Materials and Design 47 (2013) 615–623

Contents lists available at SciVerse ScienceDirect

Materials and Design

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

Numerical and experimental investigation of T-shape fillet welding of AISI 304

stainless steel plates
Farid Vakili-Tahami, Ali Ziaei-Asl ⇑
Department of Mechanical Engineering, University of Tabriz, Tabriz, Iran

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

Article history: Current study uses experimental data to develop an empirical relationship to model welding process in a
Received 13 October 2012 fillet-weld joint of AISI 304 stainless steel plates. A new Double-Ellipsoidal Heat Source (DEHS) model,
Accepted 26 December 2012 which is based on Goldak model, is developed to simulate fillet-welding process. Then, the extended
Available online 3 January 2013
model is implemented into a finite element code on which a 2D and also a 3D solutions are used to
simulate the temperature field in the weldment. A series of experimental measurements and numerical
Keywords: analyses have been carried out and the effects of temperature dependent material properties and welding
Fillet welding
heat input on temperature field and deformation are investigated.
Thermal field
T-shape joint
The results are compared with the experimental data, and the constants of the empirical relationship
Finite element analysis have been obtained using model updating method. For this purpose, an optimization computer code has
been developed to modify the initial values in order to update the model to achieve agreement with
measured data.
The results show that 2D model can only be used in thermal analysis; whereas, the developed 3D
model can predict thermo-mechanical behavior with acceptable accuracy. The major advantage of this
formulation is that the number of unknown coefficients has been reduced to only one coefficient and
other coefficients have been related to the physical or geometrical parameters that are known for each
weldment.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction out experimental measurements in welding is very difficult and

Fillet welds are extensively used in ships, bridge structures,
pressure vessels and piping. These welds are an extremely com-
mon method of connecting various elements of a welded construc-
tion, e.g. as in lap, T and cruciform joints [1]. The non-uniform
expansion and contraction of the weld and surrounding base mate-
rial, due to the heating and cooling cycle during the welding pro-
cess, lead to thermal stresses in the weld and the adjacent areas.
Also, because of the localized high temperature and severe temper-
ature gradient in welding, different types of problems may occur
during welding and after [2–4]. In many cases, such as hot tapping,
it is necessary to carry out in service welding to provide T-shape
branch connections. This process also may cause burn-through or
hot cracking. The former phenomenon has always been a major
concern in the industry [5]. Since the physical reasons behind all
these problems lay on the welding process, it is very important
to study and model this phenomenon as accurate as possible.
Due the complexity of the welding process, it is crucial to use
experimental data to develop a reliable model. Although carrying
⇑ Corresponding author. Tel.: +98 (0)411 3392461; fax: +98 (0)411 3354153.
E-mail addresses: f_vakili@tabrizu.ac.ir (F. Vakili-Tahami), ziaie_ali@yahoo.com,
a-ziaei@tabrizu.ac.ir (A. Ziaei-Asl).
expensive. On the other hand, the prediction of temperature and
stress fields during the welding process and cooling period is not
an easy task due to the complexity involved. Hence, using numer-
ical methods such as finite element based simulations gained a
considerable popularity in predicting the adverse consequences
of welding phenomenon in the last three decades [6–10].
As the simulation of the welding processes is highly computa-
tionally intensive and a large computer storage and CPU time are
required, most of the recent research works reduce computational
requirements by using simplifying assumptions in numerical
simulations [11,12]. These assumptions severely reduce the com-
putational demand at the cost of results accuracy. However, over
simplified models usually lead to unrealistic results.
Many analytical and numerical models have been proposed for
butt-weld joints, and a number of databases have been established
[13–16]. Also, limited research works which describe fillet welds
are available [17,18]. For example, a thermal and mechanical sim-
ulation was performed using a well-established two and three-
dimensional code to study the formation of the residual stresses
due to 3D effect of the welding process [19].
In general, any accurate model should consider temperature
dependence of the physical and mechanical properties of the

0261-3069/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.matdes.2012.12.064
616 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623

Table 1 Table 2
Typical chemical composition of base metal and All-weld metal. Welding conditions.

Wt.% of component C Cr Mn Ni P S Si Case Plate thickness Current Voltage Speed Length of leg
No. (mm) (A) (V) (mm/s) (mm)
Base metal 0.06 18.27 1.95 8.93 0.028 0.011 0.40
All-weld metal 0.03 19.8 0.8 10.2 – 0.021 0.8 1 5 90 28 2.34 5
2 6 90 28 2.34 5
3 8 90 28 2.34 5

