IEEE TRANSACTIONS ON RELIABILITY, VOL. R-36, NO.

4, 1987 OCTOBER 397

MIL-STD-781C: A Vicious Circle

Harold E. Ascher Notation

Naval Research Laboratory, Washington DC
s- implies: statistical(ly)
t global time.
Key Words - MIL-STD-781, Repairable system, Exponen- xi local time since failure i - 1, i = 1, 2, ..; x0o 0.
tial distribution, Homogeneous Poisson process, Force of mor- hx(x) force of mortality of time to failure.
tality, "Failure rate" hxL(xi) force of mortality of ith time between successive
failures, i = 1, 2, 3,
Reader AIds N(t) number of failures in (0, tl.
Purpose: Tutorial E{N(t)# s-expected (mean) number of failures in (0, t].
Special math needed for explanations: Probability v(t) rate of occurrence of failures at time t.
Special math needed to use results: None
Results useful to: Reliability theorists and engineers Other, standard notation is given in "Information for
Readers & Authors" at rear of each issue.
Abstract - Reliability practitioners have confused theorists
about the true area of application of MIL-STD-781C. This has
occurred because theorists have neglected to explain important 2. TRUE SCOPE OF MIL-STD-781C
basic concepts to practitioners. This paper analyzes the situation MIL-STD-781C is intended for the testing of
in detail. repairable systems. This is demonstrated in terms of four
aspects of the standard:
1. INTRODUCTION 1. Description of the action to be taken following a
MIL-STD-781C [1] is intended for the reliability failure.
testing of repairable systems. In practice, it is usually, if 2. Minimum operating time requirements on all items
not always, used solely for that purpose, rather than for on test before an accept decision can be made.
the testing of nonrepairable items. Nevertheless, the 3. The discrepancy between the maximum number of
research literature on this standard [2-10] is written as if items to be tested, relative to the maximum number of
the standard were intended solely for nonrepairable items. failures which can occur.
The results of these papers apply to the true subject of the 4. The assumption made throughout the standard,
standard, repairable systems, only under the unstated after the first sentence of its first paragraph, that complex
assumption that times between successive failures are items are being tested.
statistically independent and identically distributed (i.i.d.).
Moreover, this assumption of renewal, which would be 2.1 Action Following Failure
plausible for successive parts placed in a socket is usually
unrealistic when applied to an entire system; how is an en- Actions to be taken following failure of an item are
tire system's age reset to zero when only a very small pro- discussed in five places in the standard. In every case, the
portion of its parts - or perhaps, no parts at all - are standard makes it clear that the failed item is to be repaired
replaced during a typical repair? The reasons for the and put back on test. For example, paragraph 5.6 Failure
discrepancy between the standard and the papers which actions begins, "On the occasion of a failure, entries shall
claim to extend its applicability are the primary subject of be made on the appropriate data logs and the failed equip-
this paper, since the confusion about the standard provides ment shall be removed from test and repaired with
a particularly poignant illustration of basic misconceptions minimum interruption to the equipments continuing on
which pervade the reliability community. test. After a failed and repaired equipment has been
returned to operable condition, it shall be returned to test
Terminology ... " Similarly, the following statement, "When a failure
occurs, the equipment shall be repaired and testing re-
Global time: The total operating time of a repairable sumed, ..." appears in paragraph 5.5.4 Testing sequence;
system, regardless of the occurrence of one or more failures. and paragraph 5.1.9 Preconditioning burn-in includes the
Local time: The operating time since the most recent sentence, "Failures occurring during preconditioning are
failure of a repairable system. not chargeable for accept/reject decisions, but shall be
Force ofmortality: The ratio ofthe probability density recorded, analyzed in accordance with 5.6.2 and ap-
function divided by the survivor function of a distribution. propriate repair action taken." Paragraphs 30.4 Failure
Rate of occurrence of failures: The time derivative of a tag and 30.5 Failure report of appendix D also are worded
mean number of fallures (NOT to be confused with failure as if repair - rather than replacement of a failed
rate). nonrepairable item -always follows failure.
US Government work not protected by US copyright
398 IEEE TRANSACTIONS ON RELIABILITY, VOL. R-36, NO. 4, 1987 OCTOBER

