Anda di halaman 1dari 16

hhgfneering Frachm Mechanics Vol. 40, No. 2, pp.

355-370, 1991
Printed in GreatBritain.

0013-7944/91 $3.00 + 0.00


Pcrgamon F%%pie.

APPLICATION OF THE CYCLIC J-INTEGRAL


TO
FATIGUE CRACK PROPAGATION
OF Al 2024-T351
LESLIE BANKS-SILLS and YEHUDA VOLPERT
The Eda and Jaime David Dresxer Fracture Mechanics Laboratory, Department of Mechanics.
Materials and Structures, Faculty of Engineering, Tel Aviv University, 69978 Ramat Aviv, Israel
Ah&act-In this investigation, an experimental/numerical study of the cyclic J-integral, AJ, is
carried out in order to more clearly explain the meaning of this parameter for fatigue crack growth
studies. Constant amplitude fatigue tests with two R-ratios (R = 0.05 and R u 0.5) are performed
on compact tension specimens fabricated from Al 2024-T351. A simulation of tests is carried out
numerically by means of tbe tinite element method with the material modeled to be elasto-plastic.
From the numerical results, values of AJ are calculated from both a path independent integral and
load vs load-line displacement data. Comparisons between these values, as well as to those
determined from experimental data, are seen to be reasonable. The pammeter AJ, is seen to
properly correlate the crack growth rate under elastic and small scale yielding conditions.

1. INTRODUCTION
FATIGUE life may be considered as consisting of two stages: initiation and propagation. For many
structures, especially aircraft, crack propagation occupies a major fraction of the fatigue life. Most
existing techniques for determining crack growth are based upon a small scale yielding approach.
With such an assumption, the cyclic stress intensity factor AK, has been widely employed as the
governing parameter in predicting crack growth rate. When the plastic zone size is not small in
comparison to crack length or other geometric dimensions, so that the stress intensity factor K,
loses its meaning, the linear elastic fracture mechanics approach may not be applied. Since on many
occasions, cracks grow within an elasto-plastic field, emanating from stress concentrations suclmas
cutouts or holes, for example, a suitable parameter to treat this problem is required. The path
independent J-integral[ l] may be employed as this parameter. In this investigation, a better
understanding of the J-integral concept as applied to constant amplitude loading situations,
designated by a cyclic J-integral, AJ, is sought. In ref. [l], the path independent J-integral was
defined as

J =

Wn,-

T,,,zau,
I>

ds

(1)

where r is a path surrounding the crack tip which begins at the lower crack face and ends at the
upper face, W is the strain energy density, n, is the x, component of the outward unit normal to
r, T,,,= u@n, is the traction and u, are the displacement components, and ds is the differential arc
length along r.
The cyclic J-integral was first proposed and implemented by Dowling and Begley[2] as a
parameter which correlates with the crack growth rate, da/dN. Constant amplitude fatigue tests
with a cycle ratio R = 0 were conducted on compact tension specimens fabricated from A533B
steel. Values of AJ plotted vs corresponding crack growth rates da/dN, on a double logarithmic
scale, exhibited power law behavior similar to the Paris equationp] so that it is possible to write

It should be noted that for the cycle ratio R = 0, AJ = J_.


This work motivated other investigators to employ AJ, when plasticity effects may not be
neglected during fatigue loading. Brose and Dowling[4] studied planar size effects on fatigue crack
growth rate of annealed AISI 304 stainless steel. The cyclic J-integral method was employed in
fatigue testing of compact tension specimens with R = 0.05. The crack growth rate was well
correlated by the cyclic J-integral for this material. Mowbray[S] employed the cyclic J-integral to
355

