 Open Access
 Total Downloads : 483
 Authors : Uttam Roy, Satyabrata Podder
 Paper ID : IJERTV1IS9353
 Volume & Issue : Volume 01, Issue 09 (November 2012)
 Published (First Online): 29112012
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Neuro Genetic Optimization of Weld Metal Deposition in MAG Welding Process Using Genetic Algorithm and Adaptive Neuro Fuzzy Interference System
Vol. 1 Issue 9, November 2012
Uttam Roy1, Satyabrata Podder2
1 National Institute of Technology AP (Yupia), Arunachal Pradesh, India
2 Brainware Group of Institutions, Barasat, Kolkata, India
Abstract
This work uses Genetic Algorithm to predict the weld metal deposition in the Metal Active Gas (MAG) welding process for a given set of welding parameters; with the help of Adaptive Neuro Fuzzy Interference System (ANFIS), an artificial intelligent technique to make the inputoutput model a standard one for given set of data into its input range. Experiments are designed according to full factorial design of experiments and its experimental results are used to develop an ANFIS model. Multiple sets of data from experiments are utilized to train, test and validate the intelligent network. The trained network is used to predict the amount of weld metal deposition. The proposed ANFIS, developed using MATLAB functions, is flexible, and it scopes for a better online monitoring system. Genetic algorithm is used to optimize the predicted results which is validated with experimental results and found satisfactory. The optimization of the model shows the correct input range to achieve the optimized weld metal deposition for the desired result.
Keywords:
Genetic Algorithm, ANFIS, MAG, Weld Metal Deposition.

INTRODUCTION
Metal Active Gas (MAG) welding is a widely used industrial arc welding procedure, which needs prediction and continuous monitoring of its input and output parameters to control the process in a better way to produce consistent weld quality. MAG involves mechanicalmetallurgical features of the weldment, which depend upon the weld bead geometry, weld metal deposition, penetration and reinforcement of weld bead, wetting and fusion angle etc. These are directly related to combination and values of input process parameters of welding. Literature shows that a work has been explored since the last six decades on various aspects of modeling, simulation and process optimization in Metal Inert Gas (MIG) welding. Researchers are attempting many techniques to establish relationships between welding input parameters and weld metal deposition and weld quality leading to an optimal process by the application of different techniques like Genetic Algorithm (GA) and Adaptive Neuro Fuzzy Interference System (ANFIS) which is a rule base of fuzzy logic controller (FLC). After the development of the concept of Fuzzy Logic by Lofti Zadeh [1] [2] [3] Mamdani et al. [4] and Sugeno et al. [5] [6] extended the concept of fuzzy logic to the FLC. The theory and concept of ANFIS was developed by J.S.R. Jang [7] with an engineering application using Artificial Neural Network [8][9]. Goldberg [10] [11] explained the search technique of GA to optimize the result with a concept of local and global hybrids. Ishibuchi et al. [12] explained that how the nos. of fuzzy rules can be made minimized with the help of GA. Nozaki et al. [13] extended the concept of Ishibuchi and showed how a set of numerical data can generate fuzzy rule by heuristic method. Davi et al. [14] made a Comparison between genetic algorithms and response surface methodology in Gas Metal Arc Welding (GMAW) welding optimization. Juang and Tarng [15] adopted a modified Taguchi method to analyze the effect of each welding process parameter on the weld pool geometry and then to determine the Tungsten Inert Gas (TIG) welding process parameters combination associated with the optimal weld pool geometry. Manonmani
[16] investigated the effect of the welding parameters on the bead geometry AISI 304 stainless steel. S. Datta et al. [17] have worked on the influence of electrode stick out as an one of the important process parameters of submerged arc welding by incorporating one of the traditional methods of statistical data analysis (ANOVA). Jagdev Singh and Simranpreet Singh Gill [18] has designed and demonstrated the use of fuzzy logic based multi input and single output ANFIS model to predict the tensile strength of tubular joints, welded by the technique of radial friction welding. Manoj Singla et al. [19] have optimized the different parameters of Gas Metal Arc Welding process by using factorial design approach. The study had optimized various GMAW parameters including welding voltage, welding current, welding speed and nozzle to plate distance (NPD) by developing a mathematical model for sound weld deposit area of a mild steel specimen. P. Kumari et al. [20] has made a study on the effect of welding parameters on weld bead geometry in MIG welding of low carbon steel. J Raja Dhas and S Kannan [21] have adopted a neuro hybrid model to predict bead width in submerged arc welding. A. Biswas et al. [22] has optimized the bead geometry in Submerged Arc Welding which was conducted based on Taguchis L25 orthogonal array design with combinations of process control parameters.Different bead geometry parameters was optimized and optimal result has been verified by confirmatVoorly. .1 TIsshuies 9s,tNuodvyember 2012
proposes a hybrid intelligent technique, ANFIS, to predict weld metal deposition in a MAG welding process for a given set of welding parameters and optimization of the same using genetic algorithm to have an corrected result.

