CRACK PROPAGATION MODELING FOR

SURFACE GENERATED SCRATCHES

IN Al 2024-T3 CLAD ALLOY

M. Talia* and J.E. Talia**

(*) Mechanical Engineering Department (**)Mechanical Engineering Department

ABSTRACT:

The effects of surface generated scratches on fatigue life and fatigue crack growth rate in

Al 2024-T3 clad alloy is reported. Significant drops in fatigue life of up to 95% were observedfor

100 micron scratch depth in 0.4 inches (1.016 mm) thick specimens, which suggests that this alloy

is a scratch-sensitive material. In addition, it was observed that fatigue crack growth rates are

significantly higher in scratched specimens than in the unscratched specimens. Based on data

analysis, Paris equation type models have been developed that predict the fatigue crack propagation

behavior in scratched specimens for both long and short cracks. Scanning Electron Microscope

observations reveal that scratches generate micro fissures. The micro fissures aided by the fatigue

process coalesce to form larger concentration of micro cracks (whose performance resembles short

crack behavior) in front of the path of cracks. It is conjectured that this correlation can be linked

to the reduction in fatigue life and to the increase in fatigue crack growth rates of the alloy.

117
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Nomenclature

a: shape parameter = 0.746

β: material constant = 0.025

Θ: characteristic or scale value = 0.09

Δσ: applied stress range

σ: applied stress

af: fatigue endurance limit

omax: maximum applied stress

omin: minimum applied stress

a: crack length

b: Burgers vector = .286 nm

C: material constant = 4.41 10"10

d: scratch depth

D: grain size = 35 μηι

da/dN: crack growth rate

Aa: crack length range

ΔΚ: stress intensity factor range

AK^t^: stress intensity factor range generated by a scratch

AKth: threshold stress intensity factor range

K: stress intensity factor

Kf: fatigue stress limit

K,: stress intensity concentration factor of a sample subjected to Mode I loading

Kmax: maximum stress intensity factor

minimum stress intensity factor

10: effective crack length

n: material constant = 2.53

N: number of cycles to failure

N,: number of cycles to crack initiation at a stress amplitude o

q: notch or scratch sensitivity factor

r: scratch tip radius = 83 μηι

R: fatigue stress ratio = 0.1

s: average crack spacing at a stress amplitude σ

Se: endurance limit

Se(!CT): endurance limit of scratched samples

^e(uscr)' endurance limit of unscratched samples

S: average crack spacing at fatigue limit

w: width of the sample = 1 mm

Y: geometric factor

INTRODUCTION:

The relation describing fatigue crack propagation in metals is usually presented in a logarithm

plot of the crack growth rate (da/dN) against the logarithm of alternating stress intensity factor, Ak,

at the crack tip [1] (Figure 1). The lower limit or threshold, region one, shows the fatigue conditions

under which the crack does not propagate. The upper limit, region two, covers the crack propagation
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Figure 1. Typical fatigue crack growth rate data for small and large cracks

associated with the start of final rupture. Region two is described by power law for metals known

as the Paris equation, [1] which has the form,

n
— = C(AK) (1)
dN

where C and η are numerical constants and a is the crack length. ΔΚ can be expressed in terms of

the geometry, stress and crack size [1], as

AK = ΥΔσ ( a ) 1 / 2 (2)

Another critical fracture mechanics parameter is the threshold stress intensity factor range

(ΔΚ^). LEFM concept offers, in principle, a relation for stress magnitude and distribution near the

120
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

crack tip to the applied stress, crack size and shape, and material properties of the cracked specimens

[2]·

Region one in Figure 1 is used quite commonly in failure analysis. A number of investigators

[3-6] have proved that cracks less than 1mm in length tend to show faster growth rates and lower

thresholds than long cracks. The growth of short cracks in aluminum alloys have been investigated

by Lankford [7] and Morris et al. [8], They proposed that micro cracks are formed near the

intrusions in stage one of the crack growth. The crack tendency to grow at intrusion rather than slip

band depends on stress level and inclusion content. In addition, they suggested that a slip band crack

generally occurs from high stresses and higher purity of the alloys.

