AUTOMATIC

By Edward K. Budnick, P.E.

SPRINKLER SYSTEM
RELIABILITY
INTRODUCTION

W hen automatic fire sprinkler

systems, or any fire protec-
tion safety features, are
included in a fire protection design
package, it is assumed that, if needed,
they will perform as expected. One
measure of expected performance is a
system’s reliability. Reliability is gener-
ally reported as probabilistic, i.e., the a high level of operational reliability), changing ITM frequency on those esti-
percent success rate for a given num- and the designer (in performing proba- mates. These methods can be helpful
ber of systems over a fixed time peri- bilistic-based risk analyses for new in evaluating the reliability of newer
od. Put more simply, the reliability design projects). Accurate estimates of sprinkler technologies with relatively
(i.e., probability of success) is reliability are necessary inputs to risk- short field experience.
expressed as based analyses where failure rates and
redundancy considerations must be RELIABILITY CONCEPTS
number of successes
P ( success) = (1) evaluated.
total number of incidents
This article provides a brief summa- A detailed discussion of reliability
Several published studies provide ry of published sprinkler reliability engineering is outside the scope of this
reliability estimates for automatic sprin- studies. An attempt is made to address article. Modarres1 provides a more
kler systems. These estimates indicate uncertainties in the reported failure complete review of the subject. Brief
that, historically, sprinkler systems have rates in a systematic manner. Rather descriptions of selected elements of
been highly reliable. However, there than report the results as a single esti- reliability are listed below to orient the
are biases and limitations in the data mate of sprinkler reliability, statistical reader to the value and limitations
and the analyses that restrict their use- methods are used to average the indi- associated with published data on
fulness. Most of the published studies vidual estimates within specified confi- automatic sprinkler system reliability.
do not attempt to address these limita- dence limits. The value of this simple Reliability is normally defined as an
tions or any other elements of uncer- analysis lies in the limit estimates. estimate of the probability that a sys-
tainty in the estimates. Also, most of While calculating the “average” value tem or component will function as
the studies are more than ten years old among several available data sources designed over a designated time peri-
and, therefore, do not include more does not in itself improve the quality od. There are two components to
current sprinkler technologies in the of the estimate, the simple calculation overall reliability. Operational reliabil-
databases. of the range of possible estimates at ity is a measure of the probability that
Simple statistical methods are avail- least allows the user to perform limited a system or component will operate as
able to more accurately estimate auto- sensitivity analyses. Such analyses pro- intended when called upon. It is
matic sprinkler system reliability and vide input to risk-based assessments directly affected by the types and fre-
the uncertainty associated with such for existing or proposed fire safety quency of testing and maintenance
estimates. Some of these methods can designs. performed on the system.
handle “sparse” databases, i.e., small An illustration of the use of selected Performance reliability (i.e., capabili-
data sets from a few systems or from a statistical methods to evaluate reliabili- ty) is a measure of the adequacy of
single system over a relatively short ty for small or “sparse” data sets is also the system, once it has operated, to
period of time. These methods and the provided. As an example, limited ITM successfully perform its intended func-
resulting predictions are of value to data for several existing sprinkler sys- tion. For a sprinkler system, opera-
manufacturers (in developing new tems are analyzed. The results demon- tional reliability accounts for the
sprinkler technologies and identifying strate the potential usefulness of small “readiness” of the system components,
possible failure modes), the user (in or limited data sets in estimating the while performance reliability address-
optimizing Inspection, Testing, and reliability of a specific automatic sprin- es the “capability” of the system to
Maintenance (ITM) costs and insuring kler system as well as the effects of perform satisfactorily under specific

