Renewable and Sustainable Energy Reviews 82 (2018) 940–949

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

Advances in fatigue life modeling: A review MARK

a b,⁎
M. Kamal , M.M. Rahman
a
International Affairs and Training Division, Pakistan Atomic Energy Commission, Karachi, Pakistan
b
Automotive Engineering Research Group, Faculty of Mechanical Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia

A R T I C L E I N F O A B S T R A C T

Keywords: The purpose of this paper is to examine the state-of-the-art research efforts linked with the development of
Fatigue fatigue life estimation models. The main objective is to identify new concepts for fatigue life estimation other
Critical plane than the classical models and their hybrids. Various techniques to estimate fatigue life have been identified, such
Optimization algorithm as critical plane deviation, 5D deviatoric space enclosed surface, modified Wholer curve. However, the most
Classical fatigue models
notable one to be found is the application of evolutionary optimization algorithms for, e.g., genetic algorithms,
Hybrid models
artificial neural networking, and differential ant-stigmergy algorithms. Initially, a brief history of fatigue life
estimation and modeling is presented. In subsequent sections, some familiar classical models are discussed, and
then various innovative approaches to fatigue life prediction are reviewed. The survey is fairly detailed, and best
efforts have been made to the net in as many new methodologies as possible. The review is organized to offer
insight on how past research efforts have provided the groundwork for subsequent studies.

1. Introduction Stress analysis is conducted to correctly estimate fatigue damage by

The importance of fatigue is evident; although exact numbers are
not available, it is expected that at least half of all mechanical failures
are due to fatigue. The cost of these failures constitutes approximately
4% of the gross national product of the USA [1]. That is why it is es-
sential to understand the physics of fatigue, to create a cause and effect
relationship to reduce the probability of such failures [2]. Since the
investigations by Wohler in 1860, fatigue experiments and predictions
have played a major role in mechanical design [3,4], and researchers
are investigating the fatigue problem have made enormous efforts to
devise sound methodologies suitable for safely assessing mechanical
components subjected to time-variable loadings [5–10]. It is an ac-
knowledged fact that to estimate fatigue life accurately in real-world
scenarios is a complex task in which various variables have to be taken
into account to avoid unwanted and dangerous failures [11]. The re-
liability of a fatigue estimation technique depends on its ability to
model damage due to non-zero superimposed static stresses, the degree
of multiaxiality in the stress field and the effects of stress concentration
phenomena [12]. For the cases of cyclic and random multiaxial loading
conditions, it was difficult to estimate fatigue life as damage is depen-
dent on all the stress components and their variations during the whole
period of load application [12,13]. The fatigue assessment method
should be calibrated concerning some experimental information that
can be easily obtained through tests run by the relevant standard codes
to predict the fatigue estimation results accurately [6,11,12,14,15].
direct post-processing simple linear elastic finite element [16] models
[17–21].
In this paper, a review of fatigue estimation techniques has been
presented, with the emphasis on the newly proposed models high-
lighting new ideas to estimate fatigue life. As the models developed by
accommodating the proposed modifications to the earlier models have
limited capability of estimating the fatigue life in limited loading or
material conditions, no universally accepted model can estimate or
predict fatigue life for a range of loadings as well as material conditions
[6,22]. This review paper attempts to catalogue the novel concepts as
well as methods for fatigue life estimation that may be helpful in for-
mulating a universal model for a broad range of loadings and material
conditions.
2. Material fatigue
Fatigue is defined as a failure under a repeated load which never
reaches a level sufficient to cause failure in a single application [17].
The word fatigue originates from the Latin expression ‘fatigue’ which
means ‘to tire'. The terminology used in engineering refers to the da-
mage and failure of materials under cyclic loads, including mechanical
loads, thermal loads, and so forth [23]. Fatigue damage characterized
by nucleation, coalescence and stable growth of cracks, leading ulti-
mately to net section yielding or brittle fracture. An evaluation of fa-
tigue of structures and materials in the 20th century raises the question

⁎
Corresponding author.
E-mail address: mustafizur@ump.edu.my (M.M. Rahman).

http://dx.doi.org/10.1016/j.rser.2017.09.047
Received 13 August 2015; Received in revised form 27 July 2017; Accepted 14 September 2017
1364-0321/ © 2017 Elsevier Ltd. All rights reserved.
M. Kamal, M.M. Rahman Renewable and Sustainable Energy Reviews 82 (2018) 940–949

what may have happened in the 19th century. The fatigue of structures
became evident as a by-product of the industrial revolution in the 19th
century. Fatigue failures frequently associated with steam engines, lo-
comotives, and pumps. Systematic fatigue tests were done in a few la-
boratories, notably by August Wohler. It recognized that small radii in
the geometry of the structure would be avoided. A fundamental step
regarding fatigue as a material problem made at the beginning of the
20th century by Ewing and Humfrey [24]. The author investigated that
the fatigue crack nuclei start as microcracks in slip bands [25]. When
components stressed in the high-cycle fatigue (HCF) or very high-cycle
fatigue (VHCF) regime, most load cycles in realistic in-service loading
sequences are at stress amplitudes that are too low to cause failure
under constant amplitude loading conditions. Constant amplitude cy-
cling below the endurance limit does not lead to fracture, but it can
cause fatigue damage. Several investigations show that short fatigue
cracks could be initiated by cycling carbon steels below the endurance
limit [26,27].
In the real-world scenarios, loading conditions are variable and
complex, and the resulting stress states are also multiaxial. Multiaxial
loads, which can be in-phase (proportional) or out-of-phase (non-pro-
portional), are common for many components and structures. Even
under uniaxial loads, multiaxial stresses often exist, although typically
in-phase, for example, due to geometric constraints at notches. Such
multiaxial loads and stress states are frequently encountered in many
industries, including automotive, aerospace, and power generation,
among others [28]. Non-proportional multiaxial fatigue damage occurs
if the principal stress directions vary during the loading induced by out-
of-phase bending and torsion moments [29]. The methodologies for the
more complex case of multiaxial variable amplitude loading are not yet
well established, particularly when the loads are non-proportional [30].
3. Reported researches on fatigue life estimation methods
3.1. Classical models
strength, σuts = ultimate tensile strength.
Van [37] proposed an endurance limit criterion, also known as the
Dang Van model, based on the concept of micro-stresses within a cri-
tical volume of material, expressed in Eq. (5). Hofmann and Bertolino
[38] and Charkaluk and Constantinescu [39] revisited Dang Van model
and suggested a finer qualitative analysis to understand the ability of
the model better.
τ (t ) + aσh (t ) = b (5)
where τ(t) = instantaneous shear stress, σh(t) = instantaneous hydro-
static stress, a and b = material constants.
3.1.2. Strain-based models
Brown and Miller [40,41] and Kandil and Brown [42] proposed a
parameter based on the maximum shear strain range and normal strain
range on the plane experiencing the maximum shear strain range, as
expressed in Eq. (6).
Δγmax σ f, − 2σn, mean
+ SΔεn = A (2Nf )b + Bε f, (2Nf )c
2 E (6)
where Δγmax = maximum shear strain range, Δεn = normal strain
range, σn,mean = mean normal stress, S, A and B = material constants.
Wang and Brown [43] proposed a modification of the model pro-
posed by [41], adding the capability to handle the strain path effect.
The model is expressed in Eq. (7).
Δγˆ Δγ
= max + Sεn* = (1 + νe + (1 − νe ) S ) σ ′f (2Nf )b
2 2
+ (1 + νp + (1 − νp) S ) ε′f (2Nf )c (7)
where Δγ̂ = equivalent shear strain connection, Δγmax = maximum
shear strain range, εn* = normal strain excursion between two turning
points of γmax, νe and νp = elastic and plastic Poisson ratio, S = material
parameter representing the influence of normal strain on fatigue crack
growth.

