Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Dowling Material Properties

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Fatigue & Fracture of

Engineering Materials & Structures


doi: 10.1111/j.1460-2695.2008.01322.x

Mean stress effects in stress-life fatigue and the Walker equation


N . E . D O W L I N G 1, C . A . C A L H O U N 2 a n d A . A R C A R I 2
1 Materials

Science and Engineering Department, and Engineering Science and Mechanics Department (Jointly Appointed), 2 Engineering Science and
Mechanics Department, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA
Received in final form 12 October 2008

A B S T R A C T Mean stress effects in finite-life fatigue are studied for a number of sets of experimental

data for steels, aluminium alloys and one titanium alloy. Specifically, the agreement with
these data is examined for the Goodman, Morrow, SmithWatsonTopper and Walker
equations. The Goodman relationship is found to be highly inaccurate. Reasonable accuracy is provided by the Morrow and by the SmithWatsonTopper equations. But
the Morrow method should not be used for aluminium alloys unless the true fracture
strength is employed, instead of the more usual use of the stress-life intercept constant.
The Walker equation with its adjustable fitting parameter gives superior results. For
steels, is found to correlate with the ultimate tensile strength, and a linear relationship
permits to be estimated for cases where non-zero mean stress data are not available.
Relatively high-strength aluminium alloys have 0.5, which corresponds with the
SWT method, but higher values of apply for relatively low-strength aluminium alloys.
For both steels and aluminium alloys, there is a trend of decreasing with increasing
strength, indicating an increasing sensitivity to mean stress.
Keywords fatigue of materials; Goodman equation; mean stress effect; Morrow equation; SmithWatsonTopper equation; stress-life curve; Walker equation.
NOMENCLATURE

A = intercept constant at 1 cycle for a stress-life curve


b = exponent constant for a stress-life curve
b w = exponent constant for a Walker method stress-life fit
c = exponent constant for a plastic strain versus life curve
d = intercept constant for multiple linear regression
E = elastic modulus
m 1 , m 2 = slope constants for multiple linear regression
n = number of data points for an s z calculation
N = life for a given a for the m = 0 case
N w = value of N from the Walker method
N f = fatigue life; cycles to failure
R = stress ratio, R = min /max
s z = stress deviation; the standard deviation of z for a set of data
z = normalized stress-direction deviation of a data point relative to a stress-life
curve
 = stress range,  = 2a
a = strain amplitude
ar = strain amplitude for m = 0
f = intercept constant at 1/2 cycle for a plastic strain versus life curve

Correspondence: N. E. Dowling. E-mail: ndowling@vt.edu


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

163

164

N . E . D O W L I N G et al.

= fitting constant for the Walker method


a = stress amplitude
ar = stress amplitude for m = 0; equivalent completely reversed stress amplitude
arN = point on a ar versus N f curve at the same N f as arP
arP = value of ar for the stresses applied in a test
f = intercept constant at 1/2 cycle for a stress-life curve
fw = intercept constant at 1/2 cycle for a Walker method stress-life fit
fB = true fracture strength, corrected
m = mean stress
max = maximum stress
min = minimum stress
u = ultimate tensile strength
INTRODUCTION

The effect of mean stress in fatigue of materials has been a


subject of study for well over a century. Numerous methods have been proposed and employed for handling mean
stresses in the calculation of allowable cyclic stresses and
fatigue lives. However, despite all of this work, there is
no general consensus as to the suitability of various approaches, and methods of questionable accuracy continue
to be recommended in textbooks and given as options in
computer programs.
This paper attempts to aid the situation by reporting an
extensive study of existing fatigue data that include mean
stress effects. The analysed data are limited to engineering metals and to axially loaded, unnotched specimens.
Included are a total of 28 sets of data, with 18 of these
being for steels, nine for aluminium alloys and one for a
titanium alloy. Each set of data is compared to the behaviour expected from the modified Goodman, Morrow,
SmithWatsonTopper and Walker methods. Comparisons are made in the context of stress-life curves in the
finite-fatigue-life region. Fatigue limits, that is, stress levels below which fatigue damage is presumed not to occur,
are not specifically included in the study.
The work reported here follows a preliminary assessment of eight sets of data by one of the authors.1 In that
paper, the Walker method was found to be worthy of
special attention, which emphasis in continued here.
The nomenclature employed is defined in Fig. 1, where
a is stress amplitude, m is mean stress, and  = 2a
is the stress range. The stress ratio R = min /max is also
employed. It is useful to note some relationships among
the quantities used to describe cyclic loading.
max min
2
max + min
m =
2
max
a =
(1 R)
2
max
m =
(1 + R)
2
a =

(a)
(b)
(c)
(d).

(1)

max

0
t

a
min

Fig. 1 Constant amplitude cyclic stressing and definitions of stress


variables.

The data analysed are identified in Table 1 as to the


material, literature references,215 and tensile and other
properties. Some of the materials are listed with the AISISAE or Aluminium Association designations which are
prevalent in the United States, and others with German
nomenclature from the handbooks of Baumel and Seeger4
and of Boller and Seeger.12

MEAN STRESS RELATIONSHIPS

As a basis for comparison, it is assumed that each material


has a stress versus fatigue life relationship for completely
reversed (zero mean stress) loading of the following standard form:
ar = f (2N f )b .

(2)

The quantities f and b are fitting constants, N f is cycles


to failure and ar is the stress amplitude. The subscript r
is added to a to indicate the m = 0 special case. This
power relationship of course gives a straight line on a
loglog plot, as shown for two cases in Fig. 2.
Constants for Eq. (2), as fitted in this study to the m = 0
data in each set, are given in Table 1. A reasonable fit over
the entire range of the data was generally possible. There
were three exceptions, all aluminium alloys, where the
data extended to very short lives. In these cases, the data
at short lives fell well below the trend of the data at longer


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

2, 3
4
4
5, 6
4
4
7
7
4
4
8
5, 6
7
9
9
10
11
10

9
4, 12
13, 14
4
15
5, 6
15
15
5, 6

SAE 1015 St
GSMnNi63 Stc,d
Ck45 St
SAE 4130 St, Normd
49MnVS3 St
17MnCrMo33 St
CC 450 SS, H1150d
CC 450 SS, Soln And
50CrMo4 St, El Pol
42CrMo4 St
AISI 4340 Stc
SAE 4130 St, Hardd
CC 450 SS, H900d
PH138Mo SS, H1000d
300M Std
SAE 1045 St, 705 HB
SAE 1045 St, 55 HRC
SAE 1045 St, 595 HB

6061-T6 Ale
Al Mg4.5Mn, Cld Rlc,e,f
2024-T4 Alc,f
2024-T3 Alc,e,f
2014-T6 Ale
2024-T3 Ale
2024-T4 Ale
7075-T6 Ale
7075-T6 Ale

Ti-6Al-4V c,e
1006

276
298
303
378
438
359
335
489
521

228
312
531
647
566
833
678
1041
970
998
1103
1200
1354
1358
1634
1827
1731
1841

1034

310
363
476
486
494
497
502
567
572

415
501
790
799
840
929
1015
1065
1086
1111
1172
1241
1405
1413
1958
2082
2165
2248

Ultimate
u (MPa)

1271

420
476
631
578
581
591
666
729
736

726
846
1271
1144
1152
1446
1360
1410
1609
1525
1634
1586
1750
1758
2303
2131
2690
2717

Fracture
fB (MPa)

(14.5)

(13)
35
(17.3)
(13.6)
(20.3)
(21.4)
(16.5)
(12.3)

67.9
(26)
60
(15.2)
44
58
(23.2)
(12.2)
48.6
60
56
(8.3)
(15.3)

