Master's Degree Thesis

ISRN: BTH-AMT-EX--2013/D16--SE

FEM Simulations of Induction

Hardening Process

Heng Liu

Department of Mechanical Engineering

Blekinge Institute of Technology
Karlskrona, Sweden
2013

Supervisors: Edin Omerspahic and John Lorentzon, SKF

Ansel Berghuvud, BTH
 FEM Simulations of Induction
Hardening Process
Heng Liu
Department of Mechanical Engineering
Blekinge Institute of Technology
Karlskrona, Sweden
2013
Thesis submitted for completion of Master of Science degree in Mechanical
Engineering with emphasis on Structural Mechanics at the Department of
Mechanical Engineering, Blekinge Institute of Technology, Karlskrona,
Sweden.

Abstract:
Induction heating is the process of heating an electrically conducting
object by electromagnetic induction, where eddy currents are generated
within the metal and resistance leads to Joule heating of the metal.
Heating is followed by immediate quenching. The quenched metal
undergoes a martensitic transformation, increasing the hardness of the
part. The process is widely used in industrial operations. In this thesis,
the Finite Element Method (FEM) simulations of the process have been
studied. In practice, this means that a coupling between the electro-
magnetic, thermal, mechanical and metallurgical phenomenon have
been modeled and studied in LS-DYNA. The simulation results have
been compared to experimental results from literature. The comparison
and the software’s simulation performance have been used to evaluate
the maturity of LS-DYNA to model the real process.
Keywords:
Induction Hardening, Induction Heating, Quenching, Modeling,
Simulations, Phase Transformations, FEM, LS-DYNA.
Acknowledgements
The research work was carried out at SKF AB, Manufacturing
Development Center during the spring and summer 2013. The thesis is the
concluding part of an engineering degree from Department of Mechanical
Engineering, Blekinge Institute of Technology (BTH), Karlskrona, Sweden.
First of all, I would like to express my sincere gratitude to my supervisors
at SKF, Edin Omerspahic and John Lorentzon. Without their support and
enthusiasm, the thesis will never come out. Their guidance, insight
conversations, timely advices, and comments have been a source of
strength to me.
Many thanks also to Marcus Lilja and Mikael Schill at DYNAMORE and
Vinayak Deshmukh and Johan Facht at SKF, who always helped me when I
was in trouble during the research work.
I would also like to thank Ansel Berghuvud, my Supervisor at Blekinge
Institute of Technology, who supported me a lot during this thesis work.
Finally, I would like to thank all friends and staff at SKF for their valuable
input to the thesis and their help.
Göteborg, September 2013
Heng Liu

1
Contents
Notation........................................................................................................ 3
1. Introduction ............................................................................................ 6
1.1. Background ...................................................................................... 6
1.2. Aim .................................................................................................. 9
2. Induction and the corresponding numerical background................ 12
2.1. Induction process - Maxwell equations ......................................... 12
2.1.1. Skin effect and skin depth .................................................... 14
2.1.2. Proximity effect .................................................................... 15
2.2. Numerical basis of the induction process ...................................... 15
2.2.1. FEM model for electromagnetic field .................................. 17
2.2.2. FEM model for temperature field ........................................ 19
2.2.3. FEM model for mechanical field ......................................... 21
2.2.4. Numerical procedure ............................................................ 23
2.3. Microstructures in numerical model .............................................. 26
2.4. Numerical determination of hardness ............................................ 29
3. Simulation (FEM) model ..................................................................... 31
3.1 Initial and boundary conditions ..................................................... 31
3.2 Meshing ......................................................................................... 33
3.3. Material properties ......................................................................... 34
3.3.1. Thermal and electromagnetic properties of the bar.............. 34
3.3.2. Mechanical and metallurgical properties of the bar ............. 37
3.3.3. Material properties of the inductor ....................................... 38
3.4. Limitations ..................................................................................... 38
4. Analysis and discussion of the simulation results ............................. 40
4.1. Magnetic results ............................................................................. 41
4.2. Thermal results .............................................................................. 44
4.3. Metallurgical results ...................................................................... 45
5. Evaluation of the results ...................................................................... 50
6. Conclusions ........................................................................................... 54
Reference ................................................................................................... 55
Appendix A: Miscellaneous results ......................................................... 57
Explicit mechanical solver ..................................................................... 57
Implicit mechanical solver with curve switching between heating and
cooling ........................................................................................... 58
Implicit mechanical solver with 2 successive processes – heating and
cooling enabled by INTERFACE_SPRINGBACK_LSDYNA .... 60

2
Notation

A Vector Potential [Tm]
2
a , u Acceleration [m/s ]

B Magnetic flux density [T]

b Material coefficient [-]

C Specific Heat [J/kg K]

Cc Capacitance [F]

Cp Chemical composition [%]

c Empirical grain growth parameter [-]

cd Damping coefficient [N s/m]

E Young’s Modulus [Pa]


E Electric field [V/m]

FLorentz Lorentz force [N]

f Frequency [kHz]

fD Damping force [N]

fl Inertial force [N]

fS Elastic force [N]

2
G Energy release rate [J/m ]
Gg Grain number [-]

H Magnetic field intensity [A/m]

Hv Hardness [-]
2
h Convection coefficient [W/(m K)]
I Current [A]
 2
j Current density [A/m ]
 2
js Source current density [A/m ]

k,  Thermal conductivity [W/(m K)]

3
 ks Linear stiffness [N/m]

L Inductance [H]
m Mass [kg]
temperature of the initial martensitic
Ms [K]
transformation
n Normal direction of the boundary [-]
p Phase proportion [-]

peq   Phase proportion calculated at thermodynamic

[-]
equilibrium
Q Internal heat generation rate per unit volume [-]

Qk Activation energy [J]

q Charge [C]

qsz Heat flux vector [-]

R Resistance [Ω]

Rb Radius [m]

Ru Universal gas constant [J/(mol K)]

r Position [m]
T , Temperature [K]

Tw Surface temperature of the solid [K]

T Fluid temperature [K]

t Time [sec]
u Displacement [m]
u , v Velocity [m/s]
V Voltage amplitude [V]
Vr Cooling rate at 700 ℃ [K/s]

Xk Actual phase [-]

xk True amount of phase [-]

 Material dependent constant [-]
 Stress [Pa]

B Stefan Boltzmann constant 2 4

[W/(m K )]

4
 C Electric conductivity [1/(Ωm)]
 Strain [%]
0 Permittivity of free space [F/m]

e Surface emissivity [-]

 Scalar Potential [V]

 Skin depth [m]
 Magnetic Permeability [H/m]

0 Permeability of free space [H/m]

r Relative permeability [-]

 Density 3
[kg/m ]

c Total Charge density [C/m3]

e Electrical resistivity [Ω·m]

 Shear stress [Pa]

 R   Delay time of the transformation [sec]

 Poisson's ratio [-]

3
 Volume [m ]
   Boundary surface of volume 2
[m ]
 Pulsation [rad/s]

5
1. Introduction
1.1. Background
The induction hardening is one of the methods for heat treatment of steel
workpieces. The induction hardening can be used for both through-
hardening and to selectively harden areas of a part or assembly. When the
method is used to harden only the surface of the parts, it has been applied to
various machine parts such as automobile components and toothed gears
[1].
The classic method of hardening contains first heating to an austenitic state
(austenite has a Face Centre Cubic – FCC atomic structure) and then
cooling rapidly. Let us assume the initial phase being ferritic-pearlitic.
Ferrite has a Body Centre Cubic structure (BCC) which can hold very little
carbon; typically 0.0001% at room temperature. It can exist as either alpha
or delta ferrite. Pearlite is a mixture of alternate strips of ferrite and
cementite in a single grain. The name for this structure is derived from
pearl appearance, seen under a microscope. A fully pearlitic structure
occurs at 0.8% Carbon.
During heating, see Figure 1.2, two processes occur. Firstly, the cementite
starts to dissolve and the cementite particles to shrink. When the
temperature rises above a critical value, the ferrite starts to transform to
austenite. Austenite formation and cementite dissolution occur faster the
higher the temperature. The structure is fully austenitic above the A3 (Ac3)
or Accm line (the upper line in the Iron Carbon Diagram), Figure 1.1.

