Jurnal Kejuruteraan 32(2) 2020: 171-186

https://doi.org/10.17576/jkukm-2020-32(2)-01

Applications of Machine Learning to Friction Stir Welding Process

Optimization
Tauqir Nasir a, Mohammed. Asmaela*,Qasim Zeeshana & Davut Solyalib
a
Department of Mechanical Engineering,
Faculty of Engineering, Eastern Mediterranean University, North Cyprus, Gazimağusa, Mersin 10, Turkey
b
Electric Vehicle Development Centre,
Eastern Mediterranean University, North Cyprus, Gazimağusa, Mersin 10, Turkey
*Corresponding author: [email protected]

Received 12 June 2019, Received in revised form 7 November 2019

Accepted 25 February 2020, Available online 30 May 2020

ABSTRACT

Machine learning (ML) is a branch of artificial intelligent which involve the study and development of algorithm
for computer to learn from data. A computational method used in machine learning to learn or get directly
information from data without relying on a prearranged model equation. The applications of ML applied in the
domains of all industries. In the field of manufacturing the ability of ML approach is utilized to predict the failure
before occurrence. FSW and FSSW is an advanced form of friction welding and it is a solid state joining technique
which is mostly used to weld the dissimilar alloys. FSW, FSSW has become a dominant joining method in aero-
space, railway and ship building industries. It observed that the number of applications of machine learning
increased in FSW, FSSW process which sheared the Machine-learning approaches like, artificial Neural Network
(ANN), Regression model (RSM), Support Vector Machine (SVM) and Adaptive Neuro-Fuzzy Inference System
(ANFIS). The main purpose of this study is to review and summarize the emerging research work of machine
learning techniques in FSW and FSSW. Previous researchers demonstrate that the Machine Learning applications
applied to predict the response of FSW and FSSW process. The prediction in error percentage in result of ANN
and RSM model in overall is less than 5%. In comparison between ANN/RSM the obtain result shows that ANN is
provide better and accurate than RSM. In application of SVM algorithm the prediction accuracy found 100% for
training and testing process.

Keywords: Machine learning; Artificial Neural Network; Support Vector Machine; ANFIS; Response Surface
Methodology

INTRODUCTION (Software), New generation of Machine Learning

algorithms, Deep Learning and Reinforcement
Machine Learning (ML) is a branch of Artificial Learning, Advances in Sensor Technology (Data),
Intelligent. It is an approach, which allows High-performance and cheap sensors, Large
computers to do which comes naturally from human, amounts of data (Pokutta, 2016)
learn from experience. As the number of samples for Since 2006, deep learning emerged as
learning increase, performance of algorithm expeditiously growing research field which explore
adaptively improves (Alpaydin, 2004). ML firstly the performance in a wide range of areas like
gained concentration after (Arthur, 1959) published machine translation, image segmentation, speech
his paper “Some Studies in ML Using the Game of recognition, and object recognition. Deep learning
Checkers”. Since then, ML continuously flourish in began from ANN which is branch of a ML. Most
the field of research but also it grew with more deep learning methods implies the neural network
divers. In the field of smart manufacturing ML has architecture that why some time represented as deep
capability to solve problems of NP-complete nature neural network. Deep learning exploit the technique
(Lászlo Monostori, Jozsef Homyak, Csaba Egresits, of multiple non-linear processing layers for
1998). ML has ability to learn and adapt changes supervised or unsupervised and tries to learn from
therefore no need to predict and provide solution for hierarchical description of data. The application of
all situation (Alpaydin, 2010).The major strength of deep learning is available in all industries from
ML to learn from and adapting automatically to automated driven to medical devices (Deng, 2014).
changing environment (Lu, 1990; Simon, 1983).The (Wuest, Weimer, Irgens, & Thoben, 2016)
major factors that enhanced the capability and distinguished the supervised and unsupervised ML
accelerated the applications of ML i.e. Advances in algorithm. SVM found good for most manufacturing
Computing (Hardware), Advances in Algorithms applications because of mostly manufacturing
172

application provide labeled data. In manufacturing ML is finding applications in every field systems
the SVMs is most commonly used algorithm in some commercially available fields of study are face
supervised machine learning. ML is a powerful tool recognition, image processing, manufacturing, and
and its value will enhance more in the coming days. medical and in many more areas.

MACHINE LEARNING

SUPERVISED LEARNING UNSUPERVISED LEARNING

CLASSIFICATION REGRESSION CLUSTERING

Support Vector Linear Regression K-Means, K-Medoids

Machines GLM Fuzzy C-Means

Discriminant Analysis SVR, GPR Hierarchical

Naive Bayes Ensemble Methods Gaussian Mixture

Nearest Neighbor Decision Trees Hidden Markov Model

Neural Network Neural Network Neural Network

FIGURE 1. Machine Learning Techniques

Supervised learning in the form of regression

Recently many authors applied ML techniques (continuous output) and classification (discrete
in manufacturing (Alpaydin, 2010; Dingli, 2012; output) is an important part of statistics and ML,
Gordon & Sohal, 2001; Pham & Afify, 2005; Shiang either for data analysis or sub goal of complex
& Nagaraj, 2011; Susto et al., 2015; Thomas, Byard, problems. (Verma, Gupta, & Misra, 2018) presented
& Evans, 2012). The following are the major the methodologies of machine learning approaches,
advantages of ML in manufacturing: ML technique Gaussian process regression (GPR), SVM, and MLR
in manufacturing systems provide an improved for UTS of FSW joint to investigate the incongruity
quality control optimization (Apt, Weiss, & Grout, between the predicted and experimented outcomes.
1993) Handling of high-dimensional, multi variate The applications of ML can be enforced in the
data, extract implicit relationships within large data domains of all industries. ML approaches
sets in a complicated, dynamic, and anarchic implemented in procedural compliance,
environment (Köksal, Batmaz, & Testik, 2011; documentation of process and orientation, risk and
Rostami, Dantan, & Homri, 2015; Yang & Trewn, quality frameworks of manufacturing industry. The
2004) improve understanding of expertise to arrange ML also used in cloud computing, data science and
powerful tools for constant improvement of complex in IoT. The ability of ML to predict the failure
process (Lászlo Monostori, Jozsef Homyak, Csaba before occurrence is a useful feature and some
Egresits, 1998; Pham & Afify, 2005). Only those manufacturing firms already using in production to
ML algorithm are applicable in manufacturing minimize the financial losses, as well as risk loss
which are able to handle high dimensional data. The (Kashyap, 2017). (Yucesan, Gul, & Celik, 2018)
usability of application of algorithms enhanced due explored the furniture manufacturing industry in
to ML program. The main benefit of ML algorithm Turkey, by applying an Auto regression Integrated
to find formerly anonymous implicit expertise and Moving Average with external variables
point out implicit connection in data (Alpaydin, (ARIMAX) model develop to predict total monthly
2010; Bar-or, Schuster, & Wolff, 2005; Do, Lenca, sales of furniture product of a manufacture.
Lallich, & Pham, 2010). (Malviya & Pratihar, 2011) utilized particle swarm
(Rasmussen, 2004) provided the general optimization (PSO) method for the tuning of neural
presentation on Gaussian process regression models network by using both front and back mappings of
and focused on the role of the stochastic process and metal inert gas (MIG) welding process. (Mian et al.,
how to define a distribution over function. 2005) worked on dissimilar material, especially
173

solid-state welding techniques, which can escape linear kernel function, polynomial kernel, RBF and
many of the issues such as excessive heat input, Sigmoid. Mechanical properties model for welded
fume generation, cracking, and indigent joint joint built to use SVR network and make
properties that are commonly confront when assumption. A comparison done between the
compare with fusion welding. While aluminum and prediction based SVR result and on ANFIS. The
steel are not compatible during fusion welding, FSW obtained results marks that the anticipated precision
is consider most convenient joining method for relay on SVR with radial RBF gave higher value
various alloys as well as for the combinations of than the other three kernel functions and that depend
dissimilar metals. on ANFIS.
Based on the recent studies the popularity to
study the machine learning technique in FSW
process is increasing. Now researchers are
implementing these ML techniques in FSW/FSSW
processes to forsee the actual and predicted response
of the process parameters. The main purpose of this
review paper is to gather all the implemented and
suggested approaches in one platform.

