You are on page 1of 10

Journal of Materials Processing Technology 213 (2013) 477–486

Contents lists available at SciVerse ScienceDirect

Journal of Materials Processing Technology


journal homepage: www.elsevier.com/locate/jmatprotec

The effect of hardening laws and thermal softening on modeling residual stresses
in FSW of aluminum alloy 2024-T3
M.R. Sonne a,∗ , C.C. Tutum a , J.H. Hattel a , A. Simar b , B. de Meester b
a
Technical University of Denmark, Department of Mechanical Engineering, Section of Manufacturing Engineering, Building 425, Lyngby, Denmark
b
Institute of Mechanics, Materials and Civil Engineering, Universite catholique de Louvain, Louvain-la-Neuve, Belgium

a r t i c l e i n f o a b s t r a c t

Article history: In the present paper, a numerical model consisting of a heat transfer analysis based on the Thermal Pseudo
Received 29 May 2012 Mechanical (TPM) model for heat generation, and a sequentially coupled quasi-static stress analysis with
Received in revised form 17 October 2012 a built-in metallurgical softening model was implemented in ABAQUS. Both isotropic and kinematic rules
Accepted 1 November 2012
of hardening were used in order to study the effect of the hardening law on the residual stresses as well
Available online 9 November 2012
as on the final yield stress. This numerical model was then applied in two different cases. Firstly, a very
simple 1D Satoh test was modeled. Different combinations of either isotropic or kinematic hardening
Keywords:
together with the metallurgical softening model were applied in order to give a first impression of the
Friction stir welding
Aluminum alloy 2024-T3
tendencies in residual stresses in friction stir welds when choosing different hardening and softening
Hardening laws behaviors. Secondly, real friction stir butt welding of aluminum alloy 2024-T3 were simulated and com-
Thermal softening pared with experimentally obtained results for both temperatures and residual stresses (using the slitting
Modeling residual stresses method). The comparisons showed good agreement regarding temperatures whereas the residual stress
Finite element analysis comparisons indicated different sensitivities for the cold and hot welding conditions toward the choice
of hardening rules and especially whether including the softening model or not.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction residual stresses is very similar in all kinds of welding techniques


whether it is fusion welding or FSW. Besides the thermal gradients
Friction stir welding (FSW) is an efficient solid state joining tech- due to non-uniform heat generation, mechanical boundary con-
nique that, without melting, is intended to be used for joining of, for ditions (i.e. clamping, contact conditions between the work piece
example, high strength aluminum alloys. Also welding of dissimi- and the anvil, etc.) also play an important role for promoting plas-
lar materials, which are difficult to weld with traditional welding tic strains, hence residual stresses. Although the level of residual
techniques, can be handled with FSW (Steuwer et al., 2006; Lee stresses resulting from the FSW process in aluminum alloys is nor-
et al., 2009; Inada et al., 2010; Liyanage et al., 2009; Yamamoto mally believed to be somewhat lower as compared to traditional
et al., 2008; Gerlich et al., 2008; Karlsson et al., 2001; Nandan et al., welding techniques (Hatamleh et al., 2007; Donne et al., 2001), such
2007). In the FSW process shown schematically in Fig. 1, heat is as fusion welding (see the work by Deplus et al., 2011), Bussu and
generated by the friction between the tool shoulder and the work- Irving (2003) showed that the residual stresses play a major role for
piece surface and by the plastic deformation that occurs between the fatigue crack growth in friction stir welded structures as also
the toolpin and the workpiece. mentioned by James et al. (2007). Moreover, Murphy et al. (2007)
Owing to the non-uniform heat generation during the FSW and Bhide et al. (2006) observed also buckling behavior in welded
process, the material under the tool is heated up, consequently structures obtained using FSW. Thus, in order to understand and
expands, however it encounters some resistance by relatively control the evolution of thermal and residual stresses arising from
colder material surrounding this warmer region (i.e. mainly the FSW, much work on modeling of the process has been reported in
material in front of the tool along the welding direction). It sub- the literature. In thermomechanical modeling of residual stresses
sequently yields in compression and plastic deformation starts in FSW, the most convenient assumption is to neglect the material
developing. This fundamental mechanism of evolution of the flow during welding. This results in generally semi-coupled ther-
momechanical models in a Lagrangian frame, where the thermal
field is calculated prior to the mechanical field by separating the
∗ Corresponding author at: Produktionstorvet, Building 425, DK 2800 Kgs, Lyngby, two analyses. This means that the entire temperature history is
Denmark. Tel.: +45 45254734. calculated first and is then applied in a subsequent mechanical anal-
E-mail address: mrso@mek.dtu.dk (M.R. Sonne). ysis. By this approach, it is possible to use temperature dependent

0924-0136/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jmatprotec.2012.11.001
478 M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486

Table 1
Thermomechanical models for FE modeling of residual stresses in FSW of precipitate
hardened aluminum alloys.

Softening model/ With softening Without softening


hardening law (metallurgical model)

Isotropic Chao and Qi (1998) Shi et al. (2003)


Feng et al. (2007) Zhu and Chao (2004)
Richards et al. (2008) Li et al. (2007)
Tutum et al. (2008)
Dubourg et al. (2010)
Hattel et al. (2012)
Kinematic Bastier et al. (2008) Chen and Kovacevic
(2006)
Tutum and Hattel (2010)

Fig. 1. Schematic view of the FSW process (Tutum and Hattel, 2010).

