Engineering Structures 60 (2014) 214–222

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Ultra low-cycle fatigue behaviour of a structural steel

J.C.R. Pereira a,⇑, A.M.P. de Jesus a,b, J. Xavier c, A.A. Fernandes a,d
a
IDMEC, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
b
Universidade de Trás-os-Montes e Alto Douro, UTAD, Quinta de Prados, 5000-801 Vila Real, Portugal
c
Universidade de Trás-os-Montes e Alto Douro, UTAD, CITAB, Quinta de Prados, 5000-801 Vila Real, Portugal
d
Faculty of Engineering of University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: Steel structures subjected to extreme loading conditions (e.g. earthquakes, support settlements, indus-
Received 21 June 2013 trial plant shutdown) undergo large deformations leading to fracture, either due to monotonic loading
Revised 27 December 2013 or ultra-low-cycle fatigue (ULCF) (Nf < 100 cycles). Although developments have been made to under-
Accepted 30 December 2013
stand and to model monotonic ductile damage and low-cycle fatigue (LCF), so far ULCF is neither suffi-
Available online 24 January 2014
ciently investigated nor understood. This paper presents the results of an investigation concerning the
ULCF behaviour of the S185 structural steel. An experimental program was performed to derive ULCF data
Keywords:
for notched specimens. LCF and monotonic damage data was also derived for the material under inves-
Ultra low-cycle fatigue
Structural steel
tigation, since ULCF exhibits damage features from both cases. While LCF data was derived for smooth
Experimental techniques specimens, monotonic tensile tests coupled with image-based methods were carried out on both smooth
Finite element modelling and notched specimens. Nonlinear finite element models were used to compute the history of relevant
parameters of the investigated models for ULCF life prediction. Three existing alternative modelling
approaches for ULCF were assessed using available experimental data, and important remarks for further
enhancements proposed.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction
Extreme loads applied to steel structures can yield either to
monotonic ductile failure or to fatigue failure at very small number
of cycles (<100 cycles). This fatigue regime is called ultra-low-cycle
fatigue (UCLF) or extreme-low-cycle fatigue, in order to distinguish
it from low-cycle fatigue (LCF), since ULCF damage mechanisms are
distinctive of those typical from LCF. The ULCF fits between the
monotonic ductile damage and LCF damage, as shown in Fig. 1,
and exhibits damage features from both damaging processes. Con-
cerning the monotonic ductile damage, several models have been
proposed in literature, such as the Gurson–Tvergaard–Needleman
(GTN) model [2], based on porous plasticity [2], the Johnson–Cook
(JC) model [3], the Wilkins model [4], the Cockcroft–Latham model
[5] and the Xue–Wierzbicki model [6]. The application of these
models requires the definition of adequate experimental proce-
dures in order to calibrate them and allowing the identification
of the model constants. In these models, the plastic strain and
the stress triaxiality play a central role on damage kinematics,
but more recent approaches have also shown a dependency on
Lode angle parameter [7].
⇑ Corresponding author. Tel./fax: +351 259 350 356.
E-mail address: joao7dc@gmail.com (J.C.R. Pereira).
In contrast with monotonic ductile damage and LCF, ULCF mod-
els have been less developed. In addition, concerning ULCF testing,
the data available in literature is scarce and there are no specific
standards for mechanical testing under high strain level require-
ments. Typical smooth specimen geometries used in the LCF test-
ing exhibit instability under ULCF loading, requiring special
procedures to avoid the specimens buckling.
With respect to ULCF modelling, existing approaches reported
in the literature may be classed into coupled and uncoupled mod-
els. This classification is usual in monotonic ductile models. Cou-
pled models consider interdependency between plasticity and
damage and allows linear or non-linear damage evolution. The
coupled plasticity-damage models allow the simulation of the
crack initiation (damage onset) and crack propagation (damage
spread) [8]. An example of these formulations was proposed by
Lemaitre [9]. Also, Leblond et al. [10] proposed an extension of
the GTN model for cyclic loading, consisting in the introduction
of kinematic hardening in the porous plasticity model, leading to
the so called GTN-LPD model [11]. The coupled damage-plasticity
models are computationally very expensive and the model param-
eters identification usually constitutes a complex task due to the
interdependency between plasticity and damage. Concerning the
uncoupled damage models, damage and plasticity are assumed
independent phenomena, which results in simpler approaches
requiring less computational costs. The uncoupled approaches

0141-0296/$ - see front matter Ó 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engstruct.2013.12.039
 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222 215

1.1. Johnson–Cook model

log(σ )
The JC model for monotonic ductile damage provides the rela-
Monotonic fracture
tion between the equivalent plastic strain at fracture and a mono-
ULCF tonic function of the stress triaxiality [3], expressed as:
ef ¼ C 1 þ C 2 expðC 3 gÞ ð1Þ
LCF
where the stress triaxiality, g, is defined as the ratio between the
HCF hydrostatic pressure and the von Mises equivalent stress:
p
g¼ ð2Þ
rVM
log( N ) and C1, C2 and C3 are material constants to be determined using data
from tensile tests, covering distinct stress triaxialities. In this work,
Fig. 1. Relation of ULCF with other damage mechanisms [1].
the JC model was calibrated by means of an experimental program
of tensile tests on smooth and notched specimens. This model was
can be very efficient for crack initiation modelling, and due to the later used to assess the monotonic fracture strain for the geometries
assumed separation between plasticity and damage, they allow used in ULCF tests. Since the triaxiality in the critical region during
simpler parameters identification procedures and the use of more the tensile loading is not constant, an average value of this param-
accurate state-of-the art plasticity models. eter is used [7]:
There are some propositions in literature for uncoupled ULCF Z ef
1
models, which are supported by distinct physical assumptions. gav ¼  gðeÞde ð3Þ
As mentioned by Komotori and Shimizu [12], the damage accumu- ef 0