6
 Figure 1.1. Iron Carbon Diagram

Figure 1.2. Induction heating of a part.

During cooling from the austenitic state, several different phase

transformations may take place depending on how fast the cooling process
is. When steel is cooled sufficiently rapidly, other structures do not have
sufficient time to form and the austenite can be retained at low
temperatures, since the diffusion-dependent transformations proceed
slowly. When the temperature is sufficiently low, the tendency of austenite
to be transformed becomes so strong that the transformation takes place
without diffusion. Such a transformation is called diffusionless and can, in
principle, occur with two different mechanisms, namely massive and
martensitic transformation [2].

Figure 1.3. Quenching (cooling) of a part.

7
In a martensitic transformation, FCC structure of austenite rapidly changes
to BCC leaving insufficient time for the carbon to form pearlite. This
results in a distorted structure that has the appearance of fine needles. Only
the parts of a section that cool fast enough will form martensite; in a thick
section it will only form to a certain depth, and if the shape is complex it
may only form in small pockets. The hardness of martensite is solely
dependant on carbon content; it is normally very high, unless the carbon
content is exceptionally low. The martensitic transformation is of great
practical significance, since it is the martensite which gives steel its high
degree of hardness and strength.
In the induction hardening of our interest, the surface of the workpiece is
heated up over the austenitization temperature by the induction heating,
Figure 1.2, and transformed from the ferritic and pearlitic structure, Figure
1.4 A, to the austenite structure, Figure 1.4 B. The heating process is then
followed by immediate quenching process, Figure 1.3, and the surface of
the workpiece is transformed from the austenitic to the martensitic phase,
Figure 1.4 C, and thereby hardened. The heating condition for the
induction hardening can be determined experimentally or empirically for
the workpiece of any shape [1].

Figure 1.4. Specimen microstructures of normalized steel: A) Ferritic-

Pearlitic, B) Austenitic and C) Martensitic

8
Induction heating is the process of heating an electrically conducting object
by electromagnetic induction, where eddy currents are generated within the
metal and resistance leads to Joule heating of the metal [6].
This process is widely used in industrial operations due to its high
efficiency, precise control and more environmentally friendly properties
[3]. The induction heating has some characteristics compared to the
traditional heating methods (such as furnace heating):
 It has a precise depth of heating and the heating zone, which is
easier to control.
 It is easy to implement high power density, fast heating, high
efficiency and low energy consumption.
 It is easy to control the high heating temperature.
 The conduction and infiltration of the heating temperature will be
from the surface to the interior.
 There are no penetrating impurities since non-contact heating
method is used.
 The burned part on the workpiece is smaller.
 The process is somewhat eco-friendly.
 It is easy to accomplish the automation of heating process.
The quenching part of an induction hardening process is also an important
part. Cooling rates must be rapid in order to avoid softer undesirable
structures such as pearlite and bainite. Due to its importance, the cooling
portion of the induction hardening process deserves careful consideration,
particularly when specifying new induction equipment and processes.
Process parameters must be precisely controlled to assure consistent heat
treatment results. Excessive variation in these parameters will cause
undesirable or inconsistent process results including problems with case
depth, hardness, pattern and distortion [4]. Water quench has been used for
the problem in this thesis.

1.2. Aim
Let us go to the main objective of this work. Although the induction
hardening process has many advantages, the design of it, which is usually
based on experiments, can be tiresome, time-consuming and expensive.

9
Luckily, the fast development of the computer technology makes it possible
to model the induction heat treatment process with numerical tools,
particularly with Finite Element Method (FEM). Nowadays, a lot of
engineers pay attention to this area.
There are many FEM modeling works regarding either the heating or
quenching heat treatment in the literature. However, numerical models of
the integrated heat treatment, i.e. both the induction heating and quenching,
are still gaining ground [5]. Induction hardening is a complex physical
process, which has contributions from electrical, magnetic, thermal,
mechanical and metallurgical processes. It is obvious that the complexity of
the phenomena – including phase transformation and heat exchange makes
the FEM analysis heavy and difficult.
Different FEM softwares have been used for numerical studies of the
induction hardening process. In this study, LS-DYNA has been used for
simulations. The electromagnetic field, the eddy current and the
temperature field have been calculated with the FEM and Boundary
Element Method (BEM). In fact, FEM is used for conducting pieces only,
while the FEM-BEM coupling handles heat/electromagnetic propagation
through air, thus no air mesh is needed. The main study included:
 The mathematical description and the modeling of the induction
heat treatment process.
 Solving the induction-hardening-modeling key technical issues.
 Simulating the induction hardening process with the existing
commercial software LS-DYNA.
 Comparing the results of the simulation with the literature values,
and evaluating the software’s capability.
In short, the aim was to develop a simulation model for induction heat
treatment with coupling between the electro-magnetic, thermal, mechanical
and metallurgical phenomenon. The simulation results have been compared
to literature results for evaluation of the FEM software LS-DYNA as a tool
for simulation of induction hardening.
Here follows the model selected from a literature source [5]: the induction
heating and cooling of cylindrical workpiece. The experimental setup is
made of three parts, the coil, the bar and the cooling tool, Figure 1.5.

10
Figure 1.5. The experimental set-up.

11
2. Induction and the corresponding
numerical background
2.1. Induction process - Maxwell equations
The basic model is shown in Figure 2.1.

Figure 2.1. Induction heating principle.

The partial differential equations are used to solve the electromagnetic field
distribution.
In order to define the equations solved by the electromagnetic solver in LS-
DYNA, we start with the Maxwell equations [7]:


  B
 E   , (2.1)
t

   E
  H  j  0 , (2.2)
t

B  0, (2.3)

12
   
E  , (2.4)
0
  
j  E  j s , (2.5)
 
B  0 H , (2.6)

  
where E is electric field, B is magnetic flux density, t is time, H is

magnetic field intensity, j is current density,  0 is permittivity of free

space,  is total charge density,  is electric conductivity, j s is source
current density and  0 is permeability of free space.
The eddy current approximation used here implies a divergence-free
 current
E
density and no charge accumulation thus resulting in  0  0 and   0 .
t
Equations (2.2) and (2.4) in the "eddy current approximation" give:

  
 H  j , (2.7)
 
E  0, (2.8)
 
 j  0. (2.9)


The divergence condition given by equation (2.3) allows writing B as

  
B    A, (2.10)


where A is the magnetic vector potential [8]. Equation (2.1) then implies
that the electric field is given by


  A
E    , (2.11)
t

13
where  is the electric scalar potential.
 
