Comparative study on different pin geometries of tool profile in friction stir welding

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 –
6340(Print), ISSN 0976 – 6359(Online) Volume 4, Issue 2, March - April (2013) © IAEME
245
COMPARATIVE STUDY ON DIFFERENT PIN GEOMETRIES OF
TOOL PROFILE IN FRICTION STIR WELDING USING ARTIFICIAL
NEURAL NETWORKS
D. Kanakaraja 1
, P. Hema 2
, K. Ravindranath 3
1
PG Student 2
Assistant professor 3
Professor Mechanical Engineering Department,
Sri Venkateswara College of Engineering, S.V.University, Tirupati, Andhra pradesh- 517502,
India
ABSTRACT
Friction stir welding (FSW) is an innovative solid state joining technique and has
been employed in aerospace, rail, automotive and marine industries for joining aluminium,
magnesium, zinc and copper alloys. The FSW process parameters such as tool rotational
speed, welding speed, axial force etc., play a major role in deciding the weld quality. The
present work is a comparitive study on different pin geometries of FSW tool using ANN in
MATLAB. This work focuses on two methods such as Artificial neural networks and
Regression analysis to predict the tensile strength of friction stir welded 6061 aluminium
alloy. An artificial neural network (ANN) model was developed for the analysis of the
friction stir welding parameters of AA6061 plates. The Tensile strength of weld joints were
predicted by taking the parameters Tool rotation speed, Weld speed and Axial force as a
function. A comparison was made between measured and predicted data. A regression model
is also developed and the values obtained for the response Tensile strengths are compared
with measured values. The graphs were plotted between Regression predicted values and
Experimental data to show the accuracy of experimental results. It was found that among
these methods ANN model is easier and effective methodology in order to find out the
performance output and welding conditions.
Key words: Friction stir welding, Aluminium alloy, Tensile Strength, Artificial neural
networks, Regression analysis
INTERNATIONAL JOURNAL OF MECHANICAL ENGINEERING
AND TECHNOLOGY (IJMET)
ISSN 0976 – 6340 (Print)
ISSN 0976 – 6359 (Online)
Volume 4, Issue 2, March - April (2013), pp. 245-253
© IAEME: www.iaeme.com/ijmet.asp
Journal Impact Factor (2013): 5.7731 (Calculated by GISI)
www.jifactor.com
IJMET
© I A E M E

