Int. J. Pres. Ves.

& Piping 53 (1993) 273-298 ~ • -

Prediction of Fatigue Damage and Fatigue Life under

Random Loading

Wen-Fang Wu & Tsan-Hua Huang

Department of MechanicalEngineering, National Taiwan University,Taipei,

Taiwan 10764
(Received 26 January 1992; accepted 7 February 1992)

ABSTRACT

The objective of the present paper is to find the fatigue damage and
fatigue life of a structural component when subjected to variable-
amplitude stress loading modeled as a wide-band or narrow-band
Gaussian random process. In addition to the well-known Palmgren-
Miner's linear damage rule, a nonlinear plastic work interaction damage
rule is used as a basis for the accumulation of fatigue damage which
gives us a more conservative fatigue life estimation as compared to the
Palmgren-Miner rule. Numerical examples are performed and com-
parison is made for the fatigue damage and fatigue life evaluated under
various random loading conditions and based on different damage
rules. Several conclusions are drawn at the end of the paper which give
us simpler rules to be used in the analysis and design of fatigue
structures when subjected to random loading conditions.

INTRODUCTION

Machinery and equipment are frequently subjected to random loading.

The evaluation of fatigue life under such nonconstant-amplitude
random loading has remained a problem for many years. A common
approach to this problem has been to conduct a cumulative damage
analysis based on results of fatigue life under constant-amplitude
loading. The first cumulative damage theory was proposed by Palmgren
in 1924 and later by Miner in 1945.1 This linear theory, which is still
widely used, is referred to as the Palmgren-Miner rule or the linear
273
Int. J. Pres. Ves. & Piping 0308-0161/92/$05-00 (~) 1992 Elsevier SciencePublishers Ltd,
England. Printed in Northern Ireland
274 Wen-Fang Wu, Tsan-Hua Huang

damage rule. The Palmgren-Miner linear damage rule is very easy to

apply for the prediction of fatigue life under variable-amplitude
loading. However, based on a survey of experimental results, the
fatigue damage is often found to accumulate nonlinearly instead.
Experimental evidence also indicates that damage indexes are not only
functions of stress cycles but also functions of stress amplitude levels. 2
Therefore, several nonlinear damage rules have been proposed in the
last three decades which include the Marco-Starkey damage rule, the
Henry damage rule, the Gatts cumulative damage rule, the C o r t e n -
Dolan cumulative damage rule, the Marin cumulative damage rule, the
Mason double linear damage rule and many others. Brief descriptions
of these nonlinear damage rules can be found in Ref. 2.
Among the numerous nonlinear damage rules, Morrow has proposed
a plastic work interaction nonlinear damage rule. 3 While applying this
nonlinear damage rule, he concluded, from comparison with ex-
perimental result, that for variable-amplitude loading, the nonconser-
vative fatigue life prediction based on Miner's rule can be greatly
improved. Morrow's nonlinear damage rule modified Miner's linear
cycle ratio cumulative damage rule by multiplying a given stress level's
cycle ratio damage term by a nonlinear factor. The modifying factor is
the ratio of the given stress to the maximum overload stress raised to a
power, d, to incorporate an interactive effect. The plastic work
exponent, d, can be interpreted as the material's sensitivity to the
variable-amplitude stress history, and it was pointed out that typical
values for d are between -0-25 and -0.20. According to Morrow, this
nonlinear damage rule along with overload parameter values would
produce accurate fatigue life prediction in the long life region and
slightly conservative life prediction in the short life region from
experiments performed with several selected variable-amplitude applied
stress histories on smooth cylindrical laboratory specimens of 1020
steel. These results showed that the large stress cycles nucleate multiple
cracks that are subsequently propagated by the smaller stress cycles.
Thus, the smaller cycles could create significant damage sooner than
would be anticipated in a constant-amplitude situation.
Lambert has extended Morrow's plastic work interaction nonlinear
damage rule to the case that the stress is a narrow-band Gaussian
random process. 4 Using an assumed maximum peak which the random
stress process could have produced, he noted that the predicted fatigue
life was reduced by 10-15% from that evaluated based on Miner's
linear damage rule. He also concluded that most of the fatigue damage
was caused by those stress peaks at values between two and four times
the root-mean-square stress level. These conclusions agree reasonably
well with experimental results as demonstrated in Ref. 4.
 Prediction of fatigue under random loading 275

