1

Fatigue of Metals
mechanical components can fail at stresses well below the
tensile strength of the material if subjected to alternating loads
failure of ductile materials unter alternating loads occurs in a
quasi brittle manner, i.e. by crack propagation
failure is preceded by characteristic changes in the material
microstucture
this phenomenon is called metals fatigue
A. The Comet Disasters
A famous series of fatigue failures led to the de Havilland
Comet crashes of the early 50s
The Comet was designed and built in the UK. It was the
worlds first commercial jet airliner.
I. Case Studies (the Disasters Catalogue)
2
May 2, 1953: G-ALYV disintegrated in a thunderstorm at
10000 ft during its initial climb on a flight from Calcutta
to Delhi
January 10, 1954: G-ALYP crashed from at 27000 ft in good
weather on a Rome to London flight
April 8, 1954: G-ALYY disappeared on a flight from Rome to
Cairo. All comet aircraft were grounded.
3
By investigation of recovered wreckage from G-ALYP and
pressure cycle testing of the fuselage of G-ALYU in a water
tank, fatigue failure of the fuselage was identified as the
cause of the Comet accidents
Cause of the Comet disasters
1) The economic backdrop:
In order to provide an economically satisfactory payload and range
at the high cruising speed which the turbo-jet engines offered, it was
essential that the cruising height should be upwards of 35,000 ft.
double that of the then current airliners and that the weight of the
structure and equipment should be as low as possible (Official
accident report)
This forced engineers to go at (and beyond) the limits of then current
engineering practice in designing an aircraft operating under (for the
times) extreme conditions.
4
2) This led to engineering design flaws:
Engineers were aware of potential problems due to the novel and
(for the time) extreme service conditions and tried to ensure stability
of the pressure cabin, also against fatigue, but:
They failed to properly evaluate stress concentrations near the
corners of windows
Instead of doing calculations, design engineers relied on
practical tests (very British...)
... which, though in line with good engineering practice of the
time, were substantially flawed (use of static tests to evaluate
fatigue resilience)
(http://www.geocities.com/CapeCanaveral/Lab/8803/comgalyp.htm).
But it cant happen these days?!
B The B737-200 disasters
On August 22, 1981 the fuselage of
Far Eastern Airlines Flight 103 disintegrated in mid-flight.
On April 28, 1988, the B737-200 of Aloha Airlines Flight 243
mutated into a convertible at an altitude of 27000ft. By some
kind of miracle, only one person was killed.
5
These accidents are interesting in
view of the safety philosophy
employed (fly-till-it-breaks)
which puts emphasis on ability
to contain failures once occurred
(safe decompression) scenario
rather than on detection and
replacement of fatigued parts
But: detection programs may not help either... (Enschede 1998)
II. Fatigue Lifetime Evaluation
A. Characterization of alternating loads
A1. Reversed/repeated stress cycles: Periodic stress vs time signals

max
: maximal stress

min
: minimal stress

m
= (
max
+
min
)/2:
mean stress

r
=
max
-
min
:
stress range

a
=
r
/2:
stress amplitude
0



r


m

max

min
time
6
Maximum and minimum stress equal in magnitude:
mean stress = 0, reversed stress signal
Otherwise: repeated stress signal
But: Real stress vs time signals are hardly periodic. What to do
if the stress vs time signal is irregular?
A2: Characterization of irregular load vs time curves
1) Characterize load curve in terms of maxima and minima only
2) Introduce classes (stress intervals) such that maxima/minima
belonging to the same class are counted similarly (effectively:
round the values of
max
/
min
values
0.0 0.2 0.4 0.6 0.8 1.0
-800
-600
-400
-200
0
200
400
600
800
1
2
3
4
5
6
7


S
t
r
e
s
s

[
M
P
a
]
Time [s]


7
Signal may now be characterized in terms of
peak counting: Histogram of maxima/minima
range counting: Histogram of ranges (differences
between adjacent maxima/minima)
level crossing counting: Histogram of level crossings
rainflow counting: Conversion of irregular time series
into a sequence of cycles
Reduce the time history to a sequence of maxima and minima
Turn the graph by 90 (earliest time to the top) and imagine it
depicts a pagoda roof.
Each minimum is imagined as a source of water that "drips"
down the pagoda.
Count the number of half-cycles by looking for terminations in the
flow occurring when either:
a) It reaches the end of the time history;
b) It merges with a flow that started at an earlier minimum
c) It flows above a minimum of greater depth than its origin.
Repeat this for the maxima
Assign a magnitude to each half-cycle equal to the stress difference between
its start and termination.
Pair up half-cycles of identical magnitude (but opposite sense) to count the
number of complete cycles. Typically, there are some residual half-cycles.
Sometimes, they can be paired up to close the loop.
The rainflow algorithm
8
After counting: Characterize each cycle / residual half cycle
in terms of its mean stress and amplitude
Rainflow matrix: Characterize each cycle in terms of its start,
maximum and minimum stresses (start stress => first index)
Rainflow matrix for load signal above:
or
3
4
I 5
I I 1
7 6 5 4 3 2 1
I 2
6
7
3
4
5
I I 1
7 6 5 4 3 2 1
I 2
6
7
1
2
1
2
1
2
B. Fatigue lifetime evaluation
B1. Characterization of fatigue properties
Fatigue testing is commonly done assuming reversed / repeated
loads
Determine number N of cycles to failure as a function of
stress amplitude
a
(S)
Lifetime depends on mode of testing (eg. bending vs.
torsion) and possibly on specimen geometry (notch effects)
Representation: S-N curve (Woehler curve)
9
1 10 100 1000 10000 100000 1000000 1E7
0
200
400
600
800
1000
Number of cycles, N
T
e
n
s
i
l
e

