3241

Otherwise the geometric description and discretizetAbstract : tion of a three dimensional domain is always a very dif3~ modeling.becomes necessary for analysis of speficult task which hinders the development of the finite cia1 electromagnetic systems and also for testing the consistency of two dimensional models. In this paper, element method. We shall report in the paper various SOthe variational formulations of magnetostatic scalar and lutions to the difficult problem of data organization vector potenti'aLs are revirwed.An original energy func- and some methods for making the results visual. tional fornon linear anisotropic vector potential with a proof solution VARIATIONAL uniquenessis of ofproposed. FOFMJLATIONS The biggest problems remain data input and results reduction; these two problems are analysed and Original Maxwell's equations are written in terms solutions are given with some illustrated examples-. of two vectors: the flux densityB and the magnetic field H. The concept of potential (vector or scalar) alINTRODUCTION lows to reducephysical number functions. the unknown of Scalar magnetic potential Current trends in the design of electrical appara- The scalar magnetic potential is sf great interest tus are oriented towards an increase of specific power, for the solution of problems of magnetostatics because Of in order to optimise the machines in size, weight and its computation is always cheaper than the solution re1iability.A good design of an electrical device is the vector potential consisting o f three Components. [I have based on an accurate prediction of its performance which Zienkiewicz, Lynen and7 1Owen used itfor the requires in turn the knowledge of the field distri- analysis of the field of a transformer, Rafinejad, COUbution and its influence on : magnetic forces which lomb and Meunier 1181 applied curvilinear finite ele, act on the various parts of the machine eddy current ments to a permanent magnet Eield device (Fig. 1 ) . losses, iron losses etc... Classical methods, analytic or analog, are not satisfactory for the analysis of electrical machines composed of magnetic circuitsof complex geometrical shapes. The improvement of the performance of digital computers allowed the development of accurate numerical methods for magnetic field computation in electrical devices. During the last decade great amount of work was done a on two dimensional modeling of magnetic saturation effects in ferromagnetic materials 111-[I21 and eddy current distribution in conductors 1381-1431. These methods generally inexpensive, useful and accurate are satisfactory for most field computation problems. However the rush to performance and reliability improvements involves the need of computation methods able to take into account the three dimensional'aspects of magnetic fields and the geometry of electrical devices. This last development becomes pos1 : Permanent magnet device: a plot of sible, owing to the tremendous burst ,in the technology Fig. computers of peripherical equipotential and graphicThe devices. lines. problem of the three dimensional computation of 'the magnetic field has been attacked by the finite difference method(Sarma and a1 [131, Wolf 1141) ,the boundary integral method (Armstrong et al, 151, Ancelle and In this method the excitation field Hc created by Sabonnadisre 1161) and by the finite element method primary current and the magnetizing field due to the Hm (Zienkiewicz et a1 1171, Rafinejad et a1 [8, 1! magnetic material are computed separately. The source of Demerdash et a1 [I91 and recently by Chari, Silvester, field Hc is determined by application Biot and Savart Csendes, Konrad Palmo and 1201 ). law the from current density J. The field Hm, which is Here, we shall analyse some variational formula- the difference between the total field H and the primary gradient of a tions adapted to the computation of three dimensional field Hc is irrotational and thus is a magnetic fields giving special care to the variational reduced scalar potential 4 which must fulfill the parformulation of the magnetic vector potential [191, 1201. tial differential equation : We shall proposean energy functional chosen of the divergence the (ugrad4) of div = div (wHc) + div (Br) (1) to insure nullity the magnetic vector potential according to the Coulomb gaube i n which u is the permeability tensor and Br the resige. This functional will proved to work as well for problems of linear and non linear isotropic or anisotro- dual flux density inside permanent magnets. The materials. pic functional associated problem with (1) i s : H W' = BdH] dR t B'n@dT (2)

I [I
R

Jr

where H = Hc and B' is the Manuscript received April 1, 1981. d'Electro- sed onna part of Jean-Louis Coulomb is with the Laboratoire technique (E.R.A. 534, CNRS), ENSIEG, B.P. 46, 38402 clhditions the hes. SAINT MARTIN D'HERES, FRANCE.
0018-9464/81/1100-3241$00.75 0 1981 IEEE

grad C B = D H = U(Hc - grad 4) + Br p , () normal componentof the flux density impothe boundary. Under natural boundary surface integral over the boundary vanis-