material data, as well as implementing history dependent models


for the evolution of mechanical properties such as microstructural use of either hardening laws or metallurgical models in order to
models in the mechanical analysis. Several authors in literature numerically simulate residual stresses in FSW of precipitate hard-
have used semi-coupled thermomechanical models without taking ened Al-alloys. It is the aim of this paper to more systematically
material flow into account using the finite element method. In this investigate the effects of using isotropic or kinematic hardening,
field of modeling FSW, different elasto-plastic models have been with or without a metallurgical softening model, and compare this
used in order to predict the nonlinear behavior of the material in with measurements of residual stresses obtained from experiments
the most realistic way. Generally two different hardening laws have of FSW for the precipitate hardened Al-alloy 2024-T3.
been used for modeling residual stresses of FSW in precipitate hard- The structure of the paper is the following. First, the theory used
ened Al-alloys. The majority of the authors use isotropic hardening, for the model is presented. Here the Thermal Pseudo Mechanical
whereas others have been using a kinematic hardening law in order (TPM) model for the heat transfer analysis, the different hardening
to describe the elasto-plastic behavior of the material. The mechan- laws and the metallurgical softening model in the stress analysis
ical properties will in general be changed due to the metallurgical are introduced. This is followed by a small section about how the
evolution during the welding process as well as the subsequent theory is implemented into the FE software ABAQUS. In order to
cooling. This is also the case for FSW of age hardenable Al alloys in get a first impression of how the different hardening laws and the
which the softening is closely related to the volume fraction, the metallurgical softening model affect the residual stresses, the Satoh
size and the phase of the hardening precipitates in the alloy which test is used as preliminary application. Then in the final application,
in turn will be affected by the level of temperature as well as the experiments involving FSW of aluminum alloy 2024-T3 are per-
holding time at elevated temperatures. In FSW, the most used type formed and measurements of residual stresses using the slitting
of model for this expresses the volume fraction of the hardening method are compared with results from the FE model.
precipitates via relatively simple kinetics of precipitate dissolution
often refered to as the Myhr and Grong model (proposed by Myhr
and Grong, 1991). Chao and Qi (1998) were the first to model FSW in 2. Description of the models
a Lagrangian frame based semi-coupled thermomechanical model.
In 1998, they presented a 3D model in an in-house developed code 2.1. Thermal model (TPM)
(predecessor of WELDSIM) using bilinear isotropic hardening, and
a simplified softening model, where the yield stress could follow a The thermal model for heat generation in FSW proposed by
lower curve during cooling down compared to heating up. Zhu and Schmidt and Hattel (2008) is applied in this study. The heat gen-
Chao (2004) used WELDSIM in their model of FSW of 304L stain- eration is expressed as a surface heat flux from the tool shoulder
less steel with bilinear isotropic hardening, but without including (without the tool probe) into the workpiece; however, it is a func-
softening. Chen and Kovacevic (2006) used kinematic hardening tion of the tool radius and the temperature dependent yield stress
without softening behavior, in their parametric finite element anal- as follows
ysis of the stress evolution during FSW in but welding of aluminum
q̇surface
 2n   (T )
yield
alloy 6061-T6. Bastier et al. (2008) presented a thermomechanical (r, T ) = ωr(T ) = r √ ,
A 60 3
model with kinematic hardening and softening. Here, flow was also
taken into account as the mechanical model had a viscous part in for 0 ≤ r ≤ Rshoulder (1)
the constitutive equations. Feng et al. (2007) used isotropic hard-
ening in combination with a softening model in their modeling of where n is the number of tool revolutions per minute, r is the radial
residual stresses and property distributions in FSW of aluminum position originating in the tool center, Rshoulder is the tool shoul-
alloy 6061-T6. Richards et al. (2008) used bilinear isotropic hard- der radius,  is the shear stress, and  yield (T) is the temperature
ening with softening in their work of modeling global tensioning for dependent yield stress. The model is often denoted ‘Thermal Pseudo
the management of residual stresses in FSW. Moreover, Tutum and Mechanical’ (TPM) since the heat generation is expressed via the
Hattel (2010) recently presented the first and so far only study in temperature dependent yield stress, thus taking some mechani-
optimization of process parameters in FSW based on residual stress cal effects into account; however, it should be underlined that the
analysis, in which they used kinematic hardening and no softening model is a purely thermal model involving temperature depend-
in the thermomechanical model. Also other authors modeling FSW ent heat generation, and in that sense, it also uses a ‘calibration’
have used different hardening laws with or without metallurgical parameter like the more conventional procedures, which normally
softening behavior included. In Table 1 presents an overview of rely on experimentally obtained values for the total heat gener-
some of the different contributing authors and their use of harden- ation as done in the work by Simar and Pardoen (2007) or the
ing laws and softening behavior. This literature survey shows that friction coefficient, see the review of thermomechanical models of
there is some inconsistency and still not any clear consensus in the the FSW process by Simar et al. (2012). The heat source is used as a
M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486 479

boundary condition at the interface between the shoulder and the


work piece for the heat conduction equation

∂T
cp = ∇ · (k∇ T ) + q̇vol (2)
∂t

where  is the material density, cp is the specific heat capacity, T


is the temperature, k is the thermal conductivity and k is the volu-
metric heat source term. As for most pure thermal models of FSW,
the latter is, in this case, zero because all of the heat generation
is modeled via the boundary condition at the shoulder/workpiece
interface given by Eq. (1).

