2008:42

L ICE N T IAT E T H E S I S

Simulation of Induction
Heating in Manufacturing

Martin Fisk

Luleå University of Technology

Department of Applied Physics and Mechanical Engineering
Division of Material Mechanics
Universitetstryckeriet, Luleå
2008:42|:-1757|: -lic -- 08 ⁄42 -- 
Simulation of Induction Heating in
Manufacturing

Martin Fisk
Abstract
Induction heating has been used during the past three decades in the industry.
Because of fast heating and good reproducibility it is used in heat treatment
applications, in bulk heating processed, as well as for special applications in
the food and chemical industry. Local heat treatmend by induction on repair
welded aerospace components is one example of induction heating in manu-
facturing. Numerical simulation of induction heating enables optimisation of
the process variables. Electromagnetic-thermal modelling of induction heating
using finite element method is presented in this thesis work.
To be able to compute the temperature history within tight tolerances,
an accurate model is needed. Well controlled and documented test-cases of
electromagnetic induction heating are few in the literature. The first part
of this thesis was therefore done to validate the induction heating model. The
work was then extended to include simulation of a manufacturing process chain.

iii
Acknowledgment
This work has been carried out at the division of Material Mechanics at Luleå
University of Technology from May 2006 to December 2008. The financial
support was provided by the National Aviation Research Programme (NFFP1 )
in cooperation with Volvo Aero and University West.
First and foremost I would like to express my sincere gratitude to my supervisor,
Professor Lars-Erik Lindgren for his enthusiasm and support during the course
of this work. I would also like to express my gratitude to my co-supervisor, Pro-
fessor Hans O Åkerstedt for sharing his deep knowledge in electromagnetism.
Many thanks to all my colleges at the University for making the work more
inspiring, especially those working in my division, Material Mechanics.
Finally I would like to thank Karin for support and putting up for me. Thank
you for being there for me.

1 Nationella flygforskningsprogrammet

v
Dissertation
The thesis consists of an introduction part followed by two appended papers.
The first sections are intended to introduce the reader to the basics of induction
heating and its opportunities. The papers are thereafter briefly commented as
the full description are available in the appended papers.

Paper A
Validation of induction heating model for Inconel 718 components.
M. Fisk.
Submitted to Journal of Manufacturing Science and Engineering.
Paper B
Simulations and measurements of combined induction heating and extrusion
processes.
S. Hansson, M. Fisk
Submitted to Finite Elements in Analysis and Design.

vii
Contents
1 Introduction 1
1.1 Aim and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The induction heating process . . . . . . . . . . . . . . . . . . . 1

2 Magnetic material properties 2

2.1 Paramagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Cooperative magnetism . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 Ferromagnetism . . . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Antiferromagnetism and ferrimagnetism . . . . . . . . . 3
2.3 Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4.1 Temperature influence . . . . . . . . . . . . . . . . . . . 5
2.4.2 The Frölich representation . . . . . . . . . . . . . . . . . 7

3 Constitutive relations 9
3.1 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . 10
3.3 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4 The Poynting vector . . . . . . . . . . . . . . . . . . . . . . . . 12
3.5 The diffusion equation . . . . . . . . . . . . . . . . . . . . . . . 13
3.6 Skin depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.7 Linearisation of the permeability . . . . . . . . . . . . . . . . . 15
3.8 Units in electromagnetic fields . . . . . . . . . . . . . . . . . . . 16

4 Heat transfer 18
4.1 Heat transfer modes . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1.1 Heat conduction . . . . . . . . . . . . . . . . . . . . . . 18
4.1.2 Convection . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1.3 Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 The heat conduction equation . . . . . . . . . . . . . . . . . . . 20

5 FE-solutions and summary of papers 21

5.1 Paper A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2 Paper B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6 Discussions and future work 25

ix
1 INTRODUCTION

1 Introduction
Welding changes the microstructure of the material thereby deteriorating the
mechanical properties of the component. Today, local heat treatment by elec-
tromagnetic induction is used at Volvo Aero Corporation on repair welded parts
to restore the microstructure. For large components, local heat treatment is
favoured over global, due to economic and practical reasons.
Induction heating is a convenient method useful for local heat treatment.
The current work is aimed to enable modelling of this process and will in
turn be useful when defining individual parameters affecting the temperature
distribution. The temperature distribution in the workpiece depends primarily
on parameters like, coil position, electrical current through the induction coil,
frequency of the current, thermal properties of the workpiece etc.

1.1 Aim and scope

This work is part of a project aimed to perform finite element simulation of
manufacturing chain. Using the finite element method, it is possible to compute
the temperature distribution in a component which is subjected to induction
heating. It is then possible to optimise the various process parameters. The
goal of this work is to develop a validated model for induction heating analysis.
This can be phrased as the following research question;

How should the induction heating process be modelled?

The approach taken to develop this model is to start with a simple geometry
having all the essential features. The model is later extended to more complex
industrial cases where the temperature is mapped from the induction heating
model to subsequent models in a manufacturing chain. A commercial FE-
software extended with user sub routines has been used for the papers that are
appended in this thesis.

1.2 The induction heating process

Induction heating is a non-contact heating process used when heating an elec-
trically conducting material by electromagnetic induction. In a thin layer of the
workpiece, usually called the skin depth, the alternating electromagnetic field
generates a current which is called eddy current. The eddy current generates
heat, due to ohmic power losses and is the main heat source in an induction
heating process. In a ferromagnetic material the hysteresis effect does also
contribute to heat generation, but in a much smaller amount.
The skin depth is defined as the depth at which the magnitude of the cur-
rents drops e−1 from its surface. Its depth depends on the electric conductivity
of the material, the frequency of the applied electromagnetic field and the mag-
netic properties of the workpiece itself.

1
 2 MAGNETIC MATERIAL PROPERTIES

2 Magnetic material properties

Each electron in an atom has a net magnetic moment originating from two
sources, orbiting and spin. The first source is related to the orbiting motion
of the electron around its nucleus. The orbiting motion of the electron give
rise to a current loop, generating a very small magnetic field with its moment
through its axis of rotation. The second magnetic moment is caused by the
electron’s spin and can either be in positive (up) direction or in negative (down)
direction. Thus each electron in an atom has a small permanent orbital and
spin magnetic moment.
From the simplified Bohr atomic model, the electrons of an atom are located
in shells, denoted as K, L, M, and subshells (s, p, d), which are orbiting around
a nucleus. Each subshell has different energy states, consisting of electron pairs,
with opposite spins cancelling each others magnetic moment. This cancellation
also hold for the orbiting electron. Therefor, for atoms with incomplete electron
shells or sub shells, a net magnetic moments will occur.
When an external magnetic field, H, is applied on the body, the magnetic
moment in the material tends to become aligned with this field, resulting in
a magnetization M of the solid. Assuming a linear relation between the mag-
netization vector and the magnetic field makes it possible to write [12, 14, 16]

M = χm H + M0 (1)
where χm is a dimensionless quantity called the magnetic susceptibility. It
is more or less a measure of how sensitive (or susceptible) a material is to a
magnetic field. M0 is a fixed vector that bears no functional relationship to H
and is referred to the state of permanent magnetization.
Let the magnetic flux density, B, represent the internal field strength within
a substance subjected to a external magnetic field H. Per definition, the mag-
netic flux density and the magnetic field are related to each other according to

B = µ0 (H + M) (2)
Combining Eq. (1) and (2) gives
B = µ0 (1 + χm )H + µ0 M0 = µ0 µr H + µ0 M0 = µH + µ0 M0 (3)
The quantity µ is known as the permeability of the material and µr is
called the relative permeability, which is the ratio of the permeability of a
given material to that of free space, µ0 . The relative permeability of a material
is therefore a measure of how easily a material can be magnetized, or how much
better it conducts magnetic flux than in free space.

2.1 Paramagnetism
For some solid materials, the atoms have a permanent dipole moment due to
incomplete cancellation of the electron spin and/or orbital magnetic moments.

2
2 MAGNETIC MATERIAL PROPERTIES 2.2 Cooperative magnetism

However, due to the thermal agitation, the atoms are randomly oriented and
result in zero net magnetic moment of the solid. Applying an external field on
such material the atoms start to align, resulting in a small positive susceptibility
usually in the order of 10−5 to 10−2 . Aluminum, titanium and tungsten are
examples of paramagnetic materials. Due their low relative permeability, µr ,
paramagnetic materials are all treated as nonmagnetic. See Fig. 1 for an
illustration of the orientation of magnetic moments in a paramagnetic material.

Figure 1: Schematic illustration of the alignment of magnetic moments for

different types of materials.

2.2 Cooperative magnetism

There is a number of materials with atoms possessing permanent magnetic
moment, as in paramagnetic materials, but with the difference that the atoms
cooperate with each other. Ferromagnetism, antiferromagnetism and ferrimag-
netism are members of this family. Figure 1 shows how their magnetic moments
are aligned to each other.

2.2.1 Ferromagnetism
Ferromagnetic materials have a permanent magnetization even in the absence
of a magnetic field, and is due to the same net atomic magnetic moment as
in paramagnetism i.e. electron spins that do not cancel. The orbital magnetic
moment has also a contribution but is small compared to the spin moment.
The difference with respect to a paramagnetic material is that the atoms are
not randomly oriented; they are interacting with each other in domains. See
section 2.3. This results in a magnetic susceptibility which is in several orders
of magnitude larger than a paramagnetic material obsess. Susceptibility values
as high as 106 are possible for ferromagnetic materials where cobalt, nickel and
iron (BCC structure) are examples of ferromagnetic materials. Annealed aged
Inconel 718 is paramagnetic at room temperature.

2.2.2 Antiferromagnetism and ferrimagnetism

The magnetic moment of an atom in a antiferromagnetic material are antipar-
allel to the neighboring atoms, resulting in zero net magnetization. Nickel

3
2.3 Domains 2 MAGNETIC MATERIAL PROPERTIES

oxide (N iO), manganese oxide (M nO) and iron manganese alloy (F eM n) are
examples of this material group [16]. A material that possess such magnetic
properties react as if they where paramagnetic.
Ferrimagnetism is a combination of ferromagnetism and antiferromagnetism.
However, in ferrimagnetic materials the opposing moments are unequal in
strength, resulting in a net magnetization. This phenomena can occur when
the sublattices consist of different materials or ions, such as Fe2+ and Fe3+ in
Fe3 O4 [16].

2.3 Domains
A magnetic domain is a region where the individual magnetic moments of the
atoms are aligned with each another. A ferromagnetic or ferrimagnetic material
is composed of several domains, individually changing their alignment. In a
polycrystalline specimen the domains do not correspond with the grain in the
material as each grain can consist of more than a single domain. The magnitude
of the magnetization, M, for the entire solid is therefore the vector sum of the
magnetization for all the domains.
In a permanent magnet, such as a magnet that hold notes on a refrigerator
door, the domains stays aligned and the vector sum is non-zero even in the
absence of a magnetic field. Materials that obsess this possibility is called hard
magnetic material. On the other hand, soft magnetic materials are materials
that lose their memory of previous magnetization and can therefore not be a
permanent magnetic.

2.4 Hysteresis
When a ferromagnetic or ferrimagnetic material is exposed to an externally
applied magnetic field H, the relationship to the magnetic flux density B may
not be linear as in Eq. (3). A typical hysteresis curve showing a non-linear
relation between B and H and can be seen in Fig. 2.
Initially, domains in the unmagnetized specimen are in different directions.
When an external magnetic field are applied on the specimen, the domains start
to line up in the same direction as the applied magnetic field. This results in
the magnetization of the specimen. See Fig. 2a line 1. This orientation process
will continue until the saturation point Bs is reached and all domains are lined
up. After this point there is a linear relation between changes in the magnetic
flux and the magnetic field.
When the magnetic field is reduced, it will not follow the initial curve but
lags behind, i.e. line 2 in Fig. 2a. This phenomena is called the hysteresis.
Whenever the magnetic field is zero, B is not reduced to zero but to Br . This
point is called the remanence, or remanent flux density; the material remains
magnetized in the absence of an external field.

4
2 MAGNETIC MATERIAL PROPERTIES 2.4 Hysteresis

(a) A virgin hysteresis curve (b) A full hysteresis curve

Figure 2: When a specimen initially is magnetized, it follows the virgin curve

denoted 1 in Fig. a). If the magnetic field then is reduced it follows the upper
curve, denoted 2, and a residual magnetization Br remains if the magnetic
field is reduced to zero. Fig. b) shows a full magnetization cycle where Hc
is the coercive force giving how strong the opposite magnetic field must be to
demagnetizes a material.

