International Journal of Fatigue 33 (2011) 194–202

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Mechanisms and modelling of fatigue crack growth under combined low

and high cycle fatigue loading
C. Schweizer ⇑, T. Seifert, B. Nieweg, P. von Hartrott, H. Riedel
Fraunhofer Institute for Mechanics of Materials IWM, Woehlerstrasse 11, 79108 Freiburg, Germany

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

Article history: In this paper a mechanism-based model is presented, which is able to describe the evolution of micro-
Received 17 February 2010 cracks under pure low cycle fatigue (LCF) and combined LCF and high cycle fatigue (HCF) loading condi-
Received in revised form 30 July 2010 tions. In order to verify the model and to calibrate the model parameters, the crack length evolution of
Accepted 6 August 2010
microcracks is followed at room temperature for pure LCF and combined LCF/HCF loading in a 10%-chro-
Available online 13 August 2010
mium steel. These studies reveal accelerated crack growth rates under LCF/HCF interaction as soon as a
critical crack length is reached. The model is capable of accounting for this effect and needs only few
Keywords:
parameters, including the threshold for fatigue crack growth, whose knowledge is crucial for the accuracy
Low cycle fatigue
High cycle fatigue
of the model.
Crack-tip opening displacement (CTOD) Ó 2010 Elsevier Ltd. Open access under CC BY-NC-ND license.
Threshold stress intensity factor
Fatigue crack growth

1. Introduction to smaller values [4,5]. In Powell et al. [4] the influence of the num-
ber of HCF cycles per LCF cycle on fatigue crack growth was inves-
Many engineering components are subjected to combined low tigated systematically. No measurable fatigue crack growth
cycle fatigue (LCF) and high cycle fatigue (HCF) loadings, e.g. LCF acceleration was found for 1, 10 or 100 HCF cycles per LCF cycle.
start–stop temperature cycles (thermomechanical fatigue, TMF) For 1000 HCF cycles or more, significantly higher crack growth
in steam turbine rotors are superimposed by HCF reverse bending rates were observed for stress intensity factors above DKonset.
cycles or vibrations. In combustion engines, combustion pressure While the effect of superimposed HCF loadings on the growth of
results in superimposed HCF. long cracks and elastic material behavior has been well docu-
In the LCF regime larger strain amplitudes lead to early micro- mented in literature, up to now the effect of LCF/HCF interaction
crack initiation, such that the component’s lifetime is determined on microcrack growth under large plastic deformations has not
by the growth of these microcracks. The superimposed HCF cylces been investigated in detail. In this paper naturally grown cracks
with smaller strain amplitudes even below the endurance limit of are measured with help of the replica technique for both pure
the respective material can reduce lifetimes significantly. LCF and combined LCF/HCF loading. A model is presented, which
Several studies on aluminium and cobalt alloys, which were accounts for the accelerated crack growth rates under LCF/HCF
tested under combined TMF and HCF loading, showed that out- conditions.
of-phase TMF lives are reduced with rising HCF amplitudes [1–3].
A higher frequency of the HCF loading further decreased the num-
ber of cycles to failure [2]. 2. Analytical model
In [4–9] the effect of superimposed HCF cycles on fatigue crack
growth was studied. For long cracks and small-scale yielding con- 2.1. Basic model
ditions, the HCF cycles lead to accelerated crack growth rates in
comparison to pure LCF loading as soon as the total stress intensity Experiments by Neumann [11] and others [12,13] suggest a cor-
range DKtotal overcomes a certain threshold DKonset [6,7,10]. DKonset relation between the crack growth increment da/dN and the cyclic
can usually be attributed to the point when the stress intensity crack-tip opening displacement DCTOD for cracks under mode 1
range of the HCF cycles (DKHCF) exceeds the threshold of fatigue loading. This leads to the following crack growth law:
crack growth DKth. Larger DKHCF at constant DKtotal shift DKonset
da
¼ bDCTOD; ð1Þ
dN
⇑ Corresponding author.
E-mail address: [email protected] (C. Schweizer). where b is a proportionality constant.

0142-1123/Ó 2010 Elsevier Ltd. Open access under CC BY-NC-ND license.

doi:10.1016/j.ijfatigue.2010.08.008
 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202 195

