Part Two

Interpreting FaiIure Rates

4 Realistic failure rates

4.1 DATA ACCURACY

There are many collections of failure rate data compiled by defence, telecommunications,
industrial and other organizations. Some are published data Handbooks such as:

US MIL HANDBOOK 217

CNET (French PTT) Data
HRD (British Telecom)
RADC Non-Electronic Parts Handbook NPRD
OREDA (Offshore data)

Some are data banks (see also Appendix 10) accessible by virtue of membership or fee such
as:

SRD (Systems Reliability Department of UKAEA) Data Bank

RM Consultants Ltd (Abingdon)
Technis (Tonbridge)

Some are in-house data collections which are not generally available. These occur in:

Large industrial manufacturers

Public utilities

Failure data are usually, unless otherwise specified, taken to refer to random failures (ie
constant failure rates). It is important to read, carefully, any covering notes since, for a given
temperature and environment, a component may exhibit a wide range of failure rates owing
to:

Component source - the degree of screening, QA, etc.

Circuit tolerancing - the degree of design effort affects the proportion of failures attributable
to parametric drift.
Reliability growth - the amount of field experience fed back affects reliability and hence the
failure rate data.
Maintenance philosophy - trial and error replacement, as a means of diagnosis, artificially
inflates failure rate data.
32 Reliability, Maintainability and Risk

Failure rate values can span one or two orders of magnitude as a result of different
combinations of these factors. Prediction calculations are explained in Chapter 8 but it must be
emphasized that the relevance of failure rate data is more important than refinements in the
statistics of the calculation.
The failure rates are often tabulated, for a given component type, against ambient temperature
and the ratio of applied to rated stress (power or voltage). Failure data is presented in one of two
forms:

1. Tables: Lists of failure rates such as those in Appendices 3 and 4, with or without multiplying
factors, for such parameters as quality and environment.
2. Models: Models obtained by regression analysis of the data. These are presented in the forni
of equations which yield a failure rate as a result of inserting the device parameters into the
appropriate expression.

Because of the large number of variables involved in describing microelectronic devices, data
are usually expressed in the form of models. These regression equations involve some or all of
the following:

Complexity (number of gates, bits, equivalent number of transistors).

Number of pins.
0 Junction temperature (see Arrhenius, Section 10.2).
0 Package (ceramic and plastic packages).
Technology (CMOS, NMOS, bipolar, etc.).
0 Type (memory, random LSI, analogue, etc.).
0 Voltage.
Quality level (affected by screening and burn-in).
Environment.
Length of time in manufacture.

Although empirical relationships have been established relating certain device failure rates to
specific stresses, such as voltage and temperature, no precise formula exists which links specific
environments to failure rates. The permutation of different values of environmental factors, such
as those listed in Chapter 11, is immense. General adjustment (multiplying) factors have been
evolved and these are often used to scale up basic failure rates to particular environmental
conditions.
Because Failure Rate is, probably, the least precise engineering parameter, it is important to
bear in mind the limitations of a Reliability prediction. The resulting MTBF, Availability (or
whatever) should not be taken as an absolute parameter but rather as a general guide to the
design reliability. Within the prediction, however, the relative percentages of contribution to the
total failure rate are of a better accuracy and provide a valuable-tool in design analysis.
Owing to the differences between data sources, comparisons of reliability should always
involve the same data source in each prediction.
Total failure rates are of use in calculating the total maintenance commitment, but for a
Failure Mode Analysis it is necessary to assess the percentage of the failure rate of a device for
some specific failure mode (e.g. Short Circuit, Blocked, Fail to Release, Fail High etc.).
 Realistic failure rates 33

4.2 MICROELECTRONICS DATA

4.2. I The data

Microelectronics failure rates from the following sources were compared for a range of
temperatures and device complexities:
US Military Handbook 217E
British Telecom HRD4
French PTT, CNET data bank
Some field data collected by the author
The available data sources are described in Section 4.4. Note that MIL 217 is now at issue F and
HRD at issue 5. There appear to be no plans at this time (1998) for updating either document.
Table 4.1 presents the data ranges for four junction temperatures and 20 device type/
complexity categories. In view of the variability between sources it was not considered
necessary to consider any finer subdivision of temperatures or complexities. Table 4.1 is based
on an earlier version of FARADIP.THREE (See sections 4.3 and 4.4.1) whereas Appendix 3
provides more recent data.