Artificial Neural Network (ANN)

FIGURE 2. Model of FSW (Nataliia, Erik, Igor, &
Klaus, 2019)
Neural Network (NN) technology is substantial
branch of statistical ML and repeatedly been
implemented in various kinds of prediction tasks.
(Kutsurelis, 1998) ANN inspired by natural NN.
ANN is a computer program that develop to obtain
information in a similar manner like human brain.
Artificial intelligence, is a combination of neural
networks which developed due to research on
cognitive talent and machinery design. The ability
of ANNs to resolve forecasting problems bring
FIGURE 3. Schematic illustration of FSSW process appreciable research attention because ANN
substantially beat previous implemented techniques
Currently, FSSW consider an alternate method for anticipating based on non-linear input variables.
to resistance spot welding (RSW) due to the (Ekici & Aksoy, 2009; Kandananond, 2011; Li, Hu,
meaningful energy and cost savings. Comparatively Liu, & Xue, 2015; Mena, Rodríguez, Castilla, &
to the process of resistance spot welding, FSSW had Arahal, 2014; Qamar & Khosravi, 2015) ANNs are
created huge interest in automotive manufacturing intensely favorable at modeling the nonlinearities in
industry (Abdullah & Hussein, 2018). The demand data of many fields and have theoretically provable
of lightweight materials like aluminum alloys are ability with arbitrary precision to approximate
getting more attractions in the field of automotive, complex functions.
shipbuilding, aerospace, transport, military and (Bennell & Sutcliffe, 2003) ANN is a tool that
many other industries because of extensive features, commonly used for prediction and categorization in
like high formability, high strength to weight ratio data processing that is inspired from the attribute of
and better corrosion resistance. However biological neuron system that learns by experience.
comparatively to ferrous alloys the joining method It has many features that make him attractive for
of aluminum alloy and other light weight alloys are problems such as pricing option which has the
difficult by conventional processes due to their high capability to develop a nonlinear model relationship
thermal conductivity, hydrogen solubility, high that do not depend on the restrictive assumption
thermal expansion and aluminum oxide formation implied in parametric approach, nor does depended
(Verma et al., 2018). (Nourani, Milani, & on the specification of theory that connects the price
Yannacopoulos, 2011) the forthright and of underlying assets to the price of option. The
computationally effective methodology for successful implementation of ANN models
optimizing the FSW process parameters of 6061 considered when it has ability to learn a lesson from
aluminum alloy. The achieved results confirm that the provided data and use in new one. The ANNs
the method can be successfully used for minimizing model strength lies in relationship between the input
both the HAZ distance to the weld line and peak and output variables that may be complex and
temperature. (Shuangsheng, Xingwei, Shude, & difficult to get from mathematical formulation.
Zhitao, 2012) applied the SVR network based on (Staub, Karaman, Kaya, Karapınar, & Güven, 2015)
 174

explore the features that ANN are the most and simulation between the FSW parameters of
important tool to solve the complex nonlinear aluminum plates and mechanical properties. Two
problems. ANN modify their own values and they different training algorithms of NN utilized in this
have the ability to adapt themselves for the exact study: 1) Gradient descent with momentum
solution of the problem. During the training process, algorithm: 2) Levenberg-Marquardt (LM)
ANNs are able to create the desire response. algorithm. The obtain result exhibit that the
recommended ANN (LM) algorithm shows better
Welding
performance than other because it implies 2nd order
Speed Taylor series instead of 1st order approximation as
Rotational
with gradient descent algorithm. (Maleki, 2015)
Speed Yield investigated the modeling of FSW effective
Strengt
parameters on thirty AA-7075-T6 specimens by
Axial Tensile using ANN. The network established on back
Force Strength
propagation (BP) algorithm. In this study, the TRS,
Shoulder
Diameter Notch-Tensile TS, axial force, pin diameter, shoulder diameter and
Pin
Strength tool hardness considered as input parameters of
Diameter Welding
ANN. On the other hand, TS, YS, and welding zone
Tool Zone hardness, notch tensile strength used as outcome of
Hardnes Hardness NN see figure 4 . The acquired result demonstrate
s that the forecast hardness values of welding zone,
notch tensile strength, TS and YS have the least
FIGURE 4. A conceptual Structure of ANN model mean relative error (MRE). The connection of
anticipated result and the experimental results shows
(Yousif, Daws, & Kazem, 2008) implemented
that the ANN modeling is very effective for FSW
the ANN to predict the correlation for the analysis
parameters.

TABLE 1. Applications of ANN techniques to various FSW and FSSW processes

Author Material DOE Process Models Input Output Remarks

parameters
(Vaira AA1100 FCC FSW ANN TRS (rpm), TS,  The error Percentage
Vignesh & composi (LM) WS (mm/s) prediction found to be low in
Padmanaban te design algorithm and ANN, developed model.
, 2018) with five with feed shoulder  The overall model interaction
level forward diameter coefficient is 0.8214
variation model (mm)  Which shows closeness in
relationship between the FSW
process parameters and in TS.
(Wakchaure, AA Taguchi FSW ANN TRS, WS TS and  The hybrid Taguchi GRA of
Thakur, 6082-T6 based and tilt impact ANN Method provides grey
Gadakh, & GRA angle strength relation grade 0.508.
Kumar,  Hybrid Taguchi GRA is
2018) 9.70% higher than traditional
analysis of Taguchi grey.
(Kurtulmu & high- x FSSW Feed TRS (rpm), lap-shear  Outputs ANN models
Kiraz, 2018) density forward PD (mm), fracture compared with the actual
polyethy back DT (s) load (N) values.
lene propagatio  Best prediction performance
(HDPE) n ANN achieved with 100% training
sheets model set and 20 neurons in the
hidden layer.
(Ranjith, AA2014 x FSW ANN with Pin TS  The based on ANN model
Giridharan, T651 (LM) Diameter optimized process parameter
& Senthil, and algorithm (mm), Tool are 7mm pin diameter and 4
2017) AA6063 (mm), degree tilt angle.
T651 Geometer,  Better TS exhibits when tool
Tool Offset is offset towards advancing
side.
 ANN predict the TS with an
accuracy of 98% with 2%
error.
 175