According to Shih [14] a simple relationship exists between the tions. Furthermore, the HRR-solution for dn deviates from finite
J-integral and the crack-tip opening displacement: element calculations, if plasticity is not confined to a small zone
around the crack-tip and the uncracked ligament is subjected pri-
J
CTOD ¼ dn ; ð2Þ marily to tension. Thus, in [26] Eq. (4) was checked against results
r0 of finite element calculations. It was found that Eq. (4) agrees with
where r0 denotes the yield stress and dn depends on the hardening the numerical calculations to within a factor of 1.25. In comparison
behavior of the material. For power-law hardening materials the to analytical expressions for stress intensity factors this deviation
crack-tip fields are the Hutchinson, Rice and Rosengren (HRR) sin- is rather large, but considering the above mentioned points, the
gular fields [15,16]. In this case dn is a function of the Ramberg–Os- authors believe that the accordance between Eq. (4) and the
good hardening exponent n and was tabulated by Shih [17] for both numerical results in [26] is good.
plane strain and plane stress. The tabulated values for plane strain
given by Shih [17] can be fitted with a third order polynomial
2.2. Model for combined LCF/HCF loading
function:

dn ¼ 0:78627  4:41692n þ 6:11945n2  4:2227n3 : ð3Þ To account for the effect of superimposed HCF loadings on fati-
gue crack growth, the basic model of the previous section is ex-
If cyclic material behavior is considered, the cyclic J-integral
tended assuming that each HCF cycle leads to a change of the
(DJ) has a solid mathematical foundation if the material shows
crack-tip opening displacement and according to Eq. (1) results
Masing behavior [18]. All properties in Eq. (2) have to be replaced da 
in a crack growth increment dN HCF
. Thus, in one loading block,
by their cyclic counterparts. Following Kumar et al. [19] and Heit-
which comprises one LCF cycle and the corresponding superim-
mann et al. [20], DJ can be approximated by the sum of an elastic
posed
 HCF cylces (Fig. 1a), the total crack growth increment
(or small scale yielding) and a plastic contribution. For a small da 
dN block
is expressed as the sum of the total (peak-to-peak) loading
semicircular crack with crack length a in a flat specimen under
cycle and the sum of all HCF-cycles during one LCF cycle:
plane strain and mode 1 loading conditions, one obtains  
! da  da  X da 
Dr
2 ¼ þ  : ð5Þ
dn0 eff DrDpl dNblock dNtotal block dNHCF
DCTOD ¼ 1:45 þ 2:4 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a; ð4Þ
rcy E 1 þ 3n0
This
 summation is also employed by [4,6,7,9].
where n0 is the cyclic hardening exponent, rcy is the cyclic yield da 
is calculated by using Eqs. (1) and (4), where all quantities
dN total
stress and E is Young’s modulus. Dr and Dpl are the stress and from Eq. (4) are taken from the enveloping stress–strain hysteresis
plastic strain range respectively. The second term in brackets was loop (Fig. 1a). In order to calculate the crack growth rates of the
derived by He and Hutchinson [21] for a power-law hardening da 
HCF cycles dN HCF
; DCTOD can again be calculated by Eq. (4), where
material in the fully plastic limit, but modified by a factor 1.25 to now all quantities are  taken from the stress–strain hysteresis loops
da 
account for the fact that a surface crack is considered [22]. Crack of the HCF cycles. dN only contributes, if the maximum stress of
HCF
closure is taken into account by the use of the effective stress range the individual HCF-cycle is larger than the opening stress rop from
Dreff = rmax  rop in the elastic part of DJ, as was proposed by Heit- the enveloping hysteresis loop. In engineering applications the HCF
mann et al. [20]. rmax and rop are the maximum and crack opening loadings are typically so small that they result in elastic unloading
stress, respectively. All quantities in Eq. (4) can be determined from from the enveloping stress–strain hysteresis loop as illustrated in
stress–strain hysteresis loops except rop, which can either be esti- Fig. 1a and in a stress intensity range DKHCF close to the threshold
mated with empirical formulas given in [20,23,24] or taken from of fatigue crack growth DKth. In this case the second term in brack-
numerical calculations. Here, the crack opening stress equation by ets in Eq. (4) vanishes and the effective stress range Dreff is evalu-
Newman [23] is used, which predicts decreasing crack opening ated by using the respective load ratio rmin,HCF/rmax,HCF and the
stresses with increasing maximum stresses, as is observed under maximum stress of each HCF cycle. Additionally Eq. (4) is extended
LCF conditions [25]. by a function G, which accounts for near threshold behavior [27].
The analytical expression for DCTOD (Eq. (4)) contains several 
assumptions. The interpolation formula for the cyclic J-Integral is da 
¼ ðbDCTODÞG: ð6Þ
an engineering approach for combined elastic and plastic deforma- dNHCF

a b

Fig. 1. (a) Schematic representation of a stress–strain hysteresis loop of one loading block of combined LCF/HCF loading. rmax,HCF and rmin,HCF are the maximum and minimum
stresses of an arbitrary HCF-cycle. (b) Stabilized stress–strain hysteresis loops of specimens P11A and P1B for pure LCF loading.
196 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202

The function G is defined as Table 1

Overiew over all LCF and LCF/HCF tests.
   
