Yuh J. Chao
Mem. ASME
X. Qi
W. Tang
Department of Mechanical Engineering,
University of South Carolina
300 S. Main, Columbia, SC 29208
Heat Transfer in Friction Stir
Welding—Experimental and
Numerical Studies
In the friction stir welding (FSW) process, heat is generated by friction between the tool
and the workpiece. This heat flows into the workpiece as well as the tool. The amount of
heat conducted into the workpiece determines the quality of the weld, residual stress and
distortion of the workpiece. The amount of the heat that flows to the tool dictates the life
of the tool and the capability of the tool for the joining process. In this paper, we formulate the heat transfer of the FSW process into two boundary value problems (BVP)—a
steady state BVP for the tool and a transient BVP for the workpiece. To quantify the
physical values of the process the temperatures in the workpiece and the tool are measured during FSW. Using the measured transient temperature fields finite element numerical analyses were performed to determine the heat flux generated from the friction to the
workpiece and the tool. Detailed temperature distributions in the workpiece and the tool
are presented. Discussions relative to the FSW process are then given. In particular, the
results show that (1) the majority of the heat generated from the friction, i.e., about 95%,
is transferred into the workpiece and only 5% flows into the tool and (2) the fraction of
the rate of plastic work dissipated as heat is about 80%. 关DOI: 10.1115/1.1537741兴
Introduction
Friction stir welding 共FSW兲 is a new joining method derived
from conventional friction welding which enables the advantages
of solid-state welding. This joining technique has been shown to
be viable for joining aluminum alloys, copper, magnesium and
other low-melting point metallic materials. Research and development are progressing to explore the potential in applying the technique to harder materials such as steel and titanium. Since FSW is
essentially solid-state, i.e., without melting, high quality weld can
generally be fabricated with absence of solidification cracking,
porosity, oxidation, and other defects typical to traditional fusion
welding.
A schematic of friction stir welding operation as applied to a
butt joint of two flat plates 共workpiece兲 is shown in Fig. 1. Typically, the workpiece is placed on a backup plate and clamped
rigidly by an anvil along the far side to prevent lateral movement
during the FSW. A specially designed cylindrical tool with a pin
protruding from the shoulder rotates with a speed of several hundreds rpm and is slowly plunged into the joint line to start the
joining process. The pin may have a diameter one-third of the
cylindrical tool and typically has a length slight less than the
thickness of the workpiece. The pin is forced or plunged into the
workpiece at the joint until the shoulder contacts the surface of the
workpiece. As the tool descends further, its shoulder surface has
friction with the top surface of the workpiece that creates heat. As
the temperature of the material under the tool shoulder elevates,
the strength of the material decreases. The tool then moves along
the joint line to start the joining process. The pin of the rotating
tool provides the ‘‘stir’’ action to the materials in the two plates to
be joined together. As the tool passes, the weld 共or the stirred and
mixed part兲 cools, thereby joining the two plates together.
One of the key elements in the FSW process is the heat generated at the interface between the tool and the workpiece which is
the driving force to make the FSW process successful. The heat
flux must keep the maximum temperature in the workpiece high
enough so that the material is sufficiently soft for the pin to stir
but low enough so the material does not melt. The maximum
Contributed by the Manufacturing Engineering Division for publication in the
JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING. Manuscript received
May 2001; Revised May 2002. Associate Editor: J. Hu.
138 Õ Vol. 125, FEBRUARY 2003
temperature created by FSW process ranges from 80% to 90% of
the melting temperature of the welding material, as measured by
Tang et al. 关1兴 and Colegrove et al. 关2兴, so that welding defects
and large distortion commonly associated with fusion welding are
minimized or avoided.
The heat flux in the FSW process is primarily generated by the
friction and the deformation process. This heat is conducted to
both the tool and the workpiece. The amount of the heat conducted into the workpiece dictates a successful FSW process, the
quality of the weld, shape of the weld, micro-structure of the
weld, as well as the residual stress and the distortion of the workpiece. The amount of the heat gone to the tool dictates the life of
the tool and the capability of the tool for the joining process. For
instance, insufficient heat from the friction could lead to breakage
of the pin of the tool since the material is not soft enough. Therefore, understanding the heat transfer aspect of the FSW process is
extremely important, not only for the science but also for improving the process. In addition, the overall efficiency in energy transfer or consumption of the FSW process is of interest as well since
energy translates to cost in a production environment.
