Reliability Definition

Reliability is defined as the probability of the system meeting

adequate performance for a specific time period under the specific
operating conditions.

“If a builder builds a house for a man and does not make its construction firm, and the house
which he has built collapses and causes the death of the owner of the house, that builder
shall be put to death.”
—The earliest known law governing reliability by King Hammurabi of Babylon,
4000 years ago.
Prediction & Estimation
Reliability Prediction:
Forecasting the reliability of the system/component at any given
Stage/point of time in future.
- During design phase

Reliability Estimation:
Current assessment of the reliability value of a system/component
- During the working phase
Limitations:
• Reliability theory can’t predict when the system will fail

• A system with higher reliability value can fail-first, then the system
with lower reliability value

• The ideal reliability value of “1” can be achieved only if you don’t
use the system i.e. at t = 0 R = 1
Quality Vs Reliability
Quality Reliability

• Quality is the degree to which the

product is made to its specifications.
• Reliability is the dynamic measure
• Quality is the static measure.
• At t = 0, R = 1
• All products don’t have same quality
at t = 0 • Quality will help to achieve the
design reliability
• A product with poor quality may have
better reliability. Vice versa
Basic Reliability Models
 Failure Distribution:
Reliability also termed as the measure of the system success in providing
its function adequately during its design-life.

𝑛𝑜 is the total number of components subjected to the operation in a time

interval (𝑡 − ∆𝑡, 𝑡)
𝑛𝑠 is the number of surviving components at 𝑡
𝑛𝑓 is the number of failed components at 𝑡
𝑅 𝑡 = 𝑅𝑒𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡 (Cumulative probability function of success).
To express reliability function mathematically, let 𝑇 is the random variable
denoting time to failure of a component / System.
Then, Reliability at time 𝑡 is,

Where, 𝑅 𝑡 ≥ 0; 𝑅 0 = 1; 𝑙𝑖𝑚𝑡→∞ 𝑅 𝑡 = 0
The Cumulative distribution function (CDF) of failure is the complement of
𝑅 𝑡 .
𝑅 𝑡 +𝐹 𝑡 =1
i.e. 𝑅 𝑡 = 1 − 𝐹(𝑡)
If 𝑇, has a pdf (probability density function) as 𝑓(𝑡), then
𝑡
𝑅 𝑡 =1−න 𝑓 𝜗 𝑑𝜗
0
Taking a derivative with respect to t,
 𝑑𝑅(𝑡)
= −𝑓 𝑡
𝑑𝑡
which is also
𝑑𝑅(𝑡) 𝑑𝐹(𝑡)
𝑓 𝑡 = − =
𝑑𝑡 𝑑𝑡
The above function describes the shape of the failure distribution.

The PDF of f(t) has the following properties.

∞
𝑓 𝑡 ≥ 0; න 𝑓 𝑡 𝑑𝑡 = 1
0
and, for given PDF 𝑓 𝑡 , following functions can be derived
𝑡 ∞
𝐹 𝑡 = ‫׬‬0 𝑓 𝜗 𝑑𝜗 and 𝑅 𝑡 = ‫𝑓 𝑡׬‬ 𝜗 𝑑𝜗
R(t) is used when reliabilities are computed;
F(t) is used when failure probabilities are computed.
 Hazard rate function
Hazard rate function or failure rate is extensively used in reliability
computations.
It provides the instantaneous failure rate at any given time t.
𝑡
As we deduced earlier, 𝑅 𝑡 = ‫׬‬0 𝑓 𝜗 𝑑𝜗
and
the probability of the system/component will fail in a given interval of time
(𝑡, ∆𝑡), given that the system/component* survived till 𝑡,
𝑡+∆𝑡
න 𝑓 𝑡 𝑑𝑡 = 𝑅𝑡 − 𝑅𝑡+∆𝑡
𝑡
(* conditional probability)
 R(t) − R(t + ∆𝑡)
𝑃𝑟 𝑡 ≤ 𝑇 ≤ 𝑡 + ∆𝑡ȁ𝑇 ≥ 𝑡 =
𝑅 𝑡 . ( 𝑡 + ∆𝑡 − 𝑡)

R(t) − R(t + ∆𝑡)

