Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In submerged arc welding (SAW), weld quality is greatly affected by the weld parameters such as welding
Received 8 May 2009 current, welding speed; arc voltage and electrode stickout since they are closely related to the geometry
Received in revised form 19 March 2011 of weld bead, a relationship which is thought to be complicated because of the non-linear character-
Accepted 23 May 2011
istics. However, trial-and-error methods to determine optimal conditions incur considerable time and
Available online 30 May 2011
cost. In order to overcome these problems, non-traditional methods have been suggested. Bead-on-plate
welds were carried out on mild steel plates using semi automatic SAW machine. Data were collected
Keywords:
as per Taguchi’s Design of Experiments and regression analysis was carried to establish input–output
Submerged arc welding
Optimization
relationships of the process. By this relationship, an attempt was made to minimize weld bead width,
Weld parameters a good indicator of bead geometry, using optimization procedures based on the genetic algorithm (GA)
Genetic algorithm and particle swarm optimization (PSO) algorithm to determine optimal weld parameters. The optimized
Particle swarm optimization values obtained from these techniques were compared with experimental results and presented.
© 2011 Elsevier B.V. All rights reserved.
1568-4946/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.asoc.2011.05.041
J.E.R. Dhas, S. Kumanan / Applied Soft Computing 11 (2011) 5198–5204 5199
Table 2
Observed values from the test specimen.
Trial nos. Welding Arc voltage (V) Welding speed Electrode Weld bead
current (A) (mm/min) stickout (mm) width (mm)
Start
Initial Population
(I, V, S & E)
Roulette Selection
Mutation
New population
No
Satisfy termination
criteria?
Yes
Stop
electrode stickout are established as independent parameters and tions initially distributed over the whole function space and quickly
weld bead width as independent parameters in this study and the identifies the sub domain in which the global minimum function
relationship between the weld parameters is given in Eq. (1). The is located, by maintaining the constant exploration of the search
regression analysis used in this study was SPSS, a commercial Sta- space. Hence it is possible by genetic algorithm to reach accurately
tistical Program for Social Science software. and reliably the global optimum solution in a short computation
time.
Weld bead width (mm) = 118 + (0.0056I)
Optimization procedure for SAW using proposed genetic algo-
−(1.3167V ) − (0.1708S) − (6.33E) rithm is shown in Fig. 2. Initial population is the possible number of
solutions of the optimization problem, and each possible solution
+(0.0028IE) + (0.0667VE) + (0.0083SE) (1) is called an individual. In this work, a possible solution is formed
where I = welding current (A), V = arc voltage (V), S = welding speed by values of the welding current, arc voltage, welding speed, and
(mm/min), and E = electrode stickout (mm). electrode stickout. Evaluation of the fitness function in a GA is a
The validity of the equation is confirmed by calculating R2 -value necessary procedure to decide the survival of each individual. The
which is equal to 0.814, having high coefficient of correlation. steps used in optimization by GA are:
3. Weld parameter optimization by GA 1. Determine the objective i.e., minimization of bead width
2. Identify the system parameters that are affecting the objective
Equation solved by different analytical/iterative methods (such function.
as Newton method and Bisection method) may converge to a sta- 3. Build an analytical/experimental model of the system and
tionary point and there is no guarantee that the solution converges express the objective function as the function of a system.
to global optimum. But genetic algorithm uses a population of solu- 4. Formulate an objective function shown in Eq. (1).
J.E.R. Dhas, S. Kumanan / Applied Soft Computing 11 (2011) 5198–5204 5201
Table 3
SAW input parameters.
360 25 410 23
Process parameters Ranges
360 26 410 23
5. Formulate the constraints of the objective function as:
which are the constraints
Fig. 4. Mutation of a gene in developed GA for near optimal weld bead width.
360 ≤ welding current ≤ 390
25 ≤ arc voltage ≤ 30
Table 4
400 ≤ welding speed ≤ 420
Parameters for GA computations.
19 ≤ electrode stickout ≤ 25
Population size 20
Number of generations allowed 100
The objective function is taken from Eq. (1) to minimize bead
Mutation rate 1%
width using arc voltage, welding current, welding speed and elec- Crossover rate 80%
trode stickout as SAW process parameters. It is assumed that the Type of crossover Single point
near optimum point is within the following experimental region, Type of selection Roulette wheel
defined by the GA search ranges as given in Table 3. Initial popula-
tion is created using the welding process parameters and random
solutions are obtained for weld bead width. Fitness function values where each entry has a probability rate of being mutated. In the
of each individual are calculated for the current populations using second step, the algorithm replaces each selected entry by a random
the objective function. Genetic operators (reproduction, crossover number selected uniformly from the range for that entry. Fig. 4
and mutation) are employed to produce the next generation of the shows the mutation of a gene.
