Anda di halaman 1dari 14

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/248503511

Prediction of welding distortion and residual


stress in a thin plate butt-welded joint
Article in Computational Materials Science August 2008
Impact Factor: 2.13 DOI: 10.1016/j.commatsci.2007.12.006

CITATIONS

READS

95

537

2 authors, including:
Dean Deng
Chongqing University
137 PUBLICATIONS 1,449 CITATIONS
SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,


letting you access and read them immediately.

Available from: Dean Deng


Retrieved on: 04 June 2016

Available online at www.sciencedirect.com

Computational Materials Science 43 (2008) 353365


www.elsevier.com/locate/commatsci

Prediction of welding distortion and residual stress in a


thin plate butt-welded joint
Dean Deng a,*, Hidekazu Murakawa b
a

Research Center of Computational Mechanics Inc., Togoshi NI-BLDG 7-1, Togoshi, Shinagawa-ku, Tokyo 142-0041, Japan
b
Joining and Welding Research Institute, Osaka University, 11-1, Mihogaoka, Ibaraki, Osaka 567-0047, Japan
Received 1 October 2007; received in revised form 22 November 2007; accepted 4 December 2007
Available online 29 January 2008

Abstract
In automotive industry, thin plate parts are commonly used. During assembling process, welding technology is usually employed
because of high productivity. Welding distortion often occurs in thin plate welded structures due to relatively low stiness. The distortion
causes problems not only in the assembling process but also in the nal product quality. Therefore, prediction and reduction of welding
deformation have become of critical importance. In this study, three-dimensional, thermo-elasticplastic, large deformation nite element method (FEM) is used to simulate welding distortion in a low carbon steel butt-welded joint with 1 mm thickness. To compare
with the large deformation theory, the small deformation theory is also used to simulate the welding deformation and welding residuals
stress. Meanwhile, the characteristics of welding temperature eld, plastic strain distribution and welding residual stress in thin welded
plates are also examined numerically. Experiments are also carried out to measure the welding distortion in the thin plate butt-welded
joint. By comparing the simulation results with the measurements, it is found that the results predicted by the thermo-elasticplastic,
large deformation FEM match the experimental values well. Moreover, using the inherent strains obtained by the thermo-elasticplastic
FEM, an elastic FEM is also employed to estimate welding deformation in the same butt-welded joint. Comparing the results simulated
by the elastic FEM with those predicted by the thermo-elasticplastic FEM, it is veried that the inherent strain method can eectively
predict the welding deformation in the thin plate butt-welded joint with 1 mm thickness.
2007 Elsevier B.V. All rights reserved.
PACS: 07.05.Tp; 47.11.Fg; 65.40.De; 81.20.Vj
Keywords: Welding distortion; Numerical simulation; Thin plate; Plastic strain; Finite element; Inherent strain; Nonlinear analysis

1. Introduction
Distortion in a welded structure is the result of the nonuniform expansion and contraction of the weld and surrounding base material, caused by the heating and cooling
cycle during welding process. Welding distortion has negative eects on the accuracy of assembly, external appearance, and various strengths of the welded structures. In
many cases, additional costs and schedule delays are
incurred from straightening welding distortion. On the

Corresponding author. Tel.: +81 3 3785 3033; fax: +81 3 3785 6066.
E-mail address: deng@rccm.co.jp (D. Deng).

0927-0256/$ - see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2007.12.006

other hand, increasingly, the design of engineering components and structures relies on the achievement of small tolerance. For these reasons, prediction and control of
welding deformation have become of critical importance.
In the past decades, a lot of experiments and numerical
analyses have been conducted for predicting welding distortion, and a lot of fundamental knowledge has been also
established [17]. However, there is very limited literature
describing the prediction and measurement of welding
deformation in the thin plate welded structures especially
for these welded structure whose plate or wall thickness
is less than 3.0 mm. Recently, Liang et al. [810] have
established a number of databases of welding inherent
deformations for typical thin plate welded joint using

354

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

experimental method and inverse analysis method, and


some meaningful achievements have been obtained.
In automotive industry the thin plate parts are commonly employed, and a lot of thin plate parts are assembled
by arc welding process because of high productivity. Due to
relatively small stiness, signicant welding distortion often
occurs. To comprehensively understand the characteristics
of welding deformation in thin plate welded structure, it is
necessary to do further fundamental researches by means
of both experiment and numerical simulation.
In this study, three-dimensional, thermo-elasticplastic,
large deformation nite element method (FEM) is
employed to simulate welding distortion and welding residual stress in a low carbon steel butt-welded joint with 1 mm
thickness. To compare with the large deformation theory,
the small deformation theory is also used to simulate the
welding deformation and welding residuals stress. Meanwhile, the characteristics of welding temperature eld, plastic strain distribution and welding residual stress in thin
welded plates are also examined numerically. Experiments
are also carried out to verify the numerical simulation
method. Moreover, using the inherent strains (plastic
strains) computed by the thermo-elasticplastic FEM, an
elastic FEM, in which the large deformation is taken into
account, is also utilized to estimate the welding deformation in the same butt-welded joint.

