MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de

Eletromagnetismo

Progressive Source and Reaction Fields

for Magnetodynamic Model Refinement
via a Finite Element Subproblem Method
Patrick Dular1,2, Mauricio V. Ferreira da Luz3, Patrick Kuo-Peng3 and Laurent Krähenbühl4
1
University of Liege, Dept. of Electrical Engineering and Computer Science, ACE, Belgium 2 F.R.S.-FNRS, Belgium
3
Universidade Federal de Santa Catarina, GRUCAD, Brazil
4
Université de Lyon, Ampère (CNRS UMR5005), École Centrale de Lyon, France

Abstract—Magnetodynamic models are split into a sequence of levels of precision, e.g. considering the magnetic regions via
progressive finite element subproblems. The source fields
generated by the active conductors alone are calculated at first
perfect magnetic material BCs, or the conducting/magnetic
via either finite elements or the Biot-Savart law. The associated regions via perfect conductor or impedance BCs [1]-[2], thus
reaction fields for each added magnetic and/or conducting with the source fields acting as surface sources (SSs).
region, and in return for the source regions themselves when Avoiding to mesh their interior allows to lighten the
massive, are then calculated with finite element models, possibly
with initial perfect magnetic, conductor and/or impedance computational efforts, which is interesting for the preliminary
boundary conditions to be further corrected. The resulting stage of a design. Perfect conductor BCs are suitable for high
subproblem method allows efficient solving of parameterized conductivities or frequencies, i.e. for low skin depths [2]. For
analyses thanks to a proper mesh for each subproblem and the
reuse of previous solutions to be locally corrected. Accuracy larger skin depths, impedance BCs (IBCs) lead to a better
improvements are obtained for local fields and global quantities, accuracy but, as they are generally based on analytical
i.e. inductances, resistances, Joule losses and forces. solutions of ideal problems, they are only valid in practice far
Keywords—Eddy currents; finite element method; model from geometrical discontinuities, e.g., edges and corners. An
refinement; subproblem method. additional SP is then of interest to correct these surface models
with approximate BCs to volume models [6]. Sequences of
I. INTRODUCTION such SP solutions and/or corrections are developed for the
Instead of solving a complete magnetodynamic problem, magnetic vector potential FE magnetodynamic formulation.
including all conducting and magnetic regions, it is here They are illustrated on application examples.
proposed to perform successive finite element (FE)
II. COUPLED SUBPROBLEMS
calculations via a subproblem (SP) method (SPM) [1]-[6],
mainly by separating the regions, with the advantage of using A. Sequence of Subproblems
a different mesh at each step, or no mesh when the Biot-Savart
To allow a progression from simple to more elaborate
law is used. Source and reaction fields are considered but, at
models, a complete problem is split into a series of SPs that
the difference with the common method that adds these fields
define a sequence of changes, with the complete or total
in the whole domain to define the total field, the source fields
solution given by the sum of the SP solutions [3]-[5]. Each SP
are here to be defined only in the added regions [3]-[5]. Such a
is defined in its particular domain, generally distinct from the
support reduction is of importance for efficient calculations,
complete one and usually overlapping those of the other SPs.
especially for source fields calculated via the Biot-Savart law.
At the discrete level, this aims to decrease the problem
When acting as volume sources (VSs) in each added region, complexity and to allow distinct meshes with suitable
the source fields can be even initially reduced to its boundary, refinements and possible domain overlapping between SPs.
which is an important and useful aspect developed here.
Instead of volume projections of the source fields in the mesh B. Canonical Form of Magnetodynamic Subproblems
of the added region the source fields are rather calculated A canonical magnetodynamic SP p, to be solved at step p of
there, as a first alternative, via a FE problem with their the SPM, is defined in a domain Ωp, with boundary
boundary values as boundary conditions (BCs). Another ∂Ωp = Γp = Γh,p ∪ Γb,p. The eddy current conducting part of Ωp
general alternative aims at avoiding any source field volume is denoted Ωc,p and the non-conducting one Ωc,pC, with
projection or calculation thanks to interface conditions (ICs). Ωp = Ωc,p ∪ Ωc,pC. Massive conductors belong to Ωc,p, whereas
Intermediary SPs can tackle the added regions at various stranded conductors belong to Ωs,p ⊂ Ωc,pC. The equations,
material relations and BCs of SP p are