To reduce the magnetic flux in the specimen to zero, a magnetic field of the
magnitude Hc in the opposite direction (to the original one) must be applied.
See Fig. 2b. Hc is called the coercive force.
The size and shape of the hysteresis curve is of practical importance. The
area within a loop represent is the energy loss per magnetization cycle which
appear as heat that is generated within the body. This means that a hard ferro-
magnetic material can have a heat effect due to hysteresis losses. Nevertheless,
Rudnev et. al. [11] states that in a great majority of induction heating appli-
cations, the heat effect does not exceed 7 % compared to the heat generated by
the eddy current.

2.4.1 Temperature influence

When the temperature increases in a ferromagnetic or ferrimagnetic material,
the ability to magnetize decreases. This is due to the increase in magnitude
of the thermal vibrations of the atoms, which tend to randomize any magnetic
moment resulting in decrease of the magnetization.
The saturation point Bs has its maximum at 0 K and decreases with increas-
ing temperature. For each material there is a temperature where the ability to

5
2.4 Hysteresis 2 MAGNETIC MATERIAL PROPERTIES

magnetize abruptly drops, and the magnetic characteristics vanish. It becomes

paramagnetic. This temperature is called the Curie temperature, Tc 2 , for fer-
romagnetic and ferrimagnetic materials. For antiferromagnetic materials this
temperature is called the Néel3 temperature.
The Curie temperature varies from different materials; for example, for iron,
cobalt and nickel Tc are 768, 1120 and 335 ◦ C respectively. The phenomena
can be seen as a peak in the specific heat capacity curve of the material itself.
The peak is at approximately 500 ◦ C for SAF 2507, one of the materials in
paper B, and is therefore its Curie temperature.
In Fig. 3 the hysteresis curves for SAF 2507 at different temperatures are
shown. It can be seen that the curve above the Curie temperature is linear
and have the same slope as µ0 over the whole range of applied magnetic field.
This is also true when the magnetic flux has reached its saturation point and
the only contribution is from the linear relation between the magnetic flux and
field, i.e. µr = 1. SAF 2507 is a soft magnetic material, since the residual flux,
Br , is close to zero.

1.5

0.5
B [Vs/m ]

20
2

0 100
200
300
400
−0.5
450
480
490
−1
500
520
600
−1.5
−5 −4 −3 −2 −1 0 1 2 3 4 5
H [A/m] 5
x 10

Figure 3: The hysteresis curve for SAF 2507.

2 After Pierre Curie.

3 After Louis Néel

6
2 MAGNETIC MATERIAL PROPERTIES 2.4 Hysteresis

2.4.2 The Frölich representation

There are a number of equations that mathematically can describe the hystere-
sis curve. One example is the Frölich representation
H
B= (4)
α + β||H||

which is a good compromise between accuracy and simplicity [13, 9]. Knowing
how the variables α and β changes with the temperature, it is possible to
calculate the hysteresis curve for different temperatures. In Fig. 4 the variation
of α and β versus temperature for SAF 2507 is shown. When the material
becomes paramagnetic α tends to 1/µ0 = 1/(4π · 10−7 ) and β tends to zero.

5
x 10
0.8 8

0.7 7

0.6 6

0.5 5

0.4 4
α
β

0.3 3

0.2 2

0.1 1

0 0
0 100 200 300 400 500 600
Temperature [°C]

Figure 4: The variation of α and β in Eq. (4) versus temperature for SAF
2507.

The Frölich curves using the calibrated α and β values are compared with
the measured in Fig. 5 for SAF 2507. The used values for α and β give a good
agreement. Since there is a state of permanent magnetization in the material,
an initial magnetic flux B0 is added for the temperature of interest to the
Frölich equation.

7
2.4 Hysteresis 2 MAGNETIC MATERIAL PROPERTIES

1.4
20
400
1.2
450
480
1 490
600

0.8
B [Vs/m ]
2

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
H [A/m] 5
x 10

Figure 5: Calculated (lines) and measured (lines with stars) hysteresis curves
for SAF 2507 at different temperatures. The legend gives the temperature in
Celsius.

8
3 CONSTITUTIVE RELATIONS

3 Constitutive relations and basic concepts

The basic equations and their simplification, used in the finite element code,
are summarised below.

3.1 Maxwell’s equations

Maxwell’s equations are a set of equations that describe the electric and mag-
netic field and relate them to their sources, charge density and the vector
current density. The set consists of Faraday’s law and Ampère’s circuital law
with Maxwell’s extension. The extension term is also called the displacement
current. The equations are written as
∂B
∇×E=− (5)
∂t

∂D
∇×H=J+ (6)
∂t
They consist of four different field variables; the electric field intensity E, the
magnetic flux density B, the magnetic field intensity H and the electric flux
density D.
The differential equation
∂ρ
∇·J=− (7)
∂t
expresses the conservation of charges at any point. It is frequently referred to
as the equation of continuity where ρ is the free volume charge density and J
is the current density [8, 14].
Two further conditions can be deduced directly from Maxwell’s equations.
The divergence of Eq. (5) leads to

∂B ∂
∇· = ∇·B=0 (8)
∂t ∂t
because the divergence of the curl of any vector field is identically zero. It
follows from Eq. (8) that at every point in the field the divergence is constant.
If the field sometime in its past history has vanished, the constant must be zero
and the magnetic flux becomes solenoidal, i.e. [8, 14]

∇·B=0 (9)

which sometime is called Gauss’s law for magnetism.

Similarly, the divergence of Eq. (6) leads to

∂
∇·J+ ∇·D=0 (10)
∂t

9
3.2 Constitutive relations 3 CONSTITUTIVE RELATIONS

and if it is assume that the field sometime in its past history has vanished
∇·D=ρ (11)
which is known as Gauss’s law. Equation (9) and (11) are frequently included
as a part of Maxwell’s equations.
Among these five equations that are included in Maxwell’s equations, only
three of them are independent. Either Eq. (5) - (7), or Eq. (5), (6) and (11)
can be chosen as independent. The other two equations can be derived from
these three, and are therefore called auxiliary or dependent equations [7].

3.2 Constitutive relations

The relation between the magnetic flux and field intensity was given in Eq.
(3). A similar relation can be formulated between the electric flux and field
intensity. Therefore,
D = ε0 (1 + χe )E = ε0 εr E = εE (12)
and (Eq. (3) again)
B = µ0 (1 + χm )H = µ0 µr H = µH (13)
where ε is the permittivity of the dielectric and χe is the electric susceptibility
of the material respectively. In a electric conductive material, the permittivity
are usually set to one [12]. The constant values of permeability and permittivity
of free space4 are known; µ0 = 4π · 10−7 Vs/Am and ε0 ≈ 8.854 · 10−12 As/Vm
and are related to the speed of light in free space
1
c0 = √ (14)
µ0 ε0
There is also a relation between the current density and the electric field
density such that
J = σE (15)
where σ it the conductivity of the material. It is the continuum form of Ohm’s
law.
If the medium is homogeneous then Eq. (11) can be written as
ρ
∇·E= (16)
ε
Combining the equation of continuity, Eq. (7), and Ohm’s law, Eq. (15), we
can write
∂ρ
∇ · σE + =0 (17)
∂t
4 A concept of electromagnetic theory, corresponding to a theoretically ”perfect” vacuum,

and sometimes referred to as the vacuum of free space.

10
3 CONSTITUTIVE RELATIONS 3.3 Boundary conditions

and with Eq. (16)

∂ρ ρ
+σ =0 (18)
∂t ε
The charge density of any time is therefore
ρ = ρ0 e−τ t (19)
where τ = ε/σ is the relaxation time of the medium and ρ0 is equal to the
charge density at time t = 0. The initial charge distribution will therefore
decay exponential with a time equal to the relaxation time. A typical con-
ductivity for a metal is of the order 106 Siemens/m with a permittivity ε of
10−11 As/Vm. Consequently τ is of the order 10−17 s, resulting in that the
charges will vanish from an interior point of a metal body and appear at the
surface almost immediately. Even for such a poor conductor as distilled water
the relaxation time is not greater than 10−6 s [14]. Even at high frequencies no
charge will be accumulated inside the conductor and Eq. (7) can therefore be
approximated as
∇·J=0 (20)
This allows an important simplification of Ampère’s circuital law with Maxwell’s
extension. Assume a time harmonic field with an angular frequency of ω. Equa-
tion (6) will then have a magnitude of the current and displacement current
as J and ωǫ/σ, i.e. 1 and ωτ . Therefore, a reasonable assumption is that the
displacement current is negligible in a conductor even for high frequencies and
Eq. (6) can be written as [6],
∇×H=J (21)
This is consistent with Eq. (20), since the divergence of a curl is zero. This
important simplification makes it possible to derive the diffusion equation for
electromagnetic fields. See section 3.5 for this derivation and further details.

3.3 Boundary conditions

Depending on the electric and magnetic properties, the electromagnetic fields
can be discontinuous or continuous on each side of a common interface between
two different materials. A derivation on the boundary condition can be found
in for example Sadiku [12] or Stratton [14].
At an interface between two mediums the field must satisfy the following
conditions
n̂ × (E1 − E2 ) = 0 (22a)
n̂ · (D1 − D2 ) = ρ (22b)
n̂ × (H1 − H2 ) = Js (22c)
n̂ · (B1 − B2 ) = 0 (22d)

11
3.4 The Poynting vector 3 CONSTITUTIVE RELATIONS

where the boundary outward unit normal, n̂, is directed from medium 2 to-
wards medium 1. The boundary conditions can also be formulated in words
as:

ˆ The electric field tangential components are continuous across the inter-
face of medium 1 and 2.
ˆ The electric flux normal component is discontinuous across the interfacing
mediums with a magnitude of ρ. Note that for a electrically conductive
medium, ρ = 0.
ˆ The tangential components of the magnetic field strength are discontin-
uous across the two mediums with a magnitude of Js . In the case of a
zero surface current, as when the medium has a finite conductivity, the
tangential component is continuous [8, 14].
ˆ The magnetic flux density normal component is continuous across the
interface of medium 1 and 2.

3.4 The Poynting vector

Poynting’s theorem is a relation between the rate of change of energy stored
in the fields and the energy flow [14]. Multiply Faraday’s law, Eq. (5), and
Ampère’s law ignoring the displacement current, Eq. (6), with the electric and
magnetic field respectively, and subtract them from each other gives
∂B
I Z Z
(E × H) · n dA = − H · dv − σE 2 dv (23)
s v ∂t v

The quantity E × H is known as the Poynting vector P in watts per square

meter. Poynting’s theorem states that the sum of the ohmic loss and the power
absorbed by the magnetic field in the volume is equal the power input of the
body.
It is common to represent a harmonic variation in time as a complex function
of coordinates, multiplied by the factor eiωt . The complex Poynting vector can
therefore be written as [14, 13]
1
P= (E × H∗ ) (24)
2
where (H∗ ) is the complex conjugate of the magnetic field. Equation (23) has
then the complex form [13]
1 1
I Z Z
P · dS = −j2ω µ0 µr |H|2 dv − σE 2 dv (25)
s v 4 v 2
The real part of Eq. (25) determines the energy dissipated in heat (ohmic
losses) in the volume v and the imaginary part is equal to 2ω times the mean

12
3 CONSTITUTIVE RELATIONS 3.5 The diffusion equation

values of magnetic energy. The current from the real part is known as the eddy
current5 and is the main heat source in induction heating.

3.5 The diffusion equation

To describe a vector field completely, both the divergence and curl have to be
uniquely defined. Since the magnetic flux density satisfies a zero divergence
condition, it can be represented as the curl of another vector such that

B=∇×A (26)

where A is the magnetic vector potential. From Maxwell’s equations it is known

that
∂B ∂ ∂A
∇×E=− = − (∇ × A) = −∇ × (27)
∂t ∂t ∂t
or
∂A
∇ × (E + )=0 (28)
∂t
Because ∇ × (∇ϕ) = 0, one solution to Eq. (28) is
∂A
E=− − ∇ϕ (29)
∂t
The negative sign in front of ∇ϕ is per definition [12]. Multiply the electric
field with σ
∂A ∂A
J = σE = −σ − σ∇ϕ = −σ + Js (30)
∂t ∂t
where Js is the source current density in the induction coil [11]. It follows from
Eq. (21) that
1
∇× B=J (31)
µ
Substitute Eq. (26) and (30) we obtain
∂A
∇ × ∇ × A = −µσ + µJs (32)
∂t
for a homogeneous, isotropic medium independent of field intensity. Expand
and impose on A the Coulomb gauge condition ∇ · A = 0 [7],
∂A 1
σ − ∇2 A = Js (33)
∂t µ
5 Discovered by the French physicist Léon Foucault. Localised areas of turbulent water

known as eddies give rise to the eddy current.