246
I. INTRODUCTION
Friction stir welding (FSW) is a solid state method developed by the welding institute
(TWI) (Thomas, 1991) and now being increasingly used in the welding of aluminum.
Aluminum alloys find wide applications in aerospace, automobile industries, in ship building,
in train wagons and trams, in offshore structures and in bridge constructions due to its light
weight and higher strength to weight ratio (Dawes, 1995). FSW is a innovative solid phase
welding process in which there is no melting of the material [1]. Hence, FSW is preferred
over the commonly used fusion welding techniques for the advantages such as: there are no
voids and cracking in the weld, there is no distortion of the work piece, no need of filler
materials, no costly weld preparation required, no shielding gas is required during FSW
process (Thomas et al., 1997). It is a clean and environment friendly process because there
are no harmful effects like arc formation, radiation, release of toxic gas etc. FSW is perhaps
the most remarkable and potentially useful welding technique. However, during FSW process
using inappropriate welding parameters can cause defects in the joint and deteriorate the
mechanical properties of the FSW joints (Cavaliere et al., 2008). Although FSW consistently
gives high quality welds, proper use of the process and control of number of parameters is
needed to achieve this. To produce the best weld quality, theses parameters have to be
determined individually for each new component and alloy (Wei et al., 2007). The quality of
friction stir welded joint is controlled by three welding parameters, these are Tool’s Rotation
Speed, Welding Speed and Axial Force.
In FSW a rotating tool moves along the joint interface, generating heat and resulting
in a re-circulating flow of plasticized material near the tool surface [2-5]. This plasticized
material is subjected to extrusion by the tool pin rotational and traverse movements leading to
the formation of the so called stir zone. The formation of the stir zone is affected by the
material flow behavior under the action of the rotating tool. The FSW process is applied
presently for welding aluminum and magnesium alloys as well as copper, steel, composites
and dissimilar materials [6-10]. Welding of aluminum alloy especially heat treatable wrought
aluminum alloy of AA6XXX aluminum by FSW produces better quality [11].
II. PLAN OF INVESTIGATION
This investigation was planned to be carried out in following steps (i) Identifying the
important process parameter and finding the range of process parameter such as tool
rotational speed, welding speed and Axial force. (ii) Collection of experimental data. (iii)
Developing of Regression model (Developing mathematical model and checking the
adequacy) and predict the Tensile strength as the function of input parameters. (iv)
Developing ANN model, Training and Testing of Neural Network to predict the Tensile
Strength values. (v) Comparing and concluding about different pin profiles.
The important processes parameters (Tool rotational speed, welding speed, Axial
force) and tool probe (pin) geometry were identified based on series of trials and author’s
earlier study. Parameters such a way that the friction stirred welded joint should be free from
any visible external defect. The selected process parameters with their levels are given in
Table 1. The experiment was based on three factors with three levels of full factorial
experimental design. As prescribed in the Experimental design matrix twenty seven joints
were carried out using Two different probe geometry by considering three levels of process
parameter, namely tool rotational speed and welding speed as given in the Table 2. It is to be

247
further noted that the experiments were conducted with a constant tool rotational speed,
welding speed and axial load for two pin profiles.
Table 1: Process parameters and their levels taken for AA6061 material
Level Tool speed
(rpm)
Welding speed
(mm/min)
Axial force
(K.N)
1 1200 48 1.5
2 1600 60 2.0
3 2000 72 2.5
Table 2 Experimental design matrix and results
Trial
no.
Coded value Real value
Tensile strength
(N/mm2)
N S F
Tool
speed
(rpm)
Welding
speed
(mm/min)
Axial
force(K.N
)
For
Conical pin
For
Triangular
pin
1 1 1 1 1200 48 1.5 61.4 113.39
2 1 1 1 1200 48 2 66.7 172.47
3 1 1 1 1200 48 2.5 40.5 171.11
4 1 2 2 1200 60 1.5 58.3 187.36
5 1 2 2 1200 60 2 45.3 101.28
6 1 2 2 1200 60 2.5 38.8 168.08
7 1 3 3 1200 72 1.5 64.6 129.77
8 1 3 3 1200 72 2 65.3 44.61
9 1 3 3 1200 72 2.5 40.8 153.19
10 2 1 2 1600 48 1.5 59.6 142.25
11 2 1 2 1600 48 2 75 184.14
12 2 1 2 1600 48 2.5 79.5 156.46
13 2 2 3 1600 60 1.5 60.7 181.92
14 2 2 3 1600 60 2 65.3 156.77
15 2 2 3 1600 60 2.5 62.9 120.43
16 2 3 1 1600 72 1.5 65.6 175.49
17 2 3 1 1600 72 2 62.7 135.82
18 2 3 1 1600 72 2.5 75.6 88.92
19 3 1 3 2000 48 1.5 97.8 127.11
20 3 1 3 2000 48 2 86.2 91.82
21 3 1 3 2000 48 2.5 75.57 129.27
22 3 2 1 2000 60 1.5 86.08 137.86
23 3 2 1 2000 60 2 73.8 119.45
24 3 2 1 2000 60 2.5 85.71 108.88
25 3 3 2 2000 72 1.5 90.22 109.25
26 3 3 2 2000 72 2 89.5 124.76
27 3 3 2 2000 72 2.5 88.3 89.72