Temperature measurement in FSW for the workpiece was performed by McClure et al. 关3兴 and Tang et al. 关1兴. A simple model
for the temperature distribution in the workpiece was proposed by
Gould and Feng 关4兴. Chao and Qi 关5,6兴 developed a moving heat
source model in a finite element analysis for studying the temperature, residual stress and distortion of the FSW process. Colegrove
et al. 关2兴 performed simulation of FSW for both thermo and material flow including the pin. Russel and Shercliff 关7兴 postulated a
heat input dependent upon the shear strength of the material. Using an assumed friction coefficient, Frigaard et al. 关8兴 arrived at a
formula for heat generation in their modeling. Song et al. 关9兴 used
a melting temperature at the interface as a moving heat source to
model the thermal fields in a FSW. Despite the importance of heat
and temperature in the tool, which is a main factor in FSW tool
design, there is no heat transfer analysis or measurement for the
tool published in the open literature except the preliminary data
reported by the authors 关10兴.
In the current paper, we report our study on the heat transfer of
the FSW process for both workpiece and the tool. Aluminum alloy
2195 共AA 2195-T8兲 plates were joined together using the FSW.
The material, AA 2195-T8, is a solution heat-treated, cold worked
Copyright © 2003 by ASME
Transactions of the ASME
convection and Q 4 is the heat transferred to the machine head in
which the tool is mounted. Energy balance requires
Q 3 ⫽Q 4 ⫹q 1
Fig. 1 Schematics of the friction stir welding
and artificially aged aluminum alloy. The yield and ultimate tensile strengths are 570 MPa and 600 MPa, respectively. The chemistry of AA 2195-T8 is Al-93.98, Cu-4.0, Li-1.0, Mg-0.5, Ag-0.4,
Zr-0.12. The main reason for studying the FSW for AA 2195 is
because this particular material can be difficult to fusion weld,
especially during multiple heat repairs. Successful application of
the FSW to joining AA2195 will increase its applications to various structure components, particularly in the aerospace industry
because of its lightweight and high strength.
The heat transfer problem is first formulated as a standard
boundary value problem and then solved by using an inverse approach combining experimental and numerical studies. Two
welds, termed a hot weld and a cold weld, are studied to compare
the variation in heat flux generation from different FSW process
parameters. The results are then discussed relative to the process,
the tool and the heat transfer and temperature distribution in both
the tool and the workpiece.
Formulation of the Boundary Value Problem and
Solution Procedure
Tool. FSW normally starts from the beginning of the joint and
finishes at the other end of the joint, as shown in Fig. 1. In the
central part of the plate along the longitudinal direction, the heat
transfer process in the tool is approximately steady state. As
shown in Fig. 2, the heat flow in the tool and the machine head
involves Q 3 , Q 4 and q 1 , where Q 3 is the heat flux to the tool
from the friction between the tool and the workpiece, q 1 is the
heat lost from the surface of the tool to the environment through
Fig. 2 Heat transfer in tool and workpiece in friction stir welding „One-half of the tool model is shown due to symmetry…
Journal of Manufacturing Science and Engineering
(1)
The maximum temperature rarely exceeds 500°C and therefore
radiation is neglected in the current study.
In a typical heat transfer problem, the heat input and output to
the system are often known and the temperature distribution in the
system can then be calculated as a BVP. In the FSW, if this route
were followed, it would require a substantial effort to determine
the input and output heat fluxes shown in Fig. 2. For instance, Q 3
is likely a function of the dynamic friction coefficient, downward
force from the tool to the workpiece, temperature, and the tribology conditions of the contacting surfaces. Each of these physical
parameters has difficulties and uncertainties in its own if one
wants to determine the actual value, e.g., the dynamic friction
coefficient as a function of speed and temperature is extremely
difficult to get.
Because of all these unknowns, an engineering approach is
adopted in this study in which an inverse method is used to determine the heat flux quantities in Eq. 共1兲. In essence, the temperatures on the tool surface are measured at several locations during
the FSW. Then a steady state finite element analysis is performed
using guessed Q 3 and a convection coefficient. Since the machine
head in the experiment is very large relative to the tool, it serves
as a heat sink and is modeled as a large body with constant ambient temperature 25°C on the outside surface. The guessed
boundary conditions, i.e., Q 3 and the convection coefficient, are
adjusted to match with the measured temperatures. The set of the
guessed values yielding the best fit to the measured temperatures
is considered as the ‘‘actual’’ values. The measurement and numerical modeling are detailed in the next sections.