lation mechanism in fatigue process is different between the large

and the small plastic strain amplitude regimes. Therefore, fatigue 1.2. Coffin–Manson relation
life in ULCF regime is influenced by ductility, i.e., the large plastic
deformations achieve an important fraction of the monotonic frac- Coffin and Manson [17,18] proposed an empiric relation, which
ture strain, which may activate damage mechanisms typical of has been widely used for LCF, as follows:
monotonic damage, such as voids nucleation and growth. Based
on these assumptions, a fatigue model has been proposed by DeP
¼ e0f ð2Nf Þc ð4Þ
Kuroda [13] which divides the ULCF damage into the following 2
three parts: (a) damage due to tensile straining; (b) damage due
Eq. (4) is represented by a linear relation in a bi-logarithm diagram,
to ductility exhaustion during cyclic loading; (c) damage due crack
where DeP/2 and Nf are a uniaxial plastic strain amplitude and the
propagation. Alternatively, Tateishi and Hanji [14] defined the total
number of cycles to failure, respectively; e0f is the fatigue ductility
damage during ULCF loading as a linear summation between a
coefficient and c is the fatigue ductility exponent. Some authors
tensile ductile part and a cyclic damage. Xue proposed an exponen-
[13,14,19] have shown that the Coffin–Manson relation does not
tial damage rule for fatigue life prediction in the ULCF regime,
give a satisfactory description of the ULCF regime, for many metals.
which overcomes the overestimation limitation of the classical
They report a fatigue life over prediction when the Coffin–Manson
Coffin–Manson approach that has been cited in the literature
relation is used in ULCF domain. The original Coffin–Manson
[15]. The ULCF model based on the cyclic behaviour of micro-voids,
relation was proposed for uniaxial stress–strain conditions, but its
proposed by Kanvinde and Deierlein [16] can also be classed as an
generalisation for multiaxial stress–strain conditions may be
uncoupled damage model, which postulates the material degrada-
performed using an equivalent multiaxial strain definition. The
tion, by micro-void growth, as a function of the plastic strain
Coffin–Manson model is assessed in the present research in order
weighted with a triaxiality function.
to verify its performance. Therefore, the experimental program
Several uncoupled models were assessed and reviewed in this
includes two series of smooth specimens of S185 structural steel
paper, using ULCF data generated for the S185 structural steel. In
to be used in the identification of the Coffin–Manson constants, in
particular, the classical Coffin–Manson relation [17,18], the Kanv-
the LCF region. Then, the model is used to predict ULCF results also
inde–Deierlein model and the Xue model are considered in this
generated in this paper.
study. Besides the ULCF models, the empirical model proposed
by Johnson and Cook, for monotonic ductile damage is considered,
1.3. Kanvinde–Deierlein model
since it will be used to compute the equivalent plastic strain at
fracture, as a function of the stress triaxiality, for the investigated
Kanvinde and Deierlein proposed a model for ULCF based on
geometries. It is demonstrated that the monotonic equivalent plas-
micromechanical behaviour of voids in a plastic medium, which
tic strain at fracture of a particular detail has a significant influence
was a generalisation of a model for monotonic ductile damage
on the ULCF behaviour of that detail. Experimental image-based
[16]. Metallic materials contain voids in its microstructure, which
techniques, such as digital image correlation and features tracking
may grow under the action of the plastic deformation. Race and
methods, were coupled with monotonic tensile tests yielding full-
Tracey [20] reported that, for a single spherical void in an infinite
field measurements which allowed the inspection of the mechan-
continuum, the void growth rate can be described as:
ical tests and provided a diversity of experimental data for plastic-
ity model validation. The load–displacement experimental curves dR=R ¼ C expð1:5gÞdeP ð5Þ
were used to calibrate elastoplastic finite element models. where R is the average void radius, C is a material constant, g is the
In the next subsections a brief review of the damage models qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
that will be assessed in this paper is presented. In particular, the stress triaxiality (Eq. (2)) and dep ¼ð2=3Þdepij depij is the incremental
Johnson–Cook model for monotonic ductile damage is described. equivalent plastic strain. Integrating Eq. (5) and normalising the
The classical Coffin–Manson relation for LCF is presented and the void radius R with respect to the initial void radius R0, the following
Kanvinde–Deierlein and Xue models for ULCF are described. expression is obtained:
216 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222

Z ep
Following the same convention that previously allowed the def-
lnðR=R0 Þ ¼ C expð1:5gÞdeP ð6Þ
0 inition of the monotonic void growth index, for cycling loading the
void growth index is defined as follows:
Assuming that the void growth is the controlling parameter of
X Z eP2
the monotonic fracture, the failure criterion for monotonic damage
is based on the critical size of the void radius and can be expressed VGIcyclic ¼ expðj1:5TjÞdep
tensile cycles eP1
as: Z
X eP2
Z ecritical
p  expðj1:5TjÞdep ð12Þ
ln ðR=R0 Þcritical
monotonic ¼ C expð1:5gÞdeP ð7Þ compressive cycles eP1
0

