Anda di halaman 1dari 116


Distribution Category: RX



Ronald E. Wright
Judy A. Steverson
William F. Zuroff

Published May 1987


EG&G Idaho, Inc.

Idaho Falls, Idaho 83415

Prepared for the

Division of Reactor and Plant Systems
Office of Nuclear Regulatory Research
U.S. Nuclear Regulatory Commission
Washington. D.C. 20555
Under DOE Contract No. DE·AC07·76ID01570
FIN No. A6393

This study empirically develops frequencies of safety-significant pipe failures in

commercial nuclear power plants (NPPs). Its primary purpose is to update the pipe
break frequencies reported in the Reactor Safety Study, WASH-1400, which are used
in many risk analyses. The study involved reviewing various data sources for actual
piping failure events of significant magnitude. When extant in the documentation
reviewed, information was extracted concerning conditional factors such as the sys-
tem in which the failure occurred, operational mode of the plant, and size of the pipe
inVOlved to estimate conditional pipe break frequencies useful to risk analysts.
Because of the high quality piping used in NPPs, there have been few significant pipe
failures. An attempt was made to augment the analysis with synthetic data from a
Delphi approach, but the wide uncertainty bounds on the resulting estimates rendered
the results unsuitable for combining data.

FIN No. A6393-Data Development and Evaluation

NF!C~ 1102.
:1 lllLE ""110 ~UllITL£ J LEAVI'LANK

Pipe Break Frequency Estimation for

~~ucl ear Power Pl ants

~ AuTIoIOFlt~ May
I 1987 'f'EA~


R. E. Wright, J. A. Steverson, w. F. Zuroff May

I 1987


Idaho National Engineering Laboratory

EG&G Idaho, Inc. • FIN OA GR ...NT NUMIER

P.O. Box 1625

Idaho Falls, Idaho 83415 A6393

Offi ce of Nuclear Regulatory Research Techni cal

U. S. Nuclear Regulatory Commission
b. PERIOD COVERED (lncl"•• w",.,J
\~ashington , D.C. 20555

13 "'BSTFlACTI20()wo'd~o'·.',J

This st udy empirically develops frequencies of safety-significant pipe failures in commercial nuclear power
plants (NPPs). Its primary purpose is to update the pipe break frequencies reported in the Reactor Safety
Study, WASH-1400, which are used in many risk analyses. The study involved reviewing various data sources
for actual piping failure events of significant magnitude. When extant in the documentation reviewed,
information was extracted concerning conditional factors such as the system in which the failure occurred,
operational mode of the plant, and size of the pipe involved to estimate conditional pipe break frequencies
useful to risk analysts. Because of the high quality piping used in NPPs, there have been few significant pipe
failures. An attempt was made to augment the analysis with synthetic data from a Delphi approach, but the
wide uncertainty bounds on the resulting estimates rendered the results unsuitable for combining data.

FIN No. A6393-Data Development and Evaluation

I. DOCuMt;"'l .",. ... L'SIS • KE "wO~OS'OfSCAIPTO~S 15 ... V.IL.....Il,.ITV


Un1imi ted

tl IOEI\;TI~ IEFlS.QPEN E"'OEO TERMS U~'crassified

/TI'I., ,."orO



The overall purpose of this study is to develop, by data collection and statistical analysis, new estimates of
frequencies of safety-significant pipe breaks in commercial U.S. nuclear power plants. To date, the WASH-
1400, Reactor Safety Study, has been a main source for these pipe break frequencies. At the time these
frequencies were estimated for the WASH-1400 study, approximately ISO U.S. commercial reactor years of
operation had been experienced and no significant pipe break had occurred at any nuclear power plant. By
the December 1984 cutoff date of this study, approximately 800 years of U.S. commercial nuclear power
operation had transpired. This study will determine if additional pipe break data of safety-significant impor-
tance have occurred that could be used to update and provide more specific pipe break failure rates than those
reported in WASH-1400.
The pipe break failure rates developed for WASH-1400 are for piping in which a pipe break would result in
a loss of coolant accident (LOCA). This piping was partitioned into three pipe size categories: 1/2 to 2 in.,
2 to 6 in., and >6 in. These pipe categories were based on the functional requirements of the emergency
core cooling systems (ECCS) typically found in nuclear power plants (NPPs). Table III, 2-1 in WASH-1400
also presented values of pipe break failure rates for pipes < 3 in. and> 3 in. for non-LOCA piping. For the
purposes of comparing the pipe break failure rates developed in this study for LOCA-sensitive piping with
those of WASH-1400, the rates reported in WASH-1400 are given below:

LOCA-Initiating Rupture Rates

(per plant per year)

Pipe Rupture Size Range

(in.) (90 070 ) Median
1/2 to 2 I x 10. 4-1 X IO- e I x 10-'

2to6 3 x 10-'-3 x 10-' 3 X 10- 4

>6 I x 10- 5_ I X 10-' I X 10-4

This study developed pipe break failure rates for two piping system categories: LOCA-sensitive and non-
LOCA-sensitive piping. For this study, LOCA-sensitive piping is piping in which a pipe break results in loss of
reactor coolant.
Non-LOCA-sensitive piping is associated with piping systems that would be used to help mitigate a core
melt sequence but if a piping failure occurred, no loss of reactor coolant would result.
Several additional considerations influenced the scope of this study:

• Small leak rates were not to be considered

• Only passive piping was to be considered; active components such as valves and pumps were not to
be considered
• Because this was to be strictly a data study, fracture mechanics techniques were not to be used
• As data permit, pipe break failure rates were to be developed for conditional factors such as specific
systems and plant states.

For IDCA-sensitive systems, failure (for the purposes of this study) is a leak rate of at least 50 gpm for
pressurized water reactors (PWRs) and 500 gpm for boiling water reactors (BWRs). These rates are based on
their respective normal makeup capability to the reactor coolant system. Leakage rates below these values
can be made up by the normal makeup system used to replenish reactor coolant and, therefore, do not require
the initiation of emergency core cooling.
For non-LOCA-sensitive systems, several factors were taken into consideration before a definition of
failure could be established. These include the magnitude of leak necessary to disable a system; availability
of population data, such as the number of welds and length of pipe for the system of interest from which
conditional pipe break frequencies can be derived; and quality of the pipe break data available. Based on

these factors, all pipe break failures were collected that had at least a I gpm leak rate for pipes with at least a
2 in. dia and all leak rates:::: 50 gpm regardless of pipe size. The failure rates derived for non-LOCA-sensitive
systems are given in the" Results" section of this report.
Data on only 19 pipe failures were determined to be useful in this study with no failures occurring in the
LOCA-sensitive systems (PWRs or BWRs) with a leak rate:::: 50 gpm.
For LOCA-sensitive systems, the pipe break failure rates are given here for P\VRs and BWRs separately
and combined as light water reactors (LWRs) in terms of the frequency of exceeding a specific leak rate. As
stated above, these leak rates are different for PWRs than for B\VRs. The values for the LOCA-sensitive
systems for this study are:

Leak Rate Operating Lower Point Upper

(gpm) Failures Years Bound Estimate Bound


50 to 500 o 484.73 o 0.0005 0.0062

>500 o 484.73 o 0.0005 0.0062


500 to 5000 o 313.36 o 0.0007 0.0096

>5000 o 313.36 o 0.0007 0.0096


>50 o 798.09 o 0.0003 0.0038

Note: because no failures were observed in LOCA-sensitive systems, these rate estimates depend only on
the accumulated operating years experienced, and are the same for any range of leak rates> 50 gpm. The
data do not permit more detailed assessments.
These rates differ from those reported in WASH-I 400 in two major ways. First, the rates are based on
800 reactor years of operation of U.S. NPPs and no outside data such as failure data from fossil electrical
generating plants, chemical processing plants, or foreign nuclear reactor power plants have been incorpo-
rated that could influence the calculated failure rates. This information was sought, but most was not
available or was not usable for this study. Second, the rates are in terms of exceeding a specific leak rate as
opposed to pipe size categories. The reasoning is that the pipe size categories relate to ECCS capacity where it
is implied that a leak in a small pipe means a small impact on the plant and a leak in a large pipe implies a
large impact in the plant. However, probabilistic risk assessment results have shown that leaks in large pipes
may have a small plant impact and a rupture in sma/! pipes may have a large impact. Many factors interact to
determine the relationship between break size and the resultant effect on the plant; system pressure being a
major one. Reporting the failures in terms of leak rates as opposed to pipe size corresponds directly with the
ECCS pumping capacities.
Early in the study, it became apparent that few pipe break failure events have occurred in the nuclear
industry. As a result, a Delphi approach was used to obtain pipe break failure frequencies from recognized
piping experts to supplement the experience data obtained in the literature search. A recognized statistical
method was used to integrate the two sources of information (experience and subjective) to derive the failure
frequencies to be reported.
The subjective data, however, resulted in such wide uncertainty bounds for the pipe break failure frequen-
cies that the results were deemed unsuitable for use by risk analysts. The methods and results are included in
the appendixes of this report for those interested in the methodology, but only the results obtained from the
experience data are presented for use by risk analysts.

Abo included in the appendixes are summaries of five other U.S. Nuclear Regulatory Commission piping-
related studies. They are (a) "Probability of Pipe Failure in the Reactor Coolant Loops of Westinghouse
P\\'R Plants," (b) "Probability of Pipe Failure in the Reactor Coolant Loops of Combustion Engineering
P\\, R Plants," (c) "Characteristics of Pipe System Failures in Light Water Reactors," (d) NRC Integrated
Program for the Resolution of Unresolved Safety Issues A-3, A-4, and A-5 Regarding Steam Generator Tube
Integrity," and (e) "Steam Generator Tube Experience."

The authors wish to express their appreciation to Dr. Spencer Bush, Everett Rodabaugh,
and Cindy Gentillon for their invaluable assistance in both the engineering and statistical
input throughout this study. We also wish to thank Dave Satterwhite for his contribution to
the review of related work included in this report, Sharlene Williams and Maureen Wright
for their assistance in preparing this report, and Dick Robinson and Pat Baranowsky of the
U.S. Nuclear Regulatory Commission for their data contributions and technical input.
We also deeply appreciate the technical experts who responded to our question-
naire, and the architectural engineering firm and nuclear power plant utilities that
supplied population data.







FAILURE DATA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7


RESULTS.......................................................................... 22

REFERENCES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28


FAILURE RATES............................................................... A-I


APPENDIX C-QUESTIONNAIRE RESPONSES....................................... C-I





The estimation of risk to the general public from was determined that a statistical approach would be
loss of coolant accidents (LOCAs) from nuclear pursued that would utilize t he knowledge of recog-
power plants (NPPs) is, in part, made up of contri- nized piping sysll'm experts and obtain from them
butions from pipe break frequencies. To date, the subjective estimak' of pipe break failure frequen-
\\'ASH-1400, Reactor Safety Study,1 has been the cies. This subjeclill' data would then be used to
primary source estimates for these frequencies. At supplement the experience daTa collected from the
the time of this report,1 only 150 reactor years of available literature ll' delermine the pipe break fail-
operation had transpired and limited experience ure frequencies to b", reported in this study.
data Ilere a\ailable for estimating pipe break fail- The approach \Ias to collect subjective pipe
ure frequencies. By the cutoff date for our study, break failure cstill1dl<;, lia a detailed questionnaire
December 1984, there \\ere approximately from six to twellc piping experts and, using an
800 reactor years of operation. established Bayesian statistical analysis technique,
The objecti\es of this study are: (a) determining combine these subjlYtilc' data with the collected
if sufficient pipe break failure data hale been field data. Becau,e th" re,uIts of this part of the
reporte'd in these additional 650 reactor years to study produced data with such large uncertainty
improle the uncertainties of the pipe break failure bounds, it was d"l'ided the rates obtained by thi,
frequencies currently in use by risk analysts; and approach would not be usable by risk analysts,
(bl determining if sufficient data exist to pro\ide However, the metlwdology is a useful approach to
more specific pipe break failure frequencies for research of this nature and is included in
conditional factor" such as pipe failures for spe- Appendix A for the illtereqed reader.
cific sy;,tems and failures per weld or per foot of The follO\ling ,c'Cl ion> of 1his report detail the
pipe for these ,pecific system" and failures based o\erall methodology used in this study plus a re\ie\1
on the operational mode of the plant. of the methodology u,ed in \\'ASH-1400, a descrip-
The '>cope of the study (sec Table I) is limited to tion of the obsened failll' ' and population data,
leaks in passi\e piping that are of sufficient magni- and the results obtai Ill'd flc":~ t he statistical calcula-
tude'to be significant contributors to core melt fre- tions using the operat iOlla! data. Additional appen-
quency. Acti\e components such as pumps and dixes provide nanati\e de,criptions of the failure
\alles are not considered. e\'ents, summarie, of related pipe break studies,
Early in the study, it became evident that the and details of the poil11 ,-,slimates and bounds cal-
applicable failure data are very sparse. Therefore, it culations for the <;'xperience data.
Table 1. Scope of the pipe break study

Categories of Data

Based on Ruptures
of Given Pipe Sizes Based on \Iagnitude of Brea~
System (in. in dial (Iea~ rate in gpm)


P\V R primary 1/2 to 2 >2 to 6 >6 ::: 50 (0.2 in. dial ::: 500 H in. dial

BWR recirculation 1/2 to 2 >2 to 6 >6 ::: 500 (1.3 in. dial :::5000(10 in. dial
BWR main steam (reactor vessel to 1/2 to 2 >2 to 6 >6 ::: 500 (4 in. dial ::: 5000 ( lOin. dial
first main steam isolation valve)
BWR main feed (check valves to reactor vessel) 1/2 to 2 >2 to 6 >6 ::: 500 (1.3 in. dial :::5000 (10 in dial


PWR emergency core cooling 112 to 2 >2 to 6 >6 :::1 2: 15

PWR residual heat removal 1/2 to 2 >2 to 6 >6 :::1 2: 15
PWR chemical and volume control 1/2 to 2 >2 to 6 >6 :::1 ::: 15
PWR main steam 1/2 to 2 >2 to 6 >6 2:1 215
PWR auxiliary feedwater I12to2 >2 to 6 >6 :::1 2: 15
PWR main feedwater 1/2 to 2 >2 to 6 >6 :::1 2: 15
PWR condensate 1/2 to 2 >2 to 6 >6 :::1 ::: 15
PWR condensate 112 to 2 >2 to 6 >6 2:1 2: 15
PWR component cooling water 1/2 to 2 >2 to 6 >6 :::1 ::: 15
PWR service water 1/2 to 2 >2 to 6 >6 2:1 2: 15

BWR high pressure coolant injection 1/2 to 2 >2 to 6 >6 2:1 2: 15

BWR reactor core isolation cooling 1/2 to 2 >2 to 6 >6 :::1 ::: 15
BW R core spray 1/2 to 2 >2 to 6 >6 :::1 ::: 15
BWR residual heat removal 1/2 to 2 >2to6 >6 2:1 ::: 15
BWR standby liquid control 1/2 to 2 >2 to 6 >6 2:1 ::: 15
BWR main steam (MSIV to turbine inlet) 1/2 to 2 >2 to 6 >6 :::1 ::: 15
BWR main feedwater (condensate system 1/2 to 2 >2 to 6 >6 :::1 ::: 15
to check valves)
BWR condensate 1/2 to 2 >2 to 6 >6 :::1 ::: 15
BWR reactor building component cooling 1/2 to 2 >2 to 6 >6 :::1 ::: 15
BWR service water 1/2 to 2 >2 to 6 >6 :::1 ::: 15


As stated earlier, the purpose of this study is not LOCA-Initiating Rupture Rates
only to update the overall pipe break failure fre- Pipe (per plant per year)
quencies reported in Table IV, 6-9 of WASH-1400, Rupture
but also provide more specific, conditional failure Size Range
rates that would be useful to risk analysts, such as (in.) (90070 ) Median
pipe break failure rates for a particular system or
failures per pipe hanger. 1/2 to 2 I x 10,4_1 x IO'~ I X 10- 1
Because this study will, in part, update the pipe
2to6 3x 1O-~-3 x 10" 3 X 10,4
break frequencies in WASH-1400, a brief descrip-
tion of the methods used for that report and the
>6 I x \O'~-I X \0,_1 I X 10- 4
results are given below.
The WASH-1400 reported the frequencies of
pipe failure as an initiating event for LOCAs. The As can be seen, the WASH-1400 results suggest
study examined several different data sources to that smaller pipes have a larger overall failure fre-
obtain a frequency of pipe failure for large and quency than medium-sized pipes and medium-
sized pipes have a larger failure frequency than do
small pipe. The reason for using several data
larger-sized pipes.
sources was the interest in pipe ruptures (complete
Pipe break failure rates for non-LOCA-sensitive
pipe severances) resulting in loss of coolant, and
piping are shown in the Reactor Safety Study's
none had occurred in the 150 U.S. commercial
Data Assessment Tabulation, Table III, 2-1, for
NPP operating years considered in the study.
pipes < 3 in. and> 3 in. per pipe section per hour
Therefore, other pipe break data sources were
with a section being 12 ft of pipe.
sought for extrapolating pipe failure rates for use in
The initial approach used in this study was to
analyzing U.S. commercial NPPs.
collect all applicable failure and population data
Several different means of extrapolating the data
from U.S. commercial NPPs, U.S. military NPPs,
were de\ised because the data were given in different
foreign NPPs, and fossil electrical generating and
forms and, in general, thc quality of the data was poor.
steam processing plants. Then, a statistical analysis
Details such as leak rates, pipe size, cause of failure, the would be used on the collected data to determine
system in which the failure occurred, and other perti- the overall pipe break frequency by pipe size, as was
nent information often werc not supplied. As a result, done in WASH-1400, and also estimate the condi-
weighting factors were used to relate total plant piping tional pipe break failure frequencies that would
to LOCA-sensitive piping and to large and small pipe; meet the objectives of this study. This approach
severity factors were uscd to relate non-severance pipe and the type of failure rates to be estimated were
failure to small pipe and complete severance to large modified as the quantity and quality of the data
pipe. Finally, the failure data were placed in two pipe became known and upon the recommendations
size categories; <4 in. and >4 in. dia, with the from the piping experts consulted during this study.
former simply called sIllall and the latter called large. To determine the piping systems applicable to this
Later, for the final assessment, three categories were study, the systems analyzed in probabilistic risk assess-
established: small (1/2 in. to 2 in.), medium (2 in. to ments (PRAs),2-4 final safety analysis reports
6 in.), and large (> 6 in.). The rates obtained for small (FSARs), and the Accident Sequence Evaluation Pro-
and large pipes were retained with engineering judg- gram (ASEP) Catalog of PRA Dominant Action
ment used to obtain the rates for breaks in medium Sequence Information (NUREG/CR-3301)5 were
pipes. reviewed with respect to the criterion that only systems
The pipe break failure frequencies reported in where leakage could either be considered a LOCA-
WASH-1400 (Appendix III, Table 111,6-9) are: initiating event or could inhibit the mitigation of a core

melt sequence. The systems selected were placed into type. For PWRs, the lowest capacity reactor cool-
one of two piping categories: LOCA-sensitive or non- ant makeup system uses a positive-displacement
LOCA-sensitive. For this study, LOCA-sensitive piping charging pump with a typical pumping capacity of
is piping in which a pipe break results in a loss of reac- > 75 gpm. Allowing for small leakages, such as
tor coolant. For pressurized water reactors (PWRs), seal leakage, the pumping capacity is assumed
this is the primary coolant system excluding the steam ~ 50 gpm. Other PWRs use a centrifugal pump for
generators.a For boiling water reactors (BWRs), it con- normal plant makeup that has a pumping capacity
sists of the part of the main feedwater system down- of several hundred gpm. For BWRs, the normal
stream of the isolation check valves, the recirculation reactor coolant makeup is from the reactor core iso-
system, and that part of the main steam system lation cooling system with a typical pumping
upstream of the main steam isolation valves. Non- capacity of 500 gpm. It is assumed that leak rates
LOCA-sensitive piping is associated with piping sys- below these values will be detected by the numerous
tems that would be used to help mitigate a core melt diverse detectors such as temperature, humidity,
sequence but, if a pipe break occurred, no loss of reac- radiation, and makeup monitors with normal
tor coolant would result. For PWRs, the systems are: makeup replenishing the reactor coolant while an
high pressure injection, residual heat removal, chemi- orderly plant shutdown with subsequent repair
cal and volume control, auxiliary feedwater, conden- commences. Above these leak rates, it is assumed
sate, main feed water, component cooling water, normal makeup to the reactor coolant is insuffi-
essential raw cooling water, and main steam. For cient and actuation of the emergency core cooling
BWRs, the systems are: high pressure coolant injec- system (ECCS) would be required.
tion, standby liquid control, reactor core isolation For non-LOCA-sensitive systems, se\eral factors
cooling, core spray, residual heat removal, condensate, had to be considered in the definition of failure.
reactor building component cooling, service water, and One factor considered was whether one could
the portions of main steam and main feed water that determine the leak rate necessary to disable a sys-
are not in the LOCA-sensitive category.b Pipe break tem from performing its intended function. Know-
failure rates, overall and conditional, would then be ing the probability of having such a leak rate is of
developed for PWRs and BWRs for all the systems interest to risk analysts. However, this leak rate
listed above and for the two piping categories, LOCA- value is system and location dependent. For exam-
sensitive and non-LOCA-sensitive. ple, a system may be a closed loop system with only
Within the piping systems mentioned above, the a relatively small head tank to accommodate sys-
components of interest (for this study) are straight tem water volume expansion and contraction due to
lengths of pipe, pipe elbows, couplings, fittings, temperature changes and have only local controls
flanged joints, and welds, which are passive bound- for establishing makeup to the system in case of
ary components in piping systems. As mentioned leakage. Other systems may have a large volume
earlier, components such as pumps and valves, water source such as a condensate storage tank. For
although making up part of the piping system these systems, a particular leak rate on the dis-
boundary, are not considered. Heat exchangers, charge of a pump may be tolerable but the same
including steam generators, are being considered in leak rate near the suction of the pump could be
other studies and are not included here. However, system disabling due to the loss of the net positive
some of these studies are summarized in suction head. Because of these differences, a func-
Appendix E. tion estimating the probability of exceeding any
Failure criteria were then established for the two specified leak rate is needed. The quality of the fail-
piping categories to define the failure data to be ure data, however, was not sufficient to establish
collected. For LOCA-sensitive piping, failure is a such a function. Instead, the data were placed into
leak rate of at least 50 gpm for PWRs and 500 gpm discrete categories (> I gpm and > 15 gpm) from
for BWRs. This rate is based on the normal reactor which the probability of exceeding these two rates
coolant makeup system capacity for each plant could be determined. These two leak rate categories
were selected because the few actual known leak
a. Steam generator tube rupture, have been analyzed in other
rates reported occurred roughly in the I and
U.S. Nudear Regulatory Commi"ion (NRC) studies (see 15 gpm range. Although leak rates were not speci-
Appendi\ E). and are not ~onsidered here. fied for much of the data, a rough assessment of
b. For this slUdy. the main steam system terminates at the high
whether they exceeded 15 gpm could be made. For
pressure turbine stop lal\es. this study, the lower bound cutoff was I gpm.

