Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

145

(compare C and B in Figure 5.10); during compression the closed crack is no


stress raiser and therefore all compressive stresses and strains are elastic (elastic
strains are negligible as compared to plastic strains).

5.7. Retardation models


The complexities of the retardation phenomenon so far have precluded the
development of an all-encompassing mathematical-physical treatment of the
problem in a retardation model. More than half a dozen models have been
proposed [1-7], none of which covers all aspects of the problem.
A number of models are based on crack closure. As was explained in the
previous section, crack tip closure occurs even in constant amplitude loading.
Some closure models consider only that part of the cycle effective over which the
crack is open. However, even during that part of the cycle the crack tip material
is straining (B-C in Figure 5.13) and this straining is just as much part of the
crack-tip strain loop as is the part associated with the open crack. Hence,
considering only that part of the cycle during which the crack is open, is
physically incorrect. Certainly, closure changes after an overload and this affects
the general residual stress field at the crack tip, but this residual stress field is
changed more due to the larger plastic zone. In view of this, pure closure models
do not cover all aspects of the problem. There is no doubt that overloads affect
closure; hence there will be a correlation between closure and retardation. But
retardation and changes in closure are both consequences of the overload.
Closure changes are not the cause of retardation. Both are symptoms, and one
symptom cannot explain another; both are caused by the 'disease'. A proper
treatment of the model must consider the effect of overloads upon the total
residual stress field in the wake of the crack and ahead of the crack, and in
particular upon the plastic strain range.
Other retardation models attempt to account for the residual stress field
directly by superposing it on the stress field due to the external load. The
residual stresses themselves cause a certain stress intensity which can be added
to the stress intensity due to the applied loads. Such models have several
draw-backs and short-comings. In the first place, although the residual stress
field can be assessed qualitatively in an easy manner, the quantitative evaluation
(and thus evaluation of the resulting K) is difficult. A second problem is, that
a residual stress field already exists even in constant amplitude loading. Also this
field should be accounted for if the model is to be based on residual stresses.
(Normally, it will be automatically accounted for in the data base).
Despite the mathematical complexity of some, all models are two-dimension-
al over-simplifications of a complicated three-dimensional problem, and full of
assumptions. For example, all models must consider plastic deformation, but
even plastic deformation is not easily treated analytically, further the yield
146

strength, F;y, is arbitrarily defined (e.g. 0.2% plastic strain), while it is an


essential number in the calculation of the plastic zone. Plastic zone size
equations by themselves are subject to doubt, especially in the case of a changing
state of stress. Some models use empirical equations for changes of the closure
stress.
All of this leads to the conclusion that any model must contain one or several
unknown parameters which must be obtained from tests (i.e. adjusted empiric-
ally). In one model such a parameter was later included to make empirical
adjustment possible in the first place. There is no practical objection against the
use of empirical parameters (E, v and F;y follow from experiments and so does
after all da/dN), provided the use of these empirical parameters leads to useful
crack growth predictions for structural applications.
If all models are simplifications, none can be preferred over another.
Moreover, if they all contain adjustable parameters, they can all be made to
work if the parameters are adjusted appropriately. All claims that one model is
better than another are improper. Each model can be made to work if empiric-
ally adjusted; ifit does not work, it was not adjusted properly. Generality of the
adjustment may be a problem. In that respect some models may be somewhat
better than others. Clearly, the adjustment parameters will be material
dependent. But should they also depend strongly upon the stress history, as they
do (see Chapters 6, 7) then they cannot be used generally. Attempts to make
general use of these parameters then lead to false claims with regard to a model's
inadequacy.
There are so many retardation models that a discussion of all would be
beyond the scope of this book. Review of just a few might suggest that these
would be better than others. The interested reader is therefore referred to the
relevant literature. The general reader probably has no use for such a review at
all, because crack growth analysis with retardation requires the use of a
computer anyway and the general user will employ existing software. Such
software should have the option for the use of several retardation models. If just
one model is available generality may be hampered, because one model may be
somewhat more appropriate for certain applications and vice versa. Which
models are included in the software is rather secondary, because all can be made
to work if properly calibrated.
More important than which retardation model is used, may be the following:
(a) As demonstrated in the previous section retardation will depend upon the
relative stiffness of plastic zone and elastic material. This may become important
for larger cracks in small components. No known retardation model considers
this problem but some account could be made in the computer code.
(b) All models must consider crack tip plasticity and therefore will use
arbitrary numbers. As was explained in Chapters 2 and 3, the state of stress has
a great influence on plasticity. The plastic zone is rp = K!ax/rx.nF;; where rx. is
commonly considered to be 2 for plane stress and 6 for plane strain. Further,
147

