Risk Assessment

STP 1150
Fire Hazard and Fire Risk

Assessment
Marcelo M. Hirschler, editor
ASTM Publication Code Number (PCN)

04-011500-31
ASTM
1916 Race Street
Philadelphia, PA 19103
Library of Congress Cataloging-in-Publication Data
Fire hazard and fire risk assessment/Marcelo M. Hirschler, editor.

(STP; 1150) Based on papers presented at a symposium held in San Antonio,
Tex., on Dec. 3, 1990.
"ASTM publication code number (PCN) 04-011500-31."
Includes bibliographical references.
ISBN 0-8031-1447-8
1. Fire hazard assessment-Congresses. 2. Fire risk assessment-
Congresses. I. Hirschler, M. M. II. Series: ASTM special
technical publication; 1150.
TH9446.3.F57 1992 92-16044
628.9'22-dc20 CIP
Copyright © 1992 AMERICAN SOCIETY FOR TESTING AND MATERIALS, Phil-

adelphia, PA. All rights reserved. This material may not be reproduced or copied, in whole
or in part, in any printed, mechanical, electronic, film, or other distribution and storage
media, without the written consent of the publisher.
Photocopy Rights
Authorization to photocopy items for internal or personal use, or the internal or personal
use of specific clients, is granted by the AMERICAN SOCIETY FOR TESTING AND
MATERIALS for users registered with the Copyright Clearance Center (CCC) Transac-
tional Reporting Service, provided that the base fee of $2.50 per copy, plus $0.50 per page
is paid directly to CCC, 27 Congress St., Salem, MA 01970; (508) 744-3350. For those
organizations that have been granted a photocopy license by CCC, a separate system of
payment has been arranged. The fee code for users of the Transactional Reporting Service
is 0-8031-1447-8/92 $2.50 + .50.
Peer Review Policy
Each paper published in this volume was evaluated by three peer reviewers. The authors
addressed all of the reviewers' comments to the satisfaction of both the technical editor(s)
and the ASTM Committee on Publications.
The quality of the papers in this publication reflects not only the obvious efforts of the
authors and the technical editor(s), but also the work of these peer reviewers. The ASTM
Committee on Publications acknowledges with appreciation their dedication and contribution
to time and effort on behalf of ASTM.
Printed in Philadelphia, PA
June 1992
Foreword
This publication, Fire Hazard and Fire Risk Assessment, contains papers presented at the
symposium of the same name, held in San Antonio, TX on 3 Dec. 1990. The symposium
was sponsored by ASTM Committee E-5 on Fire Standards. Marcelo M. Hirschler of Safety
Engineering Labs in Rocky River, OH presided as symposium chairman and is the editor
of the resulting publication.
Contents
Ot'erview-M. M. HIRSCHLER 1
INTRODUCTION TO FIRE HAZARD AND FIRE RISK ASSESSMENT
~ Importance of Carbon Monoxide in the Toxicity of Fire Atmospheres-

s. M. DEBANNE, M. M. HIRSCHLER, AND G. L. NELSON 9
rife Risk Rating Schedules-J. M. WATTS, JR. 24
USE OF FIRE TESTS FOR FIRE HAZARD ASSESSMENT
£qluating the Hazards of Large Petrochemical Fires-N. R. KELTNER,

R. U. ACTON, AND W. GILL 37
adncal Cable Fire Hazard Assessment with the Cone Calorimeter-

M. M. HIRSCHLER 44
Pa-formance of Plastic Plumbing and Electrical Products in Fire Resistive

Assemblies-1. B. ZICHERMAN 66
FIRE HAZARD ASSESSMENT
Fire Hazard Assessment of Composite Materials: The Use and Limitations of

Current Hazard Analysis Methodology-p. J. DINENNO AND C. L. BEYLER 87
Fire Engulfment of Pressure-Liquefied Gas Tanks: Experiments and Modeling-

u. K. SUMATHIPALA, G. V. HADJISOPHOCLEOUS, N. U. AYDEMIR, C. M. YU,
C. M. SOUSA, F. R. STEWARD, AND 1. E. S. VENART 100
Development of a Benefit Analysis for an Onboard Aircraft Cabin Water Spray

System-R. G. HILL, c. P. SARKOS, AND T. R. MARKER 116
FIRE RISK ASSESSMENT
How to Tell Whether What You Have is a Fire Risk Analysis Model-1. R. HALL, JR. 131
Predicting Product Fire Risk: A Review of Four Case Studies-R. W. BUKOWSKI,

S. W. STIEFEL, F. B. CLARKE, III, AND J. R. HALL, JR. 136
A Reliability Methodology Applied to Halon 1301 Extinguishing Systems in

ComputerRooms-J. STECIAK AND R. G. ZALOSH 161
v
Overview
Fire has been a scourge of society for a very long time now, both in terms of its human
and economic impact: fire fatalities, fire injuries, and direct and indirect losses from fire.
North America, in particular, has the dubious distinction of hosting the highest fire fatality
rate per capita in the industrialized world.
The traditional way in which fire studies have been made is by using fire tests, of various
degrees of usefulness, which measure a particular fire property (or fire-test-response char-
acteristic, in ASTM fire parlance). The results have then been used to rank materials based
on a single fire property. Unfortunately, fire performance (response of materials or products
in fires rather than fire tests) is often poorly predicted by many tests which have usually
not been designed based on sound engineering principles.
It has now become clear that there needs to be better predictive ways to make fire safety
decisions. These predictive tools are fire models, which are used to analyze (or assess) the
danger (fire hazard) associated with burning a particular material, product, or assembly in
a specified situation (fire scenario). Thus, ASTM has defined fire hazard as "the potential
for harm to people, property, or operations" (ASTM Terminology Relating to Fire Stan-
dards, E 176-91d). However, fire hazard presupposes that a fire will take place. Fire risk
is a measure of fire loss (life, health, animals, or property) that combines (a) the potential
for harm in the various fire scenarios that can occur and (b) the probabilities of occurrence
of those scenarios, within a specified period, in a defined occupancy or situation. As such,
fire risk does not assume that a fire will take place, but it incorporates the probability of
the fire occurring. Thus, whereas fire hazard measures the potential for harm with respect
to one single scenario, fire risk measures the potential for harm in the full range of all
possible scenarios, using the probabilities of each one of those scenarios to measure the
relative importance of each of them. Therefore, a fire risk measure is a statistic derived
from an underlying probability distribution on a measure of fire hazard. It is important to
stress, however, that by its nature, a fire risk measure is not applicable to the prediction of
the occurrence or of the potential for harm of an individual fire.
With the expansion of the capability of large computers and the increased use of the
personal computer, it has become possible for many people to manipulate large amounts
of information, and to use it in order to predict fire performance. Among the consequences
of this has been the appearance of a number of fire models that can predict fire hazard or
fire risk.
The ASTM board has adopted a policy on fire standards. This policy acknowledges the
existence of three kinds of fire standards: fire-test response standards, fire hazard assessment
standards, and fire risk assessment standards. The board gave committee E-S on Fire Stan-
dards the exclusive authority to write fire hazard or fire risk assessment standards. In order
to better understand what this involves, Subcommittee E-S.3S on Fire Hazard and Fire Risk
Assessment is working on standard guides for the development of fire hazard and fire risk
assessment standards. Several other subcommittees are also working, and have made various
degrees of progress on a number of fire hazard assessment standards.
In order to aid in the understanding of fire hazard and fire risk assessment models, ASTM
Committee E-S has organized this international symposium, conceived within subcommittee
E-S.32 on Research, held in San Antonio, TX, on 3 Dec. 1990.
The 16 papers published herein can be divided into 5 broad categories: (1) Introduction
1
2 FIRE HAZARD AND FIRE RISK ASSESSMENT
to Fire Hazard and Fire Risk Assessment, (2) Use of Fire Tests for Fire Hazard Assessment,
(3) Fire Hazard Assessment, (4) Fire Risk Assessment, and (5) Fire Risk Assessment and
Building Codes.
Introduction to Fire Hazard and Fire Risk Assessment

This section includes papers that deal with important aspects that need to be considered
in order to carry out a fire hazard or a fire risk assessment.
A common misconception in the public view of fire hazard is that fire hazard is primarily
or exclusively a matter of smoke toxicity. The paper by Debanne et al. summarizes the two
most important sets of human forensic studies carried out to investigate the issue of smoke
toxicity and its implications for fire hazard. The studies showed that the victim population
distributions are very different for fires and non-fire carbon monoxide fatalities, but that
the blood carboxyhemoglobin distributions, once equal populations are compared, are very
similar. Furthermore, the importance of carbon monoxide in fire atmospheres has not
changed between the 1940s and the 1980s. The authors conclude that death in fires, by
smoke inhalation, appears to be overwhelmingly associated with carbon monoxide poisoning.
Moreover, their work shows that carbon monoxide can kill human beings (rather than test
animals) at blood carboxyhemoglobin levels much lower than was previously thought pos-
sible. The combination of this finding and the fact that most small scale smoke toxicity tests
cannot predict carbon monoxide levels adequately means that such tests, and smoke toxicity
in general, must playa small role in fire hazard assessment.
It is often thought that fire hazard and fire risk assessment is necessarily the result of
complex mathematical models. Watts shows that heuristic models of fire safety, which he
calls fire risk rating schedules, can be used as indicators of fire safety. He presents in his
paper, three examples of fire risk rating schedules which have varying degrees of sophisti-
cation. The first one is the prediction of heat release rates of upholstered furniture by using
a model that combines laboratory scale heat release measurements with various empirical
parameters. The next one is the basis for ASTM Practice for Assessment of Fire Risk by
Occupancy Classification (Commentary), E 931, which developed an occupancy classification
based on a Delphi approach and assigned various weighting values to a number of elements.
Although this is no longer accepted as a form of fire risk assessment, it is a very useful
simple means to give numerical results to common sense. The final example is a trade-off
model, again derived from a Delphi approach, to trade off various fire safety alternatives,
such as active (smoke detectors, sprinklers) and passive (products with better fire perfor-
mance) fire protection measures.
Use of Fire Tests for Fire Hazard Assessment

The papers in this section deal with means by which fire tests can be used to predict fire
hazard in a variety of fire scenarios.
One of the types of fire which has the most serious potential is the case of the high intensity
fires which can occur in petrochemical facilities, or when liquid fuels are transported. In
ASTM Subcommittee E-5.11 there is work in progress to develop a standard test method
to address such fires. Keltner et al. addresses one aspect of this problem, when dealing with
petrochemical plant fires. They show that it is essential in such cases to adequately char-
acterize the fire environment. Moreover, in that connection they discuss those cases where
fire temperature is the dominant issue to be addressed and those other cases where fire heat
flux is the more important parameter, since there is no univocal correlation between the
two. They conclude that for good fire testing practice both temperature and heat flux should
be taken into consideration before making a fire safety decision.
rue Hazard Assessment
The papers in this section address specific fire hazard assessment problems, by using
computer models, statistics, a combination of models and engineering judgment, or a com-
bination of experiments and analyses.
DiNenno and Beyler address the way in which new, unconventional materials (composites)
might be used to replace traditional materials in naval applications. The new materials had
not been expected to offer the same degree of fire protection, but were known to yield
various other advantages. Thus, the authors indicate that the results of various fire response
tests can then be used, in conjunction with a "fire hazard analysis package" to determine
an acceptable level of fire hazard. Unfortunately, the authors argue that the failings of the
existing "packages" make it necessary to use sound engineering judgment to overcome their
limitations, and to combine different approaches. Once that has been done, progress can
be made in evaluating fire hazard and in weighing the results against the other advantages
and disadvantages.
Pressure liquefied gas tanks, especially when used in transportation, need to offer par-
ticular protection because of the potential intense energy of any resulting fire. The best
known examples of such fires are the Boiling Liquid Expanding Vapor Explosions (BLEVEs)
which have caused many serious accidents. Sumathipala et al. have developed a zone fire
model, which they call PLGS-l, together with computational rules, PLGS-2D, to describe
the behavior of such tanks when there is an external pool fire in their proximity. Results
from experiments involving both midsize (40 L) and large externally-heated partially-filled
horizontal cylindrical vessels have been compared with the predictions of the computer
4 ARE HAZARD AND FIRE RISK ASSESSMENT
models. This work has focussed on heat transfer and pressure response parameters, in order
to improve the understanding of the physical phenomena and develop fire protection strategies.
Fatal fires in aircraft seldom occur. However, when such a fire occurs, often following a
survivable crash, the results can be catastrophic. Hill et al. have initiated a study of the
benefits and disbenefits of installing an on board aircraft cabin water spray system. The
study, which is currently underway, involves aviation authorities of various countries, in-
cluding the United States, United Kingdom, France, and Canada. This system has been
shown by the authors in full-scale aircraft fire tests, to decrease fire temperatures, and thus,
lower burning rates, heat release rates, and smoke emission rates. On the other hand, such
a system is generally incapable of extinguishing the fire and can cause a series of unwelcome
consequences in case of false discharges. The final result of the study will likely determine
whether such systems will be installed in commercial aircraft.
Fire Risk Assessment

This section contains papers which address the problem of fire risk, either by explaining
what a fire risk model is, or by using one to apply to a specific fire scenario.
After lengthy discussions and disagreements, both based on fundamental concepts and
terminology, the paper by Hall is an attempt to clarify what a comprehensive fire risk
assessment is and is not. The paper describes the most common misconceptions about fire
risk analysis. The paper also describes key concepts in fire risk analysis and shows how fire
risk is simply one facet of overall risk. In particular, it explains the types of fires and types
of human behavior that should not be excluded. The paper poses questions to the reader
in order to ascertain whether the model in question is a fire risk assessment model or
something else.
The concepts discussed by Hall are illustrated in the work of a research team put together
by the National Fire Protection Research Foundation (NFPRF) to develop a comprehensive
fire risk assessment methodology that could be applied to a large number of fire scenarios
and a large number of products. The paper by Bukowski et al. is one result of that program.
The authors start by explaining the methodology developed, which is based on the use of
the fire hazard model HAZARD I, followed by an 8-step procedure. In this approach,
after the product/occupancy set has been selected, representative characteristics are chosen
and incorporated into the fire model. The model is then run for a base case product, which
can represent the average of what is presently being used, or a particular product of specific
interest, for whatever reason. The fire risk assessment is then carried out for the base case.
The product characteristics are then changed to those of a new product and the fire risk
assessment carried out again. The process ends with the two results being compared. This
paper describes four cases studied: (1) upholstered furniture in residences (the single fire
scenario associated with the largest number of fire deaths), (2) carpets in offices (a very low
fire risk scenario), (3) concealed combustibles (electrical cables) in hotels (a low fire risk
scenario, but one which has been associated with public controversy), and (4) interior finish
in restaurants (a case which would address heavily regulated products and would introduce
vertical flame spread into the model). In every case, the results are compared with the fire
experience. The work succeeded in developing a methodology and applying it satisfactorily
to a variety of scenarios.
A different kind of fire risk assessment methodology is applied by Steciak and Zalosh to
the use of gaseous (Halon 1301) extinguishing systems in computer rooms. The methodology
uses occurrence probability data applied to the different failure scenarios for its calculations.
Finally, the effects of various measures, such as human intervention, inspection intervals,
dential dwellings governed by the Ontario Building Code.

