Reliability Analysis Methods

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 1

Lecture 1

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 2

1
 1. RELIABILITY DEFINITIONS

➢ Note that the variable "time" implicitly represents the agent of

failure.
➢ So, R(t) = Probability of success

R t = Pr(T > t) (i)

where
T = time to failure
t = mission time
➢ Reliability being a probability, it must lie between the range
0 ≤ R(t) ≤ 1

➢ Therefore unreliability = F t = 1 − R(t) (ii)

F(t) denotes the probability that the item will fail sometime up to time t.
Time-to-failure CDF which known as the unreliability function.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 3

RELIABILITY DEFINITIONS (cont.)

➢ Let there be N identical units that are subjected to life

tests. As time goes by, each unit will function over a
certain time and then fail. We plot the frequency of time
to failure (of non-repairable units), we get a continuous
time to failure distribution.

f(t) = probability density function (pdf) / failure density function

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 4

2
 RELIABILITY DEFINITIONS (cont.)

➢ Mean-Time-To-Failure (MTTF)
▪ In a life test of replaceable units all the sample units would
not fail at the same time. As such we get a distribution of
time-to-failure. The expected time-to-failure is called MTTF.
Mathematically,
N
1
MTTF = ෍ t i = E(t)
N
Where, i=0

N → sample size in a life test,

t i → time to failure of ith unit.
▪ More formally, for a continuous or discrete distribution expect time-to-
failure E(t), i.e., MTTF
▪ MTTF is associated with nonrepairable components.
▪ MTBF is related to the repairable components.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 5

RELIABILITY DEFINITIONS (cont.)

R(t)
1

R t = Pr(T > t)

time

F(t)
1

F t = Pr(T  t)

time

t R t =1−F t
dF(t) ∞
F t = න f x dx and = f(t)
0 dt R t = න f x dx
t

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 6

3
 RELIABILITY DEFINITIONS (cont.)

∞
➢ According to MTTF = ‫׬‬0 t f t dt another form in terms of R(t) can
be proved.
∞
If lim t 𝑓 t = 0 , then, integrating by parts, 𝐌𝐓𝐓𝐅 = ‫𝐭𝐝 𝐭 𝐑 𝟎׬‬
t→∞

➢ Conditional reliability function, is the probability that the item will

survive for time τ, given that it has survived for time t.

➢ Thus, the conditional probability of failure during the same interval

is

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 7

Failure Rate ( = )

➢ Failure Rate ()
▪ A reliability index that represents the rate at which your products
fails.

➢ Mean Time to Failure (MTTF , )

▪ The reliability index for non-repairable units represents the main
time to failure. 
➢ Mean Time Between Failure (MTBF , )
▪ The reliability index for repairable units represents the main time
between failure. 

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 8

4
 Failure Rate (cont.)

➢ Failure Rate and Hazard Rate Defined

▪ Conditional probability of failure, h(t), between a time interval
(t1 , t 2 ) under the condition that it has survived up to t1 (t1 < t 2 )
is
R t1 − R(t 2 )
Pr T < t 2 T > t1 =
R(t1 )
, (t2 = )
R t1 − R(t 2 )
therefore, R(t1 )
Failure hazard rate =
t 2 − t1

f(t)
we can say (t 2 = t1 + ∆t). Hazard Rate h t =
R(t)

H(x) = Cumulative Hazard

t
R t = e− ‫׬‬0 h x dx
= e−H(t)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 9

Failure Rate (cont.)

➢ Hazard rate is an important function in reliability analysis since it

shows changes in the probability of failure over the lifetime of a
component.
➢ The residual MTTF or simply mean residual life (MRL):
➢ MRL is the expected value associated with the conditional pdf of the
component time to failure.
➢ It is expected remaining life, T-t, given that the product, or a system
has survived to time t.
∞
1
Mean Residual Life, MRL = න R t dt
R(T1 ) T1
∞
If T1 = 0 → MTTF = න R t dt
T1
➢ In practice, h(t) often exhibits a bathtub shape and is referred to as a
bathtub curve.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 10

10

5
 Failure Rate (Example)

20 units are put on test and run at their normal operating condition for
1,000 hours. If 6 of those units fail, at the following hours (550, 480,
680, 790, 860, 620), what is the failure rate of the product?

