You are on page 1of 14

F2004/51

Mean Stress Effects in Stress-Life and Strain-Life Fatigue


Norman E. Dowling
Department of Engineering Science and Mechanics
Virginia Polytechnic Institute and State University
Blacksburg, Virginia 24061, USA

Copyright 2004 Society of Automotive Engineers, Inc.


properties from the same sources, specifically, 0.2% offset
yield strength o , ultimate tensile strength u , true stress
ABSTRACT
at fracture, ~
, and ductility, either percent reduction in
fB
Various approaches to estimating mean stress effects on area, or percent elongation, whichever is given. As indicated
stress-life and strain-life behavior are compared with test by the subscript B, the ~ f B values are corrected for hoop
data for engineering metals. The modified Goodman equa-
tion with the ultimate tensile strength is found to be highly stress due to necking according to Bridgman [16]. As noted,
inaccurate, and the similar expression of Morrow using the some values of ~ f B were unavailable and had to be esti-
true fracture strength is a considerable improvement. How- mated from those for similar material by assuming propor-
ever, the Morrow expression employing the fatigue strength tionality with the ultimate strength.
coefficient f may be grossly non-conservative for metals
TABLE 1 - Metals Studied, Sources of Fatigue Data, and
other than steels. The Smith, Watson, and Topper (SWT) Tensile Properties
method is a reasonable choice that avoids the above difficul- Ulti- Red
ties. Another option is the Walker approach, with an adjust- Yield Fracture
mate Area
able exponent that may be fitted to test data, allowing Material ~
[Data Source] o u fB (Elong)
superior accuracy. Handling mean stress effects for strain-
life curves is also discussed, including the issue of mathe- MPa MPa MPa %
matical consistency with mean stress equations expressed in
SAE 1015 Steel - [4] 228 415 726 68
terms of stress. A new and mathematically consistent
method for incorporating the Walker approach into strain- SAE 1045 Steel1 - [5] 1841 2248 2717 40.5
life curves is developed. With = 0.5, this result gives a AISI 4340 Steel2 - [6] 1103 1172 1634 56
new strain-based interpretation of the SWT method. 2014-T6 Al - [7] 438 494 580 4
(13.6)
4
INTRODUCTION 2024-T3 Al - [8, 9] 359 497 610 (20.3)
2
Mean stress effects have long been studied, as in the 2024-T4 Al - [10, 11] 303 476 631 35
early work of Gerber [1] and Goodman [2, 3], and one might 7075-T6 Al - [7] 489 567 7304 (16.5)
think that all has been said on the subject that needs to be Ti-6Al-4V3 - [12, 13] 930 978 13624 (20)
said. Nevertheless, several methods of questionable accu- 1 2
racy are currently in wide use. It is the purpose of this paper Notes: Hardness 595 HB. Fatigue specimens at longer lives
plastically strained prior to testing. 3Solution treated and vacuum
to examine the most widely used methods and to compare annealed. 4Values estimated from similar material in [14] or
their success in correlating fatigue data for engineering met- [15] by ratioing ultimate strengths.
als. The methods considered are those of Goodman, Mor-
row, Smith-Watson-Topper, and Walker. There are more With the aid of Fig. 1, let us be sure that the nomen-
than one version of some of these, and they may be used clature used herein is clear. The mean stress m is the av-
differently in the context of stress-life versus strain-life erage level of a constant amplitude cyclic loading, and the
curves.
stress amplitude a is the variation about this mean. The
Fatigue data will be analyzed for several steels and non- amplitude is also half of the overall stress range . The
ferrous metals as listed in Table 1, where references to the maximum and minimum values reached are, respectively,
sources of fatigue data are given. Table 1 also lists tensile max = m + a and min = m a . The ratio
R = min / max is also used to characterize the mean of the data flattens at short lives and the fit to Eq. 3 is not
stress situation. Further, note that very good. Table 2 gives values of f and b for the metals
to be studied here, where these values are from fitting the
max min max + min
a = , m = (a, b) (1) zero mean stress portion of the data.
2 2
max max At very short lives, stress-life data tend to approach the
a = (1 R ), m = (1 + R ) (a, b) (2)
2 2 true fracture strength from a tension test, ~ f B . If the fit to
Eq. 3 is quite good, then f , which is noted to be the inter-
Manipulating Eq. 1 into the product of max and an algebraic
expression, and invoking the definition of R, gives two addi- cept at one-half cycle, is approximately equal to the true
tional useful relationships, Eq. 2. fracture strength, f ~ f B . This is often the case for

max steels, as for SAE 1045 steel in Fig. 2.


m a 10000

a, Stress Amplitude, MPa


SAE 1045 Steel, 595 HB


0 1000
a
t
CCCCC

min
2024-T4 Al, Prestrained
100
1.E-01 1.E+01 1.E+03 1.E+05 1.E+07
Figure 1 Definitions for cyclic stressing Nf, Cycles

For the special case of stress amplitude a where the mean



Figure 2 Two stress-life curves with intercepts at cycle
stress is zero, m = 0, the notation ar is employed for the compared to true facture strengths. True fracture
amplitude. Such a situation of zero mean stress is also called f B and Intercept, f .
strength, ~
completely reversed cycling, and corresponds to R = 1.