2
38
40.5

Red. Areag
(Elong.) (%)
990
739
1217
1584
1263
1275
1453
2046
1588
1505
1963
2378
2839
2521
4214
2370
3762
3034
651
1045
2499
1611
949
1772
897
3378
4546
2452

0.1379
0.0811
0.0984
0.0845
0.0948
0.0666
0.0591
0.0600
0.0841
0.0758
0.0958
0.1182
0.0840
0.0995
0.1150
0.0750
0.1075
0.1014
0.1151
0.1372
0.1755
0.1324
0.1221
0.1544
0.1051
0.2624
0.2388
0.1285

2262

722
1045
1944
1328
1120
1602
970
4402
4324

1020
672
1267
1168
1325
1232
1365
1326
1547
1452
1722
2437
2317
2889
3612
2418
3542
3149

0.1379

0.1080
0.1394
0.1969
0.1529
0.1080
0.1626
0.09771
0.2412
0.2419

0.1336
0.0925
0.0936
0.1040
0.0899
0.0752
0.0668
0.1078
0.0899
0.0810
0.1077
0.1160
0.1083
0.0879
0.1318
0.0728
0.1147
0.0985

0.5431

0.6330
0.6681
0.5031
0.5655
0.4803
0.4596
0.4090
0.4150
0.4774

0.7352
0.8113
0.6949
0.6903
0.8492
0.6575
0.6253
0.7051
0.7776
0.7781
0.6497
0.5457
0.4758
0.5769
0.4157
0.4839
0.4286
0.5245

fw (MPa)

f (MPa)
bw

Fit all m to Walker Equation

Fit m = 0

214

5353
2800
5100
4249
1200
663
1000
1000
5,182

52
15
217
211
246
23
157
231
52
248
222
74
223
2797
684
2.5
26
20

Min (cycles)

1,100,000

83,870,000
2,000,000
1,836,000
2,700,000
7,260,000
1,404,000
29,400,000
929,000
1,992,000

1,241,000
1,900,000
620,000
1,984,000
1,710,000
2,840,000
72,950
94,613
380,000
520,000
901,430
1,023,000
90,941
1,266,000
652,600
161,250
750,791
572,000

Max (cycles)

Range of Walker Fit

a Abbreviations: St = steel, SS = stainless steel, Al = aluminium alloy, HB = Brinell hardness, HRC = Rockwell hardness, El Pol = electropolished, Soln An = solution annealed, H900, etc.
= heat treatment conditions, Norm = normalized, Hard = hardened, Cld Rl = cold rolled.
b Numbers refer to the list of References at the end of the paper.
c Samples initially overstrained before testing.
d True fracture strength estimated from + 345 MPa.
u
e True fracture strengths estimated by scaling data on similar material.
f Short life data omitted in stress-life fits; for the = 0 fits, the shortest life data points employed were N = 4750 cycles for Al Mg4.5Mn, 4200 cycles for 2024-T4 Al, and 2050 cycles for
m
f
2024-T3 Al.
g Reduction in area is listed, or if not available, elongation is listed.

Sourceb

Materiala

Yield
o (MPa)

Table 1 Materials, tensile properties and stress-life fitting constants

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION


165

166

N . E . D O W L I N G et al.

Another relationship is that of Smith, Watson and


Topper.3 Three equivalent forms, related by manipulations employing Eq. (1), are
ar =

max a

1 R
ar = max
2

2
ar = a
1 R

Fig. 2 Stress-life curves for a steel and for an aluminium alloy for
zero mean stress, showing intercepts and true fracture strengths.

lives, with the lower line in Fig. 2 being an example of


this behaviour. For these three cases, which are identified
by a footnote in Table 1, the constants given correspond
to a fit that omits the short life data that depart from the
trend of the remaining data.
To obtain the estimated fatigue life N f for cases of nonzero mean stress, an additional equation, ar = f (a , m ), is
needed to calculate ar for use with Eq. (2). In this context,
ar can be thought of as an equivalent completely reversed
stress amplitude. Below, various ar = f (a , m ) equations
are presented, as is a special data fitting procedure for the
Walker method.
Equivalent completely reversed stress amplitude
equations
The traditional Goodman relationship,16 as employed by
J.O. Smith,17 includes the ultimate tensile strength, u .
ar =

a
.
1 m /u

(3)

The equation of Morrow18 has the same form but employs


the true fracture strength fB , as corrected according to
Bridgman,19 in place of u . An alternate form that is often
used is obtained by substituting the fitting constant f , so
that the two forms are
a
a
ar =
, ar =
. (a, b)
(4)
1 m / fB
1 m / f
For Eq. (2), note that f is the stress intercept at
N f = 0.5 cycles. Interpreting a tension test as a fatigue test
with this life implies the estimate f = fB , which is the
logic leading to Eq. (4b). The f = fB estimate is often
quite good for steels, as shown by Landgraf,20 but may
be very inaccurate for aluminium alloys, as for the cases
illustrated in Fig. 2. Equation (4a) has the disadvantage
that true fracture strengths fB are often not available and
then must be estimated.

(a)
(b)
(c).

(5)

For brevity, this method will be termed the SWT


method. It has the advantage of simplicity and of not
depending on any material constant.
A somewhat similar approach is that of Walker,21 which
involves an additional materials property, here denoted .
Three corresponding equivalent forms are
1

ar = max a


1 R
ar = max
2
1

2
ar = a
1 R

(a)
(b)
(c).

(6)

Note that max = m + a , so that Eqs (5) and (6) also


provide a relationship ar = f (a , m ).
Values of for Eq. (6) can conceivably vary from 0 to 1,
but for metals are usually in the range 0.4 to 0.8. If =
0.5, then Eq. (6) is seen to reduce to Eq. (5). The quantity
(1 ) can be thought of as a measure of the materials
sensitivity to mean stress. Hence, a metal with = 0.4
is quite sensitive to mean stress, and one with = 0.8 is
relatively insensitive.
Another equation that is sometimes employed is the
Gerber parabola.
ar =

a
.
1 (m /u )2

(7)

From the historical review of Mann,22 this method dates


to 1874 and is perhaps the oldest mean stress equation. However, Eq. (7) was not included in this study,
as it is generally inaccurate and moreover usually nonconservative for tensile mean stresses. Also, it incorrectly
predicts a harmful effect of compressive mean stress, and
so would have to be employed in some modified form for
such cases. These trends can be seen in Fig. 3, where the
parabola from Eq. (7) is compared with test data for an
alloy steel and with the straight lines from Eqs (3) and
(4). (See Table 2 for an explanation of conservative and
non-conservative trends.)
Numerous additional equations for estimating mean
stress effects exist. For example, some others are discussed


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION


1/b


1 R 1
N f = max
2
A

1.6
AISI 4340 Steel
u = 1172 MPa

1.4

a/

ar

1.2

0.8
Gerber

0.0
-400

Goodman
u

400

fB

800
1200
m , MPa

y = m1 x1 + m2 x2 + d ,

'f
1600

2000

by Nihei,23 by Chu24 and by Wang.25 The methods of


Eqs (3)(6) were judged by the authors to posses various
combinations of wide acceptance, simplicity, and promise
as to make them particularly worthy of study.
Fitting the Walker equation

where

m1 =

1
,
b

x1 = log max ,

m2 =

,
b

A = f 2b .

(8)

Combine this with Eq. (6b) and solve for the dependent
variable N f .


1 R
ar = ANbf = max
(a)
2

x2 = log

1
d = log A.
b

1 R
2

(12)

The regression yields values of the fitting constants m 1 ,


m 2 and d, which in turn can be used to calculate the desired
constants f , b and .
b=

1
,
m1

= bm2 =

A = 10db = 10d /m1 ,

In applying any one of the Goodman, Morrow or SWT


relationships to make a life estimate, note that the ar
value from one of Eqs (3) to (5) is employed with Eq. (2),
which is based on fitting only the test data for completely
reversed (m = 0) loading. In contrast, the Walker relationship provides an opportunity to fit fatigue life data at
various mean stresses all together, with being obtained
as part of the fitting process. The procedure involves a
multiple linear regression. Details follow.
To proceed, it is convenient to temporarily employ an
alternate form of Eq. (2).