DCTODth p=2 DK th;eff p
G¼1 ¼1 ; ð7Þ Specimen a,LCF [%] a,HCF [%]
DCTOD DK eff
P1B 0.45 0
for DKeff P DKth,eff and is set to zero otherwise. Here DKth,eff denotes P11A 0.5 0
P12B 0.4 0.05
the effective threshold for fatigue crack growth and DKeff is the effec-
P09A 0.4 0.05
tive stress intensity factor range. For elastic materials DCTOD is pro-
portional to DK2 and thus the original formulation of Newman [27]
in terms of the stress intensity factor can be used. The exponent p
controls the transition from the threshold to the Paris-line. Consis- Table 2
tently with Eq. (4) the effective stress intensity factor range for a Averaged material parameters determined
small semicircular surface crack in a flat specimen under mode 1 from stabilized hysteresis loops.
loading is taken from the solution of a penny-shaped crack, but mod- E [MPa] rCY [MPa] n
0

ified by a factor of 1.12 in order to account for a surface crack [22]:

206,500 1213 0.163
2 pffiffiffiffiffiffi
DK eff ¼ 1:12 Dreff pa: ð8Þ
p
under combined LCF/HCF loading, the stress–strain response from
specimen P1B corresponds to the enveloping hysteresis loop. The
3. Experimental details material parameters given in Table 2 were adjusted to the hyster-
esis loops from Fig. 1b.
3.1. Material
3.3. Crack growth measurements
A 10%-chromium steel, which is used for turbine shafts served
as a test material. The ferritic–martensitic microstructure is shown
The fatigue crack length evolution of microcracks was followed
in Fig. 2. Single martensite needles are visible, which can be as long
by using the replica technique. To this end, the tests were inter-
as 300 lm and often are arranged parallely in groups. Those nee-
rupted after increments of 5% of the estimated total lifetime. A
dles are later found to have a significant influence on the initial
small force was applied in order to garantuee open crack faces
stage of microcrack growth.
when the replicas were taken. A acetate foil of 30 lm thickness
was dissolved in propanone solution and was placed on the speci-
3.2. Experimental setup men surface. After drying, the foils, which contain the negative pat-
tern of the crack on the surface of the specimen, were investigated
A servohydraulic machine with a maximum force of 35 kN was by light microscopy. In the following the crack length a denotes
used to perform the LCF and LCF/HCF tests. The fatigue specimens half of the surface crack length observed, since semicircular surface
were cylindrical with a diameter of 7 mm in the gauge length and cracks are typical for cylindrical smooth specimens.
were polished. For displacement measurements an extensometer
with a gauge length of 10 mm was employed. In the case of LCF/
HCF interaction a sinusoidal displacement was superimposed with 4. Results
a frequency of 50 Hz. All tests were strain controlled and the LCF
loadings were conducted under a strain ratio of R = 1 and with 4.1. Microcrack growth under pure LCF loading
a strain rate of 103 s1. An overview over all LCF and LCF/HCF tests
is given in Table 1, where a,LCF and a,HCF denote the strain ampli- Specimen P1B was tested at a strain amplitude of a,LCF = 0.45%.
tudes of the LCF and HCF cycles respectively. The two longest cracks were followed. Two replicas of crack 2 are
Stabilized hysteresis loops from specimens P11A and P1B are shown in Fig. 3 at different stages of the total lifetime. The percent-
shown in Fig. 1b. For specimens P12B and P09A, which were tested age of the total lifetime and the current surface crack length are gi-
ven in the pictures. Crack 2 shows considerable stage 1 crack
growth, reaching a surface crack length of 340 lm after 965
(LCF) cycles before kinking for the first time. In the following, the
number of cycles refers to the number of LCF cycles, unless stated
otherwise. The left side of Fig. 3 illustrates the crack shortly after
the transition from stage 1 to stage 2 crack growth. In the following
the crack deflects several times (see right side of Fig. 3). Crack 1
shows stage 1 crack growth until a surface crack length of
150 lm, followed by a stage 2 regime, which exhibits the same fea-
tures as for crack 2.
The crack length evolution of both cracks is displayed in Fig. 4.
The pronounced stage 1 crack growth of crack 2 is reflected in the
higher crack growth rates until 965 cycles in comparison to crack
1. A kink in the crack length evolution clearly indicates the transi-
tion point from stage 1 to stage 2 crack growth.
The crack growth curves of Fig. 4 are supposed to represent the
growth of individual cracks. However, the replicas reveal that coa-
lescence of microcracks occurs. To illustrate this, two replicas of
crack 1 at different stages of the total lifetime are shown in
Fig. 5. The microcrack, which is visible on the top of the left side
Fig. 2. Ferritic–martensitic microstructure of the 10%-chromium steel, which of the figure coalesces with the main crack during the following
contains randomly distributed martensite needles. cycles. On the right side a detail of the crack-tip of crack 1 is shown
 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202 197

Fig. 6. Accumulated crack length due to crack coalescence. The arrows indicate that
crack coalescence becomes increasingly important towards the end of lifetime. No
data is available at large cycle numbers, since too many cracks coalesce between
Fig. 3. Two replicas of crack 2 from specimen P1B after 1570 cycles (47% Nf) and
two replicas.
2973 cycles (89% Nf). The loading axis is in the horizontal direction. Additionally,
the surface crack lengths are given in the pictures.