the condition for plane strain is B > 2.5K~ax/Ft~, in which 2.5 is a somewhat
arbitrary number (Chapter 3). 'Normal' stress cycles at low Kmax may give plane
strain, but an overload may cause plane stress due to its larger Kmax. This also
causes a more extensive residual stress field and more retardation. Thus, in
accounting for retardation the computer code should assess the state of stress
in each cycle. If it does not, even the 'so-called' sophisticated retardation model
may give large errors. For example, the larger retardation at longer cracks in
Figure 5.10 may be caused by a change from plane strain to plane stress during
the overloads. But even if the state of stress is evaluated, the value of a will be
arbitrary, the factor of 2.5 is arbitrary and F,y is more or less arbitrary. Some
models use different plastic zone formulations than above, but these still contain
arbitrary numbers.
In the following crack growth analysis for variable amplitude loading will be
illustrated on the basis of the Wheeler model. This model is used here not
because it is believed to be better than any other, but because it is very simple,
so that it can be used easily for illustrations. It is worth mentioning however that
if all models are simplifications anyway, the simplest certainly is the most
appealing. If all models must be calibrated for general use, even the simplest
model can be made to work by calibration.
Wheeler introduces a retardation parameter ¢ R' It is based on the ratio of the
current plastic zone size and the size of the plastic enclave formed by an
overload (Figure 5.17a). An overload occurring at a crack, size ao will cause a
crack tip plastic zone of size
K~ p2(JJao
rPO = --2 = --2-' (5.20)
anF,y aF,y
where (Jo is the overload stress. When the crack has propagated further to a
length ai the current plastic zone size will be
p2(Jimax ai
(5.21)
aF,~
where (Jimax is the maximum stress in the i-th cycle. This plastic zone is still
embedded in the plastic enclave of the overload: the latter proceeds over a
distance g in front of the current crack (Figure 5.17b). Wheeler assumes that the
retardation factor ¢R will be a power function of rpc/g. Since
g = ao + rPO - ai' the assumption amounts to:
da
dN
with (5.22)
148

Figure 5.17. The model of Wheeler. (a) Situation immediately after overload; (b) After some crack
growth; (c) Situation after second overload.

If rpc = {!, the crack has grown through the overload plastic zone, and the
retardation factor becomes <p R = I by definition. The exponent in Equation
(5.22) has to be determined empirically. This is the adjustable calibration factor.
°
Note that if y = there will be no retardation at all under any circumstances.

°
Hence, the minimum value ofy is zero. Typically for variable amplitude loading
< y < 2, depending upon the material but also upon the spectrum.
For the case of a single overload in a constant amplitude test the retardation
factor gradually decreases to unity while the crack progresses through the plastic
enclave (Figure 15.17b). If a second high load occurs, producing a plastic zone
extending beyond the border of the existing plastic enclave, the boundary of this
new plastic zone will have to be used in the equation (Figure 5.17c), and the
instantaneous crack length will then become the new ao.
Calibration of the above model (and all other models) proceeds as follows. A
test is performed under variable amplitude loading. The test result is then
're-predicted' several times using the proper dajdN - 11K data and the proper
/3, but with different values of the adjustable parameters; (in the case of Wheeler
y-values taken are e.g. 0,0.5, 1, 1.5 etc.). The parameter value(s) that produce(s)
the best coverage of the test data, is (are) the values to be used in analysis. An
149

