Article

FSW Optimization: Prediction Using Polynomial Regression and

Optimization with Hill-Climbing Method
Piotr Myśliwiec 1, * , Paulina Szawara 2 , Andrzej Kubit 3 , Marek Zwolak 1 , Robert Ostrowski 1 ,
Hamed Aghajani Derazkola 4 and Wojciech Jurczak 5

1 Department of Materials Forming and Processing, Rzeszow University of Technology, al. Powst. Warszawy 8,
35-959 Rzeszów, Poland; [email protected] (M.Z.); [email protected] (R.O.)
2 Doctoral School of Engineering and Technical Sciences, Rzeszow University of Technology, al. Powst.
Warszawy 12, 35-959 Rzeszów, Poland; [email protected]
3 Department of Manufacturing and Production Engineering, Rzeszow University of Technology, al. Powst.
Warszawy 8, 35-959 Rzeszów, Poland; [email protected]
4 Department of Nonlinear Solid Mechanics, Faculty of Engineering Technology, University of Twente,
7500-7549 Enschede, The Netherlands; [email protected]
5 Mechanical and Electrical Engineering Department, Polish Naval Academy, 81-103 Gdynia, Poland;
[email protected]
* Correspondence: [email protected]

Abstract: This study presents the optimization of the friction stir welding (FSW) process
using polynomial regression to predict the maximum tensile load (MTL) of welded joints.
The experimental design included varying spindle speeds from 600 to 2200 rpm and
welding speeds from 100 to 350 mm/min over 28 experimental points. The resulting MTL
values ranged from 1912 to 15,336 N. A fifth-degree polynomial regression model was
developed to fit the experimental data. Diagnostic tests, including the Shapiro–Wilk test and
kurtosis analysis, indicated a non-normal distribution of the MTL data. Model validation
showed that fifth-degree polynomial regression provided a robust fit with high fitted and
predicted R2 values, indicating strong predictive power. Hill-climbing optimization was
used to fine-tune the welding parameters, identifying an optimal spindle speed of 1100 rpm
and a welding speed of 332 mm/min, which was predicted to achieve an MTL of 16,852 N.
Academic Editor: Chih-Chun Hsieh Response surface analysis confirmed the effectiveness of the identified parameters and
Received: 3 December 2024
demonstrated their significant influence on the MTL. These results suggest that the applied
Revised: 20 December 2024 polynomial regression model and optimization approach are effective tools for improving
Accepted: 17 January 2025 the performance and reliability of the FSW process.
Published: 19 January 2025

Citation: Myśliwiec, P.; Szawara, P.; Keywords: friction stir welding; polynomial regression; hill-climbing; response surface
Kubit, A.; Zwolak, M.; Ostrowski, R.; analysis; Design Expert 12; welding parameter optimization; aluminum alloy 2024-T3
Derazkola, H.A.; Jurczak, W. FSW
Optimization: Prediction Using
Polynomial Regression and
Optimization with Hill-Climbing
1. Introduction
Method. Materials 2025, 18, 448.
https://doi.org/10.3390/ Friction Stir Welding (FSW), introduced in the early 1990s by Thomas et al. [1], is a solid-
ma18020448 state joining technique that has gained widespread attention for its ability to produce high-
Copyright: © 2025 by the authors.
quality joints in lightweight and difficult-to-weld alloys, notably aluminum [2]. Operating
Licensee MDPI, Basel, Switzerland. below the melting temperature of the base material, FSW reduces common defects found
This article is an open access article in fusion welding—such as porosity, hot cracking, and distortion—resulting in improved
distributed under the terms and mechanical properties and structural integrity [3]. Due to these advantages, it has found
conditions of the Creative Commons
extensive applications in transportation, aerospace, and marine industries [4–6]. However,
Attribution (CC BY) license
determining the optimal set of parameters—rotational speed, traverse speed, axial force,
(https://creativecommons.org/
licenses/by/4.0/).
tool geometry, and cooling strategy—remains challenging, as these factors must be balanced

Materials 2025, 18, 448 https://doi.org/10.3390/ma18020448

Materials 2025, 18, 448 2 of 29

to achieve the desired mechanical strength, microstructural refinement, and minimal defect
formation [7–9].
Early optimization strategies often relied on trial-and-error approaches combined with
classical statistical tools (e.g., Taguchi, ANOVA), which provided some guidance but were
time-consuming and limited in handling the nonlinear and high-dimensional parameter
interactions of FSW [10–12]. For instance, Sabry et al. [7] and Kesharwani et al. [8] employed
statistical methodologies to identify welding conditions that improved tensile properties
and reduced defects. While these approaches could pinpoint useful parameter trends, they
typically demanded substantial experimental efforts. Similarly, preliminary models or
linear approximations struggled with complex parameter spaces, especially when multiple
performance criteria had to be considered simultaneously [13–15].
As computational resources improved, finite element modeling (FEM) and numerical
simulations contributed insights into temperature fields, material flow, and residual stresses.
He et al. [3] and Jasim et al. [16], for example, integrated FEM with data-driven methods
to propose parameter sets that enhanced joint performance. Although simulations helped
reduce the trial-and-error burden, FEM models are often computationally expensive and
may require extensive calibration, limiting their scalability and adaptation to new welding
conditions [17–21].
The advent of artificial intelligence (AI) and machine learning (ML) techniques pro-
vided new avenues for more accurate and cost-effective process optimization [6,13,22]. By
leveraging historical or experimentally obtained datasets, ML models can capture com-
plex, nonlinear parameter-response relationships. Myśliwiec et al. [22] applied Random
Forest, XGBoost, and MLP models to optimize the FSW of 2024-T3 aluminum alloy, demon-
strating that advanced ML algorithms could accurately predict mechanical outputs and
suggest improved parameter combinations. Similarly, Cho et al. [13] and Albaijan et al. [14]
showed that artificial neural network-based predictions not only improved accuracy but
also reduced guesswork in selecting parameter sets that enhanced tensile strength and
joint reliability.
Nonetheless, ML approaches often rely on extensive, high-quality datasets and can
pose interpretability challenges. Vidakis et al. [23] found that integrating data-driven mod-
els with domain knowledge (e.g., polynomial regression or RSM-based insights) reduced
the need for large-scale experimentation. Mishra et al. [19] and Sambath et al. [24] high-
lighted that while ML excels at identifying trends and relationships, its predictive power is
sensitive to data diversity and preprocessing quality. Chadha et al. [25] successfully imple-
mented ML-based defect prediction yet noted that specialized datasets and sensor feedback
were needed for real-time adaptation, adding complexity to practical implementation.
Metaheuristic and evolutionary algorithms have shown promise in exploring large
parameter spaces and identifying global optima [26–28]. These methods can balance multi-
ple performance criteria—such as tensile strength, microstructural uniformity, and energy
efficiency—more flexibly than classical optimization techniques. Prabhakar et al. [29]
and Kubit et al. [30] employed multi-objective optimization frameworks to improve joint
efficiency and processing time, outperforming baseline conditions derived from simpler
approaches. Rana et al. [26] reported that while hybrid evolutionary algorithms could iden-
tify global optima, the computational overhead and complexity in tuning these algorithms
were not trivial.
To mitigate these challenges, researchers have explored hybrid strategies combining
polynomial regression, RSM, ML, and metaheuristics to achieve interpretability, data
efficiency, and accuracy. Yaknesh et al. [31] and Rao et al. [32] demonstrated that polynomial
models, when integrated with evolutionary searches, provided robust optimization results
without the heavy computational demands typical of complex ML models. Sabry et al. [15]
Materials 2025, 18, 448 3 of 29

and Babalola et al. [33] found that using polynomial fitting as a preliminary modeling step
not only improved interpretability but also served as a strong foundation for subsequent
ML-driven refinement.
Bayesian optimization, transfer learning, and multi-criteria decision-making frame-
works further enhance adaptability, enabling parameter identification that can transfer
across materials or joint configurations with minimal additional experimentation [6,34–37].
Sengupta et al. [6], for example, demonstrated how transfer learning could reuse knowledge
from one alloy system to expedite optimization in another. Kumar et al. [11] used multi-
criteria decision-making to incorporate human expertise and manufacturing constraints
directly into the decision-making process, offering a more holistic perspective.
Despite these advancements, key gaps remain. Many studies reported substantial
improvements in specific metrics—such as achieving a 10–20% increase in tensile strength
or reduced defects [13,22,38]—but often under constrained conditions or tailored datasets.
Scalability, generalizability, and adaptability to limited data scenarios remain pressing
concerns. Likewise, sophisticated ML or metaheuristic approaches might require careful
tuning and might not always be justifiable for certain industrial environments [36,39,40].
Ensuring that enhancements in one property (e.g., tensile strength) do not compromise
others (e.g., ductility or fatigue life) is another ongoing challenge [29,40].
This study proposes a hybrid approach that combines polynomial regression modeling
with a hill-climbing optimization technique to predict and enhance the maximum tensile
load (MTL) of FSW joints. Polynomial regression can capture higher-order interactions
while maintaining interpretability and reducing complexity, and hill-climbing provides a
straightforward iterative procedure for refining parameters and approaching near-optimal
solutions. Unlike pure ML models that require extensive data or metaheuristic methods
that may be computationally demanding, this strategy aims for a balanced solution. By
leveraging polynomial modeling as a starting point and then applying hill-climbing op-
timization, the approach reduces both experimental and computational burdens while
delivering strong predictive capabilities. Thus, it can potentially match or exceed the
improvements reported in the literature while offering greater transparency, adaptability,
and reduced cost—key factors in advancing robust FSW parameter optimization.

2. Materials and Methods

The friction stir welding process was conducted using AA2024-T3 aluminum sheets
with a thickness of 1.5 mm. The experiments were performed on a Makino PS95 CNC
milling machine (Figure 1a) using a commercially available tool with the geometric pa-
rameters shown in Table 1. The plates, which were 200 mm long and 100 mm wide, were
joined along the rolling line. The joints were configured in a lap joint arrangement with
a lap width of 30 mm. A factorial design was used to plan the experiment. The design
matrix, shown in Figure 1b, included 28 experimental points with varying spindle speeds
from 600 to 2200 rpm and welding speeds from 100 to 350 mm/min. The technological
parameters for each test run, along with the measured strength of the FSW joints. The
selected range of spindle and welding speeds was determined through a comprehensive
literature review and preliminary experimental trials. Prior research on AA2024-T3 and
analogous aluminum alloys indicated that these ranges are optimal for attaining defect-free
welds with high mechanical performance. Furthermore, in-house trials were conducted to
refine the process window, thereby ensuring that the selected parameters encompassed the
regions where the maximum tensile strength of the joints could be obtained. This factorial
design allowed a comprehensive analysis of the effects of spindle and welding speeds
on the tensile strength of the FSW joints, ensuring a robust assessment of the optimum
welding conditions. Each joint was then divided into four sections for tensile testing. The
Materials 2025,18,
Materials2025, 18,448
448 4 of
4 of2929

specimens wereThe
tensile testing. precisely cut using
specimens wereaprecisely
wire electrical discharge
cut using machining
a wire electrical (EDM) technique
discharge machin-
to
ing (EDM) technique to minimize the influence of external forces on the cutting
minimize the influence of external forces on the joint structure. The process
joint structure.
followed theprocess
The cutting configuration shown
followed in Figure 1c. shown
the configuration Each ofinthe four1c.
Figure specimens wasfour
Each of the mounted
spec-
on a ZWICK/ROELL Z100 universal testing machine to evaluate the bond
imens was mounted on a ZWICK/ROELL Z100 universal testing machine to evaluate the strength. The
experimental
bond strength. results were analyzed
The experimental using
results Design
were Expert
analyzed 12 software,
using and a12
Design Expert fifth-degree
software,
polynomial regression model was developed to predict and optimize the
and a fifth-degree polynomial regression model was developed to predict and optimize technological
parameters in the FSW
the technological process.
parameters in the FSW process.

Figure1.1.(a)
Figure (a) FSW
FSW machine,
machine, (b)
(b) factor
factor plan
plan of
of the
theFSW
FSWprocess,
process,and
and(c)
(c)configuration
configurationofofthe welded
the welded
panels(unit
panels (unitmm).
mm).
Materials 2025, 18, 448 5 of 29
Materials 2025, 18, 448 5 of 29

Table 1. Geometric parameters of the FSW tool and welding conditions [22].
Table 1. Geometric parameters of the FSW tool and welding conditions [22].
Tool Parameters
Tool Parameters Value
Value Tool
ToolView
View
Shoulder Shoulder
diameterdiameter
D [mm] D [mm] 12 12
Pin diameter d [mm] d [mm]
Pin diameter 4.5 4.5
Pin height [mm]
Pin height [mm] 2.55 2.55
Tool offset [mm]
Tool offset [mm] 0.05 0.05
Dwell time [s] time [s]
Dwell 10 10
Tool tilt angle
Tool tilt angle 0◦ 0°
Tool plunge speed
Tool plunge[mm/min]
speed [mm/min] 2 2
ShoulderShoulder
profile profile Flat with spiral groove
Flat with spiral groove
Pin profile
Pin profile
Conical threaded
Conical threaded
D/d ratio of the tool 2.7
D/d ratio of the tool 2.7
Tool material H13 Steel
Tool material H13 Steel

Evaluation of the Experimental

Evaluation ModelModel
of the Experimental
The FSW Theprocess
FSW process was implemented
was implemented for afor a range
range of parameters:
of parameters: tool
tool speed
speed from600
from 600toto
2200
2200 rpm rpm
and and welding
welding speedspeed
from from
100 to100 tomm/min.
350 350 mm/min. TheThe independent
independent variableswere
variables were
coded and presented in Table 2. The dependent variable was the response or strength ofof
coded and presented in Table 2. The dependent variable was the response or strength
the FSWthelap
FSW lapThe
joint. joint.resulting
The resulting
load load capacities
capacities ranged
ranged fromfrom 1912
1912 to 15,336
to 15,336 NN (Table
(Table 3).3).