material, as well as the three-dimensional (3D) fields of the tem-

perature and stresses around the heat source [11,20,21].
In order to validate the numerical solutions, it is necessary to
compare the results with the experimental data. For example Oddy
and McDill [5] have carried out an exploratory study using 3D ther-
mal–mechanical Finite Element Analysis (FEA) of welding on pres-
surized vessels. The results of the FEA simulations were compared
with those of the physical tests.
Current study also uses an empirical relationship to model
welding process in a fillet-weld joint of stainless steel 304 plates
in different conditions. A series of numerical analyses has been car-
ried out, results are compared with the experimental data, and the
constants of this empirical relationship have been obtained using
model updating method. For this purpose, the Double-Ellipsoidal
Heat Source (DEHS) model proposed by Goldak et al. [22] is mod-
ified by introducing a new set of coefficients to simulate fillet weld
joint. Then, the extended model is implemented into the finite ele-
ment code on which a 2D and also a 3D models are used to obtain
the temperature field and deformation in the weldment.
2. Experimental procedure
Experiments were conducted to collect the welding information
to provide data for updating the FE model. Experiments have been
carried out on T shape fillet-weld joints of AISI 304 stainless steel
plates with different thicknesses. Because of superior mechanical
and corrosion properties of 304 austenitic stainless steel, it is used
widely in industry [23]. Temperature and displacement histories of
specific points on the specimens are measured continuously during
the welding and cooling periods, using an online data acquisition
system with computer interface.
2.1. Welding processing
The welding method selected for the experimental work is the
Manual Metal Arc Welding (MMAW) process and ‘‘AWS 308L-17
rutile coated core wire electrodes’’ with 2.5 mm diameter are used.
MMAW is usually used in ‘‘on-site’’ industrial applications, because
the equipment is relatively simple, portable, and inexpensive. In
order to remove moisture, the electrodes were baked in the oven
for 45 min before welding. The chemical compositions of the base
and all-weld metals are shown in Table 1.
This type of electrode is recommended for welding AISI 304
stainless steels by AWS A5.4/A5.4M [24]. The weld deposit con-
tains a maximum of 0.04% carbon. The lower carbon content min-
imizes the formation of chromium carbides; consequently, reduces
the intergranular corrosion up to +350 °C. This kind of electrode
has a silica–titania type coating. Additional silicon in the coating
acts as a wetting agent, having the effect of increasing puddle flu-
idity. This is particularly helpful with stainless steel, as it tends to
have more of a sluggish weld bead than carbon steel. Manganese is
considered to be a detrimental effect on the pitting resistance be-
cause of the formation of manganese sulfide. Cui and Lundin [25]
have used three weld-deposit compositions to describe the corro-
sion initiating phase between austenite and ferrite in 316 austen-
itic stainless steel weld metals. They concluded that because of
the primary austenite solidification mode, with the cores of the cel-
lular dendrites depleted in chromium and nickel relative to the
normal composition, austenite is the preferential phase for corro-
sion attack. Siewert [26] has showed that the toughness increases
with nickel content and decreases with ferrite number (a measure
of the ferrite phase content of austenitic stainless steels). Since
increasing the nickel content is also known to reduce the ferrite
number, nickel additions improve the toughness by both means.
A research [27] shows a linear increase in strength as the nitrogen
is increased from 0.04 to 0.28 wt.%. Lee et al. [28] while investigat-
ing the pitting corrosion behavior of welded joints of AISI 304L
using flux cored arc welding process, found that tensile and yield
strengths were increased with increasing equivalent ratio of Creq/
Nieq. Due to these facts, the type of the electrode and its cover re-
main the same in all tests.

(a) (b)