Table 4.1 Ranges (per million hours)

Up to 30°C Approx. 50°C Approx. 75°C Approx. J00"C

Bipolar
L'inear 5 trans 0.04 0.07 0.05 0.08 0.05 0.15 0.10 0.30
25 trans 0.04 0.10 0.05 0.08 0.05 0.15 0.10 0.30
100 trans 0.04 0.20 q.05 0.20 0.05 0.25 0.15 0.50
Digital 50 gates 0.02 0.10 0.02 0.20 0.03 0.60 0.05 1S O
500 gates 0.03 0.20 0.03 0.30 0.03 1 .oo 0.07 3.00
1 K gates 0.05 0.30 0.05 0.50 0.07 1.20 0.15 3.00
RAM 4K 0.05 0.10 0.05 0.20 0.07 0.50 0.15 1 .oo
16K 0.07 0.20 0.07 0.30 0.10 1 .oo 0.20 3 .OO

MOS
Linear 5 trans 0.04 0.07 0.05 0.10 0.05 0.20 0.10 0.30
25 trans 0.04 0.10 0.05 0.20 0.05 0.30 0.10 0.35
100 trans 0.04 0.20 0.05 0.50 0.05 0.40 0.10 0.40
Digital NMOS 1K 0.15 0.30 0.15 0.30 0.20 1 .oo 0.50 3.00
NMOS 2K 0.15 0.40 0.15 0.70 0.20 2.00 0.50 6.00
CMOS 400 0.10 0.30 0.10 1.00 0.15 5.00 0.30 20.00
CMOS 1000 0.15 0.40 0.15 1 .OO 0.20 5.00 0.40 20.00
CMOS 2000 0.15 0.60 0.15 2.00 0.20 10.00 0.50 40.00
RAM CMOS 4K 0.02 0.10 0.02 0.25 0.03 1 .oo 0.06 4.00
CMOS 16K 0.03 0.15 0.03 0.30 0.05 1.oo 0.10 5.00
ROM CMOS 4K 0.01 0.15 0.01 0.30 0.02 1S O 0.03 6.00
CMOS 64K 0.02 0.20 0.02 0.60 0.03 3.00 0.07 12.00
34 Reliability, Maintainability and Risk

Table 4.2 Ratios

Up to 30°C Approx. 50°C Approx. 75°C Approx. I00"C

Bipolar
Linear 5 trans 2 2 3 3
25 trans 3 2 3 3
100 trans 5 4 5 3
Digital 50 gates 5 10 20 30
500 gates 7 10 33 43
1 K gates 6 10 17 20
RAM 4K 2 4 7 7
16K 3 4 10 15

MOS
Linear 5 trans 2 2 4 3
25 trans 3 4 6 3
100 trans 5 10 8 4
Digital NMOS 1K 2 2 5 6
NMOS 2K 3 5 10 12
CMOS 400 3 10 33 67
CMOS 1000 3 7 25 50
CMOS 2000 4 13 50 80
RAM CMOS 4K 5 13 33 67
CMOS 16K 5 10 20 50
ROM CMOS 4K 15 30 75 200
CMOS 64K 10 30 100 171
4.52 9.03 23.41 41.86

In general, they represent the maximum and minimum failure rates indicated for that cell
in the original data. Some smoothing has been attempted where an obvious pattern of
progression exists or where one value is clearly inconsistent. The author's data represent a
small sample compared with the published data sources and have not been allowed to
dominate.
Table 4.2 presents each range as a ratio of the range width, together with an average for
each of the four junction temperature columns. It will be observed that the range widths vary
significantly with temperature in that they approximately double for each 25°C increment.
This is due largely to the widening of the ranges in the high-temperature and MOS
areas.

4.2.2 The most reasonable estimate

The use of the simple arithmetic mean is not satisfactory for selecting a representative number
when the two estimates are so widely spaced, since it favours the higher figure. The following
 Realistic failure rates 35

example compares the arithmetic and geometric means where: Arithmetic Mean of n values of
xi is given by
n
xi/n
I

and the Geometric Mean by

Consider two estimates of failure rate, 0.1 and 1 .O (per million hours). The Arithmetic Mean
(0.55) is five times the lower value and only a half of the upper value, thereby favouring the 1.0
failure rate. Where the range is an order or more, the larger value has significantly more bias on
the arithmetic mean than the smaller.
The Geometric Mean (0.3 16) is, on the other hand, related to both values by a multiple of 3
and the excursion is thus the same. The Geometric Mean is, of course, the Arithmetic Mean of
the logarithms and therefore provides an average of the orders of magnitude involved. It is thus
a more desirable parameter for describing the range.
In order to express the ranges as a single failure rate it is thus proposed to utilize the
Geometric Mean. Appendix 3 provides the data in three columns giving the minima, maxima
and geometric means. The quality and environmental multipliers are described in Section 4.3
below. In addition, the following packaging multipliers need to be applied.