Pearson [9] concluded that cracks the size of approximately the average grain-size, grew faster

than long cracks at identical ΔΚ values where ΔΚ is the stress intensity factor range and is defined

as, AK = K Tnax -K imn .

Several models have been proposed to study the mechanisms of short crack under plane-stress

conditions. El-Haddad [10], suggested that AK,h is a function of (na + l0)Vj, where 10 is the effective

crack length considering the small crack behavior. The standard equation for ΔΚ is modified by

adding a constant, to the crack length. This equation fulfills the basic requirements in that it tends

to a constant value at very short crack lengths, equal to the fatigue limit, and it merges with the

long-crack values once crack length becomes large compared to 10.

The stress intensity factor equation relates to cracks growing from notches in an infinite plate.

For cracks growing from the tip of an elliptical notch, K, was expressed as,

KI = KtoJna (3)

where K, is the stress intensity factor [11], The local stress near the tip of the notch is elevated by

121
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

the stress concentration. Dowling [12], used Equation 3 to derive an expression for the transition

crack length in short crack problem. For a short crack, careful attempt should be made to control

a relatively high stress for a given ΔΚ value. If the stress is too high, LEFM conditions will fail and

this will result in a larger plastic zone and faster crack propagation [13], Lankford and Hudak [14]

have observed larger crack tip strain values for short cracks at given Aa. Holm and Blom [15] used

an elastic-plastic finite element model to predict the crack-tip plastic zone. It is found to be

considerably larger in a short crack, even in the absence of micro structural effects.

EXPERIMENTAL:

Fatigue test specimens were prepared from metal sheets according to ASTM standards as

shown in Figure 2. Scratches with a uniform tip radius (r) of 83 μηι were introduced with the help

of special end mill form cutter on the specimens across the width at midway along the gage length.

(a) (b)

Figure 2. Fatigue (a) and crack propagation (b) test specimens

122
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

A specially designed fixture was used to mount the specimen on the milling machine to ensure the

uniformness of the scratch. It is worthwhile to mention here that the scratch depth regime was

carefully selected for these experiments in order to make sure that these scratch depths corresponds

to the ones normally found in service. For instance 100 microns deep scratches are common during

the fabrication and assembly procedures on critical aluminum alloy components in aircraft industries.

Once the scratch was introduced onto the specimens, the scratch depth was measured under the

optical microscope. The technique used consisted of focusing at the surface of the sample which was

taken as a reference point and then focusing on the scratch bed. The distance between these two

points can then be evaluated with the aid of proper graduation. The method is reliable and accurate

with an acceptable error of +/- 1 micron.

The unscratched surface of the fatigue crack specimens were polished using a standard

polishing technique. A small scratch was introduced at the edge and through-thickness of the samples

for the startup of the crack. Crack propagation rate was measured by observing the movement of the

surface crack on the polished face with a traveling microscope. The thickness and width for all the

fatigue crack propagation samples were 1 mm and 25 .4 mm respectively. The calculations of stress

intensity for scratched and unscratched samples were based on edge crack formula [16]:

Κ = Υσγ/i

where,

y = l . 9 9 - 0 . 4 1 — + 1 8 . 7 ( — ) 2 - 3 8 . 4S ( — ) 3 + 5 3 . 8 5 ( — ) 4
W W W W

RESULTS & DISCUSSION:

Figure 3 shews the microstructure of the Al 2024-T Clad sample with an average grain size
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Figure 3. Microstructure of Al 2024-T3 clad specimen.

