Transient analysis of coil using Ansys Maxwell (2018)

[highvoltageguypk]

Hi every one,

Anyone interested to learn about the transient analysis of a coil, let me know. 

Regards,
highvoltageguypk.

[klugesmith]

I am sure interested. Please tell us more.

Not to divert your thread, but here's another project wanting a proficient user or teacher of Ansys Maxwell.
This would be my DC electromagnet with about 100 pounds of steel.
Simple 1-dimensional reluctance model works when air gap is tiny, but fails miserably with useful air gap length.
I'm sure because of leakage flux, due to exceptionally slender geometry.   Need a real 3D simulator; it can't be adequately represented in 2D axisymmetric or planar simulator like FEMM.  It would inspire lab flux measurements to confirm the simulation. Picture shows BTI next to Mullins cyclotron magnet, in a fusor.net discussion from 2018.

[Twospoons]

You have access to Ansys transient sims?  Now I'm jealous! That is one seriously expensive bit of software.

[highvoltageguypk]

I am sure interested. Please tell us more.

Not to divert your thread, but here's another project wanting a proficient user or teacher of Ansys Maxwell.
This would be my DC electromagnet with about 100 pounds of steel.
Simple 1-dimensional reluctance model works when air gap is tiny, but fails miserably with useful air gap length.
I'm sure because of leakage flux, due to exceptionally slender geometry.   Need a real 3D simulator; it can't be adequately represented in 2D axisymmetric or planar simulator like FEMM.  It would inspire lab flux measurements to confirm the simulation. Picture shows BTI next to Mullins cyclotron magnet, in a fusor.net discussion from 2018.

Hi klugesmith,

Sure I would love to give a try but geometry is not much clear to me. If you can share some more views so that I can understand it clearly. And also information about the number of turns, core, excitation circuit will be helpful. Thanks. 

[highvoltageguypk]

You have access to Ansys transient sims?  Now I'm jealous! That is one seriously expensive bit of software.

Dear Toospoons,

People of 3rd world country don't need to worry about the price.

[klugesmith]

Welcome to the forum.  We see your personal introduction posted in March of this year.

Threads are easier to follow if short replies do not Quote the whole previous message.

Let's figure out a private-message or email channel for magnet discussions.

3rd world country?   I don't remember if it was Munir Ahmad Khan himself, whom I saw quoted about the challenge of developing an atomic bomb when domestic industry could hardly produce bicycle chains.

[highvoltageguypk]

Hi,
Let us focus on the simulation part only. I have tried to explain the procedure of simulating transient analysis in the following YouTube video.

/>
Your questions and recommendations will be appreciated. Thanks.

Regards,
highvoltageguypk.

[klugesmith]

Thank you. The video gives an idea of the user interface and problem approach.   

A bit of scripting and editing could help, but that applies to the vast majority of tutorials on youtube. It's confusing where you jump to a previous simulation model. Not clear if you are cutting and pasting things from that. It would also help to hear up front about the real-world problem to be solved.

Do you really have to individually select each of the 32 turns in coil, and define each one to be part of the "Lcoil" element?
Would the field results be much different if coil were represented as a narrow rectangle defined to have 32 turns, as you would do if (for example) the same bobbin were wound with 512 turns of 0.4 mm wire? 

The circles representing individual turns as round wire do contribute to pretty plots, and AFAIK are not available in FEMM or Quickfield.   Can you zoom in and show us the FEA mesh in coil area of Maxwell model?

I like how the finite element coil model is represented as a discrete inductor in electric circuit schematic, for interacting with external capacitor and switch etc.    In this case, would we not get the same answer using non-transient method?   1: static FEM field solver to get coil inductance and resistance, which are linear and invariant.   2: put that in plain old circuit simulator to get current waveform.   3: at any instant of time domain, take the circuit simulator value for current in Lcoil and plug that into static field solver?
Of course it is good to start with an easy case.   No nonlinear materials, eddy currents, or moving parts.

Can you say more about the simulation boundary rectangle? How you chose its size, and how does it represent an infinite world around the coil?

