Reliability Definition

In engineering, reliability is the probability that a product, system, or component will perform
its intended function adequately, without failure, for a specified period of time under stated
operating conditions.
This definition has four key parts:
Probability: A value between 0 and 1 representing the likelihood of success.
Intended Function: The specific task(s) the item is designed to do.
Specified Period of Time: The duration (e.g., hours, cycles, miles) over which reliability is
measured.
Stated Conditions: The environmental and operational conditions (e.g., temperature, load,
speed) it must operate within.

Quality and Reliability

Quality and reliability are closely related but distinct.
Quality measures how well a product conforms to its specifications at the time of delivery (a
static snapshot). It answers the question: "Does it work as intended right now?"
Reliability is essentially quality over time (a dynamic measure). It answers the question: "Will
it keep working as intended for a certain amount of time?"
A product can have high quality (it works perfectly out of the box) but low reliability (it fails
quickly). A high-reliability product implies high quality from the start.

Reliability Mathematics
Reliability is quantified using probability and statistics. Key mathematical functions describe the
failure characteristics of a component or system.
Reliability Function, .R(t).
The Reliability Function, denoted as .R(t)., is the probability that a component or system will
survive (i.e., not fail) beyond a specific time, .t..
At the beginning (.t=0.), .R(0) = 1. (it is 100% certain to be working).
As time increases, .R(t). decreases, approaching 0.
Failure Function, .F(t).
This is the Cumulative Distribution Function (CDF) of failure. It is the probability that the
component will fail by time .t..
It is the complement of the reliability function: .F(t) = 1 - R(t)..
At .t=0., .F(0) = 0. (probability of failure is zero).
As .t \to \infty., .F(t) \to 1..
Failure Density Function, .f(t).
This is the Probability Density Function (PDF) of failure. It represents the rate at which
failures occur over time. Mathematically, it is the derivative of the failure function .F(t)..
..f(t) = \frac{dF(t)}{dt} = -\frac{dR(t)}{dt}..
The probability of failing in a small interval of time .dt. is .f(t)dt..

Hazard Rate, .\lambda(t).

The Hazard Rate (also called the instantaneous failure rate) is a fundamental concept. It
represents the likelihood of failure for a component at time .t., given that it has successfully
survived up to time .t..
It answers the question: "What is the probability this item, which is working right now, will fail in the
next instant?"

The hazard rate is defined as:

..\lambda(t) = \frac{f(t)}{R(t)}..
Where:
.f(t). is the failure density function.
.R(t). is the reliability function.

Measures of Reliability
These are specific metrics used to quantify reliability:
Mean Time To Failure (MTTF): The average time a non-repairable component is expected to
operate before it fails. This applies to items that are replaced, not fixed (e.g., a lightbulb, a fuse).
Mean Time Between Failures (MTBF): The average time a repairable system operates
between one failure and the next. MTBF includes the time to repair (Mean Time To Repair, or
MTTR).
.MTBF = MTTF + MTTR. (Note: In some contexts, MTBF is used interchangeably with MTTF,
but the distinction for repairable systems is important).
Failure Rate (.\lambda.): The frequency of failures, often expressed as failures per unit of time
(e.g., failures per million hours). In the "Useful Life" period (see below), this rate is often
assumed to be constant, and in that case, .MTTF = 1/\lambda..

