Materials Today Communications 33 (2022) 104437

Contents lists available at ScienceDirect

Materials Today Communications

journal homepage: www.elsevier.com/locate/mtcomm

Full length article

Applications of data-driven approaches in prediction of fatigue and fracture

Sara Nasiri a , Mohammad Reza Khosravani b ,∗
a
SmartyX GmbH, Martinshardt 19, 57074 Siegen, Germany
b
Chair of Product Development, University of Siegen, Paul-Bonatz-Str. 9-11, 57068 Siegen, Germany

ARTICLE INFO ABSTRACT

Keywords: Since various mechanical components could undergo cyclic loadings in their operations, fatigue evaluation of
Fatigue the parts is a crucial issue. In this context, different experimental practices have been developed to investigate
Fracture fatigue strength, crack growth rate, and fracture behavior of mechanical parts. Along with conventional fatigue
3D printing
test methods, data-driven approaches can be used to investigate fatigue life and fracture behavior of metallic
Data-driven approach
and non-metallic structural elements. A data-driven analysis is a technique which uses collected data to
answer the particular demand. In this contribution, we have reviewed data-driven approaches which have
been applied to predict fatigue life and fracture of components. To this aim, we discuss the documented
results in applied data-driven techniques in evaluating fatigue performance of metallic, composite, and 3D-
printed parts. The material, fabrication process, geometric parameters, and details of data-driven approaches
have been considered in review of the reported literature. Although improvements are recorded due to using
data-driven approaches in prediction of fatigue behavior of parts, there are challenges and limitations which
have been highlighted in the present study. Moreover, this review explains the key parameters which play
crucial roles in fatigue life of metallic and non-metallic components.

1. Introduction strength is significantly affected by the presence of inhomogeneities in

a component, attention is mostly focused on identification of defect size
Since repeated loading and unloading in materials leads to many and the maximum inclusion. Indeed, according to [15], fatigue cracks
damages, fatigue can be considered as one of the major reasons of generally originate from the largest defect within the most stressed vol-
engineering failure. Therefore, numerous investigations have been per- ume. Based on the applied stress level, the fatigue life of the part can be
formed to determine fatigue behavior of materials over the years [1–5]. classified into low-cycle fatigue and high cycle fatigue. In experimental
Fatigue behavior of mechanical components and structural elements
practices, defect size and the specimen displacement can be acquired
generally depend on the intrinsic properties of the materials’ resis-
by appropriate equipment such as optical extensometers [16], digital
tance which should be considered in the design of the parts. Among
image correlation technique [17], and linear variable displacement
the main parameters which have effects on the fatigue behavior of
materials, there are three important factors: (i) processing which in- transducer [18]. Recently, in [19] the microstructural and geometrical
volves the residual stresses, (ii) geometric shape which controls the size effects on the fatigue behavior of metallic parts under different
average nominal level of the stress, and (iii) surface roughness with fatigue loading regimes have been reviewed.
has effects on the probability of the crack development [6]. Considering Due to the advances in fabrication of composites, different compos-
above-mentioned factors, fatigue problem can be solved by controlling ites have been produced with unique stiffness to weight ratio which
process, shapes, and surface finishing instead of changing the material. make them a favorable material for different industrial applications.
A review of literature confirmed that fatigue tests have been con- For instance, composites have been used as ship hulls, rotor blades in
ducted on different parts such as metallic components, composites, wind turbines, aircraft structures and so on. Such structural elements
and additively manufactured parts [7–11]. Parallel to the experimental experience different types of constant and variable amplitude fatigue
practices, numerical simulations are performed to estimate fatigue loads in their service time. Therefore, the study of fatigue and fracture
behavior of structural elements [12–14]. If the applied load is higher
behavior of composite has become of significant importance. However,
than a certain threshold, cracks are propagated due to the accumu-
composites do not yet enjoy the kind of databases which are avail-
lated microscopic damage. In engineering metals, there is a trend of
able for conventional materials. In this respect, different experimental
evolution of localized damage in the vicinity of the defects. As fatigue

∗ Corresponding author.
E-mail address: mohammadreza.khosravani@uni-siegen.de (M.R. Khosravani).

https://doi.org/10.1016/j.mtcomm.2022.104437
Received 25 July 2022; Received in revised form 1 September 2022; Accepted 10 September 2022
Available online 16 September 2022
2352-4928/© 2022 Elsevier Ltd. All rights reserved.
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

tests have been performed to predict and determine fatigue life of

composites [20–22]. For instance, fatigue behavior of ceramic matrix
composite is studied in [23]. In this context, fatigue tests via constant
and variable amplitudes are performed and based on the results the
effects of needled fiber bundle on the fatigue life dispersion is analyzed.
Recently, in [24] effects of partial bonds on the fatigue crack growth of
a composite is investigated. The experiments indicate that the particles
can exert the blocking effect to fatigue crack growth.
Additive manufacturing (AM) as an advanced technology based on
adding material in layer form is widely used in many industries [25–
27]. Considering applications of 3D-printed parts, the study of fatigue
and fracture behavior of these parts is a necessity. In [28], defects
in 3D-printed metal parts and their effects on the fatigue behavior of
these parts are reviewed. To this aim, AM defect characterization and
Fig. 1. Factors with effects on the fatigue life of metals.
statistical analysis methods are presented and influence of processing
and post-processing parameters on the defect and fatigue performance
of 3D-printed parts are explained. At the same time, fracture and fatigue
Not only factors presented in Fig. 1, but also other parameters
in additively manufactured metal parts are reviewed in [29]. In detail,
such as chemical composition and heat treatment have effects on the
researchers summarized effects of post-processing such as hot-isostatic
fatigue life of metals. Moreover, some parameters such as yield stress
pressing, annealing, and shot peening on the fatigue behavior of the
concentrations and residual stress could be considered as sub-domains
components.
of shape parameter.
Although fatigue phenomenon can be considered as one of the
Different composites have been utilized in a wide range of applica-
primary causes of component failures, may not be always feasible to
tions based on their favorable strength and unique stiffness to weight
perform fatigue test with well-defined geometry and quantity of test
ratio [40–42]. Most of the applications utilizing hybrid composites are
coupons. Particularly, brittle materials [30], small volumes like solder
operating under fatigue loading conditions where components experi-
joints [31], high entropy alloys [32], and thin films [33] are examples
ence a high number of fatigue cycles in their lifetime. Although fatigue
of cases where traditional fatigue tests may not be an appropriate
of composites has been investigated over the years, prediction of fatigue
choice. In this context, different approaches are used to estimate fatigue
life of engineering materials and composites is difficult. In Fig. 2, main
life of the components [34–36]. However, each approach has its own
parameters which have effects on the fatigue behavior of polymers and
advantages and disadvantages.
composites are summarized. In composites, combination of fibers with
In the present study, applications of data-driven approaches in
different failure strains prevents rapid crack extensions which leads to
prediction of fatigue and fracture behavior of parts have been reviewed
a slower fatigue damage growth and enhancing the fatigue life.
based on an extensive literature review of articles indexed in Scopus
Considering advantages of 3D printing, it has been used in different
and Web of Science database. To the authors’ knowledge, no review
fields in both academic and industrial research projects. The reliability
article exists on this topic, and the motivation of this paper is to
and structural integrity concerns of 3D-printed parts have attracted
address applications, reasons, and challenges in employing data-driven
attention of research into the fatigue behavior of AM parts [43–46]. As
techniques in evaluating fatigue properties of metallic, non-metallic,
3D printing is based on adding layer-upon-layer of material, additively
composite, and 3D-printed parts. To this end, the remainder of this
manufactured parts have large surface roughness and many sub-surface
paper is organized as follows: in the following section an overview of
defects. Surface roughness increases stress and it has a negative effect
influencing parameters on evaluation of fatigue life has been presented.
on the fatigue properties, especially in high cycle fatigue [47]. In
Section 3 reports efforts in evaluation of fatigue life via data-driven
a previous study [48], post-processing (e.g., mechanical milling and
approaches. In Section 4 advantages, challenges, and perspectives have vibratory grinding) is used to remove the rough surface in order to
been outlined. Finally, a conclusion has been furnished in Section 5. improve the durability of 3D-printed components. In [49], it has been
reported that fatigue performance of stainless steel 316 L fabricated by
2. Influencing parameters on fatigue life of a component the laser-based powder bed fusion process increased by 50% compared
to the behavior of the as-fabricated part.
Fatigue is a size-dependent phenomenon and fatigue strength is Internal defects is another important parameter on the fatigue be-
enhanced with decreasing the length scale, because of less probability havior of 3D-printed parts. Different reasons such as surface tension,
of defects [37]. The defects includes surface and interior defects which gaps, inappropriate processing parameters, and vapor recoil can lead
can occur during manufacturing process. In addition, fatigue crack to internal defects and decrease fatigue life of additively manufactured
growth rate is a function of geometric dimension and a long crack in parts [50–53]. Although size, shape, and location of internal defect
a small-scale specimen can be considered physically short in regular have influence on shortening the fatigue life of 3D-printed parts, the
(bulk-scale) fatigue samples [38]. As the current study deals with ap- location has a more adverse influence contrary to the size and shape. It
plications of data-driven approaches for fatigue and fracture in metallic, is noteworthy to mention that build thickness of 3D-printed parts must
composites, and 3D-printed parts, in this section we briefly summa- be considered as a parameter which change fatigue performance of
rize influencing parameters in the fatigue behavior of aforementioned the part, but it may differ from the rule obtained from conventionally-
materials. fabricated metallic materials. Indeed, the review of the published works
In metallic materials, the size effect is mainly related to specimen indicates that further research is needed to clarify the effects of build
thickness/diameter and the microstructural scale, such as grain or thickness on the fatigue behavior of 3D-printed parts.
phase size. With combining effects of geometrical and the microstruc- Numerous mechanical parts are experiencing high loading cycles in
tural scales, an aspect ratio of the specimen thickness or diameter (𝑡) their operations (e.g., compressor blades, aircraft engines, and turbine
to the grain size (𝑑) must be considered which has a crucial role in the rotor blades). Several testing equipment have been utilized to carry
mechanical properties when 𝑡 and 𝑑 values are in the same order [39]. out fatigue test, such as rotating–bending fatigue tester and servo-
Fig. 1 illustrates some factors which have influence on the fatigue life hydraulic testing machine. These types of the testing approaches are
of metals. usually regarded as conventional fatigue testers. Conventional test