(Dehabadi, AA6061 x FSW ANN Threads, Vickers  For training and test data sets
Ghorbanpou model tool tilt micro Mean absolute percentage
r, & Azimi, with Two angle , and hardness error (MAPE) did not exceed
2016) feed welding from 5.4% and 7.48%,
forward distance respectively.
(BP) from  In MAPE training process the
centerline both predict values for ANNs
were less than 4.83%
 Mathematical modeling
techniques i.e ANN can save
time, material, costs, and
results in optimized designs.
(Anand, Incoloy x FSW ANN Heating TS, micro  GANN process urged for
Barik, 800H based on pressure hardness( keeping both forward and
Tamilmanna (BBPNN), (HP), H) and reverse mappings.
n, & Sathiya, (IBPNN), heating time burn off  The averages RMSE of
2015) (QPNN), (HT), length training, testing and
(LMNN) upsetting (BOL) validation data are
and pressure 0.9628,1.2148 and 1.2196
(GANN) (UP) and respectively
upsetting  The RMSE for validation data
time (UT) is 1.2196, the coefficient of
determination is 0.9899 and
the R2 is 0.9978.
(Paoletti, Polycarb x FSSW ANN plunge rate, plunging  Strong correlation is observed
Lambiase, & onate composed TRS, and force between experimental results
Di Ilio, sheets by three DT (Fmax), and ANN predicted values.
2015) layers torque  Result confirmed by low
(Cmax), values of the Mean Absolute
temperatur Errors (MAE).
e (Tmax),  which is ~2% for Fmax, ~3%
heat for Cmax and ~5% for Fr.
resistance
(Fr) of
joint
(Ghetiya & AA8014 x FSW ANN with TRS (rpm), TS  The measured and anticipated
Patel, 2014) (BP) WS (mm/s), values almost close to each
algorithm and PD other.
(mm)  Overall R2 value for training,
validation and testing is
bigger than 0.99.
 ANN design 4-8-1 has less
than 3% error between
experiment and predicted
result.
(Shojaeefard AA7075 x FSW (ANNs) TRS (rpm), UTS and  ANN disclose a better
, Behnagh, -O / feed WS (mm/s) hardness interaction between the
Akbari, AA5083 forward predicted data and the acquire
Givi, & -O NN with data
Farhani, BP  Linear regression analysis is
2013) algorithm performed to obtain the R2
and multi among the experimental and
objective anticipated values.
particle  The R2 result for UTS and
swarm hardness at training and
optimizati testing were respectively as
on 0.999 and 0.9916 and 0.9799
(MOPSO) and 0.9891.
(Manvatkar, AA L50 FSW ANN TRS, TS, Total  The uncertainties in
Arora, De, & 7075 Taguchi pin radius, torque, prediction of ANN models
DebRoy, array tool WS, peak alter from 2.5% for peak
2012) and shoulder, temperatur temperature and 7.5% for
CCD axial force, e, bending torque, Max shear stress and
design pin length. stress and traverse force.
Max
 176

shear  Bending stress within the

stress training data range vary up to
12%.
 Training data sets values are
exceeding when calculated in
range up to 20%.
 The maximum prediction
value of uncertainties for peak
temperature is 4%.
 Maximum shear stress and
torque has 12%, traverse force
15% and 20% for the bending
stress.
(Buffa, Ti–6Al– x FSW ANN and TRS (rpm), Microstruc  Two different neural network
Fratini, & 4V multi WS ture, and trained under different
Micari, titanium objective (mm/min), Micro process parameters for the
2012) alloy optimizati Tilt angle, hardness calculation of post weld micro
on PD (mm) hardness and microstructure.
 A delightful agreement found
for the prediction of micro
hardness
 An excellent prediction
capability of neural network
achieved regarded to
microstructure.
(Okuyucu, Hot x FSW ANN (BP) TRS (rpm), Hardness  The RMS error values for
Kurt, & rolled algorithm WS (mm/s) (HV), Hardness of HAZ, weld
Arcaklioglu, aluminu with Weld metal, %EL, yield Strength,
2007) m plates numerical metal, TS are 0.0115, 0.0064,
technique %Elongati 0.0566, 0.0253, and 0.018
(SCG) and on, YS, TS respectively.
(LM)  The R2 values are bigger than
0.99 except elongation that is
0.985

optimal TRS and TS proposed by using Genetic

The application of ANN in manufacturing used algorithm (GA). Only one ANN model assigned for
like cold forging to predict the flow stress during hot five performance parameters of welding zone, TS,
deformation, for tool wear, for machining behavior YS, elongation, weld metal hardness and hardness of
prediction and manufacturing process optimization HAZ. The input were same for five ANNs (TRS and
along with other process (Ghetiya & Patel, 2014). TS). The error estimation of the ANNs were superior
(Tansel, Demetgul, Okuyucu, & Yapici, 2010) to average 0.5%. (Boldsaikhan, Corwin, Logar, &
proposed FSW operation by using ANNs and choose Arbegast, 2011) introduced a real-time novel
the optimal tool rotational speed and feed rate by technique to detect the wormhole imperfection in
using genetic algorithm. The selection of GONNS FSW in a nondestructive method. In a way by
for modeling the stir welding process founded a utilizing the discrete Fourier transformation and the
viable option for optimal solutions. (Shojaeefard, multilayer neural network to figure out the provided
Akbari, & Asadi, 2014) conduct the ANN analysis feedback forces by welding method. By trial and
to model the correlation between the tool parameters error a near optimum neural network value achieved.
(pin and shoulder diameter) and heat-affected zone, A classified testing result of 95% achieved with 60
thermal, and strain value in the weld zone. (Fratini, input unit by optimum neural network, with 9 hidden
Buffa, & Palmeri, 2009) linked ANN to a finite units, and one output unit. A validation of
element model (FEM) and predicted the average experiment conducted to proof the generality of NN
grain size values of butt, lap and T type FSW joints. to characterize the weld quality. The suggested
(Jayaraman, Sivasubramanian, Balasubramanian, & algorithm spent about 0.01 s on a 2700 MHz
Lakshminarayanan, 2008) ANN modelling machine. (Khourshid, El-Kassas, & Sabry, 2015)
predicted the TS of A356 alloy which is a high investigated the mechanical properties to show the
strength Aluminum. feasibility of FSW of Al 6061 on pipe. To conclude
(Tansel et al., 2010) applied genetically the TS, the %EL and hardness of FSW weld of
optimized neural network system (GONNS) to AA6061 aluminum ANN and RSM implanted. The
evaluate the optimal operation condition of FSW obtained results of ANN and RSM model proved
process. The characteristics of FSW operation by prosperous in term of settlement with experimental
using ANNS and the selection of parameters like result ratio of 93.5% and 90%.
177

Layer 1 Layer 2 Layer 3 Layer 4 Layer 5

ANFIS Modeling

A1 W1 ഥ𝟏
𝐖 x Y
The acronym ANFIS derives its name from adaptive
neuro-fuzzy inference system. By Utilizing a X Π Ɲ ഥ 𝟏𝐩𝟏
𝐖
provided input/output data set, the toolbox function A2
F
ANFIS build a fuzzy inference system (FIS) whose
function membership framework are tuned by B1
Σ
utilizing only a back propagation (BP) algorithm or Y
merger with a least squares type of method. This Π Ɲ
B2 ഥ 𝟐𝐩𝟐
𝐖
adaptation of fuzzy systems allow him learn from W2 ഥ𝟐
𝐖
the data they are modeling. The ANFIS learning x Y
method works similarly as neural networks. Fuzzy
modeling method obtained from neuro-adaptive
FIGURE 5. ANFIS design model of two input rules
learning technique to learn information about a data
set (Jang, 1993). Fuzzy modeling is established on
(Babajanzade Roshan et al., 2013) tested the
Fuzzy implications and interpretation are one of the
various arrangement of ANFIS model for each
most important field in Fuzzy system approach. The
response (TS, YS, and hardness). Mechanical
Fuzzy model build on input-output data that
classified into two things, first is mathematical tool properties of FSW process predicted by two stage of
to show a model system and the second is ANFIS model, training and testing. The obtained
structural model shown in figure, model that has a
identification method. The Fuzzy implication
broad number of MFs shows overfitting and not be
depend on input space of a Fuzzy partition. A linear
able to develop the ambitious value of root mean
input-output relation formed in each Fuzzy
square error (RMSE). In the accuracy of ANFIS
subspace. A fuzzy meddling system apply fuzzy if-
model the influential factor that affect is the class of
then rules which can model the qualitative aspects of
membership function. It can conclude that a good
human expertise and analysis processes without
relationship established between the predicted value
exploit accurate quantitative investigation. First time
Fuzzy modeling is systematically explore by of ANFIS and experimental value.
(Takagi & Sugeno, 1985).