Key Life Cycle Concepts

Component Mortality and the Mortality Curve
Component "mortality" describes the failure behavior of a population of components over their
entire lifespan. This is most famously represented by the Mortality Curve, also known as the
Bathtub Curve. This graph plots the hazard rate .\lambda(t). over time and has three distinct
phases:
Infant Mortality (Early-Life Failures):
The failure rate is high but decreasing.
Failures are typically due to manufacturing defects, poor-quality materials, or assembly errors.
Reliable manufacturers use a "burn-in" period to test products and weed out these early failures
before they reach the customer.
Useful Life (Random Failures):
The failure rate is low and relatively constant.
Failures are considered "random" and are caused by sudden, unpredictable stresses or events.
This is the main operating life of the product, and most reliability calculations (like MTTF) are
based on this constant failure rate.
Wear-Out (Age-Related Failures):
The failure rate increases rapidly.
Failures are caused by aging, fatigue, corrosion, or depletion of materials.
The component is nearing the end of its intended life.
Design Life
The Design Life is the period of time for which a product is designed to function within its
specified performance levels without requiring major repairs. It is the intended operational
lifespan as determined by the engineers and is often based on the onset of the wear-out period.
Useful Life
The Useful Life directly corresponds to the flat, bottom portion of the bathtub curve (Phase 2). It
is the period after infant mortality has passed and before the wear-out phase begins,
characterized by a constant, low, random failure rate.

A Priori and A Posteriori Probabilities

These terms relate to how a probability is determined.
A Priori Probability: A probability derived from deductive reasoning or prior knowledge,
before any experimental data is collected.
Example: The probability of a fair coin landing on "heads" is 0.5. This is known from the
physical properties of the coin (two sides, equal chance) without needing to flip it.
A Posteriori Probability: A probability derived empirically after collecting and analyzing
experimental data. It is an updated probability based on observed evidence.
Example: If you flip a coin 100 times and it lands on "heads" 60 times, the a posteriori
probability of heads for that specific coin is 60/100 or 0.6. This new evidence might cause you to
revise your a priori assumption that the coin was fair. This concept is a cornerstone of Bayesian
statistics.

Here is a breakdown of the concepts involved in Life Data Analysis.

Life Data Analysis (LDA)

Life Data Analysis (also called "Weibull Analysis" or "survival analysis") is a field of statistics
used to analyze "time-to-failure" data. The primary goal is to model the failure behavior of a
component or system to answer questions like:

 How reliable is this product?

 What is its average life (e.g., MTTF)?
 What percentage will fail during the warranty period?
 When does the product start to wear out?

1. Data Collection and Types

The first step is collecting failure data, which comes in several forms. The two main
classifications are completeness and grouping.

Data Type Description Example

You test 10 lightbulbs and run them until all

We know the exact time-to-failure for
Complete 10 have burned out, recording each exact
every unit in our test sample.
time.

We have incomplete information

about some or all of the failure times.
Censored
This is the most common type of
data.

You test 10 lightbulbs for 1000 hours. At the

A unit is removed from the test
Right- end, 8 have failed, but 2 are still working.
before it fails. We only know it
Censored The 2 working bulbs are right-censored at
survived at least .t. hours.
1000 hours.

You inspect a component at 100 hours and

We know a unit failed sometime
Left-Censored find it has already failed. You don't know if
before a specific inspection time.
it failed at 10, 50, or 99 hours.

You check a component at 100 hours (it's

Interval- We know a unit failed between two working) and again at 200 hours (it's failed).
Censored inspection times. The failure occurred between 100 and 200
hours.

Each failure and censoring time is Failures at 105, 120, 132 hours. Censored at
Ungrouped
recorded individually. 150, 150 hours.

0-100 hours: 5 failures 100-200 hours: 12

Grouped Data is recorded in time intervals.
failures

2. Non-Parametric Methods
These methods analyze data without assuming an underlying statistical distribution (like
Exponential or Weibull). They are used to get a direct, model-free estimate of reliability from the
data.

 For Ungrouped Data (Kaplan-Meier Estimator): This is the most common non-
parametric method. It creates a "step-function" showing the estimated reliability .R(t).
over time. The reliability estimate only changes at the time of an actual failure. It's
excellent for handling right-censored data.
 For Grouped Data (Actuarial or Life-Table Analysis): This method calculates the
probability of survival for each time interval, and the overall reliability is the product of
these interval probabilities.

3. Time-to-Failure Distributions (Parametric)

This is the more powerful approach, where we assume the failure data follows a specific
probability distribution. By fitting the data to the distribution, we can extract its parameters and
make predictions.

