A Process Model for Friction Stir Welding of Age Hardening

Aluminum Alloys
Ø. FRIGAARD, Ø. GRONG, and O.T. MIDLING

In the present investigation, a numerical three-dimensional (3-D) heat flow model for friction stir
welding (FSW) has been developed, based on the method of finite differences. The algorithm,
which is implemented in MATLAB 5.2, is provided with a separate module for calculation of the
microstructure evolution and the resulting hardness distribution. The process model is validated by
comparison with in-situ thermocouple measurements and experimental hardness profiles measured at
specific time intervals after welding to unravel the strength recovery during natural aging. Furthermore,
the grain structure within the plastically deformed region of the as-welded materials has been character-
ized by means of the electron backscattered diffraction (EBSD) technique in the scanning electron
microscope (SEM). Some practical applications of the process model are described toward the end
of the article.

I. INTRODUCTION which provide quantitative information about the

THE interest in friction stir welding (FSW) has gained
considerable momentum over the past few years.[1–4] This
is because the process has made it possible to implement
the advantage of solid-state bonding to aluminum plate and
profile joints, thus leading to new product designs previously
not feasible. In FSW, the rotating movement of the shoulder
and the pin generates the heat. The frictional heating contri-
butes to the formation of a plasticized layer of soft metal
beneath the tool shoulder and about the pin. The material
is then transported to the flow side of the tool due to the
imposed mechanical stirring and forging action before it
cools and forms a solid-state joint. At present, the basic
mechanism behind the weld formation and the underlying
physical processes that lead to microstructural changes dur-
ing FSW of aluminum alloys appear to be reasonably
well established.[5–20]
In recent years, significant progress has been made in the
understanding of physical processes that take place during
welding of aluminum alloys.[21–28] A synthesis of that knowl-
edge has, in turn, been consolidated into process models,
which provide a mathematical description of the relation
between the main welding variables (e.g., heat input, plate
thickness, and joint configuration) and the resulting weld
properties, based on sound physical principles.[21,24,25,28] The
components of such a model will be as follows:
(1) a heat flow model for prediction of the temperature-
time pattern during welding;
(2) kinetic models for prediction of the heat-affected zone
(HAZ) microstructure evolution (e.g., volume fraction of
hardening precipitates) as a function of temperature; and
(3) constitutive equations, based on dislocation mechanics,
Ø. FRIGAARD, Senior Engineer Metallurgist, is with the Material Com-
mand, Analytical Laboratory, Royal Norwegian Airforce, N-2027 Kjeller,
Norway. Ø. GRONG, Professor of Metallurgy, is with Department of Mate-
rials Technology and Electrochemistry, The Norwegian University of Sci-
ence and Technology, N-7491 Trondheim, Norway. O.T. MIDLING, Senior
resulting HAZ hardness or strength.
In the present investigation, this concept is further devel-
oped and applied to FSW of AA6082-T6 and AA7108-T79
aluminum alloys. The former alloy is commonly used for
structural applications, while the AA7108 has traditionally
been utilized in production of autoparts such as bumpers
and more recently in load bearing structures such as ship
deckings and stiffeners. Included in the investigation is also
a characterization of the subgrain structure that forms within
the plastically deformed region of the weld HAZ, using
the electron backscattered diffraction (EBSD) technique. By
combining information about the subgrain size with outputs
from the heat flow model, an estimate of the mean strain
rate within this region has been made via the Zener–
Hollomon equation.
The symbols and units used throughout the article are
defined in Appendix I.
II. MATHEMATICAL MODELING
In the following, the different components of the process
model are briefly outlined. A more detailed description of
the solution methods, algorithms, and computer programs
referred to in the article can be found in the Ph.D. thesis of
Frigaard.[29] The thesis itself is written in English and can
be obtained from the author on request or ordered through
the university library system.
A. Numerical Heat Flow Model
In FSW, the heat generation takes place mainly under the
tool shoulder, leading to variable thermal gradients in the
through-thickness direction of the plate. This, together with
the transient nature of the FSW process, implies that no
simple analytical solution to the heat flow problem can be
found. Instead, a numerical solution is sought, based on a
descretization of Fourier’s 2nd Law:[30]

冢 冣
Project Manager, is with Hydro Aluminium Maritime AS, N-4262 Avald- ⭸T ⭸2T ⭸2T ⭸2T q0
snes, Norway. ␳c ⫽␭ 2⫹ 2⫹ 2 ⫹ [1]
Manuscript submitted May 9, 2000. ⭸t ⭸x ⭸y ⭸z V

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, MAY 2001—1189

Fig. 1—Schematic diagram illustrating the rotation of the shaft relative to
the plate surface during FSW.