example of such a calibration [8] is shown in Figure 5.18. Clearly, in this case
}' = 1.4 is the parameter value to be used.
Unfortunately, the parameter calibration is not general. It depends upon the
load-history and spectrum (Chapters 6 and 7). A different spectrum with a
different mixture of high and low loads requires a different calibration factor.
E.g. the non-linear man-induced exceedance diagram (Chapter 6) requires
different calibration parameter(s) than a nature-induced log-linear exceedance
diagram; the calibration parameters are suitable for one type of spectrum, but
they cannot be generalized for all spectra. Failure to perform this re-calibration
and subsequent general use of calibration factors, is the main cause of claims
that one retardation model is better than another. If proper calibration is
performed for each spectrum type, any model can be as good as any other.
Calibration for a certain spectrum type and material generally gives good results
[8] for all variations of the same type of spectrum (Chapter 6), as shown in
Figure 5.19. This figure shows results of about 70 predictions for random
loading with the spectrum and calibration as used in Figure 5.18. More informa-
tion on model calibration can be found in Chapter 7.

5.8. Crack growth analysis for variable amplitude loading


Most structures are subjected to variable amplitude loading, i.e. of the type
shown in Figure 1.2 (For a detailed discussion of load histories see Chapter 6).
In such cases the crack growth rate dajdN varies from cycle to cycle, depending
upon AK and R of the cycle involved, and upon retardation (any cycle can be

28
lin)
O. TEST DATA

2.5

OL-------l0~O-O------2~OO-O------3~OO-O------4~O~OO------5-0~OO-----F-LlG~HTS

Figure 5.18. Calibration of Wheeler model for flight-by-flight aircraft simulation loading [8).
150
25~----------------------,

20

15

10

OL-L-LL~UL~~~~~~~~
0.4 0.5 0.6 0.7 O.S 0.9 1.0

NpREDICTED
N TEST

Figure 5.19. Accuracy of 70 predictions as compared to tests [8], Courtesy Engineering Fracture
Mechanics.
an overload cycle with respect to the subsequent cycle). Therefore, the computer
program must evaluate and add crack growth on a cycle-by-cycle basis. For
example if crack growth takes 200000 cycles, the computer must perform the
'operation' 200000 times. This may take considerable computer time. Typically,
a mainframe computer can perform at about 50000 cycles a minute, so that the
above computation would take about four minutes. A personal computer might
take as much as two to three hours for the same job (1987).
A logic diagram for the integration is shown in Figure 5.20. This is again
based on the Wheeler model because of the latter's simplicity, but it is not
essentially different for other retardation models. Naturally, in order to perform
the calculation for a structural crack, the computer must be provided with
applicable da/dN data for the material at hand (Chapter 7), and last but not
least, the stress history for the structure (Chapter 6).
A more detailed representation of the algorithm involved is shown in Table
151

°i+I-Oj
Ni+I-N j
"'i+I-O'j

Figure 5.20. Logic diagram for crack-growth computation.


5.2. A hand-calculation for a few successive cycles is shown in Table 5.3. These
tables show the basic algorithm which is quite simple; a useful computer code
with many options, especially for the complicated book keeping for stress
histories, is rather involved and will contain approximately 3000 statements
(lines ).
The accuracy of the computation depends somewhat on the retardation
model, but useful results can be obtained with any well-calibrated model. Most
influential to the accuracy are the input of the stress sequence (Chapter 6) and
material data (Chapter 7). A general discussion of accuracy and errors is
presented in Chapter 12. The accuracy problem involved in the simple algorithm
of Table 5.2 is in the addition of a very small da to a relatively large a. For
example, 8-bit personal computers evaluate a number into 8 digits. If a = I
inch and da in a given cycle is 0.000001 inch, the new crack size will become
152

Table 5.2. Crack growth analysis in variable amplitude

Subroutine or preprocessor; fJ-Library or methods of Chapter 8

Subroutine; generation of stress history (Chapter 6), random-


ization, counting if necessary (Chapter 6).