(11)

where
y = log N f ,

Fig. 3 Normalized stress amplitudemean plot for AISI 4340 steel,


showing the curves for various mean stress equations.

ar = ANbf ,

(9)

A multiple linear regression can then be performed.

Morrow

0.4
0.2

(b).

Then take the logarithm with base 10 of both sides.




1
1

1 R
log A. (10)
log N f = log max + log
b
b
2
b

1.0

0.6

167

m2
m1
f =

A
.
2b

(13)

To distinguish values of f and b obtained by this procedure from those obtained by mere fitting of zero mean
stress data, subscripts w for Walker are added, so that

the resulting three constants are denoted fw
, b w and .
Hence, the fitted stress-life curve is

ar = fw
(2N f )b w

(14)

with ar given by Eq. (6) and the fitted . Table 1 gives


the resulting constants from fits of this type for each data
set.
For the three aluminium alloys having behaviour similar
to the lower curve in Fig. 2, a number of cycles N 2 was
chosen by judgment as the separation between the long
and short life behaviour. The Eq. (14) fit was applied only
to the data with N f > N 2 , which in these cases included
all of the data with mean stresses that differed significantly
from zero.

Table 2 Explanation of trends in graphical comparisons


Trend seen on a normalized
amplitudemean plot
a /ar vs. m
or a / ar vs. m / ar

Trend seen on an equivalent completely


reversed stress amplitude versus life plot
ar vs. N f

Consequence
for life estimates

Data above line or curve


for a given model

Data above m = 0 (R = 1) line,


or above Walker fit line

Conservative: Actual fatigue life will be


longer than estimated

Data below line or curve


for a given model

Data below m = 0 (R = 1) line,


or below Walker fit line

Non-conservative: Actual fatigue life


will be shorter than estimated


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

N . E . D O W L I N G et al.

168

COMPARISONS WITH TEST DATA

1.2
2014-T6 Al
1.0
0.8
a / ar

Fatigue test data for various mean stresses can be


compared with candidate mean stress equations ar =
f (a , m ) on stress amplitude versus mean plots, as in
Fig. 3. However, a better sense of the success of such an
equation for estimating fatigue lives is provided by plotting ar values versus fatigue lives N f . All of the data sets
were compared with Eqs (3)(6) by both of these graphical
methods. In what follows in this section, the two graphical methods are described in more detail and examples
are presented, and some comments are also given on the
data sets that were analysed.

0.6
0.4
0.2
0.0
-200

200

~
fB

400

600

'f
800

1000

1200

m , MPa

Normalized amplitudemean plots

Fig. 5 Normalized stress amplitudemean plot for 2014-T6


aluminium.

The relationships of Goodman or Morrow can be compared to test data by plotting normalized stress amplitudes
a /ar versus the mean stress m . For the Goodman relationship, Eq. (3), a straight line is expected between the
points a /ar = 1 and m = u . The two forms of the
Morrow relationship, Eqs (4a) and (4b), give similar lines,
but with intercepts at m = fB and m = f , respectively. Example comparisons with test data on this basis
are shown for steels in Figs 3 and 4 and for an aluminium
alloy in Fig. 5. For the data points shown, each a and m
combination corresponds to a laboratory test, and in each
case the ar value used to normalize a is determined from
the experimental life N f using Eq. (2).
For the steels, note that the data are above the Goodman
line for tensile mean stresses, that is, higher stresses than
predicted by this method are required to cause fatigue failure. Hence, fatigue life predictions made from a Goodman ar value for tensile mean stresses will be shorter than
actual fatigue lives, and the method thus errs on the conservative side. (See Table 2.) However, for compressive
mean stresses, the opposite is true, and the method is nonconservative. Either form of the Morrow relationship is

in better agreement with the test data for both tensile


and compressive mean stresses. For the aluminium alloy,
the Goodman equation is again conservative for tensile
mean stresses, and the Morrow relationship of Eq. (4a)
with fB is in good agreement with the data. However,
the Morrow relationship of Eq. (4b) with f disagrees
with the data and is non-conservative to a drastic degree.
This trend seems to be associated with a stress-life curve
that flattens at short life, as for the aluminium alloy in
Fig. 2.
On a plot of a /ar versus m , the SWT and Walker
relationships produce a family of curves, complicating
comparisons with test data. However, both form a single
curve on a plot with the mean stress axis also normalized
to ar , that is, on a plot of a /ar versus m /ar . Comparisons on this basis are shown in Figs 68 for the same
three materials, with ar for the Walker curves being obtained from Eq. (14). For the steels, the fitted value of
for the Walker relationship gives much better agreement
with the data than = 0.5 corresponding to the SWT
1.6

50CrMo4 Steel
u = 1086 MPa

1. 2

1.2

0.6

= 0.650

0.8
0.6
0.4

0.4

0.2

0.2
u

0.0
-400

1.0

ar
a/

ar

1.0
0.8

AISI 4340 Steel


u = 1172 MPa

1.4

1.4

400

'f

800
1200
P
,
M
a
m

0.0
-1.2

fB

1600

2000

Fig. 4 Normalized stress amplitudemean plot for 50CrMo4 steel.

Data
SWT
Walker
-0.6

= 0.500

0.0

0.6
/

1.2

1.8

2.4

ar

Fig. 6 SWT and Walker amplitudemean curves for AISI 4340


steel.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

1.4
1.2
= 0.778

0.8

ar

1.0

= 0.500

0.6
0.4

Data
SWT
Walker

0.2
0.0
-0.6

-0.4

-0.2

50CrMo4 Steel
u = 1086 MPa
0.0

0.2
m/

0.4

0.6

0.8

1.0

ar

Fig. 7 SWT and Walker amplitudemean curves for 50CrMo4


steel.

1.4
Data
SWT
Walker

1.2

0.8

ar

1.0

0.6

= 0.500

0.4
0.2
0.0
-0.4

= 0.480

2014-T6 Al
0.0

0.4

0.8

1.2
/

1.6

2.0

2.4

2.8

ar

Fig. 8 SWT and Walker amplitudemean curves for 2014-T6


aluminium.

equation. However, for the aluminium alloy, the fitted


is close to 0.5, so that both the Walker and SWT methods
are in good agreement with the data.
The other steels studied had behaviour generally similar
to that of the two steels illustrated, except that the Walker
curve may be more shallow, more steep, or similar to
the SWT one, depending on the fitted value. Further,
the other aluminium alloys had behaviour similar to that
of the one illustrated, except that the Walker and SWT
curves differed considerably for the two lowest strength
alloys, for which 0.65.
Life plots
Now consider comparison of mean stress equations and
test data on a stress-life plot of ar versus N f . In this case,
ar is calculated for each test data point using one of the
ar = f (a , m ) expressions above, and these values are
plotted versus the experimental fatigue lives N f . Example
plots of this type for 50CrMo4 steel and 2014-T6 aluminium are shown in Figs 9 and 10. Note that for each