Fusion
Boundary

HAZ

Fig. 1. (a) Schematic diagram of T-shape joint and (b) snapshot of micrographs in the weld zone.
 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623
In these tests, the major dimensions of the plates for the T-
joints are: length 160 mm; breadth of the flange 80 mm; height
of the web 80 mm. To study the effect of plate thickness, different
thicknesses of 5, 6 and 8 mm are used.
First, the web and the flange are tack welded at both ends. Each
length of tack weld is 4 mm. The flange plate is firmly clamped to
the platform. Then the single pass welding is performed for each
case study using a single-sided welding process. Schematic dia-
gram of the T-shape joint is depicted in Fig. 1a. Since different
parameters affect the welding condition and some of them are very
difficult to control, for example the skills of the welder, a number
of tests have been carried out at the same condition to provide a
set of data for each case study. The welding conditions are given
in Table 2. The welded joints were sectioned and polished to show
the distribution of the weld metal and the Heat-Affected Zone
(HAZ). The penetration of weld metal and depth of HAZ were mea-
sured using OLYMPUS BX60 microscope with mercury fluores-
cence. Fig. 1b shows a snapshot of the micrographs, which have
been taken to observe the quality and geometry of the weldment.
2.2. Measurement of temperature and deflection
A data logger setup using K type thermocouples, micrometers,
and data acquisition system with computer interface is adopted
to measure the transient temperature and displacement histories
during the welding and cooling processes. The data logger is capa-
ble to record up to 10 temperatures in each second via eight chan-
nels. Temperature measuring system is designed according to
ASTM: E235/E235M-12 standard specification. Thermocouples
were fixed in the specific points in the bottom of the plate along
the welding line. The distributions of measurement points for ther-
mal cycles are shown in Fig. 2. Temperatures were measured every
0.1 s of time interval during the welding, and every 2 s during the
cooling periods. The measurements have been recorded until the
entire specimen cools down to a relatively stable temperature,
which is set to be the temperature variations less than 0.1 °C/s.
The deflections of the edges of the flange and web plates are con-
tinuously measured by using two micrometers with the accuracy
of ±0.01 mm. The test rig setup is shown in Fig. 3.
3. Computational procedure
In order to simulate the T-joint welded plates, 2D and 3D
finite element models are developed. The dimensions of the
Second welding
4 1
5 2
6 3
3mm
Fig. 2. Locations of temperature measurements.
617
models are the same as the test specimens. In the weld zone
and its vicinity, a fine mesh is adopted and the meshes become
gradually coarser away from the weld zone. Fig. 4 shows the
finite element mesh model for 6 mm-thickness plates. The min-
imum element size, number of nodes and elements are listed in
Table 3.
The solution procedure consists of two steps. At the first step,
the temperature distribution and its history are computed by the
thermal analysis. At the second step, the temperature history is
employed as a thermal load in the subsequent mechanical elasto-
plastic analysis. The temperature-dependent thermo-physical and
mechanical properties of the metal are used in the model as pre-
sented in Table 4. To accommodate the plastic deformation of
the material especially at high temperatures, a bilinear elasto-plas-
tic model has been used with the temperature dependent tangent
modulus.
3.1. Thermal analysis
During the welding, the governing equation for transient heat
transfer analysis is given by [11]:
@T !
qc ðx; y; z; tÞ ¼ r  q ðx; y; z; tÞ þ Qðx; y; z; tÞ ð1Þ
@t
in which, q is the density of the materials, c is the specific heat
capacity, T is the current temperature, t is the time, x, y and z are
the coordinates in the reference system, r is the spatial gradient
operator, ~
q is the heat flux vector and Q is the internal heat gener-
ation rate.
The non-liner isotropic Fourier heat flux constitutive equation is
also employed:
!
q ¼ krT ð2Þ
where k is the temperature-dependent thermal conductivity.
For high power density welding, the welding heat source acts
not only on the work piece surface, but also in the direction
of thickness. Therefore, it must be taken as a volumetric distri-
bution. For this purpose, an ellipsoid mode can be written as
[29]:
2
By2 Cz2 Þ
qf ðx; y; zÞ ¼ qm eðAx ð3Þ
where qm is the maximum value of the power density at the center
of the ellipsoid, and A, B, C are the distribution parameters. Since the
heat density is distributed inside the semi ellipsoid, we have
First welding
25 mm
15 mm
15 mm
3mm
25 mm
618 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623

Table 3
The minimum element, the number of node and element in numerical models.

Data acquisition Model Plate Minimum element Number of Number of

PC system No. thickness size (mm3) nodes elements
10 data per second (mm)
1 5 0.5  0.5  1 121045 105696
2 6 0.5  0.5  1 167486 143695
Displacement
6 channels of K-type 3 8 0.5  0.5  1 214817 183595
measurement
thermocouples accuracy: ±0.01mm

(a)  
pffiffiffi 2
x2 þy þðzv tÞ
2
6 3 I  V  g  ff 3 a2 b2 c2
qf ðx; y; z; tÞ ¼ pffiffiffiffi e f ð6Þ
p p abcf

pffiffiffi n o
2 2
6 3 I  V  g  fr 3 ax2 þyb2 þðzcv2tÞ
2

qr ðx; y; z; tÞ ¼ pffiffiffiffi e r ð7Þ

p p abcr

for front and rear of the welding center. In these equations, qf and qr
denote the front and rear heat flux distributions respectively, ff and
fr are the front and rear distribution fractions respectively and v is
the welding speed. a, b, cf and cr are the radii of the rear and front
ellipsoid. For the front part of the semi-ellipsoid, the heat deposi-
tion is
Z 1 Z 1 Z 1