3242

Linkage between reduced and total potentials cond term solves(6) and the minimization of their sum The previous statement of the problem pointsout provides the only solution with zero divergence (Proof some numerical difficulties in saturable ferromagnetic given in appendix). The quantity V is chosen equal to v in linear parts of the device whqre the vector sum of Hc and Hm is a field of weak streng Simkin and Trowbridge 1221 material, at v (B= 0 ) for saturated parts and 1 1 v I I tk oroposed the use of the reduced scalar potential 0 inside for anisotropic media.A choice of a constant value the conductors and the total scalar potential $, such, = 1 is also possible. For a linear problem with an that H =-Grad , everywhere else. Both potentials are ) I UO linked on the interface between the conductor region homogeneous reluctivity and a choice = v formulation v v and the non conductor region in order to insure the(8) is equivalent to the functional presented in 1201. of continuity of the normal component the flux density andthetangentialcomponentofthemagneticfield.THE 3D FINITEELEMENT METHOD Inside each region the unknown potential is determined byextremizingtheco-energyfunctional (2) accor-Inthefiniteelementmethodtheunknownfunction, ding to the usual finite element method. for instance the vector potential, is represented by an The magnetic vector potential expression : The magnetic vector potential is very popular for N two dimensional or ax?symmetricproblems because it a (X,Y,~)A i (9) A(x,y,z) = owns only omcomponent orthogonal to the planef analyo are weighted by sis. It has been used in almost all magnetic field pro- where the approximation functions blems in which the current density exists. For three di- the nodal values of the function seeke Expansion (9) d. mensional problems, all three components must be computed is put into the functional and the best solution in together and thus the vector potential becomes less at- of nodal values Ai and trial functions ai is that which tractive than magnetic scalar potential. : insures the conditions In spite of this drawback important amount of an recent is workmade on the of use 3D vector potential aF i NN (10) - = 0 for = 1 [191, 1201, [ X I . axi The magnetostatic vector equation governing the magnetic field is : total which numberNNin the is o f nodes. Equation (10) is a linear system when the problem is linear ornon linear system when saturation effect is Curl (vcurlA) = 5 - Curl fi-, (10) provides the nodal taken into account. Solution of Where A is the magnetic vector potential a tensor function of the values Ai and thus the approximate solution of A. The the reluctivity domain R is discretized into small domains named finite flux density = Curl A ) . elements. Shape functions u;(x,y,z) are piecewise contiJ is the current density rcthe coercitive magnetic field in permanent manuous over each finite element at the interfaces and between elements. Various kinds of finite elements and gnets. Natural boundary conditioniare in termsof the magnetic associated shape functions are extensions of 2D usual elements. vector potential : a) Normal component of flux density First order tetrahedron extension is four The to the node tetrahedron 3D Bn = 0 = > A t = 0 (4) problems of the popular 2D triangular element. Inside the tetrahedron the gradient of the approximated function b) Tangential component of magnetic field is constant,that allows the analytical evaluation, even of the contribution of the elefor non linear material, Rt = o => Kn = o (5) merit to the minimization of the functional.

....

Partial ( 4 ) and tential placing

differential equation (3) and boundary conditions (5) are not sufficient to insure the vector pounicity. This last requirement is obtained by : the condition usuallyknown as Coulombs gauge div A
=

/:

(6)

3!
,

The energy functional associated with this problem 5) (equations ( 3 ) , ( 4 ) and ( ) is direct extensionof the bidimensional formulation [I91 :
F(A)=j(

tetrahzdrms

triasgular prisms

JR

B dB - 5 T

] dQ.4

(A x g ) . iidr
(7)

Natural boundary conditions( 4 ) , (5) cancel the Surface integral. 2 : Typical 3D finite elements Unfortunately, infinity functions Fig. an of exit which minimize (7) (every function A = A+grad @ is also top : elements order of one a solution). This fact impliesbad a convergence the of bottom : isoparametric elements order of numerical for algorithms iterative thesolution used Of two. non linear problems.UnicitY of vector potential for vector potential will be insured if we modify ( 7 ) in or- Other first order elements t ) Eight-node triangular 6 account six-node intobrick element$or take der to prism elements are also of current use. The discretiza2 [ dB - 7 dR + V (div dR more is (8) F visualized into easilyelements than these tion the tetrahedric one, however the shape function of these elements becomes more complex [201 and for non linear material numerical integration is ComP~lsorY. In equation ( ) v is a positive quantity which is phy8 sically equivalent to a reluctivity. The minimization of An elegant solution may consist to process the visual checking on prismatic and brickelements and then (8) solves ( 3 ) , ( 4 ) and ( 5 ) , the sethe first term of