This expression can be further simplified in order to derive a frac- VGIcyclic increases and decreases during cycling, but this parameter
ture criterion based on void growth index, VGImonotonic, which is cannot assume negative values. If VGIcyclic decreases to zero, it re-
compared to its critical value, VGIcritical mains at zero until a subsequent ‘‘tensile’’ cycle increases its value
monotonic , as described in Eq. (8):
Z above zero. Similarly to monotonic ductile fracture, ULCF failure oc-
ep
ln ðR=R0 Þcritical expð1:5gÞdep curs when VGIcyclic reaches a critical value, VGIcyclic 6 VGIcritical
cyclic .
monotonic =C ¼ VGImonotonic ¼
0 Kanvinde and Deierlein proposed, for the critical cyclic void growth
6 VGIcritical ð8Þ index, VGIcritical
cyclic , an exponential decay function of the critical mono-
monotonic
tonic void growth index:
Eq. (8) is the basis of the void growth model for monotonic loading,  
in which VGIcritical
monotonic is considered a material constant. For cycling
VGIcritical cyclic accumulated
cyclic ¼ VGImonotonic exp kep ð13Þ
loading, critical regions of the component experiences alternating
– positive and negative – stress triaxialities. When triaxiality is po- where k is treated as a material-dependent damage coefficient. To
sitive, voids will grow; inversely, when triaxiality is negative, voids compute the VGIcritical
cyclic it is important to distinguish the equivalent

will shrink (average void radius will reduce). Thus, it is necessary to
consider the sign of the triaxiality to differentiate tensile from com-
pressive loading. The sign of the triaxiality governs the voids
growth or shrinkage, while the magnitude of triaxiality and the
equivalent plastic strain govern the rate of voids growth/shrinkage
[20]. For a cyclic loading, Eq. (5) can be rewritten in the following
plastic strain that increases over the entire loading and the damage
variable, eaccumulated
p , which is defined as the equivalent plastic strain
that is accumulated up to the beginning of each tensile excursion of
loading. Therefore, the increment of equivalent plastic strain during
the current tensile loading does not contribute to the cyclic damage
that occurs within that loading increment. The VGIcritical
cyclic will show a

more general form: discontinuous stepwise behaviour.

dR=R ¼ signðTÞC expðj1:5TjÞdep ð9Þ 1.4. Xue model

where the sign(T) term incorporates the influence of the hydrostatic
pressure, p. In this way, the void growth/shrinkage rate is controlled Xue [15] proposed a model for ULCF as an extension of the Cof-
by the magnitude of triaxiality and the equivalent plastic strain, fin–Manson relation and assuming linear damage summation,
whilst the sign of the triaxiality governs whether voids grow or according to Miner’s rule:
shrink. The void growth/shrinkage during cyclic loading is captured 4nDep n
by the integration of Eq. (9) over tensile compressive excursions of D¼ ¼ ð14Þ
4NDep N
loading. The loading cycles can be subdivided into tensile and com-
pressive, based on the sign of stress triaxiality. The evolution of void where Dep is the plastic strain amplitude, n is the current number
radius along the cyclic loading, for tensile and compressive excur- of applied cycles and N is the number of life cycles to failure. The
sions is expressed as: Coffin–Manson relation, Eq. (4), is conveniently expressed in the
Z alternative form:
X eP2
ln ðR=R0 Þcyclic ¼ C1 expðj1:5TjÞdep Dep  Nk ¼ C ð15Þ
tensile cycles eP1
Z eP2 where C and k are materials constants. Considering a specific type of
X
 C2 expðj1:5TjÞdep ð10Þ periodic plastic loading with Re = 0, the monotonic loading can be
compressive cycles eP1 assumed the limiting case of N = 1/2. For this case, Dep = ef and
C = ef/2k, where ef is the plastic strain at fracture. Xue also assumed
The first term of the previous equation contains the cumulative void
that the cyclic plastic damage is governed by the equivalent plastic
growth over all tensile excursions. However, each excursion re-
distortion, which may be defined using the three principal compo-
quires the computation of the integral between eP1 and eP2 plastic
nents of the plastic strain tensor as:
strains at the beginning and end of that tensile excursion. The same rffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
procedure is used for the compressive excursions, which is distin- 2 2
guished from negative sign in the second term. The negative sign
ed ¼ e þ e22 þ e23 ð16Þ
3 1
presence reflects the assumption that the negative triaxiality will
The equivalent plastic distortion should be distinguished from
cause void shrinkage. C1 and C2 are used to define the different
the equivalent accumulated plastic strain, which takes into account
growth or shrinkage rates. Taking into account the lack of experi-
the plastic strain accumulation along the loading [15].
mental evidence, Kanvinde and Deierlein considers C1 = C2 = C, from
A power law damage rule is derived for monotonic loading, by
which the following expression is obtained:
differentiating the Coffin–Manson relation and knowing that the
X Z eP2 incremental plastic distortion is the same as the equivalent plastic
ln ðR=R0 Þcyclic ¼ C expðj1:5TjÞdep strain, for each proportional loading branch as expressed:
tensile cycles eP1
Z !  ðm1Þ
X eP2 ed 1
 expðj1:5TjÞdep ð11Þ dD ¼ m dep ð17Þ
compressive cycles eP1 ef ef
 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222 217