(b) 2
0 0
qf ðx; y; zÞdxdydz
0
 
Fig. 3. Setup of the test rig for T-shape weld joint: (a) schematic diagram and (b) pffiffiffi Z 1 3 x2 þy2 þz2
6 3 Qff a2 b2 c2 1
experimental rig. ¼ 2  pffiffiffiffi e f dxdydz ¼ Q  ff ð8Þ
p p abcf 0 2
Similar integration can be carried out for the rear part.
Z 1 Z 1 Z 1
pffiffiffiffi Since gIV ¼ Q ¼ 0:5Q  ff þ 0:5Q  fr ¼ 0:5Q ðff þ fr Þ, then
q p p
Q ¼ gIV ¼ 4 qðx; y; zÞdxdydz ¼ mpffiffiffiffiffiffiffiffiffi ð4Þ ff þ fr ¼ 2. By implementing the DEHS defined by the Eqs. (6) and
0 0 0 2 ABC
(7) into Eq. (1) and assuming the initial homogeneous temperature
where I is the electric current, V is the voltage and g is the efficiency To, the temperature field is defined by
pffiffiffiffiffiffiffiffiffi pffiffiffiffi
of arc welding. This equation will give qm ¼ 2Q ABC =p p. To eval- h i
pffiffiffi Z 3y2 3x2
uate the parameters A, B and C, the semi-axes of the ellipsoid a, b 3 3Q t exp 12jðtt Þþb 2  12jðtt Þþa2
o
o
Tðx; y; z; tÞ ¼ T o þ pffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
and c in the directions x, y and z respectively, are defined such that p p 0 ½12jðt  t Þ þ a2 ½12jðt  t Þ þ b2 
o o
the heat density falls to 0.05qm at the surface of the ellipsoid [29],  
2
which gives: A ¼ 3=a2 ; B ¼ 3=b and C ¼ 3=c2 . Substituting into ff Af ð1 þ Bf Þ þ fr Ar ð1  Br Þ dt o ð9Þ
Eq. (3), the semi ellipsoid distribution is obtained as
where Af, Ar, Bf and Br are functions of z; t; t o ; cf ; cr and j = k/(qc)
pffiffiffi n o
2
6 3 I  V  g 3 ax22 þyb2 þcz22 is the thermal diffusivity. This equation gives the temperature at
qðx; y; zÞ ¼ p ffiffiffi
ffi e ð5Þ point (x, y, z) and instant t due to an instantaneous unit heat source
p p abc applied at point (xo, yo, zo) at instant to. The time integral in the Eq.
Considering the effect of welding speed, the power density (9) does not have a closed analytical expression, and recourse must
should be asymmetrically distributed ahead of the arc and behind be made to numerical integration.
it. Thus, Goldak et al. [22] proposed the double-ellipsoid heat The above DEHS model can be used for simple butt-welds
source (DEHS) which has been often used to approximate the heat which are carried out in a straight line on a horizontal flat plate,
source in common non-autogenous welding processes. The DEHS is with the electrode perpendicular to the plate. The coefficients of
expressed by the following equations: the model are dependent to the type of welding, its geometry

(a) 2D finite element model (b) 3D finite element model

Fig. 4. Finite element model of fillet weld for 6 mm-thickness plates.
 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623 619

Table 4
Temperature-dependent thermo-physical and mechanical properties of SUS 304 [11].

Temperature Specific heat (J/ Conductivity (W/ Density (g/ Yield stress Thermal expansion coefficient Young modulus Poisson’s
(°C) g °C) mm °C) mm3) (MPa) (1/ °C) (GPa) ratio
0 0.462 0.0146 0.790 265 1.70  105 198.5 0.294
100 0.496 0.0151 0.788 218 1.74  105 193.0 0.295
200 0.512 0.0161 0.783 186 1.80  105 185.0 0.301
300 0.525 0.0179 0.779 170 1.86  105 176.0 0.310
400 0.540 0.0180 0.775 155 1.91  105 167.0 0.318
600 0.577 0.0208 0.766 149 1.96  105 159.0 0.326
800 0.604 0.0239 0.756 91 2.02  105 151.0 0.333
1200 0.676 0.0322 0.737 25 2.07  105 60.0 0.339
1300 0.692 0.0337 0.732 21 2.11  105 20.0 0.342
1500 0.700 0.120 0.732 10 2.16  105 10.0 0.388

and materials. Therefore it is necessary to adopt the formulation to
each specific weldment. So in this study, a number of modifications
to the formulation of the DEHS has been made to accurately model
the T-shape joint of 304 stainless steel plates by MMAW process.
Also, a new set of coefficients related to the parameters of fusion
zone is introduced to simulate the behavior of the moving heat
source in the described arc welded T-shape joint. The resulting
equations are:
n o
½ðxx0 Þþðyy0 Þ2 ½ðxx0 Þðyy0 Þ2 ðzv tÞ2
I  V  g  ff 3Cos2 h pffi
½c 2=22
þ pffi
½c 22
qf ðx; y; z; tÞ ¼ ðkÞ e ðc 2Þ
V WeldBead
for z > v t
n o
½ðxx0 Þþðyy0 Þ2 ½ðxx0 Þðyy0 Þ2 ðzv tÞ2
I  V  g  fr 3Cos2 h pffi
½c 2=22
þ pffi
½c 22
qr ðx; y; z; tÞ ¼ ðkÞ e ðc 2Þ
V WeldBead
for z < v t
In these equations, k is a coefficient which is obtained based on
experimental data and VWeld Bead is the volume of weld bead. h is
the angle between x and x0 according to Fig. 5a. The description
of the welding, transverse and perpendicular directions (x, y, z)
are also given in Fig. 5. xo and yo show the position of the point
where the torch is aimed to.
The first and second terms of the exponent power of Eqs. (10)
and (11) represent the elliptical distribution of energy as shown
in Fig. 5. Also, the positional variation of the end of the electrode
against time is described by the third term of the exponent power
and simulates the motion of the heat source with the speed of v.
The moving heat source has a local co-ordinate system while the
stationary finite element mesh has a global co-ordinate system.
After implementing the co-ordinate transformation due to the
movement of the heat-source, the heat-flux can be calculated at
any point along the weld path. The parameters of the fusion zone:
a, b, cf and cr are determined based on the experimental results.
To simulate the weld material variation with time in T-joint fil-
let weld, new elements should be added to the mesh, while the
welding is in process. Meanwhile, the heat transfer boundary con-
ditions should also be modified after the new elements are added.
The thermal effects caused by solidification of the weld pool were
modeled by considering the latent heat of fusion.
Since 3D FE analysis requires high computational capacities,
also a 2D model has been developed with the form of:
 