Equation (2.10) leaves a mathematical degree of freedom to A (if A is
 
transformed to a given A   then Equation (2.10) remains valid).
Therefore, the introduction of a gauge i.e. a particular choice of the scalar
and vector potentials is needed. Gauge choosing denotes a mathematical
procedure for coping with redundant degrees of freedom in field variables.
The gauge chosen here is the "generalized Coulomb" gauge

 

 A  0 . (2.12)

Equations (2.5), (2.9), (2.11) and (2.12) give

 

   0 . (2.13)

Equations (2.5), (2.7), (2.11) and (2.10) give


A   1     
       A     j s . (2.14)
t  

Equation (2.13) and Equation (2.14) are the two equations constituting the

system that will be solved, where A and  are the two unknowns of the
problem [7].

2.1.1. Skin effect and skin depth

Skin effect is the tendency of an alternating electric current (AC) to become

distributed within a conductor such that the current density is largest near
the surface of the conductor, and decreases with greater depths in the
conductor [9].
Skin effect is associated with the current flowing mainly at the "skin" of the
conductor, at an average depth called the skin depth. The skin depth is

14
defined as the depth at which the electromagnetic field in a conducting
material has decreased to 0.37 of its value just outside the material, which
describes the electric and magnetic fields. The formula for the skin depth is
given by

2 2 
   503 , (2.15)
 (2f )  0  r r f

where  is the skin depth, f is the frequency,  is the average electrical

resistivity and  r is the average relative permeability.

2.1.2. Proximity effect

A changing magnetic field will influence the distribution of an electric

current flowing within an electrical conductor by electromagnetic
induction. When an alternating current flows through an isolated conductor,
it creates an associated alternating magnetic field around it. The alternating
magnetic field induces eddy currents in adjacent conductors, altering the
overall distribution of current flowing through them – the distribution of
current within the conductor will be constrained to smaller regions.
Subsequently, the resistance is increased in those regions. The resulting
current crowding is termed the proximity effect. Usually the current is
concentrated in the areas of the conductor furthest away from nearby
conductors carrying current in the same direction [10].
Thus, since in our case the inductor is a coil, the maximum current density
will be at the inner side of the coil [3]. So the inner side of the coil will be
used to heat the workpiece, which will get faster temperature increase and
will be more efficient.

2.2. Numerical basis of the induction process

All the physical phenomena encountered in engineering mechanics are
modeled by differential equations. Usually, it is difficult to obtain accurate
analytical solution of the differential equation However, the numerical
solution could be calculated, but only when boundary conditions and initial

15
conditions under specific situations were given. The following numerical
methods are used to model the induction process in LS-DYNA:
 Finite Element Method:
The FEM is today a powerful (often the most powerful) tool for numerical
solution of any differential equation, whether this arises from structural
mechanics, fluid mechanics, thermodynamics, biology, ecology or any
other field of science [11].
The finite element method is a numerical approach by which general
differential equations can be solved in an approximate manner [12]. A
domain of interest is represented as an assembly of finite elements. The
FEM is useful for problems with complicated geometries, loadings, and
material properties where analytical solutions cannot be obtained [13].
The main steps in the general FE formulation and solution of a physical
problem are [11]:
o Establish the strong form of the governing differential
equation.
o Transform this differential equation into the weak form.
o Choose trial functions for the unknown function, that is,
choose element type(s) and mesh the solution domain.
o Choose weight functions and establish the system of
algebraic equations for each element (element equations).
o Assemble these element systems into the global system of
algebraic equations.
o Introduce boundary conditions into the global system of
algebraic equations.
o Solve the system of algebraic equations and present the
results or use them for further calculations.
 Boundary Element Method:
The boundary element method (BEM) is a numerical computational method
of solving linear partial differential equations which have been formulated
as integral equations. BEM attempts to use the given boundary conditions
to fit only boundary values into the integral equation. Once this is done, the
integral equation can then be used again to calculate numerically solution at
any desired point in the interior of the solution domain. The boundary

16
element method is often more efficient than other methods, including FEM,
in terms of computational resources for problems where there is a small
surface/volume ratio. Conceptually, it works by constructing a "mesh" over
the modeled surface. However, for many problems boundary element
methods are significantly less suitable and efficient than volume-
discretization methods [14].
In numerical computations of the problem in this thesis, with LS-DYNA,
FEM is used for conducting pieces only, while the FEM-BEM coupling
handles heat/electromagnetic propagation through air.

2.2.1. FEM model for electromagnetic field

In LS-DYNA, equation (2.13) is projected on the W 0 forms (0-forms are

continuous scalar basis functions that have a well defined gradient, the
gradient of a 0-form being a 1- form) and equation (2.14) is projected on

the W 1 forms (1-forms are vector basis functions with continuous
tangential components but discontinuous normal components). They have a
well defined curl, the curl of a 1-form being a 2-form) giving after
integrating by part the following weak formulations [15]:

 
   W 0 d  0 , (2.16)

A  1 1    1
   t  W d       A   W d , (2.17)
 

   

      W 1d    n  (  A)  W 1d


where d an element of volume  and    the surface of  with n
outer normal to  .
 
The  and A decompositions on respectfully W 0 and W 1 give:

   i wi0 , (2.18)
 
A   ai wi1 . (2.19)

17
 
When replacing  and A in equation (2.16) and (2.17) by (2.18) and
(2.19), one gets:

S 0 ( )  0 , (2.20)
a 1
M 1 ( )  S 1 ( )a   D 01 ( )  Sa , (2.21)
t 

where
 the stiffness matrix of the 0-forms is given by

 
S 0 ( )(i, j )      Wi 0  W j0 d , (2.22)

 the mass matrix of the 1-forms is given by

 
M 1 ( )(i, j )    Wi1 W j1d , (2.23)

 the stiffness matrix of the 1-forms is given by

1 1    
S 1 ( )(i, j )    (  Wi1 )  ( W j1 )d , (2.24)
 

 the derivative matrix of the 0-1-forms is given by

 
D 01 ( )(i, j )    Wi 0  (W j1 )d , (2.25)

 the outside stiffness matrix is given by

18
 1 1   
 

S ( )(i, j )    n  (  Wi1 ) W j1d ,
 
(2.26)


where  is the magnetic permeability, n is the normal vector,  is the
volume and    is the boundary surface of volume.
Equation (2.20) and (2.21) form the FEM system with  and a being the
unknowns. From this system only the outside stiffness matrix cannot be
directly computed. The calculation of this matrix will be made possible
through the definition of the BEM system [7]. The BEM system is used for
the air, and will not be shown in this report. More information about it
could be found in [7].

2.2.2. FEM model for temperature field

The steady state or transient temperature field on three dimensional

geometries can also be solved by LS-DYNA. Material properties may be
temperature dependent and either isotopic or orthotropic. A variety of time
and temperature dependent boundary conditions can be specified including
temperature, flux, convection and radiation. The implementation of heat
conduction into LS-DYNA is based on the work of Shapiro [16].
The differential equations of conduction of heat in a three-dimensional
continuum is given by


c  k ij , j ,i  Q , (2.27)
t

where    ( xi , t ) is temperature,    ( xi ) is density, c   c  ( xi ,  ) is

the specific heat, k ij  kij ( xi ,  ) is thermal conductivity, Q  Q( xi ,  ) is
internal heat generation rate per unit volume  .
The boundary conditions are

   s on 1 , (2.28)

19
 kij , j ni     on 2 . (2.29)

Initial conditions at t0 are given by

    0 ( xi ) at t  t0 , (2.30)

where xi  xi (t ) are coordinates as a function of time,   is prescribed

temperature on 1 and ni is normal vector to 2 .
Equations (2.27-2.30) represent the strong form of a boundary value
problem to be solved for the temperature field within the solid continuum
[16].
The finite element method provides the following equations for the
numerical solution of equations (2.27-2.30):

 Cn 
 t  H n  n1   n   Fn  H n  n  , (2.31)
 
C    Cije     N i cN j d , (2.32)
e e e

 
H    H ije       T N i KN j d   N i N j d , (2.33)
e e e e 
 
F    Fi e      N i q g d   N id , (2.34)
e e e e 

where  ,  and  are the parameters that are different when using
different methods, like Crank-Nicolson, Galerkin and so on. The parameter
 is taken to be in the interval [0,1]. C  , H  and F  are the element
stiffness, load and boundary matrices, respectively, N is the element shape
functions, q g is the heat flow, K is the thermal conductivity tensor.

