Compurers & Srrucrures Vol. 20. No. l-3, pp. 239-246, 1985 0045-7949/85 $3.00 + .

@I
Pnnted in the U.S.A. 0 1985 PergamonPress Ltd.

MECHANICS OF MATERIALS AND STRUCTURAL THEORIES

LIFETIME PREDICTION FOR METALLIC SPECIMENS

SUBJECTED TO LOADING WITH VARYING
INTENSITY

H. H. E. LEIPHOLZ~
Departments of Civil and Mechanical Engineering, Solid Mechanics Division,
University of Waterloo, Waterloo, Ontario, Canada N2L 3Gl

Abstract-Due to the interaction between various loading blocks of varying intensity, the po-
tential of a metallic test specimen to withstand fatigue fracture is affected in a detrimental way.
Specifically, load blocks of low intensity are influenced by preceding load blocks of high in-
tensity in such a way, that they contribute more strongly to fatigue damage than they would if
they were repeated without interference by other load blocks. This phenomenon of activation
of low intensity blocks may be explained by the assumption that high intensity blocks involve
sufficiently large compressive stress peaks which unlock otherwise closed cracks, thus accel-
erating the growth of damage. Therefore, Wohler curves, here introduced as Weibull curves
for reasons explained in the paper, and obtained by cycling virgin test specimens, become
questionable as they do not take account of the loading blocks’ interaction effects. It is therefore
proposed to work instead with modified Weibull curves which reflect the reduction of fatigue
life caused by block interaction. Some numerical examples explain how modified curves can
be obtained and how well and safely the fatigue life of a test specimen can be predicted.

1. INTRODUCTION namely, s,, S2, S3, . . . , yielding the “fatigue lifes”

NI, Nz, N3,. . . . If the s, are plotted against the
In engineering practice, so-called phenomenologi-
N,, the so-called s-N curve is obtained, see Fig. 1.
cal models are used for the prediction of fatigue
In the following, curves of that nature may be called
fracture. It is generally assumed that these models
Weibull curves.
are chosen with little regard for the physical basis
Phenomenologically, Weibull [2] has described
of fatigue fracture but mostly with the aim to pro-
such a curve by
vide an easily applicable procedure for design
against fracture. It is the intent to show in this paper
that contrary to such opinion a well chosen model @*)a = $, P> 1,
does indeed reflect physical fundamentals. For this
purpose, the stress-fatigue relationship, in a form where
attributed to Weibull, and the damge accumulation
law, i.e. Miner’s rule, will be reviewed. s* = s - so, N* = N + No. (2)
There can be no question that, since the physical
constitution of the object subjected to fatigue will In (2), so is a threshold stress value which is reached
be of great importance, all statements to be made for N + 00, and No is a quantity which can be in-
can have only limited generality. They should be terpreted as expressing the existence of initial dam-
seen in the context of the situation and experimen- age inherent to the test specimen even when cycling
tal evidence provided by the specific metal used in has not been performed at all. The existence of such
certain tests, to which reference will be made. initial damage is obvious in view of the fact that the
These tests performed on Van-80 steel produced by test specimen would indeed break at a final value
Midland Steel Company have been reported in [ 11. of s, that is for the ultimate stress s,,, which is sup-
posed to be applied statically (i.e. with no cycling,
N = 0).
2. THE s-N CURVE
Let it now be shown that relationship (1) can be
Assume for the sake of simplicity, that a sinu- justified on grounds of physics. Corresponding to
soidal stress with zero mean and with intensity s,, a stress cycle, one has a hysteresis loop in the s, E-
i = 1, 2, 3, . . . , is being applied to a test specimen plane (s = stress, E = strain) as shown in Fig. 2 in
until it breaks at the number Ni of load cycles. This a schematized form.
test is applied several times for various values of s, Assume that the damage caused by a stress cycle
is proportional to the area F of the hysteresis loop
t Professor. above the horizontal e-axis. Then, F = 0 for s, 5
239
240 H. H. E. LEIPHOLZ

~~~*

Nl N, N3 N

b)
Fig. 1. (a) Sinusoidal stress; (b) s-N curve.

