You are on page 1of 123

Fatigue

Failure

1
Materials Tetrahedron
Processing
Performance

Microstructure Properties

2
Objective

 The objective of this lecture is to


explain the phenomenon of fatigue
and also to show how resistance to
fatigue failure depends on stress –
strain distribution, stress amplitude
and material microstructure.

3
sa := Alternating stress
sm := Mean stress Notation
R := Stress ratio
e := strain
Nf := number of cycles to failure
A := Amplitude ratio
∆pl := Plastic strain amplitude
∆el := Elastic strain amplitude
K’ := Proportionality constant, cyclic stress-strain
n’ := Exponent in cyclic stress-strain
c := Exponent in Coffin-Manson Eq.;
also, crack length
E := Young’s modulus
b := exponent in Basquin Eq.
m := exponent in Paris Law
K := Stress intensity
∆K := Stress intensity amplitude
a := crack length

4
Fatigue Defined
 ASTM E206-72 Definition
The Process of PROGRESSIVE
LOCALIZED PERMANENT Structural
Change Occurring in a Material
Subjected to Conditions Which Produce
FLUCTUATING Stresses and Strains at
Some Point or Points Which May
Culminate in CRACKS or Complete
FRACTURE After a Sufficient Number
of Fluctuations
5
Fatigue
 Fatigue is the name given to failure in response
to alternating loads (as opposed to monotonic
straining).
 Instead of measuring the resistance to fatigue
failure through an upper limit to strain (as in
ductility), the typical measure of fatigue
resistance is expressed in terms of numbers of
cycles to failure. For a given number of cycles
(required in an application), sometimes the
stress (that can be safely endured by the
material) is specified.

6
Fatigue: general characteristics

 Primary design criterion in rotating parts.


 Fatigue as a name for the phenomenon based on the
notion of a material becoming “tired”, i.e. failing at
less than its nominal strength.
 Cyclical strain (stress) leads to fatigue failure.
 Occurs in metals and polymers but rarely in
ceramics.
 Also an issue for “static” parts, e.g. bridges.
 Cyclic loading stress limit<static stress capability.

7
Fatigue: general characteristics
 Most applications of structural materials involve cyclic
loading; any net tensile stress leads to fatigue.
 Fatigue failure surfaces have three characteristic
features:
• A (near-)surface defect as the origin of the crack
• Striations corresponding to slow, intermittent crack
growth
• Dull, fibrous brittle fracture surface (rapid growth).
 Life of structural components generally limited by
cyclic loading, not static strength.
 Most environmental factors shorten life.

8
Three Stages of Fatigue Failure
 Crack Initiation
 Crack Propagation
• oscillating stress… crack grows, stops
growing, grows, stops growing… with crack
growth due to tensile stresses
 Fracture
• sudden, brittle-like failure

9
Fatigue Failure
 Caused by Load-
Cycling at <y
 Brittle-Like Fracture
with Little Warning
by Plastic
Deformation
• May take Millions of
Cycles to Failure 1. Crack Initiation Site(s)
2. “Beach Marks” Indicate of
 Fatigue Failure Crack Growth
Time-Stages 3. Distinct Final
Fracture Region
10
Identifying Fatigue Fractures

beachmarks

11
Fatigue Crack Nucleation
 Flaws, cracks, voids can all act as crack
nucleation sites, especially at the
surface.
 Therefore, smooth surfaces increase
the time to nucleation; notches, stress
risers decrease fatigue life.
 Dislocation activity (slip) can also
nucleate fatigue cracks.

12
Role of dislocations

 Dislocations in a material provide a mechanism by


which large numbers of atomic bonds can be
broken and re-formed
• the theoretical strength of ideal, dislocation-free
materials is much higher than that measured in practice
• also, the preparation and treatment of the material
significantly influenced the measured strength
 Recall the two main types of dislocation
• edge dislocation
• screw dislocation
13
Edge dislocation
 Under an applied stress, the edge dislocation can
move in the direction of the stress
unit step
of slip

slip plane

shear stress

 This process, leading to elastic deformation, is


called slip
14
Screw dislocation
 The screw dislocation itself moves perpendicular
to the stress direction, but the deformation ends
up the same

15
Slip systems
 In crystalline materials, the anisotropy of the structure
can mean that certain slip directions are preferred
• termed slip planes
• and these move in slip directions
 Together slip planes and directions are called slip
systems
• and these slip systems act to minimise the overall
atomic distortion caused by the motion of the
dislocation
 It follows from the above that the slip planes are those
planes in the crystal which have the highest packing
density of atoms
• by keeping these densely packed atoms together,
fewer bonds are distorted

16
(111) plane of a FCC
material, showing three
<110> slip directions

 The number of slip systems depends on the


crystal structure
• each slip system for BCC and FCC has at least 12 slip
directions
• while the maximum for HCP is 6
 Hence FCC (Cu, Al, Ni, Ag, Au) and BCC (Fe)
metals tend to be ductile and exhibit large plastic
deformation
• while HCP (Ti, Zn) are more brittle

17
Dislocation Slip Crack Nucleation

 Dislocation slip -> tendency to localize


slip in bands.
 Persistent Slip Bands (PSB’s)
characteristic of cyclic strains.
 Slip Bands -> extrusion at free surface.
 Extrusions -> intrusions and crack
nucleation.

18
Intrusions and Extrusions:
The Early Stages of Fatigue Crack Formation

 

19
Fatigue Crack Propagation
 Crack Nucleation stress intensification at crack
tip.
 Stress intensity crack propagation (growth);
- stage I growth on shear planes (45°),
strong influence of microstructure
- stage II growth normal to tensile load (90°)
weak influence of microstructure.
 Crack propagation catastrophic, or ductile
failure at crack length dependent on boundary
conditions, fracture toughness.