METAL ACTIVE GAS (MAG) WELDING
MAG, a common arc welding process has welding current, arc voltage, welding speed, electrode stick out (extension), electrode diameter, polarity, current type etc. as input variables. Welding current directly influences the weld metal deposition which gives better depth of penetration and base metal fusion. At a given current, weld metal deposition is affected by the electrode diameter. Since the weld is more brittle than the parent material, it is vital that the weld metal deposition must be minimal without disturbing desired penetration and strength. Minimization of the weld metal deposition is necessary because excessive deposited weld metal leads to wastage of the welding electrode and the process consumes more time. Therefore sufficient attention is required to select the process parameters in welding to get a minimized weld metal deposition with having desired weld quality as required.

PROPOSED METHODOLOGY

Data Acquisition
Full factorial design of experiments is a systematic application of design of experiments to improve the product quality which uses the all possible combinations of levels of the input factors to make a meticulous investigation of the nature of the output. A four factors three levels design of experiments was done where (3)4 = 81 numbers of experiments were involved in the MAG (POWERMIG T400) welding machine (Fig: 01). The experiment was conducted at M/s. Hind Engineering, Badu, Madhyamgram, West Bengal. Single pass butt welding is performed on the commercially available steel of IS2962 grade (C 0.25%, Si 0.20%, Mn 0.75% and balance Fe) on a pair of 100mm Ã— 100mm Ã— 5mm work piece. Before welding required edge preparation was done. Electrode (Dia 1.2 mm) (AWS/SFA 5.18: ER 70S6) was used with CO2 gas at 11 lit/min flow rate as shielding gas. The weights were recorded before and after welding to measure the amount of weld metal deposition on the base metal (Table I).
Fig: 01. PowerMIG T400 model (Make: Powercon Electric Company)