Solution:

• Now thing about the above result, which is a negligible number, but the
question is it a good number or bad number.
• According to customer and to understand the meaning of this result, need
to find MTTF / MTBF (which is just an inverse relationship of ).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 11

11

MTTF / MTBF () (Example)

20 units are put on test and run at their normal operating condition for
1,000 hours. If 6 of those units fail, at the following hours (550, 480,
680, 790, 860, 620), what is the mean time to failure of the product?

Solution:

• Now if the customer needs a motor to run 2000 hours without failure, thus
the above product seems to meet the reliability needed.
• But if the customer needs at least 10,000 hours, then for sure it will not
meet the reliability needed.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 12

12

6
 BATHTUB CURVE

➢ Bathtub curve shows failure over various regions of life.

➢ It is a function describing the variation of hazard rate with time (life
of a unit).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 13

13

BATHTUB CURVE (cont.)

➢ Bathtub curve can be divided into three regions:

1) Burn-in failure region exhibits a decreasing failure rate (DFR),
characterized by early failures attributable to defects in design,
manufacturing, or construction.
• It is representative of the population and not individual units as most
components do not experience the early failure characteristic.
2) Chance-failure region exhibits a reasonably constant failure
rate, characterized by random failures of the component.
3) Wear out failure, which exhibits an increasing failure rate (IFR),
is characterized mainly by complex aging phenomena.
• The component deteriorates (e.g., due to accumulated fatigue) and
is more vulnerable to outside shocks.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 14

14

7
 2. COMMON DISTRIBUTIONS IN COMPONENT
RELIABILITY

➢ When defining times-to-failure as a particular distribution, a system

or component’s reliability and MTTF may be represented by such
distributions as,
▪ Exponential
▪ Weibull
▪ Gamma
▪ Normal and
▪ Lognormal

➢ In fact, all the previously mentioned distributions have their own form
of reliability and MTTF. Each one has its own merits depending on
the form of the distribution.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 15

15

THREE DISTINCT HAZARD RATES

➢ Case I - Constant hazard rate () - (chance-failure region)

That is, h t = λ
t t t
and 𝐑 𝒕 = e− ‫׬‬0 h x dx
= e− ‫׬‬0 λdx = e−λ ‫׬‬0 dx = 𝒆−𝝀𝒕

And 𝐟 𝒕 = h t ∙ R t = 𝛌𝒆−𝝀𝒕
(exponential distribution)

So, the random failure region can be modeled by an exponential

distributions model (often the case for electronic components and
mechanical components).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 16

16

8
 Constant failure rate () (Example)

We have test 20 units and found that our MTBF is 2,996 hours. What
is the reliability of the product at 1,200 hours of operation.

The probability that the product

will perform successfully past
the 1,200 hours is approximately
67%.

Alternative interpretation: 67% of the population of units can be

expected to surpass the 1,200 hours mark.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 17

17

THREE DISTINCT HAZARD RATES (cont.)

➢ Case II - Increasing failure rate

▪ For example, linearly increasing

h t = βt ( β is some constant )
t t t 𝟏 𝟐
Therefore, 𝐑 𝒕 = e− ‫׬‬0 h u du
= e− ‫׬‬0 βudu = e−β ‫׬‬0 udx = 𝒆−𝟐𝛃𝒕
1 2 𝟏 𝟐
and 𝐟 𝒕 = R t ∙ h t = e−2βt × βt = 𝛃𝐭𝐞−𝟐𝛃𝐭
▪ This is the Rayleigh distribution. The wear out region in Bathtub
curve can be modeled by this distribution. Other distributions may
also be used, depending on failure data.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 18

18

9
 THREE DISTINCT HAZARD RATES (cont.)

➢ Case III - Decreasing failure rate

▪ This case is similar to Case II, only the hazard rate is a
decreasing function of time.
𝐡 𝐭 = −𝛃𝐭
▪ Some characteristics of exponential distribution.
∞ ∞
𝟏
(i) 𝐌𝐓𝐓𝐅 = E t = න t ∙ f t dt = න t ∙ λe−λt dt = hours
0 0 𝛌
𝟏
Therefore, 𝐡 𝒕 = 𝛌 = (for exponential distribution)
𝑴𝑻𝑻𝑭
1
(ii) Mean life for exponential distribution =
λ

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 19

19

THREE DISTINCT HAZARD RATES (cont.)