2.2. Mechanical model


Fig. 2. Uniaxial tension and compression using the two different hardening laws.
For calculation of the transient as well as the residual stress
field in the workpiece, a standard mechanical model based on the
solution of the three static force equilibrium equations is used, i.e.
2.3. Short introduction to the softening model
ij,i + pj = 0 (3)
The microstructural evolution of the material during friction stir
where pj is the body force at any point within the plate and  ij is welding is for Al-alloys in T3 condition expected to have a consid-
the stress tensor. Hooke’s law and the linear decomposition of the erable effect on the residual stress distribution. During the thermal
strain tensor as well as small strain theory are applied together with cycle of welding, dissolution, precipitation and coarsening occur in
the expression for the thermal strain the HAZ and thermomechanically affected zone (TMAZ), Genevois
et al. (2005). This results in softening of the material, with a con-
pl sequent reduction of the yield strength. Evolved numerical models
εtot
ij
= εel
ij
+ εij + ıij εth
have been developed to account for the dissolution/precipitation
el εel = C el (εtot − ε − ı εth )pl
ij = Cijkl kl ijkl ij kl kl in precipitation hardened aluminum alloys and have been applied
E
1  to FSW (Simar et al., 2012), however these are not well suited for a
el =
Cijkl (ıik ıjl + ıil ıjk ) + ıij ıkl coupling with a FE code since they are too computationally inten-
1+ 2 1−2
sive. Furthermore, it is difficult to develop such models for the 2024
(4)
1 alloy due to the requirement of accounting for the heterogeneous
εtot = (u + uj,i )
ij 2 i,j precipitation that is observed on dislocations, dispersoids and at
T2 grain boundaries in this alloy. Room temperature aging may also
occur after cooling at room temperature but that is not expected to
εth (T1 → T2 ) = ˛t dT
affect the residual stress distribution. It is therefore ignored in the
T1 present study, in which the softening model developed by Myhr and
Grong (1991) was used to predict the phase transformation during
where εij is the strain tensor, ıij is the Kronecker tensor, Cijkl is the welding as also proposed by Richards et al. (2008). This model only
elastic stiffness 4th order tensor, E is the Young’s modulus, is the accounts for the dissolution of GPB zones in the HAZ with simple
Poisson’s ratio, ˛t is the coefficient of thermal expansion and ui is analytical equations that can be introduced in the FE code ABAQUS.
the displacement vector. The model relates the fraction of dissolved hardening precipitates
Xd to the equivalent time of heat treatment, teq = t/t* (where t is
the period of time at a temperature T and t* is the time for total
2.2.1. Hardening laws precipitation dissolution at this temperature) as follows
The plastic strain is based on the standard J2 flow theory with
a temperature dependent von Mises yield surface. For isotropic  t
Ntotal 
Ntotal
ti
hardening, the elastic range expands from the initial yield stress, n
Xd = teq teq = i
=  
ti∗ Qeff
when a load above the yield stress is applied. After loading in uni- i=1 i=1 tref exp R
(1/Ti − 1/Tref )
axial tension, the load in compression must exceed the maximum
yield stress value obtained in tension to yield. For kinematic hard- f n

= 1 − Xd = 1 − teq = 1 − teq (5)
ening, instead of expanding the yield surface, it keeps its initial f0
size, but translates in stress space. This means that after loading
above the initial yield stress in compression (typical for the heat where tref is the time for total dissolution at the reference temper-
affected zone (HAZ) in welding), the yield stress in tension is some- ature Tref , R is the gas constant and Qeff is the effective activation
what lower than the initial one, see Fig. 2. This will obviously affect energy for precipitate dissolution. Eq. (5) reflects that the equiva-
the final yield stress level. Since aluminum is fcc and iron is bcc, lent time in a numerical model is found by discretizing the thermal
the Bauschinger effect is more often seen in aluminum alloys as history into small steps, calculating the equivalent time for each
compared to steel alloys. Furthermore, Fribourg et al. (2011) have step and then summing up in order to find the total equivalent
shown that in Al alloys the Baushinger effect is amplified by the time. The fraction of hardening precipitates f/f0 then relates to the
presence of strengthening non-shearable precipitates. Such precip- equivalent time teq via the fraction of dissolved precipitates Xd as
itates where observed by Genevois et al. (2005) in the heat affected shown in Eq. (5), where n is a material constant which is obtained
zone of 2024-T351 welds. Multiaxial yield criteria, for isotropic experimentally. A value of 0.5 is often used as indicated in the last
and kinematic hardening, are well-known and can be found in e.g. part of Eq. (5). Finally, the hardness and yield stress distributions
Tvergaard (2001). are predicted via linear interpolation between the original state and
480 M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486

Fig. 5. Satoh test bar and heating profile.

Surface film conditions are also applied to the top and bottom
surfaces of the plate by the equation
Fig. 3. Upper (T3 condition) and lower yield stress (after a solution treatment)
bounds as a function of temperature.
q = hA(T∞ − T ) (7)

the fully dissolved state where h, the heat transfer coefficients are different in the top and
the bottom of the plate as it is clamped to a large steel backing plate,
f f resulting in large dissipation of heat at the bottom as compared to
HV = (HVmax − HVmin ) + HVmin  = (max − min )
f0 f0 the top surface which is in contact with the steady ambient air. In
+ min (6) the quasi-static elastoplastic analysis, the metallurgical softening
model is implemented using the ABAQUS user subroutine USDFLD.
where HVmax and  max are the hardness and yield stress of the In this subroutine, the fraction of softening is calculated as a value
material in T3 condition and HVmin and  min are the hardness and between 1 (material in T3 condition) and 0 (fully softened mate-
yield stress of the solution heat treated material. This softening rial). The summation is accomplished by storing the fraction value
model has been applied to FSW by several authors. Chao and Qi of dissolved hardening precipitates for each time step in a state
(1998), Richards et al. (2008) and Feng et al. (2007) have used this variable (SDV). The material’s plasticity properties are written into
softening model in their FSW model. The yield stress at the instan- a large lookup data table (see Appendix A) where ABAQUS makes
taneous temperature is also found by an interpolation between the a trilinear interpolation between the temperature, effective plastic
upper and lower bound yield stress curves in proportion to Xd. In strain and softening fraction in order to determine the yield stress
Fig. 3, the upper and lower yield stress bounds as a function of in the given situation.
temperature are shown schematically.

3. Applications
2.4. Implementation in ABAQUS
Before the main application in determining residual stresses in a
The semi coupled thermomechanical model of the FSW process real weld in the FSW 2024-T3 aluminum alloy, a preliminary anal-
consists of a transient thermal model and a quasi-static elasto- ysis with the Satoh test (Satoh, 1972) (Fig. 5) is performed. This is
plastic mechanical model as described in Sections 2.1 and 2.2, chosen because it shows in a simple way the origin of the resid-
implemented in ABAQUS. The thermal and mechanical properties ual stresses in welded pieces. Deplus et al. (2011), used this test to
of AA 2024-T3 are given in Appendix A. In the heat transfer anal- explain the tensile stresses in the weld nugget after FSW.
ysis, the moving heat source is modeled using the user subroutine
DFLUX applying Eq. (1). The yield stress data used in Eq. (1) are
given as tabular values, but in order to get the yield stress and its 3.1. Satoh test
derivative (used in the DFLUX subroutine) in the TPM model, the
upper yield stress curve is fitted with a 3rd order polynomial, see The axial stress in the Satoh test corresponds to the longitudinal
Fig. 4. stress in the welding process. The constraints in both ends repre-
sent the resistance of the cold material against expanding material
due to the heat input. The perfect rigid constraint (see Fig. 5) is an
extreme case compared to the real welding applications. But for
benchmark testing of different theories, because of its simplicity, it
is a very useful tool for validation.