Fig. 1. Thin plate butt-welded joint.

Weld bead

Specimens

h0

h1

Block

2. Experimental procedure
In this study, a simple experiment is carried out to measure welding deformation in the thin plate butt-welded
joint. The butt-welded joint consists of two thin mild steel
sheets. The dimension of each sheet is 100 mm 
100 mm  1 mm. The welding method is gas metal arc
welding (GMAW). The shielding gas was 80%Ar +
20%CO2. The welding wire is YGW16 [11]. The detailed
welding conditions are shown in Table 1.
To imitate the welding conditions used in automobile
industry, an initial gap between the two plates is set to be
0.4 mm and a partially welding is performed in the joint.
The length of welding line is about 60 mm. The specimens
are welded without external constraints. In the experiments, three identical butt joints are performed. Fig. 1
shows the picture of a completed joint.
After welding, the deection at the center of the welding
line is measured using a simple method as shown in Fig. 2.
In the experiments, a Vernier caliper is used to measure the
values of h0 and h1. In this gure, w is the dierence
between h0 and h1. The measured values (w) of these three
joints are 2.20 mm, 1.90 mm and 2.10 mm. Approximately,

Fig. 2. Schematic image of measuring the deection in the butt-welded


joint.

w/2 can be regard as the deection at the center of the welding line. The average value (w) of these three butt joints is
2.067 mm. This means the average value of the deection
(w/2) at the welding line is 1.03 mm.
3. Prediction of temperature eld, residual stress and
deformation using thermo-elasticplastic FEM
In this section, based on ABAQUS code [12] a sequentially
coupled thermo-elasticplastic nite element computational
procedure is developed to calculate temperature eld, welding
residual stress and welding deformations. The 3D nite element analyses are performed to numerically study the welding
distortion and welding residual stress in butt-welded thin
plates. In order to capture the nonlinear geometrical behaviors
in a thin plate structure, the large deformation theory is incorporated into thermo-elasticplastic FEM.
Fig. 3 shows the nite element mesh model used in the
simulation. The dimensions of the nite element model

Table 1
Welding conditions
Parameter

Current
(A)

Voltage
(V)

Welding speed
(mm/min)

Shielding gas ow
rate (L/min)

Tip-to-work
distance (mm)

Wire diameter
(mm)

Travel
angle ()

Value

65.0

17.0

780.0

15.0

25.0

0.9

45.0

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

Thermal Conductivity

1.4

355

Thermal Conductivity
Density
Specificheat

1.2
1

c (J/g/ C)

(102g/mm3)

0.8
0.6
0.4

k (101J/mm/s/ C)

0.2

300

600

900

1200

1500

Temperature ( oC)
Fig. 4. Temperature dependent thermal physical properties.
Fig. 3. Finite element model, gap and bead shape.

are the same as those of the experimental specimen. Element meshes are denser in the vicinity of the weld centerline, while the meshes become gradually coarser away
from the weld zone. The length of each element in the welding direction is 2.5 mm. The number of division in the
thickness direction is four. The total number of 8-node
brick element is 4000.
In the present study, to consider the bead shape in the
FE model, three dimensions of the weld bead obtained in
the experiments are measured. One is the height of weld
reinforcement; and the other two are the breadths between
the two toes on the upper surface and the bottom surface,
respectively. In the FE model, the bead shape is roughly
determined based on these three parameters. The bead
shape is shown in Fig. 3.
In the thermal analysis, the welding conditions are
assumed to be the same as those used in the experiment.
3.1. Heat source and thermal analysis
Welding heat transfer analysis with given welding conditions is performed in the 3D thin plate model. In this step,
temperature histories at each element nodes are computed
during the welding process. 3D, 8-nodes, linear brick and
heat elements (DC3D8) [12] are selected for the thermal
analysis. Temperature dependent physical properties of
the mild carbon steel as shown in Fig. 4 [13] are employed
in heat transfer analysis. In this study, solid-state phase
transformation is neglected because the inuence of phase
transformation on the welding deformation and welding
residual stress is insignicant in the lower carbon steel [14].
During the welding, the governing equation for transient
heat transfer analysis is given by:
qc

oT
x; y; z; t r  qx; y; z; t Qx; y; z; t
ot

where q is the density of the materials [g/mm3], c is the specic heat capacity [J/(g C)], T is the current temperature
[C], q is the heat ux vector [W/mm2], Q is the internal
heat generation rate [W/mm3], x, y and z are the coordinates in the reference system [mm], t is the time [s], and
$ is the spatial gradient operator.
The non-linear isotropic Fourier heat ux constitutive
equation is employed:
q krT