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

169

material there are four plots, one for each of Eqs (3),
(4a), (4b) and (5). On any particular plot, the degree to
which the resulting data points consolidate with the line of
Eq. (2) gives an indication of the success of the mean stress
relationship. (Recall that the Eq. 2 constants are obtained
from fitting the m = 0 data.)
For the Walker relationship, ar values from Eq. (6) are
similarly plotted versus the N f data for comparison with
the special fit of Eq. (14), which is noted to result from
fitting the full set of data at all mean stresses. Comparisons
of this type are shown in Figs 11 and 12 for 50CrMo4 steel
and 2014-T6 aluminium.
The trends seen in the ar versus N f plots correspond
to those already noted for steels and aluminium alloys on
the two types of amplitudemean plot. On any particular
plot, where test data points lie above the line for zero mean
stress, the estimated lives are too short, and the method
errs on the conservative side. Conversely, data points below the line indicate that the method is non-conservative.
(See Table 2.) The ar versus N f comparisons shown for
50CrMo4 steel and 2014-T6 aluminium are typical of the
steels and aluminium alloys studied, respectively. Where
the Walker differs significantly from 0.5, the Walker
correlation is noticeably better than the SWT one, and
the Walker method always gives very good correlation of
the test data.
Comments on the data sets analysed
For 18 of the 28 data sets, the tests were conducted in
strain control, with stress control sometimes being substituted for tests at relatively long lives where the strains
were fully elastic. However, only stress control was employed for the seven data sets referenced to sources in
the 1950s, and also for the three data sets obtained from
MIL-HDBK-5H.9
Tabulated numerical data were utilized, usually from the
source referenced, but in a few cases from data obtained
through contacting the original authors. The only exceptions were the data from MIL-HDBK-5H, which were
digitized from stress-life plots in that publication.
In strain-controlled tests at very short lives, that is,
around N f < 50 cycles, and especially for N f < 10 cycles, the very large strains involved may cause buckling or
other anomalous behaviour that affects the life. Hence,
data points in this range that seemed to depart from the
trend of the data at longer lives were not included, specifically, by omitting all points with N f less than a value
chosen as a matter of judgment. In the data sets for stresscontrolled tests, there were sometimes points with anomalously short lives where the maximum stress was around
or even greater than the materials yield strength. Such
data points were omitted, as they are surely affected by
large amounts cycle-dependent creep (ratcheting).

170

N . E . D O W L I N G et al.

(b) Morrow, Fracture

(a) Goodman
1000

100
10

102

0
0.01
-0.01
R=0
R = -2
0 Fit

ar

ar

, MPa

0
0.01
-0.01
R=0
R = -2
0 Fit

, MPa

1000

Mean Strain
or R

50CrMo4 Steel
u = 1086 MPa

103
104
Nf , Cycles

105

106

100
10

102

Mean Strain
or R

103

50CrMo4 Steel
u = 1086 MPa
104

105

106

Nf , Cycles

(c) Morrow, Intercept

(d) SWT

1000

ar

100
10

102

, MPa

0
0.01
-0.01
R=0
R = -2
0 Fit

0
0.01
-0.01 Mean Strain
R = 0 or R
R = -2
0 Fit

ar

, MPa

1000

Mean Strain
or R

103

50CrMo4 Steel
u = 1086 MPa
104

105

106

Nf , Cycles

100
10

102

103

50CrMo4 Steel
u = 1086 MPa
104

105

106

Nf , Cycles

Fig. 9 Equivalent completely reversed stress amplitude versus life correlations for 50CrMo4 steel for the (a) Goodman, (b) Morrow fracture
strength, (c) Morrow intercept, and (d) SWT methods.

Many of the data sets had a small number of points at lives


around or beyond N f = 106 cycles that departed from the
trend of the data at shorter lives. Also, there were sometimes a few tests stopped without failure (runouts), where
the material began to exhibit fatigue limit behaviour.
There was seldom sufficient data of these types to enable any sort of systematic analysis, and so no attempt
was made to do so. Hence, in such cases, all points were
omitted with N f greater than a value chosen as a matter of judgment. As a result of these omissions, the data
points included for steels exceed N f = 2 106 cycles in
only one case, but the data included for some aluminium
alloys extend well beyond this. Table 1 gives the smallest
and largest N f values that are included in the analysis for
each data set. As indicated by a note to the table for three
aluminium alloys, short-life data points were excluded, so
that anomalous behaviour similar to that of Fig. 2 is not
included in the analysis. For these, the smallest N f values
included in the Walker fits of Eq. (14) are the values tabulated, and the smallest N f values for the m = 0 fits of
Eq. (2) are given in the note to the table.
For some data sets, the samples were subjected to several cycles of initial overstrain at a level around 1 to

2% strain. This follows the logic of Topper and coworkers,14,26 whose work indicates that such overstrains
are useful in advancing the fatigue damage process, so that
the behaviour is similar to that during service loading with
occasional severe cycles. These cases are identified by a
footnote in Table 1.
Additional data sets were located, as in MIL-HDBK5H, but were not included, as they covered only a narrow range of N f values or included only a small amount
of data for non-zero mean stresses. The data on hotrolled and normalized SAE 1045 steel of Fatemi27 were
analysed, but the results are not reported here. Relaxation during these strain-controlled tests resulted in quite
small mean stress values, which hardly exceeded the scatter in the zero mean stress data, causing difficulty with
the Walker fit. As reported by Fatemi, this set of data is
correlated well by the SWT equation. Also analysed but
not included are the data on Ti-6Al-4V generated in a
U.S. Air Force program.28 In this extensive data set, the
stress-life curve begins to flatten around N f = 3 105
cycles and the data extend to 107 cycles. It was omitted only because a more complex form of stress-life
equation would be needed than for the other data


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

171

1000

1000
-1

Approx. R

0.45

MPa

0.05

0.45

-1 Fit

(a) Goodman
2014-T6 Al

(b) Morrow, Fracture


2014-T6 Al

100
103

-0.37

0.05
-1 Fit

ar N

-1

ar ,

MPa

ar P

ar ,

Approx. R

-0.37

104

105
Nf , Cycles

106

107

1000

100
103

104

105
Nf , Cycles

106

107

1000
-1
Approx. R

-1
Approx. R

-0.37

MPa

0.45

MPa

ar ,

-1 Fit

ar ,

0.05

0.05
0.45
-1 Fit

(d) SWT
2014-T6 Al

(c) Morrow, Intercept


2014-T6 Al
100
103

104

-0.37

105
Nf , Cycles

106

107

100
103

104

105
Nf , Cycles

106

107

Fig. 10 Equivalent completely reversed stress amplitude versus life correlations for 2014-T6 aluminium for the (a) Goodman, (b) Morrow
fracture strength, (c) Morrow intercept and (d) SWT methods.

1000

Walker

Approx. R

ar

ar ,

, MPa

0
0.01
-0.01
R=0
R = -2
0 Fit

MPa

1000

Mean Strain
or R

50CrMo4 Steel
u = 1086 MPa

100
10

102

103

10
04

10
05

106

Nf , Cycles
Fig. 11 Equivalent completely reversed stress amplitude versus life
for the Walker method applied to 50CrMo4 steel, for which =
0.778.

sets. However, the SWT equation provides an excellent


correlation.
In applying the Morrow relationship in the form of
Eq. (4a), it was necessary in many cases to estimate the true


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

-1
-0.37
0.05
0.45
Fit

Walker
2014-T6 Al
100
10 3

10
04

105
Nf , Cycles

10 6

107

Fig. 12 Equivalent completely reversed stress amplitude versus life


for the Walker method applied to 2014-T6 aluminium, for which
= 0.480.

fracture strength fB , as this was not given with the data.


For most of the aluminium alloys and for the titanium
alloy, fB values for a different batch of the same or a very
similar alloy were found in the 1998 SAE Handbook29

172

N . E . D O W L I N G et al.

or in Smith et al.30 The missing values were then estimated by multiplying the known one by the ratio of
the ultimate tensile strengths u for the two batches of
material.
u2
fB2 = fB1
.
(15)
u1
However, for 6061-T6 aluminium, no value could be
found in the literature, and so one was obtained by conducting three tension tests on this alloy, averaging the
resulting properties, and then applying Eq. (15).
For steels where fB was not given with the data, the
following estimate was employed:
fB = u + 345 MPa.