Several other statements in the standard also are Classification begins, "This standard is applicable to six
worded as if the possibility of its application to the testing broad categories of equipment, distinguished according to
of nonrepairable items was not considered. For example, their field service applications: Category 1 Fixed ground
paragraph 5.8 Verifying repair reads, in entirety, "The equipment, Category 2 Mobile ground vehicle equipment
contractor shall verify the effectiveness of the repair of ..." Most nonrepairable items eg, transistors, could be used
failed equipment before resuming reliability testing." in any of the 6 listed categories. Other paragraphs which
Similar statements ie, statements which are based on the a clearly assume that only complex items are to be tested in-
priori assumption that only repairable systems are to be clude 1.4, 3.1.3, [Link], [Link], [Link], [Link], [Link],
tested, appear in paragraphs 4.4, [Link], 5.6.2, 5.10, 6.21, [Link], 4.3.1, [Link], 4.4, 5.1.5, 5.6, 5.6.2, 5.10, 5.11, as
and 20.i of appendix F. well as numerous citations in appendixes A-F.
Some complex items, eg, torpedoes or missiles,are not
2.2 Minimum Operating Time repairable under use conditions. However, very few, if any,
complex items would be nonrepairable during testing under
Paragraph 5.7.1. General (under 5.7 Determination of MIL-STD-781C conditions, since failed items are readily ac-
compliance) states, "Unless otherwise specified by the pro- cessible for repair. Hence, without even considering the
curing activity, an accept decision shall not be made at a numerous explicit references to repair discussed here in sec-
point where any single equipment of the demonstration tions 2.1-2.3 above, there would be very little, if any, loss of
sample shall have accumulated less than one-half of the generality in assuming that MIL-STD-781C, which obvious-
average creditable test time of all samples." An ab- ly applies only to complex items, applied only to repairable
breviated version of this sentence appears in paragraph systems.
[Link] Test length. This requirement obviously would be
totally unworkable if nonrepairable items were to be tested 2.5 "Up Front" Evidence
under the provisions of MIL-STD-781C.
In opposition to the large amount of evidence presented
Number of
Tailurested
Number of Failures
ItemsvsMaximhere in sections 2.1-2.4, about complex (and hence,
repairable) systems, the standard supplies few hints that it is
The number of items to be tested is discussed in meant to be used for the testing of nonrepairable items. Un-
paragraphs [Link] Equipment quantity, [Link] Reliability fortunately, most of the wording which appears to imply ap-
first article (preproduction) qualification test, [Link] plicability to nonrepairable items is "up front". Specifical-
Reliability production acceptance test and 6.2.j Ordering ly, the Standard's title uses the term "exponential distribu-
data. Paragraph [Link].b puts an upper bound of 20 on tion" and the first sentence of the first paragraph (1.1
the number of equipments to be tested and the other Scope) states, "This standard covers the requirements for
paragraphs specify very low minimums, without address- reliability ... tests ... for equipment that experiences a
ing the possibility of a potentially large number of failures distribution of times-to-failure that is exponential."
occurring before a decision can be made. For example,
[Link] states that, "Normally, at least two equipments are 2.6 Two Vicious Circles
used, depending on equipment availability, schedule and
test duration." Nowhere in the standard is there any hint MIL-STD-781C never explicitly states that it is in-
that failed equipments are to be replaced. Since Test Plan tended only for the testing of repairable systems. Never-
IC, page 66, allows for acceptance with as many as 40 theless, the preponderance of its wording shows that it is in-
failures, (and hence, for rejection with 41 failures) how are tended solely for that purpose. Moreover as shown here in
so many failures to be squeezed out of 20, much less two,
nonrepairable items? Other test plans terminate with sections 2.1-2.4 and especially in section 2.2, strictly speak-
smaller maximaa on the possible number of failures but ing the standard cannot be applied to nonrepairable items.
smostofthemall
owm forthepossibliy ofm failures, bth Nevertheless, the "up front" wordings of the standard's ti-
the number of equipments discussed in the four tle and scope have misled theorists into thinking that the
paragraphs listed above. Clearly, it is implicitly assumed in standard applied solely to nonrepairable items. Clearly, the
these paragraphs that failed equipments will be repaired, practitioners who wrote MIL-STD-781C have confused the
after failure, in order to obtain adequate sample sizes for theorists. What I will show, however, is that the theorists
acceptance or rejection. confused the practitioners first! In other words, in section 4
here I show that the vicious circle of figure 1 is a result of,
2.4 Only Complex Items are Considered and also contributes to, the related vicious circle of figure 2.
First, however, some basic concepts for the homogeneous
Many paragraphs of the standard are worded as if only Poisson process and for the exponential distribution are
complex items are to be tested. For example, paragraph 1.3 discussed here in section 3.
ASCHER: MIL-STD-781C: A VICIOUS CIRCLE 399