Workpiece. The BVP for the workpiece is schematically
shown in Fig. 2. Energy balance at any time during the FSW,
requires
Q 1 ⫽Q 2 ⫹q 2 ⫹Q
(2)
where Q 1 is the heat flux coming from the friction between the
tool and the workpiece, Q 2 is the heat conducted from the bottom
surface of the workpiece to the backing plate on the machine, q 2
is the heat lost from the surface of the workpiece to the environment through convection, and Q is the increase of the heat content
in the workpiece. Again, radiation is neglected.
The heat transfer process in the entire workpiece is transient
and never steady state because of the term Q in Eq. 共2兲 varies with
time. However, Q 1 , except in the beginning and the end of the
process, is expected to be constant since it is a function of only the
physical conditions at the interface between the tool and the workpiece and is independent of the location of the tool on the workpiece. As such, a moving heat source model with an assumed
constant Q 1 was developed for the numerical simulation. Inverse
method procedure used for the tool is used here for the workpiece
as well. Transient temperatures at several locations in the workpiece were measured experimentally during the FSW using thermocouples. Three-dimensional finite element analyses assuming
various heat inputs and outputs were performed for the workpiece.
The set of the boundary conditions ‘‘best-fits’’ with the measured
temperatures yields Q 1 , the value of interest in the study.
Note that in a FSW process, a pin stirs the material immediately
under the tool center. Plastic work associated with this process is
not modeled in the current work. However, the heat generated
from this stirring process is included in the temperatures measured
by the thermocouples. This heat is lumped into the friction heat at
the interface of the tool and the workpiece in our analysis. This
modeling procedure, although not perfect, is believed to be a reasonable approach.
FEBRUARY 2003, Vol. 125 Õ 139
Fig. 3 Workpiece dimensions and the locations of the
thermocouples
Weld Preparation
Friction stir welding was carried out on joining two AA2195
plates having the dimensions L⫻W⫻H⫽610⫻102⫻8.1 mm
(24⫻4⫻0.32 inches) each as shown in Fig. 3. Two kinds of
joints, a normal weld and a cold weld, were welded. The normal
weld was welded with 240 RPM tool rotational speed and 2.36
mm/sec linear welding speed, while the cold weld was welded
with the same tool rotational speed but with a faster welding
speed 3.32 mm/sec using the same tool. The faster welding speed
leaves less heat to the workpiece and thus is called a cold weld.
Temperature Measurement
Transient temperatures were recorded at nine locations during
the FSW process using 36 gauge K type thermocouples. The layout of the locations of the thermocouples is shown in Fig. 3. Three
rows of the thermocouples are placed roughly in the center of the
plate along the welding direction. The thermocouples in each row
are placed at a certain depth in the plate. The top row is 2 mm and
the middle row is 4 mm from the top surface, respectively, and the
bottom row is on the back surface of the plate. Each row has three
thermocouples located at 5 mm, 12.7 mm and 25.4 mm from the
centerline of the weld. These three locations correspond to the
edge of the tool pin, the edge of the tool shoulder and a far away
point from the tool.
Holes were pre-drilled from the bottom surface of the plate. The
thermocouples were beaded at the tip and stuck at the measuring
points with high thermal conductivity glue OMEGABOND. The
transient temperatures from the thermocouples were recorded using Labview on a PC at 10 HZ.
The welding tool is made of M2 tool steel and has a shoulder
diameter 25.4 mm and pin diameter 10 mm. Rectangular grooves
schematically shown in Fig. 2 are on the tool surface to dissipate
the heat from the tool more efficiently. Five thermocouples were
beaded and attached on the welding tool surface at 1.5 mm, 4.0
mm, 12.5 mm, 19 mm, and 23 mm above the tool shoulder. They
are mounted on the outside diameter of the tool and not in the
valley of the grooves. Due to the rotation of the tool, a slip ring,
model S6 from Michigan Scientific which has double channel
K-type amplifier for thermocouples, was used for taking the signal
out from the thermocouples to the data acquisition system.
Numerical Modeling
Tool. Steady state heat transfer analysis is performed for the
tool using the commercial finite element analysis 共FEA兲 code
ABAQUS. A length of 35 mm is used for the tool which is the
length protruded from the machine head. The machine head is
140 Õ Vol. 125, FEBRUARY 2003
Fig. 4 Finite element mesh used in the tool modeling
modeled as a cylinder having radius 75 mm and length 150 mm.
To ensure that the size of the machine in the modeling is sufficiently large to not affect the results at the tool, a larger model 共for
the machine head兲 with 105 mm in radius and 210 mm in length,
is also analyzed. Figure 4 shows the finite element mesh used for
the smaller model. The model comprises of 1,069 axisymmetric
quadrilateral elements 共type DCAX4兲 with 1,166 nodes.
