IJNME2003 Inductiondirect

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/229471180
A numerical model for induction heating processes coupling

electromagnetism and thermomechanics
Article in International Journal for Numerical Methods in Engineering · October 2003

DOI: 10.1002/nme.796
CITATIONS READS
170 5,539
4 authors, including:
F. Bay Yann Favennec

MINES ParisTech Polytech Nantes
99 PUBLICATIONS 1,605 CITATIONS 83 PUBLICATIONS 907 CITATIONS
SEE PROFILE SEE PROFILE
J.L. Chenot
Transvalor and ParisTech
194 PUBLICATIONS 3,471 CITATIONS
SEE PROFILE
All content following this page was uploaded by Yann Favennec on 24 October 2018.
The user has requested enhancement of the downloaded file.

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng 2003; 58:839–867 (DOI: 10.1002/nme.796)
A numerical model for induction heating processes coupling

electromagnetism and thermomechanics
F. Bay∗; † , V. Labbe, Y. Favennec and J. L. Chenot

Centre de Mise en Forme des Materiaux; Ecole des Mines de Paris; UMR 7635 CNRS; BP 207;
06904 Sophia-Antipolis Cedex; France
SUMMARY
This paper presents a mathematical and numerical model developed for coupling the various physical
phenomena (electromagnetic, thermal and mechanical) taking place in axisymmetrical induction heating
processes. All three electromagnetic, thermal and mechanical models are time dependent and take full
account of the electromagnetic and thermal non-linear eect especially with magnetic materials. The
electromagnetic problem is discretized and solved in the workpiece, air and inductors. The heat transfer
equation and the mechanical equilibrium equations are solved in the workpiece only, both using a
nite element method. The mechanical model can take into account thermoelastic–plastic behaviour for
the part. The model has been successfully applied to several cases of induction heating. Comparisons
between numerical and experimental results show an excellent agreement. Copyright ? 2003 John Wiley
& Sons, Ltd.
KEY WORDS: induction heating; electromagnetic modelling; heat transfer computations; mechanical
modelling; multiphysics coupling; nite elements
1. INTRODUCTION
1.1. The induction heating process

Induction heating has become increasingly used in the last years for various industrial-
manu-facturing processes. It can be used as well at low current frequencies (∼ 50 Hz) for
initial preheating before deformation (forging, material forming), or at higher frequencies
(∼ 104 –106 Hz) for processes involving metallurgical heat treatment (quenching, hardening,
brazing), surface coating or melting in electromagnetic crucibles [1]. Its main advantages
are the fast heating rate, great precision in heating localization (supercial heating with high
∗ Correspondenceto: F. Bay, Centre de Mise en Forme des Materiaux, Ecole des Mines de Paris, UMR 7635
CNRS, BP 207, 06904 Sophia-Antipolis Cedex, France.
† E-mail:
francois.bay@cemef.cma.fr
Contract=grant sponsor: European Community; contract=grant number: 28283

Received 16 October 2001
Revised 2 January 2003
Copyright ? 2003 John Wiley & Sons, Ltd. Accepted 3 January 2003
840 F. BAY ET AL.
Figure 1. Induction heating set-up.
frequency power supplies for surface treatment process), instant start=stop (no warm up re-
quired for each cycle) and its good reproducibility.
The basic induction setup consists in one or several inductors and metallic workpieces to
be heated (see Figure 1). The inductors are supplied with alternating current with frequen-
cies ranging from fty to several hundred thousand cycles per second. A rapidly oscillating
magnetic eld is generated and in turn induces eddy currents in the workpiece due to the
Joule eect. These currents generate ohmic heat losses inside the workpiece. Moreover, for
ferromagnetic materials, magnetic hysteresis eect also contributes to heat generation, but in
a much smaller amount. Most of the dissipated heat is produced in a thin layer of depth
under the surface of the workpiece. The skin depth is dened as the depth at which the
magnitude of the eld drops to a value equal to its surface value multiplied by e−1 :

1
= (1)
f
where f is the frequency, the electrical conductivity and the magnetic permeability. For
instance for a typical steel at 20◦ C, the skin depth may range from 3 mm at a frequency of
50 Hz down to 0:04 mm at 105 Hz. It is clear from Equation (1) why high frequencies are used
to achieve surface heating, while low frequencies are favored for initial uniform pre-heating.
1.2. Potential applications of a numerical model

Industrial use of induction heating processes aims at achieving various objectives, which are
among others:
• reaching a uniform prescribed temperature,
• achieving control of the grain size,
• getting a certain hardness level at the surface.
Industrials try to reach most of these goals by time consuming trial-and-error procedures.
Fine and accurate control of these goals may be hard to achieve most of the time. Further-
more, since induction heating processes involves and couples various physical phenomena, the
goals will be achieved only through an in-depth understanding of the industrial processes, and
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 58:839–867
NUMERICAL MODEL FOR INDUCTION HEATING PROCESSES 841
therefore through numerical modelling. The numerical model presented here has been devel-
oped having in mind the fact that a ne and ecient numerical model coupled with op-
timization techniques [2] will be a major step in determining optimal process parameters
(frequency, current density, coil geometry and velocity, etc) suitable to reach the industrial
objectives previously mentioned.
1.3. The various numerical models

Several numerical models have been developed up to this day in order to model the induction
heating process. These models carry out a simulation of the induction heating process but
with various restrictions. Most numerical models use a harmonic approximation to evaluate
the electromagnetic eld in the workpiece. This approximation is valid for linear magnetic
materials but can become inaccurate when dealing with non-linear magnetic materials [3, 4].
Most models do not carry out full coupling between electromagnetism, heat transfer and solid
mechanics, but focus only on the electromagnetism=heat transfer coupling. Concerning the
numerical methods, the coupling between inductor and workpiece is carried either through the
use of a boundary element method [5] or through nite elements [6]. In this last method, the
air domain is also meshed and the electromagnetic equation is solved on the global domain
made out of the air, the inductor and the workpiece.
2. THE MATHEMATICAL MODEL
As previously stated, induction heating processes involve three main dierent physical phe-
nomena, related to electromagnetism, heat transfer and solid mechanics. A complete mathe-
matical model for induction heating processes therefore needs to consider and tightly couple
the equations modelling these phenomena. We present here a general mathematical model
which couples:
• the Maxwell equations, in order to access the electric eld in the workpiece,
• the heat transfer equation: the heat dissipated in the part by the eddy currents being the
main source of temperature evolution in the workpiece,
• the mechanical equilibrium equations, which will enable us to access the stress and strain
rates in the workpiece.
2.1. The electromagnetic model