While the narrow-band process is a good assumption for many

random stresses, a wide-band random stress history may also occur in
some engineering fields such as the design of ocean structures subjected
to random w a v e s : Under such circumstance, Lambert's analysis cannot
be used. A probabilistic fatigue analysis based on Morrow's plastic
work interaction damage rule when a structural component is subjected
to a wide-band random stress history is therefore needed.
The application of a probabilistic method in fatigue analysis and
design has been extensively studied by Wirsching and his associates. 5-9
Their serial analyses started with the situation that the fatigue loading is
a narrow-band random stress history. Under this circumstance, the
expected fatigue damage, fatigue life and fatigue reliability can all be
found analytically without any difficulty. For other situations such that
the fatigue loading modeled as wide-band processes some modification
based on the result from narrow-band random fatigue analysis should
be made. In Ref. 6, a rain-flow cycle counting method was used by
Wirsching and Shehata to obtain from the randomly generated stress
history, statistics related to the applied wide-band random loading. A
Weibull probability density function, together with a correction factor
as necessary adjustment, were proposed to fit these stress statistics.
Fatigue design is then performed based on the corrected stress statistics
and a modified Palmgren-Miner rule. 6 In Refs 7 and 8, Wirsching and
his associates proposed other models to predict high cycle fatigue under
wide-band random stresses. These models were based on applying a
correction factor to a narrow-band stress process which was equivalent
to the applied wide-band process. Statistical variability in material
strength as well as mean stress effect were also included in those
analyses. The overall concept of the probabilistic fatigue design under
random loading can be found in Refs 5 and 9.
The objective of the present paper is to find the fatigue damage and
fatigue life of a structural component when subjected to wide-band
Gaussian random loading. In order to predict the fatigue damage and
fatigue life on the conservative side, Morrow's plastic work interaction
damage rule is applied. Due to the irregular behaviour of the
wide-band random stresses, some analytical procedures employed by
Wirsching are used in the present analysis.

MORROW'S NONLINEAR DAMAGE RULE

In a variable-amplitude stress condition, according to Morrow's plastic

work interaction damage rule, 3 the fatigue damage caused by the stress
276 Wen-FangWu, Tsan-HuaHuang
of amplitude s,. can be written as
=ni(s,] d
Oi -~i \~m / (1)

in which Sm is the maximum stress amplitude in the stress history, ni is

the number of stress peak at level si, Ni is the n u m b e r of stress peak to
cause failure if constant-amplitrde s, is considered, and d is Morrow's
plastic work interaction exponent which can be considered as the
sequence effect of the stress history on the fatigue damage. From eqn
(1), if all different cyclic stresses are considered, the total fatigue
damage is then
ni(si] d
i

and the fatigue failure would occur when D reaches one. It is

interesting to note that Morrow's plastic work interaction fatigue
damage rule has exactly the same form as Marin's cumulative damage
theory 1° although they were derived based on different arguments and
different assumptions. A very similar damage rule has also been derived
by Corten and Dolan using a third different approach. 1~ These
nonlinear damage rules all reduce to the famous P a l m g r e n - M i n e r rule
if the coefficient d is chosen to be zero as can be seen from eqns (1) and
(2).
For a variable-amplitude stress history, if only a few different stress
amplitudes occur such as those in a block-type test loading, eqns (1)
and (2) can be applied directly without any difficulty. The true
variable-amplitude stress history, however, may be a continuous
random process without a finite stress range. In such a case, eqn (1)
may be interpreted to be

D~=~ (3)

where s is the stress amplitude which is now continuously distributed,

s,n is the maximum stress peak which may occur during the applied
stress history, ns is the distribution of the n u m b e r of stress amplitudes
with respect to stress level s, and N~ is the number of stress amplitudes
at level s to cause failure. Fatigue damage is accumulated continuously
by mathematically integrating eqn (3) over all stress levels, and failure
occurs when the integrated Ds reaches one, i.e.

f D~ds = 1 (4)
 Prediction of fatigue under random loading 277

Equations (3) and (4) will be used in the following sections for the
prediction of fatigue life of a metallic component subjected to random
loading.

S I M U L A T I O N A N D STATISTICS OF R A N D O M L O A D I N G

In order to use eqns (3) and (4) to predict the fatigue damage and
fatigue life under random loading, the statistics of the random stress
history should be obtained in advance. For a narrow-band random
loading, the method used by Lambert in Ref. 4 can be adopted. For a
wide-band random loading, due to the complexity of the stress history
and the difficulty in counting the stress cycles, a different approach is
required to obtain statistical quantities needed in applying eqns (3) and
(4). In the present section, a procedure used by Wirsching and Shehata 6
will be adopted in order to obtain these quantities.
According to Ref. 6, a stationary Gaussian random stress process
X(t) can be completely characterized by its spectral density function
G(to). For a given single-side spectral density function G(to), a sample
function of X(t) can be simulated using the following formula
J

x(t) = ~] [2G(o~k)Aogk] '/2 cos (oJkt + q~,) (5)

k=l

In the above digital simulation of x(t) from G(w), the frequency is

defined over the interval (0, w.) with partitions of length Awk such that
w, = Z~,=, Aog, and q~, is a random phase angle uniformly distributed in
the interval (0, 2~r). The number of harmonic components J is selected
to be 50 in the present study. In order to avoid repetitive pattern
appearing in x(t), a small random perturbation for each w, is used
according to Ref. 12.
After a long enough random stress history x(t) is simulated, its peaks
and troughs are recorded and a sample of stress amplitudes S and cycle
n can be obtained using the rain-flow cycle counting method. ~ It should
be mentioned that a complete rain-flow cycle counting method can
record not only stress amplitudes but also the mean values of these
amplitudes. '3 In the analysis of the present section, we will follow the
method of Wirsching and Shehata and ignore the effect of these mean
stresses. The mean stress effect will, however, be considered in a later
section based on the Goodman criterion.
If the mean stress effect is omitted, the rain-flow cycle counting
method will give us a sequence of stress amplitudes which we shall call
the equivalent (discrete) stress history si (j = 1, 2 , . . . , n) out of the
278 Wen-Fang Wu, Tsan-Hua Huang
simulated random stress x(t). The mean value and standard deviation
of this equivalent stress history can be obtained from the following
formulas

