Materials and Design 53 (2014) 10521063

Contents lists available at ScienceDirect

Materials and Design

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

Numerical analysis and experimental investigation of welding residual

stresses and distortions in a T-joint llet weld
Mato Peric b,2, Zdenko Tonkovic a,,1, Alan Rodic a,1, Martin Surjak a,1, Ivica Garaic a,1, Ivanka Boras a,1,
Srecko vaic a,1
a
Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, I. Lucica 5, 10000 Zagreb, Croatia
b
Bestprojekt, Bureau of Energetics and Mechanical Engineering Ltd., Petrovaradinska 7, 10000 Zagreb, Croatia

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

Article history: This paper presents a numerical and experimental study of residual stresses and distortions induced by
Received 8 March 2013 the T-joint welding of two plates. Within the framework of numerical investigations, a thermo-mechan-
Accepted 4 August 2013 ical nite element analysis is performed by using a shell/three-dimensional modeling technique to
Available online 14 August 2013
improve both the computational efciency and the accuracy. The inuence of the choice of the local
3D model size on the temperature, residual stress, and displacement distributions is investigated. A min-
Keywords: imal 3D zone size that had both appropriate convergence of the solution and accuracy is dened. To val-
Welding residual stress
idate the numerical model, a series of experiments using a fully automated welding process are
Welding distortion
Temperature eld
conducted. A thermographic camera and an optical measurement system are used to measure the tem-
T-joint llet weld perature and displacement distributions.
Finite element analysis 2013 Elsevier Ltd. All rights reserved.

1. Introduction rate of heat generation due to mechanical dissipation energy can

It is well known that during the welding process, due to localized
heating and subsequent rapid cooling, residual stresses appear
around welding zones and cause post-weld deformations of the struc-
ture. Numerical simulations have become an important tool for pre-
dicting these phenomena, but the structures that can be simulated
and the time period over which they can be studied will always be
limited by computational power. Reduction of the computational
time and complexity of transient nonlinear problems such as the
welding process is one of the active research areas in engineering
elds. Numerous studies have been published in this area of study.
In principle, a nite element (FE) simulation of the welding pro-
cess consists of two main parts: thermal analysis and mechanical
stress analysis [15]. In thermal analysis, the temperature eld is
determined as a function of time for each integration point. This
temperature time-history is used as an input into the thermal stress
analysis. Herein, the thermal solution can be sequentially or fully
coupled with the mechanical solution of the structure. Because the
Corresponding author. Address: Institute of Applied Mechanics, Faculty of
Mechanical Engineering and Naval Architecture, University of Zagreb Ivana Lucica 5,
10000 Zagreb, Croatia. Tel.: +385 1 61 68 450/222; fax: +385 1 61 68 187.
E-mail addresses: mato.peric@fsb.hr (M. Peric), zdenko.tonkovic@fsb.hr (Z.
Tonkovic), alan.rodic@fsb.hr (A. Rodic), martin.surjak@fsb.hr (M. Surjak), ivica.gar-
asic@fsb.hr (I. Garaic), ivanka.boras@fsb.hr (I. Boras), srecko.svaic@fsb.hr (S. vaic).
URL: http://www.fsb.hr/lnm/ (Z. Tonkovic).
1
Tel.: +385 1 61 68 222.
2
Tel.: +385 1 38 45 147.
be neglected in the heat transfer analysis, a sequentially coupled
thermal-stress analysis is commonly applied for the simulation of
a welding process in which a thermal analysis is followed by a stress
analysis [69]. As presented in the literature, the use of three-
dimensional (3D) models is required for accurate prediction of
post-weld deformation and residual stress distribution. Such a com-
plex model can be computationally expensive or even oversized for
available resources [1016]. In spite of the development of faster
computers, the time duration of 3D numerical simulations of weld-
ing processes remains a substantial challenge. This challenge is cre-
ated by the industrial need to solve more and more demanding
problems that appear in engineering practice. To reduce the time
needed to numerically analyze a welding process, many solutions
are suggested. In the case of 3D numerical analysis, it is always rec-
ommended to use conditions of symmetry [3,5,7]. Whenever possi-
ble, 3D models are replaced by 2D models [6,7]. Furthermore,
complicated denitions of the heat ux are replaced by a uniform
heat ux [17], or thermal boundary conditions are simplied (i.e.
convection coefcient and radiation coefcient are taken to be con-
stant [7]). There are also alternative solutions such as the inherent
strain method [9]. For different kinds of mechanical structures, a
simultaneous discretization using a shell and 3D nite elements
can be successful in reducing the numerical analysis time [10]. This
has created the idea of applying a global/local modeling approach of
numerically simulating the welding processes in this study.
The goal of this study is to reduce the simulation time while
maintaining accuracy by using a sequential method combining