Fig. 2—Schematic representation of the heat source used in the 3-D heat
where ␳c is the volume heat capacity; x, y, and z are the flow simulations. Note that only half of the source is considered due
space coordinates; and q0 /V is the source term. The other to symmetry.
symbols have their usual meaning and are defined in Appen-
dix I.
1. Solution algorithm leading to a nonuniform heat generation during welding.
The solution algorithm is developed according to the finite These parameters are the main process variables in FSW,
difference approach, where a coupled set of ordinary differ- since the pressure P, in practice, cannot exceed the actual
ential equations is derived by integrating Eq. [1] over a flow stress of the material at the operating temperature if a
representative control volume. In the present model, an sound weld without depressions is to be obtained.
explicit solution method is used in combination with a non- In order to describe the heat generation in the numerical
uniform grid size. By this means, the computational effort model, it is convenient to divide the heat source into a fixed
can be minimized without loss of accuracy in the heat number of square grid points. According to this procedure,
flow calculations. the strength of each grid point ⌬q0 is obtained by dividing
During a weld simulation, the heat source is assumed to the net heat input q0 by the total number of grid points
move at a constant speed v in the positive x direction. For involved. By considering the individual contributions from
each time-step dt, the energy input is calculated by allowing the tool shoulder and the rotating pin, the total heat genera-
the source to advance one grid length dx after (dx/v)/dt tion during FSW can be calculated, as indicated in Figure
iterations. Following the implementation of the algorithm 2. Although the underlying assumption of a square tool
in the numerical code, the reliability of the solution with geometry in Figure 2 is not physically correct, its relevance
respect to choice of discretization (i.e., values of dx, dy, dz, to FSW is thought to be acceptable in the context of the
and dt) has been checked against outputs from the exact mathematical model being developed.
medium thick plate solution.[21] 3. Adaptive adjustment of the friction coefficient
2. Heat generation during FSW During the transient heating period, when the temperature
For the ideal case considered in Figure 1, the torque starts to build up beneath the tool shoulder, Eq. [4] is believed
required to rotate a circular shaft relative to the plate surface to provide an adequate description of the heat generation
under the action of an axial load is given by[31,32] during FSW. This is probably true as long as the individual
contributions from the plastic deformation and the variable
MR R
frictional conditions at the tool/matrix interface are
M⫽ 兰 dM ⫽ 兰 ␮P(r)2␲r
0 0
2
dr ⫽
2
3
␮␲PR3 [2] accounted for by the use of a reasonable average value for
the friction coefficient ␮. However, in thermal processing
of aluminum alloys, it is generally accepted that local melting
where M is the interfacial torque, ␮ is the friction coefficient,
will occur if the material is heated above the eutectic temper-
R is the surface radius, and P(r) is the pressure distribution
ature at a rate that does not allow the eutectic phase to fully
across the interface (here assumed constant and equal to P).
dissolve into the matrix.[23] If the conditions for local melting
If all the shearing work at the interface is converted into
are met, the presence of a liquid film at the matrix/tool
frictional heat, the average heat input per unit area and
interface will lubricate the tool shoulder resulting in the lack
time becomes[33]
of friction. In the numerical code, proper corrections have
MR R been made for the drop in the frictional coefficient at elevated
q0 ⫽ 兰 ␻ dM ⫽ 兰 ␻ 2␲␮Pr
0 0
2
dr [3] temperatures. In practice, this is done by adjusting the value
of ␮ in Eq. [4] at each time-step dt so that the temperature
beneath the tool shoulder never exceeds the chosen value
where q0 is the net power (in Watts) and ␻ is the angular of Tmax in the input file.
velocity (in rad/s).
The next step is to express the angular velocity in terms
of the rotational speed N (in rot/s). By substituting ␻ ⫽ B. HAZ Microstructure Models
2␲N into Eq. [3], we get The models used to capture the HAZ microstructure evo-
R lution during welding of 6XXX and 7XXX series aluminum
q0 ⫽ 兰 4␲
0
2
␮PNr2 dr ⫽
4 2
3
␲ ␮PNR3 [4]
alloys have been adopted from Grong,[21] Myhr et al.,[24,26–28]
and Bjørneklett et al.,[25] respectively. These models use
a combination of chemical thermodynamics and diffusion
From Eq. [4], it is obvious that the heat input depends theory to describe the dissolution, reprecipitation, and natu-
both on the applied rotational speed and the shoulder radius, ral aging kinetics occurring sequentially in time as a result of

1190—VOLUME 32A, MAY 2001 METALLURGICAL AND MATERIALS TRANSACTIONS A

 (a)
(b)
Fig. 3—Schematic diagrams showing details of HAZ microstructure evolu-
tion during welding of age hardening aluminum alloys: (a) AA6082-T6
and (b) AA7108-T6.
welding. Referring to Figure 3, dissolution of the hardening
phase (i.e., ␤ ⬙ or ␩⬘ in AA6082 and AA7108, respectively)
is the main factor contributing to strength loss during weld-
ing. At the same time, growth of the nonhardening phase
(␤ ⬘ or ␩) occurs during the cooling leg of the thermal cycle,
leading to solute depletion within the aluminum matrix. This,
in turn, reduces the precipitation potential and contributes
to the development of a permanent soft zone within the
partly reverted region of the weld HAZ after prolonged
room-temperature aging.
C. Toolbox for Simulation of FSW
Both the numerical heat flow model and the different
HAZ microstructure models are implemented in MATLAB*
*MATLAB is a trademark of The Math Works, Inc., Natick, MA.
5.2, which is a high-performance language for technical
computing. Details of the numerical code can be found in
METALLURGICAL AND MATERIALS TRANSACTIONS A
Fig. 4—Flow chart for the MATLAB simulation toolbox.
Reference [29]. Briefly, the toolbox consists of several files,
i.e., a main program for numerical integration of the differen-
tial equations and separate subroutines for the input parame-
ters, the derivatives, the time constants and the response
equations. Moreover, separate postprocessing routines have
been developed in order to present the outputs from the
computations graphically. Figure 4 shows a simplified flow
chart for the Matlab toolbox.
III. EXPERIMENTAL PROGRAM
The experimental validation of the process model is done
on the basis of special designed welding trials carried out
under controlled laboratory conditions, using a dedicated
FSW machine at Hydro Aluminium, R&D Material Technol-
ogy (Karmøy, Norway). These trials were carried out at
constant welding force F and rotational speed N of 7000 N
and 1500 rpm, respectively, at three different welding speeds,
i.e., 5, 8, and 12 mm/s.
A. Materials
The base materials, one AA6082 variant and one AA7108
variant, were received in the form of 6-mm-thick extrusions.
Their chemical compositions are summarized in Table I.
Prior to welding, the 6082.50 and 7108.50 plate materials
were subjected to standard heat treatment to obtain the
desired T6 and T79 strength, respectively. Details of the
applied heat treatment schedules are given in Table II along
with information about the resulting mechanical properties.
B. In-Situ Thermocouple Measurements
The dimensions of the plate materials used in the FSW
experiments were 100 ⫻ 300 ⫻ 6 mm. On each plate,
eight 3-mm-deep holes were drilled from the upper side at
positions that were found to be located outside the plastically
deformed region. The experimental setup is shown in Figure
5. The thermocouples were subsequently inserted into the
VOLUME 32A, MAY 2001—1191
 Table I. Nominal Chemical Compositions of AA6082 and AA7108 Extrusions in (Weight Percent)
Alloy Si Mg Zn Cu Cr Zr Mn Ti Fe Al
6082.50 0.95 0.62 0.02 0.03 — — 0.50 — 0.15 balance
1.05 0.70 — — 0.03 — 0.57 0.02 0.23 —
7108.50 — 0.70 5.20 0.10 — 0.12 0.05 — 0.30 balance
0.20 1.00 5.70 — 0.04 0.25 — 0.03 — —