Example:
Consider λ = 10−3 𝑝𝑒𝑟 hr, calculate R(t = 1000 hrs) for an exponentially
distributed time-to-failure.
−3 ×103
R t = e−λt = e−10 = e−1 = 0.368

Therefore,

F t = 1 − 0.368 = 0.632

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 20

20

10
 THREE DISTINCT HAZARD RATES (cont.)

(iii) exponential distribution in reliability has no memory.

Reliability at time 𝐑 𝐭 = 𝐞−𝛌𝐭

Therefore, unreliability 𝐅 𝐭 = 𝟏 − 𝐞−𝛌𝐭

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 21

21

EXPONENTIAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE
Review:

𝐑 𝒕 = 𝒆−𝝀𝒕

𝟏
𝐌𝐓𝐓𝐅 =
𝛌

Where, λ = failure rate

(λt is the mean number of shocks – i.e., each failure in the process
is caused by a random shock, and the number of shocks occurring
in a time interval of length t)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 22

22

11
 WEIBULL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE
• It is appropriate for a system or complex component composed of
a number of components or parts whose failure is governed by the
most severe defect or vulnerable of its components or parts.

βt β−1 − t β
f t = e α
αβ
∞ ∞
βθβ−1 − θ β θ β
α dθ
−
α
∞
R t = න f θ dθ = න e = −e ฬ
t t αβ t
𝐭 𝛃
−
𝐑 𝐭 =𝐞 𝛂
where
α = scale parameter (to adjust the MTTF value. It is usually specific to the
particular system or component being analyzed).
β = shape parameter (positive real number that affects the distribution of
failure times in the system).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 23

23

WEIBULL DISTRIBUTION AS A MODEL OF

TIME-TO-FAILURE (cont.)
➢ Weibull distribution is a good choice for
▪ Corrosion caused failure
▪ Time to failure of many types of basic component
(capacitors, ball bearings, etc.)

βt β−1 − t β
f t = e α
αβ
β
βt β−1 − αt
f(t) β e
h t = = α
R(t) t β
−
e α

𝛃−𝟏
𝛃 𝐭
𝐡 𝐭 =
𝛂 𝛂

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 24

24

12
 USES OF WEIBULL DISTRIBUTION

➢ Case I: β = 1
1 t 1−1 𝟏
That is, 𝐡 𝐭 =α =𝛂
α

1
So, it is the same as an exponential distribution with λ = and
α
MTTF = α.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 25

25

USES OF WEIBULL DISTRIBUTION (cont.)

➢ Case II: β > 1

• The Weibull distribution can be used as a model for the
wear out region.
• That is, h(t) is monotonically increasing.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 26

26

13
 USES OF WEIBULL DISTRIBUTION (cont.)

➢ Case III: β < 1

That is, h(t) is monotonically decreasing

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 27

27

USES OF WEIBULL DISTRIBUTION (cont.)

➢ For both Case II and Case III

∞
t β
MTTF = න R t dt −
R t =e α
0
𝛃+𝟏
𝐌𝐓𝐓𝐅 = 𝛂𝚪
𝛃
Γ: Gamma function is a reliability function parameters (Gamma
distribution)
▪ Note that Γ x + 1 = xΓ(x).
▪ For Γ 1 < x < 2 a table can be used and for integer
values of n, Γ n + 1 = n!
▪ For β = 3.75 Weibull represents normal distribution
▪ For β > 10 Weibull represents extreme value distribution