Ceramic 1
Plastic 1 for Quality factor 0.4
2 for Quality factors 1 or 2

4.2.3 Conclusions

1. The use of the published databases, and their stress-related regression models, implies an
unjustified precision in estimating the failure rate parameter.
2. Use of Appendix 3 as a source of failure rates is adequate precision.
3. The FARADIP.THREE software package (Appendix 10) provides the maximum and
minimum rates of Appendix 3 together with the geometric mean. The data can be viewed on
a PC and used directly in a failure mode and effect analysis as described in Chapter 8.
4. In practice, failure rate is a system level effect. It is closely related to but not entirely
explained by component failure. A significant proportion of failures encountered with
modern electronic systems are not the direct result of parts failures but of more complex
interactions within the system. The reason for this lack of precise mapping arises from such
effects as human factors, software, environmental interference, interrelated component drift
and circuit design tolerance.
5 . The primary benefit to be derived from reliability engineering is the reliability growth which
arises from continuing analysis and follow-up as well as corrective actions following failure
analysis. Reliability prediction, based on the manipulation of failure rate data, involves so
many potential parameters that a valid repeatable model for failure rate estimation is not
possible. Thus, failure rate is the least accurate of engineering parameters and prediction
from past data should only be carried out either:
0 As an indicator of the approximate level of reliability of which the design is capable, given
reliability growth in the field (see section 4.3.2).
36 Reliability, Maintainability and Risk

To provide relative comparisons in order to make engineering decisions concerning

optimum redundancy or,
As a contractual requirement.
It should not be regarded as an accurate indicator of future field reliability.

4.3 OVERALL DATA

4.3. I The data

The previous section concentrated on L..E particular case of microelectronics failure rate data.
Examination of over 20 databases, including those described in section 4.4, has resulted in the
production of ranges which are presented in three columns in the FARADIP.THREE database
(Appendix 10).
For some components there is fairly close agreement between different data banks and in
other cases there is a wide range of failure rates. The latter may be due to a number of reasons
as, for example:

Some failure rates include items replaced during preventive maintenance whereas others do
not. This can affect rates by an order of magnitude.
Failure rates are affected by the tolerance of a design and this will cause a variation in the
values.
Although nominal environmental and quality levels are described in some databases, the
range of parameters covered by these broad descriptions is large. They represent, therefore,
another source of variability.
Component parts are often only described by reference to their broad type (e.g. signal
transformer). Data are therefore combined for a range of similar devices rather than being
grouped, thus widening the range of values.

FARADIP.THREE (Appendices 4 and 10) presents each failure rate option in one of three
ways:

1. A single value, where the various references are in good agreement.

2 . Two values indicating a range. It is not uncommon for the range to be an order of magnitude
wide. The user, as does the author, must apply engineering judgement in choosing a value.
This involves consideration of the size, application and type of device in question. Where two
values occupy the first and third columns then an even spread of failure rates is indicated.
Where the middle and one other column are occupied then a spread with predominance to the
value in the middle column is indicated.
3 . Three values indicating a range. This implies that there is a fair amount of data available but
that it spans more than an order of magnitude in range. Where the data tend to predominate
in one area of the range then this is indicated in the middle column.

The most likely explanation of the range widths is the fact that some data refer only to
catastrophic failures whereas other data include degraded performance and minor defects
revealed during preventive maintenance. This should be taken into account when choosing a
failure rate from the tables.
 Realistic failure rates 37

As far as possible, the data given are for a normal ground fixed environment and for items
procured to a good standard of quality assurance as might be anticipated from a reputable
manufacturer operating to I S 0 9000. The variation which might be expected due to other
environments and quality arrangements is described in the following table of multiplying
factors.
Multiplying
Qualig factor
Normal commercial procurement 2
Items procured to some agreed specification and to a recognized 1
quality system
Items subject to 100% screening and bum-in 0.4
Environment
Dormant situation with little or no electrical or mechanical stress 0.1
(i.e. shelf or idling)
Benign as in an air-conditioned laboratory or computer area 0.5
Fixed ground with no adverse vibration or temperature cycling 1
Mobile ground as for portable or vehicle-mounted items 4