3.1.1. Temperature input for the Satoh test


The heating up and cooling down of the Satoh test are supposed
to reflect the temperature profile experienced in a point in the real
FSW piece. For this material a cut off temperature of 500 ◦ C is set for
the temperature dependent yield stress properties. Therefore the
maximum temperature in the Satoh test is set to this quantity in
order to cover the full range of the plasticity effects. As the plasticity
behavior is time independent, it is here unimportant how the max-
imum temperature of 500 ◦ C is reached. But the softening model
used in this study is time dependent, and it is hence important for
this reason to have the most correct temperature history in order to
get the fraction of softening and thereby the yield stress behavior
more realistic in the Satoh test. In literature, the theory of moving
Fig. 4. Upper and lower yield stress curves, and the fitted 3rd order polynomial for heat sources developed by Rosenthal (1946) has been commonly
2024-T3 aluminum alloy (ASM Handbook, 1993). used as the analytical solution to the temperature field of welds.
M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486 481

Fig. 6. Temperature distribution and softening fraction of Rosenthal’s solution heat up and cool down.

The Rosenthal transient 2D solution for a moving heat source in a


thin infinite plate is given by Eq. (8).
Q v
T − T0 = e−ω K0 (ωr); ω= (8)
2kd 2˛
where T0 is the initial temperature of the infinite plate, Q is the
rate of heat input, k is the thermal conductivity, ˛ is the thermal
diffusivity, v the welding speed and d the thickness of the thin plate.
K0 is the Bessel function of the second kind and zero’th order. r is
the distance from the moving heat source. To include heat losses
through the surface of the infinite plate, ω in K0 (ωr) can be modified
as given by the following formula from Karlsson (1986)

v2 + (˛2 h/kd)
ω= (9)

where h is the heat transfer coefficient. The temperature is calcu-
lated in a position 0.2 m from the starting point (i.e. x = 0.25 m) of the
moving source and 10 mm in transverse distance from the weld line Fig. 7. Satoh test emulating the “longitudinal stress” as a function of temperature
center. This point is chosen because it gives a temperature history with four different plastic behaviors.

with a peak temperature of approximately 500 ◦ C, which is the cut


off temperature for the minimum yield strength. The other welding
parameters for this preliminary Satoh test are d = 2 · 10−3 m, v =
2 · 10−3 m/s, Q = 1100 W, h = 600 W/(m2 K). The constants
for the Myhr and Grong softening model for aluminum alloy 2024-
T3 were found to be: Tref = 623 K, tref = 16 s, Qeff = 70.52 kJ/mol based
on measurements of the hardness as a function of high temperature
heat treatments out of the “Deepweld” European project (2008).
The Rosenthal solution is now calculated for a total time of 800 s in
time increments of 1 s. For each time step the fraction of softening is
calculated according to the procedure listed in Section 2.3. The ini-
Fig. 8. Plate and welding dimensions.
tial temperature is set to room temperature, i.e. 15 ◦ C. As expected,
the temperature history shows to have a significant impact on the
behavior is also expected to be seen in the following and final FSW
resulting fraction of softening, see Fig. 6. The fraction of precipitates
application.
with the Rosenthal solution based temperature history dropped
down to 0 after cooling down. Indeed, the 2024-T3 aluminum alloy
fully softens extremely rapidly as the temperature reaches 500 ◦ C. 3.2. FSW thermo-mechanical model

3.2.1. Experiment
3.1.2. Results
Butt welds of aluminum alloy 2024-T3 are performed on a 12-
The Satoh test is now performed using an in-house developed 1D
kW HERMLE milling machine. The plates are 70 mm wide and
FEM code written in MATLAB, where a single element is constrained
600 mm long. The welds are 550 mm long and the welds start with
as shown in Fig. 5, and is subjected to the Rosenthal temperature
the tool center 25 mm from the plate edge, see Fig. 8.
solution. The quasi-static analysis is then performed, using differ-
Welds are performed for both “cold weld” and “hot weld” condi-
ent hardening laws (isotropic or kinematic hardening) and with or
tions. For the cold weld conditions, the plates are 2 mm in thickness,
without the metallurgical softening model. The results from the
the tool has a 10 mm shoulder diameter, a pin height of 1.7 mm and
Satoh test show that there are major differences in the final resid-
a pin diameter of 4 mm, see Fig. 9. The cold FSW is performed with
ual stresses, depending on which hardening laws and metallurgical
a traveling speed of 200 mm/min and a tool rotation of 800 rpm. For
models are chosen, see Fig. 7. First of all, it is seen that all resulting
residual stresses are in tension. This is in good agreement with what
is known from other simulations and experiments on welding of
aluminum alloys. It is observed that a change in hardening law from
isotropic to kinematic hardening without softening will decrease
the residual stresses around 40 MPa in this case. Softening shows
in this Satoh test to have a major effect on the residual stresses
of the material. It is seen that the longitudinal stress reduces from
300 MPa with kinematic hardening without the softening model,
to just below 100 MPa for isotropic hardening with softening. This Fig. 9. Tool dimensions.
482 M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486