Exponential Distribution

 Key Feature: It has only one parameter, the constant failure rate (.\lambda.).
 Reliability Function: .R(t) = e^{-\lambda t}.
 Use Case: This distribution models the "Useful Life" (random failure) portion of the
bathtub curve. It has a "memoryless" property, meaning an old component is just as likely
to fail in the next hour as a brand-new one. If this distribution fits, the MTTF is
simply .1/\lambda..

Weibull Distribution

 Key Feature: This is the most flexible and widely used distribution in reliability. It has
two main parameters:
o .\beta. (Beta) — Shape Parameter: This is the most important parameter. It
describes the failure characteristic.
 .\beta < 1.: Decreasing failure rate. Models infant mortality (defective
parts failing early).
 .\beta = 1.: Constant failure rate. The Weibull distribution becomes the
Exponential distribution.
 .\beta > 1.: Increasing failure rate. This is the most common case and
models wear-out.
o .\eta. (Eta) — Scale Parameter (or Characteristic Life): This is the time at
which 63.2% of the population is expected to have failed.
 Reliability Function: .R(t) = e^{-(t/\eta)^\beta}.
 Use Case: Because .\beta. can be changed, the Weibull distribution can model all three
phases of the bathtub curve (infant mortality, useful life, and wear-out).
4. Probability Plotting

This is a graphical technique used to determine if a dataset fits a specific distribution and to
estimate that distribution's parameters.

1. Process: The failure times are sorted and plotted on special "probability paper." The axes
of this paper are transformed (e.g., using logarithms) so that if the data perfectly follows
the chosen distribution (like Weibull), the data points will form a straight line.
2. Analysis (for a Weibull plot):
o If the points are scattered and do not form a line, the data does not fit a Weibull
distribution.
o If the points do form a straight line:
 The slope of the line is a direct estimate of the shape parameter (.\beta.).
 The intercept of the line is used to find the characteristic life (.\eta.).

5. Goodness of Fit (GoF) Tests

While a probability plot is a good visual check, a GoF test provides a quantitative, statistical
measure of how well the line fits the data points.

 Hypothesis: These tests check the "null hypothesis" that the data was, in fact, drawn
from the chosen distribution.
 Common Tests:
o Correlation Coefficient (.R^2.): A simple measure (from 0 to 1) of how well the
points fit the straight line on the probability plot. A value close to 1 is good.
o Kolmogorov-Smirnov (K-S) Test: Compares the data's empirical cumulative
distribution (like the Kaplan-Meier steps) to the fitted distribution's cumulative
distribution.
o Anderson-Darling (A-D) Test: A more sophisticated test that is particularly
sensitive to mismatches in the "tails" of the distribution (the very early or very
late failures), making it very popular for reliability analysis.
Reliability assessment is the process of using analytical and logical models to predict the
reliability of a system, often based on the reliability of its individual components.

Different Configurations & Redundancy

Redundancy is the core concept of improving system reliability by adding extra components to
take over if a primary component fails. The way these components are arranged is called the
system configuration.

 Series Configuration: Components are arranged in a chain. The system functions only if
all components function. This is the simplest configuration but also the least reliable.
o Reliability: .R_{System} = R_1 \times R_2 \times \dots \times R_n.
o The system's reliability is always lower than that of its least reliable component.
 Parallel Configuration (Active Redundancy): Components are arranged in parallel
paths. The system functions if at least one component functions. This is the most
common form of redundancy.
o Reliability: .R_{System} = 1 - F_{System} = 1 - (F_1 \times F_2 \times \dots \
times F_n).
o Where .F = (1 - R). is the probability of failure.
o The system's reliability is always higher than that of its most reliable component.

k-out-of-n System
This is a more generalized form of redundancy. The system has .n. identical components and
functions if at least .k. of them are working.

 Series System: This is an .n.-out-of-.n. system.

 Parallel System: This is a .1.-out-of-.n. system.
 Example: A jet with four engines that can fly safely as long as at least two engines are
working (a 2-out-of-4 system).