(16)

This estimate has been widely employed elsewhere but is


commonly considered to be valid only for u up to about
1700 MPa. However, the Eq. (16) estimate was compared
with fB data for 25 steels with u fairly evenly spread over
the range 400 to 2600 MPa, and it was found to provide
a satisfactory correlation over this entire range. The data
examined include those in Table 1 of this study where no
estimate was necessary, as well as data in Landgraf 10 and
in Tables 4.2, 4.6 and 14.1 of the book by Dowling.31
DISCUSSION

Graphical comparisons as in Figs 312 provide a good


sense of the success of the various mean stress equations
under study. However, it is not feasible to present similar
plots for all 28 sets of data examined, and even if this were
done, the information provided would be in very diffuse
form. Hence, some means of quantitatively summarizing
the success of a given ar equation for a given set of experimental data is needed. In this section of the paper, we
will first describe such a method, which will then serve as
a basis for evaluating and discussing the relative success
of the mean stress relationships of Eqs (3)(6). Next, we
will examine trends in values of the fitting constant of
the Walker equation, and, finally, we will briefly consider
extensions of mean stress equations to strain-life curves.
Quantifying comparisons of mean stress equations
and test data
Consider ar versus N f plots as in Figs 9 and 10. If the (ar ,
N f ) points for the experimental data for non-zero mean
stress all lie close to the zero mean stress fit, corresponding
to Eq. (2), then the mean stress equation used to calculate
the ar values is quite successful. Conversely, the mean
stress equation is less successful to the extent that the
data scatter widely. To obtain a quantitative measure of
the scatter, consider the points arP and arN as labelled
in Fig. 10a. The quantity arP corresponds to the value
calculated from the stresses applied in the test, plotted

versus the experimental life N f . And arN is the point


on the zero mean stress fit at the same life N f . Hence,
arN is calculated by entering Eq. (2) with the N f value
corresponding to arP .
The difference (arP arN ) is a measure of the discrepancy, and a dimensionless measure is obtained by normalized this to the arN value.
z=

arP
arP arN
=
1.
arN
arN

(17)

The standard deviation of the variable z is considered to


be a measure of the overall scatter. Considering all nonzero mean stress data points in a given comparison, where
n is the number of these, we have

n
i=1 zi
.
(18)
sz =
n1
We will call this s z the stress deviation.
To similarly summarize the success of a Walker equation
fit, such as those of Figs 11 and 12, the Eq. (14) line is used
as the basis of the comparison. Hence, Eq. (14) is used to
compute arN for each N f value. All data points in the
range of the Eq. (14) fit are included in the calculation,
including the zero mean stress ones.
Success of the various mean stress equations
Values of the stress deviation s z of Eq. (18) were calculated
as just described for all 28 data sets, in each case for ar
values calculated from each of Eqs (3), (4a, b), (5) and (6).
These s z values are given in Table 3, as are the numbers
of data points employed in this analysis for each material,
both the total numbers and the numbers with approximately zero mean stress. Also, the s z values are shown in
bar chart form for steels in Fig. 13 and for aluminium
alloys and the single titanium alloy in Fig. 14. In each bar
chart, the materials are ordered according to the ultimate
tensile strength u , and the numbers on the horizontal
axes correspond to numbers given in Table 3. Of course,
smaller s z values indicate greater success and larger ones
the opposite. In Table 3 and Fig. 13, a comparison is
also shown for estimated values for steels, which will be
discussed later.
Plots similar to Figs 312 for all data sets, as summarized
by the bar charts of Figs 13 and 14 and the values in
Table 3, lead to the following observations concerning
the success of the various mean stress equations:
The Goodman relationship of Eq. (3), which employs
the ultimate tensile strength u , is quite inaccurate and is
considerably less accurate than any of the other equations
examined. Values of s z for steels are generally around
0.2, and for aluminium alloys generally around 0.3 or
above. The method is excessively conservative for tensile
mean stresses, but non-conservative for compressive ones.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

173

Table 3 Values of stress deviation for each material and mean stress equation
Number of data points

s z , Stress deviation

m 0

Goodman

Morrow
fB

Morrow
f

SWT

Sourcea

Number on
bar chartb

Total

Material

Walker
Fit

Walker
Est.

SAE 1015 St
GSMnNi63 St
Ck45 St
SAE 4130 St, Norm
49MnVS3 St
17MnCrMo33 St
CC 450 SS, H1150
CC 450 SS, Soln An
50CrMo4 St, El Pol
42CrMo4 St
AISI 4340 St
SAE 4130 St, Hard
CC 450 SS, H900
PH138Mo SS, H1000
300M St
SAE 1045 St, 705 HB
SAE 1045 St, 55 HRC
SAE 1045 St, 595 HB
Steel Average

2, 3
4
4
5, 6
4
4
7
7
4
4
8
5, 6
7
9
9
10
11
10

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

32
31
33
65
30
70
25
25
25
29
25
33
28
17
71
13
28
16

12
14
21
19
16
23
11
9
11
19
10
18
9
6
27
8
13
9

0.142
0.205
0.098
0.382
0.129
0.269
0.069
0.224
0.248
0.237
0.236
0.170
0.159
0.525
0.178
0.091
0.139
0.090
0.200

0.080
0.129
0.082
0.217
0.108
0.079
0.044
0.124
0.104
0.120
0.077
0.111
0.087
0.251
0.145
0.083
0.096
0.068
0.111

0.075
0.147
0.082
0.211
0.102
0.109
0.044
0.139
0.114
0.133
0.070
0.111
0.100
0.061
0.233
0.055
0.148
0.084
0.112

0.118
0.195
0.094
0.220
0.126
0.077
0.038
0.086
0.143
0.146
0.118
0.096
0.076
0.108
0.123
0.059
0.084
0.105
0.112

0.069
0.116
0.069
0.107
0.061
0.061
0.042
0.111
0.041
0.047
0.048
0.103
0.077
0.046
0.099
0.052
0.086
0.070
0.073

0.070
0.133
0.080
0.134
0.097
0.093
0.040
0.080
0.081
0.082
0.062
0.096
0.105
0.067
0.119
0.060
0.069
0.120
0.088

6061-T6 Al
Al Mg4.5Mn, Cld Rl
2024-T4 Al
2024-T3 Al
2014-T6 Al
2024-T3 Al
2024-T4 Al
7075-T6 Al
7075-T6 Al
Aluminum Average

9
4, 12
13, 14
4
15
5, 6
15
15
5, 6

1
2
3
4
5
6
7
8
9

50
22
45
22
32
68
37
34
112

13
11
7
6
6
14
12
9
49

0.518
0.655
0.205
0.344
0.265
0.281
0.467
0.243
0.439
0.380

0.179
0.218
0.130
0.120
0.101
0.118
0.132
0.203
0.172
0.152

0.147
0.191
0.233
0.266
0.199
0.252
0.249
0.394
0.327
0.251

0.222
0.188
0.061
0.124
0.091
0.083
0.107
0.150
0.165
0.132

0.117
0.078
0.052
0.105
0.066
0.081
0.067
0.102
0.156
0.091

Ti-6Al-4V

10

21

14

0.283

0.049

0.267

0.096

0.064

a Numbers
b Numbers

refer to the list of References at the end of the paper.


identify materials in Fig. 13 or 14.