Engineering Stress [MPa]

Number of cycles to failure (X1000)

Figure 4. Stress vs. cycles to failure of Al 2 0 2 4 - T 3 clad samples

(D) of 35 μιτι in the rolling direction. Figure 4. shows the engineering stress versus number of cycles

to failure of Al 2024-T3 clad samples at different scratch depth. It is clear that the unscratched

samples are more resistant to fatigue than the scratched samples. As the scratch depths increases,

124
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

fatigue life decreases. For example, samples with a scratch depth near ΓΟΟ μπι, demonstrate a fatigue

life reduction of 95% at a stress of 345 MPa. As the stress decreases, fatigue life of the scratch

samples of 100 μπι increases to 87% at a maximum stress of 172 MPa. However, as the scratch depth

increases from ΙΟΟμηι to 225μπι, the fatigue life reduces to 72% at higher stress (380 MPa) and to

71% at low stress (170 MPa). In addition, for scratch depths of 100 μπι and 225 μπι and for life

greater than 1,000,000 cycles, the curves level off and there are no further noticeable decrease in the

life curves. The arrows on the lower end of the curves, indicate that the samples did not break at

the corresponding number of cycles. These points are defined as an endurance limit which usually

correspond to a stress at which the life become infinite. However, experimentally only ferrous

materials have an endurance limit. For non-ferrous materials, as in this case, an endurance limit is

defined as the stress corresponding to 106 or 107 cycles. The endurance limits, presented in Figure

4, corresponds to 159 MPa for unscratched samples, 145 MPa for scratched (100 μπι) curves, 126

MPa for 150 μηι scratch curve, and 120 MPa for the scratched samples at 225 μπι.

The experimental crack growth rate data (da/dN) was obtained as a function of stress intensity

factor for both scratched and unscratched specimens (Figure 5). As expected, scratches generated

conditions favorable for crack propagation and short cracks which propagate faster than long cracks

and at a lower stress intensity factor than unscratched samples. Moreover it is clear from the same

figure that cracks in scratched samples propagates at a lower threshold intensity factor than the

unscratched samples (Kth = 1 . 1 MPa. m1/2). As the scratch depth increased, the crack growth rate

increased significantly. At a scratch depth greater than 150 μηι, no measurements were obtained

since the crack propagated freely at a very large rate in a highly defected scratch surface.

Fatigue life predictions is usually divided into two parts, crack initiation and crack

propagation. The crack initiation life largely covers the micro cracks growth period [17], Several

125
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

da/dN [m/cycle]

ΔΚ [MPa >/m]
Figure 5. Crack propagation rates of scratched samples

aspects can explain the differences between the growth mechanisms of the two types of cracks, short

and long cracks [18], However, there is no single numerical value for the transition crack length

between micro cracks and macro cracks.

LONG CRACK PROPAGATION

In applying the Paris law model (Equation 1) and introducing the stress concentration factor

(K,), da/dN can then be written as:

fN - <4>

126
Μ. Talia and J. Ε. Talia Journal of the Mechanical Behavior of Materials

AK = YKtAoJI (5)

where, K t was obtained using the notch sensitivity factor equation (q) as

Kf - 1
1 = — <*>

As defined [19], values of q range from zero (no notch effect) to unity (full theoretical effect).

Through correlation to experimental data, a number of researchers have proposed analytical

relationships for the determination of q. The most common is being proposed by Peterson [20] and

Neuber [21], Peterson showed the relationship q as a function of the notch tip radius (r, in inches),

For Aluminum alloys β = 0.025, [22], IQ which is defined as the fatigue stress limit, was

taken as the ratio of endurance limit of unscratched (Se(uscr)) to endurance limit of scratched samples

(Seiscr))· Therefore in finding an equation for the fatigue crack growth for scratched samples, the AK eff

which counts for K, has to be taken into consideration. Since it was shown earlier that the fatigue

crack growth (da/dN) increases significantly as the scratch depth (d) increases, it is evident that the

scratch depth (with a tip radius) influence the correction factor K,. Hence combining Equation (6)

and (7), an expression for K, is obtained as,

Kt = (Kf - 1) ( 1 + 3 / r ) + 1 (8)

Using the crack propagation data for unscratched samples (K, = 1) and Equation 4, the
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

material constant, C=4.41 10"10 and η = 2.53, were obtained. Keeping C and η constants, Equation

4 was used for the scratched samples. The following table shows the values obtained in computing