0261-3069/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.matdes.2013.08.011
 M. Peric et al. / Materials and Design 53 (2014) 10521063
3D solid and shell elements. Three-dimensional continuum ele-
ments are used close to the weld where stress gradients are high;
outside the welding zone, the structure is discretized with shell
elements to reduce the overall model size. Thus, this shell/3D mod-
eling technique (shell-to-solid coupling option in Abaqus [16])
combines the accuracy of the full three-dimensional solution with
the computational efciency of a shell nite element model
[11,12]. In contrast to the shell-to-solid submodeling [10], which
rstly performs a global analysis on a shell model followed by a
submodel analysis with a continuum model, the shell-to-solid cou-
pling model uses only a single analysis with 3D solids and shell ele-
ments used in different regions. For verication of the shell-to-
solid coupling model, an example of a T-joint llet welding is cho-
sen. T-joint llet welds are some of the most common types of
welds used in the fabrication of structural members in civil engi-
neering, shipbuilding, automobile manufacturing, and other indus-
tries. During the last two decades, a number of FE analyses have
been carried out to investigate the residual stresses and distortions
in T-joint llet welds. Here, it is necessary to highlight some of the
most relevant papers in this eld. Teng et al. [7] designed a thermal
elastoplastic computational model to predict residual stresses and
distortions in T-joint llet welds based on the ange thickness,
welding penetration depth, and restraint condition of welding.
Therein, the technique of element birth and death is used to simu-
late the weld ller variation with time [17]. The inuence on weld-
ing deformation of the ange thickness is investigated
experimentally and numerically by Deng et al. [1]. Here, the angu-
lar distortion generation mechanism is addressed using a numeri-
cal simulation. It is concluded that the distribution of transverse
shrinkage along the welding line is almost unaffected by the ange
thickness. Chang and Lee [18] performed 3D uncoupled thermome-
chanical FE analyses of the residual stresses in the T-joint llet
welds made of similar and dissimilar steels. Moreover, Ogawa
et al. [15] experimentally and numerically studied the characteris-
tics of the welding residual stress distribution in penetration noz-
zles welded by a multi-pass J-groove joint. In another study,
Gannon et al. [4] discussed the inuence of welding sequences
on the distribution of residual stress and distortions generated by
the welding of a at-bar stiffener to a steel plate.
However, in the above-mentioned papers, the authors performed
the time-consuming and computationally expensive three-dimen-
sional FE analysis. In contrast, this paper presents a numerical anal-
ysis technique that provides inexpensive and reasonable accuracy
predictions of welding-induced distortion in large structures. In
the FE model, it is assumed that the base metal and weld metal have
the same thermal and mechanical properties. Moreover, the creep
strains and the transformation plasticity are not taken into account.
All computations have been performed within the FE software Aba-
qus/Standard [16]. As shown in the literature, thermocouples are
widely used in various congurations for the measurement of the
temperature distribution during the welding process [9,13]. Because
the change in the temperature during welding is very fast, the inher-
ent slow and low spatial response of thermocouples represents a sig-
nicant problem [19]. Recently, infrared thermography has been
used for surface temperature measurements [20]. A detailed over-
view of the applications of the thermocouples and infrared thermog-
raphy for the measurement of welding temperature is presented in
[21]. In this study, the welding experiments are conducted using the
metal active gas (MAG) welding process. Here, the temperature and
displacement measurements with thermocouples, infrared thermo-
graphic camera and digital image correlation (DIC) system are per-
formed and the test results are compared with those obtained
from the nite element analyses.
The paper is organized as follows. Section 2 contains a descrip-
tion of the experimental investigations. The welding conditions,
thermal measurements, and displacement measurements are
1053
presented. Therein, all procedures, measurement parameters, and
equipment are described in detail. Next, in Section 3, the full 3D
numerical model and a combined shell/3D nite element model
are presented. Section 4 contains detailed quantitative compari-
sons of the experimental and numerical temperature, residual
stresses, and displacement distributions induced by the T-joint
welding of two plates. Herein, the efciency and accuracy of the
shell/three-dimensional modeling technique are demonstrated. Fi-
nally, some concluding remarks are given in the last section.
2. Experimental investigations
2.1. Welding conditions
The geometry of two plates welded into a T-joint, including the
relevant dimensions, is presented in Fig. 1. The plates are made of
nonalloyed structural steel EN 10025-2: S355JR [22] and welded by
llet welds with 7 mm leg lengths. Joint preparation is suited for
the double-sided llet welds with partial penetration and no gap
between the horizontal and the vertical plate, according to EN
ISO 9692-1 Ref. No. 4.1.3 [23]. The positioning angle between the
horizontal and the vertical plate is 90. The MAG process EN ISO
4063-135-P [24] is used. The welding of the llet welds is per-
formed by a 6-axis robot supplied by a digital inverter power
source to ensure process stability. The welding parameters chosen
for this analysis were as follows: wire diameter = 1.2 mm classi-
cation EN ISO 14341-A: G 42 4M/C 3Si1 [25], shielding gas compo-
sition of 82% Ar/18% CO2 EN ISO 14175: M21 [26], welding current
I = 270 A, arc voltage U = 29 V, and welding speed v = 400 mm/min.
The welding sequence and direction are presented in Fig. 1. After
welding of the rst llet weld, the welding jig is rotated 180
and prepared for the second pass. The time period between the
end of the rst pass and the beginning of second pass was 215 s.
For the rst and second pass, the welding parameters remained
the same without any corrections. Material transfer during MAG
welding is supported by a pulse welding current through which
short circuiting and spattering are avoided. Visual testing accord-
ing to EN ISO 17637 [27] showed no imperfections; therefore, it
is concluded that the welded joint meets the quality requirements
of EN ISO 5817 class B acceptance criteria [28].
2.2. Thermal measurements
In this study, an SC2000 infrared camera (from FLIR Systems AB
[29]) is used for the surface temperature measurement during a T-
joint welding. The sensitivity of the camera is 80 mK at 30 C, and
the eld of view is 24  18, with a minimum focal distance of
0.5 m and a spatial resolution of 1.3 mrad. The camera detector is
of the focal plane array type and consists of a non-cooled microb-
olometer that has an array of 320  240 pixels and a spectral range
of 7.513 lm. The camera was positioned at a distance of 1.5 m
from the surface of the plates under consideration.
Furthermore, to evaluate the obtained thermograms, tempera-
ture measurements are simultaneously conducted by thermocou-
ples (Type K) that are spaced 6 mm apart in the material, at the
bottom edge of the horizontal plate. The location of the thermo-
couples is shown in Fig. 2.
The recording process is conducted with pauses between the
three stages of the experiment, which consisted of two passes of
the electrode and a cooling stage. The time increment for recording
thermograms was different for the rst two stages (2 s) and for the
nal cooling stage (30 s) of the experiment. An analysis of the ther-
mograms is performed for specic time instants (Fig. 3). The results
obtained from IR imaging are compared with those obtained from
the numerical simulations for the selected proles (proles 1, 2,
and 3), which are presented in Figs. 2 and 3.
1054 M. Peric et al. / Materials and Design 53 (2014) 10521063