2
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 2. Different factors influencing fatigue resistance in composites and polymers.

methods require huge load capacities, but the resonance method is a In the hybrid prognosis model, physics-based modeling was integrated
fast and energy-efficient approach. Resonance fatigue testing involves with a data-driven approach and different types of loading conditions
exciting a specimen close to its first mode of vibration by applying a were investigated. The hybrid model provides an appropriate fracture
rotating radial force to one end. The resonance test method applies a mechanisms with a correction for material variations and uncertainty
fully alternating stress cycle and bending moment is generated in the in the parameters via updating in data-driven model. Therefore, the
specimen. However, there is no standard that defines all details of the accuracy of this hybrid prognosis framework is greater than those of
resonance fatigue test. In comparison, utilizing the conventional fatigue physics-based and data-driven models alone. The exact relationship
test setup at a frequency of 5 Hz, would take several years. Therefore, between stress intensity factor (SIF) and crack growth rate for a certain
using data-driven approaches is expected to be beneficial in this field. cycle was inferred from the available data. The crack length at a given
In the following section, a systematic literature review is presented cycle (𝑁) can be written as:
with a focus on applications, suitability, and advantages of data-driven 𝑁

approaches in prediction of fracture and fatigue of components. 𝑎𝑁 = 𝑒𝐶1 +𝐶2 𝑙𝑜𝑔(△𝑘) 𝑑𝑁 (1)
∫0
where 𝐶1 and 𝐶2 are coefficient functions and △𝐾 denotes the SIF.
3. Applications of data-driven approaches in fatigue and fracture
Considering the uncertainty, the distribution on crack length as a
function of cycle can be written as:
3.1. Data-driven approaches in fatigue of metallic parts
∑
𝑁+△𝑁
𝑝(𝑎𝑁+△𝑁 ) = 𝑒𝐶1 +𝐶2 𝑙𝑜𝑔(△𝑘) △ 𝑁 + 𝑝(𝑎𝑁 ) (2)
Using the data-driven models is beneficial, when there is a high
𝑁
complexity in material behavior or the fracture characterization is not
where 𝑝(.) is the probability distribution. For a given specimen ge-
well understood. This technique fits the experimental results utilizing
ometry under the fatigue loading, the developed hybrid approach
a mathematical model such as artificial neural networks (ANN) [54],
comprised a physics-based formulation to calculate the SIF and data-
Gaussian process regression [55], and deep learning regression [56].
driven component to determine the nonconstant coefficient 𝐶1 and 𝐶2 .
By combining experimental and numerical simulations, a different
The prediction of the crack length was performed in a probabilistic
class of data-driven damage models has been proposed. For instance,
framework. Authors evaluated efficiency of the proposed model using a
a study [57] recommended a second-order damage tensor which was
compact tension test based on ASTM E647 [61]. The crack length is pre-
evolved with model parameters defined based on the experimental dicted by hybrid model with precision of 95% which is a good possible
data. Kompalka et al. [58] utilized successive model to refine an initial scenario for a practical application. The residual useful life prediction
polynomial approximation of a damage parameter. Later, data analytics is within ±5% of the actual residual useful life for constant-amplitude
tools were used to predict the fatigue strength of steels in [59]. To this loading. It was reduced to ±2% as the training data increased. Even
aim, different physics-based and data-driven approaches were utilized if the overload points were unknown, the algorithm was still able to
to find correlations between different properties of alloys and their predict within 5% error given training data for overload behavior. The
compositions and manufacturing process parameters. In this context, results indicated that the algorithm was able to predict the crack length
researchers used twelve predictive modeling techniques, such as linear with an error of less than 7% for most of the crack growth regime. The
regression, support vector machine (SVM), decision table, and ANN. proposed methodology was compared with Paris’ law, and this method
Moreover, they used different evaluation metrics (e.g., mean absolute provide more accurate results with high confidence. It should be noted
error, explained variance, and root mean squared error) to determine that the accuracy of the hybrid framework is greater than those of
differences between predicted and actual fatigue strength values. The data-driven and physics-based models alone.
obtained results show a high predictive accuracy. At the same time, Subsequently, a date-driven damage model was formulated based in
in [60] a hybrid prognosis model was developed to predict crack nondestructive evaluation [63]. The material was examined via scan-
growth regime and fatigue life time of metallic aircraft components. ning electron microscopy (SEM) and the deformation was monitored

3
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 3. A schematic of the adopted procedure for identification of a data-driven deterministic driving force for small crack propagation [62].

based on acoustic emission and digital image correlation (DIC) tech- micromechanical variables. The relevant variables (e.g., crack propaga-
nique. The proposed model was tested on dogbone specimens subjected tion) were identified by quantifying their correlation with experimental
to monotonic loading and on a prismatic beam under a set of cyclic results, and simulations and experimental findings were consolidated
loading. The obtained results confirmed the role of the data-driven into a multimodal dataset. The researchers compared the analytical
damage model on prediction of a realistic response. model for the crack driving force to common models available in the
In the subsequent year, data science approaches were used to de- literature and data stratification was adopted. The obtained results
termine crack formation behavior in [64]. More in deep, random forest proved the effectiveness of the recommended method in prediction
and logistic regression were used to predict fatigue crack initiation in of small crack propagation. In [62], the same researchers used their
an aluminum alloy. In addition, dogbone specimens were subjected method to predict 3D fatigue crack growth rate which has a crucial
to different electrochemical exposures, and later used as test coupons role in improving fatigue life prediction.
In a study [65], a data-driven modeling approach was employed
in fatigue experiments. The results show that data science analyses
to investigate high-cycle fatigue life of nickel titanium consisting of a
cannot provide location of fatigue crack initiation accurately. Indeed,
pair of hemispherical inclusions connected by a cylindrical void. The
the lack of consistency from one replicate to another indicates that
proposed approach uses decomposition of the problem into a training
these parameters are not enough to accurately predict the location of
stage and a prediction stage. The findings indicated that the trend is a
the fatigue crack initiation.
nonlinear dependence of fatigue incubation life on inclusion diameter.
Rovinelli et al. [62], used machine learning (ML) and a data-driven It means larger inclusions lead to a lower estimated fatigue lifetime.
approach to identify the fatigue crack driving force in a titanium alloy. There is a converse effect in voids. In fact, a wider void tends to increase
In detail, Bayesian network (BN) and ML were employed for identifica- the fatigue life, which implies that larger voids may not necessarily be
tion of relevant micromechanical and microstructural variables which more dangerous. At the same time, data-driven correlation analysis was
have effects on the direction and rate of the fatigue crack propaga- performed to predict fatigue crack path of an Al–Mg–Si alloy in [66].
tion. Particularly, the relevant variables, as identified from correlations To this aim, an algorithm was developed to systematically correlate
produced from the BN, were selected, and a functional form of the the 3D micromechanical fields from simulation with the 3D fatigue
deterministic driving force metric was determined by a ML approach. failure surface from test. Numerous data points were considered and
The data were collected from the crack front by utilizing a non-local 2D the parameters include field variables and slip-based damage metrics
data mining procedure, and experimental findings were compared with computed at peak load, as well as their corresponding cyclic values.
the predictions. Fig. 3 shows a schematic of the adopted procedure, The results of the algorithm consists of data frames, each row is an
where 𝑋1 , 𝑋2 , ..., 𝑋𝑁 represent the considered microstructural and observed point from the microstructure and each column is the distance

