You are on page 1of 5

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308952943

Problems of Reliability Prediction of Electronic Equipment

Conference Paper · September 2014

CITATION READS

1 269

1 author:

Slavko Pokorni
Information Technology School
44 PUBLICATIONS   23 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Slavko Pokorni on 23 January 2019.

The user has requested enhancement of the downloaded file.


PROBLEMS OF RELIABILITY PREDICTION OF ELECTRONIC
EQUIPMENT
SLAVKO POKORNI
Information Technology School, Belgrade, slavko.pokorni@its.edu.rs

Abstract: Military handbook for the reliability prediction of electronic equipment MIL-HDBK-217 is still used by more
than 80% of engineers. But, in spite of that MIL-HDBK-217 was updated several times to keep pace with technology
advancement as well as the changes in prediction procedure, it has limitations. Beside that, problems in setting
requirements for reliability were encountered in practice. The both lead to differences between the calculated reliability
and actual reliability in use. The limitations of MIL-HDBK-217 and efforts to revise it are presented in this paper. The
problems to establish good reliability requirements in connection with the statistics of causes of failures, which shows
that big percentage of failures are non-hardvare failures and a way how to establish a good contractual reliability
requirements, are also presented.
Keywords: reliability, reliability calculation, MIL-HDBK-217, hardware, software, human, reliability requirements

1. INTRODUCTION 1961 [1], and the first version was published in 1965 [2].
The aim was to establish a consistent and uniform
Reliability as a theory and practice began to develop in methods for assessing the inherent reliability of the
the 50s of the last century. In the mid 60s the military military electronic equipment and systems. MIL-HDBK-
manual for calculating reliability of MIL-HDBK-217 217 provides the basis for forecasting the reliability of the
appeared. Even though the commercial manuals appeared military electronic equipment and systems during design
later, the MIL-HDBK-217 is still used by more than 80% and contracting. It is extensively used for military and
of engineers in calculating reliability. By the time it has non-military programs. It also provides a basis for
been shown that this manual, which is essentially based comparison and evaluation of the prognostic reliability of
on an exponential distribution of failure, has a number of certain variants of the project. It is designed to be a tool
limitations, and that other approaches are needed, such as for increasing the reliability of the equipment that is
for example, so-called physics of failure. designed. It is periodically updated to folow the change of
technology and innovations in design procedures. There
First, we briefly analyzed in this paper the limitations of are versions A, B, C, D, E and F, as well as the addition
the MIL-HDBK-217, as well as efforts on its of two variants for version F (Notice Notice 1 and 2). It
revitalization. The analysis shows that the problem is not has not been modified since 1995 [2].
only in the models underlying the failure intensity of
electronic components in MIL-HDBK-217, but also the MIL-HDBK-217 contains two calculation methods of the
problem is establishing good requirements. This is reliability: Parts Count and Parts Stress. The methods
because the statistics of the causes of failure shows that differ in the amount of information required as input data
nearly 70% of the failures belong to the so-called non- for models of failure rate of elements. The Parts Count
hardware failures, so we addressed this problem and also calculation of reliability is performed in the early stages
the way to solve it. of the system design (when the electrical diagram is not
yet known) and based on the mean values of the failure
Beside that, the problem is in the inadequate cooperation rate of the elements and environmental conditions in
of engineers who design the equipment and engineers which the system or its parts is used. Generally, serial
who calculate the equipment reliability, especially in configuration of the elements according to the reliability
inadequate estimate of the input data in the calculation of (reliability block diagram) is assumed. The Parts Stress
reliability. reliability calculation requires a large amount of
information. It is done in the later stages of design when
2. LIMITATIONS AND REVITALIZATION OF the electrical and other components that will be installed
MIL-HDBK-217 are known, it means that the real failure rate of elements
based on the actual load (electrical, temperature, etc) and
MIL-HDBK-217 is a military manual for the calculation the actual surrounding conditions must be known and the
of the reliability of electronic devices. It was developed in actual configuration of elements with respect to reliability