356

L. BA~SIL~

and Y. VOLFERT

study fatigue crack growth in Cr-MO-V steel with R = 0.1. The test results give further support
to the hypothesis that crack growth rate is controlled by AJ. El-Haddad and Mukherjee[q
performed cyclic J tests of ASTM A516 Grade 70 steel compact tension specimens of various
thicknesses. Fatigue crack growth rate data was obtained for specimens subto elastic and
el~t~pl~tic cyclic loading with - 2.0 < R c 0.5. At high load levels, AJ was seen to be better than
AKin correlating fatigue crack growth data. Tanaka et a1.[7]employed the cyclic J-integral to study
the effect of loading conditions and specimen geometry on elasto-plastic fatigue crack growth in
low-carbon steel JIS SM41B. Using thin compact tension and center cracked panel specimens, R
ratios between - 1 and 0.7, frequencies between 0.3 to 30 hertz, and high cyclic stresses, they
reported a sign&cant acceleration in crack growth which deviated from the stable relation between
growth rate and AJ. This acceleration coincided with the spread of plastic yielding across the
specimen ligament. Jolles[8]implemented the cyclic J-integral to characterize fatigue crack growth
rates when plasticity effects are important for A533B steel. In particular, it was found that for
both increasing and decreasing load gradient, AJ may be employed to correlate fatigue data.
Lambert et a1.[9]applied the J-integral concept to fatigue crack propagation of AISI 316 stainless
steel, compact tension specimens under elasto-plastic conditions. The AJ parameter once again
demonstrated its ability to correlate crack growth data. A numerical simulation of two tests was
carried out to evaluate AJ. Comparison of the results demonstrated close agreement between
empirical and numerical AJ values.
As has been noted by many authors, the use of AJ for correlating crack growth data does not
have a sound theoretical basis. The J-integral is based upon a deformtion theory of plasticity where
unloading is not permitted. Perhaps in all experimeuts, the deviation from proportional loading
is small except in [7J where plasticity spread throughout the ligament; apparently there, the
deviation from proportional loading was great.
Returning to the work of Dowling and Begley[2], the experimental results illustrate that
if cyclic plasticity is small during a test, then the experimentally determined AJ values are
approximately equal to calculated (AK)2/Evalues. This rather surprising result may be clarified by
exploiting the definition of AJ given by Ta~a~lO]. Tanaka represented the cyclic J-integral for
a standing crack subjected to a remote loading change (a,), - (a,), as

where r is the path on which AJ is calculated, AW, AT,, and Au, are the changes in strain energy
density, traction, and displacement, respectively, given by
k, )J
AW=
(4)
[ok/- h,hl h,
f &llh
(5)
AT, = (Q - (T,)r
Au, = (K& - (a,),.

(6)

In ref. [lo], path independence of AJ,,, was proven. Note that AJj,, is identified with AJ.
For linear elastic material, eq. (3) may be rewritten as

where

The integral in eq. (7) may be integrated analytically by substituting the near tip asymptotic
expressions for all quantities to obtain in plane stress
&

wE .

03)

Application of cyclic J-integral

351

The expression given in eq. (8) agrees with experimental msults obtained in ref.[2] for R = 0.
Of course, experimental veriftcation of eq. (8) for R = 0 is trivial since J = F/E. The validity of
eq. (8) for R # 0 is sought in this investigation. Since
AR,, = R, - %,

(9)

it is clear that
A$,+

w-0

J,- JP

In fact, from eq. (3) it is possible to show that


Al,,i = Jj - Ji -

f[A~,(~,,mh + hm),A,ln,

AT, f$$

+ (TA,%

II

ds.

(11)

The expression for AJ in eq. (3) cannot be employed in experimental work. Instead, the
expression developed by Merkle and Corten[l l] for the compact tension specimen shown in Fig. 1
was extended for use in cyclic loading as
~

=--2(1+a)AA
l+a*
Bb

(12)

where
a=[($r+2(2)+2]-[($)+l]

(13)

a is crack length, b is untracked ligament, B is thickness, and AA is the area under the load vs
load-line displacement curve obtained,during the loading portion of a fatigue cycle (see Fig. 2).
The aim of this investigation is to examine further the cyclic J-integral. Based upon both
experimental work and numerical analysis carried out on a compact tension specimen fabricated
from Al 2024-T351, further verification of the applicability of AJ is sought. In Section 2, the
experiments are described. Five specimens are tested, four at R Y 0.05 and one at R N 0.5. Values
of C and n in eq. (2) are determined for R z 0.05. Comparisons ma& between AJ and (AK)*/E
show good agreement for both R-ratios. In Section 3, the finite element analysis and simulation
of two experiments for determining AJ are presented. Simulations are carried out for a single crack
length with both R-ratios. In this section, comparisons are made between values of AJ as calculated
by the path independent integral in eq. (3) and that calculated from numerical load, load-line
displacement data and eq. (12), as well as to experimentally determined values. Agreement is good
with certain trends discerned.

Fig. 1. Mod&d compact tension .spechn. Alldimcnsions


are in mm. Spa5men thickness is 6.3Smm.

Fig.2.TheareaMemployedtoestimatehl:disthecrack
mouth opening displacement measured along the load-line.

