See discussions, stats, and author profiles for this publication at: [Link]

net/publication/271960054

Genetic algorithm based optimization of the process parameters for gas metal
arc welding of AISI 904 L stainless steel

Article in Journal of Mechanical Science and Technology · August 2013

DOI: 10.1007/s12206-013-0631-8

CITATIONS READS

27 1,619

3 authors:

Paulraj Sathiya Ajith P M

National Institute of Technology Tiruchirappalli College of Engineering Trivandrum
148 PUBLICATIONS 2,905 CITATIONS 14 PUBLICATIONS 189 CITATIONS

SEE PROFILE SEE PROFILE

Soundararajan .R
Sri Krishna College of Engineering and Technology
105 PUBLICATIONS 821 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Shrinkage characteristics of cast Aluminium alloys View project

Welding Distortions View project

All content following this page was uploaded by Soundararajan .R on 10 March 2020.

The user has requested enhancement of the downloaded file.

 Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465
[Link]/content/1738-494x
DOI 10.1007/s12206-013-0631-8

Genetic algorithm based optimization of the process parameters for gas metal arc
welding of AISI 904 L stainless steel†
P. Sathiya1,*, P. M. Ajith2 and R. Soundararajan3
1
Department of production Engineering, National Institute of Technology Tiruchirappalli-620015, Tamilnadu, India
2
Department of Mechanical Engineering Rajiv Gandhi Institute of Technology, Kottayam, Kerala
3
Department of Mechanical Engineering, Sri Krishna College of Engineering and Technology, Coimbatore -641008, Tamilnadu, India

(Manuscript Received August 14, 2012; Revised January 22, 2013; Accepted February 19, 2013)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Abstract

The present study is focused on welding of super austenitic stainless steel sheet using gas metal arc welding process with AISI 904 L
super austenitic stainless steel with solid wire of 1.2 mm diameter. Based on the Box - Behnken design technique, the experiments are
carried out. The input parameters (gas flow rate, voltage, travel speed and wire feed rate) ranges are selected based on the filler wire
thickness and base material thickness and the corresponding output variables such as bead width (BW), bead height (BH) and depth of
penetration (DP) are measured using optical microscopy. Based on the experimental data, the mathematical models are developed as per
regression analysis using Design Expert 7.1 software. An attempt is made to minimize the bead width and bead height and maximize the
depth of penetration using genetic algorithm.

Keywords: Gas metal arc welding; Box - Behnken design; Bead geometry; Optimization; Genetic algorithm
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

lated to arc welding process with a number of assumptions.

1. Introduction
However, it is not focused on theoretical studies for the weld
Super austenitic stainless steel 904 L is a highly-alloyed bead-geometry predictions. Super austenitic stainless steels
austenitic low carbon stainless steel with a fully austenitic are particularly interesting because they bridge between rela-
structure. Due to its high molybdenum content and specially tively cheap austenitic stainless steel and expensive Nickel
designed welding consumables with low impurity level, hot base super alloys, when high corrosion properties are required
crack formation during welding can be avoided despite the at moderately high temperatures Wallen et al. [2], Heino et al.
fully austenitic filler metal. This type of steels cannot be hard- [3], investigated the welding of the super austenitic stainless
ened by heat treatment as they are normally supplied in steel Avesta 654 SMO wires, Ames et al. [4] compared the
quench annealed condition. Major industrial applications of austenitic (316 L), super austenitic (254 SMO) and super du-
904L super austenitic stainless steel are in production and in plex (SAF 2507) weld properties produced using GTAW filler
pipe work required for general paper and allied industries, sea wire. The microstructure, mechanical properties and corrosion
water cooling equipment’s, oil and refinery components and resistance of such welds are compared to autogenous welds
in transport of sulfuric acid, sea water, condensers and heat produced in the current industry. In all cases, the welds pro-
exchangers. The quality of the weld may be dependent on a duced with GTAW filler wire exhibited equivalent or im-
number of input process parameters namely welding speed, proved microstructure and properties when compared to the
voltage, gas flow rate and wire feed rate. Several methods autogenous non-flux welds. Depending on the amount of cold
have been developed by various investigators to predict bead work present, the strength and hardness in the HAZ and FZ
geometry in welding. These methods include theoretical stud- softened due to the weld thermal cycle. Microhardeness trans-
ies, statistical analysis and others, some of which are stated verse across welds also examined. Kim et al. [5] have em-
below. Rosenthal studied the temperature distributions on an ployed factorial design to correlate the robotic GMAW proc-
infinite sheet due to a moving point heat source by consider- ess parameters (welding voltage, welding speed and arc cur-
ing the conduction mode of heat transfer [1]. It could be re- rent) to three responses (bead width, bead height and penetra-
*
Corresponding author. Tel.: +91 431 2503510, Fax.: +91 431 2500133
tion) for optimization purposes. The material was used as a
E-mail address: psathiya @ [Link] plates of AS 1204 mild steel adopting the bead-on-plate tech-
†
Recommended by Associate Editor Young Whan Park nique. Electrode wire with a diameter of 1.2 mm with the
© KSME & Springer 2013
2458 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465