20
Schematic of Fatigue Crack Initiation Subsequent Growth Corresponding
and Transition From Mode II to Mode I

Δσ
Locally, the crack grows in shear;
macroscopically it grows in tension.
Δσ

21
The Process of Fatigue

The Materials Science Perspective:


• Cyclic slip,
• Fatigue crack initiation,
• Stage I fatigue crack growth,
• Stage II fatigue crack growth,
• Brittle fracture or ductile rupture

22
Features of the Fatigue Fracture Surface of a Typical Ductile Metal
Subjected to Variable Amplitude Cyclic Loading

A – fatigue crack area


B – area of the final static
failure

23
Appearance of Failure Surfaces Caused by Various
Modes of Loading (SAE Handbook)

24
Factors Influencing Fatigue Life
Applied Stresses
 Stress range – The basic cause of plastic deformation and consequently the
accumulation of damage
 Mean stress – Tensile mean and residual stresses aid to the formation and
growth of fatigue cracks
 Stress gradients – Bending is a more favorable loading mode than axial
loading because in bending fatigue cracks propagate into the region of lower
stresses
Materials
 Tensile and yield strength – Higher strength materials resist plastic
deformation and hence have a higher fatigue strength at long lives. Most
ductile materials perform better at short lives
 Quality of material – Metallurgical defects such as inclusions, seams,
internal tears, and segregated elements can initiate fatigue cracks
 Temperature – Temperature usually changes the yield and tensile strength
resulting in the change of fatigue resistance (high temperature decreases
fatigue resistance)
 Frequency (rate of straining) – At high frequencies, the metal component
may be self-heated.

25
Strength-Fatigue Analysis Procedure

Material Component Loading


Properties Geometry History

Stress-Strain
Analysis

Damage
Analysis

Allowable Load - Fatigue Life

Information path in strength and fatigue life prediction


procedures
26
Design Philosophy:
Damage Tolerant Design

 S-N (stress-cycles) curves = basic characterization.


 Old Design Philosophy = Infinite Life design: accept
empirical information about fatigue life (S-N curves);
apply a (large!) safety factor; retire components or
assemblies at the pre-set life limit, e.g. Nf=107.
 Crack Growth Rate characterization ->
 Modern Design Philosophy (Air Force, not Navy
carriers!) = Damage Tolerant design: accept
presence of cracks in components. Determine life
based on prediction of crack growth rate.

27
Three Theories

stress-based, for high-cycle fatigue, aims to


Stress-Life prevent crack initiation

useful when yielding begins (i.e., during


Strain-Life crack initiation), for low-cycle fatigue

LEFM (Fracture best model of crack propagation, for low-


Mechanics) cycle fatigue

28
Definitions: Stress Ratios
 Alternating Stress  a
 Mean stress  m = (max +min)/2.
 Pure sine wave Mean stress=0.
 Stress ratio  R = max/min.
 For m = 0, R=-1
 Amplitude ratio  A = (1-R)/(1+R).
 Statistical approach shows significant distribution
in Nf for given stress.

29
Types of Fatigue Loading

Fully Reversed Repeated Fluctuating

   max   min stress range a


amplitude A
 alternating
ratio m
a  component
2
 min
 max   min mean stress ratio R
m  component  max
2
30
Constant and Variable Amplitude Stress Histories; Definition of a Stress
Cycle & Stress Reversal

a) Constant amplitude stress history In the case of the peak stress


max history the important parameters
are:

 peak   max
peak
  min
peak
;
Stress

m

One min
 peak  max
peak
  mipeak
cycle
 apeak   n
2 2
0 Time
Variable amplitude stress
 max   min
b) peak peak
history
 peak
m  ;
2
Stress

 mipeak
R  penak
One
reversal Time
 max
0
31
Mean Stress
 Alternating stress  a = (max-min)/2.
 Raising the mean stress (m) decreases Nf.
 Various relations between R = 0 limit and the ultimate
(or yield) stress are known as Soderberg (linear to
yield stress), Goodman (linear to ultimate) and
Gerber (parabolic to ultimate).
endurance limit at zero mean stress
a
  mean 
 a   fat
1  

 tensilestrength
tensile strength
mean
32
Stress amplitude, Sa Mean Stress Effect

sm> 0

0
sm= 0 time

sm< 0
Stress amplitude, logSa

sm< 0

sm= 0

sm> 0

No. of cycles, logN 2*106


33
The tensile mean stress is in general detrimental while the compressive mean stress is beneficial or
has negligible effect on the fatigue durability.
Because most of the S – N data used in analyses was produced under zero mean stress (R = -1)
therefore it is necessary to translate cycles with non- zero mean stress into equivalent cycles with
zero mean stress producing the same fatigue life.
There are several empirical methods used in practice:
The Hiagh diagram was one of the first concepts where the mean stress effect could be accounted
for. The procedure is based on a family of Sa – Sm curves obtained for various fatigue lives.

Steel AISI 4340,


Sy = 147 ksi (Collins)

34
Mean Stress Correction for Endurance Limit
2
Sa  S m 
Gereber (1874)   1 Sa – stress amplitude
Se  S u  applied at the mean
stress Sm≠ 0 and fatigue
life N = 1-2x106cycles.
Sa S m
Goodman (1899)  1 Sm- mean stress

Se Su Se- fatigue limit at Sm=0


Su- ultimate strength
Sa Sm f’- true stress at fracture
Soderberg (1930)  1
Se Sy
Sa Sm
Morrow (1960)  ' 1
Se  f
35
Mean stress correction for arbitrary stress
amplitude applied at non-zero mean stress
 Sm 
2
Sa
Gereber (1874)   1 Sa – stress amplitude

 Su 
applied at the mean
S ar stress Sm≠ 0 and
resulting in fatigue life of
Sa Sm N cycles.
Goodman (1899)  1 Sm- mean stress
S ar Su
Sar- fully reversed stress
amplitude applied at
Sa Sm mean stress Sm=0 and
Soderberg (1930)  1 resulting in the same
S ar Sy fatigue life of N cycles
Su- ultimate strength
Sa Sm f’- true stress at fracture
Morrow (1960)  1
'
S ar f
36
Comparison of various
methods of accounting
for the mean stress effect

Most of the experimental data lies between the Goodman and the yield line!