4.3.2 Using he ranges

The average range ratio for the entire FARADIP.THREE database is

50: 1

In all cases, site specific failure rate data or even that acquired from identical (or similar)
equipment, and being used under the same operating conditions and environment, should be
used in place of any published data.
Such data should, nevertheless, be compared with the appropriate range. In the event that it
falls outside the range there is a case for closer examination of the way in which the data were
collected or in which the accumulated component hours were estimated.
The ranges can contain:

1. A single value: The value can be used without need for judgement unless the specific
circumstances of the assessment indicate a reason for a more optimistic or pessimistic failure
rate estimate.
2. Two or three values with predominating centre column: In the absence of any specific reason
to favour the extreme values the predominating value is the most credible choice.
3. Wide ranges with ratios > 1 O : l : In Section 4.2 the use of the geometric mean was explained
and justified. In the absence of specific reasons to favour one of the extreme values the
geometric mean provides the most credible choice.

They can be interpreted as follows:

In general the lower figure in the range, used in a prediction, is likely to yield an assessment of
the credible design objective reliability. That is the reliability which might reasonably be
targeted after some field experience and a realistic reliability growth programme. The initial
(field trial or prototype) reliability might well be an order of magnitude less than this figure.
38 Reliability, Maintainabilityand Risk

The centre column figure indicates a failure rate which is more frequently indicated by the
various sources. It is therefore a matter of judgement, depending on the type of prediction being
carried out, as to whether it should be used in place of the lower figure.
The higher figure will probably include a high proportion of maintenance revealed defects
and failures. The fact that data collection schemes vary in the degree of screening of
maintenance revealed defects explains the wide ranges of quoted values.
If site, or industry, specific data is not available then the FARADIP.THREE DATA base could
be consulted in the following manner.
The prediction is carried out using the minimum failure rate quoted for each component. A
final range multiplier ( r ) will be allocated and the result of the prediction expressed as a range
of either:

100% - r X 100% of failure rate or unavailability

r X 100%-100% of mean time between failures

Because the larger value of failure rate will tend to include degradation, diagnostic
replacements, preventive maintenance, and other incidents not necessary likely to lead to the
failure in question, then the most credible design reliability prediction will lie between the
minimum and midpoint of the range. Since the midpoint of a range is best expressed by use of
the geometric mean, as was explained above in Section 4.2.2, then the square root of the range
is thus selected as the multiplier.
Thus the square root of 50 suggests a multiplier of 7. This is closely supported by the author’s
experience of calibrating predictions to field data.

4.3.3 The need for failure modes

For any reliability assessment to be meaningful it must address a specific system failure mode.
It is of little use to predict that a safety (shutdown) system will fail at a rate of, say, once per
annum since, if 90% of the failures lead to spurious shutdown and 10% to failure to respond then
the picture is vastly different if the ratios were to be reversed.
The failure rates, mean times between failures or availabilities must therefore be assessed for
defined failure types (modes). In order to achieve this, the appropriate component level failure
modes must be applied to the prediction models which are described in Chapter 8. Component
failure mode information is sparse but Appendix 5 provides a summary of what is available.

4.4 SOURCES OF FAILURE RATE DATA

The following are the 13 major sources of failure rate data.

4.4.1 FARADIRTHREE

This database, already described in section 4.3, is a summary of all the other databases and
shows, for each component, the range of failure rate values which is to be found from them.
Where a value in the range tends to predominate then this is indicated. Failure mode percentages
are also included.
 Realistic failure rates 39

It is a software database available on disk from the author at 26 Orchard Drive, Tonbridge,
Kent TNlO 4LG, UK (See Appendix lo). It includes:

0 Microelectronics
0 Discrete Semiconductors, tubes, lamps
0 Passive Electrical Components
0 Instruments and Analysers
0 Connectors, Switches, PCBs, Cables
0 Electromechanical and Rotating Machinery
0 Power Supplies, Sources and Transformers
0 Mechanical Items (incl. Pumps and Valves)
0 Pneumatic and Hydraulic equipment
0 Computers, DP and Comm’s
0 Alarms, Fire Protection, Arrestors and Fuses

