Documente Academic
Documente Profesional
Documente Cultură
Probabilistic Fatigue
1 of 108
Probabilistic Models
Probabilistic models are no better than the
underlying deterministic models
They require more work to implement
Why use them?
Probabilistic Fatigue
2 of 108
Risk / Reliability
What is the increased risk from reduced testing ?
Probabilistic Fatigue
3 of 108
Risk
10-3
10-4
Risk
10-5
Acceptable risk
10-6
10-7
10-8
10-9
Probabilistic Fatigue
4 of 108
Reliability
99 %
Expected Failures
80 %
50 %
103
10 %
104
105
106
Fatigue Life
1%
0.1 %
Probabilistic Fatigue
5 of 108
Speed
Material
Properties
Operating
Temperature
Probabilistic Fatigue
6 of 108
Stress
usage
Failures
100 %
50%
Strain Amplitude
time
0.1
10-2
10-3
10-4
1 10
102
103
104
105
106
107
materials
Probabilistic Fatigue
Strength
2003-2014 Darrell Socie, All Rights Reserved
manufacturing
7 of 108
Probabilistic Fatigue
8 of 108
Deterministic Design
Safety
Factor
Stress
Strength
Probabilistic Fatigue
9 of 108
Probabilistic Design
Stress
Strength
Probabilistic Fatigue
10 of 108
3 Approach
3 contains 99.87% of the data
P( s < S ) = 2.3 10-3
If we use 3 on both stress and strength
P( failure)
P(
S ) 5.3 10
4 .5
1.2 10
5 .7
11 of 108
Benefits
Reduces conservatism (cost) compared to
assuming the worst case for every design
variable
Quantifies life drivers what are the most
important variables and how well are they
known or controlled ?
Quantifies risk
Probabilistic Fatigue
12 of 108
Reliability Analysis
0.4
0.3
P( Failure )
0.2
0.1
-3
-2
-1
Analysis
Probability Density
Strength Variables
0.8
Probability Density
Probability Density
Stressing Variables
0.8
0.6
0.4
0.2
1
0.6
0.4
0.2
1
Probabilistic Fatigue
13 of 108
Analytical
First order reliability method FORM
Second order reliability method SORM
Probabilistic Fatigue
14 of 108
Statistical Techniques
Normal Distributions
LogNormal Distributions
Monte Carlo
Distribution Fitting
Probabilistic Fatigue
15 of 108
Failure Probability
Stress,
Strength, S
s
Let
P(
Probabilistic Fatigue
S)
16 of 108
Normal Variables
Linear Response Function
n
Z ao
ai Xi
i 1
Xi ~ N(
, Ci )
n
z
ao
ai
i 1
n
ai
2
i
i 1
Probabilistic Fatigue
17 of 108
Calculations
Stress,
Strength, S
S ~ N( 200 , 0.1 )
~ N( 100 , 0.2 )
= 20
z
= 20
z
Safety factor of 2
Probabilistic Fatigue
20 2 20 2
28.2
18 of 108
Failure Probability
Z=SFailure will occur whenever Z <= 0
Z=
z
Z
Z
=0
100
28.2
19 of 108
Failure Distribution
Stress,
Strength, S
Probabilistic Fatigue
20 of 108
Fatigue Data
Stress Amplitude
1000
'
f
100
b
1
10
1
1
10
102
103
104
105
106
107
Fatigue Life
'
f
Probabilistic Fatigue
S
2 (2Nf )b
2Nf
S
2
1
b
'
f
21 of 108
LogNormal Variables
n
Z ao
Xi
ai
i 1
median Z
ao
Xi
ai
i 1
COV C Z
1 C Xi
2
2 ai
i 1
Probabilistic Fatigue
22 of 108
Calculations
Stress,
Strength, S
2Nf
'
f
S
~ LN( 250 , 0.2 )
2
S
2
Z 2Nf
= 50
b = -0.125
S
2
Z 2Nf
COVZ
Probabilistic Fatigue
1 COV
'
f
= 100
2
S
1
b
82
8
'8
f
8
'8
f
1 COV
2
2 8
f
23 of 108
Results
S/2
'
f
2Nf
Percentile
Life
250
1000
355,368
99.9
17,706,069
COVx
0.2
0.1
4.72
5.50
245
50
6.90
995
100
11.21
73,676
1,676,831
99
95
90
50
10
4,566,613
1,363,200
715,589
73,676
7,586
lnx
0.198
0.100
1.774
b=
-0.125
5
1
0.1
3,982
1,189
307
lnx
X
x
Probabilistic Fatigue
24 of 108
'2
f
2Nf
2b
' '
f f
2Nf
'
f
b c
and
'
f
Z = 2Nf = ?