248
III. PREDICTION OF TENSILE STRENGTH
A. Development of mathematical model by Regression analysis
Tensile strength of the joints is the function of rotational speed, welding speed, and axial
force and it can be expressed as
Y = f (N, S, F) ---------(1)
Where
Y-The response, N- Rotational speed (rpm), S- Welding Speed (mm/s) , F - Axial Force
(tones).
For the three factors, the selected polynomial (regression) could be expressed as
Y = k+ aN + bS + cF ----------(2)
Where k is the free term of the regression equation, the coefficients a, b, and c are linear
terms
Table 3: Estimated regression coefficients of mathematical models
Regression
coefficients
Tensile Strength N/mm2
For Conical Pin For Triangular Pin
K -2.65 8.83
A 0.961 -0.197
B -0.014 -0.581
C -0.232 -0.228
MINITAB 15 Software Packages is used to calculate the values of those coefficients for
different responses and is presented in Table 3. After determining the coefficients, the
mathematical models are developed. The developed final mathematical model equations in
the coded form are given below:
For Conical pin profile
Tensile strength = - 2.65 + 0.961 (N) - 0.014 (S) - 0.232(F) --------- (3)
For Triangular pin profile
Tensile strength = 8.83 - 0.197 (N) - 0.581 (S) - 0.228 (F) ----------(4)
The validity of regression models developed is tested by drawing scatter diagrams. Typical
scatter diagrams for all the models are presented in Figures 6 and 7. The observed values and
predicted values of the responses are scattered close to the 45° line, indicating an almost
perfect fit of the developed empirical models [12]

249
Figure 2: Scatter diagram for conical pin Figure 3: Scatter digram for Triangular pin
B. Artificial Neural Network (ANN)
ANNs are computational models, which replicate the function of a biological
network, composed of neurons and are used to solve complex functions in various
applications. Neural networks consist of simple synchronous processing elements that are
inspired by the biological nerve systems. The basic unit in the ANN is the neuron. Neurons
are connected to each other by links known as synapses; associated with each synapse there is
a weight factor. Details on the neural network modeling approach are given in elsewhere[13].
One of the most popular learning algorithms is the back-propagation (BP) algorithm. In this
present study, BP algorithm is used with a single hidden layer improved with numerical
optimization techniques called Levenberg Marquardt (LM) [14]. The architecture of ANN
used in this study is 3-20 -1, 3 corresponding to the input values, 20 to the number of hidden
layer neurons and 1 to the output. The topology architecture of feed-forward three-layered
back propagation neural network is illustrated in Figure 4 below.
Figure 4: Architecture of feed forward three layered back propagation neural network
30
50
70
90
110
30 50 70 90
RegressionpredictedT.Svalues
Experimental T.S values
40
60
80
100
120
140
160
180
200
40 60 80 100 120 140 160 180 200
RegressionpredictedT.Svalues
Experimental T.S values

250
MATLAB 7.1 has been used for training the network model for tensile strength prediction.
The training parameters used in this investigation are shown in Table.5 The neural network
described in this paper, after successful training, will be used to predict the tensile strength of
friction stir welded joints of 6061 aluminium alloy within the trained range.
Table 4: Training parameters used in ANN
Number of input nodes 3
Number of hidden nodes 20
Number of output nodes 1
Learning rule Levenburg marquatt
Number of epochs 5000
Mu 0.01
IV. ANN RESULTS
The results obtained after training and testing of Artificial Neural Networks are shown in the
tables below
Table 5: Test conditions at different hidden neurons (TS) for Conical pin profile
Tool
speed
(rpm)
Welding
speed
(mm/min)
Axial
force
(KN)
Exp.
Value
N/mm2
Predicted Tensile Strength N/mm2
@ 20N DEV @25N DEV @30N DEV @35N DEV
1200 48 2.5 40.5 64.19 -0.585 65.01 -0.605 60.62 -0.497 48.08 -0.187
1200 60 2.5 38.8 71.46 -0.842 42.62 -0.098 59.85 -0.543 58.08 -0.497
1200 72 2.5 40.8 78.16 -0.916 55.63 -0.363 66.82 -0.638 64.95 -0.592
1600 48 1.5 59.6 73.38 -0.231 85.22 -0.430 72.85 -0.222 64.38 -0.080
1600 60 1.5 60.7 72.84 -0.200 69.09 -0.138 65.25 -0.075 64.94 -0.070
2000 48 2.5 75.57 86.58 -0.146 89.72 -0.187 87.91 -0.163 78.15 -0.034
2000 60 2 73.8 89.02 -0.206 89.09 -0.207 87.66 -0.188 86.15 -0.167
AVERAGES 55.68 76.51 -0.447 70.91 -0.290 71.56 -0.332 66.39