Table II. Summary of Applied Heat Treatment Schedules measurements were carried at a constant load of 1 kg. Sepa-
and Resulting Base Plate Mechanical Properties (Yield and rate hardness measurements were also done in the through-
Ultimate Strength) thickness (z) direction close to the centerline in order to
Temper Heat Treatment Rp0.2 Rm detect possible local hardness variations within the plasti-
Alloy Condition Schedule (MPa) (MPa) cally deformed region of the friction stir welds.
6082.50 T6 100 ⬚C for 6 h ⫹ 310 320
150 ⬚C for 6 h D. Postweld Annealing
7108.50 T79 100 ⬚C for 6 h ⫹ 290 330
170 ⬚C for 6 h In order to investigate the thermal stability of the micro-
structure within the plastically deformed region of the welds,
small samples of the friction stir-welded AA6082-T6 and
AA7108-T79 plate materials were immersed into a salt bath
at a temperature of 510 ⬚C and 475 ⬚C, respectively, for 15
minutes. Prior to the metallurgical examination in a light
microscope, the specimens were first ground and then pol-
ished down to a 1 ␮m surface finish before anodizing, using
a reagent consisting of 760 mL H2O and 40 mL HBF4.

Fig. 5—Schematic diagram showing the locations of the thermocouples
used to record the HAZ thermal program.
holes using a copper paste to improve the contact between
the thermocouples and the aluminum. In addition, separate
welding experiments were carried out to obtain quantitative
information about the thermal conditions close to the plasti-
cally deformed region. For this reason, holes were drilled
from the backside of the plates at varying depths (ranging
from 4.5 to 5.75 mm) at a constant location y ⫽ 6 mm away
from the weld centerline. The temperature-time histories
were recorded using a digital data logger with a total capacity
of eight type-K (chromel-alumel) thermocouples. The
applied sampling interval varied from 0.6 to 0.1 seconds,
depending on the circumstances.
C. Position and Hardness Measurements
After welding, the plates were sectioned to determine the
exact position of the thermocouples referred to the centerline
of the tool shoulder. At the same time, the hardness profiles
across the HAZ were measured at specific time intervals on
both sides of the weld to quantify the response of the materi-
als to the imposed thermal cycles. The Vickers hardness
E. EBSD Analyses
The cross sections of the as-welded materials examined
in the scanning electron microscope (SEM) were first ground
and then polished down to a surface finish 1 ␮m before
electropolishing, using a reagent consisting of 78 mL per-
chloric acid, 120 mL H2O, 700 mL ethanol, and 100 mL
ethylen-glycolmonobutylether (C6H14O2). This final electro-
polishing was done to remove any previous mechanical
deformation from the surface. The specially designed EBSD
experiments were done to obtain a complete mapping of the
microstructure at specific positions in the through-thickness
direction of the welds. The examination was carried out
using a JEOL* 840 SEM equipped with NORDIF EBSP
*JEOL is a trademark of Japan Electron Optics Ltd., Tokyo.
hardware and CHANNEL ⫹ EBSD software for fully auto-
mated pattern indexing. The beam was set to scan an area
of 75 ⫻ 100 ␮m2, with a constant step size of 1 ␮m. A full
description of this system and the applied experimental setup
has been reported elsewhere.[34]
IV. EXPERIMENTAL VALIDATION OF THE
PROCESS MODEL
In the following, the heat flow model and the HAZ micro-
structure models will be validated by comparison with exper-
imental data obtained from in-situ thermocouple and Vickers
hardness measurements, respectively.
A. Thermal Field
Table III contains a summary of the base plate thermal
properties and input parameters used in the heat flow model.
Notice that uniform properties for a and ␳c have been

1192—VOLUME 32A, MAY 2001 METALLURGICAL AND MATERIALS TRANSACTIONS A

 Table III. Summary of Input Data Used in the Heat
Flow Model
␳c a Tmax
Parameter (J m⫺3 ⬚C⫺1) (m2 s⫺1) ␮ (⬚C)
Value 2.4 ⫻ 106 5.8 ⫻ 10⫺5 0.4 555