4
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

to the crack surface. It is concluded that the variables, treated indepen- By using the model in an embedded system, crack lengths can be
dently, are not enough to completely describe evolution of fatigue crack estimated in real-time, as long as data acquisition is possible in real-
surfaces and it remains to be tested. time. However, further studies are required to overcome the limitations
In [67], physics-based modeling was combined with ML techniques of commercialization. Following this route, a data-driven framework
to predict crack growth in aluminum alloys subjected to biaxial in- has been developed to detect early-stage fatigue damage in aluminum
phase and out-of-phase fatigue loading conditions. In this context, alloys [72]. Indeed, U and V notched Al7075-T6 specimens were inves-
energy release rate was utilized as the physics-based parameter and tigated by confocal microscope (focus on the notch root) and a digital
Gaussian process was employed to model the complex nonlinear re- microscope (focus on the side of the notch). The authors extracted two
lationships. The ML model was developed to accurately model the features during the test: (a) crack opening displacement, and (b) the
nonlinear relationship between driving force and the crack growth rate. crack length. Moreover, ML was used to analyze the signal data and
The predictions were made by projecting the input space to the output depend on the magnitude, the new signal was labeled as healthy or
space by inferring their underlying nonlinear relationship. Since the cracked. The obtained results show that the performance of proposed
accuracy of the prediction in prognosis model depends on the choice of algorithm for crack coalescence is lower than crack propagation regime.
the kernel function, choosing an appropriate kernel function is a crucial Another application of data-driven approach for fatigue and fracture
issue. The researchers used squared exponential kernel function which of a metallic part was presented in [73]. Specifically, a data-driven
is suitable for modeling of nonlinear smooth surfaces. The direction approach and reexamining the physical background of fracture tough-
of the crack propagation was assumed to be the same as the crack ness (𝐾𝐽 𝑐 ) temperature dependence were considered for providing a
direction, and once the crack growth rate was evaluated, both one- 𝐾𝐽 𝑐 master curve which is applicable to all ferritic steels. In addition,
step ahead and multi-step ahead predictions of the crack length were Bayesian generalized linear regression was adopted and Weibull distri-
made. Results confirmed a high accuracy of the proposed technique bution was assumed. The applied data-driven approach is categorized
in prediction of single and multiple cracks under out-of-phase loading as data regression to predict 𝐾𝐽 𝑐 from the explanatory variables related
conditions. In the subsequent year, a data-driven approach was used to the tensile properties and specimen size. It is expected to inte-
to construct a quantitative relationship between fatigue cracks and grate this data-driven approach with an elastic–plastic FEM to predict
acoustic nonlinearity in A7075 aluminum specimens [68]. To this end, fracture of complicated structures with cracks.
dynamic acousto-elastic testing was carried out and microstructural In the most recent study, a probabilistic data-driven method was
features were extracted from SEM images of fatigue cracks. In detail, introduced for real-time prediction of fatigue failure [74]. In detail,
different functions in MATLAB image processing toolbox were used a hybrid architecture made by fully connected ANN and Gaussian
and fifteen crack features (e.g., open crack length, distance to crack regression was proposed to enhance predictive abilities of uncertainty
tip, and average aperture to open crack) were extracted from the qualification which is an important part of a real-time decision-making
analysis of SEM images. The researchers used three regression methods process. After training process, the performance of the developed hy-
to relate the extracted crack microstructural features and the measured brid system was evaluated based on the signal taken from the test set.
nonlinearity parameters. All three models showed good predictability The results show that the suggested method is suitable for different
performance. Although the results are in an agreement with physical physical properties such as stress and strain in mechanical components.
interpretation, it is expected to obtain a better predictability with At the same time, in [75] data-driven approach was developed to
a larger dataset and more variability in crack aperture sizes. Later, increase fracture resistance. To this end, topology optimization was
based on the extended finite element method (XFEM), Feng et al. [69] used to control the stress. Then, data-driven techniques were employed
formulated a data-driven algorithm for real-time fatigue life prediction. to find the target unit cells and stratifying the optimal distribution of
In this respect, the XFEM was used to generate large amount of datasets the stiffness tensor. Finally, phase field fracture simulation was carried
which pair the structural responses with remaining fatigue life. Then, out to illustrate the data-driven design enhanced fracture resistance.
the back propagation neural network (BPNN) was used to prepare a Although there are not too many reported data-driven methods
fatigue life prediction model based on the XFEM datasets. Fig. 4 shows for fatigue prediction of metallic parts, due to numerous applica-
an overview of the proposed method. BPNN is a type of multilayer tions of metallic components and abilities and advantages of data-
forward neural network based on error back propagation technique driven approaches, a global growth is expected in utilizing data-driven
which usually consists of an input layer, one or more hidden layers techniques for estimation of fatigue life time in metal parts.
and one output layer. Here, after training of BPNN via XFEM datasets,
the trained BPNN model was used to obtain real-time prediction of 3.2. Data-driven methods in fatigue of polymer and composite parts
structural fatigue life. The researchers evaluated the algorithm by two
case studies: (a) a square plate with center crack, and (b) three-point Carbon fiber reinforced polymer composites often fail without warn-
bending beam containing edge crack. The results showed performance ing, therefore, there is no time for maintenance and repair. Con-
stability of the algorithm for different test datasets. sequently, prediction of their fatigue life has become of significant
An extant study [70], used a data-driven machine learning method- importance. Currently, fatigue life prediction of composite under vari-
ology to predict fatigue strength from metallographic data. To this aim, able amplitude loading is mainly focused on the polymer composites.
a simplified residual neural network was used to a micromechanical Multiaxial in-plane and out-of-plane fatigue in cross-ply, unidirectional,
model for the simulation of fatigue strength of nodular cast iron. Test and other layup configuration (e.g., woven and hybrid composites)
coupons of fatigue test in a previous study were used as metallographic have been investigated over the years [76–80]. The life prediction mod-
specimens and investigated by a microscope. Moreover, shakedown the- els can be divided into three categories: (a) macro property degradation
orem was applied to micromechanical model built from micrographs of models, (b) damage models, and (c) fatigue models. The first cate-
cast iron. The reported results indicate that the simplified residual neu- gory mainly includes strength and stiffness degradation models [81].
ral network is able to predict fatigue relationships of microstructure. The progressive damage models can be employed for fatigue analysis
At the same time, a hybrid approach of data-driven and physic-based based on the correlation of stiffness or strength with damage evolu-
methods has been introduced to estimate fatigue crack growth in an tion [82]. The fatigue life models are accomplished by combination
aluminum lap joint specimen [71]. To this end, data-driven method of constant life fatigue diagram, constant amplitude S-N curve, and
based on ML and signal processing was utilized to estimate the crack damage accumulative rule [83].
lengths for a few cycles. Then, a random forest model was employed to Acoustic emission (AE), as a non-destructive testing (NDT) has been
estimate crack lengths through optimal feature selection. In addition, proposed as a potential solution to detect the damage. AEs are the
the remaining crack lengths were predicted by a physics-based method. stress waves produced by the sudden internal stress redistribution of

5
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 4. A brief description of the proposed algorithm [69].