The final paper is an application of the concepts of fire risk assessment at the community
level. Harvey, a fire chief, has been instrumental in using the concepts encompassed in a
fire risk assessment model to evaluate fire safety in his community of Boulder, CO. Local
planners, building and fire officials, owners, designers, and builders worked as a team to
use new engineering methods for enhancing fire safety in the community while minimizing
cost increases. This involved not only fire protection measures in the buildings themselves
(sprinklers were eventually mandated) but also in the number of fire stations and their
equipment. Entry into the system was voluntary at first but became mandatory when it was
found that overall cost savings were attained, with no opposition from the public who could
see the benefits.
Conclusions
The papers summarized above should provide the reader with a broad understanding of
the issues involved in fire hazard and fire risk assessment and with an overview of some of
the most interesting techniques available today. The diversity of papers is probably sufficient
to offer different perspectives and tools both for workers in the field and for other readers
concerned with fire safety. This is an area where advancements are occurring in leaps and
bounds; however, the papers presented here should serve as an excellent literature data
base. The symposium chairman thanks the other members of his committee, in particular
Dr. Harry K. Hasegawa (Lawrence Livermore National Laboratory) and Dr. James R.
Mehaffey (Forintek) for their invaluable assistance to make this publication possible. He
also acknowledges the efforts of the authors and of the ASTM personnel who made this
happen.
Marcelo M. Hirschler,
Safety Engineering Laboratories,
Rocky River, OH, 44116,
symposium chairman and editor.
Introduction to Fire Hazard and Fire Risk
Assessment
REFERENCE: Debanne, S. M., Hirschler, M. M., and Nelson, G. L., "The Importance of
c.bon Monoxide in the Toxicity of Fire Atmospheres," Fire Hazard and Fire Risk Assessment,
ASTM STP 1150, Marcelo M. Hirschler, Ed., American Society for Testing and Materials,
Philadelphia, 1992, pp. 9-23.
ABSTRACT: This paper addresses the issue of determining toxic fire hazard in a manner that
is relevant to the large-scale fire scenarios that cause most lethalities, ventilation controlled
Oashover fires.
The work involves (a) a literature study of background information on the toxicity of carbon
monoxide (CO) to humans, (b) two very extensive forensic studies (with 2241 and 2673 cases,
respectively) on human lethality involving CO in fires and non-fires, (c) statistical analyses of
the forensic data, to ensure full separation of variables, and (d) an analysis of the literature
011 the effects of different parameters in fire atmospheres and in small-scale tests.
This work addresses five issues essential to fire hazard assessment:
(I) Relative role of toxicants other than CO in causing fire fatalities and critical carboxy-
hemoglobin (COHb) values representing lethality.
(2) Differences in populations between fire victims and those dying in other CO-containing
atmospheres.
(3) Comparison of modern fire atmospheres, containing smoke from man-made materials,
with traditional fire atmospheres.
(.t) CO yields in flashover fires and effects of fuel chemistry.
(5) CO yields in small-scale toxicity tests and fuel effects.
The work has shown that:
(I) The toxicity of fire atmospheres is determined almost exclusively by the amount of CO.
The lethal CO threshold level depends on the physical condition of the victim, but
COHb values> 20% can produce lethality with no other apparent cause.
(2) Fire and non-fire CO victim populations are very different: fire victims are much older
or much younger and more infirm, and thus more sensitive to CO than those in non-
fire exposures.
(3) Replacing large amounts of natural and traditionally used materials by man-made ma-
terials has made no difference to fire atmosphere toxicity.
(4) CO concentrations in large-scale flashover fire atmospheres are determined by oxygen
availability and such variables, but are little affected by chemical composition of fuels.
(6) Small-scale tests give excessively low CO yields so that they cannot be used to predict
toxic fire hazard for ventilation controlled flashover fires, although post computational
CO concentration corrections can make these tests useful as part of fire hazard assess-
ment calculations.
KEY WORDS: carbon monoxide, carboxyhemoglobin, fire fatalities, fire gases, fire hazard,
fire scenario, flashover, smoke toxicity, toxicity test
'Associate professor, Department of Epidemiology and Biostatistics, Case Western Reserve Uni-
werntv, Cleveland, OH 44106.
0Safety Engineering Laboratories, 38 Oak Road, Rocky River, OH 44116.
'Dean, College of Science and Liberal Arts, Florida Institute of Technology, Melbourne, FL 32901-
6988.
It has been discussed in the National Fire Protection Association (NFPA) Quarterly, in
1933, that the main direct cause of death in fires was combustion product toxicity [1]. For
ASTM use, smoke is interpreted as the sum total of the gaseous, liquid, and solid airborne
products of combustion (ASTM Terminology Relating to Fire Standards, E 176). Smoke
is, thus, not a uniform material, so that its composition depends on the exact conditions
under which it was generated. However, smoke always contains two principal types of toxic
gases: asphyxiants and irritants ..
All organic (that is, carbon-containing) materials give off carbon monoxide (CO), carbon
dioxide, and water on combustion [2]. Many other combustion gases are released from
burning materials, but most of them are characteristic of particular classes of fuels. Carbon
monoxide is a highly toxic asphyxiant gas, which is odorless and tasteless. It is the individual
toxic gas associated with the largest fire hazard. Many other components of smoke are much
more toxic, but those are usually present in much smaller concentrations. On the other
hand, some gases are present at higher concentrations (typically carbon dioxide and water)
but they are of much lower toxicity. Two studies which sent fire fighters into actual buildings
on fire, while equipped with combustion product monitors, [3,4] had the same conclusions;
the overwhelming hazardous toxicant in fire is CO.
The other principal toxicants in fires are acrolein that has the second highest ratio, after
CO, of peak level found to lethal level and is emitted by polyolefins and cellulosic materials),
hydrogen cyanide (emitted by N-containing materials), and hydrogen chloride (emitted by
Cl-containing materials) [5].
The mechanism by which CO acts on mammals is by competing with oxygen for the
hemoglobin in blood and tying it up as COHb, rather than as the normal oxyhemoglobin.
The hemoglobin (Hb) fraction tied up as COHb is normally expressed as percent COHb
(which means the percentage of the total hemoglobin present as COHb rather than as
oxyhemoglobin). Carbon monoxide reacts ~21O times faster with hemoglobin than does
oxygen, so that it can lead to an oxyhemoglobin deficiency, even at low CO concentrations.
This lack of oxyhemoglobin then leads to hypoxia, which can cause cerebral damage and
eventual death by asphyxiation. However, the reaction of CO with hemoglobin to yield
COHb is reversible, and COHb levels will decrease when the CO exposure ceases.
Traditional wisdom has set a value of 50% COHb as the threshold level for human lethality
[5,6]. This has been taken to mean that blood COHb levels of 2:50% COHb inevitably
leads to death, and that if a fatality is autopsied and its COHb level was <50%, CO poisoning
could not be the sole cause of death. However, it is known that 25% COHb already causes
myocardial damage [1].
Carbon monoxide is also present in non-fire atmospheres. Two typical examples are those
resulting from malfunctioning unvented gas or charcoal heaters, or from automobile exhaust.
In these two cases, CO is widely recognized as being virtually the only toxicant of any
consequence present. This is an important difference from fire atmospheres, which contain
several other toxicants, some of which may be present at high enough concentrations to
cause senous concern.
Several fire hazard issues will be addressed in this paper, which includes a description of
some of the major results from a series of studies sponsored by the Society of the Plastic
Industry, Inc. [8].
(1) The presence of these other toxicants and their importance in the toxicity of fire
atmospheres is an issue with important implications for fire hazard assessment because
of the way in which smoke toxicity is normally determined. One of the issues that
needs to be understood is the relative role of toxicants other than CO is causing fire
fatalities and whether there is any critical value of COHb that represents lethality.
DEBANNE ET AL. ON THE IMPORTANCE OF CARBON MONOXIDE 11
(2) Fire fatalities are most commonly unwitting victims of external events that overtake
them, while many individuals who succumb to the intoxicating effects of automobile
exhaust do so at their own initiative. Thus, it is worth investigating whether the
populations that lose their lives in fire atmospheres and in other CO-containing at-
mospheres are identical. If it is found that the two populations are significantly dif-
ferent, the follow up query is whether the effects of CO can be separated from other
effects, inherent in the characteristics of the population involved. Fire hazard or fire
risk assessment models require this answer since they have to concentrate on the
population most at risk in fires.
(3) The modern world in which we live contains a large proportion of products made
from synthetic materials, for example, plastics. The combustion of these materials
can generate toxicants which are different from those traditionally generated by natural
materials. This has caused a great deal of speculation that the toxicity of modern fires
may be significantly different from that of fires of another era. Therefore, another
issue of importance to fire hazard is an understanding of the relative toxicity of fires
involving new and traditional materials.
(4) The vast majority of the fire fatalities that occur in the United States are found away
from the room of fire origin. That is a clear indication that those fires which lead to
fatalities tend to be big fires. In other words, they are flaming fires, which have
involved the entire room of origin and gone beyond it. These types of fires are often
described as flashover or post-flashover fires [9]. Any smoke toxicity assessment that
would attempt to address fire risk should, of course, address such high intensity flaming
fires. Therefore, it is of interest to fire safety to understand what types of atmospheres
are to be expected in these fires, particularly as regards their CO content.
(5) Small-scale toxicity tests are the most common means of assessing smoke toxicity of
materials or products. The final question to be asked in this study regards the CO
yields in such tests and how they compare to those in the real full-scale fires that
cause the majority of fire fatalities.
Literature Search
Dr. Gordon L. Nelson, at the Department of Polymer Science at the University of South-
ern Mississippi, carried out a literature search on information available regarding human
exposure to CO and fatalities. This study unearthed more than 100 references, and found
several major forensic studies, both of fire victims and of victims of non-fire CO exposure.
Full details of this study will be given elsewhere, together with those of other associated
studies, all sponsored by the Society of the Plastics Industry [8]. However, the 12 most
notable studies are described in Table 1. Only two of these studies have been very widely
publicized, namely, those in Maryland and Glasgow. Interestingly, none of the referenced
studies involved over 1 000 cases. This means that several of them would have had great
difficulties making any statistically valid separate analyses of individual variables.
It is worth discussing some of the most interesting results of these studies. They have
found that when various individuals are exposed to a particular atmosphere, their COHb
levels can be very different. Furthermore, some people may even survive the exposures,
and others, for no apparent reason, succumb. These studies have also found both survivors
of CO exposure who have COHb levels of well over 50% and fatalities from pure CO
exposure with COHb levels in the range of 20 to 40%. In fact, around the turn of the
century, John Scott Haldane poisoned himself with CO and measured COHb levels of over
50%, without succumbing to the experience.
The age distribution of fire victims was found to have a bimodal distribution, both in the
12 ARE HAZARD AND FIRE RISK ASSESSMENT
Maryland and Glasgow studies [19-22] (Fig. 1). In contrast, the age distribution for non-
fire suicide victims, from automobile exhaust in the Oslo study [15], is a simple distribution
with a single peak (Fig. 2).
A histogram of victim frequency versus COHb levels for non-fires shows that there is a
strong difference between accidental victims (CPSC study of gas heaters [17]) and those
causing their own deaths (Oslo study of automobile exhaust [16]). Figure 3 shows that suicide
victims end up with very high COHb levels, typically in the 65 to 85% levels. On the other
hand, accidental victims of CO poisoning end up with much lower levels, between 26 and
81%, and with a median of 45%. This is a very important finding, because it indicates that
accidental victims of CO poisoning can die at very low levels, without any additional factor
being involved.
Figure 4 compares the frequency distribution of COHb levels for fire fatality studies with
that of an accidental non-fire CO exposure forensic study. It is clear that the distributions
are very similar and that both have significant proportions of fatalities at levels well under
50% COHb, the traditional lethal level.
This portion of the study shows that 50% COHb can not be a magical number such that
it is the inevitable pass/fail lethality criterion for pure CO poisoning.
Forensic Investigation on CO Victims Across the United States

The Department of Polymer Science at the University of Southern Mississippi (USM)
carried out an extensive investigation, on a countrywide basis, of fatalities associated with
CO. The investigators were Gordon L. Nelson, Dennis V. Canfield, and James B. Larsen.
This investigation prepared a data base of 2 241 cases, originating in 37 laboratories across
the entire United States. The vast majority of the victims died over a relatively short time
frame in the 1980s. Table 2 shows the variables investigated, and Table 3 presents some
information on the data base itself. It is worth pointing out that the data base, which is
CARBON MONOXIDE AND FATALITIES

FIG. 4-COHb distributions for victims in 5 studies, one of which is a non-fire fatality study (CPSC,
(17]) while the others are fire fatality studies: MOM [23], Aircraft [11), MD [19,20], and Scot [21,22].
much more extensive than any of the earlier ones, contains roughly twice as many victims
of fires as of non-fire CO poisoning. More extensive details of this study will be given
elsewhere [8].
The cases that were retained for further investigation where those where the COHb level
was ~ 20%, because it was decided that it was very unlikely that victims with lower levels
of COHb would have died exclusively of CO poisoning. In this connection it is worth
mentioning that smokers can exhibit COHb levels of up to 14% (for example, Ref 25).
Table 3 shows that only -8% of all the cases in the data base had a known COHb blood
level of under 20% saturation, evenly distributed among fire and non-fire cases.
Over 10% of the victims in either data base (fire or non-fire) had a blood COHb level
TABLE 2- Victim variables included in the USM database.
1. Data about the Victim Itself:

• age
• gender
2. Data about the Incident:
• fire/non-fire
• source of CO
• data of incident
3. Data from the Forensic Report
• COHb level
• blood cyanide level
• blood alcohol
• presence of drugs
• survival
• cause of death
• verdict
TABLE 3-Details of USM database.
Fire Non-fire Unknown Total
ALL CASES
No. of Cases 1203 660 378 2241
% Total 54 29 17 100
% Known source 65 35 100
CASESWITH KNOWN % COHB LEVELS
No. of Cases 1037 641 372 2050
% Total 51 31 18 100
% Known source 62 38 100
CASESWITH COHB % ABOVE20%
No. of Cases 961 600 303 1864
% Cases in source 80 91 80 83
% Known source 93 94 81 91
between 20 and 50%. This proportion becomes even bigger when one excludes those victims
with alcohol in their blood.
Figure 5 has the age distribution of the fire and non-fire victims in this study. It is clear
that, just as was seen in Fig. 1, there is a bimodal distribution for fire fatalities, with most
of the victims being either very young or elderly. On the other hand, non-fire victims have
a unimodal age distribution, which peaks between 30 and 45, similar to the one shown in
Fig. 2. The data discussed earlier have indicated that there is an effect of population char-
acteristics on susceptibility to CO poisoning. Consequently, as expected, the COHb distri-
butions for both sets of data are not identical, with the fire one being centered at a somewhat
lDwer COHb level than the non-fire one (Fig. 6). The difference in age distributions, how-
ewer, means that this disparity may well be due to a comparison of unequal populations.
'Ibis will be analyzed in the next section.
This portion of the study reinforced the suggestion that CO can kill at COHb saturation
lI:vels of well under 50% without requiring an additional toxicant. This can now be considered
a firm finding.
llatistical Analysis of Forensic Investigation

A team led by Sara M. Debanne and Douglas Y. Rowland, in the Department of Bio-
llatistics and Epidemiology at Case Western Reserve University (CWRU) analyzed the
University of Southern Mississippi data base, by means of a separation of variables. The
...un variables analyzed separately were: age, alcohol level, presence of drugs and preexisting
Jlhysical condition, as well as the source of CO. Again, the full set of results will be published
elsewhere [8], but a summary will be discussed here. The results were very interesting. They
showed that the COHb frequency distributions are very similar, once these other factors
Dve been accounted for. Figure 7 shows the distributions for two ethanol-free populations
which are very comparable, all fire victims ages 6 to 20 and all non-fire victims above age
6. The similarity of the two curves indicates, or at least suggests very strongly, that CO is
dearly, the principal, if not the overwhelming cause of death in fires.
This study suggests that there is generally no need to look for an additional source of
lethality in fire atmospheres if the COHb level ranges between 20 and 50%. Since CO alone
can account for all deaths, this indicates that the effect of other toxicants in fire atmospheres
• invariably small, if not negligible, in the majority of cases. This gives a clear answer now
10 the first issue to be addressed.
This study also gives a fairly definitive answer to the second issue to be addressed; the
populations of fire and non-fire victims are generally very different. Moreover, the fire
W;tim population contains more individuals that are at higher risk of succumbing to carbon
'" COHb
FIG. 7-COHb distribution for selected victims (comparable populations) in the study carried out at
the University of Southern Mississippi (USM) on fire (ethanol-free 6 to 20 years old) and non-fire (ethanol-
free> 6 years old) CO fatalities.
monoxide poisoning than does the population as a whole or than does the population that
dies from CO poisoning in non-fire cases.
Forensic Investigation on CO victims across time in Cleveland, OH

The same team at Case Western Reserve University also created a new data base of CO
victims of fires and non-fires. This data base contains all victims of CO poisoning investigated
by the Cuyahoga County Coroner's office between the years of 1938 and 1979. Cuyahoga
County, in Ohio, is the county surrounding the city of Cleveland. The reason for the choice
of period was that this coroner's department was headed by the same man (Dr. Samuel
Gerber) over the entire period concerned. The Cuyahoga coroner's office investigated and
carried out autopsies on the victims of every fire in the county as well as of every case of
accidental death and was respected for its high scientific integrity and accuracy. This data
base, containing 2 637 cases, was somewhat different from the USM database. It contained
roughly twice as many non-fire victims as fire victims. This makes it the largest data base
ever put together on CO and fatalities. Table 4 shows some of the main features of the data
base, with more details to be shown elsewhere [8J.
There were two reasons to carry out this study. The first was to discover whether the
COHb levels at which fire victims died have changed over the years and the second was to
compare, once again, fire and non-fire data.
As discussed in the introduction, the modern world, particularly in the United States,
provides the population with a large number of products made with synthetic materials.
Such materials were first developed in the 1930s and 1940s, and did not generate a large
fraction of the overall combustible products in use until the 1960s or 1970s. Their relative
proportion of use in the United States is probably still rising.
In Fig. 8, the median level of COHb found each year for the cases in the study is shown.
The data are presented as differences with the 1941 value and two curves are presented,
issue that was to be addressed; the effect of new materials on the toxicity of fire atmospheres
appears to be negligible.
The separation of variables shows some remarkable similarities between this study and
the previous one. That is, the populations of fire and non-fire victims are very different,
and this difference can be seen to affect their sensitivity to CO poisoning.
Study of Fire Retarded and Non-Fire Retarded Products

Effect of Study on Usefulness of Small-Scale Tests
A study was carried out by the Center for Fire Research at the National Institute for
Standards and Technology (NIST) for the Fire Retardant Chemicals Association [26]. This
involved burning 5 sets of products in fire retarded and non-fire retarded versions: TV
cabinets, business machines, upholstered chairs, circuit boards, and electrical cables.
The small-scale techniques used were the cone calorimeter and the NBS cup furnace
smoke toxic potency test. In those apparatuses the individual materials were tested as mock-
up combinations. The medium-scale technique used was the furniture calorimeter, where
the entire products were tested individually. The full-scale test used involved combining
all five products together in a single room.
The results indicated that the fire retarded products were safer than the non-fire retarded
products, indicating that improved fire performance (by whatever means it is achieved) leads
to lower fire hazard. The results showed that, in the full-scale tests:
• FR materials lost less than half the mass of non-fire retarded ones.
• FR materials released Y4 the heat of non-fire retarded ones.
• FR materials generated Y3 of the toxic gases of non-fire retarded ones.
• FR and non-fire retarded materials generated similar smoke obscuration.
The results also indicated that the cone calorimeter could be used to predict full-scale
ignitability, heat release, and flame spread. It could not be used, however, to predict release
of CO. The authors found that the bench-scale tests were expected to give adequate yields
of gases other than CO, but will almost inevitably give too low yields of CO. Thus, small-
scale tests would be expected to be biased in favor of non-CO species, meaning that they
exaggerate the relative importance of such non-CO species.
The production of combustion gases in small-scale tests is, thus, heavily influenced by
the chemistry of the materials being burned. However, the production of CO in large-scale
fire tests, and thus presumably in real fires, is only somewhat influenced by the chemical
properties of the substance being burned.
Moreover, the CO production in large-scale fires appears to be much more influenced
by the availability of oxygen in the fire. This, in turn, is affected by variables such as
geometry, ventilation, configuration, turbulence, and mixing. Therefore, the authors con-
clude that the use of any less-than-room-sized tests for making CO predictions has to be
deferred until these oxygen supply variables have been sorted out.
This indicates that the toxicity of the atmosphere in real fires is likely to be little affected
by the type of materials causing the fire. The CO levels in large-scale fires are governed
by the number and size of openings, the mass loading, and the burning characteristics (that
is, ignitability, flame spread, and heat release) of the fuels.
Rough algorithms now exist for assessing the CO yields for various large-scale fire sce-
narios. These are probably not yet adequate for widespread use, but they indicate the way
forward to assess the toxic hazard associated with any large-scale fire scenarios. Further
studies are needed before these can be considered well documented [27].
Experience also shows that in full-scale tests where full flaming room involvement is
achieved (those fires that cause the majority of fire fatalities) there is ventilation control
and the oxygen levels get very close to zero. Small-scale tests are rarely carried out under
conditions of low oxygen and high heat flux, because flaming combustion does not occur
under such conditions.
Consequently, it is now possible to give answers to the last two issues to be investigated
in this study. The atmospheres in those large-scale fires that cause most lethality have a
toxicity determined mostly by the concentration of CO and by a low oxygen content. The
yield of this CO is almost independent of the type of materials being burned.
Toxic potency of smoke is a quantitative expression which relates the concentration of
smoke and the exposure time to the achievement of an adverse effect, usually lethality, on
a test animal (E 176). It has to be stressed that the toxic potency of smoke is heavily dependent
00 the conditions under which the smoke is generated, which affect both the quality and
the quantity of smoke.