37
Approximate Goodman’s diagrams for ductile and brittle
materials

Kf

Kf

Kf

38
The following generalisations can be made when discussing
mean stress effects:
1. The Söderberg method is very conservative and seldom used.
3. Actual test data tend to fall between the Goodman and Gerber curves.
3. For hard steels (i.e., brittle), where the ultimate strength approaches the
true fracture stress, the Morrow and Goodman lines are essentially the
same. For ductile steels (of > S,,) the Morrow line predicts less
sensitivity to mean stress.
4. For most fatigue design situations, R < 1 (i.e., small mean stress in
relation to alternating stress), there is little difference in the theories.
5. In the range where the theories show a large difference (i.e., R values
approaching 1), there is little experimental data. In this region the yield
criterion may set design limits.
6. The mean stress correction methods have been developed mainly for the
cases of tensile mean stress.
For finite-life calculations the endurance limit in any of the equations can be
replaced with a fully reversed alternating stress level corresponding to that
finite-life value!
39
Cyclic strain vs. cyclic stress
 Cyclic strain control complements cyclic stress
characterization: applicable to thermal fatigue, or
fixed displacement conditions.
 Cyclic stress-strain testing defined by a controlled
strain range, ∆pl.
 Soft, annealed metals tend to harden;
strengthened metals tend to soften.
 Thus, many materials tend towards a fixed cycle,
i.e. constant stress, strain amplitudes.

40
Cyclic stress-strain curve

[Courtney]

• Large number of cycles typically needed to reach


asymptotic hysteresis loop (~100).
• Softening or hardening possible.
41
Cyclic stress-strain
 Wavy-slip materials generally
reach asymptote in cyclic
stress-strain: planar slip
materials (e.g. brass) exhibit
history dependence.
 Cyclic stress-strain curve
defined by the extrema, i.e.
the “tips” of the hysteresis
loops. Cyclic stress-strain
curves tend to lie below those
for monotonic tensile tests.
 Polymers tend to soften in
cyclic straining.

42
Cyclic Strain Control
 Strain is a more logical independent variable for
characterization of fatigue.
 Define an elastic strain range as ∆el = ∆/E.
 Define a plastic strain range, ∆pl.
 Typically observe a change in slope between the
elastic and plastic regimes.
 Low cycle fatigue (small Nf) dominated by plastic
strain: high cycle fatigue (large Nf) dominated by
elastic strain.

43
Cyclic Strain control: low cycle
 Constitutive relation n
for cyclic stress-strain:   K  
 n’ ≈ 0.1-0.2
 Fatigue life: Coffin Manson relation:
 p
 
c
 f 2N f
2
 f ~ true fracture strain; close to tensile
ductility
 c ≈ -0.5 to -0.7
 c = -1/(1+5n’); large n’  longer life.

44
Cyclic Strain control: high cycle
 For elastic-dominated strains  e
at high cycles, adapt a  E  f 2N b
Basquin’s equation: 2
 Intercept on strain axis of extrapolated elastic line =
f/E.
 High cycle = elastic strain control:
slope (in elastic regime) = b = -n’/(1+5n’)
 The high cycle fatigue strength, f, scales with the
yield stress  high strength good in high-cycle

45
Low vs. High Cycle
>103 cycles, high cycle fatigue
car crank shaft – ~2.5 E8 Rev/105 miles
manufacturing equipment @ 100 rpm – 1.25 E8 Rev/year

<103 cycles, low cycle fatigue

ships, planes, vehicle chassis


46
Strain amplitude - cycles

[Courtney]

47
Total strain (plastic+elastic) life
 Low cycle = plastic control: slope = c
 Add the elastic and plastic strains.

  el  pl f
   
b c
   2N f  f 2N f
2 2 2 E
 Cross-over between elastic and plastic control is
typically at Nf = 103 cycles.
 Ductility useful for low-cycle; strength for high cycle
 Examples of Maraging steel for high cycle
endurance, annealed 4340 for low cycle fatigue
strength.

48
The Fatigue S-N method
(Nominal Stress Approach)

 The principles of the S-N approach (the nominal stress method)

 Fatigue damage accumulation

 Significance of geometry (notches) and stress analysis in fatigue


evaluations of engineering structures

 Fatigue life prediction in the design process

49
S-N Curves

 S-N [stress-number of cycles to failure] curve defines


locus of cycles-to-failure for given cyclic stress.
 Rotating-beam fatigue test is standard; also alternating
tension-compression.
 Plot stress versus the log (number of cycles to failure),
log(Nf).

 For frequencies < 200Hz, metals are insensitive to


frequency; fatigue life in polymers is frequency
dependent.

50
Testing Fatigue Properties

 Rotating Beam – most data is from this type


 Axial
• lower or higher? Why?
 Cantilever
 Torsion

51
Wöhler’s Fatigue Test

Note! In the case of smooth


components such as the
railway axle the nominal stress
and the local peak stress are
the same!
B A

S  peak

Smin

Smax

52
Fatigue testing, S-N curve
mean 3 > mean 2 > mean 1
a The greater the number of
mean 1 cycles in the loading history,
mean 2 the smaller the stress that
mean 3 the material can withstand
without failure.
log Nf

Note the presence of a


fatigue limit in many
steels and its absence
in aluminum alloys.