3.1.1. Stress-based models 3.1.3. Strain energy-based models

Sines [31], Sines [32] proposed that octahedral (von Mises) shear Smith, Watson [44] proposed a damage model also known as the
stress is used as a fatigue damage criterion as expressed in Eq. (1), but Smith Watson Topper (SWT) model, including the cyclic normal strain
this model is incapable of handling non-proportional loading. range and maximum normal stress, as expressed in Eq. (8); the critical
plane is identified as the plane of maximum normal stress. This model
Δτoct
+ α (3σh) = β was originally developed, and it is still used for mean stress correction.
2 (1)
Δε1 σ f,2
where Δτoct = octahedral stress (von Mises) range, σh = hydrostatic σn, max = (2Nf )2b + σ f, ε′f (2Nf )b + c
stress, α, and β are material parameters. 2 E (8)
Crossland [33] proposed a similar parameter to that of Sines but where σn,max = maximum normal stress, Δε1 = principal strain range,
used maximum hydrostatic stress (σhmax) instead of mean, as expressed σf’ = fatigue strength coefficient, εf’ = fatigue ductility coefficient, E =
in Eq. (2) to face problems in dealing with out-of-phase multiaxial elastic modulus, Nf = fatigue life, b = fatigue strength coefficient, c =
loading [22]. fatigue ductility exponent.
Δτoct Fatemi and Socie [45] suggested a modification to the Brown and
+ α (3σh max ) = β Miller model by replacing the normal strain term with normal stress.
2 (2)
Eq. (9) represents the Fatemi–Socie model when shear fatigue proper-
Findley [34] proposed a fatigue life parameter, as expressed in Eq. ties are used [1] and also in the form of uniaxial fatigue properties
(3) based on the combination of shear stress range and normal stress on [45,46]. Additional cyclic hardening developed during out-of-phase
the plane having the maximum value of the parameter. loading included in the normal stress term. Mean stress accounted for
Δτ by adding the normal mean stress across the maximum shear plane to
⎛ + kσn ⎞ =f
⎝ 2 ⎠max (3) the alternating normal stress across the same plane.
,
where Δτ = shear stress range, σn = normal stress, k = material Δγ ⎛ σn, max ⎞ τ f
⎜1 + k ⎟ = (2Nf )bγ + γ f, (2Nf )cγ
constant. 2 ⎝ σy ⎠ G
McDiarmid [35,36] proposed a model similar to Findley's, as ex- σ ′f ⎡
pressed in Eq. (4), in which the critical plane is identified as the plane = ⎡ (1 + νe ) (2Nf )b + (1 + νp) ε′f (2Nf )c ⎤ ⎢1
⎢ E ⎥
with the maximum shear stress range, but has a large scatter in results. ⎣ ⎦ ⎣
Δτmax σn, max σ ′f ⎤
+ =1 + k ⎛⎜ (2Nf )b⎞⎟ ⎥
2tA, B 2σuts (4) 2
⎝ σy ⎠⎦ (9)
’
where Δτmax = maximum shear stress range, tA,B = shear fatigue where Δγ = shear strain range, σy = yield stress, τf = shear fatigue

941
M. Kamal, M.M. Rahman
strength coefficient, γf = shear fatigue ductility coefficient, G = shear
modulus, bγ = shear fatigue strength exponent, cγ = shear fatigue
ductility exponent, νe and νp = elastic and plastic Poisson ratio.
Liu [47] proposed a model based on virtual strain energy (VSE), i.e.,
the product of stress and strain ranges, expressed in Eqs. (10) and (11).
The critical plane defined by the value of maximum normal work and
the VSE quantity is the sum of normal work and shear work on the
critical plane for tensile failure dominant materials, and vice versa for
shear failure dominant materials.
For tensile failure:
ΔW = (Δσn Δεn )max + (ΔτΔγ ) (10)
For shear failure:
ΔW = (Δσn Δεn ) + (ΔτΔγ )max (11)
where ΔW = virtual strain energy parameter, Δσn = normal stress
range, Δεn = normal strain range, Δτ = shear stress range, Δγ = shear
strain range.
Liu and Wang [48] revisited the VSE model and evaluated its ef-
fectiveness and stated the strengths of the method as its ability to
predict the physical characteristics of fatigue cracks, such as initiation
sites, fracture modes and crack orientations. The method under-pre-
dicted in the case of superimposed compressive mean stress for tor-
sional fatigue life. Núñez and Calvo [49] introduced a probabilistic
formulation in Liu's model based on a perturbation method to obtain
two statistical moments (mean and variance) of the random variable
fatigue lifetime. They compared their results with the analysis done
with a Monte Carlo simulation approach, and good agreement was
found. However, no experimental results are used in the study for
comparison of model performance.
Chu [50] proposed a similar parameter to Liu's model to combine
shear and normal work but replaces stress ranges with maximum
stresses to include the mean stress effect, as expressed in Eq. (12). The
critical plane is defined by the plane having the maximum fatigue
parameter value.
Δγ Δε
ΔW = ⎛τn, max + σn, max ⎞
⎝ 2 2 ⎠max (12)
where τn,max = maximum shear stress on the plane.
Glinka and Wang [51] proposed a shear strain energy model that
includes the effects of both tensile and shear mean stresses, as expressed
in Eq. (13). The critical plane is identified as the plane experiencing the
largest shear work.
Δτ Δγ ⎛ σ f, τ f, ⎞ also highlighted three forms of application of TCD: the point method,
ΔW = +
2 2 ⎜ σ f, − σn, max τ f, − τn, max ⎟ (13)
⎝ ⎠
The next section outlines the models proposed in the last 30 years
which presented new proposals and expressions to estimate multiaxial
fatigue damage.
3.2. Recent advances
The above discussion gives a brief introduction to the fatigue
models that have gained widespread acceptance. The following section
discusses some less known models, which may not be widely used but
present new concepts and fatigue parameter expressions to estimate
multiaxial fatigue life. This review is not exhaustive, but best efforts
have been made to gather as much data as possible about the devel-
opment of fatigue life models from the previously published literature.
3.2.1. Critical plane models
Susmel and Lazzarin [52] and Lazzarin and Susmel [53] proposed a
critical plane model, i.e. the plane of maximum shear stress amplitude,
then calculated both the maximum shear stress amplitude and the
maximum normal and mean stress relative to the same plane. Their
942
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
modified Wholer curve method (MWCM) is expressed in Eqs. (14) and
(15).
k τ (ρeff ) = a ρeff + b (14)
τA,Re f (ρeff ) = α ρeff + β (15)
where kτ(ρeff) = negative inverse slope of Wholer curve, ρeff = critical
plane stress ratio, τA,Ref (ρeff) = reference shear stress amplitude at
considered limit to cycles to failure, a, b, α, β = material constants.
Susmel and Taylor [54] proposed a simplified approach to apply the
theory of critical distances (TCD) [55], as defined in Eq. (16), to predict
the fatigue behavior of notched components subjected to torsional fa-
tigue loading. TCD was found to be a useful tool for assessing fatigue
life, as the experimental results generated using uniaxial testing
equipment and linear elastic FE models are the only two pieces of in-
formation needed for its application.
2
1 ⎛ ΔKI , th ⎞
L= ⎜ ⎟
π ⎝ Δσ0 ⎠ (16)
where L = material characteristic length, ΔKI,th = fatigue threshold
stress intensity factor, Δσ0 = range of uniaxial plane fatigue limit.
Susmel and Taylor [12] later proposed a method which combines
the MWCM and TCD methods to estimate fatigue life under multiaxial
loadings. They concluded that the proposed method is efficient, pro-
vided that it is calibrated by using appropriate pieces of speculative
information. Susmel [56] formulated a critical plane determination
method based on the concept. The material plane where the crack in-
itiation phenomenon takes place is the one containing the direction
along variance of the resolved shear stress reaches its maximum. Ex-
pressions defining variance are shown in Eqs. (17)–(19). The main
feature of the method is the speed of critical plane determination, as the
time needed to find the global maximum is not dependent on the length
of the load history under study.
Var [τq (t )] = dT [C ] d (17)
where d = direction cosines, [C] = matrix consisting of variance Vi,
covariance Ci,j terms defined as:
Vi = Var [σi (t )] for i = x , y, z , xy, xz , yz (18)
Ci, j = CoVar [σi (t ), σj (t )] for i = x , y, z , xy, xz , yz (19)
Susmel and Taylor [57] proposed a reformulation of the TCD
method so that it can be used to estimate the fatigue life of notched
components facing variable amplitude uniaxial fatigue loading. They
line method, and area method, of which the point method is said to be
the easiest to apply regarding calculation load. Susmel and Tovo [15]
presented the methodology of estimating fatigue life under variable
amplitude loading conditions, using MWCM along with Maximum
Variance Method (MVM) for critical plane determination. They also
introduced a material parameter critical damage sum, expressed in Eqs.
(20) and (21), determined experimentally, where this parameter can
vary with the degree of multiaxiality and non-proportionality.
DCR (ρeff ) = d1 ⋅ ρeff + d2 (20)
where d1and d2 = material fatigue properties to be determined ex-
perimentally, ρeff = critical plane stress ratio defined by
m ⋅ σn, m + σn, a
ρeff =
τa (21)
where m = mean stress sensitivity index, σn,m and σn,a = mean and
amplitude of stress perpendicular to the critical plane, τa = shear stress
amplitude.
Susmel and Taylor [6] devised a technique for fatigue assessment of
notched components subjected to variable amplitude fatigue loading.
They used MWCM along with TCD in the form of the point method,
M. Kamal, M.M. Rahman Renewable and Sustainable Energy Reviews 82 (2018) 940–949