During the 1970s and early 1980s, many small-scale tests for smoke toxic potency were
developed [6]. These tests differ in many respects, including: fire model, being static or
dynamic, use of animals or analytical tools, animal model for bioassay, and end point. Partly
due to all these differences, the tests lead to tremendous ranking variations for the smoke
of various materials. This was illustrated in a study of the toxic potency of 14 materials by
2 methods which ranked one material most toxic by one protocol and least toxic by the
other protocol. Although neither of those protocols is in common use now, this work
illustrates some of the inevitable problems of such tests.
The toxicity of smoke is a function of its composition, which depends, in turn, both on
the fuel and the combustion mode. Thus, the composition of the smoke generated by any
individual material varies broadly from test to test, and so will its toxic potency. In fact,
the toxic potency of the smoke of most ordinary materials (whether natural or synthetic) is
very similar [6]. Thus, relative smoke toxic potency rankings depend on the exact composition
of the smoke being tested, that is, the small-scale test protocol, and are of little interest
from the viewpoint of fire hazard assessment.
Moreover, small-scale smoke-toxicity tests give inadequately low yields of CO and
adequate yields of other toxicants. Thus, such tests require post computational correction
for CO before their results are directly relevant to fire hazard in the real fires that cause
most fire deaths. This correction for CO yields has been addressed in recent toxicology work
[28], where it has been applied to the results of full-scale and small-scale tests using rats
as animal models, with an excellent outcome.
From the point of view of fire hazard assessment, the work described here puts into
perspective the importance of one of the elements of fire hazard assessment, namely smoke
toxicity. It shows that care must be taken when using toxicity data as input for fire hazard
assessment.
Conclusions
This paper has addressed the five fire hazard issues presented in the introduction, which
are:
(1) (a) The toxicity of fire atmospheres is determined to a very large extent by the amount
of CO; the contribution of other toxicants is usually very small. (b) The lethal level
of CO is heavily dependent on the individual concerned, and the 50% COHb threshold
normally mentioned is not a realistic value. Both fire atmospheres and non-fire
atmospheres containing mainly CO can cause lethality due exclusively to CO at COHb
levels of 20%.
(2) The population of individuals who die in fires has a bimodal distribution, with an
excess of very young and very old and infirm people, while that of people who die
in automobile exhaust CO exposures has a unimodal distribution. Thus, it appears
that fire victims are significantly more sensitive to CO poisoning than non-fire victims
and may tend to die at lower COHb levels.
(3) The replacement of large amounts of natural and traditionally used materials by man-
made materials among the normal products in everyday use has made no difference
to the toxicity of fire atmospheres.
(4) The CO concentrations in the atmospheres of large-scale fires are determined by the
oxygen availability, ventilation, mass loading, and other such variables, but are very
minimally affected by the chemical composition of the fuels.
(5) Small-scale tests give excessively low CO yields while they can predict adequately
the concentrations of other combustion products. Thus, such tests cannot be used to
predict toxic fire hazard for the fire scenarios of greatest interest, ventilation controlled
flashover fires, which cause most of the fire fatalities. The use of post computational
correction for CO concentration, which is now becoming available, is required to
make these tests relevant to the critical fire scenarios. They can only legitimately,
however, be used as a part of a fire hazard assessment.
Acknowledgment
This work was carried out under the sponsorship of the Society of the Plastics Industry,
Inc.
(14) Cimbura, G., McGarry, E., and Daigle, J., "Toxicological Data for Fatalities due to Carbon
Monoxide and Barbiturates in Ontario-A Four Year Survey 1965-1968," Journal of Forensic
Sciences, 1972, pp. 640-642.
(15) Kishitani, K., "Study of Injurious Properties of Combustion Products of Building Materials at
Initial Stage of Fire," Journal of Faculty of Engineering University Tokyo (B), Vol. 31,1971, pp.
1-35.
(16) Teige, B., Lundevall, J., and Fleischer, E., "Carboxyhemoglobin Concentrations in Fire Victims
and in Cases of Fatal Carbon Monoxide Poisoning," Zeitschrift Rechtsmedizin, Vol. 80, 1977, pp.
17-21.
117) Consumer Product Safety Commission, Federal Register, Vol. 45 No. 182, 17 Sept., 1980, p. 61880.
(IS] Pach, J., Cholewa, L., Marek, Z., Bogusz, M., and Groszek, B., "Analysis of Predictive Factors
in Acute Carbon Monoxide Poisonings," (Toxicological Clinic and Institute of Forensic Medicine,
Krakow, Poland,) Folia Medica Cracoviensia, 1978, pp. 159-186.
119] Berl, W. G. and Halpin, B. M., "Human Fatalities from Unwanted Fires." Fire Journal, Sept.
1979, pp. 105-123.
(20) Halpin, B. M., Radford, E. P., Fisher, R., and Caplan, Y., "A Fire Fatality Study," Fire Journal,
May 1975, pp. 11-\3.
(2/) Anderson, R. A., Watson, A. A., and Harland, W. A., "Fire Deaths in the Glasgow Area: I.
General Considerations and Pathology," Medical Science Law, Vol. 21, 1981, pp. 175-83.
(22) Anderson, R. A., Watson, A. A" and Harland, W. A., "Fire Deaths in the Glasgow Area: II.
The Role of Carbon Monoxide," Medical Science Law, Vol. 21, 1981. pp. 288-94.
(23) Birky, M. M., Malek, D., and Paabo, M., Journal of Analytical Toxicology, Vol. 7,1983, p. 265.
124] Gormsen, H., Jeppesen, N., and Lund, A., 'The Causes of Death in Fire Victims," Forensic
Science International, Vol. 24, No.2, 1984, pp. 107-111.
(25) Wald, N., Howard, S., Smith, P. G., and Bailey, A" "Use of Carboxyhemoglobin Levels to
Predict the Development of Diseases Associated with Cigarette Smoking," Thorax, Vol. 30,1975,
pp. 133-40.
126] Babrauskas, V., Harris, R. H., Gann, R. G., Levin, B. C, Lee, B. T., Peacock, R. D., Paabo,
M., Twilley, W., Yoklavich, M. F., and Clark, H. M., "Fire Hazard Comparison of Fire-Retarded
and Non-Fire-Retarded Products," NBS Special Publication 749, National Institute of Standards
and Technology, Gaithersburg, MD, July 1988.
127] Pitts, W. M., "Executive Summary for the Workshop on Developing a Predictive Capability for
CO Formation in Fires," NISTIR 89-4093, National Institute of Standards and Technology, Gaith-
ersburg, MD, 1989.
(.?S) Babrauskas, V., Harris, R. H., Braun, E., Levin, B. C, Paabo, M., and Gann, R. G., 'The Role
of Bench-Scale Data in Assessing Real-Scale Fire Toxicity," NIST Technical Note 1284, National
Institute of Standards and Technology, Gaithersburg, MD, 1991.
REFERENCE: Watts, J. M., "Fire Risk Rating Schedules," Fire Hazard and Fire Risk As-
sessment, ASTM STP 1150, Marcelo M. Hirschler, Ed., American Society for Testing and
Materials, Philadelphia, 1992, pp. 24-34.
ABSTRACT: Fire risk rating schedules (FRRS) are heuristic models of fire safety. They
constitute various processes of modeling and scoring fire hazards and other risk parameters
to produce a rapid and simple estimate of relative risk. Selected examples of FRRSs are briefly
described to illustrate the nature and diversity of these methods. Following the examples,
general characteristics and common components of FRRSs are discussed. The analysis in this
paper is intended to lead to a more rational and consistent approach in the development and
evaluation of fire risk rating schedules.
KEY WORDS: fire, fire hazard, fire risk, fire risk analysis, fire risk assessment, fire safety,
rating schedules
Fire risk rating schedules (FRRSs) are heuristic, quasi-mathematical paradigms of fire
safety. They constitute various processes of modeling and scoring hazard and safety param-
eters to produce a rapid and simple estimate of relative risk. FRRSs are useful and powerful
tools that can provide valuable information on the risks associated with fire. They have
been applied to a variety of hazard and risk assessment projects to reduce fire safety costs,
set priorities, and facilitate use of technical information. FRRSs provide an important link
between the complex scientific principles of theoretical and empirical models, and the less
than perfect circumstances of non-laboratory conditions found in real world applications.
There is significant diversity in the formation and use of schedule approaches to fire safety,
yet there is very little information on methodology or evaluation. This leads to a lack of
structure or consistency in the development process, and a lack of validity or credibility in
the outcomes. This paper presents some results from a survey and study to identify char-
acteristics and components of, and criteria for, FRRSs. The approach of the study is to
locate, review, and evaluate existing FRRSs and their methods of development. Critiquing
these methods has formulated concepts to help guide future development and evaluation.
Examples described in this paper concentrate on the more technically sound approaches
which relate most closely to development of fire standards. These examples are typically
more robust than most of the schedules surveyed and the generalized criticisms of FRRSs
do not necessarily apply to the examples cited.
Following the examples, general characteristics and common components of FRRSs are
described. The analysis of these features is intended to suggest a systematic approach to the
formulation and validation of such schedules.
Examples
FRRSs come in a large variety of formats and a broad spectrum of purposes. Examples
used in this paper to illustrate the principles of FRRSs are selected for their relationship to
fire standards development.
'Director, Fire Safety Institute, PO Box 674, Middlebury, VT 05753.
24
WATTS ON FIRE RISK RATING SCHEDULES 25
Upholstered Furniture Heat Release Rates

Babrauskas (1] and Babrauskas and Walton [2] present a simplified rule for estimating
the maximum or peak heat release rate of upholstered furniture based on design factors
(Eq 1). The most accurate determination of heat release rate involves destructive testing in
a furniture calorimeter. The approximation rule uses non-destructive measures and generic
design categories which are readily determined in the field. These factors are multiplied
together implying interdependent contributions to the magnitude of the heat release rate.
Qpeak = 210.0 [fabric factor] x [padding factor] x [mass]
x [frame factor] x [style factor] (1)
where
[fabric factor] 1.0 for thermoplastic fabrics (fabrics such as polyolefin, which melt prior
=
to burning),
0.4 for cellulosic fabrics (cotton, rayon), and
0.25 for pye or polyurethane film type coverings.
(padding factor] = 1.0 for polyurethane foam or latex foam,
0.4 for cotton batting,
1.0 for mixed materials (that is, both polyurethane or latex foam and
cotton), and
0.4 for neoprene foam.
[mass] combustible mass, in kilograms.
[frame factor] = 1.66 for non-combustible,
0.58 for melting plastic,
0.30 for wood, and
0.18 for charring plastic.
[style factor] = 1.0 for plain, primarily rectilinear construction, and
1.5 for ornate, convoluted shapes and intermediate values for inter-
mediate shapes.
_ the constant 210.0 has units kW/kg.

The concept of an FRRS is to use readily determinable parameters, combined to yield a
single value which can be associated with fire risk. A significant facet of this type of ap-
proximation is the mathematical distinction between risk, or the value being approximated,
and the result of the approximation. There is not a one-to-one mapping of the model output
to the real world. In this example, while the rate of heat release for a given mass has an
infinite number of possible values, the rate of heat release for a given mass approximated
by Eq 1 has only 40 possible discrete values.
Assessing the Fire Risk of Products

In a brief paper, Gross [3] describes a process which illustrates several important concepts
of FRRSs. The suggested approach to assessing fire risk of products involves the following
five steps:
• Identify appropriate fire response characteristics.

• Identify appropriate measure(s) of the fire response characteristics.
• Normalize measures to a consistent scale.
• Assign weights indicating relative importance .

• Combine to yield assessment of fire risk.
Steps one and two are largely dependent on experience and expert judgment. Step three
uses, as example, a scale of 0 to 100, with the performance limit of a fire response char-
acteristic test equated to 100. Step four relies on fire incidence statistics and expert judgment.
Step five is not discussed by Gross.
As an illustration of this approach, Gross uses the evaluation of fire response characteristics
of curtains and drapes as shown in Table 1 below.
The NFP A 701 large-scale test is considered an appropriate test for ignitability and flame
spread and is weighted as having 90% of the relative importance. A maximum char length
of 250 mm is considered the performance limit in the test; hence, the factor of 0.4 is used
to normalize results to the 0 to 100 scale. Similarly, for a performance limit of 800 D m
(specific optical density) in ASTM E 662, a normalizing factor of 0.125 is used. Although
not discussed by Gross, one can envision how the products of the weighting factors and
normalized test results could be summed to give a fire risk index.
Assessment of Fire Risk by Occupancy Classification

ASTM Standard Practice for Assessment of Fire Risk by Occupancy Classification (E
931) presents a method for assessment of fire risk inherent in different occupancies as
essential to the development and use of fire standards for furnishings. The stated purpose
is twofold and corresponds to basic objectives of FRRSs:
1) provide a uniform procedure for assessment of fire risk; and

2) provide a logical basis for establishing levels of fire performance.
The approach is based on consideration of twelve parameters relating to the risk of death
and injury, property loss, and fire potential. The rater assigns each parameter a value
according to the degree of risk present in the occupancy. Each parameter has also been
assigned a fixed weight.
The product of the weight and risk rating yields the parameter total. Fixed multipliers
are provided for fire detection and fire suppression systems where deemed appropriate. The
sum of the parameter totals for death and injury and property loss times the total for fire
potential, provides a risk assessment index. This number is intended as a basis for both fire
risk comparisons and for placement of occupancies within risk classes. The fire risk assess-
ment rating form is illustrated in Fig. 1.
Decision Logic for Trading Between Fire Safety Measures

Harmathy [5] presents a method of quantitative decision support concerning the equiv-
alence of various fire safety alternatives. This is a comprehensive approach which involves
four distinct parts.
TABLE 1- Evaluation of fire response characteristics of curtains and drapes.
Fire Response Characteristic Test Method Normalizing Factor Weighting Factor
Ignitabilitylflame spread NFPA 701 [4] 0.4 0.9

Smoke generation ASTM E 662a 0.125 0.1
aStandard Test Method for Specific Optical Density of Smoke Generated by Solid Materials.
• A probabilistic fire loss model which yields a measure of fire risk in terms of dollars
per year per square meter of floor area.
• A data set to evaluate expected fire loss in a standard reference building.
• A set of parameters and conditions which have an impact on expected fire loss.
• A data set which quantitatively defines the effect of parameter conditions on expected
fire loss.
The probability model (Eq 2) is a straightforward analysis of likelihood that a fire will
pew or spread to a certain stage. The fire loss expected in each stage for both property
and life is also analyzed. Data for this model is gleaned from North American statistics on
fire occurrence, fire spread, and fire losses. The parameters and their corresponding quan-
titative impacts were derived through a Delphi exercise.
where
L - fire loss (human and property) expectation (dollars per year per m" of floor area);
N = expected number of fire incidents (per year per m" of floor area);
Pp = probability that, given ignition, the fire will not reach flashover;
P FN = probability that, given flashover, the fire will not spread to other compartment(s);
PFSD = probability that, given flashover, the fire will spread to other compartment(s) by
destruction;
lp = average loss (human and property) resulting from fires that do not reach flashover
(dollars per incident);
IFN = average loss (human and property) resulting from postflashover fires that do not
spread (dollars per incident);

IFsD - average loss (human and property) resulting from postflashover fires that spread
by destruction (dollars per incident);
IFsc = average loss (human and property) resulting from postflashover fires that spread
by convection (dollars per incident).
Decoupling the probability model from the parameter evaluation is an important feature.
It illustrates the use of relatively simplistic schedule formats to provide input for more
rigorous risk assessment models.
Table 2 shows the parameters and their associated quantitative values. Column two rep-
resents the incremental impact of a parameter on the probability that, given ignition, the
fire will not reach flashover. Column three is the incremental impact on the probability that,
given flashover, the fire will not spread to other compartments. For example, if rooms are
larger than 20 m2, the probability of no flashover is increased by 6.4%, and the probability
of no fire spread to another compartment is decreased by 12.4%.
These four examples were selected for their relationship to ASTM E 5 activities. They
are generally more technically sound than many other FRRSs presently in use. Review of
additional examples of schedules that have achieved considerable usage may be found in
the SFPE Handbook of Fire Protection Engineering [6].
Generalized Characteristics of Fire Risk Rating Schedules

Fire risk rating schedules have been referred to by various designations such as index
systems [7]. numerical grading [8], and point schemes [9]. They originated as insurance
rating schedules in the 19th century, but in the last few decades, the basic concepts have
appeared in a wide variety of formats. FRRSs have proliferated because of the high utility
in their relative ease of application. However, they lack validity due to the unspecified
nature of the selection of variables and their relationships.
Characteristics of FRRSs identify their diversity and usefulness. They are described here
in terms of their functional relation to other risk evaluation methods and in terms of their
applications and attributes.
Function of FRRSs
The function of an FRRS is, explicitly or implicitly, fire risk analysis. By nature of the
circumstances, fire safety decisions often have to be made under conditions where the data
TABLE 2-lncremental effect of various factors on probabilities Pp and PFN•

Factor i (ilP,.),I(P,.)" (ilP'N),I(P'N),.
Occupancy
dwellings, apartments 0 0
hotels, motels - 0.027 0.128
health care buildings 0.090 0.246
office buildings 0.064 0.149
educational buildings 0.078 0.204
Building height
I or 2 storys 0 0
3 to 8 storys -0.016 - 0.109
more than 8 storys - 0.057 - 0.230
Average floor area of compartments
less than 20 mO 0 0
more than 20 mO 0.064 -0.124
Furniture arrangement
sparse 0.107 0.123
normal (as in a dwelling) 0 0
dense -0.148 -0.147
Combustible lining in compartments
not used 0 0
used -0.213 -0.238
Combustible lining in corridors
not used 0 0
used 0 - 0.435
Combustible lining on building exterior
not used 0 0
used 0 -(J.211
Smoke detector(s)
not used 0 0
used (J.118 0.268
Sprinkler system
not used 0 0
used 0.196 0.479
Building pressurization
not used 0 0
used 0 0.199
Self-closing doors
not used 0 0
used 0.030 0.530
Municipality served by
full-time FD 0 0
composite FD -0.012 -0.110
volunteer FD - 0.033 -0.201
Distance from FD
less than 8 km 0 0
more than 8 km - 0.043 - 0.280
Access by FD
from one side of building 0 0
from more than one side of building 0.022 0.184
Average temperature in January
higher than O°C 0 0
lower than O°C - 0.029 - 0.140
NOTES: 1. The reference conditions are characterized by 0 increments.
2. Positive values indicate improved safety; negative values indicate decreased safety.
are sparse and uncertain. The technical parameters of fire risk are very complex and normally
involve a network of interacting components, the interactions generally being nonlinear and
multidirectional. However, complexity and sparseness of data do not preclude useful and
valid approaches. Such circumstances are not unusual in decision making in business or
other risk venues (the space program illustrates how success can be achieved when there is
little relevant data). Thus, development of approaches to fire risk analysis is a significant
ongoing process. Methods of fire risk analysis may be classified by evolutionary development
into four categories: narratives, check lists, schedules, and emulations.
Narratives- The bulk of our present day wisdom on fire safety consists of narrative
descriptions of various hazardous conditions and ways to reduce or eliminate them. These
appear in the form of building codes and various fire safety standards. Narratives do not
attempt to evaluate the fire risk quantitatively, rather, a risk is judged acceptable if it complies
with published recommendations. An obvious limitation to this approach is that such nar-
ratives can not hope to be interpreted and applied uniformly for the myriad conditions of
human activity. While there is much common ground among different fire hazard situations,
in detail there is considerable variation.
Check lists- A common accessory of fire safety is a listing of hazards and recommended
practices. While narratives take the form of a book, check lists represent a table of contents,
identifying complex fire safety concepts with a few words. These checklists comprise valuable
tools for identifying fire risk factors. They do not, however, distinguish among the importance
of these factors. For example, the relative value of hydrants, sprinklers, and manual fire
extinguishers is not constant, but a function of other features of a building's form and
function.
Schedules- In general, fire risk rating schedules assign values to selected variables based
on professional judgment and past experience. The selected variables represent both positive
and negative fire safety features and the assigned values are then operated on by some
combination of arithmetic functions to arrive at a single value. This single value can be
compared to other similar assessments or to a standard.
Emulation-Growing interest in analytical fire risk assessment and an increasing data
base, has led to use of more sophisticated mathematical techniques. Emulation methods
manipulate fire safety variables according to recognized theoretical principles. Among these
approaches are: computer simulation, linear regression, network analysis, and stochastic
modeling.
While both schedules and emulations are numerical approaches to fire risk assessment,
they represent different levels of depth and accuracy. FRRSs can be considered to be at
one end of a continuous quantitative assessment spectrum, where hazard and exposure are
estimated through the use of simple models. A detailed probabilistic risk analysis would be
at the other end of this spectrum, where hazard and exposure are tested, measured, and
assessed as rigorously as possible. Choosing the depth of the risk analysis is a critical decision
that depends on such factors as time and resource committment and the intended use of
the results.
Applications
For many situations where a quantitative risk assessment is desirable, an in-depth theoretic
analysis may not be cost-effective, nor appropriate. This could be the fundamental case
where greater sophistication is not required, where prioritization is the principal objective,
or where it is necessary to institutionalize for a wide base of usage.
The level of accuracy demanded for a fire risk assessment is not the same as for other
engineering purposes; often establishing an order of magnitude will suffice. Time and re-
source expenditure increases as the depth of analysis is increased. In an age where resources
are scarce, and efficiency is prized, maximizing the utility of the FRRS is clearly desirable
for the many situations where assessing fire risk is fundamental.
FRRSs have gained widespread acceptance as cost-effective prioritization and screening
tools for fire risk assessment. Risk assessment can be an expensive and labor-intensive
process, and much time and money c,!n be wasted if the products or facilities with the
~atest potential for risk and associated liability are not identified and assessed first. Without
a prioritization plan it will not be known whether a risk was worth assessing until the time
and money has been spent. In order to be cost-effective, a prioritization system must be
simple, rapid, and accurate. An appeal of FRRSs is that it can be all of these.
Perhaps the most common implicit justification of FRRSs is the need for a simplistic
process of fire risk analysis. In most applications, the target of a FRRS is a broad class of
products or facilities for which a detailed fire risk analysis of each individual case is not
ieasible. FRRSs have appeal to administrators charged with risk management decision-
making responsibilities, but who may be unfamiliar with the details and mechanics of the
risk assessment process. Wide spread implementation of a generalized approach to fire risk
is contingent on its appeal to a broad class of users including architects, code officials, and
property managers.
Attributes
From their range of application and relationship to other methods of fire risk assessment,
we abstract the following general attributes of FRRSs:
• Inclusive-FRRSs tend to treat fire phenomena as part of a larger system rather than
in terms of individual elements. That is, they tend to address a broad, systemic concept
of fire safety.
• Numerical-The objective of FRRSs is to provide a logical and consistent approach
to fire assessment through a mathematically derived ordinal grading ..
• Heuristic-FRRSs are relatively simple models which do not rely exclusively on dem-
onstrated principles of physical or management science (but their credibility is enhanced
to the extent they do employ such principles).
• Experiential-Input and structure is largely based on experienced judgment, supported
where possible or appropriate by historical data, fire tests, and more rigorous predictive
modeling.
• Surrogate-Output is usually a measure of relative risk with no significant physical
units (but it may be input to a model of absolute risk).
Components
Components of fire risk rating schedules identify elements they have in common. Most
FRRSs are found to have three basic components [8]: a list of parameters, procedures for
assigning values to the parameters, and relationships which define mathematical operations
on the parameter values to produce an assessment of hazard or risk. Various aspects and
ramifications of these three components are further discussed.
Parameters
Parameters of FRRSs, also referred to as elements, factors, variables, and so forth, identify
the ingredients of fire safety. Fire safety is a complex system with an inordinately large
number of factors which may affect it. These can range from ignitability of personal clothing,
to availability of a heliport for evacuation. It is computation ally feasible to deal with only
a relatively small number of variables. Therefore, it becomes necessary to reduce the large
number of variables to an appropriate subset.
It is intuitively appealing to postulate that safety from fire is a Paretian phenomenon in
that a relatively small number of factors account for most of the problem. This is supported
by general fire loss figures which suggest that a small number of factors are associated with
a large proportion of fire deaths.
It is necessary then, to identify some defensible combination of factors which account for
an acceptable portion of the fire risk. Such a process is not identified for most FRRSs.
Selection of parameters is apparently most often arbitrary, with correspondingly disparate
results. Even when some parameters seem comparable, their usage may differ and lack of
operational definitions leads to confusion in trying to implement the schedule.
Where this subject is addressed, the approaches to selection of parameters generally fall
into one of three categories:
• Delphi, or some less formal consensus process which relies on expert judgment.
• Fire scenarios, ideally based on loss statistics, but usually employing subjective opinion.
• Cut set of a hierarchical, success tree, providing an inclusive list.
Values
The second component of FRRSs is the quantitative measures that associate fire risk with
qualitative characteristics of the parameters. Value selection for the parameters is where
FRRSs show the greatest range of variation. Sources for these values range from fire test
data to hearsay.
The approach to selection of values can be either objective or subjective. The main
criticism of objective estimates is that they may be using past data which are not relevant
to the future conditions being considered. For some decisions, the subjective approach may
be superior when it takes into account both historical data and the decision maker's as-
sessment of present and future influences. The problem with subjective values is that, like
selection of parameters, they are often arbitrary. Although it is possible that much thought
went into the determination of parameter values in the FRRSs reviewed, in many cases it
is not evident. Where a process is identified, it takes the form of a modified Delphi exercise
or a less rigorous method of eliciting experienced judgment.
In some FRRSs, two sets of values are associated with each parameter, intensity and
importance, for example, Gross [3] and ASTM E 931. Intensity is a measure of the amount
or degree that a parameter is present in a specific application, for example: fuel load, flame
spread rating, and so forth. Importance is a weight indicating the influence or significance
of the parameter to fire safety. This is one of the notable differences between FRRSs and
check lists, which treat all parameters equally. In many FRRSs intensity and importance
are combined implicitly into a single dimensionless value.
Since the process for selecting values is undocumented in most FRRSs, there is no way
to update the schedule to account for technological change. Where Delphi or similar proc-
esses are used, they are generally too cumbersome to allow the schedule to be responsive
to new developments.
Relationships
Relationships are the mechanisms by which the parameter values are combined to yield
a measure of fire risk. In general, the intensity measure is multiplied by the importance
weight for each parameter and the results are summed, for example, Fig. 1. Most often,
safety parameters are assigned positive values and hazard parameters are assigned negative
values. This algebraic addition implies the parameters are all independent in their effect on
fire risk.
An alternative approach uses the product of all safety parameters divided by the hazard
parameters. Multiplication of parameter values implicitly assumes that all the parameters
are completely interactive. That is, the impact on fire risk of any unit change in a single
parameter will be dependent on the values of each of the other parameters.
Most of the FRRSs studied do not explicitly address the interaction of parameters. Yet
it is intuitive that certain combinations of parameters, for example, smoke detection and
automatic suppression, are not simply additive in their associated effect on fire risk.
Linearity is implicitly assumed by the relationships employed in all the methods studied.
Summary
Fire risk analysis involves a large number of multifarious factors which are difficult to
assess in a uniform and consistent way. The analysis of such complex systems is difficult but
not impossible as evidenced by activities in areas such as nuclear safety and environmental
protection. Detailed risk assessment can be an expensive and labor intensive process and
there is considerable scope for improving the presentation of results. Fire risk rating sched-
ules can provide a cost-effective means of risk evaluation which is sufficient in both utility
and validity.
With all their potential value, it is unfortunate that so little attention has been directed
toward the development process of fire risk rating schedules. While much effort may have
been expended, it is not often evident and the results frequently suggest a shortage of
analytical skill in the development process. They typically lack even a statement of underlying
assumptions, much less any other information that would be conducive to a validation. It
is paradoxical that while seeking logic and consistency in application, they show little in
development.
Some notable exceptions to these generalities have been presented here as examples. This
paper also presents generalized descriptions of the characteristics and components of fire
risk rating schedules in an effort to promote a more rational and consistent approach to
their development and use. The importance of FRRSs as tools of fire risk analysis ensures
their continued popularity. Hopefully, their general credibility will be improved through
increased consideration in formulation and evaluation.
References
(1] Babrauskas, V., "Upholstered Furniture Heat Release Rates: Measurements and Estimation,"
Journal of Fire Sciences, Vol. 1, 1983, pp. 9-32.
(2] Babrauskas, V. and Walton, W. D., "A Simplified Characterization of Upholstered Furniture Heat
Release Rates," Fire Safety Journal, Vol. 11,1986, pp. 181-192.
(3] Gross, D., "The Use of Fire Statistics in Assessing the Fire Risk of Products," INTERFLAM '85
Conference Workbook, 1985, pp. 11-18.
(4] Methods of Fire Tests for Flame Resistant Textiles and Films, NFPA 701, National Fire Protection
Association, Quincy, MA, 1989.
(5] Harmathy, T. Z., et aI., "A Decision Logic for Trading between Fire Safety Measures," Fire and
Materials, Vol. 14, 1989, pp. 1-10 ..
16] Watts, J. M., Jr., "Fire Risk Assessment Schedules," SFPE Handbook of Fire Protection Engi-
neering, Section 4, Chapter II, National Fire Protection Association, Quincy, MA, 1988, pp. 4-89
to 4-102.
(7] Rosenblum, G. R. and Lapp, S. A., "The Use of Risk Index Systems to Evaluate Risk," Risk
Analysis: Setting National Priorities, Proceedings of the Society for Risk Analysis, Houston, TX,
1987.
[8] Nelson, H. E., "Overview: Numerical Grading Systems," Report from the 1987 Workshop on
Analytical Methods for Designing Buildings for Fire Safety Design, National Academy Press,
Washington, DC, 1988.
[9] Rasbash, D. J., "Approaches to the Measurement and Evaluation of Fire Safety," presented to
the Third International Fire Protection Engineering Institute, Wageningen, The Netherlands. 1980.
Use of Fire Tests for Fire Hazard
Assessment
REFERENCE: Keltner, N. R., Acton, R. U., and Gill, W., "Evaluating the Hazards of Large
Petrochemical Fires," Fire Hazard and Fire Risk Assessment, ASTM STP 1150, Marcelo M.
Hirschler, Ed., American Society for Testing and Materials, Philadelphia, 1992, pp. 37-43.
ABSTRACT: Fire testing and fire research would be aided by the standardization of the fire
environment, test articles, sensors, and procedures for the sake of reproducibility and inter-
laboratory comparisons. Current and proposed fire standards dictate control by either "fire
temperature" or by "fire heat flux". This equivocation can lead to confusion. This article
discusses the circumstances under which temperature may be the more important control
parameter and when heat flux may be the more important control parameter. Examples are
given.
KEY WORDS: fire testing, fire research, fire temperature, fire heat flux, fire standards, fire
control, fire comparisons, hydrocarbon fires, petrochemical fires
Fires that might occur in a transportation accident involving hazardous materials or in a