(x)=

lQ joB

x]

LT J

x)

3243

in to break down these elements into tetrahedra for compu- The topological description is used in [251 2D in tation [19] and thus to gain the advantages of both fa- problems and has been implemented [321 3D ; it consist mainly in a description througha cubic grid of milies of elements. the topology of the object followed aby distorsion of the grid to fit exactly the geometric definition of the Curvilinear isoparametric elements device. Improvement of the accuracy of the models may come By this method it is possible make at once finite to from eithera finer discretization of the domain with a considerable increaseof the number of nodes or from element discretization of the object and its surrounding. a better approximation by high order finite elements.. An other advantage of this method is the possibility to get Curvilinear high order elements bring also the ad- a matrix which is nerfectly band structured. In spitc is vantage of a better fit of the geometry of the device of its advantages this method somewhat difficult for Curvilinear finite elements the user who is not accustomed to imagine the topolo[61 , [181 ,[34],[35], 1361. gy need always numerical integration either for linear or of his object prior to its discretization. a Fig. 4 shows the description of permanent magnet non linear problems. a On Fig. 3 the reader can see test example of cur-micromotor and theprogressof the discretization aof vilinear 27-node brick type finite elementson a 2D pro- two-pole slice of the motor [301. blem extended into 3D. Fig.4 : Discretization of a permanent magnet micromotor: a) perspective of permaneut magnet and teeth, b) initial mesh, c) final mesh.

(a) Discretization

(b) Scalar equipotential plotting

Fig. 3 : A 2D Dermanentmamet Droblem treated asa 3D scalar potential problem. Notice the important distorIn tion of the central curvilinear hexahedric element. spite of this fact the error potential value is 2 s ~ on 1 than 1 %

Functional minimization The solution magnetic problem of the field (2) is obtained by minimization of the functional or (8) with respect to each nodal value of the potential. A necessary condition for this minimization is the cancellation of all first derivatives the functional. of When the problem is linear, the functional is quadratic and thus the minimum is the solution 0f.a linear matrix equation. If the problemnon 1inear.a is direct solution is not possible and the minimum must be reached through iterative process. The Newton an Raphson widely used for 2D problems is also very [61 efficient for 3D applications 1261. Matrix equations coming from linear systems or Newton-Raphson's iterations are solved by pre-conditioned congugate graduent methods like 1.C.C.G [37]. DATA HANDLING

Data input 2D finite element codes specializedfor magnetic field computation are generally compact packages in Fig. 5 : Description of an electro-magnet using which the problem data processing for the geometric desbrick and cylindrical nrimitives. of cription and discretization the domain have been correctly solved [231, 1241 ,[251. For three dimensional problems the crucial difficulty comes from the discretibuilzation because for field computation the finite element The geometric description is very simulara to ding set game. The user chooses elements a in library of grid must be extended beyond the object to the surrounelements, then he handles cubic, brick, cylindrical ding space. The use of powerful input processors which amake the selected elements to build his systems by assembling This method allows a very convenient graphical wide use of graphics is obviously essential. Pre-pro- them representation of the system. : topological cessing may be made following two ways The discretization of assembled elements is very easy description or geometric description. but that of the surrounding space is rather difficult.

...

3244

In fact this method is perfect for the field computaThe solution of magnetic field problems by the tion byan hybrid method using finite elements in satura- element method does not involve very difficult finite ble parts and boundary integral method for external theoretical points. From an engineeringpoint of view, parts for which the discretization can be avoided. Fig.5 the use of finite elements will be slowed down by two [1 an coming from Ancelle's work 3] shows the outlines of important factors: computing time and user's time for electromagnet described by this way. data processing.It appears that these problems will soon find a solution: the first one by the new generaResults display tion of powerful computers with the development of A very important aspect of magnetic field computaparallel computation (well adapted finite elements); to tion liesin the results display. When the magnetic scathe second by graphic interactive softwares specially lar potential is used it is possible to draw equipotendesigned for field problem data processing and oriented tiallines on some carefully chcsen surface. The surfaces towards geometric and discretization description and of interest are sometimes the boundary of the domainalso for display of results. but, often, some internal surface;a specific region of the domain may of special interest for a designer. be [ Fig.@ coming from3 2 ] shows a plot of equipotential lines on a ferrite stick and over the permanent magnet APPENDIX isolated from the whole system. To make this plot,a set 2D finite elements has of Variational expression for the magnetostatic field been created from visible faces of the discretized object; Droblem then a 2D algorithm has been used for plotting the equipotential lines. In this appendix, we show the equivalence between () the variational problem 8 and the differential pro(4 blem (3) and ( 6 ) with its boundary conditions ) and () 5.
First part o f the demonstration: Take a variation SA of , such that SA = 0 or 6An = 0 in order to verify 4 ) and (5) ( tThis varia-