where k is the temperature-dependent thermal conductivity


[J/(mm s C)].
In this study, the heat from the moving welding arc is
applied as a volumetric heat source with a double ellipsoidal distribution proposed by Goldak [15], which is
expressed by the following equations:
For the front heat source:
p
6 3ff Qw 3x02 =a2 3y 02 =b2 3z02 =c2
0 0 0
1e
p e
Qx ; y ; z ; t
e
3
a1 bcp p
For the rear heat source:
p
6 3fr Qw 3x02 =a2 3y02 =b2 3z02 =c2
0 0 0
2e
p e
e
Qx ; y ; z ; t
a2 bcp p

where x0 , y0 and z0 are the local coordinates of the double


ellipsoid model aligned with the welded pipe; ff and fr are
parameters which give the fraction of the heat deposited
in the front and the rear parts, respectively. Because the
temperature gradient in the front leading part is steeper
than in the tailing edge, ff and fr are assumed to be 1.33
and 0.67, respectively. Qw is the power of the welding heat
source. It can be calculated according to the welding current, the arc voltage and the arc eciency. The arc eciency g, is assumed to be 80% for the GMAW welding
process. The parameters a1, a2, b and c are related to the
characteristics of the welding heat source. The parameters

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

Table 2
Parameters of the heat source
Parameter

Value (mm)

a1
a2
b
c

2.0
4.0
1.2
1.0

3.2. Mechanical analysis

of the heat source can be adjusted to create a desired


melted zone according to the welding conditions. The values of these parameters used in the present simulation are
summarized in Table 2.
When structural analysis FEM codes such as ABAQUS
and MARC are used to simulate the temperature distribution in welding process, the uid ow and solidication of
material in the weld pool cannot be directly considered
because the coupled problem between solid and liquid is
not involved in these software at present. However, the
eect of the uid ow has signicant eects on the temperature distribution and the shape of weld pool. If the eect
of the uid ow is neglected, the highest temperature in
weld pool will be very high. According to the authors
experience, the peak temperature in welding pool is higher
than 3000 C in some cases when the uid ow eect is
neglected. This phenomenon is much dierent from the
realistic situation. Okagaito et al. [16] measured the surface temperature distribution on TIG weld pool in
SUS304 steel. Their research suggests that the highest temperature on the molten pool surface is approximately
1750 C. In this study, to consider the uid ow an articially increased thermal conductivity in the weld pool is
used. The thermal conductivity is assumed to be twice as
large as the value of room temperature for temperature
above the melting point.
The thermal eects due to solidication of the weld pool
are modeled by taking into account the latent heat for
fusion. The value of the latent heat is 270 J/g [17]. The liquidus temperature TL and the solidus temperature TS are
assumed to be 1500 C and 1450 C, respectively.
Heat losses (qc) due to convection are considered for all
the surfaces using Newtons law:
qc hf T sur  T 0

5
2

where hf is lm coecient for convection [W/(mm C)],


Tsur is the surface temperature [C], and T0 is ambient temperature [C].
In this study, a temperature-dependent lm coecient
[18] is used, and the ambient temperature is assumed to
be 20 C.
Radiation heat losses (qr) are accounted for all the surfaces by using StefanBoltzman law:
qr erT 4sur  T 40

The user-dened subroutines to ABAQUS code are utilized in the heat transfer analysis to model heat uxes, convection and radiation boundary conditions.

where e is emissivity, r is StefanBoltzman constant for


radiation.
In this study, the emissivity is assumed to be 0.2 [19,20].

The same nite element models used in the thermal analyses are employed in mechanical analyses, except for the
element type and boundary conditions. The restraint conditions are shown in Fig. 3 by the arrows. The C3D8I element type [12] is used to simulate the stressstrain eld.
The analyses are conducted using the temperature history
calculated by the thermal analyses as the input
information.
For the mild steel, because phase transformation has an
insignicant eect on the welding residual stress and the
deformation, the total strain can therefore be decomposed
into three components as follows:
etotal ee ep eth

The components on the right-hand side of Eq. (7) correspond to elastic, plastic and thermal strain, respectively.
The elastic strain is modeled using the isotropic Hookes
law with temperature-dependent Youngs modulus and
Poissons ratio. For the plastic strain component, a plastic
model is employed with the following features: the Von
Mises yield surface and temperature-dependent material
properties. Fig. 5 shows the temperature dependent
mechanical properties [13]. Because the eect of work hardening is not signicant in mild steel, it is neglected in this
study.
In this study, the thickness of the specimen is only 1 mm,
so it can be expected that the geometrically nonlinear phenomenon probably occur during welding. To examine the
dierence between the numerical results computed by the
350
Yield Strength (MPa)
Young's Modulus (GMPa)
-7 o

300

Thermal expansion coefficient (10 / C)


-2

Possion's ratio (10 )

Mechanical Properties

356

250
200
150
100
50
0

300

600

900

1200
o

Temperature ( C)
Fig. 5. Temperature dependent mechanical properties.

