RELIABILITY DEFINITION, METRICS, AND PRODUCT LIFE

DISTRIBUTIONS
LECTURE 2
Reliability consist of planning, design, testing and analysis. In reliability planning a
product of concern must be defined in specified terms such as, intended function,
specified condition, and specified period of time must be adhere to, to satisfy or
exceed pertinent design requirements. Once reliability is defined, appropriate
metrics must be chosen to characterize it.Metrics that are sensitive to time and stress
and are of concern to both the manufacturer and the customer.
In this instance, the common metrics/statistical distributions are exponentials,
weibull, mixed weibull, smallest extreme values, normal and lognormal.
RELIABILITY DEFINITION/ TERMINOLOGIES
The following are terms that relate to reliability:
1 Binary State: The function of a product is either success or failure.
2 Multi state: The function of a product can be complete success, partial success, or
a failure. Performance degradation is a special case of multistate.
3 Hard failure: This is a total failure that causes complete ceasation of a function.
Such a failure mode occurs in binary state product.
4 Soft failure: This is a partial loss of function. The failure mode occurs in multistate
(degradation) product.
RELIABILITY is defined as the probability that a product performs its intended
function without failure underspecified conditions for a specified period of time. The
definition contain three main elements: underspecified condition, specified period
of time and intended functions.
RELIABILITY METRICS is the common metric used to measure reliability.
Probability Density Function (pdf): The pdf, denoted f (t) which indicate the
failure distribution over the entire time range and represents absolute failure speed.
The larger the value of f (t) the more the failures that occurs in a small time interval
of t.
Cumulative Distribution Function (cdf): The cdf, denoted F (t), which is the
probability that a product will fail by a specified time t. It is the probability of failure,
interpreted as a fraction of the populations failing by time t. Mathematically, it is
defined as
∞
F (t) = Pr (T≤ t) = ∫−∞ 𝑓 (𝑡)𝑑𝑡 ……………………………. (1)
(1) is equivalent to

f (t) =𝑑𝐹(𝑡)
𝑑𝑡
….. . …………………………………... (2)
For example if the time to a failure of a product is exponentially distributed with
parameter λ, the pdf is
f (t) = 𝜆 exp(−𝜆𝑡), t ≥ 0 ……………………………………………… (3)
1
F (t) = ∫0 𝜆 exp(−𝜆𝑡) 𝑑𝑡 = 1 − exp(−𝜆𝑡) , t ≥ 0 …………………… (4)
Reliability The reliability function, denoted R (t), is called survival function, is often
interpreted as population fraction surviving time t. R (t) is the probability of success
which is the complement of F (t). It can be written as
∞
R (t) = Pr (T ≥ t) =1 – F (t) = ∫1 𝑓(𝑡)𝑑𝑡 ……………………………. .(5)
From (4) and (5), the probability function of the exponential distribution is
R (t) = exp (-𝜆𝑡), 𝑡 ≥ 0 ……………………………………………… (6)
Hazard function: The hazard function or hazard rate, denoted h (t) and often called
the failure rate, measures the rate of change in the probability that a surviving
product will fail in the next small interval of time. It can be written as
Pr(𝑡≤𝑇≤1+Δ𝑡|𝑇≥𝑡
h (t) = lim
Δ𝑡
Δ𝑡 → 0
1 −𝑑𝑅(𝑡) 𝑓(𝑡)
= [ ]= …………………………………………………… (7)
𝑅(𝑡) 𝑑𝑡 𝑅(𝑡)

From (3), (6), and (7) the hazard rate of the exponential distribution is
h (t) = λ ……………………………………………………………..(8)
The equation (8) indicates that the hazard rate of exponential distribution is a
constant. The unit of hazard is failure per unit time. There are three types of hazard
rates 1.Decreasing hazard rate (DFR) 2.Constant hazard rate (CRF) 3.Increasing
hazard rate (IRF)
The figure below shows the bathtub hazard rate function. The curve represents the
observation that the life span of a population of products is comprised of three
distinct periods
a) Early Failure Period: The hazard rate decreases over time
b) Random Failure Period: The hazard rate is constant over time
c) Wear-out Failure Period: The hazard rate increases over time

Early Failure is caused by patent defects early in the service time. The patent
defects may be induced by manufacturing process variation, material flaws, and
design errors; customer misuse is another cause of early failure. If the life
distribution of a sub-standard products is weibull with a shape parameter less than
1, the hazard rate decrease over time and if it is greater than 1, it has an increasing
trend.
Random Failure is such that the hazard rate remains approximately constant. In this
period, failures do not follow any predictable pattern and occur at random due to
unexpected changes in stresses.
Wear-out Failure the hazard rate increases with time as a result of irreversible aging
effects. The failures are attributable to degradation or wear-out which accelerate over
time. At this point failure is imminent and to minimize these failures, preventive
maintenance or schedule replacement of product is often necessary.
CUMMULATIVE HAZARDS FUNCTION
The cumulative hazard function, denoted H (t) is defined as
𝑡
H (t) = ∫−∞ ℎ(𝑡)𝑑𝑡 …………………………………………………….. (9)