petrochemical industry accident can put workers or the public at risk. In an attempt to
IUlderstand, and possibly reduce these hazards, there is a strong interest in determining the
response and/or survivability of a variety of items when subjected to large petrochemical
fires. This effort usually involves actual testing as opposed to a purely analytical evaluation
due to complexities of modeling heat transfer in this fire environment.
Most of the fire test work done in the past in the United States, and indeed around the
world, has involved fire protection of materials and/or assemblies for buildings, not petro-
chemical fires. Standard test methods were developed by voluntary standards groups, such
as the American Society for Testing and Materials (ASTM) Committee E-5 on Fire Stan-
dards, the National Fire Protection Association (NFPA), and the International Standards
Organization (ISO). Because the testing is part of the process of developing and rating the
materials, large numbers of tests are conducted with relatively simple test assemblies. The
tests are typically conducted in a furnace. The thermal exposure is usually based on con-
trolling the furnace to a time-temperature curve, such as that given in ASTM Method for
Fire Tests of Building Construction and Materials (E 119). It is understood that this type
attest is not a simulation of a compartment or building fire environment; rather, experience
shows that performance of materials in these tests is a good indicator of performance in
actual fires [1].
For petrochemical fires, the approach has been to develop test specifications that are
intended to be reasonable simulations of the fire environment in the use situation, or to
produce damage greater than that expected to occur in a large fraction of the accidents.
Organizations around the world have made an effort to develop standardized test methods
for "hydrocarbon fires".
'Supervisor, senior members of technical staff, respectively, Sandia National Laboratories, Thermal
Test and Analysis Division, 2737, Albuquerque, NM 87185.
37
In the transportation area, national and international agencies have developed regulations
for testing of radioactive material (RAM), shipping containers (for example, United States
Nuclear Regulatory Commission, US-NRC and International Atomic Energy Agency, IAEA),
and liquefied petroleum gas (LPG) tankcars (for example, United States Department of
Transportation, US-DOT). The regulatory specifications that have been given by the IAEA
and the US-NRC are relatively simple in that a "uniform fire" boundary condition is used.
The US-NRC regulation, lOCFR Part 71.73, specifies exposure " ... to a heat flux not less
than that of a radiation source of 800°C with an emissivity coefficient of at least 0.9 ... ,
... the surface absorptivity must be either that value which the package may be expected
to possess if exposed to a fire or 0.8, whichever is greater." This regulation does not specify
the fire temperature; it specifies that the minimum acceptable heat flux to a cold wall is
55.5 kW/m2• Specifications have been given by the US-DOT, in 49CFR Part 179.105-4, for
the evaluation of LPG tank car thermal insulation systems. These specifications call for a
flame temperature of 870°C ± 55°C and, additionally, an average heat flux ranging from
35.2-30.2 kW/m2 is required for a 15.9-mm steel calibration sample to reach a temperature
of 430°C within a 12 to 14 min period.
The oil and natural gas industries are interested in test methods for evaluating the pro-
tection of personnel and large facilities such as refineries and offshore platforms [2]. Test
specifications are under development for these applications by groups such as ASTM and
the Norwegian Petroleum Directorate.
During the development of these test specifications, there has been a lot of discussion
about the relative importance of controlling either the fire temperature or the fire heat flux
[2]. The current draft of a test method for evaluating fire protection materials (under
development by ASTM Committee E-5) calls for the fire temperature to reach approxi-
mately 1l00°C in 5 min. It also calls for the heat flux to a "cold wall" to reach 158 kW/m2
in the same period. The heat flux is to be measured with a calibration specimen. An interim
method developed by the United Kingdom and Norway calls for the temperature to reach
880°C within 3 min, 1033°C after 10 min, and a peak temperature of 1l00°C in 2 h [3,4].
The testing methods allowed in these different specifications may involve furnaces (for
example, US-NRC), gas fires (for example, Mobil pit fire, US-DOT torch fire, and so forth),
or pool fires (for example, IAEA, US-NRC, and so forth); the thermal exposure is intended
to simulate that which occurs in large pool fires. Because they are expensive, only a limited
number of large or full-scale tests have been conducted [5-10]. In an effort to reduce costs,
models have been developed for predicting the heat transfer in fires so that the number of
tests required for qualifying a particular item is reduced [11-13]. Tests of components or
subsystems are used to help predict the performance of the full systems. Scale model tests
have also been tried [9]. The success of any of these approaches has been limited by the
lack of a complete understanding of the heat transfer mechanisms occurring in a fire.
Considerable work has been done in the recent past to study heat transfer in fires and
fire tests. A number of studies of furnace tests have been made; two very useful studies are
Refs 14 and 15. Transportation accidents generally assume that there is a pool fire. Studies
of the heat transfer in pool fires have been made for RAM shipping containers and other
systems; some examples for RAM containers are Refs 6,9,10, and 16, for LPG tank cars
Ref 5, and for refineries and offshore oil platforms, Refs 2 and 17.
Heat Transfer in Petrochemical Fires and Fire Tests

Most considerations of the fire test environment deal with the fire temperature. Tem-
perature is important because material operating limits are directly related to temperature
or to time above some temperature. However, the heat flux is just as important. Because
KELTNER ET AL. ON LARGE PETROCHEMICAL FIRES 39
die initial response of the test item is governed by the heat flux level, the heat flux levels
8dicate how severely an item will be thermally stressed. Thermal shock, such as can occur
ill a large pool or spill fire, can be important. Unfortunately, it has been found that there
i5 not a one-to-one correspondence between the average heat flux and the average tem-
perature in a fire [18].
Figure 1 demonstrates this lack of correspondence with a compilation of data from 9-m
~ I8-m lP-4 pool fire tests. Absorbed heat fluxes or "hot wall heat fluxes" are obtained
from temperature histories of the calorimeters. The values shown are "cold wall heat fluxes";
1bese include a correction which accounts for reradiation from the surface of the calorimeters.
The flame temperatures were measured with sheathed thermocouples mounted close to the
calorimeters. The averages are made over the duration of the tests, which is typically 30
min. The data show that there is a significant scatter of the heat flux values about the curve
which shows the blackbody heat flux at the average temperature. Although the reasons for
the scatter are not completely understood, additional details and discussion are given in
Ref 8.
Measurements made in these large pool fires using calorimeters 0f different sizes and
shapes have indicated that the heat transfer to the calorimeter is affected by the physical
size and shape of the calorimeter, as well as by its "thermal massiveness." The temperature
rise of a thermally massive device is small for a given heat input; some examples would be
a thick steel plate or a wetted tank wall. As the physical size and the thermal massiveness
of the calorimeters increases, the peak heat flux decreases. Figure 2 shows the average heat
flux versus the surface temperature at the bottom station for three sizes of cylindrical
calorimeters mounted horizontally in the fires. Due to the size differences, there should be
significant differences in the convective heat transfer; however, the stagnation point heat
transfer was estimated to be 8.5, 22, and 32 kW/m2 for the IA-m, 20-cm, and lO-cm
calorimeters, respectively [19]. Thus, the roughly 20% difference shown in Fig. 2 at ap-
proximately 200°C cannot be accounted for solely by convective heat transfer differences.
Part of the difference is believed to be due to heat transfer to the device cooling the adjacent
layer of combustion products; this layer blocks a fraction of the incoming radiation [20,21].
These studies demonstrate the difficulty of predicting the heat flux in a petrochemical fire.
It can be seen that the temperatures and heat fluxes called out in the ASTM and United
Kingdom-Norway test specifications are at the upper end of the measurements. Schneider
and Kent [7] show that under low wind conditions the average temperatures rarely exceed
1l000C. A recent study by Mansfield and Linley [22], shows that the llOOoe average tem-
perature may be exceeded for short periods of time in low wind conditions. Overall, infor-
mation available in the literature would indicate that tests run to these specifications would
produce damage greater than that expected in most fires.
Comparing the Response of Different Items