S=ILsj
nj=l
(6)

and

SS ~ ~
1=1

If we assume the amplitudes of the equivalent stress history are

distributed following a continuous Weibull probability density function
then

fs(S)= 6~ exp - (8)

in which ~ and 6 are Weibull parameters. The kth statistical moment of

the Weibull density function can be obtained directly from eqn (8) as

,k = e [ S k l w = 6 w 1 +-~ (9)

where F(.) is the Gamma function and a subscript W is affixed to E[S k]

to indicate that the moments of the equivalent stress amplitudes are
calculated based on the Weibull probability density function assump-
tion. The mean value and standard deviation of the Weibull density
function are derived by plugging k = 1 and k = 2 into eqn (9), and after
some simple operation one obtains

#s=6F 1+ (10)
and
°s=~f62[F(l+~)-F2(l+~)] (11)

The Weibull parameters ~ and 6 of the equivalent random stress

history can be obtained by replacing the left-hand side of eqn (10) with
the statistic S of eqn (6), and the left-hand side of eqn (11) with the
statistic Ss of eqn (7), respectively. In the present study, different power
spectral density functions having different irregular factors are selected
as numerical examples and will follow in a later section. It is found that
= 2 and 6 =V2ax for a narrow-band random process having a
standard deviation ox and an irregular factor c~= 1. Under this
condition, the Weibull probability density function reduces to the
 Prediction of fatigue under random loading 279

Rayleigh probability density which is exactly the theoretical peak

density function of a narrow-band random stress history. TM

P R E D I C T I O N OF F A T I G U E D A M A G E A N D F A T I G U E LIFE

Assuming the constant-amplitude fatigue test data can be applied

directly to the variable-amplitude loading (including random loading),
the fatigue life will then follow the well-known S - N curve relationship,
Ns Sm = C (12)
in which Ns is the number of cycles to cause failures under constant
stress amplitude s, m is the material's fatigue curve slope in log-log
plot, and C is the material's cyclic fatigue strength coefficient. For the
evaluation of fatigue damage caused by n random stress cycles, one
may consider that these n cycles are distributed with respect to their
stress amplitudes as
ns = nf~(s) (13)
where fs(S) is the probability density function of the amplitudes.
Substituting eqns (12) and (13) into eqn (3) and then integrating with
respect to s, one obtains the expected amount of fatigue damage (or
fatigue damage index) caused by n random stress cycles as
nE[S ~+a]
E[O(n)] Cs~ (14)

It should be mentioned that in the derivation of eqn (14) we have

assumed the independence between two variables S and n.
Theoretically, n is a random variable but can usually be replaced by the
expected number of stress cycles which can be calculated analytically
from the given power spectral density function. 6
If the probability of the random amplitudes is assumed to follow a
Weibull density function, i.e. Eqn (8), then the expected fatigue
damage caused by n stress cycles becomes

nO,.+aF/1 +-mg+- d\
)
E[o(n)lw = (15)

where, as before, a subscript W is affixed to E[D(n)] to indicate that

damage is calculated based on the Weibull probability density function
assumption for the equivalent stress amplitudes. For a narrow-band
280 Wen-FangWu, Tsan-HuaHuang
process, a~ = 1, ~ = 2 and 6 = V~o~, and eqn (15) reduces to

2('+~'%o~+"F(1+ 2 + ~)
E[D(n)lw = (16)
CSm
which is very close to eqn (13) in Ref. 4 where an upper limit was set
during the integration with respect to s.
Another special case is to let d = 0 in eqn (3), which means that
Palmgren-Miner rule is used instead of Morrow's. Under this assump-
tion, eqn (15) reduces to

E[O(n)lw = C (17)

As expected, eqn (17) is exactly the same as eqns (16) and (18) in Ref.
6 where the P a l m g r e n - M i n e r rule was considered.
In using eqn (14) to evaluate the fatigue damage caused by n stress
cycles, instead of using the assumed Weibull probability density
function to obtain E[S "+~] in eqn (14), we can also estimate E[S m+"]
from the simulated equivalent stress history based on the following
statistic

(18)
nj_l
where k = m + d if the plastic work interaction damage rule is used and
k = m if the P a l m g r e n - M i n e r rule is adopted. According to Ref. 6, the
Weibull density may be a good approximation for the probability
density function of the amplitudes in an equivalent stress history of a
random loading. However, the expected value E[S m÷a]calculated from
Weibull density may differ from that calculated from eqn (18) which is
considered to be the exact expected m o m e n t of the equivalent stress
history as long as the simulation process is carried on long enough.
Therefore, if we use the Weibull probability density function for the
equivalent stress history to predict the fatigue damage, a correction
factor Q should be used to correct the results calculated from eqns
(15)-(17). This correction factor is best if it is a function of irregular
factor ce and spectral m o m e n t k and can be obtained from