the materials caused by the changes in the internal structure. High- beam available in literature [88]. The evaluation was conducted on
sensitivity AE sensors can pick up signals at an early stage of fatigue different sizes of experimental data and performance of the approach
crack initiation. The advantage of AE method is the ability of testing was confirmed.
of the entire structure, localization of potential defects, and assessment Considering importance of early failure detection, fatigue prediction
of their risk for the structural integrity of the component. The limited of brittle materials based on data-driven approach has been developed
ability of precise identification of the defect type can be considered as in [89]. To this aim, virtual sensing nodes have been used to generate
a partial disadvantage. Therefore, it might be needed to use other NDT time-series data of the phase-field model. In addition, for damage
method. Data-driven approaches bring some advantages, for example classification, a pattern recognition scheme represents time-series data
the ability to do the test several times on the same specimen, something responses with a corresponding label, integrated with ML algorithms.
that is not possible with AE. In AE the test can only be done once and After identification of the patterns which represent different states of
only at the exact time when the damage happens. the materials, the developed ML framework used k-nearest neighbor
Algorithms which are used in ML are suitable to discover unique mi- and ANN algorithms to detect the presence and location of failure. The
crostructural configurations that accumulate fatigue deformation [84]. researchers evaluated the proposed framework by different types of
An extant study [85], a hybrid data-driven framework has been devel- failure and introducing noise to the time-series data. The reported re-
oped to enhance fracture paths in composites. The approach is based sults indicate the satisfactory performance of the developed algorithmic
on the combination of manifold learning, that is introduced to reveal framework in prediction of failure with damage phase-field time-series
the geometrical structure of crack patterns. The researchers presented data. Although the presented results are based on virtual sensing nodes,
phase-field simulation which was used for the initial dataset. Although a good performance of the developed framework with real sensor data
the proposed approach was used for synthetic data, it can be employed is expected.
for experimentally characterized cracks. In [86] combination of tensor In [90], a probabilistic fatigue life prediction framework has been
train and Gaussian process regression has been proposed for prediction developed which used a set of experimental data. In detail, physics-
of fracture in composite materials. The recommended approach is a based fatigue damage evolution relation and a data-driven ANN model
fully data-driven and non-intrusive approach which was used for a brit- have been employed to predict fatigue life in adhesively bonded joints.
tle fracture. The documented results show that the proposed technique The fatigue damage evolution was used to evaluate degradation of
can predict the evolution of crack growth with an accuracy of up to the adhesive, then it was mapped to the strain redistribution along
95%. At the same time, crack propagation has been predicted by data- the bondline by ANN model. The fatigue damage growth simulator
driven approach in [87]. Particularly, a long short-term memory has was used for fatigue simulation for a given geometric configuration.
been used to train the data and output has been improved by adding The quantified uncertainties from the proposed framework correspond
predicted different changes which are computed by hidden Markov to the intrinsic statistical material properties and can be utilized for
model. Compared to feedforward neural networks, long short-term probabilistic fatigue life prediction in any joint configuration with the
memory has both forward and backward loops connecting nodes of same adhesive material. Authors used experimental fatigue life data,
network and it is suitable for solving problems of classification, storage for various loading spectra, reported in the literature. In Fig. 5, the
and forecast time series information. The researchers evaluated their developed framework for prediction of fatigue life in adhesively bonded
proposed method based on numerical model of single-edge notched joints is illustrated. The outer loop (red line arrows) was utilized to

6
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 5. An integrated framework for fatigue life prediction in adhesively bonded joints using the experimental data [90].

quantify uncertainty in the model parameters, and the inner loop (deep developed model by different numerical examples. For instance, they
blue line arrows) simulated the fatigue damage progression, until the used an infinite square plate with initial central crack, and three-point
joint failure. bending beam with an initial edge crack to test validity of the proposed
In [91], the researchers extended the model-free data-driven ap- algorithm. The researchers used the sequential quadratic programming
proach which they previously presented in [92] to the case of rate- method and genetic algorithm to calculate the weight coefficient vector
dependent and fatigue fracture propagation. In [92], the analysis was for optimization analysis. The obtained results proved performance of
limited to quasi-static rate-independent conditions, but in [91] phe- the proposed approach which perform much better than any individual
nomenological approach of Larsen et al. [93] and Negri [94] is adopted learning algorithms.
and a rate-dependent energy dissipation is assumed to account for the
effects of non-negligible crack tip velocities. The proposed approach 3.3. Data-driven techniques in fatigue of 3D-printed parts
was examined on double-cantilever beam specimens with different
geometries showing an excellent agreement with findings obtained Although 3D printing was introduced for fabrication of prototypes,
from solving the related fracture mechanics problems. Since there are it has been currently used for fabrication of end used products. Con-
different crack branching which cannot be solved within a macroscopic sequently, the fatigue life time of 3D-printed parts has become of sig-
description of the phenomenon, the proposed data-driven formulation nificant importance. Since 3D-printed components are produced layer-
is beneficial in this field. wise, there are anisotropic properties throughout the parts. Owing to
In a recent study [95], a data-driven approach was used to predict the production process, 3D-printed parts present different mechanical
the remaining fatigue life. To this end, a large amount of datasets properties and microstructures compared to components fabricated by
was produced by XFEM, and then the remaining fatigue life prediction other processes [96,97].
model developed via ensemble learning algorithm. In this respect, Traditional fatigue test may not be a logical choice in structural
multiple ML algorithms were used to learn useful degradation pat- components with variation in local properties like additively manu-
terns of the structures from the XFEM datasets. Authors evaluated the factured parts. Therefore, in view of these challenges, using different

7
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 6. A fishbone diagram to illustrate effects of the FDM parameters on the fatigue life of 3D-printed polymers.

approaches would be beneficial. In this context, developing data-driven approach which used clustering and subsequent analysis of deforma-
models for prediction of fatigue life in 3D-printed parts can shorten the tion. The microscopic approaches collected data and two methods were
time for experimental practices and numerical simulations. used to make the predictions. In detail, non-uniform transformation
Fused deposition modeling (FDM) is a robust and cost-effective 3D field analysis, and variants of the proper orthogonal decomposition
printing technique for production of geometrically complex part. Litera- were utilized and predictions were achieved by linear combinations of
ture investigation shows that different thermoplastic polymer filaments a finite number of representative volume element modes from previ-
have been used in fabrication of 3D-printed components [98–100]. Due ously completed simulations. The presented model can capture local
to the thermal characteristics, thermoplastic-based filaments are widely microstructures and allows a higher details or a larger region of interest
used in the FDM process. Although, there are several advantages in the which is desirable for prediction of fatigue initiation within the parts.
FDM process, mechanical strength of FDM 3D-printed prats depends on In the subsequent year, Yan et al. developed data-driven models to
different factors. Fig. 6 shows a fishbone diagram of the FDM process derive process–structure–property relationships for 3D printing [109].
parameters which have effects on the mechanical properties of 3D- Particularly, the researchers developed a reduced order modeling tech-
printed parts. Since polylactic acid (PLA) is one of the most popular nique (called self-consistent clustering analysis) as an efficient solution
materials utilized in the FDM process, several research works have for analysis of plasticity and damage in additively manufactured parts.
been performed to determine fatigue and fracture of 3D-printed PLA A collection of voids were captured by 3D X-ray imaging and imported
specimens using FDM technique [101–106]. Fatigue properties of FDM to a voxel mesh. Then self-consistent clustering analysis was utilized
3D-printed components have been comprehensively reviewed in [107]. with a crystal plasticity constitutive law. The results of the study
In this context, the influences of printing parameters and printing
show capability of the developed model in improving the cycle of
material characteristics have been discussed in detail. The researchers
design–predict–optimize for 3D-printed parts.
concluded that the available data are not enough to define fatigue
An extant study [110], presents applications of data-driven ap-
mechanism of FDM 3D-printed polymers.
proaches for evaluation of fatigue in 3D-printed parts using minia-
Thermosets have increasing importance in recent decades due to
ture specimens. In this context, state-of-the-art research progress in
several advantages (e.g., environmental stability, chemical resistance,
evaluating the fatigue behavior of 3D-printed components has been
and mechanical strength). Thermosets also have good resistance to fa-
introduced. Moreover, a roadmap of data-driven evaluation has been
tigue, which is also attractive in aerospace applications. Thermosetting
established based on experimental data on a large number of miniature
polymers are the build materials for thermoset 3D printer, most of them
are a viscous liquid. 3D printing techniques which deal with a liquid specimens. ML plays a crucial role in large-scale data analysis which
form of thermosetting polymers are mainly based on VAT photopoly- includes: (a) data preprocessing, (b) data representation, (c) model
merization and material extrusion. However, 3D printing of thermosets selection, (d) model establishment, and (e) optimization by training
faces a set of challenges for the massive manufacturing. For example, and testing data. The concept of data-driven evaluation workflow for
the availability of high-performance 3D printing ink materials is very prediction of fatigue in 3D-printed parts is depicted in Fig. 7. As illus-
limited. Moreover, most of the photocurable thermosets are suffering trated, a series of individual miniature test coupons should be extracted
from brittleness and poor impact resistance due to inhomogeneous from different locations of 3D-printed parts, in order to determine
polymer architecture. Due to the above-mentioned reasons, fatigue test the location-specific and orientation-dependent fatigue properties in
on 3D-printed thermoset is limited and there is no published research geometrically complex 3D-printed components. Metal 3D printing is
works on application of data-driven approaches for prediction of fatigue becoming mature in industrial application, and as discussed in [111],
life time in 3D-printed composites with thermoset matrix. the need for qualification and certification of metal 3D-printed parts
Although traditional fatigue testing has been used for evaluation of have become of significant importance.
3D-printed components, data-driven frameworks have been developed In [112], an adaptive neuro-fuzzy-based ML method has been used
to predict fatigue life time of different additively manufactured parts. to predict high cycle fatigue of laser additive manufactured stainless
In [108], data-driven approach was used to reduce computational de- steel. A dataset consisting of fatigue life data for specimens fabricated
grees of freedom required for estimating macroscopic behaviors of 3D- under different process parameters was constructed to simulate a com-
printed metal parts. The proposed technique was based on a two-stage plex nonlinear input–output environment. Although there are various