20
The boundary conditions for temperature, flux, convection and radiation are

T  f ( x, y, z , t )
 T 
 k  q sz
 n
 T , (2.35)
 n  h(Tw  T )

 T  F (T 4  T 4 )
 n
12 1 2

where T is the temperature, k is the thermal conductivity, n is the normal


direction of the boundary, qsz is the heat flux vector, h is the convective
heat transfer coefficient, Tw is the surface temperature of the solid, T is
the fluid temperature,  is the emissivity,  is the Stefan Boltzmann
constant.

2.2.3. FEM model for mechanical field

The equations that govern analyses of the behavior of a solid continuum are
those of momentum conservation, i.e. the equations of motion. For an
analysis of small deformation of a solid continuum, these are (in tensor
form) [17]

 ij , j  bi  ui , (2.36)


where ij is the Cauchy stress tensor, bi the body force vector per unit
volume,  the density, and ui the displacement vector.
To establish a weak form from the strong one, we multiply (2.36) by an
arbitrary velocity, i.e. the test function, vi and integrate over the region  .
By introducing two boundary conditions, ui  ui* on u , and  ij n j   i on
 , where v  0 on u , the above differential equation in the weak form
[17] is given as

21
 v

i, j  ij d   vi ui d   vi bi d   vi i d .
  
(2.37)

To perform the FE discretization of the weak form (2.37) means to divide

the continuum volume,  , into sub-elements, where the displacement field
in every element is approximated by shape functions, N I ( x ) , and nodal
displacements, uiI (t ) , that is summation of their products [17]:

ui ( x, t )   uiI (t ) N I ( x) . (2.38)

By approximating the test functions with the same shape functions

(Galerkin method), we obtain:

vT ( f int  Mu  f ext )  0,

N I
f int   f (int
e)    BT d, B jI  ,
(e)
x j
M   M (e)    N
T
Nd, (2.39)
(e)


 
f ext   f (ext  N T bd  N Td ,
e)    ( e ) ( e ) 
 


which must hold for an arbitrary v , and which puts the FE equation in
order:

Mu  f ext  f int . (2.40)

For a linear material,   C , the FE equation that emerges is

22
 Mu  Ku  f ext , K   K ( e )    B CB d .
T T
(2.41)
(e)

2.2.4. Numerical procedure

For the induction hardening process, three different analyses have been
combined in one numerical procedure: mechanical, thermal-metallurgical
and electromagnetic (EM) computations. They are solved fully transiently.
Boundary conditions and material properties beside one unique geometric
model were required by each of them.
What is necessary to mention is that some characteristics of the material are
interdependent. The electric conductivity, for instance, depends on the
temperature. In addition, all thermal properties depend on the temperature
[18]. The variation of the properties with the temperature makes the system
to be non-linear.
There is a high coupling grade between thermal and EM equations, because
the electrical and magnetic properties laws depend on temperature. When
the initial temperature is known, the eddy current value is calculated, and
then used to compute the heat generated by the Joule effect [5]. At each
time step the convergence is checked. Until a steady state between the heat
and the temperature field is reached, the temperature value will be
recalculated for each magnetic sub-step.
EM solver can be coupled with the thermal and mechanical solvers in order
to take full advantage of their capabilities [7]. Both the thermal and the EM
solver run with implicit time integration. For mechanical solver, there are
two time integration methods, of explicit and implicit type.
Explicit and implicit methods are numerical schemes for obtaining
numerical solutions of time-dependent ordinary and partial differential
equations, as is required in computer simulations of physical processes.
Explicit methods calculate the state of a system at a later time from the state
of the system at the current time, while implicit methods find a solution by
solving an equation involving both the current state of the system and the
later one [19]. Here follows the difference between explicit and implicit
methods:
 Implicit method
o More accurate

23
 o It has large time step increment
o Convergence of each load step can be controlled to avoid
error accumulation
o Iteration may not converge
 Explicit method
o Less accurate
o It has small time step
o There is error accumulation and the error is difficult to
estimate
o Iteration converges
However, the implicit type has been governing the mechanical solver for
the induction process in this thesis.
Now, let us go back to the couplings. For the electromagnetic and structure
interaction, both the mechanical and the EM solver have distinct time steps.
By linear interpolation, the EM fields are evaluated at the mechanical time
step. The two solvers will interact at each electromagnetic time step. The
EM solver will communicate the Lorentz force to the mechanical solver [7]
resulting in an extra force in the mechanic equation

Du
      f ext  FLorentz , (2.42)
Dt

where  is total charge density,  is electrical conductivity, f ext is the

external force, while FLorentz is the Lorentz force. In turn, the displacements
and deformations of the conductors are returned by the mechanical solver.
When it comes to the thermal coupling, at each electromagnetic time step,
the EM solver will communicate the extra Joule heating power term and the
thermal solver will communicate the temperature.
Figure 2.2 shows the interactions between the different solvers in LS-
DYNA.

24
 Figure 2.2. Interactions between the different solvers.

For the induction heating analyses, the solver works the following way: it
assumes a current which oscillates very rapidly compared to the total time
of the process. The solver works in the time domain and not in the
frequency domain, in order to easily take into account coil/workpiece
motion as well as the time evolution of the EM parameters. An EM time
step must be compatible with the frequency (such that there are at least a
few dozens of steps in the period of the current). In practice, this means that
a full eddy-current problem is solved on a quarter-period with a "micro"
EM time step, see Figure 2.3. The number of these "micro" steps in a
quarter period can be specified by a software user.
An average of the EM fields during this half-period and the joule heating
are computed. Then it is assumed that the properties of the material do not
change for the next periods of the current. These properties depend mostly
on the temperature, therefore the assumption can be considered accurate as
long as the temperature does not change too much. No EM computation is
done during these periods, only the averaged joule heating power is given
to the thermal solver. But as the temperature changes, and thus the
electrical conductivity, the EM fields need to be updated accordingly, so
another full eddy current resolution is computed for a half-period of the
current giving new averaged EM fields and an update of the Joule heating
power [7].

25
 Figure 2.3. Inductive heating time stepping.

On the other hand, the latent heat of phase transformations couples the
thermal and metallurgical parts. In the heating process, phase
transformation will happen (for this particular material – C45) from ferrite
and pearlite to austenite at austenitization temperature, while in the cooling
process, phase transformation will happen from austenite to martensite,
provided that the cooling rate is enough high.