L. BANKS-SILLS and Y. VOLPERT

358

2. ExPBIuMENTAL

RESULTS

In this section, a description is presented of the experimental program which was carried out
on five specimens to study the cyclic J-integral as a crack growth governing parameter. In most
inv~tigations of the cyclic J-integral, steel alloys have been employed in order to create large plastic
zones. As a result of its importance and extensive use in aircraft industries, an almninum alloy,
Al 2024T351, is considered here.
The compact tension specimen employed is adapted from two ASTM standards, El647[121and
E818[13],and is illustrated in Fig. 1 with its dimensions shown. To enhance initiation of a fatigue
precrack, a chevron starter notch is utilized. The notch is oriented in the L-T direction. All
specimens were ~nst~ct~
from the same plate of Al 2024T351 and machined to a nominal
6.35 mm thickness. The mechanical properties of the material were determined by an average of
four tests carried out on ASTM standard, E 8[14], specimens. The resulting stress-strain curve is
exhibited in Fig. 3 with Youngs modulus found as 71.2 GPa. The material behavior may be defined
approximately by three distinct slopes. Values of the end points of these lines are given in Table 1
so that they may be employed in the elasto-plastic finite element analysis.
All compact tension specimens were tested in a boo-hydrau~c Instron loading machine,
model number 1341, with a 5 kN load cell. Displacement across the knife edges (see Fig. 1) was
measured with a 5 mm crack opening displacement gauge. Crack length was monitored by a
traveling microscope with a maximum magnification of 80. Load and load-line displacement data
were periodically recorded by an IBM-PC AT personal computer through an IBM analog to digital
(A/B) device. In order to mitigate possible retardation effects, the fatigue precrack was introduced
by reducing the load, step wise, from 3 kN to 2.225 kN. During the final millimeter of precrack
growth, the maximum load was 2.225 kN with the crack reaching its initial length of 12.5mm. This
load was also the maximum load employed during the test itself.
All tests were carried out at room temperature with a constant amplitude sinusoidal wave
form. For the 8rst four tests, the test frequency was 10 Hz; for the fifth specimen with R = 0.5,
the test frequency was 20 Hz until a crack length of 30.5 mm was reached and then it was lowered
to 10 Hz. Other test parameters are presented in Table 2. The frequency at which measurements
were recorded sometimes differed from the test frequency as seen in Table 2. Data was stored
beginning with a crack length of 12.5mm. As the crack propagated, at each successive millimeter,
data from several complete consecutive waves was recorded. The appropriate crack length a, and
the corresponding number of fatigue cycles N, were documented simultaneously.
As an example, a hysteresis loop from specimen 3 is shown in Fig. 4. In Fig. 4(a) the noisy
recorded data is illustrated; in Fig. 4(b), the data has undergone a noise filtering procedure in which
Table 1. Stress and strain
values at the end of each
linear portion of the
stress-strain curve shown in
Fig. 3 for Al 2024-T35t
(h&a)

f x 10-z

373.7
408.6
419.1

0.525
2.125
2.878

Table 2. Test parameters for the fatigue mack propagation


tests
Specimen
number
1
2

Fig. 3. S-train

curve for Al 2K?&T351.

P(kN)
2.225
2.225

0.05
0.05

sampling
points
51
51

Sampling
fresuencv
10
10

tCrack length a c 30.5 mm. $Crack length a > 30.5mm.

Apportion

359

of cyciic J-integral

(a)

I
0.2

I
0.4

I
0.8

I
0.8

6 (mm)

2-

f
s
a

l-

0.2

0.4

0.8

I
0.8

8(mm)