-E
E[S "~] n j=,
Q(o~,k)= - -
E[Sk]w -
akr(1+ ) k (19)
 Predictionof fatigue underrandom loading 281

Based on the present study (as will follow in a later section) as well as
that in Ref. 6, this correction factor becomes one for the case of a
narrow-band process having the irregular factor a~ = 1.
The purpose of the above analysis is to find through numerical
simulation and the rain-flow cycle counting m e t h o d the Weibull
parameters ~ and 6 and a correction factor Q for the equivalent stress
history which is considered to be the exact stress cycles of a random
loading. Once these parameters are found, we can predict directly the
fatigue damage caused by a given power spectral density function
corresponding to a certain pattern of random loading. No more tedious
numerical simulation and rain-flow cycle counting m e t h o d are needed.
In using eqns (15), (16) and (19) to predict the fatigue damage, the
maximum stress peak s,, should be known in advance. In Ref. 4
arbitrarily assumed 3ox, 4ox and 5OxS are used to represent the
maximum stress peaks. In the present study the following expected
maximum stress amplitude ~4 is used instead,

Sm=Ox( 2V~--~nq-~nn) (20)

in which 7 = 0.577 216 is the Euler constant. Equation (20) is the true
expected maximum stress amplitude for a narrow-band stress with
a~ = 1. For a wide-band stress having c~ = 0, the true expected maximum
stress amplitude should b e 14
y In I n n + I n 4~'~
s,, ='ox ~ + V~ In n 2V~Jnn / (21)
For other random loading having irregular factors between 0 and 1, the
expected maximum peaks should possess values between those given by
eqns (20) and (21). However, no closed form expression as a function
of n can be found. Because the selection of Sm is not very sensitive to
the numerical result, and also because that use of eqn (20) is on the
conservative side, we will use eqn (20) in the present analysis.
Combining all the expressions listed above, the expected fatigue
damage after n random stress cycles can be evaluated from the
following formula
m+d\
n(Sm+aF(1 + ---~)Q(o4 m + d)
E[O(n)] = (22)
+ r
%

Equation (22) can be reduced to other simpler forms under different

assumptions. For example, for a narrow-band stress with a~ = 1, we
282 Wen-Fang Wu, Tsan-Hua Huang
have Q = 1, ~ = 2 and 6 = V~ ox. Substituting these values into eqn
(22) we have

2(m+a)/2noT+aF 1 + -~ +
E[O(n) l = ,~ (23)

If we further assume that d = 0 which means that P a l m g r e n - M i n e r rule

is used instead of Morrow's rule, then eqn (23) becomes

E[O(n)] - C (24)

which is exactly the same as eqn (9-150a) in Ref. 15. Finally, if both
Palmgren-Miner rule and wide-band random stresses are considered as
used in Ref. 6, the damage is then evaluated from eqn (22) by setting
d=0.
Besides the fatigue damage, we are also interested in predicting the
fatigue life (in expected n u m b e r of stress cycles n) of a structural
component when subject to a r a n d o m stress history. Depending on the
damage rules and the spectral density functions of the random loading,
the fatigue life can be predicted by letting the expected fatigue damage
(index) E[D(n)] in eqns (22)-(24) be one and then solving these
equations.
In the above analysis, the mean stress effect of the equivalent stress
history is not considered. For a narrow-band r a n d o m stress history,
since the mean value of each stress amplitude in the equivalent stress
history is about zero, the mean stress effect is not important. For a
wide-band random stress history, we may obtain quite a few stress
cycles with small amplitudes after the rain-flow cycle counting. These
small amplitudes can have large mean values which are critical in
causing the fatigue damage. In general, the positive mean values reduce
the fatigue life while the negative mean values increase the fatigue life.
The mean stress effect can be taken into account in the fatigue analysis
using several criteria such as the G o o d m a n criterion, Soderberg
criterion, Gerber criterion and Morrow criterion. In the present study,
a modified G o o d m a n criterion is used by correcting each stress
amplitude sj in the equivalent stress history by a factor 1 - s'p/Su where
s7' represents the mean value of the stress amplitude sj and S, represents
the ultimate stress of the material. U n d e r this modified G o o d m a n
 Prediction of fatigue under random loading 283

criterion, eqn (18) will be modified to

etsq =-1~ ( ~' ~k

n,=, 1-~u)
(25)
Other analyses will then follow the same procedure derived from eqns
(19)-(24).

NUMERICAL EXAMPLES AND DISCUSSION

Based on the analysis presented in the previous section, several

numerical examples are performed. The analysis begins with the

0(= 0.2 a = 0.3 a = 0.4

1 ,..= ], ~ I I--'-I~
~=0.5 a=0.6 a=0.7

~] rn~ N rl_ r~
¢~= 0.8 a = 09 at= 0.99
Fig. la. Bi-modal spectral density functions.
284 Wen-Fang Wu, Tsan-Hua Huang

simulation of random stress histories from different given power

spectral density functions. In order to cover as many different values of
the irregular factor as possible, the studied spectral density functions
together with their irregular factors are presented in Fig. 1 which covers
(a) the bi-modal spectral density functions and (b) the uni-modal
spectral density functions. A typical simulated narrow-band random
stress history is shown in Fig. 2a while a typical wide-band random
stress history is shown in Fig. 2b. For a section of wide-band stress
history with tr = 0.2 such as that shown in Fig. 3a, the rain-flow cycle
counting method gives us a result shown in Fig. 3b which is also called
an equivalent random stress history.
According to the analysis in the previous section, so long as the
irregular factor is the same, the probability density function of the
equivalent stress amplitudes should be the same regardless of the shape