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 28

28

14
 NORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE
𝟏 f(t)
𝟏 − 𝟐 𝐭−𝛍𝐭 𝟐
𝐟 𝐭 = 𝟐𝛔
𝐞 𝐭
𝛔𝐭 𝟐𝛑 1

where -  < t < ,-  < µ < , 2 > 0 time

Normal Distribution

𝛍𝐭 = 𝐌𝐓𝐓𝐅
t−μt 2
1 ∞ − 2 𝐭−𝛍
𝐑 𝐭 = ‫ ׬‬e 2σt dt = 𝟏 − 𝛷(z) , z =
σt 2π t 𝛔

➢ Normal distribution is good model for components under high stress, which
are time-independent reliability models.
➢ Values of ϕ(z) may be obtained from standard normal distribution table.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 29

29

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 30

30

15
 NORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE (cont.)
Example:
A component’s stress to failure is normally distributed with MTTF of 20
kg/cm2 and standard deviation of 3 kg/cm2.
1) Find the reliability at a stress level of 25 kg/cm2.

25 − 20
R t =1−Φ = 1 − Φ 1.667 = 0.048
3

2) Find the probability of failure between stress levels 25 and 28

kg/cm2.

Φ 2.667 − Φ 1.667 = 0.044

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 31

31

NORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE (cont.)
3) Find the conditional probability of failure given that the component
has survived under a 25kg/cm2 stress

Pr(25 < T < 28 ∩ T > 25)

Pr 25 < T < 28 T > 25 =
Pr(t ≥ 25)

Pr(25 < T < 28) 0.044

Conditional Probability of Failure = = = 0.92
Pr(t ≥ 25) 0.048

R = 0.08

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 32

32

16
 LOGNORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE
➢ Widely used in reliability engineering.
➢ It is good for failures due to events that may vary by several orders of
magnitude, such as the time to finish a repair task.
▪ Specific applications include time to failure of components due to fatigue
cracks.
▪ Other applications are associated with failures attributed to maintenance
activities and distribution of cracks initiated and grown by mechanical
fatigue.
𝟏
𝟏 − 𝐥𝐧𝐭−𝛍𝐭 𝟐
𝐟 𝐭 = 𝐞 𝟐𝛔𝟐𝐭
𝛔𝐭 𝐭 𝟐𝛑
Where,
μt = mean of log of t’s [ μt = E(ln t) and 2 = Var(ln t) ]

➢ The failure rate for the lognormal distribution initially increases over time and
then decreases (depends on the values of the parameters μt and σt).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 33

33

LOGNORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE (cont.)
Example:
➢ Time that an operator can shutdown a system successfully
is lognormally distributed with μt = 2.0273 , σ2t = 0.4608 ,
determine probability of success by t = 20 seconds (that is:
the operator shuts down the system within 20 seconds).
➢ The failure probability = Pr(T > t), and the success
probability = Pr(T  t)

lnt − µt ln20 − 2.0273

z1 = = = 1.427
σt 0.4608
Pr T > 20 = 0.923
Pr T ≤ 20 = 0.077

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 34

34

17
 LOGNORMAL DISTRIBUTION AS
A MODEL OF TIME-TO-FAILURE (cont.)
➢ Note that you can convert mean-log value to lognormal mean
(same with variance)
σ2
μt + t
MTTS = μy = e 2
1
2 2
σy = eσt − 1 μy

➢ MTTS represents the average time it takes for a system or process to

achieve a successful outcome or complete a task. It is a measure of the
reliability or efficiency of a system in terms of achieving its intended goals.

➢ Therefore, MTTS (Mean-Time-To-Success)

0.4608
MTTS = μy = e2.0273+ 2 = 10 sec
Median = eμt = e2.0273 = 7.56 sec

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 35

35

FAILURE MODEL SELECTION

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 36

36

18
 3. FAILURE MODEL SELECTION
➢ There are ways in which we can establish whether a distribution is
a good fit. Two ways will be discussed:

▪ Non-parametric
(do not need a particular distribution function)

Probability Plotting

▪ Parametric Maximum Likelihood Estimation

(based on selected distribution)
Bayesian Estimation

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 37

37

LIFE ESTIMATION

➢ What we estimate from the data to construct parameters of the

underlying distribution model using test data or field data.