Fig. 4. Hysteresis loop recorded from specimen 3 at a crack length a = 36.5 mm, (a) raw data and
(b) filtered data.
a cubic spline is employed (see ref. [ls] pp. W-556). All data was Wered by this procedure. A
comparison made to hysteresis loops obtained simultaneously with an X-Y recorder at several
crack lengths showed excellent agreement between the two methods. Hence, confidence in both the
A/D device and filtering procedure is attained. As an example, hysteresis loops obtained as the
crack progresses in specimen number 4 are shown in Fig. 5. As crack length increases, the slope
of the load-load-one ~spla~ment curves decreases and the area of the hysteresis loop increases;
the latter indicates an increase in the plastic zone size ahead of the crack tip. In addition, AA and
hence AJ values are increasing with increasing crack length.
From these tests, values of AJ vs da/dN were determined. For AJ values, load vs load-line
displacement data was recorded at each succeeding millimeter as the crack propagated. The data
was filtered and the area under the curve determined by the trapezoidal rule. For specimens 1, 2
and 4, nine consecutive curves were recorded, for specimen 3, one, and for specimen 5, four. The
area M, illustrated schematically in Fig. 2, was determined as the average of these consecutively
recorded curves. For specimens 1, 2, and 4, the coefficient of variation was employed to compare
differences between AA values found at a particular crack length. The coefficient of variation is
defined as the standard deviation divided by the mean value, multiplied by 100%. The maximum
value obtained was 2.7% which was at the shortest crack length. The coefficient of variation was
generally found to be under 1% . Indeed, for specimen 4 in which data was recorded at 501 stations,
as opposed to 51 for specimens 1 and 2, lower coefhcient of variation values were obtained. Values
of AA are substituted into eq. (12) to obtain AJ. Crack growth rate du/dN, was obtained by passing
a cubic spline through the recorded data of crack length vs number of cycles and differentiating
(see ref.[l5] pp. 420-422, 431-433).

360

L. BANKS-SILLS and Y. VOLPERT

1.0

0.5
6 (mm)

Fig. 5. Load, displacement behavior with increasing crack length; specimen 4. The data has been filtered.

In Fig. 6, the behavior of AJ as a function of crack length is exhibited for all specimens. For
specimen 5, a slight irregularity in AJmay be observed. This variation resulted from a slight change
in applied load as shown in Table 2. Indeed, this change insignificantly affects the results. Good
agreement is achieved in the behavior of specimens with nearly the same R-ratio (see Table 2); i.e.
specimens 1 and 2 and specimens 3 and 4. There is clearly an R-ratio effect; for specimen 5, the
nominal R-ratio is 0.5. Results in the sequel will be exhibited for specimens 3 and 4 unless there
is a difference in behavior which should be illustrated. Curves of crack length a, as a function of
fatigue life for specimens 3 and 4 are shown in Fig. 7. The scatter between specimens experiencing
similar conditions is minimal. The crack growth rate da/dN vs the cyclic J-integral is plotted in
Fig. 8 for these specimens. Data was not recorded for specimen 3 when a < 28.5 mm. In a manner
similar to common practice with da/dN vs AK data, a straight line fit with, in this case little scatter,
may be carried out. With crack length greater than 32.5 mm, acceleration in crack propagation is
observed, so that stage III fatigue crack growth appears to have been reached (see also Fig. 7).
From Fig. 8, it may be discerned that AJ may be employed as a parameter to correlate crack growth
rate.

40

Specimen 5

Spdmen

40 lSpecimen 4
a-32.5mm

30

Specimen 1

30 -

Specimen 2
0 Specimen3
x Spdmen4

0
R1

(D

20-

20

101

10

20

AJ (kNlm)-

Fig. 6. The cyclic J-integral vs crack length.

1U
0

Nxl@
Fig. 7. Crack length vs fatigue life.

Application of cyclic J-integral

t
a9328mm

\*$:z

Spdman a
Spednun 4

.IX
2
l
XXX

xx

XX
xxXXx
XXXX

Fig. 8. Crack growth rate vs AJ.

Since it was shown theoretically that for small scale yielding, W is equal to (AK)z/E for all
R-ratios, it is interesting to compare them experimentally. For the compact tension specimen[16]
2
-_
0.886+4.64;-13.32
$
0

AK is obtained by substituting AP, the test load range, for P in eq. (14). Comparisons for
specimens 4 and 5 are illustrated in Figs 9(a) and (b) respectively. Reasonable agreement between
W and (AK)]E is observed in these graphs. The differences for all specimens varied between 4%
and 55% with the highest deviation recorded for the smallest crack length of a/W = 0.25. For
a/W > 0.5, the maximum difference is 23%. For quasi-static loading, it is recommended in ref. [13]
to employ eq. (12) for a/W > 0.5 since this equation is more accurate for these crack lengths.
Another possible cause for the observed difference may relate to the Youngs modulus value, E.
In four tests, E varied between 68.8 GPa and 73.6 GPa. This variation causes a change in (AQ2/E
of about 7%. Finally, a plane stress assumption has been made in determining eq. (8). The specimen
is actually devour.
By assuming the other extreme, namely plane strain, the results would
move closer together, sometimes owing.
Clearly, this effect should be tempered somewhat by
the three-dimensional nature of the problem. After the finite element analysis is presented, a further
discussion of these diEerences will be presented.
Because of the small Merence between the curves in each of Figs 9, it may be concluded that
the plastic zone size is small for all crack lengths. Thus, eq. (8) is verified experimentally for the
small values of R employed in specimens l-4, as well as by specimen 5 for R > 0.
The R-ratio effect is illustrated in Fig. 10 which is a composite graph of all test results. It may
be seen that for R near 0.05, the results may be approximated by a single straight line. A second,
nearly parallel line is given for R N 0.5. This is typical behavior when AK is employed as the
abscissa; here, the analogous behavior is obtained with AJ. For R 2 0.05, the data may be
correlated by
$