8 tion leads to a variation ( ) obtained by : of

R
We use the following vector identities : Fig. 6 : A plot of equipotential lines a on ferrite stick and a permanent magnet. div

(SSE)

= -6x.curl

%+%

. curl 6 x

(A.

2a)

div (Sx.v'div

x)

6 z

. grad

(v'div A) + v'div z.divSx

For vector of poten-2D problems,equipotential lines (A. 2b) tial are in fact constant flux lines and thus map of the and the divergence theorem to convert into : (A.1) field. For 3D vector potentia1,Vector At equipotential Sx [curl % - grad (v'div - T] f dR plotting is meaningful on some special surfaces (like a SF = only the normal component is nonplane of symmetry) where R zero, butit lbases all physical meaning otherwise.Chari et a1 [ 2 0 ] , proposed a way of displaying the direction of strength and (A.3) cones the size of which is proportional to field strengkh, The arrival of graphic color displays the market, on Variational principles allow to write the folus allows one to expect some exciting development for : lowing preliminary conclusions 3D field representation. These devices make it possible to better display E - grad ( div " - = inside (A.4) different parts of the object by different colours. It also becomes possible to generate,instead of tradi-Ht = On where At is free (A. 5) tional equipotential lines, some colour pictures in which the colour is modulated by the value of potential, field = 0 on ' where kn free I is (A. 6 ) strengtk, flux density or any other physical quantity of v' div interest. We are justnow at the beginning of this era, and it 4s certain that some useful creative work will Second part of the demonstration : soon appear in this field. Let us select a special group of variations such that : CONCLUSIOX S x = grad S$ (A.7) After a brief survey of usual variational formulation! represents the variations of a scalar funcof the problem of magnetostatic scalar and vector poten- where 6$ of tial, we have proposed an original energy functional for tion space subject to the boundary conditions non linear anisotropic vector potential with a proof of 6$ = 0 on I 7 where At = 0 uniquenessofthesolutionofthisvariationalproblem. (A 8 ) We described then different kinds of finite elements and have we shown different two ways handle to the - =0 on I ' whereA = 0 (A. 9 ) difficalt processing data of input some D-rOblem with an n illustrated examples. Finally the display of results has been analysed In order to simplify the notation we write: and we emphasized the need of a graphical vizualisa- div u = V' (A. IO) tion of results at each point of the domain.

x)

'

3245

From the vector identity curl (grad 6$) = 0 , we find MAG-15, calculations", IEEE (A.1) for 6F =

[1 8

(A. 11) electri-ofanalysisE 1 elementChari. "Finite M.V.K 9 cal machinery and devices", IEEE MAG-16,5, No Using (A.2b) again and the divergence theorem the pp. 1014-1019 - 1980. variation (A.11) may be written as :

[U

. div grad

6$

- - . grad J

A. Wexler, "Some recent developments in field 6, pp. 1659-1664, 1979.

641

dR

6F =

grad&$. [ -gradU

-7 ]dR +

a w 1r
an

[lo] vdr
(a) An (A.12)

R
and - - 0 (A.8) an

We have only two possibilities on

M.S. Sarma, "Magnetostatic field computation by finite element formulation", IEEE MAG-12, No 6, pp. 1050-1052 1976.