13
3.6 Skin depth 3 CONSTITUTIVE RELATIONS

which is called the diffusion equation. The equation can also be expressed in
the same form but for different vectors; instead of A we write J, E, H, B [6].
In the case of the magnetic vector potential, it can be related to any physi-
cally observable electromagnetic induction phenomenon such as eddy current,
induced voltage, coil impedance, flaw impedance, coil inductance etc. [5]. If
the excitation current is assumed to be sinusoidal, and the eddy current as well,
a time-harmonic electromagnetic field can be introduced in Eq. (33) and we
gain

1 2
iωσA − ∇ A = Js (34)
µ

It should be noted, that in a ferromagnetic material the harmonic approxi-

mation is not valid anymore. This is due to its the time dependent relative
permeability µr , which make the eddy current non sinusoidal [9]. See section
3.7 for how to approximate the behaviour of ferromagnetic materials, using this
harmonic approximation.

3.6 Skin depth

Maxwell’s equations can be expressed as two second order differential equations
in E and H alone. Assume that the material is linear, isotropic, homogeneous
and charge free (ρ = 0). Start with taking curl on equation (5) and (6)
 
∂(∇ × H) ∂ ∂
∇ × ∇ × E = −µ = −µ σ+ε E (35)
∂t ∂t ∂t
   
∂ ∂ ∂
∇×∇×H= σ+ε ∇ × E = −µ σ+ε H (36)
∂t ∂t ∂t
Furthermore, harmonic solutions are assumed and the relation

∇ × ∇ × a = ∇ (∇ · a) − ∇2 a (37)

is used together with the properties ∇ · H = 0 and ∇ · E = 0. These steps lead

to two second order homogeneous wave equations;

∇2 E − γ 2 E = 0 (38a)
2 2
∇ H−γ H=0 (38b)

where γ 2 = iωµ(σ + iωε) and γ is called the propagation constant of the

medium. Let the propagation constant be defined as
s  
iσ
γ = α + iβ = iω µε 1 − (39)
ωε

14
3 CONSTITUTIVE RELATIONS 3.7 Linearisation of the permeability

The solution of equation (38b) is, if it represents a uniform electric wave prop-
agating in the positive z-direction [12],

Ex (z, t) = Re[E0 e−αz e−iβz eiωt ] = E0 e−αz cos(ωt − βz) (40)

Consequently, if a electric or magnetic wave travels in a conductive medium, its

amplitude starts to attenuate exponentially by a factor e−αz . The factor α is
therefore called the attenuation constant, with the unit Np/m (Neper/meter).
A good conductor is per definition if σ >> εω why Eq. (39) can be simplified
to6
√√ 1 + ip p
γ= i µωσ = √ µ2πf σ = (1 + i) πf µσ (41)
2

The skin depth is defined as the distance for which the amplitude of a plane
wave decreases a factor e−1 = 0.368. Thus it becomes
1 1
δ= =√ (42)
α πf µσ

Therefore, when the electromagnetic waves reach the surface in a conducting

medium, a current will start to flow within the surface and decreasing expo-
nentially towards the interior of the material. The total current in a conductor
is given by [13],
0
Js Js δ
Z
π
I= J dy = = √ e−i 4 (43)
−∞ α 2

which is equal to the r.m.s.7 value of the surface current density flowing uni-
formly in a layer of δ lagging the coil current (and the magnetic field) by 45◦ .
However, it is important to remember that the electromagnetic field exists
below the skin dept.

3.7 Linearisation of the permeability

In nonlinear, like SAF 2507, materials Eq. (34) is not valid anymore. This
is due to the time dependent relative permeability µr , which makes the eddy
current non sinusoidal. There are two possibilities to overcome this; introduce
a complex permeability, or introduce a constant permeability. Labridis and
Dokopoulos [9] shows that it is possible to approximate the nonlinear material
with a fictive linear material that has a constant unknown relative permeability,
µfr . This permeability is related to the non-linear B − H characteristic curve
at every point. See Fig. 3. The condition that the linear fictitious material
π π π
6 In polar form: i = ei 2 = ei 4 ei 4 = ( 1+i
√ )2
2
7 Root mean square

15
3.8 Units in electromagnetic fields 3 CONSTITUTIVE RELATIONS

should have the same average heat density losses as the nonlinear material at
every point can be used to derive the fictive relative permeability as
w1i + w2i
µfri = f 2
(44)
µ0 ∗ (Hmi )

where w1i and w2i is the magnetic co-energy density of the material. w1i is
related to the actual B − H curve and w2i is related to the average value of the
f
slope dB/dH during a quarter of a period T . Hmi is the maximum magnetic
field intensity for a sinusoidal function in the fictitious material. The magnetic
co-energy density can be calculated as
f f
Hmi α α + βHmi
w1i = − 2 ln (45)
β β α
f 2
1 (Hmi )
w2i = f
(46)
2 α + βHmi

giving an over- and underestimation of heat losses respectively.

Since both the the relative permeability of the fictitious material, µfr , and
f
the magnetic field intensity, Hm , are unknown, an iterative process is necessary.
f
In the first step µr is set to the initial slope of the Frölich curve, i.e.

1
µfr = (47)
µ0 α

In all the next steps µfr will be calculated from Eq. (44).
The formulation is based on the Frölich representation of the nonlinear
B − H curve. See Eq. (4) and section 2.4.2 for further information.

3.8 Units in electromagnetic fields

A new system of units was after Maxwell’s discovery proposed by Giorgi8 1901
to the AEI9 , since it was clear that electric measurements could not be ex-
plained in terms of the three fundamental units of length, mass and time. The
new system had the fundamental units metre, the kilogram and the second
(m.k.s) and as a fourth unit any electrical quantity belonging to the practical
system such as the coulomb, the ampere, or the ohm. From the field equations,
it is then possible to deduce the units of every field quantity in terms of these
four fundamental units [14].
In a plenary session in 1935, the IEC10 adopted unanimously the m.k.s
system, although the fourth unit was not chosen. Giorgi himself recommended
the ohm to be the fourth unit, but it was not followed strictly. Fore example, in
8 Giovanni Giorgi, an electrical engineer who invented the Giorgi system of measurement.
9 Associazione Elettrotecnica Italiana
10 International Electrotechnical Commission

16
3 CONSTITUTIVE RELATIONS 3.8 Units in electromagnetic fields

the book written by Stratton in 1941 [14], the metre-kilogram-second-coulomb

system is used.
In 1960, at the 11th General Conference of Weights and Measures, the
International System (SI) was adopted. It is bases on seven fundamental units;
the metre, the kilogram, the second, the ampere, the kelvin, the mole and the
candela. In Table 1 the SI units and their related bas units are shown for the
most common derived quantities in electromagnetism.

Table 1: Basic quantities in electromagnetism.

Derived quantity Symbol SI unit SI base unit
Electric field intensity E V/m m kg s−3 A−1
Electric flux density D C/m2 m−2 s A
Magnetic field intensity H A/m m−1 A
Magnetic flux density B T m2 kg s−2 A−1
Current Density J A/m2 m−2 A
Magnetic flux φ Wb m2 kg s−2 A−1
Permeability µ H/m m kg s−2 A−2
Permittivity ε F/m m−3 kg−1 s4 A2
Volume charge density ρ C/m3 m−3 s A
Conductivity σ S/m m−1 kg−1 s3 A2
Resistivity ρc Ωm m3 kg s−3 A−2

17
 4 HEAT TRANSFER

4 Heat transfer
4.1 Heat transfer modes
There are three modes of heat transfer, conduction, convection and radiation.
Heat transfer by conduction occurs inside the solid and the other modes are
active in the thermal boundary conditions.

4.1.1 Heat conduction

The empirical constitutive law for heat conduction is called the “Fourier law
of heat conduction” and is written as

qcond = −k∇T (48)

where k is the thermal conductivity, T is the temperature and qcond is the heat
flux by conduction. The conductivity is usually a function of temperature.
Typically is a uniform temperature distribution along the workpiece easier to
obtain if the thermal conductivity is high.

4.1.2 Convection

Heat transferred by convection from the surface of the workpiece to the ambient
fluid or gas can be expressed as

qconv = h(Ts − T∞ )α (49)

where h is the convection surface heat transfer coefficient11 , Ts is the surface

temperature, T∞ is the ambient temperature and qconv is the heat flux denity
by convection.
The value of the film coefficient depends primarily on the thermal proper-
ties of the surrounding gas fluid, its viscosity and the velocity of the gas. In
many induction heating applications, the workpiece is moving at high speed
(e.g., heating of rotating disk, wire heating, etc.). Then the convection can be
considered as forced. This convection can be equal or exceed heat losses due to
heat radiation for low-temperature induction heat treatment. For a workpiece
in open air a good approximation is [4, 11]

qconv = 1.54(Ts − T∞ )1.33 (50)

which is compared with heat losses due to radiation in Fig. 6.

11 Also known as the film coefficient.

18
4 HEAT TRANSFER 4.1 Heat transfer modes

4.1.3 Radiation

Heat losses transferred from the hot workpiece due to electromagnetic radiation
is called thermal or heat radiation. The equation that describe this can be
expressed as

qrad = σε[(Ts )4 − (T∞ )4 ] (51)

where σ is the Stefan-Boltzman constant (σ = 5.67 · 10−8 W/m2 K 4 ) and ε is

the emissivity of the surface. The emissivity is defined as the ratio of the heat
emitted by the surface to the heat emitted by a black body. Ts is the surface
temperature and T∞ is the ambient temperature given in Kelvin.
Since the radiation loss is proportional to the fourth power of temperature
it exceeds the forced convection loss at elevated temperature, especially if the
emissivity is larger than 0.3 as is seen in Fig. 6. Note that the value of the
emissivity can vary for the same material. It varies from 0.03 for a for a polished
copper surface to 0.70 for a heavily oxidised surface [4].

140

120

100
ε = 0.8
Heat losses [kW/m ]
2

80
qradiation

60

40
ε = 0.1
q
convection

20

0
0 100 200 300 400 500 600 700 800 900 1000 1100
Temperature [°C]

Figure 6: The heat losses due to convection and radiation transfer modes. The
full lines represent the radiation heat loss due to an emissivity varying from
0.1 to 0.8 with a 0.1 increase for each line. The dashed line represents the
convection heat loss according to Eq. (50).

19
4.2 The heat conduction equation 4 HEAT TRANSFER

4.2 The heat conduction equation

The temperature distribution in a medium is governed by the heat transfer
equation [10]
∂T
ρc − ∇ · (k∇T ) = Q (52)
∂t
where ρ is the density, c is the specific heat capacity, k is the thermal conductiv-
ity and Q the energy generated in the material per unit volume and time. The
heat transfer equation determines the temperature distribution in a medium as
a function of space and time. Once the temperature distribution is known, the
heat flux within the body or by its surface can be calculated from Eq. (48).
The solution of the heat conduction equation requires initial and boundary
conditions. The boundary condition can either be prescribed12 , where the
temperature is known on the boundary, or/and as an imposed flux13 using
the radiation and convection boundary. The flux condition is related on the
boundary as
∂T
qn = −k (53)
∂n
where n̂ is the outward direction normal to the surface and qn the constant
flux given as
qn = h(Ts − T∞ )α + σε[Ts4 − T∞
4
] (54)
if both convection and radiation boundary condition, as given in sections 4.1.2
and 4.1.3, are assumed.
The similarity with the diffusion equation stated in chapter 3.5 can be
noted. Suppose the temperature is varying on the surface of the material in a
sinusoidal mode. It can then be shown that the amplitude of the variation will
be smaller with increasing depth and with some time delay, i.e. a phase angle.
If the frequency of the temperature variation is increased, the attenuation and
the phase angle will increase [6, 13]. This is the same phenomena as for the
electromagnetic wave case, but with another time scale, as the electromagnetic
field has a much shorter relaxation time. See Carslaw and Jaeger chapter 2.6
for the solution of the heat transfer equation due to sinusoidal heating of the
surface [1].
Even though the diffusion equation, Eq. (34), has the same properties as
the heat conduction equation, Eq. (52), their time scales differ. A harmonic
applied current in a induction heating equipment are usually in the frequency
range of 50 Hz − 70 KHz and for wire up to 4000 KHz [11]. Therefore, the use
of Eq. (34) is possible within each time increment in the thermal analysis as
the time scale of the electromagnetic problem is much shorter. This is utilised
in the solution strategy described in section 5.
12 The Dirichlet condition
13 The Neumann condition.