1
 MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de
Eletromagnetismo

curl hp = jp , div bp = 0 , curl ep = – ∂t bp , (1a-b-c) and for the test function a' (at the discrete level, this space is
hp = µp–1 bp + hs,p , jp = σp ep + js,p , (2a-b) defined by edge FEs; the gauge is based on the tree-co-tree
n × hp|Γh,p = jf,p , n ⋅ bp|Γb,p = ff,p , n × ep|Γe,p ⊂ Γb,p = kf,p , (3a-b-c) technique); ( , )Ω and < , >Γ denote a volume integral in
Ω and a surface integral on Γ, respectively, of the product of
where hp is the magnetic field, bp is the magnetic flux density, their field arguments. An SP p with only sources js,p in Ωs,p
ep is the electric field, jp is the electric current density, µp is has a direct solution given by the Biot-Savart formula, with no
the magnetic permeability, σp is the electric conductivity, n is need of FE calculation.
the unit normal on Γp exterior to Ωp and jf,p, ff,p and kf,p are D. Addition of Material Regions with Model Refinements
some given surface fields, defining SSs. Note that (1c) is only
defined in Ωc,p (as well as ep), whereas it is reduced to (1b) in Progressively adding some material regions, a wide variety
Ωc,pC. of SPs can be defined to allow various source and reaction
fields, together with model refinements of the added regions.
Fields hs,p and js,p in (2a-b) are VSs. They can classically be Each SP p is defined as a change or correction of a previous
remnant fields in magnets or fixed current densities in (or several) SP(s) q, (i.e., for all SPs q prior to SP p),
conductors. With the SPM, hs,p is also used for expressing without involving the already considered sources (i.e., active
changes of permeability and js,p for changes of conductivity conductors, previous VSs and SSs). It is constrained via VSs
[3], [4]. For changes from µq and σq for previous SP q to µp and SSs defined from parts of the solutions of the SP(s) q, as
and σp for SP p in some regions, the associated VSs hs,p and detailed hereafter for practical models.
js,p, nonzero only in these regions, are
A change with a significant effect on the previously solved
hs,p = (µp–1 – µq–1) bq , js,p = (σp – σq) eq . (4a-b) SPs has to be further considered as a source for these, which
thus requires iterative corrections. Also, a nonlinear SP p
These correctly define the material relations for the total requires classical nonlinear iterations.
fields, i.e., hq + hp = µp–1 (bq + bp) and jq + jp = σp (eq + ep).
E. Required Sources and their Discretization via Projections
Regarding BCs (3a-b-c), some paired portions of Γp can
define double layers, with the thin region in between exterior Each SP p requires VSs and/or SSs in some regions
to Ωp [1]-[6]. In particular, these will be associated with the ωs,p ⊂ Ωp evaluated from previous SPs q, . These
boundary of regions initially considered, in previous SP q, via sources, coming from previous meshes or Biot-Savart
simplified BCs. They are denoted γp+ and γp– and are evaluations of SPs q, have to be properly discretized in the
geometrically defined as a single surface γp with ICs, fixing mesh of SP p to assure the conformity of the sequenced FE
the discontinuities or IC-SSs ([ ⋅ ]γp = ⋅ |γp+ – ⋅ |γp–), i.e., weak formulations. They are obtained by means of Galerkin
projections [10] of the primary field aq between the meshes,
[n × hp]γp= [jf,p]γp, [n ⋅ bp]γp= [ff,p]γp, [n × ep]γp= [kf,p]γp. (5a-b-c) i.e.

C. Canonical Magnetic Vector Potential Weak Formulation aq p−mesh a ωs = aq a ωs ∀a ∈Fp ω s p

(9)
p p

The magnetic vector potential ap and the electric scalar

potential vp are defined via where Fp1(ωs,p) is a gauged curl-conform function space for
the projected source aq,p-mesh (the projection of aq on mesh p)
bp = curl ap , ep = – ∂t ap – grad vp = – ∂t ap – up, (6a-b) and the test function a'.
n × ap|Γb,p = af,p , [n × ap]γp = [af,p]γp, (7a-b) SSs associated with n × hq|Γh,p,i, with (3a) or (5a), are
involved in surface integral terms in (8) for SP p. Being of
with given surface potential af,p. The ap weak formulation of weak nature, they are to be weakly expressed from (8) written
the magnetodynamic problem is then obtained from the weak for each prior SP q, , i.e. from volume integrals
form of the Ampère equation, i.e. [3], generally limited to
1
( curl ,curl ') + ( , ,curl ')
< n× hq ,a ' >Γ = −(µ −1
q curl aq ,curl a ')Ω p =Ωq . (10)
h, p,i