Fig. 4. Crack length evolution of two microcracks under LCF conditions for Fig. 7. Fracture surface of specimen P1B after final fracture. The striations are found
specimen P1B and a,LCF = 0.45%. 1 mm below the specimen surface. The arrows indicate the origins of two fatigue
cracks.

at N = 3369 cycles. The crack density is obviously higher at this

stage of the lifetime than on the left side of the figure. The detail
of the crack on the left side of Fig. 5 is not visible on the right side
of the figure.
Every time coalescence of microcracks with the main crack is
observed, the length of the microcracks is measured separately
and is accumulated over the number of cycles. This is done for both
cracks 1 and 2 of specimen P1B. The resulting crack length is re-
fered to as acc (cc: crack coalescence) and is shown in Fig. 6. For
crack 1 no crack coalescence is observed until N = 2023 cycles,
but acc becomes increasingly larger towards the end of the lifetime.
At very large cycle numbers no reliable data is available, since too
many cracks grow together within the interval between two repli-
cas. To indicate this the arrows are shown in Fig. 6. For crack 2
crack coalescence is observed to occur earlier than for crack 1,
but never contributes more than 1/4 to the total crack length in
the regime considered.
Fig. 7 shows the fracture surface of specimen P1B. The fracture
surface is rough but shows parts where striations are clearly visi-
Fig. 5. Crack 1 from specimen P1B at different stages of the total lifetime. Both
ble. The striations shown in the figure originate from a point
pictures show microcracks in the vicinity of the crack-tip. The crack density is much 1 mm below the specimen surface. The arrows mark the origin of
larger on the right side. two fatigue cracks, one of which being crack 1.
198 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202

Fig. 8. Crack length evolution of two microcracks under LCF conditions for
specimen P11A and a,LCF = 0.5%.

Fig. 10. Two replicas of crack 1 from specimen P12B after 1426 cycles (66% Nf) and
2150 cycles (100% Nf). The loading axis is in the horizontal direction. Additionally,
the surface crack lengths are given in the pictures.

Fig. 9. Crack growth rates for all cracks under pure LCF loading. The line with
b = 0.42 illustrates the linearity between da/dN and DCTOD.

Specimen P11A was tested at a strain amplitude of a,LCF = 0.5% Fig. 11. Crack length evolution under LCF/HCF conditions in comparison to pure
under pure LCF loading conditions. The two largest cracks, which LCF loading.
lead to final rupture, are analysed and the crack lengths versus
the number of cycles are shown in Fig. 8. Both cracks nucleate after
only few percent of the total lifetime. Crack 1 grows perpendicu- gest cracks are measured. Examplarily the replicas of crack 1 are
larly to the loading axis throughout its lifetime, whereas crack 2 displayed in Fig. 10. Crack 1 initially grows under stage 1 until a
nucleates at a martensite needle and shows pronounced stage 1 surface crack length of 150 lm (see left side of Fig. 10). In contrast
crack growth until a surface crack length of 300 lm. This again to pure LCF loading, the crack stays almost perfectly in the plane
leads to initially higher crack growth rates for crack 2. perpendicular to the loading axis after the transition to stage 2
Fatigue crack growth rates received from Figs. 4 and 8 are cor- crack growth has occured (compare with right side of Fig. 10).
related in Fig. 9 with DCTOD from Eq. (4). The large scatter in the Crack 2 emerges from three microcracks, which grow together dur-
crack growth rates at low DCTOD values and small crack lengths ing the initial stage 1 period. After crack coalescence the following
is typical for microcracks. Crack 2 from specimen P11A and crack growth is qualitatively the same as for crack 1.
2 from specimen P1B both show pronounced stage 1 crack growth In Fig. 11 the crack length evolutions under pure LCF loading
before growing under stage 2. This is reflected in the crack growth and LCF/HCF interaction are compared. Initially the crack growth
rates, where a drop in da/dN is visible for both cracks at rates are almost identical for both loading conditions if crack 1
DCTOD  6  107 m (full symbols). This drop corresponds to the from specimen P1B is considered. When a crack length of approx-
transition from stage 1 to stage 2 crack growth. To illustrate the imately aonset = 100 lm is reached, the LCF/HCF interaction leads to
linearity between da/dN and DCTOD, Eq. (1) is fitted to the exper- accelerated crack growth and a reduced lifetime in contrast to pure
imental data resulting in b = 0.42. LCF loading. When crack 2 from specimen P1B is considered, the
crack length aonset cannot be identified that easily due to the pro-
4.2. Microcrack growth under combined LCF/HCF loading nounced stage 1 crack growth with initially higher crack growth
rates.
Specimen P12B was tested with the same total strain amplitude Additionally, fractographic analysis are performed. The fracture
a,total = a,LCF + a,HCF = 0.4% + 0.05% as specimen P1B. The two lon- surface of specimen P12B with a semicircular fatigue crack is
 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202 199

