Is Spalart Allmaras DES still the best turbulence model for external aero?
 

I'm new to CFD and have been running many simulation based on a refinement of the motorbike tutorial. I have come to the conclusion (that I think everyone else came to decades ago) that the k omega SST RANS model is not taking into account the effects of vortices in a large turblent wake sufficiently.

So a little googling and I came across this study done by Daimler in 2013 on CFD modeling of trucks that came to the conclusion that the Spalart Allmaras DES model was within 2% on cD compared to wind tunnel tests vs. 20% errors with the RANS model. They discount the use of an LES model due to the computational costs.

Has anything changed in the last 5 years that would make this untrue? Are there merits to the IDDES model over this one?

Since you are new in CFD, the first thing you must know is that RANS/DES/LES are different formulations not different turbulence models. That is the equations that are solved imply differente assumption on the field decomposition in averaged/filtered and fluctuations fields.

Said that, DES is a hybrid formulation where RANS is used close to the wall and LES formulation describes the wakes. However, this formulation has its advantage and disadvantage. And DES is still under improvements.

There is no way to define "the best" formulation if you do not specify before the goal of the simulation.

Thanks -

The goal is to use the CFD results to evaluate iterative design changes over a baseline to reduce drag vs. a baseline design.

Thanks.

Well, the more important thing is to be careful in the grid refinement close to the wall. If you want to compute the viscous drag you have to describe the boundary layer and that can be done by a proper distribution of the grid nodes, at least 3-4 nodes within y+<=1. Consequently, in RANS you have to adopt the congruent turbulence model.

DES shouldn't ever be less accurate than RANS since the least you can do with DES is get the RANS solution and the most accurate you can achieve is the LES solution.


What is the question here? Are you asking if the same Spalart-Allmaras model which is unchanged suddenly produces different results? Then no.