Section 7a Reliability Notes
Section 7a Reliability Notes
Section
7
Product Excellence using 6 Sigma (PEUSS)
Introduction
to
Reliability
Warwick Manufacturing Group
AN INTRODUCTION TO RELIABILITY
ENGINEERING
Contents
1 Introduction 1
2 Measuring reliability 4
3 Design for reliability 12
4 Reliability management 34
5 Summary 35
RELIABILITY ENGINEERING
1 Introduction
1.1 Definition
Most people will have some concept of what reliability is from everyday life, for example,
people may discuss how reliable their washing machine has been over the length of time they
have owned it. Similarly, a car that doesn’t need to go to the garage for repairs often, during
its lifetime, would be said to have been reliable. It can be said that reliability is quality over
time. Quality is associated with workmanship and manufacturing and therefore if a product
doesn’t work or breaks as soon as you buy it you would consider the product to have poor
quality. However if over time parts of the product wear-out before you expect them to then
this would be termed poor reliability. The difference therefore between quality and reliability
is concerned with time and more specifically product life time.
Reliability engineering has both quantitative and qualitative aspects; measurements of
reliability are necessary for customer requirements compliance. However measuring
reliability does not make a product reliable, only by designing in reliability can a product
achieve its reliability targets. These lecture notes will therefore introduce some of the
terminology used in reliability engineering. It will provide information about measuring
reliability as well as designing for reliability. Moreover it will emphasise the importance of
good engineering principles to ensure product reliability. By identifying possible causes of
failure and elimination will obviously help to improve product reliability.
The formal definition of reliability is as follows: The ability of an item to perform a required
function under stated conditions for a stated period of time. BS4778
Another definition concerns the probabilistic nature of measuring reliability, i.e. the
probability of an item to perform a required function under specified conditions for a stated
period of time. It is therefore a measure of engineering uncertainty and to quantify reliability
involves the use of statistics and more specifically probability theory. These notes will also
describe some useful probability distributions that can describe the lifetime behaviour of
products.
• Items are designed for a specific operating environment and if they are then used
outside this environment then failure can occur.
There are many reasons for failure in items the list above is a generic list.
The load and strength of an item may be generally known, however there will always be an
element of uncertainty. The actual strength values of any population of components will vary;
there will be some that are relatively strong, others that are relatively weak, but most will be
of nearly average strength. Similarly there will be some loads greater than others but mostly
they will be average. Figure 1, below shows the load strength relationship with no overlaps.
Probability
Load Strength
However if, as shown in figure 2, there is an overlap of the two distributions then failures will
occur. There therefore needs to be a safety margin to ensure that there is no overlap of these
distributions.
Failure
Probability
Load Strength
It is clear that to ensure good reliability the causes of failure need to be identified and
eliminated. Indeed the objectives of reliability engineering are:
• To apply engineering knowledge to prevent or reduce the likelihood or frequency of
failures;
• To identify and correct the causes of failure that do occur;
• To apply methods of estimating the likely reliability of new designs, and for analysing
reliability data.
These notes will discuss some of the techniques that can be used to identify failures as well as
the statistical techniques for analysing reliability data.
Unreliability has a number of unfortunate consequences and therefore for many products and
services is a serious threat. For example poor reliability can have implications for:
• Safety
• Competitiveness
• Profit margins
• Reputation
• Good will
KEY POINTS
2 Measuring reliability
2.1 Requirements
Many customers will produce a statement of the reliability requirements that is included in the
specification of the product. This statement should include the following:
• The definition of failure related to the product’s function and should cover all failure
modes relevant to the function;
• A full description of the environments in which the product will be stored, transported,
operated and maintained;
• A statement of the reliability requirement
Care must be given in defining failure to ensure that the failure criteria are unambiguous.
Failure should always relate to a measurable parameter or to a clear indication. For example, a
definition of failure could include ‘failure of a function to operate’. To be able to design for
the load of the product the design team must have accurate information concerning the
environment of the product. If an item must fully operate at high altitude with extreme
changes in temperature then the design must be robust enough to withstand such
environmental factors. Similarly if a product is stored in extreme conditions prior to use then
the design must accommodate for the storage conditions.
The reliability requirement should be stated in a way which can be verified, and which makes
sense relative to the use of the product. The simplest requirement is to state that no failure will
occur under stated conditions. Reliability requirements based on life parameters (see section
2.3) must be based on the corresponding life distributions. A common parameter used is
MTBF, when a constant failure rate is assumed.
Reliability and Maintainability case
The UK MOD has recently moved away from prescriptive reliability requirements and now
requests a reliability case from their suppliers. The reliability and maintainability (R&M) case
is defines as “A reasoned, auditable argument created to support the contention that a defined
system satisfies the R&M requirements” . DEF STAN 0042 part 3 is a document produced by
the UK MOD that gives guidance on what goes into an R&M case.
period of time represents the burn-in or debugging period where weak items are weeded out.