Temperature-dependent thermal material properties for the M2
tool steel are used in the modeling as listed in Table I. The thermal
conductivity and density at various temperatures and the specific
heat at room temperature are from 关11兴. The specific heat at elevated temperature, i.e., 550* shown in Table 1, was extrapolated
using the same ratio as that for mild steel. Linear interpolation
was used for properties at temperatures in between those listed in
Table 1.
As shown in Fig. 2, the heat input is assumed to be linearly
proportional to the distance from the center of the tool which is
derived from the assumptions 共a兲 the downward force applied to
the workpiece from the tool creates a uniform pressure between
the shoulder and the workpiece, and 共b兲 the heat is generated from
the work done by the friction force. The heat flux from the friction
is applied at the lower end of the tool and the distribution of the
rate of heat flux can then be represented by 关5兴
q共 r 兲⫽
3Q 3 r
2 ␲ r 30
for r⭐r 0
(3)
where r 0 is the outside radius of the shoulder of the tool. Q 3 is
related to other process parameters as well as discussed in 关5兴.
Transient temperatures measured at the five locations on the
tool surface are fitted with guessed heat flux Q 3 and a connective
Table 1 Thermal material properties of M2 tool steel used in
the analysis
Temperature
共°C兲
20
95
200
400
500
540
675
Thermal conductivity
共Watt/m °K兲
Specific heat
共J/Kg °K兲
Density
共Kg/m3兲
450
8100
21.3
23.5
25.6
550*
27.0
28.9
8100
Transactions of the ASME
Fig. 5 Steady state temperature distribution at the outside surface of the tool; Measurement „points…; numerical
results „lines…
heat transfer coefficient for q 1 . It was found that the heat convection is relatively insensitive to the temperature near the friction
end of the tool. The value 10 W/m2•°C was ultimately used for the
convection coefficient. Notice this number is twice the typical for
natural convection between carbon steel and air to account for the
rotation of the tool. The best fit in this procedure yields Q 3
⫽86 Watt for the normal weld and 85 Watt for the cold weld. The
difference in the numerical temperatures at the thermocouple locations from the two sizes of the machine head is less than 0.1
percent, which indicates that the smaller size used is sufficiently
large.
Figure 5 shows the steady state temperature distribution on the
outside surface of the tool from the experiment and the finite
element analysis. Temperatures at the valleys of the grooves of the
tool are not plotted in Fig. 5 and this results in the discontinuity in
presenting the numerical results. It can be seen that the ‘‘best fit’’
numerical results compare well with the measured temperatures
from the thermocouples.
Workpiece. To model the workpiece, the welding simulation
code WELDSIM is used. WELDSIM is a transient, nonlinear,
three-dimensional finite element computer code for both heat
transfer and solid mechanics analyses. The code was originally
developed to determine the transient temperature fields, residual
stress and distortion in fusion welding process with the objective
of improving the computational efficiency. Details of the WELDSIM code can be found in 关12,13兴. The WELDSIM code was
modified for FSW in the current study and is discussed in 关5兴 and
the next paragraphs. Time-dependent thermal material properties
for AA2195 are shown in Fig. 6.
The heat source Q 1 having a linear distribution of heat from the
center to the outside diameter of the tool similar to Eq. 共3兲 is
applied to the top surface of the workpiece to simulate the heat
generated from the friction. This heat source Q 1 moves along the
weld line on the top surface of the workpiece at the same speed as
the tool’s, e.g., 3.32 mm/sec in the cold weld.
In the finite element modeling, the top and bottom surfaces of
the workpiece are assumed to have two different heat convection
coefficients. The heat transfer due to radiation of the surfaces is
believed to be small and is lumped into the convection. At the top
surface, a convective heat transfer coefficient 30 W/m2•°C is used
which is typical for natural convection between aluminum and air.
At the bottom surface of the workpiece, because the workpiece is
clamped to a backup steel plate and the support of the milling
machine used in the FSW, some contact conduction resistance is
anticipated at this interface for the heat flow. Due to the lack of
physical data related to the contact resistance at this interface, the
bottom surface is modeled with a convective heat transfer coeffiJournal of Manufacturing Science and Engineering
cient to account for the heat flowing through the contact interface.
Q in Eq. 共2兲 is the heat absorbed inside the workpiece and is
included in the element formulation of the finite element model.