2.3. Microstructures in numerical model

There are five microstructures considered in the numerical model: bainite,
ferrite, pearlite, martensite and austenite. In LS-DYNA, the phase
transformation during heating is calculated by re-austenitization model
according to Oddy et al. [20]. The grain growth will be activated when the
temperature exceeds a threshold value T,

B
T , (2.43)
A  ln(GRCM a GRCC b )

where A, B, a, b are the grain growth parameters, GRCM is the grain

growth parameter with respect to the concentration of metals in the
workpiece (B+Co+Mo+Cr+Ni+Mn+Si+V+W+Cu+Al+As+Ti), and GRCC

26
is the grain growth parameter with respect to the concentration of non-
metals in the workpiece (C+P).
When the temperature exceeds the threshold value, the rate equation for the
grain growth is

Q
k  RT
g  e . (2.44)
2g

In this equation, Q is the grain growth activation energy, divided by the

universal gas constant R, while k is the growth parameter.
When the grain growth is activated, the rate equation for the re-
austenitization is given as

n 1
xeu xeu  xa
x a  n(ln( )) n
( ), (2.45)
xeu  xa  (T )
where
 (T )  c1 (T  Ts ) c . 2
(2.46)

n is the grain growth parameter for the austenite formation, c1 and c2 are the
empirical grain growth parameters, Ts is the start temperature, xeu is the
eutectoid fraction and xa is the result of a normalized eutectoid austenite
fraction multiplied by the eutectoid fraction.

The phase distributions during cooling (Li et al. [21]) are calculated by
solving the following rate equation for phase transitions (k = 2 for ferrite, k
= 3 for pearlite and k = 4 for bainite):

X k  g k (G , C , Tk , Qk ) f k ( X k ), k  2,3,4 , (2.47)
where
f k ( K k )  X k0.4 ( X k 1)
(1  X k ) 0.4 X , k  2,3,4 ,
k
(2.48)

27
and
X k  xk / xeq . (2.49)

In the Equations (2.47), (2.48) and (2.49), G is the grain size number, C is
the chemical composition, Tk is the temperature, Qk is the activation
energy, X k is the actual phase and xk is the true amount of phase.
The true amount of martensite, (k = 5), is modeled by using the true amount
of the austenite left after the bainite phase,

x5  x1 (1  e  ( MS T ) ) , (2.50)

where x1 is the true amount of austenite left for the reaction, α is a material
dependent constant and MS is the start temperature of the martensitic
reaction.
The start temperatures are calculated with respect to the chemical
composition:

Ferrite:
FS ( K )  1185  203 C  15.2 Ni  44.7 Si  104V  31.5Mo  13.1W
,
 30Mn  11Cr  20Cu  700 P  400 Al  120 As  400Ti
(2.51)
Pearlite:
PS ( K )  996  10.7 Mn  16.9 Ni  29 Si  16.9Cr  290 As  6.4W ,
(2.52)
Bainite:
BS ( K )  910  58C  35Mn  15 Ni  34Cr  41Mo , (2.53)
Martensite:
MS ( K )  812  423C  30.4Mn  17.7 Ni  12.1Cr  7.5Mo  10Co  7.5Si .
(2.54)

28
The below figure is used to define the activation energy divided by the
universal gas constant for the diffusion reaction of the austenite-ferrite
reaction, the austenite-pearlite reaction and the austenite-bainite reaction.

Figure 2.4. CCT diagram for C45 material [5].

2.4. Numerical determination of hardness

Hardness can be predicted in the heat-altered zone with the knowledge of
the thermal and metallurgical history. The rule of mixtures is used to
calculated the hardness, which is

Hv  X M Hv M  X B Hv B  ( X F  X P ) Hv F  P , (2.55)

where Hv is the hardness in Vickers; X M , X B , X F , and X P are the

volume fraction of martensite, bainite, ferrite, and pearlite, respectively;
and Hv M , Hv B , and Hv F  P are the hardness of martensite, bainite, and the
mixture of ferrite and pearlite, respectively. Empirically based formulas
developed by Maynier et al. [22] were used in the study for the calculation
of those hardnesses as functions of steel composition and cooling rate:

29
 Hv M  127  949C  27 Si  11Mn  8 Ni  16Cr  21log Vr ,
(2.56)
Hv B  323  185C  330Si  153Mn  65 Ni  144Cr  191Mo
,
 (89  53C  55Si  22Mn  10 Ni  20Cr  33Mo) log Vr
(2.57)
Hv F  P  42  223C  53Si  30Mn  12.6 Ni  7Cr  19Mo
,
 (10  19Si  4 Ni  8Cr  130V ) log Vr
(2.58)

where Vr is the cooling rate at 700℃ in degrees Celsius per hour.

One thing that needs to be mentioned is in order to obtain valid results – the
element proportions must be located within the following chemical
composition ranges: 0.1% < C < 0.5%, Si < 1%, Mn < 2%, Ni < 4%, Cr <
3%, Mo < 1%, V < 0.2%, Cu <0.5%, and 0.01% < Al < 0.05% [5].
An error with a standard deviation of about 20 Vickers is foreseen [22,23]
in Equations (2.56) and (2.57).

30
3. Simulation (FEM) model
In this chapter, the modeling of induction heating and cooling of a
cylindrical workpiece (a bar) is presented. The bar is made of C45 steel
with the radius of 10 mm and the length of 200 mm. The inductor is made
of copper with one turn, the inner ring radius of 15 mm, and the rectangular
cross section of 10 x 15 mm. The cooling tool has a constant temperature of
20˚C, the shell radius of 15 mm, and the length of 200 mm. Both the
inductor and cooling tool move together with the same speed rate of 20
mm/s.
The LS_DYNA FEM model of the inductor ring with a rectangular section,
the pipe standing for the cooling tool (sprouting water inside), and the bar is
shown in 3-dimensional Figure 3.1.

Figure 3.1. FEM model.

3.1 Initial and boundary conditions

At the beginning, the temperature of all parts is the same as the room
temperature, which is 20˚C.
A contact (CONTACT_AUTOMATIC_SURFACE_TO_SURFACE) is
used to model the thermal contact between the workpiece and the cooling
tool defining the convection coefficient of 5500 W/m2K. Radiation and
convection parameters of the air-workpiece and air-coil are also considered

31
in the calculation (BOUNDARY_CONVECTION_SET and
BOUNDARY_RADIATION_SET), see Table 3.1.

Table 3.1. Radiation and convection parameters.

Surface emissivity Stefan Boltzmann constant W/m2K4 Convection coefficient W/m2K
0.9 5.67E-08 5

The function that describes the voltage trend at the extremes of the inductor
is represented by Eq. (3.1).

VV (tt , t )  V0 (tt ) * cos(t   ) , (3.1)

where tt is the time on the thermal scale and t is the time on the
electromagnetic scale (this diversification is necessary because of the
coupling between the two different analyses,  is the pulsation and  is
the phase. In particular,   2f , where f is the frequency (being 27 KHz).
V0 (tt ) is the amplitude of the voltage (being 31.048 V).
In order to find the optimal time step, simulations with different time steps
have been done. Below follows the figure showing chosen-element
temperatures as the functions of time for four different time steps.
Considering both the calculation time and the accuracy, the conclusion was
that the time step of 0.1 s was the optimal choice.

32
 Figure 3.2. Time step optimization.

3.2 Meshing
The model is made in three dimensions as shown in Figure 3.3. Some
modeling details are different from reality, still not having significant
influences on results: solid cross section of the coil instead of the hollowed
one, and open coil instead of the closed one (only the open coil can indicate
input and output segments of the current).

Figure 3.3. Assembled and reassembled FEM model.