After the initial phase when the weak components have been weeded out and mistakes
corrected, the remaining population reaches a relatively constant hazard function period,
known as the useful life period. From figure 3 you can see that the hazard function is constant,
this shape can be modelled by the exponential distribution (see section 2.3) when failure are
occurring randomly through time. The final portion of the bath-tub curve is called the wear-
out phase, this is when the hazard function increases with time.
Hazard function
Useful Life
Time
Infant Wear Out
Mortality
Figure 3 The bath-tub curve
If you take a large number of measurements you can draw a histogram to show the how the
measurements vary. A more useful diagram, for continuous data, is the probability density
function. The y axis is the percentage measured in a range(shown on the x-axis) rather than
the frequency as in a histogram. If you reduce the ranges(or intervals) then the histogram
becomes a curve which describes the distribution of the measurements or values. This
distribution is the probability density function or PDF. Figure 4, below, shows an example
of a PDF. The area under the curve of the distribution is equal to 1, i.e.
−∞
∫ f (x)dx = 1
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 6
The probability of a value falling between any two values x1 and x2 is the area bounded by
this interval, i.e.
x2
f(x)Frequency
values
In reliability since we are usual discussing time we will change x to t, i.e. f(t). The CDF,
cumulative distribution function or F(t), gives the probability that a measured value will
fall between -∞ and t, i.e.
F (t) = ∫ f (t)dt
−∞
t
0 .8
0 .6
re q u e
nCDF,F(t)
0 .4
F
0 .2
0
1 1 2 3 3 4 5 t 5 6 7 7 9
V a lu e
In reliability engineering we are concerned with the probability that an item will survive for a
stated interval of time (or cycles or distance etc.) i.e. there is no failure in the interval (0 to t).
This is known as the survival function and is given by R(t). From the definition:
t −∞
1.2
1
(%)
0.8
0.6
R(t)
0.4
Frequenc
0.2
y
0
1 1 2 3 3 4 5t5 6 7 7 9
Value
Another important function in reliability is the hazard function h(t); it is the conditional
probability of failure in the interval t to (t+dt), given that no failure has occurred by t:
f (t ) f (t )
h(t) = R(t ) = 1− F (t )
The bath tub curve, shown in figure 3, shows 3 different hazard functions, one decreasing,
one constant and one increasing hazard function.
There are 3 continuous life distributions that are commonly used in reliability engineering;
these are called, the exponential, Weibull and lognormal distributions. The normal distribution
as discussed in both the Six Sigma and SPC lectures is not generally used in reliability
engineering (although it is sometimes used).
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 8
When an item is subject to failures that occur in random intervals and the expected number of
failures is the same for long periods of time then the distribution of failures is said to fit an
exponential distribution. The PDF, CDF and survival function is given as:
f (t) = λe−λt and F (t) = 1 − e−λt and R(t) = e−λt
λe− λt
h(t) = − λt =λ
e
Notice that the hazard function is not a function of time and is in fact a constant equal to λ..
For repaired items, λ, is called the failure rate and 1/λ is called the mean time between failures
(MTBF), sometimes denoted as θ. An important point to note is that 63.2% of items will have
failed by time t=θ.
The failure rate can be calculated as the total number of failures divided by the total operating
time.
The exponential distribution is the most commonly used distribution in reliability engineering
and models the useful life portion of the bath-tub curve.
This distribution takes account of a non-constant hazard function. The Survival function is
given by:
t β
−( )
R(t) = e η
where β is the shape parameter and η is the scale parameter or characteristic life. The
characteristic life is the life at which 63.2% of the population will have failed.
When β = 1, the hazard function is constant and therefore the data can be modelled by an
exponential distribution with η=1/λ .
When β<1, we get a decreasing hazard function and
Figure 7, below, shows the Weibull shape parameters superimposed on the bath-tub curve.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 9
function
Useful Life
Hazar
β=1
d
β<1 β>1
Time
Infant Wear Out
Mortality
Figure 7 bath tub curve and the weibull distribution
When β>3.5, the Weibull distribution is an approximation for the Normal distribution.
There is also a three parameter version of the Weibull distribution and this is called the
location parameter. It is sometimes called the failure free time or the minimum life.
Other notation often used with the Weibull distribution is the B n-life, this is the time by which
n% of the population can be expected to fail. (n is the proportion failing).
The lognormal distribution is more versatile than the normal distribution and is a better fit to
reliability data, such as for populations with wear-out characteristics. Also, it does not have
the Normal’s disadvantage of extending below zero i.e. always positive. The lognormal
distribution and the normal distribution describe situations when the hazard function is
increasing.
Usually multiple components make up a systems and we often want to know the reliability of
a system that uses more than one component. How the components are connected together
determines what type of system reliability model is used.
