Thermal Science and Engineering Progress 8 (2018) 327–339

Contents lists available at ScienceDirect

Thermal Science and Engineering Progress

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

Optimization of tensile strength in TIG welding using the Taguchi method T

and analysis of variance (ANOVA)
A. Balaram Naik , A. Chennakeshava Reddy
⁎

Department of Mechanical Engineering, Jawaharlal Nehru Technological University Hyderabad, Kukatpally, Telangana, India

ARTICLE INFO ABSTRACT

Keywords: The main criteria discussed in this paper concern the welding optimization parameters and tensile strength of
TIG welding duplex stainless steel 2205 by tungsten inert gas welding based on Taguchi method and analysis of variance.
Taguchi method Taguchi method of orthogonal L9 design experiment is carried out using orthogonal array for defining the
Analysis of variance (ANOVA) problem occur on welding process and to reduce the error occurred in the neural network for the prediction of
Orthogonal array
output. The neural network is a mathematical prediction model for the optimization process using back pro-
Design optimization
pagation algorithm. Analysis of variance (ANOVA) is a decision tool for detecting the variation of process
parameters, it is a statistical technique for find out the optimal level of factors for the verification of the optimal
design parameters through confirmation experiments. The purpose of this paper to increase the tensile strength,
hardness and depth of weld by varying the parameters such as current, time, speed, variation of oxide fluxes,
electrode diameter and gas flow rate. The Mat lab software is used for analyzing results and it shows that neural
network coupled with Taguchi method and Anova is an effective method for optimizing the weld quality of
material.

1. Introduction 1.1. Stainless steels

Metal joining plays a significant role in modern fabrication tech-
nology. Production of materials increased by day to day and competi-
tion of manufacturer is also increasing. Recent technologies in pro-
duction such as metal joining, metal welding, strengthening, material
quality and durability. Welding is the process of joining two metals by
fusing the base metal and by adding a filler material over the surface of
molten metal to form a strong bonding on metals. In tungsten arc
welding the tungsten electrode with constant weld power supply is used
to generate electric arc between the electrode and the workpiece which
create resultant heat to form the weld [1]. Welding is an efficient and
economic method for permanent joining of metals. Different number of
welding process are followed in manufacturing which are implemented
in short time. Welding process are differ from one another by usage and
type of equipment used. The tungsten inert gas arc welding is one of the
most popular welding process followed for the material production. In
tungsten inert gas arc welding an electric power supply will create arc
which melts base metal to form a molten pool [2]. The filler material is
manually added for TIG welding and the molten metal is allowed to
cool.
Stainless steels are mostly used in industries and in commercial
field. Stainless steels are used to produce automobile parts, vehicles and
in rail coach manufacturing. Stainless steels combine with composites
like nickel, chromium and magnesium to form alloys. Stainless steels
are possessing good corrosion resistance and low productivity cost.
Stainless steels are classified in to three types austenitic, ferritic and
martensitic. The austenitic stainless steel is face centered cubic crys-
talline in structure. The chemical composition consists of 0.15% carbon,
16% nickel and chromium to structure the austenitic. The ferritic
stainless steel has good mechanical properties than austenitic steels. It
has low corrosion resistance due to low composite mix of nickel and
chromium. It is a body centered cubic structure having a composition of
10.5–27% chromium with low amount of nickel content. Martensitic
stainless steel are extremely tough and highly machined to any shape
and sizes but it is not corrosion resistance as every stainless steels. It has
a composition of 12–14% chromium, 0.2–1% molybdenum, 2% nickel
and 0.1–1% carbon.
1.2. Tungsten inert gas arc welding (TIGAW)

In tungsten inert gas arc welding the arc is produced by the electric

⁎
Corresponding author.
E-mail address: abalaramnaik@jntuh.ac.in (A. Balaram Naik).