~=0.2 ~=0.3 o(=0.4

]
F,_a=0.5 ~=0.6 o(= 0.7

L r
FI_
a --. 0.8 at= 0.9 0(=099
Fig. l b . Uni-modal spectral density functions.
 Prediction of fatigue under random loading 285

30O:

2.oG

uJ
Q 1.00
:D
~--- O.OO
,.,,,,.J
0..
~E -1.oo
<

-2.00

-3.0(

TIME
Fig. 2a. A sample of narrow-band random stress history.

of the power spectral density functions. In the present study this is

found to be not always true. Figure 4a shows that the Rayleigh
probability density function (which is a special Weibull with ~ = 2 and
6 = V~ Ox) is indeed a very good fit for the equivalent r a n d o m stress
histories of two different narrow-band power spectral density functions
having the same te = 0.99. However, in Fig. 4b, the equivalent stress
amplitudes are distributed differently for two power spectral density
functions having the same irregular factor tr = 0-7. As expected, the
difference becomes more obvious as the irregular factor becomes

3.o
W 1,0

°
~'-
_J
0.0 ' "
A
< -1,o

-2.00

- 3.0C

TIME
Fig. 2b. A sample of wide-band random stress history.
286 Wen-Fang Wu, Tsan-Hua Huang

2.501
2.0C
1.5(3
1.00
LU
CI
::D 0.50
I--
0.00
_.1
O- -0.50
-tOO
-1.50
-2.0C
-2.5C

TIME

Fig. 3a. A section of stress history before rain-flow cycle counting.

smaller. Fortunately, an optimal Weibull probability density fit to both

data sets as that shown in Fig. 4 by the solid curves together with the
correction factor Q(o~, k) diminish the effect of this difference as will be
seen later.
For each given o~, the Weibull parameters ~ and ~ can be determined
either from the best fit of both data sets as just explained or from eqns
(6), (7), (10) and (11) after giving both data sets equal weights. The
values of these Weibull parameters for different irregular factors are
shown in Fig. 5 by those data points. The solid curves are the best

2.50
2.00
vA
1.50 m

J,l
0

l--
1.00
050
0.00 v
A
V ^
A v
v
V,! vAAAA^A^^
^ V
I
A
AV
,,AV
V,
V
a_
-050 - V

-1.00 VV A
v
-1.501 ^ V

- 2 .00.
A
-2501

CYCLES
Fig. 3b. A section of stress history after rain-flow cycle counting.
 Prediction of fatigue under random loading 287

o.71 | I I I I I I I I

0.6
d
II
"6 0.5 B i m o d a l PSDF
z ~, ~ * ~ * B a n d p a s s PSDF
o
0.4
Z

~_ 0.3

z
uJ
c~ 0.3
<

0.0 1.0 2.0 3.0 4.0 5.0

NORMALIZED STRESS AMPLITUDE
Fig. 4a. Probability density of the equivalent stress history for the case of o~= 0-99.

spline fits of the data points and the dashed curves are those presented
in Ref. 6 where stress range was used instead of stress amplitude to
make (~ twice as large as that presented herein. In Fig. 5, ~ is found to
be 2 and ~ = V~ ax for the case a~ = 1 which corresponds to a Rayleigh
density function as expected.
After the Weibull curve fitting is completed, the expected fatigue
damage can be calculated from eqns (15), (16) or (17). This damage,
however, is considered to be approximate to the real damage caused by
the equivalent stress history. The real damage is evaluated from eqns
(14) and (18). The correction factor Q to convert the approximate
damage to the real damage is found from eqn (19) and is a function of
(x and k. The result is shown in Fig. 6 which, again, is very similar to
the result presented in Ref. 6. In the calculation of eqn (18) it is found
that for two different shapes of spectral density function, so long as
their irregular factors are the same, the results of E[S k] differ only a
little. This is because the major difference between two equivalent
stress histories comes from those small amplitudes which usually
288 Wen-Fang Wu, Tsan-Hua Huang

1.0 'i! i i i i i i i i i

0.8 j EX]EX3D Bimodal PSDF

II
~ * *--**--_~ Triangular PSDF

Z
0
I-
IJ
0.6
z
:C)
It

I--

i./'l
0.4

02[
Z
Ill
O
,.¢.