There are different types of system reliability models and theses are typically used to analyse
items such as an aircraft completing its flight successfully.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 10
Simplest reliability model is a serial model where all the components must be working for the
system to be successful, for example:
A B Z
Figure 8 example of components connected in series
If the components can be assumed to be exponentially distributed then the system reliability
can be calculated as:
RS = e−λ t A * e
−λ t
B *....* e
−λ
Z
t
The Failure rate of the system is calculated as by adding the failure rates together, i.e.
λS = λA + λB + .......... + λZ
2.4.2 Active redundancy
One of the most common forms of redundancy is the parallel reliability model where two
independent items are operating but the system can successfully operate as long as one of
them is working, diagrammatically:
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 11
B
Figure 9 dual redundant system
The reliability of the system is equal to the probability of item 1 or item 2 surviving, e.g.
RS = RA + RB – (RA* RB)
RS = e−λ t A + e
−λ t
B − e
−(λ
A
+λ
B
)t
This example assumes each item is not repaired after failure i.e. non-maintained system.
In some active parallel redundant configurations, m out of the n items may be required to be
working for the system to function. The reliability of an m-out-of-n system, with n identical
independent items is given by:
m −1
R = 1 − ∑ n Ci Ri (1 − R)n
−i
i =0
There are other system reliability models including the standby redundancy situation but those
above are the simplest.
One of the most common methods for predicting the reliability of a system is called the parts
count method.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 12
Assuming all the parts in a system are independently exponentially distributed, i.e. one part
does not cause the other to fail then the overall system failure rate can be calculated using the
series system model shown above. For example, the failure rate of a printed circuit board is
the sum of the failure rates of each of the components.
For example:
The failure rates for components can be estimated from company in-service databases or can
be attained from published handbooks and published data.
KEY POINTS
• Weibull with shape parameter, can model decreasing and increasing hazard function.
rd
When Beta =1 is equal to exponential. Characteristic life is the 63 percentile
• Series systems modelling used for estimating system reliability by using parts count
method
The objective of design for reliability is to design a given product that meets its requirements
under the specified environmental conditions. To achieve this good sound engineering design
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 13
rules should be followed. However there are a few general principles that should observed,
these include:
• Component selection – well-established and known components should be used
(company usually have their own approved components list). If this is not he case then
analysis must be done to check the component is fit for purpose.
• Consider the load-strength relationship and ensure there is an adequate safety margin.
• Minimum complexity
• Identify any single point failures and either mitigate or design them out.
• Use lessons learned from previous products to design out any known
weaknesses. Ultimately the aim is to maximise reliability during service life by:
Each product has a life cycle, figure 10 illustrates a generic product life cycle. There are a
number of tools and techniques that are most useful at various stages of the product life cycle.
For example, at the design stage, it is most appropriate to use techniques that will be useful for
design reviews. Testing parts for fitness of purpose using accelerated life testing is also
necessary at this stage. When the product has been built it becomes costly to change the
design so all design reviews need to be done as early as possible in the product life cycle.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 14
Design
FMECA, FTA, PoF,RBD
FE,accelerated life test
Development
Development Test
Use
Field data analysis
FRACAS
Test
Manufacture
ESS, Burn-in
SPC
Development testing is used to investigate the robustness of the product and to identify any
design weaknesses with respect to the load. Development testing incorporates environmental
testing and is used for fitness of purpose of the product.
When the product has been developed, the design closed and ready for production, statistical
process control and other quality engineering tools are imperative for ensuring a good quality
product.
Environmental stress screening or burn-in is sometimes used to test all manufactured units
prior to release to the customer. The purpose of ESS is to identify any manufacturing
weaknesses in individual items.
When in-service, product performance data should be collected to check the product
reliability and also to feed forward to new product design in the form of lessons learned.
More discussion on some of these tools and techniques is given in later sections.
Some of the tools that are useful during the design stage can be thought of as tools for fault
avoidance. The fall into two general methods, bottom-up and top-down.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 15
• Undesirable single event or system success at the highest level of interest (the top event)
should be defined.
• Contributory causes of that event at all levels are then identified and analysed.
• Event-oriented method
• Used for evaluating multiple failures including sequentially related failures and common-
cause events
Some examples of top-down methods include: Fault tree analysis (FTA); Reliability block
diagram (RBD) and Markov analysis
Fault tree analysis
Fault tree analysis is a systematic way of identifying all possible faults that could lead to
system fail-danger failure. The FTA provides a concise description of the various
combinations of possible occurrences within the system that can result in predetermined
critical output events. The FTA helps identify and evaluate critical components, fault paths,
and possible errors. It is both a reliability and safety engineering task, and it is a critical data
item that is submitted to the customer for their approval and their use in their higher-level
FTA and safety analysis. The key elements of a FTA include:
– Gates represent the outcome
– Cut sets are groups of events that would cause a system to fail
The following diagram shows the flowchart symbols that are used in fault tree analysis in
order to aid with the correct reading of the fault tree.
FTA can be done qualitatively by drawing the tree and identifying all the basic events.