835
mus be determined [3, 4, 5]. engineering evaluation of the input data in the calculation
must be performed. This requires additional knowledge of
Although other industrial and commercial standards
the engineers involved in the design of devices and
appeared, MIL-HDBK-217 is still used by more than 80%
reliability calculations, and close cooperation of the
of engineers doing the calculation of reliability (on the
engineers, which in practice is not adequate, so an
second place are PRISM and Telcordia) [2]. adequate education of engineers in the field of reliability
During the last 20 years of the last century, when is necessary.
updating of the MIL-HDBK-217E was done, it was
concluded that the model of constant failure rate (it means 3. PROBLEMS OF ESTABLISHING
the exponential distribution of failures) on which it is REQUIREMENTS FOR RELIABILITY
mainly based, is not reasonably applied to each type of
the element and the system, if it is not really justified [2]. The practice of reliability calculations showed that the
This has led to the fact that the Department of Defense, problem is not only calculation of prognostic reliability,
since 1994, began to stimulate the use of the industry but also establishing the good requirements for reliability.
standards such as the ISO 9000 series of standards for the Studies have shown that the distribution of causes of
quality assurance. failures in electronic systems is as in Table 1 [9]. In the
It is believed that the MIL-HDBK-217, as the basis of practice of contracting requirements for reliability only
almost all the conventional approaches to the calculation part of this causes of failures are assumed, generally only
of the reliability, has the following limitations [2]: those relating to the failures of the elements (parts,
hardware). In fact, the eight categories of causes of
− It has not been updated since 1995 and data for most failures, in the first column of the Table 1, can be grouped
integrated circuits have not been updated since in 1991, into two categories, as in the second column of the Table
so the most recent technologies are not included or 1: the inherent failures of hardware and non-hardware
defined; failures. The inherent hardware failures are those that
− In practice, significant deviations between the values of occur because of failure of parts before its expected „end-
reliability which were obtained by calculations based on of-life“ (or wearout) is reached, or due to variations in the
MIL-HDBK-217 and values really achieving in the use, quality (random failures, typicaly based on part quality
appeared [2, 6]; variability issues). In non-hardware failures "everything
else" is counted. From the Table 1, it can be also seen
− Models have never been subjected to an independent
that, in practice, there are 20% failures whose causes of
review by an association of engineers or technical
failures are not identified („No Defect“), and 9% whose
journals;
causes of failures are software bugs.
− Price of calculation is largely compared with the value
of the results and sometimes can direct on the wrong If we start from the assumption that the system original
path in the design of reliable electronic equipment; specified MTBF (initial MTBF required by the user) is
1000 hours, then such a request was made assuming that
− Multiple failure mechanisms can appear in
in the reliability calculation all of these causes were taken
semiconductor circuits with the same load factor: into account. However, the calculation of reliability takes
electrical voltage or temperature. The exponential
only the reliability of the hardware in account, and
distribution with only one parameter is good for
everything else is not taken into account, so the actual
individual failure mechanism of an element, but not for
MTBF will be 310 hours in use, as shown in the next to
the whole equipment.
the last column of the Table 1 (Corresponding Original
Efforts to update and revitalize the MIL-HDBK-217F Operational MTBF). Based on the system original
began in 2004. Version G was planned to be finished by specified MTBF and failure statistics in Table 1, such a
the end of 2009 [1]. The goal was to determine which result (really MTBF of the system or equipment), based
tools are used in industry to gather data for MTBF (Mean on [9], can be obtained by the relation
Time Between Failure). A group of representatives of the
MTBFs = 1 (1)
U.S. Government, Department of Defense and industry is λRe q + λNon − HW
formed, and researches in the field of space equipment are
sponsored. The IEEE Reliability Society (IEEE, The where
Institute of Electrical and Electronics Engineers) also
participated. It is not known that the version G appeared, λRe q = 1 (2)
but the Handbook of 217Plus Reliability Prediction MTBFRe q
Models appeared in 2006. This handbook is no longer
free, and can be found on the website of RIAC and
(Reliability Information Analysis Center)
http://www.theriac.org/riacapps/search/?category=all%20 u Non − HW
λNon − HW = λ (3)
products&keyword=217plus (accessed 17/04/2014), u Inh. HW Re q
where so called physics of failure should be implemented
[7, 8]. where:
In addition, it is recommended that not only MIL-HDBK- λNon − HW - failure rate contribution of all Non-
217 is to be used in the calculation and serious Hardware failure categories (in failures per hour, h-1),