=
𝑅 𝑡 . ∆𝑡
This function gives the “conditional” probability of failure per unit time i.e.
failure rate.
Hazard rate is then defined as the limit of the failure rates as ∆𝑡 tends to be
zero.
Now the Hazard rate/ instantaneous failure rate is given as,

R(t) − R(t + ∆𝑡) 1 R(t) − R(t + ∆𝑡) 1 𝑑

𝜆(𝑡) = lim = = − 𝑅(𝑡)
∆𝑡→0 𝑅 𝑡 . ∆𝑡 𝑅(𝑡) ∆𝑡 𝑅(𝑡) 𝑑𝑡
 𝑑𝑅(𝑡)
from the expression 𝑓 𝑡 = − ,
𝑑𝑡
The Hazard rate is expressed as,
𝒇(𝒕)
𝝀(𝒕) =
𝑹(𝒕)
To relate /deduce the R(t) from 𝝀𝒕 :
1 𝑑 − 𝑑 𝑅(𝑡)
𝜆(𝑡) = − 𝑅(𝑡) ; 𝜆(𝑡) 𝑑𝑡 =
𝑅(𝑡) 𝑑𝑡 𝑅(𝑡)

Integrating both side of the expression,

𝑡 𝑅(𝑡) − 𝑑 𝑅(𝜗) 𝑡
‫׬‬0 𝜆 𝜗 𝑑𝜗 = ‫׬‬1 𝑅(𝜗)
= − ln 𝑅(𝑡) ; − ‫׬‬0 𝜆 𝜗 𝑑𝜗 = ln 𝑅(𝑡)
Taking “ln” out:
𝒕
− ‫𝝑𝒅 𝝑 𝝀 𝟎׬‬
𝑹 𝒕 =𝒆
Cumulative & Average failure rates
The cumulative failure rate over a time period of t is
𝑡
𝐿(𝑡) = න 𝜆 𝜗 𝑑𝜗
0
The average failure rate (AFR), defined between 𝑡1 & 𝑡2

𝑡
1 ln 𝑅(𝑡1 ) − ln 𝑅(𝑡2 )
𝐴𝐹𝑅 𝑡1 , 𝑡2 = න 𝜆 𝜗 𝑑𝜗 =
𝑡2 − 𝑡1 0 𝑡2 − 𝑡1
And if we let 𝑡1 =0 and 𝑡2 = 𝑇 , then AFR (t) is
ln 𝑅(0)−ln 𝑅(𝑡) 𝐿(𝑡)
𝐴𝐹𝑅 𝑡 = =
𝑡−0 𝑡
Central tendency of failure probability
MTTF (Mean Time To Failure)
• MTTF is one of the measures of reliability.
• MTTF is the expected time* between two successive failures when the
system/component is non-repairable.
• MTBF (mean time between failures) related to the repairable systems.
(*Expected time is the Mean of the PDF f(t))
𝑛
1
𝑀𝑇𝑇𝐹 = ෍ 𝑡𝑖
𝑛
𝑖=1
n = total number identical non-repairable systems observed
𝑡𝑖=1,2,3,….𝑛 = observed times to failure
Since 𝑇𝑖 is a random variable, the MTTF can be determined by
∞ ∞
𝑀𝑇𝑇𝐹 = ‫׬‬0 𝑡 𝑓 𝑡 𝑑𝑡 ; 𝑀𝑇𝑇𝐹 = ‫׬‬0 𝑅 𝑡 𝑑𝑡
Median:
• Median is another central tendency of the failure probability (pdf).
• Median divides the probability distribution into 2 halves, with 50% of
failures occur before median time and 50% after the median time.
𝑹 𝒕𝒎𝒆𝒅 = 𝟎. 𝟓
• This measure of central tendency is used when the pdf is highly skewed.
Mode:
• The most likely observed failure time (𝑡𝑚𝑜𝑑𝑒 ) defined as the mode of the
distribution.
• The probability of failure for a small interval of time centered around the
mode is higher than that of an interval of the same size located elsewhere
within the failure time distribution.
• The mode is obtained using the expression : 𝒇 𝒕𝒎𝒐𝒅𝒆 = 𝐦𝐚𝐱 𝒇(𝒕)
𝟎≤𝒕≤
Variance / Standard Deviation
• It is another measure of the failure distribution.
• The variance an average squared distance a failure time will be from the
mean (MTTF).
• It is the measure of spread / dispersion of the failure times about the mean
∞
𝜎 2 = න (𝑡 − 𝑀𝑇𝑇𝐹)2 𝑓 𝑡 𝑑𝑡
0
Also,
∞
𝜎 2 = න (𝑡)2 𝑓 𝑡 𝑑𝑡 − (𝑀𝑇𝑇𝐹)2
0
The square root of the variance is Standard deviation
 Conditional Reliability
Conditional reliability is a concept of assessing the reliability of a
component/system after certain time-period.
The time period may be burn-in period 𝑇0 or a warranty period 𝑇0
It can be defined as the reliability of a system given that it operated for time 𝑇0