https://doi.org/10.1016/j.tsep.2018.08.005
Received 26 July 2017; Received in revised form 4 August 2018; Accepted 7 August 2018
2451-9049/ © 2018 Elsevier Ltd. All rights reserved.
A. Balaram Naik, A. Chennakeshava Reddy
supply which forms between the tungsten electrode and the base metal.
Though other welding process the electrode melt to form weld but in
gas arc welding point of base metal where weld is carried out is
transformed in to weld pool by the arc. The thin filler metal is fed
manually in to the pool where it melts. The tungsten is very sensitive to
oxygen in atmosphere, so it is recommended working with tungsten
need good shielding with oxygen free gas is required. During solidifi-
cation the inert gas provides a stable inert atmosphere for the protec-
tion of weld pool [5].
In TIG welding the fluxes are not used in weld because it form slag
inclusion and form corrosion on weld metals. Tungsten has high tem-
perature on melting and possess good electrical conductivity thus it
make the electrode non-usable [7]. The temperature of arc is typically
around 11,000 °F and similar shielding gases used are Ar, He, N or a
mixture of two. The tungsten inert gas welding is generally performed
on stainless steel, magnesium, alloys to aluminium, copper, brass,
nickel, titanium etc. However TIG welding is one of the slower method
in arc welding. The weld quality can affect the overall property of the
metal, thus an optimized method to find out the error and minimize to
least possible. Taguchi method and analysis of variance is used to op-
timize the weld metal property by finding errors and choosing best
value from its weld parameters. As a result the improved optimization
technique can improve the DSS 2205 tensile and mechanical property
by using Taguchi method and analysis of variance is discussed in the
paper. The rest of the paper includes related work at Section 2, pro-
posed methodology in Section 3, experimentation result, performance
analysis and discussion is detailed at Section 4 and conclusion part at
Section 5.
2. Related work
The welding is practiced in the production field for the past years.
Welding is a production process to improve productivity of higher
quality material. Welding process are practiced numerously which
could affect the weld quality, productivity and cost of manufacture of a
product. Many welding techniques are proved with various welding
metals such as titanium, aluminium and stainless steels Ericsson. M and
Sandstorm. R et al. [1]. Tungsten inert gas arc welding play an im-
portant role in acceptance of aluminium for quality and structural ap-
plication. TIG welding is most accurate mode of welding in stainless
steel and many researches are carried out experiments to optimize the
properties of the metal. The activated tungsten inert gas welding pro-
cess is used for the welding of duplex stainless steel alloy 2205 with the
welding parameters is discussed here.
Stainless steels is widely used in many in product fabrication such as
automotive, vehicles bodies and in commercial fields etc., due to its
high corrosion resistance, fire resistance and better weight ratio.
Stainless steel is generally a composition of metal alloys such as chro-
mium, carbon, nickel and molybdenum. Stainless steel has more
amount of chromium so it differs from carbon steels and carbon steels
rust easily when it is exposed to air and moisture Rizvi, Saadat Ali. S. P.
Tewari et al. [2]. Welding is a process of joining two metals by using
heat and electricity. Tungsten inert gas arc welding is a joining process
and it is most effective welding process in welding stainless steel Deva
Kumar. D, and D. B. Jabaraj et al. [3]. In TIG welding process the weld
arc is formed in between the workpiece and the electrode with an inert
gas atmosphere of argon or helium. For quality welding small intense
arc is formed by the electrode is used L.S. Kim, C.E. Park, and Y.K.
Jeong et al. [11].
The TIG welding is operated in either DC or AC, a constant current
power source is essential to avoid high current when electrode is short
circuited on the workpiece surface Ladislav Grad, Janez Grum et al. [4].
Welding arc is started by scratching the workpiece surface forms a short
circuit. These short circuit will break and main weld current will flow
Fude Wang, Stewart Williams et al. [13]. There is a risk the electrode
may stick to the surface and cause a tungsten inclusion in the weld, so it
328
Thermal Science and Engineering Progress 8 (2018) 327–339
can be minimized by using lift arc technique where short circuit current
formed at very low current level H.G. Fan, H.L. Tsai et al. [5]. Elec-
trodes for DC welding generally used are tungsten with 1–4% thoria to
improve for the formation of weld arc. Other alternate method for
improve weld are lanthanum oxide and cerium oxide are used for ob-
tain better performance on welding. The shielding gas used in TIG
welding are argon, argon with 2 to 5% H2 and helium and helium argon
mixtures J. Tusek, M. Suban et al. [6].
The TIG welding is applied for the top side infrared sensing tech-
nique for the closed loop control of weld penetration in welding S.
Mukhopadhyay, T.K. Pal et al. [10]. The heat transfer on electrode may
differ during welding and different current, workpieces sizes and elec-
trode tip angle are arranged in the experiments to validate current
adjustments for the heat variation and electrode wear A.K. Lakshmi-
narayanan, K. Shanmugam et al. [12]. The shielding gas composition is
used for the welding of high strength low alloy steels with flux cored arc
welding wires which can improve the quality of weld and properties of
the welded metals Hussain, Ahmed Khalid et al. [7]. Gas metal arc
welding is mainly used for welding ferrous and non-ferrous metals for
the increase of production and quality of welding metals. In industries
the finite element method is employed for the monitor robotic welding
to control the bad geometry of welding in real time. Metal properties of
welded metals of extra high strength steels has to be found out for the
optimization of tensile properties Balasubramanian. M, V. Jayabalan
et al. [17] Y.S. Tarng, H.L. Tsai, S.S. Yeh et al. [8].
The thermal cycle generates on the weld during cooling forms
strains in joints due to thermal expansion which cause tensile stress on
weld metals. Tensile failures generally occurred by the improper joints,
power supply and gas used for shielding H.E. Beardsley, Y.M. Zhang
et al. [9]. During welding cracks are obtained due to the overheating of
metal or loss of strength in metals. To find out the crack in welded
metal local stress approach is applied to predict the crack on the welded
joints P. Johan Singh, D.R.G. Achar et al. [15]. Welding current is an-
other factor affecting the welding lack of power supply can affect the
fusion and weld metals. The welding fusion need power supply from
either DC or AC power supply to create arc for the weld Nilanjan,
Suman Chakraborty et al. [16] A. Farzadi, S. Serajzadeh, and A.H. Ko-
kabi et al. [14]. Welding fusion zones on metals lead to mechanical and
metallurgical properties. The pulsed current can beneficiary to grain
refinement on weld zone and is controlled by pulse current and fre-
quency Kishore. K, P.V Gopal Krishna et al. [18].
Taguchi method is a powerful tool for improving productivity in
welding process. This method develop high quality products at low cost
at short time period. Taguchi method is also very important tool for the
robust design Anoop. C.A, Pawan Kumar et al. [19]. Taguchi method
uses orthogonal array for analyze the parameter with least experiments.
By Taguchi L9 orthogonal array nine experiments is carried out to ob-
tain output. The artificial neural network with Taguchi method is used
for the optimization where the weights are given by back propagation
algorithm for the optimization of welding parameters Kalaiselvan. K,
Elango. A et al. [22]. To find out the most significant factor analysis of
variance is applied and to classify weld quality in TIG welding Korra,
Nanda Naik, M. Vasudevan et al. [20]. Other methods like fuzzy pattern
recognition technique to classify the weld quality of metal in tungsten
inert gas welding Tarng. Y.S, S.S. Yeh, and S.C. Juang et al. [21].
3. Proposed methodology
Welding of stainless steels are followed in the production field very
earlier and many composite mixture metals are also introduced to
joining process. Defects are occurred when the weld metal does not met
the mechanical and metallurgical properties of the metal design.
Improper of design can affect the strength and load applied on it. The
flow of gas, current and fusion heat are some of the defects occurs on
the weld. During welding the difference of input current, gas flow rate,
speed and depth parameter will make the weld metal properties lower.
A. Balaram Naik, A. Chennakeshava Reddy
Tensile strength is one of the defects occur during welding. The tensile
strength is lower when the weld metal joint has poor loading capacity
and weld quality. If the weld of metal is high and toughness can possess
better tensile strength. In this paper the duplex stainless steel 2205 is
taken for the welding process. The duplex stainless steels are strong and
good weldable and possess good corrosion resistance to atmosphere and
chemicals. These steels are a composite mixture of chromium, mo-
lybdenum and nitrogen which forms corrosion resistance to certain
chemicals like chloride. Welding technique used for welding DSS 2205
is tungsten inert gas arc welding. The tungsten inert gas arc welding
uses consumable tungsten electrode for welding.
The welding area is protected from the atmosphere contamination
by shielding gases like argon or helium. The arc is generated by a
constant power supply by either AC or DC. The main objective on this
paper is to optimize the tensile strength and hardness on metal by
maximizing/minimizing the welding parameters like gas flow rate,
current, oxide fluxes etc. During welding the error obtained on metal is
analyzed by Taguchi method and artificial neural network. In the
welding process the artificial neural network coupled with Taguchi
method of orthogonal array is also used to find the optimized value in
neural network design. The Taguchi method optimize the error on
neural network by adjusting the network weights for the outputs. The
orthogonal L9 array is an evaluation design to evaluate the problem by
conducting tests. A set of nine experiments conducted to get an opti-
mized results. A decision tool analysis of variance (ANOVA) for detect