1500

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

3.3. Simulation results


3.3.1. Characteristics of welding temperature eld
Fig. 6 shows the temperature histories at the center of
the welding line. From this gure, it can be seen that the
peak temperature at the top surface of the weld pool is
about 1800 C. In the same gure, it can also be observed
that the dierence between the peak temperature at the
top surface and that at the bottom surface is not signicant. The thermal eects due to solidication of the weld
pool are considered in the FE model, so these two cooling
curves reect the solidication phenomena of the weld
pool. Except for the peak temperature, the temperature history curves of the top surface and the bottom surface in the
fusion zone have no dierence. Fig. 7 shows the temperature histories of the top surface and the bottom surface
in the heat-aected zone (HAZ) of the mid-section. The
distance between this position and the weld centerline is
3.2 mm. This gure indicates that both the top surface
and the bottom surface in the HAZ have almost the same
peak temperature and the identical cooling rate. In the FE
model, the plate thickness of the fusion zone is less than
2 mm, and the plate thickness is only 1 mm in other parts.
It is the small plate thickness that resulted in an even temperature distribution through thickness during welding.
3.3.2. Welding residual stress
In the mechanics analysis, both the large deformation
theory and the small deformation theory are used to simu-

1000
Bottom

30

40

600

400

200

0
0

10

20

50

Time (s)
Fig. 7. Temperature histories in the HAZ.

late welding residual stress and deformation in the buttwelded joint. Fig. 8 shows the longitudinal residual stress
distributions of the middle section, which are computed
by the large deformation theory (case A). Fig. 9 shows
the longitudinal stress distributions of the middle section,
which predicted by the small deformation theory (case
B). Due to a relatively large out-of-deformation are generated after welding in case A, there is a dierence between
the longitudinal stress of the top surface and that of the
bottom surface. On the contrary, there is almost no dierence between the longitudinal stress of the top surface and
that of the bottom surface in case B.
Fig. 10 shows the transverse residual stress distributions
of the middle section predicted by case A. There is also a
signicant dierence between the transverse residual stress
of the top surface and that of the bottom surface. It is very

500

2000

Longitudinal Stress (MPa)

Top Surface
Bottom Surface

Temperature (oC)

Top

800

Temperature ( oC)

large deformation theory and that calculated by the small


deformation theory, both the two methods are used to predict welding distortion and welding residual stress. Thus, in
the mechanical analysis stage, two simulation cases (case A
and case B) are performed. The large deformation theory is
considered in case A, while the small deformation theory is
used in case B.

357

1500

1000

500

Top Surface
Bottom Surface

400
300
200
100
0
-100
-200

10

15

20

Time (s)
Fig. 6. Temperature histories in the fusion zone.

25

-300

50

100

150

200

Y-Coordinate (mm)
Fig. 8. Longitudinal residual stress in mid-section computed by Case A.

358

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

400
Top Surface
Bottom Surface

400

Transverse Stress (MPa)

Longitudinal Stress (MPa)

500

300
200
100
0
-100

Top Surface
Bottom Surface

300

200

100

-200
-100
-300

50

100

150

200

Y-Coordinate (mm)

50

100

150

200

Y-Coordinate (mm)
Fig. 11. Transverse residual stress in mid-section simulated by Case B.

Fig. 9. Longitudinal residual stress in mid-section simulated by Case B.

400

Transverse Stress (MPa)

Top Surface
Bottom Surface

300

200

100

-100

50

100

150

200

Y-Coordinate (mm)
Fig. 10. Transverse residual stress in mid-section computed by case A.

clear that in the fusion zone and the HAZ the transverse
residual stress of the top surface is much larger than that
of the bottom surface. The dierence results from the transverse bending deformation. Fig. 11 shows the transverse
residual stress distributions of the middle section computed
bye case B. This gure indicates that there is almost no difference between the top surface and the bottom surface.
3.3.3. Welding deformation
Fig. 12 shows the contours of the deection distribution
computed by the large deformation theory. From this gure, it can be observed that a large longitudinal bending
and a transverse bending are produced after welding. At
the two ends of the gaps, the maximum deection is about
1.7 mm. Fig. 13 shows the contours of the deection distribution predicted by the small deformation theory (case B).