Fig. 14. Crack growth rates for LCF and LCF/HCF loading. Additionally, striation
Fig. 12. Fracture surface of specimen P12B tested under LCF/HCF loading. The spacings are shown from specimen P09A.
fracture surface is flat (right and upper left picture) and fatigue striations are visible
(lower left picture), which stem from 1.1 mm below the specimen surface and have
a spacing of approximately 3 lm.

Fig. 15. Adjustment of the model to the experimentally observed crack length
evolution under pure LCF loading conditions for specimen P1B and crack 1.

Fig. 13. Detail of the fracture surface of specimen P09A, which was tested under
combined LCF/HCF loading. Striations are clearly visible in front of the transition to
residual fracture.

shown in Fig. 12 and is much flatter than for pure LCF loading. The
lower left side of Fig. 12 shows fatigue striations from 1.1 mm be-
low the specimen surface with a spacing of approximately 3 lm.
The fracture surface of specimen P09A, which is tested under ex-
actly the same conditions as specimen P12B, is visible in Fig. 13.
No replicas are taken in this case. Striations are clearly visible
and are assumed to stem from one complete loading block. No sub-
structure between the striations is found, which could be attrib-
uted to single HCF-cylces.
Crack growth curves for combined LCF/HCF loading are plotted
in Fig. 14 together with the data from Fig. 9. In the case of the LCF/
HCF loading DCTOD is taken from the enveloping hysteresis loop.
While the LCF/HCF data (open triangles) seems to fall into the scat- Fig. 16. Adjustment of the model to the experimentally observed crack length
terband of the LCF data (black squares) for DCTOD < 3  107 m, evolution under pure LCF loading conditions for specimen P1B and crack 2.
the HCF-cycles for larger DCTOD values lead to crack growth rates,
which are approximately a factor of two higher than under LCF
conditions. In addition, striation spacings from specimen P09A one LCF/HCF loading block. The striation spacings fall into the pic-
are displayed (open circles). At large values of DCTOD one striation ture given by the crack growth rates. Unfortunately, no striations
spacing should correspond to the total crack growth increment of are visible at smaller crack depths. This probably stems from
200 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202

Fig. 17. Adjustment of the model to the experimentally observed crack length Fig. 19. Comparison between model and experiment for LCF/HCF interaction.
evolution under pure LCF loading conditions for specimen P11A and crack 1.

gives a good description for the crack growth curves from speci-
pffiffiffiffiffi
men P12B for a < 100 lm (Fig. 19). Now DK th;eff ¼ 2:6 MPa m is
computed from Eq. (8), using the experimentally observed crack
length of aonset = 100 lm. This corresponds to a threshold in terms
of DCTOD with DCTODth ¼ dn0 DK 2th;eff ð1  m2 Þ=ðErCY Þ ¼ 9:19 
109 m. Here Poisson’s ratio m = 0.3 is used. The exponent from
Eq. (7) is set to p = 0.3, which is a typical value for steels. The model
prediction is shown in Fig. 19. The model is able to describe the
accelerated crack growth rates under combined LCF/HCF loading.
To illustrate the importance of the threshold, the model prediction
pffiffiffiffiffi
for the case that no threshold exists ðDK th;eff ¼ 0 MPa mÞ is also
shown in the figure. The model then predicts rapid crack growth
from the beginning resulting in a considerable underestimation
of the lifetime.