However to identify the probability of the top event then probabilities or reliability figures
must be input for the basic events. Using logic the probabilities are worked up to given a
probability that the top event will occur. Often the data from an FMEA are used in
conjunction with an FTA.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 16
Circle signifies a primary failure or basic fault that requires no further development
Diamond denotes a secondary failure or undesired event but not developed further
And gate denotes that a failure will occur if all inputs fail (parallel redundancy)
Or gate denotes a failure will occur if any input fails (series reliability)
Transfer event
The RBD is discussed and shown in section 2.4 above. It is however among the first tasks to
be completed. It model system success and gives results for the total system. As shown in
section 2.4, it deals with different system configuration, including, parallel, redundant,
standby and alternative functional paths. It doesn’t provide any fault analysis and uses
probabilistic measures to calculate system reliability.
• For each fault mode the corresponding effect on performance is deduced for the next
higher system level
• The resulting fault effect becomes the fault mode at the next higher system level, and
so on
• Successive iterations result in the eventual identification of the fault effects at all
functional levels up to the system level.
Some examples of bottom-up methods include: Event tree analysis (ETA); FMEA and Hazard
and operability study (HAZOP).
Event tree analysis
• used when it is essential to investigate all possible paths of consequent events their
sequence
• analysis can become very involved and complicated when analysing larger systems
Example:
Benefits include:
• Identifies systematically the cause and effect relationships.
• Gives an initial indication of those failure modes that are likely to be critical,
especially single failures that may propagate.
• Identifies outcomes arising from specific causes or initiating events that are believed
to be important.
• Provides a framework for identification of measures to mitigate risk.
• Useful in the preliminary analysis of new or untried systems or processes.
Limitations include:
• The output data may be large even for relatively simple systems.
• May become complicated and unmanageable unless there is a fairly direct (or "single-
chain") relationship between cause and effect may not easily deal with time sequences,
restoration processes, environmental conditions, maintenance aspects, etc.
• Prioritising mode criticality is complicated by competing factors involved
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 19
This type of analysis will answer most of the why, where, when, and how questions about the
life of a component. Analysis methods prove essential in understanding, determining, and
applying appropriate corrective action for root cause of failure. Understanding the root cause
of a failure is essential in today’s highly competitive market for successfully manufacturing
quality components.
The concept of accelerated testing is to compress time and accelerate the failure mechanisms
in a reasonable test period so that product reliability can be assessed. The only way to
accelerate time is to stress potential failure modes. These include electrical and mechanical
failures. Failure occurs when the stress exceeds the product’s strength. In a product’s
population, the strength is generally distributed and usually degrades over time. Applying
stress simply simulates aging. Increasing stress increases the unreliability and improves the
chances for failure occurring in a shorter period of time. This also means that a smaller sample
population of devices can be tested with an increased probability of finding failure. Stress
testing amplifies unreliability so failure can be detected sooner. Accelerated life tests are also
used extensively to help make predictions. Predictions can be limited when testing small
sample sizes. Predictions can be erroneously based on the assumption that life-test results are
representative of the entire population. Therefore, it can be difficult to design an efficient
experiment that yields enough failures so that the measures of uncertainty in the predictions
are not too large. Stresses can also be unrealistic. Fortunately, it is generally rare for an
increased stress to cause anomalous failures, especially if common sense guidelines are
observed.
Anomalous testing failures can occur when testing pushes the limits of the material out of the
region of the intended design capability. The natural question to ask is: What should the
guidelines be for designing proper accelerated tests and evaluating failures? The answer is:
Judgment is required by management and engineering staff to make the correct decisions in
this regard. To aid such decisions, the following guidelines are provided:
1. Always refer to the literature to see what has been done in the area of accelerated
testing.
2. Avoid accelerated stresses that cause “nonlinearities,” unless such stresses are
plausible in product-use conditions. Anomalous failures occur when accelerated stress
causes “nonlinearities” in the product. For example, material changing phases from
solid to liquid, as in a chemical “nonlinear” phase transition (e.g., solder melting,
inter-metallic changes, etc.); an electric spark in a material is an electrical
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 20
3. Tests can be designed in two ways: by avoiding high stresses or by allowing them,
which may or may not cause nonlinear stresses. In the latter test design, a concurrent
engineering design team reviews all failures and decides if a failure is anomalous or
not. Then a decision is made whether or not to fix the problem. Conservative decisions
may result in fixing some anomalous failures. This is not a concern when time and
money permit fixing all problems. The problem occurs when normal failures are
labeled incorrectly as anomalous and no corrective action is taken.
Accelerated life testing is normally done early in the design process as a method for testing
for fit for purpose. It can be done at the component level or the sub -assembly level but is
rarely done at a system level as there are usually too many parts and factors that can cause
failures and these can be difficult to control and monitor.
• Aging information can be obtained in a relatively short period of time. Common step-
stress tests take about 1 to 2 weeks, depending on the objective.