This gure also reects that both longitudinal bending and


transverse bending are generated in the butt-welded joint.
The deformation mode is similar to case A, however the
magnitudes of deection are much smaller than case A.
Fig. 14 shows the deection distributions of case A and
case B along the middle line (line AB), which is dened
in Fig. 12. In this gure, point A shown in Fig. 12 is
assumed to be the origin. The experimental measurement
(deection) at the center of the welding line is also plotted
in the same gure. It is clear that the deection at the center
of the welding line predicted by case A is much close to the
experimental value. From the same gure, it can be also
known that the experimental value is signicantly larger
than the numerical result computed by case B. This information suggests that when welding deformation of a thin
plate welded joint or structure is simulated numerically it
is necessary to consider geometrically nonlinear phenomenon. Otherwise, it is probable to result in a very large error.
Fig. 15 shows the Y-directional displacement distribution of the middle section predicted by the large deformation theory. Comparing the transverse shrinkage of the
top surface with that of the bottom surface, it can be
observed that the former is smaller than the latter. However, the dierence is very small. Fig. 16 shows the Y-directional displacement distribution of the middle section
computed by the small deformation theory. This gure
indicates that the transverse shrinkage through thickness
is almost uniform. Comparing Fig. 15 with Fig. 16, it can
be conclude that the Y-directional displacements predicted
by case A are close to those computed by case B on the
whole.
3.3.4. Plastic strain distribution
Fig. 17 shows the plastic strain distributions in the longitudinal direction (welding direction) of the middle section. The plastic strain values are the average ones
through thickness predicted by the large deformation the-

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

359

Fig. 12. Deection distribution of Case A.

Fig. 13. Deection distribution of Case B.

ory and the small deformation theory. From this gure, it


can be observed that the plastic strain distributions predicted by the two cases have no signicant dierence. Both
the range of the longitudinal plastic strain values and the
distribution shapes of these two cases are fairly close.
Fig. 18 shows the plastic strain distributions in the transverse direction of the middle section computed by Case A
and Case B. This gure indicates that the transverse plastic
strain distributions of the two cases are quite similar.
From Figs. 17 and 18, it is known that the range of the
longitudinal plastic strain distribution is larger than that of

the transverse plastic strain distribution. The transverse


plastic strain range concentrates almost only in the fusion
zone and the HAZ, while the longitudinal plastic strain
range distributes in a relatively large range. Generally,
the range and magnitude of the plastic strain component
are mainly governed by the peak temperature and the
restraint conditions [21]. During welding, because the
restraint intensity in the longitudinal direction (welding
direction) is larger than that in the transverse direction,
the range of the longitudinal plastic strain is relatively
large.

360

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

1.5

0.1
Top Surface
Bottom Surface

Y-Displacement (mm)

Deflection (mm)

Large deformation theory


Small deformation theory
Experiment

0.5

-0.1

-0.2

Small deformation theory

-0.3
0

0
0

50

100

150

50

200

100

150

200

Y-Coordinate (mm)

Y-Coordinate (mm)
Fig. 16. Y-directional displacement distributions of the middle section
(Case B).

Fig. 14. Deection distributions along line AB.

0.1

Longitudinal Plastic Strain

Y-Displacement (mm)

Top Surface
Bottom Surface

-0.1

-0.2

-0.001

-0.002

-0.003
Large deformation theory
Small deformation theory

Large deformation theory

-0.004

-0.3
0

50

100

150

200

Y-Coordinate (mm)

60

80

100

120

140

Y-Coordinate (mm)
Fig. 17. Longitudinal plastic strain distribution of the middle section.

Fig. 15. Y-directional displacement distributions of the middle section


(Case A).

3.3.5. Discussions
Based on the simulation results, it is known that both
the longitudinal plastic strain and the transverse plastic
strain distribute narrowly in the fusion zone and its vicinity. For mild steel, the plastic strains are mainly governed
by the thermo-mechanical behavior of the weld metal and
the base metal near the fusion zone during welding, so
the magnitudes and distributions predicted by the large
deformation theory and the small deformation theory are
fairly similar. On the contrary, the nal deformation of
the thin plate butt-welded joint shows a signicantly global
characteristic. It is very clear that even though the large
deformation theory and the small deformation theory predict similar plastic strains in the butt-welded joint, however, the nal welding deformations especially the out-of-

deformation (deection) are signicant dierent. The reason is that beyond the plastic strain zone the welding deformation is mainly governed by the elastic strain and the
straindisplacement relationship. By comparing with the
experiment, it is clear that the deection at the center of
the welding line predicted by the large deformation theory
is very close to the experimental measurement. The result
simulated by the small deformation theory is much smaller
than the measurement. This indicates that when the
thermo-elasticplastic FEM is used to predict the welding
deformation in a thin plate structure the geometrically nonlinear phenomenon should be carefully considered. It is
very interesting to note that even though the magnitudes
of the welding deformation predicted by the two theories
are much dierent, however, the deformation modes are
similar. Both Figs. 12 and 13 show that the deection dis-

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

Transverse Plastic Strain

0.05
Large deformation theory
Small deformation theory

0.025
0
-0.025
-0.05
-0.075
-0.1
80

90

100

110

120

Y-Coordinate (mm)
Fig. 18. Longitudinal plastic strain distribution of the middle section.