Amplitudemean plots, such as those of Figs 35, show the


source of these trends, as the data are above the Goodman
line for tensile mean stresses, but below for compressive
ones. (See Table 2.) Consequently, for life plots as in
Figs 9a and 10a, the ar data for tensile mean stress are
often far above the zero mean stress line, and those for
compressive mean stress are usually below.
The Morrow relationship in the form of Eq. (4a), with
the true fracture strength fB , is considerably more accurate than the Goodman relationship. Values of s z for steels
average around 0.11, and for aluminium alloys around
0.15, which still indicates reasonable agreement, but not
as good as for steels. On amplitudemean plots, such as
in Figs 35, the corresponding line tends to agree with
the data. And for ar versus N f plots as in Figs 9b and
10b, only modest scatter about the zero mean stress line
is seen.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

For steels, the Morrow relationship of Eq. (4b), with the


intercept constant f , is also reasonably accurate, which
is expected, as fB and f have similar values for most
steels. Values of s z again average around 0.11, and good
correlation can be seen on amplitudemean plots, such
as Figs 3 and 4, as well as on ar versus N f plots, as in
Fig. 9c.
However, for aluminium alloys, the Morrow f method
is quite inaccurate, with an average s z of 0.25. This situation is associated with a stress-life trend similar to that for
the aluminium alloy in Fig. 2, where the stress values at
short lives tend to approach fB , rather than the considerably larger f value obtained from fitting data at relatively
long lives. The artificially large f value causes a drastic
lack of correlation that can be seen both on amplitude
mean plots, as in Fig. 5, and on ar versus N f plots, as
in Fig. 10c. Note that the direction and magnitude of

174

N . E . D O W L I N G et al.

Fig. 13 Stress deviation values s z for steels for the (a) Goodman, (b) Morrow fracture strength, (c) Morrow intercept, (d) SWT and (e)
Walker methods. Also, s z values are given in (f) for Walker values estimated from ultimate tensile strengths. The numbers on the
horizontal axes identify the steels listed in Table 3.

the disagreement is such as to cause non-conservative life


estimates that can be dangerously inaccurate. Because it is
often the case that only relatively long life data are available for fitting Eq. (2), this difficulty with the Morrow
f method represents a fatal flaw as to its usefulness for
aluminium alloys.
Considering the SWT method, the overall agreement
for steels is similar to that for either version of the Morrow
method, with the average s z for all three being around
0.11. For aluminium alloys, the agreement is better than
for even the Morrow fB method, with the average s z
now being 0.13. Also, good correlation is seen on ar
versus N f plots, as in Figs 9d and 10d. The SWT method
does tend to give somewhat non-conservative estimates
for compressive mean stresses in steels, which can be seen
in Fig. 9d. However, considering its overall performance,
the SWT method is a good choice for general use.
The Walker method gives excellent results where data
are available for fitting to evaluate the adjustable param-

eter . The average s z values are 0.07 for steels and 0.09
for aluminium alloys. Noting these values and looking at
the bar charts of Figs 13 and 14, as well as the typical life
correlations of Figs 11 and 12, it is clear that this method
gives the best results of all those considered. This might
be expected, as the ability to vary a fitting parameter is
not present for any of the other methods. On normalized
amplitudemean plots similar to Figs 68, adjusting is
seen to allow varying the resulting curve to fit data in cases
where the SWT method is not very accurate. Moreover,
on such a plot, data tend to correlate rather tightly about
a curve that fits a particular value.
Considering the single titanium alloy studied, it is of
course not possible to draw general conclusions for titanium alloys from one set of data. Nevertheless, it is notable
that the behaviour was generally similar to that for aluminium alloys, with poor correlation for the Goodman
and Morrow f methods, and good or excellent agreement for the remaining three methods.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

175

Fig. 14 Stress deviation values s z for


aluminium and titanium alloys for the (a)
Goodman, (b) Morrow fracture strength, (c)
Morrow intercept, (d) SWT and (e) Walker
methods. The numbers on the horizontal
axes identify the alloys listed in Table 3.

Comments on errors in life


Consider either an individual z value of 0.10, or a stress
deviation s z for a set of data of 0.10. This corresponds
to an error in stress of a factor of 1.10. The error in life
associated with this value depends on the exponent b or
b w of Eq. (2) or (14), that is, on the slope of the stress-life
curve on a loglog plot. For a typical value of b = 0.10,
the corresponding error in life, termed the life factor, is
life factor = [1 + (z or s z )]1/b = 1.1010 = 2.59

(19)

Hence, the fatigue life deviates from the zero mean stress
(or Walker-fitted) line by a factor of 2.59. Of course,
when the above calculation is applied to a z value, the
result characterizes the error in life for only an individual
data point, and when applied to an s z value, the result
characterizes the overall error in life for the set of data.
For the typical b = 0.10, some additional values of the
life factor are given below.

z or s z
Life factor

0.072
2.00

0.10
2.59

0.20
6.19

0.30
13.8

0.50
57.7

Thus, for b = 0.10, the z or s z cannot exceed 0.072 for the


life to not be more than a factor of two in error. Values


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

of s z similar to the average for the Goodman method


applied to steels of 0.2, or the average for aluminium
alloys of 0.38, correspond to quite large errors in life,
and the same applies for the Morrow f method applied
to aluminium alloys. Also, in the comparisons made for
the cases just mentioned, individual data points for the
larger mean stress values often had values around z = 0.5
or larger, examples of which can be seen in Figs 9a, 10a
and 10c. Further, for the same cases, there were s z values
for several sets of data in the range 0.3 to 0.6, which
corresponds to quite extreme errors in life.
The values of b and b w in Table 1 vary considerably,
and similar calculations can of course be done for any
particular value of b or b w . For values of b smaller
than 0.10, the life factors increase relative to the values for b = 0.10, and they decrease for b larger than
0.10. Even considering the variation of b, the overall picture that emerges is that use of the less successful mean
stress equations can result in unacceptably large errors in
life.

Trends in fitted Walker values for steels


The success of the Walker method suggests that it would
be worthwhile to develop methods for estimating values
for materials where non-zero mean stress data for fitting

N . E . D O W L I N G et al.

176

Walker, Estimated

0.8

1000

0.6

, MPa

Steels

ar

0
0.01
-0.01
R=0
R = -2
0 Fit

0.4
= 0.0002000

0.2

+ 0.8818
100
10

0.0
0

500
1000
1500
2000
u , Ultimate Tensile Strength, MPa

Fig. 15 Trend of fitted values with ultimate tensile strength for


the Walker method applied to steels.

are not available. We will now explore this possibility for


steels and for aluminium alloys.
The fitted Walker values from Table 1 for steels are
plotted versus the ultimate tensile strength u in Fig. 15.
A trend of decreasing with increasing u is seen, and
fitting a linear relationship gives
= 0.0002000u + 0.8818

(u in MPa).

Mean Strain
or R

1 02

(20)

For this fit, the coefficient of determination R2 is 0.68.


A similar trend with essentially the same R2 also exists
between and yield strength o . (If desired, the reader
can obtain this fit from the data in Table 1.)
The trend of decreasing for higher strength steels indicates an increasing sensitivity to mean stress. Also, looking
at Eq. (6a), the behaviour of higher strength steels is increasingly determined by the maximum stress max , and
less by the stress amplitude a , indicating more brittle behaviour. This behaviour may be related to the analogous
trend seen in the crack growth behaviour of a wide range
of materials, as discussed by Ritchie.32
Equation (20) was employed to estimate the value for
the steels of this study from their respective u values, and
the comparisons with test data were repeated, with one
of the life plots being shown in Fig. 16. Because the
Eq. (14) fit would not be available where needs to be
estimated, the arN values for these comparisons were calculated from the Eq. (2) fits to the zero mean stress data,
and only the non-zero mean-stress data points were employed in the summation of Eq. (18) to obtain s z . In other
words, the same procedure was used as for the evaluations
of the non-Walker mean stress methods of Eqs (3)(5).
The bar chart showing all of the resulting stress deviations
s z is given as Fig. 13f, and the values are listed in Table 3.
The result is almost as good as for the individually determined values used with the Eq. (14) fits. Note that the
average s z increases to 0.088 versus 0.073 for the fitted