• Step-stress tests establish a baseline for future tests. For example, if a process
changes, quick comparisons can be made between the old process and the new
process. Accuracy can be enhanced when parametric change can be used as a measure
for comparison. Otherwise, catastrophic information is used.
• Failure mechanisms and design weaknesses can be identified along with material
limitations. Failure-mode information can provide opportunities for reliability growth.
Fixes can then be put back on test and compared to previous test results to assess fix
effectiveness.
• Data analysis can provide accurate information on the stress distribution in which the
median-failure stress and stress standard deviation can be obtained.
The goal of Reliability enhancement testing (RET) is to identify any potential failure modes
that are inherent in a design early in the design process. Identifying the root cause of the
failure mode and then incorporating a fix to the design can achieve reliability growth. This is
accomplished by designing out the possibility of potential failure modes occurring with the
customer and reducing the inherent risk associated with new product development. RET at the
unit or subassembly level utilizes step-stress testing as its primary test method. It should be
noted that Highly Accelerated Life Testing (HALT) is not meant to be a simulation of the real
world but a rapid way to stimulate failure modes. These methods commonly employ
sequential testing, such as step-stressing the units with temperature and then vibration. These
two stresses can be combined so that temperature and vibration are applied simultaneously.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 21
This speeds up testing, and if an interactive vibration/temperature failure mode is present, this
combined testing may be the only way to find it. Other stresses used may be power step-
stress, power cycling, package preconditioning with infrared (IR) reflow, electrostatic-
discharge (ESD) simulation, and so forth. The choice depends on the intended type of unit
under test and the unit’s potential failure modes.
HALT is primarily for assemblies and subassemblies. The HALT test method utilizes a HALT
chamber. Today, these multi -stress environmental systems are produced by a large number of
suppliers. The chamber is unique and can perform both temperature and vibration step-stress
testing.
If all processes were under complete control, product screening or monitoring would be
unnecessary. If products were perfect, there would be no field returns or infant mortality
problems, and customers would be satisfied with product reliability and quality. However, in
the real world, unacceptable process and material variations exist. Product flaws need to be
anticipated before customers receive final products and use them. This is the primary reason
that a good screening and monitoring program is needed to provide high quality products.
Screening and monitoring programs are a major factor in achieving customer satisfaction.
Parts are screened in the early production stage until the process is under control and any
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 22
material problems have been resolved. Once this occurs, a monitoring program can ensure that
the process has not changed and that any deviations have been stabilized. Here, the term
“screening” implies 100% product testing while “monitoring” indicates a sample test. Screens
are based upon a product’s potential failure modes. Screening may be simple, such as on-off
cycling of the unit, or it may be more involved, requiring one or more powered environmental
stress screens. Usually, screens that power up the unit, compared with non-powered screens,
provide the best opportunity to precipitate failure-mode problems. Screens are constantly
reviewed and may be modified based on screening yield results. For example, if field returns
are low and the screen yields are high (near 100 percent), the screen should be changed to find
all the field issues. If yields are high with acceptable part per million (PPM) field returns, then
a monitoring program will replace the screen. In general, monitoring is preferred for low-
cost/high-volume jobs. A major caution for selecting the correct screening program is to
ensure that the process of screening out early life failures does not remove too much of a
product’s useful life. Manufacturers have noted that, in the attempt to drive out early life
failure, the useful life of some products can become reduced. If this occurs, customers will
find wear-out failure mechanisms during early field use.
The information obtained when a product is first introduced to the Development Phase in the
HALT process enables the development of a HASS test. HALT, is a highly accelerated
reliability growth Test-Analyze-And-Fix (TAAF) process. In HASS, failures are analyzed,
and corrective actions are implemented. The test is repeated until the observed failure modes
have been fixed, and the environmental technology limits of the part are understood. This
information is used for the Production Phase. At this stage, one either develops a traditional
ESS or a HASS test. HASS test combines thermal cycling, vibration, and power stress
simultaneously. The testing range is within the operating limits that are known from prior
HALT testing performed. Similar to HALT, HASS is an aggressive screening program to help
weed out failure modes and implement corrective actions as soon as possible. This process
enables products to be moved quickly into a monitoring program. The HASS process typically
helps to reduce screen time (30 percent to 80 percent) and move a product more quickly into
the Monitoring Phase. For example, a common screen uses 168-hour burn-in, 20-hour thermal
shock, and a 60-minute vibration test. Since this is a fairly lengthy screen, it is advantageous
to work with a HASS program. In the HASS process, this test is quickly reduced to a
monitoring program. Since faster test results help in implementing product improvements and
moving to a monitoring test, cost savings can be passed onto the customer. If HASS
precipitates a failure, an immediate failure analysis is performed to determine the root cause.
A 100 percent screen is maintained until the process is in control. At that point, monitoring
can be performed. The monitoring also includes a HALT at given intervals to ensure that the
product safety margins have not deteriorated from those obtained in the Evaluation Phase.