2.1.1. The Maxwell equations. The global system of equations modelling electromagnetic
wave propagation and any kind of electromagnetic phenomena was established by Maxwell
more than a century ago. This system is made up of the four following equations:
Magnetic ux equation:
∇:B=0 (2)
Maxwell–Gauss equation:
∇:D=0 (3)
842 F. BAY ET AL.
Maxwell–Faraday equation:
@B
∇ × E=− (4)
@t
Maxwell–Ampere equation:
@D
∇ × H=j + (5)
@t
where H is the magnetic eld, B the magnetic induction, E the electric eld, D the electric
ux density, j the electric current density associated with free charges, and x denotes the
vector product—thus ∇ × E is the curl vector of E. This system of equations needs to be
completed by relations which take into account material properties (magnetic permeability ,
electrical conductivity , dielectric constant ):
D = E (6)
B = (T; H)H (7)

(T )E in the conductors
j= (8)
0 in the air
relation (8) being commonly known as the Ohm law. It should be pointed out here that the
magnetic permeability , and the electrical conductivity do strongly depend on temperature.
Furthermore, for non-linear ferromagnetic materials, the magnetic permeability may also
depend on the magnetic eld strength H. Equations (2)–(8) are valid for any range of
frequencies and for any kind of geometric congurations.
2.1.2. The magneto-quasi-static approximation. In induction heating processes, the range of

frequencies dealt with can go from 50 Hz (low frequency for homogeneous heating) to several
hundreds of MHz (high frequency for heat treatment applications). The magneto-quasi-static
approximation consists in neglecting the displacement currents @D=@t in the Maxwell–Ampere
equation (5). This approximation therefore leads to neglecting the propagation phenomena.
This hypothesis is acceptable if the distances between the source locations and the points
where the electromagnetic eld is computed are smaller than the wave length. This is the
case for the range of frequencies typically used in industrials set-up, since frequencies are
smaller than 109 Hz. This approximation is realistic here; thus the Maxwell–Ampere equation
(5) becomes
∇ × H=j (9)
Let us now divide Equation (4) by the magnetic permeability and take the rotational on both
sides of this equation, we get

1 @
∇× ∇ × E = − (∇ × H) (10)
@t
from which we can eliminate the magnetic eld by using Equation (9), which leads to

1 @j
∇× ∇ × E =− (11)
@t
The total current density being the sum of the induced current density and of the prescribed
one JS :
j = jinduced + JS (12)
and the induced current density being related to the electric eld through the Ohm law:
jinduced = E (13)
we nally get

@E 1 @JS
+∇× ∇ × E =− (14)
@t @t
The electric eld is solution in 3D computations of Equation (14), coupled with Equation (3)
and with initial and boundary conditions.
2.1.3. Axisymmetrical congurations. When dealing with axisymmetrical congurations, the

electric eld E can be expressed in cylindrical co-ordinates with a non-zero component only
in the direction:
E = (0; E (r; z); 0) (15)
This conguration is a transverse magnetic mode, see Reference [1], where the electric eld E
is directed in the -direction, whereas the magnetic eld H propagates in the (r; z) directions.
Equations (14) and (3) can then be reduced to a single one, with JS (r; z; t) = (0; JS (r; z; t); 0):

@E 1 1 E @ 1 E @J
− ∇: ∇E + − = − S (16)
@t r 2 @r r @t
where the term

@ 1 E
(17)
@r r
is usually neglected in the electromagnetic model [7]. However for ferromagnetic materials,
the inuence of the derivative of the magnetic eld can be strong and neglecting it conse-
quently aects the numerical results. The transfer of experimental power or current intensity
measurements into numerical input data is one of the most complex tasks in the simulation
of induction heating processes. The input for the electromagnetic equation is entered through
a current density term. The variation with time of the current intensity may be of any shape
(sinusoidal, impulsion or more complex shapes). For instance, for a sinusoidal source current,
we get
JS (r; z; t) = JO (r; z) cos(!t + ’) (18)
with the pulsation ! = 2f.
844 F. BAY ET AL.
Figure 2. Typical magnetization curve obtained with the Frohlich–Kenelly model. |B| is expressed in
Tesla and |H| in A/m. Coecients and are, respectively, 1.5 and 4500.
2.1.4. The boundary conditions. The choice of boundary conditions needs to be carried out
carefully: nite element computations of electromagnetic eld in the air can be altered by
‘articial reections’ on the boundary of the air domain. A fairly good approximation can be
obtained if a Robin boundary condition is applied on an enclosing ‘box’ for the surrounding
air, the dimensions of the ‘box’ being ‘large enough’ in order not to perturb the magnetic
eld lines. This Robin like condition writes in our case:
@E E
+ er : n = 0 (19)
@n r
where @E =@n means the derivative of E along the unit outward normal direction n and er
is the unit radial direction. In axisymmetrical cases, a null Dirichlet boundary condition for
the E variable is prescribed on the symmetry axis.
2.1.5. The magnetic permeability. The physical parameters involved in the electromagnetic
model are the magnetic permeability, which can be dened as the derivative of the mag-
netic induction with respect to the total magnetic eld strength: = @B=@H (see Refer-
ences [3, 8, 9]) and the electrical conductivity, both being strongly dependent on temperature.
The relative magnetic permeability can suddenly drop from several hundreds (or even thou-
sands) down to unity when temperature gets close to the Curie temperature TC . Since most
electro-thermal problems take place at high elds, a usable approximation of the magnetic
permeability—and thus magnetization curve, see Figure 2, [10]—is given by the Frohlich–
Kenelly model [11, 12]:

|B| = 0 + |H| (20)
+ |H|
where and are two constant coecients. Following the Frohlich formulation, the magnetic
permeability becomes simply

|H|
(|H|) = 0 1 + − (21)
0 ( + |H|) 0 ( + |H|)2
It is still a good approximation when temperature dependency is added. In the rst type
of models, the coecients and are temperature dependent, [13]. In the second type of
models, a separation of the temperature dependency from the magnetic eld dependency is
used [3, 14, 15] leading to the general form

|H|
(|H|; T ) = 0 − f(T ) (22)
+ |H| ( + |H|)2
where the temperature dependency f(T ) writes [16]