251
Table 6: Test conditions at different hidden neurons (TS) for Triangular pin profile
Comparison of Conical pin and Triangular pin profile tools
Figure 5: Conical pin profile vs Triangular pin profile experimental T.S values(N/mm2
)
From the above plot it is observed that Triangular pin profile tool is better than conical pin
profile tool which yields more (2 times) Tensile strength.
Tool
spee
d
(rpm)
Welding
speed
(mm/min)
Axial
force
(KN)
Exp.
Value
N/mm2
Predicted Tensile Strength N/mm2
@ 20N DEV @25N DEV @30N DEV @35N DEV
1200 60 2 101.28 174.76 -0.726 170.96 -0.688 152.03 -0.501 140.22 -0.384
1200 72 2 44.61 141.48 -2.171 143.12 -2.208 143.03 -2.206 125.42 -1.811
1600 60 2.5 120.43 155.83 -0.294 119.1 0.011 138.34 -0.149 131.54 -0.092
1600 72 2.5 88.92 134.76 -0.516 173.64 -0.953 134.74 -0.515 90.83 -0.021
2000 48 2 91.82 125.23 -0.364 137.34 -0.496 124.31 -0.354 119.23 -0.299
2000 60 2.5 108.88 124.91 -0.147 119.6 -0.098 126.25 -0.160 115.57 -0.061
2000 72 2.5 89.72 126.26 -0.407 118.49 -0.321 131.78 -0.469 118.68 -0.323
AVERAGES 92.23 140.46 -0.661 140.32 -0.679 135.78 -0.622 120.21

252
VI CONCLUSIONS
This paper has describes two models for Predicting the Tensile strength of friction stir
welded AA6061 aluminium alloy using Regression analysis and Artificial Neural
Network(ANN). From this investigation, the following important conclusions are derived.
1) a regression model is developed and the values obtained for the response strengths are
compared with measured values. it shows that the models are adequate without any
violation of independence or constant assumption.
2) ANN model has been developed for prediction of strength as a function of welding
parameters. The model has been proved to be successful in terms of agreement with
experimental results. The proposed model can be used in optimization of welding process
for efficient and economic production by forecasting the strength in welding operations.
3) The results that are obtained for the response Tensile strength in ANN model and
Regression analysis are compared with experimental data. Among these methods ANN
model is easier and effective methodology in order to find out the performance output and
welding conditions.
4) Finally it is observed that Triangular pin profile is better than Conical pin profile because
the obtained tensile strength values are higher for Friction stir welded joint.
REFERENCES
1. Ratnesh K. Shukla and Pravin K. Shah, “Comparative study of friction stir welding and
tungsten inert gas welding process” Indian Journal of Science and Technology Vol. 3
No. 6 (June 2010) ISSN: 0974- 6846
2. R. Nandan, T. DebRoy and H. K. D. H. Bhadeshia, “Recent advances in friction stir
welding–process, weldment structure and properties”, Progress in Materials Science,
vol.53, 980-1023, (2008)
3. W. M. Thomas, E. D. Nicholas, J. C. Needham, M. G. Murch, P. Temple- Smith and C.
J. Dawes, “Friction stir butt welding”, International Patent Application No. PCT/GB92/
02203, December 1991.
4. C. J. Dawes and W. M. Thomas, “Friction stir process welds aluminum alloys”, Welding
Journal 75, 41-44, (1996).
5. Jae-Hyung Cho, E. Donald, Boyce and Paul R. Dawson, “Modeling strain hardening and
texture evolution in friction stir welding of stainless steel”, Material Science Engineering
A, 398, 146-163. (2005).
6. Kalemba, S. Dymek, C. Hamilton and M. Blicharski, “Microstructural investigation of
friction stir welded 7136-T76511 aluminium”, Proceedings of the 13th International
Conference on Electron Microscopy,’EM2008’, Zakopane, 79 , (2008).
7. H. Uzun, C.D. Donne, A. Argagnotto, T. Ghidini and C. Gambaro, “Friction stir welding
of dissimilar Al 6013-T4 to X5CrNi18-10 stainless steel”, Materials and Design, 26, 41-
46, (2005).
8. P. Cavaliere, R. Nobile, F. W. Panella and A. Squillance, “Mechanical and
microstructural behavior of 2024-7075 aluminium alloy sheets joined by friction stir
welding”, International Journal of Machine Tools and Manufacture, 46, 588-594, (2006).