Another factor to consider was the availability of Furthermore, with nearly 800 reactor years of oper-
the population data necessary to support the ation of commercial U.S. NPP experience to draw
development of the conditional pipe break failure from in obtaining pipe break frequency estimates,
rates. Two large classes of population data were the the need to extrapolate such data into the U.S.
number of welds and length of piping in the sys- commercial NPP data is not compelling.
tems being considered. Numerous utilities and an Once all of the failure and population data were
architectural engineering firm provided this infor- gathered, the statistical analysis began. The pipe
mation in two forms. One was the actual count data break failure frequencies determined for this study
available from their files; the other was in the form are those for PWRs and BWRs in general-
of system isometric drawings depicting the welds frequencies of pipe breaks categorized by system,
and feet of pipe from which the data could be by operational mode, by pipe size (for non-LOCA-
counted. Howev'er, data on <2-in. dia piping was sensitive systems only), and by leak rate. These are
unavailable from either source, primarily because shown in the "Results" section, where the failure
of the enormous amount of instrumentation piping rates for LOCA-sensitive piping are given in terms
in this category. Therefore, pipe break failures in of a leak rate instead of pipe size. The reason is the
piping <2 in. ID. pipes generally were not direct correspondence between a given leak rate and
included in this study. the necessary emergency core cooling response.
A final consideration was that leak rates of This relationship for a particular pipe size rupture
50 gpm and higher were considered significant is dependent on many parameters; system pressure
enough from the viewpoint of system operability being a major one. Figure I (which is for illustra-
t hat all failure events in this category would be col- tive purposes only and not meant to represent
lected regardless of pipe size. As a result, pipe actual values) provides a reference for considering
break failure data were not collected for pipes the relation between the pipe size in which a leakage
<2 in. unless the leak vvas ~50 gpm. may occur and the impact of such a leakage on the
In summary, for this study, the definition of fail- plant. The figure applies to PWRs; a similar figure
ure for non-LOCA-sensitive systems is a leak rate of would apply to BWRs. The curved line in the figure
at least I gpm in piping that has, at a minimum, a represents the hypothetical expected maximum sus-
2 in. dia unless the leak rate was ~50 gpm, for tained leak rate possible given the operating pres-
which all such data were collected regardless of the sures and temperatures expected in a typical
pipe size. primary system as a function of the pipe size. It
As mentioned earlier, an attempt was made to shows that in the worst case of a double-ended
obtain other pipe break information than just from break, some maximum leak rate exists, and that
U.S. commercial NPPs because we felt the latter rate would be expected to increase with the diame-
data would be sparse; which was true. However, ter of the pipe. Pipe break events may be plotted on
very little data were obtained and what were this figure according to the pipe size and the corres-
obtained are not available for release and thus was ponding actual leak rate. The shaded region, "A",
not usable in this program. If potentially useful in the figure describes events occurring in small pip-
data had been obtained, several concerns regarding ing that contribute to small LOCAs. Region "B" is
the data would have to be addressed. Some of the a similar region for large piping and large LOCAs.
concerns inhibiting the usefulness of such data are: The pipe size and plant impact ideas differ, how-
ever, for events such as those in Region "C". This
• The lack of knowledge of the completeness region represents a leak rate > 50 gpm but
of reporting the failure events < 500 gpm in large piping. Such an event should
• Lack of supporting population data contribute to the small LOCA probability esti-
(welds, feet of piping, etc.) mates, but it is in large piping and also contributes
• The amount of resources available to to estimating the probability of ~ 50 gpm leaks in
determine how to use the data, given such large piping. Because smaller leaks can occur in
differences as diversities in operating con- large piping, the plant impact from pipe break
ditions, fluid composition, pipe quality, events does not necessarily correspond to the pipe
and test and installation procedures. size.


L t
- Ii
~ ii A I C
0 j.
..J %


2 4 6 8
'-- small medium large

Pipe Diameter (inches)

Figure I. Graphic example 10 illustrate the range of PWR leak rates for a given LOCA pipe break size.

Piping failures from many diverse sources were Requests for piping failure data for systems simi-
sought for this study. These sources included com- lar to those found in NPPs were made to several
mercial U.S. and foreign NPPs, military NPPs, fossil electrical generating and steam generating
and fossil electrical generating and process steam facilities. In all cases, the response was that kind of
generating plants. information is, in general, not recorded and not
For commercial U.S. NPPs, several sources of available.
data I\'ere available: As stated, 19 qualifying piping failures were
found. Nine failures occurred in PWRs, none of
• AECL-MISC 204, A Study of Piping Fail- which were in the primary system. Ten failures were
ures in U.S. Nuclear Power Reactors, by found in BWRs, with none occurring in LOCA-
P. Jansen, April 1981 sensitive piping. A summary of the 19 events is
• EGG-EA-5989, Loss of Coolant (LDCA) contained in Appendix D. In many event descrip-
Data Base Study, by D. E. Baxter et a!., tions, the total leakage was given but no time was
August 1982 associated with the leakage nor was a leak rate
• EPRI-NP-LD, Hater Hammer in BWRs, given. For such cases, an assessment of whether the
Final Report, September 1982 leak rate was at least I gpm was made. If so, the
• IAEA-S!\I-218/ 11, Reliability of Piping in event was further classified as to whether the leak
Light Hater Reactors, by S. H. Bush rate was likely 10 be > 15 gpm. If the leak rate for
• In-Plant Reliability Data System (IPRDS) an event could reasonably be ~ 1 gpm or
~ 15 gpm, it was classified as such. These failures
(Sort on pump and valve cause codes
are listed in Tables 2 (PWRs) and 3 (B\\'Rs) and
related to pipe leakage; transmitted
June 15, 1984) are also contained in Appendix D in a narrative
format, summarized from their individual sources.
• Licensee Event Reports (LERsl,
Of the nine PWR events, five had leak rates esti-
January 1976 to April 1984
mated as ~ 1 gpm and < 15 gpm and four had leak
• Nuclear Plant Reliability Data System
rates ~ 15 gpm. Six of the failures were caused by
(NPRDS) (Sort on pipe failures,
vibration; the remaining three were a result of water
April 2, 1984)
hammer. Five failures occurred in the chemical and 101-
• Nuclear Power Experience 1975 10 1984 ume control system, all in >2 to 6 in. pipes. Of the
• Nuclear Safety (Volume 10, 196910 Vol- other four failures, two were in a residual heat removal
ume 25, 1984). system; two were in main feedwater systems.
Ten piping failures occurred in BWRs. The events
The most useful I\'ere Licensee Event Reports (LERs) occurred in high pressure coolant injection, con-
and Nuclear Power Experience (NPE) reports. HO\\'- densate, main feedwater, and service water systems.
eler, el en these reports lacked information that would Nine failures resulted in ~ 15 gpm leak rates. Five
hale been quite useful for this study. For example, of failures occurred in > 6 in. pipes, two occurred in
the 19 reported piping failure events, only 5 event >2 to 6 in. pipes, and three failures were reported
descriptions gale actual leak rates or sufficient infor- for> 1/2 to 2 in. pipes. The causes of the failures
mation 10 calculate the leak rate. For those piping fail- varied from corrosion, operator error, vibration,
ures with unknown leak rates, an assignment into one and mechanical failure to pump seizure and a for-
of two leak rate categories was made: ~ 1 gpm and eign object piercing a pipe.
~ 15 gpm. Other relevant data such as the pipe size in Of these 19 failures, 4 occurred while plants
which the el'ent occurred were also sometimes difficult were in shutdown, 2 occurred during startup proce-
10 assess, dures, and 13 occurred during normal operation.
Numerous contacts with foreign personnel were Table 4 categorizes these events by the type of
made in an attempt 10 collect foreign pipe failure data, plant, pipe size, leak rate, operational mode, and
but this effort met with little success. The information failure cause.
that lIas receil'ed contained little applicable data or was The failure data are very sparse considering
not alailable for public release. (a) the amount of piping in each of the 81 commer-
\Iilitary NPP pipe failure data was unavailable cial nuclear power plants, and (b) by the cutoff
due to being classified. date of the data collection, December 1984,

approximately 800 operating years had transpired piping, the zeros for data points are still significant
for the U.S. commercial nuclear industry. Again, because of the large amount of operating experi-
however, even though the data are sparse, with no ence that now exists.
reported leak rates ~ 50 gpm in LOCA-sensitive

Table 2. PWR pipe break events

Rate Pipe Size Plant
Sy,tem Plant Date (gpm) (in. in dial Cause Condition Data Source

Chemical and Indian Point 2 11178 8 >2. s6 Vibration Normal LER 247178-032
volume control

Chemical and Indian Point 2 03177 ~I >2. s6 Vibration Normal LER 247177-002-58
volume control

Chemical and Salem 2 06/81 ~15 > 1/2. s2 Vibration Startup NPE (PWR-2 VIIl-A-538)
volume control

Chemical and Salem 2 07/84 ~I >2. s6 Vibration Normal LER 311/84-016

volume control

Chemical and Trojan 09/77 II >2. s6 Vibration Normal NPE (PWR-2 VIII-A-308)
volume control LER 344/77-038

Main feedwater Indian Point 2 11173 ~15 >6 Water hammer Startup NPE (PWR-2 VI-E-34)

Main feedwaler Maine Yankee 01/83 ~ 15 >6 Water hammer l"ormal NPE (PWR-2 VI-E-435)
LER 309/83-003

Residual heat Arkansas 08176 ~2 >6 Vibration Shutdown NPE (PWR-2 VIll-8-96)
removal Nuclear One-l

Residual heat McGuire 2 08/84 ~ 15 > 1/2. s2 Water hammer Shutdown LER 370/84-017

Table 3. BWR pipe break events

Rate Pipe Size Plant
S~'tem Plant Date (gpm) (in. in dial Cause Condition Data Source

High pre;;ure Dre,den 2 09/76 >6 Corrosion Normal NPE (BWR-2 VII-E-93)
(oolant injc(i ion

High pr\.?""ure Dre,dcn I 05171 2: 15 >6 Operator error Normal NPE (BWR-2 VII-E-9)
coolant i njc":l ion

Conden,atc Brow m Ferr) 04/78 2: 15 >6 Vibration Normal NPE (BWR-2 VII-E-147)

Condelhatc Oystcr Creek 06/82 2: 130 >1/2, os2 Mechanical fatigue Normal NPE (BWR-2 VI-E-157)

Conden,ulc Dre,den 2 0974 2: 15 >1/2, os2 Vibration Normal NPE (BWR-2 VI-E-33)

Conden,ate Dresden I 01/64 2: 15 > 2, os 6 Pump seizure Normal AECL (Appendix B)

'lain feedwater Quad Cities 2 08/75 2: 15 >6 Vibration Normal NPE (BWR-2 VI-E-45)

'lain kedwaler Quad Cit ie, 2 08/75 2: 15 > 1/2, os2 Vibration Shutdown NPE (BWR-2 VI-E-45)

'lain kedwaler Quad Cit ie, 2 11/74 2: 15 > 2, os 6 Unknown Normal Nuclear Safety
(No.2. Vol. 16, 1975)

Re,idual heat Quad Cities I 04174 50 >6 Pipe pierced by Shutdown NPE (BWR-2 VIII-C-26)
remmal construction

Table 4. Failure data categorized by conditional factors

PWRs BWRs All Plant Types

Conditional Factor Frequency 010 Frequency 010 Frequency 010

Pipe Size (dia)

1/2to 2 in. 2 0.22 3 0.30 5 0.26

>2to 6 in. 4 0.44 2 0.20 6 0.32
>6 in. 3 0.33 5 0.50 8 0.42

Leak Rate

~1 gpm, <15 gpm 5 0.56 1 0.10 6 0.32

~ 15 gpm 4 0.44 9 0.90 13 0.68

Operational Mode

Starting up 2 0.22 0 0 2 0.11

Normal operation 5 0.56 8 0.80 13 0.68
Shutting down 0 0 0 0 0 0
Shutdown 2 0.22 2 0.20 4 0.21
Transient 0 0 0 0 0 0

Failure Cause

Vibration 6 0.67 4 0.40 01 0.52

Water hammer 3 0.33 0 0 3 0.18
Mechanical fatigue 0 0 1 0.10 1 0.05
Corrosion 0 0 I 0.10 1 0.05
Operator error 0 0 1 0.10 1 0.05
External damage 0 0 1 0.10 1 0.05
Other 0 0 I 0.10 1 0.05
Unknown 0 0 1 0.10 1 0.05

TOTAL 9 10 19


Pipe break failure rates are generally given in The time period used in the failure rate is the
terms of failures per year or failures per hour. The number of years of commercial power generation
denominator of these terms, year and hour, is and was obtained from the licensed operating reac-
referred to here as population data. Other popula- tor status summary reports, NUREG-0020, com-
tion data were collected for this study in order to monly referred to as the Gray Books. A list of
estimate conditional pipe break failure rates such as plants, date of their first electrical generation, and
failures per hour per system. years of operating experience are shown in Table 5.
The population data gathered for this report are: Over all the plants, 798.09 total years of light water
U.S. commercial NPP generation time, and the reactor (LWR) operation had transpired through
amount of it spent in various plant operational December 1984. This breaks down to 313.36 for
modes; the number of plant transients; and the General Electric plants (BWRs), 81.55 for Babcock
number of welds and feet of pipe for each of the and Wilcox (PWRs), 320.73 for Westinghouse
piping system being considered. (PWRs), and 82.45 for Combustion Engineering

Table 5. Years of operating experience and number of transients for each plant

Years of
Date of First Operating Number of
Plant Electrical Generation Experience Transients


Arkansas Nuclear One-I 08/01174 10.42 58

Crystal River 3 01/30/77 7.92 66
Davis-Besse I 08/28177 7.35 67
Oconee I 05/06173 11.66 84
Oconee 2 12/05173 11.08 50
Oconee 3 09/01174 10.34 46
Rancho Seco I 10/13174 10.22 48
Three Mile Island Ia 06/19174 10.54 6
Three Mile Island 2b 04121178 2.02 2

81.55 427


Arkansas Nuclear One-2 12/26178 6.02 77

Calvert Cliffs 1 01/03175 10.00 78
Calvert Cliffs 2 12/07176 8.07 53
Fort Calhoun 1 08125173 11.36 35
Maine Yankee 11/08/72 12.15 57
Millstone 2 11/09175 8.06 82
Palisades 12/31171 13.01 108
San Onofre 2 09120/82 2.29 37
San Onofre 3 09125/83 1.27 20
St. Lucie 1 05/07176 8.66 57
St. Lucie 2 06/13/83 1.56 14

82.45 618

Table 5. (continued)

Years of
Date of First Operating Number of
Plant Electrical Generation Experience Transients


Beaver Valley I 06/14176 8.55 118

D. C. Cook I 02/10175 9.89 55
D. C. Cook 2 03/22178 6.78 51
1. M. Farley I 08/18177 7.38 84
J. M. Farley 2 OS/25/81 3.61 28
R. E. Ginna 12/02/69 15.09 41
Haddam Neck 08/07/67 17.41 91
Indian Point 1c 09/16/62 12.12 266
Indian Point 2 06/26/73 11.52 216
Indian Point 3 04127176 8.68 75
Kewaunee 04/08174 10.74 84
McGuire 1 06/30/81 3.51 41
McGuire 2 05123/83 1.62 18
North Anna I 04/17178 6.71 49
North Anna 2 08/25/80 4.36 35
Point Beach 1 11/06170 14.16 56
Point Beach 2 08/02/72 12.42 43
Prairie Island 1 12/04173 11.08 61
Prairie Island 2 12/21/74 10.03 59
H. B. Robinson 2 09126170 14.27 195
Salem I 12/25/76 8.02 85
Salem 2 06/03/81 4.08 38
San Onofre I 07/16176 17.47 51
Sequoyah 1 07122/80 4.45 37
Sequoyah 2 12123/81 3.03 17
Summer 1 11/16/82 2.13 41
Surry 1 07/04/72 12.50 144
Surry 2 03/10173 11.82 112
Trojan 12123175 9.03 83
Turkey Point 3 11/02172 12.17 125
Turkey Point 4 06121173 11.53 113
Yankee-Rowe 1 11/10/60 24.15 98
Zion 1 06128173 11.52 129
Zion 2 12126173 11.02 160

320.73 2899

TOTAL PWRs 484.73 3944


Big Rock Point I 12/08/62 22.98 15 d

Browns Ferry I 10/15173 11.22 131
Browns Ferry 2 08/28174 10.35 120
Browns Ferry 3 09/12176 8.31 81
Brunswick 1 12/04176 8.08 81
Brunswick 2 04/29175 9.68 129
Cooper Station 05/10174 10.65 66

Table 5. (continued)

Years of
Date of First Operating Number of
Plant Electrical Generation Experience Transients


Dresden Ie 04/15/60 19.63 I7 f

Dresden 2 04/13170 14.72 105
Dresden 3 07/22171 13.45 90
Duane Arnold 05/19174 10.63 58
Fitzpatrick 02/01174 9.92 59
Edwin I. Hatch I 11/11174 10.14 130
Edwin I. Hatch 2 09122/78 6.28 57
Humboldt Bayg 04/18/63 13.21 43
LaSalle I 09/04/82 2.33 46
Millstone I 11129170 14.09 86
Monticello 03/05/71 13.83 60
Nine Mile Point 11/09/69 15.15 81
Oyster Creek I 09123/69 15.28 59
Peach Bottom 2 02/18/74 10.87 58
Peach Bottom 3 09/01174 10.34 56
Pilgram I 07/19/72 12.46 113
Quad-Cities I 04/12/72 12.73 86
Quad-Cities 2 OS/23/72 12.61 73
Susquehanna I 11/16/82 2.13 15
Vermont Yankee I 09/20172 12.29 55

TOTAL BWRs 313.36 1970

a. Shut dow n indefinitely 3/29/79.

n. Shut down indefinitely 3/28/79.
c. Shut down indefinitely 10/31/74.
d. First 10 years of transient data unavailable.
e. Shut down indefinitely 10131/78.
f. First 12 years of transient data unavailable.
g. Shut down indefinitely 07/02/76.

plants (PWRs). A total of 484.73 years of commer- heat removal, standby liquid control, main steam,
cial operation pertain to PWRs. main feedwater, condensate, reactor building com-
The particular systems for which weld and pipe ponent cooling water and raw cooling water.
length data were obtained are as follows. For Weld count and pipe length data were available from
PWRs, the systems are primary reactor coolant, 18 different nuclear power plants. The data were sup-
high pressure injection, low pressure injection, plied in tabular form or in the form of isometric draw-
residual heat removal, chemical volume and con- ings from which the length and weld counts were
trol, main steam, auxiliary feed water, main feed- obtained. Thirteen of the plants are PWRs; numerical
water, condensate, component cooling water, and data were supplied on four plants and isometric draw-
essential raw cooling water. For BWRs, the systems ing were supplied for nine plants. For BWRs, isomet-
are recirculation, high pressure coolant injection, rics were obtained on four plants and numerical data
reactor core isolation cooling, core spray, residual were obtained on two plants. For the PWRs, four are

four-loop plants; one is a three-loop plant; and dght depressurized, and starting up the
are two-loop plants. Tables 6-11 show the pipe length plant from this condition.
and weld population data obtained. The information
contained in the tables shows the pipe length and weld In general, all systems, except the ECCS, are
population for three pipe size categories (2 in., in operation sometime during the startup
>2 in. to 6 in., >6 in.) for each of the systems of the when the startup is commenced from a
NSSS vendors considered in this study: General Elec- cooled-down condition. The ECCS is placed
tric for BWRs; Babcock and Wilcox, Combustion in standby. During the startup mode, the pri-
Engineering, and Westinghouse for PWRs. The tables mary system is taken to normal operating
also show the total pipe length and weld population for temperature and pressure. Criticality is
PWRs and BWRs for each of the systems for each of achieved and the secondary systems (main
the NSSS vendors. feed and condensate) are placed in operation.
As stated above, the information was supplied in Thermal power is increased such that the
different forms. Some of the data were given for nuclear plant will supply electrical power. The
each of the systems as a total count and could not systems in operation during this time are
be broken down into the pipe size categories. As a essentially the same as those during normal
result, the tables are read in the following manner. operation, except the systems are in transition
Using Table 6 as an example (pipe length popula- from a shutdown condition to a normal oper-
tion by system and vendor for PWRs), the left- ation condition (systems are undergoing both
thermal and pressure transitions).
hand column lists the systems being considered
here. The second column lists the vendor. For each
of the third, fourth, and fifth columns, three sub-
• Normal Operation-During the normal
columns show, respectively, the number of plant
operation mode, the nuclear plant is on-
units for which the information could be broken
line generating electrical power. The pri-
down into the particular pipe size, the mean or
mary coolant system is at normal
average number of counts (ft), and its standard
operating temperature, pressure, and flow.
deviation. The last column shows the total number
During this mode, the emergency systems
of plant units for which the information was sup-
are in standby and will operate on demand.
plied for that vendor, the mean count for total sys-
• Shutting Down-The shutting down opera-
tem pipe length (which may include more tion is a condition in which the nuclear plant
information than was available for the previous is taken off-line, the reactor is shut down, the
columns), and its standard deviation. As can be primary coolant system is being cooled down
seen, the count spread (feet of piping) was quite and depressurized, and most support systems
·Iarge for some systems. The information is pro- are shut down. The normal cooldown
vided for use by the reader but failure rates on a per method uses the main feedwater system or the
weld basis, or per foot of pipe, were not computed auxiliary feedwater system (PWR) to cool
in this study. This is explained further in the down the primary system to the point where
"Results" section of this report. the low pressure residual heat removal system
The plant operational modes used in this study are: can complete the cooldown. As systems are
starting up, normal operation, shutting down, shut- no longer needed during the cooldown, they
down, and transients. The hot standby and hot shut- are shut down.
down modes are contained within the starting up and • Shutdown-In this mode, the nuclear
shutting down modes because of the relatively brief plant is off-line, the nuclear reactor is sput
amount of time spent in them. The operational modes down, and the primary coolant system is
used in this study are explained below. cooled down and depressurized. Support
systems may still be in operation (e.g.,
• Starting Up-The starting up mode for decay heat removal). However, these sys-
this study encompasses two areas: tems are shut down and depressurized as
Start up of the nuclear plant from they are no longer needed. During this
shutdown/cooldown conditions time, the plant is typically involved in refu-
Maintaining the plant for brief peri- eling and maintenance operations.
ods in a shutdown condition in which • Transients-The last area of nuclear plant
the plant is not cooled down and operation of concern is transients, which