5. Discussion

Fig. 18. Adjustment of the model to the experimentally observed crack length
The microstructure of the ferritic–martensitic 10%-chromium
evolution under pure LCF loading conditions for specimen P11A and crack 2. steel contains long martensite needles, which are randomly dis-
tributed and can be as long as 300 lm (Fig. 2). In the uniaxial fati-
gue tests, some of the martensite needles are oriented under
continuous crack closure such that the information on the fracture approximately 45° to the loading axis and thus experience the
surface is partly destroyed. highest shear stresses. Those needles are likely to be responsible
for the crack initiation and the distinct stage 1 crack growth with
4.3. Prediction of fatigue crack growth under pure LCF loading initially high crack growth rates. In some cases it is observed, that
several microcracks nucleate at parallel martensite needles, which
For pure LCF loading the crack length is integrated from an ini- are situated very close to each other. In the following those micro-
tial crack length a0 to an arbitrary crack length a by using Eqs. (1) cracks grow together and the resulting single crack continues to
and (4). This leads to the exponential crack growth law: grow. Whether the crack growth occurs along the phase bound-
aries of the martensite and the ferrite or in the martensite needle
a ¼ a0 ebðDCTOD=aÞðNN0 Þ ; ð9Þ
itself cannot not be clarified.
a0 is set to the first crack length observed after N0 cycles for each After the transition from stage 1 to stage 2 crack growth, the
crack. Thus, the only unknown in Eq. (9) remains b, which is treated crack growth rate decreases and results in a drop of the da/dN
as a fitting parameter. The results are shown in Figs. 15–18. As long curve (see Fig. 9 and crack 2 from P11A and crack 2 from P1B). It
as stage 1 crack growth is not too pronounced, the model is capable seems likely that the amount of roughness induced crack closure
of describing the observed crack length evolution (see Figs. 15 and increases just after the transition from stage 1 to stage 2 crack
17). If this is not the case the model initially underestimates the growth and thus is responsible for the decreasing crack growth
measured crack lengths (see Figs. 16 and 18). The values for the pro- rates. After the crack has grown for a certain distance, the crack
portionality constant b range from 0.38 to 0.49. growth rates increase again. During the stage 2 regime crack coa-
lescence with microcracks is sometimes observed and often seems
4.4. Prediction of fatigue crack growth under combined LCF/HCF to be responsible for the crack deflection under pure LCF loading. It
loading is shown in Fig. 6 that crack coalescence is not dominant until the
very last stage of the lifetime. During the final stage of the lifetime,
In order to desribe the LCF/HCF test, b is set to 0.49, which cor- the crack density increases considerably, and thus crack coales-
responds to the fitting result for crack 1 and specimen P1B and also cence becomes more and more important.
 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202 201