Probability Plotting
Parameter
Maximum Likelihood Estimation
Estimation
Bayesian Estimation

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 38

38

19
 NON-PARAMETRIC
HAZARD-RATE ESTIMATES
➢ Time to failure data t1 ≤ t 2 ≤ ⋯ ≤ t n

probability of failure
➢ Knowing that, h t =
time

for an interval ∆t = t i+1 − t i

number of failure between t and t + ∆t

number of units surviving past t
h t =
∆t

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 39

39

NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)

t=0 i=1 2 3 4 5 n
t1 t2 t3
for small samples
1
h t =
(n − i)(t i+1 − t i )

➢ To make an unbiased estimation of h(t), corrections are necessary.

𝟏
መ
𝐡(𝐭) =
(𝐧 − 𝐢 + 𝟎. 𝟔𝟐𝟓)(𝐭 𝐢+𝟏 − 𝐭 𝐢 )

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 40

40

20
 NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)

f(t)
h t =
R(t)
n−i
R ti =
n
➢ Similarly unbiased estimators of R(t) and f(t) are
𝐧 − 𝐢 + 𝟎. 𝟔𝟐𝟓
෡ 𝐭𝐢 =
𝐑
𝐧 + 𝟎. 𝟐𝟓
𝟏
𝐟መ 𝐭 𝐢 =
(𝐧 + 𝟎. 𝟐𝟓)(𝐭 𝐢+𝟏 − 𝐭 𝐢 )
➢ Keep in mind that 0.625 and 0.25 are correction terms of a minor
importance, which result in a small bias and a small mean square
error for the Weibull distribution estimation.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 41

41

NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)
Example:

1
i
ti f̂ ( t i ) =
(hour) (n + 0.25 )(t i+1 − t i )
1
= 3.88E -5 hr -1
1 1,000 ( 4.25)( 6100) f(t)
7.6E-5 X

1
= 5.74E -5 hr -1
2 7,100 ( 4.25)( 4100)

5.7 X
1
3 11,200 = 7.59E -5 hr -1
( 4.25)( 3100) 3.8 X

1000 7100 11,200 t

1
4 14,300 = undefined
( 4.25)( ? )

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 42

42

21
 NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)
 (t ) =
( n - i + 0.625)
ti R
i
(hour)
i
( n + 0.25)

( 4 - 1 + 0.625)
1 1,000 = 0.853
( 4 + 0.25)

( 4 - 2 + 0.625)
2 7,100 = 0.617
( 4 + 0.25)

( 4 - 3 + 0.625)
3 11,200 = 0.382
( 4 + 0.25)

4 14,300 ( 4 - 4 + 0.625)
= 0.147
( 4 + 0.25)

➢ So, the hazard rate is more or less constant and the reliability is
decreasing exponentially.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 43

43

NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)
ti 1
i
(hour)
hˆ(t ) =
(n − i + 0.625)(ti +1 − ti )
1
= 4.52  10 −5
1 1,000 (4 − 1 + 0.625)(7100 − 1000)

1
2 7,100 = 9.29  10 −5
(4 − 2 + 0.625)(11200 − 7100)

1
3 11,200 = 1.99  10 − 4
(4 − 3 + 0.625)(14300 − 11200)

1
4 14,300 = undefined
(4 − 4 + 0.625)(?− 14300)

➢ So, the hazard rate is more or less constant, and the reliability is
decreasing exponentially.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 44

44

22
 NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)

Large Samples:

Nf (ti )
The estimate of probability that the component h෠ t i =
Ns (ti )∆t
will fail in the interval (ti, ti + Δt)
Ns (t i )
Nonparametric estimate of the reliability ෡ ti =
R
෡ ti
function is R N

Nf (t i )
The pdf is estimated by መf t i fመ t i =
N(∆t)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 45

45

NON-PARAMETRIC
HAZARD-RATE ESTIMATES (cont.)

Example:

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 46

46

23
 Important Time-to-Failure Distributions and
Their Characteristics

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 47

47

Important Time-to-Failure Distributions and

Their Characteristics (cont.)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 48

48

24
 PROBABILITY PLOTTING
➢ Observed data may be plotted on special probability papers such
that the resulting curve falls on a straight line.