Table 2. FSW process parameters: spindle speed (A) and welding speed (B) with their ranges, coded
Table 2. FSW process parameters: spindle speed (A) and welding speed (B) with their ranges, coded
values, averages, and standard deviations.
values, averages, and standard deviations.
Factor Name Units Type Min. Max. Coded Low Coded High Mean Std. Dev.
Factor A Name Units
Spindle Speed Type
rpm Min.
Numeric Max. 2200
600 Coded−1
Low
↔ 600Coded+1High
↔ 2200 Mean
1400 Std.536.92
Dev.
A B
Spindle Welding Speed
Speed rpm mm/min Numeric
Numeric 600 2200 350−1 ↔−1
100 ↔ 100 +1 ↔+1
600 ↔ 350 1400
2200 226.79 536.92
87.49
B Welding Speed mm/min Numeric 100 350 −1 ↔ 100 +1 ↔ 350 226.79 87.49
Table 3. Measured joint maximum tensile load (MTL) in the FSW experiment, including the number

Table 3.of observations,

Measured jointrange of values,
maximum mean,
tensile and
load standard
(MTL) deviation.
in the FSW experiment, including the number
Response of observations,
Name Units range of values, mean, and
Observations Min.standard deviation.Mean
Max. Std. Dev. Ratio
R1 MTL N 112 1912 15,336 6451.97 3071.58 8.02
Response Name Units Observations Min. Max. Mean Std. Dev. Ratio
R1 MTL N The112histogram of1912 the ultimate 15,336
maximum tensile6451.97load, hereafter
3071.58referred to as MTL,
8.02
measurements (Figure 2) from the FSW process were analyzed along with statistical tests
to assess the normality of the data distribution. The histogram shows a right-skewed dis-
The histogram
tribution of MTLof the ultimate
values, maximum
with the majoritytensile load, hereafter
of measurements referredbetween
concentrated to as MTL,
2500
measurements (Figure 2) from the FSW process were analyzed along
N and 7500 N. The peak frequency occurs around 5000 N, and there is a noticeable de- with statistical tests
to assess the in
crease normality
frequencyoftowards
the data distribution.
higher MTL values. The
To histogram
statistically shows
evaluatea the
right-skewed
normality of
distribution of MTL
the data, values, with
the Shapiro–Wilk testthe
was majority of measurements
performed. The test statistic concentrated
is 0.861 with a between
p-value of
2500 N7.62
and×7500
10−9 , N. The peak
indicating thatfrequency
the MTL data occurs around
do not follow5000 N, and
a normal there is a and
distribution, noticeable
the null
decrease in frequency
hypothesis (H0) towards
is rejected. higher MTL is
This result values.
visuallyToconfirmed
statistically
by evaluate the normality
superimposing a normal
distribution
of the data, curve on thetest
the Shapiro–Wilk histogram, which highlights
was performed. The test significant
statistic departures
is 0.861 withfrom normal-
a p-value
− 9
of 7.62 × 10 , indicating that the MTL data do not follow a normal distribution,kurto-
ity. Further analysis of the kurtosis of the data yields a value of 0.764. This positive and
the nullsis indicates that
hypothesis (H0) theisdistribution
rejected. This hasresult
heavier
is tails and aconfirmed
visually sharper peakby compared to a nor-
superimposing a
mal distribution, contributing to the observed skewness. In addition,
normal distribution curve on the histogram, which highlights significant departures from the histogram shows
several
normality. outliers
Further at higher
analysis of theMTL values,ofespecially
kurtosis the data above
yields 10,000
a valueN,offurther
0.764. supporting
This positivethe
non-normality of the data.
kurtosis indicates that the distribution has heavier tails and a sharper peak compared to
a normal distribution, contributing to the observed skewness. In addition, the histogram
shows several outliers at higher MTL values, especially above 10,000 N, further supporting
the non-normality of the data.
Materials 2025, 18, 448 6
Materials 2025, 18, 448 6 of 29

Figure2.2.Histogram
Figure Histogramof maximum
of maximum tensiletensile
load (MTL)
load results
(MTL)from the FSW
results fromprocess.
the FSW Theprocess.
analysis The ana
indicates a right-skewed distribution, showing that the data are not normally distributed.
indicates a right-skewed distribution, showing that the data are not normally distributed.
The box plot of the maximum tensile load (MTL) measurements (Figure 3a) from
the FSW Theprocess
box plotwasofanalyzed
the maximum
to identify tensile loadoutliers
potential (MTL)and measurements
assess the overall(Figure
data3a) from
FSW process
distribution. was
The boxanalyzed
plot shows to that
identify potential
the majority outliers
of the MTL andvaluesassess the overall data d
are concentrated
within
bution. The box plot shows that the majority of the MTL values areofconcentrated
the interquartile range (IQR), with the median value in the lower half the IQR, w
indicating a slight skew in the data distribution. The box plot analysis
the interquartile range (IQR), with the median value in the lower half of the IQR, ind highlights several
high MTL values that are considered outliers. However, these outliers are critical to
ing a slight skew in the data distribution. The box plot analysis highlights several
understanding the conditions that lead to exceptionally high tensile strengths in welds.
MTL values
Rather that are these
than dismissing considered outliers.
values, further However,isthese
investigation outliers
warranted are critical
to explore the to un
standing
specific the conditions
parameters that lead
and conditions thattoresulted
exceptionally
in these high tensile
superior weldstrengths
strengths. in welds. Ra
This
than dismissing
information could bethese values,
critical furtherthe
to optimizing investigation
FSW process and is warranted
achieving moreto explore
reliable the spe
and higher-quality
parameters and welds. The final
conditions thatstep in evaluating
resulted in these thesuperior
quality ofweld
the experimental
strengths. This in
model was to plot the FDS. The graph (Figure 3b) plots the
mation could be critical to optimizing the FSW process and achieving standard error mean (Stdmore
Errorreliable
Mean) as a function of the fraction of the design space (FDS). This type of graph is typically
higher-quality welds. The final step in evaluating the quality of the experimental m
used to evaluate the quality of an experimental design [41]. The X-axis represents the
was to plot the FDS. The graph (Figure 3b) plots the standard error mean (Std Error M
fraction of the design space, while the y-axis represents the standard error of the mean.
as adesign
The functionspaceofisthe fraction
defined as a of theindicating
cube, design space (FDS). This
a rectangular area oftype of graph
variable space.isThe
typically
to evaluate
radius theindicates
of 1.41421 qualitythat
of antheexperimental design
space is constructed [41]. The
considering theX-axis represents
Euclidean distance. the frac
of the
The design
analysis wasspace, whileon
performed the y-axis
a very represents
large number ofthe standard
points error
(150,026), of the
which mean. The de
increases
the
space is defined as a cube, indicating a rectangular area of variable space.the
accuracy of the evaluation. The t-Student value indicates the critical value for The radiu
t-distribution at a given confidence level, which is used to assess statistical significance. The
1.41421 indicates that the space is constructed considering the Euclidean distance.
design is well constructed and accurate in the central regions of the design space, indicating
analysis was performed on a very large number of points (150,026), which increase
good model quality in these regions. The increase in standard error at the edges is typical
accuracy
and of potentially
indicates the evaluation. The t-Student
lower model reliabilityvalue
in these indicates
regions buttherequires
criticalattention
value for at the t-d
bution
the edgesatofathegiven confidence
design space. level, which is used to assess statistical significance. The
sign is well constructed and accurate in the central regions of the design space, indica
good model quality in these regions. The increase in standard error at the edges is ty
and indicates potentially lower model reliability in these regions but requires attentio
the edges of the design space.
Materials 2025,18,
Materials2025, 18,448
448 7 of7 29
of 29

Figure 3.
Figure 3. (a) Box plot of
of maximum
maximumtensile
tensileload
load(MTL)
(MTL)results
resultsfrom
fromthe
theFSW
FSWprocess.
process.(b)(b)
Fraction of of
Fraction
design
design space
space (FDS) graph for
for standard
standard error
errormean.
mean.

3.
3. Creating
Creating aa Polynomial
Polynomial Regression
RegressionModel
Model
3.1.
3.1. Principles ofPolynomial
Principles of PolynomialRegression
Regression
Polynomial regression isis an
Polynomial regression anextension
extensionofoflinear
linearregression
regressionthat
thatis isused
usedtoto model
model thethe
relationship between a dependent variable y and one or more independent
relationship between a dependent variable 𝑦 and one or more independent variables 𝑥 by variables x by
fitting a polynomial equation to the observed data. This type of regression is
fitting a polynomial equation to the observed data. This type of regression is particularly particularly
useful
useful when
when thetherelationship
relationshipbetween
betweenthe thevariables
variablesisisnonlinear.
nonlinear.The
Thepolynomial
polynomialregression
regres-
sion model of degree 𝑛 can be written as:
model of degree n can be written as:

𝑦= β 0𝛽++β 1𝛽x 𝑥++β𝛽 𝑥2 + 𝛽 𝑥3 + ⋯ + 𝛽 𝑥 n+ 𝜖 (1)(1)

y= 2 x + β3 x + · · · + βn x + ϵ
where 𝑦 is the dependent variable, 𝑥 is the independent variable, 𝛽0, 𝛽1, 𝛽2, ..., 𝛽𝑛 are the
where
coefficients y is the dependent
of the polynomial variable, x is the independent
to be estimated from the data,variable, and 𝜖 is the β0 ,error
β1 , βterm
2 , . . ., βn are
repre-
the
sentingcoefficients of the polynomial
the difference between the to be estimated
observed from thevalues
and predicted data, and is the
[42]. ϵThe error term
polynomial
representing the difference between the observed and predicted values
regression process involves several steps. First, data points (𝑥𝑖, 𝑦𝑖) are collected for𝑖 = 1, 2, [42]. The polynomial
..., 𝑚, where
regression 𝑚 is the
process involves
numberseveral steps. First,
of observations. Next,data
thepoints
degree𝑛of (xi , ythe
i ) are collected is
polynomial i = 1,
forcho-
2,
sen, m, where
. . .,with a highermdegree
is the polynomial
number of observations. Next, but
fitting the data better degree n ofleading
the potentially the polynomial
to over-
is chosen,
fitting. A design higher𝑋degree
with amatrix is thenpolynomial
constructed fitting the data
that contains thebetter
powers butofpotentially
the independent leading
to
variable 𝑥. For aApolynomial
overfitting. design matrix X is 𝑛,
of degree then
the constructed
design matrixthat is 𝑋:contains the powers of the
independent variable x. For a polynomial of degree n, the design matrix is X:
1 𝑥 𝑥 … 𝑥
⎡ 1 𝑥 … 𝑥
⎤
𝑋 = ⎢1 x1 x𝑥12 . . . x1n ⎥ (2)
⎢ ⋮ ⋮ 2⋮ . .⋱ n⋮ ⎥
1 x x . x
⎣1 𝑥.2 𝑥.2 … 𝑥2 ⎦
 
X=  .. .. ..  (2)
. .
. .
. . . 
The coefficients 𝛽 are estimated using the least 2
squares n
method, which minimizes the
1 x xm . . . xm
sum of the squared differences between them observed values 𝑦 and the values predicted
by the polynomial. The coefficients are given by:
The coefficients β are estimated using the least squares method, which minimizes the
sum of the squared differences between 𝛽= the (𝑋observed
𝑋) 𝑋 𝑦values yi and the values predicted (3)by
the polynomial. The coefficients are given by:
where 𝑋 is the transpose of the design matrix 𝑋, and 𝑦 is the vector of observed values.
Once the coefficients are estimated, the polynomial equation can be used to predict the
β = ( X T X ) −1 X T y (3)
Materials 2025, 18, 448 8 of 29

where X T is the transpose of the design matrix X, and y is the vector of observed values.
Once the coefficients are estimated, the polynomial equation can be used to predict the
values of the dependent variable for new values of the independent variable x [43–45]. To
illustrate the process, consider fitting a fifth-degree polynomial regression model to a set of
data points (x1 , y1 ), (x2 , y2 ), . . ., (xm , ym ). The fifth-degree polynomial regression model can
be written as:
y = β 0 + β 1 x + β 2 x2 + β 3 x3 + β 4 x4 + β 5 x5 + ϵ (4)

The data points (xi , yi ) are collected for i = 1, 2, . . ., m. The design matrix X is formulated
to include the powers of the independent variable x. For a fifth-degree polynomial, the
design matrix X is constructed as follows:
 
1 x1 x12 x13 x14 x15
1 x2 x22 x23 x24 x25 
 
X=
 .. .. .. .. .. ..  (5)
. . . . . . 