In the treatment that follows, we will first briefly dis- However, where the data flatten at short lives, f may
cuss stress-life curves. Following this, we will present vari- considerably exceed ~
, as for 2024-T4 aluminum in Fig.
fB
ous methods for estimating mean stress effects, and then we
will look at the ability of these methods to correlate stress- 2. Depending on how the data are fit and the range of data
life data for various mean stresses. Next, we will consider available, the difference may be quite large. As an extreme
strain-life equations that include mean stress. Finally, con- example, consider the values in Tables 1 and 2 for 7075-T6
cluding remarks are given that are intended to interpret and aluminum, where we have ~ f B = 730 MPa. A stress-life fit
summarize the earlier portions of the paper. over a wide range of lives for data on similar material gives
STRESS-LIFE CURVES f = 1466 MPa. And a fit to the intermediate-to-long life

Stress-life curves are assumed to follow a power data to be analyzed gives f = 4402 MPa. Note that the
relationship.
two f values differ by a factor of three, and the larger
ar = f (2N f )b (3) one is six times larger than ~
. fB

where Nf is cycles to failure, and the stress variable is ar , Comparing the ~


f B values in Table 1 with the corre-
as the fitting constants f and b are determined from tests sponding f values in Table 2, the latter are typically
under zero mean stress, also called completely reversed higher than the former by a factor of three or four for the
tests. The degree to which actual data closely fit Eq. 3 var- nonferrous metals. This contrasts with the situation for the
ies. For example, in Fig. 2, data for SAE 1045 steel fit the three steels, where the values are similar.
expected straight line on this log-log plot fairly well. But
this is not the case for 2024-T4 aluminum, where the trend
AMPLITUDE-MEAN EQUATIONS
1.4

1.2 AISI 4340 Steel


Consider a set of fatigue data with cycles to failure at u = 1172 MPa
various stress amplitudes and mean stresses. The number of 1.0
cycles to failure Nf for each test can be used with Eq. 3 to 0.8

a/ar
calculate a value of completely reversed stress ar that is
0.6
expected to cause the same life as the actual combination of
0.4
amplitude and mean, a and m . One can then normalize
0.2 ~
the amplitudes to ar , plotting the ratio a / ar ,versus the u f B 'f
0.0
mean stress. For two of the data sets of current interest, such -400 0 400 800 1200 1600 2000
normalized amplitude-mean plots are given as Figs. 3 and 4.
m, MPa

TABLE 2 - Stress-Life Fitting Constants1, 2


Fit m = 0 Fit All Data to
Material Data Walker Figure 3 Normalized stress amplitude-mean plot for AISI
4340 steel.
'f b 'fw bw
SAE 1015 801 -0.113 799 -0.114 0.713 1.2

SAE 1045 3050 -0.098 (Not Done) 1.0 2024-T3 Al

AISI 4340 1758 -0.098 1963 -0.108 0.650 0.8

2014-T6 1120 -0.122 949 -0.108 0.480 a / ar


0.6

2024-T3 1602 -0.154 1772 -0.163 0.460 0.4

2024-T43, 4 1294 -0.142 2452 -0.195 0.505 0.2


u
~
fB
'f
5 0.0
7075-T6 1466 -0.143 (Not Done)
-200 0 200 400 600 800 1000 1200 1400 1600
6
7075-T6 4402 -0.262 (Not Done) m, MPa

Ti-6Al-4V 2749 -0.144 (Not Done)


Notes: 1Stress intercepts in MPa units. 2Data for Nf > 106 cycles
not included in fit if far from the log-log linear trend; runouts not Figure 4 Normalized stress amplitude-mean plot for
considered. 3Zero mean is overall fit for the same data from [17]. 2024-T3 aluminum.
4
Walker fit from Nf > 103 data. 5Overall fit for similar material
from [18]. 6Fit to data from [7]. a
ar = (5)
m
Note that the trend of such data is expected to pass through 1
a / ar = 1.0 at m = 0. A linear relationship is often as- u
sumed to occur, with the intercept along the a / ar = 0
If a given combination of stress amplitude and mean are
axis expected to be the static strength of the material.
substituted, the calculated value of ar can be thought of as
GOODMAN AND MORROW EQUATIONS - If the an equivalent completely reversed stress amplitude that is
static strength is taken as the ultimate tensile strength, the expected to cause the same life as the a , m combination.
straight line corresponds to One can then estimate the life by entering Eq. 3. Since the
fitting constants f and b are obtained by testing at zero
a m mean stress, it is not necessary to have data at non-zero
+ =1 (4) mean stress to make a life estimate. But the life estimate
ar u
does depend on the accuracy of Eq. 5. In Figs. 3 and 4, note
that the data tend to lie above the Eq. 5 line for tensile mean
This is the modified Goodman relationship as formulated by stress, causing conservative life estimates.
J. O. Smith [3]. It is useful to solve for ar .
Morrow [19] suggested modifying the Goodman rela-
tionship by employing the true fracture strength as the inter-
cept.
lustrated in Figs. 5 and 6 for the same sets of data. In Fig. 5,
a a the SWT equation ( = 0.50) deviates from the data for
ar = , ar = (a,b) (6)
AISI 4340 steel somewhat, but Walker with = 0.65 fits it
1 m 1 m
fB f quite well. For 2024-T3 aluminum in Fig. 6, the curves for
SWT and Walker with = 0.46 are close together, the dif-
where form (b) arises from estimating ~
f B as being equal to ference being less than the considerable scatter in the data.
The Walker equation has the obvious advantage of having
f . In Figs. 3 and 4, Eq. 6(a) fits the data quite well. Equa- an adjustable parameter to aid in fitting data. But this is
tion 6(b) seems to work equally well for the AISI 4340 steel of no benefit unless the value is known from mean stress
of Fig. 3, as the approximation f ~ f B works quite data for at least similar material.

well. But for 2024-T3 aluminum in Fig. 4, Eq. 6(b) com- WALKER EQUATION FITTING - To fit a set of ampli-
pletely misses most of the data due to f being far larger tude-mean-life data to the Walker equation, first write Eq. 3
than ~ . This of course arises from a stress-life curve that
fB
in the following convenient form:

does not fit Eq. 3 very well in the manner of the similar ar = AN bf , where A = f 2b (9)
aluminum alloy in Fig. 2.