same mechanical and physical properties of the base metal overcome this problem, the response surface methodology
was used. Their results showed that all process parameters uses the near-optimal values as a reference point to obtain a
influenced the responses and the models developed are able to model of the welding process and determine optimal values of
predict the responses with 0-25% accuracy. the process variables. Vidut Dey et al. [10] conducted the bead
The bead geometry that characterizes the quality of the weld -on-plate welds on austenitic stainless steel plates using GMA
is dependent on a number of input process parameters. These welding machine. Experimental data were collected as per
parameters are closely coupled in such a way that it is difficult Box - Behnken design and regression analysis was conducted
to identify the extent of contribution of these factors toward to establish input-output relationships of the process. An at-
the desired output. An expert welder from his experience of tempt was made to minimize the weldment area, after satisfy-
trial and error selects a set of parameters that may yield good ing the condition of maximum bead penetration. Thus, it was
results. However, the obtained results may not be the optimal posed as a constrained optimization problem and it was solved
one. The trial and error of the welder can be avoided, if a suit- by utilizing a genetic algorithm with a penalty function ap-
able mathematical model is developed, which can forecast the proach. The genetic algorithm is able to determine optimal
output from a set of desired parameters or vice versa. A math- weld-bead geometry and recommend the necessary process
ematical model can be made to solve the above problem using parameters for the same.
differential equations depicting the actual physical phenomena. Considering the above available literature, it seems only a
Welding is a process comprising of a number of complicated very limited literature is available on welding and its parame-
natural phenomena, none of which may be fully understood. ter optimization of super austenitic stainless steel. In the pre-
Thus, it may not be always possible to develop an appropriate sent investigation, the bead on plate welding trials are carried
differential equation of the said process. In such situations, out using a gas metal arc welding process and the bead pro-
models are made from the outcomes of experiments per- files i.e., output variables BW, BH and DP of the welds are
formed as per some statistical designs and then analyzed by measured using optical microscopy. These output variables
regression methods to predict the required output. The regres- are determined according to the variables, which are the volt-
sion equations can be either linear or non-linear. Yang and age, gas flow rate, travel speed and wire feed rate. Based on
Chandel [6], performed both linear as well as non-linear re- the input and output parameters the welding parameters are
gression analysis to model submerged arc welding process. optimized using genetic algorithms.
Yang et al. [7] observed that non-linear regression equations
are generally used to model welding phenomena but it, during
2. Genetic algorithm
modeling of submerged arc welding process that linear regres-
sion equations were found equally suitable. The above statisti- Genetic algorithm, introduced by Holland (1975), is a popu-
cal regression analysis yielded more or less satisfactory results, lation-based search and optimization tool. The GA works
while predicting the response from the process parameters. It equally well both in continuous or discrete search space. It is a
is to be mentioned that the statistical methods are mainly heuristic technique inspired by the natural biological evolu-
global in nature, that is, the usual practice is to establish a tionary process comprising of selection, crossover, mutation,
single working relationship between the input and the output etc. The evolution starts with a population of randomly gener-
for the entire domain of interest, as a result of which, it might ated individuals in first generation. In each generation, the
be possible to predict the results accurately at the anchor fitness of every individual in the population is evaluated,
points only (that is, the points used to carry out the regression compared with the best value, and modified (recombined and
analysis). If the search space is large and the objective func- possibly randomly mutated), if required, to form a new popu-
tions become highly complicated, then the computational time lation.
of a GA increases drastically and it is difficult to get solution The new population is then used in the next iteration of the
in real time. To overcome this difficulty, Kumar and Debroy algorithm. The algorithm terminates, when either a maximum
[8] showed that multiple sets of welding variables capable of number of generations has been produced or a satisfactory
producing the target weld geometry could be determined in a fitness level has been reached for the population. The fitness
realistic time frame by coupling a real-coded GA with a neural function of a GA is defined first. Thereafter, the GA proceeds
network model for gas metal arc fillet welding. Kim et al. [9] to initialize a population of solutions randomly and then im-
exploited the above mentioned benefit of regression analysis. proves it through repetitive application of selection, crossover
They used genetic algorithm and response surface methodol- and mutation operators. This generational process is repeated
ogy simultaneously, in order to find a set of welding process until a termination condition is reached.
variables that could produce the desired weld-bead geometry The major aim of this study is to develop a genetic algo-
in GMAW. A genetic algorithm does not require the objective rithm model for predicting optimum bead profiles: to mini-
function to be differentiable. It means that even if there are mize the bead height, minimize the bead width and maximize
some bad data in the search space, the model does not get the depth of penetration. Based on the constrained conditions
affected. However, this algorithm could not produce a mathe- the parameters are optimized by genetic algorithm.
matical model between the input and output variables. To The genetic algorithm is able to determine optimal weld-
 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465 2459