If all components have the same reliability (.R.), the system reliability is calculated using the
binomial distribution:

..R_{sys} = \sum_{i=k}^{n} \binom{n}{i} R^i (1-R)^{n-i}..

This formula sums the probabilities of having exactly .k. successes, .k+1. successes, ..., all the
way to .n. successes.
Complex Systems: Analysis Methods
When a system is not a simple series/parallel arrangement (e.g., it has "bridge" networks), more
advanced methods are needed.

1. Reliability Block Diagrams (RBD)

An RBD is a success-oriented graphical tool.

 It visualizes the system as a set of blocks, where each block represents a component.
 The blocks are connected in a way that shows the logical paths required for the system to
operate.
 The system is considered successful if there is at least one continuous path from the Start
node to the End node.

2. Cut and Tie Sets

These are methods used to analyze a complex RBD.

 Tie Set: A minimal set of components that, if all are working, guarantees the system
works (a "success path"). The system is considered working if at least one tie set is
working.
o Analogy: Tie sets are like parallel paths.
 Cut Set: A minimal set of components that, if all have failed, guarantees the system fails
(a "failure cut"). The system is considered failed if at least one cut set has failed.
o Analogy: Cut sets are like series components in the failure domain.

By identifying all minimal cut or tie sets, a complex RBD can be broken down into a solvable
series-parallel combination.

3. Fault Trees (FTA)

A Fault Tree is the logical inverse of an RBD. It is a top-down, failure-oriented graphical

method.

 It starts with a single, undesired "Top Event" (e.g., "System Fails").

 It then uses logic gates (like AND and OR) to break down this event into its contributing
causes, moving down to the level of basic component failures.
o OR Gate: The output event occurs if any of the input events occur. (This
corresponds to a series system, where any one failure causes the system to fail).
o AND Gate: The output event occurs only if all of the input events occur. (This
corresponds to a parallel system, where all components must fail for the system to
fail).
 FTA is excellent for identifying the root causes and combinations of failures that lead to a
critical system failure.
4. Baye's (Bayes') Approach

This uses Bayes' Theorem to update reliability estimates. It allows you to formally combine a
priori knowledge (e.g., a component's reliability from its design specifications) with a posteriori
evidence (e.g., new data from field tests).

..P(H|E) = \frac{P(E|H) \times P(H)}{P(E)}..

In reliability:

 .P(H). (Prior) = Our initial belief about the component's reliability.

 .E. (Evidence) = New test data (e.g., 5 failures in 1000 hours).
 .P(H|E). (Posterior) = Our updated belief about the reliability, having seen the new
evidence.

This is a powerful way to refine reliability predictions as a product's life cycle progresses.

Standby Systems
This is a special, highly effective type of redundancy where the backup component is not in use
until the primary one fails.

 Active Redundancy (Parallel): All components are on and sharing the load. They all
age and can fail simultaneously.
 Standby Redundancy: A primary component (.A.) operates while a secondary
component (.B.) is dormant (in "standby"). A switch detects the failure of .A. and
activates .B. to take its place.

The reliability of a standby system depends on:

1. Reliability of the Primary Unit (.R_A.)

2. Reliability of the Standby Unit (.R_B.)
3. Reliability of the Switch (.R_{SW}.)
4. Standby State:
o Cold Standby: The standby unit cannot fail while it is off (.R_B. in standby = 1).
o Warm/Hot Standby: The standby unit is powered on (or partially) and can fail,
even if not in use (though often at a lower rate).

Standby systems are generally much more reliable than active parallel systems, if the switch is
reliable and the standby unit doesn't fail while dormant.
Reliability monitoring involves a set of activities used to track, assess, and improve a product's
reliability throughout its lifecycle, from development to field operation.

📈 Life Testing Methods

Life testing is the primary way to gather reliability data. It involves running a sample of
components under specified conditions to observe their failure behavior.

