Reliability Engineering

Why Design for Reliability?

• Reliability can make or break the long-term

success of a product:
– Too high reliability will cause the product to be too
expensive
– Too low reliability will cause warranty and repair
costs to be high and therefore market share will
be lost
 Optimum reliability
Failure cost
Cost
Total cost

Reliability cost

Reliability (R)
0% 100 %

= Optimum reliability
 What is Reliability ?

• A product has 95 % reliability ?

• Reliability is:
• Elimination/avoidance of failure modes/mistakes
• The probability that a product will perform its intended function:
• Under customer operating conditions
• For a specified life
• In a manner that meets or exceeds customer expectations
• A reliable product is robust and mistake-free
Definition of Reliability:
Reliability of a component or a system is defined as the probability that
the component or the system will perform its intended function
adequately for a specified period of time under stated operating
conditions.

Reliability is a time dependent characteristics

Reliability is evaluated based on the failure data of the component or

the system
Mathematical expression of Reliability:

n = Number of tube lights = Time to failure of ith tube light

= a random variable

= Specified time period

Tube light is not reliable

Tube light is reliable

Time to failure (T)

Reliability =

Unreliability =

= Failure density function

= 90 % = 90 components survived till specified

time period out of 100 components.
R(t)

(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

Again, let, = Number of components survived till time period ‘t’

= Number of components failed before time period ‘t’

-------- (1)

Time period till all N components fail

 Hence, Reliability = -------- (2)

Now, the failure rate function or instantaneous failure rate is defined as,

------ (3)

Differentiating equation (2) with respect to ‘t’

------- (4)

From equations (3) and (4), we get

------- (5)
 Now, integrating equation (5) over specified time period, we get

------- (6)

Again, from equation (1), we get

------- (7)

Differentiating equation (6) with respect to ‘t’

------- (8)
 Therefore, ------- (9)

Relationship matrix among f(t), F(t), R(t) and

R(t) F(t) f(t)

R(t) ------

F(t) ------

f(t) ------

------
 Important Failure Distributions (Models)
• Exponential Distribution
Very commonly used, even in cases to
which it does not apply (simple);
Applications: Electronics, mechanical
components etc.
• Normal Distribution
Very straightforward and widely used;
Applications: Electronics, mechanical
components etc.
• Lognormal Distribution
Very powerful and can be applied to
describe various failure processes;
Applications: Electronics, material,
structure etc.
• Weibull Distribution
Very powerful and can be applied to
describe various failure processes;
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

(a) 0.5488 (b) 56.54 %

(c) 158.04 hrs (d) 1098, 1131

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
interval
(days)
Observed 85 56 30 18 12 6 3
frequency

(a) Fit the above data into a suitable failure distribution

(b) Calculate the reliability for a specified time period of 365 days.
Solution:
(a) Draw histogram to know the approximate pattern of distribution and then assume
a suitable distribution to represent the failure law. Then perform testing of
hypothesis using Chi-square test (test of goodness of fit)

80

60

40

20

0 100 200 300 400 500 600 700

So, assume exponential failure law to fit the data.

i = Number of intervals

= Mid-value of ith interval

= Frequency of ith interval

: The data fits into the assumed distribution

: The data does not fit into the assumed distribution

 = Obsereved frequency

= Expected frequency

If > ; Reject null hypothesis

Expected frequency =

N = Total number of observations = 210

∆t = Time interval = 100
Time 0 - 100 100 - 200 200 - 300 300 - 400 400 - 500 500 - 600 600 - 700
interval
(days)
Mid-value 50 150 250 350 450 550 650
Observed 85 56 30 18 12 6 3
frequency

Expected 89.56 50.8 28.87 16.4 9.3 5.3 3

frequency =90 =51 =29 =16 =9 =6
As per Yates’s correction the last two frequency has been merged ( since frequency < 5)

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

(a) Fit the above data into a suitable failure distribution.