½ðxx0 Þþðyy0 Þ2 ½ðxx0 Þðyy0 Þ2
8 I  V  g 3Cos
2
h p ffi
½c 2=22
þ p 2 ffi
½c 2
qðx; y; zÞ ¼  e
9 V WeldBead
All the parameters of this equation are the same as 3D formula-
tion, apart from the last exponent power term, which is used to
simulate the movement of the heat source. In other words, the heat
source is assumed to increase until t = to and then decreases. to is
an arbitrary time at which heat source reaches the 2D section
which is considered.
To simulate the combined thermal boundary condition, the to-
tal temperature-dependent heat transfer coefficient which ac-
counted for both convection and radiation [12] is used.
(
0:0668TðW=m2  CÞ When 6 T 6 500  C
h¼ ð13Þ
ð0:231T  82:1ÞðW=m2  CÞ When T P 500  C
þ p 2 ffi
; This boundary condition is employed for all free surfaces of the
2D and 3D models of the T-shape joints including the successive
ð10Þ boundaries created for the newly added elements during the
welding.
þ p 2 ffi
; 3.2. Mechanical analysis
ð11Þ
The finite element model employed for the mechanical analysis
is similar to the thermal analysis except for the type of elements
and the boundary conditions. The mechanical analysis is carried
out using the temperature histories computed by the thermal anal-
ysis as the input data.
During the welding process, because solid-state phase transfor-
mation does not occur in the stainless base metal and the weld me-
tal [11], the total strain rate can be decomposed into three
components as follows:
e_ ¼ e_ e þ e_ p þ e_ th ð14Þ
The components in the above equation correspond to elastic
strain, plastic strain and thermal strain, respectively. Since during
welding, material experiences high temperature for a very short
period of time, creep strain is neglected [30]. The elastic behavior
is modeled using the isotropic Hook’s rule with temperature-
dependent Young’s modulus and Poisson’s ratio as shown in Ta-
ble 4. The thermal strain is considered using the temperature-
dependent coefficient of thermal expansion. For the plastic strain,
a rate-independent plastic model is employed with the following
features: the Von Mises yield surface, temperature-dependent
mechanical properties, and linear isotropic hardening model.
4. Model updating
In order to obtain the coefficient of the formulation in the form
of Eqs. (10) and (11), initial values have been chosen and a set of
h i
2p v 2 ðtt02Þ
2
numerical solution have been carried out and the temperature
e ðcÞ
and deformation of the critical points have been compared with
those obtained from experimental measurements. Then, an optimi-
ð12Þ
zation computer code has been developed to modify the initial
(previous at each step-solution) values in order to reduce the dif-
ference between measured data and the calculated results. In this
code, the following concerns or restrictions are considered:
620 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623
Fig. 5. (a) Transformation of coordinate system and (b) Heat source parameters.
Fig. 6. Flowchart for model updating.
 The maximum weld pool temperature should not exceed
2500 °C and should not drop to lower than 1500 °C. This restric-
tion is based on experimental measurements and the vaporiza-
tion point of stainless steel [31].
 Calculated temperatures for all points should be within the
range of measured values. To increase the safety factor, the cal-
culated temperatures are preferred which are close or above the
average values of the experimental data.
 The geometry and penetration of the weld metal or fusion
boundary should be similar to the experimental observations.
This is carried out using micrographs as shown in Fig. 1b.
 Deformation which are calculated for the defined points should
be comparable with the test data.
The simple flowchart, which shows the major steps of the mod-
el updating, is shown in Fig. 6. The optimum values for the coeffi-
cients for the 2D and 3D simulating formulations are given in
Table 5.
In most previous models that are usually based on Goldak for-
mulation, there are number of unknown coefficients that should
be obtained based on experimental data. However, in this method,
most of these coefficients have been related to the physical or geo-
metrical parameters of the weldment such as the weld bead vol-
ume or weld leg size. Therefore, only one coefficient, k, has to be
obtained using the model updating method. The values of fr and
ff also should be obtained but they highly depend on the type of
Table 5
The value of parameters for the welding.
k g ff fr h c
0.9 0.75 1.4 0.6 45° 5 mm
welding technology. Therefore, if the type of welding changes from
MMAW, these coefficients should also be readjusted.
After obtaining the coefficients for the simulating formulation,
final FE solution has been carried out and results are given below.
5. Results and discussion
Fig. 7 shows the experimental thermal histories for six points on
the back surface of the flange plate during the welding of 6 mm-
thickness plates. A comparison of the calculated temperatures
and test data during the welding and cooling periods for point
No. 2 (shown in Fig. 2) is presented in Fig. 8. It can be seen that
the maximum temperature obtained using the 2D model is closer
to the average test data; however, the cooling rate for the 2D mod-
el is higher than the experimental measurements. In addition, this
figure shows that the maximum temperature calculated by the 3D
model is closer to the maximum temperature recorded in the test
and the cooling rate is closer to the experiments. The major con-
cerns in welding process are the temperature distribution and
cooling rate, the former shows the risk of burn-through, and, the
latter may lead to hot cracking. It can be seen that using 3D model
will be on the safe side and the calculated results are more realistic.
This is also reflected by the work of Shan et al. [32] who declared
that 3D FE models using moving heat source techniques show good
agreement with experimental measurements, and analysis using
element birth techniques, which closely simulate the welding pro-
cedure, are more accurate. Barsoum and Lundbäck [33] have also
shown that the differences in the results for 2D and 3D models
are principally based on the fact that the former is using infinite
welding speed and the latter employs a moving heat source. In
the present work, this shortcoming has been mostly overcome by
introducing the last term in Eq. (12), which is used to simulate
the movement of the heat source. However, in many research stud-
ies, the difference between the results obtained using 2D and 3D
models have been disregarded due to the reduced computation
time [34,35].
To study the effect of plate thickness, the same procedure (com-
paring the experimental measurements and the results of FE anal-
yses) has been carried out for the welding of plates with 5 and
8 mm of thickness. Since the trend of the temperature variation
for all case studies is the same, only those for 6 mm-thickness
plates is presented in the figures, to avoid any confusion. The max-
pffiffiffi pffiffiffi pffiffiffi pffiffiffi
a ¼ c 2=2 b¼c 2 cf ¼ c 2 cr ¼ 2c 2
pffiffiffi pffiffiffi pffiffiffi pffiffiffi
5 2=2 mm 5 2 mm 5 2 mm 10 2 mm
 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623 621