T
f(T ) = 1 − (23)
TC
TC being the Curie temperature and being the temperature sensitivity parameter.
2.1.6. Computation of the magnetic eld H. The magnitude of the magnetic eld H is
determined in two stages. First, the magnetic induction eld B is computed from the electric
eld values by integrating in time relation (4) with an explicit time integration scheme:


 @
 Br (t) = Br (t − t) − t
 E
@z
B(r; z; t) (24)

 t @ E @E

 BZ (t) = BZ (t − t) + (r : E ) = BZ (t − t) + t +
r @r r @r
Relations (20), (21) work well when one knows the magnetic eld H and needs simply to
compute the magnetic permeability (|H|T ), [17]. Conversely, to compute the H eld one
needs to have an equivalent relation
B
H = (25)
(B; T )
This relation is obtained from the Frohlich–Kenelly model (20) by seeking the physically
realistic solution of the second order equation in H. We get

1
H = B − − 0 + ( + 0 − B) + 40 B
2 (26)
20
Relation (26) becomes, with the temperature dependency,

1
H = B − f(T ) − 0 + (f(T ) + 0 − B) + 40 B
2 (27)
20
846 F. BAY ET AL.
2.2. The thermal model

Modelling of induction heating processes involves modelling of heat transfer, which is mainly
governed by the heat dissipated by eddy currents in the workpiece, heat diusion and heat
uxes at the interface between the workpiece and the air.
2.2.1. The heat transfer equation. Temperature evolution in the workpiece is governed by
the classical heat transfer equation:
@T
C − div(k ∇T ) = Q em (28)
@t
where is the workpiece density, C the specic heat, k the thermal conductivity, and Q em the
heat source term due to eddy currents, all temperature dependent. The heat capacity can vary
sharply with temperature—especially during metallurgical phase transitions and near Curie
temperature. The thermal conductivity decreases smoothly with respect to temperature.
2.2.2. The heat source term due to eddy currents. Q em is local heat density rate supplied by
the eddy current derived from the mean electromagnetic solution
(n+1)T
1
Q em = E2 with E2 = (t)E2 (t) d t (29)
T nT
T being the period of the electromagnetic periodic signal. Taking this average value is justied
when one compares the scale of the electromagnetic time step (less than 0:001 s) with the
average induction heating time scale (several seconds).
2.2.3. The boundary conditions. Induction heating can involve dierent kinds of boundary
conditions at the interfaces: convection and radiation at the interface between the workpiece
and the air, prescribed heat ux or temperature. A general model needs therefore to consider
all these conditions, which are summarized here:
Convection=radiation:
−k ∇T : n = h(T − Text ) + emi Ste (T 4 − Text
4
) (30)
where h denotes the convection coecient, Text the room temperature (in Kelvin), emi the
material emissivity, and Ste the Stefan constant,
Prescribed heat ux:
−k ∇T : n = prescribed (31)
Prescribed temperature:
T = Tprescribed (32)
2.3. The mechanical model

Induction heating can generate—especially for heat treatment applications—mechanical stresses
and strains in workpieces. We have therefore chosen to include a mechanical model which
can take into account thermal, elastic and plastic behaviour for the material being heated. The
model is based on the classical virtual work principle, which expresses the equilibrium of
the workpiece undergoing thermo-mechanical loads, as well as on a general constitutive law
based on the material mechanical properties.
2.3.1. The equilibrium equations. The classical equilibrium equations for a solid undergoing
mechanical loads can be expressed in a local form by
div() + f = 0 (33)
where denotes the stress eld in the solid and f the body forces per mass unit.
2.3.2. The constitutive law. The material behaviour is assumed to obey a thermo-elastic–
plastic constitutive law. The material strain rate ˙ is decomposed as the sum of three fractions:
namely, the elastic fraction ˙el , the plastic fraction ˙pl and the thermal one ˙th :
˙ = ˙el + ˙pl + ˙th (34)
The elastic fraction ˙el is related to the stress eld tensor ˙ through the Hooke law:
1+
˙elij = ˙ij − ˙kk ij (35)
E E
where E and , respectively, denote the Young modulus and the Poisson coecient. It should
be noted here that we do not need to use a Jaumann derivation for ˙ since the rotations
involved are quite small. The deformation ˙th due to thermal expansion can be expressed as
ij = Ṫ ij
˙th (36)
where denotes the volume thermal expansion coecient. The equivalent stress is dened
through the Von Mises yield criterion:
(eq )2 = 12 ((xx − yy )2 + (yy − zz )2 + (xx − zz )2 ) + 3(xy
2
+ yz
2
+ xz
2
) (37)
The equivalent stress eq needs to comply with the material yield stress Re :
eq 6Re (38)
In case the equivalent stress is equal to the material yield stress Re , plastic deformation will
occur. The relation between the plastic strain rate and the stress eld is governed by the
plastic ow law:
@f
˙pl = (); ¿0 (39)
@
where f denotes the yield criterion and the plastic multiplier. The model can include
strain-hardening, by using a power law
eq = 0 = Ks n (40)
where n denotes the strain-hardening exponent and the equivalent deformation dened by
= 2
3 ij2 (41)
i; j
848 F. BAY ET AL.
2.3.3. The physical parameters. The material parameters involved in the mechanical model
include Young’s modulus, Poisson coecient, the yield stress and the strain-hardening coe-
cient. All these parameters depend on temperature and need to be well determined, especially
when dealing with heat treatment applications where the highly localized heat source can
induce stresses and distortion that can interact with metallurgy. The evolution of these pa-
rameters can be provided in two ways: either through the use of constant values or tabulated
values of the parameters at some given temperatures, or by using analytical expressions den-
ing the evolution of these parameters with respect to temperature.
3. THE NUMERICAL METHODS
Up to this day, most numerical methods for modelling induction heating processes have mainly
focused on the electromagnetic and thermal computations. Most of these methods use one of
the three following approaches:
• nite elements method,
• boundary element method,
• mixed nite elements=boundary elements method.
The coupled electromagnetic, thermal and mechanical computations are carried out in the
following way:
• in the part, electromagnetic, thermal and mechanical computations,
• in the inductor, electromagnetic and thermal computations,
• in the air, electromagnetic computations.
We have chosen to carry out a complete nite element approach using six-nodes quadratic
triangles (P2) for these three kind of computations. This means that the coupling between
the part and the inductor for the electromagnetic computations is carried out through the
computation of the electromagnetic eld in the air. In the following subsections the specicities
and implementation details of our approach are described.
3.1. The electromagnetic problem