Li.I
ll

0.0
0.0 1.0 2.0 3.0 4.0 5.0
NORMALIZED STRESS AMPLITUDE
Fig. 4b. Probability density of the equivalent stress history for the case of oc= 0-70.

contribute little to eqn (18) as well as the fatigue damage as can be seen
from Fig. 4(b).
One of the major purposes of the present paper is to compare the
fatigue damage and fatigue life caused by the same random loading but
evaluated based on different damage rules. After accomplishing the
rain-flow counting, the Weibull density fit, and introducing the correc-
tion factor and the maximum stress peak, the fatigue damage caused by
n random stress cycles can be calculated from eqns (22), (23), (24) or
others depending on which damage rule is adopted. Since the fatigue
damage under a narrow-band random stress and evaluated based on the
Palmgren-Miner rule has an analytical solution, i.e. eqn (24), and is
well-recognized and used by many researchers, most of the comparisons
are made based on this result and presented in a form of nondimen-
sional damage ratio. Figure 7 shows the damage ratios caused by
different values of ce representing different wide-band random stress
processes. Two curves are presented in this figure of which the upper
one is calculated based on Morrow's rule while the lower curve is based
 Prediction of fatigue under random loading 289

2.5

2.0

1.5
E
/
13.
f
-5 1.0 f f
f
f
I

0.5

O.Ot-
0.0 0.2 O.Z, 0.6 0.8 1.0
Irregular Factor,~
Fig. 5. Weibull parameters as functions of irregular factor.

on the P a l m g r e n - M i n e r rule. Two conclusions can be drawn from this

figure and other similar figures which are not presented here due to
limited space. First, Morrow's plastic work interaction damage rule
predicts fatigue damage with a 10-20% higher result than that based on
the P a l m g r e n - M i n e r rule for a narrow-band process. For a wide-band
process, the difference is even more pronounced. Secondly, under the
same standard deviation trx of the original random stress and same
number of cycles, the fatigue damage is lower for wide-band r a n d o m
stresses than that for narrow-band random stresses, therefore, the
wider the bandwidth (i.e. the lower irregular factor), the lower the
fatigue damage.
Figure 8 is constructed for the damage ratios between those
calculated for wide-band stresses and based on Morrow's rule and those
from narrow-band stress and based on the P a l m g r e n - M i n e r rule. The
effect of the material constant m can be seen from this figure. Figures 7
and 8 are calculated under the assumption that n = 108 cycles. Although
the number of cycles affects the damage ratios presented in the above
290 Wen-Fang Wu, Tsan-Hua Huang

"6
0
k.
o
U

t-
O
t9

k.
o
0.1

0.2 0.4 0.6 0.8 1.0

Irregular Factor,oL
Fig. 6. Correction factor Q(ol, k) as a function of irregular factor.

two figures, it is found that the effect is not very obvious. For example,
if n = 1-5 × 104 is considered, the results in Figs 7 and 8 are found to be
at most 6% lower than the original results.
The expected fatigue life under r a n d o m loading can be predicted by
letting the left-hand side of eqns (22)-(24) be zero and then solving the
equations. As before, we are interested in finding ratios of the life
predicted based on different rules and u n d e r different wide-band
stresses over that based on the P a l m g r e n - M i n e r rule and u n d e r a
narrow-band stress process. For example, if Morrow's plastic work
interaction damage rule is considered, the fatigue lives under various
wide-band stresses having different irregular factors can be calculated
from eqn (22). These lives are then divided by the life calculated from
eqn (24) using the same parametric values. The result is shown in Fig. 9
where two sets of curves are presented. The solid curves correspond to
the case that n = 105 calculated from eqn (24) and the dashed curves
correspond to the other case that n = 108. The difference between the
 Prediction of fatigue under random loading 291

1.2 i I i i I I

(3~ Morrow's Rule

Ear~nnr~ Polmgren-Miner
1.0

0.8
o

o
(2:

0.6
o
E
o
o
0.4

0.2

0.0 I I J I I I I I J I
0.0 0.2 0.4 0.6 0.8 1.0
Irregular Factor,~,
Fig. 7. Comparison of damage ratios under different damage rules.

two curve sets is not obvious as c o m p a r e d to the influences of o t h e r

parameters such as irregular factor a~ and material constant m. The life
ratio can be as large as five if the irregular factor of a wide-band stress
process is very small which is because small stress amplitudes do not
cause serious fatigue d a m a g e but are still c o u n t e d as stress cycles in the
rain-flow cycle counting as explained before.
Finally, if the m e a n stress effect is considered and the modified
G o o d m a n criterion is e m p l o y e d , the fatigue life and fatigue d a m a g e are
then calculated from eqns (19)-(24) and eqn (25). Figure 10 shows one
of the results thus calculated w h e r e Su in eqn (25) is assumed to be
12ax. In this figure, the upper curve represents the d a m a g e ratio
obtained after considering the m e a n stress effect and the lower curve
indicates the result without considering the m e a n stress effect. Both
curves are calculated based on Morrow's rule and u n d e r different
irregular factors. The influence of m e a n stress effect on the fatigue
damage can easily be seen from this figure. It is observed that the
fatigue damage increases 10-40% if m e a n stress effect is considered.
292 Wen-Fang Wu, Tsan-Hua Huang

1.0 I t i I

m=3.2
IZ]IIKI~ m=6.2
m=9.2
0.8

.9 0.6
O
cc

c2~
0
E 0.4
0

0.2

0.0 i I ~ I a I I I i ,I
0.0 0.2 O.Z, 0.6 0.8 1.0
]rregulor Fator,~,
Fig. 8. Comparison of damage ratios among different material constants m.