W INTER 2001 Fire Protection Engineering 7

 fire exposures. The capability of the exists among the various studies. The nology, it may not be appropriate to
system is related to the scope and ade- reported reliability estimates range rely on these estimates to evaluate the
quacy of the engineering design stan- from 81.3 percent to 99.5 percent. reliability of newer technologies such
dards (e.g., NFPA 13) and the level of These differences may be attributable as quick response, residential, and
compliance of the system and its com- to any number of variations in the pro- ESFR sprinklers without addressing
ponents with the standards. tocols or the databases used by each additional factors.
Two other important concepts are study. For example, the relatively low
failed-safe and failed-dangerous. value of 81.3 percent2 as well as the LIMITED UNCERTAINTY ANALYSIS
When a sprinkler system fails safe, it somewhat higher value of 87.6 percent
operates when no fire event has reported by Kook3 appear to reflect Estimates of reliability are required
occurred. An accidental discharge of a biases in the databases. In both stud- input to fire risk assessments. The
sprinkler is an example of a failed-safe ies, the number of incidents was rela- applicability and accuracy of such esti-
condition. A failed-dangerous condi- tively small. And, while most of the mates are perpetuated through the risk
tion occurs when a system does not suppression systems in the databases analysis and directly reflected in the
function when needed, e.g., a sprin- were sprinkler systems, apparently performance outcomes. The estimates
kler fails to open, or the water supply other types of suppression systems compiled in Table 1 demonstrate vari-
is unavailable. were also included. In addition, the ability in sprinkler system reliability
Studies that rely on fire incident high-end estimates of 99.5 percent among different studies. Unless the
data to estimate automatic sprinkler reported by Maybee4 and Marryat5 parameters of a particular study match
system reliability mix both operational reflect sprinkler system performance in those of interest, reliance on the esti-
and performance reliability elements. occupancies where inspection, testing, mate of reliability from a single study
They also typically do not include and maintenance were rigorous and can incorrectly alter the results of a
failed-safe incidents in the analysis. On exceeded customary requirements for risk assessment.
the other hand, studies that rely on ITM activities. If these studies were Relatively simple statistical methods
testing and maintenance data are, for excluded from the group, the range of are employed here to provide both
the most part, providing estimates of reliability estimates for the remaining “estimates” of sprinkler system reliabil-
operational reliability. studies is from 86 percent to 97.9 per- ity and measures of “uncertainty” asso-
cent, which still represents a significant ciated with the reported estimates.
PUBLISHED STUDIES range. Uncertainty is reported as confidence
An additional limitation in the intervals, i.e., the upper and lower
Several studies have been published reported sprinkler reliability estimates bounds associated with the reliability
that report estimates of automatic sprin- is that most of the sprinkler systems estimate.
kler system “reliability.” For the most were more than 15 years old. The estimates of reliability are sim-
part, these studies provide estimates Therefore, while the reliability esti- ple calculations of the “mean” of the
based on review of actual fire incidents mates provide reasonable information values reported in Table 1. The confi-
where automatic sprinklers were pre- for conventional spray sprinkler tech- dence intervals are calculated based
sent. As a group, they vary significantly
in terms of reporting periods, the types
of occupancies, and the level of detail TABLE 1. Selected Automatic Sprinkler Reliability Studies (percent)
regarding the types of fire incidents and
sprinkler system design. Nevertheless, Reliability Value
such studies are routinely referenced Occupancy Reference (of success)
and provide some basis for estimating
sprinkler system reliability.
Milne6 96.6/97.6/89.2
Table 1 provides a summary of the NFPA7 90.8-98.2
reported reliability estimates. The three Miller7 86
occupancy categories reflect occupancy Commercial Maybee4 99.5
type variations in the reported esti- Kook3 87.6
mates. Several studies provided reliabil-
Taylor2 81.3
ity estimates for “commercial” occupan-
cies. These are grouped accordingly in Linder9 96
the table. The estimates grouped under Miller8 95.8
the “general” occupancy category were
Miller8 94.8
from studies that grouped commercial,
residential, and institutional occupan- General Powers10 96.2
cies into a single database. Richardson11 96
The estimates indicate relatively high Finucane et al.12 96.9-97.9
reliability for automatic fire sprinkler Marryat5 99.5
systems. However, significant variation