where the critical plane is determined using MVM. They concluded that where
their method provided a high level of accuracy and that real compo-
4 k2 k 4 − k2
nents can be designed against variable amplitude uniaxial/multiaxial k< ≅ 1.155: ac = + , bc = f−1
loading by direct post-processing of the relevant stress fields de- 3 2 2 (25)
termined through conventional linear-elastic FE models. Susmel and 2 2
8f−1 k 2 (4 − k 2)
4 4k ⎞
Tovo [58] investigated the MVM method to determine independently k≥ ≅ 1.155: ac = ⎛ ⎜
2
, ⎟ bc =
3 ⎝ +k ⎠
4 (4 + k 2)2 (26)
from the degree of multiaxiality and non-proportionality of the applied
loading history. The path of maximum variance of resolved shear stress where k, ac, bc = material parameter, Ca = shear stress on considered
is similarly capable of accurately estimating the orientation of Stage-I plane, Na = maximum normal stress, Nm = mean normal stress, t−1 =
crack paths. The study showed that MVM is capable of estimating the fatigue limit in fully reversed torsion, f0 = fatigue limit in repeated
direction of the Stage I crack paths under non-proportional loading with axial loading, f−1 = fatigue limit in fully reversed axial loading.
an adequate level of accuracy. Susmel and Louks [59] performed a Lu and Liu [64] proposed a critical plane model with equivalent
validation exercise to judge the effectiveness of TCD in the form of the initial flaw size expressed in Eqs. (27) and (28). The critical plane is
point method along with MWCM in estimating fatigue life for uniaxial/ determined by the maximum normal stress plane and the ratio of mode
multiaxial loading cases by directly post-processing linear elastic stress II and mode I stress intensity factor coefficients corresponding to a
fields calculated from commercial finite element software. A high level specific crack growth rate. The critical plane determination is depen-
of accuracy was observed in the calculated fatigue life results. dent on stress states as well as material properties.
Mahadevan and Liu [60] proposed a fatigue model based on the ac 1
concept of critical plane deviation, as expressed in Eq. (22), where first N= ∫a
i C [ΔK eq − ΔKth]m
da
(27)
a search for the fracture plane is conducted, and then the critical plane
is found at a certain deviation from the fracture plane. The concept is
1 k 2
based on the idea that a crack initiates on one plane and then propa- Kmixed, eq = (k1)2 + ⎛ 2 ⎞ + A (k H )2
B ⎝s⎠ (28)
gates along a distinct plane orientation. The model is tested for constant
amplitude loading and is found to give results in good agreement with where N = fatigue life, ΔKeq = equivalent stress intensity factor, ΔKth
experimental data. Later, Mahadevan and Liu [61] extended the ap- = threshold stress intensity factor, a = crack length, A, B, C, m =
plication domain of their model from isotropic materials to anisotropic material parameters, Kmixed, eq = equivalent stress intensity factor
materials and composite materials. However, due to the non-avail- under general mixed mode loading, k1, k2 and kH = loading related
ability of experimental fatigue data on anisotropic materials in the lit- parameters, s = ratio of stress intensity factors for mode II and mode I.
erature, they suggested testing the proposed model with more experi- Shang and Sun [65] proposed a damage parameter based on a cri-
mental data available in future studies. tical plane having a maximum shear strain and a higher value of normal
strain excursion, as expressed in Eq. (29). The proposed parameter
2 2
1 ⎡ ⎛ σm, c ⎞ ⎤ ⎛ fNf ⎞ 2
shows good correlation with the multiaxial fatigue lives of different
⎢σa, c 1 + η Nf ⎥ +⎜ (τa, c )2 + k (σaH, c ) = fNf materials in low cycle loading cases.
β ⎢ ⎜ fNf ⎟ ⎥ t Nf ⎟
⎣ ⎝ ⎠⎦ ⎝ ⎠ (22)
2 1/2
⎛ε *2 + υeff (2 − υeff ) ⎛ Δγmax ⎞ ⎞ σ ′f
where σa,c = normal stress amplitude on critical plane, σm,c = mean ⎜ n
2 ⎟ = (2Nf )b + ε′f (2Nf )c
⎝ (1 + υeff ) ⎝ 2 ⎠ ⎠ E (29)
normal stress on critical plane, fNf = normal stress at Nf cycles, tNf =
shear stress at Nf cycles, τa,c = shear stress amplitude on critical plane, where εn* = normal strain excursion between turning points on the
σa,cH = hydrostatic stress amplitude on critical plane. critical plane, νeff = effective Poisson ratio, Δγmax = maximum shear
Ninic and Stark [62] proposed a non-linear fatigue damage function strain range.
based on the critical plane concept expressed in Eq. (23), rightly named Li and Sun [66] and Li and Zhang [67] presented a simple critical
as the quadratic critical plane formula by Papuga [22]. The critical plane type method to assess the fatigue life of metallic materials sub-
plane is identified as the plane experiencing the maximum value of jected to proportional and non-proportional loading, and their model is
damage function. Their study concluded that the proposed fatigue da- expressed in Eq. (30). The model has maximum shear strain range
mage function is dependent on the normal stress sensitivity factor for (Δγmax), normal strain range (Δεn) and the maximum normal stress
accurate predictions, and also identified that the endurance strength (σn,max) on the maximum shear strain range plane.
ratio is critical for multiaxial fatigue analysis.
*
Δεeq Δγmax σn, max ⎞ Δεn
2 2 1/2
= + ⎛⎜1 + ⎟
⎡ τ (l, m , n) ⎞ σeq (l, m , n) ⎞ ⎤ 2 2 ⎝ σy ⎠ 2
D (l, m , n) = ⎢ ⎛ a
⎜ + k′ ⎛
⎟ ⎜ ⎟
⎥
⎣ ⎝ Te ⎠ ⎝ Se ⎠⎦ (23) σ f, σ f,
= ⎡ (2Nf )b + ε f, (2Nf )c ⎤ ⎛⎜1 + (2Nf )b⎟⎞
⎣E ⎦ ⎝ σy ⎠ (30)
where D = damage on critical plane, τa = shear stress amplitude, Te =
fatigue limit in fully reversed torsion, σeq = equivalent stress ampli- Ince and Glinka [68] proposed two different forms of fatigue da-
tude, Se = fatigue limit in fully reversed axial and torsion, (l,m,n) = mage parameter related to the maximum fatigue damage plane. The
direction cosines of vector normal to plane, k = normal stress sensi- two forms are the generalized strain energy damage parameter ex-
tivity factor. pressed in Eq. (31) moreover, the generalized strain amplitude damage
Papuga and Ruzicka [63] proposed two criteria with a similar da- parameter expressed in Eq. (32).
mage parameter with the emphasis on the effects of shear stress in
Δγ e Δτ Δγ p Δεne Δσn Δεnp ⎞
comparison to normal stress. The criteria for the search for the critical * = ⎜⎛τmax
W gen + +σ + ⎟ = f (Nf )
plane are expressed in Eqs. (24)–(26). The two methods derived here ⎝ 2 2 2 n, max 2 2 2 ⎠max (31)
are to study whether it is more efficient to integrate the load effect or to
*
Δεgen τ Δγ e Δγ p σn, max Δεne Δε p
maximize it. Both approaches provide very similar results as regards the = ⎛ max
⎜ + + + n⎞ = f (Nf ) ⎟

presented data set of 119 experimental results. 2 ⎝ Δτ /2 2 2 Δσ n /2 2 2 ⎠max (32)

where = maximum generalized strain energy, τmax = maximum
W*gen
t
ac ⋅Ca2 + bc⋅⎜⎛Na + −1 ⋅Nm ⎟⎞ ≤ f−1 shear stress, Δγe and Δγp = elastic and plastic strain range, Δτ = shear
f0 (24)
⎝ ⎠ stress range, σn,max = maximum normal stress, Δεne and Δεnp = elastic