SMITH-WATSON-TOPPER (SWT) AND WALKER Combine this with the Walker form of Eq. 8(b) and then
EQUATIONS - Numerous additional relationships have solve for Nf .
been proposed, including the widely used one of Smith,
Watson, and Topper [4].
1 R
ar = AN bf = max
2
ar = max a (a)
1/b (10)

1 R 1 R 1
ar = max (b) (7) Nf = max
2 2 A

2
ar = a (c)
1 R
1.6
where forms (b) and (c) are equivalent to (a) and are ob- AISI 4340 Steel
1.4
tained from (a) by making substitutions from Eq. 2(a). u = 1172 MPa
1.2

Another proposal is that of Walker [20], which may be 1.0


a / ar

written 0.8 = 0.650


0.6

ar = 1max a (a) 0.4
Data
SWT
= 0.500
1 R 0.2 Walker
ar = max (b) (8)
2 0.0
1 -1.2 -0.8 -0.4 0.0 0.4 0.8 1.2 1.6 2.0 2.4
2
ar = a (c) m / ar
1 R

where equivalent forms (b) and (c) are obtained from (a) by Figure 5 SWT and Walker amplitude-mean curves for
making substitutions from Eq. 2(a). The quantity is a AISI 4340 steel.
fitting constant that may be considered to be a materials
property. Obviously, all of the above forms of the Walker
equation reduce to the corresponding SWT forms for the
special case = 0.5.

Neither the SWT nor the Walker equations give a single


trend on a plot of the type of Figs. 3 and 4, forming instead
a family of curves. However, both do form a single curve if
the mean stress axis is also normalized by ar . This is il-
analysis. From Table 2 and other similar fitting for engi-
1.4 neering metals, values of seem to generally be around
Data 0.50 or above, with steels typically having higher values
1.2
SWT than nonferrous metals. Note that lower values of corre-
1.0 Walker spond to greater sensitivity to mean stress, there being no
effect for = 1, in which case ar = a
a / ar

0.8

0.6
= 0.500 CORRELATION OF STRESS-LIFE DATA
0.4

0.2 2024-T3 Al = 0.460 For a set of amplitude-mean-life data, it is useful to cal-


culate equivalent completely reversed stress amplitudes for
0.0 each test and plot these versus life. This is shown in Fig.
-0.4 0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 7(a-e) for AISI 4340 steel and in Fig. 8(a-e) for 2024-T3
m / ar aluminum. For each material, the ar values plotted are from
Eqs. 5, 6(a), 6(b), 7, or 8, respectively, for the (a) to (e) parts
Figure 6 SWT and Walker amplitude-mean curves for of each of Figs. 7 and 8. In these plots, the line fitted to Eq.
2024-T3 aluminum. 3 for the m = 0 data is shown, except for the Walker
correlations, where the line corresponds to fw and
Next, take the logarithm to the base 10 of both sides.
b w used with Eq. 3, as fitted to the entire set of data with
1 1 R 1 as listed in Table 2.
log N f = log max + log log A
b b 2 b The correlation is favorable to the extent that the data
(11) points fall very near the line. Points to the right of the line
correspond to lives longer than expected for the particular
Based on Eq. 11, we can now do a multiple linear regression ar equation, that is, conservative "predictions". Conversely,
with independent variables x1 and x2 and dependent vari- points to the left of the line indicate lives shorter than ex-
able y. pected, or nonconservative "predictions". Similar plots were
prepared for all of the metals listed in Table 2, except that
y = m1 x1 + m 2 x 2 + d (12) Walker correlations were not done for all of them, as noted.
The trends observed were similar to those seen for AISI
where 4340 steel and 2024-T3 aluminum in Figs. 7 and 8.

1 R In particular, the Goodman relationship, Eq. 5, gives


y = log N f , x1 = log max , x2 = log poor results, usually being excessively conservative for ten-
2
(13) sile mean stresses, but nonconservative for the limited data
1 1
m1 = , m2 = , d = log A available at compressive mean stresses. An exception is
b b b SAE 1045 steel, where the results are quite good, as might
be expected form u being only 17% below ~ f B .
Once the fitting constants m1, m2, and d are known, the de-
sired values are easily determined. 1000

1 m2
b= , = bm 2 = Mean Stress, MPa
m1 m1
ar , MPa

(14) 621

db d / m1 A 414
A = 10 = 10 , f = 207 Goodman
2b -207
0 AISI 4340 Steel
Table 2 gives the resulting three values, fw , b w , and , 0 Fit u = 1172 MPa

100
for five of the data sets, where subscripts w, for Walker, are
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06
added to avoid confusion with values fitted to only zero
Nf , Cycles
mean stress data.
Figure 7(a) Goodman life correlation for AISI 4340 steel.
It is significant that all of the stress-life data at all mean
stresses are now involved in the fit. Treating the data as a
single, larger set enhances the possibility of statistical
1000 1000