8 Fire Protection Engineering N UMBER 9

 Table 2. Reliability Estimates for Sprinkler Systems
on the degree of certainty required.
There are many factors that are
Commercial General Combined
involved in selecting a degree of cer-
tainty. In its simplest form, it is a mea- Lower confidence limit (95%) 88.1 93.9 92.2
sure of the likelihood that the actual
Mean (%) 93.1 96.0 94.6
mean value falls within the confidence
intervals. Assuming normal distribu- Upper confidence limit (95%) 98.1 98.1 97.1
tion, the higher the accuracy desired, Number of referenced studies 9 7 16
the wider the confidence intervals
(and the higher the required certainty). Note: Combined = Commercial and General
Mean reliability estimates and 95
percent confidence limits were calcu-
lated for each of the occupancy cate-
gories represented in the data sources. Selection of the Pilot Study Development of System and
Similar estimates were calculated for a System(s) Component ITM Database
category referred to as “combined,” An important step in the pilot study The second element of the study
which is simply combined estimates was the selection of the sprinkler sys- was the development of a database.
for both the commercial and general tems and collection of the data to be The raw ITM data collected for the
categories. The 95 percent confidence studied. This was accomplished by sprinkler systems were reviewed and
limits were selected as representative reviewing existing sprinkler system an appropriate database scheme devel-
of confidence limits typically used for ITM data, in addition to available sys- oped. The data obtained from the ITM
quality assurance estimates for manu- tem drawings and documentation. reports were placed in a spreadsheet
factured machine parts. Other confi- Detailed ITM records were obtained database in the statistical package.14
dence limits are routinely used, for a 66-month period for several This statistical package offers the abili-
depending on the required certainty sprinkler systems in the same complex ty to serve as a database and a statisti-
associated with a particular product or of buildings. cal tool for analysis.
system. Table 2 provides a summary of
the results of this analysis.
This relatively simple effort at esti-
mating variance in reported data
improves the statistical certainty of the Sprinkler System Figure 1.
reported reliability estimates. For Operational Failure Simplified Automatic Sprinkler
example, for the three occupancy cate- Fault Tree
gories presented in Table 2, the
“mean” reliability estimates range from
93.1 to 96.0 percent, a relatively small
range, and much smaller than the
range associated with the raw reliabili-
ty estimates in Table 1. Greater confi-
dence in the estimates is also provided
by reporting a range of estimates using Sprinkler
8" Alarm Failure
upper and lower confidence limits. Pipe
Check
This information reduces the uncer- Rupture
Valve
tainty in estimating the impact of
sprinkler system reliability in risk-
based design evaluations.

ANALYSIS OF SMALL DATA SETS

Field performance data for new Sprinkler Sprinkler

sprinkler technologies are limited. 1 6
Therefore, in order to estimate the reli-
Sprinkler Sprinkler
ability of these systems or components,
2 5
methods must be used that can handle
small data sets. The results from a pilot Sprinkler Sprinkler
study13 of several existing automatic 3 4
sprinkler systems are used to demon-
strate the usefulness of such analyses.