3.1.1. The weak formulation. We consider the domain as being
= part ∪ inductor ∪ air (42)
Due to the axial symmetry, we consider from now on cylindrical co-ordinates (r; ; z). We
dene the two dimensional domain axi as
axi = {(r; = 0; z) ∈ } (43)

We dene (resp. 0 ) as being the outer boundary of the axi domain (resp. part of the
outer boundary where the unknown is prescribed). The functional space V in which we are
searching the solution is dened in accordance with the regularity of the solution

V= ∈ H ( axi ); ∈ L ( axi )
1 2
(44)
r
with H 1 ( axi ) a Sobolev space:

H 1( axi ) = { ∈ L2 ( axi ); ∇ ∈ L2 ( axi )} (45)
To establish a weak formulation of the electromagnetic model, Equation (16) is multiplied by
a test function belonging to the functional space V and integrated on the whole domain.
After using the Green theorem, we get

@E
a ; + b(E ; ) = l( ) ∀ ∈ V
@t

@E @E
a ; = r ds
@t @t
(46)
1 1 1 @
b(E ; ) = ∇ E : ∇ r d s + E : d s + (E ) d s
r @r

@Js 1 @E E
l( ) = − r ds + + er : n : r d
@t − 0 @n r
where d s = d r= d z. The integral term (47) depends on the boundary conditions and vanishes
with the imposed Robin-like boundary conditions, see Section 2.1.4.

1 @E E
+ er n : d s (47)
− 0 @n r
If we assume and to be independent on the unknown eld E , this problem belongs to
the class of parabolic problems, since:
• the bilinear form a(: ; :) is continuous and positive denite,
• the linear form l(:) is continuous,
• the bilinear form b(: ; :) is continuous and V -elliptic.
Such evolution problems with mixed boundary conditions are sometimes called Cauchy–
Dirichlet–Neumann problems and the existence and uniqueness of a solution E (t; x) belonging
to L2 ([0; t]; V ) can be proven [18].
3.1.2. Space discretization. Having established a weak formulation, we can now carry out
the nite element space discretization of Equation (46). The functional space V is approached
by a discretized space V h , the test functions by h and the unknown E by E h (for the
sake of clarity, we will not carry on using for the unknown). The discretized version of
Equation (46) is then
h
@E
a (t); h + b(E h (t); h ) = l( h ); ∀ h ∈ V h (48)
@t
A basis of the discretized space is provided by the shape functions Ni associated to each node
i of the mesh:

1; i = j
Ni (node j) = (49)
0 otherwise
850 F. BAY ET AL.
This means that Equation (48) needs only to be veried for all shape functions Ni , and that
the unknown of the electrical eld E h can be expressed as
nb:nodes nb:nodes
E h (x) = E h (node j) : Nj (x) = Ejh : Nj (x) (50)
j=1 j=1
Equation (48) thus becomes, in a Galerkin approximation:

nb:nodes @E h nb:nodes
j
a (t) : Nj ; Ni + b Ejh (t) : Nj ; Ni = l(Ni ); ∀i = 1; nb:nodes (51)
j=1 @t j=1
which can be rewritten as

nb:nodes @Ejh nb:nodes
(t) : a(Nj ; Ni ) + Ejh (t) : b(Nj ; Ni ) = l(Ni ); ∀i = 1; nb:nodes (52)
j=1 @t j=1
which is equivalent to the following linear system:

@E
em
[C ] (t) + [K em ]{E(t)} = {Bem } (53)
@t
with
nb:elts
[C em ]ij = a(Nj ; Ni ) = Nj Ni r d s
elt=1 elt

nb:elts 1 1 1 @
[K em ]ij = b(Nj ; Ni ) = ∇Nj :∇Ni r d s + Nj Ni d s + (Nj Ni )r d s (54)
elt=1 elt elt r elt r @r

nb:elts @Js
{B }i = l(Ni ) =
em
− : Ni r d s
elt=1 elt @t
We wish to draw attention here to the fact that using nite elements for electromagnetic
computations in the air requires to mesh a large enough domain to avoid any articial reection
problems. It will be shown in the numerical result section that a good quality computation
must display magnetic eld lines that do not intersect the boundary of the global domain.
Another point which helps avoid articial reection problems is to have well-suited boundary
conditions for the air domain.
3.1.3. Time discretization. Since we have chosen to solve the time-dependent model, we now
need to integrate numerically in time the space-discretized electromagnetic equation (53). For
accuracy purposes, we have selected and implemented a second-order two time step nite
dierence scheme detailed here. In a rst stage, the system is solved at time t ∗ such that
t − t1 ¡t ∗ ¡t + t2 :
t ∗ = 1 (t − t1 ) + 2 t + 3 (t + t2 ) with 1 + 2 + 3 = 0 (55)
where t1 denotes the previous time step and t2 the current one. The electric eld E ∗ at time
t ∗ and its time derivative write
E ∗ = 1 E t−t1 + 2 E t + 3 E t+t2 (56)
@E ∗ E t+t2 − E t E t−t1 − E t
= + ( − 1) (57)
@t t2 t1
System (53) is written at time t ∗ : E ∗ and its derivative are replaced by expressions (56) and
(57). The system is solved for the unknown variable E ∗ :