Thus, knowledge of the HALT and HASS environmental limits, relative to a customer
specification, is very helpful in providing engineering confidence in the proper design of the
screening and the subsequent monitoring test. Such sound practices are important for
providing a highly reliable product.
Traditionally, the need for Reliability Growth planning has been for large subsystems or
systems. This is simply because of the greater risk in new product development at that level
compared to the component level. Also, in programs where one wishes to push mature
products or complex systems to new reliability milestones, inadequate strategies will be
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 23
costly. A program manager must know if Reliability Growth can be achieved under required
time and cost constraints. A plan of attack is required for each major subsystem so that
system-level reliability goals can be met. However Reliability Growth planning is
recommended for all new “platforms,” whether they are complex subsystems or simple
components. In a commercial environment with numerous product types, the emphasis must
be on platforms rather than products. Often there may be little time to validate, let alone
assess, reliability. Yet, without some method of assessment, platforms could be jeopardized.
Accelerated testing is, without question, the featured Reliability Growth tool for industry. It is
important to devise reliability planning during development that incorporates the most time-
and cost effective testing techniques available.
Reliability growth can occur at the design and development stage of a project but most of the
growth should occur in the first accelerated testing stage, early in design. Generally, there are
two basic kinds of Reliability Growth test methods used: constant stress testing and step-stress
testing. Constant stress testing applies to an elevated stress maintained at a particular level
over time, such as isothermal aging, in which parts are subjected to the same temperature for
the entire test (similar to a burn-in). Step-stress testing can apply to such stresses as
temperature, shock, vibration, and Highly Accelerated Life Test (HALT). These tests
stimulate potential failure modes, and Reliability Growth occurs when failure modes are
fixed. No matter what the method, Reliability Growth planning is essential to avoid wasting
time and money when accelerated testing is attempted without an organized program plan.
Today’s competitive market requires thorough planning, especially since platform complexity
has increased dramatically as competition and technological advances have driven down size
and costs. Traditional methodology provides us with many valuable tools, such as test
planning, growth tracking and assessment, fix-effectiveness factor estimation, corrective
action review team operations, and Test-Analyze-And-Fix (TAAF) strategies. All methods
use a FRACAS type approach to audit corrective actions.
There are numerous types of accelerated tests including HALT, Step-Stress, Highly
Accelerated Stress Screening (HASS), Environmental Stress Screening (ESS), failure-free
accelerated testing, Reliability Growth and accelerated reliability growth. These practices are
all important, since each has been used today in both commercial and industrial applications
for ensuring product reliability. The methods have not been without confusion. Confusion
exists as to when and which test method should be used and the Reliability Growth that can be
achieved with any one method. The methods should be to integrating and implemented
throughout the product development cycle using Table 1 summarizes the approach and how
these tests fit into the product life cycle.
Failure-Free Test or Post design This is also termed zero failure testing.
demonstration test This is a statistically significant reliability
test used to demonstrate that a particular
reliability objective can be met at a certain
level of confidence. For example, the
reliability objective may be 1000 FITs (1
million hours MTTF) at the 90 percent
confidence level. The most efficient
statistical sample size is calculated when
no failures are expected during the test
period. Hence the name.
ESS (Environmental production This is an environmental screening test or
Stress Screening) tests used in production to weed out latent
and infant mortality failures.
3.4 FRACAS
Data can be collected from the field and analysed for a number of reasons. These broadly
include checking that the product service reliability is meeting requirements; looking at the
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 26
service performance and identifying any systematic faults that can be fixed in this and future
products.
Exploratory Data Analysis can be used on such data to look for trends or systematic failures.
To analyse such data the data recorded should include the following basic information:
• Date of failure
• Reason for removal from field
• Serial number of product removed
• Product type
• Customer
• Length of time operating prior to failure
• Date entered service
After diagnosis of the product the following information is required:
• Root cause of failure
• Repair information – what was removed and repaired and when
• Diagnostic information
• Category of failure i.e. was is a design failure or a component failure or a
manufacturing failure or a misuse failure or a diagnostic failure or was is working and
no fault found.
• When was it shipped out after repair
This type of information allows the product support organisation within a company to analyse
the data to identify where the majority of failures are occurring.
For example:
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 27
Find ‘Root’ Cause Determine ‘Real’ Cause
Agree Categorisation
of N.F.F. Removals of all Unit Removals of Confirmed Removals
External Fault
No Fault Found
Troubleshooting
System Design
Fault Isolation Manual
Risk management applies to all new product development. Common technical risk areas
include performance, producibility, production, scheduling and resources. Risk varies
depending on whether customer requirements match technology performance capability
predictions, if field experience is available on analogous assemblies, if the technology is
revolutionary or evolutionary, if the application is new, if the intended use environment is
harsh and different from previous field experience, and so forth. Risks are often assessed in
categories. A technology management risk matrix is often used in industry as shown below:
Evolutionary Revolutionary
Same application Low Risk High Risk
New Application Moderate Risk Very High Risk
Revolutionary technologies carry a higher risk. For example, when the first airplanes were
developed in the early 1900s, flying these early machines often resulted in injury or death.