33
The inductor and the workpiece have been modeled with constant stress
solid elements (type nr. 1 in LS-DYNA), while the element type of the
cooling tool is the shell of Belytschko-Tsay type (nr. 2 in LS-DYNA).
Meshing of the coil and cooling tool is simpler and the element size is
bigger. For the coil, the solid element size is 2 x 1.5 x 1.4 mm in average,
and the total number of elements is 14371. For the cooling tool, the element
size is 2.35 x 6.67 mm, and the total number of elements is 600. For the
workpiece, in order to save the calculation time, considering the skin effect,
the solid element size is bigger at the middle axis than on the outside. The
critical outside element size (governing the time stepping) is 0.48 x 0.52 x 2
mm, and the total number of elements on the whole bar is 567 566.

3.3. Material properties

3.3.1. Thermal and electromagnetic properties of the bar

The high-quality structural steels with carbon content at least 0.45% are
mostly used in induction heat treatment. They are mainly used in the shafts,
such as crankshafts, camshafts and so on. C45 steel is the most widely used,
because the carbon content is normal and high hardness can be obtained.
C45 steel is also used in the example in this thesis. Its chemical
composition is shown below.

Table 3.2. Nominal chemical composition of ISO C45 steel.

C Si Mn S P Cr Mo Ni Al Cu Fe
0.45 0.25 0.65 0.025 0.008 0.4 0.1 0.4 0.01 0.17 Bal.

In LS-DYNA, currently it is not possible to include material property

diversity of different phases to the material model called "MAT_UHS" (to
utilize different thermal conductivities and specific heat for the different
phases). So the curves of the thermal conductivity and specific heat, shown
in Figures 3.4 and 3.5 are used for all phases.

34
 Figure 3.4. Thermal conductivity for the C45 steel.

Figure 3.5. Specific heat for the C45 steel.

So far, in LS-DYNA, it is not possible to include any temperature

dependence for permeability. Therefore, the average relative permeability is
given by the following equation considering Figure 3.6,

PS  (T2  T1 )  ( PS  PE )  (T3  T2 ) / 2  ( x  T3 )  PE
r 
x  T1
30 * (823.15  293.15)  (31) * (1023.15  823.15) / 2  ( x  1023.15)

x  293.15
x  17976.85

x  293.15
, (3.2)

35
where T1 (being 293.15 K) is the initial temperature of the workpiece, T2
(being 823.15 K) is the temperature when relative permeability starts
changing, T3 (being 1023.15 K) is the temperature when relative
permeability stops changing, PS (being 30) is the relative permeability
corresponding to T1 , PE (being 1) is the relative permeability
corresponding to T3 and x is the assumed peak temperature.

Figure 3.6. Relative permeability as function of temperature.

The procedure is to assume different peak temperature, calculate the

relative permeability in the air, and then re-calculate the temperature,
aiming in minimizing the difference between the assumed and re-calculated
peak temperature. So this procedure gives the relative permeability in air of
15.854.
The electric conductivity of C45 steel is given as a function of temperature
below.

36
 Figure 3.7. Electric conductivity of C45 steel.

3.3.2. Mechanical and metallurgical properties of the bar

The Poisson's ratio and the Young's modulus are constant, 0.3 and
2.050e+5 MPa, respectively. An average value of 7600 kg/m³ is used for
the density. When it comes to yield stress – effective plastic strain
properties, there are five figures for different plastic hardening curves
(austenite, ferrite, pearlite, bainite and martensite, respectively).

Figure 3.8. Yield stress vs. effective plastic strain for different phases.

37
The start temperature for ferritic, pearlitic, bainitic and martensitic
transformations are 1039 K, 996 K, 837 K, and 587 K, respectively. And
the temperature for instantaneous transformation of ferrite to austenite is
1185 K. The initial phase proportion is 40% pearlite and 60% ferrite.
The latent heat for the transformation of ferrite, pearlite and bainite into
austenite is 595.36 J/m3. And the latent heat for the decomposition of
austenite into martensite is 661.51 J/m3.

3.3.3. Material properties of the inductor

In order to simplify the model, the parameters of copper have been

considered temperature independent [5].

Table 3.3. Material parameters of the inductor.

Specific
Thermal Electric Young's
heat Density Poisson's Permeability
conductivity conductivity modulus
[J/(kg [kg/m³] ratio [H/m]
[W/(m K)] [S/m] [MPa]
K)]
Inductor 386 400 8930 5.9E7 0.3 2.050E5 12.57E-7

3.4. Limitations
Some limitations have to be considered in the analysis. Here is the list.
 The density can only be described by a constant value. The impact
of this will be that the accuracy will decrease if density changes
with the temperature.
 2D axisymmetrical FEM modeling is still not available in LS-
DYNA. The impact of this is that the 3D-simulations take longer
time.
 The thermal conductivity and specific heat cannot be set for
different phases. The impact of this is that the accuracy will
decrease.
 The permeability can only be described as a constant – leading to a
decreased accuracy.

38
 The CFL (Courant, Friedrich, Levy, 1928) condition is an important
convergence condition when solving partial differential equations
such as the equations in the eddy current solver. In this case it is
based on the magnetic diffusion time step size of the element. The
method that LSTC (LS-DYNA developer) is currently using to
solve the coupled FEM-BEM system has been to actually decouple
these 2 systems and solve them in an iterative way. This often
necessitates very small time steps, smaller than the CFL. LSTC is
currently working on a new method to actually solve the coupled
system using GMRES methodology which seems to allow
significantly larger time steps. GMRES is the Generalized Minimal
RESidual method i.e. an iterative method for the numerical solution
of a non-symmetric system of linear equations. However, this
method is not yet available in LS-DYNA.

39
4. Analysis and discussion of the
simulation results
In order to make simulations happen within reasonable time frame, the bar
length has been cut to contain only three cross section of it, with a length of
6 mm, as shown in Figure 4.1. The simulation results have been obtained
with MAT_UHS_STEEL material model. Within that model users have
possibility to switch the HEAT parameter to activate the heating/cooling
mechanism:
 =0 – Heating is not activated. No transformation to austenite is
possible.
 =1 – Heating is activated. Transformation to austenite is possible.
 =2 – Automatic switching between heating and cooling. LS-DYNA
checks the temperature gradient and calls the appropriate algorithm.
 <0 – Switch between heating (equal to 1) and cooling (equal to 0) is
defined by a time dependent load curve.

Figure 4.1. "New" bar - three cross section of the whole bar.

Automatic switching between heating and cooling, HEAT = 2, has been

used in the analysis.

40
4.1. Magnetic results
Figures 4.2 and 4.3 show the magnetic field distribution on the bar cross
section at three different times and the relative position in the production
process, respectively.

Figure 4.2. The magnetic field distribution of the bar at different times.

Figure 4.3. Position of the coil and workpiece at different times.

41
At the time equal to 0 s, the inductor starts moving towards the bar. First, at
the time equal to 1.5 s, the notable values of the magnetic field can be
obtained in the first two element rows. At the time equal to 3.1 s, the value
of the magnetic field reaches the highest value in the process. It can be
noticed that at this time point, the distance between the inductor and the bar
is the smallest – the bar is inside the inductor. The value of the magnetic
field decreases with the inductor moving away from the bar.
Numerical measurements of the magnetic field of four elements on the bar
(Figure 4.4) are shown in Figure 4.5. From the core to the surface, the four
elements are A, B, C and D, respectively. From the graph, it can be seen
that the more distant elements are from the core, the higher value of the
magnetic field can be obtained.

Figure 4.4. The position of the elements – A, B, C and D from the left to the
right.

42
 Figure 4.5. Magnetic field as a function of time for four elements.

Here follows the figure showing the proximity effect – the inner side of the
coil has been used to heat the workpiece. Since the inductor is the coil, the
maximum current density is noticeable at the inner side of the coil.