[C ] + [K ] {E }∗ = {Bem }∗ + c1 [C em ]∗ {E }t + c0 [C em ]∗ {E }t−t1
em ∗ em ∗
3 t2
−1 2
c1 = + + (58)
t2 t1 3 t2
1 −1
c2 = −
3 t2 t1
The two-time step scheme we have used needs to solve a non-linear equation, as the matrix
[C em ] is dependant on the magnetic eld. In order to avoid an additional non-linearity in the
system, the matrices are linearized [19] in order to depend only on their values at time t
and t − t1 :

t2 t2
[C]∗ = 1 − 3 [C]t−t1 + 2 + 3 1 + [C]t (59)
t1 t1
The second and last stage consists in the computation of {E } at time t + t2 :
1
{E }t+t2 = ({E }∗ − 1 {E }t−t1 − 2 {E }t ) (60)
3
The choice of a good value for tem is not obvious; however, a value of around (T=32) or
(T=64)—where T stands for the period of the power current supply—seems to be a good
compromise between computation time costs and results precision. We now have a complete
numerical scheme which enables us to compute the response in terms of electric eld for a
given current power supply. As the matrix of the linear system (58) is symmetric and positive
denite, the linear system is solved with an iterative pre-conditioned conjugate gradient solver,
using a diagonal preconditioner.
3.1.4. The moving inductor case. In heat treatment applications, we are often faced with the
case of inductors moving continuously along the z-axis. When carrying out a global nite
element simulation, this could be dealt with by moving the inductor outlines and mesh in the
global mesh. However, this can lead quite rapidly to large distortions of the mesh in the air
and may induce several time-consuming remeshing of the entire domain. We have adopted
here another strategy:
• the area where the inductor will move through is initially dened and meshed separately,
• the electromagnetic properties of this area are moved back and forth from air properties
to inductor material properties.
852 F. BAY ET AL.
Figure 3. Example of mesh for a simulation with a coil displacement: the area
of displacement is delimited and meshed.
The inductors are moved virtually through a continuous change of properties and location of
source terms, Figure 3. This enables an accurate simulation of the process without any mesh
distortions, and thus avoids time consuming remeshing problems.
3.2. The thermal problem

The basic approach will be the same as for the electromagnetic problem. We consider the
thermal problem on the domain TH : made out of the workpiece; the boundary TH can
be decomposed in several parts with various boundary conditions: prescribed temperature
or heat ux, convection, radiation. We then get the following equations discretized in
space:

@T
th
[C ] (t) + [K th ]{T (t)} = {Bth } (61)
@t
nb:elts
[C th ]ij = a(Nj ; Ni ) = CNj Ni r d s
elt=1 elt
nb:elts

[K th ]ij = b(Nj ; Ni ) = k ∇Nj :∇Ni r d s + hNj Ni r d (62)
elt=1 elt @elt ∩ 2
nb:elts

{Bth }i = l(Ni ) = Q em : Ni r d s + prescribed Nj r d + hText Nj r d
elt=1 elt @elt ∩ 1 @elt ∩ 2
The space-discretized thermal equation (61) is then integrated in time using the same second-
order two-time step nite dierence scheme as for the time integration of the electromagnetic
problem.
3.3. The mechanical problem

We solve the mechanical problem—for which the unknown eld is the displacement eld
u—on the domain ME restricted to the workpiece. The boundary ME is made out of the
symmetry axis, for which we have a null Dirichlet boundary condition, and of the complement
which is a stress-free surface. We then get the following space-discretized equations:
[K me ]{u(t)} = {Bme } (63)
The expressions for the matrix and loading vectors are derived from the virtual work principle.
In a purely linear elastic case for instance, these expressions are given by
nb:elts

me
[K ]ij = ll (Nj ) : kk (Ni )r d s + 2 lk (Nj ) : lk (Ni )r d s
elt=1 elt elt

nb:elts
(64)
{Bme }i = f: Ni r d s
elt=1 elt
which is evaluated for the components and for nodes i and j, while and denote the
Lame coecients. The rst term for [K ME ] is evaluated using a reduced integration procedure.
In case of non-linear mechanical behaviour, the matrix [K ME ] in system (63) depends on
the solution. A Newton–Raphson algorithm is used in that case to converge towards the
solution, [20].
3.4. The coupling procedure

Numerical modelling of the induction heating procedure requires the development of a cou-
pling procedure between:
• electromagnetic computations,
• thermal computations,
• mechanical computations.
A weak coupling procedure has been developed. The computation takes place in an incremental
way, with two dierent time steps:
• one time step tem for electromagnetic computations,
• one time step tth for thermal and mechanical computations.
At each time step, the electromagnetic eld distribution and thus the induced currents depend
explicitly on the thermal eld of the body, ruled by the heat transfer equation. Inversely, the
eddy currents calculated from the electromagnetic solution are used as the thermal sources
for the thermal nite element analysis. The handling of the coupling is carried out in two
dierent stages. Figure 4 shows the global organization of the coupling procedure. Concerning
the thermal-mechanical coupling, the thermal evolutions drive the dilatation of the part, but
the mechanical deformations do not act on the thermal eld.
854 F. BAY ET AL.
Figure 4. Organization of the coupling procedure.
3.4.1. Criterion from electromagnetic to thermal calculations. The criterion from electro-
magnetic to thermal calculations rely on the stabilization of the mean heat source term over
the electromagnetic periods. Once the electromagnetic eld has been calculated, the rate of
heat generation Q em for the heat equation is computed for all integration points of the nite
element mesh. As the electromagnetic time step is far smaller than the thermal one, we do
not consider the instantaneous Joule power calculated at a given time at every integration
point. We rather consider a mean Joule power averaged over one or several periods of the
electromagnetic eld:
nT
1
Q em (nT; int) = (int; t) |E (int; ; t)|2 d t (65)
T (n−1)T
where int is the considered integration point, T is the period of the power supply currents,
n is number of periods considered and E (int; t) is the value at time t of the electric eld
interpolated at the integration point int. At the end of each electromagnetic period, the newly
calculated mean power is compared to the one calculated at the previous period until it
stabilizes. The following convergence test is conducted at every integration point:
Q em ((n + 1)T ) − Q em (nT )