the variance in the process and it is a statistical technique for finding of
optimal results through confirmation experiments.
3.1. Data flow
In Fig. 1, a pictorial representation of the data flow for the DSS 2205
material optimization using TIG welding is shown. The diagram con-
sists of steps such as material acquisition, process input parameters
details, artificial neural network, outputs and the test validation step.
The optimization process is carried out using Taguchi method of or-
thogonal array.
4. Design of experiment and data collection
The design experiment starts with the analysis of welding of duplex
stainless steels 2205 by tungsten inert gas welding. After the welding
the base metal is subjected to testing where it forms irregularities on the
Fig. 1. Data flow for the Optimization of DSS 2205 in TIG welding.
329
Thermal Science and Engineering Progress 8 (2018) 327–339
weld like roughness, lower tensile strength. To validating the error
Taguchi method of orthogonal array is used to conduct the test for find
out the best result. Taguchi has derived a new method to perform test
on design of experiments. This method uses a set of orthogonal arrays
for conducting minimum number of experiments that give brief details
on factors affecting performance parameters. The mode of orthogonal
array is based on signal, noise and control factors. The experiment
conducted on duplex stainless steel grade 2205 of Avesta type of di-
mensions 200 × 50 × 3 mm cross section plates are used. The TIG
welding is done on base metal with filler material made of ER316L of
diameter 2.4 mm with ceramic nozzle of diameter 9.5 mm is used for
the experiment. During the welding process the constant arc is main-
tained at a gap of 3 mm and the electrode is placed perpendicular to
base metal. The joint of weld will be good if 5 mm plate with arc gap is
not exceed to 0.7 mm and 0.3 mm for 3 mm plates. The base metal is
made of duplex stainless steels of grade 2205 which has higher me-
chanical and chemical properties which offers the metal to good
weldability. This grade metal is generally used in production due to its
properties. The filler material or electrode is made of ER316L contains
2–3% molybdenum which can improve the pitting corrosion on welded
metals. The use of Molybdenum in filler can increase corrosion re-
sistance. These type of filler material is generally used in sheet metal
working corresponding to stainless steels base metals. Due to its high
corrosion resistance on atmospheric contaminants it is widely used for
weld high pressure piping and tubing.
4.1. Duplex steels
Duplex stainless steel are a composite mixture of austenitic and
ferritic. It has high amount of chromium, nitrogen and molybdenum.
These steels are high strength, high resistance to corrosion and possess
good weldability. Duplex stainless steel 2205 is a composite steel
mixture of austenite and ferrite and more resistance to stress corrosion.
These steels have higher amount of chromium, molybdenum and ni-
trogen possessing the crevice corrosion resistance in the presence of
chlorides. The SS2205 steels are mainly used in marine environments,
oil and gas extraction due to its high serviceability in environments
contain chemicals like chlorides and hydrogen. These steels work long
time on heat at a temperature of 300 °C and it extend up to 950 °C. After
the extend temperature the metal may embrittle and lower its corrosion
resistance.
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Table 1 voltage with high current using a transformer in the machine. The
Chemical Composition of Base metal and Filler metal. power supply needed for the AC welding is 80–110 V with 50–80 A
S. No. Chemical % Composition of base % Composition of filler constant current is established and power factor is kept low. In case of
Composition metal DSS2205 material ER316L DC welding power supply is generally 8–25 V with 50 A current and the
polarity is constant factor. DC welding generally has two types of
1 Chromium 22.821 18.5
polarities. They are straight polarity and reversed polarity. In straight
2 Nickel 5.75 11.5
3 Carbon 0.024 0.03
polarity the positive and negative terminals are connected to work
4 Silicon 0.457 0.45 material and electrode. In reversed polarity the terminals connected is
5 Magnesium 1.72 1.75 reversed to straight polarity. The heat is generated at positive pole
6 Phosphorus 0.019 0.03 which is one third at negative pole. Polarity is generated according to
7 Sulphur 0.017 0.03
requirement of heat at either poles.
8 Molybdenum 3.22 2.8
9 Copper 0.73 0.75
10 Iron Remaining Remaining 4.2.1.5. Welding electrodes. Electrodes are used in arc welding for
fusing workpiece by conducting current through the workpiece. The
electrode is either consumable or non-consumable based on the process
4.2. Chemical composition of base metal and filler metal and type of machine used. In case of tungsten inert gas arc welding non-
consumable electrode is used. Electrodes is classified in to coated
In Table 1, the chemical composition of the base material and the electrodes and bare electrodes. Bare electrodes are simple electrodes
filler material used is provided. The base material used is DS2205 and with no coating over them.
filler material used is ER316L. From the table it is clear that these two
materials are having common elements in their composition and the 4.2.1.6. Non-consumable welding electrodes. These electrodes are made
major difference in material composition is seen for elements such as of tungsten or carbon and during welding these electrode not melts due
chromium and nickel. to low depletion rate. The flux material is additionally needed for the
bare electrode and these electrodes are rarely used. Non-consumable
4.2.1. Step 1: Parameters influence on welding bare electrodes are used in gas shielded welding process such as MIG
4.2.1.1. Arc Welding. Electric arc welding is one of the type of fusion and TIG.
welding process followed in manufacturing. The weld is performed by
the metal is heated by an electric arc between an electrode and 4.2.1.7. Coated consumable welding electrodes. In arc welding coated
workpiece. The electric arc is produced where the electrode is made consumable electrodes are generally used and flux and filler material
to contact with the metal with a short distance which is equal to 2 mm. are not added additionally for the welding. These electrodes have a core
The machine operates at low voltage with high current which create arc of mild steel and flux is coated over the surface of the electrode. Coating
for weld by thermionic emission from electrode (cathode) to workpiece over the electrode reduce atmosphere, prevent oxidation and
(anode) Deva Kumar. D, and D. B. Jabaraj et al. [3]. The arc is constant removable slag from metal surfaces.
by the presence of thermally ionized column of gas. During welding the
arc produced at a temperature of 5500 °C or higher. By this high 4.2.1.8. Shielding gas. The shielding gas used in TIG welding for the
temperature weld a pool of molten metal (workpiece and filler metal) protection of welding area from the atmospheric gases which can
formed on the welding zone Ladislav Grad, Janez Grum et al. [4]. The defects during fusion of metals, porosity, weld metal ductility when
electrode is placed in perpendicular zig-zag motion. The molten metal they come contact with the arc or the welding metal. Shielding gas also
pool is solidified to form strong welded joint and the movement of transfers heat from electrode to the metal and helps to start a sub arc
electrode in weld is either manually or automatically by the type of [6]. Some of shielding gas used in TIG welding are
machine used.
4.2.1.9. Argon. It is most commonly used in a wide variety of metals
4.2.1.2. Selection of the welding process. Welding process is selected such as steels, stainless steels, aluminium and titanium.
based on the particular work carried out. Selection of welding is based
on the shape and size of component to be manufactured. Each welding 4.2.1.10. Argon with 2–5% H2. Argon is mixed with hydrogen to make
process is differ from its type of welding equipment, heat fusion and the gas slightly reducing to weld the metal without oxidation.
protection of weld. Welding equipments are available based on the Disadvantage is when weld metal hydrogen cracking risk to carbon
weld and material to be weld. For commercial usage welding machines steels is to forms porous in metal alloys like aluminium.
are powered by DC and AC power supply. Some of the common factors
governing the selection of welding process are availability of 4.2.1.11. Helium and Argon-Helium mixtures. Helium is added to argon
equipment, repetitiveness of the operation, quality of metal, type of to form high temperature arc which offers high welding speeds and
joint, time, cost etc. deeper weld penetration. The only disadvantage is these gas mixtures
are high in cost and difficult to start the arc.
4.2.1.3. Flux. Flux is used in weld to prevent oxidation of hot metal.
The flux converts the gases (oxides and nitrides) to slag that can be 4.2.2. Step 2: Orthogonal array
removed from welded metal easily. The formation of slag can make the In orthogonal array the independent variable has a combination of
weld weak thus flux material are used for removing slag from the metal. level setting and are equal number of times. The L9 orthogonal is used
Fluxes are different for different metal welding. For welding of metals for the testing of parameters in level 1, 2, 3 at variable 4. It is called
like copper and its alloys the sodium carbonates are used as flux. For balancing property of orthogonal array. The experiment is conducted
welding of aluminium and it alloys the fluxes like sodium chloride, with all these independent variables and experiment 1 cannot conduct
potassium, lithium and barium are used. with variable 1, level 2 and experiment 2 with variable 1 and level 1.
This is due to the array of each factor column are orthogonally to other
4.2.1.4. Power source. In welding AC and DC power supplies are used level of column values. The inner product of vector corresponds to
for the operation. For welding ferrous metals AC supply is used. For weight is zero and the above 3 levels are normalized between −1, 0, 1.
welding non-ferrous and mixture of alloys such as aluminium, stainless Therefore the weighing factor of inner product corresponds to in-
steel etc. Dc supply is used. The welding machine generally outputs low dependent variable is 1 and the independent variable 3 will be