TABLE 2. Applications of ANFIS techniques to various Friction Stir Welding processes

Author Material DOE Proce Models Input Output Remarks

ss parameters
(Shanavas AA CCCD FSW Fuzzy Tool pin Weld  Fuzzy model anticipate an
& Dhas, 5052 with 4 logic geometry, quality acceptable output with less than
2018) H32 factors, model TRS, WS, 4% error.
5 levels with four and tool tilt  Regression model foresee the
stages angle. outcome less than 7%.
and RSM  The obtained result shows
calculated F-ratio value is higher
than tabulated F-ratio at 95%
confidence level reveals that
model is adequate.
(Barath, AA202 X FSW Sugeno- TRS and TS  The ANFIS is trained by utilizing
Vaira 4- Fuzzy WS the training data in Sugeno
Vignesh, AA707 logic inference system.
& 5 mechanis  For the experimental data 80% is
Padmanab m uses used for training and testing and
an, 2018) ANFIS remaining used for testing and
validation.
 The ANFIS model demonstrate
that TRS of 1050 rpm and WS of
15 mm/min-1 are highly influential
parameters in FSW to generate
optimum heat for grain refinement
getting peak TS.
178

(Dewan, AA- X FSW ANN and TRS, UTS  For the development of ANFIS
Huggett, 2219- ANFIS WS,PD, and and ANN all four input
Warren T87 model empirical parameters were utilized and
Liao, with force index optimized and obtain result show
Wahab, & approach (EFI) that EFI has strong relation with
Okeil, of leave- UTS compared to others.
2016) one-out  ANFIS predict better results than
cross- ANN in term of RMSE and
validation MAPE 29.7 MPa and 7.7% in
(LOO- ANFIS model and 36.7 MPa and
CV) 10.09% in ANN model
respectively.
(Babajanz AA707 CCD FSW ANFIS TRS (rpm), TS, YS,  The prediction error for TS is
ade 5 with 4 with WS (mm/s), and 3.21% for single response and
Roshan et factor simulated Axial force Hardne 2.24% for multi-response.
al., 2013) 5 levels annealing (N), Tool ss  The prediction response for YS
pin and Hardness for Single and
geometry multi-response are
respectively.2.27% and 3.1% for
single and 3% and 3.8% for multi-
response.

(Satpathy, Mishra, & Sahoo, 2018) developed variables. The cooperation between dependent
the regression model, ANN, and ANFIS to simulate variable and predictor variables is develop as a linear
and predict the joint strength of Ultrasonic metal model in Eq. (1)
welding USMW of Al-Cu sheets. The result of ANN
and ANFIS investigation compared with outcome of Y=β0+∑βi Xi+ εi (1)
regression analysis. The obtained average absolute
error of TS for regression, ANN and ANFIS analysis In this formulation β0 ….. βp are the regression
are 0.47%, 0.15% and o.o7%. Similarly TP values coefficients to be predicted according to scrutiny. To
varied respectively 1.89%, 0.61% and 0.22% for prevent multicollinearity problems, interrelationship
regression, ANN and ANFIS. So result conclude between the predictors should be organized (the
that the ANFIS predict most accurate result than correlation coefficient of the descriptive variables
ANN and regression. The obtained R2 result for TS should not surpass 0.7) (Anderson, Sweeney, &
values for regression, ANN and ANFIS are 91.47%, Williams, 2011). The last term ε, designate the
99.30% and 99.98%. The R2 values for TP results in random error and is attribute as the residual for
ANFIS is 99.79% that more accurate than other two examining the overall influence of the model and
techniques. (Dewan, Huggett, Warren Liao, Wahab, each regression coefficient. Error term is
& Okeil, 2016) developed an optimized ANFIS independently and normally distributed, with a mean
model to anticipate UTS of FSW joints. Total 1200 of zero and a constant variance of σ2 (Douglas C.
models developed by changing the quantity of Montgomery & Vining, 2012). Regression models
membership function (MFs), types of MFs, and describe the relationship between the output values
mixture of input parameter which are spindle speed, and one or more input values. The multiple
plunge depth, welding speed and empirical force regression model is a parametric model. There are
index (EFI) by using MATLAB platform. An ANNs many statistic and machine learning method to
models was also develop for the comparison of UTS generate the result like, linear, generalized and
of FSW process. EFI founded a strong relation with nonlinear regression model, containing mixed effect
UTS relative to other parameters. The predicted model and stepwise models. The connection
results of both models ANFIS and ANN with three between the numeric predictor and continuous target
input variables WS, PD, and EFI resulted as approximates by simple linear regression by using
respectively in lowest RMSE 29.7 MPa and Mean straight line. Relationship between a set of P>1
Absolute percentage error (MAPE) of 7.7% in predictors and a single continuous target
ANFIS model on the other hand in ANN model approximates multiple regression modeling using a
minimum Root mean square error (RMSE) is 36.7 P-dimensional plane (Vardhan & Bayar, 2013). The
MPa and MAPE is 10.09% which is larger than main objective of regression model designing to
ANFIS model. select a best suitable regresses that can develop an
accurate response variable. Regression Trees (RT)
Regression model and ANN are ambitious techniques for modeling
regression problems. MLR is a classic method that
Regression analysis is the most frequently used provide many advantages: simplicity,
conventional prediction approach to recognize the interpretability, chances of being accommodated
connection between the dependent and independent over the transformations of the variables, and the
179

performing of reasoning, supposing the hypothesis RSM in partnership with electron back scattered
of normality, homoscedasticity and inter correlation microscopy (EBSD) and TEM to examine the
between the error ε and the predictor variables influence of parameters on tensile properties of brass
(Chakraborty, Chakraborty, & Chattopadhyay, plate.
2018). (Heidarzadeh, 2019) applied first time the