Therefore, it leaves only two parameters, Q 1 and the convection
coefficient at the bottom surface of the workpiece, to be
‘‘matched’’ with the measured temperature data. It was found that
Q 1 is more sensitive to the temperature near the top surface and
the convection coefficient at the bottom surface is more sensitive
to the temperature at the bottom surface when performing the
matching with experimental data.
In FSW the pin stirs the plate materials around the pin. This
action accelerates the heat conduction in this localized region. In
our analysis using WELDSIM, we applied larger heat conduction
coefficient, i.e., five times of the value at the corresponding temperature, for the region under the shoulder to account for this
mechanical stir action. Note ‘‘five’’ is arbitrary. However, we have
found that the results are not sensitive to this number. This simulation strategy has been used in fusion welding where a higher
heat conduction coefficient is suggested for weld pool region, e.g.
ten times higher by Dike et al. 关14兴, to account for the higher heat
convection of the melting metal.
One-half of the workpiece (L⫻W⫻H⫽610⫻102⫻8.1 mm) is
modeled in the WELDSIM due to symmetry. The model has 8,800
eight-node solid elements and 12,060 nodes. A time step of 0.05
seconds was used in the transient calculation. Analyses using
smaller time steps and finer meshes were also performed. It was
Fig. 6 Thermal properties of AA 2195 used in the finite element analysis
FEBRUARY 2003, Vol. 125 Õ 141
Fig. 7 Comparison of the temperature data in the workpiece from modeling and experiment, top layer; Measurement
„points…, numerical results „lines…
found that the mesh described is fine enough and the time step
0.05 seconds is sufficiently small for modeling the current problem. The total computation time on a PC with Pentium III 600
MHz is about 30 minutes for one case.
The heat input to the system from the ‘‘best fit’’ procedure
yields Q 1 ⫽1,740 Watts for the normal weld and 1,860 Watts for
the cold weld as well as 350 W/m2•°C for the heat convection
coefficient at the bottom surface. Figures 7, 8 and 9 show the
transient temperatures from the measurement and the numerical
modeling for the top layer, middle layer and the bottom layer,
respectively. The comparison is very good for both the top and
middle layers of the thermo-couples and not as good for the bottom layer. This could be due to that the analysis uses heat convection at the interface between the anvil support and the workpiece to simply the analysis and in fact it is heat conduction.
Overall, the ‘‘best-fitted’’ results are reasonably good considering
many practical parameters that are difficult to account for.
Interpretation of Results
The welding parameters used in both the normal and cold FSW
studied in the current paper are within the process window that
produces ‘‘good’’ welds. ‘‘Good’’ means the weld has the quality
as anticipated from this FSW process, e.g., 66% tensile strength of
the base metal for heat treatable aluminum alloys, free of void,
and no damage to the tool. The following discussions and comparisons are therefore based on the outcome that both welds are
‘‘good.’’
A summary of the rate of heat input to the tool and workpiece
from the study is provided in Table 2. It is shown that the rate of
the heat transferred to the workpiece, Q 1 , is slightly higher in the
cold weld case, i.e., about 7% higher. Note that the weld speeds
are 2.36 mm/sec and 3.32 mm/sec for the normal and cold welds,
respectively. The difference is (3.32⫺2.36)/2.36⫽41%. Some,
although minor, effect on heat input rate is anticipated with such a
significant difference in welding speed. The 7% difference obtained from the modeling seems reasonable. However, this 7%
could be due to the accuracy of the modeling and could also
because the tool moves faster in the cold weld case and therefore
there is more aluminum in volume to absorb the heat within a
fixed time period.
However, since these two cases have different welding speed,
the linear heat input to the workpiece 共i.e., the heat applied in a
fixed distance from the process, J/mm兲, a term used frequently in
fusion welding, differs substantially, i.e., 560 共cold weld兲 versus
Fig. 8 Comparison of the temperature data in the workpiece from modeling and experiment, middle layer; Measurement „points…,
Numerical results „lines…
142 Õ Vol. 125, FEBRUARY 2003
Transactions of the ASME
Fig. 9 Comparison of the temperature data in the workpiece from modeling and experiment, bottom layer; Measurement
„points…, numerical results „lines…
737 共normal兲 J/mm, which is about 76% in ratio. Assuming that
Q 1 is approximately constant from the beginning to the end of the
FSW process, total heat generated from the FSW in the cold weld
for the entire plate is therefore about 76% of that of the normal
weld.
The heat flux to the tool is nearly identical for the two cases.
