Reliability

Resources
Reference books
Ang, A. H-S., and W. H. Tang. Probability Concepts in Engineering
Planning and Design, vol. 1, Basic Principles. New York, NY: John
Wiley & Sons, 1975. ISBN: 9780471032007. In reliability terms, this function gives us the probability that a failure
occurs between time a and time b. This function completely describes
Rausand, M., and A. Hoyland. System Reliability Theory: Models,
the distribution, and is the basis for almost all of the familiar reliability
Statistical Methods, and Applications. 2nd ed. New York, NY: John
and life data functions.
Wiley & Sons, 2003. ISBN: 9780471471332.

G. Habchi and C. Barthod, An overall methodology for reliability

prediction of mechatronic systems design with industrial application. Properties of a density function:
Reliability Engineering & System Safety, 155 (2016), 236-254.
 Must be positive. Obviously a probability cannot be a negative
Website value
 Must be continuous by definition.
https://www.weibull.com.  The area under the entire curve must be equal to 1 (as the sum of
the probabilities values of a law must be equal to1).

Recall of Basic Probability Theory and

Example to better understand PDF
Statistics
Take the example of heights of students in a class. Let’s say that
the height of all the students is between 160 cm and 170 cm. It
Important Terms makes no sense to ask the probability that the height of the student
is EXACTELY 165.84 cm; that probability is zero. We can, however,
Probability density function (pdf) – A continuous representation define the probability that the height of the student lies in the
of a histogram that shows how the number of component failures infinitesimal interval [163;166] cm for example. The function that
are distributed in time. gives this probability density is referred to as the probability density
function or pdf.
Cumulative distribution function (cdf) – Also called the
probability of failure, represents the probability that a brand new
component will fail at or before a specified time.
Cumulative distribution function (cdf)

The cumulative distribution function, cdf, is a function 𝐹(𝑥) of a

From probability and statistics, given a continuous random variable
X, we denote: random variable 𝑋 , and is defined for a number 𝑥 by:

 The probability density PDF, as 𝑓(𝑥). 𝑥

 The cumulative distribution CDF, as 𝐹(𝑥) 𝐹(𝑥) = 𝑃(𝑋 ≤ 𝑥) = ∫0,−∞ 𝑓(𝑠) 𝑑𝑠
 The pdf and cdf give a complete description of the probability
distribution of a random variable.

That is, for a given value 𝑥 , 𝐹(𝑥) is the probability that the observed

Probability density function (pdf) value of 𝑋 will be at most 𝑥 . Evaluated at 𝑥 , is the probability that 𝑋
will take a value less than or equal to 𝑥 .
If X is a continuous random variable, then the probability density The cumulative distribution function, 𝐹(𝑥), denotes the area beneath
function, pdf, of 𝑋 is a function 𝑓(𝑥) such that for two numbers, a and the probability density function to the left of 𝑥 .
b with a ≤ b:

𝑏 PDF and CDF relationship

𝑃(𝑎 ≤ 𝑋 ≤ 𝑏) = ∫ 𝑓(𝑥) 𝑑𝑥 Probability looks at probability at one point, while Cumulative is the
𝑎 total probability of anything below it. As you can see in the diagram
below, the cumulative is much greater than the just probability because
In other words, the pdf defines the probability that 𝑋 takes on a value it is the sum of many, and not just of one probabilities.
in the interval [a,b] is the area under the density function from a to b.
This is represented graphically in the following plot.
Following is a graphical representation of the relationship between the
PDF and CDF.

F(t) in this particular case with the exponential distribution, is the

Failure function of a product when the failure rate is constant. The
failure density is linked to the failure rate by the following relation:
𝑓(𝑡) = λ(t). [1 − 𝐹(𝑡)]. Details are provided further.

Important Terms

Failure – The number of failures per unit time that can be expected
to occur for the product.

Lifetime – A statistical measure (or estimate) of how long a product

is expected to perform its intended functions under a specific set of
environmental, electrical and mechanical conditions. Lifetime
specifications can only describe the behaviour of a population; any
single product may fail before or after the rated lifetime.

Mean Time Between Failures (MTBF) – The average time

The following figure illustrates the relationship between the reliability between failures during useful life for repairable or redundant
function and the CDF, or the unreliability function. systems.

Mean Time To Failure (MTTF) – The average time to failure during

useful life for components or non-repairable systems.