1 xm xm2 xm3 xm4 xm5

Once the coefficients are estimated, the polynomial equation is used to predict the
values of the dependent variable for new values of the independent variable x:

ŷ = β 0 + β 1 x + β 2 x2 + β 3 x3 + β 4 x4 + β 5 x5 + ϵ (6)

The Fit Summary Table 4 compares different polynomial regression models for the
MTL data from the FSW process. Sequential p-values indicate that cubic, quartic, fifth, and
sixth polynomial terms significantly improve the model fit (p < 0.0001), meaning that the
addition of these terms provides a statistically significant improvement to the model. The
lack of fit p-value is less than 0.0001 for all models, indicating a statistically significant lack
of fit. This indicates that the models do not perfectly capture all the underlying patterns
in the data. However, it is common in complex real-world data for models to show some
lack of fit. The adjusted R2 , which accounts for model complexity, increases from 0.3877
for the linear model to 0.9752 for the sixth-degree polynomial. Similarly, the predicted R2 ,
which indicates the predictive power of the model, also increases with model complexity.
The fifth-degree polynomial model is recommended with an adjusted R2 of 0.9489 and a
predicted R2 of 0.9390, providing an excellent balance between model fit and predictive
accuracy. Although the sixth-degree polynomial has slightly higher R2 values, it is noted as
aliased, suggesting potential overfitting or multicollinearity issues. In conclusion, while the
lack of fit is statistically significant for all models, the fifth-degree polynomial model stands
out with high fitted and predicted R2 values, indicating strong model performance and
predictive power. This makes it the optimal choice for accurately modeling and predicting
MTL data in the FSW process.

Table 4. Fit summary for polynomial regression models applied to MTL data.

Source Sequential p-Value Lack of Fit p-Value Adjusted R2 Predicted R2

Linear <0.0001 <0.0001 0.3877 0.3629
2FI 0.1962 <0.0001 0.3916 0.3541
Quadratic 0.1153 <0.0001 0.4049 0.3573
Cubic <0.0001 <0.0001 0.5213 0.4544
Quartic <0.0001 <0.0001 0.8435 0.8234
Fifth <0.0001 <0.0001 0.9489 0.9390 Suggested
Sixth <0.0001 <0.0001 0.9752 0.9683 Aliased
Materials 2025, 18, 448 9 of 29

Table 5 presents key indicators of the quality of the polynomial regression model used to
optimize the FSW process. These indicators include the standard error, VIF, R2 , and power
of the model for each term. The standard error measures the variability of the regression
coefficient estimates. In a balanced design, the standard errors for different terms should be
similar. In this case, the standard errors for all terms are relatively small, indicating robust
estimates in the model. The lowest standard error is 0.1357 for term B (welding speed), and
the highest is 0.2453 for term A2 (spindle speed squared). The Variance Inflation Factor (VIF)
measures the degree of collinearity between independent variables. The ideal VIF value is 1.0,
with values greater than 10 indicating significant collinearity, which could lead to problems
with coefficient estimation. All terms have VIF values close to 1.0, indicating no significant
collinearity in the model. R2 measures the fit of the model and indicates the proportion of
variance in the dependent variable that is explained by the independent variables. For an ideal
polynomial regression model, the R2 values for individual terms should be close to 0 to avoid
overfitting. The R2 values for all terms are very low, indicating no overfitting and suggesting
that the model fits well without unnecessary complexity. The power of the statistical test
indicates the ability to detect a true effect if it exists. High power (close to 100%) indicates that
the model is highly effective in detecting the influence of process variables on joint strength.
All terms have a power level of 99.9%, indicating that the model is very effective in detecting
the effects of the process variables. The quality assessment of the polynomial regression model
for the FSW process indicates that the model is well fitted and does not have problems with
collinearity or overfitting. The low standard errors and VIF values, along with the high power
values, demonstrate the robustness and effectiveness of the model. The R2 values suggest
that the model is well calibrated without unnecessary complexity, which is beneficial for the
interpretation of results and practical application.

Table 5. Assessing the quality of the experimental model. * For a standard deviation of 1.

Term Standard Error * VIF Ri 2 Power

A 0.1416 1.00309 0.0031 99.9%
B 0.1357 1.00182 0.0018 99.9%
AB 0.1964 1.00309 0.0031 99.9%
A2 0.2453 1.01148 0.0114 99.9%
B2 0.2349 1.00965 0.0096 99.9%

3.2. ANOVA for Fifth Model

An analysis of variance was performed on the accepted fifth-degree polynomial
regression model. The ANOVA Table 6 for the 5th degree polynomial model of the MTL
data from the FSW process provides a comprehensive analysis of the sources of variation
and their statistical significance. The overall model is highly significant, with an F-value
of 104.12 and a p-value of less than 0.0001, indicating that the model effectively explains
a significant portion of the variability in MTL. Several terms in the model are identified
as statistically significant with p-values less than 0.05. These include the main effect of
spindle speed (A) and numerous higher-order interactions and polynomial terms such as
AB (spindle speed * welding speed), A2 , B2 , A3 , B3 , A4 , B4 , A5 , and B5 . The significance of
these terms suggests that both the main effects and complex interactions between spindle
speed and welding speed are critical in determining MTL. However, some terms, such as
the main effect of welding speed (B) and interactions such as A2 B, A3 B, A2 B3 , and A4 B,
are not statistically significant, as indicated by their p-values greater than 0.05. These
non-significant terms do not contribute meaningfully to the model, suggesting that they
could be excluded in future model refinement to improve simplicity without sacrificing
predictive power. The F-value for lack of fit is 49.80, with a p-value of less than 0.0001,
Materials 2025, 18, 448 10 of 29

indicating a significant lack of fit. This suggests that the model does not perfectly capture
all the underlying variability in the data and that there may be other factors or interactions
influencing MTL that are not included in the model.

Table 6. ANOVA for fifth-degree polynomial model applied to MTL data.

Source Sum of Squares df Mean Square F-Value p-Value

Model 1.003 × 109 20 5.017 × 107 104.12 <0.0001 significant
A—Spindle Speed 3.854 × 107 1 3.854 × 107 79.99 <0.0001
B—Welding Speed 7.292 × 105 1 7.292 × 105 1.51 0.2218
AB 1.535 × 108 1 1.535 × 108 318.49 <0.0001
A2 1.946 × 107 1 1.946 × 107 40.40 <0.0001
B2 2.012 × 107 1 2.012 × 107 41.77 <0.0001
A2 B 1.090 × 106 1 1.090 × 106 2.26 0.1360
AB2 4.119 × 107 1 4.119 × 107 85.49 <0.0001
A3 4.509 × 107 1 4.509 × 107 93.59 <0.0001
B3 1.284 × 107 1 1.284 × 107 26.65 <0.0001
A2 B2 3.609 × 107 1 3.609 × 107 74.91 <0.0001
A3 B 1.930 × 108 1 1.930 × 108 400.65 <0.0001
AB3 5.528 × 106 1 5.528 × 106 11.47 0.0010
A4 2.486 × 107 1 2.486 × 107 51.60 <0.0001
B4 8.687 × 106 1 8.687 × 106 18.03 <0.0001
A3 B2 1.317 × 107 1 1.317 × 107 27.33 <0.0001
A2 B3 5.118 × 104 1 5.118 × 104 0.1061 0.7454
A4 B 5.018 × 104 1 5.018 × 104 0.1041 0.7477
AB4 2.440 × 107 1 2.440 × 107 50.64 <0.0001
A5 4.439 × 107 1 4.439 × 107 92.12 <0.0001
B5 1.765 × 107 1 1.765 × 107 36.62 <0.0001
Residual 4.385 × 107 91 4.818 × 105
Lack of Fit 3.533 × 107 7 5.048 × 106 49.80 <0.0001 significant
Pure Error 8.514 × 106 84 1.014 × 105
Cor Total 1.047 × 109 111

The fit statistics for the fifth-degree polynomial model (Table 7) applied to the MTL
data from the FSW process indicate strong model performance. The model achieves a
high R2 value of 0.9581, explaining 95.81% of the variability in MTL. The adjusted R2 of
0.9489 and the predicted R2 of 0.9390 are in close agreement, indicating excellent predictive
accuracy and minimal overfitting. The standard deviation of the residuals is 694.15, and
the coefficient of variation is 10.76%, indicating low variability relative to the MTL mean
of 6451.97. In addition, the adequate precision ratio of 40.7155 far exceeds the desirable
threshold of 4, confirming a strong signal-to-noise ratio.

Table 7. Fit statistics for the fifth-degree polynomial model applied to MTL data.

Std. Dev. 694.15 R2 0.9581

Mean 6451.97 Adjusted R2 0.9489
C.V. % 10.76 Predicted R2 0.9390
Adeq Precision 40.7155

The final regression equation (Table 8) for predicting MTL in the FSW process is
expressed in terms of the actual factors, specifically spindle speed and welding speed. This
equation allows accurate predictions by substituting the specified values of these factors.
It is important to note that the coefficients are scaled to the units of each factor, and the
intercept is not centered in the design space.
Materials 2025, 18, 448 11 of 29

Table 8. Final regression equation for predicting MTL in the FSW process.

MTL =
+2.65761 × 105
−816.24439 Spindle Speed
−1633.38459 Welding Speed
−2.37849 Spindle Speed· Welding Speed
+1.42349 Spindle Speed2
+24.68195 Welding Speed2
+0.000460 Spindle Speed2 · Welding Speed
+0.016034 Spindle Speed· Welding Speed2
−0.001068 Spindle Speed3
−0.158386 Welding Speed3
−2.71047 × 10−6 Spindle Speed2 · Welding Speed2
−4.74309 × 10−8 Spindle Speed3 · Welding Speed
−0.000039 Spindle Speed· Welding Speed3
+3.80250 × 10−7 Spindle Speed4
+0.000442 Welding Speed4
+5.20073 × 10−10 Spindle Speed3 · Welding Speed2
+2.29939 × 10−10 Spindle Speed2 · Welding Speed3
−7.12925 × 10−12 Spindle Speed4 · Welding Speed
+4.38557 × 10−8 Spindle Speed· Welding Speed4
−5.22195 × 10−11 Spindle Speed5
−4.58801 × 10−7 Welding Speed5

4. Results
The applied regression model was used to predict maximum tensile load, hereafter
referred to as MTL values. In addition, a series of diagnostic tests were performed on the
applied model. The experimental results, along with the predicted MTL values from the
fifth-degree polynomial regression model and associated metrics, are presented in Table 9.
The table lists the actual MTL values from the experiments, the predicted maximum tensile
load values from the regression model, the residuals (differences between actual and
predicted values), leverage values indicating the influence of each data point on the model,
internally and externally studentized residuals for detecting outliers, Cook’s Distance for
identifying influential observations, and DFFITS values for assessing the influence of each
observation on the fitted values.

Table 9. Experimental and predicted maximum tensile load (MTL) values along with diagnostic
metrics for the fifth-degree polynomial regression model.

Rotational Welding Actual Predicted Internally Externally Influence on

Run Cook’s
Speed Speed Value of Value of Residual Leverage Studentized Studentized Fitted Value
Order Distance
[rpm] [mm/min] MTL [N] MTL [N] Residuals Residuals DFFITS
1 7165.00 7288.17 −123.17 0.248 −0.205 −0.204 0.001 −0.117
2 7299.00 7288.17 10.83 0.248 0.018 0.018 0.000 0.010
600 100
3 7057.00 7288.17 −231.17 0.248 −0.384 −0.382 0.002 −0.219
4 7329.00 7288.17 40.83 0.248 0.068 0.067 0.000 0.039
5 2754.00 3364.21 −610.21 0.213 −0.991 −0.991 0.013 −0.516
6 2800.00 3364.21 −564.21 0.213 −0.916 −0.916 0.011 −0.477
800 150
7 2671.00 3364.21 −693.21 0.213 −1.126 −1.128 0.016 −0.588
8 2974.00 3364.21 −390.21 0.213 −0.634 −0.632 0.005 −0.329
9 4746.00 4397.76 348.24 0.218 0.567 0.565 0.004 0.299
10 4885.00 4397.76 487.24 0.218 0.794 0.792 0.008 0.419
1000 100
11 5084.00 4397.76 686.24 0.218 1.118 1.120 0.017 0.592
12 5272.00 4397.76 874.24 0.218 1.425 1.433 0.027 0.757
Materials 2025, 18, 448 12 of 29

Table 9. Cont.