where m = 1/k. Integrating Eq. (17) along the plastic loading path
yields the plastic damage potential function for materials obeying 1 3
the Coffin–Manson relation:
   m
ed ed
W ¼ ð18Þ
ef ef
2
Based on experimental data, Xue proposed an alternative to the
damage potential function, valid for fatigue life prediction in ULCF 4
regime:
ed
  k
e e ef  1
W d ¼ k ð19Þ
ef e 1
where k is a damage parameter, which is determined from experi-
mental data by regression analysis. Using Eq. (19), the number of
5
life cycles, as function of strain amplitude, can be obtained by
differentiation:
1 ek  1 Fig. 2. Tensile tests on dog bone (1) (DB), plate with a circular hole, (2) (PCH),
N¼   ð20Þ smooth cylindrical specimen, (3) (SC) and notched cylindrical specimens with high,
2 k
Dep
ef (4) (SNC) and low (5) (LNC) notch severity.
e 1
However, Xue verified that neither the power law nor the expo-
The displacement and strain resolution associated to the DIC mea-
nential damage rules fits the complete range of life cycles from
surements were, respectively, in the range of 2  102 pixel and
both LCF and ULCF regimes. Therefore a new function, based on
0.02–0.04%. This estimation was based on simple rigid-body tests
m and k parameters, was suggested to predict fatigue life cycles
[25,26]. For cylindrical specimens, feature tracking method was
from 1/2  104 [15]. In order to derive a new damage potential
used [21,27], as shown in Fig. 2. Circular marks of about 2 mm
function for the entire fatigue regime, the influence of the plastic
were painted on the surface of the specimens and tracked in the
damage accumulation and the exponential damage rule for the
images recorded during tests, at an acquisition frequency of 1 Hz.
ULCF regime can be combined through the joint effect of m and k
This tracking was performed by a computational algorithm that al-
parameters:
lows the measurement of the displacements of the target features
 m
  k ed
e in a sequence of images grabbed during the object deformation by
e e 1 f
an external loading.
W d ¼ ð21Þ
ef ek  1 Images were recorded by a Baumer Optronic FWX20 digital
camera coupled with telecentric lens. The lens aperture, lighting
The damage rate is derived from Eq. (21),
intensity and shutter time were set in order to enhance image seg-
  m 
 ðm1Þ k ed mentation. Based on rigid-body translation tests, an estimation of
ed ef
mk e e dep the strain resolution was obtained in the range of 0.03–0.04%.
f
dD ¼ ð22Þ These techniques allowed the evaluation of the load–displacement
ðek  1Þ ef
experimental curves, which were used to calibrate the finite ele-
and the strain–life relationship is governed by the Eq. (23): ment models. Numerical simulations, using ANSYS 12.1Ò [28],
were performed to compute the necessary parameters to calibrate
1 ek  1
N¼  m ð23Þ the JC and KD models. The finite element models included a plas-
2 k ed
e
ticity model based on J2 yield theory, with isotropic hardening.
e f
1

2. Monotonic tensile tests

Monotonic tensile tests were conducted in order to allow the

identification of the Johnson–Cook model [3] as well as the Kanv-
inde and Deierlein model [16], for monotonic ductile fracture. Ten-
sile tests were carried out on a servohydraulic INSTRON 8801 test
machine, at room temperature. Dog-bone (DB), plate with a circu-
lar hole (PCH), smooth (SC), large notch (LNC) and small notch
(SNC) cylindrical specimens were machined from a S185 structural
steel plate with a thickness of 8 mm, as described in Ref. [21] (see
Fig. 2).
Plane specimens were painted using aerosol spray in order to
define a speckle pattern suitable for DIC measurements. The ARA-
MIS DIC 2D system was used in this work [22–24]. The optical sys-
tem was equipped with an 8-bit Baumer Optronic FWX20 digital
camera coupled with an Opto-Engineering TC 23 36 telecentric
lens. A facet size (number of pixels per subset) and facet step (dis-
tance between adjacent facet centres) were set to 15  15 pixels2
(0.27  0.27 mm2). In post-processing, the strain fields were evalu- Fig. 3. Finite elements meshes of dog bone (1), plate with a circular hole (2),
ated from the displacement fields by numerical differentiation smooth cylindrical specimen (3) and notched cylindrical specimens with high (4)
using a base strain length of 7  7 subsets (1.89  1.89 mm2). and low (5) notch severity, built in ANSYS 12.1Ò (SOLID 185 elements).
218 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222

(a) 50 (b) 70
45
60
40
35 50

Load [kN]
Load [kN]

30 40
25
20 30

15 20
10
10
5
0 0
0 2 4 6 8 0 2 4 6 8
Displacement [mm] Displacement [mm]

(c) 18 (d) 18 Numerical curve

16 16 Experimental curves
14 14

12 12

Load [kN]
Load [kN]

1 2 3 10
10
5mm
8 8 1 2 3 4
6 6
5mm 10mm 5mm
4 1 2 3 4
2 2
0 0
0 1 2 3 4 5 6 0 1 2 3 4

Displacement [mm] Displacement [mm]

(e) 14 Numerical curve

12 Experimental curves

10
Load [kN]

6 1 2 3 4

4 5mm 10mm 5mm

0
0 0.5 1 1.5 2 2.5
Displacement [mm]

Fig. 4. Experimental load–displacement experimental curves versus numerical response; (a) dog-bone (DB); (b) plate with a circular hole (PCH); (c) smooth cylindrical
specimen (SC); (d) notched cylindrical specimen with low severity (LNC) and (e) notched cylindrical specimen with high severity (SNC).

SC
2.0 SNC
Three-dimensional models, presented in Fig. 3, were built with LNC
DB
the 8-node isoparametric solid element SOLID185 [28]. The exper- PCH
imental load–displacement curves are compared with the numer- 1.6 UNC
UPCH
ical responses for the DB, PCH, SC, LNC and SNC specimens and UPCH_5
(Fracture strain)

UPCH_6
represented in Fig. 4. A very good agreement between the experi- 1.2 Johnson-Cook

mental and numerical curves is verified.