Development of ANFIS for weld metal deposition prediction
ANFIS is a fuzzy interference system which uses the framework of Neural Network. Thistechnique provides a method for fuzzy modeling procedure to learn information about a data set in order to achieve a rule base for selection of fuzzy rules. A database defines the membership functions used in the rules which creates a reasoning mechanism to carryout interference procedure on the rules and the given fact. This methodology combines the advantages of fuzzy system and Neural Network. The modeling of weld metal deposition by Metal Active Gas welding is done by considering four input parameters and one output parameter. The membership functions parameters are tuned using a hybrid system which is the combination of back propagation and the method of least squares. The parameters associated with the membership functions will change through the learning process. The computation of these parameters is facilitated by gradient vector, which provides a measure of how well fuzzy inference system is modeling the input/output data for a given set of parameters. Once the gradient vector is obtained, any of the several optimization routines could be applied in order to adjust the parameters so as to reduce some error measure. The proposed ANFIS
TABLE I
EXPERIMENTAL RESULTS FOR WELD METAL DEPOSITION
Vol. 1 Issue 9, November 2012
Exp. No.
Arc Voltage
Welding current
Welding Speed
Electrode Stick out
Weight before welding
Weight after welding
Weld Deposition
1
20
180
3.85
6
770
780
10
2
24
180
3.85
6
767
782
15
3
20
180
3.85
10
765
777
12
4
20
200
3.85
6
776
788
12
5
20
180
4.54
6
778
786
8
6
20
180
4.54
10
780
790
10
7
24
180
4.54
6
768
781
13
8
20
200
3.85
10
780
794
14
9
24
180
3.85
10
775
792
17
10
20
200
4.54
6
773
783
10
11
24
200
3.85
6
771
788
17
12
20
200
4.54
10
776
786
10
13
24
200
3.85
10
777
796
19
14
24
200
4.54
6
769
785
16
15
24
180
4.54
10
768
783
15
16
24
200
4.54
10
770
780
10
17
22
200
4.16
6
773
786
13
18
20
190
4.54
8
773
783
10
19
24
190
3.85
8
771
788
17
20
22
190
3.85
6
775
788
13
21
22
200
4.54
8
776
787
11
22
22
200
3.85
8
772
787
15
23
22
190
3.85
10
774
789
15
24
20
200
4.16
8
777
789
12
25
22
200
4.16
10
781
796
15
26
24
190
4.16
10
769
786
17
27
24
190
4.16
6
768
783
15
28
20
190
4.16
6
772
782
10
29
20
190
4.16
10
774
786
12
30
22
190
4.54
6
777
788
11
31
22
190
4.54
10
773
786
13
32
24
190
4.54
8
776
791
15
33
24
180
4.16
8
775
790
15
34
22
180
4.54
8
774
785
11
35
20
180
4.16
8
778
788
10
36
22
180
4.16
10
776
789
13
37
24
200
4.16
8
777
794
17
38
22
180
4.16
6
773
784
11
39
22
180
3.85
8
771
784
13
40
20
190
3.85
8
780
792
12
Exp. No.
Arc Voltage
Welding current
Welding Speed
Electrode Stick out
Weight before welding
Weight after welding
Weld Deposition
42
22
190
3.85
8
773
787
14
43
22
190
4.54
8
771
783
12
44
22
190
4.16
6
772
784
12
45
22
180
4.16
8
773
785
12
46
24
190
4.16
8
775
791
16
47
22
200
4.16
8
778
792
14
48
20
190
4.16
8
779
790
11
49
22
190
4.16
8
773
786
13
50
20
180
4.54
8
776
785
9
51
20
190
4.54
6
775
784
9
52
20
180
4.16
6
778
787
9
53
22
180
3.85
6
777
789
12
54
24
200
4.54
8
773
785
12
55
22
200
4.54
10
771
781
10
56
20
180
3.85
8
770
781
11
57
24
190
4.54
10
769
781
12
58
20
200
4.54
8
773
783
10
59
24
200
3.85
8
776
794
18
60
22
200
4.54
6
775
787
12
61
20
190
4.54
10
778
788
10
62
24
200
4.16
10
768
786
18
63
24
190
3.85
10
765
783
18
64
24
190
4.54
6
764
778
14
65
20
180
4.16
10
778
789
11
66
22
180
3.85
10
776
790
14
67
20
200
4.16
10
779
792
13
68
24
180
4.16
6
777
791
14
69
20
200
4.16
6
771
782
11
70
22
200
3.85
6
776
790
14
71
22
180
4.54
10
772
784
12
72
24
180
4.54
8
776
790
14
73
24
180
3.85
8
773
789
16
74
20
190
3.85
6
772
783
11
75
22
200
3.85
10
776
792
16
76
20
200
3.85
8
778
791
13
77
24
190
3.85
6
775
791
16
78
20
190
3.85
10
778
791
13
79
24
200
4.16
6
776
792
16
80
22
180
4.54
6
779
789
10
81
24
180
4.16
10
773
789
16
(Fig: 02) structure utilizes Sugeno type fuzzy interference systems and generalized Gaussian bellshaped membership function to execute a given training data set. It employs 55 nodes, 80 linear parameters, 24 nonlinear parameters, 104 total numbers of parameters 57 training data pairs, 8 checking data pairs and 16 fuzzy rules to predict weld metal deposition. ANFIS modelling process starts by obtaining an inputoutput pair of data sets and dividing it into training and checking data. The training data are used to find out the initial premise parameters for membership functions by equally spacing membership functions.
The final output of the system is the weighted average of the all rule outputs, computed as
Final output (f) = N w f / N w
.. (1)
1 i i 1 i
Where wi = firing strength of the rule fi = output level of each rule