10

d
el ld
w we

300
10

300
eb ell
W
Sh
7 7
y y
ss x ss
C pa C 50 50
st C z pa C
x Fir 0 st 3D
B z B 20 0 Fir 20
0
ss 50 B as
s ell B 0
pa 20
0 p Sh 0 50
A nd A A ell 3D on
d
A 20

t2
Flange co 50 Sh Se
c
50
Se
t1

10
10

300
300

mechanical restraint in X , Y and Z direction mechanical restraint in X , Y and Z direction

mechanical restraint in X and Y direction mechanical restraint in X and Y direction

mechanical restraint in Y direction mechanical restraint in Y direction

(a) (b)
Fig. 1. Geometry of the T-joint: (a) the full 3D model and (b) the shell/3D model.

300

150
50

TC-102
Flange
18 profile-2
Second pass
First pass

200
TC-103

TC-106
TC-105

500

18 profile-1
30
200

Web 45

TC-104

profile-3
50

18

Thermocouple

Fig. 2. Locations of the thermocouples.

Fig. 3. Thermograms; (a) 290 s after the beginning of the welding process and (b) 403 s after the beginning of the welding process.

2.3. Displacement measurements 3D digital image correlation system ARAMIS 4M (from GOM mbH)
[30]. The ARAMIS system calculates the surface strains from
In the presented experimental investigation, the displacements the measured displacements, a process which in our experi-
of the plates during welding are measured by using the noncontact ments is synchronized with the thermographic measuring to nd
 M. Peric et al. / Materials and Design 53 (2014) 10521063
Fig. 4. Measured displacement distribution.
correlations between the displacement and strain of certain points
on the plate and the temperature. For this purpose, two digital
cameras are used that are calibrated prior to measuring. Before
the start of the experiment, the specimen is prepared by applying
a random spray pattern over the mat surface to clearly allocate the
pixels in the camera images [31]. To obtain a good-quality raster,
the mixture of titanium oxide and alcohol compressed to 1 bar is
applied on the plate surface with powder spray. After recording
the series of images, the optical system ARAMIS detects the dis-
placements of the specimen through the images by means of var-
ious square or rectangular facets. From each valid facet, a
measuring point is produced after the computation. The facet size
used in the study was 15  15 pixels.
The welds on both sides of the plates are performed in se-
quence, one after the other. Consequently, the cameras of the opti-
cal system ARAMIS are placed on the opposite side of the welding
arc so that the plates need to be rotated 180 between each weld-
ing pass. Accordingly, white markers are glued near the specimens
that later served as the reference points for data transformation.
Because the default coordinate system is xed with the ARAMIS,
a new coordinate system is dened by the reference points. The Tx; y; z; 0 T 0 x; y; z
strain and displacement data of a measuring project are trans-
formed into a new coordinate system to be interpreted correctly.
The full eld displacement distribution is shown in Fig. 4.
(a)
Fig. 5. Typical FE meshes: (a) full 3D; (b) combined shell/3D.
1055
3. Numerical model
Within the framework of numerical investigations, two models
are considered: a full 3D model (Fig. 1a) and a combined shell/3D
nite element model (Fig. 1b). To discretize the full 3D geometry,
three-dimensional 8-node solid DC3D8 elements are applied for
the thermal model, and C3D8 elements are applied for the mechan-
ical stress analysis. Herein, the size of elements becomes larger as
the distance from the root of the weld increases, as shown in
Fig. 5a, which presents the nite element mesh. The 3D mesh con-
sists of 22,176 elements, which corresponds to 27,540 degrees of
freedom (DOF) in the thermal analysis and 82613 DOF in the
mechanical analysis. For the numerical simulation of the welding
process using the combination of 3D and shell elements, ve differ-
ent models are analyzed (T 12  7, T 15  12.9, T 18  19.3, T
24  26.2, and T 30  33.7). They differed in the size of the struc-
ture discretized with 3D continuum elements, t1 and t2, which
are dened in Fig. 1b. The dimensions and the number of nite ele-
ments together with the corresponding degrees of freedom for all
ve shell/3D models are provided in Table 1. To discretize the
shell/3D geometry, DC3D8 solid elements and DS4 shell elements
are used during the thermal analysis, while C3D8 solid elements
and S4 shell elements with ve degrees of freedom per node are
applied for the mechanical analysis. Herein, the portion of the T-
joint closest to the weld is discretized with 3D continuum ele-
ments using the same elements as in the full 3D model (Fig. 5b).
Both models in this study adopt the element birth and death tech-
nique to simulate the weld ller deposition. More details concern-
ing the applied shell/3D modeling technique can be found in the
authors previous papers [11,12]. In all computational studies, the
same FE discretization has been adopted for both the thermal
and mechanical calculation.
In the thermal analysis, the nite element formulation is based
on the governing equation for transient nonlinear heat transfer
analysis in the form
     