so, so that there is no damage for s1 5 so in spite causes one and the same damage,
of stress cycling, making so the threshold stress for
the curve in Fig. l(b). But, D
F= (7)
P(Ni + N*)
F = he f 3S’E (3)
holds, where pF = d is the damage of an individual
cycle, (p being a proportionality factor), D is the
for s, > sO, The quantities used in (2) are shown in
total damage the test specimen can take, and Ni and
Fig. 2.
No stand for fatigue lie and inherent initial damage,
It is easily found that
respectively.
Combining (6) and (7) yields
h = s&in OL, e = E sin (r,
E = s’(crg a’ - ctg a), s’ = s, - $0. (4)
W
$’ =I
sf + (8)
Setting PAW, + No)>’

ctg (Y’ - erg a = A, (5)

In deriving (8), the simple assumption that E =
As’ has been made according to (4) and (5). A more
and using (4) and (5) in (3), one has
general assumption is that

F = (Si - SO)‘$
+ So(Si - SCM. ‘6) F. = Ad, y< 1. (9)

Assuming further that there is no aging of tht Set, for example, y = l/n, n > 1. Then,
metal due to repeated cycling, so that each cycle .
F = (s, - s~)~+““;+ as&, - #“A (10)

Using (10) in (7) yields

(s, - S(Jn+t’n + ZSO(Si - SO)lin

20
(11)
= pA(N, + NoI

For sl > sO, (s, - so)“” 6 (si - s#+lin. Therefore,

the second term on the 1.h.s. of (11) may be neg-
lected leading to

(s, - Sg)n+“n= ,,N”+

f
N*). (12)

Using (2) in (12) and setting

n+l
K = ZDlpA, p=-->l, (13)
Fig. 2. Schematized hysteresis loop. n
 Lifetime prediction for metallic specimens 241

transforms (12) into Comparing (18) with (19), one finds that

which means that relationship (18) is more conser-

which is Weibull’s relationship (1), indicating in-
vative than Weibull’s relationship (19). It may
deed that this relationship does reflect fairly well therefore be advisable to use (18) instead of (19) for
some of the physical facts behind metal fatigue a design. The two relationships (18) and (19) are
caused by stress cycling. These facts are at this depicted in Fig. 3.
point assumed to be given by the axions
(i): Damage d caused by a stress cycle is given
by 3. ON DAMAGE ACCUMULATION

d = pF, p = factor of proportionality, The actual problem of damage accumulation

F = area of hysterisis comes up, when one is faced with loading of varying
(15)
intensity as shown in Fig. 4. There can be no ques-
loop above e-axis. tion that the specimen subjected to cycling with re-
peatedly changing stress amplitudes will suffer a no-
(ii): All cycles cause identical damage. ticeable deterioration of its internal structure. This
(iii): There is a total damage fact will have to be considered in the following the-
ory:
D = d(N, + No), (16) Consider a loading block Bi, i = 1, 2, 3, . . . ,
with stress intensity si and damage di per cycle. Let
which the metal considered can take before
the number of cycles in Bi be n,. Then, the damage
breaking.
caused by B, is
(iv): There is an inherent initial damage
Di = nid,. (21)
Do = dNo (17)

present in the metal already without any cy- Hence, forj loading blocks, the total damage is
cling being performed.
It can easily be seen that by applying axioms (i)- D = i Di = 5 n,d,.
(iv), one is in fact led to eqn (11) when using expres- r=, ,=I

sion (10).
Return to (8). Using (13), the Weibull curve cor- Suppose, the test specimen were subjected to cy-
responding to this relationship is analytically ex- cling with a constant stress amplitude s, until it
pressed by breaks at Ni,o cycles. In this test, with uniform si
as stress intensity, the total damage were
K “’
s, = (18)
” + N, + No> ’ (23)

Obviously, p = 2. Correspondingly, using (2), the The total damage Do satisfies the condition
Weibull relationship (1) yields for l3 = 2
Do I=-D. (24)

sy’ = So + (&)“t (19) This is the case, because D corresponds to a situ-

ation in which the test specimen had been weak-
ened due to the effect of varying stress intensity,
while Do corresponds to cycling with uniform stress
intensity carried out on unweakened test specimens
in the virgin state.