➢ Probability papers for many types of distributions exist.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 49

49

EXPONENTIAL PLOTTING

R t = e−λt
1
or = eλt
R(t)

Taking logarithms, we get

1
ln = λt
R(t)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 50

50

25
 EXPONENTIAL PLOTTING (cont.)

➢ Suppose n components of a small sample (e.g., n < 25) fail, i.e.,

time-to-failure are t1 , t 2 , ⋯ t i ⋯ t n for n components.

▪ order failure times t1 ≤ t 2 ≤ ⋯ ≤ t i ≤ ⋯ ≤ t n

▪ Reliability estimator (Kimball Plot Position)

(n − i + 0.625)
෡ ti =
R
(n + 0.25)

0.625 and 0.25 come from statistical analyses to give a good

estimate, thereby removing any bias.

(Slide 42)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 51

51

EXPONENTIAL PLOTTING (cont.)

➢ Plot the n values

▪ Data is plotted on a semi-log paper with arithmetic scale for ti

and logarithmic scale for

n + 0.25
n − i + 0.625
▪ If a straight line is a good fit exponential distribution is an
adequate model.

▪ Slope of the fitted line indicate λ (failure rate)

n+0.25
ti, i = 1,2, … , ณ
𝑛
n−i+0.625
ma𝑥

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 52

52

26
 EXPONENTIAL PLOTTING (cont.)

Example:
250 Germanium transistors have the following time-to-failures:
(Note that the test was terminated after 240 failures were
observed).

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 53

53

EXPONENTIAL PLOTTING (cont.)

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 54

54

27
 EXPONENTIAL PLOTTING (cont.)

➢ We may observe two or more straight lines. This happens when

there may be initially a particular failure mode and another failure
mode become predominant, and so on.

λE = λ1 + λ2
Therefore,

R = R1 ∙ R 2
= e−λ1 t e−λ2 t
= e−λEt

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 55

55

WEIBULL DISTRIBUTION
➢ The most versatile distribution in Reliability Engineering because of
its ability to model a variety of distributions.
➢ It can be used to model any failure rate of the product over the
entire lifetime.
▪ For the three cases discussed earlier (decreasing, constant, and
increasing failure rate).

➢ The Weibull Distribution is a

continuous probability distribution
that is used to analyze life data,
model failure times, and assess
the reliability of access products.
➢ This distribution can be used
largely in different fields to analyze
data.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 56

56

28
 WEIBULL PLOTTING
t β
−
➢ Weibull distribution: R t =e α

➢ How do we calculate the shape parameter β (the most important

parameter) also called a slope, and the scale parameter α.
➢ The shape parameter  dictates the shape of the distribution.
▪ When  < 1, it represents a system with a decreasing failure rate.
▪ When  = 1, it approximates the exponential distribution.
▪ When  > 1, it represents a system with an increasing failure rate.
▪ When  = 3.5, it approximates the normal distribution.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 57

57

WEIBULL DISTRIBUTION
➢ If 𝛃 = 𝟏, Weibull distribution becomes an exponential distribution,
t 1
−
because R t = e α = e−λt
1
Thus, λ =
α
➢ scale parameter α, (MTTF) for certain , a stretches the distribution
out over the time.
▪ The bigger scale parameter the more reliable the product is.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 58

58

29
 WEIBULL DISTRIBUTION (Example)
You have collected data on a component and calculated that your
product fits a Weibull distribution with a slope of 2. The data also
indicates that your scale parameter is equal to 8,000 hours.
What is the reliability of the system at 5,000 hours.