TABLE 3. Applications of RSM techniques to various FSW processes

Author Material DOE Process Models Input Output Remarks

parameters
(Jenarthanan, AA2014 Central- applying TRS (rpm), TS  RSM validated by using
Varun Varma, and Composit FSW RSM WS (mm/s), confirmation test and error found
& Krishna AA6061 e method pin diameter within ± 5%
Manohar, (CCD)  RSM is a power full tool in
2018) optimizing the FSW process
parameters.
 The difference between the
predicted and experimental strength
values are marginal± 5%. In
modelling and optimizing process
the RSM show better accuracy.
(Kadaganchi, AA 2014- CCD FSW RSM TRS (rpm), % EL,  2nd order response surface fitting
Gankidi, & T6. with four applied WS (mm/s), YS and model by using analysis of variance.
Gokhale, process Tilt angle, UTS  Regression equation developed on
2015) parameter Tool pin experimental values of YS, UTS
profile and %EL.
 Developed model utilized to predict
the response within ±10 % of
experimental values at 95%
confidence level.
(Elatharasan AA 6061- face- FSW RSM TRS (rpm), UTS,  The fitted quadratic model is
& Kumar, T6 centered WS (mm/s), TS, applied to get the response.
2013) CCD Axial force YS,  UTS, YS, and %E effectively
design (N) and predict the joint at 95% confidence
%EL level
(Karthikeyan AA2024- CCD FSSW RSM TRS (rpm), tensile  2nd order polynomial equation used
& T3 rotatable with PD (mm), shear to response the model.
Balasubraman four- Plunge rate fracture  To predict the TSFL of joint an
ian, 2010) factor, (mm/min), load empirical relationship developed
five-level DT (s) (TSFL) combine with welding parameters at
factorial 95% confidence level.
design  On TSFL plunge rate influence
greater than PD, DT, and TRS.

(Elatharasan & Kumar, 2012) applied the which disclose a good combination between the
quadratic model of RSM to evaluate the UTS, YS response data and independent variables.
and displacement of FSW joint. Multi objective
optimization by utilizing the RSM is a valuable
method to enhance the FSW parameters to achieve
optimum UTS, YS and displacement of a joint at
95% confidence level. (Srinivasa Rao & Ramanaiah,
2018) applied three factor central composite design
with five level to construct a mathematical
regression model for employing RSM. The
importance of process parameters studied by
implementing the RSM technique. The R2 values of
predicted model for hardness, UTS, %E, bending
strength, and impact strength are respectively
83.90%, 95.47%, 86.47%, 90.73% and 93.78%
180

TABLE 4. Comparison of ANN/RSM techniques to various FSW processes

Author Material DOE Process Models Input Output Remarks

parameters
(M. AA6063 Three FSW (ANNS) TRS YS,  ANN model cultivate to predict
Krishnan, -T6 and factor ,5 with (BP) (rpm), WS UTS, the exclusive input parameter
Maniraj, A319.0 level algorithm (mm/s), %EL, and reciprocal effect like TS,
Deepak, & (CCC) and RSM and axial and and hardness.
Anganan, design applied force(N) hardnes  Regression model developed on
2018) s experimental value of YS, TS,
%EL and hardness and develop
model validated for 95%
confidence level.
(Lakshminar AA7039 Three FSW Compariso TRS TS  More robust and accurate model
ayan & factor, n of RSM (rpm), WS found ANN in evolution of TS
Balasubram three and ANN (mm/s), values.
anian, 2009) level and Model and axial  When compared ANN with the
(CCC) force(N) RSM.
design  The mean errors for ANN and
RSM were 0.258, 847% and
0.769, 831% respectively
(Jayaraman Commer Central FSW ANN with TRS TS  In comparison ANN model
et al., 2008) cial composit BP (rpm), WS result are better and accurate
A356 e faced algorithm (mm/s), than RSM.
aluminu design and RSM and axial  ANN is good in estimating the
m (CCFD) with cause force(N) tensile strength values.
and effect  The obtained R2 value is
diagram 0.978398 of this model which is
only 3% less of the total
variation
 The lower value of coefficient
of variation (CV) is 2.556 which
shows improvement in
reliability and precision in
experiment.

(Dong, Cao, & Eang, 2005) enforced SVM to

Support Vector Machine (SVM)
forecast the energy utilization of buildings in a
tropical region. The obtained result have coefficient
In many machine learning tasks SVM is used, such of variance (CV) less than 3% and percentage error
as object classification, pattern recognition and in within 4%.
time series prediction, also containing forecasting of Figure 6 demonstrate the hyperplanes H1, H2,
energy consumption. SVR is a procedure for and H3 in which only H2 gain maximum margin.
regressions in support vector machine (SVMs). The p-1 dimensional hyperplanes that allocate
SVMs worked on the principle of structural risk vectors but only one hyperplane that can escalate the
belittlement. SVM build up one or more hyperplanes margin between two classes. Otherwise, the nearest
in a high dimensional space. The purpose of SVR is hyperplane between sides of this hyperplane is
diminish the probability of the model that produce maximized. Such hyperplane called maximum-
from input data set which will create an error on an margin hyperplane and recognize as the SVM
unseen data item. The objective is accomplished by classifier (Nguyen, 2017).
finding a solution which, best generalizes the
training examples. (Vapnik, 2000) SVM segregate
the data points into two classes. Each data point
apply to one of the two classes distinguished by a
linear classifier with a hyper plane. The data points
are separated into two classes by using various linear
classifier. To obtain best classification between the
two classes it is necessary to select the hyperplane
with utmost margin. SVM classify the testing data
points by choosing the hyperplane with maximum
margin. That utmost margin hyperplane is persistent
by a subset of data points called support vector.
181

linearly separable classes. Based on linear

X2 discriminant function SVM is the most popular
distributer. SVM is perfectly suitable for binary
H1 categorization. It extensively studied in data mining
H2 H3
and in pattern recognition applications. Over the past
three decades SVM turn into basic standard for
classification due to exceptional software packages
which developed consistently. (Cortes & Vapnik,
1995) explored the SVM classification problems for
two group. The machine theoretically implement an
idea: input vector non-linearly mapped to a very
high dimension feature space. Decision surface
X1 assemble in this feature space. The immense
generalization capability of the learning machine
FIGURE 6. Separating Hyperplane Adapting from ensures by special characteristic of the decision
(Nguyen, 2017) surface. Previously SVM network carried out for
limited condition where training data detached
SVM are capable of handling immense without any error. However, in this study the result
dimensionality greater than1000 very well. extended to non-separable training data. Due to the
However, accompanying concern like possible over- extension, SVM consider as a new method of
fitting has to be (Murty & Raghava, 2016; Widodo machine learning that is strong and comprehensive
& Yang, 2007; Yang & Trewn, 2004) applied the as neural networks.
SVM classification. Particularly the linear SVM
which is perfectly appropriate to pledge with

TABLE 5. Application of SVM techniques to various FSW processes

Author Material DOE Process Models Input Output Remarks

parameters
(Armans AA6061 X FSW SVM through TRS (rpm) TS  Performance evaluation and
yah & Kernel and WS (MPa) testation model developed
Astuti, function as (mm/s) for FSW.
2018) pattern  Prediction accuracy for TS
classification found 100% for training and
testing system.
(Armans AA5052 X FSSW SVM through TRS (rpm), Shear  SVM classification is
yah & -H112 Kernel PD (mm), Tensile implemented for pattern
Astuti, function as DT (s). load classification and model
2018) pattern development for system
classification model.
 The training and testing
process of FSSW joint
result found with 100%
accuracy.
(Bhat, AA1100 X FSW SVM TRS (rpm), Energy,  In comparison Gaussian
Kumari, classification WS Varianc kernel provide higher
Dutta, technique (mm/s), and e and accuracy than polynomial
Pal, & using PD(mm) Entrop kernel.
Pal, Gaussian and y  The obtain result classifying
2015) polynomial with Gaussian and
kernel polynomial kernel with
good and defective weld
with 99% and 97%
accuracy.