Table 2 applied under the entire bottom surface of the plate (uz = 0), and on
Overview of settings for hot and cold welding conditions.
the edges, all degrees of freedom are constrained (ux = uy = uz = 0).
Cold Hot In Fig. 10 the applied boundary conditions and the dimensions of
uweld 200 mm/min 100 mm/min the plate are shown.
n 800 rpm 400 rpm To get realistic residual stresses, the “clamping” boundary con-
Dpin 4 mm 5 mm ditions are released in subsequent steps.
Dshoulder 10 mm 15 mm The plate is discretized by structured 8-node brick elements. In a
hpin 1.7 mm 2.9 mm
20 mm band near the welding line, the elements have a side length
Plate thickness 2.0 mm 3.175 mm
of 1 mm. Outside this welding zone, the mesh is biased toward the
edge of the plate, see Fig. 10.
the hot weld conditions, the plates are 3.175 mm in thickness, the
tool has a 15 mm shoulder diameter, a pin height of 2.9 mm and 3.2.3. Results
a pin diameter of 5 mm. The hot FSW is performed with a trav- The results in terms of temperatures from the heat transfer
eling speed of 100 mm/min and a tool rotation of 400 rpm. The analyses are now compared with the values measured during the
denomination “cold weld” and “hot weld” is based on the heat experiments. The temperature histories at the points closest to
input computed from torque measurements Mz performed during the weld center (thermocouple 2), see Fig. 11, show relatively
welding (see Jacquin et al. (2011) for more details). Those were good agreement between the numerical heat transfer analysis and
found to be equal to 12 and 32 Nm leading to heat inputs per unit the experimentally measured values for both cold and hot weld-
length (H = Mz .ω/uweld ) equal to 800 and 290 kJ/m for respectively ing conditions. Note that the temperature difference between the
the “hot” and “cold” welds. An overview of the welding conditions advancing and retreating side cannot be captured by a symmetric
for cold and hot welding conditions is given in Table 2. model which explains why two experimental curves are compared
The welded plates are clamped to a steel backing plate with a to a single simulation result. The maximum temperatures further
thickness of 80 mm, 450 mm wide and 630 mm long. During weld- away from the weld axis tend to be slightly over-predicted for the
ing, measurements of temperatures are performed by introduced cold weld, see Fig. 11c. This is expected, since the contact conditions
0.8 mm diameter thermocouples into the aluminum plates on the between workpiece and backing plate are identical everywhere,
advancing and retreating sides, respectively. The positions of all the which is not the case in the experiments, and this will introduce
thermocouples are given in Table 3. Note, that thermocouple no. 1 some difference in measured and predicted values.
is not used. The residual stresses from the subsequent stress analysis are
After welding, Deplus et al. (2011) have measured the longitu- now compared with measured residual stresses from the experi-
dinal residual stresses in the plate using the slitting method (or ments using the slitting method published by Deplus et al. (2011).
crack compliance method) proposed by Prime (1999). The slitting The longitudinal stresses are calculated as an average over the
method is a measurement method of residual stresses averaged thickness of the plate, in order to make it comparable with the
out over the thickness of a plate. It is a destructive method where experimentally found stresses.
residual stresses are deduced from the strains resulting from the From comparison of the computed residual stresses with the
relaxation of the residual stresses originally present in the work- measured values from the experiment in cold welding conditions,
piece caused by a cut introduced along the plane normal to the see Fig. 12a, it is observed that the residual stress curves from the FE
longitudinal residual stress. All measurements were obtained on analyses with kinematic or isotropic hardening, or isotropic hard-
90 mm × 86.4 mm compact tension (CT) specimens. The cuts were ening with the metallurgical softening model included give almost
performed using a wire electric discharge machine and the induced similar results. Moreover, the results from the FE analyses are in rel-
strains were measured using strain gages (EA-13-062AQ-350/LE). atively good agreement with the measured residual stresses in the
The method consists in computing the original residual stress welded specimen. The residual stress near the weld center is a bit
distribution which matches best the strains that are actually mea- lower for kinematic hardening than for isotropic hardening with-
sured. For that cause, the measured strains are converted in residual out softening. This is due to the shift of the yield surface, which
stresses using an analytical model based on the stress intensity was also seen from the preliminary Satoh test. For both measure-
factor solutions from linear elastic fracture mechanics (LEFM), see ments and simulations in cold welding condition, the well-known
Prime (1999). Note that similar tests were performed by chang- M-shape is observed, which is characteristic of age hardenable alu-
ing the cutting method, using in particular a circular saw mounted minum alloys. Deplus et al. (2011) also observed this M-shape in
on a milling machine, and no significant effect was found on the their work of describing residual stresses in aluminum alloy friction
estimated residual stress distribution. stir welds as well as Price et al. (2007) in their paper about distor-
tion control in welding by mechanical tensioning and Altenkirch
3.2.2. Numerical model et al. (2009) in their work on residual stress engineering in friction
Simulations of FSW of the two thin plates are performed in stir welds by roller tensioning.
the cold weld and hot weld conditions. Temperatures and residual For the hot welding conditions, see Fig. 12b, the results from
stresses from the model are then compared with the experiment. the FE analyses with different hardening and softening behaviors
The heat flux is symmetric. The modeled plate is in these analyses are much more diverging. The residual stresses for calculations
therefore only one half of the total welded plate and symmet- with kinematic and isotropic hardening without the softening
ric boundary conditions are applied in the mechanical model. The model have much higher longitudinal stresses near the weld cen-
coordinate system is chosen as shown in Fig. 10. The top surface ter compared to the calculation with isotropic hardening and the
is simulated to be in contact with steady air with a tempera- metallurgical softening model included. The results for the hot
ture of 15 ◦ C, seen via a convection heat transfer coefficient of welding conditions with isotropic hardening and softening model
h = 10 W/m2 K. The bottom surface is in contact with a backing plate resemble much better the measured residual stresses compared
also with a temperature of 15 ◦ C. Here the heat transfer coefficient to the calculations without this metallurgical model. This stresses
is set to h = 1000 W/m2 K as suggested by Khandkar et al. (2003). As the importance of including a metallurgical model to insure reli-
mentioned, symmetry conditions are applied on the surface of the able predictions of the residual stresses. However the computed
weld line, which means that uy = 0 for this surface. In order to sim- M-shape is wider and has a lower quantity of longitudinal stresses
ulate realistic predictions of the clamping boundaries, rollers are near the weld center compared to the measurements. It should be
M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486 483

Table 3
Location of thermocouples for cold(C) and hot(H) welding conditions. A, advancing side; R, retreating side.