tribution of the butt-welded joint has a saddle-shaped


mode, which is generally opposite to that of a full welding
butt-welded joint [13]. Because a partially welding is performed in the butt joint, the gaps remained at the two ends
caused the saddle-shaped mode after welding.
Because relatively large longitudinal bending and transverse bending is predicted by the large deformation theory,
both the longitudinal stress and the transverse stress of case
A have signicant gradient distributions through thickness.
On the contrary, a small out-of-deformation is simulated
by the small deformation theory, so either the longitudinal
stress or the transverse stress of case B has almost no gradient through thickness in the butt-welded joint.
4. Prediction of welding deformation using inherent strain
method
4.1. Inherent strain
Although the thermo-elasticplastic FEM can be used to
simulate welding temperature eld, welding residual stress
and welding deformation, a very long computational time
is needed because the welding mechanical behavior is
highly nonlinear problem including material nonlinearity,
geometrical nonlinearity and sometimes contact nonlinearity. Besides the thermo-elasticplastic FEM, elastic FEM
based on inherent strain theory [2126] can also be utilized
to predict welding deformation. Comparing with the
thermo-elasticplastic FEM, only a very short computational time is needed to complete the simulation even for
a large and complex structure. Moreover, only the elastic
modulus and the Possions ratio at room temperature are
used in the elastic FEM, and the temperature dependent
material properties are not needed. In this study, the inherent strain method is employed to simulate the welding
deformation in the thin plate butt-welded joint.

361

Based on experimental observations and theoretical


analysis, it is found that the total welding distortion of a
weld joint is mainly produced by four components, namely
longitudinal shrinkage (dx), transverse shrinkage (dy), longitudinal bending (hx) and transverse bending (angular distortion hy). The four fundamental deformation
components are also called inherent deformations [10].
According to the numerical results obtained by the
thermo-elasticplastic FEM, the four inherent deformation
components in a cross-section of the butt-welded joint can
be calculated using the following equations [811]:
Z
1
ep dy dz
dx
8
hZ x
1
dy
9
ep dy dz
h Z y
12
epx z  h=2 dy dz
10
hx 3
h Z
12
epy z  h=2 dy dz
11
hy 3
h
where epx is the plastic strain in the welding direction (longitudinal direction); epy is the plastic strain in the transverse
direction; h is the thickness of the plate, and z is the coordinate in the thickness direction.
The source induced welding deformation and welding
residual stress is called inherent strain [22]. When the inherent deformations of a weld joint are known, they can be
transferred into the corresponding inherent strain components [26]. When the inherent strains are introduced into
an elastic FEM, it is usually assumed that each component
has a uniform distribution along the welding line. Therefore,
the average values of a weld joint are often used to approximately represent the inherent strains for the whole joint.
The average values of the four inherent deformations in
the butt-welded joint can be calculated using the following
formulae:
Z L
dx 1
dx dx
12
Lw 0
Z L
dy 1
dy dx
13
Lw 0
Z L
hx 1
hx dx
14
Lw 0
Z L
hy 1
hy dx
15
Lw 0
where Lw is the length of the welding line, L is the length of
the specimen.
4.2. Elastic nite element based on inherent strain
In the elastic FEM, the 4-node plate is used. For thin
plate deformation, deection w (x, y) is assumed to be
equal to the deection of mid-plane, w0 (x, y). When large
deformation (geometrically nonlinear phenomenon) is con-

362

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

sidered, the straindisplacement relations can be dened as


follows:
"


2 # 
ou0 1 ow0
o2 w0
i
b

ex ex ex
z 2
16
ox 2 ox
ox
"


2 # 
ov0 1 ow0
o2 w 0
i
b

17
z 2
ey ey ey
oy 2 oy
oy

 

ou0 ov0 ow0 ow0
o2 w 0

cxy cixy cbxy


2z
18
oy
ox
ox oy
oxoy
where u0, v0 and w0 are the displacement at mid-plane, ex, ey
and cxy are total strains, and eix ; eiy and cixy are in-plane
strains, and ebx ; eby and cbxy bending strains.
The curvature (jx) in a plane parallel to the xz plane
and the curvature (jy) in a plane parallel to the yz plane
and the twisting curvature (jxy), which represents the
warping of the xy plane, can be dened as follows:
o2 w0
ox2

19

o2 w0
jy  2
oy

20

o2 w0
jxy 
oxoy

21

jx 

{ }

Input *

{ f } = [B] [D]{ * }dv


T

{u} = [K ]1{ f }

{ } = [B]{u}

{ }= { } { }
e

{}

[D] { e}

Fig. 20. Analysis procedure of inherent strain method.

