Professional Documents
Culture Documents
Reliability cost
Reliability (R)
0% 100 %
= Optimum reliability
What is Reliability ?
Unreliability =
(t)
Failure function F(t)
F(t)
1.0
(t)
Reliability Terms
• Mean Time To Failure (MTTF) for non-repairable systems
• Mean Time Between Failures for repairable systems (MTBF)
• Reliability Probability (survival function), R(t)
• Failure Probability (cumulative density function ), F(t)=1-R(t)
• Failure Density Function, f(t)
• Failure Rate Function(hazard rate), λ(t)
• Mean time to repair (MTTR)
• Mean down time (MDT)
• Mean failure rate (MFR)
Relationship among f(t), F(t), R(t) and
Let, N = Total number of components
As the time progresses, the components fail
-------- (1)
Now, the failure rate function or instantaneous failure rate is defined as,
------ (3)
------- (4)
------- (5)
Now, integrating equation (5) over specified time period, we get
------- (6)
------- (7)
------- (8)
Therefore, ------- (9)
R(t) ------
F(t) ------
f(t) ------
------
Important Failure Distributions (Models)
Very straightforward and widely used; Very powerful and can be applied to
Applications: Electronics, mechanical describe various failure processes;
components etc. Applications: Electronics, mechanical
components, material, structure etc.
Exponential failure distribution
Graphs related to exponential failure distribution
Exponential failure distribution Cont.
Statistical Properties:
Problem:
The failure distribution of a component follows exponential law with mean time to
failure (MTTF) equal to 1500 hrs. Determine the followings:
a) The reliability that the component will function satisfactorily till 900 hrs.
b) The probability that the component will fail before 1250 hrs.
c) The time period so that the component will function with 90 % reliability.
d) If 2000 of such components are started at a time how many of them are likely to
survive till 900 hrs and how many of them are likely to fail till 1250 hrs.
Solution:
MTTF = 1500 hrs
Time 0 - 100 100 - 200 200 - 300 300 - 400 400 - 500 500 - 600 600 - 700
interval
(days)
Observed 85 56 30 18 12 6 3
frequency
80
60
40
20
i = Number of intervals
= Expected frequency
Expected frequency =
Now
Therefore, accept null hypothesis. So, the data fits well into the assumed distribution
Acceptance region
Rejection region
α
Problem:
Consider the following frequency distribution of failures for a component:
Time 0 - 100 100 - 200 200 - 300 300 - 400 400 - 500 500 - 600 600 - 700 700 -800
interval
(hrs)
Observed 8 24 40 78 70 42 20 10
frequency
80
60
40
20
= Shape parameter
= Characteristic life
For =0
Different functions for Weibull distribution
Statistical properties for Weibull failure Distribution
Location parameter =
Mean life =
Median life =
Significance of Weibull failure parameters
Or
Useful period
Early Life
Region/Infant Wear-Out
Mortality Region
Period/
Burn-in period
0 Time t
Significance of shape parameter
Definition and significance of Weibull failure parameters
Let, then
Estimation of Weibull Failure Parameters
Two methods: (1) Analytical method and (2) Graphical method
(1) Analytical method:
(Method of Least Squared Error)
------- (1)
------- (2)
------- (3)
Where
Method of Least Squared Error:
------- (4)
On simplification, we get
------- (8)
------- (9)
480, 700, 300, 900, 850, 1150, 600, 500, 400, 770
For =0
Where,
------- (2)
(a)
(b)
(c)
Example: Q-Q of Weibull Distribution and
Weibull Fit (works well)
• T~Weibull(1, 4000) Generate 50 data points
Weibull Probability Plot
0.99
0.96
0.90
0.75 1-e-1
0.50 0.632
P ro b a b ility
0.25
0.10
b
0.05
0.02