Kmax = I1K;/(1 - R;)


rp; = K~",/(1.nF,~, +--- Subroutine state of stress for (1..

if Xe new > X eo1d then


(! = Xe - a j

Subroutine various retardation models. Only Wheeler shown


l/!R; = (!/rp;
da
-
dN = {(11K R)
' , Subroutine or data library

da da
dN = l/!Rj dN (retarded)

I1N = 1
l1a = I x da/dN

N=N+I
if ai < ap then return to line 2
a, N
a (years, voyages, flights) Subroutine conversion
Output Subroutines
Plots Subroutines
End

a = 1.0000001. However, if da = 0.000,0001, the computer evaluates


1 + 0.000,00001 = 1, i.e. the crack has not grown due to a computer rounding
error. This means that a and da must be evaluated in double precision which in
an 8-bit personal computer provides 16 digits. Hence a = landda = IE - 15
Tuble 5.3. Example of retarded crack growth by hand calculation: on the basis of Table 5.2.
Note a: fi assumed equal to 1 throughout; b: (X = 2 (plane strain assumed); c: dujdN = 2£-9 flK2 Km " assumed: y assumed as 1.5; F,y = 50 ksi;
d: Stress history assumed as in columns 3 and 4; e: Table must be worked horizontally.

2 3 4 5 6 7 8 9 10 II 12 \3 14 15 16
Cycle a fl(T R flK KmJx r ,Ji Xe new X vo1d Xe Q 1 (dajdN) Retarded N fla a = a + flu
= 1 dajdN new a for
next line
0.1 10 0 5.61 5.61 0.0020 0.10020 - 0.10020 0.0020 3.53£-7 1 3.53E-7 0.100000353
2 0.1 ' 10 0 5.61 5.61 0.0020 0.10020 0.10020 0.10020 0.0020 3.53£-7 2 3.53 £-7 0.100000706
3 0.1 + 15 0.1 8.42 9.35 0.0056 0.10056 0.10020 0.10056 0.0056 1.32£-6 3 1.32 £-6 0.100002026
4 0.1 + 10 0 5.61 5.61 0.0020 0.10020 0.10056 0.10056 0.0056 0.21 7.41 £-8 4 7.41 £-8 0.100002100
5 0.1+ 12 0 6.73 6.73 0.0029 0.10029 0.10056 0.10056 0.0056 0.37 2.25E-7 5 2.25£-7 0.10002325
6 0.1 + 10 0 5.61 5.61 0.0020 0.10020 0.10056 0.10056 0.0056 0.21 7.41 £-8 6 7.41 £-8 0.100002399
7 0.1 + 19 0 10.66 10.66 0.0072 0.10072 0.10056 0.10072 0.0072 2.42.£-6 7 2.42£-6 0.100004822
8 0.1 + 15 0.1 8.42 9.35 0.0056 0.10056 0.10072 0.10072 0.0072 0.69 9.11 £-7 8 9.11 £-7 0.100005733
9 0.1 j 10 0 5.61 5.61 0.0020 0.10020 0.10072 0.10072 0.0072 0.15 5.17 £-8 9 5.17 £-8 0.100005785
10 0.1 + 12 0 6.73 6.73 0.0029 0.10029 0.10072 0.10072 0.0072 0.26 1.59 £-7 10 1.59 £-7 0.100005944