Figure 4.6. The contours of current density of the inductor.

43
4.2. Thermal results
At the time 0, the inductor and the cooling tool start moving together
towards the bar. The temperature of the bar is equal to the environment
temperature, 293.15 K. At the process time of 3 s, the heating process is
going on, the temperature of the bar is increasing, and the bar is near the
inductor. At the process time of 4 s, the bar starts entering the cooling tool.
Moreover the temperature of the bar reaches the peak value; the
temperature at the surface of the bar is higher than in the core.
Further on, the bar is exposed to the cooling process. The heat losses are
more obvious on the surface of the bar, so the temperature will decrease
faster the closer it is to the surface.
Temperature measurements for the same four elements on the (Figure 4.4)
have been done, see Figure 4.7. The same observation as above can be
concluded - the surface temperature reaches the peak value at the time of 4
s, which is near 1400 K. Then it starts decreasing very fast in the cooling
process. However the temperature of the core still increases until the
process time is around 7 s.

Figure 4.7. Temperatures as functions of time for the measured elements.

44
4.3. Metallurgical results
At the end of the heating process, the surface layer has been transformed to
austenite – only four outside element rows have the austenite content at this
moment. The phase proportion of the elements at other position keeps the
initial phase proportion.
At the process time of 16 s, when the cooling process is finished, the
situation is following. All the austenite has been transferred into the
martensite. The inner ferrite and pearlite have not changed at all, while no
bainite appears from the beginning to the end of the process.
This has been documented in the figures that follow. The following figure
shows the position of the six measured elements A to F.

Figure 4.8. The position of the measured elements.

The following figures show the five phase proportion contents through the
time in different elements.

45
 Figure 4.9. Phase proportions in the element A as functions of time.

The measured element A is at the surface of the bar. In the beginning, there
is 60% ferrite and 40% pearlite, then all of them transfer into austenite at
about 3 s of the process time. After 7th second, austenite starts transferring
into martensite. So at the end of the cooling process, 100% martensite can
be obtained in the elements at the surface.
Figure 4.10 shows the phase proportions in the measured element B (at the
third row of the cross section) as functions of time. As previously, in the
beginning there is 60% ferrite and 40% pearlite. Then all of them transfer
into austenite at the austenitization temperature (t = 3.5 s). At the time
about 8 s, the austenite starts transferring into the martensite. After the end
of cooling process, 100% martensite can be obtained in the element.
Though the final phase proportion is the same as for the element A, the
transformation time is different.

46
 Figure 4.10. Phase proportions in the element B as functions of time.

Figure 4.11 shows the phase proportions in the element C through the time
(at the fourth element row). All pearlite and some ferrite have been
transferred into austenite at about the process time of 4 s, austenite starts
transferring into the martensite after t = 8 s, and until the end of the cooling
process, all austenite has been transferred. The final phase proportions are
about 50% ferrite and 50% martensite.

Figure 4.11. Phase proportions in the element C as functions of time.

Figure 4.12 shows the phase proportions in the element D (at the fifth
element row). Only about 5% pearlite has been transferred into austenite
soon after 4th second, while ferrite amount have not changed. At about the
process time of 8.5 s, austenite starts transferring into martensite. After the
end of the cooling process, all austenite has been transferred into

47
martensite. The final phase proportions are about 60% ferrite, 35% pearlite
and 5% martensite.

Figure 4.12. Phase proportions in the element D as functions of time.

Figure 4.13 shows phase proportions of the element E (at the sixth element
row). From the graph it can be noticed that there are no phase
transformations in this element.

Figure 4.13. Phase proportions in the element E as functions of time.

Considering the element F (at the core), we got the same result as the
previous one – phase proportions are always 60% ferrite and 40% pearlite.
So from the sixth row to the core, the phase proportions do not change
through the whole process.

48
Still some strange results can be obtained for a couple of elements: Around
5th second the proportion of ferrite starts increasing after it has been reset to
zero, see Figure 4.14. This is something connected to a bug in the material
model.

Figure 4.14. Strange ferrite amount in some elements.

The final micro-hardness in Vickers is shown in Figure 4.15. From the

figure, we notice the maximum value of the hardness at the surface of the
bar, 703 HV. On the other side the core have not changed the initial
hardness, it is still 184 HV.

Figure 4.15. Final micro-hardness in Vickers.

Miscellaneous results of interest have been discussed in Appendix A.

49
5. Evaluation of the results
In the paper [5], Magnabosco et al. use two samples to do the experimental
tests. One of them was subjected to the annealing heat treatment before the
quenching process, while the other one was left in its normalized state [5].
After the cooling process, the micro-hardness in Vickers in the radial
direction of the specimen was measured to detect the penetration hardening.
Then, those results were compared to the numerical ones (Sysweld).
The same has been done with the numerical result from this study, but the
comparison is made only to the normalized specimen. We have to
emphasize here that there is a large difference in the bar length in our and
their [5] study. The shorter bar that we used has been justified by the wish
of reasonable running times.

Figure 5.1. Comparison between the micro-hardness measurement in the

experiment [5] and the values obtained with the numerical model.

50
The numerical values of micro-hardness are in satisfactory agreement with
experimental observations and a good prediction of the maximum value of
micro-hardness was obtained.
Magnabosco et al. [5] made SEM analyses of the final phase proportions at
different depths. The specimen microstructures of normalized steel after
induction hardening are shown in the Figure 5.2. Micrograph A shows the
microstructures at 0.5 mm distance from the surface, and we can notice that
a complete martensitic micro-structure can be obtained. Micrographs B and
C (depths of 1.6 and 2.1 mm respectively) show that an increment of
untransformed ferritic phase moves toward the specimen axis [5]. In
micrograph D (2.5 mm depth), the phase proportion has been kept the same
in comparison with the initial situation.

Figure 5.2. Specimen microstructures after induction hardening [5].

51
From the numerical model, the final phase proportions are shown in the
following pie charts (Figures 5.3 to 5.6). They show the microstructure
proportions at 0.46 mm, 1.759 mm, 2.237 mm and 2.706 mm distance from
the surface, respectively. At 0.46 mm, there is only martensite, which is in
line with the experimental results. At 1.759 mm, all the pearlite and some
of the ferrite have been transferred into martensite. At 2.237 mm, some of
the pearlite transferred into martensite, but no ferrite. While at 2.706 mm,
the phase proportion does not change from the initial situation.

Figure 5.3. Final phase proportion at 0.46 mm distance from the surface.

Figure 5.4. Final phase proportion at 1.759 mm distance from the surface.

52
Figure 5.5. Final phase proportion at 2.237 mm distance from the surface.

Figure 5.6. Final phase proportion at 2.706 mm distance from the surface.

53
6. Conclusions
LS-DYNA FEM software has been used for simulations of an induction
hardening process. The aim was to develop a simulation model for
induction heat treatment with coupling between the electro-magnetic,
thermal, mechanical and metallurgical phenomenon. The main study
included also comparison of the results of the simulation with the literature
values, for evaluation and validation of the FEM software LS-DYNA as a
tool for simulation of induction hardening.
Because of the long computational times, the numerical model of the bar
has been made considerably smaller. Apart from that, the numerical results
seem very reasonable.
Three things are more essential than others when talking about the LS-
DYNA capability to efficiently model this process:
 More stable material model (unrealistic ferritic occurrence at
cooling, see Figure 4.14)
 Access to a 2D solver
 Faster EM solver
LSTC is currently working on all these topics. However, during the time of
writing the thesis, these methods have not been available.

