Professional Documents
Culture Documents
Abstract
This paper describes the application of the computational fluid dynamics (CFD) code, FLUENT, to modelling the 3-dimensional metal
flow in friction stir welding (FSW). A standard threaded tool profile is used for the analysis and features such as the tool rake angle, heat
generation and heat flow are included. The primary goal is to gain a better understanding of the material flow around a complex FSW tool and
to demonstrate the effect of the tool rake angle, and weld and rotation speed. The model captured many of the real process characteristics, but
gave poor predictions of the welding forces. The model also generated an excessive amount of heat, which led to a large over-prediction in
the weld temperature. These shortcomings can be overcome by using a viscosity relationship that includes material softening near the solidus
or material slip at the tool interface.
© 2005 Elsevier B.V. All rights reserved.
0924-0136/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2005.03.015
P.A. Colegrove, H.R. Shercliff / Journal of Materials Processing Technology 169 (2005) 320–327 321
elasto-viscoplastic material properties and very good agree- Fig. 2. Description of the mesh used for the models.
ment was obtained between the model output and marker
experiments with the flow being largely confined to the region
tion can be neglected. Neither region includes the combined
adjacent to the pin. The temperatures from the model com-
elasto-viscoplastic response of the material.
pared favourably with those obtained experimentally. Com-
The FSW flow model considers a rotating tool with the
parisons between the weld power input and forces were not
material translating past it at the welding speed. The work-
reported.
piece material is subdivided into the five regions shown in
Other authors have analysed the 3-dimensional flow
Fig. 2. Firstly, a rotating mesh that moves at the rotation speed
around unthreaded pins using analytical [8], finite element
of the tool represents a fluid region adjacent to the tool. This is
[9–11], finite difference [12] and Arbitrary Lagrangian Eule-
embedded in a stationary mesh slightly wider than the rotating
rian (ALE) [13–15] techniques. The ALE approach used by
region, extending to the length of the plate, which models the
Schmidt and Hattel [14,15] shows particular promise and can
flow and thermal response of the material that passes through
predict void formation in the weld which is impossible with
the deformation zone. To either side a solid aluminium region
the CFD approach used in this work.
models the thermal response in the non-deforming region,
The aim of the present work is to demonstrate the benefits
which is known to be travelling at the welding speed. Finally,
and pitfalls of using CFD to model FSW. To do this a flow
the tool and backing plate are solid regions that are used to
model was developed that included:
calculate the heat loss to these components.
(a) coupled thermal/flow analysis;
(b) heat generation by viscous dissipation; 2.2. Solver type and boundary conditions
(c) full thermal boundary conditions, such as heat loss to the
backing plate and tool, and the convective heat loss on The model was solved as a steady-state problem. While
the top surface of the plate; this solution technique is valid for an axi-symmetric or cylin-
(d) the tool rake angle. drical tool, the solution will only be an approximation for
a threaded tool. In this case, the solution is a ‘snap-shot’
The CFD package FLUENT [16] was selected for the of the flow at a particular instant in time. The approxima-
modelling work because of its ability to handle the high strain- tion is reasonable provided the time-dependent terms in the
rates that occur near the tool surface. Navier–Stokes equation remain small. The benefit of this
approach over a full transient analysis is much faster solu-
tion times.
2. Description of the model The model assumed that the material sticks to the tool
surface, so the material velocity equalled the rotation speed
2.1. Model regions multiplied by the radius. The translation is prescribed by
allowing the material to flow past the tool. Therefore, the
The models were 3-dimensional and assumed that the material enters the model at a velocity equal to the welding
material stuck to the tool surface. Two different material defi- speed, and at ambient temperature.
nitions are available within the FLUENT solver. Firstly, there Both isothermal and non-isothermal models were solved.