Solution:
 = 2 ,  = 8,000 , T = 5,000

t β
−
R t =e α

5,000 2
− =0.6766
R 5,000 = e 8,000 or 67.7% Reliability

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 59

59

WEIBULL PLOTTING

1
ln ln = βlnt − βlnα
R

➢ This is a straight line of the form Y = mX + c

1
Where, Y equivalent to ln ln
R
X equivalent to ln𝑡
slope m is the shape parameter .

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 60

60

30
 Steps in Probability Plotting of complete data

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 61

61

OTHER PLOTTING POSITIONS

i−0.5
Midpoint n

i
Mean n+1

i−0.3
Median n+0.4

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 62

62

31
 NORMAL PLOT

➢ We plot t i on the ordinate axis and the percentage of failure

i − 0.375
estimated by × 100 on the abscissa.
n + 0.25

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 63

63

NORMAL PLOT OF THE EXAMPLE

t−μ t−μ
Example: Normal distribution F = Φ( ), Φ−1 F =
σ σ

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 64

64

32
 LOGNORMAL PLOT OF THE EXAMPLE
lnt−μ lnt−μ
Example: Lognormal distribution F = Φ( ), Φ −1 F =
σ σ

➢ Conclusion: In this case, Normal distribution fits better than Lognormal distribution
(ie: better R2 value for the normal plot)
Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 65

65

WEIBULL PLOTTING (Example)

For certain motor, the instances of failure and the number of days for
the failure. For example, the 1st failure occurs after 400 days, 2nd one
after 820 days. The following table shows a data to 10 failures.
Failure 1 2 3 4 5 6 7 8 9 10

Days 400 820 1050 1244 1500 1780 1920 2100 2230 2500

Step 1: Calculate Median Rank

Step 2: Determine Natural Logarithm
Step 3: Plot Distribution Chart
Step 4: Compare Equation to Determine Coefficients

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 66

66

33
 WEIBULL PLOTTING (Example)

i−0.3
Step 1: Calculate Median Rank Median, F(t) =
n+0.4
i
Failure Days Median F(t) n = 10
Arranged in ascending order

1 400 0.067
2 820 0.163
3 1050 0.260
4 1244 0.356
5 1500 0.452
6 1780 0.548
7 1920 0.644
8 2100 0.740
9 2230 0.837
10 2500 0.933

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 67

67

WEIBULL PLOTTING (Example)

1
Step 2: Determine Natural Logarithm ln ln = βlnt − βlnα
R

Failure Days Median F(t) ln(days) ln(ln(1/(1-F(t))))

1 400 0.067 5.991464547 -2.663843085
2 820 0.163 6.70930434 -1.72326315
3 1050 0.260 6.956545443 -1.202023115
4 1244 0.356 7.126087273 -0.821666515
5 1500 0.452 7.313220387 -0.508595394
6 1780 0.548 7.484368643 -0.230365445
7 1920 0.644 7.560080465 0.032924962
8 2100 0.740 7.649692624 0.299032932
9 2230 0.837 7.709756864 0.593977217
10 2500 0.933 7.824046011 0.992688929

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 68

68

34
 WEIBULL PLOTTING (Example)

Step 3: Plot Distribution Chart

• The next step is to plot a chart using ln(days) as x-axis and
ln(ln(1/(1-F(t)))) as y-axis.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 69

69

WEIBULL PLOTTING (Example)

Step 4: Compare Equation to Determine Coefficients

• The equation is y = 1.9551x - 14.663.
• After comparing with ln(ln(1/(R(t))) = βlnt - βlnα,

βlnt equivalent to 1.9551x , where lnt = x and β = 1.9551

βlnα = 14.663 , thus, α = 1807.811

• So, the Reliability Function becomes

t 1.9551
−
R t =e 1807.811

• As β > 1, the failure rate increases with time.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 70

70

35
 CENSORING
➢ Censoring refers to the situation where the exact failure time of a
product or system is not observed or recorded for all samples in the
study.
➢ Censoring occurs due to the following reasons:
▪ when a sample has not failed by the end of the testing period.
▪ when the failure time is unknown due to other reasons, such as;
• the study being terminated before all samples fail.
• the failure time being outside the observation window.
➢ it is often impractical or too time-consuming to observe the failure
time for every individual sample until it fails.
▪ Censoring allows for efficient use of resources and enables
statistical analysis to estimate the reliability characteristics of
the product or system.
Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 71