Now that flying is a mature technology, the risks of flying are very low. Evolutionary changes
to the aircraft having similar applications today carry low risks since the technology is mature.
The goal of a risk management program is to make correct decisions at key points in the
program. Technology risk management is essential to the success of any development
program. Risk issues and their consequences concern everyone involved with a program’s
success. The larger and the more undeveloped a technical program, the more important it is to
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 28
manage risks. In the case of a reasonably large and/or complex program, many technical
details can impact the system. The purpose of risk management is to identify, assess and
mitigate risks throughout the project.
Since component and subsystem risks are magnified at the system level, it is important that
program management becomes aware of issues early in the program. All potential risk areas
require identification and risk handling. Management can then direct resources to prioritized
risk areas and conserve valuable time and expenses. These benefits are best realized when
technical risk issues can be properly identified, assessed, quantified, and finally handled both
at the system and the subsystem level.
3.7 REMM
The REMM project developed a methodology that supports product reliability enhancement
and consequently is viewed primarily as a design for reliability tool although it could be used
by management and product support to view the reliability status of products. It is viewed as a
move away from ‘playing the numbers game’ using reliability predictions towards a more
considered approach. Its purpose is to encourage design for reliability by providing engineers
with reliability information and lessons learned on previous designs as well as providing a
proactive holistic approach to design
Current reliability prediction techniques for electronics have been shown to be
unrepresentative of real situations. There is a dichotomy between the way reliability is
assessed (predicted) and the way it is actually achieved.
Many aerospace companies continue to use MIL Handbook 217 for reliability predictions.
Usually the predictions are made on a parts count basis and then the overall predicted failure
rate is factored using in-service performance of previous products. So if a prediction had been
done on a product that is now in the field then a multiplicative factor can be obtained by
comparing the in-service reliability and that prediction. Such factors are therefore used in
predictions for new designs.
The problem with this approach is that the product reliability is not really considered. The
parts count prediction is based on the number of components of each type and their
corresponding failure rates and as is widely known this is all based on the assumption of
constant failure rates. In addition there is a temptation when given an MTBF requirement to
use the prediction to show that it will be met without doing any engineering analysis. In other
words the prediction is often altered rather than the design or manufacture of the product.
The R&M case was developed by the UK MOD as a move away from prescribing reliability
methods and tools towards a more measured approach to achieving reliable products. The
definition is given as “A reasoned, auditable argument created to support the contention that a
defined system satisfies the R&M requirement and is therefore concerned with providing
progressive assurance that the product will be reliable. The impetus for this move towards the
R&M case is partly driven by experience. In the past the MOD, in their contracts, prescribed
specific reliability techniques and tools to be performed. They believed that using such tools
and techniques would produce the required reliable products. However, in many cases, the
opposite occurred. Payments were made to suppliers for producing the required documents
however no evidence was necessary to show that any of the tools and techniques applied
actually affected the product design and development. In other words the focus was on
producing documents rather than showing that the product would meet the necessary
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 29
REMM Process
The REMM process was developed by the project team based on the philosophy of
evolutionary design of products i.e. any new product design can be based on an existing
product design. To design a new product the REMM philosophy urges the user to consider the
differences between the new product design and previous product designs. These differences
would include functionality, process (design and manufacture) and environmental changes to
enable the project team to concentrate on the risky aspects of the new product design and
development. Figure 11 below shows the simplified REMM process flow. As in DEF STAN
00-42 part3 the REMM process starts with analysing reliability requirements and then moves
on to identifying the reliability risks. The risks are identified in REMM by identifying the
novel aspects of the product design, manufacture, application and use.
Having captured requirements a similar product is identified and its associated data is
analysed in order to inform the project team of any reliability issues. The new design is
therefore altered in light of this data analysis. For example, if a particular component used in a
previous product and likely to be used in the new product is shown from analysis to be
causing reliability problems in the field then by highlighting such issues the project team can
make informed decisions regarding the new product design.
Capture requirements
Modify design
Service data
Capture differences between base and new design
environment etc. These differences between the new design and previous designs are captured
by the tool and fed into the REMM expert system. Capturing these differences focuses the
project team on the risky aspects of this new product.
The REMM expert system has been implemented at Goodrich engine controls and is a
knowledge base expert system. It has been populated with rules that were written by
interviewing chief engineers across the different disciplines, i.e. electronic, mechanical,
components, process and reliability. The rules consist of all the possible changes structured
hierarchically. So for example, Environmental changes can be due to vibration, thermal,
humidity and shock changes. Looking at vibration changes, the input fact could be, for
example:
• Small change in level of vibration;
• Significant change in level of vibration;
• Small change in vibration frequency range;
• Significant change in vibration frequency range.
Rules for new product design aspects have also been generated and so the differences
captured include both actual differences and novel aspects.