[Twospoons]

Because its modelled as 32 loops in an axisymmetric model, instead of a 32 turn helix in a 3d model, then yes you do have to define each turn as being part of one coil - ie setting all the elements in series.  And yes, it probably could have been defined as a rectangle with 32 turns, but you'd have to be careful with the material description so that the algorithm doesn't think its continuously conductive, which would redistribute the current incorrectly.
You can model individual turns like this in FEMM - I have done so - and FEMM will model things like skin and proximity effect.

This is not a particularly compelling example for transient simulation - a static solver could achieve the same result.  Models with non-linear elements, like iron saturation, would be better targets for transient sims.

[highvoltageguypk]

Thanks for the questions and let me answer it to the best of my knowledge:
Q1: Do you really have to individually select each of the 32 turns in coil, and define each one to be part of the "Lcoil" element?
Ans: yes I have to define them individually to be a part of "Lcoil".

Q2:Would the field results be much different if coil were represented as a narrow rectangle defined to have 32 turns, as you would do if (for example) the same bobbin were wound with 512 turns of 0.4 mm wire?
Ans:I will try it and share the results asap.

Q3:Can you zoom in and show us the FEA mesh in coil area of Maxwell model?
Ans:Image attached.

Q4:In this case, would we not get the same answer using non-transient method?
Ans:It depends upon the type of results you are interested. The basic simulation of this project used fixed current values (as compared with time varying current values) the field pattern at that specific current remains the same but I was interested in the field information w.r.t time, so, that's why I preferred co-simulation with actual circuit. And yes the suggested approach that you mentioned is highly recommended when we are targeting basic parameters like inductance, resistance, capacitance etc. but as I said we were interested in recording the time varying magnetic field, force and flux interaction.

Q5: Can you say more about the simulation boundary rectangle?
Ans: In Ansys Maxwell there is an option of Balloon boundary, it can expand itself to the required value. as it is shown in the attached images.

This is not a particularly compelling example for transient simulation

Please let me know which type of example you think will be the best. (I hope you have seen the complete video because I am not interested in the inductance of the coil but instead time varying parameters).

[highvoltageguypk]

Ans to Q2 is shown in the image. By simulating it as a lumped coil in the form of equivalent rectangle dimension is not much different as compared with individual circular coils. So, yes it saves some time (from 4 min to 1 min approx). But still I need to do some more comparative study on this topic. There are some cases (as mentioned by klugesmith) when the number of turns are much greater that individual turn cannot be treated, so, for that lumped inductor approach is the only possible solution( as the example in Ansys Maxwell documents also recommends this) but in the case when no. of turns are not much greater then this approach may not generate accurate results. 

[klugesmith]

Nice gifs.   What are the red and blue materials in the model?  They appear to have significant permeability. Or is it their conductivity and eddy currents that cause magnetic discontinuities at their boundaries?   You have not told us the real-world application.

[highvoltageguypk]

Q1:  What are the red and blue materials in the model?
Ans: Image attached.
Q2: Or is it their conductivity and eddy currents that cause magnetic discontinuities at their boundaries?
Ans: In this simulation eddy current effect was not included and can you elaborate the exact location of discontinuity??

This coil is designed for the application of Pulsed Electromagnetic Field and research paper is attached.
Thanks.

[klugesmith]

Thanks; that's interesting.  Before downloading and reading your paper...

I guess that the low amount of magnetic flux in steel shell is due to skin effect.
How is skin current handled in FEA when the skin depth is much smaller than any mesh triangle? Can we get a measure of pulse energy dissipated as heat in the steel?  What happens if steel is replaced by ferrite that does not conduct electricity?
No need to reply if the answers are in your paper.

[klugesmith]

Interesting paper there; thanks for sharing.   
You led me to learn some things about free FEMM simulator, where I'd been experimenting with AC mode in connection with induction heater coils.  With coil material defined as a copper wire gauge, I knew that frequency affects reported wire resistance and (slightly) flux linkage.
Adding an external ferromagnetic tube for the first time, I was surprised and pleased to see what looks like skin effect modeling.  This pic shows field line map where the outer sleeve is air, then ferrite, then steel, all at 92 kHz and never close to saturation.  View for steel at at 1 Hz is qualitatively like the middle picture here.  Must read the manual.


