate its sensitivity to commonly varied aspects of its usage. We demonstrate that variations in the usage of a single approach can easily be of the order of those observed when using different approaches.Previous studies, reconfirmed here, indicate that the majority of standard single point closure turbulence models are unable to provide a satisfactory prediction of the recirculating flow region aft of the body. This holds regardless of mesh resolution, model selection or numerical scheme. These models under-predict levels of turbulence over the slanted back, leading to over-prediction of the size of the separation region. DDES can offer an improved prediction although, while better than URANS, the use of DDES in its standard form still falls short of equivalent results obtained from either wall modelled or wall resolved Large Eddy Simulation. In the present work we investigate four aspects of DDES in an attempt to identify mechanisms for improving DDES for this representative case: (1) the underlying RANS model, (2) mesh resolution, (3) numerical scheme and (4) the addition of turbulent fluctuations.We observe that with insufficient mesh resolution the DDES models produce worse results than the URANS models. While first order methods are inappropriate, the more commonly selected second order upwind scheme is also demonstrated to have substantial adverse impact. As mesh resolution is increased the influence of the underlying RANS model diminishes. In a zonal RANS–DDES approach, the domain was split in two just before the rear slant, and the upstream RANS solution is used to inform a DDES calculation via the superposition of synthetic fluctuations at the interface. This technique is demonstrated to substantially improve the prediction, whilst also reducing the overall simulation cost by virtue of a smaller domain size. The injection of synthetic fluctuations provides a more accurate level of turbulence at the onset of separation and thereby overcomes the lack of resolved turbulence in the initial separated shear layer.