20
5 FE-SOLUTIONS AND SUMMARY OF PAPERS

5 FE-solutions and summary of papers

Practical problems governed by the diffusion and the heat conduction equations
are usually impossible to solve analytically, especially when the material has
temperature dependent and non-linear properties. A common approach to
obtain solutions for such systems is to use the Finite Element Method. In this
work the FE-software MSC.Marc has been used, allowing a coupling between
the harmonic electromagnetic solution and the heat transfer equation.
In order to calculate the temperature field within a body, a harmonically
oscillating current is applied in the coil and from Eq. (34) the magnetic vec-
tor potential is calculated. The generated current within the body, the eddy
current, is then calculated from Eq. (29) as

Je = −iωσA (55)

The heat source term in the heat conduction equation, Eq. (52), can then
be computed [15, 3] as
1 1
Q= |Je |2 = Je · J∗e (56)
2σ 2σ
It should be noticed, that for all nodes where the source current is zero, ∇ϕ is
set to zero. On the outer boundary of the region, the magnetic vector potential
is selected such that it is zero along the boundary (Dirichlet condition).
From Eq. (34) it is seen that the solution of the magnetic vector potential
is a steady state solution, using the material properties from the time step tn .
This is possible since the time scale for the electromagnetic problem is much
shorter than in the thermal problem. Figure 7 shows the solution strategy
during a time step. The electromagnetic problem is firs solved using the elec-
tromagnetic properties for the temperature at the time step Tn . Thereafter,
the heat load due to the eddy current is calculated and the thermal problem is
soled giving temperatures Tn+1 at the end of the time step.

21
5.1 Paper A 5 FE-SOLUTIONS AND SUMMARY OF PAPERS

Figure 7: The coupling between the electromagnetic and thermal problems

5.1 Paper A
The study done in this paper was aimed to provide a test case for validating
the induction heating model. It is the first step in a larger project where man-
ufacturing chains will be simulated. The test case consists of two experimental
arrangements; a cylinder heated along the circumference by a coil in the mid-
dle, or at its end. See Fig. 8 for the magnetic field around the coil located in
the middle.
The published works in simulation of induction heating has been limited to
keeping the current constant in the coil during the heating stage, allowing the
temperature to vary [3, 15, 2]. However, here is the current is controlled in
order to obtain a particular temperature history at a specified location of the
workpiece.
The good agreement between simulations and measurements shows that the
model, the computational approach, as well as used material properties is valid
and can be used to study the induction heating process for Inconel 718. See
Fig. 9 for the temperature agreement when the coil was located in the middle.

22
5 FE-SOLUTIONS AND SUMMARY OF PAPERS 5.1 Paper A

Figure 8: The workpiece with coil and the magnetic field generated from the
current at 23 KHz.

800
Calculated data
7
Measured data
700
6
600
Temperature [ C]

500
o

400

5
300

200

100

0
0 100 200 300 400 500 600 700
Time [s]

Figure 9: Measured and computed temperature for the coil located in the
middle of the workpiece. The numbers refers to positions of the thermo-couple.
See Paper A for further information.

23
5.2 Paper B 5 FE-SOLUTIONS AND SUMMARY OF PAPERS

5.2 Paper B
Four steps in a manufacturing process of extruded tubes are studied. Two of
the steps are induction heating and two of them are expansion and extrusion of
stainless steel tubes respectively. Axisymmetric FE-models of these steps, in-
cluding cooling at the intermediate transports have been combined. All models
were developed in the FE-software MSC.Marc and the nodal temperatures were
transferred from one model to another using an external mapping program.
Two different materials were used in the process flow simulation. One is an
austenitic stainless steel and the other is a duplex, austenitic/ferritic, stainless
steel. Since the duplex steel is ferromagnetic below the Curie temperature,
the skin effect is more pronounced compared to the paramagnetic austenitic
material. Also the magnetic non-linearity of the material properties of the bil-
let leads to differences in the induction heating modelling. Results from the
numerical models have been validated by experiments in production presses
and induction furnaces. The result of the measure temperature and the calcu-
lated i seen in Fig. 10. Since MSC.Masc uses a harmonic approximation, the
permeability has been linearised as is described in section 3.7.

1400

1200

1000
Temperature [ C]
o

800

5 mm
600
10 mm
20 mm
400 30 mm
40 mm
64 mm
200 78 mm

0
0 100 200 300 400 500 600 700 800
Time [s]

Figure 10: The measured temperatures (dashed lines) compared with calcu-
lated temperatures (full lines). The legend gives the depth in the workpiece at
which the measurements has been carried out at.

24
6 DISCUSSIONS AND FUTURE WORK

6 Discussions and future work

How should the induction heating process be modelled? was the research ques-
tion asked in the beginning of this thesis. The goal of this work has therefore
been to develop a validated finite element model for simulation of the induction
heating process. In paper A, the temperatures as well as the current shows a
good agreement between simulations and measurements. This shows that the
model and the computation approaches as well as used material properties are
valid and can be used to study the induction heating process.
The temperature agreement for the duplex steel in paper B shows that the
approach Labridis and Dokopoulos [9] proposed, to linearize the permeability,
is a good approximation. This is important since a harmonic solution with a
constant permeability is much faster than an electromagnetic non-linear solu-
tion [3]. In this work, four steps in a manufacturing process chain are simulated.
The resulting temperatures from the electromagnetic-thermal calculation are
mapped as an initial temperature to the thermo-mechanical calculation using
an external mapping tool.
In the future work, the simulations are extended to three dimensions, where
the calculated temperature field are used in thermo-mechanical calculations.
Welding changes the microstructure of the material and thereby deteriorating
the mechanical properties of the component. The microstructure needs due
to this be restored either by local or global heat treatment. The temperature
field must then be mapped from the induction heating model to the thermo-
mechanical model every time step.

25
REFERENCES REFERENCES

References
[1] H. S. Carslaw and J. C. Jaeger. Conduction of heat in solids. Oxford
University Press, Oxford, 2. ed. edition, 1959.

[2] C. Chaboudez, S. Clain, R. Glardon, D. Mari, J. Rappaz, and

M. Swierkosz. Numerical modeling in induction heating for axisymmet-
ric geometries. Magnetics, IEEE Transactions on, 33(1):739–745, 1997.
0018-9464.

[3] S. Clain, J. Rappaz, J. Swierkosz, and R. Touzani. Numerical modeling

of induction heating for two-dimensional geometries. Mathematicl Models
and Methods in Applied Sciences, 3(6):805–822, 1993.

[4] E. J. Davies. Conduction and induction heating. Number 0-86341-174-6

in IEE Power Engeneering Series 11. Peter Peregrinus Ltd., London, UK,
1990.

[5] C. V. Dodd and W. E. Deeds. Analytical solutions to eddy-current probe-

coil problems. Journal of Applied Physics, 39(6):2829–2838, 1968.

[6] Percy Hammond. Applied electromagnetism. Pergamon Press, Oxford,

1985.

[7] Jianming Jin. The Finite Element Method in Electromagnetics, volume 2.

John Wiley & sons, inc., 2002.

[8] Gerhard Kristensson. Elektromagnetisk vågutbredning. Studentlitteratur,

1999.

[9] Dimitris Labridis and Petros Dokopoulos. Calculation of eddy current

losses in nonlinear ferromagneitc materials. IEEE Transactions on Mag-
netics, 25(3), May 1989.

[10] Roland W. Lewis, Perumal Nitiarasu, and Kankanhalli N. Seetharamu.

Fundamentals of the finite element method for heat and fluid flow. Wiley,
Chichester, 2004.

[11] Valery Rudnev, Don Loveless, Raymond Cook, and Michan Black. Hand-
book of Induction Heating. Inductoheat, Inc., 2003.

[12] Mattew N.O. Sadiku. Elements of Electromagnetics, volume 3. Oxford

University Press, Inc., 2001.

[13] Richard L. Stoll. The analysis of eddy currents. Oxford University Press,
Oxford, 1974.

[14] Julius Adams Stratton. Electromagnetic theory. IEEE Press, Piscataway,

NJ, 2007.

27
REFERENCES REFERENCES

[15] K.F. Wang, S. Chandrasekar, and Henry T.Y. Yang. Finite-element simu-
lation of induction heat treatment. Journal of Materials Engineering and
Performance, 1, 1992.

[16] Jr. Willian D. Callister. Materials science and engineering: an introduc-

tion. John Wiley & Sons, Inc., 1999.

28
Paper A
 Validation of induction heating model for Inconel
718 components

Martin Fisk
Division of Material Mechanics
Department of Applied Physics and Mechanical Engineering
Luleå University of Technology
SE-971 87 Luleå, Sweden
Email: [email protected]

Solving the coupled electromagnetic-thermal problem ing the past three decades. Due to fast heating rate and
of induction heating requires simplifications in the nu- good reproducibility, it is used in applications such as
merical algorithm, mainly due to the different time hardening, tempering and annealing [1].
scales of the fields. Temperature dependent mate- Welding changes the micro-structure on the mate-
rial properties is another uncertain issue in modelling. rial and a heat treatment is needed to restore it. Fur-
However, well controlled and documented test cases are thermore, it introduces residual stresses in the material
few in the literature and therefore the current study was which will be reduced by the heat treatment. For large
done as a first step in a larger project where manufac- components, local heat treatment by induction heating
turing chains will be simulated. The induction heating is favoured over global, due to economic and practical
process is used for local heat treatment of aerospace reasons. Temperature distribution in the workpiece de-
components. This paper describes measurements and pends on parameters like, coil position, electrical cur-
simulations of induction heating of an axisymmetric rod rent through the induction coil, frequency of the cur-
made of Inconel 718. The model was found to have a rent, thermal properties of the workpiece, as well as
good accuracy. Furthermore, an approach combining many other factors.
the finite element code with a control algorithm of the
current has been implemented and is described. This The published works in simulation of induction
algorithm makes it possible to obtain the variation of heating has been limited to keeping the current constant
current needed to create a given temperature history. in the coil during the heating stage, allowing the tem-
perature to vary [2–4]. However, in this work, current
is controlled in order to obtain a particular temperature
1 Introduction history at a specified location of the workpiece. This
The basic setup of an induction heat equipment makes it possible to do a more traditional local heat
consists of a coil with one or several windings, keep- treatment with the temperature kept constant under a
ing an electrically conducting workpiece within. An certain time for full annealing processes or ageing [1].
eddy current is generated in the workpiece as a conse- Without controlling the current, it is only possible to
quence of the alternating electromagnetic field, and is temper parts; self-tempering, where residual heat in the
produced by the current carrying coil. In a thin sur- part is used, and induction tempering, where the hard-
face layer of the workpiece (skin depth), heat is gener- ened part is reheated [1].
ated due to the Joule heating effect caused by the eddy This paper describes measurements and finite ele-
current. The skin depth depends on the frequency of ment simulations of an axisymmetric rod made of In-
the electromagnetic field and the material properties, conel 718. The computational model was validated by
resulting in changing temperature distribution for dif- two experimental arrangements. A cylinder was heated
ferent frequencies. Induction heating is a convenient along the circumference by a coil in the middle or at
non-contact heating method, used in the industry dur- its end. The first case is similar to the setup in [3, 5].
Luo and Shih [5] perform inverse thermal analysis and duced. It is related to the magnetic flux density by
did not solve the electromagnetic problem. Therefore,
their approach cannot be used to design general induc- B = ∇×A (8)
tion heating setups. Wang et. al. [3] compare their finite
element analysis with analytical solutions.
The finite element analysis was performed together Maxwell’s equations can then be written so that
with a user routine; a logic for controlling the current in
order to produce specified temperature. The obtained 1
iωσA − ∇2 A = J0 (9)
current history agreed well with the measured current. µ
The reverse calculation was also done where the mea-
sured current was applied to the model and the com- which is a complex diffusion equation. A similar equa-
puted temperature was found to agree well with the tion can also be formulated in terms of the electric field
measured. quantity [6, 7].
Additionally, simulations where performed to eval- The domain for Eq. (9) is the rod, coil and sur-
uate the sensitivity of the setup. They show that a small rounding air. Additionally boundary conditions for the
change in the current give relatively large variation in interface between solids and air as well as external
the temperature of the workpiece. Frequency variation boundary are required. Since there is no surface cur-
is also shown to affect the temperature. rent in the electrically conducting material, the tangen-
tial vector on the interface between the mediums are
continuous which gives us [8]
2 The coupled electromagnetic-thermal problem
The basic equations and their simplification, used n̂ ×(E1 − E2 ) = 0 (10)
in the finite element code, are summarised below. n̂ ×(H1 − H2 ) = 0 (11)

2.1 Electromagnetism where the boundary outward unit normal is directed

from medium 2 towards medium 1. On the outer bound-
The electromagnetic wave propagation is described
ary of the region, the magnetic vector potential is se-
by Maxwell’s system of equations
lected such that it is zero along the boundary (Dirichlet
condition).
∂B
∇×E = − (1)
∂t
∂D
∇×H = J+ (2) 2.2 Heat conduction
∂t
∇·B = 0 (3) The heat conductivity in a solid is governed by
∇ · D = ρv (4)
∂T
ρc − ∇ · (κ∇T ) = Q (12)
The constitutive relations between the previous men- ∂t
tioned field quantities in an isotropic homogeneous
medium are The formulation accounts for temperature dependent
material properties. The solution of the equation re-
B = µH (5) quires initial and boundary conditions. A general model
for the boundary conditions are
D = εE (6)
J = σE (7)
qn = h(Ts − Ta ) + σs εw (Ts4 − Ta4 ) (13)