@ @T @ @T @ @T @T
kx ky kz Q qC 1
@x @x @y @y @z @z @t
where kx, ky, and kz are the thermal conductivities in the x, y and z
directions, respectively; T is the current temperature; Q is the heat
generation; q is the density; C is the specic heat capacity; and t is
the time, respectively. A general solution of Eq. (1) is obtained by
introducing the following initial and boundary conditions:
2
 
@T @T @T
kx Nx ky Ny kz Nz qs hc T  T 1 hr T  T r 0 3
@x @y @z
(b)
1056 M. Peric et al. / Materials and Design 53 (2014) 10521063

Table 1
Dimensions and number of degrees of freedom for the shell/3D models.

Shell/3D model t1/mm t2/mm Total number Total number Total number of degrees of freedom Total number of degrees of freedom
of 3D elements of shell elements in the thermal analysis in the mechanical analysis
T 12  7 12.0 7 4536 3864 9945 36,713
T 15  12.9 15.0 12.9 5712 3612 11,135 39,773
T 18  19.3 18.0 19.3 6888 3360 12,325 42,833
T 24  26.2 24.0 26.2 8064 3108 13,515 47,043
T 30  33.7 30.0 33.7 9240 2856 14,705 48,953

where Nx, Ny, and Nz are the direction cosine of the normal to the 500
boundary; hc and hr are the convection and radiation heat transfer
Youngs Modulus (GPa),
coefcients, respectively; qs is the boundary heat ux; Tr denotes (p=0.1)

Mechanical Properties
400 Thermal Expansion Coefficient (1/C)
the temperature of radiation; and T1 represents the surrounding Poisson ratio
temperature. Radiation heat losses are dominant near the weld *Yield Stress (MPa)
300
and can be expressed by the following equation: (p=0) **Yield Stress (MPa)
 
hr reF T 2 T 2r T T r 4 200
-7
(10 )

where r = 5.67  108 J/(m2 K4) is the StefanBoltzmann constant, e 100

(10-2)
is the effective emissivity, and F is the conguration factor. The
numerical simulation of the welding process is performed by apply- 0
ing a distributed heat ux to the weld elements. In so doing, the to- 0 200 400 600 800 1000 1200 1400 1600
tal heat input is given by: Temperature ( C)
gUI Fig. 7. Mechanical properties of S355JR steel [17].
Q ; 5
VH
where g represents the efciency of the heat input, I is current, U is 4. Results and discussion
the arc voltage, and VH is the volume of the activated weld bead ele-
ment. The thermal material properties depending on temperature, This section contains detailed quantitative comparisons of the
adopted from Pilipenko [17], are shown in Fig. 6. Furthermore, the experimental and numerical results. To validate the temperature
following data are assumed: the convection heat transfer coefcient and residual stress calculations performed by the shell/3D model-
hc = 10 W/m2 K, the efciency of the heat input g = 83%, and the ing technique, the results are rstly compared with those obtained
emissivity e = 0.9. The heat ux applied to the weld was uniform, from the full 3D model nite element analyses. Herein, to dene
Q = 4.0  1010 J/m3s. the optimal shell/3D model that has both accuracy and the appro-
As mentioned in the introduction section of the paper, the priate convergence of the solution, the full 3D solid element mesh
mechanical analysis is performed by assuming that the base metal is used as a reference solution. The measured temperatures and
and weld metal have the same mechanical properties. The bound- displacements are then compared with the results obtained using
ary conditions that are applied are schematically shown in Fig. 1, both the optimal shell/3D and the full 3D models.
while the temperature-dependent mechanical properties of
S355JR steel, adopted from Pilipenko [17], are presented in Fig. 7.
4.1. Thermal analysis
The material is modeled as an elasticplastic solid with kinematic
hardening. Nonlinear material behavior is modeled using incre-
Fig. 8 shows a comparison of results for the temperature distribu-
mental plasticity with the von Mises yield criterion together with
tion for the full 3D and all ve shell/3D models along the line BB
the associated ow rule and assuming geometrically nonlinear
seen in Fig. 1, 403 s after the beginning of the welding process. It
behavior of the two plates. For more details about an algorithm
can be seen from the results obtained by the shell/3D modeling tech-
for solving nonisothermal elastoplastic problems with the nite
nique and full 3D model that the size of the 3D zone (t1 and t2, seen in
element method, see the authors earlier paper [32].
Fig. 1) does not inuence the temperature eld distribution. It is

1,4
Thermal conductivity 200
Full 3D
1,2 175 T 12 x 7
Specific heat
Temperature (C)

T 15 x 12.9
Thermal Properties

Density
150
1 T 18 x 19.3
-2 3
125 T 24 x 26.2
0,8 (10 g/mm ) T 30 x 33.7
100

0,6 C (J/g/C) 75
50
0,4
25
0,2 k (10 J/mm/s/C)
-1 0
0 25 50 75 100 125 150 175 200 225 250 275 300
0 X - Coordinate (mm)
0 200 400 600 800 1000 1200 1400 1600
Temperature ( C) Fig. 8. Comparison of temperature proles for the 3D and the shell/3D models 403 s
after the beginning of the welding process on the middle surface of the ange along
Fig. 6. Thermal properties of S355JR steel [17]. the line BB seen in Fig. 1.
 M. Peric et al. / Materials and Design 53 (2014) 10521063 1057

Fig. 9. Comparison between the temperature proles obtained using the thermocouples and the IR thermographic camera 290 s and 403 s after the beginning of the welding
process.