943
M. Kamal, M.M. Rahman
and plastic normal strain range, Δσn = normal stress range, Δεgen* =
maximum generalized strain amplitude.
Carpinteri-Spagnoli [69] proposed a multiaxial high-cycle fatigue
criterion based on the critical plane approach. According to this, the
orientation of the critical plane is linked to both the averaged directions
of the principal stress axes and the fatigue properties of the material,
i.e., incorporated through a rotational angle. Then the multiaxial fa-
tigue strength estimation is performed by computing an equivalent
stress amplitude on the critical plane. A modification has also been
proposed where expressions depend on the ratio between the fatigue
limit under fully reversed shear stress and also under fully reversed
normal stress, which can then be employed for metals ranging from
mild to tough fatigue behavior.
The tension-torsion histories combine the shear and normal stress
amplitudes applied on the specimen cross section, using a stress scale
factor (SSF) polynomial function that depends on the Stress Amplitude
Ratio (SAR) between shear and normal components, as proposed by
Anes et al. [70]. The constant amplitude loading [71] and under vari-
able amplitude loading [72] successfully applied to multiaxial fatigue
life predictions. This approach is an alternative to multiaxial fatigue life
prediction by critical plane-based criteria, employed for both constant
[73,74] and variable amplitude loadings [75]. The critical plane ap-
proach calculates the damage on the plane, where the damage is
maximized while adopting SSF value that is assumed constant for a
given material. Sometimes varying with the fatigue life however not
with the SAR, or the stress amplitude level, or the loading path shape
[76].
The maximum principle stress in conjunction with Goodman model
is more suitable for brittle materials. The non-zero mean stress is
commonly used to predict the fatigue life of structure under combined
thermal-acoustic loadings [77], as per uniaxial fatigue model. However,
the metal structure under this condition is subjected to multiaxial fa-
tigue problem. Thus, the cumulative damage becomes difficult to as-
sess, and the uniaxial fatigue model leads to significant errors [78]. The
critical plane model gives a physical interpretation of the multiaxial
fatigue damage process. Thus, the critical plane model with the influ-
ence of mean stress should be used to predict the fatigue life. In this
regard, a novel critical plane model is proposed by Ge et al. [79]. The
fatigue life of metallic structures under combined thermal-acoustic
loadings is predicted based on the critical plane model, a new critical
plane model which relies on shear strain and is proposed to consider the
effect of mean stresses induced by temperature loading. The proposed
model is validated with experimental data from literature in this area
and by testing with four metal materials under various strain paths with
zero/non-zero mean stress. The results agreed well with the values
derived from experiments. Furthermore, the proposed model is applied
to predict the fatigue life of metal structures under combined thermal-
acoustic loadings and compared with the uniaxial Goodman model. The
comparison indicates that the proposed model is conservative, and the
thermal loading can significantly reduce the fatigue life.
3.2.2. Enclosed surface models
Mamiya and Goncalves [80] proposed a multiaxial high cycle fa-
tigue endurance criterion for sinusoidal iso-frequency in-phase and out-
of-phase loading conditions based on the idea of the minimum cir-
cumscribe ellipse in Ilyushin's deviatoric space, as expressed in Eq. (33).
The mean stress effect is conservative on fatigue life results compared to
other models used in the study. One limitation is that the model ap-
plication is restricted to iso-frequency loading cases.
5
∑ ai2 + kσp max ≤ λ
i=1 (33)
where ai(1–5) = amplitude of deviatoric stress, k = fatigue limit ratio,
σp,max = maximum principal stress, λ = material parameter.
Leila and Emmanuel [81] proposed the enclosing path method with
944
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
prismatic hulls on the principal axes directions, as expressed in Eq. (34).
This technique is effective in the sense that it is easy to capture the
equivalent stress parameter with different loads causing the same value
of stress parameter.
J2, a = R12 + R22 + R32 + R42 + R52 (34)
where, R1–R5 = amplitudes of stresses in principal directions in 5D
Euclidean space, J2, a = amplitude of the second invariant of stress
deviator.
Araujo and Dantas [82] proposed the maximum rectangular hull
method to define the equivalent shear stress amplitude, as expressed in
Eqs. (35) and (36), where a rectangle is fitted on a complex loading
path by maximizing its size. The method is capable of distinguishing
between proportional and non-proportional loading. The method pre-
dicted better results than the minimum rectangular hull method when
compared with experimental data.
τamax = max{τa (ϕ, θ)} (35)
where
τa = max a12 (φ) + a22 (φ) (36)
whereτamax = maximum equivalent shear stress amplitude, τa (ϕ, θ) =
equivalent shear stress on plane located by ϕ and θ, φ = angle locating
rectangular hull on plane.
Mamiya and Castro [83] proposed a fatigue life estimation model
regarding a piecewise ruled S-N surface. The first surface is defined as
the sum of deviatoric stress amplitude (τa) and maximum hydrostatic
stress (σHmax) as the exponential function of fatigue life as expressed in
Eq. (37). The second surface only the deviatoric stress amplitude is used
(Eq. (38) for cases where hydrostatic stresses have a small magnitude.
The maximum prismatic hull method is used to calculate the deviatoric
stress amplitude. The proposed model performed in the same way as the
other earlier proposed model included in the study, but better in the
case of mean normal stress.
β/δ
τ
τa + σH max = αN fβ if σH max ≥ α ⎜⎛ a ⎟⎞ − τa
⎝γ⎠ (37)
τa = γN fδ (38)
where Nf = fatigue life, α, β, γ, δ = material parameters.
Meggiolaro and de Castro [29,84] proposed an enclosing surface
model with a modified Wang–Brown rainflow counting method. They
identified discrepancies in the previously defined enclosing surface
methods. The minimum ball method has a physical foundation. The
considered portion of the path should have more than one cycle or else
when it is considered as a single cycle. The actual damage might be
underestimated, with a loss of information, as the enclosing surface
algorithms do not take into consideration the actual loading path, but
only the convex hulls associated with them. For loading histories with
more than one cycle counted by the rainflow method before applying
the moment of inertia method, in the second part they have shown the
shortcomings of the original Wang–Brown model, which can lead to
non-conservative predictions by incorrectly filtering out significant
events within a multiaxial loading cycle. The two improvements they
proposed are, first, related to choosing the starting point of the cycle so
that counting is modified, and secondly, that algorithm implementation
is simplified by formulating 5D Euclidean stress space in the form of the
deviatoric stress tensor. Marsh et al. [85] reviewed the residual data
point post-processing technique that processes the cycle count of the
data point left after the rainflow count of a random signal. It was
concluded that the most significance was found when the load process
resulted in a slowly varying mean stress was not fully accounted for by
the traditional rainflow counting methods.
M. Kamal, M.M. Rahman Renewable and Sustainable Energy Reviews 82 (2018) 940–949

3.2.3. Integral type models
Papadopoulos [86] proposed an integral type fatigue life criterion
where all components are integrated over all planes at the point of
consideration, with an additional integration of resolved shear stress
over the shear plane as expressed in Eqs. (39) and (40). It is placed a
high demand on computation time, thus preventing its use in com-
mercial fatigue solvers [22]. This model is best suited for hard metals
[87].
Papuga and Ruzicka [63] proposed two criteria with a similar da-
mage parameter with the emphasis on the effects of shear stress in
comparison to normal stress. The criteria integrate the fatigue para-
meter over all planes, as expressed in Eqs. (44) and (45). The two
methods derived here are to study whether integrating the load effect or
maximizing it is the more efficient approach. Both methods provide
very similar results as regards the presented data set of 119 experi-
mental results.

Ta2 + ασh, max ≤ β (39) 1 f

4π
∫φ ∫ψ ⎡⎢a1⋅Ca2 + bI ⋅⎛Na + t−−11 Nm ⎞ ⎤⎥ sin ψ dψ dφ−1 ≤ f
⎜ ⎟

⎣ ⎝ ⎠⎦ (44)
1 2π π 2π
Ta2 = 5
8π 2
∫φ=0 ∫θ=0 ∫χ =0 (Ta (φ, θ, χ ))2dχ sin θdθdφ (40) where

where Ta = resolved shear stress amplitude, φ and θ = angle locating 5 2

aI = k , bI = f−1 (3 − k 2)
2 (45)
the plane, χ = angle between the major axis and resolved shear stress,
α and β = material constant. where k, a1, b1 = material parameter, Ca = shear stress on considered
Lasserre and Palin-Luc [88] proposed an energy density model ex- plane, Na = maximum normal stress, Nm = mean normal stress, t−1 =
pressed in Eq. (41), based on the idea of the volumetric distribution of fatigue limit in fully reversed torsion, f0 = fatigue limit in repeated
strain energy density around the critical point considered for fatigue axial loading, f−1 = fatigue limit in fully reversed axial loading, ψ and
failure. The model predicted results with good agreement with uniaxial φ = Euler angles of planes examined in local coordinate system.
and multiaxial experimental data of smooth cylindrical specimens.
1 3.2.4. Material structure-based models
ωaD (Ci, load ) =
V * (Ci )
∫ ∫ ∫V * (C ) [Wa (x, y, z, load)
i Luo and Chattopadhyay [92] proposed a multi-scale damage cri-
terion for initial stage crack prediction, as expressed in Eqs. (46) and
− Wa* (Ci, load )] dυ (41)
(47). The local damage state is derived using optimization theory and
where ωa = volumetric mean value of strain energy volumetric den-
D then passed on to grain level. The damage for a meso representative
sity, V*(Ci) = volume at a critical point, Wa = strain volumetric energy volume element (RVE), which contains several grains, is determined.
density, Wa* = strain energy volumetric density at critical point Ci. The estimation results of RVE failure at a structural hotspot match the
Later, Palin-Luc and Banvillet [89] have improved the volumetric experimental results. The damage criterion has the capability to provide
density model by modifying the damaged part of the model to over- the potential directions for crack growth. This model can be suitable for
come the limitation of fully reversed sinusoidal loading in the [88] application in case studies, but due to its complex application method
model. Saintier and Palin-luc [90] reformulated the model, as expressed and dependence on grain structure, it may not be applicable in a design
in Eq. (42), and suggested improvements in the criterion for multiaxial process where the average performance and behavior is required to
variable amplitude loading, where previously only constant amplitude represent the life of a component.
loadings are considered. The model also improved the incremental fa- m′
(α )
tigue life assessment method for proportional and non-proportional dD (α ) =
σmr
−1 ⎛1 + σn ⎞ dY (α′)
⎜ ⎟
multiaxial variable amplitude loadings. The proposed model estimated σo ⎝ σf ⎠ (46)
fatigue life results with good accuracy within the considered experi-
mental cases, and it was also suggested that a more thorough study with 1 − δ′ (α′)
dY (α′) = δ′σn(α′) (dε p)(α′) + σs (dγ p)(α′)
a variety of materials is required. 2 (47)

6 where dD = damage parameter increment on slip system (α), dY(α)

(α)