Optimization using Genetic Algorithm
A Genetic Algorithm (GA) is a search heuristic that mimics the process of natural evolution. This heuristic is routinely used to generate useful solutions to optimization and search problems. In a Genetic Algorithm, a population of strings (called Genome or Genotype) which encode candidate solutions (called Individuals or Phenotypes) to an optimization problem evolves toward better solutions. Traditionally, solutions are represented in binary as strings of 0s and 1s, but other encodings are also possible. The evolution usually starts from a population of randomly generated individuals and happens in generations. In each generation, the fitness of every individual in the population is evaluated,
multiple individuals are stochastically selected from the current population (based on their fitness)V, oal.n1dIssmueo9d,iNfioevdember 2012
(recombined and possibly randomly mutated) to form a new population. The new populationis then used in the next iteration of the algorithm. Commonly, the algorithm terminates when either a maximum number of generations has been produced, or a satisfactory fitness level has been reached for the population.
A1
f1
X1
B1
A2
Prod
w1 w1
B2
X2
Norm
Prod
w1 f1
f
X3
C2
C1
Prod
X4
D1
Norm
w2 w2
Prod
f2
w2 f2
D2
Fig: 02. Proposed ANFIS structure for four inputs, single output to predict weld mass deposition in MAG welding process.
The present work aims to explore genetic algorithm (GA) as a method for optimizing welding parameters selection. The algorithm searches for the best solution in terms of arc voltage, welding current, welding speed and electrode stick out with the aim of optimizing an objective function i.e. weld metal deposition. Minimum Deposition is the objective here which is calculated for each input process data combination invoked as the result of multiple input single objective Genetic Algorithm. The objective function of Genetic Algorithm was written in MATLAB and the program was executed to acquire the values of population size, number of generation by taking minimum deposition as the criteria.