𝑅 𝑡ห𝑇0 = 𝑃𝑟 𝑇 > 𝑇0 + 𝑡ห𝑇 > 𝑇0

𝑃𝑟 𝑇 > 𝑇0 + 𝑡 𝑅(𝑇0 + 𝑡)
= =
𝑃𝑟 𝑇 > 𝑇0 𝑅(𝑇0 )
𝑇 +𝑡
(− ‫׬‬0 0 𝜆 𝜗 𝑑𝜗) 𝑇0 +𝑡
𝑒
= 𝑇 = 𝑒𝑥𝑝 − න 𝜆 𝜗 𝑑𝜗
(− ‫׬‬0 0 𝜆 𝜗 𝑑𝜗)
𝑒 𝑇0
Mean residual life
• MRL is a measure of the reliability characteristic of a component /system.
• It is mathematically expressed as,
𝐿 𝑡 = 𝐸 [𝑇 − 𝑡ȁ𝑇 ≥ 𝑡] , 𝑡 ≥ 0
• The mean residual function is the expected remaining life, T−t, given that the
product, component, or a system has survived to time t.
• The conditional p.d.f. for any time 𝜏 ≥ 𝑡 is
𝑓(𝑡)
𝑓𝑇ȁ𝑇≥𝑡 𝜏 = 𝑤ℎ𝑒𝑟𝑒 𝜏 ≥ 𝑡
𝑅(𝑡)
• The conditional expectation of the function 𝑓(𝜏) is
∞ ∞
𝑓(𝜏)
𝐸 [𝑇 อ𝑇 ≥ 𝑡] = න 𝜏. 𝑓𝑇ȁ𝑇≥𝑡 𝜏 𝑑𝜏 = න 𝜏. 𝑑𝜏
𝑡 𝑡 𝑅(𝑡)
• Now MRL of a system can be obtained by deducting the survival time t from
conditional expectation function. And expressed as,

𝐿 𝑡 = 𝐸 [𝑇 − 𝑡ȁ𝑇 ≥ 𝑡]

∞ ∞
𝑓 𝜏 𝑓(𝜏)
=න 𝜏−𝑡 . 𝑑𝜏 = න 𝜏. 𝑑𝜏 − 𝑡
𝑡 𝑅 𝑡 𝑡 𝑅(𝑡)
∞
1
𝐿 𝑡 = න 𝜏. 𝑓 𝜏 𝑑𝜏 − 𝑡
𝑅(𝑡) 𝑡
Residual MTTF
𝑅(𝑡ห𝑇0 ) is a reliability function with a condition (i.e system/component
survived till time 𝑇0 ), the Mean (MTTF) will be the mean remaining
lifetime/mean residual life.
Residual MTTF is given by,
∞
𝑀𝑇𝑇𝐹 𝑇0 = න 𝑅 𝑡ห𝑇0 𝑑𝑡
0