50CrMo4 Steel
u = 1086 MPa

104
103
Nf , Cycles

2500

105

106

Fig. 16 Equivalent completely reversed stress amplitude versus life


for the Walker method applied to 50CrMo4 steel, where an
estimated value of = 0.665 is employed.

values. This s z = 0.088 value is also somewhat lower


than the 0.11 values for the two versions of the Morrow
method and the SWT method.
Of course, it is true that the just-described comparisons
with data for the estimated values involve a degree of
circular logic, in that the same data were used to develop
the Eq. (20) correlation as to evaluate it. However, the
number of 18 data sets for steel is sufficiently large that
omitting any one data set and revising Eq. (20) will not
significantly affect the comparison for the omitted data
set. This contention was tested by omitting, one at a time,
several of the data points that fell farthest from the fitted
line of Fig. 15, and noting that in each case the fit was
only slightly affected.
Trends in fitted Walker values for aluminium
alloys
Considering aluminium alloys, the fitted values are
plotted versus ultimate tensile strength u in Fig. 17.
The smaller number of data points and their distribution
1.0
, Walker Mean Stress Constant

, Walker Mean Stress Constant

1.0

Aluminum alloys
0.8
0.651
0.6

6061

Al Mg4.5Mn

0.473

0.4

2014
2024

7075

0.2
0.0
300

350
u,

400
450
500
550
Ultimate Tensile Strength, MPa

600

Fig. 17 Trend of fitted values with ultimate tensile strength for


the Walker method applied to aluminium alloys.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

Extension to strain-life curves


Consider the strain-life curve, which is usually represented by the form
ar =

f
E

(2N f )b + f (2N f )c .

(21)

Here, the strain amplitude a has the subscript r added


to indicate that the curve is obtained from completely
reversed (m 0) test data. Because ar /E is the elastic
strain amplitude, with E being the elastic modulus, the
first term above is consistent with Eq. (2). The second
term, with additional fitting constants f and c, derives
from fitting plastic strain amplitudes versus life. Hence,
the two terms sum to the total strain amplitude ar .
As previously discussed in publication,1,31 Eq. (21) can
be generalized to include mean stress effects in a manner
that is mathematically consistent with any desired mean
stress equation, ar = f (a , m ), such as any of Eqs (3)(6).
In particular, it is found that Eq. (21) becomes
a =

f
E

(2N )b + f (2N )c .

(22)

The quantity N is an equivalent life value corresponding to the zero mean stress case, that is, N is simply
the life calculated from a as if the mean stress were
zero. An expression for N can be developed for any
particular ar = f (a , m ).
Because the Walker mean stress relationship has been
shown to provide better correlation with test data than
other common methods, its incorporation into the strainlife curve is of particular interest. In this case, Eq. (6b)
leads to the particular N value for the Walker method,
here denoted N w .


1 R (1 )/b
Nw = N f
(a)
2


1 R (1 )/b
(23)
(b).
N f = Nw
2


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

0.1
a , Strain Amplitude

does not seem to justify fitting an overall trend line.


Also, the trend with yield strength was examined, but the
correlation was poor. The relatively high strength alloys
2014, 2024 and 7075 had values of in the range 0.41 to
0.57, scattering around an average of 0.47. This is close
enough to = 0.50 that the SWT relationship would
appear to suffice for these alloys. The two relatively low
strength alloys, 6061-T6 and Al Mg4.5Mn, had values
near 0.65, so that this value might be used as an estimate
for alloys of similar strength.
As just noted for steels, there appears to be a trend of decreasing for higher strength aluminium alloys, although
the distribution of the present data makes the trend less
clear than for steels.

177

AISI 4340 Steel


u = 1172 MPa

0.01

Mean Stress, MPa


0
207
414
621
-207
Fit

= 0.650
0.001
102

103

104
105
106
N*w , Walker Equivalent Cycles

107

108

Fig. 18 Strain amplitude versus Walker equivalent life Nw for AISI


4340 steel.

Form (23b) is simply (23a) solved for N f . To calculate


the fatigue life, the strain amplitude a is substituted into
Eq. (22), and an iterative solution yields N w . Then the life
N f with the mean stress effect included is calculated from
Eq. (23b). Also, Eq. (22) indicates that a plot of a versus
N w is expected to provide a single curve for all values of
mean stress. This is illustrated in Fig. 18 for one of the
sets of data with strain-controlled tests at short lives.
An additional journal article is in preparation where the
above approach is explored in more detail and is then
applied to the 18 data sets of this study that include straincontrolled tests at short lives.
CONCLUSIONS

Based on the analysis and discussion of this paper, the


following conclusions can be stated regarding the methods that were examined for accounting for mean stress
effects:
1 Where life estimates are desired in fatigue analysis, the
Goodman method employing the ultimate tensile strength
u is highly inaccurate and should not be employed.
2 The Morrow equation with the true fracture strength fB
is reasonably accurate in most cases. However, this method
has the disadvantage that true fracture strengths fB are
often not available and then must be estimated.
3 The Morrow method with the intercept constant f is
also reasonably accurate for steels, but often gives grossly
non-conservative life estimates for aluminium alloys.
4 The SmithWatsonTopper (SWT) method provides
good results in most cases, and for aluminium alloys it
is somewhat more accurate than the Morrow fB equation. The SWT method has the advantage of simplicity
and is a good choice for general use.
5 Where data are available for fitting the adjustable parameter , the Walker method gives superior results. A
straightforward multiple regression procedure allows data
at various mean stresses to be fitted all together.

178

N . E . D O W L I N G et al.

6 For steels, the correlation of with u of Eq. (20) allows to be estimated with sufficient accuracy to still
give excellent results for this class of material. Relatively
high strength aluminium alloys have 0.5, which corresponds with the SWT method, but higher values of
apply for relatively low strength aluminium alloys.
7 For both steels and aluminium alloys, there is a trend
of decreasing with increasing strength, indicating an
increasing sensitivity to mean stress, and also more brittle
behaviour that is increasingly dominated by the maximum
stress max , and less affected by the stress amplitude a .

Acknowledgements
The latter portions of this data analysis study that spanned
several years were supported by the U.S. Naval Air Systems Command, Patuxent River, MD. Technical direction was provided by Nam D. Phan of NAVAIR Structures, with the aid of Trung T. Nguyen. Project administration was handled by Pamela F. Bowen of Integrated
Systems Solutions, Inc., California, MD. Gratitude is expressed to these individuals and organizations for their
able assistance.
Institutional support was provided by the Engineering
Science and Mechanics Department, and by the Materials Science and Engineering Department, Virginia Tech,
Blacksburg, VA. We are especially indebted to staff member Lisa L. Smith of the ESM Department, who was helpful in a variety of ways.
The authors wish to thank several individuals who
aided with useful advice, comments, or data, including:
Jamie T.P. de Castro (Pontifical Catholic University,
Rio de Janeiro, Brazil); Ali Fatemi (The University of
Toledo, Toledo, OH); Nagaraja Iyyer (Technical Data
Analysis, Inc., Falls Church, VA); Chih-Kuang Lin
(National Central University, Chung-Li, Taiwan) and
Surot Thangjitham (Virginia Tech, Blacksburg, VA.).
REFERENCES
1

4
5

Dowling, N. E. (2004) Mean stress effects in stress-life and


strain-life fatigue. In: Proceedings of Fatigue 2004: Second SAE
Brazil International Conference on Fatigue, Sao Paulo, Brazil,
June 2004. Also SAE Paper No. 200401-2227, SAE
International, Warrendale, PA.
Keshavan, S. (1967) Some Studies on the Deformation and
Fracture of Normalized Mild Steel Under Cyclic Conditions. PhD
Thesis, University of Waterloo, Waterloo, Ontario, Canada.
Smith, K. N., Watson, P. and Topper, T. H. (1970) A
stress-strain function for the fatigue of metals. J. Mater.,
ASTM 5, 767778.
Baumel, A. and Seeger, T. (1990) Materials Data for Cyclic
Loading, Supplement 1. Elsevier Science Pubs., Amsterdam.
Grover, H. J., Bishop, S. M. and Jackson, L. R. (1951) Fatigue
strengths of aircraft materials: axial-load fatigue tests on