Table 1. Four factors Box-Behnken design.

Exp. Nos Factor 1 Factor 2 Factor 3 Factor 4

1 2 2 3 3
2 2 3 3 2
3 2 2 1 1
4 3 2 3 2
5 2 3 1 2
6 3 2 2 1
7 1 3 2 2
8 2 2 1 3
9 1 1 2 2
10 1 2 2 3
11 1 2 1 2
12 2 2 2 2
13 2 3 2 1
14 3 2 2 3
15 2 1 2 3
16 3 1 2 2
17 1 2 3 2
18 2 1 3 2
Fig. 1. Flow chart for GA.
19 1 2 2 1
20 2 2 2 2

bead geometry and recommend the necessary process parame- 21 3 3 2 2

ters for the same. The flow chart of the genetic algorithm is 22 2 2 3 1
presented in Fig. 1. The aim of this experimental work was to 23 2 2 2 2
investigate the effects of welding parameters, and to establish 24 2 1 2 1
a correlation between input and output parameters. In order to 25 3 2 1 2
this, gas flow rate, voltage, travel speed and wire feed rate 26 2 1 1 2
were chosen as process input parameters. The experimental
27 2 3 2 3
results were analyzed with analysis of variance (ANOVA),
which is used for identifying the significant factor analysis for
a significance level of α = 0.05, i.e. for a confidence level of
95%. 4. Mathematical formulation of the problem
The main aim of the present study is to determine the set of
optimal parameters of a GMA welding process to ensure
3. Experimentation and data collection
minimum bead width and height after satisfying the condition
The experiments are carried out on 904L super austenitic of maximum depth of penetration. Based on the given con-
stainless steel of size 100x40x5 mm3. The experiments are strained conditions the mathematical statement is formulated
performed with a GMAW process. The working ranges of the as below:
welding parameters like gas flow rate (GF), voltage (V), travel
speed (S) and wire feed rate (WF) are kept fixed to (12-16
Minimize weldment area =2/3(BH+DP) BW. (1)
lpm), (28-32 V), (90-110 mm/min) and (1.5-2 m/min) respec-
tively. The direct current electrode positive polarity and argon
gas is used as a shielding media. In this study Box Benken Subject to the condition that BP takes the maximum value
design method is used and presented in Table 1. After the and
welding is done the specimens are sectioned and polished with
suitable abrasive and diamond paste. After polishing, the
GFmin ≤ GF ≤ GFmax (2)
specimens are etched with electrolytic with oxalic acid to
Vmin ≤ V ≤ Vmax (3)
clearly reveal the fused metal zone. The etched samples are
WFmin ≤ WF ≤ WFmax (4)
measured for bead width (BW), height (BH) and depth of
Smin ≤ S ≤ Smax. (5)
penetration (DP) using optical microscopy. The measured
bead profiles values are presented in Table 2.
2460 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465

Table 2. Bead profile measurements.