¡ (66)
Q em (nT )
where is the magnetic convergence parameter. If condition (66) is veried then thermal com-
putations are started with the stabilized thermal source power calculated during the (n + 1)T
period. When dealing with linear magnetic materials and sinusoidal power current supplies, the
previous procedure converges in two or three periods depending on the value of the magnetic
convergence parameter.
3.4.2. Criterion from thermal to electromagnetic calculations. The criterion from thermal
calculations back to electromagnetic calculations is based on the variations of the magnetic
and electric parameters with temperature. Thermal computations can use the same source term
derived from the electromagnetic computations as long as the physical magnetic parameters
such as the magnetic permeability and the electric conductivity do not change ‘too much’.
Their variations with temperature are tested after each new thermal computations. The fol-
lowing criteria are used for each mesh element:
n+1 n n+1 n
(Tmax ) − (Tmax ) (Tmax ) − (Tmax )
¡5% and ¡5% (67)
(Tmax )
n (Tmax )
n
n+1
where Tmax is the maximum value of the temperature eld in the mesh element at time t + d tth
n
and Tmax is the maximum value of the temperature eld in the same element at the current
time t. When the maximum relative variations reach a given threshold (5% in our case), the
previously calculated mean heat power is assumed to be irrelevant. A new electromagnetic
calculation is then therefore carried out.
4. RESULTS AND DISCUSSION
This part is devoted to the presentation of some numerical results and numerical studies. Three
dierent cases are presented here:
• a case with a static long inductor (about the same size as the workpiece),
• a case with a short inductor (compared to the size of the workpiece),
• a case with a moving short inductor—typical of a heat treatment process.
These cases have been used to check the validity of the numerical computations with re-
spect to experimental data or other numerical results. The numerical results presented here
enable to better understand the behaviour of the temperature and electric eld with respect to
the physical and numerical parameters. Results are also presented for the thermo-mechanical
coupling stage. Computations of the cases were run on a single processor Pentium3. The
856 F. BAY ET AL.
Figure 5. Static long inductor case: geometry and mesh (12 000 nodes).
Table I. Geometric data.

Billet height = 120 mm
Billet diameter = 44 mm
Billet centre hole diameter = 3 mm
Thickness of Kaowool insulation = 3 mm
Height of ceramic former = 120 mm
Internal diameter of ceramic former = 50 mm
External diameter of ceramic former = 60 mm
Ceramic material = Fused Alumina
Number of layers for copper windings = 3
Number of turns on inner layer = 55
Number of turns on middle layer = 54
Number of turns on outer layer = 55
Diameter of copper wire = 2 mm
Length of coil on former = 117 mm
computational times were between than 5 min for the linear case and hours for the non-linear
cases for meshes of around 12 000 nodes.
4.1. A static long joint-spires inductor case

4.1.1. Description of the case. The case deals with a part in ferromagnetic EN3 mild steel
annealed at 930◦ C. In this case, the workpiece is heated by an inductor of the same length
as the workpiece. The geometry consists in a cylindrical EN3 part surrounded with Kaowool
insulation, a ceramic tube and nally a three-layers coil. The geometry and the mesh used
for this case are displayed in Figure 5 and in Table I. The multi-turn coil is modelled as a
continuous single coil with a uniform initial current density. Two thermocouples have been
Figure 6. Electrical conductivity, specic heat and thermal conductivity versus temperature.
placed: the rst one on the surface on the middle of the part and the second one in a hole
bored in the centre of the part so as to measure temperature inside the part.
The physical data are given in Figure 6 by a set of physical values for dierent temperatures
and computed at a given temperature by linear interpolation. The magnetic permeability of the
EN3 steel is dependent on the magnitude of the magnetic eld and follows model (21), with
the Frohlich coecients: = 2:31695 and = 5769:55. These data did not take into account
the temperature dependency of the magnetic permeability. Concerning the process parameters,
their values are: frequency: 500 Hz, current density: 8 × 108 A= m2 , electromagnetic time step:
T=64 = 3:125 × 10−5 s, thermal time step: 0:5 s.
4.1.2. Results and validation. In order to enable a visualization of the dissipated joule power,
an eective electrical eld has been dened at a given node as the integration over an elec-
tromagnetic period of the square of the real instantaneous electrical eld:

(n+1)T
1
Ee = E 2 (t) d t
T nT
This eective electrical eld is also proportional to the amplitude of the electric eld in a
harmonic complex formulation. Figure 7 shows the eective electric elds isovalues. Due to
the high value of the permeability, we can notice the concentration of the magnetic eld along
the workpiece surface; this is a well-known physical eect since magnetic eld lines tend to
concentrate on material surface when the magnetic permeability gets high. It is indeed a strong
point of our model that it can take into account non-linear magnetic eects. The non-linear
magnetic electromagnetic properties of steel tend to distort the shape of the current or voltage
waveforms. Both have been measured in the coil (whether it is the voltage that becomes
distorted or the current depends on the nature of the power supply). Figure 8 displays the
evolution with respect to time of the experimental intensity measured in the coil and of the
electric eld in the part. The wave’s shapes show the same tendencies with similar bumps
after the maximum and minimum peaks.
858 F. BAY ET AL.
Figure 7. Long inductor case: eective electric eld isovalues.
Figure 8. Evolution with respect to time of: (a) the experimental intensity in the coil; and (b) the
numerically computed electrical eld in the part.
Figure 9. Long inductor case: comparison between experimental and computed temperature evolutions.
Figure 9 shows the excellent agreement between experimental and computed temperatures
on the surface and inside the workpiece. On one hand, the computed temperatures on the
surface provide a good way to validate the dissipated heat due to eddy currents inside the
workpiece—and thus the electric eld computation and the whole electromagnetic procedure.
On the other hand, the good agreement observed between the experimental and numerically
computed delay on heat propagation to the internal part of the workpiece validates the heat
transfer diusion model and the thermal coupling procedure. Figure 10 shows the temperature
isovalues at two given time steps. Heat propagates here rather uniformly from the surface
towards the center. This is typical of a global homogeneous heating application. Figure 11
shows the results obtained in terms of velocity eld for the coupled mechanical computa-
tion. One can observe the dilatation eects taking place in the workpiece while it is being
heated.
4.2. A static short inductor case