Probabilistic Fatigue
25 of 108
Simple Example
Probability of rolling a 3 on a die
1/6
fx(x)
1
Probabilistic Fatigue
26 of 108
Computer Simulation
1. Generate n random numbers between 1 and 6,
all integers
2. Count the number of 3s
Let Xi = 1 if 3
0 otherwise
P3
Probabilistic Fatigue
1 n
Xi
ni1
27 of 108
EXCEL
=ROUNDUP( 6 * RAND() , 0 )
=IF( A1 = 3 , 1 , 0 )
=SUM($B$1:B1)/ROW(B1)
Probabilistic Fatigue
5
3
4
4
5
6
1
3
3
6
0
1
0
0
0
0
0
1
1
0
0
0.5
0.333333
0.25
0.2
0.166667
0.142857
0.25
0.333333
0.3
28 of 108
Results
probability
0.4
0.3
0.2
0.1
100
200
300
400
500
trials
Probabilistic Fatigue
29 of 108
Evaluate
y
P( inside circle )
-1
r2
4
=4P
-1
x = 2 * RAND() - 1
y = 2 * RAND() - 1
IF( x2 + y2 < 1 , 1 , 0 )
Probabilistic Fatigue
30 of 108
4
3
2
1
0
1
100
200
300
400
500
trials
Probabilistic Fatigue
31 of 108
Strength, S
2Nf
'
f
S
~ LN( 250 , 0.2 )
2
b = -0.125
S
2
1
b
'
f
= 100
= 50
Probabilistic Fatigue
32 of 108
Generating Distributions
Fx(X)
1
RAND
0
x
Randomly choose a value between 0 and 1
x = Fx-1( RAND )
Probabilistic Fatigue
33 of 108
Probabilistic Fatigue
34 of 108
EXCEL
Probabilistic Fatigue
'
f
S
2
893
1102
852
963
1050
1080
965
1073
1052
954
965
993
1191
831
204
301
285
173
283
265
313
213
226
322
240
207
368
210
2Nf
134,677
32,180
6,355
929,249
35,565
77,057
8,227
420,456
224,000
5,878
68,671
277,192
11,967
59,473
35 of 108
Simulation Results
99.9 %
Cumulative Probability
99 %
90 %
50 %
103
104
105
106
107
108
Life
10 %
1%
0.1 %
Probabilistic Fatigue
36 of 108
Summary
Simulation is relatively straightforward and simple
Obtaining the necessary input data and distributions
is difficult
Probabilistic Fatigue
37 of 108
99.9 %
Cumulative Probability
99 %
Median 431
COV 0.34
90% confidence COV 0.41
90 %
50 %
100
200
500
1000
10 %
1%
0.1 %
Probabilistic Fatigue
38 of 108
0.004
LogNormal
0.003
0.002
0.001
200
400
600
800
1000
1200
Maximum Load
Uncertainty in Variance is just as important,
perhaps more important than the choice of the distribution
Probabilistic Fatigue
39 of 108
Probability Density
0.003
LogNormal
Weibull
Gumble
0.002
Normal
0.001
200
400
600
800
1000
Maximum Load
Probabilistic Fatigue
40 of 108
Distributions
Normal
Strength
Dimensions
LogNormal
Fatigue Lives
Large variance in properties or loads
Gumble
Maximums in a population
Weibull
Fatigue Lives
Probabilistic Fatigue
41 of 108
as n
Probabilistic Fatigue
/ n
42 of 108
Translation
When there are many variables affecting the outcome,
The final result will be normally distributed even if the
individual variable distributions are not.
Probabilistic Fatigue
43 of 108
Example
Probability of rolling a die
1/6
fx(x)
1
44 of 108
Results
Frequency
500 trials
60
XZ = 21.12
50
CZ = 0.20
40
30
20
10
0
6
12
18
24
30
36
45 of 108
X2
X3
X4
.. Xn
Normal as n increases
Products: Z = X1 X2 X3 X4 .. Xn
Z
LogNormal as n increases
46 of 108
Key Points
All variables are random and can be characterized
by a statistical distribution with a mean and variance.
The final result will be normally distributed even if
the individual variable distributions are not.