The increase is m o r e obvious for wide-band processes with smaller

irregular factors. It can be attributed to the fact that there are m o r e and
more small stress amplitudes with large m e a n values as the irregular
factor becomes smaller. Although these small amplitudes do not cause
serious fatigue damage based on either the P a l m g r e n - M i n e r rule or
Morrow's rule, their m e a n values do have some effect on the fatigue
behaviour according to G o o d m a n ' s criterion.

CONCLUSIONS

The following conclusions are drawn from the present study:

1. A n equivalent stress history can be obtained by applying the
rain-flow cycle counting m e t h o d to a r a n d o m stress history
simulated from a given power spectral density function. It is
found that the amplitude distribution of the equivalent stress
history depends on the shape as well as the irregular factor of the
 Prediction of fatigue under random loading 293
6.0 | J I l I I

OOCX30 m=3.2
13r~13nr~ m = 6.2
5.0 AZ~AZ~ m = 9.2

o 4.0
o

Q;

._1
3.0

o
IL
2.0

1.0

0.0 I I J . 1 i I J
0.2 0.4 0.6 0.8 r.o
Irregular F a c t o r ot
Fig. 9. Comparison of fatigue life ratios among different material constants m.

random stress history. In general, for power spectral density

functions having the same spectral area (i.e. same Orx), if (a) the
irregular factor is smaller, (b) the spectral shape is higher for a
uni-modal spectrum, or if (c) the high frequency content is more
to the right for a bi-modal spectrum, the small stress amplitudes
appear more frequently.
. Because of the appearance of small amplitudes mentioned above,
the Weibull probability density fit for the equivalent stress
amplitudes may not always be perfect, even for two different
spectral shapes having the same irregular factor. Fortunately,
these small amplitudes do not cause serious fatigue damage
according to the Palmgren-Miner rule or Morrow's rule, and it is
found that the fatigue damages caused by two different spectral
density functions are almost the same as long as the two spectral
density functions have the same irregular factor.
. The fatigue damage and fatigue life caused by random loading
can be predicted either based on the Palmgren-Miner rule or
294 Wen-Fang Wu, Tsan-Hua Huang
1.0

AZ~AAA With rneon stress correction

OOOOO Without meon stress correction

0.8

o 0.6
0
t'~

o'I
o
E 0.4
t~
O

0.2

0.0~-
0.0 0.2 0.4 0.6 0.8 1.0
Ir regular Foctor .~
Fig. 10. The effect of mean stress on the fatigue damage.

based on Morrow's plastic work interaction d a m a g e rule. For

narrow-band r a n d o m loading, the damage evaluated based on
Morrow's rule is about 10-20% higher than that based on the
P a l m g r e n - M i n e r rule. For wide-band r a n d o m loading, the
difference can be as much as 40% depending on values of the
irregular factor of the r a n d o m loading.
. The fatigue d a m a g e depends primarily on the statistical m o m e n t
of the equivalent stress amplitudes E[S ~] no m a t t e r w h e t h e r
Morrow's plastic work interaction d a m a g e rule or the P a l m g r e n -
Miner rule is used. If the value of E[S k] is calculated from the
fitted Weibull probability density function, then a correction
factor Q should be used to convert it to the real value evaluated
directly from the statistics of the equivalent stress history.
. The m e a n stress effect and G o o d m a n criterion can be included in
the present analysis. It is found that fatigue d a m a g e increases
10-40% if the m e a n stress effect is considered.
 Prediction of fatigue under random loading 295

According to conclusion 4, the value of E[S k] plays an important role

in the prediction of fatigue damage and fatigue life under random
loading. It is interesting to note that in Ref. 9 Wirsching and Haugen
have used the theoretical peak density function of a continuous random
process ~5 to evaluate the fatigue damage and fatigue life. They assumed
that the negative stress peaks do not cause any fatigue damage and
obtained the statistical moments of the stress peaks by the following
integration,

E[S k] = fo~ skfo(s, ol) ds (26)

where fp(s, or) is the theoretical peak density function which is also a
function of the irregular factor of the continuous random process. It is
found that when k = 0, the result of the above integration is approxi-
mately equal to (1 + tr)/2. Furthermore, when k = 1 and k = 2, the
results of the integration are approximately equal to those obtained
from the statistics of the equivalent stress history. For higher values of
k the approximation becomes worse but a certain ratio exists for each
given irregular factor as will be explained later.
The theoretical peak density function for a narrow-band random
process is a Rayleigh function defined in the domain (0, ~). According
to Ref. 9, the fatigue damage ratio between wide-band random loading
and narrow-band random loading is determined from

ff s%(s, oc)ds
I(o~, k) = (27)

Equation (27) can be calculated either analytically or numerically.

Some results for several parametric values are shown in Fig. 11.
According to Fig. 11, I(a,, k) is approximately equal to a~ and is
influenced little by the parameter k. The same conclusion has also been
drawn in Ref. 9 by Wirsching and Haugen.
In fact, Linsley and Hillberry ~6 have also used the above theoretical
peak density function to evaluate the fatigue life under random loading.
However, in performing the integration of eqn (26), they selected the
lower integration limit to be - ~ instead of zero. For the case where
a~ = 0.3, the result calculated is 10% higher than that calculated using
eqn (26). Linsley and Hillberry concluded in their paper that the fatigue
life is the same so long as two peak density functions of the random
stresses have the same irregular factor. This conclusion has also been
296 Wen-Fang Wu, Tsan-Hua Huang
1.2 . . . . . . . .