Under pure LCF loading the fracture surface is rougher than un-
der combined LCF/HCF loading. Obviously superimposed HCF-cy-
cles force the crack to stay on the plane perpendicular to the
loading axis. Striations are observed under both LCF and LCF/HCF
loading conditions. This indicates that crack growth predominantly
occurs by crack-tip blunting and resharpening as is implied by the
DCTOD concept. Striation spacings are measured and compared to
the crack growth rates received from the replicas. For combined
low and high cycle fatigue loading, both crack growth rates and
striation spacings indicate accelerated fatigue crack growth by
approximately a factor of two in comparison to pure low cycle fa-
tigue loading, as soon as a certain crack length is reached. This
shows that fatigue crack growth rates received from the specimen
surface can be related to the evolution of the crack depth in this
case.
The model prediction for pure LCF loading is shown in Figs. 15–
18. The model, which makes use of a linear correlation between da/
dN and DCTOD, is able to describe the microcrack length evolution Fig. 20. Predicted influence of the HCF-frequency on da/dN. DCTOD is evaluated
as long as stage 1 crack growth is not dominant. This is the case for from the enveloping hysteresis loop.
the cracks in Figs. 15 and 17. The agreement of the model and the
experiments is somewhat surprising with regard to the very small
The threshold is only introduced in the HCF-part of Eq. (5), since
crack lengths. Small cracks are often stopped at microstructural
the applied LCF strain amplitudes even for crack lengths <10 lm
barriers such as grain boundaries, because the plastic zone cannot
result in a DCTOD, which is larger than the threshold of
spread into the next grain. Under large plastic strains the plastic
DCTODth = 9.19  109 m (Fig. 9).
zone size is spread over the whole specimen and grain boundaries
It should be noted though that the point when the HCF-cycles
no longer pose barriers for microcracks in this sense. Further the
become active is not clearly visible due to the larger scatter in fa-
model would imply a Paris-law of da/dN / DKm with a Paris-expo-
tigue crack growth rates for small cracks in comparison to typical
nent of m = 2 since DCTOD / DK2. In the literature most Paris-
long crack data. A possible solution would be to increase the total
exponents for metallic materials range between two and four.
number of HCF-cycles during one loading block by either increas-
The 10%-chromium steel obviously has a slope which is close to
ing the HCF frequency or extending the LCF cycle time. The cur-
m = 2, even though the scatter in data is large. For materials with
rently used HCF frequency of 50 Hz with a cycle time of 18 s,
m – 2, the model can be easily extended by introducing an expo-
results in 900 HCF-cycles per loading block. The effect of frequency
nent in Eqs. (1) and (6).
on da/dN is studied in Fig. 20 using the adjusted model. For 1, 10
When stage 1 crack growth becomes more dominant, the model
and 100 HCF-cycles per loading block, the acceleration in compar-
cannot capture the initially fast crack growth. The model does not
ison to pure LCF loading is almost negligible. For fHCF/fLCF = 900,
include any mode 2 components, which cannot be neglected if the
da/dN increases strongly above DCTOD = 3.6  107 m. These
crack grows under 45° to the loading axis. A possible extension of
findings qualitatively agree with the results found in Powell et al.
the model taking into account mode 2 components could be for-
[4], where it was found for long cracks that 1000 HCF-cycles per
mulated as follows: da/dN = b(DCTODI + DCTODII). Additionally,
loading block were enough to clearly identify DKonset, while fHCF/
stage 1 cracks growing on a single slip system should show less
fLCF 6 100 resulted in no measureable increase in fatigue crack
plasticity induced crack closure as stage 2 cracks, since pure shear
growth rates. It has to be emphasized though that all HCF cycles
does not lead to opening components at the crack-tip. The crack
were applied during a dwell time in tension only, whereas in the
opening stress equation by Newman [23] which is employed here,
present study the HCF-cycles are active during the whole LCF-
was developed for cracks under mode 1 loading and plasticity in-
cycle. As was shown by Beck et al. [3] the damaging effect of the
duced crack closure. Thus it does not contain any information
HCF-cycles during the compressive part of a TMF-cycle is only very
about the difference between mode 2 and mode 1 controlled fati-
small. This means that the amount of the damaging HCF-cycles per
gue crack growth. This difference in the crack opening behavior
loading block is less than 1000 in our case. Considering the larger
for stage 1 cracks would result in larger effective cyclic J-integrals
scatter in fatigue crack growth data for small cracks compared to
in Eq. (4) and hence also in larger crack growth rates. In order to
the scatter for long cracks, it has to be concluded that more than
properly describe the drop of da/dN after the transition from stage
1000 HCF-cycles per loading block are desirable in future works
1 to stage 2 due to roughness induced crack closure, complex
to clearly identify the point, when the HCF-cycles become active.
numerical simulations are necessary, which are beyond the scope
The proposed model for LCF/HCF describes the higher crack growth
of this paper.
rates under LCF/HCF interaction. Still the model does not account
All values of the fitting parameter b lie between 0.38 and 0.49
for load interaction effects, which could strongly influence the
and thus have a reasonable value in comparison with physical
crack opening stresses during a complete loading block. The use
models of crack growth by crack-tip blunting. Since b is reasonably
of a modified strip yield model as proposed in [27] or other
constant for different cracks and loading amplitudes, it could be
simulation methods could give valueable information about the
used as a material constant for lifetime prediction models.
development of the crack opening stresses during one cycle.
Under combined LCF/HCF loading accelerated crack growth
rates are observed as soon as a certain crack length is reached.
The model can describe this effect by introducing the effective fa- 6. Conclusions
tigue threshold. The threshold can be roughly estimated from the
crack length evolution and the resulting effective threshold In the present work the evolution of microcracks was followed
pffiffiffiffiffi
DK th;eff ¼ 2:6 MP m corresponds to typical values for steels. With- for a ferritic–martensitic 10%-chromium steel by using the replica
out the introduction of the threshold the model predicts much fas- technique. Several cracks were observed under pure LCF and com-
ter crack growth rates than what is observed in the experiments. bined LCF/HCF loading. An analytical model was presented, which
202 C. Schweizer et al. / International Journal of Fatigue 33 (2011) 194–202