8
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 7. Roadmap of data-driven approach for evaluation of fatigue in 3D-printed parts based on the experiments on miniature specimens [110].

fatigue and fracture properties, the proposed model indicated a good and intra-layer bonding. Consequently, in [114] data-driven approach
prediction compared to the test data. as a non-destructive technique has been used to investigate surface
Porosity and bonding between adjacent filaments (within and be- roughness of Ti–6Al–4V specimens produced by a laser beam powder
tween layers) are important parameters which have effects on the me- fusion system. Since as-built surface roughness can accelerate surface
chanical strength of extrusion-based 3D-printed parts. In this respect, crack initiation and lead to early fatigue failure of metallic 3D-printed
a hybrid physic-based and data-driven approach has been proposed parts, influence of surface roughness on the mechanical strength and
to characterize filament bonding and porosity variation in extrusion- fracture behavior of these parts must be characterized before using
based 3D printing [113]. To this end, the geometry of interest is in fatigue-critical applications. Experimental investigations show that
decomposed into anisotropic voxels, and an analytical heat transfer as-built and half-polished specimens have similar fatigue lives among
modeling approach is used to estimate the temperature history at all cycle regimes. The proposed fatigue model was used to predict
the center of each voxel. The, ML-based model is made to deter- fatigue performance and comparison of the results with experimental
mine the deformation of micro geometries and their influences on data confirmed the validity of the model. However, it is recommended
the bonding degree. Later, the calculated inter-layer and intralayer to use surface analysis for longed profile lines.
bonding values, and void boundaries are updated based on the law of As we discussed in [115], different manufacturing defects can be
conservation of mass and obtained filament deformation values. The occurred during fabrication 3D-printed components. Defects in 3D-
proposed model shows high accuracy for characterizing both inter-layer printed parts significantly reduce structural integrity of the parts, thus,

9
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

Fig. 8. ML process flowchart for prediction of fatigue life time [119].

study their influences on the fatigue performance of the parts is a research works, the accuracy of the predicted fatigue curve is restricted
necessity. In this regard, Bao et al. [116] recommended a ML fatigue by the quality of the original dataset. Most recently, efficiency of
life prediction approach to explore effects of defects on the fatigue ML for evaluation of fatigue in a 3D-printed AlSi10Mg specimen has
life of metallic 3D-printed parts. To this aim, researchers combined been investigated in [120]. To this aim, authors developed a deep
synchrotron X-ray tomography and high cycle fatigue postmortem ex- learning based approach to predict fatigue behavior of the specimens
amination to obtain the geometric features of the critical defects, which which have been investigated experimentally in their previous re-
were trained via SVM. In fact, three defect features (location, size, and search. The obtained results confirmed that deep learning is an effective
morphology) were tentatively trained into the SVM model. To acceler- tool for prediction of fatigue behavior and determining the effect of
ate the optimization process and fitting the model parameters, the grid post-processing on 3D-printed parts.
search approach was used. In addition, a series of high cycle fatigue Reviewed research works in applications of data-driven approaches
tests was carried out to obtain the critical defect-determined fatigue in prediction of fracture and fatigue in 3D-printed parts, confirmed that
life. High accuracy of the results proved strong training ability of the 3D printing can use benefits of data-driven approaches to achieve the
SVM model. At the same time, a data-driven fatigue life prediction of optimized microstructure and higher fatigue resistance.
additively manufactured titanium alloy has been presented in [117].
4. Challenges and perspectives
In this context, damage mechanics based fatigue models are presented
for fatigue life prediction of 3D-printed titanium alloy. Then, a com-
Although data-driven approaches are of significant interest espe-
putational methodology is demonstrated. Performance of the proposed
cially in arriving at extreme value properties such as cyclic fatigue,
framework is evaluated by tests on smooth and notched specimens. All
there are some challenging issues for the current theoretical and practi-
predicted values verified the applicability of the approach. cal applications. Considerable progress has been achieved in this field,
In [118], we have reviewed applications of ML as a subfield of AI in but the following challenges still remain:
predicting the mechanical behavior of 3D-printed components. Indeed, – A full prediction of fatigue failure based on the microstructure
we have analyzed applications of ML in characterization of polymeric dependent scatter in the early stage of crack propagation, is still a
and metallic 3D-printed parts and discussed limitations of the field. challenging issue. The fatigue scatter depends on the applied load
In a recent study, [119], Monte Carlo simulation is used to enlarge and further researches are required for characterization of material
dataset size and a ML method has been recommended to investigate variability through probabilistic models.
the synergic influence of build orientation, defect size, and depth on – By acquisition of data in real-time, there is the possibility to
fatigue behavior of Ti–6Al–4V. In Fig. 8, the flowchart of ML process use data-driven approaches to predict the crack length in real time.
for fatigue life prediction is depicted. This technique can be used in different engineering systems such as
The documented results indicate that BPNN with the Levenberg– wind turbines and rotors, but due to the manual procedures in signal
Marquardt algorithm shows a good fault tolerance in processing of processing and dependency on hardware performance further attempts
very high cycle fatigue properties. However, similar to other reviewed are required.