Rotational Welding Actual Predicted Internally Externally Influence on

Run Cook’s
Speed Speed Value of Value of Residual Leverage Studentized Studentized Fitted Value
Order Distance
[rpm] [mm/min] MTL [N] MTL [N] Residuals Residuals DFFITS
13 6138.00 5629.89 508.11 0.139 0.789 0.787 0.005 0.316
14 5938.00 5629.89 308.11 0.139 0.478 0.476 0.002 0.191
1000 200
15 6264.00 5629.89 634.11 0.139 0.984 0.984 0.007 0.395
16 6017.00 5629.89 387.11 0.139 0.601 0.599 0.003 0.241
17 13,477.00 13,585.93 −108.93 0.141 −0.169 −0.168 0.000 −0.068
18 13,438.00 13,585.93 −147.93 0.141 −0.230 −0.229 0.000 −0.093
1000 300
19 13,602.00 13,585.93 16.07 0.141 0.025 0.025 0.000 0.010
20 13,511.00 13,585.93 −74.93 0.141 −0.116 −0.116 0.000 −0.047
21 5692.00 5464.73 227.27 0.141 0.353 0.352 0.001 0.143
22 5538.00 5464.73 73.27 0.141 0.114 0.113 0.000 0.046
1200 150
23 5462.00 5464.73 −2.73 0.141 −0.004 −0.004 0.000 −0.002
24 5206.00 5464.73 −258.73 0.141 −0.402 −0.400 0.001 −0.162
25 7612.00 8243.00 −631.00 0.128 −0.973 −0.973 0.007 −0.373
26 7547.00 8243.00 −696.00 0.128 −1.074 −1.075 0.008 −0.411
1200 250
27 7803.00 8243.00 −440.00 0.128 −0.679 −0.677 0.003 −0.259
28 7735.00 8243.00 −508.00 0.128 −0.784 −0.782 0.004 −0.299
29 4984.00 5902.63 −918.63 0.173 −1.455 −1.464 0.021 −0.669
30 5059.00 5902.63 −843.63 0.173 −1.336 −1.342 0.018 −0.613
1400 100
31 4939.00 5902.63 −963.63 0.173 −1.526 −1.538 0.023 −0.702
32 5326.00 5902.63 −576.63 0.173 −0.913 −0.912 0.008 −0.417
33 5560.00 6117.56 −557.56 0.106 −0.850 −0.848 0.004 −0.292
34 5602.00 6117.56 −515.56 0.106 −0.786 −0.784 0.003 −0.270
1400 200
35 5714.00 6117.56 −403.56 0.106 −0.615 −0.613 0.002 −0.211
36 5623.00 6117.56 −494.56 0.106 −0.754 −0.752 0.003 −0.259
37 12,271.00 10,669.27 1601.73 0.131 2.475 2.549 0.044 0.989
38 12,121.00 10,669.27 1451.73 0.131 2.243 2.295 0.036 0.891
1400 300
39 12,189.00 10,669.27 1519.73 0.131 2.348 2.410 0.040 0.935
40 11,405.00 10,669.27 735.73 0.131 1.137 1.139 0.009 0.442
41 11,064.00 11,140.53 −76.53 0.191 −0.123 −0.122 0.000 −0.059
42 9862.00 11,140.53 −1278.53 0.191 −2.047 −2.084 0.047 −1.011
1400 350
43 9916.00 11,140.53 −1224.53 0.191 −1.961 −1.992 0.043 −0.967
44 10,018.00 11,140.53 −1122.53 0.191 −1.797 −1.820 0.036 −0.883
45 4556.00 3732.42 823.58 0.141 1.280 1.285 0.013 0.521
46 4525.00 3732.42 792.58 0.141 1.232 1.236 0.012 0.501
1600 150
47 4989.00 3732.42 1256.58 0.141 1.953 1.985 0.030 0.805
48 4734.00 3732.42 1001.58 0.141 1.557 1.569 0.019 0.636
49 5028.00 4620.71 407.29 0.128 0.628 0.626 0.003 0.240
50 4914.00 4620.71 293.29 0.128 0.452 0.450 0.001 0.172
1600 250
51 4870.00 4620.71 249.29 0.128 0.385 0.383 0.001 0.147
52 4966.00 4620.71 345.29 0.128 0.533 0.531 0.002 0.203
53 5092.00 4625.74 466.26 0.218 0.760 0.758 0.008 0.401
54 4889.00 4625.74 263.26 0.218 0.429 0.427 0.002 0.226
1800 100
55 5102.00 4625.74 476.26 0.218 0.776 0.774 0.008 0.409
56 4612.00 4625.74 −13.74 0.218 −0.022 −0.022 0.000 −0.012
57 4465.00 4913.36 −448.36 0.139 −0.696 −0.694 0.004 −0.279
58 4357.00 4913.36 −556.36 0.139 −0.864 −0.863 0.006 −0.347
1800 200
59 4599.00 4913.36 −314.36 0.139 −0.488 −0.486 0.002 −0.195
60 4741.00 4913.36 −172.36 0.139 −0.268 −0.266 0.001 −0.107
61 3250.00 4666.11 −1416.11 0.141 −2.201 −2.250 0.038 −0.912
62 3612.00 4666.11 −1054.11 0.141 −1.638 −1.654 0.021 −0.670
1800 300
63 3341.00 4666.11 −1325.11 0.141 −2.060 −2.098 0.033 −0.850
64 3493.00 4666.11 −1173.11 0.141 −1.823 −1.847 0.026 −0.749
65 4472.00 3473.27 998.73 0.202 1.611 1.625 0.031 0.818
66 3922.00 3473.27 448.73 0.202 0.724 0.722 0.006 0.363
1800 350
67 3816.00 3473.27 342.73 0.202 0.553 0.551 0.004 0.277
68 4715.00 3473.27 1241.73 0.202 2.003 2.037 0.048 1.025
69 3611.00 4131.38 −520.38 0.213 −0.845 −0.844 0.009 −0.440
70 3511.00 4131.38 −620.38 0.213 −1.008 −1.008 0.013 −0.525
2000 150
71 3761.00 4131.38 −370.38 0.213 −0.602 −0.600 0.005 −0.312
72 3987.00 4131.38 −144.38 0.213 −0.235 −0.233 0.001 −0.122
Materials 2025, 18, 448 13 of 29

Table 9. Cont.

Rotational Welding Actual Predicted Internally Externally Influence on

Run Cook’s
Speed Speed Value of Value of Residual Leverage Studentized Studentized Fitted Value
Order Distance
[rpm] [mm/min] MTL [N] MTL [N] Residuals Residuals DFFITS
73 5235.00 5184.34 50.66 0.194 0.081 0.081 0.000 0.040
74 4987.00 5184.34 −197.34 0.194 −0.317 −0.315 0.001 −0.154
2000 250
75 6166.00 5184.34 981.66 0.194 1.575 1.588 0.028 0.778
76 6125.00 5184.34 940.66 0.194 1.509 1.520 0.026 0.745
77 5828.00 5831.36 −3.36 0.224 −0.006 −0.005 0.000 −0.003
78 5552.00 5831.36 −279.36 0.224 −0.457 −0.455 0.003 −0.245
2200 300
79 6836.00 5831.36 1004.64 0.224 1.643 1.659 0.037 0.892
80 5555.00 5831.36 −276.36 0.224 −0.452 −0.450 0.003 −0.242
81 5228.00 6187.20 −959.20 0.242 −1.587 −1.601 0.038 −0.905
82 5928.00 6187.20 −259.20 0.242 −0.429 −0.427 0.003 −0.241
2200 350
83 6865.00 6187.20 677.80 0.242 1.122 1.123 0.019 0.635
84 6133.00 6187.20 −54.20 0.242 −0.090 −0.089 0.000 −0.050
85 5738.00 5860.18 −122.18 0.234 −0.201 −0.200 0.001 −0.111
86 5759.00 5860.18 −101.18 0.234 −0.167 −0.166 0.000 −0.092
2200 200
87 6288.00 5860.18 427.82 0.234 0.704 0.702 0.007 0.388
88 5890.00 5860.18 29.82 0.234 0.049 0.049 0.000 0.027
89 1912.00 1992.21 −80.21 0.248 −0.133 −0.133 0.000 −0.076
90 2065.00 1992.21 72.79 0.248 0.121 0.120 0.000 0.069
2200 100
91 2058.00 1992.21 65.79 0.248 0.109 0.109 0.000 0.062
92 1951.00 1992.21 −41.21 0.248 −0.068 −0.068 0.000 −0.039
93 10,620.99 10,887.32 −266.33 0.224 −0.436 −0.434 0.003 −0.233
94 10,933.00 10,887.32 45.68 0.224 0.075 0.074 0.000 0.040
600 300
95 10,641.00 10,887.32 −246.32 0.224 −0.403 −0.401 0.002 −0.216
96 10,884.00 10,887.32 −3.32 0.224 −0.005 −0.005 0.000 −0.003
97 8110.00 7771.51 338.49 0.234 0.557 0.555 0.005 0.307
98 8014.00 7771.51 242.49 0.234 0.399 0.397 0.002 0.220
600 200
99 7917.00 7771.51 145.49 0.234 0.239 0.238 0.001 0.132
100 8436.00 7771.51 664.49 0.234 1.094 1.095 0.017 0.606
101 5529.00 5453.20 75.80 0.194 0.122 0.121 0.000 0.059
102 5168.00 5453.20 −285.20 0.194 −0.458 −0.456 0.002 −0.223
800 250
103 5127.00 5453.20 −326.20 0.194 −0.523 −0.521 0.003 −0.255
104 5193.00 5453.20 −260.20 0.194 −0.417 −0.416 0.002 −0.204
105 15,336.00 14,230.29 1105.71 0.202 1.783 1.805 0.038 0.908
106 14,590.00 14,230.29 359.71 0.202 0.580 0.578 0.004 0.291
1000 350
107 14,274.97 14,230.29 44.68 0.202 0.072 0.072 0.000 0.036
108 13,972.00 14,230.29 −258.29 0.202 −0.417 −0.415 0.002 −0.209
109 5009.00 5290.96 −281.96 0.242 −0.467 −0.465 0.003 −0.263
110 5514.00 5290.96 223.04 0.242 0.369 0.367 0.002 0.208
600 350
111 5435.00 5290.96 144.04 0.242 0.238 0.237 0.001 0.134
112 5219.00 5290.96 −71.96 0.242 −0.119 −0.118 0.000 −0.067

4.1. Model Diagnostics

For the accepted regression model, diagnostic tests were performed to assess its
validity. The first diagnostic test is the Normal Probability Plot of Residuals (Figure 4),
which assesses whether the residuals from the fifth-degree polynomial regression model
for the MTL data are normally distributed. In this plot, the externally studentized residuals
are plotted on the x-axis, while the corresponding normal cumulative probabilities are
plotted on the y-axis. Most of the residuals are close to the red reference line, indicating
that they follow a normal distribution. However, some deviations from normality are
observed, especially in the tails of the distribution. Some residuals, especially those on
the far right, deviate significantly from the line, indicating potential outliers or deviations
from normality.
Materials 2025,18,
Materials2025, 18,448
448 14 of 29 14 of 29
Materials 2025, 18, 448 14 of 29

Figure
Figure4.4.Normal probability
Normal plot plot
probability of residuals for thefor
of residuals fifth-degree polynomial
the fifth-degree regressionregression
polynomial model model
Figure 4. Normal probability plot of residuals for the fifth-degree polynomial regression model
applied
appliedtotoMTL
MTLdata.
data.
applied to MTL data.
Thesecond
The seconddiagnostic
diagnostic testtest is the
is the residuals
residuals versusversus predicted
predicted valuesvalues plot (Figure
plot (Figure 5), 5),
The evaluates
which second diagnostic
whether test
the isresiduals
the residuals
are versus predicted
randomly valuesindicating
distributed, plot (Figure
a 5), fit for
good
which evaluates whether the residuals are randomly distributed, indicating a good fit for
which evaluates whether the residuals are randomly distributed, indicating a good fit for
theregression
the regressionmodel.
model. TheThe
plotplot shows
shows the the externally
externally studentized
studentized residuals
residuals on theon the y-axis
y-axis
the regression model. The plot shows the externally studentized residuals on the y-axis
and
andthethepredicted
predicted MTLMTL values
valueson onthethe
x-axis. MostMost
x-axis. of the
ofresiduals are randomly
the residuals scattered
are randomly scattered
and the predicted MTL values on the x-axis. Most of the residuals are randomly scattered
around the horizontal line at zero, indicating that the model captures the
around the horizontal line at zero, indicating that the model captures the data well with data well with
around the horizontal line at zero, indicating that the model captures the data well with
no
noobvious
obvious patterns.
patterns.TheThe
absence of a clear
absence of apattern or trendor
clear pattern in trend
the residuals
in the indicates
residualsthatindicates
no obvious patterns. The absence of a clear pattern or trend in the residuals indicates that
there is no
thatisthere significant nonlinearity
is no significant or
nonlinearityheteroscedasticity (nonconstant
or heteroscedasticity variance).
(nonconstant This ran-
variance). This
there no significant nonlinearity or heteroscedasticity (nonconstant variance). This ran-
dom scatter supports the assumption that the residuals are independent and identically
random scatter supports the assumption that the residuals are
dom scatter supports the assumption that the residuals are independent and identicallyindependent and identically
distributed. The plot includes red lines at ±3.65 standard deviations. Note that none of the
distributed.
distributed. TheThe
plotplot includes
includes red at
red lines lines
±3.65 ±3.65 standard
atstandard deviations.
deviations. Noteofthat
Note that none the none of
residuals exceed these thresholds, indicating that there are no extreme outliers. The con-
the residuals exceed these thresholds, indicating that there are no
residuals exceed these thresholds, indicating that there are no extreme outliers. The con- extreme outliers. The
centration of the residuals around the zero line, with no discernible patterns, confirms that
concentration
centration of the residuals
of the residuals around the around the zero
zero line, withline, with no discernible
no discernible patterns,
patterns, confirms thatconfirms
the model is a good fit for most of the data points.
that
the the is
model model
a goodis fit
a good fit for
for most most
of the dataof points.
the data points.

Figure 5. Residuals vs. predicted values plot for the fifth-degree polynomial regression model applied
to MTL data.
Materials 2025, 18, 448 Figure 5. Residuals vs. predicted values plot for the fifth-degree polynomial regression model
15 ap-
of 29
plied to MTL data.