(Fleming et al., 2007) worked on fault detection in TS class which demonstrates the effectiveness and
FSW. Fault such as tool misalignment and excessive accuracy of this technique that can be implemented
flash can reduce the weld quality of the weld. SVM in a variety of other FSW fault detection scenarios.
based method implemented to identifies the (Zhu Lingyun, Cao Changxiu, Wu Wei, & Xu
presence of gaps and determine the gap depth. The Xiaoling, 2003) applied a new method of
predicted result accuracy found 100% for each computation intelligent using SVM to envision the
training and testing system, either for low TSor high bond welding quality. To develop SVM Classifier a
182

RBF picked as kernel function. The weld quality of Structures, 14(5), 908–922.
the FSW joints by SVM classifier is completely Alpaydin, E. (2004). Introduction to Machine
feasible. The new method perform exceptional than Learning. The MIT Press Cambridge,
traditional assessment methods with benefit of low Massachustts London,England.
cost, better efficiency and simple implementation on Alpaydin, E. (2010). Introduction to Machine
line. In precise prediction and generalization the Learning. Second Edition. The MIT Press
SVM classifier proved with better result than RBF Cambridge, Massachustts London, England
neural networks. This technique provides a novel Anand, K., Barik, B. K., Tamilmannan, K., &
approach for evaluation of nondestructive Sathiya, P. (2015). Artificial neural network
characteristic of friction welding joints modeling studies to predict the friction
welding process parameters of Incoloy 800H
DISCUSSION AND CONCLUSION joints. Engineering Science and Technology,
an International Journal, 18(3), 394–407.
Machine learning is a waste field which is https://doi.org/10.1016/j.jestch.2015.02.001
implemented in every field some commercially Anderson, D. R., Sweeney, D. J., & Williams, T. A.
available fields of study are face recognition, image (2011). David R. Anderson, Dennis J.
processing, manufacturing, and medical and in many Sweeney, Thomas A. Williams Essentials of
more areas. Mostly in the application of ANN Modern Business Statistics with Microsoft
algorithm in FSW process parameters the major Excel. Cengage Learning.
material that used is aluminum 5xxx, 6xxx and 7xxx Armansyah, & Astuti, W. (2018). Development of
series. The design of experimented (DOE) technique Prediction System Model for Mechanical
applied by few author which shows a lack of Property in Friction Stir Welding Using
systematical approach in process parameters. The Support Vector Machine ( SVM ). Journal of
error percentage in prediction the result in ANN and Mechanical Engineering, 5(5), 216–225.
RSM methods is less than 5% in overall. In case of Arthur, S. (1959). Some Studies in Machine
comparison between RSM/ANN it found that the Learning Using the Game of Checkers. IBM
ANN is more robust and accurate than RSM. The Journal, 3(3), 210–229.
result prediction accuracy in training and testing in Babajanzade Roshan, S., Behboodi Jooibari, M.,
SVM is 100% approximately in all present cases. In Teimouri, R., Asgharzadeh-Ahmadi, G.,
machine learning popularity the SVM technique Falahati-Naghibi, M., & Sohrabpoor, H.
overtook the ANN technique. In comparison (2013). Optimization of friction stir welding
between the regression model and ANFIS it found process of AA7075 aluminum alloy to
that ANFIS is more suitable in prediction the output achieve desirable mechanical properties
with error percentage less than 4%. In using ANFIS models and simulated
manufacturing the most commonly used algorithm annealing algorithm. The International
in supervised machine learning is SVMs. Journal of Advanced Manufacturing
In this study, an attempt made to highlights Technology, 69(5–8), 1803–1818.
all the machine-learning approaches that recently Bar-or, A., Schuster, A., & Wolff, R. (2005).
implemented in FSW process to predict the response Decision Tree Induction in High
of the process parameters. The recent work on Dimensional , Hierarchically Distributed
machine learning algorithms, which implemented in Decision Tree Induction in High
FSW and FSSW process parameter are ANNS, Dimensional , Hierarchically Distributed
ANFIS, Regression model, and SVM classification. Databases Faculty of Computer Science.
During this study found that the Deep Learning has Proceedings of the 2005 SIAM International
not yet been applied to FSW or FSSW. There is Conference on Data Mining, (April), 466–
much scope and gap of research available in the 470.
application of ANFIS and SVM method to apply in Barath, V. R., Vaira Vignesh, R., & Padmanaban, R.
FSW or FSSW process parameters. In the (2018). Analysing the strength of friction stir
knowledge of prescribed work that there is much welded dissimilar aluminium alloys using
need of implementation in machine learning Sugeno Fuzzy model. IOP Conference
techniques to predict the behavior of process Series: Materials Science and Engineering,
parameters in FSW or FSSW. 310(1), 012043.
Bennell, J. A., & Sutcliffe, C. M. (2003). Black-
REFERENCES
Scholes Versus Artificial Neural Networks in
Abdullah, I. T., & Hussein, S. K. (2018). Improving Pricing FTSE 100 Options. SSRN Electronic
the joint strength of the friction stir spot Journal, 260(2004), 243–260.
welding of carbon steel and copper using the Bhat, N. N., Kumari, K., Dutta, S., Pal, S. K., & Pal,
design of experiments method. S. (2015). Friction stir weld classification by
Multidiscipline Modeling in Materials and applying wavelet analysis and support vector
machine on weld surface images. Journal of
183

Manufacturing Processes, 20, 274–281. Elatharasan, G., & Kumar, V. S. S. (2012).