253
9. L. Litynska, R. Braun, G. Staniek, C. Dalle Donne and J. Dutkiewicz, “TEM study of the
microstructure evolution in a friction stir-welded Al Cu Mg Ag alloy”, Materials
Chemistry and Physics, 81, 293-295, (2003).
10. C. Yeni, S. Sayer, O. Ertugrul and M. Pakdil, “Effect of post-weld aging on the
mechanical stir and microstructural properties of friction welded aluminium alloy 7075”,
Archives of Materials Science and Engineering, 34, 105-109, (2008).
11. C. Hamilton, S. Dymek and M. Blicharski, “Mechanical properties of al 6101-T6 welds
by friction stir welding and metal inert gas welding”, Archives of Metallurgy and
Materials, 52, 67-72, (2007).
12. R. Palanivel, P. Koshy Mathews and N. Murugan “Development of mathematical model
to predict the mechanical properties of friction stir welded AA6351 aluminum alloy”
Journal of Engineering Science and Technology Review 4 (1) (2011) 25-31.
13. ZHANG Z, FRIEDRICH K. Artificial neural networks applied to polymer composites:
A review [J]. Composites Science and Technology, 2003, 63: 2029−2044.
14. ARCAKHOGLU E, CAVUSOGLU A, ERISEN A. Thermodynamic analyses of
refrigerant mixtures using artificial neural networks [J]. Applied Energy, 2004, 78:
219−230.
15. D.Muruganandam and Dr.Sushil lal Das, “Friction Stir Welding Process Parameters for
Joining Dissimilar Aluminum Alloys”, International Journal of Mechanical Engineering
& Technology (IJMET), Volume 2, Issue 2, 2011, pp. 25 - 38, ISSN Print: 0976 – 6340,
ISSN Online: 0976 – 6359.
16. U.S.Patil and M.S.Kadam, “Effect of the Welding Process Parameter in MMAW for
Joining of Dissimilar Metals and Parameter Optimization using Artificial Neural Fuzzy
Interface System”, International Journal of Mechanical Engineering & Technology
(IJMET), Volume 4, Issue 2, 2013, pp. 79 - 85, ISSN Print: 0976 – 6340, ISSN Online:
0976 – 6359.

Comparative study on different pin geometries of tool profile in friction stir welding

More Related Content

What's hot (18)

Viewers also liked (11)

Similar to Comparative study on different pin geometries of tool profile in friction stir welding (20)

More from IAEME Publication (20)

Recently uploaded (20)

Comparative study on different pin geometries of tool profile in friction stir welding