Mean Stress, MPa


ar , MPa

ar, MPa
621 Mean Stress, MPa
414 621
207 Morrow, Fracture 414
-207 207 Walker
AISI 4340 Steel -207 AISI 4340 Steel
0
u = 1172 MPa 0 u= 1172 MPa
0 Fit
Fit
100 100
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.0E+02 1.0E+03 1.0E+04 1.0E+05 1.0E+06
Nf , Cycles Nf, Cycles

Figure 7(b) - Morrow ~


f B life correlation for AISI Figure 7(e) Walker life correlation for AISI 4340 steel
with = 0.650.
4340 steel
1000
1000
Goodman
2024-T3 Al
Mean Stress, MPa
ar , MPa

621

ar, MPa
414
207 0.6
Morrow, Intercept
0.4
-207 0.02
0 AISI 4340 Steel -0.3
0 Fit u = 1172 MPa -0.6 R-ratio
-1
-1 Fit
100 100
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
Nf , Cycles Nf, Cycles

Figure 7(c) - Morrow f life correlation for AISI 4340 Figure 8(a) Goodman life correlation for 2024-T3 alumi-
steel. num.
1000 1000
Morrow, Fracture
2024-T3 Al

Mean Stress, MPa


ar , MPa

ar , MPa

621
0.6
414 0.4
207 SWT 0.02
-207 -0.3
AISI 4340 Steel -0.6 R-ratio
0 -1
u = 1172 MPa -1 Fit
0 Fit
100
100 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Nf , Cycles
Nf , Cycles
Figure 8(b) - Morrow ~
f B life correlation for 2024-T3 alu-
Figure 7(d) SWT life correlation for AISI 4340 steel.
minum.
1000 1000
0.6 0.6
0.4 R-ratio 0.4
R-ratio
0.02 0.02
-0.3
-0.3
-0.6
ar, MPa

-0.6
-1

ar, MPa
-1 Fit -1
Fit

Morrow, Intercept Walker


2024-T3 Al 2024-T3 Al
100
100
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
1.0E+02 1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07
Nf, Cycles
Nf, Cycles
Figure 8(c) Morrow f life correlation for 2024-T3 alu- Figure 8(e) Walker life correlation for 2024-T3 aluminum
minum. with = 0.460.
1000
0.6
0.4
R-ratio
0.02 1000
-0.3 0
Morrow, Fracture Mean Stress, MPa
-0.6
ar , MPa

34.5
-1 SAE 1015 Steel
69
-1 Fit
103

ar , MPa
-34.5
-69
0 Fit
SWT
2024-T3 Al
100
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
Nf , Cycles
100
Figure 8(d) SWT life correlation for 2024-T3 aluminum. 1.E+04 1.E+05 1.E+06
Nf , Cycles
The Morrow expression of Eq. 6(a) with the true fracture
Figure 9(a) - Morrow ~
f B life correlation for SAE 1015
strength ~
f B gives considerably better results than Good-
steel.
man in all cases, except for SAE 1045 steel, where the re-
sults are similar. For the three steels, the Morrow form of 1000
0
Eq. 6(b) with the intercept constant f gives essentially the SWT
SAE 1015 Steel Mean Stress, MPa 34.5

same result as Eq. 6(a), as expected due to ~ and 69


fB f 103
ar , MPa

having similar values. However, Eq. 6(b) gives very poor -34.5
-69
and nonconservative values for the nonferrous metals due to 0 Fit
the high values of f .

The SWT relationship of Eq. 7 gives good results in all


cases. For steels, it is not quite as good as Morrow with ~
f B 100
1.E+04 1.E+05 1.E+06
or f , and it tends to be nonconservative for compressive Nf , Cycles
mean stresses. But SWT is consistently better than Morrow Figure 9(b) - SWT life correlation for SAE 1015 steel.
with ~ f B for the nonferrous metals. The correlations for
Morrow with ~ and SWT, Eqs. 6(a) and 7, are given in
fB
The Walker expression of Eq. 8 always gives an excel-
lent correlation, as might be expected from its ability to ad-
Figs. 9(a,b) through 13(a,b) for SAE 1015 steel, SAE 1045 just the value of . Correlations have already been pre-
steel, 2014-T6 aluminum, 7075-T6 aluminum, and titanium
sented in Fig. 7(e) for AISI 4340 steel and Fig. 8(e) for
6Al-4V, respectively.
2024-T3 aluminum. In addition, Fig. 14 gives the plot for
SAE 1015 steel, and Fig. 15 for 2014-T6 aluminum.
Real data obey the above mathematical forms only im-
perfectly. To aid with accurately fitting each curve under
10,000
Morrow, Fracture these circumstances, it is recommended that the above two
SAE 1045 Steel equations be fitted separately to stress-strain-life test data,
595 HB
with theoretical relationships among the six fitting constants
not being invoked. See Landgraf [21] for discussion of Eqs.
ar , MPa

1,000 15 and 16, and see Dowling [22, 23] for descriptions and
0 and -138 to +138 discussion of the overall strain-based approach for making
690 life estimates for notched components.
-345 Mean Stress, MPa
1000
0 Fit
Approx. R
100 0.45
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 0.05
Nf , Cycles -0.37

ar, MPa
-1.0
Figure 10(a) - Morrow ~
f B life correlation for SAE 1045 -1.0 Fit

steel.
10,000 Morrow, Fracture
SWT 2014-T6 Al
SAE 1045 Steel
595 HB 100
1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
ar , MPa