Thethird
The thirddiagnostic
diagnostic test
test is the
is the residuals
residuals versus
versus runrun orderorder
plotplot (Figure
(Figure 6), which
6), which as-
assesses whether the residuals are randomly distributed across the
sesses whether the residuals are randomly distributed across the sequence of observa- sequence of observations,
indicating
tions, the absence
indicating the absence of temporal
of temporal or or
sequence
sequence biasbiasininthe
themodel.
model. The plot plotplots
plotsthethe
externally studentized residuals on the y-axis and the run number on the x-axis. Most ofof
externally studentized residuals on the y-axis and the run number on the x-axis. Most
theresiduals
the residualsare arescattered
scatteredaround
aroundthe thehorizontal
horizontallinelineatatzero,
zero,indicating
indicatingthat thatthere
thereisisnono
clear pattern or trend over the sequence of observations. Although
clear pattern or trend over the sequence of observations. Although there is some variation there is some variation
ininthe
theresiduals,
residuals, there
there is is
nono consistent
consistent pattern
pattern or trend
or trend thatthat would
would suggest
suggest systematic
systematic er-
errors related to run order. This lack of a discernible trend suggests
rors related to run order. This lack of a discernible trend suggests that the residuals are that the residuals are
independentof
independent ofrun
run order,
order, supporting
supporting the the assumption
assumptionthat thatthere
thereisisnonoautocorrelation
autocorrelation in in
the
the residuals. The Durbin-Watson statistic, which tests for the presence of autocorrelationin
residuals. The Durbin-Watson statistic, which tests for the presence of autocorrelation
inthe
theresiduals,
residuals,isis0.8776
0.8776withwith an
an autocorrelation value of of 0.5609.
0.5609.AADurbin-Watson
Durbin-Watsonvalue value
close to 2 indicates no autocorrelation, while values significantly lower
close to 2 indicates no autocorrelation, while values significantly lower or higher indicate or higher indicate
positiveorornegative
positive negativeautocorrelation,
autocorrelation,respectively.
respectively.The Theobserved
observedvaluevalue(0.8776)
(0.8776)indicates
indicates
some positive autocorrelation, which may require further investigation.
some positive autocorrelation, which may require further investigation. The plot includes The plot includes
redlines
red ±3.64
linesatat±3.64 standard
standard deviations,
deviations, andand none
none ofofthe
theresiduals
residualsexceed
exceedthese
thesethresholds,
thresholds,
indicating that there are no extreme
indicating that there are no extreme outliers. outliers.

Residualsvs.
Figure6.6.Residuals
Figure vs.run
runorder
orderplot
plotfor
forthe
thefifth-degree
fifth-degreepolynomial
polynomialregression
regressionmodel
modelapplied
appliedtoto
MTL data.
MTL data.

Thefourth
The fourthdiagnostic
diagnostictest
testisisthe
theCook
Cookdistance
distanceplot
plot(Figure
(Figure7),
7),which
whichevaluates
evaluatesthe
the
influenceofofeach
influence eachobservation
observationon onthetheregression
regressionmodel.
model.Cook’s
Cook’sDistance
Distancemeasures
measureshow
how
much the regression coefficients change when a particular observation
much the regression coefficients change when a particular observation is removed, help- is removed, helping
to identify
ing influential
to identify datadata
influential points that that
points may may disproportionately
disproportionatelyaffectaffect
the model [46]. Cook’s
the model [46].
Distance for the i-th observation is calculated using the following
Cook’s Distance for the 𝑖-th observation is calculated using the following formula: formula:

e𝑒i2 hℎi
!
D𝐷i == (7)
(7)
p𝑝· MSE
∙ 𝑀𝑆𝐸 ((1 1−− hℎ ))2
i
where 𝐷 is the Cook’s Distance for the 𝑖-th observation. 𝑒 is the residual for the 𝑖-th ob-
where Di is the Cook’s Distance for the i-th observation. ei is the residual for the i-th
servation (i.e., the difference between the observed and fitted values for that observation).
observation (i.e., the difference between the observed and fitted values for that observation).
𝑝 is the number of parameters in the model, including the intercept. 𝑀𝑆𝐸 is the mean
p is the number of parameters in the model, including the intercept. MSE is the mean
squared error of the regression model. hi is the leverage of the i-th observation, which is the
Materials
Materials2025,
2025,18,
18,448
448 16
16 of
of 29
29

Materials 2025, 18, 448 16 of 29

squared
squarederror
errorof
ofthe
theregression model. ℎℎ isisthe
regressionmodel. theleverage
leverageof the𝑖-th
ofthe 𝑖-thobservation,
observation,which
whichisis
the𝑖-th
the
i-th 𝑖-th diagonal
diagonal
diagonal element
element
element of
ofthe
of the hathat
the hatmatrixH𝐻
matrix
matrix =𝑋(𝑋′𝑋)
=𝐻=X(X𝑋(𝑋′𝑋)
′ X) −−−1𝑋′.
1X ′
1𝑋′..Most
Mostobservations
Most observations
observations have
have Cook’s
haveCook’s
Cook’s
Distance
Distancevalues
Distance valuesclose
closeto zero,
tozero, indicating
zero,indicating that
indicatingthat they
thatthey have
theyhave minimal
haveminimal influence
minimalinfluence on
influenceonon the
the
the regres-
regres-
regression
sion
sionmodel.
model. model. The
The red red
Theline line
redatline at
atCook’s
Cook’s Cook’s Distance
DistanceDistance of
of 0.976of0.976
0.976serves
serves serves as
asaathreshold;
as a threshold; points points
threshold; pointsabove
above above
this line
this line
lineare
areconsidered
thisconsidered
are considered highly
highlyinfluential.
highly influential. influential. In
Inthis
In this plot,this plot,
none ofnone
plot, none of
ofthe
theobservations
exceedexceed
observations
the observations exceed the
the
the Cook’s
Cook’s
Cook’s Distance
Distance Distance
threshold threshold
of 0.976,of
threshold of 0.976,
0.976, indicating
indicating that therethat
indicating that there
are nothere are
are no
highly no highly
highly influential
influential pointsdata
influential
data data
in the
points
points
data in
inthe
set. thedata
dataset.
set.

Figure Cook’sDistance
Figure7.7.Cook’s
Figure Cook’s Distanceplot
Distance plot
plot for
for
for the
the
the fifth-degree
fifth-degree
fifth-degree polynomial
polynomial
polynomial regression
regression
regression model
model
model applied
applied
appliedto to MTL
toMTL
MTL data.data.
data.

The
The fifth
The fifth diagnostic
fifth diagnostic testisis
diagnostic test
test isthe
thepredicted
the predictedvs.
predicted vs.actual
vs. actualplot
actual plot(Figure
plot (Figure8),
(Figure 8),which
8), whichevaluates
which evaluates
evaluates
how
how well
how wellthe
well theregression
the regressionmodel
regression model
model predicts thethe
predicts
predicts observed
the observeddata.
observed TheThe
data.
data. distribution
The of points
distribution
distribution of along
of points
points
the line
along
alongtheindicates
theline a strong
lineindicates correlation
indicatesaastrong between
strongcorrelation predicted
correlationbetween
betweenpredictedand
predictedandactual values,
andactual confirming
actualvalues,
values,confirm-
confirm-the
model’s
ing
ingthe ability to
themodel’s
model’s accurately
ability
abilityto predictpredict
toaccurately
accurately MTL. MTL.
predict MTL.

Figure 8. Predicted vs. actual values plot for the fifth-degree polynomial regression model applied to
MTL data.
 Figure 8. Predicted vs. actual values plot for the fifth-degree polynomial regression model applied
Materials 2025, 18, 448 to MTL data. 17 of 29

The sixth diagnostic test is the DFFITS vs. Run Number plot (Figure 9), which exam-
inesThe
the sixth
influence of each
diagnostic testobservation
is the DFFITS onvs.the
Runfitted values
Number of (Figure
plot the regression
9), whichmodel. DFFITS
examines
(Difference
the influencein ofFits)
eachmeasures
observation how onmuch an observation
the fitted affects
values of the the fitted
regression valueDFFITS
model. and is used
to identifyininfluential
(Difference points
Fits) measures howthat may
much andisproportionately
observation affects the affect thevalue
fitted model.
and DFFITS
is used is a
to identify influential points that may disproportionately affect the model.
diagnostic measure that quantifies the influence of a single observation on the fitted DFFITS is values
a
diagnostic measuremodel.
of the regression that quantifies the influence
It calculates the change of a single
in the observation
predicted valueon thewhen
fittedan
values
observa-
of the regression model. It calculates the change in the predicted
tion is excluded from the model. The formula for DFFITS is given by: value when an observation
is excluded from the model. The formula for DFFITS is given by:
𝑦 −𝑦( )
𝐷𝐹𝐹𝐼𝑇𝑆 = (8)
ŷi − ŷ𝑠i(− i) ℎ
DFFITSi = √( ) (8)
s(−i) hi
where 𝑦 is the predicted value with all observations included. 𝑦 ( ) is the predicted
valueŷwith
where i is the 𝑖-th observation
thepredicted value withexcluded. 𝑠( ) is included.
all observations the standardŷi(−ierror of the regression with
) is the predicted value
the 𝑖-th
with observation
the i-th observation excluded.ℎ s(−
excluded. isi)the leverage
is the standard the 𝑖-th
of error observation.
of the Manytheof the
regression with
DFFITS values lie around izero, indicating that individual observations generally
i-th observation excluded. h is the leverage of the i-th observation. Many of the DFFITShave a
values
minimallie around zero,
influence onindicating
the fitted that
valuesindividual observations
of the model. generally have
As observations move a minimal
further from
influence on the
the zero line, fitted
they values
have of theimpact
a greater model.on Astheobservations
model’s fittedmove furtherThe
values. from the zero blue
horizontal
line, they have a greater impact on the model’s fitted values. The horizontal blue lines
lines at ±1.29904 serve as thresholds, calculated based on the formula q
p ±2 , where 𝑝 is
at ±1.29904 serve as thresholds, calculated based on the formula ±2 n , where p is the
the number
number of predictors
of predictors (including
(including the intercept),
the intercept), and
and n is the 𝑛 is the
number of number of observations.
observations. Values
Values these
beyond beyond these
lines linesinfluential
suggest suggest influential
observationsobservations [47].shows
[47]. The plot The plot
thatshows
none ofthat
thenone
of the observations
observations exceed
exceed these these thresholds,
thresholds, indicatingindicating
that therethat
are there are no
no highly highly influential
influential data
points in the data
data points set.data
in the Thisset.
suggests that the model
This suggests is robust
that the modeland not unduly
is robust and influenced
not undulybyinflu-
any single observation.
enced by any single observation.

Figure9.9.DFFITS
Figure DFFITSvs.vs.run number
run plot
number forfor
plot thethe
fifth-degree polynomial
fifth-degree regression
polynomial model
regression applied
model to
applied to
MTL data.
MTL data.

The next diagnostic test is the DFBETAS vs. Run Number plot (Figure 10), which
The next diagnostic test is the DFBETAS vs. Run Number plot (Figure 10), which
examines the influence of each observation on the estimated regression coefficients. DF-
examines the influence of each observation on the estimated regression coefficients.
BETAS, which stands for “Difference in Beta”, is a diagnostic measure that assesses the
DFBETAS,
impact which
of each standsdata
individual for point
“Difference in Beta”, is
on the estimated a diagnostic
regression measureSpecifically,
coefficients. that assesses
it the
Materials 2025, 18, 448 18 of 29
Materials 2025, 18, 448 18 of 29

impact ofthe
measures each individual
change data pointcoefficient
in a regression on the estimated regression
when an coefficients.
observation Specifically,
is omitted from the
it measures the change in a regression
analysis. The formula for DFBETAS is: coefficient when an observation is omitted from the
analysis. The formula for DFBETAS is:
β̂ j 𝛽− −
β̂ j𝛽
(−i )
DFBETAS ( )
ij = = q
𝐷𝐹𝐵𝐸𝑇𝐴𝑆 (9)
s(−i) ( X ′ X )− 1 (9)
𝑠( ) (𝑋′𝑋)jj

where
whereβ̂𝛽j isisthe
theestimated
estimatedcoefficient
coefficientfor predictorj 𝑗with
forpredictor withall
allobservations
observationsincluded.
included.β̂𝛽j(−( i) )
isisthe
theestimated
estimatedcoefficient
coefficient for predictor jj𝑗jwith
for predictor the i-th
with the 𝑖-th observation excluded. s𝑠(−i) isisthe
observation excluded. ( the
)
standard
standard error
errorof
ofthe
theregression
regressionwith
withthe 𝑖-thobservation
thei-th observationexcluded. X ′ X )−
excluded.( (𝑋′𝑋)
1
jj is
is the 𝑗-th
the j-th
diagonal element of the inverse of the design matrix X ′ X.
diagonal element of the inverse of the design matrix 𝑋′𝑋.

Figure 10.DFBETAS
Figure10. DFBETASvs.
vs.run
runnumber
numberplot
plotfor
forthe
thefifth-degree
fifth-degreepolynomial
polynomialregression
regressionmodel
modelapplied
applied
to
toMTL
MTLdata.
data.