54
 Reference
[1] Yutaka Toi and Masakazu Takagaki, (2008), Computational Modeling of Induction Hardening
Process of Machine Parts, WCECS, San Francisco, USA.
[2] Torsten Holm, Pelle Olsson and Eva Troell, (2012), Steel and Its Heat Treatment, Mölndal,
Sweden.
[3] J.F. Wu, (2004), Numerical Simulation of Induction Heating Based on ANSYS, J. Manuf. Sci.
Eng. 32(l):58-62, Hangzhou, China.
[4] Daniel Williams, (1995), Quench Systems for Induction Hardening, Metal Heat Treating,
Cleveland, Ohio, USA.
[5] I. Magnabosco,P. Ferro,A. Tiziani,F. Bonollo, (2006), Induction heat treatment of a ISO C45
steel bar: Experimental and numerical analysis, Computational Materials Science 35 (2006) 98–
106, Vicenza, Italy.
[6] Induction heating. Available from: < [Link] >.
[7] Pierre L’ Eplattenier, Inaki Caldichoury, (2011), Electromagnetism and Linear Algebra in
LSDYNA, EM Theory manual, LSTC-LS-DYNA-EM-THE1.1-1, Gramat, France.
[8] O. Biro and K. Preis, (1989), On the use of the magnetic vector potential in the finite element
analysis of three dimensional eddy currents, IEEE Transaction on Magnetics, 25 No. 4 (1989),
pp. 3145–3159, USA.
[9] Skin effect. Available from: < [Link]
[10] Proximity effect (electromagnetism). Available from:
<[Link]
[11] Göran Broman, (2003), Computational Engineering, Karlskrona, Sweden.
[12] Ottosen N. S. and Petersson H, (1992), Introduction to the Finite Element Method, Prentice Hall,
Michigan, USA.
[13] G. P. Nikishkov, (2004), Introduction to the Finite Element Method, 2004 Lecture Notes.
University of Aizu, Aizu-Wakamatsu 965-8580, Japan.
[14] Boundary element method. Available from: <
[Link]
[15] R. Rieben and D. White, (2006), Verification of high-order mixed finite element solution of
transient magnetic diffusion problems, IEEE Transaction on Magnetics, 42 No.1 (2006), pp.
25–39, USA.
[16] John [Link], (1998), LS-DYNA Theoretical Manual, Livermore Software Technology
Corporation, California, USA.
[17] Edin Omerspahic, (2005), Modeling of Inelastic Effects in Metal Sheets and Identification of
Material Parameters, Chalmers University of Technology, Göteborg, Sweden.
[18] Sven Wanser, Laurent Wenbiihl, Main Nicolas, (1994), Computation of 3D Induction
Hardening Problems by Combined Finite and Boundary Element Methods, IEEE transactions
on magnetics, Vol.30, No.5, France.
[19] Explicit and implicit methods. Available from:
<[Link]

55
[20] [Link], J.M.J. McDill and L. Karlsson, (1996), Microstructural predictions including
arbitrary thermal histories, reaustenitization and carbon segregation effects, Can. Metall. Quart.,
1996. 35(3): p. 275-283, Ottawa, Canada.
[21] M.V Li, D.V Niebuhr, L.L Meekisho and D.G Atteridge, (1998), A Computatinal model for te
prediction of steel hardenability, Metallurgical and materials transactions B, 29B, 661-672,
Portland, USA.
[22] P. Maynier, J. Dollet, and P. Bastien, (1978), Hardenability Concepts with Applications to
Steels, D.V. Doane and J.S. Kirkaldy, eds., AIME, New York, NY, pp. 518-44, Chicago, USA.
[23] SYSTUS and SYSWELD 2000 manuals, (2000), ESI GROUP, Lyon, French.
[24] ASM Handbook, (1998), Heat treating, vol. 4, 1998.
[25] J. Yuan, J. Kang, Y. Rong, and R.D. Sisson, Jr., (2003), FEM Modeling of Induction Hardening
Processes in Steel, JMEPEG (2003) 12:589-596, Massachusetts, USA.
[26] Conjugate gradient method. Available from: <
[Link]
[27] Finite element method. Available from: <[Link]
[28] Equations of motion. Available from: < [Link]
[29] Klaus Weimar, (2001), LS-DYNA User's Guide, Rev. 1.19, Hamburg, Germany.
[30] Heath, Michael T., (2002) Scientific computing, an introductory survey, 2nd ed., Mc Graw Hill,
Illinois, USA.
[31] Ivan Oskarsson, (2006), A new h-adaptive method for shell elements in LS-DYNA, Lund
University, Lund, Sweden.
[32] Materials Algorithms Project. Available from: <[Link]
[33] H. Porzner (ESI Group), (2001), Private communication, USA.

56
Appendix A: Miscellaneous results
The simulation results in the result relating chapters have been obtained
with HEAT = 2 – Automatic switching between heating and cooling, and
run implicitly on the mechanical solver. In order to check the LS-DYNA’s
capability to model the induction hardening process, some other modeling
techniques and features have been tested. Here follows some results from
the other methods and comparison with the results described in the previous
chapter.

Explicit mechanical solver

Figure A.1 shows the ferrite distribution on the bar after the cooling process
with the auto switching between heating and cooling, and running explicitly
on the mechanical solver. It can be seen that the ferrite distribution at the
surface is inhomogeneous. If we check the ferrite amount in one element at
the surface, as shown in Figure A.2, there is a strange phenomenon that
some ferrite appears at very high cooling rate. The reason for this error is
probable bugs in the material model for the explicit software.

Figure A.1. The ferrite distribution of the bar after cooling.

57
 Figure A.2. Ferrite phase transformations for the measured element.

Implicit mechanical solver with curve switching between heating and

cooling

As shown in Figure A.3, this curve has been used as the HEAT parameter
in MAT_UHS_STEEL material model to define the heating (=1) and
cooling process (=0) for the bar.

Figure A.3. Curve used to switch between heating and cooling.

Ferrite proportion was checked in the elements of the first and second row,
as shown in Figures A.4 and A.5.

58
Figure A.4. Amount of the ferrite as the functions of time in the measured
elements – 1st row.

Figure A.5. Amount of the ferrite as the functions of time in the measured
elements – 2nd row.

59
While all the measured elements in the first row show the realistic result, all
checked elements in the second row have the unrealistic ferrite increase
after 6th second (very high cooling rate). Other elements in the cross section
do not show this strange behavior. The reason is probably the mentioned
bug.

Implicit mechanical solver with 2 successive processes – heating and

cooling enabled by INTERFACE_SPRINGBACK_LSDYNA

This attempt was about separating the whole process into two parts, and
running first the heating process (with an output which has been the input
for the new process), and then the cooling process. This has been made
possible by using INTERFACE_SPRINGBACK_LSDYNA keyword
where the number of history variable (NSHV) should be high enough to
map all variables between the calculations.
The result with the automatic switching is better than the one with 2
successive processes. Figure A.6 shows it clearly, illustrating ferrite
proportion in all the elements of the middle cross section. We can see that
the number of elements that have the unrealistic increase of ferrite is bigger
than the one resulting from the automatic switching method.

60
Figure A.6. Comparison of the amount of ferrite in the measured elements
between different methods.

To sum up, by comparing all the results, the best ones have been achieved
with the auto switching between heating and cooling, running implicitly the
mechanical solver.

61
School of Engineering, Department of Mechanical Engineering Telephone: +46 455-38 50 00
Blekinge Institute of Technology E-mail: info@[Link]
SE-371 79 Karlskrona, SWEDEN