Table 6. Pipe length population by system and vendor for PWRs


1--------0---- -I' 1 1 1 I I I I I I I I
IPRI"~RY 'atll 1 01 , I 01 I 1 01 I I 01 I I
I, I---------+----+---.---+--+--.---~-..----+_---.--+__-..___I

I 1----------.--_--_-_--.---_--_--__-_-_1

I 11 401

1 ZI Z981 69.31

ZI 3011 Z94.91
1 II
II 3601

I 11 5091

51 Z0991 681.41

I - - - - - - - - ...- + - - - - . . . . - - - + - - . - - - + - - - + - - - - + - - - . . . . . - - . - - - _ 1
IHIGH PRESS '8'11 1 0' I I 11 11341 '01 I I 01 I 1
Il/IIJfCT 1----------.--------.---.----------------1
,I 'CE I ZI &lSI 621.91
'llEST 1 ZI 5671 270.81 ZI 14071 614.51
ZI 16131 944.71
ZI 30Sl 420.01
ZI 19ZJlun.1I
61 303412339.01
ZI "36112699.71

lUll I 01 1
'IIEST I II 3281 178.ZI
I 11 156' I II U71
ZI 348' ZI.ZI
I 01 I 1
21 170Z' 25.51 61 18231 905.5'
I 01 I 1 . , Z3Z91 I 01 I I 01 I
_ I1
ICE I 01 I 01 I
I 01
I 01
1 - - - - - - - + - - - + - - - + - - - - + - - -..... -+----.
, I I


. . -.---_--+---.---+---_-_-_--.--_-_1
ZI Z80l1 n.6'
11 1133'
ZI 991
ZI JZ061 35Z.1I
61 5ZUI5073.71
8'l1 ' Z I U70In09.5I ZI 328414639.31

IAUX fEEn '8''' I 01 I I
11 7411 I 01 I
I - - - - - - - - + - - - . - - - . - - - . - - + - - - . - - -.. -~--..
ICE I 11 H' I 11 9D61 1 01 1




'WF.ST 1
l' HI I zi 5421 Z16.41
ZI 1511 160.5' ., 305813406.81
IMAIN FfF.D IR'II I 01 1 ' 1 1 334' 1 11 1"'761
'01 1 I

,, ,---------....--·----+------·---+---·---+---..---.---.--.....
ICE I 01 I I 01 I I 01 I I 01
, I
'01 1 I ZI 310' 343.71
11 6771 1 . 1 219412"'55.11
ICONOENSATE 'UII '01 I I 11 1760' I
11 Z6261 I 01 1 I
II 1-------+---+----+-----+--.--.---+--+----..--.--.......-.--1
IllEST 1 01 I '01 I 1 01 , 1 31 406312761.31
ICOMPONFNT IR'II 1 01 I I 11 30871 I 11 21"'1' I 01 , 1
I CDOli NG ,----------+--+---+----.----.---.------_--_-+--_-_1
I IllEST I 11 1811 1 ZI 9831 717.71 ZI 315ZI3683031 51 UZ8IZ7Z9.ZI


I - - - - - - - - - - + - - - - - - - - - ... --.---~--...
I 0' I I 11 unl I 11 4Z081 I 01 I
ICE '01 1 I 11 9881 I 11 )8UI I 01 1 I
IllEST I 11 1147' 1 ZI 23651 979.31 ZI 289011...5."" 41 ""01lZ161.91

are defined as plant conditions or plant Table 12 shows the general plant conditions (tem-
upsets that result in a reactor trip. These perature and thermal power) that determine what
events cause both the secondary systems mode the nuclear plant is in at any given time. The
and the primary system to undergo moder- last column also contains the population results in
ate (within design conditions) temperature terms of what portion of time the plant on the aver-
and pressure transients. age spends in each mode.

Table 7. Pipe length population by system for PWRs

, S v S IE" I I I I I I I I I 1 I I I
,----------------------------- I I I I I I I I I I I I I
1PRI MARY I 31 2121 156.81 31 3381 218.21 21 2091 214.31 61 18341 890.41
'HIGH PRESS INJECT I 41 6961 422.11 51 14351 597.01 41 111311164.91 81 336612307.91
'RHR I 21 3l81118.21 31 2841111.91 31 14HI 499.71 6118231905.51
ICVCS I 21 28021 63.61 31 29HI 564.21 21 991 12.81 61 522215013.71
IMAIN STEAM I 11 11331 I 11 8921 I 21 221013209.51 21 328414639.31
0\ IAUX feED I 21 251 0.71 41 6831 215.81 21 1511 160.51 41 305813406.81
'MAIN fEED I 01 I I 31 3181 243.41 21 10171 565.01 41 219412455.81
ICONDENSATE I 01 1 I 11 17601 1 11 26261 I 31 406312761.31
ICOMPCNENT COOLING I 11 1811 I 31 168~11316.81 31 281512669.01 51 442812729.21
ISERVICE WATER I 11 11471 I 41 19131 860.81 41345611163.11 41480112861.91
Table 8. Pipe length population by system for BWRs


I I N IMEAN I sro I N IMEAN' sro • N 'MEAN' SJO • N 'MUN' STO •
ISYSTEM , I I I • • I I I • I • I
,------------ - I • • • I I I I I • I • I
IREelRe I 01 I I 11 201 I 11 1131 I 21 416148.11
IHPCI I 01 I I 11 .usl I 11 25081 I 21 1011 151.31
IRCle • 01 I I 38S1 •
l' 11 236' • 21 3911 234.1'
leORE SPRAY I 01 I I 11 181 I 11 5181 I 2' 5531 15.0'
1------------------ +---+----+---+--+--+--+----+--------+--+---+----1
IRHR I 01 • • 11 15861 I 11 13111 I 21 16851 445.51
........ I-----------------+---+---+--+----+_--+---+_-+_----+_---+--+_-+---I
ISlCS I 01 • I 11 421 I 01 I • o. . .
ICONDENSATE I 01 • I 11 9291 I II 1023' • O. • I
'RBecws I I
o. I 11 24111 I 8511 I' I 01 I •
1-- ---- -+---+---+---+--+----+--+----+----+----+----+---+----.
IMAIN STEAM I 11 26031 • 11 18131 I 11 35011 I 21 4611'4615.4'
IFEEOWATER • O. I • 11311 I'I 11 1531 I I' 4251 •
Table 9. Weld population by system and vendor for PWRs

I 1 2 IN. PIPES 1>2 TO 6 IN. PIPl:S1 >6 IN. PIPES I TOTA\. I

1 1--------+-------_-----_------1
1------------------..--.---..---.--.----.---...-.-.--.....-.-.--..-- .-.---1
1--------.--------1 I I I I I I I I
IP~IIIARY 111'11 I 11 lZl I 11 2171 I 11 HI I 11 2561 I
1 1-------....--.--_.--.---.---+--...---..-.---.--..--+---1
I-------..---.---+---+--.__ -+
ICE I 21 1071 119.'1 21 931 89.81
21 1591 1'15.91 21
3581 ZZ5.61
I IllEST I 11 2651 I 1I Z-.ll I 11 561 I 51 7321 156.01
'HIGH P~ESS lUll I 11 3351 I ZI 9501 90Z.31 01 I I 01 I 1
I INJECT 1 - - - - - - - _ - - + -_ _- _ - _ - - _ - + - - - + _ _ - - _ - - - _ _- - . - - - 1
, ICE I 31 -.121 189.21 31 -.911 UO.51 31 6Z1I Z53.-'1 31 15261 5Z6.91
, J ... --.----....---_+---+--+--+---+---+--..--.----.
1 IllEST I 31 Z361 "O.ZI 31 2891 U7.71 31 3331 212.-'1 71 10~01 899.11
I~H~ 18'11 I 1I 1361 I 21 1821 168.31 21 -.ZOI 179.61 11 9... 1 I
I 1--------+---.--.---+---..----...----.---....-.-.----..--+--+-----1
I IllEST I ZI 1021 118.81 21 1271 33.21 ZI "291 11.31 61 ~9Z1 Z36.61
ICVCS IUW I 01 I I 11 11061 1 01 I I 01 I I
I 1-------..----.-.-.----.---.---+---....--.---.----+-+--+----1
I ICE 1 11 7801 I 11 7061 1 a1 I I aI I I
I 1-------...--..---..--.---+--+---+---....--+----.--..---+----,
I IllEST I 31 7521 65Z.21 31 7391 556.61 21 291 -'.91 61 HJ7IZ036.01
I"AIN STEAM IWEST I 11 7381 I 11 2,.11 I 11 7531 I 21 279111-.97.71
IAUX FEEO IUW I 01 I I 21 29Z1 2.81 11 Z61 I or I I
1 .- ... --...---+----+---....-.--+---+----+---.-.--..--+--+----1
I ICE I 21 951 109.61 21 2991 73.51 11 991 I 11 6221 I
I 1 - - - - - - -...--+---...---+-+---.---.......--.-.--...---+--+--+-----1
I IllEST I 11 HI I 21 3251 11>-..81 21 HI 17.01 -., 8281 811.31
'".lIN FF.EIllUll I 01 1 I 11 1621 I 11 3061 I 01 1 I
I 1 - - - - - - + - - _ - _ - -_ _- - + - - - _ - - + _ _ - _ - - - _ - _ - + - - - 1
I ICE I 11 581 I 11 UI I 11 1851 I 11 Z891 I
I I ....--.--..-..--+--+---+---_-+--.....-.-....-.-...--.----.
I IWEST I 01 I I 11 701 I 31 lHI 58.-'1 -'1 6551 U3.01
1---------+---------+---+----·--....---·--+---+----....- - - + - - + - - - 1
ICCI<CENSATE IUw 1 01 I I 11 -.361 I 11 u81 I 01 I I
1--------+--.---.---.---.---....- -....-+----+---_-.--+----.
I 01 1 I 01 I 1 01 I I 31 8361 585.81
IClJ" PIlHENT lUll I 01 I I 11 13HI I 11
-'121 I 01 I I
ICO::L1Mi 1--------+---...--.-..---..--+---+---.--.....---..----.--..... -....---1
I IWEST I 11 991 I 21 3891 350.01 21 5181 620.11 51 195311Z15.61
ISEkVIl,;E: ."TER lUll I 01 I I 11 ~8l1 I 11 8361 I 01 I 1
1 ~....·------·t----,-----t------r---,_--t---r--,...--'------I- ----1----- 1- - ----I
I I.EST I 11 H51 I 21 7501 317.51 21 -.921 308.31 -.1 11"1 89Z.01

Table 10. Weld population by system for PWRs


1------------+---------+---------- ---I
I SVS TEM I I I , , I , , , , , , I
1-----------------------------1 1 I I , , , , , , , , ,
IPRIMARY I 4' 123' 125.61 4' 161' 94.81 4' 100' 132.3' 8' 519' 258.41
IHIGH PRESS INJECT I 71 3251 142.31 8' 5311 4U.31 6' 411' 261.8' 101 1185' 809.8'
I 3' 1131 86.JI '" 1541 104.1J 4' 425' lCK.OI 11 5621 285.11
ICvCS , ,,' 159' 532.71 5' 806' 428.1' 2' 29' 4.9' 6' 2431i2036.0'
IMuN STEAM I 11 138' ' 1 1 2"11 ' 1 1 153' ' 2 ' 219111491.1'
\0 1---------------------------+----+----+-----+----+-----+------+----1 • -+-+___--+--1
'AUX FEED I 31 771 83.01 61 305' 82.1' 4' 48' 35.1' 5' 181' 108.6'
IMAIN FEED 1 11 581 '3' 931 61.2' 5' 1131
- -+----+------,
89.2' 5' '82' 139.1'
Ir.ONDENSATE I 0' I 1 11 4361 '11 U8' ' 3 ' 836' 585.81
ICOMPllNENT COOLING I 11 99' ' 3 1 710' 609.9' 3' 522' 448.8' 5' 195311215.6'
'SfRlIICE WATER , II 145' ' 4 1 586' 268.31 4' 685' 287.0' 4'. 1156' 892.01
Table 11. Weld population by system for BWRs


ISYSTff4 1 I I I I I I I I I I I I
1-------------------------------1 I I I I I I I I , I , ,
IRECIRe 1 0' I I 0' I I 11 521 1 21 11~1 15.61
IHPCI I 01 I I 11 1581 1 11 ~111 I 21 1821 1.~1
IRCIC I 01 I 1 11 IH' 1 11 721 ' 2 1 1161 31.8J
ICORE SPRAY I 01 1 1 11 521 I 11 2051 I 21 1711 22.61
IRHR I 01 I 1 114961 I 11472' I 21 3921 76.~1
o 1-------------------------------+----+-----+------+----+-----+------+----+-----+------+----+-----+------1
1SlC S I 0I I I 11 391 I 01 I I 0I 1 I
ICONOENSATE '01 I 1 11 2591 I 11 ~331 I o. • I
IRBCCWS 1 0' 1 I 1'1671 I 11 SIS' 1 01 • ,
IMAIN STEAM I 11 1)701 I 11 3441 I I' 3~21 I 21 8851 9~8.91
IHEOWATER '01 1 I 11 104' I 11 2281 • 11 791 I
Table 12. Operational mode criteria a

Average Percent
Temperature of Year
Mode Thermal Power (Primary/Recirc) Spent in Mode b

Starting upc S 5010 >200°F 2.5

Normal operation >5010 ~350°F 62.5

Shutting down 15010 > 200°F 0.8

Shutdown 0 s200°F 33.6

Transients 8 transients/y

a. These are reasonable guidelines for both plant types but are not exact definitions of reactor modes.

b. Percentages are based on the actual operating experience of U.S. commercial PWR and BWR plants in 1983.

c. Starting up for the purposes of this study continues until the reactor is at normal operating temperature and pressure.


The pipe break failure frequencies presented in this estimates. Table 14 shows the frequencies of pipe
section are those based on actual observed failure data. breaks categorized by pipe size. This is applicable for
The only pipe breaks observed occurred in non-WCA- non-LOCA-sensitive systems only, because the defini-
sensitive systems. Frequencies obtained by combining tion of failure is different for LOCA-sensitive systems
the subjective and the experience data are described in than for non-LOCA-sensitive systems. The same pipe
Appendix A. size categories used in WASH-I400 are used here for
The following tables show the pipe break failure fre- convenience. Table 15 shows the frequencies of pipe
quencies categorized by: plant type, Table 13; pipe breaks categorized by leak rate for both the LOCA-
size, Table 14; leak rate, Table IS; system, Table 16; and non-LOCA-sensitive systems. The LOCA-sensitive
and operational mode, Table 17. The rates reflect the leak rate categories used here, although containing no
number of pipe failures per year. Rates for selected failure events, are suggested categories for any possible
operational modes are per year spent in that mode of
future pipe failure events. Table 16 shows the frequen-
operation; rates for transients are per transient. Meth-
cies of pipe breaks categorized by system. In this table,
ods for the point estimate and bounds calculations are
the failures are associated with the same respective sys-
explained in Appendix F.
tem and number of reactor years of operation as
Table 13 shows the frequencies of pipe breaks cate-
assigned in Table 13. In Table 17, the frequencies of
gorized by plant type (PWR and BWR). There were
pipe breaks are categorized by operational mode, and
nine failures found for PWRs and ten failures found
also separated by reactor type, with the number of fail-
for BWRs. The sum of the number of years since first
ures being associated with the mode of operation in
electrical power generation for the plant units (exclud-
ing plants that were shut down indefinitely) is also which they occurred.
shown for the two reactor types: 484.73 for PWRs; In the initial stages of this study, developing pipe
313.36 for BWRs. The difference in reactor years break failure rates on a per weld and per foot of pipe
accounts for the different failure frequency basis was planned for the piping systems selected for

Table 13. Frequencies of pipe breaks categorized by plant typea

Lower Bound Estimate Upper BOllnd
Plant Numerator Denominator
Type (N failures) (T years) x~().();. cs/ 2T (NIT) \~II')< 2'. . .:;/2T

PWR 9 484.73 0.0097 0.0186 0.0324

BWR 10 313.36 0.0173 0.0319 0.0541

LWR 19 798.09 0.0156 0.0238 0.0350

a. Note that no failures were observed in LOCA-,en,iti,e piping.

Table 14. Frequencies of pipe breaks categorized by pipe sizea

Pipe Size Numerator Denominator Lower Bound Estimate Upper bound
(in. in dia) (N failures) (T years) X~O.05. 2N/ 2T (NIT)
X~O.95. 2N + 2/ 2T


1/2 to 2 2 484.73 0.0007 0.0041 0.0130

>2 to 6 4 484.73 0.0028 0.0083 0.0189
>6 3 484.73 0.0017 0.0062 0.0160



1/2 to 2 3 313.36 0.0026 0.0096 0.0247

>2 to 6 2 313.36 0.0011 0.0064 0.0201
>6 5 313.36 0.0063 0.0160 0.0335

a. Note that no failures were observed in LOC A-sensitive piping.

Table 15. Frequencies of pipe breaks categorized by leak rate

Leak Numerator Denominator Lower Bound Point Upper Bound

Rate (N failures) (T operating years) XfOO5.2,../2T Estimate a Xf09<. 2' _2/ 2T



~ 1, < 15 gpm 5 484.73 0.0041 0.0103 0.0217

~ 15 gpm 4 484.73 0.0028 0.0083 0.0189


50 to 500 gpm 0 484.73 0 0.0005 0.0062

>50 gpm 0 484.73 0 0.0005 0.0062



~ 1, < 15 gpm 1 313.36 0.002 0.0032 0.0151

~ 15 gpm 9 313.36 0.0150 0.0287 0.0501


500 to 5000 gpm 0 313.36 0 0.0007 0.0096

> 5000 gpm 0 313.36 0 0.0007 0.0096

a. Point estimate l\ = NIT if N > 0; if N = 0, l\ = X;050. IN _ 1/2T.

b. Non-lOCA systems are those systems that if disabled, could not mitigate a lOCA (see Table 1).
c. lOCA systems are those systems susceptible to piping failures that could result in loss of reactor coolant (see Figure 1).

Table 16. Frequencies of pipe breaks categorized by system

Numerator Denominator Lower Bound Point Upper Bound

System (N failures) (T years) XZO.05, 2N/ 2T Estimatea XZ0.95. 2N + 2/2T


Residual heat 2 484.73 0.0007 0.0041 0.0130


Chemical and volume 5 484.73 0.0041 0.0103 0.0217


Main feedwater 2 484.73 0.0007 0.0041 0.0130

All other PWR 0 484.73 0 0.0005 0.0062



High pressure 2 313.36 0.0011 0.0064 0.0201

coolant injection

Condensate 4 313.36 0.0045 0.0128 0.0292

Main feed water 3 313.36 0.0026 0.0096 0.0247

Service water 313.36 0.0002 0.0032 0.0151

All other BWR 0 313.36 0 0.0007 0.0096


a. Point estimate A = NIT if N > 0; A = \~(}jO, 1N + t/2T.

Table 17. Frequencies of pipe breaks categorized by operational mode

Operational Numerator Denominator Lower Bound Point Upper Bound

Mode (N failures) (T years)a X~O.05. ~N/2T Estimate b \fO.95.~" _ ~/2T


Starting up 2 12.60 0.0282 0.1587 0.4997

Normal 5 305.38 0.0065 0.0164 0.0344

Shutting down 0 3.39 0 0.0671 0.8837

Shutdown 2 163.36 0.0022 0.0122 0.0385

Transient 0 3944 0 0.0001 0.0008


Starting up 0 9.40 0 0.0242 0.3187

Normal 8 195.22 0.0204 0.0410 0.0739

Shutting down 0 3.13 0 0.0727 0.9571

Shutdown 2 105.61 0.0034 0.0189 0.0596

Transient 0 1970 0 0.0001 0.0015

a. For the transient operational mode. the denominator is the number of transients over all operating years.

b. Point estimate A = NIT if N > 0; if N = O. A = 'To.5o. 2N + 1/2T.

analysis, which are listed in the "Overall Methodol- to permit calculating rates on a per weld or per foot of
ogy" section. However, the piping experts consulted pipe basis.
during this study stated that knowing the total number The "Overall Methodology" section shows the
of welds in a system was not as crucial as knowing WASH-14oo pipe break failure frequencies for
certain attributes of the welds, such as their location. LOCA-sensitive piping. These rates are based on
That is, there is not as strong a correlation between the 150 reactor years of LWR operation with no pipe
number of pipe failures and the number of welds in a ruptures occurring in LOCA-sensitive systems. sup-
system as there is with the type and location of the plemented by additional data from other sources.
welds. Because the data sources available did not corre- As of the cutoff date for this study,
late (or even address) weld attributes with pipe failures, December 1984, 798.09 U.S. commercial LWR
failure rates per weld were not calculated. Rates per years had transpired with no failures; i.e., no leak
foot of pipe also were not calculated. The populations rates> 50 gpm experienced in LOCA-sensitive sys-
for both welds and lengths of pipe are extremely varia- tems for PWRs or BWRs. As shown in
ble so that rates required on a per weld or per length Appendix F, using a chi-squared distribution on
basis ideally should be estimated on plant-specific zero failures for 798.09 reactor years results in a
terms, if such rates are desired. However, population point value of 3.0E-4 with an upper bound of .f.OE-3
information on pipe length and weld counts, including and a lower bound of zero. This result is compa-
indications of their variability, was developed and is rable to that of WASH-1400 for medium breaks
summarized in the "Population" section of this report and has virtually the same uncertainty bounds.