10
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

– As training has a crucial role in accuracy and the results of the References
data-driven based system, applicability of these systems are limited to
the experimental space where they are trained. Therefore, a correct [1] T.S. Srivatstan, M.A. Imam, Fatigue of Materials: Advances and Emergences in
Understanding, Wiley, 2010.
choice of representative input variables and creating a large database
[2] S.L. Filho, C.T. Garcia, M.V. Donadon, F. Scapa, T.H. Panzera, The impact
for training would play a key role in application of developed systems. behaviour of hybrid fibre-particle composites based on a full factorial design,
– Since microstructure design is a crucial issue, designing more Mater. Today Commun. 31 (2022) 103459.
practical microstructures can be achieved by AI approaches. It can over- [3] X. Liu, X. Wang, Z. Liu, Z. Chen, Q. Sun, Continuum damage mechanics based
probabilistic fatigue life prediction for metallic material, J. Mater. Res. Technol.
come some limitations and improves mechanical properties (e.g., ten- 18 (2022) 75–84.
sile strength, fatigue behavior) of mechanical parts. In this respect, a [4] G. Kang, H. Luo, Review on fatigue life prediction models of welded joint, Acta
deeper understanding of microstructures and fabrication mechanisms Mech. Sin. 36 (2020) 701–726.
is a necessity. [5] Q. An, R. An, C. Wang, H. Wang, Ductile-to-brittle transition in fracture
behaviors of common solder alloys over a temperature range down to -150
– The are numerous processing parameters which have significant C, Mater. Today Commun. 29 (2021) 102962.
effects on the mechanical behavior of 3D-printed parts. Data-driven [6] M.C. Tanzi, S. Fare, G. Candiani, Foundations of Biomaterials Engineering,
approaches can be used to determine process–structure–property re- Academic Press, 2019.
lationships to make the most efficient use of the raw materials. In [7] M. Kiani, G.T. Fry, Fatigue analysis of railway wheel using a multiaxial
strain-based critical-plane index, Fatigue Fract. Eng. Mater. Struct. 41 (2018)
this context, further experimental studies are encouraged to obtain a 412–424.
profound knowledge on theory and common physics behind different [8] F. Iacoviello, V. Di Cocco, S. Natali, A. Brotzu, Grain size and loading conditions
3D printing processes. influence on fatigue crack propagation in a Cu-Zn-Al shape memory alloy, Int.
J. Fatigue 115 (2018) 27–34.
Due to the benefits and advantages of data-driven approaches,
[9] C.W. Ziemian, R.D. Ziemian, Residual strength of additive manufactured ABS
the expectations are high in future applications of data-driven frame- parts subjected to fatigue loading, Int. J. Fatigue 134 (2020) 105455.
works for prediction of fatigue behavior of metallic, composites, and [10] G.S. Pereira, T.R.F. Cavalcante, M.C. Cid, M.F. Oliveira, M.H. Miyazaki, J.A.
3D-printed components. Diaz, W.W. Filho, Effect of saline environment on the fatigue crack growth
resistance of WE43 Mg alloy, Mater. Today Commun. 31 (2022) 103788.
[11] I. Hanhan, A.M. Ortiz Morales, J.J. Solano, M.D. Sangid, Slow crack growth
5. Conclusions in laminate composites via in-situ X-ray tomography and simulations, Int. J.
Fatigue 155 (2022) 106612.
[12] X. Qian, C.T. Nguyen, Y. Petchdemaneengam, Z. Ou, S. Swaddiwudhipong,
Although numerous fatigue tests have been performed over the P. Marshall, Fatigue performance of tubular X-joints with PJP+ welds: II -
years, they are expensive and time-consuming experiments. In this Numerical investigation, J. Constr. Steel Res. 89 (2013) 252–261.
[13] M. Surendran, S. Natarajan, G.S. Palani, S.P.A. Bordas, Linear smoothed
context, data driven approaches have been employed to predict fatigue
extended finite element method for fatigue crack growth simulations, Eng. Fract.
behavior of different structural elements. This technique is beneficial Mech. 206 (2019) 551–564.
when a full physics-based understanding of relevant phenomena is [14] G. Meneghetti, L. Vecchiato, A. Campagnolo, P. Rech, M. Cova, Numerical
too complex and obtaining results via experiment is costly. In the calibration of the direct current potential drop (DCPD) method in fracture
mechanics fatigue tests, Procedia Struct. Integr. 28 (2020) 1536–1550.
present study, we have reviewed applications of various data-driven
[15] Y. Murakami, Metal Fatigue: Effects of Small Defects and Nonmetallic
approaches and developed frameworks utilized for prediction of fatigue Inclusions, Academic Press, 2019.
life time of metallic and non-metallic components. Review and analysis [16] S. Zhnag, S. Mao, D. Arola, D. Zhang, Characterization of the strain-life fatigue
of proposed approaches confirmed that instead of complex experiments, properties of thin sheet metal using an optical extensometer, Opt. Lasers Eng.
60 (2014) 44–48.
XFEM can be used to generate datasets for predictions. In addition,
[17] Q.Y. Duan, H.W. Xue, Y.H. Yang, Y.Z. Zhou, H.M. Xie, W. He, Study on fracture
due to the defect distribution in geometrically complex 3D-printed behavior of nickel-based single crystal superalloy subjected to high temperature
parts, utilizing miniature test coupons rather than standard specimens fatigue using digital image correlation, Int. J. Fatigue 155 (2022) 106598.
is encouraged. Our discussion in this paper shows the advantages, and [18] L. Garijo, J.J. Ortega, G. Guiz, A. De La Rosa, X.X. Zhang, Effect of loading
frequency on the fatigue life in compression of natural hydraulic lime mortars,
limitations of the proposed approaches. Based on the current review,
Theor. Appl. Fract. Mech. 118 (2022) 103201.
we conclude that several challenges in fatigue testing can be addressed [19] Z. Zheng, M. Zhan, M.W. Fu, Microstructural and geometrical size effects on
through data-driven approaches. Since previous research works indi- the fatigue of metallic materials, Int. J. Mech. Sci. 218 (2022) 107058.
cate successful applications of data-driven approaches for prediction [20] P. Zuo, D.V. Srinivasan, A.P. Vassilopoulos, Review of hybrid composites
fatigue, Compos. Struct. 274 (2021) 114358.
of fatigue life time, further significant developments in this field are
[21] B.F. Sørensen, S. Soutianos, L.P. Mikkelsen, S. Fæster, Fatigue damage growth
expected. and fatigue life of unidirectional composites, Compos. Sci. Technol. 211 (2021)
108656.
[22] W. Song, W. Fan, T. Liu, X. Chen, S. Wang, Y. Zhao, C. Song, P. Mi, H.
CRediT authorship contribution statement Dou, X. Gao, Flexural fatigue properties and failure propagation of 3D stitched
composites under 3-point bending loading, Int. J. Fatigue 153 (2021) 106507.
[23] G. Hou, D.G. Shang, L.X. Zuo, L.F. Qu, M. Xia, S. Wu, G.C. Hao, Fatigue
Sara Nasiri: Methodology, Formal analysis, Investigation, Writing
life prediction of needled ceramic matrix composite under variable amplitude
– original draft. Mohammad Reza Khosravani: Conceptualization, loading, Int. J. Fatigue 156 (2022) 106690.
Formal analysis, Writing – review & editing. [24] J. Geng, Y. Li, F. Wang, C. Zhang, D. Chen, M. Wang, H. Wang, Revealing
the complex effects of particle bands on fatigue crack growth in an extruded
aluminium matrix composite, Int. J. Fatigue 157 (2022) 106720.
Declaration of competing interest [25] R.D.P. Reddy, V. Sharma, Additive manufacturing in drug delivery applications:
A review, Int. J. Pharm. 589 (2020) 119820.
[26] S. Nasiri, M.R. Khosravani, Progress and challenges in fabrication of wearable
The authors declare that they have no known competing finan- sensors for health monitoring, Sensors Actuators A 312 (2020) 112105.
cial interests or personal relationships that could have appeared to [27] E.J. Parry, J.M. Best, C.E. Banks, Three-dimensional (3D) scanning and additive
influence the work reported in this paper. manufacturing (AM) allows the fabrication of customised crutch grips, Mater.
Today Commun. 25 (2020) 101225.
[28] N. Sanaei, A. Fatemi, Defects in additive manufactured metals and their effect
Data availability on fatigue performance: A state-of-the-art review, Prog. Mater. Sci. 117 (2021)
100724.
[29] T.H. Becker, P. Kumar, U. Ramamurty, Fracture and fatigue in additively
No data was used for the research described in the article. manufactured metals, Acta Mater. 219 (2021) 117240.