10

11

12

13

14

15

16
17

18

19
20

21

unnotched sheet specimens of 24S-T3 and 75S-T6 aluminum


alloys and of SAE 4130 steel. NACA TN 2324. National
Advisory Committee for Aeronautics, Washington, DC.
Illg, W. (1956) Fatigue tests on notched and unnotched sheet
specimens of 2024-T3 and 7075-T6 aluminum alloys and of
SAE 4130 steel with special consideration of the life range
from 2 to 10,000 cycles. NACA TN 3866. National Advisory
Committee for Aeronautics, Washington, DC.
Lin, C. K. and Chu, C. C. (2000) Mean stress effects on
low-cycle fatigue for a precipitation-hardening martensitic
stainless steel in different tempers. Fatigue Fract. Eng. Mater.
Struct. 23, 545553.
Dowling, N. E. (1973) Fatigue life and inelastic strain
response under complex histories for an alloy steel. J. Testing
Evaluat. 1, 271287.
MIL-HDBK-5H. (1998) Military Handbook: Metallic Materials
and Elements for Aerospace Vehicle Structures, 2 Vols. U.S. Dept.
of Defense, Wright-Patterson AFB, OH.
Landgraf, R. W. (1966) Effect of Mean Stress on the Fatigue
Behavior of a Hard Steel. Report No. 662. Dept. of Theoretical
and Applied Mechanics, University of Illinois, Urbana, IL.
Wehner, T. and Fatemi, A. (1991) Effects of mean stress on
fatigue behavior of a hardened carbon steel. Int. J. Fatigue 13,
241248.
Boller, C. and Seeger, T. (1987) Materials Data for Cyclic
Loading, Part D: Aluminum and Titanium Alloys. Elsevier
Science Pubs, Amsterdam.
Endo, T. and Morrow, J. (1969) Cyclic stress-strain and
fatigue behavior of representative aircraft metals. J. Mater.
ASTM 4, 159175.
Topper, T. H. and Sandor, B. I. (1970) Effects of mean stress
and prestrain on fatigue damage summation. In: Effects of
Environment and Complex Load History on Fatigue Life, ASTM
STP 462. Am. Soc. for Testing and Materials, Philadelphia,
PA, pp. 93104.
Lazan, B. J. and Blatherwick, A. A. (1952) Fatigue properties
of aluminum alloys at various direct stress ratios: Part 1, rolled
alloys. WADC TR 52307, Part 1. Wright Air Development
Center. Wright-Patterson AFB, OH.
Goodman, J. (1919) Mechanics Applied to Engineering.
Longmans, Green and Co., London, pp. 631636.
Smith, J. O. (1942) The Effect of Range of Stress on the Fatigue
Strength of Metals. Bulletin No. 334. University of Illinois,
Engineering Experiment Station, Urbana, IL, Feb. 1942. See
also Bulletin No. 316, Sept. 1939.
Morrow, J. (1968) Fatigue properties of metals, section 3.2. In:
Fatigue Design Handbook, Pub. No. AE-4. Soc. of Automotive
Engineers, Warrendale, PA. Section 3.2 is a summary of a
paper presented at Division 4 of the SAE Iron and Steel
Technical Committee, Nov. 4, 1964.
Bridgman, P. W. (1944) The stress distribution at the neck of
a tension specimen. Trans. Am. Soc. Metals 32, 553574.
Landgraf, R. W. (1970) The resistance of metals to cyclic
deformation. In: Achievement of High Fatigue Resistance in
Metals and Alloys, ASTM STP 467. Am. Soc. for Testing and
Materials, Philadelphia, PA, pp. 336.
Walker, K. (1970) The effect of stress ratio during crack
propagation and fatigue for 2024-T3 and 7075-T6 aluminum.
In: Effects of Environment and Complex Load History on Fatigue
Life, ASTM STP 462. Am. Soc. for Testing and Materials,
Philadelphia, PA, pp. 114.


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

MEAN STRESS EFFECTS IN STRESS-LIFE FATIGUE AND THE WALKER EQUATION

22

23

24

25

26

27

Mann, J. Y. (1958) The historical development of research on


the fatigue of materials and structures. J. Aust. Inst. Metals. 3,
222241.
Nihei, M., Heuler, P., Boller, C. and Seeger, T. (1986)
Evaluation of mean stress effect on fatigue life by use of
damage parameters. Int. J. Fatigue 8, 119126.
Chu, C. C. (2000) Comparison of mean stress correction
methods for fatigue life prediction. SAE Paper No.
200001-0778. SAE International, SAE 2000 World
Congress, Detroit, MI, March 2000.
Wang, S. J., Dixon, M. W., Huey, C. O. Jr. and Chen, S. C.
(2000) The Clemson limit stress diagram for ductile parts
subjected to positive mean fatigue loading. J. Mech. Design,
Trans. ASME 122, 143146.
Watson, P. and Topper, T. H. (1970) The effects of
overstrains on the fatigue behavior of five steels. 1970 Fall
Meeting of the Metallurgical Society of AIME, Cleveland, OH,
Oct. 1970. See also: Watson, P (1971). The Effect of Mean
Stress and Overstrains on the Fatigue Behavior of Structural
Steels. PhD Dissertation, Department of Civil
Engineering, University of Waterloo, Ontario, Canada.
Fatemi, A. and Stephens, R. I. (1987) Tensile mean stress
effects on uniaxial fatigue behavior of 1045 HR steel. In:
Fatigue 87, Vol. I, Proceedings of the Third International
Conference on Fatigue and Fatigue Thresholds (Edited by R. O.
Ritchie and E. A. Starke Jr.), Engineering Materials Advisory


c 2009 Blackwell Publishing Ltd. Fatigue Fract Engng Mater Struct 32, 163179

28

29

30

31

32

179

Services Ltd. (EMAS), Warley, West Midlands, UK,


pp. 537546.
Gallagher, J. P. et al. (2001) Improved high cycle fatigue
(HCF) life predictions. AFRL-ML-WP-TR-20014159. Final
Report on Contract F3361596-C-5269, Air Force Research
Laboratory, Wright-Patterson AFB, OH.
SAE. (1998) Technical report on fatigue properties: SAE
J1099. In: 1998 SAE Handbook, Vol. 1, Materials, Fuels,
Emissions, and Noise. Soc. of Automotive Engineers,
Warrendale, PA, pp. 3.1523.161. (More recent editions omit
true fracture strength data.) See also: Tucker, L. E., Landgraf,
R. W. and Brose, W. R. (1974). Proposed technical report on
fatigue properties for the SAE handbook. SAE Paper No.
740279. Soc. of Automotive Engineers, Automotive
Engineering Congress, Detroit, MI, Feb. 1974.
Smith, R. W., Hirschberg, M. H. and Manson, S. S. (1963)
Fatigue behavior of materials under strain cycling in low and
intermediate life range. NASA TN D-1574. National
Aeronautics and Space Administration, Washington, DC.
Dowling, N. E. (2007) Mechanical Behavior of Materials:
Engineering Methods for Deformation, Fracture, and Fatigue,
3rd edn. Pearson Prentice Hall, Upper Saddle River,
NJ.
Ritchie, R. O. (1999) Mechanisms of fatigue-crack
propagation in ductile and brittle solids. Int. J. Fracture 100,
5583.

You might also like