Thus, the only sense in which this result represents a would be the same as shown above; 3.0E-4 with the
reduction in the uncertainty bounds of WASH-I400 is same upper and lower bound.
that this study is based directly on U.S. LWR opera- At the time of this study, other NRC-sponsored pro-
tional data for LOCA-sensitive systems, and does not grams utilized probabilistic techniques and fracture
contain uncertainties from extrapolating from other mechanics theory to determine the probability of a
data sources. pipe failure in the reactor coolant system of a PWR
A significant difference between the results pre- induced by fatigue crack growth resulting from the
sented here and those presented in WASH-l 400 is that, combined effects of thermal, pressure, seismic, and
for the leak rate categories established here (for LOCA- other cyclic loads. One such study is reported in
sensitive piping), the failure rates are the same; the fail- NUREG/CR-3663,7 and is summarized in
ure rates for the various pipe size categories shown in Appendix E. The best-estimate probability reported in
\\i\SH-1400 are different. The reason for this differ- that study is a range from 1.5E-8 to 2.3E-8 events per
ence is that WASH-I400 used failure data from non- year. These values differ from the 5.0E-4 events per
commercial U.S. reactors and this study did not. year for PWRs reported here by 4 orders of magnitude.
Among data from LOCA-sensitive systems for U.S. However, this value does not necessarily contradict the
commercial NPPs, there is no evidence that the rates value reported in the fracture mechanics study because
differ by pipe size or by leak rate categories. Therefore, the frequencies reported here are based on an actual
using only U.S. commercial reactor failure data, the operating history of approximately 800 reactor years
pipe break failure frequencies for the pipe size catego- and will continually become smaller over time, assum-
ries of 112 in. to 2 in., 2 in. to 6 in. and >6 in. ing no failures occur.


1. N. C. Rasmussen et aI., Reactor Safety Study, an Assessment of Accident Risks in U.S. Commercial
Nuclear Power Plants, WASH-1400, NUREG-75/014, October 1975.

2. N. G. Catheyet aI., Catalog of PRA Dominant Accident Sequence Information, NUREG/CR-330 I,

EGG-2259, July 1985.

3. G. 1. Kolb et aI., Reactor Safety Study Methodology Applications Program: Oconee #3 PWR Power
Plant, NUREG/CR-1659/2 of 4, SAND80-1897/2 of 4, Revised May 1981.

4. A. A. Garcia et aI., Crystal River-3 Safety Study, NUREG/CR-2515, SAND81-7229,

December 1981.

5. S. W. Hatch et aI., Reactor Safety Study Methodology Applications Program: Grand Gulf #1 BWR
Power Plant, NUREG/CR-1659/4 of 4, SAND80-1897/4 of 4, October 1981.

6. U.S. Nuclear Regulatory Commission, Licensed Operating Reactors, NUREG-0020, 1982-1984.

7. T. Lo et aI., Probability ofPipe Failure in the Reactor Coolant Loops ofCombustion Engineering PWP
Plants, Volume 2: Pipe Failure Induced by Crack Growth, NUREG/CR-3663, January 1984.





INTRODUCTION. . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-3

Eliciting Subjective Opinion A-3

Failure Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-4

Yes/No to Dependency of Rates on Operational Mode A-4
Yes/No to Change in Failure Rates for Newer Vintage Plants A-4
Failure Causes Associated with the Given Conditions A-6
Personal Confidence Rating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-6


Combining Experience Data With Subjective Data Using Bayesian Methods. . . . . . . . . . . . . . A-8

Prior Distribution A-8

Posterior Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-9

Analysis of Variance of the Failure Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-IO

Model and Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-IO

Significance Tests. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-II


Significant Terms in the Model..................................... A-12

Discussion of Significant Interaction Terms A-12

Final Conditional Factor Multipliers and Failure Rates A-13

Variance of Estimates A-I?

Failure Causes A-18

Discussion of Combined Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-19

REFERENCES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-31

(Figure A-5 on microfiche attached to inside backcover.)





Because very few actual plpmg failures have gained in the process of becoming an expert. This
occurred in the nuclear power industry, this study study, therefore, does not "resort to" subjective
supplements the experience data with subjective data, but chooses to tap those resources of experts
data. This appendix discusses the process used and and elicit information on failures in piping systems
its results. The section, "Eliciting Subjective Opin- in nuclear power plants.
ion," discusses the methods used in collecting opin- Little improvement is gained by using more than
ions from the piping experts. six experts. A-I Nine respondents were confirmed
A major portion of the appendix addresses the for this study because it was anticipated that some
methods used to analyze the actual and subjective experts would not assess the failure frequencies for
data. The data are combined using Bayesian statistical all the given conditions. All nine respondents are
methods. The resulting Bayesian probabilities of fail- well-known recognized experts with strong back-
ure are then treated as weights on the rates submitted grounds in materials and piping. The experts are
by the experts. The discussion concludes with details of either consultants, some owning and operating
the final procedure, called analysis of variance, used to firms, or are employed by national engineering lab-
produce the frequencies of interest. oratories. Of the nine respondents confirmed, five
The "Results" section presents significant terms complete responses were received for analysis.
from the procedure. From the analysis of variance, It was expected, and actually hoped, that the
the final predicted rates and uncertainties are deter- group would hold as diverse viewpoints as possible
mined. The information from the questionnaire on on the subject in question. Seaver A-2 concluded
failure causes is also presented. Finally, results of that " ... a large diversity of individual opinion
this procedure are discussed. among group members will lead to greater super-
iority of the group judgment over individual judg-
ments." Martz A -3 also notes that the diversity of
Eliciting Subjective Opinion opinion tends to decrease the statistical dependency
among the individual opinions.
Quantifying risks is difficult at best, with a lack The fact that experts have the most difficulty assess-
of significant data. Experiencing zero failures or a ing rare events has been discussed repeatedly in the lit-
negligible number of failures, however, should not erature. This is not to say it cannot be done or the
be considered a lack of data. A zero is valid and resulting responses are meaningless. If careful and
significant if: (a) the definition of "failure" was scrutable means are used in the elicitation process, the
concise, specific, and clearly understood, leaving responses are indeed useful. Therefore, much care was
little or no question if an event qualified as a fail- taken to ensure that all respondents were supplied with
ure; (b) all relevant data sources were carefully and clearly understood such variable terms as failure
searched; and (c) all failures, as defined, were and rupture; diagrams, descriptions, and operating
reportable events and were submitted to a data parameters for each of the systems concerned were
source. Nonetheless, zero failures make the quanti- supplied to minimize the assumptions the respondents
fication of probabilities of risk difficult. When the may have made; and a completed sample questionnaire
evidence is overwhelmingly strong, e.g., when very form was included with an interpretation of the form
few failures are seen and many years of operating to ensure that the respondents understood how their
experience have passed, experts are guided by and assessments would be interpreted. The respondents
use this information. That is, the judgments of were also supplied with potential failure mechanisms,
experts encompass this information when it is testing and inspection routines, material and length of
appropriate. By the virtue of their expertise, they pipes within the systems, physical conditions of the
give information based on what laymen can systems, and descriptions of the operational modes. As
observe, augmented with knowledge they have experts, the respondents could carefully weigh each

piece of information to determine its significance to the sidered and are listed in Table A- I. Within each sys-
situation they were assessing. tem, rates were requested for each of fi\e
A portion of the information provided to the operational modes: startup, normal operation,
respondents is presented in Appendix B. Defini- shutting down, shutdown, and transients. Rates
tions, descriptions of the operating modes, and the were elicited on this level because the operational
operating parameters for each system are also in mode of the plant was seen as a potential source of
Appendix B. The well-known Delphi technique A-4 variation in the failure rates. Then, within each sys-
was used to elicit the subjective data. A detailed tem and each operational mode, rates for rupture
questionnaire was developed as a tool to implement events were categorized by pipe size. The reason for
the Delphi technique. The basic approach was to this was twofold: (a) the pipe sizes were, again, a
select six to twelve respondents, issue the question- potential source of failure rate variation, and
naire, compile the responses and associate each (b) one of the original task objectives was to update
respondent with an identification number, and reis- the rates in WASH-1400, which were based on the
sue the questionnaire. With this final step, the same breakdown of pipe sizes. For all systems, the
respondents were invited to review, evaluate, and breakdown was constant: 1/2 to 2 in. in dia, >2
modify their own responses if they felt it was appro- to 6 in. in dia, and> 6 in. in dia.
priate. This is known as a noninteractive feedback More rates were requested on system by opera-
method of the Delphi; the method allows for feed- tional mode by leak rate criteria. Two leak rates
back without physically assembling the experts, an were always given but were not constant among sys-
approach infeasible for this study, and tends to tems. The magnitude of the leak rate depending on
result in significant reductions in error in the esti- the system was discussed in the "Overall Methodol-
mates (Reference A-3). ogy" section of the main report. Table A-I details
Basically, five types of information were an association of the leak rates with the various
requested from the experts: systems. The respondents were asked to respond to
many sets of conditions. Considering the rates
• Failure rates for piping, by system, under based on pipe size (of which there are three), a
given conditions respondent could potentially submit
• Yes/No response to a question concerning 23 (systems) x 5 (operational modes) x 3 (pipe
the dependency of failure rates on the sizes) or 345 rates. Considering the rates based on
operational mode of the plant leak rate (of which there are two), he could poten-
• Yes/No response to a question concerning the tially submit 23 x 5 x 2 (leak rates) or 230 rates, for
change in failure rates for newer vintage a sum of 575 total failure rates elicited.
plants, i.e., those built after the mid-1970s
• Failure causes associated with given condi-
Yes/No to Dependency of Rates on Operational
Mode. The experts were asked to respond to the
• The expert's personal confidence rating for
each of the failure rates he has submitted. question "Do you feel that the frequency of pipe
failures in this system during the 40-year lifetime of
Each of these items is discussed below and can be iden- the plant is dependent on the operational state of
tified in the sample questionnaire form in Figure A-I. the plant?" See letter B on Figure A-I. If their
Although this form is designated for a Residual Heat opinion was "no", they simply entered one overall
Removal (RHR) system, the basic format was identical series of rates rather than one series for each of the
five modes given. Not only did this offer a shortcut
for all systems.
to responding to the questions when they did not
feel a dependency exists, but it also imposed a train
Failure Rates. Failure rates were requested based
of thought upon them. That is, the question brings
on three categories: system, operational mode, and
to mind the differences in potential failure condi-
severity. Two sets of rates were requested based on
tions for piping; e.g., during startup versus normal
severity; one set concerns ruptures by pipe size, and
the other concerns leaks by given leak rates. All
rates were requested on a per 4O-year plant lifetime
basis. See letter A on Figure A-I. Yes/No to Change in Failure Rates for Newer
Rates were elicited on a system level to supply Vintage Plants. For each system, respondents were
PRA analysts with a more specific rate. Ten PWR asked, "Would your response differ considering a new
and thirteen BWR safety-related systems were con- generation plant?" See letter C on Figure A-I. They


110 yo" feel lllit the frequO!flCY of pipe hlh,res In thts syste. dllrlng the 40 yr Ilfetl"", of th~ plant Is d~pendent on the operatIonal
,tate of Ihe plant? B
FaHurt! Condit Ions: Auptures In ¥arious Pipe Sizes
1/2 to Z In. >2 to Ii In. >6 In.

__!.!!!!.L."!lde _ Frequencl Ca"sels)b Confidence frequencl Cause( sib Confidence Freg"enc~ Causels)b Confidence
Starting liP A
( including 1101 standby) -o- --
E - ---
Hnrmal operat Ion

ShuLL Ing do....

( inc IlIdlng refueling)
Trans lent

Fat1ure ConditIons: Exceeding Various leak Aates

=1> 11 gp,.c 115 gp",d

Frequencl Causels)b Confidence Freqllencl Cause(s)b Confidence Add Itiona I QuesllonSlA....rks
Start tng up 1. IIpuld your response differ
( Inc Iud Ing hot standby) cOnS Idering a new generation
plant? yes no C operat Ion doii'TInow - -
Increase 1
Decrease [0 _ "
ShuLL Ing down

(Including refueling)
lrans lent

a. Failure frequency per 40 yr plant life, expressed as a • loeb.

b. likely contrlhutlng ca"sels!: see list.
c. lllis leak rate would be caused by an approxlute pipe break of 1.8 x 10- 5 tt 2 (0.06 In. dlaooeter).

d. This leak rate would be caused by an approxlNte pipe break of 2.7 x 10- 4 fl2 (0.2 In. dl_ter).

Figure A-I. Sample questionnaire form.

Table A-'. Scope of the pipe break study

Categories of Data

Based on Ruptures
of Given Pipe Sizes Based on \Iagnitude of Break
System (in. in dial (leak rate in gpm)


PWR primary 1/2 to 2 >2 to 6 >6 2: 50 (0.2 in. dial 2: 500 (4 in. dial

BWR recirculation 1/2102 >210 6 >6 2: 500 (1.3 in. dial 2: 5000 (10 in. dial
BWR main steam (reactor vessel to 1/2102 >210 6 >6 2:500 (4 in. dial 2: 5000 (10 in. dial
first main steam isolation valve)
BWR main feed (check valves to reactor vessel) 1/2 to 2 >2 to 6 >6 2:500 (1.3 in. dial 2: 5000 (10 in. dial


PW R emergency core cooling 1/2102 >210 6 >6 2:1 2: 15

PWR residual heat removal 1/2102 >2[06 >6 2:1 2: 15
PW R chemical and volume control 1/210 2 >210 6 >6 2:1 2: 15
PWR main steam 1/2 to 2 >210 6 >6 2:1 2: 15
PWR auxiliary feedllater 1/2102 >2 to 6 >6 2:1 ~ 15
PWR main feedllater 1/2 to 2 >2 to 6 >6 2:1 2: 15
PWR condensate 1/2102 >2 to 6 >6 2:1 2: 15
PWR condensate 1/2 to 2 >2 to 6 >6 2:1 2: 15
PWR component cooling water 1/2102 >210 6 >6 2:1 2: 15
PW R service water 112102 >210 6 >6 2:1 2: 15

BWR high pressure coolant injection 1/2 to 2 >210 6 >6 2:1 2: 15

BWR reactor core isolation cooling 1/2to 2 >2 to 6 >6 2:1 2: 15
BW R core spray 1/2 to 2 >2 to 6 >6 2:1 2: 15
BWR residual heat removal 1/2102 >2 t06 >6 2:1 2: 15
BWR standby liquid control 1/2102 >2t06 >6 2:1 2: 15
BWR main steam (\ISIV to turbine inlet) 1/2102 >210 6 >6 2:1 2: 15
BWR main feedwater (condensate system 112 10 2 >210 6 >6 2:1 2: 15
to check valves)
BWR condensate 1/2102 >210 6 >6 2:1 2: 15
BWR reactor building component cooling 1/2 to 2 >2 to 6 >6 2:1 2: 15
BWR service water 1/2 to 2 >2 to 6 >6 2:1 2: 15

were instructed to consider a new generation plant to be or were invited to enter their own failure cause if it
one built after the mid-1970s. If a respondent answers is not on the list. The suggested causes were: corro-
"yes", he was then asked to quantify the change, that sion (stress corrosion cracking, corrosion fatigue),
is, to indicate the percent increase or decrease in the thermal fatigue, erosion, mechanical fatigue, water
failure rate. hammer, external damage, manufacturing/design
error, and weld defect.

Failure Causes Associated with the Given Con-

ditions. For each system, operational mode, and Personal Confidence Rating. Along with each
pipe size or leak rate combination, the respondents failure rate an expert submitted, he was asked to indi-
were asked to indicate any dominant cause(s) for a cate his confidence in his response. See letter E on
failure. See letter D on Figure A-I. They were sup- Figure A-I. He was instructed to enter a "I" if his
plied with eight likely causes from which to choose, experience had not dealt with the particular set of

conditions, i.e., if he did not feel strongly about the confident with his response, while a "3" indicates a
frequency he had indicated. A "2" implies he felt fairly strong feeling of confidence about the response.


Combining Experience Data With severity conditions. The 40 years, on the average,
consist of the following for both PWRs and B\VRs:
Subjective Data Using Bayesian
Methods • 1.08 years in startup
• 25 years in normal operation
Once failure data, operating experience times, • 0.32 years shutting down
and the subject data were collected, the final pro- • 13.6 years in shutdown
cess in determining piping failure rates involved • 320 transients.
combining the sources. Bayesian statistical meth-
Thus, for calculating and estimating yearly rates
ods were used to perform this integration of data
for the various operational modes, the reported
sources. Martz and Waller A-5 note that statistical
rates were divided by the factors stated above. For
inferences based on Bayes' methods are usually less
transients, a rat<: per transient was thus obtained.
restrictive than those based on sampling theory due
More details on these unit conversions are found in
to the exclusive use of sampling data for the latter.
Appendix C.
The degree to which more informative inferences
occur using relevant past experience depends upon
Prior Distribution. As described above, plpmg
the quality of the assessments embodied in that
experts responded to a questionnaire that elicited fail-
experience. Martz and Waller go on to note that
ure rates for pipes under certain conditions. The infor-
"the Bayesian method usually requires less sample
mation sent with the questionnaire did not include the
data to achieve the same quality of inferences than
sample data located in this study. The resulting subjec-
the method based on sampling theory ... This is
tive probabilities formed the prior distribution. The
an especially important consideration in those
prior distribution of the parameter Acan be defined as
areas of application where sample data may be
a probability function or probability density function
either expensive or difficult to obtain, such as relia-
expressing the degree of belief about the value of A
bility" (Reference A-5).
The unknown parameter here is the rate of fail-
prior to observing a sample of the random variable x:
the number of failures, whose distribution function is
ure of pipes. The degrees of belief held by the
indexed by A.
(questionnaire) experts about this failure rate, A, is
As an example, assume five respondents were
used in Bayesian techniques. These degrees of
asked to assess the frequency of pipe failures per
belief, or subjective probabilities, describe what is
year in a certain system and the following responses
called the prior distribution of the parameter A. The
were submitted:
term "distribution" implies that A is a random vari-
able. "Prior", in the terminology, refers to a distri- Respondent (i) Response (\)
bution of A'S described before observing the sample
data, or number of pipe failures. After the subjec-
tive probabilities are summarized, the sample data I I x 10-"
2 I X 10- 3
are collected. The prior and the sample data are
3 I x 10-"
brought together to produce a posterior distribu- 4 I X 10-4
tion, posterior referring to after the data collection. 5 I X 10- 3
So the posterior distribution contains not only
information collected from a sampling procedure, The prior subjective assessments of the failure rate,
but is augmented through the prior with relevant A, are distributed as follows:
information from our respondents. Inferences can
then be made from this posterior distribution. A PIJA)
Before presenting further details about how the
prior distributions obtained from the respondents I x 10-" 215
were combined with the data to form posterior dis- I x 10- 3 2/5
tributions, some data conversions need to be men- I x 10- 4 1/5
tioned. The respondents gave rates for the time
during 40 years that would be spent in each opera- Here, p,(A) re:ers. to the prio: .distribution; p,v(\) is
tional mode for each system and set of pipe break the pnor (subJective) probability that A equals x.

A minor adjustment to the prior distribution was Given the prior distribution for A, expressed as
made when the confidence levels submitted by the Px(A), and the conditional density of the elements
respondents were taken into consideration. The lev- of a sample, f'lx(xIA), the joint (unconditional)
els were coded 1 to 3 with a 1 indicating a low confi- density for the sample and the parameter is the
dence and a 3 indicating high confidence. Note in product of the two:
the example continued below, that while two
respondents submitted rates of 1 x 10-2 and two (A-I)
other respondents submitted rates of 1 x 10- 3, one
submitting the latter did so with a higher confi-
dence of 2. The A = 1 X 10- 3 was thus weighted to where the conditional density of the elements of a
give it a slightly higher prior probability than that sample, given A, is
for A = 1 X 10- 2 • Similarly, a response with a con-
fidence of 3 would be weighted the same as three f'lx(xIA) = [probability of x failures in
responses with a confidence of 1.
Adjusted Prior
Respondent (i) A, Confidence Weight A p,(A) time t, given Al (A-2)
1 x 10-2 1/6

1 x 10,2
For example, assume that x = I failure was
2 1 x 10-) 2 2/6 2/6
observed in a given system over t = 750 accumu-
3 I x 10-2 1/6 1 x 10-) 3/6 lated reactor years. Then we have:

4 1 x 10-4 1/6 1 x 10-4 1/6

I x 10-) 1/6
f X,A (I , A)I = [e-\' 750 . (Ai' 750)' ]

These adjusted prior probabilities were com-

. Px("A;) for each Ai (A-3)
pared at this point with the ratio of number of fail-
ures (n), to time (t). In order for the actual data to
have the opportunity to influence the resulting pos- Given the prior distribution discussed above, the
terior probabilities, the ratio of actual failures to joint density for these data would be
time had to lie within the range of prior probabili-
ties for the given set of conditions. If it did not, a ,
X PI, (Xi) f
'.' 11.,,)
new A was created and assigned the value of nit.
1 , 10- 2 2/6 0.0014
The prior probability associated with it was initially
set equal to the minimum existing prior probability 1 , 10- 3 3/6 0.1771
for that set of conditions, i.e., system, pipe size, 1 , 10,4 1/6 0.0116
and operational mode; then all the probabilities
were again normalized (as in the example above) so The second step is to compute the marginal den-
their sum was 1. sity of the sample values, independent of A. This is
the sum of the joint density over the range of A's.
Thus, f,(x) = Ef,.x(X,A). For this example, the
Posterior Distribution. The sample data enters probability that x = I failure is
at this point in what is called the likelihood func-
tion, which is a conditional probability, expressing f,(1) 0.0014 + 0.1771 + 0.0116
in this case the probability that x failures occur
over time, t, given a failure rate of A. That is, the 0.1901 . (A-4)
number of failures is described by a Poisson distri-
bution, conditional on a given failure rate. The pos- Because the posterior probabilities for the failure
terior distribution is based on turning this rate, A, are conditional on the sample values
conditional probability around and seeking a dis- observed, i.e., number of pipe failures, the last
tribution for the failure rate conditioned on the stage in the computation is to evaluate
observed number of failures. This transition occurs
in three stages, as described below. (A-5)

In t he example normal operation to shutdown. That is, the failure
rates may not behave similarly over the operational
1',,(1 \ 10. 2 '1) 0.0014/0.1901 = 0.0074, modes for each system. The failure rate for plants
in a particular mode may depend 0/1 the system; or
f,,(i \ 10')11) 0.1771/0.1901 = 0.9316, (/\-6) vice versa, the failure rates for a gilen system may
depend on the operational mode of the plant. This
1', ,II \ 10'.\' I) = 0.0 I 16 / O. I90 I = 0.0610. dependence is reflected in the interaction terms.
An analysis of variance (ANOVA) was per-
The posterior density is simply the conditional den- formed on the natural logs of the failure rates sub-
sity of the failure rate A, given the number of fail- mitted by the respondents. It was proposed that the
ures observed; the prior density expresses the dependent variable, y, the natural log of the subjec-
degree of belief of the location of the value of A tive failure rate, was a function of or II as influenced
prior to sampling, and the posterior density by the system, operational mode, and pipe size (or
expresses the degree of belief of the location of A, leak rate, as the data case may be). Interaction
given the results of the sample. terms were also tested for signi ficance: system by
operational mode, system by pipe size (leak rate),
and operational mode by pipe size (leak rate). :\
Analysis of Variance of the third order interaction term exists, system by mode
Failure Rates by pipe size (leak rate), but was assumed negligible
and so was pooled (combined) with the error term.
The rate of pipe failures is potentially dependent The dependent variable was "adjusted" before
on the system under consideration, current opera- its variation was studied. In particular, the subjec-
tional mode of the plant, and size of the pipes in tive failure rates Ilere Ileighted by the posterior
question (or, in the case of leaks, the magnitude of probabilities determined from the Bayes method.
the leak rate). Of course, many other factors may The subjective rates were weighted Ilith informa-
also contribute to the failure rate, but those factors tion from the actual failures observed. The \I eights
that are most important, most informative, and represent a relative accuracy of the responses. A
most feasible to measure are desirable. These fac- Ileight is more influential, or heavier, Ilhen the sub-
tors are called effects in the statistical technique jective data agrees with the actual failure data in the
known as analysis of variance. Bayesian method.
The analysis of variance methodology is used to
determine which of these effects are important;
that is, which effects are most influential on the Model and Assumptions. The response lariable,
dependent variable, the failure rate. The technique y, in the analysis of variance model represents natural
recognizes the various effects to be operating simul- logarithms of the subjective rates, weighted by poste-
taneously on the failure rate. rior probabilities determined from the Bayesian analy-
Analysis of variance is just that, a study of the sis. The following equation was used to model the
variation among the failure rates. One source of the response variable, referred to elsewhere in the results as
variation could be the nuclear power plant system t'nrate:
and could be quantified; another portion of the
variation could be partitioned and attributed to the
pipe size. After all reasonable sources of variation
are accounted for, some variation in the failure (A-7)
rates inevitably remains. This remaining portion is
attributed to random, inherent variation in the fail- where
ure rates. The significance of a specific measurable
effect is determined by comparing the specific vari- YiJld is the log of the response submitted by
ation due to that effect with the random variation. expert 1concerning system i, mode j, and pipe
An additional concept in analysis of variance size (or leak rate) k
that needs to be developed is interaction. A specific
example may best demonstrate interaction: failure iJ. is the ol'erall mean ('mate for the data set
rates for the primary system may decrease from
normal operation to shutdown, while rates for the Q
i is the effect on the overall mean (mate of the
residual heat removal system may increase from ith system such that 7Q ; = 0

/3 j is the effect on the overall means ('nrate of
the jth mode such that 1:/3, J
'h is the effect on the overall mean ('nrate of the
kth pipe size (or leak rate in Cases III through because E(e jju ) = O. 02(yij,l) = 0 2, the variance of
VI) such that l \ = a the error term, e jjkl , because it is the only random
term on the right-hand side of the model. Thus, the
(a/3)'J is the interaction effect on the overall Yijkl are· assumed to be normal with mean
mean I'nrate of the ith system and jth mode ~ + a; + (3j + 'k + (a/3)jj + (a');k + (/3')jk
such that 1:(a/3)
1 I)
= 1:(a/3)..
J I)
=a and variance 0 2 for significant tests.