0, when X is exponentially distributed. I refer to hx(x) as

the force of mortality, since as stressed in [11] and [12, pp
_CONFUSED 153-157] other terminology has led to confusion.
CONCEPTS The force of mortality is a relative rate since it is
defined as the derivative of the cumulative distribution
function, normalized by the time varying survivor func-
tion, 1 - Fx(x). In contrast, the probability density
SOME||CONFUSED, function, - Fx(x), is an absolute rate since, by definition,
SUBTLE CONFOUNDED
CONCEPTS TERMINOLOGY it is an "unnormalized" derivative.

3.2 Basic concepts for stochastic point processes

Some fundamental concepts for modeling the times

between successive failures of a repairable system are
CONFUSED, sketched below. These basic concepts are presented in
CONFOUNDED much greater detail in [12, pp 17-29].
NOTATION A stochastic point process is a mathematical model
for a physical phenomenon characterized by highly local-
Fig. 1. A Vicious Circle ized events occurring randomly in a continuum. In the
present application, the continuum is time and the highly
localized events are failures which are assumed to occur at
instants within the time continuum.
TH EO RISTS Figure 3 shows a portion of a sample path of a
CONFUSE stochastic point process, representing the successive
/t
|PRAiCTITIONERS N X failures of a single system. Repair times are not con-
PRACTITIONERS sidered, ie, repairs are assumed to be performed instan-
taneously. Note the distinction between global time, t,
which measures the total time since the system was first put
into operation, and local time xi, i = 1, 2, 3, ..., where xi is
the local time since the (i - 1)st failure, x0 - 0. Define the
PRACTITIONERS random variable N(t) to be the number of failures which
CON FUSE occur
(mean)innumber
the interval (0, t]. inThen
of failures (0, E{N(t)}
t] and the rates-expected
is the of occur-
THEORISTS rence of failures is:
d
Fig. 2. Another Vicious Circle v(t) d E{N(t)}.

3. PROPERTIES OF EXPONENTIAL DISTRIBUTION The rate of occurrence of failures is obviously an absolute

AND HOMOGENEOUS POISSON PROCESS rate, since it is an unnormalized derivative.
3.1 Exponential Distribution
t
It is well known that when a random variable, X, is xx2 l_ 3 x4
exponentially distributed, its cumulative distribution func -______ x
T1 3
Ft(x)e Pr{X6x}=1-e , >O0, x>O.
It is also well known that the function h(x)e(x)

d dxFx(x) s
~FA(x) < 1, reduces to the constant, -y, for x T4--
1 -F(x{) Fig. 3. System Arrival and Interarrival Times
400 IEEE TRANSACTIONS ON RELIABILITY, VOL. R-36, NO. 4, 1987 OCTOBER

3.3 Homogeneous Poisson process (HPP) failures is very small, however, this is at least as realistic as
assuming, for example, that the number of rotten eggs in
Among several equivalent definitions of the HPP, the an ordinary carton is Poisson distributed, which results in
most straightforward one to introduce here is that it is a positive probability of thirteen or more rotten eggs.)
nonterminating sequence of statistically independent and
identically distributed (i.i.d.) exponential random 3.4 Terminating HPP
variables, ie, each of these independent exponential ran-
dom variables has force of mortality, -y. It is well known A sequence of i.i.d. exponential distributions, which
that in the case of the HPP. terminates after a known finite number, M, of events, is
called a terminating HPP. This process is not equivalent to
d an HPP. In fact, the very different properties of these two
v(t) - E{N(t)} = 'Y t > 0. models can be used to emphasize the difference between
dt
the relative rate of an exponential distribution and the ab-
Therefore, under the HPP, solute rate of an HPP. It is shown in [12, pp 61-62] that for
a sequence of M i.i.d. exponentials, M known a priori, the
hxi(xi) = y = v(t), i = 1, 2, ..., absolute rate v(t) _d E{N(t)} is
dt
but hxi(xi) and v(t), though numerically equal, are not
equivalent. That is, v(t) = 'ye t1 + eyt + ... + (I0
L ~ (M- 1)! i
dxi
- Fxi(x1) ( d In the special case M - 00, v(t) = -ey'
e"t - -y, as an-
1- Fxi(x) --hxi(xi) v(t) _-E{
=