The displacement current is neglected and a harmonic

which is related to the gradient of the temperature field
solution is assumed. The latter means that the excita-
in the normal direction of the surface, ∂T /∂n.
tion current and its solutions are sinusoidal functions.
Furthermore, the magnetic vector potential A is intro- In the case of a harmonic electromagnetic loading,
the heat source term can be written as [2, 3], constants the coil current is regulated giving an output
temperature closed to the prescribed temperature.
1 1
Q= |Je |2 = Je · J∗e (14)
2σ 2σ
4 Experimental setup
where Je is the induced current density in the conductor, The induction heating equipment is a voltage-fed
calculated as Je = −iωσA. The heat source term, Q, is series load with a capacitor at the input of the inverter
then solved for each node point whenever Eq. (9) is and a series-connected output circuit. To transfer max-
solved, and used as input to Eq. (12) for a solution of imum power from the DC-source to the load, the trans-
the heat distribution. The use of Eq. (14) is possible as former needs to work in its resonance frequency. There-
the time scale of the electromagnetic problem is much fore during, the first 12 s of operation, the induction
shorter than in the thermal problem. heating equipment searches for its resonance frequency.
See [1] for more information. The induction heating
equipment used in the experiments is Minac 18-5.
3 Control logic for current
In order to obtain a desired temperature, a Propor-
tional Integral Derivative (PID) controller is used to
regulate the current in the computational model. The
deviation between the prescribed and actual tempera-
ture is the error e. This is used to compute a correction
u, to the current according to

1 t
 
de
Z
u(t) = K e(t) + e(τ)dτ + TD (15)
TI 0 dt

Fig. 2. The geometry of the workpiece and position of the thermo-

The algorithm is often illustrated by the block diagram
couples. See Table 1 for the distances. The coil has a dimension of
in Fig. 1
8x4 mm.

Table 1. Positions of thermo-couples.

Measurement Distance from Distance from

location center [mm] the end [mm]
Fig. 1. Block diagram of the PID controlling system. The process,
1 0 0
G, represents by the FE-program, e is the error and u the correction.
2 8 0

The discrete version of Eq. 15 is implemented as 3 16 1

4 16 12
" #
∆t n
TD
∑ ei + ∆t (en − en−1 )
5 16 39
un = K en + (16)
TI i=0
6 16 60
7 16 70
This is used to modify the current for the next increment

Pn+1 = Pn + un (17) Fig. 2 shows the experimental setup; the coil is

placed in position A (at the middle of the workpiece) or
By adjusting the length of the time step and the different at B (at the end of the workpiece). The rod is 140 mm
long and has a diameter of 32 mm. The numbers in cir- tion B. An average frequency of 23 KHz was used in all
cles denote the different thermo-couples with locations the simulations.
given in Table 1.
The prescribed temperature history in the two dif-
ferent cases consist of three phases:
1. Linear heating at a rate of (760/180) °C/s
2. Temperature is held constant at 760 °C for 120 s.
3. Current turned off, and the workpiece is allowed to
cool down to a uniform temperature.
The temperature was measured using thermo-
couples and recorded at a frequency of two samples per Fig. 3. The mesh created for coil position A. Air, coil and workpiece
second. Each test was repeated three times to ensure are modelled.
the reproducibility. In the following result figures, the
mean value of the temperature for the three tests are
plotted. The maximum temperature difference between
the replicated tests was 7.3 °C at time 735 s for the coil
located at position A and 7.4 °C at time 176 s for the 29

coil at position B. The thermo-couples were calibrated 28

before and after their use.
27
The difference did not exceed 1.1 °C or 0.4% of
its measured value. Thus, the maximum error in the
Frequency [KHz]

26
Coil position A
measurement did not exceed 3.0 °C for 760 °C. 25
Coil position B

The induction heating equipment uses a PID con-

24
troller to regulate the current to obtain a specified tem-
perature history. Therefore, a separate measurement 23

thermo-couple was fixed at the workpiece for feedback

22
to the induction device. Its location was diametrically
opposite to thermo-couple 7 for case A and opposite 21
0 50 100 150 200 250 300
Time [s]
thermo-couple 4 in test B.
The workpiece material is Inconel 718, and its
Fig. 4. Measured frequency of the heating equipment.
properties can bee seen in Appendix 7. Inconel is para-
magnetic and therefore has a permeability of unity.
The heat generation is concentrated near the surface
5 Computational model due to the skin effect and can be estimated to 3.6 mm
with [9]
A commercial finite element program MSC.
Marc(2005r3 64 bit) was used to solve the computa-
tional model. The finite element mesh shown in Fig. 1
δ= √ (18)
3 consist of four-noded axisymmetric quadrilateral el- π f µσ
ement. The different domains, air, coil and workpiece,
were given different material properties. Electrical con- Therefore, a fine mesh was created near the surface of
ductivity of the coil and air were ignored. The current the workpiece and gradually coarsened towards the in-
was prescribed in the coil. terior.
The frequency was registered from the heating Both natural convection and radiation boundary
equipment and varied for the coil in position A from condition was taken into account during the simulation.
26 KHz, in the beginning of the experiment, to 22 KHz. The film coefficient h = 1.9 and the potential α = 1.3
A similar behaviour was also registered when the coil was used on all boundaries between the solid and the
was in position B. See Fig. 4. The average frequency air [1, 10]. The curved surface of the cylinder was non-
after 50s was calculated and was 22.4 KHz for the coil polished, and is therefore assumed to have an emissivity
in position A and 23 KHz for the coil located in posi- factor of εw = 0.5. The ends of the cylinder were on the
other hand polished and an emissivity of εw = 0.25 is
180
assumed here.
160
The following parameters were used in the PID
controller logic together with a simulation time step of 140

∆t = 0.5s. See chapter 3 for description of the controller 120

Current Amplitude [A]

logic. 100

80

K = 6000 60

TI = 1080 40

TD = 210 20

0
0 50 100 150 200 250 300 350
Time [s]

6 Results
The two tests, A and B, differ with respect to the Fig. 5. Symbols denote measured current and the line denote cal-
location of the coil. See Fig. 2. Table 1 gives the loca- culated current using a PID-regulator. The current is the peak value
tion of the different thermo-couples. Thermo-couples of the alternating current. This is for test case A.
5-7 were used in case A and no. 1, 4 and 5 for case B.

800
Calculated data
7
6.1 Controlling current for obtaining specified 700
Measured data

temperature 6
600
The PID controller logic was used to calculate the
current needed in the model to obtain the desired tem-
Temperature [ C]

500
o

perature at a node located at the same place as thermo- 400

couple 7 in case A and 4 in case B. The calculated cur- 5
300
rent is then compared with the measured current from
the experiment. The current and temperature at position 200

A is given in Fig. 5 and 6 respectively. For position B, 100

the current is given in Fig. 7 and Fig. 8 shows the tem-
0
perature history. 0 100 200 300 400 500 600 700
Time [s]

6.2 Prescribing measured current to model Fig. 6. Measured and computed temperature for case A. The num-
bers refer to measurement locations in Tab. 1. The PID regulator
The current profile given from the measurement
uses the temperature at position 7 as input.
was used as input current density (i.e. J0 (t)) in this
case. The calculated temperature was then compared
with measured temperature. See Fig. 9 for the temper- 7 Conclusions
ature history when the coil is located in position A and The good agreement between simulations and mea-
Fig. 10 for the coil located in position B. surements shows that the model, the computational ap-
The frequency influence on the temperature was proach as well as used material properties, is valid and
also simulated. Three different frequencies, 22, 23 and can be used to study the induction heating process. The
24 KHz was used as excitation frequencies. The tem- current can be controlled with a PID-regulator or with
perature was calculated using a prescribed current am- a measured current history profile. Regulating the cur-
plitude. Fig. 9 shows the temperature history for differ- rent work particular well aiming to find the process pa-
ent frequencies. rameter for a specified temperature history. This will
In order to study the influence of current on the be useful when designing the induction heating process
temperature, a small change in the current amplitude needed to create a wanted temperature history.
was performed; 1% reduction in case B. The experi- For test case A, the analysis shows significant dif-
ment followed the same procedure as in section 6.2 and ference in temperature between the thermal position 6
the result is seen in Fig. 10. and 7. This confirms that the induction heating process
 150 900
Calculated data
24 KHz Measured data
800
23 KHz
125 7 22 KHz
700
Current Amplitude [A]

100 600

Temperature [oC]
6
500
75

400

50
300
5

25 200

100
0
0 50 100 150 200 250 300 350
Time[s] 0
0 100 200 300 400 500 600 700
Time [s]

Fig. 7. Symbols denote measured current and the line denote cal-
Fig. 9. Calculated and measured temperatures for case A when ap-
culated current using a PID-regulator. The current is the peak value
plying measured current to the model. The current amplitude is seen
of the alternating current. This is for test case B.
in Fig. 5. The frequency is changed to 24 KHz and 22 KHz and
compared with the assumed frequency of 23 KHz.

800
Calculated data
Measured data
700 800
Calculated data
4
Measured data
600 700
0.99 x curr. amp.
4
500 600
Temperature [oC]

1
Temperature [ C]

400 500
o

1
300 400

5
200 300

5
100 200

0 100
0 100 200 300 400 500 600 700
Time[s]
0
0 100 200 300 400 500 600 700
Time [s]
Fig. 8. Measured and computed temperature for case B. The num-
bers refer to measurement locations in Tab. 1. The PID regulator
Fig. 10. Calculated and measured temperatures for case B when
uses the temperature at position 4 as input.
applying measured Curran to the model. The current amplitude is
seen in Fig. 7. The current amplitude is reduces by 1% and com-
pared with the original one.
can produce a desired temperature profile at a selected
region. The temperature drops fast outside the region of
the coil, showing that local heat treatment is possible. Nomenclature
The sensitivity of the temperature of different pro- A magnetic vector potential
cess parameters (current and frequency) was evaluated. B magnetic flux density
When the current amplitude was reduced by 1% of its D displacement current
original value, the maximum temperature difference is E electric field density
estimated to be 15 °C. The difference in temperature H magnetic field density
between the highest and lowest used frequencies is ap- J current density
proximately 50 °C. Although the skin depth is small J0 source current density in the induction coil
compared to the workpiece diameter, it still has an in- Je induced current in conductors
fluence on the temperature history. This shows that it J∗e conjugated complex value of J
may be important to be able to vary the frequency dur- K proportional gain
ing a heating cycle in the computational model. P power output
T temperature Journal of Manufacturing Science and Engineer-
Q heat source density generated in the material per ing, 127, August, p. 555–563.
unit volume and time [6] Bay, F., Labbe, V., Favennec, Y., and Chenot, J. L.,
Ta ambient temperature in Kelvin 2003. “A numerical model for induction heating
Td derivative time processes coupling electromagnetism and thermo-
Ti integral time mechanics”. International Journal for Numeri-
Ts surface temperature in Kelvin cal Methods in Engineering, 58(6), p. 839–867.
c specific heat capacity 10.1002/nme.796.
e error between the measured and desired temperature [7] Drobenko, B., Hachjevych, O., and Kournyts’kyi,
T., 2006. “A mathematical simulation of high
f frequency
temperature induction heating of electrocunduc-
h film coefficient
tive solids”. International Heat and Mass Trans-
i complex representation
fer.
n̂ normal vector
[8] Stratton, J. A., 2007. Electromagnetic theory.
(n) current time step IEEE Press, Piscataway, NJ.
(n − 1) previous time step [9] Sadiku, M. N., 2001. Elements of Electromagnet-
qn the outgoing heat flux ics, Vol. 3. Oxford University Press, Inc.
u controller output [10] Davies, E. J., 1990. Conduction and induction
∆t time step length heating. No. 0-86341-174-6 in IEE Power Enge-
δ skin depth neering Series 11. Peter Peregrinus Ltd., London,
ε permittivity UK.
εw emissivity of the workpiece
κ the thermal conductivity
µ permeability APPENDIX
ω excitation frequency The used material properties for Inconel 718. Fig.
11 shows the electric and thermal conductivity and Fig.
ρ density of the material
12 the specific heat capacity. The electric conductivity
ρv free volume charges
is expressed in Kilo Siemens per meter.
σ electrical conductivity
σs Stefan Boltzmann’s constant
30 840