Fig. 10. Temperature proles obtained using the nite element method 290 s and 403 s after the beginning of the welding process.

245 245

220 220

195 Abaqus, 290 s 195 Abaqus, 403 s

Temperature (C)

Temperature (C)

IR camera, 290 s IR camera, 403 s

170 170
TC-103, 290 s TC-103, 403 s
145 145
TC-105, 290 s TC-105, 403 s
120 120
TC-106, 290 s TC-106, 403 s
95 95

70 70

45 45
20 20
0 15 30 45 60 75 90 105 120 135 150 0 15 30 45 60 75 90 105 120 135 150
X-Coordinate (mm) X-Coordinate (mm)
(a) (b)
Fig. 11. Comparison between the temperature proles obtained using the thermocouples, the IR thermographic camera, and the nite element method, prole-1; (a) 290 s
after the beginning of the welding process and (b) 403 s after the beginning of the welding process.

obvious that the temperature distribution at the transition between
3D and shell elements is smooth, without large differences in the 3D
and shell nodes on the shell-to-solid interface. Therefore, it could be
concluded that compared to the 3D model, all ve shell/3D models
analyzed in this study are equally suitable for describing the temper-
ature eld distribution.
Fig. 9 shows the comparison between temperature proles ob-
tained using the thermocouples and IR thermographic camera
290 s and 403 s after the beginning of the welding process at the
locations of the installed thermocouples. The associated FE results
are presented in Fig. 10.
A comparison between the temperature proles obtained using
the thermocouples, the IR thermographic camera and the FE meth-
od 290 s and 403 s after the beginning of the welding process are
shown in Figs. 1113. As seen from the temperature proles
(prole-1, prole-2, and prole-3, seen in Fig. 2) in Figs. 11a, 12a,
and 13a, the temperatures of all three proles are approximately
equal because the proles show temperatures 215 s after the pas-
sage of the rst electrode (prolonged cooling time). In contrast,
Figs. 11b, 12b, and 13b show the temperature proles obtained
only 38 s after the passing of the second electrode, which is not
sufcient for the equalization of the temperature distribution at
the above proles. The same conclusion can be drawn from the
thermograms shown in Fig. 3 and from the FE simulation results
shown in Figs. 14 and 15.
As may be seen from Figs. 913, the results obtained by the IR
thermographic camera are in good agreement with the results ob-
served experimentally by the thermocouples. In contrast, it is evi-
dent that the temperatures obtained by numerical simulation are
somewhat lower than those obtained by experimental measure-
ments. This may have occurred because the use of simplied mod-
els in this study to simulate the heat ux distribution and thermal
boundary conditions. A more realistic but more complicated deni-
tion of the heat ux distribution, such as a combination of the
Gaussian surface heat ux and volumetric heat ux [1,4], is re-
placed by a uniform heat ux per weld volume. Furthermore, the
convection heat transfer coefcient and the emissivity factor are
assumed to be constant, i.e., independent of temperature.
1058 M. Peric et al. / Materials and Design 53 (2014) 10521063

245 245

220 220

195 195
Temperature (C)

Temperature (C)
Abaqus, 290 s Abaqus, 403 s
170 170
IR camera, 290 s IR camera, 403 s
145 145
TC-102, 290 s TC-102, 403 s
120 120

95 95

70 70

45 45

20 20
0 15 30 45 60 75 90 105 120 135 150 0 15 30 45 60 75 90 105 120 135 150
X-Coordinate (mm) X-Coordinate (mm)
(a) (b)
Fig. 12. Comparison between the temperature proles obtained using the thermocouples, the IR thermographic camera, and the nite element method, prole-2; (a) 290 s
after the beginning of the welding process and (b) 403 s after the beginning of the welding process.

245 245
220 220
195 195
Temperature (C)

Temperature (C)

Abaqus, 290 s Abaqus, 403 s

170 170
IR camera, 290 s IR camera, 403 s
145 145
TC-104, 290 s TC-104, 403 s
120 120
95 95
70 70
45 45

20 20
0 15 30 45 60 75 90 105 120 135 150 0 15 30 45 60 75 90 105 120 135 150
X-Coordinate (mm) X-Coordinate (mm)
(a) (b)
Fig. 13. Comparison between the temperature proles obtained using the thermocouples, the IR thermographic camera, and the nite element method, prole-3; (a) 290 s
after the beginning of the welding process and (b) 403 s after the beginning of the welding process.

Fig. 14. Temperature eld distribution 290 s after the beginning of the welding process: (a) full 3D model and (b) shell/3D model.
 M. Peric et al. / Materials and Design 53 (2014) 10521063 1059

Fig. 15. Temperature eld distribution 403 s after the beginning of the welding process: (a) full 3D model and (b) shell/3D model.