53
Infinite life

Part 1 Part 2 Part 3


Fatigue S-N curve
Su
Stress amplitude, Sa (ksi)

S103

Sy

Se

Number of cycles, N

Fully reversed axial S-N curve for AISI 4130 steel. Note the break at the LCF/HCF transition and
the endurance limit
Characteristic parameters of the S - N curve are:
Se - fatigue limit corresponding to N = 1 or 2106 cycles for
steels and N = 108 cycles for aluminum alloys,
S10 - fully reversed stress amplitude corresponding to N = 10
3 3S CN
a
m
 10  N m
A

cycles
m - slope of the high cycle regime curve (Part 2)
54
Fully Reversed Empirical Data
An S-N Curve

Wrought Steel

55
Fully Reversed Empirical Data

Aluminum
56
Most of available S - N fatigue data has been obtained from fully reversed rotational bending tests.
However, material behavior and the resultant S - N curves are different for different types of loading.
It concerns in particular the fatigue limit Se.
S103

Relative stress amplitude, Sa/Su 1.0


Se

0.5
Bending
Axial
0.3
Torsion

0.1
103 10 105 106 10
4 7
Number of cycles, Log(N)
The stress endurance limit, Se, of steels (at 106 cycles) and the fatigue strength, S103 corresponding
to 103 cycles for three types of loading can be approximated as (ref. 1, 23, 24):
S103 = 0.90Su and Se = S106 = 0.5 Su - bending
S103 = 0.75Su and Se = S106 = 0.35 - 0.45Su - axial

S103 = 0.72Su and Se = S106 = 0.29 Su - torsion


57
Endurance Limit
S e A stress level below which a material can be cycle
infinitely without failure
Many materials have an endurance limit:
low-strength carbon and alloy steels, some stainless steels, irons,
molybdenum alloys, titanium alloys, and some polymers

Many other materials DO NOT have an endurance limit:


aluminum, magnesium, copper, nickel alloys, some stainless steels,
high-strength carbon and alloy steels
for these, we use a FATIGUE STRENGTH defined for a certain
Sf number of cycles (5E8 is typical)

58
Endurance Limits
 Some materials exhibit endurance limits, i.e. a stress
below which the life is infinite:
• Steels typically show an endurance limit, = 40% of
yield; this is typically associated with the presence
of a solute (carbon, nitrogen) that pines
dislocations and prevents dislocation motion at
small displacements or strains (which is apparent
in an upper yield point).
• Aluminum alloys do not show endurance limits;
this is related to the absence of dislocation-pinning
solutes.
 At large Nf, the lifetime is dominated by nucleation.
• Therefore strengthening the surface (shot
peening) is beneficial to delay crack nucleation
and extend life.
59
Approximate endurance limit for various materials:
Magnesium alloys (at 108 cycles) Se = 0.35Su
Copper alloys (at 108 cycles) 0.25Su< Se <0.50Su
Nickel alloys (at 108 cycles) 0.35Su <Se < 0.50Su
Titanium alloys (at 107 cycles) 0.45Su <Se< 0.65Su
Al alloys (at 5x108 cycles) Se = 0.45Su (if Su ≤ 48 ksi) or Se = 19 ksi (if Su> 48 ksi)
Steels (at 106 cycles) Se = 0.5Su (if Su ≤ 200 ksi) or Se = 100 ksi (if Su>200 ksi)
Irons (at 106 cycles) Se = 0.4Su (if Su ≤ 60 ksi) or Se = 24 ksi (if Su> 60 ksi)

S – N curve
1 1 A 1
 
S a  C  N m  10 A  N m or N  C m
 Sa  m C m
 Sa  m

 S
  
2
1  103 
S  103 
m   log   and A  log 
3  Se  Se 
 
60
Fatigue Limit – Modifying Factors
For many years the emphasis of most fatigue testing was to gain
an empirical understanding of the effects of various factors on
the base-line S-N curves for ferrous alloys in the intermediate
to long life ranges. The variables investigated include:

- Rotational bending fatigue limit, Se’, Fatigue limit of a machine


part, Se
- Surface conditions, ka,
- Size, kb,

- Mode of loading, kc, Se = ka kb kc kd ke kf·Se’


- Temperature, kd
- Reliability factor, ke
- Miscellaneous effects (notch), kf

61
Surface Finish Effects on Fatigue Endurance Limit
The scratches, pits and machining marks on the surface of a material add stress concentrations to the
ones already present due to component geometry. The correction factor for surface finish is sometimes
presented on graphs that use a qualitative description of surface finish such as “polished” or “machined”.

Below a generalized empirical graph


is shown which can be used to
estimate the effect of surface finish
in comparison with mirror-polished
specimens [Shigley (23), Juvinal
ka

(24), Bannantine (1) and other


textbooks].

Effect of various surface finishes


on the fatigue limit of steel.
Shown are values of the ka, the
Ca

ratio of the fatigue limit to that


for polished specimens.

(from J. Bannantine, ref.1)

62
Size Effects on Endurance Limit
Fatigue is controlled by the weakest link of the material, with the probability of existence (or density) of a
weak link increasing with material volume. The size effect has been correlated with the thin layer of
surface material subjected to 95% or more of the maximum surface stress.
There are many empirical fits to the size effect data. A fairly conservative one is:

Se 1 if d  0.3 in 
kb   
Se  0.869d 0.097
'
if 0.3 in  d  10.0 in 
or
Se 1.0 if d  8 mm 
kb  '   
Se 1.189d 0.097 if 8  d  250 mm 
• The size effect is seen mainly at very long lives.
• The effect is small in diameters up to 2.0 in (even in bending and torsion).
Stress effects in non-circular cross section members
In the case of non-circular members the approach is based on so called effective diameter, de.
The effective diameter, de, for non-circular cross sections is obtained by equating the volume of material
stressed at and above 95% of the maximum stress to the same volume in the rotating-bending
specimen.
63
max
The effective diameter, de, for members
0.95max
with non-circular cross sections
+

The material volume subjected to stresses


  0.95max is concentrated in the ring of
0.05d/2 thick.

