Abstract :
[en] Magnetic components are essential in power electronic systems, where their behavior under
high-frequency excitation plays a critical role in overall system performance. However, ac-
curately modeling these components remains a challenge due to complex geometries, inter-
actions between electromagnetic phenomena and frequency-dependent loss mechanisms.
This thesis addresses the research problem of developing a general and robust methodology
to derive accurate equivalent circuit models of magnetic components from physical prin-
ciples, enabling fast and reliable simulations in frequency and time domains. The primary
objective is to generalize the modeling of magnetic components through localized equiva-
lent circuits across different levels of discretization, from the turn level to the winding level.
The approach allows to identify circuit parameters representing both magnetodynamic and
electrostatic behaviors by means of numerical resolutions. The method also makes an in-
tensive use of ladder-type circuits in order to equip the model with frequency dependent
behavior and still comply with the strong constraint of localized constant circuits. Data used
in the study include detailed material characterizations (permeability, permittivity, losses)
of ferrite cores and dielectric materials, as well as impedance measurements of various in-
ductor and transformer prototypes studied under diverse configurations. The finite element
method (FEM) is employed to extract local field quantities, which are then transformed into
global circuit parameters: resistances, inductances, capacitances, and conductances. The
method is first validated on inductors built with simple geometries (essentially exploiting
rotational symmetry) and made of the materials characterized beforehand. It is then applied
to multi-winding components with more complex geometries and design, such as industrial
transformers. Key findings demonstrate that the proposed method and equivalent circuits
can accurately predict the first few resonances. The tunable levels of discretization (by turn,
by section, by winding) offer great flexibility to the designer, providing a trade-off between
computational efficiency and model accuracy.