Reliability – A statistical measure (or estimate) of the ability of a

product to perform its intended functions under a specific set of
environmental, electrical, and mechanical conditions, for a specific
period of time. Reliability estimates for the entire useful life phase
of a product are commonly reported using MTBF or MTTF.

Serviceability – The ability of a product to be repaired by regular

maintenance personnel, typically through replacement of a
subsystem or one or more associated components

Example with the exponential distribution

The probability density function (PDF) of exponential distribution is: Reliability Functions
f(t) = λ. e(−λ.t) Reliability
The exponential cumulative distribution function can be derived from The term “reliability” in engineering refers to the probability that a
its PDF as follows : product, or system, will perform it’s designed functions under a given
set of operating conditions for a specific period of time. It is also known
as the “probability of survival”.

Reliability is defined as the probability that an individual unit of the

product, operating under specified conditions, will work correctly for a
specified length of time. Hence reliability is the probability of failure in
the flat central part of the familiar bathtub curve shown below.
Several distributions defined by parameters are commonly used for
reliability models, including:

 Exponential
 Weibull
 Gamma
 normal (Gaussian)
 log-normal
 log-logistic.

The choice of parametric distribution for a particular application can be

made using graphical methods or using formal tests of fit.

The exponential law is well adapted when the failure rate (λ) is Scale parameter η
constant, (i.e. it is independent of time). The normal law represents
behaviours where the lifetime of the population is homogeneous, the The scale parameter, or spread, η, sometimes also called the
probability of failure is centred and symmetrical. It can be used to characteristic life, represents the typical time-to-failure in Weibull
model maturity behaviour (without breakdowns) and then rapid wear analysis. It is related to the Mean-Time-to-Failure (MTTF). In Weibull
and tear. The Weibull's law models each of the three phases of a analysis, η is defined as the time at which 63.2% of the products will
material's life. It generalises the two previous laws but is more difficult have failed
to use and interpret.
Location parameter γ
Exponential model It is expressed in units of time. It indicates the date of the occurrence
of the failure mode characterised by β.
The exponential distribution is most widely used distribution in
reliability and risk. Weibull functions
𝑡−𝛾 𝛽
Weibull model Reliability [−(
𝜂
) ]
R(t) = e
Named after the Swedish professor Waloddi Weibull (1887-1979), the 𝑡−𝛾 𝛽
Failure [−( ) ]
𝜂
Weibull distribution is one of the most widely used life distributions in F(t)=1-R(t)= 1 − e
reliability analysis.  1
  t    
Failure probability dF (t )   t   
f (t )     exp    
density dt         
Weibull analysis is able to answer many life cycle engineering
problems such as mean life estimation, reliability of products at any
operational time and warranty cost estimation etc. The advantages of
Weibull model in life data analysis can be extended to facilitate Failure rate
decision-making processes in many remanufacturing practices such
as prediction of the number of cores returned for remanufacturing, A way to look at the failure behaviour in time is to examine the failure
estimation of spare parts or remedy resources needed for each failure rate. Failure rate is function of time. For instance, over a year, a
mode, and so on. bearing used in a car might have a higher failure rate in summer than
The Weibull distribution has three parameters: in winter.

• Shape parameter β (beta) Failure rates can be expressed using any measure of time, but hours
is the most common unit in practice. Other units, such as km,
• Scaling parameter η (eta)
revolutions, etc., can also be used in place of "time" units.
• Position parameter γ (gamma)
Mathematically, the failure rate function is a conditional form of the pdf,
Among the two parameters, the slope of the Weibull distribution, β, is as seen in the following equation:
very important as it determines which member of the family of Weibull
failure distributions best fits or describes the data. It also indicates the
class of failures in the “bathtub curve” failure modes. 𝑓(𝑡)
λ(t) =
The Weibull shape parameter β indicates whether the failure rate is 𝑅(𝑡)
increasing, constant or decreasing.
• β < 1 indicates infant mortality;
under the condition is that no failure has occurred before time t.
• β = 1 means random failures (i.e. independent of time);
• β > 1 indicates wear-out failures. then
The information about the β value is extremely useful for reliability
centered maintenance planning and product life cycle management.
This is because it can provide a clue to the physics of the failures and
tell the analyst whether or not scheduled inspections and overhauls
are needed. For instance, if β is less than or equal to one, overhauls
are not cost effective. With β greater than one, the overhaul period or
scheduled inspection interval can be read directly from the plot at an over a time interval Δt = (t2 − t1) from t1 (or t) to t2.
acceptable or allowable probability of failures. For wear-out failure
modes, if the cost of an unplanned failure is much greater than the cost
of a planned replacement, there will be an optimum replacement This curve is usually called a bathtub curve after its characteristic
interval for minimum cost. shape. The failure rate is often high in the initial phase. This can be
explained by the fact that there may be undiscovered defects (known
as “infant mortality”) in the items; these soon show up when the items
are activated. When the item has survived the infant mortality period,
the failure rate often stabilizes at a level where it remains for a certain
amount of time until it starts to increase as the items begin to wear out.
From the shape of the bathtub curve, the lifetime of an item may be
divided into three typical intervals: the burn-in period, the useful life
Nt
period and the wear-out period. The useful life period is also called the Average ranks if 20<N<=50 F (t ) 
chance failure period. N 1
N
Constant Failure Rate Reliability Models Cumulative frequencies if N>50 F (t )  t
N