assumed for both alloys, i.e., AA6082 and AA7108, in order

to simplify the calculation procedure.
1. Choice of friction coefficient
In the simulations, it is convenient to keep the friction
coefficient constant, although the value is changing continu-
ously during the welding cycle (i.e., from unity at the dry
sliding start toward zero when the temperature for local
melting is reached at the tool/base-material interface). In
previous work done on modeling of continuous drive friction
welding, the frictional coefficient was assumed to be con- (a)
stant and equal to 0.5.[31] This value is close to that reported
for sticky friction.[31,33] Moreover, results obtained from dry
sliding tests at ambient temperatures yield a coefficient of
about 0.25 for steel sliding on aluminum.[35] Based on these
findings, an average value for the friction coefficient ␮ of
0.4 seems reasonable for both alloys during the transient
heating period, as the temperature starts to build up beneath
the tool shoulder.
2. Choice of maximum temperature
Moreover, in order to indicate a reasonable average value
for the maximum temperature, computed thermal cycles with
different input values for Tmax have been compared with
those recorded for the high peak temperature zone of the
weld adjacent to the plastically deformed region. The com-
parison shows that a maximum temperature of 555 ⬚C is a
reasonable compromise under the prevailing circumstances,
as indicated in Figure 6. This value agrees well with the
reported eutectic temperature of Al-Mg-Si alloys[36] and falls
(b)
between the eutectic and solidus temperature of the AA7108
alloy, stated to be 475 ⬚C and 606 ⬚C, respectively.[37,38] In Fig. 6—Comparison between measured and computed thermal histories
the following, the value of Tmax has been fixed to 555 ⬚C for thermocouples located adjacent to the plastically deformed region
beneath the tool shoulder for a constant input value of Tmax of 555 ⬚C: (a)
throughout the calculations, independent of the applied AA6082 and (b) AA7108.
experimental conditions. This procedure is similar to the
one adopted in the earlier study of heat flow phenomena
in FSW.[39] than the measured ones. Besides a minor discrepancy
3. Predictive power of the heat flow model between predicted and measured peak temperatures, which
Figure 7 shows a comparison between predicted and mea- can be attributed to the steep thermal gradients within this
sured HAZ thermal cycles for three different welding speeds, region, the overall impression is that the thermal model
using input data from Table III. These examples are represen- adequately reproduces the heating and cooling conditions
tative of the degree of accuracy obtained in the present within the central parts of the HAZ. This is evidenced by
investigation.[29] the fact that the predicted and measured thermal cycles tend
There seems to be a general trend that the computed peak to be parallel and just slightly shifted in time, as shown in
temperatures are consistently higher than the measured ones. Figure 7.
This result is to be expected, as the hot side is used as a Moreover, by comparing the thermal cycles from the shear
basis for comparison when the position of the thermocouple and the flow side of the weld (not shown here), they all
falls between two grid points. As shown in Figure 8, the show symmetry across the weld centerline indicating an
thermal gradient in the transverse ( y) direction of the plate axisymmetric heat source.[29] This observation is not surpris-
is of the order of 15 ⬚C to 25 ⬚C per mm within the high ing, considering the relatively short time required for one
peak temperature region of the weld HAZ. Considering the tool rotation compared to the total tool/base-plate interaction
fact that the diameter of the fused thermocouple head is 1.2 time. This implies that conductive heat transfer quickly
mm (which means that the sampling volume is even larger), it evenly distributes any circumferencial variation in the heat
is not surprising to find that the computed peak temperatures, generation, in agreement with the underlying assumptions
with some exceptions, are consistently 20 ⬚C to 30 ⬚C higher of the heat flow model.

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, MAY 2001—1193


(a)
(b)
Fig. 7—Comparison between computed and measured thermal cycles for
three different welding speeds: (a) AA6082 and (b) AA7108.
Fig. 8—Computed peak temperature contour plot in the transverse ( y)
direction of the plate at a constant travel speed of 8 mm/s. Numbers refer
to temperature in degrees Celsius. Note that only half the plate is considered
due to symmetry.
B. Microstructure Fields
In the following, the outputs from the microstructure mod-
els will be compared with the measured HAZ hardness pro-
files at different times of natural aging, based on the input
data given in Tables IV and V.
1. AA6082-T6 weldments
In general, the process model is seen to predict adequately
the response of the base material to FSW, as shown in Figure
9. An exception is welding at high speeds, i.e., 12 mm/s
(Figure 9(c)), where parts of the HAZ fall within the plasti-
cally deformed region beneath the tool shoulder. In that case,
the predicted dissolution kinetics are too slow to account
for the observed strength loss. The deviation arises probably
1194—VOLUME 32A, MAY 2001
from the fact that a plastically deformed material will
respond much quicker to reheating than an undeformed
material because the dislocation tangles within the substruc-
ture will act as high diffusivity paths for the solute atoms.
In the present process model, this “short circuiting” can be
simulated by reducing the input value for the time constant
t*r1 in Table 4 from 200 to, say, 20 seconds so that full particle
dissolution is achieved within a narrow region beneath the
tool shoulder during the welding operation (i.e., about ⫾2
mm from the weld centerline). If such an empirical correction
is included, a satisfactory agreement between theory and
experiments is also obtained for the highest welding speed,
as shown by the dotted hardness profile in Figure 9(c).
2. AA7108-T79 weldments
Figure 10 shows a comparison between measured and
predicted hardness profiles for the AA7108-T79 welds. In
general, the overall agreement between theory and measure-
ments is good in the sense that the observed strength recovery
after welding is adequately captured in the simulations. An
interesting feature is that these welds do not reveal the char-
acteristic drop in hardness within the partly reverted region
of the HAZ, as frequently observed during fusion welding
of similar materials.[25] It follows from Figure 10 that the
strength recovery due to natural aging occurs more or less
evenly across the entire soft zone of the welds. This suggests
that the thermal conditions existing in FSW are different
from those normally experienced during conventional fusion
welding, where the total heat dissipation is significantly
greater.
V. CHARACTERISATION OF THE
DEFORMATION MICROSTRUCTURE
In the literature, both dynamic recovery and dynamic
recrystallization have been proposed as mechanisms for the
microstructure evolution within the plastically deformed
region of friction stir aluminum welds.[2,40–42] This situation
probably reflects different materials and experimental condi-
tions (e.g., welding and rotational speed, tool geometry, and
plate thickness). It is therefore of interest to characterize the
specific deformation microstructure that forms under the
prevailing circumstances and to clarify to what extent work
hardening contributes to the overall strength of the weld
following subsequent natural aging.
A. Response to Postweld Heat Treatment
The results from postweld annealing experiments are
shown in Figures 11 and 12. When comparing the micro-
structures in the as-welded condition (Figure 11) with the
microstructures obtained after postweld annealing (Figure
12), it is evident that some regions of the weld have recrystal-
lized whereas other regions seem unaffected. This probably
reflects differences in the temperature and deformation pat-
tern during welding.
B. EBSD Orientation Maps
Based on the results obtained from the postweld annealing
experiments, it is possible to divide the plastically deformed
part of the weld into two distinct regions, i.e., one recrystal-
lized and one nonrecrystallized. In order to reveal possible
METALLURGICAL AND MATERIALS TRANSACTIONS A
 Table IV. Summary of Input Data Used in the AA6082-T6 Microstructure Model[24] (Symbols and Abbreviations Are Defined
in Appendix I)
Parameter Value Comments
Qs 30 kJ mol⫺1 reasonable value for the metastable solvus boundary enthalpy
Qd 130 kJ mol⫺1 activation energy for diffusion of Mg in aluminum
t*r1 (at 375 ⬚C) 200 s* obtained by calibration to experimental data
A0 3.6 ⫻ 108 J K2 mol⫺1 obtained from an analysis of experimental literature data
Ts 520 ⬚C estimated value for the metastable ␤⬘ solvus temperature
r2 (at 350 ⬚C)
t* 3s obtained by calibration to experimental data
HVmax 110 VPN obtained from hardness measurements
HVmin 42 VPN obtained from hardness measurements
[26]
*The value is taken from Ref.

