Vous êtes sur la page 1sur 6

International Journal of Applied Engineering

Research and Development (IJAERD)

ISSN 2250-1584
Vol. 3, Issue 3, Aug 2013, 1-6
TJPRC Pvt. Ltd.

Professor, Department of Mechanical Engineering, Sreenidhi Institute of Science & Technology, Hyderabad,
Andhra Pradesh, India

In submerged arc welding (SAW), weld quality is greatly affected by the weld parameters such as welding current,
welding speed; arc voltage and electrode stickout since they are closely related to the geometry of weld bead, a relationship
which is thought to be complicated because of the non-linear characteristics. However, trial-and-error methods to
determine optimal conditions incur considerable time and 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 as per Taguchis Design of Experiments and analysis of variance(ANOVA) was carried to
establish inputoutput relationships of the process. By this relationship, an attempt was made to minimize weld bead width,
a good indicator of bead geometry, using optimization procedures based on the ANN models to determine optimal weld
parameters. The optimized values obtained from these techniques were compared with experimental results and presented.
KEYWORDS: Weld Bead, ANOVA, Artificial Neural Networks, Sensitivity Analysis
Submerged arc welding (SAW) is widely used welding process in most fabrication industries. It requires a non-
continuously fed consumable solid or tubular (flux cored) electrode. The molten weld and the arc zone are protected from
atmospheric contamination by being submerged under a blanket of granular fusible flux consisting of lime, silica,
manganese oxide, calcium fluoride, and other compounds. When molten, the flux becomes conductive, and provides a
current path between the electrode and the work. This thick layer of flux completely covers the molten metal thus
preventing spatter and sparks as well as suppressing the intense ultraviolet radiation and fumes that are part of the shielded
metal arc welding (SMAW) process as shown in figure 1. SAW is normally operated in the automatic or mechanized mode,
however, semi-automatic (hand-held) SAW guns with pressurized or gravity flux feed delivery are available. Semi
automatic with gravity flux feed SAW machine is used in the present work.

Figure 1: Submerged Arc Welding Process
2 K. Srinivasulu Reddy
Total volume of the weld bead is one of the important bead parameters controlled by most of the other bead
parameters. Hence, the total volume, if optimized (minimized), obviously minimizes most of the other bead quality
parameters such as heat input, dilution, reinforcement, bead width and penetration. But for a sound and strong weld, bead
penetration should be maximized. Hence, in optimizing the total volume of the weld bead, the penetration, included as a
constraint, should be set at its maximum value. Minimizing the size of the weld bead reduces the welding cost through 1)
reduced consumption of consumables such as electrodes and flux 2) reduced heat input and energy consumption and 3)
increased welding productivity through a high welding speed.
Because of these advantages, the total volume of the weld bead should be optimized, having other bead
parameters as constraints, rather than optimizing all the bead parameters individually. The total volume of the weld bead is
the area of the weld bead cross section multiplied by the length of the weld bead. To reduce the complexity of the problem,
the length of the bead is assumed as unity, which simplifies the equation. Now the total volume is equal to the total area of
the weld bead. Hence, the optimization of the total area of the weld bead cross section, subject to minimum reinforcement,
minimum bead width, and maximum penetration is one objective and developing artificial neural network (ANN) models
to predict the weld bead properties accurately along with sensitivity analysis is also the prime objective of this paper.
Taguchi Methodology
Essentially, traditional experimental design procedures are too complicated and not easy to use. A large number of
experimental works have to be carried out when the number of process parameters increases. To solve this problem, the
Taguchi method uses a special design of orthogonal arrays to study the entire parameter space with only a small number of
experiments. The greatest advantage of this method is the saving of effort in conducting experiments; saving experimental
time, reducing the cost, and discovering significant factors quickly.
Step 1: Identification of important process variables
Step 2: Development of process plan
Step 3: Conducting experiments as per the plan
Step 4: Recording the responses (e.g. bead geometry)
Step 5: Testing the welded job.
Step 6: Finding out the optimized values of the parameters
Step 7: Presenting the main and significant effects of process parameters.
Important process variables identified are shown in table 1. The effect of many different parameters on the
performance characteristic in a condensed set of experiments can be examined by using the orthogonal array experimental
design proposed by Taguchi. Once the parameters affecting a process that can be controlled have been determined, the
levels at which these parameters should be varied must be determined. Determining what levels of a variable to test
requires an in-depth understanding of the process, including the minimum, maximum, and current value of the parameter.
If the difference between the minimum and maximum value of a parameter is large, the values being tested can be further
apart or more values can be tested. If the range of a parameter is small, then less value can be tested or the values tested
can be closer together. Levels of process parameters identified are shown in table 2. Knowing the number of parameters
Optimization & Prediction of Welding Parameters and Bead Geometry in Submerged Arc Welding 3
and the number of levels, the proper orthogonal array can be selected. L-8 Orthogonal array as shown in table 3 is used in
the present work. Furthermore, Taguchi has created a transformation of the repetition data to another value, which is a
measure of the variation present. The transformation is known as signal-to-noise (S/N) ratio.
The S/N ratio consolidates several repetitions (at least two data points are required) into one value, which reflects
the amount of variation present. There are several S/N ratios available depending on the type of characteristic; lower is
better (LB), nominal is best (NB), or higher is better (HB). The S/N ratio for each level of process parameters is computed-
based on the S/N analysis.
Regardless of the quality of the quality characteristic, a large S/N ratio corresponds to a better quality
characteristic. Therefore, the optimal level of the process parameters is the level with the highest S/N ratio as shown in
table 3. A statistical analysis of variance (ANOVA) is performed to see which process parameters are statistically
Table 1: Welding Parameters
Input Parameters Output Parameters
1.Weld current 1.Depth
2.Voltage 2.Penetration
3.Weld speed 3.Width
4.Electrode stick out 4.Weld bead hardness