, (8) At the discrete level, the volume integral in (10) is limited to

one single layer of FEs touching Γh,p,i (thus on one side of
where Fp1(Ωp) is a curl-conform function space defined on Γh,p,i), because it involves only the associated trace n × a'|Γh,p,i.
Ωp, gauged in Ωc,pC, and containing the basis functions for ap The source aq, initially in the mesh of SP q, has to be

2
 MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de
Eletromagnetismo

projected in the mesh of SP p via (9) only in the FE layer as Gauss point in (9) is avoided and replaced by a physical
the projection region ωs,p, which thus decreases the problem solution of similar computational weight. From
computational effort of the projection process. aq,p-mesh |ωs,p, both bq,p-mesh |ωs,p and eq,p-mesh |ωs,p can be
determined for VSs (4a-b), thus with no need to separately
SSs associated with n ⋅ bq|Γb,p,i, with (3b-c), (5b-c) or (7a-b),
evaluate (11a-b).
are of strong nature. They are to be directly defined in
function space Fp1(Ωp), with source aq to be projected only on B. Perfect Magnetic Material BC (region-PMBC)
Γb,p,i, or at most in a FE layer touching Γb,p,i.
An SP p ≡ region-PMBC is defined in a new domain Ωp by
III. PROGRESSIVE MODELS AS SPS considering some added magnetic regions Ωr,p,i (i is the region
index; Ωr,p,i ⊂ Ωc,pC or Ωr,p,i ⊂ Ωc,p) as being perfect, with
A. Active Conductor with FEs (COND-FE) or Biot-Savart infinite permeability (µp →∞ ) [1]. Its solution can serve as a
(COND-BS) Formula reference solution for any finite µp further considered. The
Considering each active conductor Ωs,p or Ωc,p, fed by interior of Ωr,p,i, with zero magnetic field hp inside, is
external circuits, without any other region in a domain Ωp, extracted from the studied domain Ωp and treated in (8) via
with some possible symmetries that do not exist anymore in BC (3a) fixing a zero trace of total magnetic field
the complete problem, offers advantages in mesh operations, h = hp + Σq < p hq on boundaries Γr,p,i = ∂Ωr,p,i, thus coupling
especially in parameterized analyzes on positions and both the unknown fields and the fields from previous SPs q,
dimensions. The field it generates can serve as a source field , acting as weak SSs via (10), i.e.
for further SPs. It can be calculated via an FE SP p ≡ COND-
FE in the mesh of Ωp [3] (Fig. 2). It can also be calculated via n × hp|Γr,p,i= – Σ q < p n × h q|Γ r,p,i
. (14)
the Biot-Savart formula [9] with a given js,p (even initially
applied to a simplified wire geometry of Ωs,p), being a direct A non-zero trace n ⋅ bp|Γr,p,i will be part of the solution of (8),
solution of this SP p ≡ COND-BS. BS source fields bp and ap, thus giving a discontinuity [n ⋅ bp]Γr,p,i = n ⋅ bp|Γr,p,i to be further
with ep given via (6b) (with vp = 0), are then defined via considered as a strong SS for a correction SP.
integrals
C. Perfect Conductor BC (region-PCBC)
µ0 An SP p ≡ region-PCBC is defined in Ωp by considering
bp ( x P ) = ∫ ( js, p × r) / r 3 dxQ , (11a)
4π Ωs, p some added conductors Ωc,p,i (i is the conductor index) as
µ0 being perfect, with σp →∞ [1]-[2] (Fig. 1). Its solution,
a p ( xP ) =
4π ∫ Ωs, p
js, p / r dxQ , (11b) independent of the conductivity, can serve as a reference
solution for any finite conductivity further considered. This
results in a zero skin depth and surface currents. The interior
with x P ∈Ω p the calculation point position vector, xQ ∈Ωs, p
of Ωc,p,i, with zero fields inside, is extracted from the studied
the integration point position vector and r= x P − xQ . The
domain Ωp and treated in (8) via BCs (3b) fixing a zero trace
source fields are to be calculated afterward only in some
of total magnetic flux density b = bp + Σq < p bq on boundaries
particular regions, for a change to a volume conductor or when
Γc,p,i = ∂Ωc,p,i. This thus couples both the unknown bp and the
adding other regions.
fields from previous SPs q, , acting as SSs [1]-[2],
When used as a VS in a region ωs,p ⊂ Ωp for an SP p, a [5]-[6], i.e.
Biot-Savart source, aq,BS from (11b), gains at being projected
only on the boundary ∂ωs,p of ωs,p via n ⋅ bp|Γc,p,i= – Σ q < p n ⋅ b q|Γ , (15a)
c,p,i