= 0.131 x IO-3(AJ)*~

where u is measured in millimeters and Win l&l/m. There is insuacient information to determine
crack grovyth material parameters for R N 0.5.
Information on crack growth conditions may be further elucidated by examining the specimen
fracture surfaces. In Fig. 11, for specimens l-4, shear lips begin to appear when a/W cy 0.5. It
would seem that for a/W < 0.5, plane strain conditions dominate. Thus, in this range, calculated

362

L.BANKS-SILLSand Y. VOLPERT

(b)

l AJ

l8x

Y (AK)2/E

ox

*X

g
2
g

OX

10-3=

*XX
l X
l X
lo4

**Xx
OX\

4% 1

l *XX
*XxX

l*
I

104
0.1

ww

> 0.5

Ill1111

,I&

1.0
AJ,(AK)~/E(kN/m)

Fig. 9. Comparisonof W and (AK)2/Efor specimens(e) 4(R Y 0.05) and (b) S(R cy0.5).

values of {AK)z/Eshould be multiplied by 1 - vz, bringing the two curves in Fig. 9(a) for specimen 4
closer together. The shear lips become a sign&ant portion of the specimen thickness for
a/W 3 0.66. For larger crack lengths, the plastic zone incre@es in size and nonlinear effects in the
load, ~spla~ment behavior are observed (see Fig. 9 for specimen 4). On the other hand, for
specimen 5, it may be observed in Fig. 11 that shear lips did not develop to a significant size, in
part explaining the linear behavior and absence of hysteresis loop development seen even for large

lo-2 z_
:
f
I!
s

10-3

46

$04

10-s
0.1

0 R'O.5
l R=0.05

I I llllll

1 III~l
1.0
AJ (kN/m)

0
:*

II
10.0

Fig. 10. Composite graph of all test rwults.

Application of cyclic J-integral

Fig. Il. Fracture surfaces of the tested specimens. The lower arrow indicates shear lip initiation, the upper
arrow, catastrophic failure.

363

365

Application of cyclic J-integral

Elastic A

Ebto-~l=tic

La
Fig. 12. Finite element mesh of the compact tension specimen.

cracks. For specimen 5 it would appear that over the entire range of crack propagation, plane strain
conditions dominate, explaining, perhaps, why the curves in-Fig. 9(b) remain parallel, whereas
those in Fig. 9(a) as well as for specimens l-3 approach one another for large crack lengths.
In the next section, a numerical simulation of the tests is carried out for comparison to the
experimental results.
3. NIJMERICAL

SIMULATION

A finite element analysis is carried out on the compact tension specimen shown in Fig. 1 with
the program ADINA[17]. As a result of symmetry, a mesh of one-half the specimen is constructed
(see Fig. 12). The mesh contains 345 plane stress, eight noded, serendipity, isoparametric elements.
Rectangular, quarter-point, singular elements are employed at the crack tip. A concentrated load
is applied at the hole to represent the pin loading.
In order to demonstrate numerical accuracy, the material was first modeled to be linear elastic.
The stress intensity factor K, was determined from the J-integral of eq. (1) for 0.3 < a/W < 0.8
at twelve crack lengths along contour 1 in Fig. 12. Details of the calculation are given in ref. [18].
Comparison was made to eq. (14) which is said to be accurate to + 0.5%. Differences were found
to be less than 1%.
Next, the elasto-plastic analysis is presented. The material properties are taken from the
stress-strain curve exhibited in Fig. 3 and Table 1, yielding a piecewise linear model. The
quarter-point, singular element is appropriate for this material behavior. To avoid undesirable
plasticity effects at the concentrated load, the material in that region is modeled to be linear elastic
as illustrated in Fig. 12.
Tests of specimens 3 and 5 are simulated numerically at a crack length of 36.5 mm. For
specimen 3, two fatigue cycles are imposed, for specimen 5, three. Isotropic hardening is assumed.
Since the Bauschinger effect is neglected, ,the size of the hysteresis loop and the reversed plastic zone
will be somewhat smaller than in reality. Results are shown in Figs 13. For specimen 5, results for
the second and third cycles are identical. Moreover, these cycles yield a linear relation between load
and load-line displacement.
After the first cycle, values of AJ are calculated employing displacements and stresses
determined from the finite element analysis which are substituted into the line integral of eq. (3).
Along the three paths in Fig. 12, results differed by less than 0.5% for each specimen. Nondimensional values from path 1 are shown in the first line of Table 3 where trl is the proportional
limit (see Fig. 3 and Table 1). In the second line of Table 3, values of (AK)z/uu: are presented.
These are obtained from eq. (14) for the stress intensity factor of the compact tension specimen.
They differ by 0.9% for specimen 3 and 1.2% for specimen 5. The difference may be explained
by the small plastic zone at the crack tip. In the third line of Table 3, nondimensional AJ values are
exhibited which are evaluated with eq. (12) from the numerical load, load-line displacement data,
simulated as an experiment. The differences between these values and those determined by the path
independent integral in eq. (3) are 5.0% for specimen 3 and 5.6% for specimen 5. This comparison
exemplifies the approximate nature of eq. (12) [l l] which is employed to experimentally determine