Thermocouple number Longitudinal distance from Transverse distance to Depth below the upper Side
start of welding [mm] the weld center [mm] surface [mm]

C+H C+H C H C H C+H

2 345 6 8.9 1 1.6 R


3 355 7 10 1 1.6 R
4 365 8 12 1 1.6 R
5 375 9 14 1 1.6 R
6 385 10 16 1 1.6 R
7 345 6 8.9 1 1.6 A
8 355 7 10 1 1.6 A
9 365 8 12 1 1.6 A
10 375 9 14 1 1.6 A
11 385 10 16 1 1.6 A

Fig. 10. Dimensions and BC’s of the FSW model plate, and the mesh schematically shown.

Fig. 11. Comparison of temperatures from measurements and FE heat transfer analyses for: (a) cold welding conditions at 6 mm from the weld centreline, (b) hot welding
conditions at 8.9 mm from the weld centreline. (c) Comparison of maximum temperatures. A = advancing side, R = retreating side.

noted that if a strain hardenable alloy was to be modeled, e.g. a 5xxx From the mechanical model the stress history in a certain point
series aluminum alloy, the residual stress distribution predicted can be shown. For both cold and hot welding conditions, the
without softening would be most appropriate. stresses in longitudinal and transverse directions of the welding
From the microstructural model, it was possible to back out the path extracted for a point in the middle of the plate and on the
hardness via equation (6) with HVmax = 161 HV1 and HVmin = 120 weld center line are plotted for the different hardening laws as a
HV1, and compare it with measured values, see Fig. 13. The pre- function of the temperature, see Fig. 14. In these plots it is possi-
dicted hardness (following Grong’s model) does not increase in the ble to follow how the stresses in that point develop as the moving
weld center because the model described in Section 2.3 does not heat source passes by. It is seen that while heating up, stresses in
account for natural aging after welding. But as seen from Fig. 13 the compression are build up, and these drop as the von Mises stresses
model allows to predict the decrease in hardness in the softening reaches the temperature dependent yield stress. Then during cool-
zone, both for hot and cold welding conditions. ing down, tensile stresses are built up and the plate ends up having

Fig. 12. Comparison of longitudinal, normal residual stresses from measurements using the slitting method in experiments with results from finite element stress analysis
for: (a) cold welding conditions, (b) hot welding conditions.
484 M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486

as the clamps are released and the thin sheet is free to distort,
which is also seen from the results in Fig. 14. Especially for the
FSW simulation with hot welding conditions, this Satoh-plot way of
showing the stress history is in good agreement with the simple 1D
Satoh test performed in the previous application. It shows that the
difference in residual stresses by choosing kinematic or isotropic
hardening rule plays a minor role in the resulting residual stresses.
However, the microstructure model shows in hot welding condi-
tion to have a major impact on the resulting residual stresses, which
was also seen in the 1D Satoh test application. For cold welding con-
ditions, the Satoh-plot in Fig. 14a, does not show large differences in
residual stresses, which also correspond with the observed results
in Fig. 12a.
In continuation of the above-mentioned results, it is natural to
discuss why the metallurgical softening model only shows its effect
during simulation with hot welding conditions. To give a reasonable
explanation of this, the stress history in the time domain should be
investigated. Fig. 15 compares the cold and hot welding conditions
Fig. 13. Comparison of experimental and predicted Vickers hardness. giving the temperature history, fraction of precipitates and com-
pares the Mises stress with the yield stress as a function of time.
Fig. 15 shows that as the FSW tool passes by, the yield strength is
large tensile stresses before release in both cases. It is observed
decreasing as the temperature increases. Subsequently, after cool
that for the cold welding conditions (Fig. 14a) strains during cooling
down, the yield strength in both cold and hot welding conditions is
down are purely elastic. For the hot welding conditions (Fig. 14b), it
increased again, however to a lower level than the initial value due
is seen that with the softening model included, the tensile stresses
to the effect of the precipitate dissolution on the material soften-
are cut off as the von Mises stress reaches the yield stress in ten-
ing. In terms of fraction of precipitates, the cold welding conditions
sion. Another phenomenon observed is that for this welding case
show a partial precipitate dissolution after welding while the hot
the largest stresses in compression during heat up are found in the
welding conditions shows a complete precipitate dissolution after
transverse direction (S22). As the plates are rigidly clamed to the
welding. This is due to the observable difference in cooling rate
backing plate, these large transverse stresses are build up during
welding. However the transverse stresses are reduced substantially

Fig. 14. Longitudinal (S11) and transverse (S22) stresses plotted as a function of the temperature in the center point at the weld center line of the welds for kinematic and
isotropic hardening, and isotropic hardening with the metallurgical softening model activated; (a) Cold welding conditions, (b) Hot welding conditions. Compare with the
corresponding Satoh-test plots in Fig. 7.

Fig. 15. (a) Temperature distribution and softening fraction Xd as a function of time. (b) Longitudinal and Mises stresses plotted as a function of time together with the yield
strength in the weld centreline.
M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486 485

between both welds, i.e. the cold weld cools down much faster than to the larger precipitate dissolution, the Mises stress reaches the
the hot weld due to a lower traveling speed. yield strength and the material starts to yield during cool down, as
As a consequence, for the cold welding conditions, the Mises also seen from the Satoh-plot of the real FSW application in Fig. 15.
stress does not reach the higher yield strength during cool down, This shows that even though the results for the cold welding con-
while the stresses in tension remain in the elastic region. For the hot ditions with or without the softening model included did show the
welding conditions, as the aluminum alloy has softened more due same results regarding residual stresses, taking the softening model
into account is still important if a realistic prediction of the final
yield stress is the goal.
Table 4
Thermal properties of aluminum alloy 2024T3 (European Deepweld Project, 2008).
4. Conclusion
Temperature [◦ C] Density [g/cm3 ]