28
References 830

[1] Rudnev, V., Loveless, D., Cook, R., and Black, 26

820
Thermal conductivity [W/m K]

M., 2003. Handbook of Induction Heating. In-

Electric conductivity [KS/m]

24
810
ductoheat, Inc. 22
800
[2] Clain, S., Rappaz, J., Swierkosz, J., and Touzani, 20
790
R., 1993. “Numerical modeling of induction heat- 18
ing for two-dimentional geometries”. Mathemat- 16
780

icl Models and Methods in Applied Sciences, 3(6), 770

14
p. 805–822.
12 760
[3] Wang, K., Chandrasekar, S., and Yang, H. T.,
10 750
1992. “Finite-element simulation of induction 0 200 400 600
Temp [°C]
800 1000 1200

heat treatment”. Journal of Materials Engineer-

ing and Performance, 1.
Fig. 11. The electric and thermal conductivity for Inconel 718.
[4] Chaboudez, C., Clain, S., Glardon, R., Mari, D.,
Rappaz, J., and Swierkosz, M., 1997. “Numeri-
cal modeling in induction heating for axisymmet-
ric geometries”. Magnetics, IEEE Transactions
on, 33(1), p. 739–745. 0018-9464.
[5] Luo, J., and Shih, A. J., 2005. “Invers heat transfer
solution of the heat flux due to induction heating”.
 700

650
Specific Heat Capacity [J/Kg K]

600

550

500

450

400
0 200 400 600 800 1000 1200
Temp [°C]

Fig. 12. The heat capacity in constant pressure for Inconel 718.
Paper B
Simulations and measurements of ombined indu tion heating and extrusion pro esses
S. Hansson a,b,∗
, M. Fisk

a Sandvik Materials Te hnology, 811 81 Sandviken, Sweden

b Dalarna University, 781 77 Borlänge, Sweden
Luleå University of Te hnology, Division of Material Me hani s, 971 87 Luleå, Sweden

Abstra t
The manufa turing pro ess hain at glass-lubri ated extrusion of stainless steel tubes is simulated using the nite element
method. The developed model in ludes sub-models of indu tion heating, expansion and extrusion. An in-house mapping
tool is used to transfer the temperature elds between the ele tromagneti -thermal and thermo-me hani al analyses.
The model is su essfully applied to two ases of tube extrusion; one using an austeniti stainless steel and one using
a duplex, austeniti /ferriti stainless steel. It is shown that the indu tion heating model predi ts the temperatures
obtained experimentally from thermo ouples pla ed in the steel billets during heating. The agreement between models
and experiments regarding extrusion for e and expansion for e are satisfa tory.
Key words: Eddy urrent, Stainless steel, Finite Element Method, Extrusion, Ferromagneti , Magneti hysteresis.

1. Introdu tion from previous sub-model, i.e. the material is assumed to

be fully re overed at the initial stage of ea h model.
Seamless stainless steel tubes an be manufa tured by Extrusion of two dierent stainless steel tubes was stud-
hot extrusion using glass as lubri ation. Before extrusion ied. One was made of an austeniti grade, Sandvik 353
the billet is prepared in several operations, in luding in- MA, and one of a duplex, austeniti /ferriti alloy, Sandvik
du tion heating, lubri ation and expansion. SAF 2507. Sin e the duplex steel is ferromagneti below
Simulation of extrusion using the nite element method the Curie temperature, the skin ee t is more pronoun ed
(FEM) has been performed extensively during the last ompared to the paramagneti austeniti material. Also
de ade. The majority of these simulations on ern extru- the magneti non-linearity of the material properties leads
sion of aluminum but work regarding steel and titanium to dieren es in the indu tion heating modelling. Results
extrusion has also been reported. The most ommon ap- from the numeri al models have been validated by exper-
proa h in these simulations is to use a uniform billet tem- iments in produ tion presses and indu tion furna es.
perature as initial ondition in the extrusion model. The
initial stages of heating and transport are often ignored.
However, the major ee ts of extrusion variables on pres-
sure in steel extrusion are related to their ee t on the
temperature of the billet, and hen e the ow stress of the
material [3℄. Knowledge of the billet temperature prole
before extrusion is therefore signi ant to obtain a urate Figure 1: Simulated pro ess ow. Ea h box refers to one sub-model.
simulation results. It is also important to understand the
heating pro ess in order to eliminate unwanted tempera-
ture gradients.
In this study, axisymmetri FE-models of indu tion 2. Pro ess ow
heating, expansion and extrusion, in luding ooling at the A s hemati diagram of the pro ess ow an be seen in
intermediate transports, have been ombined. The pro- Fig. 1. The material is re eived as a hot rolled round bar
ess ow was modeled using seven dierent sub-models, as whi h is turned and ut into lengths based on the material
shown in Fig. 1. All models were developed in the om- volume required for the extruded tube. A pilot bore hole
mer ial FE-software MSC.Mar and the nodal tempera- is drilled in ea h billet; a nose radius is ma hined at the
tures were transferred from one model to another using front edge and a one at the ba k end. After this, the billet
an external mapping program. No strains were mapped is transported for manufa turing, where the rst step is to
heat the billet in an indu tion furna e. After heating, the
∗ Correspondingauthor
billet is rolled down a glass- overed tray until the surfa e
Email addresses:sofia.e.hanssonsandvik. om (S. Hansson), is oated with a thin layer of powdered glass. This glass
martin.fiskltu.se (M. Fisk) is used as lubri ant in the following expansion pro ess.
Preprint submitted to Elsevier November 13, 2008
 The heated billet is inserted nose down in a verti al onne ted ondu tor formulation must be employed [1℄.
expansion press, in whi h the pilot bore hole is expanded This is not implemented in MSC.Mar and this ee t is
using a shaped nose. The billet is elongated during this approximated as des ribed below.
pro ess. A glass pad (a dis of ompa ted glass) is pla ed The resistan e in an indu tion heating system is the
in front of the nose and melts during the operation, thereby sum of the resistan e of the oil and the ree ted resistan e
assuring su ient lubri ation. In the expansion pro ess, it of the workpie e [8℄. It is thus possible to represent the
is desirable to have a ontrolled front temperature of the oil and workpie e resistan e as a series ir uit.
billet. This is to avoid unwanted extrusion at the bottom This allows us to al ulate a heat e ien y fa tor
of the press.
The expanded billet is reheated, lubri ated and ex- Rw
η= (1)
truded to the spe ied nal dimension. A glass pad is Rw + Rc
pla ed between the billet and the die to ensure lubri a- where Rc is the resistan e of the oil and Rw , is the re-
tion. In this operation, a hot leading end is desired in sistan e of the workpie e. Details an be found in [10, 2℄.
order to redu e the peak extrusion for e. The heat e ien y fa tor is used to redu e the heat sour e
The billet is heated in an indu tion furna e with the term in the heat ondu tion equation.
aim to provide a uniform surfa e-to- ore temperature. How- The resistan e of a ylindri al workpie e and its oil
ever, in indu tion heating the ore is heated slower than its an be written as [2℄,
surfa e. This is due to the skin ee t. The skin depth de-
pends on the frequen y of the indu tion heating power as
well as the material properties of the billet [5℄. It should be Rw = K(µr pAw ) (2)
noti ed that it is di ult to a hieve uniform temperature kr πdc δc
in stainless steel due to its poor thermal ondu tivity. Rc = K( ) (3)
2
The basi setup of an indu tion furna e onsists of a
oil with one or several windings, keeping an ele tri ally Here is K a onstant, µr is the relative permeability, p
ondu ting workpie e within. The urrent arrying oil is the power loss onstant, Aw is the ross se tion area of
produ es an alternating ele tromagneti eld whi h in- the workpie e, kr is the orre tion fa tor allowing spa ing
du es eddy urrent in the surfa e of the workpie e. Heat between oil turns, dc is the oil diameter and δc is the
is generated due to the Joule heating ee t aused by this skin depth of the oil. The skin depth of a material an
urrent. be al ulated as [6℄
1
δ=√ (4)
πf µσ
where f is the ex itation frequen y, µ the permeability
and σ the ele tri al ondu tivity of the material.
The e ien y fa tor, η , an now be al ulated for a
spe i oil and workpie e geometry and used to multiply
the heat sour e term in the heat ondu tion equation
∂T
ρcp − ∇ · (κ∇T ) = ηQ (5)
∂t
Figure 2: The pro ess ow for produ tion of stainless steel tubes. where ρ is the density of the material, cp is the spe-
i heat apa ity and κ is the thermal ondu tivity of
the workpie e. In the ase of a harmoni ele tromagneti
loading the heat sour e term due to the eddy urrent an
3. The ele tromagneti -thermal problem be written as [1, 9℄
Simpli ations employed in the al ulation of the heat 1 1
e ien y fa tor in indu tion heating and linearisation of Q= |J|2 = J · J∗ (6)
2σ 2σ
the ferromagneti material are summarised below.
where J is the indu ed eddy urrent in the workpie e and
3.1. Cal ulation of the heat e ien y fa tor the star (*) denotes the onjugated omplex value of the
The magneti eld of the oil in the indu tion furna e urrent J. The urrent is genereted within the workpie e
indu es voltage in the workpie e as well as in the oil itself. as J = −iωσA where A is the magneti ve tor potential
The indu ed voltage is in su h dire tion that it tends to al ulated from the diusion equation. See Eq. (7) in
weaken the urrent owing in the oil. In order to simulate se tion 3.2. All material properties are fun tions of tem-
this phenomena using the nite element method, a double perature.

2
3.2. Magneti o-energy density and non-linearity
In the nite element implementation, time variation
1.5

of the eld quantities are assumed to be sinusoidal. To

be able to use this harmoni approximation, the magneti 1

material property must be onstant. Combining Maxwell's

equations, the linear diusion equation an be formulated 0.5

[7℄ as

B [Vs/m2]
20
100
1 0
iωσ A − ∇2 A = J0 (7) 200
µ 300
400
where A is the magneti ve tor potential, ω is the angular
−0.5
450

frequen y of the magneti eld, µ is the permeability of 480

490
the material, and J0 is the urrent applied in the oil. −1
500

The ferromagneti material SAF 2507 has nonlinear 520

600
magneti properties and therefore Eq. (7) is not valid. The −1.5
−5 −4 −3 −2 −1 0 1 2 3 4 5
hysteresis urve for SAF 2507 at dierent temperatures H [A/m] 5
x 10

an be seen in Fig. 2. The magneti ux density, B,

represents the internal eld strength within a substan e, Figure 3: Magneti hysteresis urve for dierent temperatures in
whi h is subje ted to an external magneti eld, H. This SAF 2507. The Curie temperature starts approximately at 500 ◦ C.
hysteresis behavior an be modeled by the Fröli h equation
H x 10
5

B= (8)
0.8 8

α + β||H||
0.7 7
where α and β are variables dependent on temperature.
Figure 3 shows the values of α and β for SAF 2507. 0.6 6

The nonlinear ele tromagneti behavior of the material

an be linearised following Labridis and Dokopoulos [4℄.
0.5 5

They have shown that it is possible to approximate the 0.4 4

α
β

nonlinear material with a  tive linear material that has

onstant relative permeability, µfr . The linear  titious 0.3 3

material should have the same average heat density loss as

that of the nonlinear material at every point i. It an be 0.2 2

shown that the  tive relative permeability is [4℄ 0.1 1

w1i + w2i
µfri = f 2
(9) 0
0 100 200 300 400 500 600
0
µ0 ∗ (Hmi ) Temperature [°C]

where w1i and w2i is the magneti o-energy density of the

material. w1i is related to the a tual B − H urve and w2i Figure 4: The variation of α and β for dierent temperatures. See
is related to the average value of the slope dB/dH during Eq. (8).
a quarter of a period T . Hmi f
is the maximum value of
the magneti eld intensity in the  titious material. The 4. Materials
magneti o-energy density an be al ulated as
Two dierent grades of stainless steels, Sandvik 353
f
Hmi f
α α + βHmi MA and Sandvik SAF 2507, were onsidered in this study.
w1i = − 2 ln (10) 353 MA (UNS: S35315) is an austeniti hromium-ni kel
β β α
f 2
steel mainly used in high temperature appli ations. SAF
w2i =
1 (Hmi )
(11) 2507 (UNS: S32750) is a high alloyed duplex grade, onsist-
2 α + βHmif
ing of approximately 50% austenite in a matrix of ferrite.
The nominal hemi al ompositions of 353 MA and SAF
Sin e both the relative permeability of the  titious 2507 are given in Table 1 and Table 2, respe tively.
material and the magneti eld intensity are unknown, an
iterative pro ess is ne essary. The starting value of µfr is
set to the initial slope of the Fröli h urve, i.e.
4.1. Flow stress
The deformation behavior at high temperature was de-
1 termined by me hani al testing in a number of ompres-
µfr = (12)
µ0 α sion tests. These tests were performed in a Gleeble 1500
Thereafter µfr is al ulated from Eq. (9). thermome hani al simulator at several temperatures and
3
 Table 1: Chemi al omposition (nominal) of 353 MA [%℄ 350