s
t
r
e
s
s

a
m
p
l
i
t
u
d
e

[
M
P
a
]
Typical S-N curve for a high strength steel
Lifetime decreases with increasing stress amplitude
Existence of an endurance limit (infinite lifetime below EL)

m
=-400MPa

m
=400 MPa
Little or no influence of cycling rate
Influence of mean stress:
> tensile mean stress reduces lifetime (crack opening)
> compressive stress increases lifetime (crack closure)
10
B2: Lifetime evaluation for loads with nonzero mean stress:
1) If lifetime data with mean stress are available:
Read lifetime off S-N curve (interpolate if necessary)
2) If only lifetime data without mean stress are available:
Use some empirical estimate to convert true amplitude
and mean stress into effective stress amplitude and
use the corresponding lifetime from S-N curve
11
B3: Lifetime evaluation for irregular load vs time signals
Perform rainflow classification of load vs time signal
Determine fatigue lifetimes N
i
for all occurring types
of cycles (for all elements of the rainflow matrix)
Let n
i
be the number of cycles of type i
Then failure occurs if
(Palmgren-Miner rule)
1
i
i
i
n
N
=

Example:
For previously discussed stress vs time signal: When will
the material of the example S-=N curve fail if subjected
to this periodically repeated signal?
4 classes of cycles.
Two (1-2-1 and 2-3-2) below fatigue limit: N=
1-6-1: N ~ 800
1-5-1: N~3000
1 1
[s] 1
3000 800
631s
T
T
(
+ =
(

=
12
C. Statistics of fatigue
Measured lifetimes exhibit huge statistical scatter.
Consequence: need for statistical description to predict
failure safety margins
Some probabilistic notations:
probability density function
probability that a part fails (under given conditions)
between times t, t+dt
cumulative probability (distribution function)
probability of failure before time t
( )d p t t
0
( ) ( ')d '
t
F t p t t =

Mean lifetime:
Variance of lifetimes:
Example:
Gaussian distribution, mean <t>, standard deviation s
t
2 2
2
2 2
( )d
( )d ( )d
t
t t p t t
s t t t p t t t p t t
=
(
= =

=
2
2
2
2
) (
exp
2
1
) (
t
t
s
t t
s
t p

13
The corresponding cumulative probability can be expressed
in terms of the so-called error function erf (x)
The cumulative distribution of the Gaussian is then
dt
t
x
x
(

=
0
2
2
exp
2
) ( erf

(
(

|
|

\
|
+ = =

t
t
s
dt t p t F
2
t - t
erf 1
2
1
' ) ' ( ) (
Fatigue lifetimes are usually not described by Gaussian
distributions (Why?)
Instead one usually uses a log-normal distribution, ie., the
logarithm of the lifetime (in cycles) is assumed to be Gaussian
distributed
Instead: Log-normal distribution
This can be used to evaluate time-dependent
failure/survival probabilities (see tutorials)
( )
2
2
2
log
log
log log
1
(log ) exp
2
2 N
N
N N
p N
s
s
(

(
=
(

14
Log-normal distribution:
This can be used to evaluate time-dependent
failure/survival probabilities (see tutorials)
( )
2
2
2
log
log
log log
1
(log ) exp
2
2 N
N
N N
p N
s
s
(

(
=
(

If fatigue lifetime is governed by weakest link in a
chain of identical elements (eg. in cables, Forth
Road Bridge):
Lifetimes obey Weibull distribution
Probability for failure before N cycles:
(N
0
, ) parameters of the Weibull distribution
NB: What is the variance of this distribution?
0
( ) 1 exp
N
F N
N

(
| |
( =
|
(
\

15
II. Microstructural Aspects of Fatigue
Fatigue leads to characteristic microstructure changes on the
level of the dislocation arrangement.
Dislocation microstructure in a fatigued Ni polycrystal
(SEM, channeling contrast)
Microstructure depends on stress amplitude:
At stress amplitudes below endurance limit: Patchy matrix
pattern with very little plastic activity (plastic strain amplitude
<10
-4
At the endurance limit: Formation of lamellar persistent slip
bands (PSB) parallel to slip plane with highest resolved shear
stress. Local plastic activity in PSBs is about 100 times higher.
Fatigue cracks develop where the PSB hits the surface and
grow along the PSBs
16
There seems to be a one-to-one correspondence between PSB
occurrence and fatigue failure (but we dont know why PSB occur...)
Important for practice: Fatigue cracks nucleate at surfaces
Possibility to deal with fatigue problems by surface treatment
(Grain refinement, large plastic deformation, hard coatings)