Vl
W
154
will be evaluated correctly, but if da = IE - 16 and a = 1, the computer will
still not recognize growth. There is nothing that can be done about this rounding
problem. Usually it is not serious, but it may become a problem in evaluating
retardation. Computer programs working in single precision may be one cause
of claims regarding the accuracy of retardation models.
It should be noted, that the above does not change when a is evaluated in
meters. For example if a = 0.01 m and da = l-EI6m, the addition will be
performed properly in double precision because leading zeros do not count.
Mainframe computers already carry 16 decimals in single precision; they carry
32 in double precision. Even in that case rounding errors may occur, but they
are even less significant.
Although the algorithm in Table 5.2 is simple, computer codes are generally
rather complicated [e.g. 9], because there must be
(a) preprocessors for p, or a p library.
(b) options for various rate equations, da/dN table and/or library of data;
(c) options for various retardation models;
(d) accounting for state of stress;
(e) options for random loading;
(f) accounting procedures for stress history;
(g) options for cycle counting.
If all of the above are included, the main code of 3000 statements as
mentioned above can easily triple or quadruple in size. Of the above (e), (f) and
(g) may be the most important and most involved; they are discussed separately
in Chapter 6. Further discussions of the subject are found in Chapter 7 (data and
calibration) and Chapter 12 (errors and accuracy).
It is well-known that fatigue predictions, in general, have a low accuracy. In
the case of crack propagation, a linear integration (without interaction effects
or retardation) will generally yield results which are on the safe side. As shown
in Figure 5.10 negative loads reduce the retardation caused by positive loads,
but the net effect is usually a deceleration of crack growth, so that retardation
models must be used.
Figure 5.21 shows results of crack growth in rail steel [10] under simulated
train-by-train (Chapter 6) loading. Retardation hardly plays a role in rail steels,
therefore predictions were made by means of linear integration. The figure
shows that they are within a factor 2 of the experimental data.
Better accuracy can be obtained in general, provided the retardation model
is adjusted. Predicted crack growth for a titanium alloy sUbjected to aircraft
service loading [8] is shown in Figure 5.22 together with experimental data.
Generally, part of the discrepancy between computation and test may be caused
by scatter in crack growth properties. Most retardation models can be empiric-
ally adjusted. In this respect, the Wheeler model is attractive, because it contains
only one adjustable constant.
155
42.5~---------------------------,


40.0 •

37.5 •

E
E 35.0 •
en
~ •
'" e 32.5 •
<.)

30.0 •

25.0L-_-L_ _...L.._ ___'I..-_....L._ _.l-_---':~-....L---I...--....L.---J


o 2 4 6 20
Million Gross Tons
Figure 5.21. Predictions and test data for service simulation loading in rail steel [10].

2.5

v •• LL=55 ksi ...


Test doto Ti-6AI-4V, Spectrum 8

0 LL =60 ksi 60 ksi


• 65 ksi 55 ksi
LL=65 ksi

0
2.0 " ... LL=70 ksi •
- - Wheeler, "j'=1.6

~
. v
.~ 1.5
C
N

.g
V

(f)

'"eu 1.0 v
v
<.)

vv
v
0.5

oL-_ _ _ _ _L-________L-__________I..-________ ___'~ ___________'

o 1000· 2000 3000 4000 5000


Flights

Figure 5.22. Predicted crack growth and test data for aircraft spectrum [8]. Courtesy Engineering
Fracture Mech.

Apparently, a crack growth prediction can be substantially more accurate


than a fatigue life prediction. Admittedly, a few experiments are necessary with
a spectrum of certain shape to empirically adjust the retardation model
parameters. From then on, predictions can be made for the same general
spectrum shape and variations thereof, for structural parts subjected to lower
156