This numerical result is consistent with the measured temperature
data from the tool where the time to reach the steady state is
nearly identical from the two cases.
Peak temperature experienced at a specific position during
welding is one of the important factors that determine the material
microstructure and therefore mechanical properties of the welded
joint. Figure 10 shows the maximum temperatures obtained from
the measurement and from the numerical modeling at various locations near the weld. In general it is shown that 共a兲 the maximum
temperature is slightly less than 450°C in both welds, 共b兲 temperaTable 2 Rate of heat input of cold weld and normal weld
Heat input to
the workpiece Q 1
Welding process
Normal weld
Cold weld
Heat input to
the welding tool Q 3
Watt or J/sec
J/mm
Watt or J/sec
1740
1860
737
560
86
85
ture difference between the top surface and the bottom surface is
about 60°C near the center of the weld, 共c兲 this temperature difference decreases as the position moves away from the weld centerline, and 共d兲 at 25 mm from the weld centerline, which is about
twice of the tool shoulder radius, temperature becomes uniform in
the thickness direction.
The duration of remaining at a certain high temperature, or the
cooling rate, at a specific position during welding is an important
factor in welding study because it affects the material microstructure formation, results in different joint mechanical properties and
induces distortion to the welded structure. Table 3 provides the
time duration for three locations at the middle layer. The temperature 250°C was chosen for discussion because aluminum typically
recrystalizes between 200°C and 300°C. It is shown in Table 3
that the time duration for the materials stayed at temperatures over
250°C in the normal weld is 1.6 times to 2.14 times longer than
those in the cold weld from a position of 5 mm to 12.7 mm from
the center of the weld, despite that the peak temperature of the
two welds is very close to each other. This conclusion is consistent with the welding speed, i.e., the cold weld is 1.4 times faster
than that of the cold weld.
Temperature contours from the surface of the workpiece are
shown in Fig. 11. Relatively, it is seen that faster welding speed
共i.e., the cold weld兲 makes higher 共lower兲 temperature gradient in
Fig. 10 Maximum temperatures experienced in the workpiece, AA 2195
Journal of Manufacturing Science and Engineering
FEBRUARY 2003, Vol. 125 Õ 143
Table 3 Time remains at temperature above or near 250°C at
the three thermocouple locations from the middle layer
Welding method
Cold weld
Normal weld
5 mm from
the center
11.2 sec
16.4 sec
12.7 mm from
the center
8.7 sec
14.3 sec
25.4 mm from
the center
⬍250°C
⬍250°C
the front 共back兲 of the tool. Furthermore, the high temperature
region near the tool is slightly bigger for the normal weld because
it has more time to conduct the heat to the surrounding materials.
This could also mean a larger weld nugget or heat affected zone.
For the tool, both the temperature distributions and the maximum temperature in the tool for the two cases are very close to
each other, as demonstrated in Fig. 12. This is consistent with the
experimental observation that the time to reach the steady state as
well as the actual transient temperatures at the thermocouples is
nearly identical for the two cases. In Fig. 12, only the tool portion
of the model is presented. Since the cold FSW generates nearly
the same maximum temperature as in the normal weld but the
process is much faster, it is thus preferred from a production point
of view.
The grooves in the tool appear to be very effective in reducing
the temperature through convection. As shown in Fig. 12, the
temperature is decreased by more than 100°C right after the first
groove from the friction surface. An optimal design for the geometry of the grooves for both the temperature and strength as well
as using forced convection, such as blowing with compressed air,
to the lower part of the tool can reduce the high temperature in the
lower portion of the tool and therefore enhance the tool life.
The temperature distribution in the tool also demonstrates that
except for the immediate region near the contact surface, the temperatures are lower than the tampering temperature of the steel,
e.g., around 400°C. Therefore the heat generated from the friction
does not reduce the strength and wearing resistance of the majority part of the tool. At bottom surface of the tool, surface coating
and even local heat treatment could then enhance the life of the
tool.
The most striking finding from this work is the distribution of
the heat flow from the friction process to the tool and to the
Fig. 11 Temperature contours from the top surface of the workpiece
Fig. 12 Temperature contours of the tool at steady state
144 Õ Vol. 125, FEBRUARY 2003
Transactions of the ASME
workpiece. As shown in Table 2, slightly less than 5% of the total
heat generated from the friction at the interface of the tool and the
workpiece flows to the tool and 95% of the total heat flows to the
workpiece, e.g., 86/((86⫹1740)⫽4.7% for the normal weld. Several factors could be attributed to this result. First, aluminum has
higher thermal conductivity than steel, i.e., 90 versus 20 W/m°C
at room temperature. As such, heat flows faster to the aluminum
side from the contacting surface. Secondly, the workpiece is very
large relative to the tool and therefore it serves as a very effective
heat sink to absorb most of the energy. It should be noted that
although only 5% of the heat is transferred to the tool, the peak
temperature in the tool is about the same as the corresponding
peak temperature in the workpiece, as shown in Figs. 10 and 12.
