Hi,

I computed the resistance of a ship using CFD. When I refine the grid, the resistance becomes smaller (which is closer to the experiment). My question is, is this a general thing? If it is, why? If it isn't, what is the general thing you can expect by refining the grid. FYI, I used k-omega turbulence model without transient motion (steady state), which should be close to reality as the Froude number is low enough to generate wave and the Reynolds number is high enough to suppress any transient motion such as vortex shedding.

Best regards, -Arman-

That sounds pretty realistic. Somebody asked yesterday about the drag prediction workshop, and I think that still remains a very valuable resource for making mesh fineness decisions. Of course, the situation is more complex for a ship. One should really go through the full exercise of refining the mesh until the answer doesn't change any more, or you can try to extrapolate to find what level of mesh refinement is sufficient. However, as the DPW has shown, this is not as simple as it is sometimes made out to be, and it may be important to figure out just how far you need to take this process. A lot of the time you can make the design decisions that you need based on consistent meshes rather than on mesh-independent answers.

Arman, the accuracy of a CFD code is related to the size of the mesh. As you make the mesh size smaller, the error in the solution gets smaller. This error due to mesh size is called discretization error. It is always good practice to run your analysis on smaller and smaller grid spacings until the answer does not change. This is called obtaining a grid independent solution. It is important to note that this does not necessarily mean that the solution get more accurate with respect to an experiment. Other errors in the analysis (accurate geometric representation, turbulence model, boundary conditions etc) also affect the accuracy. In fact it is fairly common to have the discretization error offset the error to say turbulence modeling. Then as you refine the mesh and reduce the discretization error, you agreement with the experiment gets worse.

well, in computing resistance, or Drag, there are many parameters which affect the results. one of the most important of which is the level of accuracy the solution has detected the viscous flow near the body, which generally is better using finer meshes. this is true generally in coarse meshes which usually are used. but when the mesh is very fine, well sth else may happen!