4.2.1. Description of the case. The geometry of the case studied here has been presented in
Reference [7]. It details the analytical solutions for two dierent constant magnetic perme-
ability of the part: = 0 and = 90 ∗ 0 . The analytical expression is valid for a relative
magnetic permeability of one. For a relative magnetic permeability greater than 1, a sup-
plementary due to the radial derivative of the magnetic permeability needs to be added. It
considerably modies the electromagnetic eld, especially on the interface between the part
and the air. The electromagnetic and temperature elds are shown as well as the mechanical
deformations. Figure 12 shows the mesh which has been rened along the workpiece surface
860 F. BAY ET AL.
Figure 10. Long inductor case: temperature eld at two given time steps.
Figure 11. Long inductor case: nodal velocities in the mesh at a given time step.
Figure 12. Static short inductor case: geometry and mesh.
Table II. Physical parameters.

Steel (part) Copper (coil) Air
( −1 =m1 ) 3 × 106 60 × 106 0

relative 1 or 90 1 1
c (J=kg1 =K 1 ) 4:87 × 106 3:43 × 106 1:3 × 103
k (W=m =K )
1 1
35 389 2:63 × 10−2
Dilatation 107
in order to have several elements within the electromagnetic skin-depth. The physical param-
eters are constant and summarized in Table II. The process parameters are: frequency: 60 Hz,
current density: 7 × 109 A m2 , electromagnetic time step: T=32 = 6:25 × 10−4 s and thermal time
step of 1 s.
4.2.2. Results and validation. The electromagnetic eld converges in three periods (relative
= 1) or in 8 periods (relative = 90) with a magnetic convergence parameter of 0.0001%. The
maximum of the eective electric eld is located on the surface of the part for a magnetic
part.Figure 13 shows the time evolution of the electrical eld for a non-magnetic material
at three dierent locations of the entire domain. We can see that the eld is for this ma-
terial maximum in the coil. A slight delay exists between the eld in the part and those
in the air or in the coil. We can observe on Figure 14 the displacement of the maximum
value of the eective electric eld towards the surface when the part becomes magnetic.
862 F. BAY ET AL.
Figure 13. Evolution of the electric eld versus time for a non-magnetic material at three given locations
in the coil, the part and in the space in between. The electrical eld is maximum in the coil.
Figure 14. Static short inductor case: eective electric eld isovalues for two
dierent permeabilities of the part at a frequency of 60 Hz: (a) non-magnetic
part: relative = 1; and (b) magnetic part: relative = 90.
Figure 15. Eective electrical eld proles on the radial axis for three dierent fre-
quencies f = 700, 1000 and 5000 Hz: (a) the part is a non-magnetic material (rela-
tive magnetic permeability of the part equals 1); and (b) the part is a ferromagnetic
material (relative magnetic permeability of the part equals 90).
Figure 16. Static short inductor case at 60 Hz: magnetic lines for r = 1 and 90.
Figure 15 shows here again, that whatever the nature of the material—magnetic or not-, the
electric eld tends to concentrate on the surface as the frequency increases. This well-known
eect is used for instance in heat treatment applications where very high frequencies are used.
Figure 16 provides a display of magnetic eld lines for both a magnetic and non-magnetic
material. Here again, we can see the concentration of the magnetic lines when dealing with
a magnetic material.
The temperature computations are carried over the whole domain. Heat propagates from the
surface towards the rest of the workpiece and is also conducted in the air. Figures 17 and 18
show that for a given frequency, temperature distributions can be totally dierent depending
on the nature of the material. A thermo-mechanical coupled analysis has been carried out, see
864 F. BAY ET AL.
Figure 17. Temperature at two time steps for a non-magnetic part

(relative = 1): (a) t = 5 s; and (b) t = 25 s.
Figure 18. Temperature at two time steps for a magnetic part (relative = 90): (a) t = 1 s; and (b) t = 5 s.
Figure 19.As it can be expected, the plot of the nodal velocity eld shows dilatation of the
part in the heated regions and on both edges.
4.3. A moving short inductor case