Probabilistic Fatigue
47 of 108
Sources of Variability
Stress
customers
Stress,
usage
Strength, S
Strain Amplitude
1
0.1
10-2
10-3
10-4
1 10
102
103
104
105
106
107
materials
Probabilistic Fatigue
Strength
2003-2014 Darrell Socie, All Rights Reserved
manufacturing
48 of 108
Probabilistic Fatigue
49 of 108
Sources of Variability
Stress Variables
Loading
Customer Usage
Environment
Strength Variables
Material
Processing
Manufacturing Tolerance
Environment
Probabilistic Fatigue
50 of 108
Sources of Uncertainty
Statistical Uncertainty
Incomplete data (small sample sizes)
Modeling Error
Analysis assumptions
Human Error
Calculation errors
Judgment errors
Probabilistic Fatigue
51 of 108
Modeling Variability
Central Limit Theorem:
Products: Z = X1 X2 X3 X4 .. Xn
Z
LogNormal as n increases
lnX
1.5
1.0
lnX
~ COVX
0.5
0.5
1.0
COVX
1.5
2.0
52 of 108
68.3%
95.4%
99.7%
0.05
1.05
1.11
1.16
0.1
1.10
1.23
1.33
0.25
1.28
1.66
2.04
0.5
1.60
2.64
3.92
2.30
5.53
11.1
COVX
Probabilistic Fatigue
53 of 108
Probabilistic Fatigue
54 of 108
Maximum Force
Maximum force from 42 automobile drivers
99.9 %
Cumulative Probability
99 %
Median 431
COV 0.34
90 %
50 %
100
200
500
1000
Force
10 %
1%
0.1 %
Probabilistic Fatigue
55 of 108
Fatigue Lives
107
106
105
104
103
0
200
400
600
800
1000
Maximum Load
Probabilistic Fatigue
56 of 108
Loading Variability
54 Tractors / Drivers
Load Range
48
32
16
10
102
103
104
105
Cumulative Cycles
Probabilistic Fatigue
57 of 108
Variability in Loading
Cumulative Probability
99.9 %
99 %
54 Tractors/Drivers
COV 0.53
90 %
50 %
1
10
Equivalent Load
10 %
S eq
1%
Sn
0.1 %
Probabilistic Fatigue
58 of 108
Structures
100
n=5
10
Crack Nucleation
104
n = 10
103
1
10
100
100
101
102
103 104
Cycles
105
Crack Growth
106
107
100
n=3
1
103
104
105
106
107
Total Fatigue Life, Cycles
108
10
K,MPa m
Equivalent Load, kN
Probabilistic Fatigue
1
10-7 10-8 10-9 10-10 10-11 10-12
Crack Growth Rate, m/cycle
59 of 108
Cumulative Probability
99 %
n
3
5
10
COV
0.53
0.43
0.38
90 %
50 %
0.1
10
Equivalent Load
10 %
1%
0.1 %
Probabilistic Fatigue
60 of 108
Probabilistic Fatigue
61 of 108
Cumulative Probability
99.9 %
COV 0.12
99 %
90 %
50 %
0.1
10
Equivalent Load
10 %
1%
0.1 %
Probabilistic Fatigue
62 of 108
Cumulative Probability
99.9 %
COV 0.32
99 %
90 %
50 %
0.1
10
Equivalent Load
10 %
1%
0.1 %
Probabilistic Fatigue
63 of 108
Variability in Environment
Inclusions
Pit depth
Probabilistic Fatigue
64 of 108
Barter, S. A., Sharp, P. K., Holden, G. & Clark, G. Initiation and early growth of fatigue cracks in an aerospace
aluminium alloy, Fatigue & Fracture of Engineering Materials & Structures 25 (2), 111-125.