330
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

( 1* 1 + 1*0 + 1*1) + (0*0 + 0*1 + 0* 1) + (1*0 + 1*1 + 1* 1) . Step 1: In Taguchi analysis the S/N ratio obtained have to normal-
=0 ized in range of 0–1
The parameters used for the experiment of orthogonal array is taken Larger the Better: In grey relational the S/N ratio is normalized to
from the below table. The table shows the TIG welding of DSS2205 is larger the better to find the larger optimum quality of the material. The
carried out by different oxide fluxes with varying time, current and grey relational is computed by taking the average of grey relational
electrode diameter. Minimum experiments to be conducted by Taguchi coefficient corresponding to every processes. The larger the better value
method can be calculated which is based on degree of freedom ap- in grey relational is found by the equation below.
proach
Yij Yij min
NV Xij =
NTaguchi = 1 + (Li 1)
Yij max Yij min (5)
(1)
i=1
where, Xij is the normalized S/N ratio, Yij is the S/N ratio analyzed by
where, Li is sum of orthogonal array used in the Taguchi method, NV is Taguchi, Yij max and Yij min are the maximum and minimum S/N ratio
number of generation carried out in Taguchi method. obtained.
Smaller the Better: The smaller the better value is the normalized S/N
4.2.3. Step 3: Signal to noise ratio ratio in the grey relational for obtaining smaller error in the quality
The quality of the product can be quantified in terms of response to standards. The grey relational analyzed for the prediction of small er-
noise factor and signal factor. The product quality can be improved by rors in quality by the following equation.
maximizing the signal to noise ratio for the respective product. The Yij max Yij
computation of signal to noise ratio can be done by Taguchi robust Xij =
design of experimental design module. The S/N ratio for any design
Yij max Yij min (6)
operation can be exewelded by resulting the value of design available in where, Xij is the normalized S/N ratio, Yij max and Yij min are the maximum
the experimental design module. and minimum S/N ratio obtained, Yij is the S/N ratio analyzed by
Taguchi.
4.2.3.1. Smaller the better type. The number of flaws in the welding is Step 2: The grey relational grade in the process analysis the values
measured as y variable and analyzed by S/N ratio. The effect of signal are obtained indicating relational degree between every sequence of
factor is zero and it shows the flaws on the welding is zero. The S/N operation. The grey relational grade of optimal quality is found by the
ratio is an expression of the assumed quadratic nature of the loss given equation below.
function and by increasing the ratio will increase the quality of the
metal. In this paper the S/N ratio to find the small least error in welded min + max
GCij =
metal is found out by the given equation below. ij + max (7)

S Yi2 where, GCij is the grey relational grade, is the assumed value 0.5 for
= 10 log the larger and smaller the better, max and min are the maximum and
N smaller n (2)
minimum absolute difference.
where, n is the number of trial test, Yi is the ith number of quality Step 3: By finding the average grey relational coefficients the grey
characteristics. relational grade is found out by the following equation below.

4.2.3.2. Bigger the better type. The quality of the product should be Gi = (1/ m) GCij (8)
satisfies and the S/N ratio of product quality should be greater as
where, GCij is the grey relational grade, m is the number of response
possible. Welding quality of product will be higher and there is no
variables, Gi is the grey relational grade.
adjustment of factor for the S/N ratio. If error occur it transformed in to
smaller the better type for the quality characteristics. The S/N ratio for
evaluating the higher quality is derived from the equation below. 4.2.4. Step 4: Optimization using artificial neural networks
The artificial neural networks is optimization tool for the optimi-
1
S Yi2
zation of input parameters with the hidden layers and desired output
= 10 log will be an optimized value. An artificial neural network is a mathe-
N Bigger n
(3) matical model for optimization based on biological neural networks. A
neural network consists of artificial neurons for the optimization of
where, n is the number of trial test, Yi is the ith number of quality
input parameters. The artificial neural networks are recently used for
characteristics.
the optimization of welding process using back propagation algorithm
approaches. The back propagation algorithm is a neural network al-
4.2.3.3. Nominal the best type. The S/N ratio is used when the quality of gorithm for the training of neural network for the optimization process.
the product is equalize with particular nominal value. The quality of The back propagation algorithm generally operates with multiple input
product is nearer to the specification as possible to ensure high quality. unit with several input variables are used. In this paper back propa-
The nominal best type for the quality equalize in welding can be gation algorithm based artificial neural network is used for the opti-
derived from the following equation [23]. mization of welding material properties.
The Fig. 2, describes the process of artificial neural network used in
_
2
S Y
= 10 log this paper to optimize the weld property of metal after the welding
N No min al s2 (4) process. The input parameters taken for the optimization are varying
where, Y_is the mean value quality characteristics, s is the sum of oxide fluxes, current for the welding operation, time taken for weld,
variance on the test value. electrode diameter. Each input node is connected with the hidden layer
for the processing and output is executed.
4.2.3.4. Grey relational analysis. Multiple performance on the system
can be analyzed using grey relational method. In this method the 4.2.5. Step 5: Mathematical modelling of proposed ANN
multiple performance parameters are converted in to single grey 4.2.5.1. Input Function. The input function for the artificial neural
relational grade. network is explained in equations below

331
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Fig. 2. Artificial Neural Network for Optimization of TIG Weld.