and higher stresses and for cracks of different types (different fJ). Crack growth
properties of most materials show considerable scatter. The rail steels that are
the subject of Figure 5.21 showed variations of almost a factor often in constant
amplitude crack growth (see Chapter 14). Therefore, discrepancies between
predicted and experimental crack growth are not a shortcoming of the predictive
method per se, but are due to anomalies in material behaviour.
Fortunately, most materials are well-behaved in comparison, by showing less
scatter in crack growth. Nevertheless, there is enough scatter that predictions
will always have some uncertainty. This would still be the case if better retarda-
tion models were developed.
However, the prediction procedure in general contains many more uncertain-
ties, which may be just as detrimental to the final results as are the shortcomings
of the retardation model. These are:
(a) Uncertainty in the local stress level.
(b) Uncertainty in the stress intensity calculation.
(c) Insufficient knowledge of the load spectrum.
(d) Possible environmental effects.
Consider first the uncertainty in stress level and stress intensity. In the case of
a complex structure consisting of many elements, an error of five percent in the
stress analysis would be quite normal. The subsequent determination of the
stress intensity can easily add another five percent, especially in the case of
corner cracks or surface flaws. Thus, the final inaccuracy of the stress intensity
may be in the order of 10%. If the crack growth rate is roughly proportional to
the fourth power of 11K, the error in the crack growth prediction will be on the
order of (1.1)4 = 1.45 (45%); see also Chapter 12.
Despite extensive load measurements, the prediction of the load spectrum is
still an uncertain projection in the future. Slight misjudgements of the spectrum
can have a large effect on crack growth.
Even if possible environmental effects are disregarded, the errors in crack
growth prediction due to uncertainties in stress analysis and loads analysis can
be just as large or larger than the errors due to the crack growth integration.
Development of better crack growth integration techniques will not improve
this situation. Therefore, the shortcomings of the retardation models can hardly
be used as an argument against crack growth predictions.
Taking into account all errors that can enter throughout the analysis, it is
obvious that a safety factor should be used. This safety factor should not be
taken on loads or stresses or dajdN data. Doing this would make some
predictions more conservative than others. The complexity of crack growth
behaviour does not permit an easy assessment of the degree of conservatism
attained through the application of such safety factors. A safety factor should
rather be applied to the final result, i.e. to the crack growth curve, by dividing
157

the number of cycles to any given crack size by a constant factor. The problem
of accuracy and sources of error is discussed further in Chapter 12.

5.9. Parameters affecting fatigue crack growth rates

When predictions of crack propagation have to be made, data should be


available relevant to the conditions prevailing in service. Such data may be hard
to find (for pragmatic solutions see Chapter 7). Fatigue crack propagation is
affected by an endless number of parameters, and the circumstances during the
test will seldom be the same as in service. The influence of the environment is
the most conspicuous.
The effect of environment on crack growth rates has been the subject of many
investigations on a variety of materials; the rate of fatigue crack propagation in
wet air may be an order of magnitude higher than in vacuum, the effect being
attributed to water vapour. The influence of salt water (seawater) is of particular
interest to marine structures. An example of its effect will be shown in Chapter
7. It is generally accounted for in the crack growth analysis by submitting the
computer program with the actual data in tabular form. It is then assumed that
during each cycle in a variable amplitude sequence the rate will immediately
adjust to the one found in the constant amplitude test data at the same 11K. This
is assuming that chemical/load equilibrium will be immediately attained. More
elaborate accounting can be implemented however. No single model can explain
the influence of the environment on the rate of propagation of fatigue cracks.
Different explanations apply to different materials. The effect is certainly a result
of corrosive action and as such it is time-dependent. Therefore, the environmental
effect is dependent upon the cycling frequency.
Among the many factors that affect crack propagation, the following should
be taken into consideration for crack growth predictions:
(a) thickness;
(b) type of product;
(c) heat treatment;
(d) cold deformation;
(e) temperature;
(f) manufacturer;
(g) batch-to-batch variation;
(h) environment and frequency.
For the factors lower in this list it is less likely that they can be properly
accounted for. No attempt will be made to illustrate the effects of all these factors
with data, because some have greatly different effects on different materials.
Many of these effects cannot be accounted for properly in the analysis of
structural cracks, primarily because the data are simply not available. A
pragmatic approach to solve the problem is discussed in Chapter 7. At this point
158
It IS sufficient to note that the necessary use of estimated data may be of
considerable influence on the accuracy of the analysis. With this in mind, the
acclaimed inaccuracies of e.g. retardation models may well become secondary
(see also Chapter 12).
In sheets there is a systematic, effect of thickness on crack propagation,
especially before the fracture mode transition. Fatigue cracks in sheets start
perpendicular to the sheet surface. When the crack grows the size of the plastic
zone increases and plane stress develops. This causes the fatigue crack to change
to single or double shear, as depicted in Figure 5.23. Plane stress develops when
the size of the plastic zone is in the order of the sheet thickness (Chapter 3). In
thi"ker sheets the transition will require a large plastic zone and occur at a
greater length of crack. The data suggest that crack growth is slower in plane
stress than in plane strain at the same stress intensity.
Although the effect of thickness on crack growth has been recognized for over
20 years, little effort has been expended in developing a useful model for
everyday damage tolerance analyses. Figure 5.24 emphasizes the necessity to
include this effect in the crack growth analysis. A tentative semi-empirical model
has been proposed [11], but the best way to account for thickness effects is
probably to submit the proper data to the computer code. A factor of two error
(due to thickness) in da/dN data may overshadow any effects of 'inaccurate'
retardation models.
Many investigators hold that there is a threshold for fatigue cracking: below
a certain 11K the rate da/dN is supposed to be essentially zero. At least one
conference, resulting in a two-volume book [12] was devoted to this subject. The
threshold would be reflected in a vertical da/dN - 11K curve at low 11K, as
shown in Figure 5.25. The threshold is usually determined by gradually
decreasing the stress in a test until crack growth comes to a halt. In view of
possible retardation this procedure is subject to some doubt. Besides, the
threshold is definitely crack size dependent; it is not unique. However, if one
accepts the presence of a threshold, the practical question is "what is the effect
on predicted crack growth". Figure 5.26 shows what it may amount to: the two
curves are indistinguishable. Generally speaking, the effect is hardly worthwhile
considering, but of course, in each case one would have to use judgement. If for
example the initial 11K is below threshold no growth occurs at all. In this respect
Figure 5.26 is somewhat deceiving; one could select a case where the effect is
larger. However in random loading many cycles will be above threshold and the
effect on life (in years or hours) is small, especially when there is retardation.
Most computer codes provide an option to use a threshold. Referring back to
Section 5.8 it should be noted that the computer uses a threshold automatically
because it rounds any da smaller than a certain value, depending upon the
current crack size.
There is often much concern regarding the anomalous behaviour of very small
159

