Learning Forum

[meiermat]

I am modeling a small motor in Icepak, linked to Maxwell through the Workbench.  I am having convergence problems in Icepak and I think it's a meshing issue.  When I generate a mesh using the defaults I get poor face alignment.  In a different thread it was suggested that all CAD objects be meshed using the multilevel mesher, but that doesn't seem to help.  Sometimes the multilevel mesher actually generates a different error wherein I get negative cell volumes, but sometimes it does complete...just with poor face alignment.

The culprit cells are in many different places in my model...some in the hollow region inside the steel, some in the steel itself, and some where the steel meets the air.

I've tried to "massage" this away with different mesh settings, but I'm not really sure what to try.  Nothing in this model is an Icepak object except the cabinet, which I'm really just using for the fluid boundary conditions, so I'm somewhat limited in what type of mesh I can even use.

Is there anything in particular I should be looking into, or am I reaching a brick wall here?

Thanks,

Matt

[Ankit Adhiya]

Hello,

 

>> In a different thread it was suggested that all CAD objects be meshed using the multilevel mesher, but that doesn't seem to help.

Multi Level options helps to capture objects of different scales with reduced mesh count using hanging node mesh option. Attach image.

 

 

Mesh quality highly depends on quality of CAD. If the quality of CAD is good, I can suggest on

1) Try changing the global mesh parameters to see if this improves the quality

2) We can also select the "Set Uniform mesh" option in Global mesh panel

In the latest version you can also import the external mesh (e.g. ANSYS Mesher) and use it in Icepak.

Best Regards,

[meiermat]

Thanks for checking in.  I think your CAD quality comment is on target.  I adjusted how I'm running this simulation and it has helped significantly.  Now I'm building the cabinet in Icepak within the AEDT environment, instead of as a standalone program in Workbench.  Then I copy-and-paste all the relevant parts from the Maxwell model (which I draw entirely within Maxwell) into the Icepak model, so I don't have to go through the geometry editor.  This is a slight disappointment because now the models aren't linked for temperature-related properties, but I can iterate that manually.  This seems to have solved the meshing errors, and the mesh quality is significantly better due to the objects being generated in the same environment.

However, I'm still having the same convergence problems.  I do have some regions that are much thinner than others.  In Maxwell I have to take care to refine the mesh in the air gap and at the tooth/magnet tips, and anywhere that the magnetic circuits might get complicated.  I assume the same holds true where the heat flows are intricate, like in the slot liner.  However, when I refine the mesh specifically in this region I get this series of errors:

Solver exited with error.

Failed to copy local file c:.../setup1.res to file m.../fields.res.  Error:  The system cannot find the file specified.

Failed to copy local file c:.../setup1.nc_cas to file m.../current.nc_cas.  Error:  The system cannot find the file specified.

Simulation completed with execution error on server:  Local Machine.

If I make the entire, global mesh about one quarter the default size then it takes forever to run and I still get convergence problems, although the convergence is better.  I can't take things too much further or I'll run out of disk space on the university servers, where all the simulations have to run.

Matt

[Ankit Adhiya]

Hello Matt,

 

I am not sure on size on small gap. But if you need good localize mesh control, did you also explore ANSYS/Fluent mesher and Fluent solver for this application. As you may need good localize mesh control.

 

 

Regards

[meiermat]

The gap is going to vary as a parameter in my analysis, but it's 1mm right now.  It might go down to 0.5mm because the design is pretty small (2" in diameter), but this will depend on the stresses on the rotor.

I specifically am using Icepak because I can integrate it with Maxwell, and was told that Icepak was basically the same thing as Fluent, just tailored to integrate with electromagnetics work.  Is this not true?  If not, is there a way to integrate the Fluent mesher with AEDT, or do I have to go back to Workbench and go Maxwell => Geometry => Fluent => Icepak?

Matt

[Ankit Adhiya]

Hello Matt,

As the gap is small 1mm and lower than that. I do not expect much convection in this region, the mode of heat transfer in this region would be conduction. So I can suggest following for Icepak. You can fill/remove the gap, and create a surface in region where there was small gap in pre-processor. In Icepak you can define this surface as thin conducting plate and specify the material as air-solid. Where air-solid is user defined material and having thermal conductivity of air. You can read more on thin conducting plate in Icepak documentation. There wont be much difference in result, with this approach.  

Hope this helps. 

Regards,

 

[meiermat]

I think I understand where you were going with that, and I've done quite a bit in the past few days since your suggestion, so this might be a little long.