0,5

0,0
0 50 100 150 200 250 300
-0,5

-1,0
Deflection (mm)

-1,5

-2,0

-2,5
Full3D
-3,0 T 12 x 7
T 15 x 12.9
-3,5
T 18 x 19.3
-4,0 T 24 x 26.2
T 30 x 33.7
-4,5
X-Coordinate (mm)

Fig. 16. Deections in the middle surface of the ange after the welding process
along the line BB in Fig. 1.

In most previously published works, infrared thermography has

been applied to measure the welding temperature distribution in
models with simple geometries, such as the butt joint of two plates
[19]. This study shows that thermography can also be an effective
tool for the measurement of the welding temperature eld in com-
plex geometries. In contrast to the spot temperature measure-
ments provided by commonly used thermocouples [9,13],
infrared thermography provides the temperature distribution over
the entire surface of the T-joint llet weld (Fig. 3). This allows the
verication of the numerical models on a whole-eld basis (Figs. 14
and 15). Therefore, infrared thermography represents a useful
method for the evaluation of the input data in a thermo-mechani-
cal determination of residual stresses and distortions induced by
the welding process.
4.2. Mechanical analysis
The next step is to test the shell/3D modeling technique to pre-
dict the out-of-plane deformation and residual stresses in a T-joint
llet weld. The plate deections in the y direction in the middle
surface of the ange along the line BB seen in Fig. 1 after comple-
tion of the welding and cooling process is shown in Fig. 16. As may
Fig. 17. Residual stress eld r2 in the weld direction on the upper surface of the
ange (detail A in Fig. 5).
be seen, the deections obtained by the T12  7 and T15  12.9
models differ considerably from those obtained using the full 3D
model. Only when a sufcient width (zone described by t1 and t2,
seen in Fig. 1) is modeled with 3D elements (in this case, the
T18  19.3, T24  26.2, and T30  33.7 models) are approximately
the same results achieved as are obtained with the full 3D model.
In contrast, the results for residual stress (Fig. 17) show that only
the T30  33.7 model has a smooth stress distribution on the
shell-to-solid interface that is in accordance with the full 3D mod-
el. Furthermore, the other models show discontinuities in the
stress on the line between the shell and the 3D model. The reason
for such behavior is that the shell-to-solid interfaces of these mod-
els lie too close to the weld where the large stress gradients occur.
An extremely discontinuous distribution from the 3D elements to
the shell elements is visible using the smallest T12  7 model. In
this way, we can conclude that with the full 3D and the shell/3D
models, we can obtain approximately an equal deection
1060 M. Peric et al. / Materials and Design 53 (2014) 10521063

500

Longitudinal residual stress z (MPa)

0,5
400
0
0 50 100 150 200 250 300
-0,5 Full 3D 300
Deflection (mm)

Full 3D
Experiment
-1
T 30 x 33.7 T 30 x 33.7
200
-1,5

-2 100
-2,5
0
-3 0 50 100 150 200 250 300

-3,5 -100
X-Coordinate (mm) X-Coordinate (mm)
(a) (b)
Fig. 18. (a) Deections on the middle surface of the ange along the line BB shown in Fig. 1. (b) Residual stresses in the weld direction along the line BB shown in Fig. 1, on
the upper surface of the ange.

Longitudinal residual stress z (MPa)

0,5 500

0,0
0 50 100 150 200 250 300 400
-0,5 Full 3D
Deflection (mm)

300
-1,0 Experiment Full 3D
T 30 x 33.7 T 30 x 33.7
-1,5 200

-2,0
100
-2,5
0
-3,0 0 50 100 150 200 250 300

-3,5 -100
X-Coordinate (mm) X-Coordinate (mm)
(a) (b)
Fig. 19. (a) Deections on the middle surface of the ange along the line AA shown in Fig. 1. (b) Residual stresses in the weld direction along the line AA shown in Fig. 1, on
the upper surface of the ange.

500
Longitudinal residual stress z (MPa)

0,5

0,0 400
0 50 100 150 200 250 300
-0,5 Full 3D
Deflection (mm)

Experiment 300
-1,0 Full 3D
T 30 x 33.7
200 T 30 x 33.7
-1,5

-2,0
100
-2,5
0
-3,0 0 50 100 150 200 250 300

-3,5 -100
X-Coordinate (mm) X-Coordinate (mm)

(a) (b)
Fig. 20. (a) Deections on the middle surface of the ange along the line CC shown in Fig. 1. (b) Residual stresses in the weld direction along the line CC shown in Fig. 1, on
the upper surface of the ange.

distribution with a smaller size of the 3D zone in relation to the est T30  33.7 shell/3D model (t1 = 30 mm, t2 = 33.7 mm), which
residual stresses. For this reason, to obtain a continued stress demonstrates a good match for deections and stresses in compar-
transfer across the shell-to-solid interface, the size of the 3D zone ison with the full 3D model.
must be chosen according to the stress criterion. Thus, in the text A comparison of the numerical results and experimental mea-
that follows, the numerical results are presented only for the larg- surements, obtained by the optical system ARAMIS 4M, for the
 M. Peric et al. / Materials and Design 53 (2014) 10521063 1061