= 0

[111

(b)

= 0 and U=v'divX = 0 (A.6) ;

M.S. Sarma, J.C. Wilson , "Accelerating the magnetic field iterative solutions", IEEE, MAG-12, NO 6, pp. 1042-1044 - 1976. Okuda, T. Kawamura and M. Nishi, I' Finite element solution of magnetic field and Eddy current problems in the end zone of turbine generators", IEEE, Winter meeting, N.Y. Paper No A 76 141-2 1976. Sarma, J.C. Wilson, P.J. Lawrenson and A.L. Joki, "End winding leakage of high speed alternators by three dimensional field determination", IEEE, Trans. PAS.90, No 2, pp. 465-477, 1971. WoLff, "Three dimensional eddy current calculations", Proc. of Compumag conference, OXFORD 1976.

in all cases the surface integralin (A.12) vanishes. The remaining volume integral converted using is : (A.2b) and the divergence theorem once more 6F
=

E121

- H.

j
R

6$ [Av+divy] dR

-I

r.

64.

r aIJ s

+ Jn]

dT

(A.13) [131

- M.S.

Using current properties (div = 0 everywhere and 7 J = 0 on r where 64 is allowed to be different n from zero (HL = 0)) and variational principles we may L write : [141

AU = 0

inside 0 on

- W.
-

au - -- 0
an

r where At

is free we find (A. 1

[151

From equations (A.14), (A.6) and (A.15)

V'

div

= 0

everywhere

A.G. Armstrong, C.J. Collie and J. Simkin, "The solution of 3D magnetostatic problems using scalar potentials", Proc. of Compumag conference, Grenoble - 1978.
E. Ancelle and J.C. Sabonnadilre, "Numericalo l u s

With the new relation (A.16), (A.4) becomes identical 8 to (A.3) and the validity of ( ) is provided. REFERENCES

[16]

tion of 3D magnetic field problems using boundary integral equations", IEEE Trans. MAG-16, No 5, pp. 1089-1091 - 1980. [17] 0.C Zienkiewiez, J. Lyness and D.J.R. Owen, "Three dimensional magnetic field using a scalar potential", IEEE Trans. MAG-13, No 51977.

E11

- E.A. -

Erdelyi and E.F. Fuchs, "Nonlinear Magnetic 11, Field Analysis of DC Machines", Part.1, 111, IEEE Trans. Pas.89, No. 7, pp. 1546-1583, 1970.

C181

- P.

[2]

0.W Andersen, "Laplacien Electrostatic Field Calculation by Finite Elements with Automatic Grid Generation", IEEE trans. Pas.92, N0.5 pp. 1485-1492 - 1973. [191 Chari and P. Silvester, "Analysis Turboof Alternator Magnetic Fields by Finite Elements", IEEE Trans. PAS90, N0.2, pp. 454-464 - 1971. [20] N.A. Demerdash, H.B. Hamilton and G.W. Brown, "Simulation for Design Purposes Magnetic of Fields in Turbogenerators with Symmetrical and Asymmetrical Rotors", Part. I and 11, IEEE, Trans. PAS.91, N0.5, pp. 1985-1999 - 1972. [21] A. Foggia, J.C. Sabonnadisre and P. Silvester, "Finite Element solution of saturated travelling magnetic field problems", IEEE, Trans. PAS.94, 1975. N0.3 [22]

Rafinejad, J.L. Coulomb, G. Meunier, "Permanent magnet 3D field computation by curvilinear finite elements", h o c . of Compumag conference, Grenoble - 1978.

- N.A,

[31

- M.V.K. -

Demerdash, T.W. Nehl and F.A. Fould, "Finiof te element formulation and analysis three dimensional magnetic field problems", IEEE MAG.16, NO 5, pp. 1092-1094 - 1980. Chari, Z . J . Csendes, P. Silvester, A. Konrad and M.A.Palmo, "Three dimensional magnetostatic field analysisof electrical machinery by the finite element method", IEEE, Winter meeting, Atlanta, Georgia, paper81 WM 175-9 - 1981. Zienkiewicz, "The finite element method in engineering science", New-York : MC Graw-Hill 1971.

141

- M.V.K

[5]

- O.C. - J.

[61

P. Silvester and P. Rafinejad, "Curvilinear finite elements for two dimensional saturable magnepp. tic fields", IEEE, Trans. PAS.93, N0.6, 18611870 1974.

Simkin and C.W. Trowbridge, "Which potential ? A comparisonof the varrous scalar and vector of potentials for the numerical solution the nonlinear Poisson problem", Rutherford Laboratories, 1978. RL - 78-C09/B

[23]
A . Konrad and P. Silvester, "Triangular finite

[7]

- E.

elements for the generalized Bessel equation of order m", Int.J. Numer. Methods eng., Vo1.7, pp. 43-55 - 1973.