new population. In GA procedure, population size, crossover rate and mutation
Selection: The selection process is based on spinning the rate are important factors which determine the performance of the
roulette wheel n times. The individuals selected from the roulette algorithm. Large population size or a higher crossover rate allows
wheel technique selection process are stored in a mating pool. exploration of solution space and reduces the chances of settling
Crossover: After selection, crossover proceeds in three steps. for poor solution. The parameters for GA computations are given
First, two strings (referred to as parents) are selected randomly in Table 4. The mutation rate is chosen as 1% and crossover rate is
from the mating pool. Second, an arbitrary location (called the 80%, i.e., 80% of the pairs are crossed, whereas the remaining 20%
crossover site) in both strings is selected randomly. Third, the por- are added to the next generation without crossover. The final value
tions of the strings following the crossover site are exchanged of the objective function is constant for last 10 generations shows
between two parent strings to form two offspring strings. This that it is the best value for the problem.
crossover does not occur with all strings, but is limited by the
crossover rate. Fig. 3 shows the typical example for single point 4. Weld parameter optimization by PSO
crossover in GA in which two mating chromosomes are cut at one
point and the cuts are exchanged between the two parents. Evolutionary technique PSO is proposed to optimize SAW
Mutation: Mutation rate is set at a low value to avoid losing good parameters to achieve good quality of weld. Optimization proce-
strings. It also provides information that did not exist in the initial dure for SAW using PSO is shown in Fig. 5. Initial population is the
stage. Uniform mutation is a two-step process. First, the algorithm possible number of solutions (particles) of the optimization prob-
selects a fraction of the vector entries of an individual for mutation, lem, and each possible solution is called an individual. In this work,
Parent 2
Parent 1
Offspring 1 Offspring 2
No
Termination criteria met
Yes
Result
a possible solution is formed by values of the welding current, 4.1. Proposed algorithm for the developed PSO model
arc voltage, welding speed, and electrode stickout. It is assumed
that the near optimum point is within the following experimental 1. Initialize each particle.
region, as given in Table 1. The objective function to minimize weld 2. Calculate the fitness value for each particle. If the fitness value
bead width takes the form from Eq. (1). is better than the best fitness value (Pbest) in history. Set the
Initial population is created using the welding process param- current value as the new Pbest.
eters (welding current, arc voltage, welding speed and electrode 3. Choose the particle with the best fitness value of all the particles
stickout) and random solutions are obtained for weld bead width. as the Gbest.
Fitness function values of each individual (particles) are calcu- 4. For each particle, calculate the particle velocity according to Eq.
lated for the current populations using the objective function. The (2). The particle position is updated by Eq. (3). C1 , C2 are learning
best fitness value is stored as Pbest from history. From all parti- factors. C1 = C2 = 1.4.
cles choose the best fitness value called Gbest. For each particle, 5. Particle velocities in each dimension are clamped to a maximum
calculate the velocity of the particle by velocity Vmax . If the sum of acceleration would cause the velocity
on that dimension to exceed Vmax , which is a parameter specified
Velocity[] = w Velocity[] + C1 rand1 (Pbest[] − present[]) by the user, the velocity in that dimension is limited to Vmax .
+C2 x rand2 (Gbest[] − present[]) (2) 4.2. Numerical illustration for developed PSO model
Table 6
Comparison of results from developed GA and PSO models with experimental results.
Solution by Welding Arc voltage (V) Welding speed Electrode Near optimal weld
current (A) (mm/min) stickout (mm) bead width (mm)
15.5
Arc voltage is calculated by
15
Arc voltage, V = Vmin + (Vmax − Vmin ) − rand()
Fitness value
14.5
13.5
Welding speed, S = Smin + (Smax − Smin ) − rand()
13
Electrode stickout is calculated by
12.5
0 20 40 60 80 100
Electrode stickout, E = Emin + (Emax − Emin ) − rand()
Number of Iterations
These values are substituted in Eq. (1) and weld bead width are Fig. 6. Convergence of PSO for optimal weld bead width.
calculated.