For exponential distribution, we have

H (t) =𝜆𝑡...……………………………………………………………………… (10)
From (7) and (9) the relationship between H (t) and R (t) can be written as
R (t) = exp [- H (t)]……………………………………………………………… (11)
If H (t) is very small, a Taylor series expansion result in the following approximation
R (t) ≈ 1 − 𝐻(𝑡) …………………………………………………… (12)
PERCENTILE, denoted tp is the time by which specified fraction p of the
population fails. It is the inverse of F (t): namely
tp = F-1(t) …………………………………………………………… (13)
For exponential distribution, we have
1 1
tp = ln( ) ………………………………………………………. (14)
𝜆 1−𝑝

MEAN TIME TO FAILURE (MTTF) is the expected life E (t) of a no repairable

product. It is defined as:
∞
MTTF=E (T) =∫−∞ 𝑡𝑓 (𝑡)𝑑𝑡 …………………………………………. (15)
If the range T, is positive then (15) can be written as
∞
MTTF =∫−∞ 𝑅 (𝑡)𝑑𝑡 ………………………………………………………… (16)
EXPONENTIAL DISTRIBUTION
For the exponential distribution, MTTF is
∞ 1
MTTF = ∫−∞ exp(−𝜆𝑡) 𝑑𝑡 = ……………………………………….. (17)
𝜆
VARIANCE is denoted Var (T) is a measure of a spread of a life distribution,
defined as:
∞
Var (T) = ∫−∞[𝑡 − 𝐸 (𝑡)]2𝑓(𝑡)𝑑𝑡 .................................................................. (18)
For exponential distribution, the variance is
∞ 1 1
Var (T) = ∫0 (𝑡 − )2𝜆 exp(−𝜆𝑡) 𝑑𝑡 = ( )2 ……………………………... (19)
𝜆 𝜆
The memory less property can be express mathematically as,
Pr (T> t +t0 ∕ T> t0) = Pr (T> 𝑡) …………………………………………. (20)
Where t0 is the time spent already.
WEIBULL DISTRIBUTION
The pdf of the weibull distribution is
𝛽 𝑡
f (t) = tβ-1 exp [- ( ) β ] , t > 0 ………………………………………….. (21)
𝛼𝛽 𝛼
The weibull cdf is
𝑡
F (t) =1- exp [- ( ) β], t > 0 ……………………………………………. (22)
𝛼
The hazard function is
𝛽 𝑡
h (t) = ( ) β-1 ………………………………………………………………. (23)
𝛼 𝛼
The cumulative hazard function is
𝑡
H (t) = ( ) β , t > 0 …………………………………………………………… (24)
𝛼
The 100pth percentile is
tp = ∝ [−𝐼𝑛(1 − 𝑝)1/β …………………………………………………………… (25)
The mean and variance are
 1
E (T) = ∝ 𝑇(1 + )
𝛽
2 1
Var (T) = ∝2 [r (1+ ) – r2 (1+ )] ………………………………………. (26)
𝛽 𝛽

∞
r (x) = ∫0 𝑍x-1 exp (-Z) dz
In the weibull formulae above, 𝛽 is the shape parameter and X is the characteristic
life, both are positive. X is also called the scale parameter, equals the 63.2th
percentile.
Example
The life of an automotive component is weibull with ∝= 6-2x 105 miles and β =
1.3. Calculate F (t), R (t), and h (t) at the end of warranty period (36,000 miles) and
t0.01
Solution
Time, t =36,000 mile
Alpha, ∝ = 6.2 x 105 miles
Beta, β = 1.3
𝑡
F (t) = 1- exp [- ( ) B ]
∝
36,000
F (36,000) = 1-exp [-( ) 1.3
6.2𝑥 105
F = 0.024
This indicates that 2.4% of the component population will fail by the end of
warranty period. The reliability at 36,000 miles is
R (t) = 1- F (t)
R (36,000) = 1- 0.024 = 0.976
That is 96.7% of the component population will survive the warranty period.
𝛽 𝑡
The hazard rate at 36.000 miles is h (t) = ( )β-1
∝ ∝
 1.3 36,000 1.3-1
h (36,000) = ( ) = 0.89 x 10-8 failures per mile.
6.2𝑥105 6.2𝑥105
Because β is greater than 1(β> 1) the hazard rate increases with mileage.
The 100pth percentile is tp = ∝ [−𝐼𝑛(1 − 𝑃)1/β. The mileage by which 1% of the
population will fail
t0.01 = 6.2x105 x [-In (1-0.01)] 1/ 1.3
= 18,014 miles.