RESULTS AND DISCUSSION
As there is a considerable variation in the input data range in terms of numerical value, the input data is normalized to a uniform scale for input to the ANFIS model and this has been achieved by normalizing, using (2) for a range varying from 0.1 to 0.9.
(2)
The normalized input parameters along with the error comparison between experimental results and ANFIS prediction is given in Table II.
TABLE II
NORMALIZED PREDICTED RESULTS BY ANFIS FOR WELD METAL DEPOSITION WITH ERROR
Exp. No.
Arc Voltage
Welding current
Welding Speed
Electrode Stick out
Weld metal deposition
ANFIS
Prediction
Error
1
0.1
0.1
0.1
0.1
0.25
0.27
0.02
2
0.9
0.1
0.1
0.1
0.61
0.62
0.01
3
0.1
0.1
0.1
0.9
0.39
0.37
0.02
4
0.1
0.9
0.1
0.1
0.39
0.38
0.01
5
0.1
0.1
0.9
0.1
0.1
0.08
0.02
6
0.1
0.1
0.9
0.9
0.25
0.21
0.03
7
0.9
0.1
0.9
0.1
0.46
0.47
0.01
8
0.1
0.9
0.1
0.9
0.54
0.61
0.07
9
0.9
0.1
0.1
0.9
0.75
0.77
0.01
10
0.1
0.9
0.9
0.1
0.25
0.2
0.04
11
0.9
0.9
0.1
0.1
0.75
0.79
0.04
12
0.1
0.9
0.9
0.9
0.25
0.21
0.03
13
0.9
0.9
0.1
0.9
0.9
0.77
0.13
14
0.9
0.9
0.9
0.1
0.68
0.49
0.19
15
0.9
0.1
0.9
0.9
0.61
0.61
0
16
0.9
0.9
0.9
0.9
0.25
0.21
0.04
17
0.5
0.9
0.46
0.1
0.46
0.46
0
18
0.1
0.5
0.9
0.5
0.25
0.16
0.09
19
0.9
0.5
0.1
0.5
0.75
0.69
0.07
20
0.5
0.5
0.1
0.1
0.46
0.45
0.01
21
0.5
0.9
0.9
0.5
0.32
0.26
0.06
22
0.5
0.9
0.1
0.5
0.61
0.58
0.03
23
0.5
0.5
0.1
0.9
0.61
0.58
0.03
24
0.1
0.9
0.46
0.5
0.39
0.36
0.03
25
0.5
0.9
0.46
0.9
0.61
0.5
0.11
26
0.9
0.5
0.46
0.9
0.75
0.6
0.16
27
0.9
0.5
0.46
0.1
0.61
0.59
0.02
28
0.1
0.5
0.46
0.1
0.25
0.25
0.01
29
0.1
0.5
0.46
0.9
0.39
0.36
0.03
30
0.5
0.5
0.9
0.1
0.32
0.27
0.05
31
0.5
0.5
0.9
0.9
0.46
0.29
0.17
32
0.9
0.5
0.9
0.5
0.61
0.43
0.18
33
0.9
0.1
0.46
0.5
0.61
0.6
0.01
34
0.5
0.1
0.9
0.5
0.32
0.31
0
35
0.1
0.1
0.46
0.5
0.25
0.23
0.01
36
0.5
0.1
0.46
0.9
0.46
0.47
0
37
0.9
0.9
0.46
0.
0.75
0.6
0.16
38
0.5
0.1
0.46
0.1
0.32
0.33
0.01
39
0.5
0.1
0.1
0.5
0.46
0.45
0.01
40
0.1
0.5
0.1
0.5
0.39
0.36
0.03
41
0.5
0.5
0.46
0.9
0.54
0.46
0.07
Exp. No.
Arc Voltage
Welding current
Welding Speed
Electrode Stick out
Weld metal deposition
ANFIS
Prediction
Error
42
0.5
0.5
0.1
0.5
0.54
0.49
0.04
43
0.5
0.5
0.9
0.5
0.39
0.28
0.11
44
0.5
0.5
0.46
0.1
0.39
0.38
0.01
45
0.5
0.1
0.46
0.5
0.39
0.39
0
46
0.9
0.5
0.46
0.5
0.68
0.58
0.1
47
0.5
0.9
0.46
0.5
0.54
0.46
0.08
48
0.1
0.5
0.46
0.5
0.32
0.28
0.03
49
0.5
0.5
0.46
0.5
0.46
0.41
0.06
50
0.1
0.1
0.9
0.5
0.17
0.14
0.03
51
0.1
0.5
0.9
0.1
0.17
0.13
0.04
52
0.1
0.1
0.46
0.1
0.17
0.2
0.03
53
0.5
0.1
0.1
0.1
0.39
0.4
0.01
54
0.9
0.9
0.9
0.5
0.39
0.35
0.05
55
0.5
0.9
0.9
0.9
0.25
0.21
0.04
56
0.1
0.1
0.1
0.5
0.32
0.3
0.02
57
0.9
0.5
0.9
0.9
0.39
0.41
0.02
58
0.1
0.9
0.9
0.5
0.25
0.2
0.05
59
0.9
0.9
0.1
0.5
0.83
0.75
0.08
60
0.5
0.9
0.9
0.1
0.39
0.32
0.07
61
0.1
0.5
0.9
0.9
0.25
0.2
0.04
62
0.9
0.9
0.46
0.9
0.83
0.55
0.28
63
0.9
0.5
0.1
0.9
0.83
0.74
0.09
64
0.9
0.5
0.9
0.1
0.54
0.46
0.08
65
0.1
0.1
0.46
0.9
0.32
0.3
0.02
66
0.5
0.1
0.1
0.9
0.54
0.54
0
67
0.1
0.9
0.46
0.9
0.46
0.46
0
68
0.9
0.1
0.46
0.1
0.54
0.54
0
69
0.1
0.9
0.46
0.1
0.32
0.32
0
70
0.5
0.9
0.1
0.1
0.54
0.54
0
71