final
failure

tensile mode

(b)

Figure 5.23. The transition of a fatigue crack in sheet. (a) Transition of fatigue cracks to double
shear (top) and single shear (center and bottom) in AI-alloy specimens; (b) Single shear (A) and
double shear (B).

cracks [13]. As shown in Figure 5.27 small cracks tend to show growth rates
much higher than would be expected on the basis of the acting 11K. Various
explanations have been put forward but the 'cure' proposed is mainly artificial
use of an apparent crack size a + ) .where A is a fixed quantity determined
empirically.
Most of the short crack data stem from strain control fatigue tests at R = - I
160

i
30.----------------------
crock size (mm)

sheet
thickness 0.6 mm
20

10

sm =8 kg/mm2

sa = 6.5 kg/ mm 2

10 20 30 40
___ N (103 cycleS)
Figure 5.24. Effect of sheet thickness on crack growth.

E
dN

(1-R)K 1C !;K

Figure 5.25. Threshold in rates (schematic).

on small notched coupons (usually with central holes). Consider a short crack
at such a hole as in Figure 5.28a and compare it with a long crack at the same
/).K (Figure 5.28b). By the nature of the test, the plastic zone at the hole is much
larger than the crack tip plastic zone. Since completely reversed plastic strain is
enforced in most of these fatigue tests, the crack tip will also be subjected to
completely reversed strain (R, = -1). A larger crack at R = -1 will close
during compressive loading, so that its strain range will still be as if R, = O. The
small crack at the hole is experiencing a strain range twice as large as a regular
crack at Ra = O. Hence, the small crack should show a rate of growth as if its
/).Kwere approximately twice as large as the calculated value. This is indeed the
161