is the fluid region where both the momentum and heat equa- The isothermal models used a representative temperature of
tions are solved. Secondly, there is the solid region where only 800 K, while the non-isothermal models included the heat
the heat equation is solved. This is used where the material equation and calculated the heat generation from viscous
being modelled is either a solid, or where material movement dissipation. In these models, the heat loss through the tool
is sufficiently small that solution of the momentum equa- was calculated by defining a 1-dimensional conduction zone
322 P.A. Colegrove, H.R. Shercliff / Journal of Materials Processing Technology 169 (2005) 320–327
Table 1
Thermal properties of the materials used in the thermal model
Material name Density (kg/m3 ) Specific heat (J/kg K) Thermal conductivity (W/m K)
5083/7075-T6 Aluminium 2660/2800b Millsa Millsa
Steel (0.5% C steel) at 150 ◦ C 7833c 465c 50c
H13 tool steel (5% chrome) at 300 ◦ C 7833c 460c 36c
a Ref. [22].
b Ref. [23].
c Ref. [24].
P.A. Colegrove, H.R. Shercliff / Journal of Materials Processing Technology 169 (2005) 320–327 323
Table 2 model’. The other models are described where their results
Summary of the welding conditions used for the models are significantly different.
Rotation speed Welding speed (mm/min)
(rpm) 3.1. Arrow plots and deformation region
60 90 120
750 5651-raked
500 5651-raked 5651-raked 5651-raked Fig. 4 shows vector plots of in-plane velocity on horizontal
5651-normal (+ a model planes at positions z = 0.1, 3.175 and 6 mm from the bottom
that uses 7075 aluminium face, for the standard model. Referring to Fig. 4a, the material
with isothermal under the dome of the tool (root) is influenced by the rotat-
properties)
Unthreaded tool
ing tool above. At mid-thickness (Fig. 4b) there is a large
250 5651-raked rotating region around the tool. Overlayed on this image is a
Both isothermal (800 K) and thermal models were solved using the properties
line where the strain-rate equals 2/s, which approximates the
of 5083 aluminium alloy. boundary between the rotating and stationary flow regions.
Finally, Fig. 4c shows the rotating material under the shoul-
der. Note that in all cases, there is a stagnation point on the
were created around the tool in accordance with Section 2.1, advancing side, and all the material up to this point flows
which were then meshed. round the retreating side of the tool.
Two models of the 5651 tool were generated: in one the Fig. 5 shows a longitudinal vertical section that compares
tool was normal to the workpiece (5651-normal) and in the the size of the deformation regions between an isothermal
other it was raked at 2.5◦ away from the direction of travel and a thermal model. Interestingly very little difference is
(5651-raked). Several alternative methods for raking the observed between the two, which contrasts with the dramatic
tool were attempted, however, the mesh shown in Fig. 2 reduction observed in a model for 7075-T6 that included a
which used a ‘prow’ around the shoulder was the only one thermal profile in Colegrove and Shercliff [27]. In the latter
that was successful. A final model (unthreaded tool) was an work, a large increase in material strength occurs outside
unthreaded version of the ‘5651-normal’ model described the deformation region, but this is far more gradual with the
above. Sellars and Tegart material properties for 5083 used in the
As seen in Fig. 2, the inner part of the tool protruded above current work.
the shoulder. This avoided the creation of highly distorted Streamlines of the flow around the tool are shown in Fig. 6.
elements around the fillet where the pin meets the shoulder. Note that streamlines are instantaneously tangential to the
Tetrahedral elements were used and a spacing of 0.3 mm was flow and are shown for a particular tool orientation. Parti-
used on the surface of the tool. cle tracks are much more difficult to calculate and require
In the ‘5651-raked’ model the plate thickness was a full transient analysis. The planar view (Fig. 6a) demon-
6.35 mm, while the other two models where the tool was strates how the material is swept around the retreating side
normal to the surface used a plate thickness of 6.95 mm. of the tool. The deformation zone is large compared with 2-
This is because the tool protrudes further into the plate when dimensional models [28], experimental results [29,30] and
it is normal to the surface. All the models used a back- later modelling work [27]. The deformation zone is larger
ing plate with a thickness of 40 mm and width of 140 mm. than the 2-dimensional models because of the additional
Finally, all the models extended 70 mm ahead of the weld and stirring induced by the shoulder. The later modelling work
100 mm behind. In this preliminary study, tests on the influ- uses a different constitutive model and includes slip at the
ence of mesh density and the effect of plate size were not tool/material interface. The large tetrahedral mesh used for
undertaken. the model may also expand the flow region.