From (23),

di = +. (25)
(I91
1.0

is obtained. Using (25) in (22) yields

i-4 ’ n,Do
I “Jo D=z+.
i= 1 1.0
(26)

Fig. 3. Comparison of Weibull curves following from

relationships (18) and (19). This expression can be transformed into
242 H. H. E. LEIPHOLZ

S
block I
n,= 3 “2
=2 n3= 2
53---_

Fig. 4. Loading with varying intensity.

(27) (31)

which can be seen as an analogue to the usual form Clearly, the interaction effect involved in loading
of Miner’s rule. with varying intensity suggests replacing (31) by
Quanity N,.,,D/Do may be discussed further: (30).
First. Let

WCN (32)
r.O*
DO

by virtue of (24). Moreover, this quantity may be be the frequency of cycles belonging to block B, in
written as the total loading history with N cycles. Let N be
the total number of cycles leading to fatigue frac-
ture. That means, N is the “fatigue life”. Using (32)
N, OD
d = Gt’Nc.01. GfNi.0) < Nr.07 (29) in (30) yields
Do

where G is an appropriate operator mapping N,,o (33)

into itself. This is so, because D/Do is not a constant
ratio but rather a function depending on varying
index i. A possible form of (29) is shown in Fig. 5. From (33). fatigue life can be determined as
With (29), relationship (27) becomes

N= [&zkJ’. (34)
If one would disregard the interaction effect pres-
ent in foading with varying intensity and work with
while Miner’s rule in its classical form would read (31), one would find for fatigue life the expression

C(Nip) (35)

I respectively. Yet, due to G(N,,) < N,.o, also N <

fi. Hence (34) yields safer fatigue life estimates than
(35). Consequently, one should work with (34) and
not with (35).

4. THE CODIFIES WEIBULL CURVE

I Values of Ni.o are available from the Weibull
N3,o Ni,o curve. This curve has been obtained by subjecting
Fig. 5. G(N,,d curve. virgin test specimens to consecutive cycling with
 Lifetime prediction for metallic specimens 243

Modified Weibull

Modified Wtibull

A
Nl ^N2 ^N3 h
Fig. 6. --- Weibull curve for virgin test specimens + ;
- modified Weibull curve to be used for damage ac-
Fig. 7. The s-fi relationshlps. &‘, = G(N,,o).
cumulation A.

uniform loading intensities Si, i = 1, 2, 3, . . . , see the physics behind the steps which led to the in-
Fig. 6. Using the G(Ni,o)-curve in Fig. 5, one can troduction of the concept of this curve.
find for every NI,Ovalue, say, N3,a, the correspond- The G(N,J curve is the tool which serves to
ing G(N,,,) value, say, G(NjJ as indicated in Fig. transform the original Weibull curve (Fig. 6) into
5. Plotting the so obtained G(N,,J values against its modification, namely, the curve representing the
the N-axis as shown in Fig. 6 and bringing the ver- s-&relationship (Fig. 7). This transformation is nec-
tical lines through the G(N,,o) points to an inter- essary to account for the detrimental effect of load-
section with the horizontal lines through the s, val- ing with varying intensity on the fatigue endurance
ues yields points which are marked in Fig. 6 by of the metal specimen.
triangles. Drawing a curve through these triangles Fatigue experiments using virgin test specimens
results in the modified Weibull curve. yield the various points in the s, N-plane which es-
It is for the sake of procedural simplicity rec- tablish the original Weibull curve. Points on this
ommendable to introduce a change of notation by curve corresponding to relatively small s values
setting (stress intensities) have relatively large abscissae
corresponding to large N values (fatigue lifes). This
G(Ni.0) = A%“,. (36) is so, because crack closure occurs, making cycles
with small stress intensities less damaging, since
Redrawing the modified Weibull curve from Fig. 6 due to closure the cracks do not completely open
in terms of s,, and A, yields the ‘*s-a relationship” up during a load cycle. However, if loading involves
shown in Fig. 7. With (36), eqn (34) assumes the preceding blocks with sufficiently large stress in-
form tensities, the sufficiently large compressive stresses
occurring will unlock the cracks, so that even
blocks with small stress intensities will contribute
(37)
essentially to damage, reducing fatigue life. That
means: smaller N values. Therefore, the points in
which corresponds completely to (35). the s, N-plane are shifted to the left, and the smaller
One is now able to make the following statement: the s-ordinate the greater this shift will be. The re-
In the case of loading with varying intensity, de- sult of this process is the modified Weibull curve
termine the G(N,,&curve shown in Fig. 5. Use which reflects in this way the influence of loading
this curve to obtain through a simple transjior- with varying intensity.