Probabilistic Fatigue
65 of 108
Crawford et.al.The EIFS Distribution for Anodized and Pre-corroded 7010-T7651 under Constant Amplitude Loading
Fatigue and Fracture of Engineering Materials and Structures, Vol. 28, No. 9 2005, 795-808
Probabilistic Fatigue
66 of 108
Frequency
40
Mean = 230
COV = 0.32
30
20
10
100
200
300
area
Probabilistic Fatigue
400
500
67 of 108
Cumulative Probability
99 %
90 %
50 %
10
Pit Depth, m
100
10 %
1%
0.1 %
68 of 108
Variability in Materials
Tensile Strength
Fracture Toughness
Fatigue
Fatigue Strength
Fatigue Life
Strain-Life
Crack Growth
Probabilistic Fatigue
69 of 108
Number of heats
100
75
50
25
500
550
600
650
700
70 of 108
Fracture Toughness
Cumulative Probability
99.9 %
99 %
90 %
50 %
60
10 %
70
80
90
100
KIc, Ksi in
1%
0.1 %
Kies, J.A., Smith, H.L., Romine, H.E. and Bernstein, H, Fracture Testing of Weldments, ASTM STP 381, 1965, 328-356
Probabilistic Fatigue
71 of 108
Fatigue Variability
Fatigue life
Fatigue strength
Stress Amplitude
1000
100
10
1
1
10
102
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
72 of 108
= 123,000 cycles
COV = 0.25
10
Number of Tests
40
= 134,000 cycles
COV = 0.27
x
20
10
50
100
150
Life x 10
200
250
73 of 108
Percent Survival
98
s=280
s=315
s=245
s=210
90
50
10
2
104
105
106
107
108
Cycles to Failure
Sinclair and Dolan, Effect of Stress Amplitude on the Variability in Fatigue Life of 7075-T6 Aluminum Alloy
Transactions ASME, 1953
Probabilistic Fatigue
74 of 108
Probabilistic Fatigue
COV
440
315
280
245
210
14,000
25,000
220,000
1,200,000
12,000,000
0.12
0.38
0.70
0.67
1.39
75 of 108
S'f ( Nf )b b
n
COV C
1 C Xi
0.085
2 ai
i 1
CS'
1 1.39
2
2 ( .085 )
0.088
Probabilistic Fatigue
76 of 108
Strain Amplitude
29 Data Sets
10-2
10-3
10-4
1
10
102
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
77 of 108
Median 820
COV 0.25
99 %
90 %
50 %
10 %
300
1000 2000
99.9 %
Cumulative Probability
Cumulative Probability
99.9 %
99 %
90 %
50 %
-0.12
-0.08
-0.04
10 %
1%
1%
0.1 %
0.1 %
Probabilistic Fatigue
Mean -0.09
COV 0.25
78 of 108
Median 0.57
COV 1.15
99.9 %
99 %
90 %
50 %
0.1
10 %
10
Cumulative Probability
Cumulative Probability
99.9 %
99 %
90 %
50 %
-0.8
-0.6
-0.4
-0.2
10 %
1%
1%
0.1 %
0.1 %
Probabilistic Fatigue
Mean -0.62
COV 0.23
79 of 108
Probabilistic Fatigue
L,
2N f
b N,
b, b
'
f
L,
2Nf
c N,
b, b
80 of 108
Simulation Results
Strain Amplitude
100
10
1
0.1
0.01
0.001
0.0001
10
100
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
81 of 108
Correlation
0.2
1.0
0.15
0.75
0.1
0.25
0.05
0
0.5
102
103
104
0.01
'
f
= -0.828
Probabilistic Fatigue
0.1
10
'
f
= -0.976
82 of 108
( rand() )
z2 =
( rand() )
z3
'
f
z1
exp(
b
Probabilistic Fatigue
z1= N(0,1)
z2 1
ln
'
f
ln
'
f
z1 )
z3
83 of 108
Correlated Properties
Strain Amplitude
100
10
1
0.1
0.01
0.001
0.0001
10
100
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
84 of 108
Curve Fitting
1
'
f
Strain Amplitude
'
f
0.1
(2Nf )b
'
f
(2Nf )c
10-2
10-3
10-4
1
10
102
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
85 of 108
Property Distribution
'
f
99.9 %
99 %
COV 0.12
b = -0.086
90 %
50 %
100
10 %
1000
Cumulative Probability
99.9 %
Cumulative Probability
'
f
99 %
Median 0.26
COV 0.42
90 %
c = -0.51
50 %
0.1
10 %
1%
1%
0.1 %
0.1 %
Probabilistic Fatigue
86 of 108
Correlation
1500
1000
'
f
500
0
0
Probabilistic Fatigue
0.1
0.2
0.3
'