Nf, Cycles
1,000
0 and -138 to +138 Figure 11(a) - Morrow ~
f B life correlation for 2014-T6
690
-345 Mean Stress, MPa
aluminum.
0 Fit
100 1000
1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 Approx. R
Nf , Cycles 0.45
0.05
Figure 10(b) SWT life correlation for SAE 1045 steel. -0.37
ar, MPa

-1.0
-1.0 Fit
STRAIN-LIFE EQUATIONS WITH MEAN STRESS

Mean stress adjustments are needed in making strain-


SWT
based fatigue life estimates. The materials properties needed 2014-T6 Al
to apply a strain-based approach are obtained from tests
100
under completely reversed controlled strain, so that mean
1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
stresses in the tests are at or near zero. Such test results pro- Nf, Cycles
vide a cyclic stress-strain curve and a strain-life curve,
which are usually represented by Figure 11(b) SWT life correlation for 2014-T6 aluminum.
1000
a a 1/ n Morrow, Fracture
a = + (15) 7075-T6 Al
E H
f
ar, MPa

ar = (2 N f ) b + f (2 N f ) c (16) 100 Approx. R


E 0.45
0.05
The quantity E is the elastic modulus. Both of these equa- -0.37
-1.0
tions represent summation of elastic strain (/E) and plastic -1.0 Fit
strain terms. Equation 15 gives the cyclic stress-strain curve, 10
in which H' and n' are fitting constants. For the strain-life 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08

curve of Eq. 16, the quantities f and b are the same as in Nf, Cycles

Figure 12(a) - Morrow ~


f B life correlation for 7075-T6
Eq. 3, and 'f and c are additional fitting constants for the
plastic strain term. aluminum.
1000 1000
SWT
7075-T6 Al

ar, MPa
ar, MPa

Approx. R
100 Approx. R
-1.0
0.45
0 to -0.78
0.05
0.1 to 0.2
-0.37 SWT
0.5
-1.0 Ti-6Al-4V
-1.0 Fit
-1.0 Fit
100
10
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08
Nf, Cycles Nf, Cycles

Figure 12(b) SWT life correlation for 7075-T6 aluminum. Figure 13(b) SWT life correlation for titanium 6A1-4V.

STRAIN-LIFE EQUATIONS FOR NONZERO MEAN MATHEMATICALLY CONSISTENT FORMS - We


STRESS - Since Eq. 16 gives only the life for zero mean can extend the ar equations such that the strain-life curve
stress, it needs to be generalized to include mean stress ef- is generalized in a mathematically consistent manner. Con-
fects. One equation that is often used for this purpose is an sider the general case of an amplitude-mean equation ex-
extension of the Morrow equation used with f , Eq. 6(b). pressed in terms of stress.

ar = f (a , m ) (19)
f m
a = 1 ( 2 N f ) b + f ( 2 N f ) c (17)

E f Simple substitutions based on the definitions of the various
stress variables of Fig. 1, such as Eq. 2, allow this relation-
ship to be expressed in terms of max and R, or in terms of
Another is an extension of the SWT relationship, obtained
a and R, so that Eq. 19 can be any of Eqs. 5 to 8.
by replacing a in Eq. 7(a) with a . Combining this with
Eq. 16 gives the following relationship for determining life.
1000
Walker 0
( f )2 + c
SAE 1015 Steel Mean Stress, MPa 34.5
max a = (2 N f )2b + f f (2 N f )b (18) 69
103
E -34.5
ar, MPa

-69
Although Eqs. 17 and 18 may give reasonable life estimates, Fit

neither is mathematically consistent with their parent ar


equations expressed in terms of stress. Also, in view of the
discussion above, the quantity m / f in Eq. 17 should be
100
replaced by / ~
for nonferrous metals, so that it is more
m fB 1.0E+04 1.0E+05 1.0E+06
Nf, Cycles
similar to Eq. 6(a).
1000 Figure 14 Walker life correlation for SAE 1015 steel with
= 0.713.

We can now manipulate Eq. 3 to obtain a relationship


between stress amplitude and life, with the mean stress ef-
ar, MPa

Approx. R
fect included on the Nf side of the equation. To proceed,
-1.0
0 to -0.78
first combine Eqs. 3 and 19 to obtain
0.1 to 0.2
0.5 f ( a , m )
-1.0 Fit
Morrow, Fracture
ar = f (a , m ) = a = f (2 N f )b (20)
100
Ti-6Al-4V
a
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07
Nf, Cycles Then solve for stress amplitude a and manipulate the
Figure 13(a) - Morrow ~
f B life correlation for titanium stress quantities on the right side of the equation to be
6Al-4V.
within brackets with Nf , allowing us to define an equivalent PARTICULAR CASES - Now let us consider particular
life N*. cases of ar = f ( a , m ) . For the Morrow form of Eq.
1000
6(a), substitution into Eq. 22 gives
Approx. R
0.45 1/ b
0.05 m
*
=N
-0.37 N mf f 1 ~ (25)
ar, MPa

-1 f B
Fit

Note that subscripts mf are added to N* to specify the Mor-


Walker
row equation based on the true fracture strength.
2014-T6 Al
100 0.100
1.0E+03 1.0E+04 1.0E+05 1.0E+06 1.0E+07 Mean Stress, MPa
AISI 4340 Steel
621
Nf, Cycles u = 1172 MPa
414

a, Strain Amplitude
Figure 15 Walker life correlation for 2014-T6 207
-207
aluminum with = 0.480. 0.010 0
Fit

1/ b b

a
a = f 2 N f = f (2 N *)b (21) = 0.650
f ( a , m )