B1 = X1 W11I + X2 W21
I
+ ...........Xi WinI (9) N
o
Yk = Hj W jk
where, B1 is the input function in neural network design, X1 is the input j =1 (14)
value for the initialization of neural network. W11I , W21I
is the input
where, Yk is the actual output, Hj is the input transig function in neural
weight acting on each node for the adjusting the input data in the
network. W jko
is the total weights act on the output layer of neural
hidden layer for the optimization process.
network.
M
Bi = Xi WijI 4.2.7. Step 6: Analysis of variance (ANOVA)
i=1 (10)
Analysis of variance otherwise called ANOVA is a decision making
Bi is the total sum of input value in neural network design, M is the tool for analyzing the average difference in performance of test con-
number of hidden layer acting on neural network, i is the ith number of ducted. The anova test the variables by mean squaring and estimate the
inputs for network design processing. Xi is the total sum of inputs and experimental errors at specific levels. Analysis of variance generally
WijI is the sum of weights acts on hidden layers in the input nodes of used tool named F-test to analyze the particular design has any sig-
neural network for the optimization. nificant change in quality standards. In analyzing the parameters the F-
test tool ratio of mean square and residual error is used to find the
4.2.5.2. Transig function. This function is basically used for train the significance of a factor. The analysis of variance conducting the average
neural network for the optimization purpose. The transig function for of test is evaluated by using the equation below.
the optimization process is explained in equation below. n
1 SST = ( i nm ) 2
Hj = M i=1 (15)
1 + exp ( i=1
Xi WijI ) (11)
where, SST is total sum of squared deviation, nm is the total mean S/N
Hj is the transig function in neural network, Xi is the as the sum of ratio, n is number of experiments conducted in orthogonal array, i is
inputs in neural network, WijI is the total sum of weights in hidden layers the mean S/N ratio for the experiment.
in artificial neural network.
4.2.8. Step 7: Percentage in ANOVA
4.2.5.3. Output function. The output layer the weight is also applied for
The percent in analysis of variance is defined as the rate of quality
the output node for the adjusting the value to be optimized.
based on the significant rate of process parameters. The percentage
N number in the welding depends on the flux, current, time, gas flow rate
o
Bk = Hj W jk and electrode diameter. The percentage validation can be observed
i=1 (12)
from the Table 2 as the parameters listed above is machined in the
Bk is the output function in ANN, Hj is the transig function in artificial tungsten inert gas arc welding of duplex stainless steel 2205 grade
neural network, W jk
o
is the output weights act with hidden layer for the composites with filler material ER316L. The percentage of machined
output function. material can be found out by the given equation below.
ssd
4.2.6. Activation function P=
ssT (16)
4.2.6.1. Purelin function:. This function is generally a transfer function
for the calculation of hidden layer output value from the total input Thus the paper briefly shows the analytical method for the opti-
value which can be formulated from the given equation. mization of weld quality is improved by the varying the flux materials,
current, gas flow rate etc. The base metal made of DSS2205 and filler
Y = Bout × 1 (13)
metal of ER316L is made to TIG welding then the weld metal is sub-
where, Y is the net output calculated by purelin function, Bout is the jected to tensile testing. The metal forms irregularities on the weld
output value calculated from its input value. The actual output is quality, so to optimize property of weld metal Taguchi optimization
calculated by sum of input transig function and net weight act on the method of orthogonal array with artificial neural network and analysis
output layer which is explain in the equations below. of variance is carried out to improve quality and tensile strength of

332
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Table 2 section. The depth of weld recorded for TiO2 in the predicted part is
Experimental Data for TIG welding of DSS2205. 6.52 mm and the value obtained in the experimental section is 6 mm,
Experiment Oxide Flux Time Current (A) Electrode Diameter which is the maximum depth of weld.
(seconds) (mm) Table 4, displays the results obtained after the grey relational-or-
thogonal L9 array experimentation for SiO2. The results were calculated
1 SiO2 100 150 2
for a current input ranging from 100 to 250 A. The depth of weld was
2 SiO2 130 200 2.4
3 SiO2 160 250 3
maximum at 250 A and the minimum was obtained at 100 A respec-
4 TiO2 100 200 3 tively. From the table it can be summarized that as the current input
5 TiO2 130 250 2 increased the ultimate tensile strength and hardness increased pro-
6 TiO2 160 250 2.4 portionally.
7 Mixture 100 250 2.4
The Table 5, presents the results obtained after the grey relational-
8 Mixture 130 150 3
9 Mixture 160 200 2 orthogonal L9 array experimentation for TiO2. The results were con-
sidered for a current input vacillating from 100 to 250 A. The data from
the table shows that for TiO2 in the final two sets of analysis as the
metals. current input is increased to 250 A the hardness is seen to decrease. In
the case of ultimate tensile strength, the value is decreasing for the
5. Experimental data collection mixture.
The Table 6, presents the results obtained after the grey relational-
The experiment is carried out by duplex stainless steel 2205 of orthogonal L9 array experimentation for mixture. The results were
Avesta type with 200 × 50 × 3 mm cross section plates is used. The measured for a current input varying from 100 to 250 A. The hardness
electrode material is made of tungsten of diameter 2.4 mm and nozzle of the mixture is lower when compared to the first two set of data. The
of material ceramic with diameter of 9.5 mm. The arc formation is depth of weld increased from 4.68 to 5.92 mm and the UTS raised from
constant through the welding process maintaining an arc gap of 3 mm 706.115 to 774.369 KN/mm2.
and the electrode is placed perpendicular to the base plate. The welding In Fig. 4, optimal mean comparison of oxide mixtures and its per-
process is carried out by different oxide fluxes like SiO2, TiO2, mixture formance of mean depth, ultimate tensile strength and hardness is
of two and other input parameters like time, current and electrode elaborated in this given figure respectively.
diameter are used. By fluctuating the input parameters the weld quality
of material can be improved. 6. Performance analysis results and discussion
In Table 2 the experimental data regarding the TIG welding of
DSS2205 is provided, which includes the type of oxide flux used, time, The proposed system for the analysis is implemented in the working
current and also the diameter of the electrode used. The table also gives platform of MATLAB with the following system specification.
information about the variation in the experimental data for oxide
fluxes such as SiO2, TiO2 and the mixture. For each flux the current is Processor: Intel Core 2 Quad @ 2.5 GHz
applied in the range of 150–250 A, time of application is between 100 RAM: 3 GB
and 160 s and the electrode used are having diameters 2, 2.4 and 3 mm Operating system: windows 7
respectively. Mat lab version: R 2014a version 8.3

