Discussion Forum

1. I encourage you to post your model to the forum to be examined by others.
2. How different are your computed values for the inductance?

1. I encourage you to post your model to the forum to be examined by others.
2. How different are your computed values for the inductance?

1. Sorry Robert, I cannot upload the model for confidentiality reasons, but I can tell you that my model is a litz coil
2. In the first case the inductance is 32.9uh, in the second case the inductance is 0.135uh, and in the third case it is 8.17uh Anyway, thanks for your suggestion

1. Those are very large differences, so at least two of the computations are totally wrong. Ideally, you should be able to eliminate two (at least) based on analytic rough order of magnitude (aka, "back-of-the-envelope") estimates.
2. I haven't done inductance calculations recently, but I will offer some comments, since others don't seem to be weighing in (yet). (Note: It may be that Comsol has added some features that make this easier than it used to be, so my comments here are based on older knowledge.) Computation of inductance by means of L = (Flux thru the loop)/(Current in the loop) can always be a bit problematic with a spatially distributed (non-filamentary) current, because it can be a bit tricky (though not impossible) to define the loop paths and flux regions. If your computational volume is large enough (and it needs to be!) so that the magnetic field is small at any external boundary surface, i.e., the volume contains essentially all of the surrounding magnetic flux, then I would trust the energy based method most, with a well-defined dc current applied. I would make sure that I had a sufficiently fine mesh in any/all areas where the magnetic induction is strong, and especially anywhere it has a strong gradient. I would be especially careful in meshing about/around thin current-carrying paths. If I wanted to be sure, and as a sanity check, I might re-run the model with other meshes and also varying the discretization order (linear, quadratic, etc.) of the elements. Anyway, then I would employ the rule that 0.5xLxI^2 = total stored magnetic energy. I'm assuming you know the exact total "circuit" current, I. Find the total stored magnetic energy by integrating the magnetic energy density over the whole volume. Then solve for L. If the result isn't strongly dependent on mesh density or element order, and if it is also reasonably close to your back of the envelope estimate (which you really, really, should always do!), then it is probably correct!
3. Perhaps this is needless to say, but all the currents in your model should ideally be closed loops inside your computational space, and not near the boundaries of your computational space. (Otherwise, you are going to have to be very careful in regard to establishing symmetry and identifying various factors of 2.)

others don't seem to be weighing in (yet).

1. I don't know what a Leeds coil is, and neither does the web.
2. In the past I have calculated inductance both ways and did not see significant disagreement.
3. If you can't post your model, I suggest constructing a model using simple (but different) coil geometries and solve that one. Then if the disagreement persists we would have something to look at.

Do you mean "Litz" coil?

I don't know what a Leeds coil is, and neither does the web.

Perhaps something like one of these "Leeds & Northrup" antiques? https://www.uvm.edu/~dahammon/museum/l&nselfinductance.html

Do you mean "Litz" coil?

yes sir, it is a "litz" coil I am Sorry for everyones confusion caused by my spelling errors

1. Those are very large differences, so at least two of the computations are totally wrong. Ideally, you should be able to eliminate two (at least) based on analytic rough order of magnitude (aka, "back-of-the-envelope") estimates.
2. I haven't done inductance calculations recently, but I will offer some comments, since others don't seem to be weighing in (yet). (Note: It may be that Comsol has added some features that make this easier than it used to be, so my comments here are based on older knowledge.) Computation of inductance by means of L = (Flux thru the loop)/(Current in the loop) can always be a bit problematic with a spatially distributed (non-filamentary) current, because it can be a bit tricky (though not impossible) to define the loop paths and flux regions. If your computational volume is large enough (and it needs to be!) so that the magnetic field is small at any external boundary surface, i.e., the volume contains essentially all of the surrounding magnetic flux, then I would trust the energy based method most, with a well-defined dc current applied. I would make sure that I had a sufficiently fine mesh in any/all areas where the magnetic induction is strong, and especially anywhere it has a strong gradient. I would be especially careful in meshing about/around thin current-carrying paths. If I wanted to be sure, and as a sanity check, I might re-run the model with other meshes and also varying the discretization order (linear, quadratic, etc.) of the elements. Anyway, then I would employ the rule that 0.5xLxI^2 = total stored magnetic energy. I'm assuming you know the exact total "circuit" current, I. Find the total stored magnetic energy by integrating the magnetic energy density over the whole volume. Then solve for L. If the result isn't strongly dependent on mesh density or element order, and if it is also reasonably close to your back of the envelope estimate (which you really, really, should always do!), then it is probably correct!
3. Perhaps this is needless to say, but all the currents in your model should ideally be closed loops inside your computational space, and not near the boundaries of your computational space. (Otherwise, you are going to have to be very careful in regard to establishing symmetry and identifying various factors of 2.)

yes， we made a physical model and the measured value is 38.9uh and very close to the fisrt case value, but I want to know why the other two values are so different.Your suggestions gave me some ideas, thank you very much！

others don't seem to be weighing in (yet).

1. I don't know what a Leeds coil is, and neither does the web.
2. In the past I have calculated inductance both ways and did not see significant disagreement.
3. If you can't post your model, I suggest constructing a model using simple (but different) coil geometries and solve that one. Then if the disagreement persists we would have something to look at.

sorry sir, it`s a litz coil . yes i used a solid coil of equal cross-sectional area instead of simulation. The calculated value is 32.68uh very close to 32.9uh, but I am not sure whether a solid coil of equal cross-sectional area can be used instead of the Litz coil.Because the copper wire with a diameter of 2.5mm has a skin effect, and the litz wire is not obvious, it is not clear whether the mutual inductance or proximity effect of each wire will affect the inductance of the entire coil.