dt
N(t)}. ticipated, since for M -X the process is an HPP. For any
M < oo, however, v(t) is strictly decreasing for 0 < t < oo
In other words, the relative rate of the exponential and lim v(t) = 0. In the special case M = 1, the process
distribution and the absolute rate of the HPP have dif- reduces to an exponential distribution of time to failure.
ferent interpretations. For example, Even in this case v(t) does not reduce to the relative rate of
the exponential distribution. In fact, for M = 1,
10t
v(y)dy = s-expected (mean) number of failures in (0, t]
v(t) = ye
but
That is, when the process terminates after one failure its
10' hxi(y1)dyi absolute rate reduces to the probability density function of
the exponential distribution, rather than to the force of
cannot be interpreted as an s-expected number of failures mortality of that distribution. How could the absolute rate
in (0, xi]. of a process consisting of a sequence of M i.i.d. exponen-
Specifying an HPP with absolute rate y, therefore, is tial distributions be equivalent to the force of mortality of
equivalent to specifying a nonterminating sequence Of an exponential distribution, when these quantities are not
s-independent exponential distributions, each with relative even of the same functionalform for any M < 00?
rate -y. On the other hand, specifying that time to first It should come as no surprise that for M = 1, v(t)
failure is exponentially distributed, with relative rate 'y reduces to the probability density function, rather than the
does NOT imply an HPP. In fact, specifying that time to force of mortality. Recall that the probability density func-
first failure is exponential by itself does NOT specify tion is the absolute rate' of change of the cumulative
ANYTHING about what happens after that first failure, distribution function.
eg, it is possible that the process of failures terminates after
the first failure. The times between successive failures must
be s-independent and identically exponentially distributed 4. CAUSES OF VICIOUS CIRCLE ONE
(ie, they must have the same parameter and the process
-y)
mustconinuinefintel, wth o uper oun onthe4.1 Will the Real Failure Rate Please Stand Up?
possible number of failures, under an HPP. Of course, in Section 3 here, shows that the exponential distribution
practice, a repairable system will not undergo an unending and the HPP are distinct models. Nevertheless, the
sequence of failures but assuming that the number of numerical equality between the force of mortality of the
failures in the interval (0, t] is Poisson distributed (a well exponential distribution and the absolute rate of occur-
known property of the HPP) implies positive probability rence of the HPP lends credence to the erroneous belief
of any finite number of failures, no matter how large. (If that these models are interchangeable. Moreover, the fact
the probability of an unrealistically large number of that in practice, the term constant failure rate is usually
ASCHER: MIL-STD-781C: A VICIOUS CIRCLE 401