800
CH2
CH3 CH1 CH2
700
CH1
600
CH3 CH4

Temperature (C)
500

400 CH5 CH6

CH6
300
CH5
200
CH4
100

0
0 20 40 60 80 100 120 140 160 180 200
Time (s)

Fig. 7. Experimental results for thermal histories-6 mm plate weldment. Welding direction and location of the probes are shown in the inset figure (according to the
experimental test as shown in Fig. 2).

800

700 FEM 3D
Experimental Data
600
Experimental
Temperature (C)

500 (average)

400

300

200
Experimental (average)
100 FEM 2D
FEM 2D
FEM 3D
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130
Time (s)

Fig. 8. Experimental and numerical results for thermal histories at the point No. 2 – 6 mm plate weldment. Experimental data are shown by point marks.

Table 6
The comparison of maximum temperature at the point No. 2 in experiment and numerical analysis.

Plate thickness (mm) Welding speed (mm/s) Maximum temperature at the point (°C) Error (%)
2D FEM 3D FEM Experiment (average) 2D FEM 3D FEM (%)
6 2.34 644 725 603 6% 16
6-Fine mesh 2.34 – 747 603 – 19
5 2.34 815 893 733 10% 18
5 4.69 – 747 611 – 18
8 2.34 428 500 448 4% 10

imum-recorded temperature for point No. 2 is compared with the
calculated values obtained from 2D and 3D FE analyses for all case
studies; and, the results with the error percentage are presented in
Table 6. It can be seen that the coefficients of simulating formula-
tion have been obtained so that they can be used for relatively
wide range of plate thicknesses with good accuracy. The results
in Table 6 depict that by increasing the weld speed from 2.34 to
4.69 mm/s in 5 mm plate, the temperature of point No. 2 increases
20%. Also, by increasing the plate thickness to 8 mm, the tempera-
ture of the same point reduces by 27%. The same trend is also ob-
served by Deng et al. [18] who showed that by increasing the
thickness of shipbuilding steel SM400A plates in a T weldment
from 9 mm to 12 mm, the maximum temperature of the same
point decreases by 28%. This is also highlighted by Gery et al.
[36] and Long et al. [37] showing that the welding speed and plate
thickness have considerable effects on welding characteristics.
To study the mesh sensitivity of the FE solutions, the size of ele-
ments have been reduced until the convergence in the results has
been obtained. For example, the results for the FE mesh with
143,695 elements and 167,486 nodes for plates with 6 mm thick-
ness are given in Table 6. For this mesh, the minimum element is
0.5 mm  0.5 mm  1.0 mm. The second row of this table gives
the results for a finer mesh with minimum element
0.25 mm  0.25 mm  0.5 mm and 1,088,147 number of nodes. It
622 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623

3.0

2.5 Experimental
(average)

Displacement (mm)
2.0
FEM 3D

1.5
Experimental Data

1.0

0.5 Experimental (average)

FEM 3D
The point
0.0
0 10 20 30 40 50 60 70 80 90 100 110 120
Time (s)

Fig. 9. Experimental and numerical results for displacement histories at the point-6 mm plate weldment.