Wgeqdam (M ) = Wgeq (M ) − ∑ Pl (M ) αM , l W g* = plastic strains energy increment on slip system (α), σmr = memory
l=1 (42) stress, σo = endurance limit, σn(α) = normal stress on slip system (α), σf
= true fracture stress, m, δ = material constants, dεp = plastic strain
where Wgeqdam = damaging part of strain work density, Wgeq =
increment corresponding to σn(α), σs(α) = shear stress on slip system (α),
equivalent strain work density, Pl = proportionality factor, αM= re-
dγp = plastic strain increment corresponding to σs(α).
presents the evolution of stress/strain tensors and sequence duration,
W*g = minimum strain work volumetric density to create irreversible
damage. 3.2.5. Stress invariants-based models
Zenner and Simburger [91] proposed a fatigue life estimation cri- Horstemeyer and Gokhale [93] proposed a void crack nucleation
terion based on an integral type approach and shear stress intensity model for ductile metals with second phases, as expressed in Eq. (48).
expressed in Eq. (43), which can result in good estimation in the case of The model is a function of the fracture toughness of the aggregate
complex periodical loadings. However, due to stress quantities, this is material, length scale parameter, the volume fraction of the second
not valid for a low-cycle fatigue regime, as strain-based parameters are phase, strain level and stresses state. Later Lugo and Jordon [94], using
required to deal with plastic deformations [91]. an acoustic emission method, showed the effectiveness of the proposed
model.
σequ, a
⎛ ε (t ) d1/2 ⎧ ,⎡ 4 J2 J3 I1 ⎫ ⎞
15 1/2
− 33 ⎤ + b, 3/2 + c,
= { 8π
π 2π
∫γ=0 ∫φ=0 [aτγφa
2 2
(1 + mτγφm 2
) + bσγφa (1 + nσγφm)]sin γdγdφ }
η (t ) = Ccoeff exp ⎜
⎝
KIC f 1/3
⎩
a
⎢
⎨ ⎣ 27 J2 ⎥
⎦ J2 J2 ⎬ ⎟
⎭⎠ (48)
(43) where η(t) = void nucleation at time t, Ccoeff = material constant, ε(t)
where σequ,a = equivalent stress amplitude, τγφm, σγφm = static shear = strain at time t, d = length scale parameter, KIC = critical stress
and normal stress on plane γφ, τγφa, σγφa = alternating shear and intensity factor, f = equivalent stress amplitude, a′, b′ and c′ = ma-
normal stress on plane γφ, a, m, b, n = obtained from the tensile- terial constants determined from different stress states, I1 = stress in-
compressive fatigue strength, torsional fatigue strength, pulsating ten- variant, J2, J3 = deviatoric stress invariants.
sile strength, and pulsating torsional strength, respectively. Vu and Halm [95] proposed a fatigue life criterion based on stress