0.01
10
-5
10
0
h 5
10
Data
Example: Q-Q Weibull Distribution and T-
Normal Fit (obviously wrong functional form)
• T~Normal(100, 400) Generate 50 data points
Weibull Probability Plot
0.99
0.96
0.90
0.75
0.50
P ro b a b ility
0.25
2
10
Data
Graphical Model Validation
• Use a Q-Q Plot for checking normality
• Plot Probability Quantiles (ln(ln(1/median rank)/ vs. Model
Distribution Quantiles
• The result should be linear in ln(time) in the middle of the
plot
http://www.qualitydigest.com/jan99/html/body_weibull.html
Weibull Using MS Excel
CCL1 ln(ln(1/(1-
hours to 1/(1-Median Median ln(Design A
failure Rank Median Ranks Rank) Rank))) Cycles)
0.75 1 0.067307692 1.072164948 -2.66384309 -0.28768207
0.9 2 0.163461538 1.195402299 -1.72326315 -0.10536052
20.3 3 0.259615385 1.350649351 -1.20202312 3.01062089
73.4 4 0.355769231 1.552238806 -0.82166652 4.29592394
91.8 5 0.451923077 1.824561404 -0.50859539 4.5196123
97.2 6 0.548076923 2.212765957 -0.23036544 4.57677071
578.9 7 0.644230769 2.810810811 0.032924962 6.36112975 1.5
ln(Design A Cycles) Line Fit Plot
609.2 8 0.740384615 3.851851852 0.299032932 6.41214662 1
912.2 9 0.836538462 6.117647059 0.593977217 6.81585926 0.5
2115 10 0.932692308 14.85714286 0.992688929 7.65681009 0
SUMMARY OUTPUT -0.5
-1
Regression Statistics -1.5
Multiple R 0.969525286
R Square 0.93997928 -2
Adjusted R Square
Standard Error
0.93247669
0.289744238 -2.5
Observations 10
-3
-1 0 1 2 3 4 5 6 7 8 9
ANOVA ln(Design A Cycles)
df SS MS F Significance F
Regression 1 10.51808512 10.51808512 125.2873043 3.63718E-06
Residual 8 0.671613788 0.083951723
Total 9 11.18969891
Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0%
Intercept -2.217586473 0.176953141 -12.53205486 1.53932E-06 -2.625641148 -1.809531798 -2.625641148 -1.809531798
ln(Design A Cycles) 0.391732899 0.034997459 11.19318115 3.63718E-06 0.311028614 0.472437184 0.311028614 0.472437184
Regression Statistics
Multiple R 0.931357
ln(Design A Cycles) Line
R Square
Adjusted R Square
Standard Error
0.867426
0.834283
0.319986
Fit Plot
Observations 6
0
8.5 9 9.5 10 10.5 11
ANOVA -0.5
Significanc
df SS MS F eF -1
ln(ln(1/(1-Median Rank)))
Regression 1 2.679769 2.679769 26.17191 0.006906
Residual 4 0.409564 0.102391
Total 5 3.089333 -1.5
-2 ln(ln(1/(1-
Coefficient Standard Lower Upper Lower Upper Median
s Error t Stat P-value 95% 95% 95.0% 95.0% -2.5
Intercept -12.9954 1.889141 -6.87899 0.00234 -18.2405 -7.75029 -18.2405 -7.75029 Rank)))
ln(Design A Cycles) 0.968949 0.189401 5.115849 0.006906 0.443086 1.494812 0.443086 1.494812
Beta (shape Parameter) 0.968949
-3
Alpha 667861.2 77.65828 Predicted
Hrs Years -3.5 ln(ln(1/(1-
RESIDUAL OUTPUT Beta=0.97 -4
Median
Rank)))
Predicted
Alpha=78 Yrs. -4.5
ln(ln(1/(1-
Median
Observation Rank))) Residuals -5
1 -4.34514 -0.34292
2 -4.2235 0.42938 ln(Design A Cycles)
3 -3.03485 -0.28994
4 -2.99881 -0.00412
5 -2.82923 0.072387
6 -2.69221 0.13521
Life test data
Life test data can broadly be divided into different categories as follows:
Exact time failure data: The exact time of failure of each component is given
Interval data: The failure data is given in the form of frequency distribution
Suspended data: When test is terminated for some components due to some
unknown reasons although the testing time has not been reached
Censored data: There may be some components which did not fail till the end
of the experiment
Problem:
Consider the following frequency distribution of failures for a component :
1000-2000 1500 16 21 13
2000-3000 2500 24 45 33
4000-5000 4500 14 88 81
5000-6000 5500 8 96 92
For =0
Where,
------- (2)
(a)
(b)
(c)