10 Fire Protection Engineering N UMBER 9

 The database spreadsheet set up each system with boundaries at the riser Reliability Estimates
inspection form as an individual case. and sprinkler grid assumed that the Table 4 provides the reliability esti-
The ITM results were entered for each water supply was 100 percent reliable. mates and associated uncertainty of the
component in each system identified by system fault tree model calculations.
the test record. The results for all com- Component Failure Rates The analysis was performed using the
ponent tests were either “pass” or “fail.” The model used to develop compo- existing ITM frequencies, checking
nent failure rates was the Exponential manual valve positions, and sprinkler
Development of System Fault Trees Model for Life Testing (EMLT).13 The and pipe inspections conducted each
Once component failure rates were EMLT model defines the estimate of month. The existing frequency tests all
developed, system schematics were the mean life (µ) of a component as other system components quarterly.
used to develop fault trees for individ- The system fault tree models were then
Tr
ual systems. Figure 1 provides a sum- µ= (2) used to estimate the reliability of the
marized version of the fault tree r sprinkler system if the monthly inspec-
design. The fault tree structures were where tions were extended to quarterly fre-
programmed into spreadsheets. The Tr = accumulated time on test, and quencies. In addition to the actual
spreadsheet programs allowed failure r = number of component fires. component failure data, industry com-
probability and reliability information ponent failure data for similar compo-
to be propagated through the systems The confidence interval about the nents were used in the fault trees. The
using the fault tree models, resulting in mean is given by reported confidence intervals were also
an overall system reliability estimate. propagated to allow for comparison.
2 Tr 2T
Once the baseline system reliability
2
<µ< 2 r (3) The uncertainty intervals reported
information was obtained, further Xα / 2 X1−α / 2 for the system reliability estimates
analysis was performed to examine reflect the propagation of the 95 per-
testing and inspection intervals and where Xα2 / 2 is dependent on the cent confidence interval about the
how altering these intervals affected degrees of freedom (DF) and found mean component failure rates through
the system’s reliability. The use of fault in statistical tables, and the fault tree models. This was accom-
trees provided information that offers DF = 2(r). plished by quantifying the component
insight into testing and inspection fre- failure rate distributions as fuzzy sets
quencies and established a means to The 95 percent confidence interval and using interval arithmetic in the
track system performance. about the estimated mean was calcu- fault tree model. Singer16 presents the
Fault trees were constructed for lated for each component failure rate. theory and methodology for this quan-
each sprinkler system. For the fault The individual system components’ tification and propagation.
tree models, the system’s boundaries failure rates are provided in Table 3 The mean reliability estimates illus-
were defined as the base of the system along with the industry reported fail- trate the reduction in system reliability
riser to the sprinkler grid. Defining the ure rates for similar components. that occurs when ITM frequencies are

Table 3. Component Failure Rates

Number of Total Hours Pilot Systems ITM Data Industry
Component Components in System 95% Confidence Interval Failure Rate
in Database Failure Rate (failures/hour) (failures/hour)

PIV 10 480,480 0 N/A

ACV 10 480,480 0 1
4.0 x 10-6
OSY 172 8,264,256 7.5x10-8 < 3.6x10-7 < 8.7x10-7 1
4.0 x 10-6
Main Drain 10 480,480 0 4.0 x 10-6
Inspector’s Test 10 480,480 2.3x10-6 < 8.3x10-6 < 1.8x10-5 4.0 x 10-6
Flow Alarm 10 480,480 5.8x10-6 < 1.5x10-5 < 2.7x10-5 2
4.6 x 10-8
Motor Gong 10 480,480 4.1x10-5 < 2.5x10-5 < 1.3x10-5 2
2.0 x 10-6
Fire Department Connection 10 480,480 0 N/A
Piping (gasket failure) 10 480,480 5.0x10 < 4.0x10 < 1.2x10
-7 -6 -5 1
1.0 x 10-6

Note: 1 from Finucane and Pickney12

2 from WASH-140015

W INTER 2001 Fire Protection Engineering 11

 Table 4. Sprinkler System Reliability Estimates and 95% Confidence Limits for Six Existing Sprinkler Systems13

System Reliability System Reliability System Reliability

System Reliability Using Industry for All ITM at for All ITM at
System at Current Component Failure Rate Data 3-Month Frequency 3-Month Frequency
Component Testing and Current Testing Using Pilot Component Using Industry Component
Frequency1 Frequency Failure Rate Data Failure Rate Data

System 1 0.949<0.993<0.999 0.864<0.984<0.997 0.854<0.978<0.998 0.684<0.971<0.996

System 2 0.949<0.993<0.999 0.840<0.971<0.994 0.854<0.978<0.998 0.665<0.963<0.993
System 3 0.949<0.993<0.999 0.864<0.984<0.997 0.854<0.978<0.998 0.684<0.971<0.996
System 4 0.949<0.993<0.999 0.864<0.984<0.997 0.854<0.978<0.998 0.684<0.971<0.996
System 5 0.949<0.993<0.999 0.864<0.984<0.997 0.854<0.978<0.998 0.684<0.971<0.996
System 6 0.948<0.993<0.999 0.856<0.981<0.996 0.852<0.978<0.997 0.665<0.963<0.993

Note:1 Monthly tests of manual valves, sprinklers, and piping; quarterly frequency for other components.