K, using Equations 6, 7, and 8,

Table 1

Scratch depth, d (pm) 0 40 70 100 150

Endurance limit, Sc (MPa) 159 156 152 145 126

Fatigue stress limit, Kf 1 1.017 1.045 1.095 1.26

Stress concentration, K, 1 1.16 1.41 1.86 3.4

Scratch sensitivity, q 0 0.11 0.11 0.11 0.11

It is important to notice that when scratch depth is zero (unscratched, K, = 1), Equation 4

would be reduced to Paris law. A close agreement can be seen in Figure 6 between theoretical and

experimental results of unscratched samples. As the scratch depth increased, da/dN will increase

significantly. Figures 7 show the crack growth rate predictions of the experimental data of scratched

samples at 40 pm, 70 pm, 100 pm, and 150 pm. Whereas, Figure 8 shows the actual value K t (Table

1) used when scratch depth is present. As the scratch depth increases, stress concentration, Κ,,

increases. However at scratch depth of 40 pm, there is a small change in K,. This is due to the

presence of clad layer (50 pm thick) which was taken into consideration in scratching the surface of

the material. As expected scratching of the clad layer has a detrimental effect on the fatigue life of

the samples.

SHORT CRACKS PROPAGATION:

The crack growth rate of scratched samples behaved differently than the unscratched sample.

Scanning electron microscope micrographs shown that scratches generated defects (micro fissures).

Such fissures in combination with cycle fatigue produce micro cracks which assist the propagation
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

d a / d N [m/cycle]

AKeff [MPa / m ]
Figure 6. Theoretical model for unscratched samples

d a / d N [m/cyclel

Δ K e f f [MPa 7m]
Figure 7. Long crack model for scratched samples

129
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Stress Concentration Factor

Scratch Depth [microns]

Figure 8. Stress concentration factor as a function of scratch depth

of cracks (see Figure 9). It is also noticeable from da/dN curves of scratched samples (Figure 5) that

cracks generated from scratches grow more rapidly than would be predicted by the LEFM data

(Equation 5). Furthermore, based on a large body of research [23], it is evident that the sort fatigue

crack problem is a multi-faceted one. In particular, any unified theoretical treatment must consider

a number of factors, including cyclic stress amplitude, crack size, grain size, plastic zone size,

crystallographic orientation, and crack closure. However it was shown [24,25] that short cracks

experience a larger crack tip effect due to the high stress factor and reduced crack closure.

To predict the growth rate of short crack and in turn the fatigue life of structures containing

defects, such as micro fissures, micro cracks, etcetera, a method is needed that can reflect Akeff at the

tip of the short crack. Mazumdar used an alternate approach in which he considered the defects-

crack interactions [26], Convincing evidence [27] shows that the average crack spacing, in smooth

130
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

Figure 9. SEM micrograph of scratched surface showing micro fissures

specimens, initially decreases due to continued nucleation of micro cracks. A fatal crack develops

once this crack spacing with cycling assumes a steady state value when life is almost near exhaustion.

Using this approach, Mazumdar derived a stress-life model for crack initiation in terms of some

fundamental parameters,

(N.)
1
= (2) (_!) (9)
b sf

where N;: is the number of cycles to crack initiation at a stress amplitude o .

The crack spacing ratio (s/sf) which varies between 0 and 1, is an important element of this

equation. A smaller ratio implies the presence of an increased crack population. Whereas, a ratio of

one mean crack densities that are independent of stress or load. As this ratio could not be determined

due to experimental difficulties, an indirect approach should be used to obtain the s/sf values. The

best fits of Equation 9 to the experimental life data in the high cycle regime are presented in Figure

10. Using Mazundar's approach a good agreement between the predictions and the life data of both

131
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

(ö/öf)2 - 1

Unscratched
Ο Scratched (100 gm)
V Scratched (150 pm)
A Scratched (225 pm)

I I I I II I I
10 100 1000 10000
N u m b e r of c y c l e s to failure ( X 1 0 0 0 )
Figure 10. B e s t f i t s of E q u a t i o n 9 to the e x p e r i m e n t a l d a t a

unscratched as well as scratched samples containing 100 μηι, 150 μπι, and 225 μηι deep scratch is

obtained. With a presence of deeper scratch, the ratio s/sf decreases to a minimum value. To

compute the crack spacing ratio s/sf, for any scratch, a relationship as a function of scratch depth is

needed. Weibull distribution was used in Figure 11 which fitted the actual values obtained as,