Table 2 describes the models and welding parameters used. The longitudinal streamlines in Fig. 6b show very little
Note that unless otherwise stated, two versions of each were deviation in height as the material passes the tool. This is
solved: one was an isothermal (800 K) model and the other in contrast to the quite large vertical movement of mate-
was a thermal model. Both used the material properties of rial observed by Colligan [31]. The problem is caused
5083 aluminium alloy. The 7075 properties were only used by the model’s over-prediction of the deformation region
for one isothermal model, since the full thermal model could size.
not be solved due to the high heat generation and excessive
temperatures produced with these properties. 3.2. Pressure plot
Fig. 4. Arrow plots across the (a) z = 0.1, (b) z = 3.18 and (c) z = 6 planes for
the 5651-raked thermal model. Note: (b) includes an overlay of the strain-
rate = 2/s boundary; although advancing and retreating side labels are shown
only for (a) they apply equally to (b) and (c). Fig. 6. Streamlines for 5651-raked thermal model (a) planar from a start
height of 3 mm; (b) transverse longitudinal section.
P.A. Colegrove, H.R. Shercliff / Journal of Materials Processing Technology 169 (2005) 320–327 325
Acknowledgements
through the material. It is unclear whether the lower force is Appendix A. Determining the viscosity from the
real or an artefact of the model. normal stress and strain-rate
The unthreaded tool had a lower traversing force than
the threaded version. This result is consistent with the 2- The equation for the effective stress is given by Bathe [38]:
dimensional stick model analysis in Colegrove and Shercliff
[28], which showed that tools with features had a higher 3
traversing force than their cylindrical counterparts. Slip mod- σ̄ = S·S (A.1)
2
els of the same profiles indicated that the tools with features
assisted the movement of material round the tool, which low- where S is the deviatoric stress tensor. The equation for the
ered the traversing force. The latter result is more consistent effective strain-rate is given by Bathe [38]:
with experimental findings. Finally, the model that used 7075
2
properties had an even higher traversing force due to the ¯ε̇ = ε̇ · ε̇ (A.2)
3
higher flow stress of this material.
where ε̇¯ is the strain-rate tensor given by:
1
4. Conclusions ε̇ = ∇v + (∇v)T (A.3)
2
This work has demonstrated how the CFD package, FLU- where v is the velocity vector. The equation relating the devi-
ENT can be used to analyse material flow in FSW. The atoric stress tensor to the strain-rate is given in FLUENT [16].
models primarily used a threaded tool, which was raked 2.5◦ (Note that the bulk viscosity term is ignored.)
away from the direction of travel. This model was compared
against ones that used a normal threaded tool and a normal S = 2µε̇ (A.4)
unthreaded tool. Two model types were solved: an isothermal where µ is the viscosity. The deviatoric stress is defined as:
model and one that included the viscous heat generation and
solved the resulting thermal profile. S = + pI (A.5)
All the models predicted that the material in line with the
deformation zone was swept around the retreating side of the where is the stress and p is the pressure (positive in com-
pin. The amount of material swept around the pin increased pression). Note that p is defined as:
at locations closer to the shoulder. It was noted, however, that 1
the size of the deformation zone was much larger than that σm = σii = −p (A.6)
3
observed experimentally.
In addition to the excessive deformation zone size, the Substitute equation (A.4) into (A.1):
models experienced several additional problems. Firstly,
σ̄ = 6µ2 ε̇ · ε̇
using representative material properties for 5083 and 7075
(without viscosity softening near the solidus), there is a large
over-prediction of the weld temperature and the weld power. 2
σ̄ = µ 9 × ε̇ · ε̇
Secondly, the rotation speed had a large impact on the peak 3
welding temperature, which is relatively small in practice. Substituting from equation (A.2) gives:
Thirdly, the traversing force was found to be an order of
magnitude lower than that measured experimentally. Many σ̄ = 3µε̇¯
P.A. Colegrove, H.R. Shercliff / Journal of Materials Processing Technology 169 (2005) 320–327 327