(a,);, is the interaction effect on the overall

mean ('nrate of the ith system and kth pipe size Significance Tests. In evaluating the signifi-
(or leak rate) such that r(a');j = pa')jk = a cance of terms in the model, one considers whether
the data could come from simple models lacking
(/3) )j, is the interaction effect on the overall these terms (i.e., from models where the coeffi-
mean ('nrate of the jth mode and kth pipe size cients for the terms are zero). One evaluates
(or leak rate) such that 1:(/3')J'
= 1:(/3')j,
, .
=a whether the data are rare given a simple model; if
so, the term that was left out in order to make the
e ijU is the random error term. The e;jU are model simple is a significant one. Terms in the
assumed to be independent normal with mean model were considered significant at a level of
zero and variance 0 2 . 0.001. That is, the test statistic for the significant
terms were all large enough that the probabilities of
Note that the expected value of the responses satis- obtaining such large values given models without
fies the significant terms were all <0.001.


The final results of the study presented in this Discussion of Significant

section are based on six different sets of data.
Interaction Terms
Recall that the data were collected first on the basis
of ruptures of various sized pipes, and second, on
the basis of leak rates of given magnitudes. Interaction terms typically are not easy to interpret;
Case I represents data collected on a pipe rupture they can, however, be graphically displayed to aid the
basis from PWRs; Case II represents the same type analyst in noting obvious deviations in otherwise well-
data from BWRs. behaved data sets. Figures A-2, A-3, and A-4 are such
Cases III through VI categorize the data on a leak diagrams. Note in Figure A-3, as in the others, the log-
rate basis. Again, the data sets distinguish reactor arithms of the rates (i.e., the ('nrates) on the y-a\is are
types. Cases III and IV represent data from non- plotted for the five modes of operation; a single line is
LOCA PWR and BWR systems, respectively; i.e., plotted for each system.
those systems that, if disabled, could not mitigate an Analyzing the interaction terms of system by
accident. Case V involves PWR leak rate data from the mode involves comparing, over the systems, mean
primary system; a significant pipe break in this system ('nrates for each of the five modes. I I' the interaction
would initiate a LOCA. Case VI is comparable in that is not significant, the situation would be this: for all
it represents leak rate data from BWR LOCA-sensitive systems, mean ('nrates would progressively increase
systems. These six cases can be summarized in the fol- or decrease from one mode to another through all
lowing tabulation. the modes. That is, with insignificant interaction,
the rates of failure could be different from one
mode to another, but the pattern of di fferences
Rupture data Case I Case II would be consistent for each system. A discussion
follows for the three significant interaction terms
Leak rate data
noted in Table A-2. Generic explanations for these
Case III Case IV interactions are not attempted due to the lack of
LOCA Case V Case VI information within this study to support such con-
In the sections below, the analysis of variance Case I (PWR rupture data). Figure A-2 is a dia-
results are presented and composite rates based on gram of ('nrates of pipe ruptures versus operational
both the observed data and the responses of the modes plotted for each of IO numbered PWR sys-
experts are presented. Recall that the observed data tems. System numbers in circles indicate that actual
(by itself) is presented in the main test and a listing failures occurred in that system and mode. One
containing all the questionnaire responses in deviation to be noted in the patterns of ('nrates of
Appendix C. Finally, results summarizing what the pipe ruptures over operational modes involves the
experts stated about likely causes of pipe break high pressure injection (2), residual heat removal
events is presented. (3) and main feedwater (6) systems during shutting
down. In general, the t'nrates of failure tend to be
lower during shutting down than normal opera-
Significant Terms in the Model tion. The t'nrates for the three noted systems, how-
ever, are much higher here. The t'nrate for essential
raw cooling water (10) is also slightly higher as the
The significance of the terms in the model was plant shuts down versus during normal operation.
tested for each of the six data cases. Case V con- A final deviation in the pattern of l'nrates of ruptures
tains data for PWR LOCA systems, of which there for this data case concerns transients. The ('nrates per
is only one, the primary system. Therefore, the transient are typically less than the t'nrates for the other
model for Case V is reduced to only contain opera- four modes. For the auxiliary feedwater (6) and main
tional mode, leak rate, and mode by leak rate inter- feedwater (7) systems, however, the ('nrates are not
action terms. minimal. In reviewing the responses of the question-
Significant terms are indicated in Table A-2 by naire, it is apparent that the experts feel these systems
an asterisk. are more subject to water hammer during transients

Table A-2. Significant terms in the model for each data case (a = 0.001)

System System Mode

System Mode Size a Mode Size a Size a

Case I * *
Case II * * *
Case III * *
Case IV *
Case V NA * NA NA

Case VI *

a. For Cases III through VI, size implies leak rate.

and, therefore, are subject to more failures than are down than for normal operation. Note that for the
other systems. circled systems (i.e., those where actual failures
Case II (BWR rupture data). System by opera- occurred), the fmates are high.
tional mode was again the only significant interac- Also similar to the rupture data, is the high {'mate of
tion term in the model for BWR pipe ruptures. leaks in the auxiliary feed water system (6) during tran-
Figure A-3 is a diagram of this interaction. The sients. The system comes on line in this mode and,
high rates for the high pressure coolant therefore, increases its risk of leaks. Unlike the rupture
injection (12) and condensate (21) systems during data, the main feed water system (7) is apparently not
normal operation are explained by failures having believed to be at an increased risk of leaks during tran-
occurred. The portion of the feedwater system sients. Perhaps the difference in failure conditions
(20) from the condensate system to the check valves (rupture versus leak) can account for this. That is,
also has a high fmate. A casual review of the failure while the main feed water system is more likely than the
causes submitted by the experts for this system other systems to experience a rupture during a tran-
under normal operation indicates they feel water sient, it is no more likely than other systems to experi-
hammer and mechanical fatigue to be prime causes ence a leak.
of ruptures. Cases IV, V, and VI. No significant interaction
Another deviation in the pattern of fmates of terms existed in these data sets.
ruptures is in the shutdown mode. The service
water system (23) has a fmate higher than all oth-
ers. And during transients, the experts felt that the
Final Conditional Factor
LOCA-sensitive portions of the main steam (17) Multipliers and Failure Rates
and main feed water (19) systems were at a higher
risk of pipe ruptures than during shutdown. Although the interaction terms discussed in the
Case III (PWR non-LOCA leak rate data). Fig- previous section were significant terms in the
ure A-4 illustrates the interaction in fmates model, their effects are not incorporated into the
between system and mode for leaks in PWR non- final rates presented in this section. From a user's
LOCA-sensitive systems. Because it was deter- standpoint, the presentation of rates is greatly sim-
mined that the frequency does not differ plified, and more useful, when the interaction
significantly for> I and> 15 gpm leaks, the dia- effects are excluded. The intent of the previous sec-
gram generalizes over both categories. tion was to highlight the items that were responsible
As in the PWR rupture data (Case I), fmates for for causing the interactions to be significant; the
systems 2 (HPI), 3 (RHR), and IO (essential raw user should review the figures and noted items in
cooling water) are much higher during shutting that section and be aware that some interactions






-14 -I



_ -20
.5 j


-2 Syst...
, p,l "
2 high pre..u... '.J.ctlOlI
J ...sldull , ......,
4 chtollctl Ind .01_ co.tro'
5 .. t. st...
I lUI 11 It" ,._to,
7 .. t. tt.d"Uo,
8 condenuto
, C_M.t cool t., Wltor
10 o....,tlll ,.. coo"" Wlter


-34.....- - - - - - - - - - - - - - - -
St.,t., No...l
Shvtd.... t.,
TrI.II•• ts

Oper.tlo••' IIOde

Figure A-2. System by mode interaction (PWR pipe ruptures).




-14 ·8



_ -20
~ "E
. j

11 rec: I",ullt Ion
-24 12 high pressure coollnt InjectIon
13 re.cto, co,e lsoI.tlon cooling
14 co,e sproy
15 ,esldu.l hut r....v.l
16 st.ndby liquId control
17 .. in ste"ILOCAl
18 .. tn stu. non-LOCAl
19 feed•• te, ILOCA)
20 feedWi te, non-lOCAl
21 condenSlte
22 ,tic tor building c_ent
-12 cooling Wlter
23 senlce Wlter

o f.llure(s) occurred In this syst. and operltlo..l

St.rtup 110,..1 Shutting Shutd.... 'rlnslents
optr.tlon do....

Figure A-3. System by mode interaction (BWR pipe ruptures).







-I. -6



_ -zo
,! i

Z hlqh pressure I.Jectlon
3 restdu.l helt r_.11
4 ch.-tcil Ind ••1_ control
5 .. t. st...
6 .ulll t.r1 fHdwater
7 .. ,. fe_ter
a c••dens.te
-12 9 c _ t coolt., ..tor
10 esse.tlil ra. cool t., ..tor

o f.llurels) occurm 1. this S1St.- .nd .perltlonal IIllde


St.rtup 110...1 St,uttl., Shutdown Yr.n,lents
....rUI•• down
Operatton.l lIllde

Figure A-4. System by mode interaction (non-LOCA PWR leaks).

exist. So, although the interactions are recognized, magnitude differences in the rates (the multipliers
the effects are not incorporated into the final rates. are generally within 0.1 and 10). This, however, is
The models, therefore, are reduced at this point not the situation for the conditional factor,
to contain only main effects, i.e., system, mode, "mode". For most data cases, the rate of failures
and pipe size (or leak rate). In Case V, only mode for a solid year spent in startup is generally two
and leak rate are present in the model because the orders of magnitude higher than the overall rate
data set only contains the primary system. An anal- and the rate of failures per transient is generally
ysis of variance was run on each of these models to three to four orders of magnitude lower than the
determine estimates of the main effects. The main annual rate of occurrence of failures.
effect of, say, the jth mode is simply the difference Because the effects of system and pipe size (or
between the mean fnrate for the jth mode and the leak rate) were not significant (except in Case II),
overall mean fnrate. In symbols, in the model rates are presented based only on operational
mode. That is, the effects of system and pipe size
fnrate = J.L + cx sySlern + {3rnode + 'Y pipe size' on the fnrates were not, in general, significantly
different from zero. When this was the case, the
the overall mean fnrate is system (i) and pipe size (k) terms were assumed to
be zero and the model is reduced again, to
fnrate = J.L + cxrnode j' A predicted fnrate is calcu-
fl = (J + uSYSIern + iJrnode + I'pipe size (A-9) lated for each mode and is applicable to each sys-
tem and each pipe size. Tables A-4, A-6, A-8, A-IO,
where u, 7J and I' are averages of the parameters A-12, and A-14 contain these predicted values and
produced by the analysis of variance for system, their uncertainties. Case II (BWR rupture data) is
mode, and pipe size, respectively. an exception in that the pipe size was a significant
The mean Cnrate for the jth mode is conditional factor in addition to operational mode.
(Although the effects or multipliers for the rates
flj = (J + uSYSIern + {3rnode j + I' pipe size (A-IO) only vary from 3.0 for small pipes to 0.3 for large
pipes, the overall variation in this data case is rela-
The effect then, of the jth mode is tively small. The significance of the factor is deter-
mined by comparing the variation in the effect
estimates to the overall variation; a resulting large
ratio thus implies the term is significant). So for
- «(J + U + j3 + 1') Pj - j3 (A-II) Case II, predicted fnrates, rates, and their uncer-
tainties are calculated by taking into consideration
Estimates of the effects for the six data cases are both mode and pipe size.
displayed in Tables A-3, A5, A7, A-9, A-II, and Consider Table A-4, Aggregated Rates for Case I
A-I3. The last column of these tables presents esti- (PWR Pipe Ruptures). The predicted mean fnrate
mates of the effects for rates (rather than for fnrates), of pipe ruptures for startup is -14.091 and the pre-
denoted e<f,J - ~). These values can be interpreted as mul- dicted mean rate is e- 14 .091 or 7.6E-7. This value
tipliers or factors that indicate how the rates of pipe would apply to any system and pipe size because
ruptures or leaks vary among the levels of the condi- neither had a significant effect on the overall mean
tional factors. fnrate of pipe ruptures.
Consider Table A-3 for example. The 66.1 in the
final column across from startup indicates that
during this mode, the frequency of pipe ruptures Variance of Estimates. The uncertainty associ-
for PWRs is expected to be approximately 70 times ated with the predicted rates is the sum of two
greater than the overall rate of pipe ruptures per terms. One term is the variance of the mean (o~).
year. The multipliers for operational mode provide This value expresses variation in the coefficient
estimates of how the failure rates for a full year estimates used to determine the predicted fnrates.
spent in that mode compare with the overall mean The second term is the Bayes variance (o~). It
rate for that data case. reflects variation among experts and hopefully cap-
A review of these tables shows that in most cases, tures, in some sense, the between-plant variation,
differences among the systems and among pipe the experts having had different perspectives on
sizes (or leak rates) do not account for order of viewing power plant performance.

The Bayes variance, a~ i' for a group i defined by The failure cause data was elicited from the experts
a system, mode, and pipe size, is calculated by tak- for informational purposes only; no statistical analysis
ing the difference was performed on this data. Figure A-5 (on microfiche
on the inside back cover) illustrates the relative ranking
E(fnrate 2 ) - [E(fnrate)]2 where, of the failure causes submitted by the experts for each
system. Each page contains two tables. The upper table
E(fnrate 2) = l:(fnrate 2 . posterior shows the relative ranking of causes of pipe ruptures
probability) and (A-I2) for a specific system and the lower figure displays
causes of leaks for that system.
E(fnrate) = l:(fnrate . posterior probability). The scale used in the figures represents the fre-
quency that the causes were submitted. More than
The summation is over experts. If a variance is to be one cause could have been indicated for a pipe
determined for a given mode, but over all systems failure. If the respondent listed more than one, the
and all pipe sizes, the a~,i's are pooled for those
percentage attributed to each failure cause was
requested from the respondent. If the percentage
was not provided, then it was distributed equally.
For ruptures in PWRs, the experts felt the domi-
nant failure causes for most systems (7 of 10) were
where the summation is over all the groups and
mechanical fatigue and corrosion. For the chemical
n· and volume control system, corrosion was felt to be
7ri = -'- (note that ~7ri = I); the dominant failure cause with mechanical fatigue
~ni i second. For main feed water and auxiliary feed-
water, the dominant failure cause was water ham-
Y i = E(fnrate) (see the equation above), (A-14) mer followed by mechanical fatigue.
For failures defined as leaks in PWRs, the failure
causes parallel those for failures defined as rup-
tures with the exception of the condensate system
whose dominant failure causes were mechanical
The variance of the mean, ~l' is an output of the fatigue and water hammer.
analysis of variance software. For rupture failures in five BWR systems [High
Pressure Coolant Injection (HPCl), Reactor Core
Isolation Cooling (RCIC), core spray, Residual
Failure Causes Heat Removal (RHR), and the LOCA-sensitive
portion of main steam), the dominant failure
causes were seen as water hammer and corrosion.
The respondents to the questionnaire were asked to Five other systems [recirculation, standby liquid
identify the dominant causes of pipe failures for each control, the non-LOCA portion of the main steam
operational mode (or overall if failures were deemed system, the LOCA portion of the feedwater system,
independent of operational mode) for each system. and the Reactor Building Closed Cooling
Eight failure causes were provided in the question- Water (RBCCW) system] have mechanical fatigue
naire as a guideline but the respondents were free to and corrosion viewed as dominant causes of rup-
indicate others if they so desired. The eight failure tures. Respondents felt the feedwater system on the
causes provided in the questionnaire were: corrosion non-LOCA side was subject to water hammer and
(stress corrosion cracking, corrosion fatigue); thermal mechanical failure. They felt that pipe ruptures in
fatigue; erosion; mechanical fatigue; water hammer; the condensate system would most likely be due to
external damage (physical abuse); manufacturing/
mechanical fatigue or erosion and problems in the
design error; and weld defect.
service water system would be attributed to corro-
Two additional categories were added by the
sion and erosion. Experts felt that a dominant
respondents. One was pitting/thinning used in the
cause of leaks in all BWR systems under consider-
service water system only; the second was a combi-
ation is mechanical fatigue. Other leading causes of
nation of provided causes, corrosion-assisted
leaks were seen to be corrosion, water hammer, and
fatigue due to pressure/temperature cycling of a
weld defect.

Discussion of Combined Results ing failures. (The rate varies somewhat, although not
tremendously, depending on the failure condition
being a rupture or leak.) Per transient, it is estimated
The goal of this project was to update and that most piping systems would experience approxi-
expand, if possible, the study on pipe break failure mately IE-12 failures (ruptures or leaks).
frequencies reported in WASH-1400. Frequencies The magnitude of the variance term for the ratio
of piping failures were determined in the current estimates, u~ + u§, in even-numbered Tables A-4
study by supplementing sparse actual data, i.e., through A-14 is unsettling. The variance of the
observed failures and operating experience, with mean (mate, u~, is negligible relative to the Bayes
subjective data supplied by piping experts. Point variance, u~; unfortunately, the variance term must
estimates and uncertainties incorporating both include this variability among experts (u~).
sources of data have been presented in the section, The upper limit in most cases is typically seven to
"Final Conditional Factor Multipliers and Failure eight orders of magnitude greater than the point
Rates." estimate. For transients, 10 orders of magnitude is
In general, the failure rates were characterized pri- not uncommon. Because of these huge uncertain-
marily by operational mode. Initially, rates were cate- ties, it is concluded that the rates presented in
gorized by system and pipe size (or leak rate) as well. Tables A-3 through A-14 are not useful input to
Through an analysis of variance, however, it was deter- probability risk assessments (PRAs).
mined that differences in the rates among systems and The resulting wide variation in the experts' opin-
pipe sizes were negligible. The resulting rates for a given ions perhaps indicates that the failure situations
operational mode, then, are applicable to any system were not assessable. This is an especially feasible
and pipe size in that data set. hypothesis because no piping failures have been
The rates themselves vary by approximately six observed in the majority of the systems under con-
orders of magnitude over the five operational cern. The experts, therefore, are in a position of
modes. Failure rates for startup, normal operation, conjecturing the failure situation and attempting to
and shutting down are generally highest. Estimates quantify it. The differences in backgrounds of the
for these modes range roughly from IE-6 to IE-7 experts possibly compounds the problem. Without
failures per solid year in the modes with shutting actual failures to focus on, the experts are guided by
down having the smaller rates. (This statement gen- their backgrounds irito various directions, thus
eralizes mer ruptures and leaks; the rates do vary, arriving at the diverse assessments.
as shown in the previous set of tables, for the two From PRA analysts' point of view, the results
failure conditions.) It is not surprising that rates for of the combined data (subjective and actual) are
these modes are higher than the others considering not meaningful considering the large uncertainties
the transient state, although controlled, of the associated with the rates. Again, the reason for
plant during startup and shutting down. The rela- this must be taken into careful consideration.
tively high rates for normal operation are also not Experts were asked to assess situations that have
curious considering the large amounts of time the not yet been experienced in the nuclear power
piping systems are in normal operation. industry. So, although the high reliability of pip-
Failure rates for shutdown and transient are consid- ing is cause for great difficulty in determining
erably smaller. It is estimated that in a solid year of meaningful estimates of failure rates, the nuclear
shutdown, a plant would experience roughly IE-9 pip- power industry benefits.