MIXED WEIBULL DISTRIBUTION

A mixed distribution comprises of two or more distributions. It is mathematically
indicated as:
𝑡 𝑡
F (t) = PF1 (t) + (1- p) F2 (t) = 1- p exp [- ( ) β1] – (1-p) exp [- ( ) β2]
∝1 ∝2
Example
An automotive component population produced in the first two months of
production contains 8% defective units. The mileage to failure of the components
has a bimodal weibull distribution with β1, = 1.3 ∝1 = 12,000 miles β2 =2.8 and
∝2 = 72,000 miles. Calculate the probability of failure at the end of warranty period
(36,000)
Solution
36,000 1.3 36,000 2.8
F (36,000) = 1- 0.08 x exp [- ( ) ] – 0.92 x exp [- ( ) ]
12,000 72,000

= 0.202
This indicates that 20.2% of the component population produced in the first two
months of production will fail by 36,000 miles.

NORMAL DISTRIBUTION
The pdf of the normal distribution is
 1 (𝑡−𝜇)2
f (t) = exp[- ] , -∞ < 𝑡 < ∞
√2𝜋𝜎 2𝜎2

THE NORMAL CFD

The normal cdf is
𝑡 1 (𝑦−𝜇)2
F (t) = ∫−∞ exp [- ] dy, -∞< 𝑡 < ∞
√2𝜋𝜎 2𝜎2
The mean and variance are
E (T) = µ and Var (T) =𝜎2 .When µ = 0 and𝜎 = 1, the normal distribution is
called the standard normal distribution, Then the pdf becomes
1 𝑧2
∅(𝑧) = Exp (- ), -∞ < 𝑧 < ∞
√2𝜋 2

The cdf of the standard normal distribution is

𝑧 1 𝑦2
∅(𝑧) = ∫−∞ √2𝜋exp (- )dy, -∞ < 𝑧 < ∞11
2

Example
An electronic circuit contains three resistors in series. The resistances (R 1+ R2+
R3) in ohms of normal distributions R1~ 𝑁(10,0.32 ), R2~𝑁(15, 0.52) and
R3~𝑁(50, 1.8 2 ). Calculate the mean and standard deviation of the total resistance
and probability of the total resistance being within the tolerance range 75 =/- 5%
Solution
Total resistance R0 = R1 + R2+ R3.
The mean of the total resistance N= 10 + 15 +50 = 75 ohms
The standard deviation of the total resistance is
𝜎= (0.32 +0.52 + 1.82) 1/2 = 1.89 ohms
The probability of the total resistance being within 75+/- 5%
78.75−75 71.25−75
Pr (71.25≤ 𝑅 0≤ 78.75) = ∅( ) - ∅( )
1.89 1.89

= 0.976- 0.024 = 0.952

LOGNORMAL DISTRIBUTION
The pdf of the lognormal distribution is
1 (𝐼𝑛(𝑡)−𝜇)2 1 𝐼𝑛(𝑡)−𝜇
f (t) = exp[- ]= ∅[ ] , where ∅(.) is the standard
√2𝜋𝜎𝑡 2𝜎 2 𝜎𝑡 𝜎
normal pdf. The lognormal cdf is
1 1 [𝐼𝑛(𝑦)−𝜇 2 𝐼𝑛(𝑡)−𝜇
F (t) = ∫0 √2𝜋𝜎𝑦exp { } dy =∅[ ], t> 0
2𝜎 2 𝜎

Where ∅(.) the standard normal cdf. The 100pth percentile is

tp = exp (𝜇 + 𝑧𝑝𝜎)
E (T) = exp (𝜇 + 0.5𝜎2) and Var (T) = [exp (2𝜇 + 𝜎2) exp (𝜎2) – T]

Example
The warranty plan of a car population covers 36 months in service or 36,000 miles
whichever comes first. The accumulated mileage U of the car population by a given
month in services can be modelled with the lognormal distribution with scale
parameter 6.5+In (t) and shape parameter 0.68, where t is the months, in services of
the vehicles.
Calculate the population fraction exceeding 36,000 miles by 36 months.
Solution

The probability that a car exceeds 36,000 miles by 36 months is

𝐼𝑛(36,000)−6.5−𝐼𝑛(36)
Pr (U≥ 36,000) = 1 − ∅[ ]
0.68

= 0.437
This means by the end of the warranty period, 43.7% of vehicles will be leaving
the warranty coverage by exceeding the warranty mileage limit.