The equivalent plastic strain at fracture and the stress triaxiality 0.8
(Eq. (3)) were computed for each type of specimens used in the
f

tensile tests, allowing the computation of the material constants,

0.4
C1 = 0.85, C2 = 2.44 and C3 = 4.67 for the JC model. The ductility/ f = 0.85 + 2.44 exp (−4.67 )
damage curve of the S185 structural steel is presented in Fig. 5.
The geometries selected for the monotonic tensile tests were 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
not suitable for the ULCF tests, since specimens showed plastic
Triaxiality
instability due to inappropriate slenderness. New specimen geom-
etries were prepared for ULCF testing, which will be described lat- Fig. 5. Ductility curve of the S185 structural steel.
ter (see Fig. 8). The monotonic fracture properties of the specimens
proposed for ULCF tests were assessed from the JC model. Some
deviations from the JC model mean line were allowed since the 3. LCF tests
experimental data also showed some scatter around that line
(see Fig. 5). Concerning the KD modelling, the VGIcritical
monotonic parame- In this section, the LCF behaviour of the S185 structural steel is
ter was computed for specimens used in ULCF tests, by means of evaluated using fatigue tests on two series of smooth specimens,
Eq. (8), the fracture strain being computed from the JC model. according to the ASTM E 606 standard [29], as described by Xavier
 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222 219

1.0E+3 Table 1
CLCF Cyclic strain hardening coefficient and exponent obtained for the S185 steel grade.
log( ) = 0.1841log(ΔεP /2) + 2.8836
R² = 0.9839 LCF
Set of specimens K0 (MPa) n0 R2
CLCF 802.5 0.1948 0.9877
LCF 705.5 0.1673 0.9928
Δσ/2 [MPa]

Global results 764.9 0.1841 0.9839

n'
1.0E-2
LCF
1 ( /2) = 0.0033 (2Nf)-0.1024+0.1955 (2Nf)-0.4891
CLCF

1.0E+2 1.0E-3
1.0E -5 1.0E -4 1.0E -3 1.0E -2
ΔεP/2 [-]

Δε/2
1.0E-4
Fig. 6. Cyclic curve of the S185 structural steel.

1.0E-5
et al. [21]. Cylindrical (CLCF) and dog-bone (LCF) type specimens E/2)
( =0.0033 (2Nf)-0.1024
were used for the first and second test series, respectively. Fatigue ( P/2 ) = 0.1955 (2Nf)-0.4891
tests were performed on a servohydraulic INSTRON 8801 universal 1.0E-6
test machine, at room temperature, under strain control with 1.0E+1 1.0E+2 1.0E+3 1.0E+4 1.0E+5
strain ratio, Re = 1. A clip gauge with a base length of 12.5 mm 2Nf
and limit displacements of 2.5 mm was used to control the strain.
Fig. 7. Global strain–life curves for the S185 structural steel, according to the
Morrow’s model.
3.1. Cyclic elastoplastic behaviour

The stress and strain amplitudes of the stabilised hysteresis
cycles were used to compute the cyclic curve of the material. The
cyclic curve was expressed through the mathematical relation
proposed by Ramberg–Osgood:
 1=n0
De DeE DeP Dr Dr
¼ þ ¼ þ
2 2 2 2E 2K 0
where E is the Young modulus, K0 and n0 are the cyclic strain hard-
ening coefficient and exponent, respectively; Dr and De are, respec-
tively, the stress and strain ranges; DeE and DeP are, respectively, the
elastic and plastic strain ranges. The cyclic curve of the material was
fitted by using both series of specimens, as shown in Fig. 6. These
results were used to compute the cyclic plastic constants of the
S185 structural steel. The correlation of the global results resulted
in a high value of the determination coefficient (R2 = 0.9839). Table 1
summarises the cyclic strain hardening coefficient and exponent,
obtained for each test series and for combined test series results.
3.2. Strain–life behaviour
LCF test data are usually presented using relations between the
strain amplitude and the number of reversals to failure, 2Nf. The ef-
fects of the elastic and plastic strains are accounted separately by
the Basquin [30] and Coffin–Manson [17,18] relations, respectively.
The combination of these relations results the Morrow [31] rela-
tion, expressed as follows:
De DeE DeP rf
0
¼ þ ¼ ð2Nf Þb þ e0f ð2Nf Þc
2 2 2 E
where r0f is the fatigue strength coefficient, b is the fatigue strength
exponent and E is the Young modulus. The other constants have
been introduced before. The constants of Morrow’s relation were
computed separately from the elastic strain versus life and plastic
strain versus life relations. The plastic strain range was determined
from the width of the stabilized hysteresis loops; regarding the
elastic strain range, it was obtained from the difference between
the total and the plastic strain ranges. The strain versus life curves
for both LCF test series correlated together are shown in Fig. 7. For
each strain component (total, plastic and elastic), one curve is
represented, which allowed the evaluation of the constants of the
Morrow’s relation, presented in Table 2. The two specimen geome-
tries did not influenced the LCF behaviour of the material. Both
geometries correspond to similar uniaxial stress conditions in the
LCF regime. The number of cycles to failure, in these tests, was in
the range of 102 to 104 cycles, which is in agreement with the LCF
ð24Þ
regime. These smooth specimens did not allow ULCF testing due
to instability under compressive loading. These LCF results will be
used in fatigue life predictions, for notched specimens tested under
fatigue lives below 102 cycles.
4. ULCF tests
ULCF tests on four series of notched specimens (see Fig. 8) were
performed on the same machine used in the other tests presented
in this research, at ambient conditions of the LCF tests, under local
(gauge) displacement control (gauge length, L0 = 12.5 mm), with
displacement ratio equal to 1. Each test was also simulated using
ANSYS 12.1Ò, using isoparametric 20-noded solid elements,
SOLID185 [28]. The finite element models included a plasticity
model based on von Mises yield criterion, with multilinear kine-
matic hardening based on Besseling proposal [32].
LCF test data was used to derive the cyclic curve of the material,
by means of the Ramberg–Osgood relation, which was used to cal-
ibrate the plasticity model and represented in Fig. 9. Finite element
simulations of the cyclic tests were performed to compute the
stress–strain field histories at the critical regions, and thereby to
ð25Þ compute the required parameters for the assessment of the ULCF
models, described in Section 1.
The Coffin–Manson relation was used to predict the number of
fatigue cycles for specimens tested in ULCF regime, using a multi-
axial strain definition according to the ASME [33]:
pffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Depeq ¼ 3
2
ðDp11  Dp22 Þ2 þðDp22  Dp33 Þ2 þðDp33  Dp11 Þ2 þ 32 ðDp212 þ Dp223 þ Dp231 Þ
Dpij ¼ pð2Þ ð1Þ
ij pij
ð26Þ
ð2Þ ð1Þ
where pij ; pij
are the plastic strain components at the reversals
points t2 and t1.
220 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222