Callegher, B. Morel, J.C. Sabonnadisre, B. An-. celle, J.L. Coulomb, P. Masse et G. Meunier, "Mode d'emploi du systsme FLUX", Rapport interne, Laboratoire d'slectrotechnique de Grenoble 1980

3246 [241 - D . A . Lowther, P.P. Silvester and E.M. Freeman, "The use of interactive postprocessing in the design of electromagnetic devices", IEEE, Trans. PlAG.16, No 5, pp. 803-805 - 1980. [25]

[ 0 - A. Konrad, J . L . Coulomb, J.C. Sabonnadisre and 41 P. Silvester, "Eigen value analysis of steadystate skin effect", Numerical methods in electrical and magnetic field problems, Santa Margherita Ligure, pp. 75-81 - 1976.

S.J. Polak and A. Wachters, "MAGGY 2 and PADDY program packages for and 3 dimensional magneto- [411 2 static problems", Compumag conference, Grenoble 1978.
J.L. Coulomb, "Analyse tridimensionnelle des champs dlectriques et magnstiques par la mzthode des &l&ments finis", Thesis. Grenoble 1981 [421 F. Young. "On the validity of characterizing eddv current phenomena by parabolic oartial differen. tial eauations". IEEE. MAG.14. No 6 DD. 11851186 - 1978.

[26]

J.L. Coulomb, P. Mass6 and J.C SabonnadiSre, "How to build your own CAD system : the cascade to program and its application finite elements program design", Numerical methods electrical in and magnetic field problems, Santa Margherita Ligure - 1976. A. Foggia et J.C. SabonnadiSre, "Les dquations de Maxwell en Glectrotechnique et leur rdsolution num&rique par une mdthode d'GlGments finis", Revue de physique appliquge, Tome 14, pp.439-444, 1979.

C271

ele[281 - C.J. Carpenter. "Comaaraison of alternative formu- [43] p. Rafinejad and J.C. sabonnadisre, I I F ~ ~ lations of 3-dimensional magnetic field and eddv ment computer programs in designof electromagnecurrent problems at Dower freauencies". Proc. IEEE tic devices", IEEE, MAG. 12, NO 5, pp. 575-578 Vol. 124. No 11. up. 1026-1034 - 1977. 1976.
[291

K. Ito. T. Tokumasu. S. Nagano. M. Tari and S.Dai. "Simulation for design Durposes of magnetic fields in turbine driven generator end region". IEEE Trans. PAS.99.No 4, KID. 1586-1596.
J.L. Coulomb and G. Meunier. "Finite elements of a seff starting micromsotor". Intermag conference, Grenoble - 1981.
E. Ancelle, "Definition an drawing of electromaenetics devices for field calculation", Compumag con1978. ference, Grenoble

[30]

[31]

[32]

P. Rafinejad, J.L. Coulomb, "Interactive computer techniques in three-dimensional modelisation of field problems by finite element method", Compumag conference, Grenoble- 1978.

[33] - M.J. Newman, "Experience with a distributed computing system for magnetic field systems", Compumag conference, Grenoble- 1978. [34] - B. Ancelle, J . L . Coulomb, P. Massd, G. Meunier and J.C. Sabonnadisre, "Software for computed aided analysis of electromagnetic fields", IEEE MAG. 16 No 6, pp. 1435-1437 - 1980. [35] - B. Ancelle, J . L . Coulomb, P. Mass&, G. Meunier and J.C. SabonnadiSre, "Computer methodsfor electrical byand magnetic devices designed field qnalysis", IEEE, MAG 15, No 6, PP. 1671-1673 - 1979. [36]

R.H. Gallegher, "Finite element analysis : fundamentals", Prentice-Hall, Englewood Cliffs, New Jersey - 1975.

[37] - D.A,H. Jacobs, "A rewiew of some recent developments in the solution of large systems of equations", Compurnag conference, Grenoble 1978. [381 - P. Silvester,. A . Konrad, J.L. Coulomb and J.C. Sabonnadisre, "Modal network representation of Stot Embedded conductors", Electric machines and Electromechanics, 1 : 107-122 - 1977. [391

J.C. Sabonnadisre, "Numerical analysis of eddy currents problems", in finite elements electriin cal and magnetic field problems, edited by M.V.K. Chari and P.P. Silvester, John Wiley and Sons, 1980.