mation of the actual Weibull curve the modified One can now conclude that the G(Ni.0) curve
one yielding the s-h relationship in Fig. 7. By could be drawn, either systematically or by trial and
means of this relationship, findfor the intensities error, so as to allow the shifting of the original Wei-
s, of the various load blocks B, the corresponding bull curve into a modified form which corresponds
values fi,. Applying these RI, values to the com- well enough to the situation following from loading
mon form (37) of Miner’s rule, calculate the fa- with varying intensity.
tigue life N of the specimen. Before proceeding to a description of a system-
atic approach to the modified Weibull curve, let
first another question be answered: What is the
range of applicability of a specific modified curve?
5. THEG(N,,o) CURVE
Will this curve only be valid for the specific loading
Basic for all preceding considerations was the history for which it was derived or will it have va-
knowledge of the G(Ni.0) curve. In order to estab- lidity for a broad class of loading cases? Only if the
lish a procedure for its derivation, one must review latter is true, the concept of the modified Weibull
244 H. II. E. LEIPHOLZ

curve will have reievance for a practical applica- the simultaneous equations
tion.
In [31, Topper and Au have reported on the re-
25,960 = [g + g]-‘,
markable effect that compressive stress peaks have
on crack growth. If a compressive stress peak has
sufftcient intensity, it has a long lasting influence, 42,860 = [E + %I-‘, (38)
accelerating crack growth. It can be found in [3]
that such a stress peak will affect the situation over
a range of 200,000 cycles. For the case under con- with the solutions
sideration in this paper, these facts can be inter- RI = 1,)4.“64, &2 = 106.268,
preted in the following way: Any loading histories (3%

which involve compressive stress peaks exceeding

a certain intensity level and having a frequency be- One can now plot in the e, N-plane (see Fig. 8)
yond a certain lower frequency limit, will have the two points
more or less the same effect on fatigue damage:
P* = (0.5%,104"*), P‘f= (O.l%, 106.*68). (40)
They all unlock equally and permanently the closed
cracks making the original Weibull curve to the
Obviously, P* is to the right of the original Weibull
same degree shift to the left. Hence, for the broad
curve. This is unacceptable, as an increase in fa-
class of loading histories involving stress cycles
tigue life is improbable. Therefore, point P* is re-
with 1s / > Slimgtand PI > Plimit, (where -sI,~,~ is
placed by the point
the compressive stress to be exceeded and QIlmliis
the minimum frequency for those cycles with in-
tensities exceeding s,,,,,,~). the modified Weibull P, = (0.5%.ro‘y, (41)
curve, and therefore also the G(N,,o) curve, are es-
sentially valid overall. This has, for example, been which is still on the original Weibull curve.
experimentally justi~ed in [I]. One can therefore Now, the first version of the modi~ed Weibull
infer that a modified Weibull curve,derived using curve will be introduced as follows: to the left of
a certain fatigue test, will have validity for a whole P,,the WeibuH curve and its modification are as-
spectrum of other fatigue tests whose loading his- sumed to coincide. To the right of P, , the modified
tories satisfy the intensity and frequency require- Weibull curve is assumed to be the straight line
ments mentioned above. through the points PI and Pd.This may seem to be
a rather rough approximation. However, as a sub-
sequent calculation will show, it may be a quite ac-
6. NUMERICAL EXAMPLES ceptable one: Another test involving two loading
blocks with the strain amplitudes el = 0.5% and ~2
In many cases, it is convenient to derive the mod-
= 0.1% and the frequencies p, = Br = 0.5 gen-
ified Weibull curve directly instead of determining
erated the fatigue life N = 21,803. Let this value
first the G(N,,) curve and then, subsequently, the
now be predicted using the first version of the mod-
rn~i~ed Weibull curve using for this end the pre-
ified Weibull curve. One has N, = 104, N, =
viously obtained G(N,,a). Such procedure is not
106268. By virtue of (37),
only advisable but even necessary in the case that
the modified Weibull curve is to be extended below
0.5 -’ = N
the original fatigue limit, which may have been low-
ered or completely removed due to the effect of
crack unlocking.