Factor 1 Factor 2 Factor 3 Factor 4 Response 1 Response 2 Response 3

Runs Travel speed Voltage Wire feed rate Gas flow rate Bead height Bead width Depth of penetration
(mm/min) (Volts) (m/min) (lpm) (mm) (mm) (mm)
1 110 30 2 14 3.412 12.19 3.502
2 110 32 1.75 14 3.19 12.03 3.415
3 90 30 1.5 14 3.475 12.05 3.505
4 110 30 1.75 16 3.042 12.29 3.507
5 90 32 1.75 14 3.21 11.85 3.214
6 100 30 1.5 16 3.23 12.1 3.512
7 100 32 1.75 12 3.12 12 3.08
8 90 30 2 14 3.12 11.82 3.3
9 100 28 1.75 12 3.06 12.09 3.2
10 100 30 2 12 3.132 12.14 3.134
11 90 30 1.75 12 3.108 12.16 3.17
12 100 30 1.75 14 3.09 11.99 3.345
13 100 32 1.5 14 3.1 12.01 3.211
14 100 30 2 16 3.201 12.11 3.503
15 100 28 2 14 3.22 11.29 3.289
16 100 28 1.75 16 3.275 12.19 3.527
17 110 30 1.75 12 3.107 12.26 3.124
18 110 28 1.75 14 3.402 11.95 3.49
19 100 30 1.5 12 3.15 12.02 3.155
20 100 30 1.75 14 3.12 11.97 3.34
21 100 32 1.75 16 3.085 11.95 3.245
22 110 30 1.5 14 3.14 11.89 3.32
23 100 30 1.75 14 3.12 11.99 3.34
24 100 28 1.5 14 3.251 11.46 3.294
25 90 30 1.75 16 3.588 12.18 3.715
26 90 28 1.75 14 3.13 11.9 3.306
27 100 32 2 14 3.3 12.23 3.342

5. Results and discussion Expert 7.1 software.

The model F-value of 135.29 implies the model is
5.1 Mathematical modeling
significant. There is only a 0.01% chance that a "model F-
Regression analysis was carried out using Design Expert value" of this large could occur due to [Link] of "Prob >
7.1 software on the experimental data collected as per the 27 F" less than 0.0500 indicate model terms are significant. In
welding conditions given by Box - Behnken design (Table 1). this case A, B, D, AB, AC, BC, BD, CD, B2, C2, D2, A2C, AC2,
The experimental bead height (BH) was expressed in a coded B2C, B2D are significant model terms. Values greater than
form as a non-linear function of (V), travel speed (S) and wire 0.05000 indicate the model terms are not [Link] there
feed rate (WF), represented by A, B, C and D respectively. are many insignificant model terms (not counting those
required to support hierarchy), model reduction may improve
Regression model for bead height the [Link] "lack of fit F-value" of 0.59 implies the lack of
fit is not significant relative to the pure error. There is a
BH=-226.93812-4.49660A+17.15963B-0.015625AB+ 75.07% chance that a "lack of fit F-value" this large could
0.018763AC-0.11428BC-3.39450BD+0.062700CD- occur due to noise. Non-significant lack of fit is good, we
0.00338208C2+1.21667D2-0.00315000A2C+ want the model to fit. The model summary statistics of bead
0.00184375B2C + 0.058500B2D. (6) height is given in Table 4.
From Table 4, the "Pred. R-squared" of 0.9451 is in
ANOVA results for the response surface quadratic model reasonable agreement with the "Adj R-squared" of 0.9887.
are given in Table 3. The results were obtained using Design "Adeq precision" measures the signal to noise ratio. A ratio
 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465 2461

Table 3. ANOVA table of Bead height.