0.5
0.1
0.9
0.9
0.39
0.39
0
72
0.9
0.1
0.9
0.5
0.54
0.54
0
73
0.9
0.1
0.1
0.5
0.68
0.68
0
74
0.1
0.5
0.1
0.1
0.32
0.32
0
75
0.5
0.9
0.1
0.9
0.68
0.68
0
76
0.1
0.9
0.1
0.5
0.46
0.46
0
77
0.9
0.5
0.1
0.1
0.68
0.68
0
78
0.1
0.5
0.1
0.9
0.46
0.46
0
79
0.9
0.9
0.46
0.1
0.68
0.68
0
80
0.5
0.1
0.9
0.1
0.25
0.25
0
81
0.9
0.1
0.46
0.9
0.68
0.68
0
The first 48 data sets are used for training the ANFIS model, next 16 data sets are used to check the model and the last 17 data sets are used to validate the network.
ANFIS predicted data were used to optimize the process using Genetic Algorithm. The objective function of Genetic Algorithm was written in MATLAB and the program was executed to acquire the values of population size, number of generation by taking minimum deposition as the criteria. The graphs of population size v/s minimum average response and number of generation v/s minimum average response was generated as an output of the executed program. The best suitable values of number of generation and population size, could be found as 65 (Fig: 03) and 50 (Fig: 04) respectively where the minimum deposition is in its minimum values.
The genetic algorithm converges to the best suitable minimum value of weld metal deposition in the selected generation. The generation was selected as 50. The next result (Fig. 05) shows how the genetic algorithm is fitted in generation with beat fitness and mean fitness.
With the help of multi input single objective Genetic Algorithm we can get the optimized welding condition to make the weld metal deposition minimized. Following (Table III) is the set of input conditions for the optimization. It is observed at lower values of arc voltage, welding current and electrode stick out and higher values of welding speed weld metal deposition is minimized.
Fig: 03. Variation of Minimum Deposition with no. of generations
Fig: 04. Variation of Minimum Deposition with population size
Fig: 05. Fitness values v/s generation graph
The result of the Genetic Algorithm in MATLAB environment is shown in following Fig: 06 where forVtohle. 1nIossrume a9,liNzoevdember 2012 values of the input parameters the MATLAB program has estimated the weld metal deposition for the minimum deposition criteria. A comparison between Fig: 06 and table III gives the idea of the validated result of the optimization
with the experimental result. It is seen that there is a little variation of input data range of the data set of experiment no. 05 and the optimization result from the Genetic Algorithm.
Fig: 06. MATLAB result for Genetic Algorithm optimization
TABLE III
Exp No.
Arc Voltage
Welding Current
Welding Speed
Electrode Stick out
Arc Voltage
Welding Current
Welding Speed
Electrode Stick out
Arc Voltage
Welding Current
Welding Speed
Electrode Stick out
Experimental Weld Metal Deposition
Normalized Value of Weld Metal Deposition
ANFIS Prediction for Weld Metal Deposition
GA output for Weld Metal Deposition
Experimental Input Data
Normalized Input Data
Optimized Input Data
5
20
180
4.54
6
0.1
0.1
0.9
0.1
0.1051
0.1000
0.8998
0.1055
8
0.1
0.08
0.0847