$+
[
106’68
1
Moreover, instead of characterizing a load cycle and
by its stress amplitudes, one may prefer using the
strain amplitudes. This is indeed recommendable, 1854286
N=-= 19,895.
as one is better able to control the load by strain 93.2
than by stress.
In the foliowing, the data will in fact be based on This value is comfortably close to the experimental
strain, and also the modified Weibull curve will be value, the difference being only 8.8%. Also, the the-
derived without making a detour via the G(N,e) oretical predicted value of N is on the safe side.
curve. The data that will be used are taken from A test with two loading blocks having the strain
El Menoufy’s thesis, see 111. peaks E, = 0.4%, l2 = 0.1% and the frequencies
Consider two-block tests with strain amplitudes pi = 0.4, pZ = 0.6 led to the fatigue life N = 49,486.
of e1 = 0.5% and e2 = 0.1%. For the first test, Let the assumption be made. that the 0.4% block
l3f” = BY’ = 0.5, and the fatigue life is N = 25,160 has a co~sponding detrimental effect on the 0.1%
cycles. For the second test, 13i2)= 0.3, @-’ = 0.7, block as the preceding 0.5% block had. Then, a sec-
and the fatigue life is N = 42,860 cycles. ond version of the modified Weibull curve is ob-
These data must satisfy eqn (37). Thus, one has tained by connecting point
 Lifetime prediction for metallic specimens 245

Weibull curve and its

modification coinciding

QS%
0.4 % I curve
a3 %

Weibull
curve
-Ps
0.1% \

IO IO~*~~*N
Fig. 8. --- Weibull curve: (log z + 2.69)(log N + 4.3)‘.s32= 3.542 x 104; -
modified Weibull curve (first version); .-.-. Modified Weibull curve (second version);
..-..-.. modified Weibull curve (third version).

Pz = (0.4%, 104.*j Again, it is assumed that the 0.3% block did affect
the 0.1% block similarly as the previous 0.5%
in Fig. 8 with point P4, the portion of the original block. Therefore, the third version of the modified
Weibull curve above of Pz remaining unchanged. A Weibull curve is obtained by connecting point
prediction of N by means of (37) yields now

,
P3 = (0.3%, 1ti9)

in Fig. 8 with point P4, while the portion of the

original Weibull curve above of P3 is to remain un-
and changed.
Predicting N by using (37) results in

=
N = 1854286
- = 39,120.
47.4
N,

The predicted value is again on the safe side, and

it differs from the experimental value by 21%, and in
which is still acceptable.
Finally, a test with two loading blocks will be 1854286
N=-= 153,247.
considered which had the strain peaks E, = 0.3%, 12.1
= O.l%, the frequencies f3r = pz = 0.5, and
2hich produced N = 154,476. This value is still on the safe side and at the same
246 H.H.E. LEIPHOLZ

time very close to the experimental value, differing REFERENCES

from it only by 0.8%.
I. M. El Menoufy, Fatigue life evaluation for a metal sub-
It may be concluded that according to the pre- jetted to a certain class of random loading using the
ceding nume~~al examples the concept of the modified life-law concept. Thesis, University of Wa-
“modified Weibull curve” seems indeed to be a terloo ( 1982).
useful tool for fatigue life prediction in the case of 2. W. Weibull, Fatigue Testing and Analysis ofResults,
loading with varying intensity. p. 174. Pergamon Press, New York (1961).
3. T. Topper and P. Au, Cyclic strain approach to fatigue
Acknow/edgement-The support of this research through in metals. AGARD-Lecture Series-118, NTIS, pp. 1I-
NSERC Grant No. A7297 is gratefully acknowledged. 1 to 1l-25. Springfield, Virginia (1981).