Boldsaikhan, E., Corwin, E. M., Logar, A. M., & Modelling and optimization of friction stir
Arbegast, W. J. (2011). The use of neural welding parameters for dissimilar aluminium
network and discrete Fourier transform for alloys using RSM. Procedia Engineering,
real-time evaluation of friction stir welding. 38(December), 3477–3481.
Applied Soft Computing Journal, 11(8), Elatharasan, G., & Kumar, V. S. S. (2013). An
4839–4846. Experimental Analysis and Optimization of
https://doi.org/10.1016/j.asoc.2011.06.017 Process Parameter on Friction Stir Welding
Buffa, G., Fratini, L., & Micari, F. (2012). of AA 6061-T6 Aluminum Alloy using
Mechanical and microstructural properties RSM. Procedia Engineering, 64, 1227–
prediction by artificial neural networks in 1234.
FSW processes of dual phase titanium alloys. Fleming, P., Fleming, K., Lammlein, D., Wilkes, D.,
Journal of Manufacturing Processes, 14(3), Bloodworth, T., Cook, G., … Bement, M.
289–296. (2007). Automatic Fault Detection in
Chakraborty, T., Chakraborty, A. K., & Friction Stir Welding. Materials Science and
Chattopadhyay, S. (2018). A novel Technology-Association for Iron and Steel
distribution-free hybrid regression model for Technology-, 5(May 2014), 3309–3316.
manufacturing process efficiency Fratini, L., Buffa, G., & Palmeri, D. (2009). Using a
improvement. (February 2019). Retrieved neural network for predicting the average
from http://arxiv.org/abs/1804.08698 grain size in friction stir welding processes.
Cortes, C., & Vapnik, V. (1995). Support-Vector Computers and Structures, 87(17–18), 1166–
Networks. Machine Learning, 20(1995), 1174.
273–297. Ghetiya, N. D., & Patel, K. M. (2014). Prediction of
Dehabadi, V. M., Ghorbanpour, S., & Azimi, G. Tensile Strength in Friction Stir Welded
(2016). Application of artificial neural Aluminium Alloy Using Artificial Neural
network to predict Vickers microhardness of Network. Procedia Technology, 14, 274–
AA6061 friction stir welded sheets. Journal 281.
of Central South University, 23(9), 2146– https://doi.org/10.1016/j.protcy.2014.08.036
2155. Gordon, J., & Sohal, A. S. (2001). Assessing
Deng, L. (2014). Deep Learning: Methods and manufacturing plant competitiveness-An
Applications. Foundations and Trends® in empirical field study. International Journal
Signal Processing, 7(3–4), 197–387. of Operations & Production Management,
Dewan, M. W., Huggett, D. J., Warren Liao, T., 21(1/2), 233–253.
Wahab, M. A., & Okeil, A. M. (2016). Heidarzadeh, A. (2019). Tensile behavior,
Prediction of tensile strength of friction stir microstructure, and substructure of the
weld joints with adaptive neuro-fuzzy friction stir welded 70/30 brass joints: RSM,
inference system (ANFIS) and neural EBSD, and TEM study. Archives of Civil and
network. Materials & Design, 92(3), 288– Mechanical Engineering, 19(1), 137–146.
299. https://doi.org/10.1016/j.acme.2018.09.009
Dingli, D. J. (2012). The Manufacturing Industry – Jang, J.-S. R. (1993). ANFIS: adaptive-network-
Coping with Challenges (Working Paper No . based fuzzy inference system. IEEE
2012 / 05). 47. Retrieved from Transactions on Systems, Man, and
https://www.msm.nl/resources/uploads/2014 Cybernetics, 23(3), 665–685.
/02/MSM-WP2012-05.pdf https://doi.org/10.1109/21.256541
Do, T., Lenca, P., Lallich, S., & Pham, N. (2010). Jayaraman, M., Sivasubramanian, R.,
Classifying Very-High-Dimensional Data Balasubramanian, V., & Lakshminarayanan,
with Random Forests of Oblique Decision A. K. (2008). Prediction of Tensile Strength
Trees. Advances in Knowledge Discovery of Friction Stir Welded A356 Cast
and Management, 292, 39–55. Aluminium Alloy Using Response Surface
Dong, B., Cao, C., & Eang, S. (2005). Applying Methodology and Artificial Neural Network.
support vector machines to predict building Journal for Manufacturing Science and
energy consumption in tropical region. Production, 9(1–2), 45–60.
Energy and Buildings, 37, 545–553. Jenarthanan, M. P., Varun Varma, C., & Krishna
Douglas C. Montgomery, E. A. P., & Vining, G. G. Manohar, V. (2018). Impact of friction stir
(2012). Introduction to Linear Regression welding (FSW) process parameters on tensile
Analysis. In Wiley (5th ed.). wiley. strength during dissimilar welds of AA2014
Ekici, B. B., & Aksoy, U. T. (2009). Prediction of and AA6061. Materials Today: Proceedings,
building energy consumption by using 5(6), 14384–14391.
artificial neural networks. Advances in Kadaganchi, R., Gankidi, M. R., & Gokhale, H.
Engineering Software, 40(5), 356–362. (2015). Optimization of process parameters
184

of aluminum alloy AA 2014-T6 friction stir ANN. Materials Today: Proceedings, 5(1),
welds by response surface methodology. 716–723.
Defence Technology, 11(3), 209–219. Maleki, E. (2015). Artificial neural networks
Kandananond, K. (2011). Forecasting Electricity application for modeling of friction stir
Demand in Thailand with an Artificial Neural welding effects on mechanical properties of
Network Approach. Energies, 4, 1246–1257. 7075-T6 aluminum alloy. IOP Conference
Karthikeyan, R., & Balasubramanian, V. (2010). Series: Materials Science and Engineering,
Predictions of the optimized friction stir spot 103(1).
welding process parameters for joining Malviya, R., & Pratihar, D. K. (2011). Tuning of
AA2024 aluminum alloy using RSM. neural networks using particle swarm
International Journal of Advanced optimization to model MIG welding process.
Manufacturing Technology, 51(1–4), 173– Swarm and Evolutionary Computation, 1(4),
183. 223–235.
Kashyap, P. (2017). Industrial Applications of Manvatkar, V. D., Arora, A., De, A., & DebRoy, T.
Machine Learning. In Machine Learning for (2012). Neural network models of peak
Decision Makers (pp. 189–233). temperature, torque, traverse force, bending
Khourshid, A. M., El-Kassas, A. M., & Sabry, I. stress and maximum shear stress during
(2015). Integration between Artificial Neural friction stir welding. Science and Technology
Network and Responses Surfaces of Welding and Joining, 17(6), 460–466.
Methodology for Modeling of Friction Stir Mena, R., Rodríguez, F., Castilla, M., & Arahal, M.
welding. International Journal of Advanced R. (2014). A prediction model based on
Engineering Research and Science neural networks for the energy consumption
(IJAERS), 2(8), 67–73. of a bioclimatic building. Energy &
Köksal, G., Batmaz, I., & Testik, M. C. (2011). A Buildings, 82, 142–155.
review of data mining applications for quality Mian, A., Newaz, G., Vendra, L., Rahman, N.,
improvement in manufacturing industry. Georgiev, D. G., Auner, G., … Herfurth, H.
Expert Systems with Applications, 38(10), (2005). Laser bonded microjoints between
13448–13467. titanium and polyimide for applications in
Kurtulmu, M., & Kiraz, A. (2018). Arti cial neural medical implants. Journal of Materials
network modelling for polyethylene FSSW Science: Materials in Medicine, 16(3), 229–
parameters. Scientia Iranica, 25, 1266–1271. 237.
Kutsurelis, J. E. (1998). Forecasting financial Murty, M. N., & Raghava, R. (2016). Support
markets using neural networks: an analysis Vector Machines and Perceptrons.
of methods and accuracy. Monterey, https://doi.org/10.1007/978-3-319-41063-0
California. Naval Postgraduate School. Nataliia, D., Erik, G., Igor, Z., & Klaus, Z. (2019).
Lakshminarayan, A. K., & Balasubramanian, V. Mathematical modeling of friction stir
(2009). Comparison of RSM with ANN in welding considering dry and viscous friction.
predicting tensile strength of friction stir Applied Mathematical Modelling, 67, 1–8.
welded AA7039 aluminium alloy joints. Nguyen, L. (2017). Tutorial on Support Vector
Transactions of Nonferrous Metals Society of Machine (SVM). Applied and Computational
China, 19(1), 9–18. Mathematics, (June), 1–13.
Lászlo Monostori, Jozsef Homyak, Csaba Egresits, Nourani, M., Milani, A. S., & Yannacopoulos, S.
Z. J. V. (1998). Soft Computing and Hybrid (2011). Taguchi Optimization of Process
AI Approaches to Intelligent Manufacturing. Parameters in Friction Stir Welding of 6061
Computer and Automation Research Aluminum Alloy: A Review and Case Study.
Institute, Hungarian Academy of Sciences, Engineering, 03(02), 144–155.
(i). Okuyucu, H., Kurt, A., & Arcaklioglu, E. (2007).
Li, K., Hu, C., Liu, G., & Xue, W. (2015). Building ’ Artificial neural network application to the
s electricity consumption prediction using friction stir welding of aluminum plates.
optimized artificial neural networks and Materials & Design, 28(1), 78–84.
principal component analysis. Energy & Paoletti, A., Lambiase, F., & Di Ilio, A. (2015).
Buildings, 108, 106–113. Optimization of Friction Stir Welding of
Lu, S. C. Y. (1990). Machine learning approaches to Thermoplastics. Procedia CIRP, 33, 562–
knowledge synthesis and integration tasks for 567.
advanced engineering automation. https://doi.org/10.1016/j.procir.2015.06.078
Computers in Industry, 15(1–2), 105–120. Pham, D. T., & Afify, A. A. (2005). Machine-
M. Krishnan, M., Maniraj, J., Deepak, R., & learning techniques and their applications in
Anganan, K. (2018). Prediction of optimum manufacturing. Journal Engineering
welding parameters for FSW of aluminium Manufacture, 219, 395–412.
alloys AA6063 and A319 using RSM and Pokutta, S. (2016). Machine Learning in
185

Engineering Applications and Trends. NASA Engineering, 29, 1471–1475.