Table 2
Morrow constants obtained for the S185 steel grade.

Set of specimens r0f (MPa) b R2 e0f c R2

CLCF (uniaxial) 597.99 0.0885 0.8613 0.1515 0.4549 0.9653

LCF (uniaxial) 722.43 0.1090 0.8889 0.2297 0.5112 0.9593
Global results (uniaxial) 676.39 0.1024 0.8531 0.1955 0.4891 0.9588
Global results (multiaxial) – – – 0.1661 0.4579 0.9456

0.40 y = e-0.252x UPCH

y = e-0.358x R² = 0.4846
0.35 UPCH_5
R² = 0.9622
PLF_6
0.30
UNC
0.25

R VGI
0.20
y = e-0.628x
0.15
R² = 0.5012
0.10
y = e-0.257x
0.05 R² = 0.9029
0.00
0 2 4 6 8 10
Damage variable, accumulated equivalent plastic strain

Fig. 8. Specimens used in ULCF tests: (1) UPCH, (2) UPCH_5, (3) UPCH_6 and (4) Fig. 11. Cyclic damage resistance evolution with the accumulated plastic strain.
UNC test series.

Table 3
900
Summary of Kanvinde and Deierlein modified model parameters.
800
Set of specimens k ef VGIcritical
monotonic
700
UNC 0.628 0.83 5.08
True stress [MPa]

600 Cyclic curve (Ramberg-Osgood relation) UPCH 0.257 1.16 2.54

Numerical fit UPCH_5 0.252 1.31 2.82
500
UPCH_6 0.358 1.07 2.26
400

300

200 Table 2 which also contains the CM constants obtained from a uni-
axial approach. The required multiaxial equivalent strains were ob-
100
tained from numerical simulation of the LCF and CLCF specimens,
0
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40
run in ANSYS 12.1Ò.
Fig. 10 compares the predictions, based on the CM model, with
True strain [-]
the experimental results, obtained with multiaxial CM approach.
Fig. 9. Cyclic curve of the S185 structural steel with numerical fit. According to Fig. 10, the CM relation does not give a satisfactory
life prediction in ULCF regime. The fatigue life is underestimated,
which does not agree with literature that states that CM model
Using the multiaxial equivalent strain definition, new CM tends to overestimate the fatigue life in the ULCF regime. A reason
constants were evaluated using the data from the low cycle fatigue for this apparent discrepancy is the use of different geometries in
tests performed on LCF and CLCF series, and were included in ULCF and LCF tests. While in the LCF tests, smooth specimens were
used, for the ULCF tests, notched specimens were performed,
which provides higher levels of stress triaxiality.
100000 With respect to the KD model, the RVGI ¼ VGIcritical critical
cyclic =VGImonotonic
was computed, supported on numerical simulation of the ULCF
10000 tests, for each specimen, and results were plotted against the accu-
mulated equivalent plastic strain, in Fig. 11. It is verified a distinct
Nf (experimental)

1000 relation for each tested geometry, which means that Eq. (13) is not
unique, or that the k parameter is not a material constant, but is
100 CLCF
LCF dependent on geometry. A new exponential damage function is
10
UPCH proposed to define RVGI ¼ VGIcritical critical
cyclic =VGImonotonic , with the following
UPCH_5
UPCH_6 form:
1 UNC h i
2x VGIcritical critical
 accumulated
cyclic ¼ VGImonotonic  exp C 1 expðC 2 ef Þep ð27Þ
0.5x
0.1
0.1 1 10 100 1000 10000 where C1 and C2 are material parameters and k becomes dependent
Nf (CM relation with multiaxial approach) on the fracture strain of the component, which in turn depends on
the triaxiality level of the component. It was already verified that
Fig. 10. Experimental results versus predictions by Coffin–Manson relation using a
multiaxial strain definition approach. the VGIcritical
monotonic also changes with the geometry, as expressed in
 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222 221

1.0
UPCH 5000
UPCH_5 m=0.9801
0.8 λ=1.2238
λ (KD model parameter)

UPCH_6

Nf (experimental)
0.6 UNC 500 UPLF
UPCH_5
PLF_6
UNC
0.4 CLCF
LCF
λ=3.4944e-2.107εf 50
1.33x
0.2 R2=0.9229 0.75x
2x
0.5x
0.0 5
0.60 0.80 1.00 1.20 1.40
5 50 500 5000
εf (Fracture strain) Nf (Xue model)

Fig. 12. Cyclic damage parameter versus fracture strain for each set of specimens. Fig. 15. Experimental number of cycles versus calculated number of cycles by Xue
model.