Source Sum of squares Df Mean square F value p-value Prob > F

Model 0.46802062 17 0.027530624 135.2887017 < 0.0001 Significant
A-gas flow rate 0.01353013 1 0.013530125 66.48861335 < 0.0001
B-voltage 0.00924075 1 0.00924075 45.41012399 < 0.0001
C-travel speed 0.00046225 1 0.00046225 2.271550449 0.1660
D-wire feed rate 0.0021125 1 0.0021125 10.38107155 0.0105
AB 0.015625 1 0.015625 76.7830736 < 0.0001
AC 0.07425625 1 0.07425625 364.903879 < 0.0001
BC 0.021316 1 0.021316 104.7493118 < 0.0001
BD 0.01334025 1 0.01334025 65.55554544 < 0.0001
CD 0.09828225 1 0.09828225 482.970447 < 0.0001
A2 5.0704E-05 1 5.07037E-05 0.249163918 0.6296
2
B 0.00404556 1 0.004045565 19.88037766 0.0016
C2 0.05351126 1 0.053511259 262.9605733 < 0.0001
A2C 0.031752 1 0.031752 156.0330338 < 0.0001
2
AC 0.01045838 1 0.010458375 51.39367535 < 0.0001
B2C 0.01087813 1 0.010878125 53.45637584 < 0.0001
B2D 0.009126 1 0.009126 44.8462291 < 0.0001
Residual 0.00183146 9 0.000203495
Lack of fit 0.00123146 7 0.000175923 0.58640873 0.7507 Not significant
Pure error 0.0006 2 0.0003
Cor total 0.46985207 26

Table 4. Model summary statistics of bead height. Regression model for bead width
Std. Dev. 0.01426518 R-squared 0.996102053 BW= 889.84708–27.64000B–11.23192C+0.19500BD
Mean 3.199185185 Adj. R-squared 0.988739265 +0.053000CD+1.75323A2–0.032396B2+0.055479C 2
C.V. % 0.445900425 Pred. R-squared 0.945134978 –1.35333D2–0.056875A2 B–0.00182500BC2. (7)
PRESS 0.025778444 Adeq. precision 46.87712903
The ANOVA results for bead width using response surface
quadratic model are given in Table 5.
From Table 5, the model F-value of 127.26 implies the
model is significant. There is only a 0.01% chance that a
"model F-value" of this large could occur due to noise. Values
of "Prob > F" less than 0.0500 indicate model terms are
significant. In this case B, C, D, AB, BC, BD, CD, A2, C2, D2,
A2B and BC2 , BC2 are significant model [Link] greater
than 0.0500 indicate the model terms are not significant. If
there are many insignificant model terms (not counting those
required to support hierarchy), model reduction may improve
the model. The "lack of fit F-value" of 5.83 implies the lack of
fit is not significant relative to the pure error. the "lack of fit F-
Fig. 2. Predicted bead height Vs Actual bead height. value" of 5.83 implies the lack of fit is not significant relative
to the pure error. There is a 15.50% chance that a "lack of fit
F-value" of this large could occur due to noise. Non-
greater than 4 is desirable. The value of Adeq precision is significant lack of fit is good. The model summary statistics of
46.877 that indicates an adequate signal. This model can be bead width is given Table 6.
used to navigate the design space. From Table 6, the "Pred R-squared" of 0.9461 is in
reasonable agreement with the "Adj R-squared" of 0.9865.
"Adeq precision" measures the signal to noise ratio. A ratio
2462 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465

Table 5. ANOVA table of bead width.

Sum of Mean F p-value

Source Squares df Square Value Prob > F
Model 1.259593519 15 0.083972901 127.2608836 < 0.0001 Significant
A-gas flow rate 0.001875 1 0.001875 2.841561424 0.1200
B-voltage 0.555025 1 0.555025 841.1400689 < 0.0001
C-travel speed 0.035208333 1 0.035208333 53.35820896 < 0.0001
D-wire feed rate 0.005208333 1 0.005208333 7.893226177 0.0170
AB 0.005625 1 0.005625 8.524684271 0.0139
AD 0.003025 1 0.003025 4.584385763 0.0555
BC 0.004225 1 0.004225 6.402985075 0.0280
BD 0.038025 1 0.038025 57.62686567 < 0.0001
CD 0.070225 1 0.070225 106.4259472 < 0.0001
A2 0.188334259 1 0.188334259 285.4204618 < 0.0001
B2 0.089556481 1 0.089556481 135.7227963 < 0.0001
C2 0.028356481 1 0.028356481 42.97423141 < 0.0001
D2 0.038156481 1 0.038156481 57.82612578 < 0.0001
A2B 0.41405 1 0.41405 627.4925373 < 0.0001
BC2 0.26645 1 0.26645 403.804822 < 0.0001
Residual 0.007258333 11 0.000659848
Lack of fit 0.006991667 9 0.000776852 5.826388889 0.1550 Not significant
Pure error 0.000266667 2 0.000133333
Cor total 1.266851852 26