Important Terms Example (N = 100)  Cumulative frequencies

N – total number of products put into operation at time t0

(sample size)
N0 – number of failures at t0 (generally equal to 0)
N1 – number of failures at t1
N2 – number of failures at t2
DN – number of failures on the interval [t1, t2]
Dt – time interval [t1, t2]
Reliability Estimators: R(t)
Point estimator at time t

Whatever the value of N, first compute the estimator of F and then

apply the complementarity relation
If the failure rate is constant for all times, λ(t) = λο, often referred to as
“failure rate on average”, the corresponding reliability and probability R(t) = 1 - F(t)
density function are exponential functions given by:
Example (N = 100)
(−λ0 .t)
R(t) = e

f(t) = λ0 . e(−λ0 .t)

The mean time to failure is readily obtained as:

1
MTTF = 𝐸(𝑡) =
𝜆0

While MTBF can be a useful metric to estimate reliability, especially

when comparing components from multiple vendors, the trap many
Density estimators: f(t)
people fall into is wrongly assuming that the MTBF figure directly
Interval estimator (failure/product and unit of time)
equates to the expected life of the component.

N
Median ranks if 1<N<=20 f t1;t 2 
Estimators ( N  0.4)  t
N
It is obviously not realistic to calculate failure rates by building many
Average ranks if 20<N<=50 f t1;t 2 
units and running them for many hours, under expected operating
conditions. This is especially true for well-designed and properly built
( N  1)  t
supplies, with extremely low failure rates, where the number of
N
supplies and hours required to get valid results would be in the Cumulative frequencies if N>50 f t1;t 2 
thousands. Instead based on representative samples of a population N  t
statistical analysis techniques can be used to estimate failure rates.

Reliability function estimators depend on the value of N (number of Example (with N = 100)
products put into operation)

Estimator N
Median ranks 1<N<=20
Average ranks 20<N<=50
Cumulative frequencies N>50

Failure Estimators: F(t)

Point estimator at time t

N t  0.3
Median ranks if 1<N<=20 F (t ) 
N  0.4
Predictive Reliability
If the components are in parallel, system performs if any one
Predictive reliability (predicted or theoretical) is calculated on the component remains operational. The corresponding reliability block
basis of a mathematical model defined from the functional diagram is:
decomposition of the system into subsets, components, etc. and the
estimated or predicted reliability of its components. The analysis and
calculation steps are usually the following:

 Decompose the system into components (parts, subsystems ...)

and establish the functional links between the components
 Identify component reliability models or collect reliability at a given
time
 Search for a model of the system and calculate its reliability

Reliability block diagram (RBD)

A reliability block diagram is a success-oriented network describing the
function of the system. It shows the logical connections of (functioning)
components needed to fulfill a specified system function
If there are n components in parallel, where the probability of failure of
Series Structure the i-th component is denoted by Fi, the probability of system failure is:

In a series configuration, a failure of any component results in the 𝑛

failure of the entire system. Since the components are independent, 𝐹𝑠 = 𝐹1 × 𝐹2 × ⋯ × 𝐹𝑛 = ∏ 𝐹𝑖
the corresponding reliability block diagram is:
𝑖=1

the system reliability is then:

𝑛 𝑛

𝑅𝑠 = 1 − ∏ 𝐹𝑖 = 1 − ∏(1 − 𝑅𝑖 )
𝑖=1 𝑖=1
In other words, for a pure series system, if there are n components in
series, where the reliabilities are independent of the i-th component is This model assumes that there is statistical independence among
denoted by Ri, the system reliability is: failures of the subsystems. This assumption is important because
dependence between subsystems can have a significant effect on
𝑛 system reliability.
𝑅𝑠 = 𝑅1 × 𝑅2 × ⋯ × 𝑅𝑛 = ∏ 𝑅𝑖
𝑖=1