f
0.4
0.5
0.6
0.7
87 of 108
Simulation
10
Strain Amplitude
0.1
0.01
0.001
0.0001
1
10
100
103
104
105
106
107
Fatigue Life
Probabilistic Fatigue
88 of 108
Strength Coefficient
Cumulative Probability
99.9 %
99 %
1500
90 %
1000
50 %
500
10 %
5000
K'
500
1%
'
f
COV 0.14
500
1000
1500
2000
2500
K'
0.1 %
= 0.863
Probabilistic Fatigue
89 of 108
Crack Length, mm
50
40
30
20
10
0
0
50
100
150
200
Cycles x103
250
300
350
Virkler, Hillberry and Goel, The Statistical Nature of Fatigue Crack Propagation, Journal of Engineering Materials
and Technology, Vol. 101, 1979, 148-153
Probabilistic Fatigue
90 of 108
Cumulative Probability
99.9 %
99 %
90 %
50 %
300,000
10 %
68 Data Points
1%
0.1 %
Median 257,000
COV 0.07
Fatigue Lives
Probabilistic Fatigue
10-2
10-3
10-4
10-5
5
10
50
Stress intensity, MPa m
91 of 108
Cumulative Probability
99.9 %
99 %
C Km
Median 5 x 10-8
COV 0.44
99.9 %
90 %
50 %
10-7
10 %
10-6
Cumulative Probability
da
dN
99 %
90 %
50 %
1%
0.1 %
0.1 %
Probabilistic Fatigue
10 %
1%
Median 3.13
COV 0.06
92 of 108
Simulated Data
Cumulative Probability
99.9 %
99 %
data
simulation
90 %
50 %
105
106
107
10 %
1%
0.1 %
Probabilistic Fatigue
93 of 108
af
1 m/2
C Sm
a
m
2
1 m/2
i
1 m/2
0.07
COVC
0.44
Probabilistic Fatigue
94 of 108
Correlation
4.0
3.5
= -0.99
3.0
2.5
2.0 -8
10
10-7
10-6
C
Probabilistic Fatigue
95 of 108
Calculated Lives
Cumulative Probability
99.9 %
Computed from
C and m pairs
99 %
experimental
90 %
50 %
105
106
af
10 %
Nf
ao
da
C
1%
0.1 %
Probabilistic Fatigue
96 of 108
Manufacturing/Processing Variability
Bolt Forces
Surface Finish
Drilled Holes
Probabilistic Fatigue
97 of 108
Cumulative Probability
99 %
90 %
50 %
Force
200 Data Points
Median 130
COV 0.14
100
1000
Bolt Force, kN
10 %
1%
0.1 %
98 of 108
Cumulative Probability
99 %
90 %
Surface Finish
125 Data Points
Median 43
COV 0.10
50 %
10
Surface Finish, in
100
10 %
1%
0.1 %
Probabilistic Fatigue
99 of 108
Drilled Holes
Fighter Spectrum
Cumulative Probability
99.9 %
99 %
Median 126,750
90 %
50 %
105
Cycles
10 %
180 drilled holes in a single plate
1%
0.1 %
From: J.P. Butler and D.A. Rees, "Development of Statistical Fatigue Failure Characteristics of 0.125-inch
2024-T3 Aluminum Under Simulated Flight-by-Flight Loading," ADA-002310 (NTIS no.), July 1974.
Probabilistic Fatigue
100 of 108
Analysis Uncertainty
Miners Linear Damage rule
Strain Life Analysis
Probabilistic Fatigue
101 of 108
Miners Rule
99.9
964 Tests
99
COV = 1.02
90
50
0.01
0.1
10
100
10
1
0.01
102 of 108
SAE Specimen
Bracket
Suspension
Transmission
Fatigue Under Complex Loading: Analysis and Experiments, SAE AE6, 1977
Probabilistic Fatigue
103 of 108
Analysis Results
Strain-Life analysis of all test data
99.9 %
Cumulative Probability
99 %
48 Data Points
COV 1.27
90 %
50 %
1
10
100
Probabilistic Fatigue
104 of 108
Material Variability
Material
Analysis
99.9 %
Cumulative Probability
99 %
90 %
50 %
1
10
100
105 of 108
Modeling Uncertainty
Analysis Uncertainty CU = ?
The variability in reproducing the original strain life data
from the material constants is CM ~ 0.44
n
COV C
1 C Xi
2
2 ai
i 1
1 CU
1 CNf
1 CM
CU = 1.09
90% of the time the analysis is within a factor of 3 !
99% of the time the analysis is within a factor of 10 !
Probabilistic Fatigue
106 of 108
COV C
1 C Xi
2
2 ai
i 1
107 of 108
COV C
1 C Xi
2
2 ai
i 1
108 of 108