945
M. Kamal, M.M. Rahman
invariants, and they introduced a quantity J2,mean, shown in Eqs.
(49)–(51), which captures the shear stress effect and phase shift effect.
The model performed well within the considered set of experimental
data. It suggested that a more detailed study is required to validate the
model's reliability.
f= γ1 J2′ (t )2 + γ2 J2,2 mean + γ3 If (I1, a, I1, m) ≤ β
where
for low strength metals
If (I1, a, I1, m) = I1, a + I1, m
for high strength metals
f−1
If (I1, a, I1, m) = I1, a + I1, m
t−1
where γ1, γ2, γ3, β = material parameters, J2 = second invariant of
'
deviator of the amplitude of stress tensor, J2,mean = mean value of J2'(t)
over a period, If = function of I1,a and I1,m, I1 = first stress invariant (a
– amplitude and m – mean), f−1 = fatigue limit in fully reversed axial
and bending, t−1 = fatigue limit in fully reversed torsion.
3.2.6. Statistical assessment models
Pinto and De Jesus [96] proposed a Weibull regression model based
on the Weibull model for statistical assessment of stress–life data for
probabilistic definition of the strain–life field. It provides an analytical
probabilistic definition of the whole strain–life field as quantile curves,
in both the low-cycle and high-cycle fatigue regions. The proposed
model deals directly with the total strain, without the need to separate
its elastic and plastic strain components, which represents a significant
advantage over the classical approaches, permitting it to deal with run-
outs, and it can be applied for probabilistic lifetime prediction using
damage accumulation.
3.2.7. Plasticity framework models
Emuakpor and George [97] developed a fatigue life assessment
expression for multiaxial loading, shown in Eq. (52), incorporating non-
linear plastic stress–strain relations in distortion theory for a case of
cyclic loading. The idea behind the criterion is that the physical damage
quantity for failure is equal to the accumulated strain energy in a
monotonic fracture, which is also equal to the strain energy during
fatigue failure.
3
σC ⎛
σE = ln ⎜ ∑ e 2σP / σC ⎞
⎟
2 ment with numerically predicted results.
⎝
p=1 ⎠ (52)
where σE = non-linear equivalent stress, σC = material parameter for
cyclic stress, σp = principal stress (p = 1, 2, 3).
Chaussumier and Mabru [98] proposed a fatigue life prediction
model based on experimental results including the multi-site crack
consideration, coalescence between neighboring cracks, a short crack
growth stage and a long crack propagation stage. This model is devel-
oped from experimental topography measurements of pickled surfaces
which make it possible to detect the pits, and to characterize their sizes.
The model was developed specifically for 7050 aluminum alloy, and its
application and effectiveness should be tested for a greater variety of
materials before considering it for widespread use. Khandelwal and El-
Tawil [99] proposed a damage mechanics-based model to simulate
ductile fracture in structural steels. The model expressed in Eq.
(53)–(55), is based on the concept of the principle of effective stress and
strain equivalence in a plasticity framework. The proposed model is
implemented in finite element code, the model parameters are mesh-
dependent, and the model has to be recalibrated if a different material
or mesh size is employed.
α̇ d = a0 T1 (α p) α̇ P + a1 Τ T1 (α P ) α̇ P + a2 exp(a3 α P ) T2 (α P ) α̇ P
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
P
⎧T1 (α , εn, Δε ) = 0 if α P < εn
⎪ (α P − εn)
2
2(α P − εn)
T1 (α P , εn, Δε ) = ⎡3 − ⎤ if εn < α p < εn + Δε
⎨ Δε 2 ⎣ Δε ⎦
⎪ and T (α P , ε , Δε ) = 1 if α P ≥ εn + Δε
⎩ 1 n (54)
P if α p < εc
⎧T2 (α , εc , Δε ) = 0
(49) ⎪ (α p − εc )2 2(α p − εc )
T2 (α p, εc , Δε ) = ⎡3 − ⎤ if εc < α p < εc + Δε
⎨ Δε 2 ⎣ Δε ⎦
⎪ and T2 (α , εc , Δε ) = 1
p if α p ≥ εc + Δε (55)
⎩
(50) Where α and α =damage and plastic internal variables, T1 and T2
d p
= threshold function, T = stress tri-axiality, a0, a1, a2, a3 = material
parameters, εn = nucleation strain, Δε = smoothing factor, εc = coa-
lescence strain
(51)
3.3. Application of optimization algorithm
Bukkapatnam and Sadananda [100] proposed a framework to
model crack growth dynamics in materials under alternative service
environments without resorting to extensive experimentation. The
model structures are derived based on the unified approach (including
all the observed effects of load ratio, tiny cracks, shielding of disloca-
tions, overload, and under-load, and surface crack), and a genetic al-
gorithm is used to parameterize the models. The GAs allow re-
conciliation of the known, complicated physical relationships with
empirical observations. The presence of physically motivated mathe-
matical structures renders these phenomena predictive and easier to
analyze than the use of purely empirical approaches such as neural
networks. A fatigue crack growth model derived from the application of
the proposed framework was found to predict crack growth rates to
within 12% error. Thus they offer a potential to reduce the extent of
experimentation needed for fatigue crack growth analysis.
Liu [23] proposed a simulation-based calculation procedure for
multiaxial fatigue life prediction, which combines a Monte Carlo si-
mulation technique with stochastic process theory and a response sur-
face method. The proposed method can include randomness in material
properties, applied loading and geometry. Time-dependent failure
probability is evaluated, where the failure is defined as occurring when
the accumulated damage exceeds an acceptable value or the crack ex-
ceeds a critical value. A Monte Carlo simulation method is used to
calculate the probabilistic life distribution. A response surface method
combined with the design of experiments is used to obtain a clear
empirical formula for the damage accumulation process, considering
several sources of variation. Field failure data was in excellent agree-
Vassilopoulos and Georgopoulos [101] proposed a procedure using
an artificial neural network to model the fatigue life of multidirectional
composite laminates made of GFRP composite materials and tested
these under constant amplitude loading patterns. The prediction accu-
racy of the artificial neural network (ANN) was validated using ex-
perimental data available from the literature. Only 50% of the experi-
mental data is required by the ANN to model the fatigue life of the
material in comparison to the data needed by the conventional method.
Vassilopoulos and Georgopoulos [102] presented a comparative study
of fatigue life estimation of composite materials using the GA tool, with
conventional methods. They concluded that the fatigue life results are
more accurate than the results produced by traditional methods. The
advantages of GA highlighted in this paper are that the modeling is not
based on any assumption, such as that the data follow a specific sta-
tistical distribution, or that the SN curve follows a power curve equa-
tion. It is a material-independent data-driven method which correlates
input with output to establish a model describing the relationship be-
tween them. Thus the model can be applied to any material, provided
that an adequate amount of data be available. For future research re-
lated to GA-based fatigue modeling, they proposed to use more complex
(53) genetic programming configurations by introducing models with
946
M. Kamal, M.M. Rahman
multiple inputs, e.g. stress amplitude, maximum stress, stress ratio, and
off-axis angle, and to try to assign a corresponding number of cycles to
failure to every set of input.
Franulovic and Basan [103] proposed a material model describing
the elasto-plastic behavior of materials under cyclic loading. The ma-
terial constitutive model for the description of low-cycle fatigue beha-
vior is highly non-linear, and therefore its parameter identification re-
quires a complex numerical procedure, such as a genetic algorithm. A
GA is a stochastic search method for obtaining good approximate so-
lutions to complex problems. The calculation resulted in the identifi-
cation of material parameters that are validated by comparing the
material response of the numerical solution with experimental data.
The GA for the parameter identification in a particular problem, using a
finite element method to simulate the materials’ response, proved to be
an excellent choice. The use of suitable genetic operators in the GA
calculation procedure made it possible to achieve very fast and reliable
convergence to accurate results.
Krishnapillai and Jones [104] proposed a structural optimization
procedure that integrates geometrical modeling, structural analysis,
and optimization into one complete and automated computer-aided
design process. They illustrated a procedure for the development of
lightweight structures using fatigue-based optimization in conjunction
with a genetic algorithm. It provides a robust methodology and also has
the potential to be applied to structures with complex structural con-
figurations with multiple optimum peaks. Brighenti and Carpinteri [11]
proposed a continuum mechanics based endurance function which
quantifies the damage accumulation in the material up to final failure,
under an arbitrary multiaxial loading history. Several parameters
characterize the proposed model, so a genetic algorithm is employed to
numerically evaluate their values once the effects of some experimental
complex stress history on the fatigue life are known. The GAs have some
advantages concerning classical techniques since they allow us to solve
problems characterized by both multiple minima and non-convexity
properties, avoiding numerical instabilities and the risk of missing the
global optimum. Furthermore, the GAs can handle any objective func-
tion and simply operate by using basic concepts such as generation of
random numbers, choice, switching and combinations of such gener-
ated numbers. No critical plane or loading cycle counting algorithm is
needed, as the proposed model simply analyzes damage accumulation
during the loading process. The fatigue life was assumed to be domi-
nated by crack-nucleation, i.e., the fatigue life for crack propagation is
negligible concerning the total life. They concluded that the fatigue life
results using the advanced endurance function show satisfactory
agreement with the experimental data. Kamal et al. [105] studied the
endurance function model and suggested simplification by reducing the
number of parameters needed to calibrate the model and also devel-
oped a methodology to use FE analysis stress results to estimate fatigue
life using the model.
Roux and Lorang [106] have proposed a method to define Equiva-
lent Fatigue Loads (EFL) from in-service load measurement in the case
of a structure subjected to multiple variable fatigue loadings. EFL can
be used to define tests on full-scale structures for an experimental va-
lidation approach. A mathematical method is proposed to find the EFL,
which uses a genetic algorithm to compute an accurate EFL for the
whole structure. This approach to finding the best EFL for a structure
can be used to perform a validation test on a bench and can also be used
during the design phase to optimize the structure geometry for a spe-
cific use. Lotfi and Beiss [107] showed the application of an ANN as an