0.001
1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08
An explicit expression for N* is thus
N*w , Walker Equivalent Cycles

1/ b Figure 16 Strain amplitude versus Walker equivalent life


a
N* = Nf (22) N*w for AISI 4340 steel with = 0.650.
f ( a , m )

Hence, one can determine the life N* that is expected for a 0.100
given stress amplitude a under zero mean stress, and then = 0.505 0
Mean Stress, MPa 72
estimate the life Nf , as affected by a nonzero mean stress,
a, Strain Amplitude

140 to 230
by solving Eq. 22 for Nf . 290
-70 to -135
0.010 Fit
1/ b
a
Nf = N* (23)
f ( a , m )
2024-T4 Al
Prestrained
The effect on life must be the same regardless of 0.001
whether one employs a stress-life or a strain-life curve. This 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08
permits Eq. 16 to be generalized to N*w, Walker Equivalent Cycles

f Figure 17 Strain amplitude versus Walker equivalent life


a = (2 N *)b + f (2 N *)c (24) N*w for 2024-T4 aluminum with = 0.505.
E

where N* is the life calculated from the strain amplitude a Where f ~f B might be a good approximation, as for
as if the mean stress were zero, and then Nf as affected by
steels, Eq. 6(b) applies, and Eq. 22 yields
the nonzero mean stress is obtained from Eq. 23. Also, on a
strain-life plot, data plotted as a versus the equivalent life 1/ b
m
N* are expected all fall together along the curve for zero *
N mi = N f 1
(26)
mean stress, Eq. 16. This is demonstrated in Figs. 16 and 17 f
for AISI 4340 steel and 2024-T4 aluminum, respectively,
where the f ( a , m ) is in this case based on the Walker The added subscripts now indicate Morrow and intercept.
expression, Eq. 8. Note that Eqs. 24 and 26 combined are not the same as Eq. 17,
as the latter has no mean stress adjustment for the second
term.
TABLE 3 - Additional Constants for Strain-Life Curve
For the Walker relationship, the form of Eq. 8(c) and Eq. *
Fit N w vs. P Cyclic - Curve1, 2
22 give Material
1 R
(1 ) / b 'fw cw H' n'
*
Nw = Nf (27) AISI 4340
2 0.624 -0.620 207,000 1655 0.131
Steel3
where subscript w of course specifies Walker. Letting = 2024-T4 Al4 0.632 -0.858 73,100 738 0.080
0.5 gives us the special case of this for the SWT equation. Notes: 1Units are MPa for E and H'. 2Values from [6] for AISI 4340
and [17] for 2024-T4. 3Fit to P > 1.5 104 and Nw < 5 105. 4Fit to
1/( 2b) P > 1.5 104 and Nw < 104.
* 1 R
N swt = Nf (28)
2 First, employ the cyclic stress-strain curve, Eq. 15, to
estimate strain amplitudes a from stress amplitudes a for
It is useful to write as a single equation the form provided by any tests in the data set where strain was not measured.
Eqs. 24 and 27 for the Walker relationship. (This is often the case for tests run in stress control at rela-
(1 ) / b b (1 ) / b c
tively long lives.) Next, using b w and , calculate values
f
1 R

1 R
a = 2 N f + f 2 N f *
of N w from Eq. 27 for each data point. Then use a and
E 2 2

*
(29) Nw to calculate values of the second (plastic strain) term of
Eq. 24 by subtracting the first (elastic strain) term from a .
Similarly, Eqs. 24 and 28 give the corresponding form for
the SWT equation.
* cw
fw * bw
P = fw (2 N w ) = a (2 N w ) (31)
f 1/(2b) b 1/(2b ) c E
1 R 1 R
a = 2N f + f 2 N f (30)
E 2 2 where the values of the plastic strain term are denoted P as a
convenience. Now do a least squares fit using these P and
Note that the latter is not the same as Eq. 18, but instead * *
Nw values by first solving for 2 N w .
represents a different generalization of the strain-life equa-
tion that is consistent with SWT expressed in terms of stress,
1/ c
Eq. 7. P w
*
2N w =
(32)
CORRELATION OF STRAIN-LIFE DATA fw

As already noted, strain-life correlations based on the Taking logarithms of both sides of this equation gives
Walker mean stress equation are given in Figs. 16 and 17 for
AISI 4340 steel and 2024-T4 aluminum, respectively. In 1 1
each case, the data points are seen to agree closely with the
*
log (2 N w )= log P log fw (33)
cw cw
strain-life curve for zero mean stress, Eq. 24, indicating suc-
cess for Eq. 27. Note that the strain values plotted are simply A linear regression can now proceed using
strain amplitudes, a . Correlation of the data for various
mean stresses is achieved by plotting on the horizontal axis y = mx + d (34)
*
Walker equivalent lives N w
from Eq. 27, in which b w and
where
from Table 2 are employed.
*
y = log (2 N w ), x = log P
The curves shown in Figs. 16 and 17 correspond to the
usual form of strain-life equation for zero mean stress, Eq. (35)
1 1
24, but new values of the fitting constants are employed. m= , d = log fw
This is done to take advantage of the ability of the Walker cw cw
equation to include all of the data at all mean stresses in the
fit, as previously described. The constants used with Eq. 24 Once the fitting constants m and d have been determined, the
are the Walker stress-life fitted values, fw and b w from desired values are easily obtained.