836
u Non − HW - percent contribution (decimal) of Non- requirements for the reliability of software is often not
Hardware failure categories to overall system failure rate, adequately specified and the requirement for human
reliability almost not specified at all. Reliability of the
u Inh. HW - percent contribution (decimal) of Inherent system, considering hardware reliability, software
Hardware failure categories to overall system failure rate, reliability and human reliability, supposing that failures
are mutually independent, can be obtained by the next
λRe q - operational failure rate required by the end user (in
relation
1/hour)
RS ( t ) = RHW ( t ) RSF ( t ) RH ( t )
MTBFRe q - operational MTBF required by the end user
(in hours).
where RHW , RSF and RH hardware reliability,
So if we need MTBF during use of the system of 1000
software reliability and human reliability, respectively [1,
hours, MTBF, which should be specified (required) for
the inherent hardware reliability in the application, can be 2, 10].
obtained by relation Author's experience has also shown that the calculated
prognostic MTBF should be at least or about twice of the
MTBFInh. HW Req = 1 (4)
uInh. HW ⋅ λReq required MTBF in order to have the operational (actual,
correct) MTBF equal to the required (original) MTBF [11
] and that was applied as a rool when Parts Count
MTBFInh. HW Req = 1 = 3226 h reliability calculation has been made [12, 13, 14].
0,31⋅ 0,001
Calculation of the hardware reliability is also faced with a
and MTBF that would apply to non-hardware failures can number of problems. In [15] it is stated that there is no
be obtained by relation standard method for creating hardware reliability
prediction, so predictions vary widely in terms of
MTBFNon−HW. Req = 1 (5) methodological rigor, data quality, extent of analisys, and
uNon−HW ⋅ λReq uncertainty, and documentation of the prediction process
employed is often not presented. Because of that IEEE
MTBFNon−HW. Req = 1 = 1449 h has created a standard IEEE Std.1413 (Standard
0,69 ⋅ 0,001 Framework for the Reliability Prediction of Hardware) in
2009. The IEEE 1413 Standard focuses not on selecting
Table 1 shows that 9% of the failures is caused by the or using any specific prediction methodology, but on rigor
software, and the 20% failures is with undetermined of what methodology is selected. This standard provides a
causes (No Defect), in which, possibly, there are those clear set of the guidelines that, when followed, will
which causes are human errors. Although there are many provide the user of the prediction a better undestanding of
models of software reliability and human reliability, the the true value of the prediction.
Table 1: Original and Correct System MTBF
Corresponding
Original Correct Contribution to Corresponding
Original
Failure Category Specified Specified Operational Correct
Operational
MTBF MTBF Reliability Operational MTBF
MTBF
Parts Inherent 22% 3226 hours
1000 hours 1000 hours
Wearout Hardware, 31% 9%
System
4%
Management
Design 1000 hours 9%
Non-Hardware,
Software N/A 9% 450 hours 1449 hours
69%
Manufacturing 15%
Induced 12%
No Defect 20%
TOTAL System 100% 310 hours 1000 hours

4. SOME PROBLEMS WITH SOFTWARE in different nature of software compared to hardware.


RELIABILITY Although defined as probabilistic function software
reliability is not a direct function of time [16]. Another
Software reliability is an important attribute determining problem is that techniques for software reliability
the quality of the software as a product. As we said, there prediction are rarely used as routine software engineering
are many models of the software reliability, but the practices and it is the case for the army programs. It calls
requirements for the reliability of the software are often for collaboration between software and reliability subject
not adequately specified if specified at all. The problem is matter experts to take appropriate steps to include
software into the reliability case for the system [17].