Table A-3. Estimates of conditional factor effects for Case I (rupture data for PWR systems)

t. = overall mean fnrate = -18.282

e# = 1.15E-8

Mean Effect
etA! -~}
System (i) fnrate (t. j ) f).)
(J,Li -

I. Primary -18.3597 -0.0774 0.9

2. High Pressure -20.4505 -2.1682 0.1


3. Residual Heat -15.7889 2.4934 12.1


4. Chemical and -17.1029 1.1794 3.3

Volume Control

5. Main Steam -18.9535 -0.6712 0.5

6. Auxiliary Feedwater -18.3286 -0.0463 1.0

7. Main Feedwater -18.0606 0.2217 1.2

8. Condensate -18.7069 -0.4246 0.7

9. Component Cooling -19.2453 -0.9630 0.4


10. Essential Raw -17.8262 0.4561 1.6

Cooling Water

Mean Effect
Mode (j) fnrate (P) (t.j - p.) e(t J - ~l

I. Starting Up -14.0906 4.1917 66.1

2. Normal -15.7990 2.4833 12.0

3. Shutting Down -15.2407 3.0416 20.9

4. Shutdown -20.3559 -2.0736 0.1

5. Transients -25.9253 -7.6430 0.0005

Mean Effect
("J,Lk - J,L)
,.. e(~~ ~)
Pipe Size (k) fnrate (t. k )

A. Small (1/2 to 2 in.) -17.9494 0.3329 1.4

B. Medium (> 2 to 6 in.) -17.6217 0.6606 1.9

C. Large (> 6 in.) -19.2758 -0.9935 0.4

Table A-4. Aggregated rates for Case I (PWR pipe ruptures)

Variance Bayes
Mean of Mean Variance Upper Limit Rate Upper Limit
Mode fnrate (~) (a~) of fnrate a (ei'nrale) of Rate b

Starting Up -14.091 0.439 81.045 3.963 7.6 E-7 5.3 E+ 1

Normal Operation -15.799 0.439 86.629 2.863 1.4 E-7 1.8E+I

Shutting Down -15.241 0.439 92.731 4.064 2.4 E-7 5.8E+I

Shutdown -20.356 0.439 116.460 1.268 1.4 E-9 3.6 E + 0

Transients -25.925 0.439 117.973 -4.162 5.5 E-12 1.6 E-2

a. Upper limit of [nrate = mean I'nrate + Vo?:l + a~.

b. Upper limit of rate = exp(upper limit of [nrate).

Table A-5. Estimates of conditional factor effects for Case II (rupture data for BWR systems)

~ = overall mean fmate = -16.907

e~ = 4.55E-8
Mean Effect
e(A i -~l
System (i) fmate (ll j ) (~j - 0.)
11. Recirculation -16.4480 0.4585 1.6
12. High Pressure -15.2673 1.6392 5.2
Coolant Injection

13. Reactor Core -17.1614 -0.2549 0.8

Isolation Cooling

14. Core Spray -16.8693 0.0372 1.0

15. Residual Heat -16.3961 0.5104 1.7

16. Standby Liquid -18.0649 -1.1584 0.3

17. Main Steam (LOCA) -17.4736 -0.5671 0.6
18. Main Steam (Non-LOCA) -18.4686 -1.5621 0.2
19. Feedwater (LOCA) -18.0189 -1.1124 0.3
20. Feedwater (Non-LOCA) -15.3174 1.5891 4.9

21. Condensate -15.1724 1.7341 5.7

22. Reactor Building -18.6090 -1.7025 0.2
Component Cooling Water

23. Service Water -16.5175 0.3890 1.5

Mean Effect
Mode U> fmate (ll j ) e(~J -;l
(Ii; - jl)

1. Starting Up -12.9752 3.9313 51.0

2. Normal -13.0144 3.8921 49.0

3. Shutting Down -14.2208 2.6857 14.7

4. Shutdown -19.4558 -2.5493 0.1

5. Transients -24.8662 -7.9597 0.0003

Mean Effect
Pipe Size (k) fmate (~k) (, 1\ e(t k · ~)
J.Lk - J.L)

A. Small (112 to 2 in.) -15.8392 1.0673 2.9

B. Medium (> 2 to 6 in.) -16.6698 0.2367 1.3

C. Large(>6 in.) -18.2105 -1.3040 0.3

Table A-5. Aggregated rates for Case" (BWR pipe ruptures)

Variance Bayes
Pipe Size Mean of Mean Variance Upper Limit Rate Upper Limit
Mode (in.) fnrate (~) (u~ fnrate a efnrale of Rate b

Starting Up 1/2 to 2 -12.560 0.439 83.661 5.742 3.4 E-06 3.1 E+02
>2 to 6 -13.391 0.439 57.082 1.778 J.5 E-06 5.9E+00
>6 -14.931 0.439 55.446 0.002 3.3 E-07 1.0 E+oo

Normal 112 to 2 -12.599 0.439 86.457 6.045 3.4 E-06 4.2E+02

>2to6 -13.430 0.439 58.088 1.871 J.5 E-06 6.5 E+oo
>6 -14.970 0.439 76.390 2.560 3.2 E-07 1.3E+OI

Shutting Down II2to2 -13.806 0.439 102.376 6.474 1.0 E-06 6.5 E+02
>2 to 6 -14.636 0.439 82.758 3.606 4.4 E-07 3.7 E+OI
>6 -16.177 0.439 74.895 1.182 9.4 E-08 3.3E+00

Shutdown 1/2 to 2 -19.041 0.439 132.670 4.034 5.4 E-09 5.6E+OI

2 to 6 -19.871 0.439 101.261 0.298 2.3 E-09 1.3 E+oo
6 -21.412 0.439 98.857 -1.483 5.0 E-IO 2.3 E-OI

Transients 1/2 to 2 -24.451 0.439 147.054 -1.617 2.4 E-II 8.5 E-OI
>2 to 6 -25.282 0.439 123.392 -3.026 1.0E-II 4.9 E-02
6 -26.822 0.439 102.697 -6.511 2.2 E-12 1.5 E-03

a. Upper Ii mil of fnrale = mean fnrate + V~I + a~.

b. Upper limit of rate = exp(upper limit of fnrate).

Table A-7. Estimates of conditional factor effects for Case III (leak rate data for non-LOCA
PWR systems)

~ = overall mean fnrate = -17.433

e~ = 2.69E-8
Mean Effect
e(Oi' ~)
System (i) fnrate (fl) (~i - M

2. High Pressure -20.2105 -2.7781 0.1


3. Residual Heat -16.3958 1.0367 2.8


4. Chemical and Volume -14.0649 3.3676 29.0


5. Main Steam -17.7085 -0.2760 0.8

6. Auxiliary Feedwater -16.8313 0.6012 1.8

7. Main Feedwater -17.0497 0.3828 1.5

8. Condensate -18.1057 -0.6732 0.5

9. Component Cooling -18.6539 -1.2214 0.3


10. Essential Raw -16.8722 0.5603 1.8

Cooling Water

Mean Effect
eaLj - nl
Mode (j) fnrate (~) (~j - M

I. Starting Up -12.3874 5.0451 155.3

2. Normal -14.4575 2.9750 19.6

3. Shutting Down -13.6242 3.8083 45.1

4. Shutdown -19.9313 -2.4988 0.1

5. Transients -26.7621 -9.3296 0.0001

Mean Effect
e(~\..· ~J
Leak Rate (k) fnrate (~k) (~k - ~)

D. >1 gpm -17.6242 -0.1917 0.8

E. >15gpm -17.2408 0.1917 1.2

Table A-S. Aggregated rates for Case III (leaks in non-LOCA sensitive PWR systems)

Variance Bayes
Mean of Mean Variance Upper Limit Rate Upper Limit
Mode fnrate (~ (<7~) fnrate a e"nrate of Rate b

Starting Up -12.387 0.786 98.710 7.562 4.2 E-6 1.9E+3

Normal Operation -14.458 0.786 104.550 6.069 5.3E-7 4.3 E+2

Shutting Down -13.624 0.786 100.925 6.546 1.2 E-6 7.0E+2

Shutdown -19.931 0.786 117.912 1.859 2.2 E-9 6.4 E+O

Transients -26.762 0.786 130.854 -3.815 2.4 E-12 2.2 E-2

a. Upper limit of fmate = mean [mate + V~l + ,,~.

b. Upper limit of rate = exp(upper limit of fnrate).

Table A-g. Estimates of conditional factor effects for Case IV (leak rate data for non-LOCA
BWR systems)

~ = overall mean fnrate = -17.278

e~ = 3.14E-8

Mean Effect
e(~i 01
(~i - ~)
System (i) fnrate (~i)

12. High Pressure -15.9988 1.2787 3.6

Coolant Injection

13. Reactor Core -16.4645 0.8130 2.3

Isolation Cooling

14. Core Spray -17.7648 -0.4873 0.6

15. Residual Heat -17.8904 -0.6129 0.5


16. Standby Liquid -17.0390 0.2385 1.3


18. Main Steam -18.0506 -0.7731 0.5

20. Feedwater -16.7764 0.5011 1.7

21. Condensate -17.4849 -0.2074 0.8

22. Reactor Building -19.0704 -1.7929 0.2

Component Cooling

23. Service Water -16.2351 1.0424 2.8

Mean Effect
Mode (j) fnrate (~j) (~j - ~) e(~j'~

1. Starting Up -13.4351 3.8424 46.6

2. Normal -13.2508 4.0267 56.1

3. Shutting Down -13.5907 3.6868 39.9

4. Shutdown -19.8696 -2.5921 0.1

5. Transients -26.2411 -8.9636 0.0001

Mean Effect
e(~k' C,
Leak Rate (k) fnrate (~k) (~k - ~)

D. >1 gpm -17.3205 -0.0430 1.0

E. > 15 gpm -17.2345 0.0430 1.0

Table A·10. Aggregated rates for Case IV (leaks in non-LOCA sensitive BWR systems)

Variance Bayes
Mean of Mean Variance Upper Limit Rate Upper Limit
Mode fnrate (~) (a~) fnrate a efnrale of Rate b

Starting Up -13.435 0.670 86.916 5.282 1.5 E-06 2.0 E+02

Normal Operation -13.251 0.670 75.892 4.249 1.8 E-06 7.0E+01

Shutting Down -13.591 0.670 94.078 5.877 1.3 E-06 3.6E+02

Shutdown -19.870 0.670 111.903 1.350 2.3 E-09 3.9E+00

Transients -26.241 0.670 136.554 -2.812 4.0 E-12 6.0 E-02

a. Upper limit of fnrate = mean fnrate + 2Ja?:1 + <7~.

b. Upper limit of rate = exp(upper limit of fnrate).

Table A-11. Estimates of conditional factor effects for Case V (leak rate data for LOCA
PWR systems)

t = overall mean fnrate = -17.161

e~ = 3.52E-8

(Note: System = Primary)

Mean Effect
System (i) fnrate (~) (~i - ~)

I. Starting Up -10.280 6.881 973.6

2. Normal -12.283 4.878 131.4

3. Shutting Down -16.593 0.568 1.8

4. Shutdown -22.791 -5.630 0.004

5. Transients -23.858 -6.697 0.001

Mean Effect
Leak Rate (k) fnrate (~k) ()lk - ~) e(t k ·AJ

D. >50 gpm -15.927 1.233 3.4

E. >500 gpm -18.393 1.233 0.3

Table A-12. Aggregated rates for Case V (leaks in PWR primary system)

Variance Bayes
Mean of Mean Variance Upper Limit Rate Upper Limit
Mode fnrate (%) (u§) fnrate a et'nrate of Rateb

Starting Up -10.280 4.511 30.035 1.475 3.4 E-05 4.3E+OO

Normal Operation -12.283 4.511 20.013 -2.379 4.6 E-06 9.3 E-02

Shutting Down -16.593 4.511 109.255 4.739 6.2 E-08 1.1 E + 02

Shutdown -22.791 4.511 85.381 -3.829 1.3 E-IO 2.2 E-02

Transients -23.858 4.511 108.995 -2.550 4.4 E-II 7.8 E-02

a. Upper limit of fnrate = mean fnrate + 2J~1 + u~.

b. Upper limit of rate = exp(upper limit of fnrate).

Table A·13. Estimates of conditional factor effects for Case VI (leak rate data for LOCA
BWR systems)

Ii = overall mean fnrate = -19.078

e~ = 5.18E-9
Mean Effect
(M e{Ai-~)
System (i) fnrate (~i - fi)

11. Recirculation -20.1534 -1.0722 0.3

17. Main Steam (Reactor -19.5160 -0.4381 0.6

Vessel to MSIV)

19. Main Feedwater -17.5675 1.5103 4.5

(Check Valves to
Reactor Vessel)

Mean Effect
Mode (j) fnrate (~j) (~j - m e(t j -:1

1. Starting Up -14.0203 5.0576 157.2

2. Normal -15.5257 3.5522 34.9

3. Shutting Down -17.0214 2.0565 7.8

4. Shutdown -22.9062 -3.8283 0.02

5. Transients -25.9150 -6.8371 0.001

Mean Effect
fnrate (fl k) e(t k - ~,
Leak Rate (k) (~k - [1)

D. >500 gpm -18.0863 0.9916 2.7

E. >5000 gpm -20.0695 -0.9916 0.4

Table A-14. Aggregated rates for Case VI (leaks in BWR LOCA-sensitive systems)

Variance Bayes
Mean of Mean Variance Upper Limit Rate Upper Limit
Mode fnrate (<tt) (a~) fnrate a el'nral. of Rate b

Starting Up -14.020 1.846 81.494 4.238 8.2 E-07 6.9E+Ol

Normal Operation -15.526 1.846 73.407 1.824 1.8 E-07 6.2 E +00

Shutting Down -17.021 1.846 99.204 3.084 4.1 E-08 2.2 E+OI

Shutdown -22.906 1.846 105.952 -2.141 1.1 E-IO 1.2 E-Ol

Transients -25.915 1.846 124.232 -3.458 5.6E-12 3.2 E-02

a. Upper limit of I'nrate = mean I'nrate + 2';% + <7~.

b. Upper limit of rate = exp(upper limit of enrate).


A-I. C. A. S. Stael van Holstein, "Probabilistic Forecasting: An Experiment Related to the Stock Mar-
ket," Organizational Behavior and Human Performance, 8, 1972, pp. 139-158.

A-2. D. A. Seaver, Assessment of Group Preferences and Group Uncertainty for Decision Making, (SSRI
Research Report 76-4). Los Angeles: University of Southern California, Social Science Research
Institute, 1976.

A-3. H. F. Martz, Quantification of Informed Opinion, (LANL Internal Report No. S-I/84-332), Los
Alamos, NM, Los Alamos National Laboratory, May 1984.

A-4. K. Brockoff, The Delphi Method: Techniques and Applications, Reading, MA: Addison-Wesley,

A-5. H. F. Martz and R. A. Waller, Bayesian Reliability Analysis, New York: John Wiley and Sons, 1982.




DEFINITIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-4


Starting Up B-5

Normal Operation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-5

Shutting Down. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-5

Shutdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B-5

Transients ........................................................ B-5






The information contained in this appendix is iden- information contained in the questionnaire packet, but
tical to that supplied to the questionnaire respondents. not presented in this appendix, included descriptions
The purpose of providing experts with such informa- of all systems in consideration with diagrams of the
tion was to minimize the amount of assumptions they systems; completed sample questionnaire forms and
may have made and to provide a uniform basis from interpretations of those responses; and testing and
which to respond to the questionnaire. Additional inservice inspection routines.


• Frequency: An estimate of the expected Valve disc breaking loose

number of failures in a 40-year plant life- Valve bonnet leakage
time of a typical PWR or BWR plant that Valve stem packing or seal leakage
became operational in the mid- I970s. a Pump seal leakage
Pump/valve body leakage
• Failure: The escape of fluid through the Rubber and expansion joint leakage
piping pressure boundary. Heat exchanger leakage
Instrument leakage.
Types of failures intended to be included in
this study are: In addition, failure is further qualified by
severity as follows:
Through-wall cracks or breaks in pip-
ing pressure boundaries Hole area equaling or exceeding speci-
Flanged joint gasket blow-out caused fied amounts
by bolt breakage. Leak rates equaling or exceeding spec-
ified amounts a
Types of failures not intended to be Rupture of a pipe of specified diameter.
included in this study are:
• Failure conditions: Constraints on the sever-
Valve seat leakage ity of the failure being considered and/or on
the size of piping in which it occurs.

a. In the questionnaire, these estimates are requested for • Pipe Rupture: A virtually instantaneous
10 PWR and 11 BWR plant systems. That is, the expected num- catastrophic pipe failure that results in a
ber of failures estimated will be the number of failures (meeting pipe severance or a hole in the pipe greater
the failure conditions) expected to occur in each specified sys-
tem. In addition, the respondent has the opportunity to break than or equal to the cross-sectional area of
down each frequency further and indicate the number of failures the pipe.
expected to occur during the part of its 40-year lifetime that a
plant spends in each of the reactor operational modes as
defined. For example, a respondent might expect 10- 5 failures
during the 25 years of the plant's lifetime spent in normal opera- a. In the questionnaire, hole areas corresponding to various
tion and 10- 3 failures during the 300 to 400 transients that might system pressures and temperatures are provided to assist respon-
be expected to occur during a plant lifetime. dents in evaluating expected leak rates.


This project is concerned with pipe failures over the ture, pressure, and flow. During this mode, the
4O-year expected life of a nuclear plant. This expected emergency systems are in standby, and will operate on
life includes various plant operational modes. These demand.
modes have been categorized into the following: start-
ing up, normal operation, shutting down, shutdown,
and transients.
Shutting Down

The shutting down operation is a condition in which

Starting Up the nuclear plant is taken off-line, the reactor is shut
down, the primary coolant system is cooled down and
The starting up mode for this study encompasses depressurized, and all support systems are shut down.
two operational areas: The normal cooldown method uses the main feed water
system or the auxiliary feedwater system (PWR) to
• Startup of the nuclear plant from cool down the primary system to the point where the
shutdown!cooldown conditions low pressure residual heat removal system can complete
• Maintaining the plant for brief periods in a the cooldown. As systems are no longer needed during
shutdown condition in which the plant is the cooldown, they are shut down.
not cooled down and depressurized, and
starting the plant up from this condition.
In general, all systems, except the emergency core
cooling system (ECCS), are in operation sometime In this mode, the nuclear plant is off-line, the
during the startup when the startup is commenced nuclear reactor is shutdown, and the primary coolant
from a cooled down condition. The ECCS is placed in system is cooled down and depressurized. Support sys-
standby. During the startup mode, the primary system tems may still be in operation (i.e., decay heat
is taken to normal operating temperature and pressure. removal). However, these systems are shut down and
Criticality is achieved and the secondary systems depressurized as they are no longer needed. During this
(mainfeed and condensate) are placed in operation. time, the plant is typically involved in refueling and
Thermal power is increased such that the nuclear plant maintenance operations.
will supply electrical power. The systems in operation
during this time are essentially the same as those during
normal operation, except the systems are in transition
from a shutdown condition to a normal operational
condition (systems are undergoing both thermal and The last area of concern in nuclear plant operation is
pressure transitions). transients, which are defined as plant conditions or
plant upsets that result in a reactor trip. These events
cause both the secondary system and primary system
Normal Operation to undergo moderate (within design conditions) tem-
perature and pressure transients.
During the normal operation mode, the nuclear Table B-1 shows the general plant conditions
plant is on-line, generating electrical power. The pri- (temperature and thermal power) that determine
mary coolant system is at normal operating tempera- what mode the nuclear plant is in at any g"iven time.

Table B-1. Operational mode criteria a

Average Percent
Thermal Temperature of Year
Mode Power (Primary/Recirc) Spent in Mode b

Starting upc ~5t1Jo >200°F 2.5

Normal operation >5010 2:350°F 62.5

Shutting down 15% >200°F 0.8

Shutdown 0 ~200°F 33.6

Transients 8 transients/y

a. These are reasonable guidelines for both plant types but are not exact definitions of reactor modes.

b. Percentages are based on the actual operating experience of U.S. commercial PWR and BWR plants in 1983.

c. Starting up, for the purposes of this study, continues until the reactor is at normal operating temperature and pressure.

8-6 .

Tables B-2 and B-3 contain information charac- operating temperature and pressure, flow (gpm),
terizing PWR and BWR systems, respectively. The fluid, number of water hammer events, amount,
information includes such items as typical number size, and type of piping, and approximate number
of trains, pumps per train, type of fluid system, of welds.

Table B-2. PWR system information

Typical Total Size of Approximate Approximate

Number Typical Type of Operating Operating Water Largest Length of Number of
of Number of Fluid Pressure Temperature Flow Piping Hammer Pipe Pipe Welds in
System Trains Pumps/Train System (psig) (OF) (gpm) Material a Fluid Eventsb,c (in.) (ft)d System d

Primary 2,3, 1 Closed 2200 580 88,000 per Stainless steel Borated water 5 36 2500 973
Reactor or 4 pump

Core Cooling

a. HPSI 2 I Closed 200 to 2200 100 to 200 700 per pump Stainless steel Borated water 3 10 1050 559
at 600 psig or carbon (2000 ppm 1113 372
steel clad boran cone.)
stainless steel
b. LPSI 2 I Closed <200 100 to 200 5000 per pump I 10 173 122
at 100 psig 1702 468

Residual Heat 2 I Closed <200 100 to 200 5000 per pump Stainless steel Borated water I 14 173 172
Removal at 100 psig or carbon steel 1702 468
clad with (2000 ppm
stainless steel boran cone.)