∞ 𝑅(𝜗)
= ‫) 𝑇(𝑅 𝑇׬‬ 𝑑𝜗
0 0

1 ∞
= ‫𝜗𝑑)𝜗(𝑅 𝑇׬‬ ; where 𝜗 = 𝑡 + 𝑇0
𝑅(𝑇0 ) 0
Bathtub curve

Decreasing failure Constant failure rate - CFR Increasing failure rate - IFR
- rate - DFR
Summary of the reliability relationship
Constant failure rate models
CFR models (based Exponential Distribution)
• A failure distribution that has a constant failure rate is called an exponential
distribution
• This distribution dominates during the useful-life of the system/component.
• Failures occur here due to random events.
In exponential failure model, the hazard rate 𝜆 𝑡 = 𝜆; where 𝑡 ≥ 𝑜 & 𝜆 > 0
𝑡
𝑅 𝑡 = 𝑒𝑥𝑝 − ‫׬‬0 𝜆 𝑑𝜐 = exp[−𝜆𝑡], 𝑡 ≥ 0
𝑓 𝑡 = 𝜆. exp[−𝜆𝑡]
∞
1
𝑀𝑇𝑇𝐹 = න 𝑒 −𝜆𝑡 𝑑𝑡 =
0 𝜆
∞
2
1 2 −𝜆𝑡 1
𝜎 = න (𝑡 − ) 𝜆𝑒 𝑑𝑡 = 2
0 𝜆 𝜆
• Contd…
From the previous expressions: Standard deviation is 𝜎 2
For Exponential failure rate …
Standard deviation (𝜎) = 1Τ𝜆 i.e. MTTF
So, R(MTTF) = ? Find….
Design-life 𝑡𝑅 is
𝑅(𝑡𝑅 ) = exp −𝜆𝑡𝑅 = 𝑅
1
𝑡𝑅 = − ln 𝑅
𝜆
When R= 0.5 i.e. at 𝑡𝑚𝑒𝑑 ,
1
𝑡𝑚𝑒𝑑 = ln 0.5 = ? Find…
𝜆
The median is always less than the mean in exponential distribution, and so it
is skewed to the right.
Memory-lessness
• In CFR, Time to failure of a system / component is not dependent on how
long the system / component is operating.
• Failure or operating history is irrelevant in this CFR models.
• Mathematically,
𝑅(𝑡+𝑇0 ) exp(−𝜆(𝑡+𝑇0 )
𝑅(𝑡 ቤ𝑇0 ) = = = ? Find…
𝑅(𝑇0 ) exp(−𝜆𝑇0 )
𝑅(𝑡ȁ𝑇0 ) = 𝑅(𝑡)
Time to failure depends only on the length of the operating time (t) and on
the prior-history/age.
Failure modes:
• A system (with different components) i.e. a complex system or component
(different characteristics) displays various “modes of failures”.
Eg.:
Load-carrying bearing or a cutting tool, may fail due to wear-out or when the
applied stress exceeds the design strength of the component material,
resulting in catastrophic-failure*.
(*Catastrophic failure is a failure that destroys the system, such as a missile
failure).
• In failure analysis, the approach should be to segregate the failures according
to the mechanisms (in components) & (in systems) according to components
causing failures.. These are referred as failure modes.
There are three necessary conditions for this model:
(i) failure modes are independent of each other.
(ii) The unit fails when the first of all failure mechanisms reaches the
failure state.
(iii) Each failure mode has its own failure time distribution.
Consider,
A unit that exhibits n failure modes and that the time to failure 𝑇𝑖 due to failure
mechanism i is distributed according to 𝐹𝑖 (𝑡) , i = 1, 2,…, n.
The failure time of the unit is the minimum of {T , T , …, T } and the failure
1 2 n

distribution function F(t) is

The reliability function is,

𝑛

𝑅 𝑡 = ෑ 𝑅𝑖 (𝑡)
𝑖=𝑛 𝑡

𝑅𝑖 𝑡 = exp[− න 𝜆𝑖 𝜗 𝑑𝜗]
0
R(t) is defined as the product of the probabilities, is the probability that none
of the n failure modes occurs before time t.
 𝑛 𝑡
𝑅 𝑡 = ෑ exp[− න 𝜆𝑖 𝜗 𝑑𝜗]
𝑖=𝑛 0

𝑡 𝑛
= exp[− න ෍ 𝜆𝑖 𝜗 𝑑𝜗]
0 𝑖=1

The failure/hazard rate function of 𝑅 𝑡 is,

𝑛

𝜆 𝑡 = ෍ 𝜆𝑖 𝑡
𝑖=1
i.e.,
The hazard rate function of the n independent failure modes can be
determined by summing the n hazard rate functions.
Failure modes with CFR models