CONCLUSION
Proposed ANFIS is based on first order Sugeno fuzzy interference system and developed to predict weld metal deposition in a MAG process. Corrected set of input data range has been achieved from the Neuro Genetic modeling and optimization. The difference can be seen in Table III where it is observed that there is a little deviation between the normalized input data and optimized input data. This correction is leading the system to achieve precision results which cant be done by the normal human observations. The residue between the experimental and optimized data can be used as feedback to the system to minimize error. This may lead to the further development of this present work. The present work also could be extended with the involvement of more welding parameters such as electrode diameters, base metal thickness, material type and their effect on the weld metal deposition. The Genetic Algorithm may be extended to multi objective GA to optimize weld metal deposition with depth of penetration, weld strength etc. simultaneously to make the system environment, a more practical one.
REFERENCES

L. A. Zadeh, Fuzzy Sets, Information and Control, Vol.8, pp.338353, 1965.

L. A. Zadeh, Outline of a new approach to the analysis of complex systems and decision processes, IEEEVoTlr. a1nIssasucteio9,nNs oovnember 2012
Systems, Man, and Cybernetics, Vol. 3, No. 1, pp. 2844, Jan 1973.

L. A. Zadeh, Fuzzy Logic, IEEE Computer, pp. 8389, 1988.

E.H. Mamdani and S. Assilian, An experiment in linguistic synthesis with a fuzzy logic controller. International Journal of ManMachine studies, 7(1):113,1975.

T. Takagi and M. Sugeno, Fuzzy identification of systems and its applications to modeling and control. IEEE Transactions on Systems, Man, and Cybernetics, 15:116132, 1985.

M. Sugeno and G. T. Kang, Structure identification Fuzzy model. Fuzzy Sets and Systems, 28:1533, 1988.

JyhShing Roger Jang, ANFIS: AdaptiveNetworkBased Fuzzy Interference System, IEEE transactions on systems, man, and cybernetics, Vol. 23, no. 3, pp. 665685, May/June 1993

J. Hopfield, Neural networks and physical systems with emergent collective computational abilities, Proc. Natl. Acad. Sci. USA, Vol. 79, pp. 25542558, 1982.

J. Hopfield and D. Tank, Computing with neural circuits: A model, Science, Vol. 233, pp. 625633, 1986.

D.E. Goldberg, Genetic algorithms for search, optimization and machine learning, Reading, MA: AddisonWesley, 1989.

D.E. Goldberg, Optimizing globallocal search hybrids, Proceedings of the Genetic and Evolutionary Computation Conference (GECCO), pp. 220228, 1999.

H. Ishibuchi, T. Murata, Minimizing the fuzzy rules and maximizing its performance by a multi objective genetic algorithm., Proceedings of the Sixth International Conference on Fuzzy Systems, pp. 259264, 1997.

Ken Nozaki, Hisao Ishibuchi, Hideo Tanaka, A simple but powerful heuristic method for generating fuzzy rules from numerical data, Fuzzy Sets and Systems 86 (1997) Elsevier Science pp. 251270, 1995.

Davi Sampaio Correia, Cristiene Vasconcelos GonÃ§alves, Sebastiao Simoes da Cunha, Jr., Comparison between genetic algorithms and response surface methodology in GMAW welding optimization, Journal of Materials Processing Technology 160, pp. 7076, 2005.

Y.S Tarng., S.C. Juang, C.H. Chang, The use of greybased Taguchi methods to determine submerged arc welding process parameters in hardfacing, Journal of Materials Processing Technology,128,I6, 2002.

K. Manonmani, N. Murugan, G. Buvanasekaran Effect of process parameters on the weld bead geometry of laser beam welded stainless steel sheets, Mt J Joining Mater 2005:l7(4):l03 9, 2005.

S. Datta et al., Effect of electrode stick out on quality and performance of submerged arc weldment – experimental and statistical analyses, International Conference on Mechanical Engineering 2005 Dhaka, Bangladesh ICME05AM39, 2005.

Jagdev Singh and Simranpreet Singh Gill, Multi input single output fuzzy model to predict tensile strength of radial friction welded GI pipes, International Journal of Information and Systems Sciences, Volume 4, Number 3, pp. 462477, 2006.

Manoj Singla, Dharminder Singh, Dharmpal Deepak, Parametric optimization of gas metal arc welding processes by using factorial design approach, Journal of Minerals & Materials Characterization & Engineering, Vol. 9, No.4, pp.353363, 2010.

P. Kumari, K. Archna and R.S. Parmar, Effect of Welding Parameters on Weld Bead Geometry in MIG Welding of Low Carbon Steel, International Journal of Applied Engineering Research ISSN 09734562 Volume 6, Number 2 (2011) pp. 249 258, 2011.

Dhas J. E. R. and Somsundaram Kumanan, Neuro hybrid model to predict weld bead width in submerged arc welding process, Journal of Scientific & Idustrial Research Vol. 69, May 2010, pp. 350355, 2010.

A. Biswas, Bead geometry optimization of Submerged Arc Weld: Exploration of Weighted Principal Component Analysis (WPCA), 2nd International Conference on Mechanical, Industrial and Manufacturing Technologies (MIMT 2011), Singapore 2011.