Measurements made in our large fires indicate that neither temperature or heat flux
measurements alone are sufficient to characterize the fire test. To examine this further,
finite difference models of three hypothetical test samples were set up using the SINDA
code [23]. The models were for a 305-mm mild steel cylinder, a pipe of the same outside
diameter with a 1O.3-mm wall, and the pipe with 51-mm-thick insulation covered by a thin
steel skin (0.9 mm). The boundary condition involves black body radiation from a source
with time varying temperature. The temperature rises to 812°e in 3 min and to 1095°e after
5 min.
The surface temperatures calculated for the three models are shown in Fig. 3. It can be
seen that the surface temperature of the insulated pipe tracks the fire temperature very
closely. The absorbed heat fluxes are shown in Fig. 4. Because the surface temperature of
the insulated pipe tracks the fire temperature, the absorbed heat flux is nearly zero after
. just a few minutes. For devices like this, the fire temperature is the more important parameter.
The solid cylinder responds slowly enough that the absorbed heat flux at 5 min. approaches
the blackbody heat flux at the peak fire temperature of 199 kW/m2• It takes approximately
60 min for the surface temperature to be within lOoDe of the fire temperature. At this point,
the total heat transfer is still over 25% of the peak heat flux. For devices like this, the heat
flux is the more important parameter.
The bare pipe represents an intermediate case where both the temperature and the heat
flux are important. During the first 5 min, the heat flux tracks fairly close to that for the
solid cylinder. Then it begins to fall as the temperature approaches the fire temperature.
After approximately 20 min, it tracks the fire temperature.
These different response characteristics have been utilized in making measurements in
fires. The behavior of the insulated pipe, which is similar to that of the shrouded plate
calorimeter in Fig. L is the basis of the design of sensors to measure the "effective fire
temperature," such as the Directional Flame Thermometer described by Fry [24] and the
Plate Thermocouple described by Wickstrom [25]. Devices like the pipe or cylinder have
been used to make hot wall heat flux measurements in fires because they respond fairly
slowly [8,19].
Discussion
Test results and analyses indicate that there are cases when the fire temperature may be
the more important control parameter and other cases when fire heat flux is the more
important parameter. Furthermore, it has been found that there is not a one-to-one cor-
respondence between the average heat fluxes and the average temperatures measured in
large pool fires.
These results could have a significant impact on the formulation of test specifications. For
example, in the testing of fire protection materials, the test material generally has a relatively
low thermal inertia (thermal conductivity, density, specific heat). When the material is
exposed to a fire, the surface heats very quickly and tracks the fire temperature. In this
case, the testing method called out in a specification should focus on the temperature as it
is probably the more important parameter. On the other hand, a large thermally massive
object, such as a wetted wall in a storage tank, will respond very slowly to a fire. In this
case, the heat flux is the more important parameter as the surface may never even approach
the fire temperature and the object itself may affect that heat transfer; the test method here
should focus on the heat flux.
While the focus might vary depending on the item under test, good practice suggests that
both the temperature and heat flux should be determined to characterize the hazard posed
by the fire test environment. In most cases, the fire test environment can not be adequately
characterized by using just one of these parameters.
Acknowledgment
Sandia National Laboratories is operated by AT&T for the United States Department of
Energy (USDOE) under Contract DE-AC04-76DP00789.
References
[1] Harmathy; T. Z., "The Fire Resistance Test and Its Relation to Real World Fires," Fire and
Materials, Vol. 5, No.3, 1981, pp. 112-122.
[2] Shipp, M. P., A Hydrocarbon Fire Standard-An Assessment of Existing Information, OT/R/
8294, Fire Research Station, Building Research Establishment, Borehamwood, United Kingdom,
Jan. 1983.
KELTNER ET AL. ON LARGE PETROCHEMICAL FIRES 43
(3] Olstad, H. T., "Norwegian Petroleum Directorate's Requirements to Passive Fire Protection
Materials," International Conference on Offshore Passive Fire Protection, The Plastics and Rubber
Institute, London, 31 March 1987.
(4] Danielsen, V., "Spalling Effects on High Strength Concrete in High Intensity Fires-Protection
Methods," American Concrete Institute Convention, Orlando, FL, March 1988.
(5] Anderson, C., Townsend, W., Zook, J., and Cowgill, G., "Effects of a Fire Environment on a
Rail Tank Car Filled with LPG," Report FRA-OR&D 75-31, V.S. Department of Transportation,
Federal Railroad Administration, Washington, DC, Sept. 1974.
(6] Gregory, J. J., Mata, R., Jr., and Keltner, N. R., "Thermal Measurements in a Series of Large
Pool Fires," SAND85-0196, Sandia National Laboratories, Albuquerque, NM, Aug. 1987.
(7] Schneider, M. E. and Kent, L. A., "Measurements of Gas Velocities and Temperatures in a Large
Open Pool Fire," Fire Technology, Vol. 25, No.1, Feb. 1989, pp. 51-80.
(8] Schneider, M. E., Keltner, N. R., and Kent, L. A., "Thermal Measurements in the Nuclear Winter
Fire Test," SAND88-2839, Sandia National Laboratories, Albuquerque, NM, Jan. 1989.
(9] Wachtell, G. P. and Langhaar, J. W., "Fire Test and Thermal Behavior of 15-Ton Lead Shielded
Cask," TlD-4500, E. 1. DuPont de Nemours and Company, Oct. 1966.
PO] Yasuda, M., Kobayashi, S., Yamakawa, H., and Abe, H., "Results of Fire Resistance Test of
Spent Fuel Transport Container (50-Ton Type Cask) Vsing a Furnace and Investigation of a Method
of a Numerical Analysis," in Proceedings of Seventh International Symposium on Packaging and
Transportation of Radioactive Materials (PATRAM), Conference 830528, Oak Ridge National
Laboratory, Oak Ridge, TN, New Orleans, May 1983.
Ill] Bader, B. E., "Heat Transfer is Liquid Hydrocarbon Fuel Fires," Chemical Engineering Progress
Symposium Series, Vol. 61, No. 56, 1965.
Ill] Burgess, M. H., "Heat Transfer Boundary Conditions in Pool Fires," in PA TRAM '86 Proceedings,
(Davos: PATRAM '86) IAEA-SM286, Vienna, Austria, 16-20 June 1986.
(13] Fry, C. J., "A Consideration of Some Simple Models for Assessing Heat Transfer to Objects
Engulfed in Pool Fires," AEE Winfrith, AEEW-M 2321, Winfrith, Great Britain, March 1985.
(14] Babrauskas, V. and Williamson, R. B., "Temperature Measurements in Fire Test Furnaces," Fire
Technology, Vol. 14, 1978, pp. 226-238.
(15] Sultan, M. A., Harmathy, T. Z., and Mehaffey, J. R., "Heat Transmission in Fires Test Furnaces,"
Fire and Materials, Vol. 10, 1986.
(16] Bainbridge, B. L. and Keltner, N. R., "Heat Transfer to Large Objects in Large Pool Fires,"
Journal of Hazardous Materials, Vol. 20, 1988, pp. 21-40.
(17] Schonbucher, A., et aI., "Investigation of Hazardous Fires of Liquid Hydrocarbons as a Contri-
bution to the Safety of Chemical Plants," Chem-lng Tech, Vol. 57, No. 10, 1985, pp. 823-834.
(An English translation of this document is available from British Gas, Central Index of Trans-
lations, Translation BG/MRS/MI4988) ..
(18] Keltner, N. R. and Moya, J. L., "Defining the Thermal Environment in Fire Tests," Fire and
Materials, Vol. 14, No.4, Dec. 1989, pp. 133-138.
(19] Gregory, J. J., Keltner, N. R., and Mata, R., Jr., "Thermal Measurements in Large Pool Fires,"
Journal of Heat Transfer, Vol. 111, May 1989, pp. 446-454.
(..'0] Nicolette, V. F. and Larson, D. W., "The Influence of Thermally Massive Objects on Engulfing
Fire Environments," 5th AIAA/ASME Thermophysics and Heat Transfer Conference, Seattle,
WA, 18-20 June 1990.
(..'1] Keltner, N. R., Nicolette, V. F., Brown, N. N., and Bainbridge, B. L., "Test Vnit Effects on
Heat Transfer in Large Fires," Journal of Hazardous Materials, Vol. 25, Nos. 1 and 2, Oct. 1990,
pp. 33-47.
(22] Mansfield, J. A. and Linley, L. J., "A Statistical Analysis of Temperature Measurements in Large
Fuel Spill Fires," JANNAF Propulsion Systems Hazards Subcommittee Meeting, Chemical Pro-
pulsion Information Agency, Laurel, MD, March 1990.
(23] SINDA User's Manual, Network Analysis Associates, Fountain Valley, CA, Feb. 1986.
(14] Fry, C. J., "Pool Fire Testing at AEE Winfrith," PATRAM '89- The 9th International Symposium
on the Packaging and Transportation of Radioactive Materials, Conference 890631, Oak Ridge
National Laboratory, Oak Ridge, TN, Washington, DC, June 1989.
(..'5] Wickstrom, V., "The Plate Thermometer-A Simple Instrument for Reaching Harmonized Fife
Resistance Tests," SP Report 1989:03, Swedish National Testing Institute, Fire Technology, Boras,
Sweden, 1988.
REFERENCE: Hirschler, M. M., "Electrical Cable Fire Hazard Assessment with the Cone
Calorimeter," Fire Hazard and Fire Risk Assessment, ASTM STP 1150, Marcelo M. Hirschler,
Ed., American Society for Testing and Materials, Philadelphia, 1992, pp. 44-65.
ABSTRACT: A series of full-scale cable tray fire tests have been done and the results have
been expressed in terms of heat release (rate and amount), smoke release (rate and amount),
mass loss and gas emissions, as well as the standard properties of flame spread and extent of
charring. These tests were carried out in two different full-scale facilities.
The same cables have also been tested in the cone rate of heat release calorimeter, (ASTM
Method for Heat and Visible Smoke Release Rates for Materials and Products Using an
Oxygen Consumption Calorimeter, E 1354), in a horizontal orientation, and the same prop-
erties (except for gas emissions) are measured. Moreover, the combustible materials which
make up many of the cables have also been tested in both the cone calorimeter and the Ohio
State University (OSU) rate of heat release calorimeter (ASTM Method for Heat and Visible
Smoke Release Rates for Materials and Products, E 906).
The rate of heat release results of the cone calorimeter tests on cables were well correlated,
linearly, with the results of the full-scale tests. This was particularly true when the cone was
used at an incident flux of 20 kW1m'. A model has therefore been devised to predict full-scale
cable tray results.
The amount of smoke obscuration resulting from all full-scale cable tests was heavily
dependent on the extent of burning of the cables. Those cables that did not burn extensively
and released very little smoke. Similarly, those cables that did not burn extensively released
low amounts of combustion gases, notably CO and HC\.
As far as smoke release is concerned, total smoke released in the full-scale fires correlated
very well with smoke factors measured in the small-scale cone calorimeter tests. The total
smoke released in the small-scale tests, following complete sample combustion, was a much
less reliable measure of full-scale smoke release than the smoke factor.
The two small-scale rate of heat release instruments correlated well with each other, on all
properties, except for time to sustained burning.
Some fire properties of the cable jacket material alone, in the cone calorimeter at 20 kWI
m2, can be used to give a priori indication of likely cable full-scale fire performance in a
certain scenario. The properties most appropriate for this purpose are the peak rate of heat
release and the smoke factor. The OSU calorimeter was a somewhat less reliable small-scale
predictor than the cone calorimeter, based on jacket material results only.
Fire tests with the cone calorimeter can thus be used for preliminary fire hazard assessment
of electrical cables when installed in vertical cable trays.
KEY WORDS: cable tray fire test, char length, cone calorimeter, fire hazard, flame height,
ignitability, Ohio State University (OSU) calorimeter, rate of heat release, smoke obscuration,
rate of smoke release, smoke factor, total heat release, total smoke release
It has been shown, for some time now, that the rate of heat released by a product is the
most important property predicting hazard in a fire situation, because the rate of heat release
'Manager, Fire Sciences, BFGoodrich, Technical Center, Avon Lake, OH 44012; Present address;
Safety Engineering Laboratories, 38 Oak Road, Rocky River, OH 44116.
44
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 45
(RHR) controls the intensity of a fire [1,2]. Furthermore, it has been established that fire
test results from the cone calorimeter rate of heat release instrument correlate with those
from full-scale fires [3-5].
The traditional way of measuring fire properties is by determining a variety of properties
independently in various small-scale instruments. The cone calorimeter provides a means
of measuring simultaneously a variety of properties in the same experiment. If, therefore,
it could be shown that the results from the cone calorimeter correlate with those of a
particular full-scale scenario, this type of instrument would be ideally suited for use to
assess fire hazard of products in that specific scenario.
General purpose cables requiring some degree of fire performance are often assessed by
running full-scale vertical cable tray fire tests. Such tests are generally quite expensive.
There are two reasons. First, there is the cost of the tests themselves (particularly setup and
execution time), and second, it is costly to construct cables with experimental materials of
unknown compositional effectiveness.
The traditional output of standard cable tray tests is purely the extent of flame spread
due to the cables themselves, measured as the length of charring of the cables. However,
cable tray fire tests can also be used to measure other fire properties, of greater interest in
terms ottire hazard assessment [6- 11]. The main additional properties of interest are those
associated with heat and smoke release.
This work was undertaken with the following objectives:
• Investigate the full-scale fire performance of some advanced vinyl compounds, con-
structed into cables, and compare them with that of traditional vinyl compounds.
• Establish whether a small-scale laboratory test (the cone calorimeter rate of heat release
instrument [12]) can predict the fire performance in the full-scale tray test.
• Establish whether the output from two different full-scale cable tray testing facilities
could be compared satisfactorily.
It was hoped that an additional objective would be achieved:
• Determine whether rate of heat release calorimeter test results on individual compounds
in the cone calorimeter, or the OSU instruments [13] would be sufficiently reliable to
predict fire performance in full-scale cable tray tests.
Experimental Procedures
Test Methods Used
Small-Scale Tests-I) Cone Calorimeter Rate of Heat Release Instrument (exposed area:
-0.01 m2) (cone) [12]. 2) OSU Rate of Heat Release Calorimeter (exposed area: -0.02
m2) [13].
The measurements made were as follows:
The main parameters that can be reported from small-scale RHR instrument tests are:
peak rate of heat release (Pk RHR, in kW/m2), the time to sustained combustion, or time
to ignition (TTI, in s), the total heat released (THR, in MJ/m2), the smoke factor (SmkFct,
ill MW/m2), the peak rate of smoke release (Pk RSR, in 1/s), the total smoke released (TSR,
DOn-dimensional), the mass loss rate parameter (MLRP, in g/m2s2) and the ratio of time to
ignition to peak rate of heat release (TTI/RHR, in s m2/kW). The OSU used did not allow
continuous mass measurements to be made and MLRP is, thus, not reported from it. Some
of these variables may not be generally known. For example, the smoke factor is a relatively
new concept, as is the TTI/RHR ratio; the others are often not measured in the cone
calorimeter but are always measured in the OSU. Thus, the variables used will all be
explained briefly.
Rate of heat release (RHR)- The rate of heat released is a measure of instantaneous
amount of heat released per nominal sample surface area. For each experiment, the max-
imum RHR value is the most significant one and is recorded here. In order to avoid
misleading information resulting from spurious single scans, a simple 5 point running average
is used to smooth the data. The RHR values are calculated from differences between
measured values of oxygen concentration and atmospheric background oxygen, by using
the principle of oxygen consumption [14,15].
Total heat released (THR)- The total heat released in each experiment is the integral of
RHR data as a function of time, per unit nominal sample surface area. The value reported
here is the one at 15 min.
Rate of smoke release (RSR)- The rate of smoke release is calculated in different ways
in both small-scale calorimeters, for traditional reasons:
OSU RSR (V * OD)/Light Path Length
Cone: RSR (V * OD * In1O)/(Sample area * Light Path Length)
where V is the volumetric flow rate (in m'/s, corrected for the relative locations of the flow
measurement device and the light measurement device and for the elevated temperature in
the duct), OD is the optical density, the Light Path Length is 0.1095 m (cone) or 0.134 m
(OSU) and the sample area is 0.0100 m' in the cone and 0.0213 m2 in the OSU. The units
are 1/s in the cone and m2/s in the OSU.
Smoke factor (SmkFct)-Smoke factor is an empirical smoke/fire hazard variable used
to estimate the potential realistic amount of smoke that a product would generate under
full-scale fire conditions. It takes into account both the potential for smoke obscuration
after full-sample destruction and the potential to cause other products to burn and release
smoke in real fires. It does so by incorporating the burning rate (as peak RHR) [16,17].
This is based on the fact that products made from materials with low peak RHR are unlikely
to burn up totally in a fire, and will, moreover, cause less smoke to be generated as a result
of the ignition of other products. Smoke factor is a continuous function calculated as the
product of total smoke released and peak RHR. The single value presented here is that at
5 min. The total smoke released is calculated as the time integral of the rate of smoke
release.
Time to ignition (TTJ)- The time to ignition is the time, in s, until the entire surface of
the sample burns with a sustained luminous flame.
Mass loss rate parameter (M LRP)- The MLRP [9,18] is the ratio of a) the average mass
loss rate between those times at which the sample loses 10 and 90% of the total mass lost
during the test and b) IT!. It gives an indication of the amount of "smoke" (combustion
products) generated in a given amount of time and, thus, of the toxic hazard.
Time to ignition/peak rate of heat release (TTI/RHR)- This parameter is proportional to
the time to flashover, that is, it may, thus, be the best individual indicator of overall fire
hazard [/9-21].
Full-Scale Tests [22,23]-1) CSA FT-4 Cable Tray Test (70 000 BTU/h: 20.50 kW) [24].
2) UL 1581 Cable Tray Test (70 000 BTU/h: 20.50 kW) [25]. 3) ICEA T-29-520 Cable Tray
Test (210 000 BTU/h: 61.50 kW) [26].
The area of cable exposed to the flame in the cable trays is different for each test (see
Table 1).
Official failure criteria for cable tray tests address char length: if the entire cable tray
TABLE 1- Test parameters for full-scale cable tests,
CSA-FT-4 UL 1581 ICEA T-529-520
Harne length, m 0,25 0,30 0,75

Cable width, m 0.15 0,11 0,11
Cable length above flame, m 2.50 1.98 2,13
Directly exposed area, m' 0,0375 0.033 0.083
Exposed cable area, m' 0,375 0.218 0.234
length (UL 1581 and lCEA T-29-520) or a length over 1.50 m (CSA Ff-4) has charred, the
cable fails the test.
The parameters reported from the full-scale tests include:
1. THR: Total heat release (after 15 and 20 min), after subtracting the heat contributed
by the burner, in MJ/m2•
2. TSR: Total smoke released (after 15 and 20 min). This was obtained by integrating
rate of smoke release, m2•
3. Pk RHR: Peak rate of heat release, kW/m2.
4. Pk RSR: Peak rate of smoke release, m2/s. The equation used was:
Tray: RSR = OD/(Light Path Length)
5. Gas concentrations (CO, CO2, HCl), in ppm or %.

6. Gas yield ratios (non-dimensional).
All cables were burnt in the cone calorimeter and in the CSA Ff-4 cable tray test. Some
of these cables (all of them experimental power cables) were also burnt in the UL 1581 or
ICE A T-29-520 test and the jacket and insulation materials for all the experimental power
cables were also burned in the small-scale RHR tests.
One series of full-scale cable tray test experiments was carried out at Underwriters'
Laboratories (UL) in Northbrook, IL, involving all three cable tray tests. The heat release
equipment used was an oxygen depletion calorimeter, based on oxygen paramagnetic anal-
ysis, in the 0.406 m diameter (16 in.; 0.1297 m2 area) duct, downstream from the burner
and tray, while the smoke release equipment was a white light system, with a light beam,
which measured smoke vertically in the duct, Subsequent full-scale cable tray tests (using
only the CSA Ff-4 protocol) were run at the BFGoodrich (BFG) fire test facility in Avon
Lake, OH. The BFGoodrich equipment for measuring heat and smoke release was con-
structed by Fire Testing Technology Ltd., East Grinstead, United Kingdom, and was also
installed in the duct. Heat release was measured with an oxygen paramagnetic analyzer,
but smoke obscuration was measured horizontally, with a laser system.
All small-scale RHR tests were run at BFGoodrich in Avon Lake, OH.
Materials
A total of 16 cables were used. These included 10 experimental power cables, all based
on vinyl compounds, 4 commercial cables of different types, and two experimental com-
munications cables. The experimental power cables were of two constructions: THHN and
TW. The THHN construction requires the use of a nylon film, extruded over the vinyl
insulation, beneath the vinyl jacket. These cables were all made with vinyl compounds
containing significant levels of fire retardants, the only exception being those compounds
designated "1" (11, 11), that contained none or very low levels. Dimensional requirements
for THHN or TW cables can be found in UL bulletins [27,28]. The chemical composition
of the cables used varied broadly: poly(vinyl chloride) (PVC), chlorinated poly(vinyl chlo-
ride) (CPVC), polychloroprene, vinyl thermoplastic elastomer alloys, polyolefins, and fluo-
roploymers.
Experimental Vinyl Power Cables:
Commercial Cables:
Plenum- Western Electric Omaha NEC-SOO-3D

Tray-XLP/CU Black Jacket 14 pr #6, Super flex
Other- Yellow Ultragard Type SOD 90°C, Super Trex 14/4
Essex THHN 600 V, 4 AWG, single conductor
Experimental Communications Cables:
IBM Type I
IBM Type II
Results and Discussion

Full-Scale Test Results
Tables 2 to 4 show the results of all the full-scale vertical cable tray tests carried out at
UL, as normally indicated for such tests, viz. pass/fail, peak flame height and total char
length. The additional results presented are the mass loss (expressed as a percentage of the
combustible mass present). Tables 5 to 7 present some gas emission results: peak concen-
trations of carbon monoxide (CO), carbon dioxide (C02), hydrogen chloride (HCl), as well
as carbon oxide ratios. The HCI results are shown because HCI is an important fire gas and
all those cables contained chlorinated materials.
It is important to observe that the peak concentrations of gases (carbon oxides or HC!)
were a function of the severity of the test and of whether the cable passed or failed the test,
rather than of the chemical composition of the cable coating materials. All cables that passed,
released only small amounts of combustion products and lost only less than 50% of the
combustible mass. Peak flame heights and char lengths were not useful magnitudes to assess
the relative performance of cables that either failed or passed. As a curious sideline, it is
interesting to compare the HCI emission from two vinyl cables, similar in all respects, but
TABLE 2-Results of UL 1581 cable tray tests.
Cable Peak Flame Height, em Total Char Length, em Combined Loss, % Pass/Fail
11 11 THHN >270 244 69 Fail

31 11 THHN >270 244 57 Fail
11 11 TW 135 122 15 Pass
31 31 TW 105 94 11 Pass
31 41 THHN 105 107 15 Pass
31 31 THHN 105 97 14 Pass
TABLE 3-Results of 1CEA T-29-520 cable tray tests.
Cable Peak Flame Height, em Total Char Length, em Combined Loss, % Pass/Fail
U 11 THHN >270 244 74 Fail

U 11 TW 225 198 47 Pass
31 11 TW 150 140 35 Pass
31 31 THHN 143 124 41 Pass
U31TW 150 124 41 Pass
31 2J TW 135 140 29 Pass
S24 31 TW 120 130 27 Pass
]I 31 TW 120 124 28 Pass
TABLE 4-Results of CSA FT-4 cable tray tests.
Cable Peak Flame Height, em Total Char Length, em Combined Loss, (%) Pass/Fail
U 11 THHN >250 265 81 Fail

U11TW 175 133 23 Pass
UllTW 150 132 21 Pass
]I 41 THHN 100 79 15 Pass
]I 2J TW 50 60 12 Pass
TABLE 5-Gas analysis results of UL 1581 cable tray tests.
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCl, ppm
U 11 THHN 1245 1.03 0.112 1121

]111 THHN 1102 0.86 0.143 1216
UllTW 229 0.13 0.092 40
]I 41 THHN 186 0.10 0.169 37
]I 31 THHN 156 0.10 0.110 62
]I 31 TW 121 0.06 0.133 26
tIDe of which has HCl retention catalysts (low acid jacket) while the other one has fire
Ktardant additives (FR jacket). The one with the FR jacket releases much less HCl (Figs.
I and 2).
Tables 8 to 10 have heat and smoke release data for all the cables in the full-scale tests.
II is clear that cables failing a test, release more heat and more smoke than those passing
tile test. Furthermore, the release rates for heat and smoke are also much higher for the
failing cables. Moreover, the peak rate of heat release indicates which of the cables that
TABLE 6-Gas analysis results of ICEA T-29-520 cable tray tests.
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCl, ppm
11 11 THHN 2966 1.92 0.091 >2326

11 11 TW 949 0.67 0.085 463
31 11 TW 853 0.41 0.086 493
31 31 THHN 885 0.34 0.106 87
11 31 TW 683 0.40 0.098 358
31 21 TW 723 0.30 0.085 387
524 31 TW 766 0.26 0.130 624
31 31 TW 641 0.32 0.091 326
TABLE 7-Gas analysis results of CSA FT-4 cable tray tests.
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCI, ppm
11 11 THHN 4251 2.60 0.133 >2332

11 11 TW 637 0.33 0.189 547
11 11 TW 552 0.29 0.178 587
31 41 THHN 406 0.16 0.267 578
31 21 TW 332 0.10 0.220 204
passed the test were possibly marginal and which ones passed comfortably. Analogously to
the case of the gas release, the heat and smoke release is heavily dependent on whether the
cable passes or fails the test and on the severity of the test (for example, Figs. 3 to 6, CSA
FT-4, address release rates of heat and smoke, and amounts of CO and HCI released).
It is worth restating an obvious fire hazard fact not often fully appreciated. These results
show that smoke release depends primarily on the amount of material burnt, as well as
TABLE 8-Heat and smoke release results from UL 1581 cable tray tests."
THR 15 PK RHR PK RSR

THR5 MJ/m2 THR 20 TSR 5 TSR 15 TSR 20 kW/m2 1/s
-.ckground 6.3 19.3 25.8 1 3 4 24.7 0.02

• U THHN 14.6 42.3 43.0 737 2046 2061 157.2 7.11
]I( U THHN 12.8 30.4 30.7 742 1877 1880 124.9 6.99
DUTW 3.0 7.4 7.9 333 635 639 13.0 1.93
]I( .u THHN 3.4 12.3 14.7 224 671 689 19.9 1.71
]I( 3J THHN 1.1 3.3 3.4 195 492 535 8.8 1.38
]I( 3J TW 1.7 4.7 5.4 164 302 321 6.0 0.75
-Background burner heat was subtracted out from total heats reported. The TSR results have been
...mplied hy the cross sectional area of the duct.
TABLE 9-Heat and smoke release results from 1CEA T-29-520 cable tray tests."