Workshop Machine Learning Technologies Simon, H. A. (1983). WHY SHOULD MACHINES
and Their Applications to Scientific and LEARN ? In R. S. Michalski, J. G. Carbonell,
Engineering Domains Workshop. & T. M. Mitchell (Eds.), Machine Learning
Qamar, M., & Khosravi, A. (2015). A review on An Artificial Intelligence Approach (pp. 25–
artificial intelligence based load demand 35).
forecasting techniques for smart grid and Srinivasa Rao, M., & Ramanaiah, N. (2018).
buildings A review on arti fi cial intelligence Optimization of Process Parameters for FSW
based load demand forecasting techniques of Al-Mg-Mn-Sc-Zr Alloy Using CCD And
for smart grid and buildings. Renewable and RSM. Journal of Mechanical Engineering,
Sustainable Energy Reviews, 50(June 2015), 68(3), 195–224.
1352–1372. Staub, S., Karaman, E., Kaya, S., Karapınar, H., &
Ranjith, R., Giridharan, P. K., & Senthil, K. B. Güven, E. (2015). Artificial Neural Network
(2017). Predicting the tensile strength of and Agility. Procedia - Social and
friction stir welded dissimilar aluminum Behavioral Sciences, 195, 1477–1485.
alloy using ann. International Journal of Susto, G. A., Schirru, A., Pampuri, S., Mcloone, S.,
Civil Engineering and Technology, 8(9), Member, S., & Beghi, A. (2015). Machine
345–353. Learning for Predictive Maintenance : A
Rasmussen, C. E. (2004). Gaussian Processes in Multiple Classifier Approach. IEEE
Machine Learning. Transactions on Industrial Informatics,
https://doi.org/10.1007/978-3-540-28650- 11(3), 812–820.
9_4 Takagi, T., & Sugeno, M. (1985). Fuzzy
Rostami, H., Dantan, J.-Y., & Homri, L. (2015). Identification of Systems and Its
Review of data mining applications for Applications to Modeling and Control. IEEE
quality assessment in manufacturing Transactions on Systems, Man and
industry: support vector machines. Cybernetics, SMC-15(1), 116–132.
International Journal of Metrology and Tansel, I. N., Demetgul, M., Okuyucu, H., & Yapici,
Quality Engineering, 6(4), 401. A. (2010). Optimizations of friction stir
Satpathy, M. P., Mishra, S. B., & Sahoo, S. K. welding of aluminum alloy by using
(2018). Ultrasonic spot welding of genetically optimized neural network.
aluminum-copper dissimilar metals: A study International Journal of Advanced
on joint strength by experimentation and Manufacturing Technology, 48(1–4), 95–
machine learning techniques. Journal of 101.
Manufacturing Processes, 33(May), 96–110. Thomas, A. J., Byard, P., & Evans, R. (2012).
Shanavas, S., & Dhas, J. E. R. (2018). Quality Identifying the UK’s manufacturing
Prediction of Friction Stir Weld Joints on Aa challenges as a benchmark for future growth.
5052 H32 Aluminium Alloy Using Fuzzy Journal of Manufacturing Technology
Logic Technique. Materials Today: Management, 23(2), 142–156.
Proceedings, 5(5), 12124–12132. Vaira Vignesh, R., & Padmanaban, R. (2018).
Shiang, L. E., & Nagaraj, S. (2011). Impediments to Artificial neural network model for
innovation: Evidence from Malaysian predicting the tensile strength of friction stir
manufacturing firms. Asia Pacific Business welded aluminium alloy AA1100. Materials
Review, 17(2), 209–223. Today: Proceedings, 5(8), 16716–16723.
Shojaeefard, M. H., Akbari, M., & Asadi, P. (2014). Vapnik, V. N. (2000). The Nature of Statical
Multi objective optimization of friction stir Learning Theory.
welding parameters using FEM and neural Vardhan, H., & Bayar, R. (2013). Regression
network. International Journal of Precision Modeling. In Rock Engineering Design (pp.
Engineering and Manufacturing, 15(11), 91–104).
2351–2356. Verma, S., Gupta, M., & Misra, J. P. (2018).
Shojaeefard, M. H., Behnagh, R. A., Akbari, M., Performance evaluation of friction stir
Givi, M. K. B., & Farhani, F. (2013). welding using machine learning approaches.
Modelling and Pareto optimization of MethodsX, 5(May), 1048–1058.
mechanical properties of friction stir welded Wakchaure, K. N., Thakur, A. G., Gadakh, V., &
AA7075/AA5083 butt joints using neural Kumar, A. (2018). Multi-Objective
network and particle swarm algorithm. Optimization of Friction Stir Welding of
Materials & Design, 44, 190–198. Aluminium Alloy 6082-T6 Using hybrid
Shuangsheng, G., Xingwei, T., Shude, J., & Zhitao, Taguchi-Grey Relation Analysis- ANN
Y. (2012). Prediction of Mechanical Method. Materials Today: Proceedings,
Properties of Welded Joints Based on 5(2), 7150–7159.
Support Vector Regression. Procedia Widodo, A., & Yang, B. S. (2007). Support vector
186

machine in machine condition monitoring

and fault diagnosis. Mechanical Systems and
Signal Processing, 21(6), 2560–2574.
Wuest, T., Weimer, D., Irgens, C., & Thoben, K. D.
(2016). Machine learning in manufacturing:
Advantages, challenges, and applications.
Production and Manufacturing Research,
4(1), 23–45.
Yang, K., & Trewn, J. (2004). Multivariate
Statistical Methods and Quality. McGraw-
Hill.
Yousif, Y. K., Daws, K. M., & Kazem, B. I. (2008).
Prediction of Friction Stir Welding
Characteristic Using Neural Network. JJMIE
Jordan Journal of Mechanical and Industrial
Engineering, 2(3), 151–155.
Yucesan, M., Gul, M., & Celik, E. (2018).
Performance Comparison between
ARIMAX, ANN and ARIMAX-ANN
Hybridization in Sales Forecasting for
Furniture Industry. Drvna Industrija, 69(4),
357–370.
Zhu Lingyun, Cao Changxiu, Wu Wei, & Xu
Xiaoling. (2003). A novel approach based on
support vector machine to forecasting the
quality of friction welding. Proceedings of
the 4th World Congress on Intelligent
Control and Automation (Cat. No.02EX527),
1, 335–339.