366

L. BANKS-SILLS and Y. VOLPERT

(a)
2.0-

0.8

0.4

6 (mm)

08
2.0 -

0.8

0.4

Q(mm)
Fig. 13. Numerical finite clement simulation of aevcral fatigue cycles at a crack length a = 36.5 mm of
experimental colKlitions for spechns (a) 3 and (b) 5.

AJ vahtes. But more importantly, the comparison i&u&rates that the path integral of eq. (3)[10]
may be employed to determineAJ for complicated safe
in fatigue crack prounion
studies.
Moreover, it is clearly seen that AJ as calculated in eq. (12) and AJ,,, of eq. (3) are equivalent.
Finally, in the last row of Table 3, the experimental nondimensional AJ values are presented.
These differ from those calculated with the finite element results and eq. (12) by 2.4% and 9.2%,
Table 3. Comparison of dldcwmined from qcriment
numcricA simulation for a = 36.5 mm
spchKxl3
Method

R ~~0.05

SptChllS

R ~~0.5

EAJ
a1
cq.(3) elite element

0.222

0.0680

0,220

0.0672

0.211

0.0642

0.2Q6

0.0583

(W
auf
eq. (14)

EAJ
2#I
eq.(12) ii&e elcmcnt
EM
P
MI
eq.02)
cxporimcntaldata

and

367

Application of cyclic J-integral

(a)
2.0 -

1
0

0.4

0.8

6 (mm)

1.0 '
0.4

Fig.

14. Comparison

I
0.8

I
0.8

of numerical simulation and experimental-load vs load-line displacement curves at


crack length a = 36.5mm for specimens (a) 3 and (b) 5.

respectively. This last comparison is a true comparison of the differences between the numerically
and experimentally obtained AJ values. These differences are not great and are probably related
to the plasticity model employed in the numerical calculations which neither models properly the
Bauschinger effect nor the plastic wake behind the crack tip, for example. This is confirmed in
Figs 14 which compare load, load-line displacement curves of the experiments and the numerical
calculations. It may be observed that in Fig. 14(a) the numerically calculated hysteresis loop is
smaller than that which was experimentally determined. This difference in size indicates smaller
plasticity effects in the analysis. This could be anticipated by the choice of isotropic hardening for
the numerical analysis. Moreover, in Fig. 14(b) for specimen 5, the analysis does not provide for
a hysteresis loop at all.
In addition, differences may be observed in Figs 14 for the crack opening displacement at Pmi.
as determined experimentally and numerically. For specimen number 3 in Fig. 14(a), the
numerically determined crack opening displacement is smaller than that obtained experimentally.
This difference may be explained by the plastic wake left as the crack propagates in the specimen,
increasing the crack opening displacement. In the finite element analysis, only a second loading was
carried out, so that a plastic wake is not present. For specimen number 5, the results are reversed
(see Fig. 14b) with a smaller difference than for specimen number 3. Since there is no compressive
plastic zone, the effect of the residual plastic strain is reduced.
Returning to Table 3, the differences in rows two and three, which are both calculated
numerically, explain the tendency of the experimentally determined AJ to be smaller than (AQ/E.
This comparison helps to explain the differences observed in Figs 9.
Now, eqs (10) and (11) are considered. As an example, calculations are performed on
specimen 5 with a crack length of 36.5 mm. The nondimensional value of J, determined from eq. (1)

L. BANKS-SILLS and Y. VOLPERT

-a*

I---a----CI
(a)

(a)

i----a4
(b)

(4
Fig. IS. Plastic zones determined from finite element analy
sis of q&men 3. (a) Loading (P,, = 2.16 kN), (b) unloading (P_= 0.15 kN), (c) reloading (P,, = 2.16 kN) and
(d) unloading (P,, = 0.15 kN).