1.0

0.8
E
£
0.6

0.4
3.2

0.2 A I [ I I I I I t
0.2 0.4 0.6 0.8 1.0 1.2
Irregular Factor,c~
Fig. l l . l(ol, m) as a function of irregular factor.

proved experimentally in their paper. In the present paper as well as

the aforementioned Ref. 9, this same conclusion is also observed.
However, if we consider the fatigue damage and fatigue life caused by
each individual random stress history we can find that the damage ratio
given by eqn (22) over eqn (24) is in fact different from that given by
eqn (27) as proposed in Ref. 9 or its equivalence proposed in Ref. 16.
This can be seen by comparing Fig. 7 with Fig. 11. It is found that
although the two figures look very similar and have the same trend, the
damage ratios are different. For each irregular factor c~, if we divide the
ratio read from Fig. 7 by the ratio from Fig. 11, a new diagram can be
constructed as shown in Fig. 12. The reason for the construction of Fig.
12 is to provide us with an easy way to estimate fatigue damage and
fatigue life caused by random loading and based on the analysis
presented in this paper, but without going through tedious simulation,
rain-flow counting, Weibull curve fitting and other correction factors.
For example, the damage caused by a certain random loading can be
estimated by the integration of eqn (26) and then multiplying the result
by the factor shown in Fig. 12.
 Prediction of fatigue under random loading 297

1.2 I I I' I I I

1.0
t~
,,o
t-
O

0
t.) 0.8

0.6 t I I ,1 r 1 l I

0.0 0.2 0.4 0.6 0.8 .0

I r r e g u l o r Foctor, 0(
Fig. 12. Correction factor to eqn (27) for the evaluation of fatigue damage.

ACKNOWLEDGEMENT

The research is supported by the National Science Council of the

Republic of China under Grant NSC79-0401-E002-19. The writers are
grateful for this support.

REFERENCES

1. Miner, M. A., Cumulative damage in fatigue. ASME Journal of Applied

Mechanics, 12 (1945) 159-64.
2. Collins, J. A., Failure of Materials in Mechanical Design. John Wiley, New
York, 1981, pp. 243-68.
3. Morrow, J. D., The effect of selected subcycle sequences in fatigue loading
histories. In Random Fatigue Life Predictions. ASME Publication PVP-
Vol. 72, 1986, pp. 43-60.
4. Lambert, R. G., Plastic work interaction damage rule applied to narrow-
band Gaussian random stress situations. ASME Journal of Pressure Vessel
Technology, 110 (1988) 88-90.
298 Wen-Fang Wu, Tsan-Hua Huang
5. Wirsching, P. H., Fatigue reliability analysis in offshore structures.
Proceedings of the Symposium on Probabilistic Methods in Structural
Engineering, St Louis, Missouri, ed. M. Shinozuka & J. T. P. Yao. 1981,
pp. 295-307.
6. Wirsching, P. H. & Shehata, A. M., Fatigue under wide band random
stresses using the rain-flow method. ASME Journal of Engineering
Materials and Technology, 99 (1977) 205-11.
7. Wirsching, P. H. & Haugen, E. B., A general statistical model for random
fatigue. ASME Journal of Engineering Materials and Technology, 96
(1974) 34-40.
8. Wirsching, P. H. & Light, M. C., Fatigue under wide band random
stresses. ASCE Journal of the Structural Division, 106 (1980) 1593-607.
9. Wirsching, P. H. & Haugen, E. B., Probabilistic design for random fatigue
loads. ASCE Journal of Engineering Mechanics Division, 99 (1973)
1165-79.
10. Marin, J., Mechanical Behavior of Engineering Materials. Prentice-Hall,
Englewood Cliffs, NJ 1962, pp. 186-98.
11. Corten, H. T. & Dolan, T. J., Cumulative fatigue damage. Proceedings of
the International Conference on Fatigue of Metals. Institution of Mechani-
cal Engineers and American Society of Mechanical Engineers, London,
1956, pp. 235-46.
12. Shinozuka, M. & Jan, C.-M., Digital simulation of random processes and
its applications. Journal of Sound and Vibration, 25 (1972) 111-28.
13. Dowling, N. E., Estimation and correlation of fatigue lives for random
loading. International Journal of Fatigue, 10 (1988) 179-85.
14. Cartwright, D. E. & Longuet-Higgins, M. S., The statistical distribution of
the maxima of a random function. Proceedings of the Royal Society of
London, Series A, 237 (1956) 212-32.
15. Lin, Y. K., Probabilistic Theory of Structural Dynamics. Robert E.
Krieger Publication, Huntington, New York, 1976, pp. 332-7.
16. Linsley, R. C. & Hillberry, B. M., Random fatigue of 2024-T3 aluminum
under two spectra with identical peak-probability density functions.
Probabilistic Aspects of Fatigue, ASTM STP 511. American Society for
Testing and Materials, Philadelphia, PA, 1972, pp. 156-67.