/ \α
d
— = exp
I eJ (10)

The relation can be better understood as the probability to find an area without a micro crack

generated by the scratch. As the ratio decreases, the area (where no micro crack exists) decreases

and thereafter the life of the scratched material is reduced. This is not unreasonable in a sense that

a scratch by its presence generate micro cracks diminishing the effective area and acting as stress

132
Μ. Talia and J. Ε. Talia Journal of the Mechanical Behavior of Materials

C r a c k Spacing Ratio

S c r a t c h D e p t h [microns]

Figure 11. C r a c k spacing ratio as a function of s c r a t c h depth

concentrator. However, the principal mechanism appear to be the formation of micro cracks ahead

of the crack. In addition, microscopic examination of the scratched tip area, for example sample of

100 μηι deep scratched fatigue at regular intervals, revealed the development of first surface micro

cracks around 10000 cycles (Figure 12). Continuous examination of the scratch surface, showed also

the presence of additional micro cracks besides the growth of some existing ones. The imperfections

introduced, micro fissures, by the scratch magnify the probability of formation of micro cracks.

Therefore, the micro fissures in combination with the cyclic loads coalesce to form a favorable

structure for cracks propagation.

Comparison of Edge Scratch to Across the Width

A comparison of an edge-scratched to an across-the-width-scratched specimen can be seen

133
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Figure 12. Micro crack formation at 10,000 cycles (a) and at 15,000 cycles (b)

in Figure 13. Both samples showed similar crack propagation rates except in the short crack region

(da/dN < 3 10"10) where the crack path is affected by the scratch. Therefore, the difference is a

matter of initiation. It is not a surprising result since it was shown (SEM analysis) that the presence

of scratches generated micro cracks that weaken the material helping the short cracks to propagate

several times faster than long cracks. Consequently, any scratch (regardless of its length) generates

fissures/microcracks on the surface of the material that will shorten the life of the material.

134
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

d a / d N [m/cycle]
10
Δ edge scratched
Ο scratched (100 pm)

10

- 6
10

-7 J I I I I L J | I I I I II
10
10 100
Δ Κ [MPa /m]

Figure 13. Crack propagation rates for scratched and edge-scratched samples

In comparing the results obtained for short cracks using Equation 9 to da/dN results, several

steps has to be taken.

First, solving for o from Equation 9 gives,

b s{ (11)
cr = +1
N~1

where o is the stress amplitude required for crack initiation.

Second, rearranging for ΔΚ in Equation 4 as,

da
= c(YKtAa/a)n (12)
dN

and knowing Δσ = σ ^ - σ ^ , and R = σ ^ / σ ^ , then Δσ becomes,

135
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Δσ =
(13)

where omax is the same as σ in Equation 11.

Third, combining Equations 11 and 13, and substituting for Δσ in Equation 12, da/dN would

be obtained as.

rte n
— „J
= C (AK scratch ( 1x 4 )'
>

where,

b sf
AK scratch = Υ Κt +1 Of (1 -R) yß
\ N.1

Hence, Equation 14 (of Paris equation form) indicates that for short crack da/dN is a complex

function of several factors geometry factor (Y), stress loading, o=f(d,Ni), and crack length (a).

Figure 14 shows a good agreement of the predicted values of both short (Equation 14) and long

(Equation 4) crack compared to the experimental results. A careful examination of the development

of micro cracks must be considered and compared with Mazumdar model, before any attempt to

predict the da/dN values using Equation 12.

CONCLUSIONS:

This study showed that Al 2024-T clad is a scratch sensitive material. As the scratch depth

increased, the fatigue life of the material decreased significantly. Scratches generated micro fissures

which coalesce assisted by cyclic fatigue generating micro cracks. Micro crack weaken the material

ahead of the crack tip assisting cracks to propagated at higher rates than cracks in unscratched

136
Μ. Talia and J. Ε. Talia Journal of the Mechanical Behavior of Materials

da/dN [m/cycle]

Figure 14. Models prediction of scratched specimens

samples and at lower stress intensity factor. Using the experimental data, mathematical models of the

form of Paris equation were developed to predict both the short and long crack propagation in

scratched samples.