Table 2, along with fw and c w as listed in Table 3. Fitting 1


cw = , fw = 10 dcw = 10 d / m (36)
of fw and c w to the test data needs to be described. m
In performing these fits, it was found that the data ex- Additional study may allow generic values of to be
hibited extreme scatter at low P values corresponding to developed for various classes of alloy, so that any disadvan-
* tage of the Walker equation is removed. For example, based
long equivalent lives N w . Accordingly, each fit was re-
stricted to the region of well behaved data by a dual crite- on the limited study done so far, 0.50 may be a good
rion, specifically, a lower limit on P and an upper limit on choice for aluminum alloys. (Note that if = 0.50 exactly,
* this method is equivalent to SWT.) For steels, a higher ge-
Nw , both of which had to be satisfied for a data point to be
used. See the notes to Table 3 for the actual values chosen neric value appears to be appropriate, perhaps 0.65.
for these limits. Also, higher strength steels are generally more sensitive to
mean stress that lower strength ones, so that it may be possi-
DISCUSSION AND SUMMARY ble to develop a correlation between and say ultimate
tensile strength.
We have compared various stress amplitude-mean equa-
tions as to their ability to correlate fatigue data. It is clear Stress amplitude-mean equations may be incorporated
that the Goodman equation employing the ultimate tensile into strain-life equations in a mathematically consistent
strength u is inaccurate. The Morrow equation using the manner. (Some strain-life or related equations in common
true fracture strength ~ works well, but has the disadvan-
fB use are not mathematically consistent with their parent
stress-based equations.) The procedure for doing so for any
tage that values of ~
f B are not always available. An empiri- stress amplitude-mean relation is given, as well as the par-
cal study to develop estimates of this quantity from other ticulars for those studied. This logic leads to Eq. 30, which
tensile properties would enhance its usefulness. is a new strain-life extension of the SWT mean stress equa-
tion.
If the Morrow equation is instead employed with the in-
tercept constant f , the results are still quite good for Similarly extending the Walker mean stress relation
gives an entirely new version of the strain-life equation, Eq,
steels. However, this is not the case for aluminum alloys or
29. This strain-life equation includes SWT as the special
for the one titanium alloy studied, where Morrow with f
case for = 0.50. It should be further evaluated and com-
is seen to be highly inaccurate and nonconservative. This pared to experimental data. Data fits for a steel and for an
difficulty is associated with stress-life behavior that does not aluminum alloy give excellent results.
fit a power law very well, in particular, with stress-life
curves that tend to flatten at short lives. As a result, f CONCLUSIONS
values may be quite large, far exceeding ~ , and causing
fB The following conclusions are drawn from the discus-
the highly inaccurate behavior. Hence, Morrow with f sion above:
should not be used for aluminum alloys or for other metals 1) The Goodman mean stress equation employing the ul-
where such stress-life behavior occurs. timate tensile strength u is inaccurate, being exces-
The Smith, Watson, and Topper (SWT) equation gave sively conservative for tensile mean stresses.
good results for all cases studied. If it is desired to choose
2) The Morrow mean stress equation using the true frac-
one simple stress amplitude-mean equation for all metals,
SWT would be the preferred choice. Another option would ture strength ~
f B works well for various metals, but has
be to use SWT except where Morrow with f is known to the disadvantage that values of ~ are not always
fB
give good results, as for steels. available.
The Walker equation has the advantage of enhanced 3) The Morrow mean stress equation with f works well
ability to fit data by its use of the adjustable parameter ,
for steels. But it is highly inaccurate and nonconserva-
which may be considered to be an additional materials prop- tive for materials with log-log stress-life behavior that
erty. However, there is the accompanying disadvantage in flattens at short lives. Thus, it should not be used for
that values are not generally known unless fatigue data at aluminum alloys.
various mean stresses are available for a given material.
Where is known, the Walker equation appears to be quite 4) The Smith, Watson, and Topper (SWT) mean stress
accurate and is the best mean stress equation of those stud- equation is a good choice for general use. It is quite ac-
ied. curate for aluminum alloys, and for steels it is accept-
able, although not quite as good as Morrow with f .
5) The Walker mean stress equation with adjustable con- Steel," NACA TN 2324, National Advisory Commit-
stant gives superior results where is known or can tee for Aeronautics, Washington, DC, March 1951.
be estimated.
[9] Illg, W., "Fatigue Tests on Notched and Unnotched
6) Any future work on mean stress equations should con- Sheet Specimens of 2024-T3 and 7075-T6 Aluminum
centrate on the Walker relationship, such as identifying Alloys and of SAE 4130 Steel with Special Consid-
generic values of for various classes of metal, or de- eration of the Life Range from 2 to 10,000 Cycles,"
veloping correlations for estimating from tensile NACA TN 3866, National Advisory Committee for
Aeronautics, Washington, DC, Dec. 1956.
properties.