5.1. Grey relational analysis results
In Table 3, the experimental and ANN predicted values are tabu-
lated. The table provides data about the depth of weld, ultimate tensile
strength and hardness. From the table it can be seen that the maximum
depth of weld is 6 mm for experimented case where as it is 6.52 mm for
ANN predicted case. Maximum hardness is predicted for SiO2 at 250A
condition and the experimented hardness at 250 A for SiO2 showed the
maximum among the whole experimental data.
The Fig. 3, represents the graphs of experimented and predicted
results of grey relational analysis. The graph is plotted against the ul-
timate tensile strength (UTS), depth of weld and hardness of SiO2, TiO2
and the mixture. For SiO2 the UTS was maximum for 3 mm electrode
diameter and the value was found to be 988 KN/mm2 in the predicted
The optimization process is done by selecting the parameters in
welding such as current, gas flow rate, flux coating, electrode diameter,
time for welding etc. The duplex stainless steel 2205 is subjected to TIG
welding by varying oxide fluxes for better results. The filler material
and fluxes used for the analysis of weld are filler material made of
ER316L and fluxes used some are SiO2, TiO2 and mixture of both fluxes
SiO2 and TiO2. The titanium dioxide TiO2 and silicon dioxide SiO2 are
widely used flux has better result on performance while comparing with
other oxide fluxes. The test evaluation of the welding process with input
parameters and analysis of variance is elaborated in the following
sections. Experimentation of analysis of variance and orthogonal L9
array are elaborated in the appendix section respectively.
The Table 7, shows the test evaluation of ANOVA and the test is
conducted with three level of sets with input current 150 A, 200 A,

Table 3
Experimented and ANN Predicted Values.
S. No. Flux Time (s) Current(A) Electrode diameter (mm) Experimented ANN Predicted Error Grey Relational

Depth (mm) UTS Hardness Depth (mm) UTS (KN/mm2) Hardness % Coefficient Grade (Gi )

1 SiO2 100 150 2.0 4.78 715 92 4.9 745 93 0.016 0.266 0.369
2 SiO2 130 200 2.4 4.97 909 94 5.3 956 96 0.074 0.453 0.481
3 SiO2 160 250 3.0 5.50 920 95 6.01 988 97 0.097 0.516 0.519
4 TiO2 100 200 3.0 4.81 680 90 5.01 745 92 0.067 0.712 0.402
5 TiO2 130 250 2.0 6.00 788 91 6.52 821 93 0.031 0.453 0.428
6 TiO2 160 150 2.4 4.12 653 90 5.25 701 95 0.078 0.426 0.491
7 Mix 100 250 2.4 4.64 706 93 5.66 782 96 0.133 0.429 0.543
8 Mix 130 150 3.0 4.10 654 91 5.13 731 92 0.069 0.409 0.390
9 Mix 160 200 2.0 5.92 774 94 6.26 802 95 0.055 0.428 0.301

333
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Fig. 3. Experimented and Predicted Results.

Table 4 Table 6
Grey Relational-Orthogonal L9 Array Experimentation for SiO2. Grey Relational-Orthogonal L9 Array Experimentation for Mixture.
Test No Current (A) Time (s) Electrode Depth UTS (KN/ Hardness Test No Current (A) Time (s) Electrode Depth UTS (KN/ Hardness
Diameter (mm) mm2) Diameter (mm) mm2)
(mm) (mm)

1 100 100 2.0 4.78 715.874 81 1 100 100 2.0 4.68 706.115 93
2 200 130 2.4 4.80 720.445 84 2 200 130 2.4 4.72 715.425 92
3 250 160 3.0 4.82 734.684 87 3 250 160 3.0 4.92 725.994 94
4 100 100 2.0 4.86 781.546 90 4 100 100 2.0 5.04 736.874 89
5 200 130 2.4 4.88 794.654 93 5 200 130 2.4 5.10 740.985 94
6 250 160 3.0 4.97 825.445 96 6 250 160 3.0 5.25 751.961 97
7 100 100 2.0 5.15 910.741 94 7 100 100 2.0 5.36 762.985 92
8 200 130 2.4 5.28 914.118 96 8 200 130 2.4 5.74 768.415 95
9 250 160 3.0 5.50 920.984 97 9 250 160 3.0 5.92 774.369 94

Table 5 7. ANOVA analysis of welding parameters

Grey Relational-Orthogonal L9 Array Experimentation for TiO2.
Test No Current (A) Time (s) Electrode Depth UTS (KN/ Hardness
Welding process parameters can be investigated by ANOVA to verify
Diameter (mm) mm2) the parameters that significantly affected the quality characteristic. We
(mm) conducted the ANOVA analysis for a standard DSS 2205 grade material
and the properties were obtained as follows
1 100 100 2.0 4.81 680.745 90
2 200 130 2.4 4.86 674.148 92
In Tables 8–10, the Anova result of depth, UTS and hardness is
3 250 160 3.0 4.88 618.684 94 analyzed and its efficiency measures are tabulated in the given section
4 100 100 2.0 4.90 673.126 89 respectively. In this result tabulation the P-value, sum of squares, means
5 200 130 2.4 4.92 678.891 94 square, F-statistic values are also noted.
6 250 160 3.0 5.20 682.746 97
Table 11, describes the ANOVA analysis result of a standard spe-
7 100 100 2.0 5.45 689.951 92
8 200 130 2.4 5.75 675.963 95 cimen and the results regarding the tensile strength, yield strength,
9 250 160 3.0 5.90 653.984 90 elongation and hardness is tabulated. From the table it can be identified
that ANOVA results are much similar with the grey relational analysis
results.
250 A with metal diameter 2 mm, 2.4 mm, 3 mm respectively. During The quality of the weld product can be quantified in terms of noise
test each level increase of parameters like penetration depth, weld pool factor and signal factor is defined in the Fig. 5 above. In this graph the
geometry and some levels have no effect. By conducting the test can quality of weld is quantified by taking the mean of S/N ratio to the
improve the mechanical properties like tensile strength, hardness, welding parameters such as flux, current, time and electrode diameter.
ductility and quality of weld etc. The S/N ratio is taken for oxide flux material used in welding process.
The result shows the titanium dioxide flux has −2.87 signal to noise
ratio with the time induced of 100 s for the processing. The amount of

334
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Fig. 4. Comparison of Test and Predicted Results.