(aq, p−mesh ,a ')∂ω =(aq,BS ,a ')∂ω ,∀a '∈Fp1(∂ω s, p ). (12) or, in terms of the primal unknown ap, with the strong BC
s, p s, p

Then, its surface projection aq,p-mesh |∂ωs,p defines a BC for a n × ap|Γc,p,i= n × grad wp|Γc,p,i – Σ q < p n × a q|Γ c,p,i
, (15b)
physical problem of form (1)-(3) in ωs,p , with the weak form
(8) here reduced to with wp an unknown surface scalar potential; explicitly
defining wp instead of ap on Γc,p,i fixes (15a). A non-zero
1
(µ −1
q curl aq, p−mesh ,curl a ')ω = 0 , ∀a ' ∈ Fp (ω s, p ) , (13) trace n × hp|Γc,p,i will be part of the solution of (8), thus giving
p
a discontinuity [n × hp]Γc,p,i = n × hp|Γc,p,i to be further
considered as a weak SS for a correction SP.
to determine the volume extension aq,p-mesh |ωs,p. In this way,
the heavy Biot-Savart evaluation that would be needed at each

3
 MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de
Eletromagnetismo

D. Impedance BC (region-IBC)
js,p = (σp – σp–1) Σ q < p eq , (19b)
In an SP p ≡ region-IBC, some conductors can also be
extracted from Ωp by using IBCs relating the tangential traces where µp–1 and σp–1 are the lastly considered material
of total magnetic and electric fields on their boundaries Γc,p,i characteristics before their changes to these of the actual
(actually the outer boundary Γc,p,i+ of Ωc,p,i) [5]-[6] (Figs. 1 conductor.
and 3), thus also coupling both unknown and previous (as SSs)
solutions q, via the BC

n × hp|Γc,p,i= Zc,p,i–1 n × (n × ep)|Γc,p,i

+ Σq < p Zc,p,i–1 n × (n × eq)|Γ c,p,i

– Σ q < p n × h q|Γ c,p,i
, (16)

with Zc,p,i the surface impedance for conductor Ωc,p,i, i.e., Fig. 2. Field lines for an inductor alone COND-FE (b1, left) and for the added
core CORE-VOL-VS (b2, µr,core = 100) (right); distinct meshes are used for
problems 1 and 2 [3].
Zc,p,i = (σp δp)–1 (1 + j) , with δ p = 2 / (ωσ p µ p ) , (17a-b)
F. Volume Region from SSs (region-VOL-SS)
with ω the angular frequency (ω = 2π f, with f the frequency) In case previous solutions q aimed at zeroing the fields in
and j the imaginary unit (∂t ≡ j ω in the frequency domain); δp Ωc,p,i, e.g., with Ωc,p,i previously considered via PMBC,
is the skin depth. BC (16) is then to be expressed in (8), with PCBC or IBC (Figs. 1, 3, 4 and 5), the VSs in (19a-b) are zero.
Γc,p,i ⊂ Γh,p\γp, in terms of the trace of the primal unknown ap In such cases, all the fields being carried in the double layer of
with its boundary, trace discontinuities of both hq and eq (aq) occur.
Their opposite values then define SSs for SP p in (5a), weakly
n × (n × ep) |Γc,p,i= ( n × (∂t ap+up) ) × n |Γc,p,i . (18) expressed via (10), and (5b) (and (7b)), strongly expressed in
function space Fp1(Ω).
The solution of (8) in this SP contains non-zero traces
n × hp|Γc,p,i and n × ap|Γc,p,i on Γc,p,i+. The traces on the inner
boundary Γc,p,i– being zero (the IBC model implying zero
inner volume fields), trace discontinuities through the
resulting double layer Γc,p,i thus occur. They can further
define weak and strong SSs for a correction SP.