7) The incorporation of the Walker equation into the [10] Topper, T. H., and B. I. Sandor, Effects of Mean
strain-life curve, Eq. 29, is a promising approach that Stress and Prestrain on Fatigue Damage Summation,
should be further evaluated and employed. Effects of Environment and Complex Load History on
Fatigue Life, ASTM STP 462, Am. Soc. for Testing
REFERENCES and Materials, West Conshohocken, PA, 1970, pp.
93-104.
[1] Mann, J. Y., The Historical Development of Re-
search on the Fatigue of Materials and Structures, [11] Endo, T., and J. Morrow, Cyclic Stress-Strain and
The Journal of the Australian Institute of Metals, Fatigue Behavior of Representative Aircraft Metals,
Nov. 1958, pp. 222-241. Journal of Materials, ASTM, Vol. 4, No. 1, March
1969, pp. 159-175.
[2] Goodman, J., Mechanics Applied to Engineering,
Longmans, Green and Co., London, 1919, pp. 631- [12] Gallagher, J. P., et al., Improved High Cycle Fatigue
636. (HCF) Life Predictions, AFRL-ML-WP-TR-2001-
4159, Air Force Research Laboratory, Wright-
[3] Smith, J. O., The Effect of Range of Stress on the Patterson Air Force Base, OH, Jan. 2001.
Fatigue Strength of Metals, Bulletin No. 334, Uni-
versity of Illinois, Engineering Experiment Station, [13] Knipling, K., High-Cycle Fatigue/Low-Cycle Fa-
Urbana, IL, Feb. 1942. See also Bulletin No. 316, tigue Interactions in Ti-6Al-4V, MS Thesis, Materi-
Sept. 1939. als Science and Engineering Department, Virginia
Tech, Blacksburg, VA, Jan. 2003.
[4] Smith, K. N., P. Watson, and T. H. Topper, "A Stress-
Strain Function for the Fatigue of Metals," Journal of [14] Conle, F. A., R. W. Landgraf and F. D. Richards,
Materials, ASTM, Vol. 5, No. 4, Dec. 1970, pp. 767- Materials Data Book: Monotonic and Cyclic Proper-
778. ties of Engineering Materials, Ford Motor Co., Scien-
tific Research Staff, Dearborn, MI, 1984.
[5] Landgraf, R. W., Effect of Mean Stress on the Fa-
tigue Behavior of a Hard Steel, Report No. 662, [15] SAE, Technical Report on Low Cycle Fatigue Prop-
Dept. of Theoretical and Applied Mechanics, Univer- erties of Wrought Materials, SAE J1099, Informa-
sity of Illinois, Urbana, IL, Jan. 1966. tion Report, Society of Automotive Engineers,
Warrendale, PA, 1989. See also L. E. Tucker, R. W.
[6] Dowling, N. E., Fatigue Life and Inelastic Strain Landgraf and W. R. Brose, Proposed Technical Re-
Response under Complex Histories for an Alloy port on Fatigue Properties for the SAE Handbook,
Steel, Journal of Testing and Evaluation, ASTM, SAE Paper No. 740279, Automotive Engineering
Vol. 1, No. 4, July 1973, pp. 271-287. Congress Detroit, MI, 1974.

[7] Lazan, B. J., and A. A. Blatherwick, "Fatigue Proper- [16] Bridgman, P. W. The Stress Distribution at the Neck
ties of Aluminum Alloys at Various Direct Stress Ra- of a Tension Specimen, Trans. of ASM International,
tios: Part 1, Rolled Alloys," WADC TR 52-307, Part Vol. 32, 1944, pp. 553-574.
1, Wright Air Development Center, Wright-Patterson
AFB, OH, Dec. 1952. [17] Dowling, N. E., and A. K. Khosrovaneh, Simplified
Analysis of Helicopter Fatigue Loading Spectra, J.
[8] Grover, H. J., S. M. Bishop, and L. R. Jackson, "Fa- M. Potter and R. T. Watanabe, eds., Development of
tigue Strengths of Aircraft Materials: Axial-Load Fa- Fatigue Loading Spectra, ASTM STP 1006, Am.
tigue Tests on Unnotched Sheet Specimens of 24S-T3 Soc. for Testing and Materials, West Conshohocken,
and 75S-T6 Aluminum Alloys and of SAE 4130 PA, 1989, pp. 150-171.
[18] Dowling, N. E., Mechanical Behavior of Materials: tance in Metals and Alloys, ASTM STP 467, Am.
Engineering Methods for Deformation, Fracture, and Soc. for Testing and Materials, West Conshohocken,
Fatigue, 2nd ed., Prentice Hall, Upper Saddle River, PA, 1970, pp. 3-36.
NJ, 1999, p. 655.
[22] Dowling, N. E., and S. Thangjitham, "An Overview
[19] Morrow, J., Fatigue Properties of Metals, Section and Discussion of Basic Methodology for Fatigue,"
3.2 of Fatigue Design Handbook, Pub. No. AE-4, So- Fatigue and Fracture Mechanics: 31st Volume,
ciety of Automotive Engineers, Warrendale, PA, ASTM STP 1389, G. R. Halford and J. P. Gallagher,
1968. Section 3.2 is a summary of a paper presented eds., Am. Soc. for Testing and Materials, West Con-
at a meeting of Division 4 of the SAE Iron and Steel shohocken, PA, 2000, pp. 3-36.
Technical Committee, Nov. 4, 1964.
[23] Dowling, N. E., "Local Strain Approach to Fatigue,"
[20] Walker, K., "The Effect of Stress Ratio During Crack Chapter 4.03, Volume 4, Cyclic Loading and Fatigue,
Propagation and Fatigue for 2024-T3 and 7075-T6 R. O. Ritchie and Y. Murakami, volume editors; part
Aluminum," Effects of Environment and Complex of the 10-volume set, Comprehensive Structural In-
Load History on Fatigue Life, ASTM STP 462, Am. tegrity, B. Karihaloo, R. O. Ritchie, and I. Milne,
Soc. for Testing and Materials, West Conshohocken, overall editors, Elsevier Science Ltd., Oxford, Eng-
PA, 1970, pp. 1-14. land, 2003
[21] Landgraf, R. W., The Resistance of Metals to Cyclic
Deformation, Achievement of High Fatigue Resis-

You might also like