Table 7 Brinell hardness tester. The flux materials of SiO2, TiO2 and mixture of
Test Evaluation of ANOVA. two is taken for the reference. The hardness of SiO2 flux material has
S. No Flux Time (s) Current (A) Electrode Diameter (mm) S/N Ratio
higher hardness value of 93.7 and TiO2 has 90.1 and mixture of 92.8 in
the hardness tester. The hardness of weld material can be improved by
1 SiO2 100 150 2 −2.93 the varying the time, current and electrode diameter of the weld ma-
2 TiO2 130 200 2.4 −2.87 terial. The parameters of current, time and electrode diameter range
3 Mixture 160 250 3 −2.89
higher of 92.8, 93.1, 92.3 respectively. The Brinell hardness number is
calculated using the equation below
feed current has S/N ratio of −2.9 with 200 A output current and the 2F
electrode diameter of 2 mm has higher S/N ratio for the optimum BHN =
D (D D2 d2 ) (17)
quality of weld.
The mean of means is the total mean for the response in the ex- where, F is force applied on the weld metal, D is the indenter diameter
periments are considered. The Fig. 6 shows the mean for the welding and d is the indentation diameter.
parameters such as fluxes, time, current, and electrode diameter is The above Fig. 8 shows the ultimate tensile strength of weld metal
taken. The mean value has a range of 350–450. The mean for the ti- using welding parameters such as flux material, time, current, electrode
tanium dioxide flux has lower in the range closer to 350 and higher for diameter is taken for the tensile testing. The testing is carried out in
silicon dioxide of range 425. The time mean for the welding process has universal testing machine. The oxide fluxes forms a major role in tensile
higher of 400 with time of 125 s. The mean for the weld current induced strength of weld material. The SiO2 flux has higher tensile strength of
on the process has higher of 410 with maximum current of 250 A and 820 N/mm2 with varying time 100 s, input current closer to 150 A and
the electrode diameter with higher mean of 387 with diameter of 2 mm. electrode diameter of 2 mm.
The Fig. 7 shows that the hardness of weld material is carried in the The depth of penetration of weld pool in tungsten inert gas arc

Table 8
ANOVA Result of Depth.
Source Degree of freedom (DF) Sum of squares (SS) Mean Square (MS) F-Stat P-Value

SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture

B/W Groups 3 74532.4 74509.8 74544.4 24844.1 24836.6 24848.1 14.75 14.7 14.7 0.00216 0.0013 0.0013
Within Groups 8 13467.1 13467.1 13467.2 1683.3 1683.3 1683.4
Total 11 87999.5 87976.9 88011.6

335
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Table 9
ANOVA Result of UTS.
Source Degree OF FREEDOM (DF) Sum of squares (SS) Mean square (MS) F-Stat P-Value

SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture

B/W Groups 4 1,069,435 876785.3 10459.9 26,735 219,196 261,478 195.70 138.6 191.5 0 0 0
Within Groups 10 1366 15805.8 13,647 1366 1580.5 1364.7
Total 14 1,083,096 892591.1 105,956

Table 10
ANOVA Result of Hardness.
Source Degree OF freedom (DF) Sum of squares (SS) Mean square (MS) F-Stat P-Value

SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture SiO2 TiO2 Mixture

B/W Groups 4 7456 74846.1 74941.1 18,641 18711.5 18735.2 13.824 13.885 13.909 0.0004 0.0004 0.0004
Within Groups 10 13,485 13475.1 13469.2 13458.5 1347.5 1346.9
Total 14 88053.1 88321.2 88410.3

Table 11 regression equations. Results of the neural network-based approaches

ANOVA Analysis Results for The Hardness of TIG welded joints.
Grade Tensile Strength Yield Strength Elongation (%)
(MPa) (MPa)
DSS 2205 621 448 25
welding is described in the Fig. 9. Generally the weld pool in TIG
welding ranges from 2.5 mm to 3.5 mm. In this paper the SiO2 flux
usage in the welding forms the depth of penetration ranging from
5.5 mm to 6 mm with welding time, current and electrode diameter are
100 s, 150 A and 2 mm respectively.
The depth of penetration for time, current, and electrode diameter
are ranged from 4 mm to 5.5 mm respectively.
During the learning of neural networks, a batch mode of training
had been adopted. As the set of 32 input-output data (with the help of
which the regression analysis had been carried out) might not be suf-
ficient to provide proper training to the network, one thousand input-
output data had been generated at random by using the above
developed to model the input–output relationships in TIG welding
HRC process are stated and explained below. The performance of a back-
propagation neural network (BPNN) depends on the quality and
quantity of data used in training. It is also dependent on its architecture,
31
connecting weights, learning rate, momentum constant, coefficients of
transfer functions (TFs), bias value. To determine an optimal set of the
above parameters, a study was carried out, by varying one parameter at
a time and keeping the other parameters unaltered. In the first stage,
the number of neurons to be lying in the hidden layer was varied in the
range of 5–30, keeping the other parameters, viz. learning rate η, mo-
mentum constant α, coefficient of transfer function of the input neurons
a1, coefficient of transfer function of the hidden neurons a2, coefficient
of transfer function of the output neurons a3, and bias value fixed to
0.5, 0.5, 1.0, 1.0, 1.0 and 0.0005, respectively. It is interesting to note
that the NN with 22 neurons lying in its hidden layer showed the best
performance in terms of mean squared deviation in prediction. Thus, in
the second stage and on-wards, the number of hidden neurons was kept
fixed to 22. In the similar way, the optimal/near-optimal values of η, α,
a1, a2, a3 and bias value were determined in stages. As the parameters

Fig. 5. Mean S/N Ratio vs Inputs.

336
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Fig. 6. Mean of Means vs Inputs.

Fig. 7. Mean Hardness vs Inputs.

were determined in stages (but not simultaneously), there is no guar-
antee that the obtained network will be globally optimal but it could be
a near-optimal one.
The Table 12, demonstrates the parameters of the optimal network,
which includes number of neurons in the hidden layer, learning rate,
momentum constant, coefficient of TF of the input neurons, coefficient
of TF of the hidden neurons, coefficient of TF of the output neurons and
bias value.
8. Conclusions
The optimization of ultimate tensile strength, hardness and depth of
weld are the main objectives discussed in this paper. Defects are mainly
occurred due to the improper fusion of weld form on the metal surface.
Welding of DSS 2205 is carried out by tungsten inert gas arc welding by
the filler material made of ER316L is discussed in this paper. After the
weld the metal is subjected to tensile testing by universal testing ma-
chine where the metal brittle by the poor quality of weld. It is mainly
occur due to current, gas flow, fusion time and oxide fluxes. To opti-
mize the tensile strength of weld metal Taguchi method of orthogonal
array and analysis of variance is carried out to optimize the tensile
strength and weld quality of the base metal. The final step is analysis
which is undergone by Mat Lab where the tested parameter of tensile
testing and welding is taken for the reference of analysis and result are
plotted in simulated plot graphs. For the forthcoming of welding the
base metal and filler can be replaced with good composition grade and

337
A. Balaram Naik, A. Chennakeshava Reddy Thermal Science and Engineering Progress 8 (2018) 327–339

Fig. 8. Mean of UTS vs Inputs.