11
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

[30] Y. Gan, H. Zhang, B. Savija, E. Schlangen, K.V. Breguel, Static and fatigue tests [59] A. Agrawal, P.D. Deshpande, A. Cecen, G.P. Basavarsu, A.N. Choudhary, S.R.
on cementitious cantilever beams using nanoindenter, Micromachines 9 (2018) Kalidindi, Exploration of data science techniques to predict fatigue strength of
1–15. steel from composition and processing parameters, Integr. Mater. Manuf. Innov.
[31] D. Chicot, K. Tilkin, K. Jankowski, A. Wymysłowski, Reliability analysis of 3 (2014).
solder joints due to creep and fatigue in microelectronic packaging using [60] R.K. Neerukatti, K.C. Liu, N. Kovvali, A. Chattopadhyay, Fatigue life prediction
microindentation technique, Microelectron. Reliab. 53 (2013) 761–766. using hybrid prognosis for structural health monitoring, J. Aerosp. Inf. Syst. 11
[32] Z. Wang, C. Wang, Y.L. Zhao, J.J. Kai, C.T. Liu, C.H. Hsueh, Fatigue stud- (2014).
ies of CoCrFeMnNi high entropy alloy films using nanoindentation dynamic [61] ASTM E647 - 15 Measurement of Fatigue Crack Growth Rates da/dN, Standard,
mechanical analyses, Surf. Coat. Technol. 410 (2021) 126927. American Society for Testing Materials, West Conshohocken, USA, 2015.
[33] O. Kraft, C.A. Volkert, Mechanical testing of thin films and small structures, [62] A. Rovinelli, M.D. Sangid, H. Proudhon, Y. Guilhem, R.A. Lebensohn, W.
Adv. Eng. Mater. 3 (2001) 99–110. Ludwig, Predicting the 3D fatigue crack growth rate of small cracks using mul-
[34] G. Zambelis, T. Da Silva Botelho, O. Klinkova, I. Tawfiq, G. Lanouette, A timodal data via Bayesian networks: In-situ experiments and crystal plasticity
new approach in testing fatigue fracture mechanics properties in asymmetrical simulations, J. Mech. Phys. Solids 115 (2018) 208–229.
bonded composite/metal assemblies, Compos., Part B 158 (2019) 390–399. [63] K.P. Baxevanakis, B. Wisner, S. Schlenker, H. Baid, A. Kontsos, Data-driven
[35] T. Gu, K.S. Stopka, C. Xu, D.L. McDowell, Prediction of maximum fatigue damage model based on nondestructive evaluation, ASME J. Nondestr. Eval.
indicator parameters for duplex Ti6Al4V using extreme value theory, Acta Diagn. Progn. Eng. Syst. 1 (2017) 1–12.
Mater. 188 (2020) 504–516. [64] N. Co, D.E. Brown, J.T. Burns, Data-science analysis of the macro-scale features
[36] A. Mohabeddine, J. Correia, P.A. Montenegro, A. De Jesus, J.M. Castro, R. governing the corrosion to crack transition in AA7050-T7451, JOM 70 (2018)
Calcada, F. Berto, An approach for predicting fatigue life of CFRP retrofitted 1168–1174.
metallic structural details, Int. J. Fatigue 154 (2022) 106557. [65] O.L. Kafka, C. Yu, M. Shakoor, Z. Liu, G.J. Wagner, W.K. Liu, Data-driven
[37] G.I. Barenblatt, Scaling phenomena in fatigue and fracture, Int. J. Fract. 138 mechanistic modeling of influence of microstructure on high-cycle fatigue life
(2006) 19–35. of nickel titanium, JOM 70 (2018) 1154–1158.
[38] M.M. Islam, S.I. Shakil, N.M. Shaheen, P. Bayati, M. Haghshenas, An overview [66] K.D. Pierson, J.D. Hochhalter, S.D. Spear, Data-driven correlation analysis
of microscale indentation fatigue: Composites, thin films, coatings, and between observed 3D fatigue-crack path and computed fields from high-fidelity,
ceramics, Micron 18 (2021) 103110. crystal-plasticity, finite-element simulations, JOM 70 (2018) 1158–1167.
[39] C. Keller, E. Hug, X. Feaugas, Microstructural size effects on mechanical [67] R.K. Neerukatti, A. Chattopadhyay, N. Iyyer, N. Phan, A hybrid prognosis model
properties of high purity nickel, Int. J. Plast. 27 (2011) 635–654. for predicting fatigue crack propagation under biaxial in-phase and out-of-phase
[40] C.E. Bakis, L.C. Bank, V. Brown, E. Cosenza, J.F. Davalos, J. Lesko, A. loading, Struct. Health Monit. 17 (2018).
Machida, S. Rizkalla, T. Triantafillou, Fiber-reinforced polymer composites for [68] J. Jin, P. Shokouhi, A data-driven approach to construct a quantitative rela-
construction - state-of-the-art review, J. Compos. Constr. 6 (2002) 73–78. tionship between microstructural features of fatigue cracks and contact acoustic
[41] A.F. S. V. Mortazavian, Effects of fiber orientation and anisotropy on tensile nonlinearity, AIP Adv. 9 (2019) 1–12.
strength and elastic modulus of short fiber reinforced polymer composites, [69] S.Z. Feng, X. Han, Z.J. Ma, G. Krolczyk, Z.X. Li, Data-driven algorithm for real-
Composites, Part B 72 (2015) 116–129. time fatigue life prediction of structures with stochastic parameters, Comput.
[42] M.R. Khosravani, K. Weinberg, Experimental investigations of the environmental Methods Appl. Mech. Engrg. 372 (2019).
effects on stability and integrity of composite sandwich T-joints, Materialwiss. [70] C. Gebhardt, T. Trimborn, F. Weber, A. Bezold, C. Broeckmann, M. Herty,
Werkstofftech. 48 (2017) 753–759. Simplified ResNet approach for data driven prediction of microstructure-fatigue
[43] M.R. P. Edwards, Fatigue performance evaluation of selective laser melted relationship, Mech. Mater. 151 (2020) 103625.
Ti-6Al-4V, Mater. Sci. Eng., A 598 (2014) 327–337. [71] H.B. Kong, S.H. Jo, J.H. Jung, J.M. Ha, Y.C. Shin, H. Yoon, K.H. Sun, Y.H.
[44] M. Kahlin, H. Ansell, J.J. Moverare, Fatigue behaviour of additive manufactured Seo, B.C. Jeon, Hybrid approach of data-driven and physics-based methods for
Ti6Al4V, with as-built surfaces, exposed to variable amplitude loading, Int. J. estimation and prediction of fatigue crack growth 11, 2020, pp. 1–12.
Fatigue 103 (2017) 353–362. [72] S. Dharmadhikari, C. Bhattacharya, A. Ray, A. Basak, A data-driven framework
[45] O.H. Ezeh, L. Susmel, On the notch fatigue strength of additively manufactured for early-stage fatigue damage detection in aluminum alloys using ultrasonic
polylactide (PLA), Int. J. Fatigue 136 (2020) 105583. sensors, Machines 9 (2021) 1–22.
[46] R. Baptista, M. Guedes, Porosity and pore design influence on fatigue behavior [73] T. Meshii, Data-driven approach to construct a fracture toughness master curve
of 3D printed scaffolds for trabecular bone replacement, J. Mech. Behav. for ferritic steels based on tensile properties, Fatigue Fract. Eng. Mater. Struct.
Biomed. Mater. 117 (2021) 104378. 45 (2021) 617–619.
[47] J. Dzugan, M. Seifi, R. Prochazka, M. Rund, P. Podany, P. Konopik, J.J. [74] M. Farid, Data-driven method for real-time prediction and uncertainty quantifi-
Lewandowski, Effects of thickness and orientation on the small scale fracture cation of fatigue failure under stochastic loading using artificial neural networks
behaviour of additively manufactured Ti-6Al-4V, Mater. Charact. 143 (2018) and Gaussian process regression, Int. J. Fatigue 155 (2022) 106415.
94–109. [75] D. Da, Y.C. Chan, L. Wang, W. Chen, Data-driven and topological design of
[48] J. Pegues, M. Roach, R.S. Williamson, N. Shamsaei, Surface roughness effects structural metamaterials for fracture resistance, Extreme Mech. Lett. 50 (2022)
on the fatigue strength of additively manufactured Ti-6Al-4V, Int. J. Fatigue 101528.
116 (2018) 543–552. [76] Y. Tomita, K. Morioka, M. Iwas, Bending fatigue of long carbon fiber-reinforced
[49] A.B. Spierings, T.L. Starr, K. Wegener, Fatigue performance of additive epoxy composite, Mater. Sci. Eng., A 319 (2001) 672–682.
manufactured metallic parts, Rapid Prototyp. J. 19 (2013) 88–94. [77] A.R. Bezazi, A. El Mahi, J.M. Berthelot, B. Bezzazi, Flexural fatigue behavior
[50] O. Holzmond, X. Li, In situ real time defect detection of 3D printed parts, Addit. of cross-ply laminates: An experimental approach, Strength Mater. 35 (2003)
Manuf. 17 (2017) 135–142. 149–161.
[51] J.A. Travieso-Rodriguez, M.D. ZAndi, R. Jerez-Mesa, J. Liuma-Fuentes, Fatigue [78] R. Sakin, I. Ay, R. Yaman, An investigation of bending fatigue behavior for
behavior of PLA-wood composite manufactured by fused filament fabrication, glass-fiber reinforced polyester composite materials, Mater. Des. 29 (2008)
J. Mater. Res. Technol. 9 (2020) 8507–8516. 212–217.
[52] M.R. Khosravani, T. Reinicke, Fracture behavior of intact and defected [79] G. Stegschuster, K. Pingkarawat, B. Wendeland, A.P. Mourtiz, Experimental
3D-printed parts, Procedia Struct. Integr. 31 (2021) 105–110. determination of the mode I delamination fracture and fatigue properties of
[53] Y. Tao, F. Kong, Z. Li, J. Zhnag, X. Zhao, Q. Yin, D. Xing, P. Li, A review on thin 3D woven composites, Compos., Part A 84 (2016) 308–315.
voids of 3D printed parts by fused filament fabrication, J. Mater. Res. Technol. [80] A.C.S. Trapani, T.A. Barreto, J.R. Trapani, Fatigue failure analysis of riveted
15 (2021) 4860–4879. fibre-metal laminate lap joints, Eng. Fract. Mech. 239 (2020) 107275.
[54] M.R. Khosravani, S. Nasiri, D. Anders, K. Weinberg, Prediction of dynamic [81] M. Eftekhari, A. Fatemi, Variable amplitude fatigue behavior of neat and short
properties of ultra-high performance concrete by an artificial intelligence glass fiber reinforced thermoplastics, Int. J. Fatigue 98 (2021) 176–186.
approach, Adv. Eng. Softw. 127 (2019) 51–58. [82] M. Brod, A. Dean, R. Rolfes, Numerical life prediction of unidirectional fiber
[55] J. Gao, G. Wang, Z. Xu, J. Wang, S. Yan, Z. Wang, Gaussian process regression composites under block loading conditions using a progressive fatigue damage
based remaining fatigue life prediction for metallic materials under two-step model, Int. J. Fatigue 147 (2021) 106159.
loading, Int. J. Fatigue 158 (2022) 106730. [83] C.M. Manjunatha, N. Jagannathan, K. Padmalatha, Improved variable-amplitude
[56] X.Q. Li, L.K. Song, G.C. Bai, Deep learning regression-based stratified proba- fatigue behavior of a glass-fiber-reinforced hybrid-toughened epoxy composite,
bilistic combined cycle fatigue damage evaluation for turbine bladed disks, Int. J. Reinf. Plast. Compos. 30 (2011) 1783–1793.
J. Fatigue 159 (2022) 106812. [84] S.K. Jha, R.A. Brockman, R.M. Hoffman, V. Sinha, A.L. Pilchak, W.J. Porter,
[57] E. Barbero, G.F. Abdelal, A. Caceres, A micromechanics approach for damage D.J. Buchanan, J.M. Larsen, R. John, A data analytics approach to discovering
modeling of polymer matrix composites, Compos. Struct. 67 (2005) 427–436. unique microstructural configurations susceptible to fatigue, JOM 70 (2018)
[58] A.S. Kompalka, S. Reese, O.T. Bruhns, Experimental investigation of damage 1147–1153.
evolution by data-driven stochastic subspace identification and iterative finite [85] J. Guilleminot, J.E. Dolbow, Data-driven enhancement of fracture paths in
element model updating, Arch. Appl. Mech. 77 (2007) 559–573. random composites, Mech. Res. Commun. 103 (2020) 103443.