can be seen that the calculated temperatures obtained using these

two FE meshes differ only by 3% while the number of nodes has in- Table 7
The comparison of maximum displacement at the edge point in experiment and
creased 6.5 times.
numerical analysis.
In order to investigate the reliability of the simulation in pre-
dicting the mechanical behavior of the weldment, the deforma- Plate thickness Speed Maximum displacement at the edge Error
(mm) (mm/s) point (mm) (%)
tions of some critical points have been measured during the
welding and cooling periods. These data were also compared with 3D FEM Experiment (average)

those that have been obtained from the FE based numerical analy- 6 2.34 2.4 2.66 9
sis. Since in the FE analysis, stresses and strains are obtained from 5 2.34 2.36 2.53 6
5 4.69 2.75 2.53 8
resulting displacement values, the good agreement of the displace-
8 2.34 2.4 2.43 1
ments with the test data ensures the accuracy of the simulation.
The displacement at the edge of the flange plate for 3D FE
mechanical analysis is plotted in Fig. 9. This figure also compares
the displacement of the same point measured in the experiments
with those obtained from 3D FE solution. The same pattern for
model is incapable of producing correct results in mechanical solu-
deformation is also observed by Deng et al. [18] for T shape fillet
tion. On the other hand, the developed 3D model can predict ther-
weld of shipbuilding steel SM400A plates. As it can be seen in
mo-mechanical behavior with acceptable accuracy.
Fig. 9, the results from the 3D FE solution are in good agreement
Also a set of coefficients for the simulating formulation have
with the experimental data. It has to be highlighted that the dis-
been obtained using the model updating method and this approach
placement predicted by 2D model is completely in contradiction
can be used for other materials provided that a sufficient set of
with the experimental observations. In practice, the edge of the
experimental data were available. The major advantage of this for-
flange plate moves upward, towards the web plate, and this is ob-
mulation is that the number of unknown coefficients has been re-
served in the results of 3D model; whereas, 2D model shows ex-
duced to only one coefficient. For this purpose, other coefficients
actly the opposite trend. This verification shows that although 2D
have been related to the physical or geometrical parameters that
model can be used in thermal analysis, but it leads to incorrect dis-
are known for each weldment.
placement and stress–strain results. Fig. 9 also shows that most of
the distortion of the plate is happening during the welding process,
which lasts 32 s, and after this period, it becomes stable. In Table 7, Acknowledgements
the maximum displacements at the edge point of the flange plate
obtained from 3D FE analysis are compared with the experimental The authors would like to express their gratitude for the techni-
measurements and good agreement is observed. cal and financial support provided by the Tabriz Petrochemical
Company and for the help of R&D center of this plant for contribu-
tion in carrying out the experimental work, infield observations
6. Conclusion and technical data.