This case deals with a moving short inductor—a case representative of what takes place in an
induction heat treatment process. As explained previously, the displacement of the inductor is
modeled through a continuous variation of physical properties of the mesh. Figure 20 shows
the geometry and the initial location of the inductor in the mesh. Figure 21 shows an example
Figure 19. Nodal velocities in the workpiece due to dilatation eects: (a) non-magnetic material
(relative = 1); and (b) non-magnetic material (relative = 90).
Figure 20. Moving inductor case: geometry and initial mesh.
866 F. BAY ET AL.
Figure 21. Moving inductor case: temperature isovalues at a given displacement step of the inductor.
of temperature distribution at a given time step; one can notice the inuence of the inductor
displacement on the temperature eld in the workpiece.
5. CONCLUSION
The strategy developed in this paper for coupling electromagnetic and thermo-mechanical
computations has proven to be very eective in terms of precision on results and computa-
tional eciency. It has enabled us to model induction heating processes for various ranges
of frequencies—low frequency applications for initial heating as well as higher frequencies
for heat treatment applications. The solving procedure for the electromagnetic problem has
enabled us to consider and model precisely the case of non-linear magnetic materials. There
is an excellent agreement between numerical and experimental results.
Moreover, the strategy developed has two other main advantages:
• it is well suited for integration in global optimization procedures,
• it enables the use of ecient parallel computation techniques based on a global domain
partitioning procedure.
Development of a parallel version to deal with larger cases has been carried out. This strategy
has also been used with an optimal control approach in order to ensure that some industrial
goals are reached (reaching a prescribed homogeneous temperature, or a certain level of
hardness, etc). Further developments will deal with the extension of this strategy to the
modelling of three-dimensional problems.
ACKNOWLEDGEMENT
The authors would like to thank the European Community for the nancial support provided for this
work (Esprit project No. 28283).
REFERENCES
1. Davies EJ. Conduction and Induction Heating. Peter Peregrinus Ltd.: London, 1990.
2. Favennec Y, Labbe V, Bay F. Induction heating processes optimisation—a general optimal control approach.
Journal of Computational Physics 2001, submitted.
3. Chaboudez C, Clain S, Glardon R, Mari D, Rappaz J, Swierkosz M. Numerical modelling in induction heating
for axisymmetric geometries. IEEE Transactions on Magnetics 1997; 33(1):739–745.
4. Rappaz J, Swierkosz M. Mathematical modelling and simulation of induction heating processes. Appl. Math.
and Comp. Sci. 1996; 6(2):207–221.
5. Salon SJ, Peng JP. A Hybrid nite element-boundary element formulation of Poisson’s equation for axisymmetric
vector potential problems. Journal of Applied Physics 1982; 53(11):8420–8422.
6. Quateroni A, Valli A. Numerical approximation of partial dierential equations. Springer Series in Computa-
tional Mathematics, vol. 23. Springer-Verlag: Berlin, 1994.
7. Wang KF, Chandrasekar S, Yang HTY. Finite element simulation of induction treatment. Journal of Materials
Engineering and Performance 1992; 1(1).
8. Chaboudez C, Clain S, Glardon R, Rappaz J, Swierkosz M, Touzani R. Numerical modelling in induction
heating of long workpieces. IEEE Transactions on Magnetics 1994; 30(6):5028–5037.
9. Guerin C. Determination des pertes par courants de Foucault dans les cuves de transformateur. These de Docteur-
Ingenieur, INP Grenoble, 1994.
10. Robert P. Electrical and Magnetic Properties of Materials. Artech House Inc., London, 1988.
11. Bossavit A, Verite JC. The Trifou’ code: solving the 3-D eddy-currents problem by using H as state variable.
IEEE Transactions on Magnetics 1983; 19(6):2465–2470.
12. Labridis D, Dokopoulos P. Calculation of eddy current losses in nonlinear ferromagnetic materials. IEEE
Transactions on Magnetics 1989; 25(3):2665–2669.
13. Bossavit A. Chauage d’un cylindre d’acier par induction. Modeles numeriques. Note EDF/DER, code:
HI2825-02, 1978.
14. Skoczkowski TP, Kalus MF. The mathematical model of induction heating of ferromagnetic pipes. IEEE
Transactions on Magnetics 1989; 25(3):2745–2750.
15. Jufer M, Radulescu M. Inclusion of magnetic hysteresis losses and eddy current dynamic losses in the modelling
of induction heating for axisymmetric geometries. International Report LEME, Ecole Polytechnique Federale
de Lausanne, 1994.
16. Bleuvin H. Analyse par la methode des e lements nis des phenomenes magneto-thermiques. Application aux
systemes de chauage par induction. These de Docteut-Ingenieur, INPG, 1984.
17. Labbe V. Modelisation numerique du chauage par induction. Approche e lements nis et calcul parallele. These
de l’Ecole des Mines de Paris, 2002.

18. Raviart PA, Thomas JM. Introduction a l’Analyse Numerique des Equations aux Derivees Partielles. Masson,
1983.
19. Zlamal M. Finite element for non-linear parabolic equations. RAIRO Analyse Numerique 1977; 11(1):93–107.
20. Chastel Y, Magny C, Bay F. An elastic-viscoplastic nite element model for multimaterials: formulation and
experimental validation. Engineering Computations 1998; 15(1):139–150.
View publication stats

IJNME2003 Inductiondirect

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IJNME2003 Inductiondirect

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

A numerical model for induction heating processes coupling

Article in International Journal for Numerical Methods in Engineering · October 2003

F. Bay Yann Favennec

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

A numerical model for induction heating processes coupling

F. Bay∗; † , V. Labbe, Y. Favennec and J. L. Chenot

1.1. The induction heating process

Contract=grant sponsor: European Community; contract=grant number: 28283

Figure 1. Induction heating set-up.

1.2. Potential applications of a numerical model

1.3. The various numerical models

2. THE MATHEMATICAL MODEL

2.1. The electromagnetic model

2.1.2. The magneto-quasi-static approximation. In induction heating processes, the range of

2.1.3. Axisymmetrical congurations. When dealing with axisymmetrical congurations, the

where the temperature dependency f(T ) writes [16]

Relation (26) becomes, with the temperature dependency,

2.2. The thermal model

2.3. The mechanical model

3. THE NUMERICAL METHODS

3.1. The electromagnetic problem

axi = {(r;  = 0; z) ∈ } (43)

with H 1 ( axi ) a Sobolev space:

Equation (48) thus becomes, in a Galerkin approximation:

which can be rewritten as

which is equivalent to the following linear system:

t ∗ = 1 (t − t1 ) + 2 t + 3 (t + t2 ) with 1 +  2 +  3 = 0 (55)

3.2. The thermal problem

3.3. The mechanical problem

[K me ]{u(t)} = {Bme } (63)

3.4. The coupling procedure

Figure 4. Organization of the coupling procedure.

Q em ((n + 1)T ) − Q em (nT )

4. RESULTS AND DISCUSSION

Table I. Geometric data.

4.1. A static long joint-spires inductor case

Figure 7. Long inductor case: eective electric eld isovalues.

4.2. A static short inductor case

Figure 12. Static short inductor case: geometry and mesh.

Table II. Physical parameters.

( −1 =m1 ) 3 × 106 60 × 106 0

Figure 17. Temperature at two time steps for a non-magnetic part

4.3. A moving short inductor case

Figure 20. Moving inductor case: geometry and initial mesh.

View publication stats

You might also like

axi = {(r; = 0; z) ∈ } (43)

t ∗ = 1 (t − t1 ) + 2 t + 3 (t + t2 ) with 1 + 2 + 3 = 0 (55)

Q em ((n + 1)T ) − Q em (nT )