DFBETAS
DFBETASvaluesvaluesindicate
indicatehowhowmuch
muchan anindividual
individual observation
observation influences
influences the
the regres-
regres-
sion coefficients. Large DFBETAS values indicate that the observation
sion coefficients. Large DFBETAS values indicate that the observation has a significant has a significant
impact
impacton onthe
thecorresponding
corresponding coefficient,
coefficient, potentially
potentiallyindicating an influential
indicating data data
an influential pointpoint
that
may disproportionately affect the model’s estimates. In the plot, the DFBETAS
that may disproportionately affect the model’s estimates. In the plot, the DFBETAS values values for
the
forintercept are plotted
the intercept on theon
are plotted y-axis, and theand
the y-axis, runthe
number is plottedison
run number the x-axis.
plotted on theMany of
x-axis.
the DFBETAS values are around zero, indicating that most observations
Many of the DFBETAS values are around zero, indicating that most observations have have little to no
effect
little on
to the estimated
no effect regression
on the estimatedcoefficients.
regressionThecoefficients.
horizontal blueThe lines at ±0.284573
horizontal serve
blue lines at
√
as thresholds, calculated using the formula ± 2/ n where n is the number
±0.284573 serve as thresholds, calculated using the formula ±2/√𝑛 where 𝑛 is the num- of observations.
Values
ber of beyond these lines
observations. Valuesindicate
beyond that the corresponding
these lines indicate that observations have a significant
the corresponding observa-
impact
tions have a significant impact on the model’s coefficients [48]. The plot fall
on the model’s coefficients [48]. The plot shows that all observations within
shows thatthe
all
thresholds of ± 0.284573, indicating that none of the data points have an undue
observations fall within the thresholds of ±0.284573, indicating that none of the data points influence
on
havethean
regression coefficients.
undue influence Thisregression
on the indicates that the model
coefficients. Thisis stable, andthat
indicates thethe
coefficients
model is
are not overly sensitive to any single observation.
stable, and the coefficients are not overly sensitive to any single observation.
Both measures are useful for identifying influential data points, but they provide
Both measures are useful for identifying influential data points, but they provide dif-
different perspectives on how an observation affects the regression model. DFFITS is
ferent perspectives on how an observation affects the regression model. DFFITS is con-
concerned with overall prediction accuracy, while DFBETAS is concerned with the stability
cerned with overall prediction accuracy, while DFBETAS is concerned with the stability
of the regression coefficients.
of the regression coefficients.
Materials 2025, 18, 448 19 of 29

4.2. Principles of Hill-Climbing Algorithm

Hill climbing is an optimization algorithm used to find the best solution to a problem
by iteratively improving the current solution based on a fitness function. Starting with
an arbitrary initial solution, the algorithm evaluates its fitness and generates neighboring
solutions by making small changes. It then selects the neighbor with the best fitness as
the new current solution. This process is repeated until a stopping criterion is met, such
as a fixed number of iterations, a time limit, or when no better neighbors are found. Hill
climbing is a local search algorithm that focuses on improving the current solution and
takes a greedy approach by always moving to the best neighbor solution. However, it
can become stuck in local optima, i.e., solutions that are better than their neighbors but
not the best overall. Variants such as Steepest Ascent, Stochastic, and First-Choice Hill
Climbing help to solve this problem. Hill climbing is easy to implement and efficient
for small problem spaces, making it useful in applications such as artificial intelligence,
operations research, and machine learning.
The process begins with an initial guess

S0 = initial guess. (10)

The fitness of a solution S is evaluated using a fitness function f (S), which maps the
solution to a real number indicating its quality.

f : S → R.

The algorithm generates a set of neighboring solutions N(S) by making small pertur-
bations to the current solution S.

N (S) = S′ S′ is a neighbor o f S

(11)

Among the generated neighbors, the algorithm selects the neighbor S′ with the optimal
fitness value, either the highest for maximization problems or the lowest for minimiza-
tion problems.
S′ = argmax S′ ∈ N (S) f (S′ ) (for maximization)
(12)
S′ = argminS′ ∈ N (S) f (S′ ) (for minimization)

If the best neighbor S′ improves the fitness function compared to the current solution
S, the algorithm updates S to S′ .

If f (S′ ) > f (S) (for maximization), thenS = S′

(13)
If f (S′ ) < f (S) (for minimization), thenS = S′
The algorithm iterates through the steps of generating neighbors and selecting the best
one until a predefined stopping criterion T is met. This criterion could be a fixed number
of iterations, a time limit, or the absence of further improvements [49,50].

5. Optimization
For our model, hill-climbing optimization was performed to fine-tune the welding
parameters to achieve the best possible MTL. The process involved starting with an initial
set of welding parameters and iteratively adjusting these parameters to explore their
neighboring values. At each step, the fitness function, defined as the MTL, was evaluated
to identify the optimal combination of welding parameters. By selecting the parameters
that maximized the MTL, the algorithm iteratively moved toward the best solution, thereby
improving the overall performance and reliability of the welding process. This approach
Materials 2025, 18, 448 20 of 29
Materials 2025, 18, 448 20 of 29

ensured that the parameters were effectively optimized, resulting in the highest achievable
MTL for the
through thisgiven welding
process conditions.
are a spindle speed ofThe
1100optimal
rpm andparameters identified
a welding speed of 332through
mm/min,this
process
which are
are apredicted
spindle speed of 1100
to achieve rpm and
an MTL a welding
of 16,852 speed11ofshows
N. Figure 332 mm/min, which are
the optimization
predicted to
parameters. achieve an MTL of 16,852 N. Figure 11 shows the optimization parameters.

Figure 11.Optimization
Figure11. Optimization results
results for welding
welding parameters
parametersusing
usinghill
hillclimbing.
climbing.

The
Theresponse
responsesurface
surfaceplot
plot(Figure
(Figure12)
12)shows
showsthe
therelationship
relationshipbetween
betweenMTL
MTLand
andtwo
twokey
welding parameters:
key welding spindle
parameters: speed
spindle (rpm)
speed andand
(rpm) welding
weldingspeed (mm/min).
speed (mm/min).

Figure12.
Figure 12.Response
Responsesurface
surface plot
plot (a)
(a) and
and contour
contourplot
plot(b)
(b)for
forMTL
MTLasas
a function of of
a function spindle speed
spindle andand
speed
welding speed. Different colors indicate MTL values.
welding speed. Different colors indicate MTL values.

Thesurface
The surfaceplot
plot has
has regions
regions of
of different
differentheights
heightsthat
thatindicate how
indicate howchanges in spindle
changes in spindle
speed and welding speed affect the MTL. The highest region of the surface corresponds
speed and welding speed affect the MTL. The highest region of the surface corresponds to
the maximum MTL values, represented by the red and yellow areas on the plot. The plot
Materials 2025, 18, 448 21 of 29

shows a peak MTL value around the spindle speed of approximately 1100 rpm and welding
speed of approximately 332 mm/min, which is consistent with the optimal parameters
determined by the hill-climbing optimization.

6. Confirmation Test
The specimen subjected to the confirmation test was FSW in a lap joint configuration
using the optimal parameters identified by the optimization process: a spindle speed of
1100 rpm and a welding speed of 332 mm/min. The test procedure followed the same
protocol as described in the Materials and Methods section to ensure consistency in the
evaluation of the welds. The measured load capacity for the FSW sample in the confirmation
test ranged from 16,200 to 17,300 N, which is in good agreement with the predicted values
from the optimization model. This confirms the accuracy and reliability of the optimization
process for these specific parameters.

6.1. Macro and Microstructure Analysis

Macro and microstructure analyses were performed on the specimen to further validate
the quality of the weld (Figure 13). The image provided illustrates these analyses, with
six different regions marked and examined in detail. The specimen has been electropolished
to reveal its microstructural features, and the image shows a cross-section of the FSW joint.
• Region 1 (Base Metal (BM), Heat-Affected Zone (HAZ), Thermomechanically Affected
Zone (TMAZ), and Stir Zone (SZ)). This region shows the transition from the BM
through the HAZ and the TMAZ into the SZ. The microstructure shows a gradual
refinement of the grains from the BM to the SZ, indicating effective thermal and me-
chanical processing during FSW. The distinct zones highlight the gradient of thermal
and mechanical effects on the material [51].
• Region 2 (TMAZ, HAZ): Similar to Region 1, this region provides a detailed view
of the microstructural changes within the TMAZ and the HAZ. Grain refinement is
evident as the material moves toward the stir zone, showing the progressive effect
of the welding process on the material structure. The shape of the grains is a direct
result of the compression process, which flattens them into small fractions and causes
further grain refinement. A similar evolution of the microstructure was shown in the
work of Orlowska et al. [52].
• Region 3 (SZ): The stir zone exhibits a uniform and refined grain structure, indicating
effective material mixing and recrystallization during the welding process. This region
confirms the high quality of the stir zone, which is critical to the integrity and strength
of the weld.
• Region 4 (SZ, TMAZ, HAZ): This region illustrates the microstructural characteristics
at the interface between the stir zone (SZ), thermomechanically affected zone (TMAZ),
and heat-affected zone (HAZ). The boundaries are well defined and demonstrate the
effectiveness of the welding parameters in producing a strong joint with distinct zones
that contribute to the overall mechanical properties of the weld.
• Region 5 (SZ—Hooking): This section shows a hooking defect within the SZ. The
hooking defect is characterized by a curved, hook-like shape at the interface between
the joined materials. [53]. Despite the presence of this defect, the overall grain structure
remains consistent with the expected characteristics of a properly welded stir zone. The
hooking defect is identified during mechanical testing as a potential crack initiation
site that can compromise the structural integrity of the weld.
• Region 6 (SZ—Material Flow Lines): The microstructure in this region shows material
flow lines within the stir zone (SZ). The visible lines are likely to flow lines of the
material, with changes in shading possibly reflecting the presence of “onion rings”
Materials 2025, 18, 448
Materials 2025, 18, 448 22 of2229
of 29

that are characteristic of FSW. These features indicate effective stirring and mixing of
that are characteristic of FSW. These features indicate effective stirring and mixing of
the
the materialwithout
material withoutthe
the presence of cracks,
presence of cracks,confirming
confirmingthe
theoverall
overall quality
quality of of
thethe weld
weld
ininthis
thisregion
region[54].
[54].

Figure 13. Macro and microstructure analysis of FSW sample. The colorful points represent different
crystallographic orientations of the material’s grains.
 Figure 13. Macro and microstructure analysis of FSW sample. The colorful points represent di
ent crystallographic orientations of the materialʹs grains.

Materials 2025, 18, 448 23 of 29

Figure 14 presents the view of the specimen after failure, showing crack initia
and propagation through the identified flaw. The confirmation test validates the effec
nessFigure
of the14optimized
presents theparameters by achieving
view of the specimen the predicted
after failure, loadinitiation
showing crack capacities
andand dem
strating robust
propagation weld
through quality through
the identified flaw. Thedetailed macro-
confirmation and microstructural
test validates the effectivenessexamina
of theconsistency
The optimized parameters
between by achievingand
predicted the predicted load capacities
actual performance and demonstrat-
underscores the reliabili
ing robust weld quality through detailed macro- and microstructural examination.
the optimization process used in this study. The observed macrostructure and The micros
consistency between predicted and actual performance underscores the reliability of the
ture are typical of AA2024-T3 alloys, as reported in the publication by Myśliwiec
optimization process used in this study. The observed macrostructure and microstructure
[55], which discusses the butt welding of thin AA2024 sheets. The application of adva
are typical of AA2024-T3 alloys, as reported in the publication by Myśliwiec et al. [55],
optimization
which discussesmethods for FSW
the butt welding of process parameters
thin AA2024 sheets. Theusing commercial
application software, suc
of advanced
Design Expert
optimization 12, has
methods for been successfully
FSW process demonstrated
parameters in thesoftware,
using commercial studies of Myśliwiec
such as
[56]. Expert 12, has been successfully demonstrated in the studies of Myśliwiec et al. [56].
Design

Figure FSW
Figure14.14. sample
FSW afterafter
sample tensile test. test.
tensile
6.2. The Microhardness Analysis
6.2. The Microhardness Analysis
The next step in evaluating the formed lap FSW joint was to measure the Vickers
The nextinstep
microhardness in evaluating
the cross-section. Thethe formed lap
measurement FSW and
method jointresults
was to aremeasure
shown in the Vic
Figure 15. The microhardness
microhardness profile of theThe
in the cross-section. lap measurement
FSW joint for themethodAA2024-T3 andalloy shows
results are show
the following key features: the microhardness of the parent material
Figure 15. The microhardness profile of the lap FSW joint for the AA2024-T3 alloy sh is at 130–140 HV,
increasing the temperature in the heat-affected zone causes a gradual increase in micro-
the following key features: the microhardness of the parent material is at 130–140
hardness to a value of 160–170 HV. The peak microhardness (210 HV) in the center of
increasing the temperature in the heat-affected zone causes a gradual increase in m
the weld indicates that the refined grains in the weld nugget (the mixing zone) have a
hardness
higher to a value
hardness of 160–170
and, therefore, HV.strength
higher The peak than microhardness
the base material. (210TheHV) in theincenter o
change
weld indicates
microhardness thatweld
in the the nugget
refinedis grains in the
significant, weld
which nugget
is typical for(the
FSWmixing
joints ofzone) have a hi
2024-T3
hardness and, therefore, higher strength than the base material. The change in microh
alloy, but we mostly observe a local decrease in microhardness in this region caused mainly
by thein
ness dissolution
the weldofnuggetAlMgCu reinforcing phase
is significant, which particles due for
is typical to high
FSWtemperatures [57]. alloy
joints of 2024-T3
Another factor in the decrease of microhardness is also the movement of dislocations due
we mostly observe a local decrease in microhardness in this region caused mainly by
to intense mechanical deformation [58]. However, in this particular case, we observe the
dissolution of AlMgCu reinforcing phase particles due to high temperatures [57]. Ano
phenomenon of a significant increase in microhardness in the weld nugget. On the one
factorthe
hand, in mechanism
the decrease of microhardness
of grain reduction due to is intense
also theplastic
movement of dislocations
deformation is responsibledue to int
mechanical
for deformation
this. According [58]. However,
to the Hall-Pecha in this
relationship, particular
grain reduction case, we in
results observe
increasedthe pheno
non of aand
hardness significant increase
strength [59]. Another in microhardness in the weld
phenomenon is probably due nugget.
to the high Onwelding
the one hand
speed and lower temperature in the weld nugget by which the strengthening
mechanism of grain reduction due to intense plastic deformation is responsible phase was not for
dissolved [60]. In addition, the overlap configuration causes an increase
According to the Hall-Pecha relationship, grain reduction results in increased hard in the cross-section
of the weld, which leads to better cooling and heat transfer to the environment and tooling.
and strength [59]. Another phenomenon is probably due to the high welding speed
This results in higher longitudinal stresses in the joint, as shown in Staron et al. [61]. The
lower temperature in the weld nugget by which the strengthening phase was not
solved [60]. In addition, the overlap configuration causes an increase in the cross-sec
of the weld, which leads to better cooling and heat transfer to the environment
Materials 2025, 18, 448 24 of 29

Materials 2025, 18, 448 24 of 29

tooling. This results in higher longitudinal stresses in the joint, as shown in Staron et al
[61].appearance
The appearance
of these of thesecan
stresses stresses
have a can haveeffect
negative a negative effect
on fatigue andon fatigue
crack and crack
propagation in prop
agation in this area. The result can be the appearance of a hook defect in the weld.
this area. The result can be the appearance of a hook defect in the weld.