The surface area of such a ring is:


0.05d/2
 2
A0.95 max  d   0.95d    0.0766d 2
2
-
4 

* rectangular cross section under bending

A  Ft
A0.95 max  Ft  0.95t  0.05 Ft
0.95t
t

Equivalent diameter

0.0766d e2  0.05Ft
F d e  0.808 Ft
64
Loading Effects on Endurance Limit

The ratio of endurance limits for a material found using axial and rotating
bending tests ranges from 0.6 to 0.9.

Se ( axial )  (0.7  0.9) Se ( bending )


kc  0.7  0.9 ( suggested by Shigley kc  0.85)
The ratio of endurance limits found using torsion and rotating bending tests
ranges from 0.5 to 0.6. A theoretical value obtained from von Mises-Huber-
Hencky failure criterion is been used as the most popular estimate.

 e ( torsion )  0.577 Se ( bending )


kc  0.57( suggested by Shigley kc  0.59)

65
Temperature Effect

From: Shigley and Mischke, Mechanical Engineering Design, 2001

Su,T Su,T
Se ,T  Se , RT kd  Se , RT ; kd 
Su, RT Su, RT
66
Reliability factor ke
The reliability factor accounts for the scatter of reference data such
as the rotational bending fatigue limit Se’.
The estimation of the reliability factor is based on the assumption that
the scatter can be approximated by the normal statistical probability
density distribution.

ke  1  0.08  za
The values of parameter za associated with various levels of
reliability can be found in Table 7-7 in the textbook by Shigley et.al.

67
S-N curves for assigned probability of failure; P - S - N curves

68
Fatigue Parameters
 Recall Fatigue Testing (RR Moore Tester)
specimen compression on top

motor counter
flex coupling
 m   max   min  2
tension on bottom
 S   max   min  2
 Stress Varies with Time; max
Key Parameters m S
• m  Mean Stress (MPa) min time
• S  Stress Amplitude (MPa)
 Failure Even though  Cause of ~90% of
max < c Mechanical Failures
69
Fatigue Design Parameter
S = stress amplitude
 Fatigue (Endurance) case for
unsafe steel (typ.)
Limit, Se’ in MPa
• Unlimited Cycles if S S e’
safe
< Se’
103 105 107 109
N = Cycles to failure

 Some Materials will S = stress amplitude


case for
NOT permit unsafe Al (typ.)
Limitless Cycling
• i.e.; Se’ = ZERO safe

103 105 107 109


N = Cycles to failure
70
Fatigue Parameters
 Fatigue Cracks Grow INCREMENTALLY
during the TENSION part of the Cycle
 Math Model for Incremental Crack Extension

typ. 1 to 6
da
 
 K
m K I ~  a 
dN Opening-Mode (Mode-I) Stress Intensity Factor

increase in crack length per loading cycle

 Example: Austenitic Stainless Steel


da
dN
 12

m / cyc   5.6 10  K Mpa m 
3.25

71
Fatigue Crack Propagation

 Crack Length := a.
Number of cycles := N
Crack Growth Rate := da/dN
Amplitude of Stress Intensity := ∆K = ∆√c.
 Define three stages of crack growth, I, II and III,
in a plot of da/dN versus ∆K.
 Stage II crack growth: application of linear elastic fracture mechanics.
 Can consider the crack growth rate to be related to the applied stress
intensity.
 Crack growth rate somewhat insensitive to R (if R<0) in Stage II
 Environmental effects can be dramatic, e.g. H in Fe, in increasing
crack growth rates.

72
Fatigue Crack Propagation
 Three stages of crack da/dN
growth, I, II and III.
 Stage I: transition to a
finite crack growth rate I
from no propagation
below a threshold value II ∆Kc
of ∆K.
 Stage II: “power law” III
dependence of crack
growth rate on ∆K.
 Stage III: acceleration of
growth rate with ∆K,
approaching catastrophic
fracture.
73 ∆Kth ∆K
Paris Law
dc
 Paris Law:  A(K)m
dN
 m ~ 3 (steel); m ~ 4 (aluminum).
 Crack nucleation ignored!
 Threshold ~ Stage I
 The threshold represents an endurance limit.
 For ceramics, threshold is close to KIC.
 Crack growth rate increases with R (for R>0).

74
Striations- mechanism
 Striations occur by development of slip bands
in each cycle, followed by tip blunting,
followed by closure.
 Can integrate the growth rate to obtain cycles
as related to cyclic stress-strain behavior.

cf cf
dc dc
N II   N II  
c dc / dN
0 c0 A m
 c
m

75
Striations, contd.
 Provided that m>2 and is constant, can integrate.

A1  m 1m / 2  1m / 2 


N II 
(m / 2)  1

c0 cf 
 If the initial crack length is much less than the final
length, c0<cf, then approximate thus:

A1  m 1m / 2 
N II  c0
(m / 2)  1

 Can use this to predict fatigue life based on known


crack

76
Damage Tolerant Design
 Calculate expected growth rates from
dc/dN data.
 Perform NDE on all flight-critical
components.
 If crack is found, calculate the expected
life of the component.
 Replace, rebuild if too close to life limit.
 Endurance limits.

77
Geometrical effects
 Notches decrease fatigue life through stress
concentration.
 Increasing specimen size lowers fatigue life.
 Surface roughness lowers life, again through stress
concentration.
 Moderate compressive stress at the surface
increases life (shot peening); it is harder to nucleate a
crack when the local stress state opposes crack
opening.
 Corrosive environment lowers life; corrosion either
increases the rate at which material is removed from
the crack tip and/or it produces material on the crack
surfaces that forces the crack open (e.g. oxidation).
 Failure mechanisms
78
Stress concentration factor, Kt, and the
notch factor effect, kf

Fatigue notch factor effect kf depends on the stress


concentration factor Kt (geometry), scale and material
properties and it is expressed in terms of the Fatigue Notch
Factor Kf.