REFERENCES:

1. P. C. Paris and E. Erdogan, " A Critical Analysis of Crack Propagation Laws," Trans. ASME, J.

Basic Eng., VoLD85, 1963, pp. 528-534.

2. R. A. Smith, Fatigue Crack Growth 30 Years of Progress Ed., University of Cambridge, UK, Ed.

137
Vol. 8, No. 2, 1997 Crack Propagation Modeling for Surface Generated Scratches in Al 2024-T3
Clad Alloy

Pergamon Press, 1986.

3. J. C. Newman, Jr., Μ. H. Swain, and E. P. Phillips, "An Assessment of the Small-Crack Effect for

2024-T Aluminum Alloy," Small Fatigue Cracks, "Proc. of the Sec. Eng. Found Int.

Conf./Workshop, Santa Barbara California, 1986.

4. D. P. Wilhem, "Investigation of Cyclic Crack Growth Transitional Behavior," ASTM STP 415

,1967, pp. 363-368

5. S. J. Hudak Jr., Journal of Engineering Materials and Technology, 1988,pp. 103.

6. P. J. E. Forsyth and P. M. Powell, Journal ofMaterials Science, 1989, pp. 18.

7. J. Lankford, Fat. Eng. Mats, and Strucs. ,6, 1983, pp. 15-32.

8. W. L. Morris, O. Buck, and H. L. Marcus, (1976). Met.Trans. A7. 1161-1165.

9. S. Pearson, (1975). Eng. Frac. Mech. 7. 235-247.

10. M. H. El Haddad, N. F. Dowling, Τ. H. Topper, and Κ. N. Smith, (1980) International Journal

of Fracture, 16, 15.

11. J. M. Barsom, and S. T. Rolfe, Fracture & Fatigue Control in Structures, 2nd Edition, p. 42.

12. Ν. E. Dowling, "Fatigue at Notches and the Local Strain and Fracture Mechanics Approaches,"

in Fracture Mechanics, ASTM STP 677, C. W. Smith (ed.), American Society for Testing and

Materials, Philadelphia, 1979, pp. 247-273.

13. D. Taylor, "Fatigue Thresholds", p. 139, 1989.

14. J. Lankford, and S. J. Hudak, (1987) International Journal of Fatigue, 9, 87.

15. D. K. Holm, and A. F. Blom, (1984) "In the Fourteen Congress of the International Council of

Aeronautical Sciences, ICAS and AIAA, p.383:

16. D. Broek, Elementary Engineering Fracture Mechanics, 4th edition, 1986, p. 85.

17. J., Schijve, Engineering Fracture Mechanics, Vol. 11, 1979, pp. 167-221.

138
Μ. Talia and J.Ε. Talia Journal of the Mechanical Behavior of Materials

18. J., Schijve, Fatigue Thresholds, Fundamentals and Engineering Applications, Vol II, EMAS,

1982, pp. 881-908.

19. J. A. Bannantine, J. J. Corner, and J. L. Handrock, Fundamentals of Metal Fatigue Analysis,

1990, pp. 126.

20. R. E. Peterson, "Relation Between Life Testing and Conventional Tests of Materials, "Bulletin

ASTMNo. 133, 1945.

21. H. Neuber, Theory of Notch Stresses: Principle for Exact Stress Calculations, Edwards, Ann

Arbor, Mich., 1946.

22. G. Sines, J. L. Waisman, Metal Fatigue, 1959, pp. 300-302.

23. K. S. Chan and J. Lankford; Acta Metallurgica 36(1988)193.

24. K. S. Chan, J. Lankford, and D. L. Davidson; Journal of Engineering Materials and Technology

108(1986)201.

25. Κ. T. Rao, W. Yu, and R. O. Ritchie; Scripta Metallurgica 20(1986)1459.

26. P. K. Mazumdar; A model for High Cycle Fatigue, Engineering Fracture Mechanics, 1992.

27. Β. T. Ma and C. Laird, Acta Metallurgica 37(1989)349.

139