(4
Fig. 16. Plastic zones determined from Snite element analysis of specimen 5. (a) Loading (P,, = 2.22 kN), (b) unloading (P,. = 1.10 kN), (c) reloading (Pm = 2.22 kN) and
(d) unioading (P,, = 1.10 kN).

at the maximum cyclic load is 0.3113, whereas Ji at the minimum cyclic load is 0.0883. The
difference 4 - JI is equal to 0.223 and clearly not A,&,,which is 0.0680.
Of further interest are the plastic zones obtained by the finite element, numerical simulation
of the tests. These are illustrate in Figs 15 and 16 for specimens 3 and 5, respectively. At the
minimum load, specimen 3 exhibits a small compressive plastic zone which is illustrated in
Fig. 15(b). At the maximum load achieved by reloading, the plastic zone in tension remains smaller
than the initial plastic zone [compare Fig. 15(a) and (c)l, but returns to the same compressive zone
after unloading [compare Fig. 15(b) and (d)]. For specimen 5, P,, is slightly larger than that
of specimen 3. The plastic zone size seen in Fig. 16(a) is somewhat larger than that in Fig. 15(a).
Since Pti is relatively large (R = OS), a compressive plastic zone does not develop for P = P&,,
[see Fig. 16(b) and (d)], although the stresses are compressive. The plastic zone in Fig. 16(c) for
P = Pm on the second cycle is larger than the one in Fig. 15(c) but again smaller than the initial
plastic zone of Fig. 16(a). For both specimens, the size of the plastic zone decreases after repeated
loading. Under anti-plane strain conditions and elastic, perfectly plastic material behavior, Rice[l9]

Application of cyclic J-integral

369

showed that the cyclic plastic zone size is one-fourth that of the monotonic plastic zone size when
Ge= - ur, the yield stress. Specimen 5 was numerically unloaded and reloaded again. The stresses
in the elements adjacent to the crack tip and the plastic zone size are identical to the second loading
cycle. Lack of a plastic zone for unloading causes the load vs load-line displacement curve to be
linear as may be seen in Fig. 13(b) for specimen 5.
4. CONCLUSIONS
A numerical/experimental investigation of the cyclic J-integral was carried out. Compact
tension specimens fabricated from Al 2024-T351were tested under constant amplitude loading with
two R-ratios of approximately 0.05 and 0.5. Cyclic J-integral, AJ, values were determined from
load vs load-displacement data. Values of (AK)*/E were calculated, as well. For crack lengths
a/W 2 0.5, dilIerences of between 4% and 23% were observed for all tests. These differences were
primarily due to the approximate nature of the specimen calibration equation for AJ, as well as
the three dimensional nature of the specimen as compared to the two dimensional assumption
~n~ming (AK)2fS
An elasto-plastic finite element simulation of the fatigue process was carried out at a crack
length a/W of 0.72. The maximum load was essentially the same for both specimens 3 and 5; the
nominal R-ratio for the former was 0.05 and for the latter, 0.5. Values of AJ were calculated in
two ways: (1) by means of a path independent AJ-integral and (2) from load vs load-line
displacement data as would be done for an experiment. The difference between these two methods
was found to be 5.0% for specimen 3 and 5.6% for specimen 5. A comparison between the
experimental and numerical results, as computed from load vs load-line displacement data, yielded
differences of 2.4% for specimen 3 and 9.2% for specimen 5.
Since AJ, as calculated by a path independent line integral from the finite element analysis,
and (AK)/E, also a calculated value, produce results which differ by no more than 1.2%, it would
seem that AJ may be exploited as a parameter for correlating the crack growth rate, &/dl\r, for
elastic and small scale yielding conditions. For two R-ratios, it has been shown experimeutally
under small scale yielding conditions that AJ determined from load versus load-line displacement
data closely approximates (AK)z/E. This implies that AJ should be based upon the area under the
loading portion of the load, load-line displacement curve. Moreover, double logarithmic plots of
da/dN versus AJ exhibited nearly parallel behavior for tests with different R-ratios.
Thus, the cyclic J-integral may be considered as a crack growth parameter and employed to
correlated crack growth rate data for small scale yielding. It is important to note that the line
integral may be employed to evaluate AJ values of more complicated geometries. Hence, with
material properties determined form laboratory specimens, such as the compact tension specimen,
for example, a crack growth analysis under elasto-plastic conditions may be carried out.
The presence of stable hysteresis loops from repeated cycles at a given crack length allows
extension of results determined from monotonic loading to cyclic loadin~l9]; that is, extension of
the J-integral to AJ. Moreover, since in this investigation, as well as most others, deviation from
proportional loading appears to be small, AJ as defined in eqs (3) and (12) may be soundly
employed as a crack growth parameter.