Fig. 19 shows schematically a butt-welded joint. In this


joint, the average longitudinal shrinkage (dx ) can be transformed into the in-plane strain component (ex ) in longitudinal direction; the average transverse shrinkage (dy ) can
be changed into the in-plane strain component ey in
transverse direction; the average angular distortion hy
can be converted into the curvature jy along the y-axis;
and the average longitudinal bending hx can be transformed into the curvature jx along the x-axis. These
inherent strain components are introduced into the elastic
FEM as initial strains, and the total welding distortion
can be estimated through elastic nite element analysis procedure [26] as shown in Fig. 20.
In Fig. 19, {e*}, {e}, {ee}, {f}, {u} and {r} are vectors of
inherent strain, total strain, elastic strain, equivalent nodal
load, nodal displacement, and residual stress, and [B], [D],
and [K] are straindisplacement matrix, elastic stressstrain
matrix and stiness matrix, respectively.

4.3. Elastic nite element model


The elastic nite element model, the area introduced
inherent strains, and the initial gaps are shown in Fig. 21.
According to the simulation results computed by the
thermo elasticplastic FEM the range of the longitudinal

Welding line

x
Fig. 19. Butt-welded joint.

Fig. 21. The FE model, the area introduced inherent strains and the initial
gap.

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

plastic strain is large than that of the transverse plastic


strain. In the elastic analysis, it is assumed that all the four
inherent strain components are introduced into the same
area. The four average inherent strain components are
determined by the following equations:
22


hx

h
y

24

23


hx =Bis
y =Bis
h

25

where Bis is the breadth of the elements in which the inherent strains are introduced.
4.4. Simulation results and discussions
Fig. 22 shows the contour of nal deection distribution. By comparing with Fig. 11, it can be concluded that
the deection distribution predicted by the elastic FEM
matches that computed by the thermo-elasticplastic
FEM well. Fig. 23 shows the deection distribution along
the line AB as dened in Fig. 20. The corresponding deection distribution predicted by the thermo-elasticplastic is
also plotted in the same gure. It is clear that the dierence
between the two curves is very small.
Fig. 24 shows the Y-directional displacement distributions along line AB predicted by the elastic FEM and the
thermo-elasticplastic FEM. The average values of the
top surface and the bottom surface simulated by the
thermo-elasticplastic FEM are plotted in this gure. This
gure means that the two distributed curves are quite similar. Through carefully observation, it can be found that

1.5
Elastic FEM
Thermo-elastic-plastic FEM

1.2

Deflection (mm)

ex 
dx =Bis

y =Bis
e d

363

0.9
0.6
0.3
0
-0.3

50

100

150

200

Y-Coordinate (mm)
Fig. 23. Deection distributions in the middle section.

the transverse shrinkage predicted by the thermo-elastic


plastic FEM is slightly larger than that estimated by the
elastic FEM. In the elastic FE model, the average transverse inherent strain of the whole joint is used. In the
thermo-elasticplastic FE model, the transverse shrinkages
have non-uniform distribution along the welding line and
the maximum value is at the compared location (line
AB). These two factors are main reasons which resulted
in a slightly dierence between the two curves shown in
Fig. 24.
When the elastic FEM is used to predict welding deformation, the attention is mainly paid to the total deforma-

Fig. 22. Deection distribution predicted by elastic FEM.

364

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365

Y-Directional Displacement (mm)

0.1
Elastic FEM
Thermo-elastic-plastic FEM

-0.1

-0.2

-0.3
0

50

100

150

200

Y-Coordinate (mm)
Fig. 24. Y-directional displacement distributions in the middle section.

tion rather than the local deformation. From this viewpoint, it can be concluded that the transverse shrinkage
predicted by the large deformation elastic FEM matches
that computed by the thermo-elasticplastic FEM well on
the whole.
From Figs. 12, 22, 23 and 24, it can be concluded that
the elastic FE model has accurately reproduced the welding
deformation, which is simulated by the thermo-elastic
plastic FE model. Based on the present study, it can be
inferred that when the inherent strains of each joint
involved in a thin plate structure are known, the proposed
elastic FEM can eectively predict welding deformation.
Comparing the computational time, it is found that the
computational time to complete the elastic FE analysis is
far shorter than that used by the thermo-elasticplastic
FE analysis. In the present study, the total computational
time of both the thermal analysis and the thermo-mechanical analysis is approximately 12 hours for the thermo-elasticplastic FE model, whereas the computational time of
the elastic FE model is shorter than 1 min. On the aspect
of computational time, the elastic FEM has a signicant
advantage over the thermo-elasticplastic FEM. Thus, this
method is a promising alternative to predict welding deformation for practical large thin-plate welded structures.

5. Conclusions
In this study, the thermo-elasticplastic FEM is used to
simulate the welding temperature eld, residual stress and
distortion in a thin plate butt joint. Meanwhile, experiments are carried out to measure the welding deformation.
Moreover, the elastic FEM based on inherent strain theory
is also utilized to simulate welding deformation in the same
butt joint. According to the numerical and experimental
results, the following conclusions can be drawn.