C Si Mn P S Cr Ni N Ce
max max 300
1000°C, 10/s

0.07 1.6 1.5 0.040 0.015 25 35 0.16 0.05 250

1000°C, 1/s

True Stress [MPa]

200
1100°C, 10/s
Table 2: Chemi al omposition (nominal) of SAF 2507 [%℄
C Si Mn P S Cr Ni Mo N 150
1100°C, 1/s
max max max max max
100 1000°C, 0.01/s
0.030 0.8 1.2 0.035 0.015 25 7 4 0.3
50 1100°C, 0.01/s

strain rates, up to a total strain of 60%. Cylindri al spe i- 0

0 0.1 0.2 0.3 0.4 0.5 0.6
mens of diameter 10 mm and height 12 mm were ma hined True Strain [−]

from hot rolled round bars, with the rolling dire tion of
the bar orresponding to the loading axis of the test spe - Figure 5: Compression tests of SAF 2507 at temperatures 1000 and
imens. 1100 ◦ C.
Flow stress urves of SAF 2507 at 1000, 1100, 1200
and 1300 ◦ C and strain rates 0.01, 1 and 10 s−1 are given 140
in Fig. 4 and Fig. 5. The jerky ow and the serrations
that are visible at higher deformation rates are dynami 120 1200°C, 10/s
ee ts due to the testing ma hine.
No me hani al testing was ondu ted for 353 MA. Its 100
material properties were assumed to be equal to those of
True Stress [MPa]

Sandvik Sani ro 28, a similar material. A omparison of 80 1200°C, 1/s

the hemi al ompositions of the materials an be seen in

Table 1. Flow stress urves of Sani ro 28 at 1000, 1100, 60 1300°C, 10/s
1200 and 1300 ◦ C and strain rates 0.01, 1 and 10 s−1 are 1300°C, 1/s
given in Fig. 6 and Fig. 7. 40

The similarity in hot strength of Sani ro 28 and 353 1200°C, 0.01/s

MA are illustrated in Fig. 8. These tests were arried out 20 1300°C, 0.01/s

in tension at a onstant rate of 50 mm/s. The test spe -

imens were ylindri al rods, length 50 mm and diameter 0
0 0.1 0.2 0.3 0.4 0.5 0.6
6.3 mm, whi h were strained to fra ture at dierent tem- True Strain [−]

peratures. The maximum for e was re orded and used as

an estimation of the hot du tility in the material. Figure 6: Compression tests of SAF 2507 at temperatures 1200 and
1300 ◦ C.

5. Experiments
on again for a nal heating before the billet is eje ted and
5.1. Indu tion furna e transported for lubri ation and expansion. To y le the
Verti al indu tion furna es with water ooled oil sur- power leads to more uniform surfa e-to- ore temperature,
rounding the billet are used in the experiments. The billet whi h is desirable in the extrusion pro ess. The same type
is inserted into the oil using a hydrauli ram. Flux on- of furna e is used for the heating between expansion and
entrators are applied at both ends of the billet in order extrusion (Heating 2), although dierent number of a tive
to obtain a required temperature uniformity through the oil and no power y ling are used. See Fig. 1 for the
length of the billet. Using ux on entrators, the oil over- simulation steps.
hang an be minimized and the possible billet length in a The oil liner has a thi kness of 19 mm with an inner
given furna e an be maximized [5℄. The number of oil diameter of 360 mm. The furna e onsists of refra tory
turns an be varied to obtain a desired temperature prole. material and a metal shell to prote t the oil liner from
The billet is heated by applying an alternating urrent damage and redu e the heat loss to the environment. This
to the oil. When the surfa e temperature rea hes a spe - redu es the inner diameter of the furna e to 260 mm. The
ied value (usually between 1170 − 1210 ◦ C), the furna e frequen y of the urrent is 50 Hz.
is swit hed o, letting the billet ool down inside the fur- The temperature of the billet during heating was mea-
na e. After a number of se onds, the furna e is turned sured ea h se ond with K-type thermo ouples using a PC-

4
 400 11000

10000 Sandvik 353 MA

350 1000°C, 10/s
Sandvik Sanicro 28
9000
300 1000°C, 1/s
8000
True Stress [MPa]

250

Max Force [N]

1100°C, 10/s 7000

200 1100°C, 1/s 6000

5000
150 1000°C, 0.01/s

4000
100
3000
1100°C, 0.01/s
50
2000

0 1000
0 0.1 0.2 0.3 0.4 0.5 0.6 850 900 950 1000 1050 1100 1150 1200 1250 1300
True Strain [−] Temperature [oC]

Figure 7: Compression tests of Sani ro 28 at temperatures 1000 and Figure 9: Comparison of hot du tility in Sani ro 28 and 353 MA.
1100 ◦ C.

220

200
1200°C, 10/s
180

160
1200°C, 1/s
True Stress [MPa]

140

120 1300°C, 10/s Figure 10: Pla ement of thermo- ouples (x) and geometri properties
100 of the billet.
1300°C, 1/s
80

60 1200°C, 0.01/s a ordan e with the dimensions of the nal tube. These
40 measurements are given in Tables 4 and 5. Transient for e
20
1300°C, 0.01/s
and ram speed were re orded during the extrusion pro ess.
0
0 0.1 0.2 0.3 0.4
True Strain [−]
0.5 0.6
6. Computational models
Separate models for expansion, extrusion and indu -
Figure 8: Compression tests of Sani ro 28 at temperatures 1200 and tion heating pro esses are used. The temperatures of the
1300 ◦ C.
pre eding pro ess step is mapped with an in-house ode
to the initial mesh of the model in the following step. The
logger. The thermo ouples were pla ed in drilled holes at ommer ial nite element program MSC. Mar 2007r1 64
dierent heights and depths of the billet. See Fig. 9 and bit is used for all al ulations.
Table 3 for the pla ement of the thermo ouples. The oil
urrent was re orded ea h se ond using a PC-logger and
an be seen in Fig. 10. Table 3: Pla ement of the thermo ouples.
Steel Billet dimension a b
5.2. Expansion and extrusion type LxRxr [mm℄ [mm℄ [mm℄
The expansion and extrusion pro esses are ondu ted
in a 12 MN verti al hydrauli press and a 33 MN horizontal 353 MA 738x118.5x18.5 100 10, 20, 72
hydrauli press, respe tively. All experimental data was 369 10, 20, 94
re orded during produ tion in these presses. In the expan- 638 10, 50, 94
sion press, the ontainer diameter was 240 mm with a max-
imum nose diameter of 112 mm for 353 MA and 118 mm SAF 2507 946x118.5x22.5 473 5, 10, 20, 30,
for SAF 2507. The transient for e during expansion was 40, 64, 78
re orded for SAF 2507 and the peak for es for 353 MA.
The ontainer diameter in the extrusion press was
250 mm. The mandrel and die geometry was hosen in
5
 this temperature is a fun tion of time. The oil liners have
4500
a onstant temperature of 60 ◦ C.
4000
535 MA − Heating 1
In order to simulate the ee t of the ux on entrators,
the material properties from SAF 2507 are used. The non-
3500
SAF 2507 − Heating 1 linear ele tromagneti behavior of the ux on entrator
3000 material is also linearised. The permeability is, however,
Current Amplitude [A]

multiplied with a fa tor of twenty, sin e the ux on en-

trators must have a higher permeability than SAF 2507
2500
535 MA − Heating 2
2000 [5℄.

6.1.1. Sandvik 353 MA

SAF 2507 − Heating 2
1500

1000 The workpie e resistan e and the oil resistan e are

500
al ulated using Eq. (2) and Eq. (3). With numbers they
give
0
0 100 200 300 400 500 600 700 800 900
Time [s] Rw = K(µr pAw ) = K(1 · 0.35 · 0.043) =
= K0.015 (14)
Figure 11: Nominal urrent amplitude in the indu tion oil during
 
kr πdc δc
the heating pro edures. Rc = K =
2
 
1.05 · π · 0.292 · 0.009
Table 4: Billet and tube dimensions used in 353 MA simulations. = K =
2
Billet Outer diameter [mm℄ 237 = K0.0045 (15)
Inner diameter [mm℄ 37
Length [mm℄ 738 The heat e ien y fa tor an then be al ulated from Eq.
Expanded billet Outer diameter [mm℄ 240 (1), giving η ≈ 0.78.
Inner diameter [mm℄ 112 The oil overhang (the distan e from the rst a tive oil
turn to the front end of the workpie e) during Heating 1
Extruded tube Outer diameter [mm℄ 141 and Heating 2 are 115 mm and 185 mm. Number of a tive
Wall thi kness [mm℄ 22.0 oil turns are 48 and 58, respe tively. The nominal urrent
in the oil is seen in Fig. 10. In Heating 2, the e ien y
fa tor is al ulated to be η = 0.74.
Table 5: Billet and tube dimensions used in SAF 2507 simulations.
Billet Outer diameter [mm℄ 237
Inner diameter [mm℄ 45
Length [mm℄ 946
Expanded billet Outer diameter [mm℄ 240
Inner diameter [mm℄ 118
Extruded tube Outer diameter [mm℄ 114
Wall thi kness [mm℄ 10.0 Figure 12: FE-model of the indu tion furna e for Heating 1 of 353
MA. The smallest element size is 2 mm.

6.1. Indu tion heating 6.1.2. SAF 2507

The FE-models are meshed with four-noded axisym- In a ferromagneti material, the initial skin depth is
metri quadrilateral elements. See Fig. 11 for how the smaller than in a paramagneti material, resulting in a
mesh is reated for Heating 1 of 353 MA. The radiation higher e ien y fa tor. Above the Curie temperature, the
heat ux is ignored as the refra tory material inside the e ien y fa tor will hange to that of a paramagneti ma-
furna e is assumed to ree t the heat radiated from the terial, be ause of de reasing magnetism for in reasing tem-
billet. peratures. However, a onstant e ien y fa tor (η = 0.85)
The heat transfer by onve tion from the surfa e of the is used throughout the Heating 1 simulation.
workpie e to the ambient air is taken into a ount. A good The oil overhang in Heating 1 is 95 mm and in Heating
approximation for a workpie e in open air [5, 2℄ is, 2 155 mm. Number of a tive oil turns are 59 and 72. The
qconv = 1.54(Tx − Ta )1.33 (13) nominal urrent in the oil is seen in Fig. 10. In Heating
2 the e ien y fa tor was al ulated to be η = 0.73.
where Ta is the ambient temperature. Sin e the ambient
temperature inside the furna e is hanging during heating,
6
6.2. Cooling
Both radiation and onve tion heat losses are taken
into a ount in the ooling stages. The ambient tempera-
ture is xed to Ta = 303 K . The onve tion losses are the
same as in se tion 6.1 and the surfa e emissivity is ε = 0.3.
The total ux an therefore be expressed as where σs is Figure 14: FE-model of expansion of 353 MA (nal stage).
the Stefan Boltzmann's onstant.
The transport time from the indu tion furna e to the
expansion press (Cooling 1) is 70 s. The expanded billet
is transported 60 s to the nal heating (Cooling 2), and
thereafter 70 s to the extrusion press (Cooling 3). See Fig.
1 for the simulation pro ess ow.
Figure 15: FE-model of extrusion of 353 MA (initial stage).
6.3. Expansion and extrusion
The expansion and extrusion models are built up using 6.3.1. Material model
four-node axisymmetri quadrilateral elements. The tools The materials are modeled as elasti -plasti with iso-
are treated as rigid bodies with heat transfer properties. tropi hardening using the von Mises yield riterion. The
This means that the heat transfer between the billet and ow stress dependen e on temperature, plasti strain and
the tools is aptured but the dee tion of the tools is ig- strain rate is tabulated. The underlying experimental data,
nored. The isolating properties of the lubri ating glass lm see Figs. 4-7, in ludes temperatures between 800−1300 ◦ C
implies low onta t heat transfer relative to other lubri- and strain rates between 0.01 − 10 s−1 . Linear interpola-
ants. A onta t heat transfer oe ient of 1.5 kW/m2 K tion is used inside this domain. Outside the temperature,
is used at the glass lubri ated onta t areas. The tools strain, and strain rate limits, the extreme values are used
in the models have thermal properties equal to the ma- and no extrapolation is performed.
terial AISI H13 and an initial temperature of 400 ◦ C at
expansion and 300 ◦ C at extrusion. The Coulomb fri tion 6.4. Mapping between indu tion heating and extrusion
oe ient between the billet and the tools is assumed to An in-house mapping tool for moving data between two
µC = 0.02 at all glass lubri ated onta t areas. In extru- nite element models is used to map temperature elds be-
sion, the fri tion oe ient between the billet and the ram tween the ele tromagneti -thermal and thermo-me hani al
is µC = 0.35. analysis. First a material identi ation is performed. Ea h
The ram (nose) speed in the expansion pro ess is material of the re eiving model obtains temperatures only
160 mm/s for 353 MA and 200 mm/s for SAF 2507. In from orresponding material of the sending model. Mate-
the extrusion pro ess, the ram speed is 260 mm/s for 353 rials in the re eiving model whi h have no orresponding
MA and 210 mm/s for SAF 2507. material in the sending model, are given a user dened
Generation of heat due to plasti deformation in the temperature.
material is taken into a ount and 95 % of the plasti work The mapping method is based on the interpolation
is assumed to be onverted into thermal energy. Large de- te hnique used in nite element formulations. The nodal
formations in the Lagrangian ode implies that ontinuous temperatures of the re eiving model are obtained by
remeshing is needed during the analysis. This is ontrolled
by automati remeshing s hemes available in the software. Tnew = Nold (Sloc )Told (16)
The rst and last in rement of the 353 MA expansion
model an be seen in Fig. 12 and Fig. 13, respe tively. The where Nold is the shape fun tion of the element in the
FE mesh for extrusion of the same material is shown in Fig. sending model evaluated at the lo al oordinate, Sloc . In
14. The models for SAF 2507 expansion and extrusion are this pro edure it is required to determine where a node
identi al to those with 353 MA apart from dieren es in of the re eiving mesh is lo ated in the sending mesh. A
geometry (see Tables 4 and 5), ram speed and material sear h is performed to determine in whi h element of the
model. sending (old) mesh the node is lo ated, and whi h lo al
oordinate it orresponds to. The use of isoparametri
nite elements gives
Xnew = Nold (Sloc )Xold (17)
where Xold is the nodal oordinates of the sending element
and Xnew is the oordinate of the node whi h re eives the
Figure 13: FE-model of expansion of 353 MA (initial stage). temperature.
Eq. (17) denes a one-to-one mapping between the lo-
al oordinate system and the global oordinate system of
7
the mesh. However, in the general ase, it does not exist
an analyti al inverse mapping in whi h the lo al oordi- 1400
10 mm
nate, Sloc , an be written expli itly. Thus Sloc must be 20 mm
evaluated numeri ally by solving
1200
94 mm