3.00 VGI cyclic Finally, the Xue model is used to predict both the ULCF and LCF
VGI cyclic critical tests. This model was originally presented for test data with a
2.50
strain ratio Re = 0. However, this approach can be changed for
Re = 1 (strain ratio adopted in the present research) if Eq. (23) is
Void Growth Index

2.00
modified according to the Eq. (28), which takes into account the
1.50
four reversals of the plastic distortion:
Failure instant 1 ek  1
1.00 N¼  m ð28Þ
4 e
k ed
e f
1
0.50
The ½ or the 1=4 coefficients appearing in the Xue model (respec-
0.00 tively in Eqs. (23) and (28)) are necessary adjustments to account
0 2 4 6
for the strain ratio, if the current plastic distortion (always posi-
Damage variable, accumulated equivalent plastic strain
tive!), given by Eq. (16), is used, as originally proposed by Xue. This
Fig. 13. Damage evolution at the centre of a notched specimen during a cyclic strain ratio parameter dependency can be considered a limitation of
loading (UPCH specimens). the original Xue model. This limitation may be overcome if the cur-
rent plastic distortion is replaced by the equivalent plastic strain
range, as defined by ASME [33]. In this case, there is no need for
50 any adjustment in the Xue model and the number of cycles can
be computed using the following equation, for any strain ratio:
ek  1
N¼  m ð29Þ
Nf (experimental)