used interchangeably for both of these rates seems to cor- reliability problems can be tackled with a distribution
roborate the impression that these models are equivalent. function, it would be difficult to devise a better means of
Consider the following quotation from paragraph confusing such practitioners, if one intended to confuse
[Link] of MIL-STD-781C, "The exponential assumption them! Finally, most theorists understand the distinction
infers [sic] a constant failure rate; therefore, these test between the relative rate of a distribution and the absolute
plans are not appropriate, and will not be applied for any rate of a stochastic process. Theorists almost invariably
test planned for the specific purpose of eliminating design define the term failure rate to be equivalent to what I have
defects or infant mortality type failures. " Since a sequence called force of mortality. Nevertheless, theorists are almost
of exponential distributions, with smaller and smaller asproneaspractitionerstousetheterm,failurerateinter-
forces of mortality, can, as has been, used for modeling changeably for both hx(x) and v(t). After all,failure rate is
reliability growth, [13], it is clear that the wording in the a reasonable term to describe the derivative of an
standard is based on an erroneously assumed equivalence
between the exponential distribution and the HPP. That is, s-expected number of failures.
the exponential distribution's constant failure rate sup- of theIntitle opinion,
my and the practitioners' is a direct wording
scope of MIL-STD-781Cmisleading result of
posedly implies the HPP's constant failure rate. As
pointed out in [14], "...'failure rate' is used to mean so the theorists' neglect of many basic, important issues.
many things that it has become meaningless."
4.4 Summary
4.2 Will the Real Model of MIL-STD-781C
Please Stand Up? Theorists have not made it clear to the practitioners
that the exponential distribution and the HPP are not in-
As shown in section 2 here, MIL-STD-781C is really terchangeable models. In turn, the practitioners have writ-
concerned with times between successive failures of ten MIL-STD-781C in such a way as to confuse theorists
repairable systems, rather than times to failure of into believing that the standard applies solely to
nonrepairable items. In spite of what is stated in the stan- nonrepairable items. The theorists then write papers [2-101
dard's title and scope, therefore, its real model is the HPP which depend crucially on the i.i.d. assumption without
- but this model is never mentioned in the Standard. ever stating the assumption, since it would almost
Given its concentration on the exponential distribution, automatically hold if nonrepairable items were tested.
rather than the HPP, it is essential for the standard to state Then, since the theorists usually use the term system, and
formally the assumption that repair restores a failed in practice most systems are repairable, especially if tested
system to its new condition. This was not done, however, under MIL-STD-781C conditions, practitioners think that
since most practitioners erroneously believe that the ex- any system can be modeled with a distribution of time to
ponential distribution's constant failure rate implies failure, rather than a stochastic process consisting of times
renewal during repair. between successive failures. In short, the vicious circle just
described has perplexed most members of both groups for
4.3 Will the Theorists Address Fundamental Issues? the entire history of the reliability discipline. Of course,
neither group intends to confuse the other but, in practice
Most practitioners misunderstand many basic concepts, "... confusion can, and has, reigned." [16]
including the distinction between the exponential distribu-
tion and the HPP. In part, this is the result of close, and 5. WIDER RAMIFICATIONS
often very subtle, connections between concepts and models
for nonrepairable and repairable items [12, pp 32-33 and The vicious circle of figure 2 is only one aspect of the
162-168]. These close connections should have prompted more general vicious circle displayed in figure 1. This latter
theorists into making special efforts to clarify and em- vicious circle is discussed in [12, chapter 8] which em-
phasize the very important distinctions between the concepts phasizes the widespread misunderstanding of many basic
and models for nonrepairable items (distribution functions) issues within the reliability community. A review of [12],
and repairable items (stochastic point processes). In actuali- published in [17], states, "The thrust of the message comes
ty, however, theorists (including, but certainly not restricted in chapter eight where a multitude of misconceptions,
to, those who wrote [2-10]) have usually ignored these many already mentioned in the review chapters, are re-
distinctions, concentrating instead on models and tech- vealed. Having reached this point in the book readers can
niques for nonrepairable items. Moreover, even when be excused for thinking that perhaps everything they are
theorists do broach the subject of repairable systems, they currently doing in the reliability area is wrong!" Another
often phrase things largely or predominantly in terms of review [18] adds, "If and when more analyses of real world
distribution functions [15]. This can be technically correct, data are performed, they should be performed correctly -
even though the underlying model is a stochastic point pro- if practitioners follow standard texts, however, analyses of
cess, because of some of these subtle connections [12, pp repairable systems failure data will not be performed
32-33]. However, since most practitioners believe that all correctly."
402 IEEE TRANSACTIONS ON RELIABILITY, VOL. R-36, NO. 4,1987 OCTOBER