Table V. Summary of Input Data Used in the AA7108-T79 Microstructure Model[25] (Symbols and Abbreviations Are
Defined in Appendix 1)
Model Parameter Value Comments
Thermodynamic model ␥ 0.25 J m⫺2
fixed for all precipitates
Vm 2 ⫻ 10⫺5 m3 mol⫺1 fixed for all precipitates
r0(␩⬘) 3.7 ⫻ 10⫺9 m reasonable average value
r0 (GP/zones) 1 to 1.5 ⫻ 10⫺9 m typical range reported for GP zones[37]
Kinetic model D0 1.05 ⫻ 10⫺4 m2 s⫺1 reasonable average value[38]
Q 127.9 kJ mol⫺1 reasonable average value[38]
Cp 85 wt pct fixed for all precipitates
f0 0.016 chosen
fm 0.018 calculated from a mass balance, assuming stoichiomet-
ric compound formation
Strength model Somax 120 VPN hardness in T79 temper condition
Somin 50 VPN hardness in fully reverted condition
Co 5.4 wt pct typical Zn content in base material
Com 0.56 wt pct calculated value

differences in the subgrain structure between these two C. Contribution from Work Hardening
regions, detailed EBSD mapping of the as-welded specimens
The results from the local Vickers hardness measurements
was performed. This was done at the locations indicated in
are shown in Figure 16. These measurements were done in
Figure 13.
the through-thickness direction of the welds close to the
Figures 14 and 15 show representative EBSD orientation
centerline. It follows from Figure 16 that the local hardness
maps for the AA6082-T6 and AA7108-T79 weldment,
within the plastically deformed region of the welds is not
respectively. Figures 14(a) and 15(a) show the areas that
significantly affected by the observed variation in the sub-
did not recrystallize after postweld annealing, while Figures
grain size in the through-thickness direction. This means
14(b) and 15(b) show the microstructure within the recrystal-
that precipitation strengthening due to natural aging will
lized regions. Superimposed on these maps is also the spatial
completely over-ride the contribution from work hardening
misorientation between the individual grains. Here, low-
when the subgrain size is, say, 2.5 ␮m or larger. Both alloys
angle grain boundaries are characterized by misorientations
show the same behavior, but the strength recovery following
between 1.5 to 15 deg and are indicated by the thin lines
natural aging is most significant in the 7108-T79 weldment.
in the maps. Conversely, high-angle grain boundaries are
defined by misorientations greater than 15 deg and are
marked as thick solid lines. VI. CASE STUDIES
It is evident from these orientation maps that the micro-
structure within the plastically deformed region of both In the following, the aptness of the process model will
welds consists of a mixture of low- and high-angle grain be illustrated in two different case studies.
boundaries, as frequently observed after dynamic recov-
ery.[43] Moreover, there is a clear difference in the subgrain
A. Estimation of Local Strain Rates During FSW
structure between the two regions examined, i.e., the finest
subgrain size is observed within the area that did recrystallize In general, the Zener–Hollomon (Zh) parameter provides
after postweld annealing. This difference is quantified in a basis for evaluating the evolution of the subgrain structure
the captions of Figures 14 and 15, which contain average during hot deformation. As shown by McQueen and
numbers for the subgrain size in regions A and B, as deter- Jonas,[44], the following relationship exists between the sub-
mined by means of the mean linear intercept technique. grain diameter ds and the Zh parameter in aluminum alloys:

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, MAY 2001—1195

 (a) (a)

(b) (b)

(c)
Fig. 9—Comparison between measured and predicted hardness profiles in
AA6082 friction stir weldments after 270 h of natural aging: (a) v ⫽ 5
mm/s, (b) v ⫽ 8 mm/s, and (c) v ⫽ 12 mm/s. Solid and broken curves in
Fig. c refer to a value of t*r1 of 200 and 20 s, respectively. Input data as in
Table IV.
(c)
Fig. 10—Comparison between measured and predicted hardness profiles
in AA7108 friction stir weldments after 140 and 2600 h of natural aging,
respectively: (a) v ⫽ 5 mm/s, (b) v ⫽ 8 mm/s, and (c) v ⫽ 12 mm/s. Input
data as in Table V.

and Tp is the peak temperature of the thermal cycle in a

ds ⫽ [⫺0.60 ⫹ 0.08 log Zh]⫺1 [5] given position (in Kelvin).
1. Predicted values
where By inserting the measured values for the subgrain size

冋 册
from Figures 14 and 15, and using the three-dimensional
18,772 (3-D) heat flow model to calculate the peak temperature
Zh ⫽ ␧˙ exp [6]
Tp distribution during FSW, it is possible to estimate the local