Calculation of Pbest value: The minimum weld bead width for 5. Results and discussion
each initial solution is considered as the Pbest value. This is the best
value of the particular solution only. Experiments are designed as Taguchi’s principles and are con-
Calculation of Gbest value: The minimum weld bead width for ducted in a semi automatic SAW machine using L8 orthogonal
the initial solution or whole iteration is considered as the Gbest array. Data observed from experiments is used to develop an objec-
value. tive function. Evolutionary computing techniques GA and PSO are
Let us consider that the particle sequence at iteration ‘n’ is invoked to optimize the parameters of SAW, for optimal weld qual-
ity. Convergence of the developed PSO model for optimal weld bead
Present [] = 370 27 405 23
width is shown in Fig. 6. From the figure it is evident that the min-
imal weld bead width is obtained at the 81st iteration. The weld
where n = iteration number.
bead width is gradually increasing up to the 81st iteration. Then
Let the particles best (Pbest) sequence be
the weld bead width is constant for further iterations.
Pbest = 390 28 415 24 The performances of proposed and developed GA and PSO algo-
rithms to optimize parameters of SAW are compared. Table 6 shows
Let the particles global best (Gbest) sequence till the present the comparison of results from developed GA and PSO model with
iteration be experimental values. PSO model gives optimal results than GA
Gbest = 380 29 410 25 model. Also it converges to the global optimum solution at the faster
rate. Further number of iterations refines the optimal value and
The velocity of the particle is found out by the Eq. (2). gives more accurate results. The reason is that PSO searches solu-
Where C1 = C2 = 1.4. tion in the search space different from GA. PSO maintains internal
rand1 is randomly generated as 0.456 memory to store the Gbest and Pbest solutions. Each individual
rand2 is randomly generated as 0.539 in the population will try to emulate the Gbest and Pbest solu-
tions in the memory through updating by two PSO equations. But
Pbest –Present = 20 1 20 1 GA, iteratively searches for several good individuals in the popula-
tion, and make the population to emulate the best solutions found
Gbest –Present = 10 2 5 2 in that generation through reproduction, crossover and mutation
operators. GA requires substantial computational time to perform
Substituting all values in Eq. (2) by taking previous velocity as decision making. Hence the effectiveness of PSO in finding the true
10 2 5 1 and inertia weight w = 0.5
V [] =0.5* (10 2 5 1 ) + 1.4 * 0.456 (20 1 20 1) +1.4 * 0.539 (10 2 5 2)
= 25.37 3.13 16.04 3.27
New sequence = Present[n] + V []
= global optimal solution is competent than GA. Fig. 7 illustrates a
395.3 30.13 421.1 26.27 good conformability of the developed models to the real process.
Overall, the results indicate that both PSO and GA algorithms
Generated new particles in the new sequence beyond the limit can be used in the optimizing the parameters of a SAW machine.
are scaled and are brought within the bounds before fitness is eval- The target values can be set as per the requirement based on the
uated. application. The models are developed in the MATLAB platform,
Gbest value for overall iteration: Gbest value is obtained at the which is highly reliable, adaptable and very user friendly. Further-
81st iteration. more, control parameters and objective function are involved in
5204 J.E.R. Dhas, S. Kumanan / Applied Soft Computing 11 (2011) 5198–5204
Weld bead width [3] D.S. Correia, C.V. Goncalves, S.C. Sebastiao Junior, V.A. Ferrares, GMAW welding
16 optimization using genetic algorithms, J. Braz. Soc. Mech. Sci. Eng. 26 (2004)
Predicted values, mm
29–33.
15 [4] J. Kennedy, R. Eberhart, Particle swarm optimization, in: Proceedings of IEEE
GA PSO International Conference on Neural Networks, vol. 4, Perth, Australia, 1995, pp.
14 1942–1948 (IEEE Service Centre, Piscataway, New Jersey).
[5] S. Fukuda, H. Morita, Y. Yamauchi, I. Nagasawa, S. Tsuji, Expert system for deter-
13 mine welding condition for a pressure vessel, Iron Steel Inst. Jpn. Int. 30 (1990)
150–154.
12 [6] D. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning,
Addison Wesley, MA, USA, 1989.
11 [7] V. Gunaraj, N. Murugan, Prediction and optimization of weld bead volume for
the submerged arc process – part 1, Weld. J. (2000) 286-s–294-s.
10
[8] V.K. Gupta, R.S. Parmar, Fractional factorial techniques to predict dimensions
10 11 12 13 14 15 16 of the weld bead in automatic submerged arc welding, J. Inst. Eng. (India). 70
(1986) 67–71.
Experimental values, mm [9] B. Howard Cary, Modern Welding Technology, Prentice Hall Inc., Englewood
Cliffs, New Jercy, 1979.
Fig. 7. Comparison of weld bead width. [10] J. Lee, K. Um, A comparison in a back-bead prediction of gas metal arc welding
using multiple regression analysis and artificial neural network, Opt. Lasers
Eng. 34 (2000) 149–158.
[11] D. Kim, S. Rhee, Optimization of arc welding process parameters using a genetic
these optimization techniques, and appropriate selection of these algorithm, Weld. J. 80 (2001) 184–189.
