Computation of a Simplified Centrifugal Blood Pump

[SRIB]

...the listing

[Yvan Fournier]

Hello,

Good to know you have a first calculation.

I would not worry too much about the Fourier number: it is related to heat transfer, and as you are not using heat transfer, you can ignore it (most of our computations use a thermal model, but we'll have to check if we can simply deactivate computation of the Fourier number when it is not useful).

The values of fraction an plane may have an impact on mesh quality, so they indirectly affect the results, but it is not easy to give recommendations: a higher fraction leads to less small faces, so probably better orthogonality, but more highly warped faces, which might lead to a locally degraded convergence order. If possible, staying in the 01. - 02. range is recommended.

As for the boundary layer, your listing says it is too refined in some areas, but for a better idea, postprocess the boundary y+ value. Depending in the turbulence model used, refining or coarsening is recommended (for turbulence, the best practice guidelines on this website probably provide a few simple recommendations, but for low-Reynolds models, you want to be around y+=1, for high Reynolds, I believe 5-30 is better, but I'll let you check).

Best regards,

Yvan

[SRIB]

Hello,

ok, i will try to find it out. Rij-SSG is a high Reynolds model, right?! I mean, i have a Reynolds number of arround 209338.

Regards,

Sebastian

[SRIB]

Excuse me, this was stupid. It is a TURBULENCE model

[SRIB]

Hello Yvan,

just something for your information and related to my questions about the relation between fraction and computed variables. I ran another computation with a fraction of 0.15 and a plane of 25. It crached after 60 time steps. But nevertheless i could compare the first 60 time steps with them of my first computation. About 95% of the variables are the same in both cases. The residual 5% divers in the region of 1e-03. So should be o.k.

Best regards,

Sebastian

[SRIB]

Hello Yvan,

in the BPG i could read that the y+ distance should be >20 for high Reynolds models. One opportunity would be to create an additional outer cell layer with a width of 0,1 x hydraulic diameter. What do you think about this?

I already searched in the internet but could not find something...i would like to visualize the velocity profile at the inlet in paraview (velocity over radial coordinate). Do you know how to do this?

Best regards,

Sebastian

[Yvan Fournier]

Hello,

If you are now familiar enough with SALOME's MESH module to add a cell layer (easy with the "viscous layers" hypothesis), do not hesitate to do it, as it will also help have a more regular mesh near the boundaries. The only thing you should try to avoid is to have rapid mesh refinement variations, so you should try to ensure the thickness out the outer cell layer and neighboring inner cells to be relatively similar.

Also, to adjust for y+, using the hydraulic diameter for a first estimate is good, but I recommend also visualizing the near-boundary y+ under ParaView (don't forget to "extract blocks" / "boundary"), check the effective y+ you obtain, and iterate if necessary.

Finally, I am not sure how to visualize velocity over a radial coordinate (I remember doing this with EnSight years ago, but have not checked which combination of filters you need to do this with ParaView). You might want to use a Code_Saturne 1D profile for this (in the GUI, you can define a profile over a curve defined by an equation), and choose to output it in csv format, which ParaView can read.

Best regards,

Yvan

[SRIB]

Hello Yvan,

one question, please. Can i adjust the y+ with help of the roughness height for rough walls in the GUI? I am still not able to create a viscous layer. It does not work. Don't know why.

Best regards,

Sebastian

[SRIB]

Hello Yvan,

how are you? As you know I had certain problems with the blood pump with respect to mesh joining. To fix the problems I meshed the interface region quadrangular to make it more regualar. Attached you can find a picture. The result is that I do not have to increase the fraction anymore (0.1 by default). But for some reason the courant number starts to increase dramatically after a few timesteps. I attached the listing. The only thing I thoud about is that the faces in the joining region are larger compared to the old mesh. Do you have an idea what could be reason for the high courant numbers?

Best regards,

Sebastian

[Yvan Fournier]

Hello,

Looking at your logs, it seems strange that the CFL number increases (actualy, the maximum increases, the minimum decreases), while the velocity remains stable (the latter being reassuring).

There might be an error in the ways the CFL is computed. Could you postprocess the CFL number to check where its behavior is strange (I assume at the joining interface, but it may be best to check).

Best regards,

Yvan