1196—VOLUME 32A, MAY 2001 METALLURGICAL AND MATERIALS TRANSACTIONS A


Fig. 11—Optical micrographs showing the plastically deformed region in
the as-welded condition: (a) AA6082-T6 and (b) AA7108-T79.
Fig. 12—Optical micrographs showing the microstructure within the plasti-
cally deformed region following postweld heat treatment: (a) AA6082-T6
and (b) AA7108-T79.
Fig. 13—Schematic diagram showing the recrystallized and nonrecrystal-
lized regions within the plastically deformed material after postweld anneal-
ing. The corresponding locations of the EBSD measurements are also
indicated.
METALLURGICAL AND MATERIALS TRANSACTIONS A
(a)
(b)
Fig. 14—EBSD orientation maps for AA6082-T6 in the as-welded condi-
tion: (a) area that is unaffected by postweld annealing and (b) area that
did recrystallize after postweld annealing. The average subgrain size within
regions A and B are 3.5 and 4.8 ␮m, respectively.
strain rate within the plastically deformed region, as shown
in Table VI. Notice that the reported differences in Tp
between the two weldments are due to the different welding
speeds used in the experiments, i.e., 8 mm/s for the AA6082-
T6 alloy and 12 mm/s for the AA7108-T79 alloy.
The calculations suggest that the local strain rate within
the plastically deformed region varies from about 1 to 20
s⫺1. These values are quite low compared with the angular
velocity of the tool, which, at least, is one order of magnitude
higher. It is important to realize that the estimated values in
Table 6 are conservative in that they represent upper limits
for the local strain rate. This is because the peak temperature
of the thermal cycle in each position is used as input to the
Zener–Hollomon equation. In practice, the evolution of the
subgrain structure is the result of a series of events that
occur during heating and cooling. Hence, in a real welding
situation, both the effective temperatures and the real strain
rates may be significantly lower than the values indicated
in Table VI.
2. Practical implications
The results in Table VI suggest that the condition for
sticky friction at the matrix/tool interface is not met under
the prevailing circumstances, probably because of asperity
(or local) melting. The phenomenon of local melting is com-
monly observed in such aluminum alloys during rapid heat-
ing.[23] A requirement is that the material is heated above
the eutectic temperature at a rate which does not allow the
eutectic phase to fully dissolve into the matrix.[21,23] If the
VOLUME 32A, MAY 2001—1197
 Table VI. Estimated Strain Rates within the Plastically
Deformed Region of the Experimental Friction Stir
Welds
Alloy Position* Tp (⬚C) ds (␮m) Zh ␧˙ (s⫺1)
6082-T6 B 550 4.8 1.28 ⫻ 1010 1.6
A 540 3.5 1.10 ⫻ 1011 10.3
7108-T79 B 500 3.8 5.82 ⫻ 1010 1.7
A 450 2.5 3.26 ⫻ 1012 17.3
*Defined in Fig. 13.

(a)

(a)

(b)
Fig. 15—EBSD orientation maps for AA7108-T79 in the as-welded condi-
tion: (a) area that is unaffected by postweld annealing and (b) area that
did recrystallize after postweld annealing. The average subgrain sizes within
regions A and B are 2.5 and 3.8 ␮m, respectively.
Fig. 16—Hardness profiles in the through-thickness (z) direction of the
welds close to the centerline. The hardness measurements are done in the
as-welded condition after 12 months of natural aging.
conditions for local melting are met, the presence of a liquid
film at the matrix/tool interface will lubricate both the shoul-
der and the pin, thus resulting in the lack of friction. Conse-
quently, local melting may have an indirect effect on the
observed deformation pattern during FSW by reducing the
effective strain rate in the vicinity of the rotating tool.
(b)
Fig. 17—Predicted isothermal contours in the length direction of the plate
under different welding conditions: (a) use of an insulated backing plate
and (b) use of a steel backing plate. Numbers refer to temperature in
degrees Celsius.
B. Optimization of Welding Conditions
During FSW, a considerable amount of heat may be con-
ducted through the steel backing plate if it is not properly
insulated from the aluminum extrusion by means of a protec-
tive ceramic layer. In certain cases, this may be desirable
(e.g., when welding is carried out at very low speeds) because
the use of a steel backing plate will reduce the HAZ strength
loss and thereby improve the mechanical integrity of the
joint. In other cases (e.g., in production of ship deckings),
where a high welding speed is required, the heat generation
itself will be a limiting factor.[39] Under such conditions,
thermal insulation of the backing plate is a better solution
for the reasons given subsequently.
In the simulations, the heat generated by the rotating
action of the tool is allowed to diffuse into a 25-mm-thick
steel backing where the lower surface is kept at a constant
temperature of 25 ⬚C. Conversely, welding on an insulated
backing plate is simulated by assuming that both the upper
and the lower surfaces of the extrusion are impermeable to
heat. The operating conditions are the same as those
employed above for welding at a constant speed of 8 mm/s.
2. Simulation results
Figure 17 shows how the thermal conditions in the front
of the tool are influenced by the use of a steel backing plate.
Here, the position x ⫽ 0 defines the center of the tool
shoulder. It is obvious that the heat diffusion into the steel
backing plate reduces the peak temperatures beneath the tool
shoulder at the same time as the thermal gradients ahead of
the tool become considerably steeper.
Because the use of a steel backing plate implies that the