Table 2: Levels of Process Parameters
Symbol Welding Parameters Level 1 Level 2
A Welding current(amp) 360 390
B Arc voltage(volts) 25 26
C Welding speed (mm/min) 400 420
D Electrode stick out(mm) 19 25

Table 3: L8 Orthogonal Array
Stick Out
Bead Width
1 1 1 1 1 13 169 22.28
2 1 1 1 2 11 121 20.83
3 1 2 2 1 12.5 156.25 21.94
4 1 2 2 2 13.5 182.25 22.61
5 2 1 2 1 14.5 210.25 23.23
6 2 1 2 2 14.0 196 22.92
7 2 2 1 1 14.5 210.25 23.23
8 2 2 1 2 15 225 22.52

Analysis of Variance (ANOVA)
ANOVA is a collection of statistical models and their associated procedures, in which the observed variance in a
particular variable is partitioned into components attributable to different sources of variation. In its simplest form,
ANOVA provides a statistical test of whether or not the means of several groups are all equal, and therefore generalizes t-
test to more than two groups.
Doing multiple two-sample t-tests would result in an increased chance of committing a type I error. For this
reason, ANOVAs are useful in comparing three, or more means. From ANOVA results shown in table 4, it can be stated
that contribution of current is maximum and contribution of speed and electrode stick out are minimum in optimum bead
4 K. Srinivasulu Reddy
Table 4: ANOVA Results of Bead Width
Sum of
Degree of
% of
CURRENT 8.0 1 8.0 12.0 0.013 66.67
VOLTAGE 1.125 1 1.125 0.621 0.461 9.38
SPEED 0.125 1 0.125 0.063 0.810 1.04
0.125 1 0.125 0.063 0.810 1.04
ERROR 2.625 3 21.825
TOTAL 12.0 7

ANN Prediction Model
ANN based models are data driven, adaptive, fast in response and have good accuracy if they are trained with
proper data. These ANN-based models can be used for real time simulation, monitoring, prediction of weld bead
parameters. Analogous to the human brain and its learning process, ANN utilizes connectivity and transfer functions
between input, hidden, and output neurons, and has been applied successfully to non-linear systems or systems with
unclear dynamics.
In particular, different from mathematical models such as regression models, ANN models have adaptability
through a self-tuning process, so can decide accurately without outside expert intervention when unusual perturbations,
disturbances, and/or changes in building background conditions occur.
This paper has presented the application of Taguchi technique to determine the optimal process parameters for
SAW process. Experimentation was done according to the Taguchis design of experiments. Using the signal-to-noise ratio
and the ANOVA technique the influence of each welding parameters are studied and the prediction of the bead geometry is
done. Then it is used to predict the SAW process parameters for any given welding conditions.
From the available 8 data sets, 7 data sets are used to train the network as given in table 5. Eigth data set is used
for production as given in table 6.Various ANN models are trained and found that modular network model is the best suited
model which can predict the values with less than 5 % error as given in table 7.
Table 5: Training Data Sets
Stick Out
Depth of
Weld Bead
1 360 25 400 19 13.0 2.0 3.0 37
2 360 25 400 25 11.0 2.0 3.5 40
3 360 26 420 19 12.5 3.0 3.5 42
4 360 26 420 25 13.5 1.5 4.0 34
5 390 25 420 19 14.5 2.0 5.0 52
6 390 25 420 25 14.0 2.5 4.5 48
7 390 26 400 19 14.5 2.0 4.0 49

Table 6: Production Data Set
Stick Out
Depth of
Weld Bead
8 390 26 400 25 13.27 1.56 3.67 36.14