effort to predict the effect of various powder metallurgy processing
parameters on the endurance limit of powder metallurgy steel speci-
mens. The study applies the existing data which was collected from
published experimental investigations. Fabrication and testing para-
meters together with corresponding fatigue limit records were used as
sets of data for network training. A genetic algorithm was also included
to optimize the experimental conditions, subject to practical limita-
tions, to achieve the desired fatigue strength values. The GA with a
947
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
trained neural network is found to be a powerful optimization and cost
reduction tool as it makes it possible to select optimum material com-
positions and processing conditions for a specific fatigue strength.
Niesłony and Böhm [108] presented a stress-based approach to take
into account the influence of the mean stress value on the fatigue
strength of constructional materials. The proposed fatigue life estima-
tion model implemented a locally developed optimization algorithm to
ensure maximum precision in the fatigue life calculations. The key as-
sumption in the proposed solution is the use of fatigue strength am-
plitudes for calibration, gained for two boundary states: tension and
compression with the stress ratio R = 1. Another one with a significant
mean stress value, e.g. the popular unilateral tension R = 0, which is
also as the main limitation. Klemenc and Fajdiga [109,110] proposed a
technique for estimating ε-N curves and their scatter. They introduced
five parameters: four parameters of the Coffin–Manson equation for the
scale parameter of the Weibull distribution and the shape parameter of
the Weibull distribution. Then they used GA and the differential ant-
stigmergy algorithm [111] to estimate the five parameters by the
known fatigue–life data (mainly median EN curves) to obtain not only
the trend of the ε-N curve but also its scatter. Both algorithms were
found to be capable of estimating the five parameters correctly with as
small a set as 25 data points of experimental data as input. GA was
found to be faster than DASA.
4. Conclusions
In this review, an attempt is made to collect information about the
development of fatigue life estimation models, where the author has
tried to include as much data as possible concerning this objective. The
main purpose of this review is to collect the different ideas put forward
to model fatigue behavior, i.e., using various combinations of stress and
strain quantities, mean stress effects, critical planes, microstresses,
virtual strain energy (VSE), enclosing surface methods, 5D Euclidean
space, optimization algorithms, and so forth. New concepts developed
to model the fatigue behavior of materials are reviewed, and it is ob-
served that the proposed models are still limited to certain conditions,
i.e., high-cycle or low-cycle fatigue, uniaxial and multiaxial loading,
constant and variable amplitude loading and material type, such as
brittle, hard and ductile materials. However, no model can yet be de-
clared as a universal or generalized model for various loading and
material conditions, so additional research and development work is
still needed for accurate and reliable multiaxial fatigue life estimation.
The main objective of this review is to explore the new concepts and
methods to estimate fatigue life, unlike the classical ones which do not
lead to any universal or generalized fatigue models. The most promising
methods for fatigue life estimation from the recently proposed models
involve the application of evolutionary optimization algorithms like
GA, ANN, DASA, and so forth. The most attractive feature of these types
of models is the inherent flexibility in model training/calibration due to
the application of evolutionary algorithms. It can lead to the more
flexible formulation of model expressions with a higher number of
parameters. In future research, the concept of using optimization al-
gorithms should be explored in detail to utilize the full potential of this
simple, versatile and easy-to-implement technique for fatigue life esti-
mation modeling.
References
[1] Stephens RI, Fatemi A, Stephens RR, Fuchs HO. Metal fatigue in engineering. New
York: John Wiley and Sons, Inc; 2000.
[2] Sangid MD. The physics of fatigue crack initiation. Int J Fatigue 2013;57:58–72.
[3] Lee Y, Pan J, Hathaway R, Barkey M. Fatigue testing and analysis: theory and
practice. New York: Elsevier Butterworth Heinemann; 2005.
[4] Manson SS, Halford GR. Fatigue and durability of structural materials. Materials
Park, Ohio: ASM International, Materials; 2006.
[5] Abdullah S, Al-Asady NA, Ariffin AK, Rahman MM. A review on finite element
analysis approaches in durability assessment of automotive components. J Appl Sci
2008;8:2192–201.
M. Kamal, M.M. Rahman
[6] Susmel L, Taylor D. A critical distance/plane method to estimate finite life of
notched components under variable amplitude uniaxial/multiaxial fatigue loading.
Int J Fatigue 2012;38:7–24.
[7] Rahman MM, Ariffin AK, Jamaludin N, Haron CHC. Influence of surface treatments
on fatigue life of a free piston linear generator engine component using random
loading. J Zhejiang Univ Sci Part A 2006;7:1819–30.
[8] Rahman MM, Ariffin AK, Jamaludin N, Haron CH. Finite element based durability
assessment of a free piston linear engine component. Struct Durab Health Monit
2007;3:1–13.
[9] Rahman MM, Kadirgama K, Noor MM, Rejab MRM, Kesulai SA. Fatigue life pre-
diction of lower suspension arm using strain-life approach. Eur J Sci Res
2009;30:437–50.
[10] Rahman MM, Ariffin AK, Rejab MRM, Kadirgama K, Noor MM. Multiaxial fatigue
behaviour of cylinder head for a free piston linear engine. J Appl Sci
2009;9:2725–34.
[11] Brighenti R, Carpinteri A. A notch multiaxial fatigue approach based on damage
mechanics. Int J Fatigue 2012;39:122–33.
[12] Susmel L, Taylor D. The modified Wohler curve method applied along with the
theory of critical distances to estimate finite life of notched components subjected
to complex multiaxial loading paths. Fatigue Fract Eng Mater Struct
2008;31:1047–64.
[13] Macha E, Niesłony A. Critical plane fatigue life models of materials and structures
under multiaxial stationary random loading: the state-of-the-art in Opole Research
Centre CESTI and directions of future activities. Int J Fatigue 2012;39:95–102.
[14] Atzori B, Berto F, Lazzarin P, Quaresimin M. Multi-axial fatigue behaviour of a
severely notched carbon steel. Int J Fatigue 2006;28:485–93.
[15] Susmel L, Tovo R. Estimating fatigue damage under variable amplitude multiaxial
fatigue loading. Fatigue Fract Eng Mater Struct 2011;34:1053–77.
[16] Chiu Y-W, Jang J-Y. 3D numerical and experimental analysis for thermal-hydraulic
characteristics of air flow inside a circular tube with different tube inserts. Appl
Therm Eng 2009;29:250–8.
[17] Bishop NWM, Sherratt F. Finite element based fatigue calculations. Netherlands:
NAFEMS Ltd.; 2000.
[18] Rahman MM, Ariffin AK, Abdullah S, Noor MM, Bakar RA, Maleque MA. Finite
element based fatigue life prediction of cylinder head for two-stroke linear engine
using stress-life approach. J Appl Sci 2008;8:3316–27.
[19] Rahman MM, Ariffin AK, Jamaludin N, Abdullah S, Noor MM. Finite element based
fatigue life prediction of a new free piston engine mounting. J Appl Sci
2008;8:1612–21.
[20] Rahman MM, Ariffin AK, Abdullah S, Noor MM, Bakar RA. Durability assessment
of cylinder block for two stroke free piston linear engine using random loading.
Am J Appl Sci 2009;6:726–35.
[21] Scattina A, Alovisi M, Paolino DS, Pasqualini D, Scotti N, Chiandussi G, et al.
Prediction of cyclic fatigue life of nickel-titanium rotary files by virtual modeling
and finite elements analysis. J Endodont 2015;41:1867–70.
[22] Papuga J. A survey on evaluating the fatigue limit under multiaxial loading. Int J
Fatigue 2011;33:153–65.
[23] Liu Y. Stochastic modeling of multiaxial fatigue and fracture [Ph.D. Thesis]. USA:
Vanderbilt University; 2006.
[24] Ewing JA, Humfrey JC. The fracture of metals under repeated alterations of stress.
Philos Trans R Soc 1903;A200:241–50.
[25] Schijve J. Fatigue of structures and materials in the 20th century and the state of
the art. Int J Fatigue 2003;25:679–702.
[26] Chapetti MD, Katsura N, Tagawa T, Miyata T. Static strengthening and fatigue
blunt-notch sensitivity in low carbon steels. Int J Fatigue 2001;23:207–14.
[27] Mayer H. Fatigue damage of low amplitude cycles in low carbon steel. J Mater Sci
2009;44:4919–29.
[28] Fatemi A, Shamsaei N. Multiaxial fatigue: an overview and some approximation
models for life estimation. Int J Fatigue 2011;33:948–58.
[29] Meggiolaro MA, de Castro JTP. An improved multiaxial rainflow algorithm for
non-proportional stress or strain histories – Part I: enclosing surface methods. Int J
Fatigue 2012;42:217–26.
[30] Ghalami TZ, Fatemi A. Multiaxial fatigue and life prediction of elastomeric com-
ponents. Int J Fatigue 2013;55:90–101.
[31] Sines G. Failure of materials under combined repeated stresses with superimposed
static stress. National Advisory Committee for Aeronautics; 1995. Technical Note
3495, p. 69.
[32] Sines G. Behavior of metals under complex static and alternating stresses. In: Sines
G, Waisman JL, editors. Metal fatigue. New York: McGraw-Hill; 1959.
[33] Crossland B. Effect of large hydrostatic pressure on the torsional fatigue strength of
an alloy steel. In: Proceedings of the international conference on fatigue of metals.
London: Institution of Mechanical Engineers; 1956, p. 138–49.
[34] Findley WN. A theory for the effect of mean stress on fatigue of metals under
combined torsion and axial load or bending. J Eng Ind 1959:301–6.
[35] McDiarmid DL. A general criterion for high cycle multiaxial fatigue failure. Fatigue
Fract Eng Mater Struct 1991;14:429–53.
[36] McDiarmid DL. A shear stress based critical-plane criterion of multiaxial fatigue
failure for design and life prediction. Fatigue Fract Eng Mater Struct
1994;7:217–26.
[37] Van KD. Macro-micro approach in high-cycle multiaxial fatigue. In: McDowell DL,
Ellis R, editors. Advances in multiaxial fatigue. West Conshohocken, PA: American
Society for Testing Materials; 1993. p. 120–30.
[38] Hofmann F, Bertolino G, Constantinescu A, Ferjani M. Numerical exploration of
the dang van high cycle fatigue criterion: application to gradient effects. J Mech
Mater Struct 2009;4:293–308.
[39] Charkaluk E, Constantinescu A, Maïtournam H, Van KD. Revisiting the Dang Van
948
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
criterion. Proc Eng Mesomech 2009;1:143–6.
[40] Brown M, Miller KJ. Theory for fatigue under multiaxial stress–strain conditions.
Proc Inst Mech Eng 1973:745–56.
[41] Brown MW, Miller KJ. Two decades of progress in the assessment of multiaxial
low-cycle fatigue life. In: Amzallag C, Leis BN, Rabbe P, editors. Low-cycle fatigue
and life prediction. West Conshohocken, PA: American Society for Testing and
Materials; 1982. p. 482–99.
[42] Kandil FA, Brown MW, Miller KJ. Biaxial low-cycle fatigue fracture of 316 stainless
steel at elevated temperatures. London: The Metals Society; 1982. p. 203–10.
[43] Wang CH, Brown MW. A path independent parameter for fatigue under propor-
tional and non proportional loading. Fatigue Fract Eng Mater Struct
1993;16:1285–98.