Filling the gap, I anticipate, was to remove any "difficult" regions for Icepak, especially when it comes to air flow.  The air in the gap is moving fast (the rotor for this machine is ~80,000rpm), but modeling it as thin conductive layer is quite standard.  Sadly it didn't affect the convergence problem, but it sent me off along several paths that, while they may not be fruitful, should provide some insight.  I'll summarize everything I've done to this point since my first post back in February, so it's all in one place.  I can't seem to post this in one piece, but it is letting me post segments, so this will span the next couple spots.

[meiermat]

1. I'm building the model in Maxwell, and have dropped Workbench.  When I'm ready for CFD analysis, I copy-and-paste the objects into Icepak within AEDT, and the mesh generated by Icepak shows a good quality (I think...I'm not a mesh expert).

2.  I can run the model and it "converges", at least as far as the pre-defined convergence limits.  More on this later...it doesn't actually converge, but because the simulation actually reached the convergence limits at the time I thought the model was OK.

3.  I need to model the heat flow restrictions between the coils and between the coils and steel, but I'm having a lot of difficulty getting them set correctly.  I mentioned this in a different post recently, but I haven't gotten any response yet, so this is still an open issue.  To work around this, I redrew the motor with 0.025mm insulation between the coils and steel and between adjacent coils.  This model is ridiculously large in Maxwell because of the mesh in the insulation, but it does run.

4.  When I copy this one into Icepak I get convergence problems.  The simulation initially starts converging, but before it reaches the limits the Energy part starts to diverge wildly, and the velocity components do the same soon after.  If I catch the simulation at the best convergence point then the results seem logical.  However, if I stop the model 10 iterations earlier or later then the results vary too much for me to trust them.  Because this has been my "working" model, we've discussed it at length in my group meetings and with anyone who would listen.

5. At that point, we figured the obvious thing was the difference in the models...the insulation.  The natural step was to try and refine the mesh in the insulation, but that's when I got the errors I mentioned above.  One of the other guys in my lab is an ME, so he had some insight into the CFD part.  At his suggestion about a month ago, I removed all sharp corners from the model, thinking that was messing up the CFD computation.  Qualitatively it didn't help, but this does go along with your suggestion that I block off the air gap.

6.  I went back to the original model, so I could look at the overall system temperature.  That's when I noticed that the model without insulation also doesn't converge well.  I see the Energy part converge quickly, but before the flow parts reach their limit, the energy part has started to rise.  I tightened the tolerance, and this model also diverges just like the one with the insulation layer.  I just didn't notice because it stopped at its convergence settings, so I just assumed it was alright.

7.  If I let the model diverge, I get ridiculous temperatures.  The hot spots are in the thousands of degrees, while the cold spots are thousands of degrees below zero.  They keep getting worse the longer the simulation runs, but I can isolate the extreme hot/cold spots.  They're mostly at a point on the underside of the coils, near where the coils meet the stator, but if I stop the simulation at the right spot then I can see some cold ridges along the coil outside corners as well.

8.  Again, I went back to the geometry and smoothed out the corners, including where the hot spots are, but no luck.  Trying to refine the mesh, I improved the mesh on the coils, but I get the aforementioned error.

9.  If I use the finest default settings for the global mesh the solution reaches a nadir for convergence and then rises, but doesn't continue to diverge.  Instead, it reaches a steady state.  The bottom is around 35 iterations.  The steady state is reached by iteration 100, and stays there for at least another 500, so I know it's stable.  It doesn't give insane results, but it also doesn't give physical results.

10.  After your suggestion, thinking back to what Steve had said, I isolated the thermal and flow parts.  If I just simulate the fluid flows (just air at this point), it makes no difference whether the gap is filled or open, or whether there are sharp corners in the model.  The CFD-only model converges just fine to a perfectly reasonable solution.  I can vary the incoming air speed up to 100m/s and the convergence is very fast and consistent.  If I do thermal only then the model gets really hot because now it's just conduction through air, but the whole machine is close to the same temperature, which makes sense.

11.  I did find that the errors seem to appear when I have the default global mesh and a very fine mesh on just one object, so I refined the global mesh and then targeted the mesh in the coils.  However, this created a model that was too large and I got warned by the university after they had to abort the simulation.  I know this isn't a viable approach, but I was hoping it would continue to give some direction.

That's where I'm at.  I hope that provides some more insight to what's going wrong.  In the meantime I've been just running the simulation with a really short leash, stopping it at 35 iterations before it can run away, so I think I can use that for comparing different machines.  However, we're at the point now where I need to take something from the virtual world to the real, and before that happens I'll need to figure this out.

Thanks again for the time,

Matt