Fig. 9. Mean Depth vs Inputs.

Table 12 standards in coating over the flux material and base metal. Thus the
The Parameters of Optimal Network. recent technology in metal joining process discussed on this paper can
No. of neurons in the hidden layer 22 improve the hardness and by increasing the tensile strength of metal. By
comparing recent research on welding the neural network and Taguchi
Learning rate, η 0.35 method is an effective method in optimization of welding process.
Momentum constant, α 0.40
Coefficient of TF of the input neurons 0.15
Coefficient of TF of the hidden neurons 1.6
Coefficient of TF of the output neurons 0.90 10. Conflict of interest statement
Bias value 0.0005
We authors A. Balaram Naik and A. Chennakeshava Reddy stated
that there is no conflict of interest between us in the preparation and
welding equipment is upgraded for the better surface finish of the the submission of manuscript.
metal. The experiment on metal welding can be continued by opti-
mizing the corrosion formation on the weld area by optimizing of better

338
A. Balaram Naik, A. Chennakeshava Reddy
References
[1] M. Ericsson, R. Sandstorm, Influence of welding speed on the fatigue of friction stir
welds, and comparison with MIG and TIG, Int. J. Fatigue. 25 (12) (2003)
1379–1387.
[2] S. Rizvi, S.P. Ali, W. Tewari Ali, Weldability of steels and its alloys under different
conditions - a review, Int. J. Sci. Eng. Technol. Res. 2 (3) (2013) 539.
[3] D.D. Kumar, D.B. Jabaraj, Research on gas tungsten arc welding of stainless steel-an
overview, Int. J. Sci. Eng. Res. 5 (1) (2014) 1612.
[4] L. Grad, J. Grum, I. Polajnar, J.M. Slabe, Feasibility study of acoustic signals for on-
line monitoring in short circuit gas metal arc welding, Int. J. Mach. Tools Manuf. 44
(5) (2004) 555–561.
[5] H.G. Fan, H.L. Tsai, S.J. Na, Heat transfer and fluid flow in a partially or fully
penetrated weld pool in gas tungsten arc welding, Int. J. Heat Mass Transf. 44 (2)
(2001) 417–428.
[6] J. Tusek, M. Suban, Experimental research of the effect of hydrogen in argon as a
shielding gas in arc welding of high alloy stainless steel, Int. J. Hydr. Energy 25 (4)
(2000) 369–376.
[7] H.A. Khalid, A. Lateef, M. Javed, T. Pramesh, Influence of welding speed on tensile
strength of welded joint in TIG welding process, Int. J. Appl. Eng. Res. 1 (3) (2010)
518.
[8] Y.S. Tarng, H.L. Tsai, S.S. Yeh, Modelling, optimization and classification of weld
quality in tungsten inert gas welding, Int. J. Mach. Tools Manuf. 39 (9) (1999)
1427–1438.
[9] H.E. Beardsley, Y.M. Zhang, R. Kovacevic, Infrared sensing of full penetration state
in gas tungsten arc welding, Int. J. Mach. Tools Manuf. 34 (8) (1994) 1079–1090.
[10] S. Mukhopadhyay, T.K. Pal, Effect of shielding gas mixture on gas metal arc welding
of HSLA steel using solid and flux-cored wires, Int. J. Adv. Manuf. Technol. 29 (3–4)
(2006) 262–268.
[11] L.S. Kim, C.E. Park, Y.K. Jeong, J.S. Son, Development of an intelligent system for
selection of the process variables in gas metal arc welding process, Int. J. Adv.
Manuf. Technol. 18 (2) (2001) 98–102.
[12] A.K. Lakshminarayanan, K. Shanmugam, V. Balasubramanian, Effect of autogenous
339
Thermal Science and Engineering Progress 8 (2018) 327–339
arc welding processes on tensile and impact properties of ferritic stainless steel
joints, Int. J. Iron Steel Res. 16 (1) (2009) 62–116.
[13] F. Wang, S. Williams, M. Rush, Morphology investigation on direct current pulsed
gas tungsten arc welded additive layer manufactured Ti6Al4V alloy, Int. J. Adv.
Manuf. Technol. 57 (5–8) (2011) 597–603.
[14] A. Farzadi, S. Serajzadeh, A.H. Kokabi, Modelling of heat transfer and fluid flow
during gas tungsten arc welding of commercial pure aluminium, Int. J. Adv. Manuf.
Technol. 38 (3–4) (2008) 258–267.
[15] P.J. Singh, D.R.G. Achar, B. Guha, H. Nordberg, Fatigue life prediction of gas
tungsten arc welded AISI 304L cruciform joints with different LOP sizes, Int. J.
Fatigue 25 (1) (2003) 1–7.
[16] Nilanjan Chakraborty, S. Chakraborty, P. Dutta, Modelling of turbulent transport in
arc welding pools, Int. J. Numer. Meth. Heat Fluid Flow 13 (1) (2003) 7–30.
[17] M. Balasubramanian, V. Jayabalan, V. Balasubramanian, Optimizing pulsed current
parameters to minimize corrosion rate in gas tungsten arc welded titanium alloy,
Int. J. Adv. Manuf. Technol. 39 (5–6) (2008) 474–481.
[18] K. Kishore, P.V. Gopal Krishna, K. Veladri, G.K. Kumar, Analysis of defects in gas
shielded arc welding of AA 6351 using Taguchi methods, Int. J. Appl. Eng. Res. 5
(3) (2010) 393–399.
[19] C.A. Anoop, P. Kumar, Application of Taguchi methods and ANOVA in GTAW
process parameters optimization for aluminium alloy 7039, Int. J. Eng. Innov.
Technol. (IJEIT) 2 (2013) 11.
[20] K.N. Naik, M. Vasudevan, K.R. Balasubramanian, Multi-objective optimization of
activated tungsten inert gas welding of duplex stainless steel using response surface
methodology, Int. J. Adv. Manuf. Technol. 77 (1–4) (2015) 67–81.
[21] Y.S. Tarng, S.S. Yeh, S.C. Juang, Fuzzy pattern recognition of tungsten inert gas
weld quality, Int. J. Adv. Manuf. Technol. 13 (6) (1997) 387–392.
[22] K. Kalaiselvan, A. Elango, N.M. Nagarajan, Artificial neural network application on
Ti-Al joint using laser beam welding – a review, Int. J. Mech., Aerospace, Industrial,
Mechatronic Manuf. Eng. 8 (8) (2014) 1477–1481.
[23] Paulraj Sathiya, S. Aravindan, A.N. Haq, Optimization for friction welding para-
meters with multiple performance characteristics, Int. J. Mech. Mater. Des. 3 (4)
(2006) 309–318.