A robust and reliable formulation has been developed in this re-

search work that can be used to simulate the welding process and
to predict the thermo-mechanical behavior of the T-shape weld-
ments. This model is developed based on the experimental data
and its formulation has been obtained using optimum model
updating method. This model is used to simulate the welding of
two stainless steel 304 plates with different thicknesses. The re-
sults of numerical solution have been compared with test data
and good agreements have been observed. The results show that
2D model can only be used in thermal analysis provided that a
new term, which simulates the movement of the heat source, has
been added to the formulation. It has to be highlighted that 2D
References
[1] Mellor BG, Rainey RCT, Kirk NE. The static strength of end and T fillet weld
connections. J Mater Des 1999;20:193–205.
[2] Sabapathy PN, Wahab MA, Painter MJ. Numerical models of in-service welding
of gas pipelines. J Mater Process Technol 2001;118:14–21.
[3] Wang R, Zhang J, Serizawa H, Murakawa H. Study of welding inherent
deformations in thin plates based on finite element analysis using interactive
substructure method. J Mater Des 2009;30:3474–81.
[4] Shanmugam NS, Buvanashekaran G, Sankaranarayanasamy K, Kumar SR. A
transient finite element simulation of the temperature and bead profiles of T-
joint laser welds. J Mater Des 2010;31:4528–42.
[5] Oddy AS, McDill JMJ. Burn-through prediction in pipeline welding. Int J Fract
1999;97:249–61.
 F. Vakili-Tahami, A. Ziaei-Asl / Materials and Design 47 (2013) 615–623
[6] Hibbitt HD, Marcal PV. A numerical, thermo-mechanical model for the welding
and subsequent loading of a fabricated structure. J Comput Struct
1973;3:1145–74.
[7] Deng D. FEM prediction of welding residual stress and distortion in carbon
steel considering phase transformation effects. J Mater Des
2009;30(2):359–66.
[8] Kong X, Yang Q, Li B, Rothwell G, English R, Ren XJ. Numerical study of
strengths of spot-welded joints of steel. J Mater Des 2008;29(8):1554–61.
[9] Lee CH, Chang KH. Three-dimensional finite element simulation of residual
stresses in circumferential welds of steel pipe including pipe diameter effects. J
Mater Sci Eng A 2008;487(1–2):210–8.
[10] Murugan N, Narayanan R. Finite element simulation of residual stresses and
their measurement by contour method. J Mater Des 2009;30(6):2067–71.
[11] Deng D, Murakawa D. Numerical simulation of temperature field and residual
stress in multi-pass welds in stainless steel pipe and comparison with
experimental measurements. J Comput Mater Sci 2006;37(3):269–77.
[12] Brickstad B, Josefson BL. A parametric study of residual stresses in multi-pass
butt-welded stainless steel pipes. Int J Press Vessels Pip 1998;75(1):11–25.
[13] Deng D, Murakawa H. Prediction of welding residual stress in multi-pass butt-
welded modified 9Cr–1Mo steel pipe considering phase transformation
effects. J Comput Mater Sci 2006;37:209–19.
[14] Tsirkas SA, Papanikos P, Kermanidis T. Numerical simulation of the laser
welding process in butt-joint specimens. J Mater Process Technol
2003;134(1):59–69.
[15] Verhaeghe G. Predictive formulate for weld distortion—a critical
review. England: Woodhead Publishing; 1999.
[16] Lee CH, Chang KH. Temperature fields and residual stress distributions in
dissimilar steel butt welds between carbon and stainless steels. J Appl Therm
Eng 2012;45–46:33–41.
[17] Teng TL, Fung CP, Chang PH, Yang WC. Analysis of residual stresses and
distortions in T-joint fillet welds. J Pressure Ves Pip 2001;78(8):523–38.
[18] Deng D, Liang W, Murakawa H. Determination of welding deformation in fillet-
welded joint by means of numerical simulation and comparison with
experimental measurements. J Mater Process Technol 2007;183(2–3):219–25.
[19] Barsoum Z, Lundback A. Simplified FE welding simulation of fillet welds – 3D
effects on the formation residual stresses. J Eng Fail Anal 2009;16(7):2281–9.
[20] Smith MC, Bouchard PJ, Turski M, Edwards L, Dennis RJ. Accurate prediction of
residual stress in stainless steel welds. J Comput Mater Sci 2012;54:312–28.
[21] Goldak J, Asadi M, Alena RG. Why power per unit length of weld does not
characterize a weld? J Comput Mater Sci 2010;48(2):390–401.
623
[22] Goldak J, Chakravarti A, Bibby M. A new finite element model for heat sources.
J Metall Trans B 1984;15B:299–305.
[23] Özyürek D. An effect of weld current and weld atmosphere on the resistance
spot weldability of 304L austenitic stainless steel. J Mater Des
2008;29(3):597–603.
[24] Specification for stainless steel electrodes for shielded metal arc welding.
Document number: ANSI/AWS A5.4/A5.4M:2012. American Welding Society;
2012.
[25] Cui Y, Lundin CD. Austenite-preferential corrosion attack in 316 austenitic
stainless steel weld metals. J Mater Des 2007;28(1):324–8.
[26] Siewert TA. Predicting the toughness of SMA austenitic stainless welds at 77 K.
J Weld 1986;65(3):23–8.
[27] McCowan CN, Siewert TA, Reed RP, Lake FB. Manganese and nitrogen in
stainless steel SMA welds for cryogenic service. J Weld 1987;66:84–92.
[28] Lee DJ, Jung KH, Sung JH, Kim YH, Lee KH, Park JU, et al. Pitting corrosion
behavior on crack property in AISI 304L weld metals with varying Cr/Ni
equivalent ratio. J Mater Des 2009;30:3269–73.
[29] Wu CS. Welding thermal processes and weld pool behaviors. London: Taylor &
Francis Group; 2010.
[30] Choi J, Mazumder J. Numerical and experimental analysis for solidification and
residual stress in the GMAW process for AISI 304 stainless steel. J Mater Sci
2002;37:2143–58.
[31] Thorslund T, Kahlen FJ, Kar A. Temperatures, pressures and stresses during
laser shock processing. J Optics Lasers Eng 2003;39(1):51–71.
[32] Shan XY, Tan MJ, O’Dowd NP. Developing a realistic FE analysis method for the
welding of a NET single-bead-on-plate test specimen. J Mater Process Technol
2007;192–193:497–503.
[33] Barsouma Z, Lundbäck A. Simplified FE welding simulation of fillet welds – 3D
effects on the formation residual stresses. J Eng Fail Anal 2009;16:2281–9.
[34] Heinze C, Schwenk C, Rethmeier M. Numerical calculation of residual stress
development of multi-pass gas metal arc welding under high restraint
conditions. J Mater Des 2012;35:201–9.
[35] Heinze C, Schwenk C, Rethmeier M. Numerical calculation of residual stress
development of multi-pass gas metal arc welding. J Construct Steel Res
2012;72:12–9.
[36] Gery D, Long H, Maropoulos P. Effects of welding speed, energy input and heat
source distribution on temperature variations in butt joint welding. J Mater
Process Technol 2005;167:393–401.
[37] Long H, Gery D, Carlier A, Maropoulos PG. Prediction of welding distortion in
butt joint of thin plates. J Mater Des 2009;30:4126–35.