Redundancy
i.e. the system reliability is equal to the product of the reliabilities of its
constituent components. Redundancy is the surest method of preventing system failure. There
are four kinds of redundancy: diverse, homogenous, active, and
passive.
Example
Active redundancy guards against both system and element failure. It
An electronic system has 1000 components in reliability series. also allows for element failure, repair, and substitution with minimal
The reliability of each component is 0.999. Determine System disruption of system performance.
reliability.
Passive redundancy is ideal for noncritical elements, but it will result in
Solution system failure when used for elements critical to system operation.
Passive redundancy is the simplest and most common kind of
1000 redundancy.
𝑅𝑠 = ∏ 0.999 = (0.999)1000 = 0.368
𝑖=1 Diverse redundancy is resistant to a single cause of failure, but is
complex to implement and maintain. For the train, a single cause is
Commentary unlikely to result in a cascade failure in all three braking systems.
1.Even though a component reliability of 0.999 sounds good, Homogenous redundancy is relatively simple to implement and
the sheer number of these components causes a low system maintain but is susceptible to single causes of failure — i.e., the type
reliability. of cause that results in failure in one element can result in failure of
other redundant elements.
2.Even though 0.999~1.0, the difference is crucial. For high
reliability values, the probability of failure often gives a clearer
picture. For example, increasing the component reliability from
0.999 to 0.9999 re-quires a ten-fold reduction of the failure
probability.

Parallel Structure
The reliability of a system containing independent subsystems can
usually be increased by using subsystems in a redundant way.
Use diverse redundancy for critical systems when the probable causes n
of failure cannot be anticipated. Use homogenous redundancy when The parallel system (n
components)
R (t )  1   (1  e (  i t ) )
the probable causes of failure can be anticipated. Use active i 1
redundancy for critical systems that must maintain stable performance
in the event of element failure or extreme changes in system load. Use The parallel system (2
components) R(t )  e1t  e2t  e(1 2 )t
passive redundancy for noncritical elements within systems, or
systems in which performance interruptions are tolerable.

Important Terms Classification of failures

Redundancy – Use of more elements than necessary to Parial: the system has a degraded performance, but is still able to
maintain the performance of a system in the event of failure of perform its essential functions.
one or more of the elements.
Complete: failure of one or several items cause immediate cessation
Active Redundancy – Application of redundant elements at of the system ability to perform the required function.
all times (e.g., using multiple independent pillars to support a
roof). All components are active from the start. Only one Failure Classification in ISO TR 12849
component working allows the functioning of the whole
ISO TR 12489 adds to the category random failures, to capture that
system.
some systematic faults (induced by human errors) can be regarded as
Passive Redundancy – Application of redundant elements random
only when an active element fails (e.g., using a spare tire on a
vehicle in the event of a flat tire). Only the first component is
activated. If it fails, then the second component is activated.
The whole system fails when the last component fails.

Diverse Redundancy – use of multiple elements of different

types (e.g., high-speed trains often have diverse redundancy
in their braking systems—one electric brake, one hydraulic
brake, and one pneumatic brake)

Homogenous Redundancy – Use of multiple elements of a

single type (e.g., use of multiple independent strands to
compose a rope).
Other mean of failure classifications

K-out-N redundancy system

A system that is functioning if and only if at least k of the n components
are functioning

R1, R2, ... Rn are the elementary independent reliabilities of the

components of the system at a given time.
𝑛

𝑅𝑆 = ∑( 𝐶𝑛𝑖 . 𝑅 𝑖 . (1 − 𝑅)𝑛−𝑖 )
𝑖=𝑘

n: total number of components

𝑅: reliability of a component Failure classification (adapted from Blanche and Shrivastava 1994).
k: number if components to operate simultaneously
𝑛!
with: 𝐶𝑛𝑖 =
𝑘!(𝑛−𝑘)!

⚠ This relation (using binomial law) is applicable only if the R are

identical, otherwise, application of the truth table.

Mixed Models
One system configuration that is often encountered is one in which
subsystems are in series, but redundancy (active) is applied to a
certain critical sub-system(s). A typical block diagram follows:

Application to the exponential model

R(t) = e(−λ0 .t)

F(t) = 1−e(−λ.t) with: λ = 1/E(t) (failure rate)

Serial system
( t  i )
R (t )  e i 1
 e (  t S )