Figure 15. Distribution of microhardness for FSW joint.

Figure 15. Distribution of microhardness for FSW joint.
7. Discussion
7. Discussion
The findings of this study provide a comprehensive understanding of the influence
ofThe
spindle speed of
findings and welding
this studyspeed on the
provide a mechanical
comprehensive performance and microstructural
understanding of the influence
characteristics of lap friction stir welded (FSW) joints in AA2024-T3 aluminum alloy. The
of spindle speed and welding speed on the mechanical performance and microstructura
results clearly demonstrate the nonlinear effects of process parameters on the maximum
characteristics of lap friction stir welded (FSW) joints in AA2024-T3 aluminum alloy. The
tensile load (MTL), with significant interaction terms identified through the fifth-order
results clearlyregression
polynomial demonstrate model. the nonlinear
Such complexityeffects of process
in parameter parameters
interaction has beenonsimilarly
the maximum
tensile
notedload (MTL),
in other with
recent workssignificant
applying interaction terms identified
advanced modeling through the
to FSW optimization fifth-order
[62–64].
polynomial regression
This highlights model.optimal
that achieving Such complexity in parameter
conditions requires interaction
not only controlling has been simi
individual
factors but also understanding their combined effects, a challenge addressed
larly noted in other recent works applying advanced modeling to FSW optimization [62– increasingly
64]. by data-driven
This highlights andthat
hybrid optimization
achieving methods
optimal [65,66]. requires not only controlling indi
conditions
The macro- and microstructural analyses confirm that the optimized parameters
vidual factors but also understanding their combined effects, a challenge addressed in
effectively refine the grain structure and produce defect-free regions essential for high
creasingly by data-driven and hybrid optimization methods [65,66].
tensile strength. Grain refinement and improved joint properties following optimized
The macro-
parameter and microstructural
selection have been reported analyses confirm
previously, that the
supporting the optimized
influence ofparameters
proper ef
fectively
thermalrefine the grain structure
and mechanical conditions and produce defect-free
on recrystallization regionsNevertheless,
kinetics [67,68]. essential forthehigh ten
sile hooking
strength. defect
Grain observed here, often
refinement seen in lap FSW
and improved jointconfigurations, remains problematic.
properties following optimized param
eter selection have been reported previously, supporting the influence of proper therma
and mechanical conditions on recrystallization kinetics [67,68]. Nevertheless, the hooking
defect observed here, often seen in lap FSW configurations, remains problematic. Recen
studies have highlighted how hooking can serve as a stress concentrator and limit the
Materials 2025, 18, 448 25 of 29

Recent studies have highlighted how hooking can serve as a stress concentrator and limit
the joint’s mechanical integrity [69,70]. This defect arises from complex material flow
patterns and tool geometry factors, including insufficient plunge depth or inappropriate
tool tilt angle, which disturb uniform material flow [53]. The interplay of tool design, heat
input, and material plasticity in creating or mitigating hooking defects has become a focal
point in recent investigations aiming to reduce defect prevalence [71,72].
In addition, the microhardness analysis revealing increased hardness up to 210 HV in
the stir zone is consistent with work showing that refined grain structures and controlled
thermal cycles enhance mechanical properties [73]. Minor hardness variations, particu-
larly near the TMAZ, may indicate localized thermal gradients and partial dissolution of
strengthening precipitates, aligning with reported thermal effects in similarly optimized
FSW joints [63,67].
The use of polynomial regression coupled with a hill-climbing algorithm successfully
identified optimal parameters (spindle speed = 1100 rpm and welding speed = 332 mm/min)
that maximize joint strength. Confirmation tests validated these optimal conditions, pro-
ducing MTL values in good agreement with model predictions and corroborating the
efficacy of integrating statistical modeling with iterative optimization for FSW parameter
tuning. Such approaches reduce reliance on extensive trial-and-error experimentation and
can lead to accelerated development cycles and more reliable welding protocols.
Limitations include the controlled laboratory conditions that may differ from industrial
settings where factors like tool wear, material batch variations, and dynamic temperature
profiles can affect performance. Future work could integrate real-time monitoring and
advanced numerical simulations, such as coupled thermo-mechanical finite element models
or machine learning-driven adaptive control systems, to enhance understanding and
predictability of material flow and defect formation under variable conditions. Expanding
the parameter search space and employing global optimization algorithms or evolutionary
strategies could further improve weld quality and robustness.

8. Conclusions
• The findings of the conducted research led to the following conclusions:
• A fifth-degree polynomial regression model was developed to predict the maximum
tensile load (MTL) of friction stir welded (FSW) lap joints, achieving high predictive
accuracy with minimal overfitting.
• The experimental results demonstrated a range of MTL values, from 1912 N to 15,336 N,
across the tested ranges of spindle speed and welding speed.
• The hill-climbing optimization algorithm identified the optimal welding parameters,
which were a spindle speed of 1100 rpm and a welding speed of 332 mm/min, resulting
in an MTL of 16,852 N.
• The results of the response surface analysis corroborate the significant interaction be-
tween spindle speed and welding speed, delineating regions of maximum MTL values.
• The confirmation tests served to validate the optimized parameters, which were shown
to achieve high load capacities and to demonstrate robust weld quality through macro-
and microstructural analyses.
• The microhardness profile revealed a peak hardness of approximately 210 HV in the
weld center, which can be attributed to grain refinement and the absence of phase
dissolution. This indicates that the joint exhibits superior strength.
Materials 2025, 18, 448 26 of 29

Author Contributions: P.M.: conceptualization, methodology, writing—original draft. A.K.: writing-

review and editing, supervision. P.S.: data curation. M.Z.: formal analysis. R.O.: visualization,
resources. H.A.D.: validation, project administration. W.J.: supervision, funding acquisition. All
authors have read and agreed to the published version of the manuscript.

Funding: This research received no external funding.

Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Not applicable.

Data Availability Statement: The original contributions presented in this study are included in the
article. Further inquiries can be directed to the corresponding author.

Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Thomas, W.M.; Nicholas, E.D. Friction stir welding for the transportation industries. Mater. Des. 1997, 18, 269–273. [CrossRef]
2. Mishra, R.S. Friction Stir Welding and Processing: Science and Engineering; Springer Cham: New York, NY, USA, 2014.
3. He, X.; Gu, F.; Ball, A. A review of numerical analysis of friction stir welding. Prog. Mater. Sci. 2014, 65, 1–66. [CrossRef]
4. Chumaevskii, A.; Amirov, A.; Ivanov, A.; Rubtsov, V.; Kolubaev, E. Friction Stir Welding/Processing of Various Metals with
Working Tools of Different Materials and Its Peculiarities for Titanium Alloys: A Review. Metals 2023, 13, 970. [CrossRef]
5. Ahmed, S.; Rahman, R.; Awan, A.; Ahmad, S.; Akram, W.; Amjad, M.; Yahya, M.; Rahimian Koloor, S.R. Optimization of Process
Parameters in Friction Stir Welding of Aluminum 5451 in Marine Applications. J. Mar. Sci. Eng. 2022, 10, 1539. [CrossRef]
6. Sengupta, S.; Basak, S.; Peters, R.A. Particle Swarm Optimization: A Survey of Historical and Recent Developments with
Hybridization Perspectives. Mach. Learn. Knowl. Extr. 2019, 1, 10. [CrossRef]
7. Sabry, I.; Gadallah, N.; Abu-Okail, M. Optimization of friction stir welding parameters using response surface methodology. IOP
Conf. Ser. Mater. Sci. Eng. 2020, 973, 012017. [CrossRef]
8. Kesharwani, R.K.; Panda, S.K.; Pal, S.K. Multi Objective Optimization of Friction Stir Welding Parameters for Joining of Two
Dissimilar Thin Aluminum Sheets. Procedia Mater. Sci. 2014, 6, 178–187. [CrossRef]
9. Verma, S.; Misra, J.P.; Singh, J.; Batra, U.; Kumar, Y. Prediction of tensile behavior of FS welded AA7039 using machine learning.
Mater. Today Commun. 2021, 26, 101933. [CrossRef]
10. Maleki, E. Artificial neural networks application for modeling of friction stir welding effects on mechanical properties of 7075-T6
aluminum alloy. IOP Conf. Ser. Mater. Sci. Eng. 2015, 103, 012034. [CrossRef]
11. Kumar, S.D.; Pugazhenthi, R.; Ajith Arul Danial, S.; Swaminathan, G. Optimization of Dissimilar aluminum alloy by Friction stir
welding Process Control variables with Multiple Objectives. J. Phys. Conf. Ser. 2021, 2040, 012042. [CrossRef]
12. Khourshid, A.M.; El-Kassas, A.M.; Hindawy, H.M.; Sabry, I. Optimization of Friction Stir Welding Parameters for Joining
Aluminum Pipes Using Regression Analysis. RG Rep. 2016, 2, 1–12. [CrossRef]
13. Cho, M.; Gim, J.; Kim, J.H.; Kang, S. Development of an Artificial Neural Network Model to Predict the Tensile Strength of
Friction Stir Welding of Dissimilar Materials Using Cryogenic Processes. Appl. Sci. 2024, 14, 9309. [CrossRef]
14. Albaijan, I.; Ahmed, M.M.Z.; El-Sayed Seleman, M.M.; Touileb, K.; Habba, M.I.A.; Fouad, R.A. Optimization of Bobbin Tool
Friction Stir Processing Parameters of AA1050 Using Response Surface Methodology. Materials 2022, 15, 6886. [CrossRef]
[PubMed]
15. Sabry, I.R.S.; Idrisi, A.H.; Mourad, A.H.I. Friction Stir Welding Process Parameters Optimization Through Hybrid Multi-Criteria
Decision-Making Approach. Int. Rev. Model. Simul. (IREMOS) 2021, 14, 75–84. [CrossRef]
16. Jasim, S.; Saleh, N.; Jasim, R. Numerical Simulation and Optimization of Friction Stir Welding Parameters. Basrah J. Eng. Sci. 2023,
23, 72–80. [CrossRef]
17. Marian, M.; Tremmel, S. Current Trends and Applications of Machine Learning in Tribology—A Review. Lubricants 2021, 9, 86.
[CrossRef]
18. Prabhu, S.R.; Shettigar, A.; Herbert, M.A.; Rao, S.S. Parameter investigation and optimization of friction stir welded AA6061/TiO2
composites through TLBO. Weld World 2022, 66, 93–103. [CrossRef]
19. Mishra, A.; Sefene, E.M.; Tsegaw, A.A. Process parameter optimization of Friction Stir Welding on 6061AA using Supervised
Machine Learning Regression-based Algorithms. arXiv 2021, arXiv:2109.00570.
20. Constantin, M.A.; Iordache, M.D.; Nitu, E.L.; Diakhaté, M.; Demmouche, Y.; Dhondt, M.; Bădulescu, C. An efficient strategy for
3D numerical simulation of friction stir welding process of pure copper plates. IOP Conf. Ser. Mater. Sci. Eng. 2020, 916, 012021.
[CrossRef]
Materials 2025, 18, 448 27 of 29