Table 6. Model summary statistics of bead width. Regression model for depth of penetration
Std. Dev. 0.025687516 R-squared 0.994270574
DP = -58.52500-17.80410A+0.75412AB-0.00510000AC–
Mean 12.00407407 Adj R-squared 0.986457721
0.10354BC-2.21950BD+0.038700CD-0.20825B2+
C.V. % 0.213989983 Pred R-squared 0.946051725
0.00224014C2-0.011625A2B-0.00169375A2C-
Press 0.068344472 Adeq precision 50.10722913 0.00731250AB2 +0.000252500AC2+ 0.00230000B2C. (8)

The calculated ANOVA result for depth of penetration is

presented in Table 7.
From Table 7, The model F-value of 508.50 implies the
model is significant. There is only a 0.01% chance that a
"model F-value" this large could occur due to noise. Values of
"Prob > F" less than 0.0500 indicate model terms are
significant. In this case A, AB, AC, BD, CD, A2, B2, C2, A2B,
A2C, AB2, AC2, B2C, B2D, BC2 are significant model terms.
Values greater than 0.0500 indicate the model terms are not
significant. If there are many insignificant model terms (not
counting those required to support hierarchy), model reduction
Fig. 3. Predicted bead width vs actual bead width.
may improve the model. The "lack of fit F-value" of 10.54
implies there is a 8.89% chance that a "lack of fit F-value" of
greater than 4 is desirable. The value of Adeq Precision this large could occur due to noise. Lack of fit is bad -- we
50.107 indicates an adequate signal. This model can be used want the model to fit. This relatively low probablity (< 10%)
to Navigate the design space. is troubling. The model summary for depth of penetration is
The predicted response vs. actual response graph, shown in presented in Table 8.
Fig. 3 confirms the adequacy of the model to predict a very From Table 8, the "Pred R-squared" of 0.9574 is in
much close value at nearly all conditions. reasonable agreement with the "Adj R-squared" of 0.9973.
"Adeq precision" measures the signal to noise ratio. A ratio
 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465 2463

Table 7. ANOVA for depth of penetration.

Sum of Mean F p-value

Source Squares df Square Value Prob > F
Model 0.629300801 19 0.033121095 508.5049736 < 0.0001 Significant
A-gas flow rate 0.131769 1 0.131769 2023.036747 < 0.0001
B-voltage 0.000225 1 0.000225 3.454403298 0.1054
C-travel speed 7.225E-05 1 7.225E-05 1.109247281 0.3272
D-wire feed rate 0.000351125 1 0.000351125 5.390788258 0.0533
AB 0.006561 1 0.006561 100.7304002 < 0.0001
AC 0.006561 1 0.006561 100.7304002 < 0.0001
BC 7.225E-05 1 7.225E-05 1.109247281 0.3272
BD 0.004624 1 0.004624 70.991826 < 0.0001
CD 0.03744225 1 0.03744225 574.8472528 < 0.0001
A2 0.003239501 1 0.003239501 49.73574881 0.0002
2
B 0.018555574 1 0.018555574 284.881939 < 0.0001
C2 0.024395001 1 0.024395001 374.5340998 < 0.0001
A2B 0.017298 1 0.017298 265.5745256 < 0.0001
2
AC 0.009180125 1 0.009180125 140.9415737 < 0.0001
AB2 0.0068445 1 0.0068445 105.0829483 < 0.0001
AC2 0.0051005 1 0.0051005 78.30748454 < 0.0001
2
BC 0.016928 1 0.016928 259.8939512 < 0.0001
B2D 0.003876042 1 0.003876042 59.50849385 0.0001
BC2 0.002346125 1 0.002346125 36.01983083 0.0005
Residual 0.00045594 7 6.51343E-05
Lack of fit 0.000439273 5 8.78546E-05 10.54255556 0.0889 Not significant
Pure error 1.66667E-05 2 8.33333E-06
Cor total 0.629756741 26

Design-Expert® Software Predicted vs. Actual

Table 8. Model summary statistics for depth of penetration. Depth of Penetration
3.7 2

Color points by value of

Depth of Penetration:

Std. Dev. 0.00807058 R-squared 0.999276006 3.715

3.08 3.5 6

Mean 3.336481481 Adj R-squared 0.997310881

Predicted

C.V. % 0.241888945 Pred R-squared 0.957400664 3.4 0

Press 0.026827219 Adeq precision 91.90476224 3.2 4

3.0 8

3.0 8 3 .24 3 .40 3.5 6 3 .72

greater than 4 is desirable. The value of Adeq precision is Actual

91.905 indicates an adequate signal. This model can be used

Fig. 4. Predicted depth of penetration vs actual depth of penetration.
to Navigate the design space.
The predicted response vs. actual response graph, Fig. 4
confirms the adequacy of the model to predict a very much algorithm graph is presented in Fig. 5. From Fig. 5, the GA
close value at nearly in all conditions. could find the minimum area of Weldment as 47.2046 mm2.
The GA obtained results and best individual parameters ef-
5.2 Genetic algorithm (GA) fects are presented in Fig. 6. From Fig. 6 shows that the
number of variables Vs best individual variables, which
In this present study MATLAB tool box is used to obtain respresents best individual parameters of GA major role in the
the GA results. The optimal GA parameters are obtained and welding parameters are wire feed rate and followed by the
the parameters are found to yield the best results: N = 20; G = voltage, travel speed and gas flow rate. The relationship
30 Tolerance limit or Termination = 1*10-25 and Mutation is between the maximum constraint and generation is plotted in
Adaptive feasibility. A uniform crossover scheme is utilized the graph and presented in Fig. 7.
for the said purpose. The convergence plot of genetical Fig. 7 shows a straight-line graph and it means there is no
2464 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465

Table 9. Optimized working parameters and bead geometries.

Gas flow rate 13.3716 lpm

Voltage 28.0004 V
Travel speed 94.8679 mm/min
Wire feed rate 1.9999 m/min
Bead height 3.0400 mm
Bead width 11.4223 mm
Depth of penetration 3.1589 mm
Weldment area 47.2046 mm2

Fig. 5. Convergence plot of genetical algorithm.

Table 10. Comparative results of conformity test.

Optimized weld-
Experimen-
ing parameters
Sl. Nos Parameters tally ob- % of Error
with predicted
served values
values
Gas flow rate
1 13.3716 13 ----
(lpm)
2 Voltage (V) 28.0004 28 ----
Travel speed
3 94.8679 95 ----
(mm/min)
Wire feed rate
4 1.9999 2 ----
(m/min)
Fig. 6. Current best individual parameters of GA. Bead height
5 3.0400 2.981 1.94
(mm)
Bead width
6 11.4223 11.33 0.786
(mm)
Depth of pene-
7 3.1589 3.187 -0.889
tration (mm)
Weldment area
8 47.2046 47.00 0.433
(mm2)

Fig. 7. Constraint vs generation.

violation of the constrains in each generation. The optimal

values of process parameters and bead geometry values ob- (a) (b)
tained by the GA are presented in Table 9.
Fig. 8. Comparative bead profiles (a-predicted bead profile, b-
experimental bead profiles).
5.3 Confirmation tests

The optimized parameter is validated by conducting con- served that, percentage of error has less than 2% i.e., with in
firmatory welding trials. For confirmatory trials, new welding the acceptable range of percentage errors. So the optimized
process parameters, which have not been used for the prelimi- parameters has prone to give the good results i.e., to maximize
nary welding set of trials are considered for welding. The the depth of penetration and minimize the bead width and
welding trials are carried out and the bead profile is measured dead height. This may be due to high wire feed and lower
(Fig. 8) and area of the weld and the test results are compared voltage, as welding current is directly proportional to the wire
to the predicted values as presented in Table 8. feed rate. So the more amount heat is placed on the weld pool
The percentage of errors are calculated by using the Eq. (9) area, which implies the high depth of penetraction and less
and the values are presented in the same. Table 10. It is ob- bead width.
 P. Sathiya et al. / Journal of Mechanical Science and Technology 27 (8) (2013) 2457~2465 2465