12
S. Nasiri and M.R. Khosravani Materials Today Communications 33 (2022) 104437

[86] P.L.T. Doung, S. Hussain, M.H. Jhon, N. Raghavan, Data driven prognosis of [105] S. Cicero, V. Martinez-Mata, L. Castanano-Jano, A. Alonso-Esebanez, B. Arroyo,
fracture dynamics using tensor train and gaussian process regression, IEEE Analysis of notch effect in the fracture behaviour of additively manufactured
Access 8 (2020). PLA and graphene reinforced PLA, Theor. Appl. Fract. Mech. 114 (2021)
[87] D.H. Nguyen-Le, Q.B. Tao, V.H. Nguyen, M. Abdel-Wahab, H. Nguyen-Xuan, 103032.
A data-driven approach based on long short-term memory and hidden Markov [106] A. Milovanovic, A. Sedmak, A. Grbovic, Z. Golubovic, M. Milosevic, Influence
model for crack propagation prediction, Eng. Fract. Mech. 235 (2020) 107085. of second-phase particles on fracture behavior of PLA and advanced PLA-X
[88] A. Carpinteri, S. Valente, G. Ferrara, G. Melchiorr, Is mode II fracture energy material, Procedia Struct. Integr. 31 (2021) 122–126.
a real material property? Comput. Struct. 48 (1993) 397–413. [107] V. Shanmugam, O. Das, K. Babu, U. Marimuthu, A. Veerasimman, D.J. Johnson,
[89] J. Pegues, M. Roach, R.S. Williamson, N. Shamsaei, Data-driven failure predic- R. Esmaeely Neisiany, M.S. Hedenqvist, S. Ramakrishna, F. Berto, Fatigue
tion in brittle materials: A phase field-based machine learning framework, J. behaviour of FDM-3D printed polymers, polymeric composites and architected
Mach. Learn. Model. Comput. 2 (2021) 65–89. cellular materials, Int. J. Fatigue 143 (2021) 106007.
[90] K. Lyathakula, F.G. Yuan, A probabilistic fatigue life prediction for adhesively [108] Z. Liu, O.L. Kafka, C. Yu, W.K. Liu, Data-driven self-consistent clustering
bonded joints via ANNs-based hybrid model, Int. J. Fatigue 151 (2021) 106352. analysis of heterogeneous materials with crystal plasticity, in: E. Onate, D. Peric,
[91] P. Carrara, M. Ortiz, L. De Lorenzis, Data-driven rate-dependent fracture E.S. Neto, M. Chiumenti (Eds.), Advances in Computational Plasticity, Springer,
mechanics, J. Mech. Phys. Solids 155 (2021) 104559. 2017, pp. 221–242, Ch. 10.
[92] P. Carrara, L. De Lorenzis, L. Stainier, M. Ortiz, Data-driven fracture mechanics, [109] W. Yan, S. Lin, R.L. Kafka, Y. Liaan, C. Yu, Z. Liu, J. Yan, S. Wolff, H. Wu, E.
Comput. Methods Appl. Mech. Engrg. 372 (2020) 113390. Agbor, M. Mozaffar, K. Ehmann, J. Cao, G. Wagner, W.K. Liu, Data-drivenmulti-
[93] C.J. Larsen, M. Ortiz, C.L. Richardson, Fracture paths from front kinetics: scale multi-physicsmodels to derive process–structure–property relationships for
Relaxation and rate independence, Arch. Ration. Mech. Anal. 193 (2009) additive manufacturing, Comput. Mech. 61 (2018) 521–541.
539–583. [110] H.Y. Wan, G.F. Chen, C.P. Li, G.P.Z. X. B. Qi, Data-driven evaluation of fatigue
[94] M. Negri, From rate-dependent to rate-independent brittle crack propagation, performance of additive manufactured parts using miniature specimens, J.
J. Elasticity 98 (2010) 159–187. Mater. Sci. Technol. 35 (2019) 1137–1146.
[95] S.Z. Feng, X. Han, Z. Li, A. Incecik, Ensemble learning for remaining fatigue [111] M. Seifi, M. Gorelik, J. Waller, N. Harbe, N. Shamsaei, S. Daniewicz, J.J.
life prediction of structures with stochastic parameters: A data-driven approach, Lewandowski, Progress towards metal additive manufacturing standardization
Int. J. Fatigue 101 (2022) 420–431. to support qualification and certification, JOM 69 (2017) 439–455.
[96] L.E. Murr, Metallurgy principles applied to powder bed fusion 3D print- [112] M. Zhang, C.N. Sun, X. Zhang, P.C. Goh, J. Wei, D. Hardacre, H. Li, High cycle
ing/additive manufacturing of personalized and optimized metal and alloy fatigue life prediction of laser additive manufactured stainless steel: A machine
biomedical implants: An overview, J. Mater. Res. Technol. 9 (2019) 1087–1103. learning approach, Int. J. Fatigue 128 (2019) 105194.
[97] M.R. Khosravani, F. Berto, M.R. Ayatollahi, T. Reinicke, Fracture behavior of [113] A. Haghighi, L. Li, A hybrid physics-based and data-driven approach for char-
additively manufactured components: A review, Theor. Appl. Fract. Mech. 109 acterizing porosity variation and filament bonding in extrusion-based additive
(2020) 102763. manufacturing, Addit. Manuf. 36 (2020) 101399.
[98] M. Hikmat, S. Rostam, Y.M. Ahmed, Investigation of tensile property-based [114] S. Lee, B. Rasoolian, D.F. Silva, J.W. Pegues, N. Shamsaei, Surface roughness
aguchi method of PLA parts fabricated by FDM 3D printing technology, Results parameter and modeling for fatigue behavior of additive manufactured parts:
Eng. 11 (2021) 100264. A non-destructive data-driven approach, Addit. Manuf. 46 (2021) 102094.
[99] M. Azadi, A. dadashi, S. Dezianinan, M. Kianifar, S. Torkman, M. Chiyani, [115] M.R. Khosravani, Z. Bozic, A. Zolfagharian, T. Reinicke, Failure analysis of 3D-
High-cycle bending fatigue properties of additive-manufactured ABS and PLA printed PLA components: Impact of manufacturing defects and thermal ageing,
polymers fabricated by fused deposition modeling 3D-printing, Forces Mech. 3 Eng. Fail. Anal. 136 (2022) 106214.
(2021) 100016. [116] H. Bao, S. Wu, Z. Wu, G. Kang, X. Peng, P.J. Withers, A machine-learning
[100] A. Szust, G. Adamski, Using thermal annealing and salt remelting to increase fatigue life prediction approach of additively manufactured metals, Eng. Fract.
tensile properties of 3D FDM prints, Eng. Fail. Anal. 132 (2022) 105932. Mech. 242 (2021) 107508.
[101] F.S. Senatov, K.V. Niaza, A.A. Stepashkin, S.D. Kaloshkin, Low-cycle fatigue [117] Z. Zhan, W. Hu, Q. Meng, Data-driven fatigue life prediction in additive
behavior of 3D-printed PLA-based porous scaffolds, Compos., Part B 97 (2016) manufactured titanium alloy: A damage mechanics based machine learning
193–200. framework, Eng. Fract. Mech. 252 (2021) 107850.
[102] B. Gong, S. Cui, Y. Zhao, Y. Sun, Q. Ding, Strain-controlled fatigue behaviors of [118] S. Nasiri, M.R. Khosravani, Machine learning in predicting mechanical behavior
porous PLA based scaffolds by 3D-printing technology, J. Biomater. Sci., Polym. of additively manufactured parts, J. Mater. Res. Technol. 14 (2021) 1137–1153.
Ed. 28 (2017) 2196–2204. [119] J. Li, Z. Yang, G. Qian, F. Berto, Machine learning based very-high-cycle fatigue
[103] M. Hikmat, S. Rostam, Y.M. Ahmed, Tensile failure strength and separation life prediction of Ti-6Al-4V alloy fabricated by selective laser melting, Int. J.
angle of FDM 3D printing PLA material: Experimental and theoretical analyses, Fatigue 158 (2022) 106764.
Compos., Part B 188 (2020) 107894. [120] E. Maleki, S. Bagherifard, S.M.J. Razavi, M. Bandini, A. du Plessis, F. Berto,
[104] M.R. Khosravani, T. Reinicke, Effects of raster layup and printing speed on M. Guagliano, On the efficiency of machine learning for fatigue assessment of
strength of 3D-printed structural components, Procedia Struct. Integrity 28 post-processed additively manufactured alsi10mg, Int. J. Fatigue 160 (2022)
(2020) 720–725. 106841.

13