[l] J. R.

Rice, A path independent integral and the approximate analysis of strain concentration by notches and cracks.
J. appi. Mtxh. 35, 379-386 (1968).
[2] N. E. Dowling and J. A. Begley, Fatigue crack growth during gross plasticity and the J-integral. ASTM-STP WI,
82-103 119761.
131 P. Pisand
5. Erdogan, A critical analysis of crack propagation laws. J. Basic Engng 85, 528-534 (1963).
[41 W. R. Brose and N. E. Dowling, Size etlects on the fat&e crack growth rate of type 304 stainlm steel. ASTM STP
648.720-735 (19791.
[S] D. P. Mowb&y,
Uk of a compact-type strip specimen for fatigue crack growth rate testing in the high-rate regime.
RSTM STP 666, 736-752 (1979).
[6] M. H. El-Haddad and B. Mukhezjee, Elasti~plastic fracture mechanics analysis of fatigue crackgrowth. ASThf STP
803, H-689-11-707 (1983).
[7l K. Tanaka, T. Hoshide and M. Nakata, Elastic-plastic crack propagation under high cyclic stress. ASTM STP 803,
I&708-11-722 (1983).

370

L. BANKS-SILLS and Y. VOLPERT

[8] M. JolIes, Effects of load grad& on applicability of a fatigue crack growtb rate-cyclic J relation. ASTM STP 866,
381-391 (1985).
[9] Y. Lambert, P. SaiBard aud C. Bathias, Application of the J conceptto fatigue crack growth in large-scale yielding.
ASTM STP 969, 318-329 (1988).
[lo] K. Tanaka, The cyclic J-integral as a criterion for fatigue crack growth. Znr. J. Fracture 22,91-104 (1983).
[ll] J. G. Merkle and II. T. Corten, A J-inte8ral analysis for the compact specimen, considering axial force as well as
beuding effects. J. Press. Vess+ Technol. 96, 286-292 (1974).
[12] Standard test method for ~~~nt-l~d-~p~~~
fatigue crack growth rates above 10e8 m/cycle, E 647-86. AAntrat
Z?o& ofASTM St&r&,
Vol. 03.01,714-736. American Society for Testing and Materials, philadelpbia, PA (1986).
[131 Standard test method for JI,, a measure of fracture touglmeas, E 8 13-8 1. Am& Book of ASTM Stan&u&, Vol. 03.01
768-786. American Society for Testing and Materials, Philadelphia, PA (1986).
[14] Standard methods of tension testing of metallic materials, E &85b. AtwuuziBook of ASTM Stan&r& Vol. 03.01,
124-145. American Society for Testing and Materials, ~l~p~a,
PA (1986).
[Is] ZMSL, Users Mamud, Math/Library, FORTRAN Subroutines for Matbematicai Applications, Version 1.0 (1987).
[16] Staudard test method for plane-strain fracture toughness of metallic materials, E 399-83. Ann& Boo& of ASTM
Stanaiar~, Vol. 03.01, 522-564. American Society for Testing and Materials, Philadelphia. PA (1986).
[17] K. J. Bathe, ADZNA-Automatic Dynwnic Zmmmental Nonlinear Analysis System. Adina Engineeriug Inc., Watertown,
MA (1981).
[18] L, Bank&Us and D. Sherman, Comparison of methods for calculating stress intensity factors with quarter-point
elements. Znt. J. Fracture 32, 127440 (1986).
[lP] J. R. Rice, Mechanics of crack tip deformation and extension by fatigue. ASTM STP 415, 247-311 (1967).
(Receiued 19 September 1990)

Anda mungkin juga menyukai