4.5
U;
W
I
u 4.0
z
SOLID LINE: PARIS: COEFnCIENT 2£-9: EXPONENT 3.5

w 3.5 CASH-DOT SAME WITH THRESHOLO OF 5 KSI RT IN


N
U1
CENTER CR .... CK: W - 6 INCH; DELTA-SIGM.-\ - 12 KSJ; R 0
3.0 m

""
u
-<
rr
u 2.5

2.0

1.5

1.0

0.5

10 20 30 40 50 60 70 80 90
LIFE (1000 CYCLES)
Figure 5.26. Effect of using threshold in computation; threshold is 5 ksi; Curves indistinguishable.

case as can be judged from Figure 5.27. When the small crack grows, it will
gradually move away from completely reversed straining so that the growth rate
decreases. Once the crack tip is outside the plastic zone of the hole, the small
crack behaves as a normal crack.
The small crack behaviour depends upon the loading and upon the ratio
between ligament and hole diameter (notch depth). A hole, unlike a crack, does
not close in tension and it is equally as much a stress raiser in compression as
in tension, so that even under load control reverse plastic strains can occur.
Therefore, similar, but smaller, effects should be anticipated under general
loading conditions. However, if the hole is filled with a fastener, it will essentially
'close' and little or no small crack effects should be anticipated. In technical
problems many cracks start from holes, but these are often "filled" fastener holes,
so that the 'small crack problem' may be of little relevance. In most other cases
these so-called 'small cracks' are well below the detection limit in practical
inspections. They are then irrelevant because crack growth below the detectable
crack size is hardly of interest. Naturally, one can always give examples of cases
where the problem might appear, but generally speaking, it is an interesting
research subject, but its technical relevance is small.
As a final note in this section consider the use of J (Chapter 4) as opposed to
K for representing da/dN data. By far the larger part of fatigue crack growth
lives is spent at low AK. Let e.g. AK = IOksiy'iil and R = O. Taking a (low)
yield strength of F;sy = 40 ksi, the plastic size would be r = 10/(6 x
n x 402 ) = 0.003 in. If this is not a small enough plastic zone, none will ever be.
162

Kmolil ' ksi../i"-


10

Growth Rate versus Kmaa in


SAE 1015 St~1

Stress and Edge Strain


Control: R •• R".=-I

Rc=-I -K, R.. =-I


0 2_5
0 4.4 <>
-I A 6_2 V
10

Note: Solid symbols denote cracks


~ 01 length less than 250l'm
~

~ -2
10
Z
'I!
..,
0

!F
~
: 10-3
~
'"~
u

-<
10

-5
10

10 '00
K""",. MPa,/iii

Figure 5.27. Crack growth rates of small cracks [13] (Courtesy EMAS).

Naturally, at large crack sizes the plastic zone will be larger, but does this justify
the use of EPFM. Consider Figure 5.29. The crack growth life might be either
N], N2 or N 3 , but practically this makes little difference on the total. The
difference is only caused at the high AK. During most of the crack growth life,
the stress intensity is well below Klc (fortunately). Thus, the use of K for the
analysis is well justified. Finally, the champions of using J for representing
da/dN data have always used approximations of J rather than actual J values
obtained properly (Chapter 4). By using 'appropriate' approximations, one can
always make data look better. However, should the data at high AK indeed be
better represented by J, the differences in analysis results (Figure 5.29) would
still not justify the use of a more complicated parameter; a parameter moreover
that has more drawbacks than advantages (see Chapter 4). In short, the use of
163

Notch zone
~crock lip zone ---e
Crock lip
Plastic zones Plaslic zone

E E

Figure 5.28. Stress-strain loop for small crack at notch. (left) as compared to regular (large crack;
right); both R = 0 loading. [7, 13]. Copyright ASTM. Reprinted with permission.

a
I J
II
I I
/,
1/
/1.
/-/.
.;<"

Figure 5.29. Effect on life of using different parameter in high rates regime.

J for fatigue crack growth analysis is another interesting research subject, but
not of practical interest.

5.10. Stress corrosion cracking

Crack growth can occur by other mechanisms than fatigue (Chapter 1). The
most important one is stress corrosion cracking. Given a specific material-
environment interaction the stress corrosion cracking rate is governed by the
stress intensity factor. Specimens with the same initial crack but loaded at

You might also like