(b) Calculate the reliability for a specified time period of 425 hrs.
(c) Determine specified time period for a required reliability of 95 %.
(d) Calculate probability of failure and failure rate for a specified time period of 350 hrs.
Solution:
(a) Draw histogram to know the approximate pattern of distribution and then assume
a suitable distribution to represent the failure law. Then perform testing of
hypothesis using Chi-square test (test of goodness of fit)

80

60

40

20

0 100 200 300 400 500 600 700 800

Steps to be followed:
1) Draw histogram of the failure data. It seems to be normal distribution. So, assume
normal distribution.
2) Estimate mean and SD from the data (parameter of the distribution)
3) Calculate expected frequency for all intervals
4) Estimate Chi-square value for the given data
5) Compare the estimated value with table value
6) Decide whether to accept null hypothesis or alternative hypothesis

(b) Calculate R(t=425) = P(t≥425), using normal distribution

(c) Calculate T’ such that P(t≥T’) = 0.95

where T’ is the required specified time period
 Weibull failure Distribution
The three parameter Weibull distribution can be written as,

= Location parameter/Threshold period/guaranteed life

= Shape parameter

= Characteristic life

The two parameter Weibull distribution can be written as,

For =0
Different functions for Weibull distribution
Statistical properties for Weibull failure Distribution

Location parameter =

Mean life =

Median life =
 Significance of Weibull failure parameters

Shape parameter (β):

Shape parameter indicates the shape of various functions like failure density function,
failure distribution function, reliability function and failure rate functions which are
used for reliability analysis.
For, Decreasing failure rate (hyper-exponential distribution)

For, Constant failure rate (exponential distribution)

For, Increasing failure rate (normal or other distribution)

 Bath Tub Curve
Failure Rate Function:

Constant Failure Rate Region

Failure Rate

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)

Transforming the above equation into a linear equation, we get

------- (2)

------- (3)

Where
Method of Least Squared Error:

Let, the observed values are represented by the following equation,

------- (4)

Sum of Squared Error (SSE) = ------- (5)

= ith observed value of dependent variable Y

= ith observed value of independent variable X

= Error component for ith observation of dependent variable Y

= Estimated value of constant A

= Estimated value of constant β

 Now, differentiating equating (5), we get

------- (6) ------- (7)

On simplification, we get

------- (8)

------- (9)

Karl-Pearson’s correlation coefficient (r) is given by, ------- (10)

Now, solving (8) and (9), the values of shape parameter and
characteristic life can be found out.
For three parameter Weibull distribution as given below,

The Variable X is replaced by

Problem:

Consider a two parameter Weibull distribution and estimate the values

of Weibull failure parameters by method of least squared error. The
time to failures for 10 components are as follows (hrs):

480, 700, 300, 900, 850, 1150, 600, 500, 400, 770

Also determine the followings:

(a)Reliability of the component for specified time period of 825 hrs.
(b) Probability of failure between 400 hrs and 900 hrs
(c)Failure for specified time period of 850 hrs
 Solution:
The two parameter Weibull distribution can be written as,

For =0

Where,

F(t) can be found out using an empirical formula given below

Median Rank = Where, i = Order number

N = Sample size
Order No Time to
(i) failure (t)
1 300 5.704 0.0673 -2.664

2 400 5.991 0.1635 -1.723

3 480 6.174 0.2596 -1.202

4 500 6.215 0.3558 -0.821

5 600 6.397 0.4519 -0.509

6 700 6.551 0.5481 -0.230

7 770 6.646 0.6442 0.033

8 850 6.745 0.7403 0.299

9 900 6.802 0.8365 0.594

10 1150 7.047 0.9327 0.993

 ------- (1)

------- (2)

Solving the above two equations, we get,

(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

0.10 Weibull model

0.05 is invalid
0.02
0.01

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

• Estimate at for ti using Bernard’s Formula

For n observed failure time data

 Weibull in Excel

http://www.qualitydigest.com/jan99/html/body_weibull.html
 Weibull Using MS Excel