This information is extremely important in the design of the tool
for FSW.
In a separate experiment using a custom made MTS friction stir
welding machine, with the same material, dimensions, tool, and
welding parameters, we obtained the mechanical power input to
the FSW process as 2,535 W in the cold and 2,218 W in the
normal weld case. The power was calculated using the pressure
drop across the hydraulic spindle motor, efficiencies of the motor
and the gearbox, and the torque from the free running spindle at
the particular RPM. With these information, an excel spreadsheet
was programmed to calculate the power applied to the process.
Comparing this power consumption with the total heat generated,
i.e., 1,860⫹85⫽1,945 W in the cold and 1,740⫹86⫽1,826 W in
the normal weld, the plastic work dissipated as heat or ␤ factor is
therefore 1,945/2,535⫽0.77 in the cold weld and 1,826/2,218
⫽0.82 in the normal weld. In other words, about 20% of the
mechanical energy input to the process is stored in the metal
workpiece, neglecting the part goes to the tool. Recent experimental 关15兴 and theoretical work 关16兴 shows that the ␤ factor is typically a function of the plastic strain and the plastic strain rate.
Experimental ␤ factor for AA 2195 is not readily available. However, the data for AA 2024-T3 关17兴, which is close to AA 2195-T8
studied, indicates that the ␤ factor varies from 0.3 to 1.0 as the
plastic strain increases from 0.05 to 0.6. In FSW, the plastic strain
ranges from zero in region far away from the weld to perhaps a
few hundreds at the weld. Our result in this paper, 0.77 and 0.82
for the cold and normal weld, respectively, reveals an overall heat
dissipation rate of the FSW process.
As a final note, an inverse approach is used in the current study
to determine the total heat input to the workpiece and to the tool
in FSW. Although interesting, the uniqueness aspect of the solution in such an inverse problem in the current case was not investigated. The authors concluded, after evaluating the FSW process
abide without rigorous proof, that this inverse method procedure
is perhaps the best ‘‘engineering approach’’ for studying such a
complex manufacturing process.
Conclusion
An integrated experimental and numerical analysis is performed to study the heat transfer aspect of the friction welding
process in a normal and a cold FSW weld. The temperature and
heat flow in both the tool 共tool steel兲 and the workpiece 共aluminum alloy 2195兲 are obtained. Discussions are given for the heat
flux, temperature distribution, heat dissipation rate as well as tool
life improvement.
The most striking result from the current study is that only
about 5% of the heat generated by the friction process flows to the
tool and the rest flows to the workpiece. The ‘‘heat efficiency’’ in
FSW is thus 95%, which is very high relative to the traditional
fusion welding where the heat efficiency is typically 60 to 80%.
However it is understood that the energy transfer in FSW is from
Journal of Manufacturing Science and Engineering
mechanical to heat as well as deformation. The term ‘‘heat efficiency’’ in FSW is not quite the same as that quoted in the traditional fusion welding.
Acknowledgment
This work is sponsored by the National Science Foundation,
Cooperative Agreement No. EPS-9630167 at the University of
South Carolina, US Air Force Research Laboratory 共AFRL兲 contract No. F33615-96-D-5835, DO#0083, and US Office of Naval
Research 共ONR兲 Agreement Number N00014-00-3-0008 which
MTS Systems Corporation is the Prime Contractor. The encouragement from Dr. Kumar Jata of AFRL and Dr. George Yoder of
ONR is greatly appreciated. The authors are also indebted to Dr.
Kenneth Miller for assisting in the temperature measurement and
Mr. William Arbegast 共Lockheed Martin兲 for sharing the AA2195
thermal material properties.
References
关1兴 Tang, W., Guo, X., McClure, J. C., Murr, L. E., Nunes, A., 1988, ‘‘Heat Input
and Temperature Distribution in Friction Stir Welding,’’ Journal of Materials
Processing and Manufacturing Science, 7共2兲, pp. 163–172.