reduced. Not only does system reliabili- REFERENCES 10 Powers, R.W. “Sprinkler Experience in
ty decrease with reduced ITM frequen- High-Rise Buildings (1969-1979),” SFPE
cies, but also the uncertainty associated 1 Modarres, M. Reliability and Risk Technology Report 79-1, Society of Fire
in the lower reliability direction of the Analysis, University of Maryland, Marcel Protection Engineers, Boston, MA, 1979.
uncertainty interval becomes larger. This Dekker, Inc., New York, NY, 1993.
11 Richardson, J.K. “The Reliability of
is an element of system reliability analy- 2 Taylor, K.T. “Office Building Fires...A Automatic Sprinkler Systems,” Canadian
sis that is often overlooked but greatly Case for Automatic Fire Protection,” Fire Building Digest, Vol. 238, July 1985.
affects the interpretation of the results Journal, 84 (1), January/February 1990,
12 Finucane, M, and Pickney, D. “Reliability
and clearly demonstrates the limitations pp. 52-54.
of Fire Protection and Detection
of a given database. As the database is 3 Kook, K.W. “Exterior Fire Propagation in Systems,” United Kingdom Atomic
expanded, uncertainty associated with a High-Rise Building,” Master’s Thesis, Energy Authority, University of
the reliability estimates will be reduced. Worcester Polytechnic Institute, Edinburgh, Scotland.
The results of this effort suggest that Worcester, MA, November 1990.
13 Hughes Associates, Inc. “Pilot Study:
meaningful reliability estimates can be 4 Maybee, W.W. “Summary of Fire Automatic Sprinkler System Reliability
obtained for sprinkler systems with Protection Programs in the U.S. Analysis,” Hughes Associates, Inc.,
limited data. This capability is helpful Department of Energy – Calendar Year Baltimore, MD, September 30, 1998.
in addressing specific types of systems 1987,” U.S. Department of Energy,
14 SYSTAT, SYSTAT 6.0 for WINDOWSTM,
(including those using newer technolo- Frederick, MD, August 1988.
SPSS, Inc.
gies) or systems exposed to similar 5 Marryat, H.W. Fire: A Century of
15 WASH-1400, “Reactor Safety Study, an
environments. Based on the results of Automatic Sprinkler Protection in
Assessment of Accident Risks in U.S.
such analysis, ITM frequencies or sys- Australia and New Zealand 1886-1986,
Commercial Nuclear Power Plants,”
tem components can be tailored to Australian Fire Protection Association,
WASH-1400, United States Atomic
achieve a desired reliability based on Melbourne, Australia, 1988.
Energy Commission, August 1974.
the specific system in question rather 6 Milne, W.D. “Automatic Sprinkler
16 Singer, D. “A Fuzzy Set Approach to
than general industry values. Protection Record,” Factors in Special
Fault Tree and Reliability Analysis,”
Fire Risk Analysis, Chapter 9, pp. 73-89.
Fuzzy Sets and Systems, 34, pp. 145-155.
ACKNOWLEDGMENTS 7 NFPA. “Automatic Sprinkler Performance
17 Schemel, C.F., and Budnick, E.K. “Pilot
Tables, 1970 Edition,” Fire Journal, July
Study: Analyzing Fire Protection System
This article is based on work sup- 1970, pp. 35-39.
Reliability Using Limited Databases,”
ported by the U.S. Department of 8 Miller, M.J. “Reliability of Fire Protection Proceedings of the Fire Suppression and
Energy and the National Institute of Systems,” Loss Prevention ACEP Detection Research Applications
Standards and Technology. The author Technical Manual, 8, 1974. Symposium, Fire Protection Research
thanks Mr. Christopher Schemel, who Foundation, Quincy, MA, February 1999.
9 Linder, K.W. “Field Probability of Fire
worked on both of these projects and Detection Systems,” Balanced Design
performed analyses that were relied Concepts Workshop, NISTIR 5264, R.W.
upon in preparing this article. Bukowski (Ed.), Building and Fire
Research Laboratory, National Institute of
Edward K. Budnick, P.E., is with Standards and Technology, September For an online version of this article, go
Hughes Associates. 1993. to www.sfpe.org.

12 Fire Protection Engineering N UMBER 9