 Failure-Terminated Test: The test is run until a pre-determined number of failures

(.r.) occurs.
o Pro: You get a fixed amount of failure data, which simplifies statistical analysis.
o Con: The test duration is unknown. If the product is highly reliable, the test could
take an impractically long time.
 Time-Terminated Test: The test is run for a pre-determined amount of time (.t.).
o Pro: The test schedule and cost are fixed, which is good for project planning.
o Con: The number of failures is variable. If the product is very reliable, you might
get zero failures, which provides less statistical insight than a failure-terminated
test.
 Sequential Testing: This is an adaptive method where the sample size is not fixed in
advance. After each component (or group of components) is tested, a statistical analysis
is performed to make one of three decisions:

1. Accept: The reliability goal has been met.

2. Reject: The product has failed to meet the reliability goal.
3. Continue Testing: The data is not yet sufficient to make a decision.

o Benefit: This method is highly efficient and, on average, requires a smaller

sample size than fixed-time or fixed-failure tests to reach a conclusion.

📊 Reliability Growth Monitoring

Reliability Growth is the intentional, positive improvement in a product's reliability during its
development. This is achieved through a structured process:

1. Test: Run the product under operational loads.

2. Find: Identify failure modes when they occur.
3. Analyze: Perform root cause analysis (RCA) on the failures.
4. Fix: Implement design or manufacturing changes to eliminate the root cause.
5. Verify: Continue testing to ensure the fix was effective and didn't introduce new
problems.
This process is tracked using Reliability Growth Models (like the Duane or AMSAA-Crow
models), which plot the product's reliability (e.g., MTBF) over cumulative test time. The goal is
to show a clear upward trend, demonstrating that the product's reliability is "growing" toward its
target.

🎯 Reliability Allocation
Reliability Allocation is a top-down planning activity done early in the design phase. It
involves:

1. Starting with the overall system-level reliability requirement (e.g., the car must have
99% reliability for a 5-year mission).
2. Apportioning or "allocating" this top-level requirement down to the individual
subsystems and components.

For example, the 99% system reliability goal might be allocated as:

 Engine: 99.9%
 Transmission: 99.8%
 Electronics: 99.5%

The allocation is based on factors like component complexity, criticality, cost, and the state of
the art. This process gives each design team a specific, quantitative reliability target to meet.

💻 Software Reliability
This field focuses on the probability of failure-free software operation for a specified period in
a specified environment.

Key Differences from Hardware:

 Software doesn't "wear out." It fails due to hidden defects (bugs) in the code.
 Failures are not caused by physical stress but by encountering a specific, unexpected
combination of inputs or states that trigger a fault.
 Reliability is improved by finding and removing bugs (faults), not by replacing a
physical part.

Software reliability is often modeled using Software Reliability Growth Models (SRGMs),
which track the failure rate as bugs are found and fixed during the testing phase.
🧑 Human Reliability
Human Reliability is the probability that a person will correctly perform a required task under
specified conditions, without error.

Human Reliability Analysis (HRA) is a field dedicated to assessing and mitigating human
error. This is critical in systems where human actions can lead to catastrophic failure (e.g.,
nuclear power plants, aviation, medicine).

HRA involves:

 Identifying critical tasks performed by people.

 Analyzing the ways a person could fail at that task (e.g., omission, commission, incorrect
sequence).
 Quantifying the probability of that error (Human Error Probability, or HEP).
 Identifying performance-shaping factors (PSFs) like stress, poor training, confusing
procedures, or a bad user interface, and implementing improvements to reduce the
likelihood of error.
🔧 Reliability Improvement through Maintainability and
Availability
Reliability improvement is strongly supported by enhancing Maintainability, the measure of
how quickly and easily a failed system can be returned to service. The concepts below link the
time a system is "up" (Reliability) with the time it takes to get it "back up" (Maintainability) to
determine overall Availability.