837
The real issue with reliable software is that the critical [3] Pokorni S., Reliability and Maintainability of
function fails safe. Failing safe is often misunderstood technical systems, Military Academy, Belgrade,
and is often misinterpreted as never failing. Software 2002. (in Serbian)
safety and software reliability are allies in the realization [4] Pokorni S., Reliability of information systems,
of their mutual goal of developing safe and reliable textbook, Information Technology School, Belgrade,
software. And again there is a need for cooperation 2014. (in Serbian)
between software and reliability engineers. But, a few
[5] Pokorni S., Approach to determining the reliability of
educational institutions or industry professionals teach the
electronic devices in the operating conditions of the
basics of software reliability and its dependence upon
aircraft, magister thesis, Faculty of electrical
software safety to be effective [18].
engineering, Sarajevo, 1985. (in Serbian)
5. CONCLUSION [6] Pokorni S., “Reliability and Maintainability of
technical systems: theory and practice“, ICDQM
2009, proceedings, pp. 44-57, Belgrade, 25-26. june
Although the military manual for the calculation of the
2009. (in Serbian)
reliability of electronic devices MIL-HDBK-217, updated
six times up to now (latest version is F), is used for almost [7] Pokorni S., Ramović R., “The role of physics of
50 years, and is still used by more than 80% of engineers, failure in assessment of reliability of up to date
it has a number of limitations and the use of exponential technical systems”, OTEH 2005, Belgrade 06-07
distribution on which MIL-HDBK-217 is mainly based, is December 2005. (in Serbian)
not always justified. MIL-HDBK-217 has not been [8] Pokorni S., Ramović R., “Assessment of high
updated since 1995 and the efforts to revitalize it reliability of military equipment by physics of
indicated that other approaches must be combined, such failure”, OTEH 2007, Belgrade, 03-05 October 2007.
as the so-called physics of failure, although physics of (in Serbian)
failure is not applicable for all the periods of failure rate [9] Nicholls D., Lein P., “When Good Requirements
curve or equipment life time, but is applicable for wearout Turn Bad”, Reliability and Maintainability
region of the failure rate curve. Symposium, RAMS 2013.
The cause of the discrepancy between the calculated and [10] [10] Pokorni S., “Reliability human-technical
actual reliability during use is not only MIL-HDBK-217, systems”, Novi Glasnik 6/1999, pp. 11-17 (in
but also inadequate requirements for reliability, where it Serbian)
is not taken into account that a significant percentage of [11] Pokorni S., Pavlović S., Study of checking the
the total number of failures, so called non-hardware reliability of the device for improving the stability
failures (which includes not only failures due to defects in and controllability of aircraft, VZ "ORAO",
the software that is an integral part of the device). In Rajlovac, Sarajevo, 1990. (in Serbian)
practice, the reliability calculation is based only on
hardware failures (only hardware failures are taken into [12] Pokorni S., Parts count reliability calculation of
account). Using an example, it has been shown how this controllable camera mount (project), Energoinvest,
could be taken into account in this paper. Sarajevo, 1990. (in Serbian)
[13] Pokorni S., Milanović V., Reliability calculation of
In setting the requirements for reliability to adequately electronic switch for car headlights AUTO LIGHTS
take the reliability of the software into account is 12-24 for Elektrometal - Niš, Faculty of electrical
necessary (the software is becoming more prevalent and engineering, Belgrade, 2008. (in Serbian)
performs important functions), and also reliability of
human and access software and human reliability. [14] Pokorni S., Reliability calculation of microwave low-
noise amplifier, MTT INFIZ, Belgrade, 2014. (in
A serious engineering evaluation of the input data in the Serbian)
calculation is also important, and this requires close [15] Elerath G. J., Pecht M., “IEEE 1413: A Standard for
cooperation of engineers involved in the design of the Reliability Predictions”, IEEE Transactions on
equipment and engineers who deal with the reliability Reliability, Vol. 61, No. 1, March 2012, pp. 125-129
calculations, and this requires certain education from the
reliability for design engineers. [16] Kapur P. K., „Measuring Software Quality (State od
the Art)“, 5th DQM International Conference Life
Cycle Engineering and Management ICDQM 2014,
REFERENCES
proceedings, pp 3-45 27-28, Belgrade, 27-28 June
[1] Gullo L., “The Revitalization of MIL-HDBK-217”, 2014
IEEE Reliability Society 2008 Annual Technology [17] Payne S. R., “A practical Approach to Software
Report (part of), IEEE, 2008. Reliability for Army Systems”, Reliability and
[2] White M., Bernstein J. B., Microelectronics Maintainability Symposium, RAMS 2014.
Reliability: Physics-of-Failure Based Modeling and [18] Naylor W., Joyner B., “A Discourse on Software
Lifetime Evaluation, National Aeronautics and Space Safety and Software Reliabiliy”, Reliability and
Administration, Jet Propulsion Laboratory California Maintainability Symposium, RAMS 2014.
Institute of Technology, 2008.

838

View publication stats

You might also like