The vicious circle about MIL-STD-781C, where [7] E. R. Montagne, N. D. Singpurwalla, "Robustness of sequential
theorists and practitioners unintentionally - but very ef- MIL-STD-781C life testing procedures", Theory ofReliability, eds.
fectively take turns in confusing each other is Serra & Barlow, North-Holland, 1986, pp 169-183.
.[8] N. D. Singpurwalla, "Some current research in reliability: an over-
a fec
par-
y -
ticularly poignant example of why the reliability field has view", Reliability in the Acquisitions Process, eds. Depriest &
gotten into the mess described above. The concepts, Launer, Marcel Dekker, 1983, pp 101-113.
models, and techniques for nonrepairable items do not [91 N. D. Singpurwalla, "Military standards for fixed-length life tests",
carryover
carry over orepirabl systems.
to repairable systms. IIn a sense,
sese, iti is ufor-
unfor- Encyclopedia of Statistical Sciences, eds. Kotz & Johnson, vol 5,
1985, pp 488-490.
tunate that one often gets the right answer under the [10] N. D. Singpurwalla, "Military standards for sequential life testing",
mistaken impression that the model being invoked is an ex- Encyclopedia of Statistical Sciences, eds. Kotz & Johnson, vol 5,
ponential distribution when the correct model is an HPP; 1985, pp 490-493.
this certainly has contributed to the confusion. Neverthe- [11] W. A. Thompson Jr., "On the foundations of reliability",
Technometrics, vol 23, 1981, pp 1-13.
less, even in the most special case of the HPP, one does not [12] H. E. Ascher, H. Feingold, Repairable Systems Reliability: Model-
always get the correct answer under the incorrect belief ing, Inference, Misconceptions and Their Causes, 1984, Marcel
that the underlying model is the exponential distribution, Dekker.
see [19] and [12, p 151 and p 160]. [13] Z. Jelinski, P. B. Moranda, "Software reliability research",
seem[19] Statistical
Almost anything that isiS called
anything that or equipment
system, orequipment
called a system,
a
Academic Computer Performance Evaluation, ed. Freiberger, 1972,
Press, pp 465-484.
as opposed to apart is repairable, at least under MIL-STD- [14] M. Shaked, Review of [12], Zentralblatt fur Mathematik, 1985, p
781C test conditions (and usually under operational condi- 543.
tions as well). Because of this fact, after the first sentence [15] H. W. Block, W. S. Borges, T. H. Savits, "Age-dependent minimal
of its scope, MIL-STD-781C is written as if the possibility repair", J. Applied Probability, vol 22, 1985, pp 370-385.
of its application to nonrepairable items simply did not oc- [161 R. G. Easterling, Review of [12], Technometrics, vol 27, 1985, pp
439-440.
cur to those who wrote it. [17] A. Veevers, Review of [12], J. Royal Statistical Society Series C, vol
In the real world, repairable systems dominate, as op- 76, 1986, p 76.
posed to the reliability literature which concentrates on [18] G. W. A. Dummer, Review of [12], Microelectronics and Reliabili-
models for nonrepairable items. It is time to break the ty, vol 26, 1986, p 993.
..ciou,
Vpicous circle whereby theorists pay little attention to
s circle whereby theorists little
pay attention to [19] H. E. Ascher, H. Feingold, "The aircraft air conditioner data
revisited", 1979 Proc. Ann. Reliability and Maintainability Symp.,
repairable systems, which leads practitioners to pose pp 153-159.
repairable system issues as if nonrepairable items were in-
volved. Since theorists have confused practitioners so THOR
thoroughly, it is up to them to take the first step by clarify- AU
ing basic concepts. Harold E. Ascher; Naval Research Laboratory; Code 5326; Washington,
DC 20375-5000 USA.
Harold Ascher is an Operations Research Analyst at the Naval
REFERENCES Research Laboratory, Washington DC. He has participated in reliability
programs on the Poseidon and Trident missiles and has performed
[1] US Department of Defense, Reliability Design Qualification and reliability growth analyses on marine gas turbines such as those installed
Production Acceptance Tests: Exponential Distribution, MIL-STD- in Spruance class destroyers. He was a member of the Steering Committee
781C: US Government Printing Office, Washington, DC, USA, of the Electronic Systems Reliability Workshop sponsored by the Joint
1977 October. Logistics Comanders in 1975. He was also a member of a tri-Service Com-
[2] H. L. Harter, A. H. Moore, "An evaluation of exponential and mittee for the preparation of MIL-HDBK-189 on reliability growth
Weibull test plans", IEEE Trans. Reliability, vol R-25, 1976, pp management. Mr. Ascher received his BS in Physics from City College of
100-104. New York in 1956 and his MS in Operations Research from New York
[3] H. L. Harter, A. H. Moore, R. P. Wiegand, "Sequential tests of University in 1970. He has written, with Mr. Harry Feingold, the first
hypotheses for system reliability modeled by a 2-parameter Weibull book which provides extensive coverage of repairable systems (Repairable
distribution", IEEE Trans. Reliability, vol R-34, 1985 Oct, pp Systems Reliability, published by Marcel Dekker). Mr. Ascher is also the
352-355. author of over 25 reliability oriented papers and is a member of the
[4] R. D. Neathammer, W. R. Pabst, Jr., C. G. Wigginton, "MIL- American Statistical Association, IEEE, Society of Reliability Engineers,
STD-781B reliability tests: Exponential distribution", J. Quality Sigma Xi, and the Washington Operations Research and Management
Technology, vol 1, 1969, pp 58-67. Sciences Council.
[5] J. M. Mogg, "The optimum number of test units for Military Stan-
dard 781-B", J. Quality Technology, vol 6, 1974, pp 101-107. Manuscript TR85-150/200 received 1985 December 30; revised 1987
[6] E. R. Montagne, N. D. Singpurwalla, "Robustness of sequential ex- March 20.
ponential life-testing procedures", J. American Statistical Associa-
tion, vol 80, 1985, pp 715-719. IEEE Log Number 15637 4 TR >