22 2.77 The following conditions could be drawn from the work:


100 2.76
200 2.73
300 2.71 • Results from the heat transfer analysis showed reasonably good
400 2.69 agreement between simulated and measured temperatures dur-
500 2.67
ing FSW.
• The choice of hardening law was not as determining as com-
Temperature [◦ C] Specific heat [J/gK]
pared to including a softening model in the simulation of residual
46 0.90 stresses.
100 0.91
• Under cold welding conditions, the softening model decreased
200 0.95
300 1.02 the yield stress, but this did not affect the residual stresses since
400 1.10 the von Mises stress during cooling remained lower than the
current yield stress.
Temperature [◦ C] Thermal conductivity [W/mK] • During hot welding conditions, it was seen that the decrease of
55 126 yield stress caused by softening very much affected the residual
102 129 stresses near the weld center line. This is due to a lower yield
152 139
stress after welding associated to a larger precipitate dissolution.
202 153
• Even though the softening model did not have any effect on the
252 170
302 178 residual stresses under cold welding conditions, it is still impor-
353 177 tant to include this metallurgical model in the simulation, as
403 187 prediction of the yield stress is a major factor when determining
453 194
the load carrying capacity of a welded structure.

Table 5 Acknowledgements
Mechanical properties of aluminum alloy 2024T3 (ASM Handbook, 1993).

Young’s modulus E [Gpa] Poisson’s ratio v Temperature T [◦ C] Benjamin Ducoeur and Kevin Deplus are acknowledged for their
Elasic
contribution to the experimental work. A.S. is a scientific collabo-
72 0.33 25 rator of the FRS-FNRS Belgium.
71 0.33 100
63 0.33 200
53 0.33 300
Appendix A.
45 0.33 400
See Appendix Tables A.1 and A.2.
Yield strength Equivalent plastic Temperature T Softening
 yield [MPa] strain εp [m/m] [◦ C] fraction Xd
References
Plastic
62 0 25 0 Altenkirch, J., Steuwer, A., Withers, P.J., Williams, S.W., Poad, M., Wen SW, 2009.
101 0.02 25 0 Residual stress engineering in friction stir welds by roller tensioning. Science
59 0 100 0 and Technology of Welding and Joining 14 (2), 185–192.
98 0.02 100 0 The ASM Speciality Handbook: Aluminium and Aluminium Alloys, 1993.
45 0 200 0 Bastier, A., Maitournam, M.H., Roger, F., Dang Van, K., 2008. Modelling of the residual
78 0.02 200 0 state of friction stir welded plates. Journal of Materials Processing Technology
29 0 300 0 200, 25–37.
35 0.02 300 0 Bhide, B., Michaleris, P., Posoda, M., DeLoach, J., 2006. Comparison of buckling dis-
8 0 400 0 tortion propensity for SAW, GMAW, and FSW. Welding Journal 85, 189–195.
Bussu, G., Irving, P.E., 2003. The role of residual stress and heat affected zone prop-
11 0.02 400 0
erties on fatigue crack propagation in friction stir welded 2024-T351 aluminum
4 0 500 0
joints. International Journal of Fatigue 25, 77–88.
5 0.02 500 0 Chao, Y.J., Qi, X., 1998. Thermal and thermo-mechanical modeling of friction stir
306 0 25 1 welding of aluminum alloy 6061-T6. Journal of Materials Processing and Man-
345 0.02 25 1 ufacturing Science 7, 215–233.
261 0 100 1 Chen, C.M., Kovacevic, R., 2006. Parametric finite element analysis of stress evolu-
300 0.02 100 1 tion during friction stir welding. Proceedings of the Institution of Mechanical
152 0 200 1 Engineers. Part B: Journal of Engineering Manufacture 220 (8), 1359–1371.
185 0.02 200 1 Deplus, K., Simar, A., Haver, W.V., de Meester, B., 2011. Residual stresses in
57 0 300 1 aluminium alloy friction stir welds. International Journal of Advanced Manu-
63 0.02 300 1 facturing Technology 56, 493–504.
13 0 400 1 Donne, C.D., Lima, E., Wegener, J., Pyzalla, A., Buslaps, T., 2001. Investigations on
16 0.02 400 1 residual stresses in friction stir welds. In: 3rd International Symposium on Fric-
tion Stir Welding, Kobe, Japan.
5 0 500 1
Dubourg, L., Doran, P., Larose, S., Gharghouri, M.A., Jahazi, M., 2010. Prediction
6 0.02 500 1
and measurements of thermal residual stresses in AA2024-T3 friction stir
486 M.R. Sonne et al. / Journal of Materials Processing Technology 213 (2013) 477–486