(1) Based on the simulation results, it is known that there


almost is no temperature gradient through thickness
in the thin plate butt joint during welding.
(2) Although the large deformation theory and the small
deformation theory predict the similar plastic strains,
however, the nal welding deformations especially
the out-of-deformation computed by the two methods are much dierent. By comparing with experiment, it is clear that the numerical result calculated
by the thermo-elasticplastic, large deformation
FEM is in a good agreement with the measurement.
Therefore, to precisely predict welding deformation
in a thin plate structure, it is necessary to consider
the geometrically nonlinear problem.
(3) Because of relatively large longitudinal bending and
transverse bending deformations, both the longitudinal and the transverse residual stresses predicted by
the thermo-elasticplastic, large deformation FEM
have a gradient through thickness.
(4) The elastic FEM with considering large deformation
can be used to predict precisely welding deformation
in the thin plate butt joint. Moreover, the computational time is much shorter than that used in the
thermo-elasticplastic FEM. For the automotive
industry application, the elastic FEM based on the
inherent strain theory is a promising method to predict welding deformation.

References
[1] G. Verhaeghe, Predictive Formulate For Weld Distortion A Critical
Review, Abingto Publishing, 2000.
[2] D. Radaj, Welding Residual Stress and Distortion Calculation and
Measurement, Woodhead Publishing Ltd., DVS Verlag, 2003.
[3] D. Deng, H. Murakawa, Y. Ueda, International Journal of Oshore
and Polar Engineering 14 (2) (2004) 138144.
[4] G. H Jung, C.L. Tsai, Welding Journal 83 (6) (2004)
177s187s.
[5] ZhiLi Feng, Processes and Mechanisms of Welding Residual Stress
and Distortion, Woodhead Publishing Ltd., Cambridge, UK, 2005.
[6] L.-E. Lindgren, Computer Methods in Applied Mechanics and
Engineering 195 (4849) (2006) 67106736.
[7] D. Deng, W. Liang, H. Murakawa, Journal of Materials Processing
Technology 183 (2-3) (2007) 219225.
[8] W. Liang, S. Sone, M. Tejima, H. Serizawa, H. Murakawa,
Transactions of JWRI 33 (2004) 4551.
[9] W. Liang, D. Deng, H. Murakawa, Transactions of JWRI 34 (1)
(2005) 113123.
[10] W. Liang, D. Deng, S. Sone, H. Murakawa, Welding in the World 49
(11/12) (2005) 3039.
[11] R. Suzuki, T. Nakano, Kobe Steel Engineering Reports 52 (3) (2002)
7478.
[12] ABAQUS/Standard, vols. 1, 2 and 3, Version 6.4, Hibbitt, Karlsson
& Sorensen Inc., 2003.
[13] D. Deng, Theoretical prediction of welding distortion in thin curved
structure during assembly considering gap and misalignment, Doctoral Thesis, Osaka University, 2002.
[14] D. Deng, Y. Luo, H. Serizawa, M. Shibahara, H. Murakawa,
Transactions of JWRI 32 (2) (2003) 325333.
[15] J. Goldak, A. Chakravarti, M. Bibby, Metallurgical Transactions B
15 (1984) 299305.

D. Deng, H. Murakawa / Computational Materials Science 43 (2008) 353365


[16] T. Okagaito, T. Ohji, F. Miyasaka, Quarterly Journal of Japan
Welding Society 22 (1) (2004) 2126.
[17] W. Zhang, J.W. Elmer, T. DebRoy, Materials Science and Engineering A 333 (2002) 320325.
[18] P. Michaleris, A. DeBiccar, Welding Journal 76 (4) (1997) 72s180s.
[19] S.B. Brown, H. Song, Journal of Engineering for Industry 114 (1992)
441451.
[20] B. Taljat, B. Radhakrishnan, T. Zacharia, Material Science and
Engineering A 246 (1998) 4554.
[21] H. Murakawa, Y. Luo, Y. Ueda, Journal of the Society of Naval
Architects of Japan 180 (1996) 739751.

365

[22] Y. Ueda, M. G Yuan, Journal of Engineering Materials and


Technology 115 (10) (1993) 417423.
[23] M.G. Yuan, Y. Ueda, Journal of Engineering Materials and
Technology 118 (4) (1996) 229234.
[24] Y. Luo, H. Murakawa, Y. Ueda, Journal of the Society of Naval
Architects of Japan 182 (1997) 783793.
[25] Y. Luo, H. Murakawa, Y. Ueda, Journal of the Society of Naval
Architects of Japan 183 (1998) 323333.
[26] D. Deng, H. Murakawa, W. Liang, Computer Methods in Applied
Mechanics and Engineering 196 (4548) (2007) 4613
4627.

Anda mungkin juga menyukai