plate deections in the y direction along the lines BB, AA, and
CC after completion of the welding and cooling process is given
in Figs. 18a, 19a, and 20a. As shown in Fig. 1, line CC is near the
rst weld start, line BB is in the middle of the ange, and line
AA is near the rst weld end. Herein, the full 3D numerical results,
together with the results obtained by the shell/3D modeling tech-
nique (model T30  33.7), are presented. As may be seen, the dif-
ferences between the simulated results and the experimental
measurements are not signicant. Fig. 21 shows the full eld ver-
tical deection distribution after the welding and cooling process
completion obtained by the shell/3D modeling technique and full
3D model. It is clear from both this gure and Fig. 4 that the sim-
ulated results are in good agreement with the results obtained
using the Aramis 3D-surface DIC system. The full-eld displace-
ment measurements provided by the DIC system demonstrate that
DIC is an effective tool for distortion study during the welding pro-
cess. In addition, this technique has proven to be a suitable
verication tool for FE welding simulations. Unlike the high-tem-
perature strain gauges [8], which measure the strain in a small area
of the welded specimen, the DIC system allows the displacements
and strains over an entire eld to be observed. Moreover, the DIC
system does not require temperature compensation and allows
displacement measurements close to the weld bead [31].
The residual stresses in the weld direction (along the lines BB,
AA, and CC, seen in Fig. 1b) on the upper surface of the ange
and the residual stress eld distribution after the welding and
cooling process completion obtained by the shell/3D modeling
technique and the full 3D model are presented in Figs. 18b, 19b,
20b, and 22. As seen from the gures, the tensile stresses at the
weld line reach the yield strength of the material, which then de-
creases to zero, nally becoming compressive at distances far from
the weld line. It should be mentioned that almost the same results
are obtained using the full 3D nite element model and the shell/
3D model. It can therefore be concluded that the shell/3D nite

Fig. 21. Vertical deection distribution: (a) full 3D model and (b) shell/3D model.