21. Buffa, G.; Fratini, L.; Ingarao, G.; Di Lorenzo, R.; Arregi, B.; Luz Penalva, M. An optimization procedure for the friction stir welding
FEM model of corner fillet joints. In Proceedings of the 14th International Conference on Metal Forming, METAL FORMING
2012, Krakow, Poland, 16–19 September 2012; Volume 2012, pp. 567–570. Available online: https://www.researchgate.net/
publication/288560277_An_optimization_procedure_for_the_friction_stir_welding_FEM_model_of_corner_fillet_joints (accessed
on 19 December 2024).
22. Myśliwiec, P.; Kubit, A.; Szawara, P. Optimization of 2024-T3 Aluminum Alloy Friction Stir Welding Using Random Forest,
XGBoost, and MLP Machine Learning Techniques. Materials 2024, 17, 1452. [CrossRef] [PubMed]
23. Vidakis, N.; Petousis, M.; Mountakis, N.; Kechagias, J.D. Optimization of Friction Stir Welding Parameters in Hybrid Additive
Manufacturing: Weldability of 3D-Printed Poly(methyl methacrylate) Plates. J. Manuf. Mater. Process. 2022, 6, 77. [CrossRef]
24. Sambath, Y.; Natarajan, R.; Babu, P.K.; Raju, K.R.; Alahmadi, A.A.; Alwetaishi, M.; Khan, S.A. Comparative Analysis of Predictive
Modeling Techniques for Mechanical Properties in Dissimilar Friction Stir Welding of AA6061 and AZ31B. J. Mater. Eng. Perform.
2024, in press. [CrossRef]
25. Chadha, U.; Selvaraj, S.K.; Gunreddy, N.; Babu, S.; Mishra, S.; Padala, D.; Shashank, M.; Mathew, R.M.; Kishore, S.R.; Panigrahi,
S.; et al. A Survey of Machine Learning in Friction Stir Welding, including Unresolved Issues and Future Research Directions.
Mater. Des. Process. Commun. 2022, 2022, 1–28. [CrossRef]
26. Rana, A.; Hooda, N.; Kumar, R.; Deshwal, S.; Gahlot, P.; Phanden, R.K. Friction stir welding process parameters optimization and
weld structure analyses to study mechanical characteristics using hybrid artificial intelligent soft computing techniques. Int. J.
Interact. Des. Manuf. (IJIDeM) 2024, in press. [CrossRef]
27. Fuse, K.; Venkata, P.; Reddy, R.M.; Bandhu, D. Machine learning classification approach for predicting tensile strength in
aluminium alloy during friction stir welding. Int. J. Interact. Des. Manuf. (IJIDeM) 2024, in press. [CrossRef]
28. Elsheikh, A.H. Applications of machine learning in friction stir welding: Prediction of joint properties, real-time control and tool
failure diagnosis. Eng. Appl. Artif. Intell. 2023, 121, 105961. [CrossRef]
29. Prabhakar, D.A.P.; Korgal, A.; Shettigar, A.K.; Herbert, M.A.; Chandrashekharappa, M.P.G.; Pimenov, D.Y.; Giasin, K. A Review of
Optimization and Measurement Techniques of the Friction Stir Welding (FSW) Process. J. Manuf. Mater. Process. 2023, 7, 181.
[CrossRef]
30. Kubit, A.; Trzepieciński, T.; Kluz, R.; Ochałek, K.; Slota, J. Multi-Criteria Optimisation of Friction Stir Welding Parameters for EN
AW-2024-T3 Aluminium Alloy Joints. Materials 2022, 15, 5428. [CrossRef] [PubMed]
31. Yaknesh, S.; Rajamurugu, N.; Babu, P.K.; Subramaniyan, S.; Khan, S.A.; Saleel, C.A.; Soudagar, M.N.-E.-A. A technical perspective
on integrating artificial intelligence to solid-state welding. Int. J. Adv. Manuf. Technol. 2024, 132, 4223–4248. [CrossRef]
32. Rao, T.B.; Rao, C.S.P.; Baki, N. Multi-objective optimisation of friction stir welding parameters: Integration of FEM and NSGA-II.
Int. J. Manuf. Res. 2021, 16, 82–101. [CrossRef]
33. Babalola, S.A.; Dutta, S.; Murmu, N.C.; Akinnuli, B.O. In-situ process reliability monitoring strategy for friction stir welding
machine. Mater. Today Proc. 2022, 66, 3883–3889. [CrossRef]
34. El-Kassas, A.M.; Sabry, I. Using Multi Criteria Decision Making in Optimizing the Friction Stir Welding Process of Pipes: A Tool
Pin Diameter Perspective. Int. J. Appl. Eng. 2019, 14, 3668–3677.
35. Maheshwari, S.; Kar, A.; Alam, Z.; Kumar, L. Deep Neural Network-Based Approach for Modeling, Predicting, and Validating
Weld Quality and Mechanical Properties of Friction Stir Welded Dissimilar Materials. JOM 2023, 75, 4562–4578. [CrossRef]
36. Kahhal, P.; Ghasemi, M.; Kashfi, M.; Ghorbani-Menghari, H.; Kim, J.H. A multi-objective optimization using response surface
model coupled with particle swarm algorithm on FSW process parameters. Sci. Rep. 2022, 12, 2837. [CrossRef]
37. Zhou, Y.; Wang, J.; Wang, Z. Bearing Faulty Prediction Method Based on Federated Transfer Learning and Knowledge Distillation.
Machines 2022, 10, 376. [CrossRef]
38. Matitopanum, S.; Pitakaso, R.; Sethanan, K.; Srichok, T.; Chokanat, P. Prediction of the Ultimate Tensile Strength (UTS) of
Asymmetric Friction Stir Welding Using Ensemble Machine Learning Methods. Processes 2023, 11, 391. [CrossRef]
39. Sun, Y.; Gong, W.; Feng, J.; Lu, G.; Zhu, R.; Li, Y. A Review of the Friction Stir Welding of Dissimilar Materials between Aluminum
Alloys and Copper. Metals 2022, 12, 675. [CrossRef]
40. Di Bella, G.; Favaloro, F.; Borsellino, C. Effect of Process Parameters on Friction Stir Welded Joints between Dissimilar Aluminum
Alloys: A Review. Metals 2023, 13, 1176. [CrossRef]
41. Zhang, F.; Li, L.; Han, Z.; Wang, X. Determination method of binary fractions by the integrated spectrum. Mon. Not. R. Astron.
Soc. 2024, 531, 3468–3478. [CrossRef]
42. James, G.; Witten, D.; Hastie, T.; Tibshirani, R. An Introduction to Statistical Learning; Springer: New York, NY, USA, 2013.
[CrossRef]
43. Li, Y.; Li, M.; Zhang, L. Evolutionary polynomial regression improved by regularization methods. PLoS ONE 2023, 18, e0282029.
[CrossRef] [PubMed]
44. Marasco, S.; Marano, G.C.; Cimellaro, G.P. Evolutionary polynomial regression algorithm combined with robust bayesian
regression. Adv. Eng. Softw. 2022, 167, 103101. [CrossRef]
Materials 2025, 18, 448 28 of 29

45. Cheng, C.-L.; Tsai, J.-R.; Schneeweiss, H. Polynomial regression with heteroscedastic measurement errors in both axes: Estimation
and hypothesis testing. Stat. Methods Med. Res. 2019, 28, 2681–2696. [CrossRef] [PubMed]
46. Sulthan, A.; Jayakumar, G.S.D.S. Exact distribution of Cook’s distance and identification of influential observations. Hacet. J.
Math. Stat. 2014, 44, 165–178. [CrossRef]
47. Li, J.; Valliant, R. Linear Regression Influence Diagnostics for Unclustered Survey Data. J. Off. Stat. 2011, 27, 99–119.
48. Bahadir, B.; İnci, H.; Karadavut, U. Determination of Outlier in Live-Weight Performance Data of Japanese Quails (Coturnix
Coturnix Japonica) By Dfbeta and Dfbetas Techniques. Ital. J. Anim. Sci. 2014, 13, 3113. [CrossRef]
49. Kuznetsov, A.; Frontoni, E.; Romeo, L.; Poluyanenko, N.; Kandiy, S.; Kuznetsova, K.; Beňová, E. Optimizing Hill Climbing
Algorithm for S-Boxes Generation. Electronics 2023, 12, 2338. [CrossRef]
50. Peker, F.; Altun, M. A Fast Hill Climbing Algorithm for Defect and Variation Tolerant Logic Mapping of Nano-Crossbar Arrays.
IEEE Trans. Multi-Scale Comput. Syst. 2018, 4, 522–532. [CrossRef]
51. Qiao, J.; Shi, Q.; Chen, S.; Yang, C.; Han, Y.; Chen, G. Visualization of vertical transfer of material through high-velocity rotating
flow zone during friction stir welding. Mater. Lett. 2024, 367, 136599. [CrossRef]
52. Orłowska, M.; Pixner, F.; Hütter, A.; Kooijman, A.; Jasiński, C.; Gonzalez-Garcia, Y.; Enzinger, N.; Lewandowska, M. Local
changes in the microstructure, mechanical and electrochemical properties of friction stir welded joints from aluminium of varying
grain size. J. Mater. Res. Technol. 2021, 15, 5968–5987. [CrossRef]
53. Ji, S.; Li, Z. Reducing the Hook Defect of Friction Stir Lap Welded Ti-6Al-4V Alloy by Slightly Penetrating into the Lower Sheet. J.
Mater. Eng. Perform. 2017, 26, 921–930. [CrossRef]
54. Yang, C.; Chen, G.; Qiao, J.; Wu, C.; Zhou, M.; Zhang, G.; Shi, Q. Material flow during dissimilar friction stir welding of Al/Mg
alloys. Int. J. Mech. Sci. 2024, 272, 109173. [CrossRef]
55. Myśliwiec, P.; Śliwa, R.E.; Ostrowski, R. Friction Stir Welding of Ultrathin AA2024-T3 Aluminum Sheets Using Ceramic Tool.
Arch. Met. Mater. 2019, 64, 1385–1394. [CrossRef]
56. Mysliwiec, P.; Śliwa, R.E. Friction Stir Welding of Thin Sheets of the AA2024-T3 Alloy with a Ceramic Tool: RSM and ANOVA
Study. Key Eng. Mater. 2022, 926, 1756–1761. [CrossRef]
57. Salih, O.S.; Neate, N.; Ou, H.; Sun, W. Influence of process parameters on the microstructural evolution and mechanical
characterisations of friction stir welded Al-Mg-Si alloy. J. Mater. Process. Technol. 2020, 275, 116366. [CrossRef]
58. Laska, A.; Szkodo, M.; Cavaliere, P.; Moszczyńska, D.; Mizera, J. Analysis of Residual Stresses and Dislocation Density of AA6082
Butt Welds Produced by Friction Stir Welding. Metall. Mater. Trans. A 2023, 54, 211–225. [CrossRef]
59. Sato, Y.S.; Urata, M.; Kokawa, H.; Ikeda, K. Hall–Petch relationship in friction stir welds of equal channel angular-pressed
aluminium alloys. Mater. Sci. Eng. A 2003, 354, 298–305. [CrossRef]
60. Kosturek, R.; Śnieżek, L.; Torzewski, J.; Wachowski, M. Research on the Friction Stir Welding of Sc-Modified AA2519 Extrusion.
Metals 2019, 9, 1024. [CrossRef]
61. Staron, P.; Kocak, M.; Williams, S. Residual stresses in friction stir welded Al sheets. Appl. Phys. A 2002, 74 (Suppl. S1),
s1161–s1162. [CrossRef]
62. Sreenivasan, K.S.; Satish Kumar, S.; Katiravan, J. Genetic algorithm based optimization of friction welding process parameters on
AA7075-SiC composite. Eng. Sci. Technol. Int. J. 2019, 22, 1136–1148. [CrossRef]
63. Anandan, B.; Manikandan, M. Machine learning approach with various regression models for predicting the ultimate tensile
strength of the friction stir welded AA 2050-T8 joints by the K-Fold cross-validation method. Mater. Today Commun. 2023, 34,
105286. [CrossRef]
64. Malde, M. Thermomechanical Modeling and Optimization of Friction Stir Welding. Master’s Thesis, Louisiana State University,
Baton Rouge, LA, USA, 2009. [CrossRef]
65. Das, P.P.; Chakraborty, S. Optimization of friction stir welding processes using hybrid-taguchi methods: A comparative analysis.
Int. J. Interact. Des. Manuf. (IJIDeM) 2023, 17, 1021–1038. [CrossRef]
66. Verma, S.; Gupta, M.; Misra, J.P. Performance evaluation of friction stir welding using machine learning approaches. MethodsX
2018, 5, 1048–1058. [CrossRef] [PubMed]
67. Lipińska, M.; Grocholewski, A.; Zakrzyk, A.; Duda, P.; Golombek, M.; Rogala, T.; Jarz˛ebska, M.; Domagała-Dubiel, J.; Przybyłek, P.
The Influence of Microstructure Evolution on the Mechanical and Electrochemical Properties of Dissimilar Welds from Aluminum
Alloys Manufactured Via Friction Stir Welding. Metall. Mater. Trans. A 2024, 55, 4373–4390. [CrossRef]
68. Ghorbanzade, T.; Soltanipour, A.; Dehghani, K.; Chabok, A. Microstructural evolutions and mechanical properties of friction stir
welded AA2024. Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 2016, 230, 75–87. [CrossRef]
69. Naik, B.S.; Chen, D.L.; Cao, X.; Wanjara, P. Microstructure and Fatigue Properties of a Friction Stir Lap Welded Magnesium Alloy.
Metall. Mater. Trans. A 2013, 44, 3732–3746. [CrossRef]
70. Liu, H.; Hu, Y.; Peng, Y.; Dou, C.; Wang, Z. The effect of interface defect on mechanical properties and its formation mechanism in
friction stir lap welded joints of aluminum alloys. J. Mater. Process. Technol. 2016, 238, 244–254. [CrossRef]
Materials 2025, 18, 448 29 of 29

71. Aldanondo, E.; Vivas, J.; Álvarez, P.; Hurtado, I. Effect of Tool Geometry and Welding Parameters on Friction Stir Welded Lap
Joint Formation with AA2099-T83 and AA2060-T8E30 Aluminium Alloys. Metals 2020, 10, 872. [CrossRef]
72. Lacki, P.; Derlatka, A.; Wi˛eckowski, W.; Adamus, J. Development of FSW Process Parameters for Lap Joints Made of Thin 7075
Aluminum Alloy Sheets. Materials 2024, 17, 672. [CrossRef] [PubMed]
73. Soto-Diaz, R.; Sandoval-Amador, A.; Escorcia-Gutierrez, J.; Unfried-Silgado, J. Influence of Process Parameters on the Mechanical
Properties and Corrosion Resistance of Dissimilar Friction Stir Welded Joints of AA2024-O and AA6061-O Aluminum Alloys.
Metals 2024, 14, 664. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.