Fig. 3. Field lines near a conductor corner for different frequencies

(f = 12.5 kHz (column 1); f = 3.125 kHz (column 2), f = 0.781 kHz (column
3)), from top to bottom: complete solutions, CORE-PCBC initial solution
(row 1, column 4), CORE-IBC correcting CORE-PCBC, CORE-VOL-SS
Fig. 1. Field lines near a conductor corner, from top to bottom: complete corrections showing field discontinuities at core interface (µr,core = 1,
solution, left: CORE-PCBC, CORE-IBC models, right: associated volume σcore = 106 S/m); same scale for direct comparisons [6].
corrections CORE-VOL-SS; same scale for all maps to point out the decrease
of the required correction [6].

E. Volume Region from VSs (region-VOL-VS)

A volume region Ωc,p,i ⊂ Ωc,p is considered in an SP
p ≡ region-VOL-VS (Fig. 2) via the VSs (4a-b) [3], [5]-[6]
using all previous solutions q, , i.e.
Fig. 4. Field lines near a conductive magnetic core, from left to right:
complete solution, CORE-PCBC initial solution, CORE-IBC correcting
hs,p = (µp–1 – µp–1–1) Σq < p bq , (19a) CORE-PCBC, CORE-VOL-SS correction (µr,core = 16, σcore = 106 S/m,
f = 0.781 kHz, δ = 4.5 mm); same scale for direct comparisons [6].

4
 MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de
Eletromagnetismo

SP p

SP q SP pa + SP pb
Ωq Ωp IN Ωvol,p Ωp IN Ωvol,p

approx. µp

Γvol,p
model σp

n hpa= – hq hq+hpa= 0
hq bpa= – bq bq+bpa= 0
bq OUT OUT ! No VS

hq hpa = 0 µp=µq No change

bq bpa = 0 σp=σq outside

a (COND-BS) a (CORE-PCBC) a (CORE-IBC) a (CORE-VOL-SS) Fig. 6. From an approximate model (SP q) to a fine volume (SP p) FE
representation of a conductor: SP p is split into SPs pa and pb, simultaneously
solved, SP pa removing the volume solution q inside the conductor and SP pb
considering the actual volume conductor properties (of all its subregions),
with no need of VSs for change of properties, but with SSs for unified SP p.

b (COND-BS b (COND-BS j (CORE-IBC) j (CORE-VOL-SS)

+CORE-PCBC) +PCBC+IBC)
Fig. 5. Field lines (a), magnetic flux density (b) and eddy current density (j) in
a 2-D rail induction heating system for progressive SPs (COND-BS, CORE-
PCBC, CORE-IBC, CORE-VOL-SS); f = 1 kHz, µr = 1, d = 16 mm [5].

Another mean to avoid VSs and to solely use SSs is via the
Y Y Y Y Y
Z X Z X Z X Z X Z X

(a) (b) (c) (d) (e)

general method described hereafter, which usually strongly
lightens the computational process. A previous model of Ωc,p,i Fig. 7. Field lines in the surrounding of an inductor (portion of a full
geometry, wire position shown): (a) Biot-Savart field from all wire inductors
could be a Biot-Savart filament model (e.g., wire conductor) COND-BS (b1, actually not calculated in the whole domain), (b-c) static
[11] or an homogenized model using equivalent material volume correction COND-VOL-SS (b2,sta) and total field, (d-e) dynamic
volume correction COND-VOL-SS (b2,dyn) and total field. The volume
properties (e.g., coil, foil winding, lamination stack). correction gives the total field in the volume inductor, is discontinuous
through the inductor boundary and quickly decreases outside [11].
Corrections can be done for the whole volume region or
some of its portions, e.g., a selection of conductors in a coil or Wire Biot-Savart field bq
30
Field b (µT)

in a lamination stack where significant edge effects can occur. 20

FE vol. corr. field bp
STATIC Total field
Subdomain Ωvol,p denotes the region to correct (Fig. 6) which 10
is now defined with its actual volume, made of both 0
-10
conducting and non-conducting materials (e.g., conductors and
-60 -50 Wire Biot-Savart
-40 -30
field bq -20 -10 0
insulations), that can be finely meshed, and its surrounding 30
Field b (µT)