The REMM tool is used at the concept stage to design the new product using information on
previous designs. One of the outputs of the tool is a task list generated from the expert system.
The tasks suggested may well be tasks that are already planned but it gives a starting point for
developing the reliability plan as it focuses on the high level risk areas in the new product
design. The task list can therefore be used to develop the reliability programme and plan
The tool developed at Goodrich establishes the skeleton of the product reliability case when
the expert system is run. The skeleton reliability case consists of the product description in
terms of functionality, installation, use, environment and technology, it lists the hardware and
provides a list of all the differences and novel aspects of the design. It also contains the task
list and provides references to guidance material or procedures related to specific tasks. When
a task has been completed the case is updated with a reference to a company report detailing
the outcome and solution. To close the loop the tool ought to be linked to the FRACAS
system to ensure that the solution found is implemented to improve the reliability of the
product.
At present the REMM tool is useful at the beginning of the project and really only identifies
fairly high level tasks but by implementing the statistical model the project team can identify
lower level, specific risks to update the skeleton reliability case document.
analysis. This means that alternative scenarios can be investigated and the impact on
reliability estimated. Therefore the REMM statistical model provides a tracking system to
help analyse how reliability will and does evolve throughout the lifecycle by integrating the
numerical estimates with the engineering understanding of reliability. This is a deliberate
move to a proactive approach to design for reliability.
The REMM statistical model requires engineering concerns about potential faults in the item
of equipment to be identified and estimates when they are likely to occur as failures under
different scenarios. This model is therefore a Bayesian type statistical model.
Two basic types of input are required to populate the model.
• Engineering concerns with the new design and an estimate of their probability of
occurrence in service assuming no actions taken
• Reliability profile of the engineering concern or fault class to which it belongs
The data therefore comes from two sources – structured engineering judgement and historical
event data.
The basic output from the REMM statistical model will be the estimated reliability function,
also known as the survival function. This provides an estimate of the probability of surviving
until a specified time. Figure 13, below shows the model formulation as the reliability of the
new design is a combination of expert judgement and event data with respect to pre-defined
categories.
Expert
ND NC NB
Judgement
Event
RD(t) RC(t) RB(t)
Data
In the individual interviews each expert was asked to concentrate on the new product and
consider the following questions:
• Do you have any concerns about any aspect of the product?
• What are they?
• What is the likelihood that your concerns will cause a failure in the field? This was
done by asking the designers and specialist experts to rate risks on a scale of 0-1.
• What mitigation action could be taken to avoid this concern occurring as a failure?
The resulting information is used to provide a list of concerns for the project team as well as
an input to the statistical model.
The concern list on its own can be used as a risk assessment and the project team can decide
what actions to take to reduce the risk and improve the reliability. The concern list can be
added to the skeleton reliability case and updated as mitigating actions are taken and concerns
resolved. This process is an iterative process and would be implemented at key stages in the
design and development process. The reliability case would therefore be a living document
that grows as more information is gathered.
Running the model provides more information about the reliability of the product at a specific
time in the product development process. The reliability estimate can be compared with the
reliability requirement. If the estimate is lower then closer inspection of the results can help to
identify the key tasks to improve the reliability. For example, in figure 13, assuming the
overall reliability estimate is unacceptable then the data for the category contributing most to
this estimate could be investigated further. In figure 14, looking at the data for the preload
concerns can help guide the project team towards mitigating actions that may improve the
reliability. ‘What if’ analysis can be done to identify those actions that would give the biggest
improvement in reliability and therefore providing guidance to the project team.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 33
Estimated F-Lynx
All the analysis undertaken can be added to the reliability case to show how the product is
being progressively improved by identifying and subsequently reducing product risks.
Reliability Measurements
From the discussion above the REMM tool is used to help build the reliability programme
plan and the skeleton reliability case and is carried out at the concept design stage. The
statistical model is implemented at key points in the design and development process. The
concerns are updated throughout as more data is gathered.
The reliability estimates can therefore be plotted at each of the key stages to show how the
reliability estimate is changing throughout the development of the product and therefore how
close the estimate is to the requirement. Figure 15, below, shows an example of reliability
changing throughout the product life cycle.
Warwick Manufacturing Group
Introduction to Reliability Engineering Page 34
Updated
Required
Current
estimate
Lifecycle stage
4 Reliability management
• Management of suppliers
Reference
5 Summary
These lecture notes provide information about why reliability is important, moreover they
give an overview of all the aspects of reliability engineering, including:
• Need for Reliability requirements;
• Need for planning to achieve reliability throughout the product life cycle;
• Other factors that are affected by poor reliability such as safety, competitiveness,
goodwill, maintenance costs and ultimately profit.
• Design for reliability;
• Managing risk;
• Using FRACAS;
This lecture is an overview of reliability engineering and therefore it gives an appreciation for
the topic. It is not expected to be a reliability-engineering manual but gives a flavour for the
importance of the topic and how it fits into the design, development and use of a product.
Warwick Manufacturing Group