THR5 MJ/m2 THR 20 TSR 5 TSR 15 TSR 20 kW/m' lis
--=tground 17.5 54.1 72.3 15 46 62 64.8 0.06

U 11 THHN 48.5 65.2 69.8 1790 2141 2154 266.2 11.85
•• 11 TW 16.9 34.9 39.5 1326 1880 1906 93.4 6.87
]111 TW 12.6 25.7 30.8 1010 1450 1534 50.8 5.41
]I 3J THHN 6.1 13.7 15.9 903 1377 1421 40.7 5.56
U31TW 7.6 24.7 30.0 597 1367 1497 47.9 3.78
]I 2J TW 8.3 20.5 25.3 1095 1585 1690 38.1 5.12
s::!43J TW 6.8 19.8 25.4 601 1531 1720 28.8 3.81
]I 31 TW 7.5 20.7 26.1 549 1091 1218 36.1 2.70
"Background burner heat was subtracted out from total heats reported. The TSR results have been
.ultiplied by the cross sectional area of the duct.
TABLE lO-Heat and smoke release results from CSA FT-4 cable tray tests."
THR5 MJ/m2 THR 20 TSR 5 TSR 15 TSR 20 kW/m2 1/s
Background 6.3 19.3 25.8 1 3 4 24.7 0.02
11 11 THHN 38.1 98.0 100.5 930 3135 3166 378.5 14.62
11 11 TW 6.6 25.7 27.5 955 2261 2275 57.1 5.25
11 11 TW 7.8 27.1 29.3 1055 2109 2124 52.4 5.50
31 4J THHN 2.4 14.0 16.1 244 1496 1532 30.4 4.12
31 2J TW 3.8 11.0 13.1 772 1384 1414 18.1 3.39
"Background burner heat was subtracted out from total heats reported. The TSR results have been
multiplied by the cross sectional area of the duct.
tkpending somewhat on the smoke producing tendency of the material itself. Thus, less
smoke is released in a full-scale fire if the material burns less readily.
The CO/C02 ratios measured show the same pattern for all tests and, in fact, the averages
were very high, at ~O.1. The ratios were particularly high during the first 5 min of the
tests (Fig. 7, CSA FT-4). The initial CO/C02 levels are highest for the better fire performing
cables, although the CO yields are much lower (Fig. 5). These results can be related to the
type of fire. Carbon oxide ratios are expected to be low for very intense fires if oxygen is
abundant and high for: (a) low intensity fires with high oxygen; or (b) very high intensity
fires with very low oxygen. In this case; low CO/C02 ratios should be expected, in principle.
However, in the early stages of the burn, the cables will all release abundant combustion
products, and cause severe local oxygen depletion. Moreover, the poorer fire performing
cables will burn more intensely and more completely, yielding higher CO2 yields and lower
CO/C02 ratios. Later on, the only cables with significant CO/C02 ratios are those which
fail the test.
An interesting comparison can be made between the results of two series of tests, viz.
those on cables 11 11 THHN, 11 31 THHN, 31 11 THHN, and on cables 11 11 TW, 11 31
TW, and 3111 TW. When the fire performance of the insulation is improved, but not that
of the jacket, the resulting fire performance changes much less than if the jacket is the one
that is improved.
The peak rate of heat release is the best discriminant between passing and failing cables:
there is a clear gap between the peak RHR of the cables that pass and those that fail any
test. There is also a gap for other parameters, but the only one as clear cut is rate of smoke
release.
Tables 11 and 12 show full-scale cable tray test data from the BFG experiments, restricted
to the CSA FT-4 protocol. The observations made in the UL facility can be transported,
both qualitatively, and, in most cases, quantitatively to the other facility.
The only important differences between the results in both test facilities are found for a
borderline cable: 11 11 TW failed at BFG and passed at UL.
The tables show that the cables tested can be subdivided into three categories:
(i) Cables that are clear failures;

(ii) Cables that are borderline in passing or failing the test; and
(iii) Cables performing better than needed to pass the test.
Category i contains 4 of the cables tested: two commercial ones (Essex THHN and
Ultragard SOO) and two experimental (1111 THHN and 31 11 THHN, both with a non-
fire retarded jacket and nylon).
Category ii contains two of the cables tested: one commercial (XLP/CU) and one exper-
imental (1111 TW).
TABLE 12-Additional cable tray test results from the

BFGoodrich CSA FT-4 tests.
THR at 15 THR at 20 TSR at 15 TSR at 20
Material MJ MJ m' m'

Essex THHN 61 61 165 168.6
1IIlJ-THHN 85 86 413 415.9
Ultragard SOO 110 120 455 459.5
3I1lJ-THHN 98 125 413 435.8
1IIlJ-TW 50 68 541 616.9
1IIlJ-TW 49 70 569 639.2
1IIlJ-TW 52 80 549 609.0
XLPE/CU 25 31 343 404.5
XLPE/CU 23 35 307 402.4
3I1lJ-TW 16 17 283 287.9
3I14J-THHN 12 16 149 181.4
3I13J-THHN 9 11 164 169.3
1II3J-TW 9 10 185 195.4
3I13J-THHN 8 11 142 144.0
1II3J-TW 10 12 189 197.4
3I12J-TW 5 8 227 237.2
3I13J-TW 7 9 164 172.4
524/3J-TW 8 10 168 181.5
3I13J-TW 6 8 158 167.7
IBM Type 11 5 6 74 79.1
IBM Type I 5 8 51 56.2
Plenum 4 5 8 10.2
Blank 19 26 0.068 0.091
NOTE- The THR data has had the blank heat value (caused
by the burner itself) subtracted. The TSR values are as measured,
because the blank TSR is negligible.
d) The CO/C02 ratios were close to 0.1 in all cases, which is similar to what is expected
in full room involvement fires.
e) Improving the fire performance of the jacket is more important for overall fire per-
formance than improving the fire performance of the primary insulation.
f) The relative severity of the tests investigated is:
UL 1581 « CSA < Ff-4 ICEA T-29-520
Cone Calorimeter Cable Test Results

All cables were also tested, in the horizontal mode, in the cone calorimeter, at 20 and
40 kW/m2 incident flux (Table 13). The cone calorimeter (heat and smoke) data are very
consistent with cable tray data, with the possible exception of data for borderline cables.
All cables that fail a cable tray test give the highest RHR and THR values at 20 kW/m2,
followed by borderline cables. However, at 40 kW/m2, one of the class ii cables tested (XLP/
CU) becomes virtually indistinguishable from class i cables while the other one (II 11 TW)
is significantly better. However, the latter failed one of the tray tests and the former passed.
In terms of the most indicative fire index, TTURHR, all cables that stand out at 40 kW/m2
(the communications cables, 31 3J THHN, 31 4J THHN, II 3J TW, 31 3J TW, and 524 3J
TW) are iii(c) cables.
This suggests, clearly, that the cone calorimeter is capable of giving a good a priori
indication of whether a cable will pass or fail, the cable tray tests studied (CSA Ff-4, UL
1581, or ICEA 529 T-20). It appears, for example, that if the peak RHR is significantly
over 100 kW/m2, at an incident flux of 20 kW/m2, the cable will fail a cable tray test. When
the cone calorimeter test is carried out at higher incident fluxes the peak RHR cut-off point
is higher: it appears to be near 200 kW/m2 at 40 kW/m2• However, borderline cases (class
ii cables) are still a problem. The number of available intermediate data (that is, close passes
or close fails in a cable tray test) is still insufficient to determine the degree of predictability
of the cone calorimeter for cables with marginal fire performance.
Correlations Between the Cone Calorimeter and a Cable Tray Test

The first approximation is to assume a linear correlation model:
full scale = (small scale x slope) + intercept (1)
It is also possible to obtain correlation models which predict full-scale results from a
combination of small-scale results at various fluxes. A simple means to do such an analysis
uses empirical models and predictions based on a linear combination of results at 20 and
40 kW/m2• The relevant model is, thus, Eq 2:
full scale = (Small scale,20 x slope,20) +
+ (Small scale,40 x slope,40) + intercept (2)
Results for the correlations based on the full-scale tests carried out at UL (with the cone
at 20 kW/m2 only, and at both fluxes), gave excellent correlation coefficients (Tables 14
to 16).
After further tests were run at the BFGoodrich facility, a new set of linear correlations
was carried out, including not only vinyl cables (like those in the UL tests), but also materials
with different chemistries. The basis for this correlation (model 20) was simply the cone
calorimeter data at 20 kW/m2• The model for this series of correlations was then compared
with the old model, obtained purely from vinyl cables, also using cone calorimeter data at
20 kW/m2• Figure 8 presents data as well as correlations obtained with both facilities (only
5 data points for the cables tested at UL with the CSA FT-4 test). The degree of agreement
found between the data for the two full-scale test facilities was very reasonable. This is
TABLE 14-Correlation between the cone calorimeter and the UL 1581 cable tray test. See Eq 2.
Correlation Coefficient
Property Flux R2/Adj R2 P
THR at 15 20 0.93/0.90 0.002
THR at 15 40 0.56/0.45 0.088
THR at 15 20,40 0.96/0.90 0.019
Peak RHR 20 0.77/0.71 0.022
Peak RHR 40 0.77/0.71 0.022
Peak RHR 20,40 0.86/0.77 0.050
Peak RSR 20 0.70/0.62 0.039
Peak RSR 40 0.75/0.68 0.027
Peak RSR 20,40 0.87/0.78 0.048
SmkFct 20 0.93/0.92 0.002
SmkFct 40 0.86/0.83 0.008
SmkFct 20,40 0.93/0.89 0.017
TABLE IS-Correlation between the cone calorimeter and the ICEA T-29-520 cable tray test.
See Eq 2.
TIlR at 15 20 0.90/0.88 0.000

TIlR at 15 40 0.57/0.50 0.030
TIlR at 15 20,40 0.94/0.92 0.001
Peak RHR 20 0.73/0.69 0.007
Peak RHR 40 0.83/0.80 0.002
Peak RHR 20,40 0.84/0.77 0.011
Peak RSR 20 0.85/0.82 0.001
Peak RSR 40 0.57/0.50 0.030
Peak RSR 20,40 0.86/0.80 0.008
SmkFct 20 0.66/0.60 0.D15
SmkFct 40 0.72/0.67 0.008
SmkFct 20,40 0.72/0.61 0.040
TABLE 16-Correlation between the cone calorimeter and the CSA FT4 cable tray test.
TIIR at 15 20 0.98 /0.97 0.001

TIIR at 15 40 0.43 /0.24 0.232
TIIR at 15 20,40 0.986/0.97 0.014
Peak RHR 20 0.91 /0.83 0.011
Peak RHR 40 0.65 /0.53 0.100
Peak RHR 20,40 0.92 /0.85 0.076
Peak RSR 20 0.68 /0.57 0.088
Peak RSR 40 0.19 /0.00 0.457
Peak RSR 20,40 0.98 /0.96 0.018
SmkFct 20 0.93 /0.91 0.008
SmkFct 40 0.71 /0.62 0.071
SmkFct 20,40 0.93 /0.86 0.071
ftry important, because it has long been assumed that exact replication of every minute
detail is essential to be able to replicate full-scale facility data. The facility used at BFGood-
rich is very similar, in most respects, to the one at UL, but it differs in a few details. The
old model had given excellent results for the tests carried out at UL, but was not sufficiently
accurate for some of the other tests. The new model gave very good predictions for all data.
The correlations based on multiple fluxes give, of course, much better predictions than
1bose based on a single flux, but at the expense of the requirements for much more testing.
The advantages gained by using complex multi-flux models are small compared to the
additional difficulty and expense involved and are not really essential. In particular, there
is virtually no need to test cables at 70 kW/m2, if it is desired to predict results associated
with occupancies modelled by the cable tray tests.
In terms of smoke release, it is important to notice the excellent correlation found between
IOtal smoke released in the full-scale tests and the smoke factor in the cone calorimeter.
T~ting of Compounds in Small-Scale Tests
Cone and as U Rate of Heat Calorimeter Test Instruments- The results of the tests carried
FIG. S-Peak rate of heat release results from all tests carried out using the CSA FT-4 apparatus. UL
FT-4: Experiments carried out at the UL facility. BFG FT-4: Experiments carried out at the BFGoodrich
facility. Predictions from old model: predicted peak RHR values from model based on UL data only.
Predictions from 20: predicted RHR values from model based on all data.
out for the materials (compounds and not cables) in the two instruments, cone and OSU
RHR, respectively, at 20 and 40 kW/m2 incident flux levels, are shown in Tables 17 and 18.
Table 19 indicates that there are excellent correlations between the results.
Three main differences exist between the two instruments:
(a) ignition source (spark igniter in cone and gas flame in OSU);
(b) sample orientation (horizontal in cone and vertical in OSU); and
(c) heat release measurement technique (oxygen depletion in cone and thermal couple
temperature measurement in OSU).
Secondary differences also exist, which are the sample size and smoke obscuration mea-
surement technique. A more extensive discussion of the differences between the two in-
struments has been given elsewhere [29].
The most interesting aspect is the excellent correlation not only between the smoke factors
in the two instruments at both fluxes, but also between the two fluxes for either instrument.
This indicates that the smoke factor is a more robust measure of smoke release than direct
smoke measurement techniques.
This good correlation between the two RHR test instruments, parallels one that has been
shown before, involving 17 materials of very varied physical characteristics and fire per-
formance [30]. The present series involves a number of compounds with a common trait;
all are vinyl wire and cable compounds.
Fire Hazard Assessment
The cable tray fire tests represent the potential fire hazard in a situation where cables are
exposed to a fairly intense source of energy, probably in a concealed compartment, where
the cables are the sole (or almost only) combustible present. The most important means of
decreasing fire hazard in that room is decreasing the peak rate of heat release from the
burning products. This avoids flame spread to other combustibles or, more likely, into other
compartments.
TABLE 19-Correlations in heat release between cone and OSU calorimeters.
8Ise Apparatus Flux Property R2 Constant Coefficient
Cone 20 PK RHR 0.786 30.99 0.509

Cone 40 PK RHR 0.905 0.041 0.623
Cone 20 THR 0.744 12.57 0.681
Cone 40 THR 0.704 19.81 0.347
Cone 20 ITI 0.575 162.35 0.106
Cone 40 ITI 0.494 23.59 0.052
Cone 20 TSR 0.673 - 484.45 1.000
Cone 40 TSR 0.882 8.056 0.872
Cone 20 PK RSR 0.771 -1.95 0.992
Cone 40 PK RSR 0.619 - 1.50 0.969
Cone 20 SmkFct 0.858 20.12 0.606
Cone 40 SmkFct 0.965 16.20 0.689
Cone 20 vs. 40 SmkFct 0.836 174.50 2.405
OSU 20 vs. 40 SmkFct 0.840 94.61 2.588
References
[1] Thomas, P. H., "How Heat Release Influences Fire Hazard," International Conference on FIRE:
Control the Heat-Reduce the Hazard, Paper I, Fire Research Station, Borehamwood, United
Kingdom, 24-25 Oct., London, 1988.
[2] Babrauskas, V., "Effective Measurement Techniques for Heat, Smoke and Toxic Fire Gases:'
International Conference on FIRE: Control the Heat- Reduce the Hazard, Paper 4, Fire Research
Station, Borehamwood, United Kingdom, 24-25 Oct., London, 1988.
[3] Babrauskas, V., "Upholstered Furniture Room Fires- Measurements, Comparison with Furniture
Calorimeter Data, and Flashover Predictions," Journal of Fire Sciences, Vol. 2, No.5, 1984, pp.
5-19.
[4] Babrauskas, V. and Krasny, J., "Prediction of Upholstered Chair Heat Release Rates from Bench
Scale Measurements," Fire Safety. Science and Engineering, ASTM STP 882, T. Z. Harmathy.
Ed., American Society for Testing and Materials, Philadelphia, 1985, p. 268.
[5] Mulholland, G. W., Henzel, V., and Babrauskas, V., "The Effect of Scale on Smoke Emission,"
2nd International Fire Safety Science Symposium, T. Wakamatsu, Y. Hasemi, A. Sekizawa.
P. G. Seeger, P. J. Pagni, and C. E. Grant, Eds., Hemisphere Publishing, Washington, DC, 13-
16 June, Tokyo, 1988. pp. 347-357.
[6] Babrauskas, V., Fire Materials, Vol. 8, No. 81,1984, p. 81.
[7] Babrauskas, V., "Bench-Scale Methods for Prediction of Full-Scale Fire Behavior of Furnishings
and Wall Linings," Technology Report 84-10, Society of Fire Protection Engineers, Boston, 1984.
[8] Fowell, A. J., Fire Technology, Vol. 21, No.3, 1985, pp. 199-212.
[9] Hirschler, M. M., "Fire Hazard and Toxic Potency of the Smoke from Burning Materials," Journal
of Fire Sciences, Vol. 5, No. 289, 1987.
[10] Tewarson, A., "Generation of Heat and Chemical Compounds in Fires," in Handbook of Fire
Protection Engineering, P. di Nenno, Ed., Chapter 1/13, National Fire Protection Association
(NFPA), Quincy, MA, 1988, pp. 1-179.
[11] Hirschler, M. M., "Fire Performance of Poly(Vinyl Chloride) and its Relation to Fire Hazard,"
31st IUPAC Microsymposium on Macromolecules Poly(Vinyl Chloride), Prague, 18-21 July, 1988;
Makromolekulare Chemie, Macromolecular Symposia Vol. 29, pp. 133-153, 1989.
[12] Babrauskas, V., "Development of the Cone Calorimeter. A Bench-Scale Heat Release Rate
Apparatus Based on Oxygen Consumption:' NBSIR 82-2611, National Institute of Standards and
Technology (N1ST), Gaithersburg, MD, 1982.
[13] Smith, E. E., "Ignition, Heat Release and Noncombustibility of Materials," Ignition, Heat Release
and Noncombustibility of Materials, ASTM STP 502, A. F. Robertson, Ed., American Society for
Testing and Materials, Philadelphia, 1972, p. 119.
[14] Parker, W. J., "Calculations of the Heat Release Rate by Oxygen Consumption for Various
Applications," NBSIR 81-2407, National Institute of Standards and Technology (NIST), Feb.
1982.
[15] Huggett, C, Fire Materials, Vol. 4, No. 61, 1980, p. 61.
(16] Babrauskas, V., Journal of Fire Flammability, Vol. 12, No. 51, 1981.
[17] Hirsch1er, M. M. and Smith, G. F., "Determination of Fire Properties of Products by Rate of
Heat Release Calorimetry: Use of the National Bureau of Standards Cone and Ohio State Uni-
versity Instruments," Fire Safety Progress in Regulations, Technology and New Products, Fire
Retardant Chemicals Association Fall Conference, Monterey, CA, 1987, pp. 133-146.
(18] Babrauskas, V., 'Toxic Hazard form Fires: A Simple Assessment Method," International Con-
ference on "FIRE: Control the Heat-Reduce the Hazard," Paper 16, Fire Research Station,
Borehamwood, United Kingdom, London, 24-25 Oct., 1988.
[19] Wickstrom, U. and Goransson, U. J., Journal of Testing and Evaluation, Vol. 15, No.6, 1987.
[20] Hirschler, M. M., "Smoke and Heat Release and Ignitability as Measures of Fire Hazard from
Burning Carpet Tiles," Fire Safety Journal, Vol. 18, 1992, pp. 305-324.
[21] Hirschler, M. M. and Poletti, R. A., "Latex Backcoatings on Polypropylene Carpets: Fire Per-
formance Testing," Journal of Coated Fabrics, Vol. 19,1989, pp. 94-111.
[22] Ebert, T. R., "Preliminary Modified Vertical Tray Flame Tests," E41877, 89NKI4704, Under-
writers' Laboratories, Inc., Northbrook, IL, 30 Aug., 1989.
[23] Underwriters' Laboratories letter for release of publication of results, Underwriters' Laboratories,
Inc., Northbrook, IL, 1989.
[24] Vertical Flame Test: Cables in Cable Trays, C22.2, No. 0.3-MI985, Section 4.11.4, Canadian
Standards Association, Toronto, Ontario, Canada, Updated Aug. 1988.
[25] Reference Standard for Electrical Wires, Cables and Flexible Cords, Section 1160, Vertical Tray
Flame Test, 1st Edition, August 1983 (periodically updated), UL 1581, Underwriters' Laboratories,
Inc., Northbrook, IL., Aug. 1983 (periodically updated).
I26J Publication T-29-520, Insulated Cable Engineers' Association, Inc., Sept., 1986.
127J Thermoplastic Insulated Wires and Cables, 9th Edition, UL 83, Tables 15.3 and 15.5, Underwriters'
Laboratories, Inc., Northbrook, IL, Sept. 1983 (periodically updated).
I28J Electrical Power and Control Tray Cables with Optional Optical Fiber Members, 1st Edition, UL
1277, Table 10.24, (Updated Oct. 1988), Underwriters' Laboratories, Inc., Northbrook, IL, Jan.,
1986.
(29J Hirschler, M. M., "Smoke in Fires: Obscuration and Toxicity," Plenary Lecture, BCC Conference
on Flame Retardancy of Polymeric Materials, G. Kirshenbaum and M. Lewin, Eds., Norwalk, CT,
15-17 May, 1990, pp. 70-82.
POJ Hirschler, M. M., "The Measurement of Smoke in Rate of Heat Release Equipment in a Manner
Related to Fire Hazard," Fire Safety Journal, Vol. 17, 1991, pp. 239-258.
Joseph B. Zicherman1
Performance of Plastic Plumbing and