||Xnew − Nold (Sloc )Xold || = 0 (18) 1000

Temperature [°C]
This is done by the Newton iterative method where the 800

initial value for the lo al oordinate is guessed.

600

7. Results and dis ussion 400

The measured and omputed values of temperatures 200

and for es are plotted in the Figures below. Fig. 15 - 17
show the temperatures for 353 MA and Fig. 18 to 19 show 0
0 100 200 300 400 500 600 700
the temperatures for SAF 2507. All the temperatures are Time [s]

measured during manufa turing step Heating 1 with the

thermo ouples pla ed a ording to Fig. 9. Figure 17: The measured temperatures (dashed lines) ompared with
The expansion for e versus time for 353 MA is shown in al ulated temperatures (full lines) at position a = 369 mm in 353
Fig. 20. The extrusion for e versus the ram displa ement MA.
for the same material is shown in Fig. 21. Figure 22 and
Fig. 23 give the expansion for e versus time and extrusion 1400
for e versus ram displa ement for SAF 2507. 10 mm
50 mm

7.1. Indu tion heating

1200
94 mm

The results from the indu tion heating pro ess show 1000

good agreement between measured and omputed values.

Temperature [°C]

For the material 353 MA the temperature is measured at 800

three dierent height and at three dierent depths as de-

s ribed in se tion 5.1. At the depth of 50 − 72 mm in the 600

workpie e, the measured and omputed values show max-

imum deviation.
400

200

1400
10 mm 0
0 100 200 300 400 500 600 700
20 mm Time [s]
1200
72 mm

1000
Figure 18: The measured temperatures (dashed lines) ompared with
al ulated temperatures (full lines) at position a = 638 mm in 353
Temperature [°C]

800 MA.

600
the approa h of Labridis and Dokopoulos [4℄ to linearize
400
the permeability is valid. This approximation makes it
possible to employ a harmoni formulation for non-linear
200 magneti material. This is important sin e a harmoni so-
lution with a onstant permeability is mu h faster than a
0 transient ele tromagneti non-linear solution [1℄.
In Fig. 18 it an be seen that the heating rate is greater
0 100 200 300 400 500 600 700
Time [s]

when the material is ferromagneti . When the material

Figure 16: The measured temperatures (dashed lines) ompared with
be omes paramagneti , the heating rate de reases. This
al ulated temperatures (full lines) at position a = 100 mm in 353 an be seen as a knee in the temperature graph that starts
MA. at dierent times for dierent depths, depending on when
the Curie temperature is rea hed.
For the material SAF 2507 the temperature is mea- Figure 19 shows the last 140 s of the heating simulation
sured in the middle of the workpie e at six dierent depths. of SAF 2507. From the gure it an be seen that the mea-
The temperature agreement for the duplex steel shows that sured and al ulated temperatures losest to the surfa e
8
(10 mm) are parallel during the power y ling. This shows before extrusion, ompared to 738 mm and 890 mm in the
that the heat ux boundary ondition is valid. simulations. The dieren e in length an be seen in Fig.
21. The temperature measurement during Heating 1 is
performed on the shorter billet.
1400
10 mm
20 mm
1200 4.5
30 mm
40 mm
64 mm 4
1000
78 mm
3.5
Temperature [°C]

800

Expansion Force [MN]

3

600
2.5

400 2

1.5
200

1
0
0 100 200 300 400 500 600 700 800 900 0.5
Time [s]

0
0 1 2 3 4 5
Time [s]
Figure 19: The measured temperatures (dashed lines) ompared with
al ulated temperatures (full lines) at position a = 473 mm in SAF
2507.
Figure 21: Cal ulated and measured for e for extrusion of 353 MA.

10 mm 35
20 mm
1300
30 mm
30
40 mm
64 mm
1250 78 mm 25
Temperature [°C]

Extrusion Force [MN]

1200 20

15
1150

10
1100

1050
660 680 700 720 740 760 780 800
Time [s] 0
0 200 400 600 800 1000
Ram Displacement [mm]

Figure 20: The measured temperatures (dashed lines) ompared with

al ulated temperatures (full lines) at position a = 473 mm in SAF Figure 22: Cal ulated (bla k line) and measured for e (blue line) for
2507. extrusion of 353 MA.

The predi tion of the initial peak for e during extru-

7.2. Expansion and extrusion sion is higher in the models than in the experiments. This
Experimental values were re orded for 14 tubes of 353 is probably due to errors in the temperature estimation at
MA. Only the peak for es ould be re orded during ex- the leading end. The ow stress of 353 MA has pronoun ed
pansion and were measured to be 4.2 ±0.5 MN and 4.6 temperature dependen e and a small hange in tempera-
±0.4 MN. This is in agreement with the model predi - ture will result in a large dieren e in stress. However, if
tions as shown in Fig. 20, although the se ond peak is on the 353 MA extrusion simulation is run using a homoge-
the lower side. neous initial billet temperature of 1200 ◦ C, the resulting
The 353 MA billets in the models were shorter than the peak for e is only 1.5 MN higher than the ow for e at
ones on whi h the for e measurements were arried out. steady-state. In the experiments the dieren e between
The experimental for e re ordings were performed on bil- initial peak and ow for e is approximately 9 MN as seen
lets with lengths of 810 mm before expansion and 960 mm in Fig. 21. This laries that a orre t front fa e temper-
9
ature of the billet is important in determining the initial 8. Con lusions
peak.
Figure 21 and Fig. 23 reveal that the start-up phase A ombined model of indu tion heating and expan-
of extrusion is not fully aptured in the model. The main sion/extrusion, onsisting of seven dierent FE-models,
reason is the un ertainties of the onditions at the begin- was developed in order to study the evolution of tempera-
ning of the extrusion pro ess. The glass-pad deforms in ture gradients in stainless steel billets for tube manufa tur-
onta t with the metal, making it di ult to predi t the ing. Simulations were arried out for an austeniti stain-
interfa e between the billet, the glass pad and the die. In less steel, Sandvik 353 MA, and for a duplex (austeniti /
the model this is negle ted and a onstant die shape and ferriti ) stainless steel, Sandvik SAF 2507. The indu -
a onstant fri tion oe ient is used. Furthermore, grav- tion heating models were validated by heating experiments
ity is negle ted in the axisymmetri model. In the real using thermo ouples pla ed in holes drilled at dierent
produ tion ase, the billet is pla ed at the bottom of the positions in the billet. The agreement between model
horizontal press and gravity will inuen e how the metal and experiments was good, both for the paramagneti
lls the ontainer. austeniti material and the ferromagneti duplex material.
The agreement for SAF 2507 onrms that the approa h to
linearize the permeability is valid. The expansion and ex-
3
trusion models were able to apture the for e development
during the pro esses for both materials. When omparing
2.5 to experimental re ordings of expansion for e versus time,
and extrusion for e versus ram displa ement, the results
were satisfa tory.
Expansion Force [MN]

1.5 A knowledgements
1
The authors are grateful to prof. Lars-Erik Lindgren
at the Division of Material Me hani s at Luleå Univer-
sity of Te hnology, Luleå, Sweden for programming the
mapping algorithm and to Christofer Hedvall and Berndt
0.5

Ferm at Sandvik Materials Te hnology, Sandviken, Swe-

0
0 1 2 3 4 5 den for their assistan e in the experimental work. The -
Time [s]
nan ial support from the Swedish Knowledge Foundation
and the Swedish Steel Produ ers' Asso iation is gratefully
Figure 23: Cal ulated (bla k line) and measured for e (blue line) for a knowledged.
expansion of SAF 2507

Referen es
35
[1℄ Clain, S., Rappaz, J., Swierkosz, J., Touzani, R., 1993. Numer-
i al modeling of indu tion heating for two-dimensional geome-
tries. Mathemati al Models and Methods in Applied S ien es
30 3 (6), 805822.
[2℄ Davies, E. J., 1990. Condu tion and indu tion heating. No. 0-
25 86341-174-6 in IEE Power Engeneering Series 11. Peter Peregri-
nus Ltd., London, UK.
Extrusion Force [MN]

[3℄ Hughes, K., Nair, K., Sellars, C., 1974. Temperature and ow
20
stress during the hot extrusion of steel. Metals Te hnology 1 (4),
161169.
15 [4℄ Labridis, D., Dokopoulos, P., May 1989. Cal ulation of eddy
urrent losses in nonlinear ferromagneit materials. IEEE Trans-
10
a tions on Magneti s 25 (3).
[5℄ Rudnev, V., Loveless, D., Cook, R., Bla k, M., 2003. Handbook
of Indu tion Heating. Indu toheat, In .
5 [6℄ Sadiku, M. N., 2001. Elements of Ele tromagneti s. Vol. 3. Ox-
ford University Press, In .
0 [7℄ Stratton, J. A., 2007. Ele tromagneti theory. IEEE Press, Pis-
0 200 400 600 800 1000 1200
Ram Displacement [mm] ataway, NJ.
[8℄ Tudbury, C. A., 1960. Basi s of Indu tion Heating. Vol. 2. John
F. Rider, In .
Figure 24: Cal ulated (bla k line) and measured for e (blue line) for [9℄ Wang, K., Chandrasekar, S., Yang, H. T., 1992. Finite-element
extrusion of SAF 2507 simulation of indu tion heat treatment. Journal of Materials
Engineering and Performan e 1.
[10℄ Zinn, S., Semiatin, S. L., 1988. Elements of Indu tion Heating.
Ele tri Power Resear h Institute, In .

10
A. Data for Sandvik 353 MA and Sandvik SAF
2507

6
x 10
1.3
353 MA
SAF 2507
1.2
Electric Conductivity [S/m]

1.1

0.9

0.8

0.7
0 200 400 600 800 1000 1200 1400
Temperature [oC]

950

Figure 25: The ele tri ondu tivity for Sani ro 39 and SAF 2507. 900 353 MA
SAF 2507
850
Specific Heat Capacity [J/Kg K]

800
35
750
353 MA
SAF 2507 700
30
650
Thermal Conductivity [W/mK]

600
25
550

500
20
450
0 200 400 600 800 1000 1200 1400
Temperature [oC]

15

Figure 27: The heat apa ity in onstant pressure for Sani ro 39 and
SAF 2507.
10
0 200 400 600 800 1000 1200 1400
Temperature [oC]

Figure 26: The thermal ondu tivity for Sani ro 39 and SAF 2507.

11