1198—VOLUME 32A, MAY 2001 METALLURGICAL AND MATERIALS TRANSACTIONS A

material is not adequately preheated before it reaches the D0 diffusion coefficient (m2 s⫺1)
tool shoulder, the flow stress will be so high that the material ds subgrain diameter (m)
cannot adequately accommodate the subsequent plastic dx, dy, dz discretization parameters in x, y, and z direc-
deformation. This will eventually lead to pin fracture and tions (m)
the formation of weld defects. This “cold” weld problem f0 initial volume fraction of precipitates in base
is well known from ordinary production welding, but can material
perhaps be overcome by simply employing an isolated back- fm maximum possible volume fraction of hardening
ing plate, as illustrated in the case study. The increased precipitates that can form at absolute zero
accumulation of heat does, however, lead to a reduced HAZ HVmax hardness in the temper condition (VPN)
strength level, particularly in the case of Al-Mg-Si alloys HVmin hardness in the fully reverted condition (VPN)
welded in the T6 temper condition.[45] M interfacial torque (Nm)
N rotational speed (rot⭈s⫺1)
P pressure (Pa)
VII. CONCLUSIONS P(r) pressure distribution across the interface (Pa)
The basic conclusions that can be drawn from the present Q activiation energy for diffusion (J mol⫺1)
investigation are the following. Qd activation energy for diffusion of Mg in Al (J
mol⫺1)
1. In a calibrated form, the 3-D numerical heat flow model Qs activation energy for diffusion of the less mobile
yields a temperature-time pattern that is consistent with constitutive atom of the precipitates (J mol⫺1)
that observed experimentally. Typically, the computed q0 net power (W)
peak temperatures are 20 ⬚C to 30 ⬚C higher than the R tool radius (m)
measured ones. This discrepancy is acceptable and within r two-dimensional radius vector (m)
the accuracy of the thermocouple measurements. r0 initial particle radius (m)
2. The results from the model simulations show that thermal Somax hardness or strength in age-hardened base mate-
effects are the main cause for the observed strength loss rial (VPN or Pa)
during FSW of age hardening aluminum alloys. Only in Somin hardness or strength in fully reverted condition
cases where the welding speed is high will the plastic (VPN or Pa)
deformation introduced by the rotating tool accelerate T temperature (⬚C or K)
the dissolution of the hardening precipitates. This is Tmax maximum temperature (⬚C or K)
because the HAZ and the plastically deformed region Tp peak temperature (⬚C or K)
overlap at low heat inputs. Ts phase boundary solvus temperature (⬚C or K)
3. The indications are that the deformation microstructure, t time (s)
as revealed by means of the EBSD technique, consists t*r1 maximum hold time for complete dissolution at
of a mixture of low- and high-angle grain boundaries reference temperature (s)
in the as-welded condition. Typically, the subgrain size t*r2 time taken to precipitate a certain fraction of ␤⬘-
within the plastically deformed region varies between Mg2Si at a chosen reference temperature (s)
2.5 and 5 ␮m. This shows that dynamic recovery is a v welding speed (m s⫺1)
significant softening process in FSW of age hardening V unit volume (m3)
aluminum alloys. Vm molar volume (m3 mol⫺1)
x x-axis/welding direction (m)
y y-axis/transverse direction (m)
ACKNOWLEDGMENTS Zh Zener–Hollomon parameter (s⫺1)
z z-axis/thickness direction
The authors acknowledge the financial support provided ␳c volume heat capacity (J m⫺3 ⬚C⫺1)
by Hydro Aluminum, Kværner, and the Norwegian Research ␧˙ strain rate (s⫺1)
Council through the Norwegian Aluminum in Ships pro- ␥ interfacial energy (J m⫺2)
gram. They are also grateful to Mr. Sverre Guldbrandsen- ␮ friction coefficient
Dahl and Professor Jarle Hjelen, Department of Materials ␻ angular velocity (rad/s)
Technology and Electrochemistry, The Norwegian Univer-
sity of Science and Technology, for the assistance provided
during the EBSD measurements. REFERENCES
1. O.T. Midling: Proc. 4th Int. Conf. on Aluminium Alloys—Their Physi-
APPENDIX I cal and Mechanical Properties, Atlanta, GA, Sept. 1994, T. Sanders
and E.A. Starke, eds., Georgia Institute of Technology School of
LIST OF SYMBOLS Material Science and Engineering, Atlanta, GA, vol. I, pp. 451-58.
A0 material constant related to the potency of the 2. C.J. Dawes and W.M. Thomas: Welding J., 1996, vol. 75 (3), pp. 41-45.
3. O.T. Midling and H.G. Johansen: 6th Int. Aluminium Technology Semi-
heterogeneous nucleation sites in actual alloy nar & Exposition, ET96, Chicago, IL, May 14–17, 1996, R.I. Werner,
(J mol⫺1) R. Peacock, and S. James, eds., Aluminium Association and the Alu-
a thermal diffusivity (m2 s⫺1) minium Extruders Council, IL, USA, vol. II, pp. 451-58.
Co total alloy content (wt pct) 4. O.T. Midling: Proc. Aluminium 97 Conf., Sept. 24–25, 1997, Essen,
Germany, Argus Business Media Ltd., United Kingdom, and Rumrest
Com matrix solute content in stabilised base material GmbH, Germany, DMG Business Media Ltd., London, United King-
(wt pct) dom, pp. 26/1-26/6.
Cp solute concentration within the particle (wt pct) 5. C.J. Dawes: Welding Met. Fabrication, 1995, vol. 63, pp. 13-16.

