[Jelmer]

Hi all,

I've been trying to do 2D car simulations with Fluent, but I keep getting the same problems.

I am using the k-w SST model with low-Re correction. We need this to accurately predict the separation behaviour of the flow. The y+ values on the car and ground are both order 1. However, the residuals do not properly converge to low values (> 1e-4 ), and quite often, the velocity field contains strange non-realistic artefacts.

We don't really understand what's going on. With coarser meshes, no low-Re corrections and y+ ~ 100 we do get realistic results (however residuals still are higher than we would like), and drag trends due to changing the carshape are realistic too.

Some things about the mesh:
- We use a partially structured mesh, made in Gambit 2.4.6 (see attachments). It is a sample geometry (4500x1200 LxH) with an offset layer of 10cm, which represents the boundary layer. This layer is meshed structured.

- Around the car, a rectangular box is fitted. The area in between is meshed with a pave scheme (quad pave in this example). The rest of the domain is structured mesh again.

- There could be a problem with the maximum element aspect ratio. The elements on the car/ground are fractions of a millimeter thick, while their length is much larger ( AR > 10^3). I have heard however that high AR can be ok if the gradient in length direction (streamwise direction) is small. In the coarser mesh this AR is much lower.


Any help would be much appreciated! What is causing my simulation to perform badly?

Greetings,
Jelmer

[Far]

Originally Posted by Jelmer

Hi all,

I've been trying to do 2D car simulations with Fluent, but I keep getting the same problems.

I am using the k-w SST model with low-Re correction. We need this to accurately predict the separation behaviour of the flow. The y+ values on the car and ground are both order 1. However, the residuals do not properly converge to low values (> 1e-4 ), and quite often, the velocity field contains strange non-realistic artefacts.

We don't really understand what's going on. With coarser meshes, no low-Re corrections and y+ ~ 100 we do get realistic results (however residuals still are higher than we would like), and drag trends due to changing the carshape are realistic too.

Some things about the mesh:
- We use a partially structured mesh, made in Gambit 2.4.6 (see attachments). It is a sample geometry (4500x1200 LxH) with an offset layer of 10cm, which represents the boundary layer. This layer is meshed structured.

- Around the car, a rectangular box is fitted. The area in between is meshed with a pave scheme (quad pave in this example). The rest of the domain is structured mesh again.

- There could be a problem with the maximum element aspect ratio. The elements on the car/ground are fractions of a millimeter thick, while their length is much larger ( AR > 10^3). I have heard however that high AR can be ok if the gradient in length direction (streamwise direction) is small. In the coarser mesh this AR is much lower.


Any help would be much appreciated! What is causing my simulation to perform badly?

Greetings,
Jelmer

You need to work on your mesh. Even better try ICEM for high quality structured mesh.

[Jelmer]

Dear Far,

Thank you for your response. What kind of improvements would you suggest on the mesh? I can reduce the aspect ratio, but then the amount of cells will be enormous ( now its already ~100k )