Optimization & Prediction of Welding Parameters and Bead Geometry in Submerged Arc Welding 5
Table 7: Comparison of Actual and Predicted Values

Depth of
Weld Bead
Experimental value 13.2 1.5 3.6 36
Predicted value 13.27 1.56 3.67 36.14
% error 0.53 4.00 1.94 0.39

Modular Neural Network Model
Modular feed forward networks are a special class of MLP. These networks process their input using several
parallel MLPs, and then recombine the results. This tends to create some structure within the topology, which will foster
specialization of function in each sub-module. In contrast to the MLP, modular networks do not have full interconnectivity
between their layers. Therefore, a smaller number of weights are required for the same size network (i.e. the same number
of PEs). This tends to speed up training times and reduce the number of required training exemplars. There are many ways
to segment a MLP into modules. It is unclear how to best design the modular topology based on the data. There are no
guarantees that each module is specializing its training on a unique portion of the data. Modular network predicted data is
given in table 8. The network is trained for 1000 epochs in three runs and least mean square error obtained is 0.00075 as
given table 9. Average of minimum MSEs in all the three runs with corresponding standard deviation values are given in
table 10.
Table 8: ANN Predicted Data
Stick Out
Bead Width
ANN Predicted
ANN Predicted
Depth of
ANN Predicted
Weld Bead
360 25 400 19 12.97 2.00 3.08 37.06
360 25 400 25 11.09 1.99 3.47 39.76
360 26 420 19 12.51 2.97 3.50 42.02
360 26 420 25 13.52 1.58 4.01 34.80
390 25 420 19 14.57 2.00 4.90 51.29
390 25 420 25 13.95 2.49 4.53 47.99
390 26 400 19 14.54 2.00 4.00 49.07

Table 9: Min. MSE & No. of Epochs
Best Network Training
Run # 3
Epoch # 1000
Minimum MSE 0.000752133
Final MSE 0.000752133

Table 10: Final MSEs & Standard Deviations
All Runs
Training Standard
Average of Minimum MSEs 0.000900985 0.000163801
Average of Final MSEs 0.000900985 0.000163801

Sensitivity Analysis
Sensitivity analysis, also known as the post optimality analysis, is the study of what happens to the value of the
objective function if the limit of each of the constraints is changed from optimum value. Optimum solution for any
function lies in a boundary or zone and, hence, it is not a single constant value[3]. This provides flexibility in fixing the
limits for the constraints. Also, for every value of each of the constraints there is a possibility for change in the value of the
6 K. Srinivasulu Reddy
objective function as well as other constraints. Sensitivity analysis was carried out for the total volume of the weld bead by
varying the limit of one of the constraints at a time and the effect of the change in the constraint limit on the value of the
total volume and other constraints were recorded and given in figure 2 and Table 11.

Figure 2: Sensitivity Analysis of SAW Process Parameters
Table 11: Sensitivity Values of SAW Process Parameters
Depth of
Weld Bead
Welding current 0.5848 0.0802 0.4141 4.6233
Arc voltage 0.1584 0.1207 0.1154 2.9008
Welding speed 0.2327 0.1414 0.4472 3.2095
Electrode stick out 0.4949 0.3427 0.0293 5.2937

Design of experiments was carried out to establish the relationship between process variables and optimization
tools are used to find an optimal solution. It is observed that the developed ANN model is a powerful tool in experimental
welding optimization, even when experimenter does not have to model the process. Modular network model predicts
accurately and corresponding sensitivity analysis revels that bead width is highly sensitive to welding current, weld
reinforcement and bead hardness are sensitive to electrode stick out and depth of penetration is sensitive to welding speed.
1. V.Gunaraj, N.MuruganPrediction and Optimization of Weld Bead Volume for the Submerged Arc Process Part
2 welding research supplement 331-338-s | Nov 2000
2. S Kumanan, J Edwin Raja Dhas & K Gowmtham Determination of submerged arc welding process parameters
using Taguchi method and regression analysis India Journal of Engineering &Material Sciences vol 14, June
2007, pp. 177-183
3. J. Edwin Raja Dhas, S. Kumanan Optimization of parameters of submerged arc welding using non conventional
techniqyes Applied Soft Computing, Elsevier , May 2011
4. P. Kanjilal, T.K. Pal, S.K. Majumdar, Combined effect of flux and welding parameters on chemical composition
and mechanical properties of submerged arc weld metal. Journal of Materials Processing Technology, 2006. 171:
p. 223-231.
5. Saurav Datta and Siba Sankar Mahapatra Multi-Objective Optimization of Submerged Arc Welding Process The
Journal of Engineering Research Vol. 7, No. 1, (2010) 42-52