Characterizing Reliability Data

Using the 3-Parameter

Weibull Distribution
Daniel Sillivant
[email protected]
PhD Student
Reliability and Failure Analysis Lab
 Objective
• Fit Reliability Data to the 3-Parameter Weibull
– Calculate Reliability Functions
– Plot Reliability Functions

• Why?
 Background
• Most failure and all repair models have a
region from 0 – γ in which no sample data
exists.
• Many distributions begin at 0.
– Exponential, lognormal, & 2-Parameter Weibull
– Normal distribution have negative values for data
that can only be positive.
• The 3-Parameter Weibull addresses this issue.
 3-Parameter Weibull

t
1
t
f ( t) e

• β (Beta) – Shape Parameter

• η (Eta) – Scale Parameter
• γ (Gamma) - Location Parameter
 Misconceptions

Probability Density Function

Distribution Shape (β) Scale (η)
2-Parameter
8.46 23.86
Weibull
0.1

Distribution Shape (β) Scale (η) Location (γ)

0.05 3-Parameter
8.46 23.86 10
Weibull

0
0 10 20 30

10
 Misconceptions
Probability Density Function
0. 2
2P Weibull
3P Weibull
0. 15

0. 1

0. 05

0
0 10 20 30 40 50

Distribution Shape Scale Threshold

2-Parameter Weibull 8.46 23.86 0

3-Parameter Weibull 8.46 23.86 10

 Correctly Fit
Probability Density Function

2P Weibull
3P Weibull
0. 1

Error in Fit
0. 05

0
0 10 20 30

Distribution Shape Scale Threshold

2-Parameter
8.46 23.86 0
Weibull
3-Parameter
2.211 6.834 16.557
Weibull
 Why 3-Parameter Weibull?
0. 15
2P Weibull
6
3P Weibull
0. 1 Histogram
4

0. 05
2

0 0
0 10 20 30
 Why 3-Parameter Weibull?
Reliability Function Hazard Function
1 1
2P Weibull
3P Weibull 0.1

0.01
3
0.5 1 10
4
1 10
5
2P Weibull
1 10
3P Weibull
6
0 1 10
0 20 40 0 10 20 30
 Why 3-Parameter Weibull?
Reliability of Parts in System
Single Part System Reliability
2-Parameter Weibull 0.9991 100 0.9139
3-Parameter Weibull 0.9995 100 0.9512

• System Requirement: Reliability of 0.95

• Parts are in series
 Tools
• Excel – Common Data analysis tool
• Minitab – Statistical software package
• Mathcad – Mathematical program
• Relex – Reliability Modeling program
• Reliasoft – Reliability Modeling program
 Excel Example
Gamma Rsquared
Guess 12 94.2586%

Gamma R Squared Data Y X = ln(TTF-γ)

0 91.1925% 18.0000 -3.2497 1.791759469
1 91.3363% 20.0000 -2.3336 2.079441542 R Squared vs Gamma
2 91.4921% 20.0000 -1.8408 2.079441542 97.00%
3 91.6615% 20.0000 -1.4939 2.079441542
96.00%
4 91.8463% 21.0000 -1.2209 2.197224577
5 92.0487% 21.0000 -0.9922 2.197224577 95.00%
6 92.2713% 21.0000 -0.7924 2.197224577 94.00%
7 92.5172% 21.0000 -0.6123 2.197224577 93.00%
8 92.7902% 22.0000 -0.4459 2.302585093
92.00%
9 93.0949% 22.0000 -0.2890 2.302585093
10 93.4368% 23.0000 -0.1380 2.397895273 91.00%
11 93.8222% 23.0000 0.0102 2.397895273 90.00%
12 94.2586% 23.0000 0.1586 2.397895273 0 5 10 15
13 94.7533% 24.0000 0.3110 2.48490665 Gamma Values
14 95.3097% 24.0000 0.4725 2.48490665
15 95.9133% 25.0000 0.6519 2.564949357
16 96.4709% 27.0000 0.8678 2.708050201
17 96.4292% 31.0000 1.1845 2.944438979
 Example 1
Probability Plot of C3
Shape Scale Threshold r Minitab Solution
3-Parameter Weibull - 95% CI
99

Minitab 1.11 145.90 2285.00 90

80
70
Mathcad 1.34 187.65 2262.91 94.69% 60
50
40
Relex 1.50 180.19 2262.82 94.65% 30

Percent
20

10
Probability Plot of C3
Relex Weibull
3-Parameter Solution
- 95% CI
Shape
Scale
1.109
145.9
5
99 Thresh 2285
3 N 10
90 2 AD 0.691
P-Value 0.077
80
70 1
60 1 10 100 1000
50
40 C3 - Threshold
30
Percent

20

10
Shape 1.5
5 Scale 180.2
Thresh 2263
3 N 10
2 AD 0.661
P-Value >0.250
1
10 100 1000
C3 - Threshold
 Example 2
Shape Scale Threshold r
Minitab 6398 1183489 -1180822
Probability Density Function
3
Mathcad 10.27 2672.77 3.65E-10 97.48% 2 10
Relex 1
Relex 23.27 5489.07 -2820.07 97.56%
Mathcad
3
Relex (2) 10.84 2665.85 0.00 97.50% 1.5 10 Relex 2

3
1 10

Relex Weibull package 5 10

4

allows the user to assign

fixed values for the
Weibull distribution. 0
3 3 3 3 3
1.5 10 2 10 2.5 10 3 10 3.5 10
Example 2
 Caution
• Results from Software packages
– Not always applicable
– Check For Sanity
 Tools
• Excel – Long Iterative Process. Solution is not exact.
• Minitab – Good Statistical software package. Minitab’s solution for
fitting a 3-Parameter Weibull is suspect.
• Mathcad – Statistical tools are lacking. The built-in 2-Parameter
Weibull function is not well defined and does not solve for the
parameters. I wrote a program to solve for the 3-Parameter
Weibull.
• Relex – Expensive Reliability software package which produces valid
results and most of the graphs for reliability functions, can give
gamma values that are suspect. Can fix by checking one box.
• Reliasoft – Expensive Reliability software package Reliability
modeling tool. Will fit data to the 3-Parameter Weibull. Do not
have 1st hand experience with the software.
Questions

Daniel Sillivant
[email protected]
http://rfal.uah.edu/