Chemical I 3 Closed 2200 < 120 Normal 55 Stainless steel Borated water 2 6 3177 928
Volume and Maximum 99 19
Control 100 per pump

Main Steam 2,3, - Closed 700 to 1000 545 1.5 x 106 Carbon steel Dry saturated 7 42 6000 2177
or 4 steam

Auxiliary 2 2 or 3 Open 110 to 1200 40 to 120 1500 per Carbon steel Water 1 12 524 159
Feedwater pumps total train 100 48

Main I 2 Closed < 1575 <520 38,000 per pump Carbon steel Water 13 20 5900 1900
Table B-2. (continued)

Typical Total Size of Approximate Approximate

Number Typical Type of Operating Operating Water Largest Length of Number of
of Number of Fluid Pressure Temperature Flow Piping Hammer Pipe Pipe Welds in
System Trains Pumps/Train System (psig) (°F) (gpm) Material a Fluid Eventsb,e (in.) (fl)d System d

Condensate I 3 Closed <625 <380 25,000 per pump Carbon steel Water 4 16 noo 1500

Component 1 5 Closed < \50 <220 8000 Carbon steel Water I 12 857 504
Cooling (normal 90) 315\ 1155

Essential Raw 2 2 Open 20 to 175 40 to 95 36,000 (total) Carbon steel Water 2 30 1092 1719
Water 3944 710

t::O a. All waler systems are chemically treated 10 inhibit corrosion. The raw waler system is trealed (0 inhibit organic attack.
b. Does nOI include steam generator ',,"'3ter hammer evenls.

c. Reference: NUREG-0927 Rev. I, EWIIUGliO/1 of UtileI' Hammer Oc('urrel1ce in Nuclear POh'er Plums. /969 to Ala.\' 198/.

d. A set of two numbers represents pipes 210 6 in. and >6 in. in dia. One number represents only the average total numher of weld, or the a\'Cragc total length 01" pipe in a ">y"itcrn.
Table B-3. BWR system information

Typical Total Size of Approximate Approximale

Number Typical Type of Operating Operating Water Largest Length of Number of
of Number of Fluid Pressure Temperature Flow Piping Hammer Pipe Pipe Welds in
System Trains Pumps/Train System (psig) (OF) (gpm) Material a Fluid Evenls b (in.) (fl)c System C
--- -
Recirculation 2 I Closed 1100 <560 45,000 per Stainless steel Water 0 28 20 96
pump 173

High Pressure I I Closed 150 to 1120 <212 pump 5000 Stainless/carbon Water 20 24 404 101
Coolant suction steel 2508 401
Injection temp

Reactor Core I I Closed < 1135 100 10 150 600 Carbon steel Water 6 18 284 49
Isolation 236 160

Core Spray 2 2 Closed <450 <350 3125 per Stainless steel! Waler 9 16 73 51
pump 6250 carbon stccl 593 205

Residual Heal 2 2 Closed <450 <350 o to 12,000 Carbon steel Water 23 24 1310 215
Removal 1371 360

Standby 2 I Open <1500 70-100 39 PCI' pump Carbon stecl Boratcd water 0 4 60 39

Main Steam 4 - Closed 1100 560 13.4 x 106 Stainless/carbon Dry saturated 6 26 1400 214
Ib/h steel steam

Feedwater I 3 Closed 1000 375 to 400 "" 10,800 Primary Water 3 24 1029 51
per pump system 753 276
stainless steel.
Rcmainder of
system carbon
Table B-3. (continued)

Typical Total Size of Approximate Approximate

Number Typical Type of Operating Operating Water Largest Length of Numher of
of Number of Fluid Pressure Temperature Flow Piping Hammer Pipe Pipe Welds in
System Trains Pumps/Train System (psig) (OF) (gpm) Material a Fluid Events b (in.) (rt)e System C
--- ---
Condensate I 3 Closed 300 300 to 350 10,800 per Carbon steel Water 3 30 606 175
pump 1023 433

Reactor I 3 Closed <100 < 150 1700 per Carbon steel Water 0 24 2069 608
Building pump 851 515
Component ~3400

Cooling required
Water for normal

Raw Cooling I 4 Open < 150 70 to 110 7350 per Carbon steel Water 10 24 Unknown Unknown
, Water pump

a. All water systems are chemically treated to inhibit corrosion. The raw water system is treated 10 inhibit organic attack.

b. Reference: NUREG-0927, Rev. I, Evaluation "f Water Hammer Occurrence in Nuclear Power Plants, t969 to May t9R t.

c. A set of two numbers represents pipes 2 lO 6 in. and> 6 in. in dia. One number represents only the average total number of welds or the average lotallcnglh of pipe in a system.

As was discussed in the section, "Scope of the Pipe tion of equivalent break area, the temperature of
Break Study," specific leak rates were equated to hole the water that is used to compensate for leakage
sizes in the pipes. The purpose was to provide the study (make-up water) is assumed to be 280°F. This tem-
with a more meaningful question to pose to the respon- perature is near the maximum design high pressure
dents. That is, a question concerning a specific leak make-up limit. The leakage rate from which the
rate in a system was more difficult to address than a break area is calculated is based upon the mass flow
hole size because of the numerous dynamic factors rate of the make-up water. Once the mass flow rate
associated with a leak rate. is determined, the break area can be determined
For any given pipe break or rupture, the fluid from the data provided by Reference B-1 using the
loss in gallons per minute or pounds mass per hour following equation:
(Ibm/h) is determined primarily by the temperature
and pressure of the fluid within the piping system. (x gpm) (7.7433 Ib/gal H 20 at 280°F)
Break sizes were calculated for leak rates of I, 15,
50, 500, and 5000 gpm. Because the piping systems (l min/60 s) (l kgl2.20462 Ib) = y kg/s, (B-1)
considered in this study are not all subject to the
same pressure and temperature conditions, three where x is the given leak rate from Table B-4 and y
pressure and temperature combinations were used is the leak rate expressed in kg/so
to calculate the break areas for systems whose fluid From Reference B-1, a mass flux is obtained.
is subcooled water. Two combinations were used The mass flux is given in units of kg/s m 2 and is
where the fluid is saturated steam. determined in part by the temperature and pressure
In any pipe or weld, a failure very seldom occurs that exists at the break location. Reference B-1
in a regular shape. The break is most often jagged required that the pressure and temperature be
and irregular. Consequently, the geometric shape expressed in KPa and K or °C respectively. The
of a pipe break is extremely difficult to predict. break area can be determined using the mass flow
Therefore, the break areas used in the question- rate from Equation B-1 and the mass flux.
naire are assumed to be uniform circular breaks or
holes in the pipes. The homogeneous equilibrium
critical flow model was used to calculate the vari- mass flow rate (kg/s)
Break area (m 2) --------"-----. (B-2)
ous pipe break areas listed in Table B-4. The data mass flux (kg/s m 2)
required to calculate the break area were taken from
Reference B-1.
All nuclear power plant water systems have some The area in square meters is then converted to
means of compensating for leakage. In the calcula- square feet.

Table 8-4. Approximate hole size a (ft2)

Subcooled Water b Saturated Steam b

Leak Rate Hole Area Hole Diameter Leak Rate C Hole Area Hole Diameter
(gpm) Temperature/Pressure (ft2) (in.) (Ibm/h) Temperature/Pressure (ft2) (in.)

100/100 1.8 x 10- 5 0.06 467 550/1030 6.2 x 10- 5 0.1

400/1000 6.7 x 10-6 0.035
\5 100/\00 2.7 x 10-4 0.2 7,000 550/1030 9.4 x 10-4 0.4
400/\000 1.0 x 10-4 0.14
50 520/1100 5.2 x 10-4 0.3 23.300 550/1030 3.1 x 10- 3 0.8
9' 55012200 2.9 x 10-4 0.23
w 500 520/1100 9.2 x 10- 3 1.3 233,000 460/460 7.0 x 10-2 3.6
55012200 2.3 x 10- 3 0.6 550/1030 3.1 x W- 2 2.4

5000 410/450 6.6 x 10-2 3.5 2,330,000 460/460 0.71 11.4

520/\\00 5.2 x 10-2 3.1

a. Calculate using equilibrium critical flow parameters for water/steam.

b. All calculations assume 280°F makeup water. Exit water via the break assumes the temperature/pressure conditions given.

c. Pounds mass per hour expulsion rates for saturated steam are equivalent (in mass) to the corresponding gpm leak rates for subcooled water.

B-1. D. G. Hall, L. S. Czapary, Tables of Homogeneous Equilibrium Critical Flow Parameters for IVater
in 51 Units, EGG-2056, September 1980.






(Figures C-I, C-2, and C-3 on microfiche attached to inside back cover.)



This appendix contains the actual responses sub- were outlined in this discussion in order to give
mitted by the experts. Any modifications the them a clear understanding of how their responses
experts made in their responses during the iteration would be handled. If, for any reason, an expert did
step of the Delphi technique are incorporated into not intend for his response to be interpreted as
this set of data. The original instructions are fol- stated, he was invited to modify his response or
lowed by an interpretation of the responses. The indicate an adjustment to the assumption for his
experts received this interpretation when they were case. Following this discussion, responses are pre-
invited to review all the responses. All assumptions sented in Figures C-l, C-2, and C-3.


1. Please read the definitions enclosed (see 5. For each frequency indicated, provide a "con-
Appendix B). fidence" rating on a scale of one to three. Enter
a "I" if your experience has not dealt in this
2. Respond "yes", "no", or "don't know" to the
system and/or operational mode, i.e., if you
question, "Do you feel that the frequency of
do not feel strongly about the frequency you
pipe failures in this system is dependent on the
have indicated. A "2" would imply you feel
operational state of the plant?"
fairly confident with your response, while a
"3" would indicate you feel strongly about
3. If you responded positively to the question
your response.
above, please assign the values you feel repre-
sent the frequencies of failure during each
operational mode over a 40-year plant lifetime. 6. After entering the frequencies, respond to the
If you responded negatively or answered question on whether your estimates would
"don't know", simply enter your single fre- change if you were considering the lifetime of a
quency estimate for the specified failure condi- newly operational plant instead of one that
tion across from "Normal operation". became operational in the mid-1970s. If your
estimates would change, please indicate how in
4. For each frequency indicated, list the dominant the spaces provided.
failure cause(s) resulting in the described fail-
ure. A list of typical piping failure causes is 7. If you have other comments concerning pipe
enclosed. If you list more than one dominant breaks that may be of value to us, please indi-
cause, provide a rough quantification of its rel- cate this in the additional question/remarks
ative likelihood on a percentage basis. section.


I. If a respondent answered "no" to the question 2. If the response to the dependency question was
concerning the dependence of the failure fre- "yes" and separate rates were given for each
quencies for a system on the plant's opera- mode, the following was assumed:
tional mode, the following was assumed:
a. The responses could be summed over the
a. The sole response was a frequency reflect- modes and could be interpreted as the fre-
ing the rate of pipe breaks pooled over all quency of events per 4O-year plant lifetime.
operational modes; i.e., a rate of I x 10-4 When the rates were pooled in this manner
implied the respondent would expect into an overall frequency, the FLAG variable
I x 10-4 events in a 40-year plant lifetime. was assigned an asterisk, meaning again that
this is a calculated figure.
b. Of the rate given, 2.711lo would occur dur-
ing startup as 2.711lo of the plant's lifetime b. When no frequency was entered across
is spent in startup. from an operational mode, the expert was
implying that the rate was negligible;
c. Using the same rational as above, it is therefore, a zero for this rate was entered.
assumed that the frequency of events in a These rates are omitted from the printout
given mode is proportional to the amount of compiled responses as this situation
of time the plant is in that mode. The cov- occurred often and would, therefore, sig-
erage breakdowns in the "Operational nificantly increase the already voluminous
Modes Criteria" figure accompanying the printout. The respondents were invited to
original questionnaire (Table B-1 in enter a non-zero rate if so desired; however,
Appendix B) were used to determine these a value of I x 10- 13 was assigned to these
times for a 40-year lifetime. So if a respon- rates eventually, as the analysis is per-
dent indicated an overall response of formed on natural logarithms of the rates.
I x 10-4 , it was assumed he would expect to
see: 2.7 x 10- 6 events during the plant's c. Confidence levels and failure causes asso-
one year in startup; 6.3 x 10- 5 events dur- ciated with rates submitted for each mode
ing the plant's 25 years in normal opera- could not be carried into the overall cate-
tion; 8.0 x 10- 7 events during the plant's gory. Therefore, if a respondent said "yes,
0.32 years shutting down; 3.4 x 10- 5 events the frequency of events depends on the
during the plant's 13.7 years in shutdown; operational mode of the plant," those
and 0 events during transients, as the rates indicated were entered in each mode
amount of time in this mode is assumed category without asterisks (as they are not
negligible. calculated), but an overall rate was calcu-
lated and is flagged with an asterisk. No
When this situation occurred, i.e., when failure causes or confidence levels were
an overall frequency was given and fre- associated with this calculated overall rate.
quencies were calculated for each mode,
an asterisk (*) was assigned to the FLAG 3. When a respondent answered "don't know" to
variable. These calculated rates are used in the dependency question, he typically gave one
the analysis. response; this answer was interpreted as an
overall rate that is not split out into rates
d. Also, when an overall rate was given, the among the modes.
corresponding failure causes and confi-
dence level associated with that rate were 4. Most respondents chose to indicate likely
assigned to each mode. causes for failures. The maximum number of
causes given for a single type of event was five.
e. Note that even if an overall rate is given, Some chose to indicate the level of likelihood
different failure causes could be entered for those causes; e.g., if "cause I (2511lo) and
for the different operational modes. cause 2 (7511lo)" was a response, that expert

implied he felt causes I and 2 were dominant Figure C-I contains a compilation of the YeslNo
causes, but number 2 was three times more answers to the question concerning the dependency of
likely than cause I. If no quantification of the rates on the operational mode and the question
these causes was entered, it was assumed that concerning the change in frequency for a new genera-
they were equally likely; i.e., 50070 and 50070 tion plant. The responses are sorted by system.
would have been assigned to the above causes. Figure C-2 contains the frequencies, confidence
Note that, in the printout, the causes are levels, and failure causes submitted by the experts
columns labeled CI through C5 and the asso- for the case of pipe rupture. The responses are
ciated quantification values are in columns sorted first by system, then by operational mode,
labeled QI to Q5. and by pipe size where "small" implies 1/2 to 2 in.
pipes, "medium" indicates > 2 to 6 in. pipes, and
5. On each questionnaire form, the respondent "large" indicates> 6 in. pipes. They are sorted last
was asked if the rates for that system would by expert number. Items numbered I and 2 in the
differ for a new generation plant, i.e. one built paragraphs above explain the presence of an aster-
later that 1975. If he responded "yes", he was isk by the values entered.
asked to indicate an amount by which the rates Figure C-3 is comparable to Figure C-2 in that
would change. He was cautioned to note that the responses are sorted by system and operational
the form says the rates would either "increase mode. They are then sorted by leak rate, of which
_ 070" meaning "increase by _070", or two were given on each page of the questionnaire
"decrease to _070." forms.




EVENTS........................................................................... D-4

ADDITIONAL PLANT INFORMATION. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D-8

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D-9



Nineteen events involving pipe failures were nonnuclear sources such as fossil and chemical plants
extracted from the data sources listed in the main do not, in general, record such events. A consistent
text. The events were from those nuclear power and complete set of piping failure data, therefore, did
plants listed in Table 5 of the main text and cover not exist. Military nuclear experience records could
approximately 800 years of operating experience. not be obtained because of the classified status, and
This appendix contains a brief summary of each of reliable information on foreign pipe failure events
the events. could not be physically obtained. A draft document
All events occurred in u.s. commercial nuclear of foreign pipe failure analysis was received with the
power plants; initially, additional data were to be acknowledgment that the information contained in it
extracted from nonnuclear sources, nuclear military was incomplete; no foreign data, therefore, are
experience, and foreign nuclear experience. However, included in the study.


DATA SOURCE: LER 247/78-32 DATA SOURCE: LER 311/84-016

During normal operation, a routine inspection During normal power operations, a leak was dis-
revealed a leak emanating from a socket weld of a covered on the common suction line to the charging
drain valve to the seal injection header. The leak pumps in the vicinity of a vent valve. A crack had
rate increased from 0.5 to 8 gpm. The leak is formed in the 8-in. line and was due to fatigue,
believed to be the result of vibration induced which probably was caused by normal system
stresses. vibration of the vent valve piping.

EVENT: 2 An actual leakage rate was not given in the report

PLANT: INDIAN POINT 2 and is assumed to have been greater than I gpm.
During normal operation, local radiation alarms CONTROL
and visual inspection confirmed a weld leak at a DATA SOURCE: NPE (PWR-2, VIII-A-308)
90 degree, 2-in. socket weld connection in the seal LER 344/77-038
injection line. The event apparently was caused by
excessive vibration in the system. During a routine inspection, leakage was found
from a 4-in. dia common suction piping for a
No leakage rate was given in the report but it was charging pump. The leak rate was approximately
assumed to be greater than 1 gpm due to the loca- 11 gpm. The cause of the pipe cracking was due to
tion and the alarming of the radiation detection a "T-bar" piping support welded to the 4-in. pip-
system. ing, which caused local raising of stress levels and
vibration induced fatigue.
(PWR-2 VIII-A-538)
The plant was returning to power from a mainte-
It was determined that a possible reactor coolant nance outage. The plant was at approximately 7lJlo
leak existed. Investigation revealed that the leak power. A high level occurred in No. 23 steam gener-
was in the chemical and volume control system at a ator resulting in a turbine trip. All steam generator
weld for a vent valve. The combination of excessive levels began decreasing. Auxiliary feed water system
vibration and a pressure transient caused the weld operation restored proper steam generator levels,
to fail. It is estimated that 3000 gal leaked from the except for steam generator No. 22. During the
system; therefore, for this event, it was assumed the event, shaking and loud noises were heard by the
leak rate was in excess of 15 gpm. feedwater line to No. 22 steam generator. The plant

was placed in cold shutdown and a containment crack was in the 10 in. discharge line downstream
entry was made. A visual inspection of the carbon of decay heat cooler E35B. The probable cause of
steel, 18-in. dia line inside containment indicated a the crack was excessive vibration.
fracture adjacent to a filled weld between the feed-
water line and the end plate that is welded into the EVENT: 9
penetration sleeve. The fracture extended PLANT: McGUIRE 2
180 degrees around the pipe. The specific cause of SYSTEM: RESIDUAL HEAT REMOVAL
the event was not found, but three possible mecha- DATA SOURCE: LER 370/84-017
nisms were identified:
The unit was in cold shutdown at the time of the
• Water hammer caused by sudden closure event and the operators were in the process of veri-
of the feed water valve fying the residual heat removal valve lineup before
• Water slug loading caused by check valve a reactor fill. Leakage at a socket weld in the resid-
malfunction, which results in feedline ual heat removal letdown line was discovered. The
back flow from the steam generator and pipe was completely separated and the total leakage
subsequent quench of steam voids was estimated at between 3000 and 7000 gal.
• Water slug loading caused by water-steam
reaction in the horizontal feedwater line The cause of the leakage was reported to be water
adjacent to steam generator No. 22, which hammer due to air being left in the line during a
followed the uncovering of the steam gen- previous work activity.
erator feed ring and partial drainage of the
connected feed piping. EVENT: 10
For this study, the leak rate is assumed to be greater SYSTEM: HIGH PRESSURE
PLANT: MAINE YANKEE During normal power operations, testing of the
SYSTEM: MAIN FEEDWATER high pressure coolant injection return line was in
DATA SOURCE: NPE (PWR-2 VI-E-435), LER progress; it was discovered that contaminated
309/83-003 demineralized water was leaking from the under-
ground pipe into the storm sewer. The return line
The unit tripped from full load and main feedwater was cut and an internal inspection was conducted
was not available; so steam generator restoration was using a TV camera. A poorly made weld, which
accomplished by auto start operation of the auxiliary was located just outside the building, apparently
feed pumps. Approximately 15 min later, a loud noise resulted in the leakage. The approximate leak rate
was heard in the shop located beneath the main feed was 5 gpm.
lines. A containment fire detector alarmed and con-
tainment humidity began to rise. A containment entry EVENT: 11
was made and a feed line was found to be leaking PLANT: DRESDEN 1
severely near the No. 2 steam generator inlet nozzle. SYSTEM: HIGH PRESSURE
The leak occurred as a result of water hammer event COOLANT INJECTION
causing the ultimate failure of what was probably an DATA SOURCE: NPE (BWR-2 VII-E-9)
existing crack in the feed pipe.
A high pressure coolant injection return line to the
EVENT: 8 condensate storage tank ruptured during a test,
PLANT: ARKANSAS NUCLEAR allowing some of the condensate storage tank waste
ONE-l to flow into a storm sewer. The cause of the rupture
SYSTEM: RESIDUAL HEAT of the 18 in. line was improper valve lineup leaving
REMOVAL a manual valve in the line closed. The event is
DATA SOURCE: NPE (PWR-2 VIII-B-96) assumed to be caused by operator error.

A leak of 2 gpm was discovered in the heat affected It is estimated that approximately 15,000 gal
zone of the boss to line weld for DH-1019. The leaked from the system.

Following a reactor scram, the supply line to the Misc-204
condensate ring header in the torus room failed at a
welded joint at the junction of a 20 in. and 24 in. Very little information is available for this event.
pipe. Approximately 80,000 gal of condensate The occurrence is a pipe split due to a condensate
leaked from the system. The apparent cause of the pump seizure. It is assumed that the leak rate would
event was weld fatigue caused by line movement have exceeded 15 gpm.
during repetitive operations of the high pressure
coolant injection system. EVENT: 16
SYSTEM: CONDENSATE During a startup, a transfer from the low flow feed
DATA SOURCE: NPE (BWR-2 VI-E-157) regulating valve to the main feed regulating valve
was in progress when the feedwater vibration alarm
A 1/2 in. pipe connecting the pressure gage to a sounded. The unit was scrammed due to indication
condensate transfer pump had sheared off at the of the low flow line starting to sever.
threaded joint just before the 1/2 in. isolation
valve. The leakage is estimated to have occurred Subsequent inspections revealed additional cracks
in the low flow piping at the low flow riser junction
over a 1.25-hour period with loss of a maximum of
to the main feedwater line and in the reducer
9700 gal. The apparent cause of the failure was
upstream of the regulating valve. The equipment
fatigue of the pipe due to pump vibration caused by
failures were due to vibration at the feedwater regu-
imbalance in the rotating assembly.
lating station during the transfer. The estimate of
water released in the plant from the damaged line
EVENT: 14 was approximately 8500 gal.
SYSTEM: CONDENSATE For this study, the leak rate is assumed to be greater
DATA SOURCE: NPE (BWR-2 VI-E-33) than 15 gpm.

A condensate booster pump vent line was ruptured PLANT: QUAD CITIES 2
filling the condensate pump room to approximately SYSTEM: MAIN FEEDWATER
20 in. of water. The reactor was scrammed and DATA SOURCE: NPE (BWR-2 VI-E-45)
after the condensate system was shut down, the
pump was isolated. The cause was failure of the While decreasing load during an orderly unit shut-
pipe nipple coming out of the casing, which down, a feedwater vibration alarm was received.
resulted from a reduction of wall thickness coupled Due to an increase in reactor water level, the turbine
with vibration of the condensate feed water system. tripped causing a reactor scram. The piping failures
were due to high vibration.
Due to the 20 in. of water collecting in the conden-
sate pump room, it is estimated that the leak rate No estimate of leakage from the system was given
was in excess of 15 gpm. but is assumed to be greater than 15 gpm due to the

necessity of a turbine trip and a reactor scram and pipe reducers. It was estimated that less that
caused by decreasing reactor water level. 4000 gal were lost.

(Vol. 16, No.2, March-April, The unit was shut down for a refueling outage.
1975) Leakage was observed from an underground 16 in.
service water line to the residual heat removal sys-
The plant was at approximately 200 MW(e) and tem and estimated to be 50 gpm. The pipe was
increasing power. The feedwater vibration alarm inspected by a TV camera and drawings were
actuated and after shutdown, the operator verified reviewed. It was surmised that the pipe had been
that the feed water bypass line was vibrating. pierced by temporary construction supports that
Cracks were found in the 4 in. valve bypass valve were not removed during backfill.