PositionFE vol. corr.

x (mm) (wirefield bp
or inductor center at 0)
Ωvol,pC is kept homogenized, thus coarsely meshed. The 20
DYNAMIC Total field
10
correction solution is calculated via an SP p in 0
Ωp = Ωvol,p ∪ Ωvol,pC with adequate sources. The key is to -10
suppress the previous solution in Ωvol,p while keeping it -60 -50 -40 -30 -20 -10 0
unchanged outside (SP pa), simultaneously with the Position x (mm) (wire or inductor center at 0)
consideration of the actual µp and σp in Ωvol,p (SP pb). This is Fig. 8. Magnetic flux density versus distance from conductor (half-width in
[–10,0] mm): wire conductor Biot-Savart field COND-BS, FE volume
done by defining both tangential and normal trace correction COND-VOL-SS and total fields (static and dynamic problems)
discontinuities of the correction field through the boundary of [11].
Γvol,p = ∂Ωvol,p as SSs equal to the corresponding traces of the
previous solution q along Γvol,p (that can exist or not in SP q) ,
i.e.,

[n × hp]Γvol,p = n × hq|Γvol,p , [n ⋅ bp]Γvol,p = n ⋅ bq|Γvol,p . (20a-b)

Y Y Y Y
Z X Z X Z X Z X

This requires a mesh projection of solution q in a layer of (a) total sol. (b) BS SF (c) sol. w/BS SF (d) vol. correct.
FEs along the boundary of Ωvol,p. Each volume correction thus Fig. 9. Current source in a slot with air gap: field lines for (a) full model
gives the total field in Ωvol,p, is discontinuous through its solution, (b) BS SF with its projection limited to the core boundary (COND-
BS), (c) total solution with BS SF, (d) volume correction of coil and its
boundary and quickly decreases outside (Figs. 7, 8, 9, 10, 11), surrounding (COND-VOL-SS), pointing out the field trace discontinuities.
which justifies the use of a coarse mesh in the outer region.

5
 MOMAG 2014: 16º SBMO - Simpósio Brasileiro de Micro-ondas e Optoeletrônica e 11º CBMag - Congresso Brasileiro de
Eletromagnetismo

to the total solution, mainly as source and reaction fields. With

the SPM, the source fields are efficiently defined so as to act
in reduced supports for the reaction field calculation, which is
of particular importance for Biot-Savart models. A source
field proper to one region can be used as a source not only for
another region but also for an improved model of this source
(a) (b) (c) (d) region, which is of interest for model improvements, e.g.,
from Biot-Savart wire models to FE volume models of
Fig. 10. Lamination stack (half core, number of lam. = 10, lam. thickness
0.5 mm, insulation 20 µm, µr,lam = 2500, σlam = 5 106 S/m; with air gaps, for conductors, or from homogenized FE models to fine FE
longitudinal and transverse fluxes): homogenized solution (a) and corrected
solutions CORE-VOL-SS in 1 (b), 2 (c) and all laminations (d) (frequency models. Each reaction field can become a source field for its
1 kHz) (field lines; eddy current density in elevation). own region or other regions. Progressive corrections can be
104 done from static to dynamic models, thus for accurate skin and
Homogenized sol.
103 Corrected sol. proximity effects. The proposed approach allows to go one
Pf / P50Hz

102 step further than the classical method using Biot-Savart source
10
1 fields, offering the possibility to focus afterwards on the actual
100 volume conductor with FEs, with a local refined mesh, not
only in statics but also in dynamics. Accurate determination of
Pf relative correction (%)