accounts for accelerated crack growth rates under LCF/HCF condi- [4] Powell BE, Duggan TV, Jeal RH. The influence of minor cycles on low cycle
fatigue crack propagation. Int J Fatigue 1982;4:4–14.
tions as soon as a certain crack length is reached.
[5] Powell BE, Duggan TV. Predicting the onset of high cycle fatigue damage: an
The following conclusions can be drawn from this work: engineering application for long crack fatigue threshold data. Int J Fatigue
1986;8:187–94.
 Long martensite needles were often responsible for the crack [6] Powell BE, Hawkyard M, Grabowski L. The growth of cracks in Ti–6Al–4V plate
under combined high and low cycle fatigue. Int J Fatigue 1997;19:167–76.
initiation and the following stage 1 crack growth regime for [7] Hall RF, Powell BE. Crack growth in imi 829 at 550 °C under combined high and
the 10%-chromium steel. low cycle fatigue. Mater High Temp 2002;19:1–8.
 Stage 1 cracks were found to reach surface crack lengths of [8] Ding J, Hall RF, Byrne J, Tong J. Fatigue crack growth from foreign object
damage under combined low and high cycle loading, Part I: experimental
more than 300 lm before kinking for the first time. Those studies. Int J Fatigue 2007;29:1339–49.
cracks showed initially higher crack growth rates than cracks [9] Hawkyard M, Powell BE, Hussey I, Grabowski L. Fatigue crack growth under the
with less pronounced or no stage 1 regimes. After the transition conjoint action of major and minor stress cycles. Fatigue Fract Eng Mater Struct
1996;19:217–27.
to stage 2 a drop in the crack growth rates was observed. [10] Byrne J, Hall RF, Powell BE. Influence of LCF overloads on combined HCF/LCF
 LCF/HCF interaction led to accelerated fatigue crack growth of crack growth. Int J Fatigue 2003;25:827–34.
microcracks in comparison to pure LCF loading as soon as a cer- [11] Neumann P. New experiments concerning the slip processes at propagating
fatigue cracks. Acta Metall 1973;22:1155–65.
tain crack length was reached. This crack length could be iden- [12] Laird C, Smith GC. Crack propagation in high stress fatigue. Philos Mag
tified with the point when the stress intensity of the HCF-cycles 1962;7:847–57.
exceeded the threshold of fatigue crack growth for the first [13] Pelloux RMN. Crack extension by alternating shear. Eng Fract Mech
pffiffiffiffiffi 1970;1:697–704.
time. The resulting effective threshold of DK th;eff ¼ 2:6 MPa m
[14] Shih CF. Relationships between the J-integral and the crack opening
corresponds to typical values for steels. displacement for stationary and extending cracks. J Mech Phys Solids
 The proposed model, which makes use of a linear correlation 1981;29(4):305–26.
between da/dN and DCTOD, can describe the crack length evolu- [15] Hutchinson JW. Singular behaviour at the end of a tensile crack in a hardening
material. J Mech Phys Solids 1968;16:13–31.
tion under pure LCF loading if stage 1 crack growth is not too [16] Rice JR, Rosengren GF. Plane strain deformation near a crack tip in a power-law
pronounced. hardening material. J Mech Phys Solids 1968;16:1–12.
 The model can describe microcrack growth under LCF/HCF con- [17] Shih CF. Tables of Hutchinson-Rice-Rosengren singular field quantities. Tech.
rep. Brown University Report MRL E-147; 1983.
ditions and only needs few fitting parameters. [18] Wuethrich C. The extension of the J-integral concept to fatigue cracks. Int J
Fract 1982;20:35–7.
[19] Kumar V, German MD, Shih CF. An engineering approach for elastic–plastic
fracture analysis. Tech. rep. Report NP-1931 on Project 1237-1 for Electric
Acknowledgments Power Research Institute, Palo Alto, California; 1983.
[20] Heitmann HH, Vehoff H, Neumann P. Advances in fracture research 84. In:
The authors thank the Stiftung Stahlanwendungsforschung im Valluri SR, et al., editor. Proc of ICF6, vol. 5. Oxford and New York: Pergamon
Press Ltd.; 1984. p. 3599–606.
Stifterverband für die Deutsche Wissenschaft e. V. for the financial [21] He MY, Hutchinson JW. The penny-shaped crack and the plane strain crack in
support of the present research. Special thanks to the study-group an infinite body of power-law material. J Appl Mech 1981;48:830–40.
of the Forschungsvereinigung Verbrennungskraftmaschinen e. V. [22] Riedel H. Fracture at high temperatures. 1st ed. Berlin, Heidelberg, New
York: Springer-Verlag; 1987.
(FVV) for continuous and helpful discussions.
[23] Newman JC. A crack opening stress equation for fatigue crack growth. Int J
Fract 1984;24:131–5.
References [24] Schijve J. Some formulas for the crack opening stress level. Eng Fract Mech
1981;14:461–5.
[1] Moalla M, Lang K-H, Loehe D. Effect of superimposed high cycle fatigue [25] Vormwald M, Seeger T. The consequences of short crack closure on fatigue
loadings on the out-of-phase thermal-mechanical fatigue behaviour of crack growth under variable amplitude loading. Fatigue Fract Eng Mater Struct
CoCr22Ni22W14. Mater Sci Eng A 2001;319–312:647–51. 1991;14:205–25.
[2] Beck T, Loehe D, Luft J, Henne I. Damage mechanisms of cast Al–Si–Mg alloys [26] Schweizer C, Seifert T, Riedel H. Simulation of fatigue crack growth under large
under superimposed thermal-mechanical fatigue and high-cycle fatigue scale yielding conditions. J Phys: Conf Ser 2010;240:012043.
loading. Mater Sci Eng A 2007;468–470:184–92. [27] Newman JC. A crack-closure model for predicting fatigue crack growth under
[3] Beck T, Henne I, Loehe D. Lifetime of cast AlSi6Cu4 under superimposed aircraft spectrum loading. In: Chang JB, Hudson CM, editors. Methods and
thermal-mechanical fatigue and high-cycle fatigue loading. Mater Sci Eng A models for predicting fatigue crack growth under random loading, ASTM STP
2008;483–484:382–6. 748. American Society for Testing of Materials; 1981. p. 53–84.