Electrical Products in Fire Resistive
Assemblies
REFERENCE: Zicherman, J. B., "Performance of Plastic Plumbing and Electrical Products
in Fire Resistive Assemblies," Fire Hazard and Fire Risk Assessment, ASTM STP 1150, Marcelo
M. Hirshchler, Ed., American Society for Testing and Materials, Philadelphia, 1992, pp. 66-
83.
ABSTRACT: The plastic pipe, tube, and conduit products used in the United States evolved
from transplanted European technologies over the past 30 years. Though originally only used
in non-fire rated construction, different versions of these products have qualified for use in a
variety of fire rated assemblies and building types after appropriate fire endurance tests. End
uses for these products include electrical power, signal distribution, and various plumbing and
mechanical applications.
Plumbing systems showing acceptable performance in fire resistive assemblies include drain,
waste, vent, and supply piping (domestic hot and cold water, sprinkler piping). Approved
electrical applications include PYC-based rigid non-metallic conduit and a flexible polyvinyl-
chloride (PYC) tubing product called electrical non-metallic tubing. Tested approaches avail-
able to safely install plastic pipe, tube, and conduit include passive and active fire stopping
systems based on insulations, intumescent systems, and even mechanical "cut-off" devices for
thermoplastic piping systems. Fire test results for these products and systems, and the designs
that resulted from those testing programs, are widely available. Plastics used to manufacture
these products include PYC, chlorinated PYC, acrylonitrile-butadiene-styrene (ABS), poly-
butylene (PB), and polypropylene (PP).
Fire risk assessments of potential performance of these products must take into account
properties of the resin-polymer systems involved, the nature of the product itself (that is,
drain, waste, and vent versus supply piping versus electrical conduit or tubing), where the
product is installed (for example, within a cavity wall versus penetrating or encased in a slab),
and the type of installation detailing used. Because of the number of possible combinations
for end use, analytical approaches based on (1) standard test results, (2) field data on fire
performance of the plastic pipe, tube, and conduit products, and (3) fire resistive assemblies
into which they are installed all contribute to inferences drawn in conducting hazard
assessments.
Specifically, this paper reviews information available to assist in preparing more specialized
risk assessments for these products than the general studies cited above, under foreseeable,
end use configurations in fire resistive assemblies. An attempt has been made to utilize ap-
proaches suggested for general fire hazard assessment procedures that have been under ex-
tended study within the ASTM Committee E-5 on Fire Standards over the past several years,
but which have not reached the status of standards.
Most field performance data on plastic pipe come from residential occupancies where these
products have been used for over 20 years, and where they constitute 95% of pipe installed.
In addition, there have been numerous installations in non-combustible buildings (including
high rises) here and outside the United States. This database represents a large population
for evaluating the fire risk associated with using plastic piping products. The residential data
are most interesting because (1) numerically, they represent the largest number of installations,
(2) this occupancy type is the most fire prone, based on frequency of occurrence, and (3) these
I President, Integrated Fire and Failure Technologies, Inc., 2322 Sixth Street, Suite 107, Berkeley, CA
94710. 66
ZICHERMAN ON PLASTIC PLUMBING AND ELECTRICAL PRODUCTS 67
structures are constructed using the least sophisticated building technology. Similarly, electrical
tubing and conduit products (having met standard fire endurance test conditions) have been
used widely in the field without adverse effect since their incremental acceptance into the
National Electrical Code (NEe) over the past decade.
For the above reasons, the three model building codes most widely used in the United States
(which themselves represent an ad hoc risk assessment process) have accepted these products
after requiring demonstration of both acceptable fire performance and availability of reasonable
installation methodologies.
KEY WORDS: plastic pipe, fire endurance, through-penetration, hazard assessment, fire re-
sistive assembly, intumescent.
The plastic pipe, tube, and conduit (hereafter called plastic pipe, unless otherwise noted)
technology used in the United States and Canada has evolved to include many new uses
following its introduction from Europe over the past 30 to 40 years. Originally permitted
only in non-fire resistive construction, plastic pipe has been demonstrated as acceptable for
fire resistive end uses after substantial testing and field evaluation.
Risk assessment studies addressing general use of plastic pipe and plastic electrical tubing
have been published. Both were conducted as part of code adoption processes. One dealt
with possible increases in fire hazards when allowable uses for un plasticized PVC electrical
raceways were expanded in the National Electrical Code (NEe) [1). The second included
an assessment of fire related hazards associated with use of plastic plumbing pipe in fire
resistive construction. That study can be found in the draft environmental impact report
published recently by SRI International as part of the work associated with the state of
California plumbing code adoption process [2].
Both of these studies considered the combustible nature of the plastic products and found
their use appropriate in fire resistive construction if appropriate installation detailing
occurred.
Hazard Assessment Bases

In considering fire hazard elements associated with use of plastic pipe, it is necessary to
determine whether a proposed use will result in levels of safety which exceed, are consistent
with, or are less than existing products performing the same function. In addition, methods
to assess safety of new or non-classical uses of those products in fire resistive construction
are also needed.
Potentially undesirable consequences needing examination may include rapid fire spread
and unusual smoke threats. Failure or compromise of fire resistive assemblies prior to their
design limits which lead to unanticipated levels of damage from fire also need to be evaluated.
Considering this topic from the standpoint of the draft, "Standard Guide for the Devel-
opment of Fire Hazard Assessment Standards for Products," produced by the ASTM E5.35.01
Task Group on Fire Risk and Fire Risk Assessment, the document instructs the analyst that
"one needs to describe the product, how it is used, and its environment" [3]. Initially
examining "its environment" is key to several questions related to both occupant safety and
fire spread where uses of plastic pipe are concerned, and description of this environment
will ultimately define what products are suitable for use. Intuitively, it can also be seen that
the other two ("the product and how it is used") lend themselves to evaluation by tests that
can lead to fairly specific but broadly applicable results. For these reasons "environment"
will be considered first.
End Use Environments for Plastic Pipe, Tube, and Conduit in Fire Resistive Construction
Plastic pipe, tube, and conduit are typically installed within the envelope which forms the
boundaries of fire resistive spaces. Since these products are installed behind such wall and
floor/ceiling surfaces, their longevity and resistance to involvement in a growing fire situation
substantially exceeds human tenability in an affected space. Additionally, because this class
of products are typically installed behind barriers, changes in preflashover characteristics of
the space as compared to when metallic piping components are used will be minimal.
Exceptions come in the form of minor amounts of plastic pipe found in plumbing traps, and
so forth, whose quantity is minimal in comparison with total amounts of normally occurring,
combustible room contents.
Further, for most plastic piping products and in the case of PYC electrical raceways,
installation occurs within assemblies whose surface membranes (usually gypsum wallboard)
possess a minimum 15 min finish rating. Thus, it will take at least 15 min for such materials
to be exposed to temperatures of 32SOF(165°C) or greater, a temperature range relied upon
in judging the performance of protective membranes (as detailed in sections 46 and 47 of
ASTM Method for Fire Tests of Building Construction and Materials), and one which is
substantially lower than the ignition temperature for members of this generic class of ma-
terials. Because of this mode of installation, harm to occupants of such spaces from com-
bustion products associated with these materials is minimized in an origin area. Additionally,
because the nature of the fire resistive construction is intended to prevent migration of
combustion products, smoke threats elsewhere will be limited.
As is well known, fire resistive construction depends on barriers at selected locations to
prevent fire spread. In theory, if these barriers (walls, floor/ceilings, and so forth) function
as intended, building occupants will have adequate time to avoid injury by exiting safely or
evacuating to areas of refuge. Fire spread will also be limited. However, since actual struc-
tures contain openings such as doors, ducts for heating, ventilating, air conditioning, and
channels for distributing all sorts of utilities and services, the concept that such barriers are
monolithic in nature is generally only theoretical. In practice, most contain numerous open-
ings which must be protected.
Thus, the environment being considered involves the envelope formed by fire resistive
components with plastic pipe, tube, or conduit installed within them and/or passing through
them.
No assumptions are made as to changes in fuel load which these products may provide.
Rather, through testing, performance of assemblies which include combustible plastic ele-
ments are compared with tested designs which do not contain those materials (or details
needed for their installation) and evaluations are made whether or not plastic inclusions
reduce .the fire endurance rating of the original design. Fire exposure conditions are pre-
scribed in ASTM E 119, for full assemblies, and its derivative, ASTM E 814, Method for
Fire Test of Through-Penetration Fire Stops, for testing of a penetration detailing itself.
Description of Product
The "product" in this case is various plastic pipe, tube, and conduit materials manufactured
from a variety of generic polymer resins. These include unplasticized polyviny1chloride
(PYC), chlorinated PYC (CPYC), acrylonitrile-butadiene-styrene (ABS), polypropylene
(PP), polybutylene (PB). Each of these has its own physical properties which include melt
point, melt viscosity, fuel content, ease of ignition, and so forth. Additionally, application
of different manufacturing technologies has led to versions with solid or foam core in PYC
and ABS drain, and waste and vent (DWY) pipe products.
Piping and raceway configurations used with these assemblies fall in one or more of the
categories following:
• Through-penetrating elements
• Partially penetrating elements
• Included elements
• Combinations of A, B, and C
Given the preceding background information, a 3-dimensional matrix to visualize vari-

ables in the evaluation is suggested. The three axes, (a) project fire environment, (b) product
being considered, and (c) use format in project, form in the three axes as represented in
Fig. 1.
Hazard Assessment Data for Plastic Pipe, Tube, and Conduit Based on ASTM E 119 and
ASTM E 814 Testing
Results from well defined standard testing of assemblies are useful as input when con-
ducting hazard assessments for plastic pipe, tube, and conduit installations. The scope for
this paper defines the environment for these products as fire resistive construction of one or
more hours expected fire endurance. Thus, when specific plastic pipe products are tested as
part of an assembly designed for use in that environment, applicable test data results. The
following sections describe and discuss such standard test results and provide references for
them.
In overviewing ASTM E 119 and ASTM E 814 results for plastic pipe, derating of fire
resistive assemblies and impact of penetrating element diameter vs. thickness of assembly
are important.
Concerning derating of assemblies (reduction in fire endurance of the assembly containing
the plastic pipe product as compared to one without that product), the concept of finish
rating can be applied to compare performance of assemblies. Z In the case of the 1- and
2-h gypsum wallboard clad wall assemblies, direct comparisons can be made between finish
rating (time) with and without plastic pipe, tube, and conduit installed. Such a series of
comparisons are shown in Table 1 and reflect no substantial variation between specimens
with pipe and/or conduit installed, and those without those products present. These results
are particularly important, since early failures of fire resistive assemblies are the most critical
and potentially catastrophic ones.
A second element of fire endurance behavior for assemblies containing plastic pipe is the
relationship between penetration diameter and assembly thickness. This was explored years
ago in ad hoc testing in Europe, where different diameters of plastic pipe were installed in
concrete slabs of differing thicknesses [9]. Recently, the author conducted similar tests which
were 2 h in duration. Four direct through penetrations were evaluated in 114.3- and 165.1-
mm (4Yz- and 6Yz-in.) thick concrete slabs. PVC-DWV sizes tested included 38.1,50.8,76.2,
and 101.6 mm (lYz, 2, 3, and 4 in.). The pipe was installed in 6.35-mm (Y4-in.) oversized
holes with ceramic fiber packing as firestopping. In both thicknesses of slab, the 38.1-mm
(1Yz-in.) pipe performed satisfactorily for the full 2-h test duration and the 50.8- and 76.2-
mm (2- and 3-in.) pipe installations performed satisfactorily in the 165.1-mm (6Yz-in.) thick
slab. The 101.6-mm (4-in.) pipe failed in both thickness of slab with the generic firestopping
used, as did the 50.8- and 76.2-mm (2- and 3-in.) pipe in the thinner slab. Overall, failure
times in the thin slab for the 2-h test were inversely proportional to pipe diameter and all
failure times for all pipe diameters were proportionally longer in the thicker slab [10]. This
relationship between slab thickness and pipe diameter is an expression of what could be
called an "aspect ratio" as relates to plastic pipe use and allows one to generalize, not
surprisingly, that larger openings are less desirable than smaller ones, and thinner assemblies
are less desirable than thicker ones, where fire endurance is concerned.
The impact of elevated atmospheric pressure on the performance of plastic pipe, tube,
and conduit in fire affected rooms is of recognized importance. Of these, vented applications
(as found in plumbing uses) are the most critical. Conversely, minimal air flows can be
expected within electrical raceways which contain substantial fill and are essentially closed
at terminating locations in junction, switch, or outlet boxes. Additionally, electrical raceways
are used typically in substantially smaller diameters.
The actual pressure gradients which exist during postflashover fires begin with negative
values in the range of - 0.508 to - 0.254 mm ( - 4.98 to - 2.49 Pa.) of water at floor level
and increase to zero at about one-third the height of an affected room. They continue to
increase such that, near the ceiling, positive pressure in the range of + .508 to + 1.02 mm
( + 4.98 to + 9.97 Pa.) of water may exist [10].
The effect of these pressure gradients on possible fire travel through assemblies which
include penetrations has been debated extensively in code forums, where pressures utilized
in test furnaces have been at issue. In response, code agencies have stipulated that some
degree of positive pressure be applied to assemblies tested under the ASTM E 814 standard.
ZA confusingsituation existsin the field, in that finishrating measurementsfor dissimilarassemblies

are taken with thermocouples which may be located at vastly different locations, that is, individually
under the gypsummembranes for walltests above vs. at wood joists at the top of a plenum space when
rated hung ceilingsare tested. Both measurements, however, represent a "finish rating."
TABLE I-Fire endurance and finish ratings of I- and 2-h walls including PVC pipe,
tube. and conduit.
Finish Rating, Fire Endurance.

Included Element ffim h Reference
none 20.0 I [4]

none 19.5 I [4]
none 19.0 I [4]
PVC-DWV 19.5 I [5]
EMT section 67.0 2 [6]
PVC ENT section 67.0 2 [6]
EMT section 60.0 2 [7]
PVC ENT section 60.0 2 [7]
none 66.0 2 [8]
This is a conservative assessment, however, since the most common plumbing wall pene-
tration of a fire resistive assembly (a pipe trap at a sink) is under negative pressure during
a postflashover fire. Nevertheless, positive pressure is justified in testing since piping also
may pass through fire resistive assemblies high in a room (as at floor penetrations or above
a hung ceiling). Sprinkler piping, for example, is often run overhead, as is DHCW.'
Finally, most of the test results which follow are for plastic pipe products installed using
common, generic firestopping materials to ensure fire resistive integrity which is usually the
simplest and most cost effective installation solution. However, because plastic pipe, tube,
and conduit do melt and burn, in some applications such as larger diameters of pipe or with
oversized penetrations, installation details based on proprietary firestopping materials (such
as intumescent rubbers, proprietary fiber insulations) and mechanical firestopping devices
are necessary to provide superior and/or cost effective protection. Test results involving such
detailing (which are usually listed and sometimes labeled by third party organizations) are
included where needed to illustrate particular points. For the most extensive listings of such
products, the reader is referred to the Underwriters Laboratory, Inc., Fire Resistance
Directory [8].
Standard Fire Testing of Plastic Pipe

Most plastic plumbing pipe is run vertically in walls and shaftways in fire resistive con-
structions where pipe installations are bounded by gypsum wallboard. For this reason, most
of the early fire test work was directed at evaluating the integrity of such generic plumbing
walls with plastic pipe installed in various ways.
Because of a lack of baseline data for classically used steel and copper pipe, the initial
fire test study conducted included these materials as well as PVC and ABS plastic DWV
pipe [13]. To date, no other standard test results for metal plumbing products have been
published, which makes the question of performance criteria for them in conducting hazard
assessments somewhat ill defined. In the other early tests of plastic DWV, generic plumbing
walls with commonly available fire stopping materials (such as grout and thermal insulations)
were evaluated [14-21]. These are similar approaches t<;>that followed in the initial series
'Interestingly, in Canada the building code calls for positive pressure testing to be conducted at a
level which is 10 times that called for in the United States [I2]. The justification for this is, in part.
based upon "stack effect" arguments. One can question the requirement for such high pressure dif-
ferentials at test, based on the presence of engineered smoke control pressurization systems in high
rise buildings constructed today. Conversely, in low rise, fire resistive buildings, the stack effect is not
a factor.
at the National Institute for Standards and Technology's Center for Fire Research (formerly
National Bureau of Standards).
Test results produced and the plumbing and fire stopping configurations that resulted
from these tests reflect the evolution in understanding of needed installation technology
that occurred from the initial series through the most recent tests of one and two h metal
framed wall systems. The latter are found in typical non-combustible buildings [5 ,22J. Many
of the assemblies tested in these series were designed to simulate a plumbing wall undergoing
a fire exposure, on a floor between two similar fire resistive assemblies above and below
the test floor. This testing was used to assess possible vertical fire spread where vented
DWV systems are installed as well as horizontal fire spread across such assemblies. The
latter is the most obvious behavior expected in a fire resistive assembly with combustible
through penetrating elements. Vertical spread has not been observed in properly constructed
test assemblies due to both (a) the propensity of the DWV materials to melt and close off
openings at protected locations within cavity walls and (b) the functioning of fire stopping
or normal clearance penetrations at simulated floor lines.
Consistent with the above, failure modes observed in testing DWV systems have included
early horizontal burn through in test assemblies where installations included oversized an-
nular spaces at through penetration locations and/or plumbing fittings bearing on gypsum
wallboard membranes [13 ,23J.
With regard to amount of plastic pipe tested on fire performance, the most conservative
trials have included 4.877 m (16 ft) of 50.8 mm (2 in.) and 76.2 mm (3 in.) PVC DWV
installed together in a 139.7 mm (5Y2 in.) deep, 2.048 x 3.658 m (10 x 12 ft) metal frame
plumbing wall. The installation included two through penetrations. The same plumbing
configuration was successfully tested in one and two hour wall designs. The test assembly
included six vertical wall cavities and simulated upper and lower floors, two of which included
the DWV pipe. A schematic of the test assembly can be seen in Fig. 2 [24J. For ABS-DWV
pipe, a single plumbing cavity, in a 3.048 x 3.658 m (10 x 12 ft) 1 h 50.8 x 152.4 mm
(2 x 6 in.) wood framed test wall with the same plumbing configuration shown above
represents the most pessimistic condition tested to date [19J.
In order to assess the impact of foam core technology in which coextrusion techniques
are used to produce plastic DWV pipe with solid exterior layers and a low density foam
core, a test on the penetration detail only for the assembly described above was conducted
accorded to the ASTM E 814 test standard. Test results showed no difference in performance
for the foam core product as compared to the solid core test result [25J. Table 2 summarizes
the test series discussed above for DWV pipe in gypsum wallboard clad 1- and 2-h metal
and wood framed assemblies.
Specialized, proprietary systems for DWV through penetration of gypsum shaft walls or
masonry walls by plastic pipe have been developed and tested. These include guillotine-
like systems or intumescent closures for 2- or 3-h assemblies for plastic pipe diameters up
to 152.4 mm (6 in.). [27-31J.
From a standpoint of fire performance of plastic pipe used for sprinklers or DHCW
applications as compared to DWV (which is largely empty and of greater diameter), the
latter poses a greater threat to fire spread. However, PB sprinkler and DHCW pipe up to
50.8 mm (2 in.) in diameter have been tested in both metal and wood framed assemblies
under the criteria of the ASTM E 814 standard [32,33J.
Although these applications contain water, this was not possible for testing. Thus, test
results are conservative since the presence of water will increase fire endurance.
Because far less plumbing pipe is found in fire resistive floor/ceiling assemblies, fewer
fire endurance tests have been run with such systems. Test results available are summarized
in Table 3.
TABLE 3-Examples of fire endurance test results for floor/ceiling assemblies.
Hourly Penetrating Structural