关2兴 Colegrove, P., Painter, M., Graham, D., and Miller, T., 2000, ‘‘3 Dimensional
Flow and Thermal Modeling of the Friction Stir Welding Process,’’ Proceedings of the Second International Symposium on Friction Stir Welding, June
26 –28, Gothenburg, Sweden.
关3兴 McClure, J. C., Tang, T., Murr, L. E., Guo, X., and Feng, Z., 1998, ‘‘A Thermal
Model of Friction Stir Welding,’’ Trends in Welding Research, J. M. Vitek,
et al., eds., Proceedings of the 5th International Conference, Pine Mountain,
GA, June 1–5, pp. 590–595.
关4兴 Gould, J., and Feng, Z., 1998, ‘‘Heat Flow Model for Friction Stir Welding of
Aluminum Alloys,’’ Journal of Materials Processing & Manufacturing Science,
7共2兲, pp. 185–194.
关5兴 Chao, Y. J., and Qi, X., 1998, ‘‘Thermal and Thermo-Mechanical Modeling of
Friction Stir Welding of Aluminum Alloy 6061-T6,’’ Journal of Materials Processing & Manufacturing Science, 7, pp. 215–233.
关6兴 Chao, Y. J., and Qi, X., 1999, ‘‘Heat Transfer and Thermo-Mechanical Analysis of Friction Stir Joining of AA6061-T6 Plates,’’ Proceedings of the First
International Symposium on Friction Stir Welding, June 14 –16, Rockwell Science Center, Thousand Oaks, California.
关7兴 Russel, M. J., and Shercliff, H. R., 1999, ‘‘Analytical Modeling of Microstructure Development in Friction Stir Welding,’’ Proceedings of the First International Symposium on Friction Stir Welding, June 14 –16, Rockwell Science
Center, Thousand Oaks, California.
关8兴 Frigaard, O., Grong, O., Bjorneklett, B., and Midling, O. T., 1999, ‘‘Modeling
of the Thermal and Microstructure Fields During Friction Stir Welding of
Aluminum Alloys,’’ Proceedings of the First International Symposium on
Friction Stir Welding, June 14 –16, Rockwell Science Center, Thousand Oaks,
California.
关9兴 Song, M., and Kovacevic, R., 2002, ‘‘A New Heat Transfer Model for Friction
Stir Welding,’’ Transactions of North America Manufacturing and Research
Institute, Society of Manufacturing Engineering, Vol. XXX, pp. 565–572.
关10兴 Chao, Y. J., Qi, X., and Tang, W., 2000, ‘‘Heat Transfer Analysis of Friction
Stir Joining of AA2195 Plates,’’ AeroMat 2000, 11th Advanced Aerospace
Materials and Processes Conference, 26 –29 June.
关11兴 Metals Handbook, 1990, 10th edition, ASM International Handbook, p. 775.
关12兴 Chao, Y. J., and Qi, X., 1999, ‘‘Three-dimensional Modeling of Gas Metal Arc
Welding Process,’’ Transaction of North America Manufacturing and Research
Institute, Society of Manufacturing Engineering, Vol. XXVII, pp. 117–122.
关13兴 Chao, Y. J., Zhu, X., and Qi, X., 2000, ‘‘WELDSIM-A WELDing SIMulation
Code for the Determination of Transient and Residual Temperature, Stress, and
Distortion,’’ Advances in Computational Engineering and Science, Vol. II,
Atluri, S. N., and Brust, F. W., eds., pp. 1207–1211.
关14兴 Dike, J., Cadden, C., Corderman, R., Schultz, C., and McAninch, M., 1995,
‘‘Finite Element Modeling of Multipass GMA Welds in Steel Plates,’’ Proceedings of the 4th International Conference of Trends in Welding Research,
Gatlinburg, TN, June 4 – 8.
关15兴 Macdougal, D., 2000, ‘‘Determination of the Plastic Work Converted to Heat
Using Radiometry,’’ Exp. Mech., 40共3兲, pp. 289–297.
关16兴 Rosakis, P., Rosakis, A. J., Ravicahandran, G., and Hodowany, J., 1999, ‘‘A
Thermodynamic Internal Variable Model for the Partition of Plastic Work into
Heat and Stored Energy in Metals,’’ J. Mech. Phys. Solids, 48, pp. 581– 607.
关17兴 Hodowany, J., Ravicahandran, G., Rosakis, A. J., and Rosakis, P., 2000, ‘‘Partition of Plastic Work into Heat and Stored Energy in Metals,’’ Exp. Mech.,
40共2兲, pp. 113–123.
FEBRUARY 2003, Vol. 125 Õ 145