🛑 Analysis of Downtime
Downtime is the total time a system is non-operational. Analyzing it is the first step toward
improvement. Downtime is categorized as:

 Scheduled Downtime: Planned interruptions for preventive maintenance, inspections, or

upgrades. This is necessary and controlled.
 Unscheduled Downtime (Breakdown): Unexpected stoppages due to failure. This is the
focus of reliability/maintainability improvement.

The analysis involves tracking and performing Root Cause Analysis (RCA) on the unexpected
downtimes to identify and eliminate the underlying failure causes. Key metrics include:

 Mean Downtime (MDT): The total time an asset is down divided by the number of
downtime events. This includes administrative time, logistics time, and repair time.

Repair Time Distribution and Maintainability

Repair Time Distribution

The time required to perform a corrective maintenance action (repair) is a random variable, just
like the time to failure. The time it takes to complete a repair is modeled using a Repair Time
Distribution.

 For many systems, the repair time distribution is assumed to be log-normal or

exponential.
 The key parameter derived from this distribution is the Mean Time To Repair (MTTR),
which is the average time spent on the active repair phase (diagnosis, disassembly,
replacement, reassembly, and checkout).

..\text{MTTR} = \frac{\sum (\text{Corrective Maintenance Time})}{\text{Total Number of

Failures}}..
System Repair Time

For a complex system composed of multiple components, the system's repair time is the time
required to diagnose the failure and repair or replace the failed component(s). This is driven by:

 Fault Detection and Isolation (FD&I): How quickly and accurately the maintenance
crew can find the failed item.
 Component Accessibility: How easy it is to reach the failed component.
 Modular Design: Using easily replaceable modules to minimize swap-out time.

📏 Measures and Prediction of Maintainability

Maintainability Prediction

This is an analytical technique, often performed early in the design stage (using standards like
MIL-HDBK-472), to forecast how long a system will take to repair. It involves:

1. Breaking down a maintenance action into a sequence of elementary tasks (e.g., open
panel, diagnose, remove bolt, replace part).
2. Estimating the time required for each task.
3. Summing these task times to predict the total repair time and calculate the expected .\
text{MTTR}. for the system.

Measures of Maintainability

The most common quantitative measure is Mean Time To Repair (MTTR). Other measures
include:

 Mean Corrective Maintenance Time (MCMT): Average time for

unscheduled/corrective maintenance. Often used interchangeably with MTTR.
 Mean Preventive Maintenance Time (MPMT): Average time for scheduled/preventive
maintenance.
 Mean Man-Hours per Repair (MMH/Repair): Maintenance labor effort required for
one repair action.

🧐 Inspection Decisions
Inspection Decisions are part of a proactive maintenance strategy (Preventive or Predictive
Maintenance). They involve determining:
  When to inspect: Based on reliability data (e.g., failure probability) or condition
monitoring data (e.g., vibration level, temperature).
 What to inspect: Focusing on critical components identified through Failure Mode and
Effects Analysis (FMEA) or high-risk areas.
 The criteria for intervention: When an inspection reveals a defect, the decision must be
made whether to repair immediately, schedule a repair, or continue operation.

The goal is to find defects before they cause a functional failure (Condition-Based
Maintenance) to minimize unscheduled downtime.

🟢 System Availability
System Availability is the probability that a system or component is performing its required
function at a stated instant or over a stated period of time. It is the ultimate metric linking
reliability and maintainability.

For systems that can be immediately repaired upon failure (restorable systems), Inherent
Availability (.\text{A}_i.) is typically calculated using the following steady-state formula:

..\text{A}_i = \frac{\text{MTBF}}{\text{MTBF} + \text{MTTR}}..

Where:

 MTBF (Mean Time Between Failures): The measure of Reliability.

 MTTR (Mean Time To Repair): The measure of Maintainability.

To improve System Availability, you must:

1. Increase Reliability (MTBF): Reduce the frequency of failure.

2. Increase Maintainability (Reduce MTTR): Decrease the time it takes to recover from a
failure.

The availability of .99.999\%. ("five nines") is a common goal for high-reliability systems,
which translates to a total downtime of less than 5.26 minutes per year.