welds as a function of welding parameters. Materials Science Forum 638–642, Murphy, A., McCune, W., Quinn, D., Price, M., 2007. The characterization of friction
1215–1220. stir welding process effects on stiffened panel buckling performance. Thin-
European Deepweld Project, 2008. Detail Multi-Physics Modelling of Friction Stir Walled Structure 45, 339–351.
Welding, Final report. Myhr, O.R., Grong, O., 1991. Process modeling applied to 6082-T6 aluminum weld-
Feng, Z., Wang, X., David, S.A., Sklad, P., 2007. Modeling of residual stresses and ments. Part 1: Reaction kinetics. Part 2: Applications of model. Acta Metal 39
property distributions in friction stir welds of aluminum alloy 6061-T6. Science (11), 2693–2708.
and Technology of Welding and Joining 12 (4), 348–356. Nandan, R., Prabu, B., De, A., DebRoy, T., 2007. Improving reliability of heat transfer
Fribourg, G., Brechet, Y., Deschamps, A., Simar, A., 2011. Microstructure-based mod- and fluid flow calculations during friction stir welding of dissimilar aluminium
elling of isotropic and kinematicstrain hardening in a precipitation-hardened alloys. Welding Journal 86, 313s–322s.
aluminium alloy. Acta Materialia 59, 3621–3635. Price, D.A., Williams, S.W., Wescott, A., Harrison, J.C., Rezai, A., Steuwer, A., Peel, M.,
Genevois, C., Deschamps, A., Denquin, A., Doisneau-cottignies, B., 2005. Quantitative StaronP, Koak, M., 2007. Distortion control in welding by mechanical tensioning.
investigation of precipiatation and mechanical behaviour for AA2024 friction stir Science and Technology of Welding and Joining 12 (7), 620–633.
welds. Acta Materialia 53, 2447–2458. Prime, M.B., 1999. Residual stress measurement by successive extension of a slot:
Gerlich, A., Su, P., Yamamoto, M., North, T.H., 2008. Material flow and intermixing the crack compliance method. Applied Mechanics Reviews 52 (2), 75–96.
during dissimilar friction stir welding. Science and Technology of Welding and Richards, D.G., Pragnell, P.B., Williams, S.W., Withers, P.J., 2008. ‘Global mechanical
Joining 13, 254–264. tensioning for the management of residual stresses in welds’. Materials Science
Hattel, J.H., Nielsen, K.L., Tutum, C.C., 2012. The effect of post-welding conditions in and Engineering A A489, 351–362.
friction stir welds: from weld simulation to ductile failure. European Journal of Rosenthal, D., 1946. The Theory of Moving Sources of Heat and Its Application to
Mechanics A/Solids, http://dx.doi.org/10.1016/j.euromechsol.2011.10.004. Metal Treatments. The American Society of Mechanical Engineers.
Inada, K., Fujii, H., Ji, Y.S., Sun, Y.F., Morisada, Y., 2010. Effect of gap on FSW joint for- Satoh, K., 1972. Transient thermal stress of weld heat affected zone by both-ends-
mation and development of friction powder processing. Science and Technology fixed bar analogy. Transactions of the Japan Welding Society 3, 125–134.
of Welding and Joining 15, 131–136. Schmidt, H., Hattel, J., 2008. Thermal modelling of friction stirwelding. Scripta Mate-
Jacquin, D., de Meester, B., Simar, A., Deloison, D., Montheillet, F., Desrayaud, C., 2011. rialia 58, 332–337.
A simple Eulerian thermomechanical modeling of friction stir welding. Journal Shi, Q., Dickerson, T., Shercliff, H., 2003. Proceedings of the 4th International Sym-
of Materials Processing Technology 211 (1), 57–65. posium on FSW , TWI, Utah, USA.
James, M.N., Hughes, D.J., Chen, Z., Lombard, H., Hattingh, D.G., Asquith, D., Yates, Simar, A., Pardoen, T., de Meester, B., 2007. Effect of rotational material flow on tem-
J.R., Webster, P.J., 2007. Residual stresses and fatigue performance. Engineering perature distribution in friction stir welds. Science and Technology of Welding
Failure Analysis 14, 384–395. and Joining 12 (4), 324–333.
Hatamleh, O., Rivero, I.V., Lyons, J., 2007. Evaluation of surface residual stresses in Simar, A., Bréchet, Y., de Meester, B., Denquin, A., Gallais, C., Pardoen, T., 2012.
friction stir welds due to laser and shot peening. Journal of Materials Engineering Integrated modeling of friction stir welding of 6xxx series Al alloys: process,
and Performance 16 (5), 549–553. microstructure and properties. Progress in Materials Science 57, 95–183.
Karlsson, L., 1986. Thermal Stresses in Welding. Department of Mechanical Engi- Steuwer, A., Peel, M.J., Withers, P.J., 2006. Dissimilar friction stir welds in
neering, Luleå University of Technology. AA5083–AA6082: the effect of process parameters on residual stress. Materials
Karlsson, L., Bergqvist, E.L., Larsson, H. ‘Application of friction stir welding to Science and Engineering A A441, 187–196.
dissimilar welding’, Eurojoin 4 conference presentation, Dubrovnik, Croatia, Tutum, C.C., Hattel, J.H., 2010. Optimization of process parameters in friction stir
May 24–26 2001. Available from: http://www.msm.cam.ac.uk/phase-trans/ welding based on residual stress analysis: a feasibility study. Science and Tech-
2004/fsw3/fsw3.htm (accessed 12.04.10). nology of Welding and Joining 15 (5), 369–377.
Khandkar, M.Z.H., Khan, J.A., Reynolds, A.P., 2003. Prediction of temperature dis- Tutum, C.C., Schmidt, H.B., Hattel, J., 2008. Assesment of benchmark cases for mod-
tribution and thermal history during friction stir welding: input torque based eling of residual stresses and distortions in friction stir welding. In: Proc. 7th Int.
model. Science and Technology of Welding and Joining 8 (3), 165–174. Symp. on ‘Friction Stir Welding, Awaji Island, Japan, May, TWI.
Lee, C.Y., Choi, D.H., Yeon, Y.M., Jung, S.B., 2009. Dissimilar friction stir spot welding of Tvergaard, V., 2001. Plasticity and Creep in Structural Materials. Department of
low carbon steel and Al–Mg alloy by formation of IMCs. Science and Technology Mechanical Engineering, Solid Mechanics, Technical University of Denmark.
of Welding and Joining 14, 216–220. Yamamoto, M., Gerlich, A., North, T.H., Shinozaki, K., 2008. Cracking in dissimilar Mg
Li, T., Shi, Q.Y., Li, H.K., 2007. Residual stresses simulation for friction stir welded alloy friction stir spot welds. Science and Technology of Welding and Joining 13,
joint. Science and Technology of Welding and Joining 12 (8), 634–640. 583–592.
Liyanage, T., Kilbourne, J., Gerlich, A.P., North, T.H., 2009. Joint formation in dissimilar Zhu, X.K., Chao, Y.J., 2004. Numerical simulation of transient temperature and resid-
Al alloy/steel and Mg alloy/steel friction stir spot welds. Science and Technology ual stresses in friction stir welding of 304L stainless steel. Journal of Materials
of Welding and Joining 14, 500–508. Processing Technology 146, 263–272.

You might also like