30
(Pf,corrected - Pf,homog.) / Pf,corrected
25
20
inductances, resistances, Joule losses and forces can thus be
15 obtained in a large frequency range. The method, tested in 2-
10
5 D, is directly applicable in 3-D.
0
-5
102 103 104
Frequency f (Hz)
REFERENCES
Fig. 11. Joule losses in lamination stack versus frequency f for homogenized [1] P. Dular, R.V. Sabariego and L. Krähenbühl, “Subdomain Perturbation
and corrected solutions: ratio (in log scale) with losses at 50Hz (top), relative Finite-Element Method for Skin and Proximity Effects”, IEEE Trans.
correction strongly increasing with f (bottom), pointing out its importance. Magn., vol. 44, no. 6, pp. 738-741, 2008.
[2] P. Dular, R. V. Sabariego, L. Krähenbühl. “Subdomain perturbation
G. Inductance and resistance calculation finite element method for skin and proximity effects in inductors,”
COMPEL, Vol. 27, No. 1, pp. 72-84, 2008.
The self inductance of a BS conductor, and the possible [3] P. Dular, R. V. Sabariego, C. Geuzaine, M. V. Ferreira da Luz, P. Kuo-
mutual inductances with other BS conductors, can be Peng and L. Krähenbühl, “Finite Element Magnetic Models via a
calculated via double integral Neumann formulas [8]-[9]. The Coupling of Subproblems of Lower Dimensions,” IEEE Trans. Magn.,
vol. 46, no. 8, pp. 2827-2830, 2010.
resistance can be approximated as well (Fig. 11). After a [4] P. Dular, L. Krähenbühl, R.V. Sabariego, M. V. Ferreira da Luz, P. Kuo-
volume correction SP p, the corrected inductance is Peng, C. Geuzaine. “A finite element subproblem method for position
change conductor systems,” IEEE Trans. Magn., vol. 48, no. 2, pp. 403-
advantageously obtained with the solution ap only in Ωs,p and 406, 2012.
Ωc,p, i.e., via [5] P. Dular, V. Péron, L. Krähenbühl, C. Geuzaine. “Progressive Eddy
Current Modeling via a finite element subproblem method,”
Φs,p = (js,p , ap)Ωs,p , Φc,p = (jp , ap)Ωc,p , (21a-b) Proceedings of ISEF 2013 (International Symposium on
Electromagnetic Fields in Mechatronics, Electrical and Electronic
Engineering), paper OS5-279_1/6, 6 pp., 2013.
defining the total linkage magnetic fluxes Φs,p and Φc,p [4], [6] P. Dular, V. Péron, R. Perrussel, L. Krähenbühl, C. Geuzaine. “Perfect
thus as the new global value without any reference to the BS conductor and impedance boundary condition corrections via a finite
inductance approximation. This is a valuable key feature of element subproblem method,” IEEE Trans. Magn., vol. 50, no. 2, paper
7000504, 4 pp., 2014.
the proposed SPM. [7] Z. Badics, Y. Matsumoto, K. Aoki, F. Nakayasu, M. Uesaka, K. Miya,
“An effective 3-D finite element scheme for computing electromagnetic
An added region in an SP gives an inductance change that is field distorsions due to defects in eddy-current nondestructive
calculated by a volume integral limited to the added region evaluation,” IEEE Trans. Magn., Vol. 33, No. 2, pp. 1012-1020, 1997.
(via the reciprocity theorem [7]), which is another key [8] R. Dengler, “Self inductance of a wire loop as a curve integral,”
arXiv:1204.1486 [physics.class-ph], 14 pp., 2013.
advantage of the SPM. [9] C. Dumont de Chassart, M. Van Beneden, V. Kluyskens, B. Dehez,
“Fully analytical determination of inductances in windings with axial
IV. CONCLUSION and azimuthal wires”, Proceedings of ISEF 2013 (International
Symposium on Electromagnetic Fields in Mechatronics, Electrical and
The developed FE-SPM allows to split magnetodynamic Electronic Engineering), paper PS2-074_1/7, 7 pp., 2013.
[10] C. Geuzaine, B. Meys, F. Henrotte, P. Dular, W. Legros, “A Galerkin
problems, gathering active and passive conductors and projection method for mixed finite elements,” IEEE Trans. Magn., Vol.
magnetic regions, into SPs of lower complexity regarding 35, No. 3, pp. 1438-1441, 1999.
meshing operations and computational aspects, with reuse of [11] P. Dular, V. Péron, L. Krähenbühl, C. Geuzaine. “Subproblem Finite
Element Refinement of Inductors from Wire to Static and Dynamic
shared solutions. Each considered region, with its surrounding, Volume Models,” Proceedings of CEFC 2014, Annecy, May 2014.
is defined with its proper FE mesh, and gives its contributions