4.4.2 US Mihary Handbook 2 17

This is the most well known of the data sources. It is currently at issue F: 1991 having
previously been published as:

217A - 1965 (Field Data)

217B - 1970 (Field Data)
2 17C - 1979 (Accelerated Life Test)
217D - 1982 (Accelerated Life Test)
217E - 1986 (Accelerated Life Test)
217E Notice 1 - 1990

It is produced by RADC (Rome Air Development Centre) under contract to the US Department
of Defense and is an electronic failure rate data book. It covers:

Microelectronics
Discrete semiconductors
Tubes (thermionic)
Lasers
Resistors and capacitors
Inductors
Connections and connectors
Meters
Crystals
Lamps, fuses and other miscellaneous items

The Microelectronics sections present the information as a number of regression models. For
example, the Monolithic Bipolar and MOS Linear Device model is given as:

Part operating failure rate model (Ap):

A, = + C2nE)nL Failures/106 hours

40 Reliability, Maintainability and Risk

where

A, is the device failure rate in F/106 hours,

is the quality factor,
nrq
nT is the temperature acceleration factor, based on technology,
nv is the voltage stress derating factor,
nE is the application environment factor,
C , is the circuit complexity factor based on transistor count as follows:

For 1 to 100 transistors, CI = 0.01

For >IO0 to 300 transistors, CI = 0.02
For >300 to 1000 transistors, C , = 0.04

C, is the package complexity failure rate,

nL is the device learning factor.

Two reservations arise from this approach. First, it is not possible to establish the original
application of the items from which the data are derived and it is not clear what mix of field and
test data pertains. Second, a regression model both interpolates and extrapolates the results of
raw data. There are similar models for other microelectronic devices and for discrete
semiconductors. Passive components involve tables of failure rates and the use of multipliers to
take account of Quality and Environment.
The trend in successive issues of MIL 217 has been towards lower failure rates, particularly
in the case ofmicroelectronics. This is also seen in other data banks and may reflect the steady
increase in manufacturing quality and screening techniques over the last 15 years. On the other
hand, it may be due to re-assessing the earlier data. MIL 217 is available (as MILSTRESS) on
disk from ITEM software (see Appendix 10). There appear to be no plans at this time (1998) for
updating the document.

4.4.3 Nonelectronic Parts Reliability Data Book - NPRD95 ( I 995)

This document is also produced by RADC and was previously published as:

NPRD 1 - 1978
NPRD 2 - 1982
NPRD 3 - 1985

It contains many hundreds of pages of failure rate information for a wide range of
electromechanical, mechanical hydraulic and pneumatic parts. Failure rates are listed for a
number of environmental applications. Unlike MIL 217, this is field data. It provides failure rate
data against each component type and there are one or more entries per component type
depending on the number of environmental applications for which a rate is available.
Each piece of data is given with the number of failures and hours (or operations/cycles). Thus
there are frequently multiple entries for a given component type. Details for the breakdown of
failure modes are given.
NPRD 5 is available on disk.
 Realistic failure rates 41

4.4.4 Handbook of Reliability Data for Electronic Components Used in

TelecommunicationsSystems HRD5 1I 994)

This document is produced, from field data, by British Telecom’s Laboratories at Martlesham
Heath. It was previously published as:

HRDl - 1977
HRD2 - 1980
HRD3 - 1984
HRD4 - 1986

It offers failure rate lists for Integrated Circuits, Discrete Semiconductors, Capacitors, Resistors,
Electromechanical and Wound Components, Optoelectronics, Surge Protection, Switches,
Visual Devices and a Miscellaneous section (eg microwave).
The failure rates obtained from this document are generally optimistic compared with the
other sources, often by as much as an order of magnitude. This is due to an extensive ‘screening’
of the data whereby failures which can be attributed to a specific cause are eliminated from the
data once remedial action has been introduced into the manufacturing process. Considerable
effort is also directed towards eliminating maintenance-induced failures from the data.

4.4.5 Recueil de Donnes de Fiabilitk du CNET (1983)

This document is produced by the Centre National d’Etudes des Telecommunications (CNET).
It was previously issued in 1981 and the latest edition is 1983. It has a similar structure to US
MIL 217 in that it consists of regression models for the prediction of component failure rates as
well as generictables. The models involve a simple regression equation with graphs and tables
which enable each parameter to be specified. The model is also stated as a parametric equation
in terms of voltage, temperature, etc. The French PTT use the CNET data as their standard.

