[Trinxo123]

Hello,

I have been performing university work where I have to validate some CFD simulations of a 3D axial fan that serves ventilation purposes. The fan is formed by an external circumferential structure, an impeller with 8 blades and a motor, see schematic picture 1 attached.

The simulations I am conducting underestimate the thrust given by the fan by 25% and underestimate the torque in the impeller walls (which determines the engine power) by 35%. Such bigger differences seem a bit weird to me when I have always found high similarities between experimental and CFD data using proper CFD simulations.

I have intended MRF SimpleFoam with K-w SST simulations with a wall-modelled approach (30<y+<300), the same with a wall-resolved method (y+ approx. 1) and also Sliding Mesh approach, as I was worried that the differences were due to the 'false swirl' effect of the MRF. Unfortunately, the 3 gave similar results very off from experimental values. A Grid study has also been done to eliminate mesh sensitivity. At this point, I am only presenting the case, with the geometry and boundary conditions to see if anyone with experience in turbomachinery can help me out a bit.

See picture 2 for the simulation domain and boundary conditions utilized, note that the fan is working in calm air and it's the fan itself creating suction to speed up the flow. It seems as if the CFD is unable to predict the amount of suction, resulting in underestimating both the thrust and power required.

[JBeilke]

Maybe there are some differences between the CAD an the real geometry. How did you mesh the impeller?

[Trinxo123]

Originally Posted by JBeilke

Maybe there are some differences between the CAD an the real geometry. How did you mesh the impeller?

First, thanks for your reply.

The likeness of the real model, the original CAD and the CFD model (mesh from a modified CAD) is a fascinating topic. The similarities between the CAD and the real model (manufacturing process vs CAD) haven't been established yet, however, they should be very very similar.

Secondly, there are differences between the CAD and the CFD model made on purpose, the real engine is quite complex (full of cooling fins) that are highly simplified in the CFD model (revolution body), the impeller is not simplified, while the hub is slightly simplified to get a better quality mesh. Nevertheless, simplifying the engine is in theory reducing the losses of the fan, therefore, if any discrepancies, more thrust and less torque should appear. Making the engine more complex and like the real physical model has been intended, resulting in even more discrepancies with experimental values and has been discarded.

Lastly, the mesh is created in Fluent Meshing with polyhedron generation (meshes vary between 2-6 million cells) and then converted into foam with fluent3DMeshToFoam.
Thanks,
Marc

[JBeilke]

If you have the mesh in Fluent, you can also try a simulation there to make sure the results remain consistent.

Adding the extra details of the casing will probably increase the torque and also affect the laminar-turbulent behavior of the boundary layer.

Can you post a picture of the mesh around the leading edge? There are some good articles on the GridPro blog about meshing impellers. Maybe you can get some ideas for the required mesh resolution from that.

https://blog.gridpro.com/multiblock-...hip-propeller/

[Trinxo123]

Hi Joern,

The simulations have also been done with Fluent and the results are consistent with OpenFoam simulations.

Regarding the fidelity of the impeller in the manufacturing process, the impeller is a twisted steel sheet, so there are no casting imperfections and almost no roughness. I have attached some images of the leading edge and also the inflation layers, maybe you see something wrong. I did check the blog you posted, but I believe I am not able to do this kind of mesh with the software I have (maybe with Ansys Meshing instead of Fluent Meshing I could try), but I hardly believe that is going to make a difference, as very fine meshes have been already tried.

Thanks for your help.

Marc

[JBeilke]

Hi Marc,

The mesh looks quite good at first glance. If there are no differences in the result between the different resolutions, it is most likely due to something else.

How do you measure the thrust in the experiment and in the simulation? Is it really the forces or do you determine the thrust via the momentum in the outlet plane of the fan? In this case, we also have to take into account the part of the flow that flows around the outside of the housing.

[Trinxo123]

Thanks again Joern for your answer,

I haven't been in the experimental trials since they were done long ago by my university. Still, I've been told that the fan was mounted on a table with a moving platform attached to a force sensor, basically measuring the force forward done by the air exiting the outlet section.

In the CFD simulation, I am using just the average velocity outlet to calculate the force knowing the area section and the density. I could introduce also the pressure contribution (Poutlet-Po)*A, but it's very small and will not fix the discrepancies.

Marc

[JBeilke]

If the forces were measured in the experiment, I would do the same in the simulation. Simply add up all the forces acting on the housing and the impeller.

[Trinxo123]

Originally Posted by JBeilke

If the forces were measured in the experiment, I would do the same in the simulation. Simply add up all the forces acting on the housing and the impeller.

Just tried it and the thrust results in a slightly different number (probably due to the outlet pressure I didn't take into account and other small forces in some walls). Nevertheless, the discrepancies are a bit higher now in the thrust as it has been reduced, so a bit more underestimated. By the way, thanks for the tip as it's a lot simpler to calculate this way and I did not think about that.

Marc