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 32A, MAY 2001—1199

 6. C.J. Dawes: Proc. 6th Int. Symp. on “The Role of Welding Science
and Technology in the 21st Century”, Nagoya, Japan, 1996, Japan
Welding Society, Tokyo, Japan, pp. 711-18.
7. S. Kallee, D. Richardson, and I. Henderson: Schweissen & Schneiden
(Welding & Cutting), 1997, vol. 49, pp. 904-09.
8. K.E. Knipström and B. Pekkari: Welding J., 1997, vol. 76, pp. 55-57.
9. J. Hagström and R. Sandström: Sci. Technol. Welding Joining, 1997,
vol. 2, pp. 199-208.
10. M. Enomoto: J. Light Met. Welding Constr., 1998, vol. 36, pp. 25-29.
11. W.M. Thomas and E.D. Nicholas: Mater. Desember, 1997, vol. 18
(4–6), pp. 267-73.
12. M.B. Ellis and M. Strangwood: Mater. Sci. Technol., 1996, vol. 12,
pp. 970-77.
13. C.G. Rhodes, M.W. Mahoney, W.H. Bingel, R.A. Spurling, and C.C.
Bamton: Scripta Mater., 1997, vol. 36, pp. 69-75.
14. M.W. Mahoney, C.G. Rhodes, J.G. Flintoff, R.A. Spurling, and W.H.
Bingel: Metall. Mater. Trans. A, 1998, vol. 29A, pp. 1955-64.
15. G. Liu, L.E. Murr, C.-S. Niou, J.C. McClure, and F.R. Vega: Scripta
Mater., 1997, vol. 37, pp. 355-61.
16. L.E. Murr, G. Liu, and J.C. McClure: J. Mater. Sci., 1998, vol. 33,
pp. 1243-51.
17. Y.S. Sato, H. Kokawa, M. Enomoto, and S. Jogan: Metall. Mater.
Trans. A, 1999, vol. 30A, pp. 2429-37.
18. Y.S. Sato, H. Kokawa, M. Enomoto, S. Jogan, and T. Hashimoto:
Metall. Mater. Trans. A, 1999, vol. 30A, pp. 3125-30.
19. M.J. Russel and H.R. Shercliff : Proc. Inalco’98 7th Int. Conf., TWI,
Cambridge, United Kingdom, April 16, 1998, The Welding Institute
(TWI), Cambridge, United Kingdom, vol. 2, pp. 185-95.
20. A.P. Reynolds, W.D. Lockwood, and T.U. Seidel: Proc. 7th Int. Conf.
on Aluminium Alloys—Their Physical and Mechanical Properties,
Charlottesville, VA, Apr. 2000, E.A. Starke, T.H. Sanders, and W.A.
Cassada, Trans Tech Publications, Zuerich, Switzerland, NH, Part 3,
pp. 1719-24.
21. Ø. Grong: Metallurgical Modelling of Welding, 2nd ed., The Institute
of Materials, London, 1997.
22. B.I. Bjørneklett, Ø. Grong, O.R. Myhr, and A.O. Kluken: Acta Mater.,
1998, vol. 46, pp. 6257-66.
23. B.I. Bjørneklett, Ø. Grong, O.R. Myhr, and A.O. Kluken: Sci. Technol.
Welding Joining, 1999, vol. 4, pp. 161-69.
24. O.R. Myhr and Ø. Grong: Acta Metall. Mater., 1991, vol. 39, pp. 2693
and 2703-08.
25. B.I. Bjørneklett, Ø. Grong, O.R. Myhr, and A.O. Kluken: Metall.
Mater. Trans. A, 1999, vol. 30A, pp. 2667-77.
26. O.R. Myhr, Ø. Grong, S. Klokkehaug, H.G. Fjær, and A.O. Kluken:
Sci. Technol. Welding Joining, 1997, vol. 2, pp. 245-53.
27. O.R. Myhr, Ø. Grong, S. Klokkehaug, H.G. Fjær, and A.O. Kluken:
Welding J., 1998, vol. 77, pp. 286-92.
28. O.R. Myhr, Ø. Grong, S. Klokkehaug, H.G. Fjær, and A.O. Kluken:
Proc. 5th Int. Conf. on Trends in Welding Research, Pine Mountain,
1200—VOLUME 32A, MAY 2001
GA, June 1–5, 1999, ASM INTERNATIONAL, Materials Park, OH,
1999, pp. 233-38.
29. Ø. Frigaard: Ph.D. Thesis, Norwegian University of Science and Tech-
nology, Trondheim, 1999, IME-Report No. 1999-5.
30. H.S. Carslaw and J.C. Jaeger: Conduction of Heat in Solids, Oxford
University Press, Oxford, United Kingdom, 1959.
31. O.T. Midling and Ø. Grong: Acta Metall. Mater., 1994, vol. 42, pp.
1595-1609 and 1611-22.
32. B. Crossland: Friction Welding, Cont. Phys., 1971, vol. 12 (6), pp.
559-74.
33. H.S. Kong and M.F. Ashby: “Case Studies in the Application of
Temperature Maps for Dry Sliding,” Engineering Department Report,
Cambridge University, Cambridge, United Kingdom, 1991.
34. J. Hjelen: Proc. 3rd Int. Conf. on Aluminium Alloys—Their Physical
and Mechanical Properties, Trondheim, Norway, June 1992, The Nor-
wegian Institute of Technology, Trondheim, vol. II, pp. 408-13.
35. Friction Data Guide: An Engineering Study of Coefficient of Friction
of Materials and Coatings, Slide-chart presentation from General Mag-
naplate Corp., Linden, NJ, 1988.
36. O. Reiso: Proc. 3rd Int. Aluminium Extrusion Technology Seminar,
Atlanta, GA, 1984, Aluminium Association, Washington, DC, 1984,
vol. 1, pp. 31-40.
37. J.E. Hatch: Aluminium, Properties and Physical Metallurgy, ASM,
Metals Park, OH, 1984.
38. P.E. Drønen and N. Ryum: Metall. Mater. Trans. A, 1994, vol. 25A,
pp. 521-30.
39. Ø. Frigaard, Ø. Grong, B. Bjørneklett, and O.T. Midling: Proc. 1st
Int. Symp. on Friction Stir Welding, Thousand Oaks, CA, June 1999,
The Welding Institute (TWI), Cambridge, United Kingdom, 1999.
40. O.V. Flores, C. Kennedy, L.E. Murr, D. Brown, S. Pappu, B.M. Nowak,
and J.C. McClure: Scripta Mater., 1998, vol. 38 (5), pp. 703-08.
41. H.S. Yang: Proc. 6th Int. Conf. on Aluminium Alloys, ICAA-6, Toyo-
hashi, Japan, July 5–10, 1998, T. Sato, S. Kumai, T. Kobayashi, and
Y. Murakami, eds., The Japan Institute of Light Metals, Tokyo, vol.
3, pp. 1483-88.
42. A.F. Norman, I. Brough, and P.B. Pragnell: Proc. 7th Int. Conf. on
Aluminium Alloys—Their Physical and Mechanical Properties, Char-
lottesville, VA, Apr. 2000, E.A. Starke and W.A. Cassada, eds., Trans
Tech Publications, Zuerich, Switzerland, NH, Part 3, pp. 1713-18.
43. F.J. Humphereys and M. Hatherly: in Recrystallization and Related
Annealing Phenomena, Pergamon, Elsevier Science Ltd., Oxford,
United Kingdom, 1996, pp. 363-92.
44. H.J. McQeen and J.J. Jonas: Plastic Deformation of Materials, Aca-
demic Press, New York, NY, 1975, vol. 6, pp. 393-493.
45. Ø. Frigaard, B.I. Bjørneklett, Ø. Grong, O.R. Myhr, and O.T. Midling:
Proc. 6th Int. Conf. on Aluminium Alloys—Their Physical and
Mechanical Properties, Toyohashi, Japan, July 1998, T. Sato, S.
Kumai, T. Kobayashi, and Y. Murakami, eds., The Japan Institute of
Light Metals, Tokyo, 1998, vol. III, pp. 1477-82.
METALLURGICAL AND MATERIALS TRANSACTIONS A