Fig. 22. Residual stress eld distributions in the welding direction: (a) full 3D model and (b) shell/3D model.
1062 M. Peric et al. / Materials and Design 53 (2014) 10521063
Table 2
Summary of the computational requirements for the T-joint welding of two plates.
Model Thermal analysis CPU/s
T 30  33.7 174,678
Full 3D 324,332
element modeling technique presented in this study can be effec-
tively used to predict the residual stresses and distortions induced
by the T-joint welding of two plates and large welded structures.
Because our aim is to develop a computational procedure that
provides a signicant reduction in the computational space and
the CPU time needed for simulation of the welding process, we
present the CPU time consumed during the T-joint welding of
the two plates. Table 2 summarizes the typical computational
requirements for the full 3D solid and the shell/3D thermal and
mechanical analyses. The results are given for the case using the
T30  33.7 model. All numerical simulations of the welding pro-
cesses shown in this study are performed on a Sun three-node clus-
ter with 48 cores in total, 32 GB of RAM, and 500 GB of storage
each. As seen, there is a considerable savings in the calculating
time when the shell/3D model, obtained by the combination of
the shell and the 3D elements, is used instead of the full 3D model.
The total computational time savings for the analysis in this case is
approximately 42%.
5. Conclusions
In the presented research, both an experimental method and -
nite element simulations are used to investigate welding distortion
in a T-joint llet weld. With the goal of increase the accuracy and
efciency of the welding simulation of large structures, the numer-
ical analysis is carried out by applying the shell/3D nite element
technique, which is based on the transition from the shell model to
the 3D model discretized with a ner mesh of elements. The inu-
ence of the 3D zone size, or the distance from the weld zone to the
shell-to-solid interface, on the convergence of the results is inves-
tigated. For that reason, the distance from the weld zone that is
needed to transition from the 3D elements to the shell elements
is dened by comparing the resulting temperatures, residual stres-
ses, and deections obtained from the shell/3D model and the full
3D model. Five different shell/3D models are considered. In addi-
tion, the welding experiments are conducted and the measured
temperatures and displacements are compared with the results ob-
tained by the numerical analysis. It is shown that the numerical re-
sults agree well with the experimental results.
From the numerical results shown in this study, it can be con-
cluded that the size of the 3D zone in the shell/3D model does
not inuence the temperature eld distribution. It is shown that
the temperature distribution at the shell-to-solid interface is
smooth, without large differences in the 3D and the shell nodes.
Furthermore, the minimal size of the 3D zone is selected by mon-
itoring deection and residual stress distributions. The simulated
results demonstrate that the size of the 3D zone on the ange that
is necessary to obtain a satisfactory solution for deections is
approximately 2 wall thickness, and for stresses, it is approxi-
mately 3 wall thicknesses. In this way, we can conclude that to ob-
tain a continued stress transfer across the shell-to-solid interface,
the size of the 3D zone must be chosen according to the stress
criterion.
The FE simulation results have proved that the proposed shell/
3D technique is an effective method to predict welding distortions
and residual stresses for large structures. Moreover, the shell/3D
analysis can be completed in a short time. It is shown that the total
computational time savings is approximately 42% when the
Mechanical analysis CPU/s Total CPU/s
104,238 278,916
156,763 481,095
shell/3D model is used instead of the full 3D model. Thus, it is
meaningful to use this method for practical engineering analysis.
Acknowledgments
The authors express their gratitude to the Ministry of Science,
Education, and Sports of the Republic of Croatia for their nancial
support.
References
[1] Deng D, Liang W, Murakawa H. Determination of welding deformation in llet-
welded joint by means of numerical simulation and comparison with
experimental measurements. J Mater Process Technol 2007;183:21925.
[2] Lee CH, Chang KH. Three-dimensional nite element simulation of residual
stresses in circumferential welds of steel pipe diameter effects. Mater Sci Eng A
2008;487:2108.
[3] Deng D. FEM prediction of welding residual stress and distortion in carbon
steel considering phase transformation effects. Mater Des 2009;30:35966.
[4] Gannon L, Liu Y, Pegg N, Smith M. Effect of welding sequence on residual stress
and distortion in at-bar stiffened plates. Mar Struct 2010;23:385404.
[5] Long H, Gery D, Carlier A, Maropoulos PG. Prediction of welding distortion in
butt joint of thin plates. Mater Des 2009;30:412635.
[6] Barsoum Z, Lundbck A. Simplied FE welding simulation of llet welds 3D
effects on formation residual stresses. Eng Fail Anal 2009;16:22819.
[7] Teng TL, Fung CP, Chang PH, Yang WC. Analysis of residual stresses and
distortions in T-Joint llet welds. Int J Press Vess Pip 2001;78:52338.
[8] Shan X, Davies CM, Wangsdan T, ODowd NP, Nikbin KM. Thermo-mechanical
modeling of a single-bead-on-plate weld using nite element method. Int J
Press Vess Pip 2009;86:11021.
[9] Wang R, Zhang J, Serizawa H, Murakawa H. Study of welding inherent
deformations in thin plates on nite element analysis using interactive
substructure method. Mater Des 2009;30:347481.
[10] Marenic E, Skozrit I, Tonkovic Z. On the calculation of stress intensity factors
and J-integrals using the submodeling technique. J Press Ves Technol
2010;132(4):04120341212.
[11] Peric M, Stamenkovic D, Milkovic V. Comparison of residual stresses in butt-
welded plates using software packages Abaqus and Ansys. Sci Tech Rev
2010;60(34):226.
[12] Peric M, Tonkovic Z, Karaj I. Numerical analysis of residual stresses in
welding process using a shell/3D modeling technique. In: Proceedings of
international conference on advances in welding science and technology for
construction, Energy and Transportation Systems (AWST-2010), Istanbul,
Turkey; 2010.
[13] Attarha MJ, Sattari-Far I. Study on welding temperature distribution in thin
welded plates through experimental measurements and nite element
simulation. J Mater Process Technol 2011;211:66894.
[14] Wang S, Goldak J, Zhou J, Tchernov S, Downey D. Simulation on thermal cycle
of a welding process by space-time convection-diffusion nite element
analysis. Int J Thermal Sci 2009;48:93647.
[15] Ogawa K, Deng D, Kiyoshima S, Yanagida N, Saito K. Investigations on welding
residual stresses in penetration nozzles by means of 3D thermal elastic plastic
FEM and experiment. Comput Mater Sci 2009:103142.
[16] Abaqus/Standard 6.10.1, Dassault Systemes. Simulia; 2010.
[17] Pilipenko A. Computer simulation of residual stress and distortion of thick
plates in multi-electrode submerged arc welding. Their mitigation techniques.
Doctoral Thesis, Trondheim, 2001.
[18] Chang KH, Lee CH. Finite element analysis of the residual stresses in T-joint
llet welds made of similar and dissimilar steels. Int J Adv Manuf Technol
2009;41:2508.
[19] Camilleri D, Comlekci T, Gray TGF. Use of thermography to calibrate fusion
welding procedures in virtual fabrication applications. In: Proceedings of
inframation 2004 conference, Las Vegas, USA, 2004.
[20] vaic S, Boras I. Application of control volume numerical method in
thermographic analysis of relative material loss. Proceedings of QIRT06
International Conference Quantitative InfraRed Thermography, Padova, Italy,
2006.
[21] Bagavathiappan S, Lahiri BB, Saravanan T, Philip J, Jayakumar T. Infrared
thermography for condition monitoring a review. Infrared Phys Technol
2013;60:3555.
[22] Hot rolled products of structural steels part 2: technical delivery conditions
for non-alloy structural steels. EN 10025-2;2004.
 M. Peric et al. / Materials and Design 53 (2014) 10521063
[23] Welding and allied processes recommendations for joint preparation, Part 1:
Manual metal-arc welding, gas-shielded metal-arc welding, gas welding, TIG
welding and beam welding of steels. EN ISO 9692-1;2003.
[24] Welding and allied processes nomenclature of processes and reference
numbers. EN ISO 4063;2010.
[25] Welding consumables wire electrodes and weld deposits for gas shielded
metal arc welding of non alloy and ne grain steels classication. EN ISO
14341;2011.
[26] Welding consumables gases and gas mixtures for fusion welding and allied
processes. EN ISO 14175;2008.
[27] Non-destructive testing of welds visual testing of fusion-welded joints. EN
ISO 17637;2011.
1063
[28] Welding-fusion-welded joints in steel, nickel, titanium and their alloys-quality
levels for imperfections. EN ISO 5817;2007.
[29] ThermaCAM SC2000, Operator0 s Manual, FLIR Systems AB, 2000.
[30] Aramis User Manual Software, GOM mbH, 2009.
[31] De Strycker M, Lava P, Van Paepegem W, Schueremans L, Debruyne D.
Measuring welding deformations with the digital image correlation technique.
Weld J 2011;90:10712.
[32] Tonkovic Z, Soric J, Krtzig WB. On nonisothermal elastoplastic analysis of
shell-components employing realistic hardening responses. Int J Solid Struct
2001;38:501939.