1) Order by number of cycles (in accelerators, hours to failure)

2) Calculate the “Median Rank”=((B2-0.3)/(10+0.4))
3) Be sure that the Analysis ToolPak Add-In is loaded into Excel.
While on the page you just created, from the menu bar, select
Tools and Data Analysis. Scroll down and highlight "Regression"
and click OK. A data-entry window will pop up.
2.Under "Input Y Range," type: $E$1:$E$11.
3.For "Input X Range," type: $F$1:$F$11.
4.Click to add a checkmark in the box for "Labels."
5.For "Output Options," select "New Worksheet Ply."
6.Click to add a checkmark in the box for "Line Fit Plots."
7.Click OK. Excel will perform the regression and place the output on a new worksheet.
Note the:
Shape Parameter β ˃1 Terminal Mortality
Characteristic Lifetime, where 1/e = 62.3%
have failed.
 ln ( ln ( 1 /( 1 - M e d ia n R a n k ) ) )
SNS RF High Voltage Converter Modulator 2008

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

Beta =0.39 (Infant Failures)

Beta (or Shape Parameter) = 0.391732899
Alpha (or Characteristic Life) = 287.4260525

RESIDUAL OUTPUT Alpha = 287

Observation Predicted ln(ln(1/(1-Median Rank))) Residuals
Adjusted R square =0.93
1 -2.330281006 -0.33356208
2 -2.258859654 0.535596503
3 -1.038227226 -0.16379589
4
5
6
-0.534731736
-0.447105645
-0.424714814
-0.286934779
-0.061489749
0.194349369
The λ for an Exponential model = 475!!
7 0.274277326 -0.241352364
8 0.294262312 0.00477062
9 0.452409836 0.141567381
10 0.781837942 0.210850988
 VME Crate Power Supplies (2009)
VME hours to 1/(1-Median ln(ln(1/(1-Median ln(Design A
failure Rank Median Ranks Rank) Rank))) Cycles)
7536 1 0.009162304 1.009247028 -4.688058902 8.927446816
8544 2 0.022251309 1.022757697 -3.794124242 9.052984561
29136 3 0.035340314 1.036635007 -3.324794914 10.2797298
30240 4 0.048429319 1.050894085 -3.002931896 10.31692083
36024 5 0.061518325 1.065550907 -2.756842175 10.49194066
41496 6 0.07460733 1.080622348 -2.556998447 10.63335232
SUMMARY OUTPUT

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:

Life test data

Complete failure data Incomplete failure data

Exact time Suspended Censored

Interval data
failure data data data

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 :

Time 0 -1000 1000 - 2000 - 3000 - 4000 - 5000 - 6000 -

interval 2000 3000 4000 5000 6000 7000
(hrs)
Frequenc 5 16 24 29 14 8 4
y of
failure
(f)

(a) Consider two parameter Weibull distribution to estimate the failure

parameters
(b) Calculate the reliability and failure rate of the component for a specified time
period of
2800 hrs
(c) Calculate the failure probability between the time intervals 2500 hrs and
4800 hrs.
Solution:
The order number (i) is calculated by the following equation and other steps
remain same.
F = Cumulative frequency upto that interval
Order number of ith interval =
f = Frequency of that interval
Time Mid- f F i F(t) X = ln(t) Y XY
intervals value
(t)
0-1000 500 5 5 2.5 0.022 6.215

1000-2000 1500 16 21 13

2000-3000 2500 24 45 33

3000-4000 3500 29 74 59.5

4000-5000 4500 14 88 81

5000-6000 5500 8 96 92

6000-7000 6500 4 100 98

 Solution:
The two parameter Weibull distribution can be written as,

For =0

Where,

F(t) can be found out using an empirical formula given below

Median Rank = Where, i = Order number

N = Sample size
 ------- (1)

------- (2)

Solving the above two equations, we get,

(a)

(b)

(c)