4.4.6 Electronic Reliability Data - INSPEC/NCSR 1I 98 1)

This book was published jointly by the Institution of Electrical Engineers and the National
Centre of Systems Reliability (Warrington) in 1981. It is now somewhat dated and the failure
rates pessimistic due to the growth in semiconductor reliability over the last decade. It consists
of simple multiplicative models for semiconductor and passive electronic components with
tables from which to establish the multipliers according to the environmental, temperature and
other parameters.

4.4.7 OREDA - Offshore Reliability Data (1984/92/95/97)

This data book was prepared and published in 1984 and updated currently to OREDA 1997, by
a consortium of:

BP Petroleum Development Ltd Norway

Elf Aquitaine Norge AIS
Norsk Agip AIS
42 Reliability, Maintainability and Risk

AIS Norske Shell

Norsk Hydro a.s
Statoil
Saga Petroleum a.s
Total Oil Marine plc.

OREDA is managed by a steering committee made of from the participating companies. It is a

collection of offshore failure rate and failure mode data with an emphasis on safety-related
equipment. It covers components and equipment from:

Fire and gas detection systems

Process alarm systems
Fire fighting systems
Emergency shut down systems
Pressure relieving systems
General alarm and communication systems
Evacuation systems
Vessels
Valves
Pumps
Heat exchangers
Compressors
Pigging
Power generation and switching
Cranes
Drilling systems and equipment

Updated (OREDA 95) data is now available as a PC package, but only to members of the
participating partners. The empty data base, however, is available for those wishing to collect
their own data.

4.4.8 Reliability Prediction Manual for Guided Weapon Systems - DX/99/0 13- 700
Issue 2 1980

This was prepared under contract by Rex Thompson and Partners (Appendix 10) for the UK
MOD. It contains tables of failure rates, for various environmental conditions, and covers:

Discrete electronic and electro-mech

Connectors
Microelectronics
One-shot devices
Mechanical items

For some discrete components the failure rates are presented in tables which permit the selection
of temperature and stress ratios. Failure rate models are also offered for microelectronic devices
including hybrids. A section covers the probability of success for one-shot devices (e.g. ejector
cartridge, igniter, rocket motor, fusible switch).
 Realistic failure rates 43

4.4.9 RMC Reliability Data Handbook

This is 12 pages of failure rates, with ranges, and is based on the H A R E database. It was last
updated in 1991 (issue 2) and is available from R M Consultants Ltd (Appendix 10).

4.4.70 Green and Bourne - Reliability Technology, WileF 7972

This book contains part failure rate data obtained mostly from US and UK atomic energy
sources. There are some failure mode percentages also included. The data are rather old and
extremely pessimistic but are still occasionally quoted.

-
4.4.7 1 Professor frank Lees Loss Prevention in the Process Industries,
Bufferworth-Heinemann, 7 996

Lees published a series of articles in technical journals between 1971 and 1973 culminating in
an article in 1976 giving failure rates and repair times from three unidentified chemical process
plants. Also in 1976 he published a paper at the IChemE symposium (Series No. 47, A Review
of Instrument Failure Data). Some of the Green and Bourne data are included together with
UKAEA and USAEC data. It is not clear whether he obtained the US data from the SRS bank
or from some of the early NUREG reports or from both. In 1980 his two-volume book Loss
Prevention in the Process Industries was published. This has a database using the process plant
data, but failure rate data for mechanical components are from Green and Bourne. A feature is
that the failure rates are some five times those experienced by the author. This is probably due
to their including defects with failures.

4.4.12 UK Atomic Energy SRD Data Bank

This data bank is maintained by the SRD (Systems Reliability Department) at the UKAEA at
Culcheth, Warrington, Cheshire. It is now available on disk to members and can be used on a
suitable PC. The UKAEA data appear from time to time in equipment studies and papers. They
are then put into the SRD data bank which is intended to be contributed to by SRD
members.

4.4.13 National & International Reliability Conferences

Every two years (odd dates) until 1991, there were UK National Reliability Conferences
sponsored by the Institute of Quality Assurance and SRD of UKAEA. The other years (even
dates) there were Advances in Reliability Technology Symposia sponsored by Bradford
University and SRD of UKAEA. From 1992 onwards there have been annual European
Reliability Conferences which circulate around Europe. Each year the UK Safety & Reliability
Society organises a Symposium.
The published papers for these conferences contain a limited amount of failure data where
authors quote original failure rates.