71

CENSORING (cont.)
➢ The concept of censoring is closely associated with the likelihood
function, which is used to estimate the parameters of a statistical
model based on observed data.
▪ the likelihood function incorporates the information from both
the observed failure times and the censored data to estimate
the reliability parameters.
➢ Here are a few common types of censoring and their corresponding
likelihood functions:
▪ Right Censoring.
▪ Left Censoring.
▪ Interval Censoring.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 72

72

36
 TYPES OF CENSORING
➢ Right Censoring:
▪ Samples are tested until they fail or until a predetermined study
duration is reached.
• If a sample has not failed by the end of the study, its failure
time is considered right-censored.
▪ The likelihood function is based on the observed failure times
and the known censoring times.
• It takes into account the probabilities of observing the
failure times and the probabilities of right-censoring for the
remaining samples.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 73

73

TYPES OF CENSORING (cont.)

➢ Left Censoring:
▪ It occurs when the failure times are known to be less than a
certain value, but the exact failure time is unknown.
• This type of censoring is observed when the study begins
after some units have already failed or when the failure
times are only known to be earlier than a specific event or
threshold.
▪ The likelihood function incorporates the probabilities of
observing the failure times that are greater than the censoring
threshold and the probabilities of left-censoring for the
remaining samples.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 74

74

37
 TYPES OF CENSORING (cont.)
➢ Interval Censoring:
▪ It occurs when the failure times are only known to lie within
certain intervals.
▪ This type of censoring is often encountered when the failure
times can only be observed at discrete time points or during
specific inspection intervals.
▪ The likelihood function considers the observed intervals and the
probabilities of the failure times falling within those intervals.
• It also incorporates information on the censoring intervals
for the remaining samples.

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 75

75

LIFE TESTING (or Reliability Testing)

➢ A method used to assess the reliability and durability of a product or

system over an extended period of time.
➢ It involves subjecting the item under consideration to various stress
levels, operating conditions, or environmental factors to determine
its performance and failure characteristics.
➢ It is done to get failure data for reliability estimation methods.

Type I (time terminated) With replacement

Life Testing
Type II (failure terminated) Without replacement

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 76

76

38
 LIFE TESTING (cont.)

➢ Type I (time terminated or cycle dependent performance) with

replacement:

▪ N components are placed under test.

▪ 𝑡0 time at which the test is terminated.
▪ T accumulated component test hours.
𝑇 = 𝑁 𝑡0

𝐹
λ෠ =
𝑇
Where, F is number of failures up to 𝑡0 .
T
෣ =
MTTF
F
and number of units actually in test (𝑁′) is 𝑁 ′ = 𝑁 + 𝐹

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 77

77

LIFE TESTING (cont.)

➢ Type I without replacement:

𝑇 = 𝑁𝑡1 + 𝑁 − 1 𝑡2 − 𝑡1 + ⋯ + 𝑁 − 𝐹 + 1 𝑡𝐹 − 𝑡𝐹−1 + (𝑁 − 𝐹)(𝑡0 − 𝑡𝐹 )

𝑇 = ෍ 𝑡𝑖 + 𝑁 − 𝐹 𝑡0
i=1

accumulated component test accumulated component test

hours of components failed hour of non-failing components

𝐹
λ෠ =
𝑇

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 78

78

39
 LIFE TESTING (cont.)

➢ Type II (failure terminated) with replacement:

▪ N components placed on test.

▪ 𝑇𝑓 the time after which test is terminated when the 𝑓 𝑡ℎ failure
has occurred.
▪ So, 𝑓 𝑡ℎ failure time is specified by 𝑇𝑓 and is a random
variable.
𝑇 = 𝑁 𝑇𝑓

𝑓 𝑇
λ෠ = , ෣ =
MTTF
𝑇 𝑓

▪ Total units put to test (N'), that is, 𝑁 ′ = 𝑁 + 𝑓 − 1

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 79

79

LIFE TESTING (cont.)

➢ Type II without replacement:

T = Nt1 + N − 1 t 2 − t1 + ⋯ + N − r + 1 t r − t r−1
r−1

= ෍ t i + N − r + 1 Tr
i=1
r

T = ෍ t i + N − r Tr
i=1
and

N′ = N

Prepared by Dr. Osama Al-Jamal Reliability Analysis Methods - 80

80

40