[12] I.S. Kim, J.S. Son, C.E. Park, Lee, C.W. Yarlagadda, K.D.V. Prasad, A study on pre-
is a key point for success. diction of bead height in robotic arc welding using a neural network, J. Mater.
Process. Technol. 131 (2002) 229–234.
[13] D. Kim, M. Kang, S. Rhee, Determination of optimal welding conditions with a
6. Conclusions
controlled random search procedure, Weld. J. (2005) 125–130.
[14] S. Kumanan, J. Edwin Raja Dhas, K. Gowthaman, Determination of submerged
Design of experiments was carried out to establish the rela- arc welding process parameters using Taguchi method and regression analysis,
tionship between process variables and optimization tools GA Indian J. Eng. Mater. Sci. 14 (2007) 103–111.
[15] J.I. Lee, K.W. Um, A prediction of welding process parameters by prediction of
and PSO are used to find an optimal solution. It is observed that back-bead geometry, J. Mater. Process. Technol. 108 (2000) 106–113.
the developed PSO algorithm is a powerful tool in experimental [16] J.L. Lee, S. Rhee, Prediction of process parameters for gas metal arc welding by
welding optimization, even when experimenter does not have to multiple regression analysis, Proc. Inst. Mech. Eng. (B) 214 (2000) 443–449.
[17] D.S. Nagesh, G.L. Datta, Prediction of weld bead geometry and penetration in
model the process. These tools find a good scope in the welding shielded metal-arc welding using artificial neural networks, J. Mater. Process.
shop floor environment to set the initial process parameter for Technol. 123 (2002) 303–312.
the weld. Sometimes conventional computational intelligence (GA, [18] S.V. Nadkarni, Modern Welding Technology, Oxford and IBH Publishing Co. Pvt.
Ltd., New Delhi, 1988.
ANN, PSO, etc.) produces extra burden on the user while formulat- [19] J. Revaandra, R.S. Parmar, Mathematical model to predict weld bead geometry
ing a problem with respect to the method used. When information for flux cored welding, Met. Construct. 19 (1987) 31R–35R.
is missing or uncertain, there are methods to deal with the opti- [20] P. Sathiya, S. Aravindan, A. Noorul Haq, K. Paneerselvam, Optimization of
friction welding parameters using evolutionary computational techniques, J.
mization problems without destroying the user-friendliness and
Mater. Process. Technol. 209 (2009) 2576–2584.
transparency. In this case decision-making approach using point- [21] K. Siva, N. Murugan, R. Logesh, Optimization of weld bead geometry in plasma
cloud-based can be applied. transferred arc hardfaced austenitic stainless steel plates using genetic algo-
rithm, Int. J. Adv. Manuf. Technol. 41 (2009) 24–30.
[22] G. Taguchi, Introduction to Quality Engineering, Asian Productivity Organiza-
Acknowledgements tion, Tokyo, 1990.
[23] Y.S. Tarng, W.H. Yang, Optimization of the weld bead geometry in gas tung-
sten arc welding by the Taguchi method, Int. J. Adv. Manuf. Technol. 14 (1998)
Authors express sincere thanks to the Ministry of Human 549–554.
Resource Development, Government of India for the sponsorship [24] L.J. Yang, R.S. Chandel, An analysis of curvilinear regression equations for mod-
under the research and development programme to undertake this eling the submerged-arc welding process, J. Mater. Process. Technol. 37 (1993)
601–611.
research work.
Dr. J. Edwin Raja Dhas is a Professor and Head in the Department of Automobile
Engineering at Noorul Islam University, India He obtained his Doctoral Degree from
References National Institute of Technology, Tiruchirappalli, India. His research interests are
Non traditional simulation and optimization of Manufacturing Systems.
[1] S.S. Amar Patnaik, S. Biswas, Mahapatra, An evolutionary approach to param-
eter optimisation of submerged arc welding in the hardfacing process, Int. J. Dr. S. Kumanan is a Professor and Dean in the Department of Production Engi-
Manuf. Res. 2 (2007) 462–483. neering at National Institute of Technology, Tiruchirappalli, India. He obtained his
[2] N. Baskar, P. Asokan, R. Saravanan, G. Prabhaharan, Optimization of machining Doctorate Degree in Manufacturing Management from Indian Institute of Technol-
parameters for milling operations using non-conventional methods, Int. J. Adv. ogy, Madras, India. His research interests are Intelligent Manufacturing Systems,
Manuf. Technol. 25 (2005) 1078–1088. Modelling, Simulation and Optimization of Manufacturing Systems.