[44] Smith KN, Watson P, Topper TH. A Stress-Strain function for the fatigue of metals.
JMater 1970;5:767–78.
[45] Fatemi A, Socie DF. A critical plane approach to multiaxial fatigue damage in-
cluding out of phase loading. Fatgiue Fract Eng Mater Struct 1988;11:149–66.
[46] Fatemi A, Gladskyi M. Notched fatigue behavior including load sequence effects
under axial and torsional loadings. Int J Fatigue 2013;55:43–53.
[47] Liu KC. A method based on virtual strain energy parameters for multiaxial fatigue
life prediction. In: McDowell DL, Ellis R, editors. Advances in multiaxial fatigue.
West Conshohocken, PA: American Society for Testing Materials; 1993. p. 67–84.
[48] Liu KC, Wang JA. An energy method for predicting fatigue life, crack orientation,
and crack growth under multiaxial loading conditions. Int J Fatigue
2001;23:S129–34.
[49] Núñez JL, Calvo S, Canales M, Gómez C, Valdés JR. Probabilistic formulation of
the multiaxial fatigue damage of Liu. Int J Fatigue 2011;33:460–5.
[50] Chu CC. Fatigue damage calculation using the critical plane approach. J Eng Mater
Technol 1995;117:41–9.
[51] Glinka G, Wang G, Plumtree A. Mean stress effects in multiaxial fatigue. Fatigue
Fract Eng Mater Struct 1995;18:755–64.
[52] Susmel L, Lazzarin P. A bi-parametric modified Wohler curve for high cycle
multiaxial fatigue assessment. Fatigue Fract Eng Mater Struct 2002;25:63–78.
[53] Lazzarin P, Susmel L. A stress-based method to predict lifetime under multiaxial
fatigue loadings. Fatigue Fract Eng Mater Struct 2003;26:1171–87.
[54] Susmel L, Taylor D. A simplified approach to apply the theory of critical distances
to notched components under torsional fatigue loading. Int J Fatigue
2006;28:417–30.
[55] Taylor D. Geometrical effects in fatigue: a unifying theoretical model. Int J Fatigue
1999;21:413–20.
[56] Susmel L. A simple and efficient numerical algorithm to determine the orientation
of the critical plane in multiaxial fatigue problems. Int J Fatigue 2010;32:1875–83.
[57] Susmel L, Taylor D. The theory of critical distances to estimate lifetime of notched
components subjected to variable amplitude uniaxial fatigue loading. Int J Fatigue
2011;33:900–11.
[58] Susmel L, Tovo R, Socie DF. Estimating the orientation of stage I crack paths
through the direction of maximum variance of the resolved shear stress. Int J
Fatigue 2014;58:94–101.
[59] Susmel L, Louks R, Gerin B, Draper J, Askes H. On the multiaxial fatigue assess-
ment of complex three-dimensional stress concentrators. Int J Fatigue
2014;63:12–24.
[60] Mahadevan S, Liu Y. Multiaxial high-cycle fatigue criterion and life prediction for
metals. Int J Fatigue 2005;27:790–800.
[61] Mahadevan S, Liu Y. A unified multiaxial fatigue damage model for isotropic and
anisotropic materials. Int J Fatigue 2007;29:347–59.
[62] Ninic D, Stark H. A multiaxial fatigue damage function. Int J Fatigue
2007;29:533–48.
[63] Papuga J, Ruzicka M. Two new multiaxial criteria for high cycle fatigue compu-
tation. Int J Fatigue 2008;30:58–66.
[64] Lu Z, Liu Y. Crack growth-based multiaxial fatigue life prediction. In: Proceedings
of the 12th international conference on fracture (ICF-12). Ottawa, Canada; 2009.
p. 1–10.
[65] Shang D-G, Sun G-Q, Bao M. Multiaxial fatigue damage parameter and life pre-
diction under low cycle loading for GH4169 alloy and other structural materials.
Int J Fatigue 2010;32:1108–15.
[66] Li J, Sun Q, Zhang ZP, Li CW, Zhang DW. A new multiaxial fatigue life prediction
model under proportional and nonproportional loading. J Eng Mater Technol
2010;132:8.
[67] Li J, Zhang ZP, Sun Q, Li CW. Multiaxial fatigue life prediction for various metallic
materials based on the critical plane approach. Int J Fatigue 2011;33:90–101.
[68] Ince A, Glinka G. A generalized fatigue damage parameter for multiaxial fatigue
life prediction under proportional and non-proportional loadings. Int J Fatigue
2014;62:34–41.
[69] Ronchei C, Carpinteri A, Fortese G, Spagnoli A, Vantadori S, Kurek M, et al. Life
estimation by varying the critical plane orientation in the modified Carpinteri-
Spagnoli criterion. Frat ed Integr Strutt 2015;34:74–9.
[70] Anes V, Reis L, Li B, Fonte M, De Freitas M. New approach for analysis of complex
multiaxial loading paths. Int J Fatigue 2014;62:21–33.
[71] Anes V, Reis L, Li B, De Freitas M. New cycle counting method for multiaxial
fatigue. Int J Fatigue 2014;67:78–94.
[72] Anes V, Reis L, Li B, Freitas M, Sonsino C. Minimum Circumscribed Ellipse ((MCE))
and Stress Scale Factor ((SSF)) criteria for multiaxial fatigue life assessment. Theor
Appl Fract Mech 2014;73:109–19.
[73] Lopez-Crespo P, Moreno B, Lopez-Moreno A, Zapatero J. Study of crack orientation
and fatigue life prediction in biaxial fatigue with critical plane models. Eng Fract
Mech 2015;136:115–30.
[74] Mohammadi M, Zehsaz M, Hassanifard S, Rahmatfam A. An evaluation of total
M. Kamal, M.M. Rahman
fatigue life prediction of a notched shaft subjected to cyclic bending load. Eng
Fract Mech 2016;166:128–38.
[75] Carpinteri A, Spagnoli A, Ronchei C, Scorza D, Vantadori S. Critical plane criterion
for fatigue life calculation: time and frequency domain formulations. Proc Eng
2015;101:518–23.
[76] de Freitas M, Reis L, Meggiolaro MA, de Castro JTP. Stress scale factor and critical
plane models under multiaxial proportional loading histories. Eng Fract Mech
2017;174:104–16.
[77] Liu L, Guo Q, He T. Thermal–acoustic fatigue of a multilayer thermal protection
system in combined extreme environments. Adv Mech Eng 2014;2014:18.
[78] Ge J, Sun Y, Zhou S, Zhang L, Zhang Y, Zhang Q. A hybrid frequency–time domain
method for predicting multiaxial fatigue life of 7075‐T6 aluminium alloy under
random loading. Fatigue Fract Eng Mater Struct 2015;38:247–56.
[79] Ge J, Sun Y, Xu J, Yang Z, Liang J. Fatigue life prediction of metal structures
subjected to combined thermal-acoustic loadings using a new critical plane model.
Int J Fatigue 2017;96:89–101.
[80] Mamiya EN, Goncalves CA, Araujo JA. Multiaxial fatigue: a stress based criterion
for hard metals. Int J Fatigue 2005;27:177–87.
[81] Leila K, Emmanuel P, Sylvain L. A measure of the equivalent shear stress amplitude
from a prismatic hull in the principal coordinate system. Int J Fatigue
2010;32:1977–84.
[82] Araujo JA, Dantas AP, Castro FC, Mamiya EN, Ferreira JLA. On the characteriza-
tion of the critical plane with a simple and fast alternative measure of the shear
stress amplitude in multiaxial fatigue. Int J Fatigue 2011;33:1092–100.
[83] Mamiya EN, Castro FC, Algarte RD, Araújo JA. Multiaxial fatigue life estimation
based on a piecewise ruled S–N surface. Int J Fatigue 2011;33:529–40.
[84] Meggiolaro MA, de Castro JTP. An improved multiaxial rainflow algorithm for
non-proportional stress or strain histories – Part II: the modified Wang–Brown
method. Int J Fatigue 2012;42:194–206.
[85] Marsh G, Wignall C, Thies PR, Barltrop N, Incecik A, Venugopal V, et al. Review
and application of rainflow residue processing techniques for accurate fatigue
damage estimation. Int J Fatigue 2016;82:757–65.
[86] Papadopoulos IV. A new criterion of fatigue strength for out-of phase bending and
torsion of hard metals. Int J Fatigue 1994;16:377–84.
[87] Papadopoulos IV, Davoli P, Gorla C, Filippini M, Bernasconi A. A comparative
study of multiaxial high-cycle fatigue criteria for metals. Int J Fatigue
1997;19:219–35.
[88] Lasserre S, Palin-Luc T. An energy based criterion for high cycle multiaxial fatigue.
Eur J Mech – A/Solids 1998;17:237–51.
[89] Palin-Luc T, Banvillet A, Lasserre S. A volumetric energy based high cycle multi-
axial fatigue criterion. Int J Fatigue 2003;25:755–69.
[90] Saintier N, Palin-luc T, Bénabes J, Cocheteux F. Non-local energy based fatigue life
calculation method under multiaxial variable amplitude loadings. Int J Fatigue
2013;54:68–83.
[91] Zenner H, Simburger A, Liu J. On the fatigue limit of ductile metals under complex
multiaxial loading. Int J Fatigue 2000;22:137–45.
[92] Luo C, Chattopadhyay A. Prediction of fatigue crack initial stage based on a
multiscale damage criterion. Int J Fatigue 2011;33:403–13.
949
Renewable and Sustainable Energy Reviews 82 (2018) 940–949
[93] Horstemeyer MF, Gokhale AM. A void_crack nucleation model for ductile metals.
Int J Solids Struct 1999;36:5029–55.
[94] Lugo M, Jordon JB, Horstemeyer MF, Tschopp MA, Harris J, Gokhale AM.
Quantification of damage evolution in a 7075 aluminum alloy using an acoustic
emission technique. Mater Sci Eng: A 2011;528:6708–14.
[95] Vu QH, Halm D, Nadot Y. Multiaxial fatigue criterion for complex loading based on
stress invariants. Int J Fatigue 2010;32:1004–14.
[96] Pinto Hn, De Jesus AlMP, Fernández-Canteli A, Castillo E, Pereira HlFSG. Analysis
of constant and variable amplitude strain-life data using a novel probabilistic
Weibull regression model. J Press Vessel Technol 2010;132:061401.
[97] Emuakpor OS, George T, Cross C, Wertz J, Shen MHH. A new distortion energy-
based equivalent stress for multiaxial fatigue life prediction. Int J Non-Linear Mech
2012;47:29–37.
[98] Chaussumier M, Mabru M, Shahzad M, Chieragatti R, Rezai-Aria F. A predictive
fatigue life model for anodized 7050 aluminium alloy. Int J Fatigue
2013;48:205–13.
[99] Khandelwal K, El-Tawil S. A finite strain continuum damage model for simulating
ductile fracture in steels. Eng Fract Mech 2014;116:172–89.
[100] Bukkapatnam S, Sadananda K. A genetic algorithm for unified approach-based
predictive modeling of fatigue crack growth. Int J Fatigue 2005;27:1354–9.
[101] Vassilopoulos A, Georgopoulos E, Dionysopoulos V. Artificial neural networks in
spectrum fatigue life prediction of composite materials. Int J Fatigue
2007;29:20–9.
[102] Vassilopoulos A, Georgopoulos E, Keller T. Comparison of genetic programming
with conventional methods for fatigue life modeling of FRP composite materials.
Int J Fatigue 2008;30:1634–45.
[103] Franulovic M, Basan R, Prebil I. Genetic algorithm in material model parameters'
identification for low-cycle fatigue. Comput Mater Sci 2009;45:505–10.
[104] Krishnapillai K, Jones R. Three-dimensional structural design optimisation based
on fatigue implementing a genetic algorithm and a non-similitude crack growth
law. Finite ElemAnal Des 2009;45:132–46.
[105] Sundar LS, Farooky MH, Sarada SN, Singh MK. Experimental thermal conductivity
of ethylene glycol and water mixture based low volume concentration of Al2O3
and CuO nanofluids. ICHM – Int Commun Heat Mass Transf 2013;41:41–6.
[106] Roux C, Lorang X, Maitournam H, Nguyen-Tajan M-L, Quesson B. Multi-parameter
fatigue equivalence loadings for specification applications. Proc Eng
2013;66:393–402.
[107] Lotfi B, Beiss P. Application of neural networking for fatigue limit prediction of
powder metallurgy steel parts. Mater Des 2013;50:440–5.
[108] Niesłony A, Böhm M. Mean stress effect correction using constant stress ratio S–N
curves. Int J Fatigue 2013;52:49–56.
[109] Klemenc J, Fajdiga M. Estimating S–N curves and their scatter using a differential
ant-stigmergy algorithm. Int J Fatigue 2012;43:90–7.
[110] Klemenc J, Fajdiga M. Joint estimation of E–N curves and their scatter using
evolutionary algorithms. Int J Fatigue 2013;56:42–53.
[111] Sridhar G, Dasappa S, Sridhar HV, Paul PJ, Rajan NKS. Gaseous emissions using
producer gas as fuel in reciprocating engines. SAE Tech Pap 2005. http://dx.doi.
org/10.4271/2005-01-1732.