I have set up a simulation on a cross flow fan in an attempt to estimate the flow rate at 1500 RPM. I have environment pressure at the inlet and static (atmospheric)  pressure at the outlet. I have totally rebuilt the fan impeller as a solid body and encased it in a cylinder (which I have selected as the rotating body and not included it in the analysis). The housing is also a solid body. Because it is a cross flow fan I should be able to run 2d simulations on it. Neither 2d or 3d seem to be working. Particularly disturbing is:

A) The distinct boundary line between the impeller body and the surrounding air. That tells me that there is very little interaction going on at the boundary.

B) The flow velocity vectors seem to all go towards the fan!

C) It takes a very long time to run in 2d.

Capture1 and Capture9 are pistures of the 2d simulation solver.

1) What should I use for the conditions at the inlets and the outlet of the simulation?

2) How can I get rid of the boundary between the fan body and the surrounding air? I already reduced it to only 5 thou larger than the blades. Better question, do I even have to play the game of encasing the fan in a rotating body rather than just rotating the fan? Are there solvers out there that can do this? It really reduces my confidence that what I'm simulating is realistic. I understand that the boundary should go away as iterations progress but it doesn't seem to.

3) Meshing? I have tried narrow channel refinement, small solid body refinement, local initial meshes for the impeller, pretty much every advanced mesh option and all kinds of different mesh sizes. What is best for this type of study?

4) The blades are sheetmetal features since that's how they are made. Could this be a problem?

As I write this i tried 3d meshing and got decent results (capture8_3d). (flow vectors flow the right direction). But it takes a while and I'm trying to optimize the design so it would be preferable to have a fast 2d simulations so I can batch run with all my configurations. With 2d can get a mesh thats actually smaller than my fan blades which makes me feel better about the accuracy of the study.