Be interesting to see how the Maxwell simulation, which approaches 10 T of field strength inside the coil, handles magnetic saturation in the steel shell.  The deeply penetrating field in your animated GIFs is what made me ask about the material.
I can now attest that FEMM's library model for "1018 steel" has saturation. There are currents and frequencies where we can see skin depth greater in middle of sleeve than near ends of sleeve.   But what about the harmonic frequencies from non-sinusoidal magnetization? Another reason for me to RTFM instead of playing with knobs to see what happens.

We can keep this thread close to Ansys Maxwell tool if you prefer.

[highvoltageguypk]

Thanks klugesmith for a comparative analysis and made me learn more about skin effects. Can you share the dimensions of the simulated coil so, that I could do the same simulation using Ansys Maxwell and see the effect of eddy currents. Following gif shows the effect of eddy current on all conductors (For pulsed magnetic design). Do you think that it is telling that steel is in saturation mode by enabling eddy effects feature ? And for time being I have no idea how to measure the heat dissipation in steel casing. I will simulate the effect of mesh size on the eddy effects. I cant see any difference between 1mm and 0.1mm mesh sizes or am I missing something. Thanks.

[Benbmw]

And for time being I have no idea how to measure the heat dissipation in steel casing. I will simulate the effect of mesh size on the eddy effects. I cant see any difference between 1mm and 0.1mm mesh sizes or am I missing something. Thanks.

You can use the fields calculator to integrate the “EMLoss” (or just “OhmicLoss” unless you have hysteresis losses enabled) over the steel object. If you are totally unfamiliar with the fields calculator let me know and I can grab some screenshots to demonstrate.

You could use the “Skin Depth Based” mesh operation on the external lines of the steel to ensure it captures the effects well. It has a built in tool that will calculate the skin depth at a given frequency in the material you are assigning it to.

-Ben

[highvoltageguypk]

Thanks Ben for introducing calculator, I will give a try and will share the results.

[klugesmith]

Thank you, Ben, for pitching in with expert hints.   
Don't know if FEMM offers skin depth based meshing. I found it easy to make the mesh finer on one side by adding points along material boundary.  Could also divide shape into layers with different mesh settings.  Looks like coarse mesh is plenty good for seeing the _external_ consequences of skin effect in relatively thick steel. (Colors represent |B| in snips below.)

Here are charts of eddy current density along straight contours in the coarse and less-coarse areas. Peak value is less than 10% of current density in copper coil.  This fixed-frequency AC model captures the 90° phase difference between coil current and eddy current at steel surface, and progressive phase lag with depth.  Apparent small increase in |B| and |J| at back surface seems to be real. It looks the same in a run with 0.1mm mesh in whole steel part, which didn't greatly increase the run time.

The model here has 2 mm thickness of coil and steel sleeve, and is running at 200 Hz for mesh comparison pictures.
At 2000 Hz (like hvpk's pulse simulation) the eddy current density at steel surface is comparable to J in coil, but we could still say more than 90% of capacitor discharge energy will heat the coil.
At 20 kHz the |Je| in steel at surface approaches 10x coil current density, as we have moved into induction heating territory.  Between 100x and 1000x power density of Joule heating, even after coil conductance is derated for skin effect and proximity effect. (Those are factored in by FEMM when the material in coil rectangle is declared to be a standard copper wire gauge.)

My cylinder radii are 16 to 18 and 22 to 24 mm, with lengths 32 and 44 mm, and 20 A in 14 turn coil. Might it be more productive for me to set up hvpk's model in FEMM, so we could both compare results with COMSOL simulation reported in the electro-magnetoporation paper?

[Twospoons]

With FEMM you can override the mesh generator and specify mesh size for lines, arcs and regions - select the item of interest and open the properties box. Uncheck "choose mesh automatically" and enter a value for the local element size.

Ansys used to do automatic mesh refinement based on field gradients, rerunning the analysis and refining the mesh until the element to element delta was under some predefine value. The idea was to minimize the number of elements by having large meshes in fairly homogenous areas, and fine meshes only where the field was changing rapidly, greatly cutting down on processing time for large problems.  This is particularly important for 3d cartesian problems where the number of elements can get very large very quickly.