1
kf 
Kf

79
F
2 Stresses in axisymmetric
2 notched body
C 2
peak C

2
2
F
2
2
n n  S 
3 A
A, B 3

and
11 1

 peak  K t n
A B
3 D
3
2
D 3
2

11

3
3
F

80
Stresses in prismatic notched body

2 F
2 A,
B, C 2
F
n  S 
peak
C
A
2
2 and
D 2
2

11 n  peak  K t n


11

 Kt S
E
A
2 D B
3
1
E
2
3
3

3 F
3
81
Stress concentration factors used in fatigue analysis
S M

peak peak

Stress
Stress

n, S
 peak n
Kt 
r n r x
0 x  peak 0

dn S dn
W W

82
Stress concentration
factors, Kt, in shafts

S=

Bending load

S=

Axial load

83
S=

84
Similarities and differences between the stress field near the notch
and in a smooth specimen
S P

Stress peak

n=S
peak= n=S

r x
0

dn
W

85
The Notch Effect in Terms of the Nominal Stress

S
n1 N (S)m = C
Smax

S
n2 S2

n3 Sesmooth
Stress range

Sesmooth
n4
Senotched 
Kf

np N2 N0 cycles

Kf  Fatigue notch factor! K f  K t !!!


86
Definition of the fatigue notch factor Kf
 22   peak
S M
  esmooth
2 2
22= peak peak

Stress
Stress

Nsmooth
 notched
Se
2  enotched
e

1 1 1
3

Nnotched Nnotched

 esmooth
K f  notched for N smooth  N notched
e
87
PETERSON's approach

Kt  1
K f  1  1  q  K t  1
1 a r
1.8
1  300 
  10  in.
3
a
q ; a – constant,
 Su 
1 a r r – notch tip radius;
for S u in  in.

NEUBER’s approach

Kt  1
K f  1 ρ – constant,

1 r r r – notch tip radius

88
The Neuber constant ‘ρ’ for steels and aluminium alloys

89
Curves of notch sensitivity index ‘q’ versus notch radius
(McGraw Hill Book Co, from ref. 1)

90
Illustration of the notch/scale effect
Plate 1
W1 = 5.0 in

d1 = 0.5 in.
pea pea
k k Su = 100 ksi
m1
m2 Kt = 2.7
q = 0.97
Kf1 = 2.65

d2 Plate 2
d1 r W2= 0.5 in
r
d2 = 0.5 i
W1 W2 Su = 100 ksi
Kt = 2.7
q = 0.78
Kf1 = 2.32

91
Procedures for construction of approximate fully reversed S-N curves for
smooth and notched components

f ’
Collins method

Juvinal/Shigley method

Nf (logartmic)

Sar, ar – nominal/local


stress amplitude at zero
mean stress m=0 (fully
reversed cycle)!
Nf (logartmic)

92
Procedures for construction of approximate fully reversed S-N curves for
smooth and notched components

0.9Su Manson method


Sar (logartmic)

Se’kakckbkdke

Se’kakckbkdkekf
Se

100 101 102 103 104 105 106 2·106


Nf (logartmic)

Sar, ar – nominal/local stress amplitude at zero mean stress m=0


(fully reversed cycle)!

93
NOTE!

• The empirical relationships concerning the S –N curve data are


only estimates! Depending on the acceptable level of uncertainty
in the fatigue design, actual test data may be necessary.

• The most useful concept of the S - N method is the endurance


limit, which is used in “infinite-life”, or “safe stress” design
philosophy.

• In general, the S – N approach should not be used to estimate


lives below 1000 cycles (N < 1000).

94
Microstructure-Fatigue Relationships
 What are the important issues in microstructure-fatigue
relationships?
 Answer: three major factors.
1: geometry of the specimen (previous slide); anything on the
surface that is a site of stress concentration will promote
crack formation (shorten the time required for nucleation of
cracks).
2: defects in the material; anything inside the material that can
reduce the stress and/or strain required to nucleate a crack
(shorten the time required for nucleation of cracks).
3: dislocation slip characteristics; if dislocation glide is confined
to particular slip planes (called planar slip) then dislocations
can pile up at any grain boundary or phase boundary. The
head of the pile-up is a stress concentration which can
initiate a crack.

95
Microstructure affects Crack Nucleation

 The main effect of da/dN


microstructure (defects,
surface treatment, etc.) is
almost all in the low stress I
intensity regime, i.e. Stage I.
Defects, for example, make II ∆Kc
it easier to nucleate a crack,
which translates into a lower III
threshold for crack
propagation (∆Kth).
 Microstructure also affects
fracture toughness and
therefore Stage III.

96 ∆Kth ∆K
Defects in Materials
 Descriptions of defects in materials at the sophomore level
focuses, appropriately on intrinsic defects (vacancies,
dislocations). For the materials engineer, however, defects
include extrinsic defects such as voids, inclusions, grain
boundary films, and other types of undesirable second phases.
 Voids are introduced either by gas evolution in solidification or
by incomplete sintering in powder consolidation.
 Inclusions are second phases entrained in a material during
solidification. In metals, inclusions are generally oxides from the
surface of the metal melt, or a slag.
 Grain boundary films are common in ceramics as glassy films
from impurities.
 In aluminum alloys, there is a hierachy of names for second
phase particles; inclusions are unwanted oxides (e.g. Al2O3);
dispersoids are intermetallic particles that, once precipitated, are
thermodynamically stable (e.g. AlFeSi compounds); precipitates
are intermetallic particles that can be dissolved or precipiated
depending on temperature (e.g. AlCu compounds).
97
Metallurgical Control: fine particles
 Tendency to localization of flow is deleterious to the initiation of
fatigue cracks, e.g. Al-7050 with non-shearable vs. shearable
precipitates (Stage I in a da/dN plot). Also Al-Cu-Mg with
shearable precipitates but non-shearable dispersions, vs. only
shearable ppts.