6. Conclusions Nomenclature------------------------------------------------------------------------

In this study, bead-on-plate weld runs are performed at a GF : Gas flow rate lpm
GMAW setup. Experiments are carried out as per Box- V : Voltage volt
Behnken design and the data can be employed to develop S : Travel speed mm/min
mathematical models for predicting weld-bead geometry. The F : Wire feed rate m/min
possibility of a GMAW welding optimization procedure using FZ : Fusion zone
genetic algorithm is investigated in this work; more specifically, ANOVA : Analysis of variance
the determination of the near-optimal GMAW process parame- BH : Bead height
ters, welding voltage (V), wire feed speed (WF) travel speed DP : Bead penetration
(S) and gas flow rate (GF). The search for the optimum is BW : Bead width
based on the minimization of an objective function, which CCD : Central composite design
takes into account the geometric characteristics (depth penetra- G : Number of generations
tion, bead width and bead height) of the bead. It is found that GA : Genetic algorithm
the GA can be a powerful tool in experimental welding optimi- GMA : Gas metal arc
zation. The confirmation test is carried out and the predicted GMAW : Gas metal arc welding
results are very closer to the experimental results (error < 2%). MIG : Metal inert gas
N : Population size
Smax : Maximum travel speed of welding
References
Smin : Minimum travel speed of welding
[1] D. Rosenthal, Mathematical theory of heat distribution dur- TIG : Tungsten inert gas
ing welding and cutting, Weld. J., 20 (1941) 220-234. V : Voltage
[2] B. Wallen, M. Liljas and P. Stenvall, Proceedings of the Vmax : Maximum voltage
Conference on Applications of Stainless Steel, Stockholm Vmin : Minimum voltage
(1992) 9-11. WFmax : Maximum wire feed rate
[3] S. Heino, E. M. Knutson-Wedel and B. Karlsson, Mater. Sci. WFmin : Minimum wire feed rate
Technol., 15 (1999) 101-108. α : Probability
[4] N. Ames, M. Ramberg, M. Johnson and T. Johns, Stainless df : Degrees of freedom
Steel World, KCI Publishing BV (2001).
[5] Kim IS, Son KJ, YS Yang and Yaragada, Sensitivity analy-
sis for process parameters in GMA welding process using a
factorial design method, Int. J. Mach Tools Manuf., 43 P. Sathiya is currently an Associate
(2003) 763-9. Professor in Department of Production
[6] L. J. Yang and R. S. Chandel, An analysis of curvilinear Engineering, National Institute of Tech-
regression equations for modeling the submerged-arc weld- nology, Tiruchirappalli, Tamilnadu,
ing process, Journal of Materials Processing Technology 37, India. In 1994 he received his B.S. in
(1993) 601-611. Mechanical Engineering, Government
[7] L. J. Yang, M. J. Bibby and R. S. Chandel, Linear regression College of Engineering, Salem, Univer-
equations for modeling the submerged-arc welding process, sity of Madras, Tamilnadu, India. In
Journal of Materials Processing Technology, 39 (1993) 33-42. 1996 he completed his M.S. in Welding Engineering, Re-
[8] A. Kumar and T. Debroy, Tailoring fillet weld geometry gional Engineering College, Bharathidasan University, Ti-
using a genetic algorithm and a neural network trained with ruchirappalli, Tamil nadu, India. By 2006 he got his Doctorate
convective heat flow calculations, Welding Journal, 92 on Friction welding of similar stainless steels and Evaluation
(2007) 26-33. of processed joints, Bharathidasan University, Tiruchirappalli,
[9] D. Kim, S. Rhee and H. Park, Modeling and optimization of Tamilnadu, India. His research interests include welding tech-
a GMA welding process by genetic algorithm and response nology, solid state joining, materials behaviour subjected to
surface methodology, International Journal of Production welding, similar and dissimilar materials welding, failure
Research, 40 (2002) 1699-1711. analysis of weldments, modeling, simulation of welding proc-
[10] Vidyut Dey, Dilip Kumar Pratihar, G. L. Datta, M. N. Jha, esses and welding parameter optimization. He received young
T. K. Sahab and A. V. Bapat, Optimization of bead geome- technology award on 2009 from Indian Welding Society, In-
try in electron beam welding using a genetic algorithm, dia, and also received young scientist award from Department
Journal of Materials Processing Technology, 209 (2009) of Science and Technology, New Delhi, India. He Published
1151-1157. fifty five papers in international and national reputed journals.

View publication stats