Depeq
k e
e f
1
UPCH
The Xue model was originally derived and applied to cyclic data
UPCH_5
from smooth specimens. In that case, ef is the uniaxial plastic strain
UPCH_6
at fracture that is assumed a material property. In this research, the
UNC
Xue model is extended for notched specimens, and ef is considered
1.33x
the equivalent accumulated plastic strain at fracture ef , which is a
0.75x
5 function of the stress triaxiality, as supported by JC model
5 50
Nf (KD model modified) formulation.
Numerical analyses performed in ANSYS 12.1Ò provided the
Fig. 14. Experimental number of cycles versus calculated number of cycles by plastic strain component histories that were used to determine
modified KD model.
the equivalent plastic strain ranges. Concerning the strain at frac-
ture, it was obtained from monotonic finite element model analy-
ses, for the specimens used in ULCF tests. The equivalent plastic
Table 3. Fig. 12 illustrates the cyclic damage parameter k against the strain was measured at the critical region of the finite element
fracture strain, ef , for each series of specimens, a good correlation models of the ULCF specimens. Typically, for monotonic tensile
being found, when an exponential function is used to describe the tests, this region corresponds to the interior of the material. The
relation between these variables. strain at the fracture was evaluated for each specimen by using
The curves shown in Fig. 13 demonstrate how the VGIcyclic var- the Eq. (3) iteratively, since this strain depends on average triaxial-
ies during the cyclic loading and the VGIcritical
cyclic decreasing behaviour ity, which in turn depends on the fracture strain itself. The pro-
– the intersection of the two curves indicates the failure point. As posed failure points were plotted in Fig. 5 (solid symbols). Both
referred above, VGIcritical
cyclic reduces in a stepwise way, based on the m and k parameters were computed using the least squares meth-
accumulation of the plastic strain at the initiation of each tensile od. Comparisons between experimental Nf and predicted Nf (Xue
load incursion. In Fig. 14, the results obtained with the new model) are plotted in Fig. 15 for both ULCF and LCF tests carried
proposal of the KD model are compared with the experimental out in this study. In general, it was verified that Xue’s model is able
data. Generally, it is verified that the proposed modified KD model to predict fatigue life in both LCF and ULCF regimes, providing very
produces good predictions of the fatigue life, in the ULCF regime. satisfactory results for both fatigue regimes.
222 J.C.R. Pereira et al. / Engineering Structures 60 (2014) 214–222
5. Conclusions
The ULCF behaviour of S185 structural steel was investigated
and existing damage models were assessed using available exper-
imental data. The following conclusions resulted from the pre-
sented research:
– In contrast with the literature, in this work it was verified that,
the Coffin–Manson model underestimates the fatigue life of the
notched details under ULCF. However, the tested geometries in
ULCF and LCF regimes were distinct, showing different levels of
stress triaxialities. The trend observed in literature was based
on same geometries tested in both ULCF and LCF regimes.
– Concerning the KD model, a new exponential damage function
was proposed for VGIcritical
cyclic , which provides the relation between
cyclic damage parameter k and the monotonic equivalent strain
at fracture ef , which depends on the stress triaxiality. This new
formulation allows improved predictions in the ULCF domain.
– A generalisation of the Xue model for any strain ratio and mul-
tiaxial strain fields was proposed. A multiaxial definition of the
fracture strain at failure, dependent on the stress triaxiality was
proposed. In addition, the use of the equivalent plastic strain
range definition according to the ASME, allows the model to
be applied for any strain ratio, without the need for any modi-
fication in its formulation.
– Both modified Xue and Kanvinde and Deierlein models resulted
in satisfactory predictions in the ULCF regime.
Acknowledgements
The author would like to acknowledge the Fundação para a
Ciência e Tecnologia for their financial support through the SFRH/
BD/80091/2011 Grant and Ciência2008 program. The European
Commission is also acknowledged through the Research Fund for
Coal and Steel that is funding the ULCF project.
References
[1] Bleck W, Dahl W, Nonn A, Amlung L, Feldman M, Schäfer D, Eichler B.
Numerical and experimental analyses of damage behaviour of steel moment
connection. Eng Fract Mech 2009;76:1531–47.
[2] Gurson AL. Continuum theory of ductile rupture by void nucleation and
growth. Part I: Yield criteria and flow rules for porous ductile media. J Mater
Eng – Trans ASME 1977;99(1):2–15.
[3] Johnson GR, Cook WH. Fracture characteristics of three metals subjected to
various strains, strain rates, temperatures and pressures. Eng Fract Mech
1985;21(1):31–48.
[4] Wilkins ML, Streit RD, Reaugh JE. Cumulative-strain-damage model of ductile
fracture: simulation and prediction of engineering fracture tests. Technical
Report UCRL-53058. Lawrence Livermore National Laboratory; October 1980.
[5] Cockcroft MG, Latham DJ. Ductility and the workability of metals. J Inst Metals
1968;96:33–9.
[6] Wierzbicki T, Xue L. On the effect of the third invariant of the stress deviator on
ductile fracture. Impact and Crashworthiness Lab Report 136. Cambridge, MA:
Massachusetts Institute of Technology; 2005.
[7] Wierzbicki T, Bao Y, Lee Y-W, Bai Y. Calibration and evaluation of seven
fracture models. Int J Mech Sci 2005;47:719–43.
[8] Dufailly J, Lemaitre J. Modeling very low cycle fatigue. Int J Damage Mech
1995;4(2):153–70.
[9] Lemaıtre J. A continuous damage mechanics model for ductile fracture. J Eng
Mater Technol – Trans ASME 1985;107:83–9.
[10] Leblond JB, Perrin G, Devaux J. An improved Gurson-type model for hardenable
ductile metals. Eur J Mech – A/Solids 1995;14:499–527.
[11] Steglich D, Pirondi A, Bonora N, Brocks W. Micromechanical modelling of cyclic
plasticity incorporating damage. Int J Solids Struct 2005;42:337–51.
[12] Komotori J, Shimizu M. Fracture mechanism of ferritic ductile cast iron in
extremely low cycle fatigue. In: Rie KT, Portella PD, editors. Low cycle fatigue
and elasto–plastic behaviour of materials. Elsevier Science Ltd.; 1998. p.
39–44.
[13] Kuroda M. Extremely low cycle fatigue life prediction based on a new
cumulative fatigue damage model. Int J Fatigue 2001;24:699–703.
[14] Tateishi K, Hanji T. Low cycle fatigue strength of butt-welded steel joint by
means of new testing system with image technique. Int J Fatigue
2004;26:1349–56.
[15] Xue L. A unified expression for low cycle fatigue and extremely low-cycle
fatigue and its implication for monotonic loading. Int J Fatigue
2007;30:1691–8.
[16] Kanvinde AM, Deierlein GG. Cyclic void growth model to assess ductile
fracture initiation in structural steels due to ultra-low cycle fatigue. J Eng
Mech – ASCE 2007;133(6):701–12.
[17] Coffin Jr LF. A note on low cycle fatigue laws. J Mater JMLSA
1971;6(2):388–402.
[18] Manson SS. Behavior of materials under conditions of thermal stress. NACA-
TR-1170. National Advisory Committee for Aeronautics; 1954.
[19] Hatanaka K, Fujimitsu T. Some considerations on cyclic stress–strain relation
and low cycle fatigue life. Trans Japan Soc Mech Eng – Part A
1984;50(451):291–300.
[20] Rice JR, Tracey DM. On the ductile enlargement of voids in triaxial stress fields.
J Mech Phys Solids 1969;35:201–17.
[21] Xavier J, Pereira JCR, de Jesus AMP. Characterisation of steel components under
monotonic loading by means of image-based methods. Opt Lasers Eng
2014;53:142–51.
[22] ARAMIS. User Manual – Software – v6.0.2-6. GOM (<http://www.gom.com>);
2009.
[23] Xavier J, de Jesus AMP, Morais JJL, Pinto JMT. Stereovision measurements on
evaluating the modulus of elasticity of wood by compression tests parallel to
the grain. Constr Build Mater 2012;26(1):207–15.
[24] Sousa AMR, Xavier J, Morais JJL, Filipe VMJ, Vaz M. Processing discontinuous
displacement fields by a spatio-temporal derivative technique. Opt Lasers Eng
2011;49(12):1402–12.
[25] Sousa AMR, Xavier J, Morais JJL, Filipe VMJ, Vaz M. Cross-correlation and
differential technique combination to determine displacement fields. Strain
2011;47(Suppl. 2):87–98.
[26] Xavier J, Sousa AMR, Morais JJL, Filipe VMJ, Vaz M. Measuring displacement
fields by cross-correlation and a differential technique: experimental
validation. Opt Eng 2012;51:043602.
[27] Chiassi B, Xavier J, Oliveira D, Lourenço P. Application of digital image
correlation in investigating the bond between FRP and masonry. Compos
Struct 2013;106:340–9.
[28] ANSYS. Swanson Analysis Systems Inc. Houston, Version 12.1; 2009.
[29] ASTM E606. Standard practice for strain controlled fatigue testing. In: Annual
book of ASTM standards, Part 10, ASTM; 1998; p. 557–71.
[30] Basquin OH. The exponential law of endurance tests. Proc Am Soc Testing
Mater 1910;10:625–30.
[31] Morrow JD. Cyclic plastic strain energy and fatigue of metals. Int Friction
Damping Cyclic Plast ASTM STP 1965;378:45–87.
[32] Besseling JF. A theory of elastic, plastic, and creep deformations of an initially
isotropic material showing anisotropic strain-hardening creep recovery and
secondary creep. J Appl Mech 1958:529–36.
[33] ASME. ASME boiler and pressure vessel code. New York: ASME; 2004.