The overall goal of this study is to present rates of drawings were provided from which the welds and
pipe failures in commercial nuclear power plants. lengths of pipe were counted; in other cases, com-
The numerator of such a rate, of course, is repre- pleted tallies of the population were received. Both
sented by the frequency of pipe breaks under a shop welds and field welds are included but are not
given set of conditions. The denominator, however, differentiated.
is not as obvious. Four nuclear steam supply system (NSSS) ven-
In the initial stages of the study, the project plan dors are represented in this population data. Infor-
included developing rates on a per weld and per mation is comprised from seven Westinghouse, two
pipe length basis. Population information from Babcock and Wilcox, four Combustion Engineer-
13 plants was, therefore, collected on this basis. In ing, and five General Electric plants; that is, thir-
the course of the study, however, it became clear teen PWRs and five BWRs are represented.
that knowing the total population of welds in a The data are categorized by pipe sizes of 2 in.,
plant was not as crucial as knowing attributes of >2 to 6 in., and >6 in. in diameter. Data on
welds, such as their location. Unfortunately, it was < 2 in. pipes were not included, simply because the
amount of < 2 in. piping, especially instrumenta-
not feasible for the study to differentiate weld
tion lines, prohibited any data collection effort.
attributes. In most cases, when a pipe break
Tables 6 through II of the main report, there-
occurred at a weld or weld repair, it was impossible
fore, contain weld and pipe length population cate-
to determine the type and location of the weld that
gorized by plant type, NSSS vendor, system, and
failed. Secondly, the populations for both welds
pipe size. For a given combination of these catego-
and lengths of pipe are extremely variable so that ries, the tables present the average number of welds
rates required on a per weld or per length basis (or length of pipe), the number of plants over
should be estimated on plant-specific terms. In which the welds and lengths are averaged (N), and
spite of these drawbacks, however, population the standard deviation of the number of welds or
information on pipe lengths and weld counts, lengths of pipes.
including indications of their variability, is summa- The values are totaled over pipe size; in some
rized and displayed in this appendix on a system- cases, however, the plants submitting information
by-system basis for those interested in calculating on 2 in. pipes were not the same plants as those
rates on a weld or pipe length basis. submitting data on > 6 in. pipes. The plants repre-
The pipe length and weld population informa- sented in the TOTAL column are only those plants
tion in Tables 6 through 11 of the main report was that either (a) supplied data for all three size cate-
obtained from an architectural engineering firm gories, or (b) supplied a figure representing total
and nuclear utility owners. In some cases, isometric piping not broken down by pipe size.


D-l. L. J. Bain, Statistical Analysis ofReliability and Life-Jesting Models, New York: Marcel Dekker, Inc.,
p. 157.

D-2. NPRDS 1978 Annual Reports of Cumulative System and Component Reliability, NUREG/CR-0942,
Southwest Research Institute, September 1979.

D-3. G. E. P. Box and G. C. Tiao, Bayesian Inference in Statistical Analysis, Reading, MA: Addison-
Wesley, 1973.

D-4. N. R. Mann, R. E. Shafer, N. D. Singpurwalla, Methods for Statistical Analysis of Reliability and
Life Data, New York: John Wiley and Sons, Inc., 1974.

D-5. N. L. Johnson and S. Kotz, Discrete Distributions, New York: John Wiley and Sons, Inc., 1969,
pp. 58-59 and 96.






Intent E-4

Procedure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E-4

Results for Westinghouse Plants East of the Rocky Mountains E-4

Results for Combustion Engineering PWR Plants E-6

Summary and Cautions E-6

STEAM GENERATOR TUBE INTEGRITY. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E-8

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E-9



Three studies addressing pipe failures induced by concerns a document discussing steam generator
crack growth in reactor coolant loops are summa- degradation in PWRs and the program designed to
rized first in this appendix. The summary includes resolve these concerns. £-4
an overview of two documents prepared by An additional report, Steam Generator Tube
Lawrence Livermore National Laboratory Experience, £-5 although not summarized in this
(LLNL), £-1, £-2 and a mention of Reference £-3, Appendix, provides an overview of tube degrada-
prepared by £PRI. In these studies, probabilistic tion problems in PWR steam generators. Particu-
calculations were made of pressurized water lar emphasis is placed on recent operating
reactor (PWR) primary system piping (coolant experience and short- and long-term corrective
loops only) leakage and double-ended guillotine actions being pursued by the industry to resolve
breaks stemming from weld cracks. the degradation problems. Steam generator
Although steam generator tubes are passive com- inspection and repair requirements that have been
ponents and are indeed pipes, they have been established by the Nuclear Regulation Commis-
excluded from the present study simply because sion to ensure the continued safe operation of
numerous other concluded studies have addressed PWR steam generators are discussed in this docu-
the reliability of these tubes. The second review ment.


Intent ity during the lifetime of the plant. The probabilistic

theory accommodated the random nature of events
and parameters considered in the studies. This analyti-
References E-I and E-2 concern the probabilis-
cal process was divided into two parts. The first
tic calculation of PWR primary system piping
involves the calculation of a conditional leak or DEGB
(coolant loops only) leakage and double-ended
probability at individual weld joints, given that a crack
guillotine break (DEGB) frequencies stemming
exists at that joint and a seismic event of specific inten-
from weld cracks. These calculations consider the
sity occurs at the site at a specific time. The second
effects of inspection and various thermal and seis-
part, system failure probability analysis, is related to
mic loadings on the piping system failure probabili-
the estimation of a leak or DEGB probability for the
ties throughout an assumed 40-year plant design
entire reactor coolant loop piping system, taking into
life. The references then use these calculations to
consideration all of the associated weld joints.
demonstrate that leakage and DEGB, with or with-
In the first part of the analysis, see Figure E-I, a
out associated seismic events, are extremely
Monte Carlo simulation technique was used. The
unlikely events; therefore, the system design
simulation starts with the random selection of
requirements associated with these events, particu-
crack sizes from a distribution of crack sizes. Frac-
larly DEGB, may be relaxed or eliminated without
ture mechanics theory was then applied to calculate
compromising plant safety.
the growth of these cracks under normal and
abnormal loading conditions, including earth-
Procedure quake load, and to determine if pipe fracture
(either leak or DEGB) would occur as the cracks
The reactor coolant loop (RCL) piping failure grow during the lifetime of the plant. Various plant
probabilities were calculated by considering three activities related to crack and leak detections, such
scenarios of interest. The thrust of the studies was as preservice inspection, hydrostatic proof test, and
to assess earthquake contributions to RCL piping leak detection, are simulated. The seismic hazard
failure probabilities. Therefore, the failure (a leak information related to the earthquake intensity and
or DEGB) probabilities were separated into the fol- the occurrence probability was folded into the sec-
lowing scenarios related to RCL piping failure: ond part of the analysis.
Two types of analyses were performed: a best-
I . An RCL piping failure occurs, but no earth- estimate analysis and an uncertainty analysis. The
quake occurs during the plant lifetime former considers only the best-estimate models of
2. An RCL piping failure occurs before the relevant parameters; the latter takes into account
first earthquake during the plant lifetime the uncertainty of the models.
3. An RCL piping failure and the first earth-
quake during the plant lifetime occur
Results for Westinghouse Plants
East of the Rocky Mountains
The sum of the probabilities for these three sce-
narios give the probability for the failure of the Reference E-I examined Westinghouse PWR
RCL piping during a plant's 40-year lifetime. plants east of the Rocky Mountains. Seventeen
A possible fourth event is one or more earthquakes sample plants, representing thirty-three reactors at
during plant lifetime and a pipe failure after the first nineteen sites, were selected to represent the larger
earthquake. This event was of little interest because the population of Westinghouse plants in the region
plant would have been shut down for complete inspec- under consideration.
tion and repair after an earthquake, by which time the The Monte Carlo simulation for single weld and pip-
plant condition would be altered. ing system failure probability analyses were performed
The studies assume that failures can occur only in for each of the sample plants. The best-estimate results
welds where a crack exists at the beginning of plant life. for piping leakage indicated that over the 40-year plant
A probabilistic fracture mechanics approach was used lifetime, Scenarios I and 2 yielded about the same
to estimate crack growth and to assess the crack stabil- cumulative failure probabilities ranging from "v2 x 10- 7

Probability that one crack exists at a weld
• weld volume

Monte Carlo simulation process

• N samples
• stratified sampling scheme

Random selection of a crack

• initial crack size distribution

Pre-service inspection and hydrostatic proof test

• crack detection probability
• hydrostatic proof pressure

Crack growth
• crack growth characteristics
• loading events
• loadings and their frequencies

In-service inspection
• crack detection probability

Failure criteria
• leak
- leak detection threshold
• break

Continuation of Monte Carlo simulation

process until completion of N samples

Probability of failure at a given weld

as a function of time

Figure E-l. Computation flow chart for estimating the failure probability at a given weld.

to '\,3 X 10-6 • Scenario 3, where the leakage and the culations for the CE plants is basically the same as
earthquake occur simultaneously, had leak probabili- outlined in the previous section, only the results
ties several orders of magnitude lower, on the order of will be given.
10- 12 to 10-9• The contributions from the three scenarios The overall best-estimate probability of failure
were then combined to produce the estimated cumula- for leakage in CE plants has a range of 1.5 x 10- 8 to
tive probability for a leak at each of the sample plants. 2.3 X 10- 8 per plant year. Similarly, the overall
This yielded failure probabilities ranging from 5 x 10-7 DEGB failure rate has a range of 5.49 x 10- 14 to
to 6 X 10-6 over the 4O-year lifetime. 4.53 X 10- 13 per plant year. This latter range is
The sample plant leak probabilities were aver- larger than for leakage, but as the values are so low,
aged to produce yearly rates and used to construct the range is not considered significant. The CE
an empirical cumulative probability distribution. study did not construct cumulative distribution
Examination of this distribution indicates that the curves. Therefore, 99th-percentile values are not
99th-percentile value of yearly failure probability is available for the best estimate.
1.6 x 10- 7 • This can be inferred to mean that 990/0 The uncertainty analysis to establish an absolute
of the Westinghouse plants east of the Rocky upper bound on the failure probabilities was per-
Mountains have a less than 1.6 x 10- 7 per plant year formed, as in the previous section, using a Latin hyper-
probability that an RCL piping leak will occur. cube approach. An analysis was performed for each of
The best-estimate DEGB failure probabilities the System 80 plants and the composite. An approxi-
were calculated using the same procedure as for mate upper bound for leakage probability in all CE
leakage. The results indicate that, as with leakage, plants was determined to be 2.0 x 10-7 per plant year.
Scenarios I and 2, being about equal, are the Similarly, the upper bound for RCL failure probability
prime contributors to the overall RCL failure prob- due to DEGB was determined to be approximately
ability. Scenario 3 is lower by several orders of 10- 10 per plant year.
magnitude. For the 17 sample plants, the failure
estimates range from '\,4 x 10- 11 to 2.5 X 10- 10 over
the assumed 40-year plant life. The cumulative Summary and Cautions
probability distribution constructed using the sam-
ple plant data indicated a 99th-percentile probabil- Comparison of the two studies indicate results are
ity of a DEGB of 6.8 x 10- 12 per plant year. very similar. The RCL leak failure probability upper
An error analysis was performed for two sample bound is on the order of 2.0 x 10-7 per plant year; the
plants using a Latin hypercube sampling technique. DEGB failure probability upper bound is on the order
The sampling technique was used to establish bound- of 10- 10 per plant year. The best-estimate values for the
ing values for the analysis parameters that were consid- CE plants tend to be about an order of magnitude
ered to contain uncertainties. The results of the lower than those obtained for the Westinghouse plants.
sampling were then used in the Monte Carlo simula- Because the reports demonstrate that seismic events
tion process to establish the upper uncertainty bounds contribute very little to system failure, this difference in
for the leakage and DEGB failure rates. The results of the best estimate is probably due to the ferritic steel
this error analysis indicate that, with 90% confidence, used in CE plants vs the stainless steel used in the
the RCL piping leak and DEGB rates for all Wes- Westinghouse plants.
tinghouse plants east of the Rocky Mountains are The results and conclusions presented in the two
bounded by and lower than 2.4 x 10-7 and 7.5 x 10- 11 documents must be taken in the proper perspective.
per plant year, respectively. The following caveats are to be used in interpreting
the LLNL results:
Results for Combustion 1. The studies concern pipe failures caused by
Engineering PWR Plants crack growth in the pipe weld joint locations.
No external causes such as failure of support
Reference E-2 examined Combustion Engineer- systems are considered. The crack is assumed
ing (CE) PWR plants. The results for the CE plants to be present in the weld at the time of con-
were reported as a composite for earlier model east- struction. Crack formation later due to stress
ern plants, and individually for the later model so- fatigue, etc., is not considered.
called "System 80" plants. Seismic considerations 2. The results for the Westinghouse plants
did not allow a composite for the System 80 plants. apply only to those east of the Rocky
As the methodology for the RCL piping failure cal- Mountains. The differing seismic hazard

characteristics of the west were not consid- 3. Reference E-3 is an EPRI report addressing
ered in this report. However, due to the observed piping failure. Some of the informa-
conclusion of the CE study that seismic tion presented in the report is salient to inter-
events contribute very little to the overall preting the LLNL results. This study
failure probability, it is judged that the concludes that for piping failures in general
results for the Westinghouse plants in the (other than just coolant loop piping),
west would not be significantly different (a) 54UJo of the failures occur in the welds or
from that reported for the east. Another weld heated zones, (b) 40% occurred in the
LLNL report, not available at this time, pipe wall (base meta!), and (c) 6UJo occurred
addresses this issue. in the threaded pipe joints.


This Section presents the steam generator tube tor year. This corresponds to the 50070 confidence value
rupture failure rates from NUREG-0844, "NRC for the upper bound point-estimate for an event that
Integrated Program for the Resolution of Unre- has not yet been experienced (i.e., no multiple SGTRs
solved Safety Issues A-3, A-4, and A-5 Regarding for 353 reactor years of operation for PWRs). The
Steam Generator Tube Integrity." probability of many tube ruptures occurring simulta-
For cases of single steam generator tube rupture neously is extremely remote. For this analysis, it was
(SGTR), an initiating event probability of 2 x 10-2 per assumed that the probability of 10 or more tube fail-
reactor year is used. This is based on actual operating ures is 2 x 1<J4 per reactor year.
experience (i.e., four tube ruptures in Westinghouse For the case of multiple tube ruptures, the prob-
plants in 240 reactor years of operation for domestic ability that they may occur in more than one steam
Westinghouse plants; sufficient operating experience is generator is less that 1.0 because degradation will
not available for Combustion Engineering and often be more advanced or more widespread in
Babcock & Wilcox plants to justify a smaller value for one steam generator than in the other steam gener-
those plants). It is highly improbable that two or more ators. It was assumed that a probability of 0.5
tubes could rupture simultaneously as an initiating multiple ruptures will involve more than one
event during normal steady-state operations. However, steam generator as a reasonable upper bound esti-
multiple tube ruptures are credible during transient mate.
conditions when the loadings on the tubes become Note that the estimated probability of single
more severe over a short period of time. For purposes SGTR events is based on actual operating experi-
of this discussion, single and multiple tube ruptures are ence, whereas the assumed probabilities for multi-
considered to be "initiating events" in cases where the ple tube ruptures in one or more steam generators
actual "initiating" transient does not involve a loss of are estimates of upper bound probabilities that are
primary or secondary system integrity. Tube rupture believed to be very conservative. Although conserv-
occurrences as a consequence of a loss of primary or ative, these estimates do not lead to high estimates
secondary system integrity are discussed below. of public risk from SGTR related causes, and the
The probability of a multiple tube rupture as an use of these estimates is, therefore, considered
"initiating event" was assumed to be 2 x 10-3 per reac- appropriate for this analysis.


E-l. H. H. Woo et aL, Probability of Pipe Failure in the Reactor Coolant Loops of Westinghouse PWR
Plants, Volume 2: Pipe Failure Induced by Crack Growth, NUREG/CR-3660, August 1984.

E-2. T. Lo et aL, Probability ofPipe Failure in the Reactor Coolant Loops ofCombustion Engineering PWR
Plants, Volume 2: Pipe Failure Induced by Crack Growth, NUREG/CR-3663, January 1984.

E-3. Characteristics of Pipe System Failures in Light Water Reactors, EPRI NP-438, August 1977.

E-4. Division of Licensing, NRC Integrated Program for the Resolution of Unresolved Safety Issues A-3,
A-4, and A-5 Regarding Steam Generator Tube Integrity, NUREG-0844, July 1983.

E-5. C. y. Cheng, Steam Generator Tube Experience, NUREG-0886, February 1982.




METHODS USED .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F-3

REFERENCES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F-5




This appendix shows the methods used to esti-

:5 "(transient)
mate the pipe failure rates using the observed fail-
ure data and the operating experience. F- I
The following well-known statistical methods were
used for Type-I censored data with replacement. F-2 XLI2(2N + 2)
:5 (F-4)
The general methods for estimating rates on a yearly 2D
basis and on a transient basis are where

X~(b) the chi-square variate at cumula-

A(yearly) (F-I) tive probability "a", with "b"
T degrees of freedom.

In these equations, a is the fraction left out of the

N intervals. For example, with 90070 confidence lim-
A(transient) (F-2) its, a is 0.10, a/2 is 0.05, and upper limit uses the
95th percentile.
If D, the number of transients, is small, then the
where Poisson approximation of the binomial distribu-
tion is not adequate, and 100(1 - a)07o confidence
limits for the transient failure rate are
" estimated failure rate

N number of reported pipe failures :5 (transient)

D -N + I + NFL
T total number of operating years expe-
rienced (N + l)Fu
D - N + (N + I)F u
D total number of transients.

Confidence limits for yearly failure rates were based where

on the assumption that the underlying pipe failure time
distributions are exponential and, therefore, that the Fa/2 (2N, 2D - 2N + 2)
resulting data can be represented by a Poisson process.
In transient evaluations, N is assumed to be binomially F)_al2 (2N + 2, 2D - 2N)
distributed. However, because the probability of failure
is small, the Poisson distribution may be used to F variate at cumulative probabil-
approximate this variable for cases where the number ity "a", with "b" and "c" degrees
of transients is large. The generalized formulas for esti- of freedom.
mating 100(1 - a)07o confidence limits on the failure
rates are As before, for 90070 confidence limits, the 0.05 and
0.95 quantities are used (a = 0.10).
In this study, yearly rate confidence limits were
X;.PN) :5 ,,(yearly) :5
XLI2(2N + 2)
always based on Equation (F-3). Transient rate
2T 2T confidence limits were based on Equation (F-4) if
D - N ~ 100, and on Equation (F-5) otherwise.
The lower limits in Equations (F-3), (F-4), and (F-5)
and are not defined in cases where no failures are observed

(N = 0). Zero is the appropriate lower limit in these The estimates of Equations (F-6) through (F-8) can
cases. However, Equations (F-l) and (F-2) also give be obtained in two ways. The first is to consider shrink-
zero as the point estimate when N = O. More realistic ing the confidence intervals of Equations (F-3), (F-4),
point estimates for such cases are and (F-5) to the case where a = 1.00 and both a/2
and I - a/2are 0.5. Because of the differing degrees of
freedom, the intervals do not shrink to a single point.
X6.50(2N + I)
A= (F-6) The equations use an average for the differing degrees
2T of freedom. Because the estimates use 50th percentiles,
they are related to medians.
2N + I) The second way of considering Equations (F-6)
A = (F-7)
20 through (F-8) uses the medians directly. In a
Bayesian context, A is regarded as a random varia-
and ble. With Poisson sampling and a noninformative
conjugate prior distribution, the posterior distribu-
(2N + I)F M tion for the occurrence rate has a gamma distribu-
A= (F-8)
20 - 2N + I + (2N + l)F M tion with parameters F-4

(a,m = (N + 1/2, liT). (F-9)
FM (2N + 1,20 - 2N + I) and the F
and X 2 distribution percentile and
degree of freedom notations are as
defined above. Because the gamma distribution with parameters
(N,2) is identical to the chi-square distribution with
Equation (F-6) applies for yearly rates and is 2N degrees of freedom, F-5 Equation (F-6) can be
used with the upper bound from Equation (F-3). shown to be the median of the distribution
described by Equation (F-9). Using a similar rela-
For transients, Equation (F-7) is used with the
tion between (3 and F distributions, Equation (F-8)
upper bound from Equation (F-4) and
can be derived as the median of the posterior failure
Equation (F-8) is used with Equation (F-5). Equa- rate distribution obtained in sampling from a bino-
tions (F-6) and (F-7) are applicable to events occur- mial distribution with a noninformative conjugate
ring according to a Poisson distribution regardless prior distribution.
of the number of failures observed. A similar com- In summary, Equations (F-6), (F-7), and (F-8)
ment applies to Equation (F-8) and the binomial describe median-point estimates for the failure
distribution. Typical estimates from Equation (F-6) rate. They can be used when N = 0, and are more
are in the following table for comparison with conservative in that case than the point estimates
Equation (F-1). Equation (F-6) has been used in given in Equations (F-l) and (F-2). In this work,
other failure data studies, such as Reference F-3. they are used with the upper confidence limits in
Equations (F-3), (F-4), and (F-5), respectively,
N A whenever no failures are observed.
In estimating the above confidence limits, all
30 30.15fT components in the sample were assumed to have
20 20.15fT exactly the same true failure rate. No effort was
10 1O.15fT made to account for possible variations arising
5 5.15fT from the mixture of populations having different
2 2.18fT true failure rates. For further discussion of the
1 1.19fT assumptions and limitations of these confidence
0 0.23fT limits, see References F-2 through F-6.


F-I. S. R. Brown, M. Trojovsky, Data Summaries ofLicensee Event Reports ofInverters at U.S. Commer-
cial Nuclear Power Plants January 1, 1976 to December 31, 1982, NUREG/CR-3867, Idaho National
Engineering Laboratory, August 1984.

F-2. L. 1. Bain, Statistical Analysis of Reliability and Life-Testing Models, New York: Marcel Dekker, Inc.,
p. 157.

F-3. NPRDS 1978 Annual Reports of Cumulative System and Component Reliability, NUREG/CR-0942,
Southwest Research Institute, September 1979.

F-4. G. E. P. Box and G. C. Tiao, Bayesian Inference in Statistical Analysis, Reading, MA: Addison-
Wesley, 1973.

F-5. N. R. Mann, R. E. Shafer, N. D. Singpurwalla, Methodsfor Statistical Analysis ofReliability and Life
Data, New York: John Wiley and Sons, Inc., 1974.

F-6. N. L. Johnson and S. Kotz, Discrete Distributions, New York: John Wiley and Sons, Inc., 1969,
pp. 58-59 and 96.