98
Coarse particle effect on fatigue
 Inclusions nucleate cracks cleanliness (w.r.t. coarse particles)
improves fatigue life, e.g.
 7475 improved by lower Fe+Si compared to 7075:
0.12Fe in 7475, compared to 0.5Fe in 7075;
0.1Si in 7475, compared to 0.4Si in 7075.

99
Alloy steel heat treatment
 Increasing hardness tends to raise the endurance limit for high
cycle fatigue. This is largely a function of the resistance to
fatigue crack formation (Stage I in a plot of da/dN).

Mobile solutes that pin


dislocations fatigue
limit, e.g. carbon in steel

100
Casting porosity affects fatigue

Gravity cast
versus
squeeze cast
versus
wrought
Al-7010

 Casting tends to result in porosity. Pores are effective sites


for nucleation of fatigue cracks. Castings thus tend to have
lower fatigue resistance (as measured by S-N curves) than
wrought materials.
 Casting technologies, such as squeeze casting, that reduce
porosity tend to eliminate this difference.
101
Titanium alloys

 For many Ti alloys, the proportion of hcp (alpha) and bcc (beta) phases depends
strongly on the heat treatment. Cooling from the two-phase region results in a two-
phase structure
 Rapid cooling from above the transformation in the single phase (beta) region
results in a two-phase microstructure with Widmanstätten laths of (martensitic)
alpha in a beta matrix.
 The fatigue properties of the two-phase structure are significantly better than the
Widmanstätten structure (more resistance to fatigue crack formation).
 The alloy in this example is IM834, Ti-5.5Al-4Sn-4Zr-0.3Mo-1Nb-0.35Si-0.6C.
102
Design Considerations
 If crack growth rates are normalized by the elastic
modulus, then material dependence is mostly
removed!
 Can distinguish between intrinsic fatigue for
combined elastic, plastic strain range for small crack
sizes and extrinsic fatigue and for crack growth rate
controlled at longer crack lengths.
 Inspection of design charts, shows that ceramics
sensitive to crack propagation (high endurance limit
in relation to fatigue threshold).

103
Design Considerations: 2

 Metals show a higher fatigue threshold in relation


to their endurance limit. PMMA and Mg are at
the lower end of the toughness range in their
class.
 Also interesting to compare fracture toughness
with fatigue threshold.
 Note that ceramics are almost on ratio=1 line,
whereas metals tend to lie well below, i.e. fatigue
is more significant criterion.

104
Fatigue property map

105
Fatigue property map

106
*Variable Stress/Strain Histories
 When the stress/strain history is stochastically
varying, a rule for combining portions of fatigue life is
needed.
 Palmgren-Miner Rule is useful: ni is the number of
cycles at each stress level, and Nfi is the failure point
for that stress.

ni
 1
i N fi

107
*Fatigue in Polymers
 Many differences from metals
 Cyclic stress-strain behavior often exhibits
softening; also affected by visco-elastic effects;
crazing in the tensile portion produces
asymmetries.
 S-N curves exhibit three regions, with steeply
decreasing region II.
 Nearness to Tg results in strong temperature
sensitivity.

108
Fatigue: summary
 Critical to practical use of structural
materials.
 Fatigue affects most structural
components, even apparently statically
loaded ones.
 Well characterized empirically.
 Connection between dislocation
behavior and fatigue life offers exciting
research opportunities, i.e. physically
based models are lacking!
109
Improving Fatigue Performance
S = stress amplitude

1. Impose a
Compressive
near compressive, m
zero ortensile,
moderate
Surface Stress (to
larger tensile,
m m
Suppress Surface
N = Cycles to failure
cracks from growing)
• Method 1: shot peening • Method 2: carburizing (interstitial)
shot
C-rich gas
put
surface
into
compression

2. Remove Stress bad better


Concentrating sharp
corners bad better

110
S-N Data for 2014-T6 Al

19.5 ksi

111
Procedure for Fatigue Damage Calculation
n1 1
Ni(i)m = a
n2 2
Stress range, 

n3 3
e
n4

'e

N1 N2 N3 No cycles
NT
n1 cycles applied at  1 n2 cycles applied at  2 ni cycles applied at  i
D  ....  ...
N 1 cycles to failure at  1 N 2 cycles to failure at  2 N i cycles to failure at  i
i
n1 n2 ni n
D  Dn1  Dn 2  .....  Dni    ...   i
N1 N 2 Ni 1 Ni

1 1
LR  
D n1 N1  n2 N 2  .... ni N i
112
n1 - number of cycles of stress range 1
n2 - number of cycles of stress range 2
ni - number of cycles of stress range i,

1
D1  - damage induced by one cycle of stress range 1,
N1
n1
Dn1  - damage induced by n1 cycles of stress range 1,
N1
1
D2  - damage induced by one cycle of stress range 2,
N2
n2
Dn 2  - damage induced by n2 cycles of stress range 2,
N2
1
Di  - damage induced by one cycle of stress range i,
Ni

ni
Dni  - damage induced by ni cycles of stress range i,
Ni
113
Total Damage Induced by the Stress History

n1 cycles applied at  1 n cycles applied at  2 n cycles applied at  i


D  2 ....  ... i
N 1 cycles to failure at  1 N 2 cycles to failure at  2 N i cycles to failure at  i

i
n1 n2 ni n
D  Dn1  Dn 2  .....  Dni    ...   i
N1 N 2 Ni 1 Ni

It is usually assumed that fatigue failure occurs when the cumulative damage
exceeds some critical value such as D =1,
i.e. if D>1 - fatigue failure occurs!
For D < 1 we can determine the remaining fatigue life:

1 1
LR   LR - number of repetitions of
D n1 N 1  n2 N 2  .... ni N i the stress history to failure

N  LR  n1  n2  n3  .....  ni  N - total number of cycles to failure

114
 
m
N j = a  j if j > e .
It is assumed that stress cycles lower than the fatigue limit, j < e, produce no damage (Nj=) in
the case of constant amplitude loading however in the case of variable amplitude loading the
extension of the S-N curve with the slope ‘m+2” is recommended. The total damage produced by
the entire stress spectrum is equal to:
j
D   Dj
j 1

It is assumed that the component fails if the damage is equal to or exceeds unity, i.e. when D  1.
This may happen after a certain number of repetitions, BL (blocks), of the stress spectrum, which
can be calculated as:

BL = 1/D.
Hence, the fatigue life of a component in cycles can be calculated as:

N = BLNT,
where, NT is the spectrum volume or the number of cycles extracted from given stress history.

NT = (NOP - 1)/2
If the record time of the stress history or the stress spectrum is equal to Tr, the fatigue life can be
expressed in working hours as:

T = BL Tr.
115
Main Steps in the S-N Fatigue Life Estimation Procedure
 Analysis of external forces acting on the structure and the component
in question,
 Analysis of internal loads in chosen cross section of a component,
 Selection of individual notched component in the structure,
 Selection (from ready made family of S-N curves) or construction of S-
N curve adequate for given notched element (corrected for all effects),
 Identification of the stress parameter used for the determination of the
S-N curve (nominal/reference stress),
 Determination of analogous stress parameter for the actual element in
the structure, as described above,
 Identification of appropriate stress history,
 Extraction of stress cycles (rainflow counting) from the stress history,
 Calculation of fatigue damage,
 Fatigue damage summation (Miner- Palmgren hypothesis),
 Determination of fatigue life in terms of number of stress history
repetitions, Nblck, (No. of blocks) or the number of cycles to failure, N.
 The procedure has to be repeated several times if multiple stress
concentrations or critical locations are found in a component or
structure.
116
Example #2
An unnotched machine component undergoes a variable amplitude
stress history Si given below. The component is made from a steel
with the ultimate strength Suts=150 ksi, the endurance limit
Se=60 ksi and the fully reversed stress amplitude at N1000=1000
cycles given as S1000=110 ksi.
Determine the expected fatigue life of the component.

Data: Kt=1, SY=100 ksi Suts=150 ksi, Se=60 ksi, S1000=110 ksi

The stress history:

Si = 0, 20, -10, 50, 10, 60, 30, 100, -70, -20, -60, -40, -80, 70, -30,
20, -10, 90, -40, 10, -30, -10, -70, -40, -90, 80, -20, 10, -20, 10, 0

Stress History
117
Si = 0, 20, -10, 50, 10, 60, 30, 100, -70, -20, -60, -40, -80, 70, -30,
20, -10, 90, -40, 10, -30, -10, -70, -40, -90, 80, -20, 10, -20, 10, 0

Stress History

100
80
60
40
20
Stress (ksi)

0
-20 0 5 10 15 20 25 30 35
-40
-60
-80
-100

Reversal point No.

118
N S a   C
m

log N  m log S a  log C

log1000  m log110  log C



log10 6
 m log 60  log C

3  m log110  log C

6  m log 60  log C

C  1.886  1026 m  11.4


119
S-N Curve

1000
Stress amplitude (ksi)

100

10
1000 10000 100000 1000000 10000000
No. of cycles

120
Goodman Diagram

SY
100
90
Stress amplitude (ksi)

80
70
60
50
40
30
20
10 Suts
0
-100 -50 0 50 100 150
Mean stress (ksi)

1
Sa Sm  S 
 1 for fatigue endurance S a ,r  at Sm 0  1  m  S a
Se Suts  Suts 

Sa S Sa
 m  1 for any stress amplitude S a ,r  at Sm 0 
S a ,r Suts S
1 m
Suts
121
Calculations of Fatigue Damage

a) Cycle No.11
Sa,r=87.93 ksi
N11= C( Sa,N)-m= 1.866102687.93-11.4=12805 cycles
D11=0.000078093

b) Cycle No. 14
Sa,r=75.0 ksi
N14= C( Sa,N)-m= 1.866102675.0-11.4=78561 cycles
D14=0.000012729

c) Cycle no. 15
Sa,r=98.28 ksi
N14= C( Sa,N)-m= 1.866102698.28-11.4=3606 cycles
D14=0.00027732

122
Results of "rainflow" counting Damage
No. S Sm Sa Sa,r (Sm=0) Di=1/Ni=1/C*Sa-m
1 30 5 15 15.52 2.0155E-13 0
2 40 30 20 25.00 4.6303E-11 0
3 30 45 15 21.43 7.9875E-12 0
4 20 -50 10 10.00 1.3461E-15 0
5 50 -45 25 25.00 4.6303E-11 0
6 30 5 15 15.52 2.0155E-13 0
7 100 20 50 57.69 6.3949E-07 0
8 20 -20 10 10.00 1.3461E-15 0
9 30 -25 15 15.00 1.3694E-13 0
10 30 -55 15 15.00 1.3694E-13 0
11 170 5 85 87.93 7.8039E-05 7.80E-05
12 30 -5 15 15.00 1.3694E-13 0
13 30 -5 15 15.00 1.3694E-13 0
14 140 10 70 75.00 1.2729E-05 1.27E-05
15 190 5 95 98.28 0.00027732 0.000277
n0=15 D= 0.00036873 3.677E-04

D=0.000370 D=3.677E-04
LR = 1/D =2712.03 L R = 1/D =2719.61
N=n0*LR=15*2712.03=40680 N=n0*LR=15*2719.61=40794

123

You might also like