Rating Material" System" Testing Agency" Reference
2 pye concrete slab A [34]

2 pye wood frame A [35a,b,c]
1 h, 56 min.c ABS, pye concrete slab on metal B [36]
deck with hung ceil-
ing detail
1 h, 30 min. ABS wood frame A [35a]
"Maximum pipe sizes-3 or 4 in. (7.6 or 10 cm) depending on specific design tested.
bKey to agencies:
A - Underwriter's Laboratories
B-Ohio State University
"In the test of this 2-h assembly, an apparently random failure occurred in the protective hung ceiling
lower membrane, leading to slightly premature failure.
Standard Fire Testing of Unplasticized PVC Products for Electrical End Uses
ASTM E 119 testing of fire resistive assemblies with unplasticized PVC electrical raceway
(and boxes) installed has been conducted. Initially, this test work was directed at demon-
strating the equivalency of the PVC products with competing metal products. Subsequently
testing took the form of more usual performance testing based on the ASTM E 119 standard
alone. Once again, scant test results existed for the metal products with no tests available
for fire resistive wall assemblies including rigid metallic conduit, for example. Tests of wall
assemblies containing various PVC materials are summarized in Table 4.
Based upon analysis of the preceding test results, the Conference of American Building
Officials (CABO) established that PVC raceway can be used in combustible and non-
combustible 2-h assemblies based on 50.8 x 101.6 mm (2 x 4 in.) nominal framing [37].
Three different floor/ceiling designs containing PVC electrical components have been
tested successfully. One, a 3-h design with concrete deck and metal structural system,
included a plenum in which three different combinations of PVC raceway and boxes were
tested. A fourth, control section, included metallic raceway and boxes. The lower membrane
was a fire rated, inorganic tile [38]. This testing verifies that PVC boxes (see also Ref [39]
for an evaluation of PVC box impact in combustible floor/ceiling assemblies) and raceway
can be successfully used in floor/ceiling assemblies without compromising fire performance.
In particular, it addresses the hazard posed when PVC materials melt onto a lower membrane
which has a critical protective function as where, in the test cited above, the lower membrane
provided protection for steel members supporting a concrete deck.
TABLE 4- PVC electrical raceway tested in gypsum clad, fire resistive wall assemblies."
Hourly
Rating pye Material Structural System Reference
2 liz-in. (12.7-mm) rigid non- 2 x 4-in. (50.8 x 101.6 -mm) [6]

metallic conduit nominal wood studs or 3%-
in. (92.08-mm) 22 gauge
steel studs
2 'I.-in. (19.05-mm) electrical non- 3%-in. (92.08-mm) 25 gauge [7]
metallic tubing steel studs
"Both tests run at Underwriter's Laboratories, Inc.

The other two systems tested were based on poured in place slabs of 2-h design. One was
of normal weight concrete 114.3-mm (4lJ2-in.) thick and contained three runs of 25.4-mm
(1 in.) PVC electrical non-metallic tubing (ENT) raceway [40]. The other system, based on
lightweight concrete, was formed on a proprietary corrugated metal pan system [41). Both
of these tests were conducted on 1.829 x 2.438-m (6 x 8-ft) speCimens accordmg to the
ASTM E 119 time temperature curve. Testing was primarily to determine temperature rise
above the ENT on the unexposed surface of the slab or where the non-metallic tubing exited
the unexposed face. Figures 3a and b illustrates test specimens.
Test results for the two unprotected concrete slabs address possible fire risks in using
PVC raceway in a cast in place format. Where possible derating of such assemblies is at
issue, two failure modes are potentially important.
• possible increases in backface temperature rise due to the presence of PVC (electrical
raceways) and
• possible fire spread along PVC raceway, leading from exposed to the unexposed side
or from internal ignition of raceway within the slab to the unexposed side.
In relation to the first item, data from thermocouples placed directly above ENT runs in
these tests showed lower readings than in areas where concrete only was present. This is a
reflection of the lower thermal conductivity of the PVC raceway as compared to the thermal
conductivity of the concrete itself. Figures 4a and b illustrates this for the normal weight
FIG. 4(b)-Comparative time temperature curves for thermocouples associated with metallic (Fig. 4a)
and non-metallic (Fig. 4b) electrical raceways tested in concrete slabs.
slab. Similar results were also obtained with the lightweight test slab which included the
25.4-mm (l-in.) ENT raceway.
Regarding possible fire spread along the PYC raceway, testing showed that no such fire
spread occurred. Rather, the material within the slab charred gradually and temperatures
at locations where ENT exited the test assemblies were below failure criteria for ASTM E 814
(Fig. 4b).
Conversely, an identical specimen, tested with cast in place, thin wall electrical metallic
tubing (EMT) showed higher backface temperatures above the metallic raceway during testing
than in the field of the concrete (Fig. 4a). In addition, at exit locations on the unexposed side,
temperatures also exceeded allowable levels for ASTM E 8 14. Test assembly design was
identical to that of the normal weight slab shown in the preceding Fig. 3. Back face time
temperature curves showed higher temperatures above the EMT as well as at outlet locations on
the unexposed surface of the slab [42]. These readings are a consequence of the high thermal
conductivity of the steel tubing as compared to both the concrete and the PYC raceway.
Nonstandard Sources of Hazard and Risk Assessment Data

In addition to results from standard testing, hazard and risk assessment procedures often
must draw on other data to reach reasonable conclusions. Often a standard test may not
address a foreseeable consequence of the use of a product or the intended use may differ
from the standard assembly result. Thus, engineering judgment, a non-standard test, and/
or field data will be required to provide needed answers. An inescapable fact is that so
many potential end-use configurations exist for pipe, tube, and conduit products (as with
other widely used construction products) that one is precluded from testing of all combinations
of these making some form of hazard assessment a necessity.
In the case of PYC electrical raceway products, for example, portions of the hazard
msessment findings in Ref 1 were based on non-standard test results used in concert with
modeling results to assess probable impact of combustion products created from fire ex-
posures of these products. The findings demonstrated that no unusual hazards existed, based
IIpOn direct exposure of PYC electrical raceway to flame, as did later findings based on
rodent exposures to PYC smoke from the same products [43].
Likewise, potential ignition hazards posed by all electrical raceway systems should be
mnsidered and evaluated although these are not readily subjected to standard test conditions.
However, reports are available on this topic which show that PYC raceways provide intrinsic
safety factors not found in metallic raceway systems due to their inability to fault to ground
[44.45].
It is obvious that redundant fire protection features required in many contemporary
buildings impact fire risk levels there. This was demonstrated for PYC electrical products
in a test conducted when sprinkler protection was added to an older, fire resistive building
in a demonstration of upgrading capabilities of modern construction materials and techniques
[46]. In this case, the active protection provided by sprinklers based on plastic pipe prevented
damage to PYC-ENT installed within fire resistive wall assemblies to such an extent that
electrical continuity was maintained after even the fire test exposure had been conducted.
In actual situations like this test scenario, actual fire loads rarely challenge properly installed
barriers (such as single or multiple layers of wallboard) and the presence of sprinklers further
reduces hazard levels. The presence of these construction features should be taken into
account in preparing a fire hazard assessment of such structures.
The emergence of CPYC and PB as listed materials for fire sprinkler pipe has been
reviewed [47] and reports on the testing to qualify them for listing are also available [48].
Those test results also have implications for pipe uses in DHCW supply based on the same
generic plastics which are qualified for use as sprinklers. For example, in test protocols for
plastic pipe sprinkler systems, the piping, charged with water, is exposed to a wood crib
heptane fire. When the test is begun and following initiation of sprinkler discharge, water
flow is continued through the plastic pipe which is itself exposed to the ongoing wood crib/
heptane fire. This provides adequate evidence of piping integrity and resistance to flame
impingement under conditions where water is present and should relieve those concerned
that CPYC and PB for DHCW uses will derate fire resistive assemblies in which they are
installed.
Concerning non-standard postflashover testing of through penetration of slabs and par-
titions by plastic DWY, two early Canadian studies showed mixed results for tests in which
attempts were made to follow the ASTM E 119 time temperature curve, for small specimens.
These tests also presented different approaches to through penetration fire stopping, in-
cluding application of metal sleeving at through penetration locations to protect pipe for
periods beyond one h [49,50].
Australian researchers conducted tests on both unprotected PYC pipe penetrating concrete
slabs and protected situations where PYC pipes were installed with intumescent fire stopping
systems. Their results reinforced the finding that fire stopping protection for direct, through
penetrations of PYC pipe through slabs was required to maintain fire resistance ratings.
These researchers also conducted tests with metal pipes which showed high unexposed face
temperatures but no transmission of fire [51-53].
Finally, no bona fide field data documenting the existence of identified fire hazards due
to plastic pipe, tube, or conduit as currently used is available. This includes an Australian
study of the subject [54] and searches and inquiries in the United States to various incident
reporting systems and databases by the author have failed to develop field data demonstrating
a fire hazard. Likewise, a search of field data on installations in fire resistive buildings made
under alternate materials and methods sections in the building codes since the 1970s did not
show a recognizable fire incidence related to plastic pipe use. Similar inquiries related to
field performance of electrical raceways based on PVC raceway, whose use increased first
with the 1987 NEC acceptance and incrementally further with the 1990 NEC, have been
negative.
Discussion
The preceding section presented data which will assist in projecting expected fire per-
formance of plastic pipe, tube, and conduit in specific situations. Generalizations which can
be drawn from these data include the following below.
Although the different generic materials used for plastic pipe, tube, and conduit have
been tested to varying degrees in both combustible and non-combustible construction as-
semblies of widely differing types, such testing has not been conducted on every type of
included or penetrating element in every possible fire resistive assembly. The testing which
has been conducted in both standard fire endurance formats (as well as for uses in fire
resistive buildings which include sprinklers based on plastic pipe) supports the following
general ranking where test results for a specific combination of pipe assembly are unavailable.
CPVC > PVC > fire retardant PP > PP, PB, and ABS
The basis for this ranking can be found by comparing tabulated test results in reports
such as Ref 13 and fire endurance periods presented in listing documents such as Ref 8
where differing generic resins, used to fabricate plastic pipe, tube, and conduit have been
tested in identical installations. From a standpoint of measured fire resistance times in side
by side testing, both of those references present data that support this ranking for products
used in plumbing and electrical applications. Likewise, properties such as resistance of
polymer resins to ignition (as expressed by numerical results in ASTM D 2863 testing) are
of importance. Finally, presence or absence of fire retardants (as in the polypropylene
materials cited) and differing melt and charring tendencies of the classes of resins used for
pipe, tube, and conduit will impact fire endurance.
Model Code Acceptance

Each of the three model building codes, the Uniform Building Code (UBC), Standard
Building Code (SBC), and the National Building Code (NBC); upon which most locally
adopted building codes are based, provide for the use of plastic pipe, tube, and conduit in
fire resistive construction in their latest versions. Other specialized or regional codes (such
as the NEC and the National and Standard Plumbing Codes) allow for unlimited use of
plastic pipe, tube, and conduit, as specified. Likewise the ANSI S-40 committee recently
developed a plumbing code allowing for use of plastic pipe in fire resistive construction and
the Conference of American Building Officials, Board for the Coordination of Model Codes
(CABO-BCMC) issued guidelines for use of plastic pipe, tube, and conduit in shaft, duct,
and at through penetration locations several years ago. However, differences exist from
community to community regarding adoption and amendment of these codes. Certain major
'UBC, promulgated by the International Conference of Building Officials(ICBO), Whittier, CA;

SBC, promulgated by the Southern BuildingCode CongressInternational (SBCCI), Birmingham,AL;
NBC, promulgated by the BuildingOfficialsand Code AdministratorsInternational (BOCA), Country
Club Hills, IL.
II) Benjamin, I. A., "Toxic Hazard Analysis: Electrical Non-Metallic Tubing," Journal of Fire Sci-
ences, Vol. 5, Jan.-Feb. 1987, pp. 25-49.
(1) "Plastic Plumbing Pipe," including "Technical Appendices," Draft Environmental Impact Report,
SRI International, (Menlo Park, CA), Department of Housing and Community Development,
State of California, Sacramento, Draft I, 1983, and Draft II, Aug. 1989.
(3) "Standard Guide for the Development of Fire Hazard Assessment Standards for Products," ASTM
Task Group E5.35 on Fire Hazard and Fire Risk Assessment, Committee E-5 on Fire Standards,
Project X-95-7, (draft), 20 April 1990.
(4) "Results of 3 ASTM E 119 I-Hour Tests of Gypsum Wallboard/Steel Stud Assemblies," (from L.
Laymon), UL NC505, Project 76 NK1268, Underwriters Laboratories, Inc., Northbrook, IL, 20
April 1976.
(5) "Report on the Fire Endurance and Hose Stream Tests of PVC Plastic Pipe Penetrations Through
a 7'14' Thick, I-Hour Rated Gypsumboard and Steel Stud Wall," and "Fire Performance of PVC
Pipe in Non-combustible Construction," File WHI-495-PSV-0580 and WHI-495-PSV-0635, War-
nock Hersey International, Inc., Pittsburg, CA, Jan. 1989.
(6) "PVC and Rigid Metallic Conduit and Metallic Outlet Boxes in a Nonbearing Partition Assembly,"
UL NC546-1,-2, Project 73 NK7657, Underwriters Laboratories, Inc., Northbrook, IL, 21 Dec.
1973.
(7] "Electrical Nonmetallic Tubing (ENMT), Electrical Metallic Tubing (EMT). and Metallic Outlet
Boxes in a Nonbearing Partition Assembly," UL R8326-4, Project 80 NK11747, Underwriters
Laboratories, Inc., Northbrook, IL, 17 Sep. 1980.
(8) Fire Resistance Directory 1990, Underwriters Laboratories, Inc., Northbrook, IL, 1990.
(9) Danielsson, c., Linder, P., Lundin, 0., and Prenell, P. A., "Fire Tests with Plastic Tubes Carried
Out at the Research Station in Studsvik, Spring 1963," Report 18:1966, National Swedish Institute
for Building Research, Studsvik, Jan. 1963.
[10] Zicherman, J. B., Unpublished test data, IFf Technical Services, Inc., Berkeley, CA, 1988.
[11] Fang, J. B., "Static Pressures Produced by Room Fires," NBS1R 80-1984, Centerfor Fire Research,
National Institute of Standards and Technology (formerly National Bureau of Standards), Gaith-
ersburg, MD, 1984.
(12] Choi, K. K., "Fire Stops for Plastic Pipe," Fire Technology, Vol. 23, No.4, Nov. 1987, pp. 267-
279.
(13] Parker, W. J., Paabo, M., Scott, J. T., Gross, D" and Benjamin, I. A .. "Fire Endurance of
Gypsum Board Walls and Chases Containing Plastic and Metallic Drain. Waste and Vent Plumbing
Systems," National Bureau of Standards Building Science Series 72, U.S. Department of Com-
merce, Sept. 1975.
(14) Bletzacker, R. W. and Birle, J. G" "Standard ASTM Fire Endurance Test and a Fire and Hose
Stream Test on Duplicate Non-load Bearing Plumbing Wall Assemblies," Report 5615, Ohio State
University, Building Research Laboratory, College of Engineering, Jan. 1974.
(15) Bletzacker, R. W. and Birle, J. G" "Standard ASTM Fire Endurance Test and Fire and Hose
Stream Test on Duplicate Load Bearing Polyvinylchloride Plumbing Wall Assemblies." Report
5561, Ohio State University, Building Research Laboratory, College of Engineering. May 197.t.
[16] Bletzacker, R. W. and Birle, J. G., "Standard ASTM Fire Endurance Test and Fire and Hose
Stream Test on Duplicate Non-load Bearing Polyvinylchloride (PYe) Plumbing Wall Assemblies,"
Report 5474, Ohio State University, Building Research Laboratory, College of Engineering, April
1973.
[17] Bletzacker, R. W. "Standard ASTM Fire Endurance Test and Fire and Hose Stream Test on
Duplicate Load-Bearing Acrylonitrile-Butadiene-Styrene Plumbing Wall Assemblies," Report 5560,
Ohio State University, Building Research Laboratory, College of Engineering, 1974.
[18] Bletzacker, R. W. "Standard ASTM Fire Endurance Test and Fire and Hose Stream Test on
Duplicate Non-Load Bearing Acrylonitrile-Butadiene-Styrene (ABS) Plumbing Wall Assemblies,"
Report 5560, Ohio State University, Building Research Laboratory, College of Engineering, 1973.
[19] Draemel, R. B. and Williamson, R. B., "Fire Tests of Six Inch Wood Stud One-hour Fire-rated
Walls with Plastic DWY Plumbing Systems," Service to Industry Report 76-7, University of Cal-
ifornia, Berkeley, Department of Civil Engineering, Structural Research Lab, Nov. 1976.
[20] Draemel, R. B. and Williamson, R. B., "Fire Test of a Six Inch Wood Stud One-Hour Fire-Rated
Wall with a Fire Retardant Polypropylene DWY Plumbing System," Service to Industry Report
7702, University of California, Department of Civil Engineering, Structural Research Lab, Berke-
ley, 1977.
[21) Williamson, R. B., "Fire Test of a Six Inch Wood Stud One-hour Fire-rated Wall with a Poly-
vinylchloride DWY Plumbing System, Structures, Materials and Fire Research," University of
California, Berkeley, Department of Civil Engineering, Structural Research Lab, Feb. 1985.
[22] "Report on the Fire Endurance and Hose Stream Tests of PYC Plastic Pipe Penetrations Through
an 8'12' Thick, 2-Hour Rated Gypsumboard and Steel Stud Wall," and "Fire Pcrformance of PYC
Pipe in Non-combustible Construction," Files WHI-495-PSY-0580 and WHI-495-PSY-0635, War-
nock Hersey International, Inc" Pittsburg, CA, Jan. 1989.
[23] Plumbers' and Steamfitters Union Local 467/Goldberg Research and Development Associates
Corp., "The Effects of ABS Plastic Plumbing Runs on a Wood Stud and Gypsum Wallboard
Partition Exposed to a Standard Fire Endurance Test," File WHI-495-0456, Warnock Hersey
International, Inc., Pittsburg, CA, 8 Oct. 1982.
[24] Zicherman, J. B., "Fire Safety of PYC-Based Plastic Piping," Fire Journal, Yol. 84, No.6, Nov.!
Dec. 1990, pp. 36-42.
[25] "Standard Fire Endurance Test Program To Evaluate and Compare the Fire Resistance of Two
Types of ABS Plastic Pipe When They Penetrate a Fire-Resistive Wall," 1FT Technical Services,
Inc., Report 5213, Coquitlam, B.C., Canada, Warnock Hersey Professional Services, Ltd., Pitts-
burg, CA, 16 Jan. 1987.
[26] Plastic Pipe and Fittings Association (PPFA), Plastic Pipe in Fire Resistive Construction, The Design
and Installation Manual for Plastic Pipe in Fire Resistive Construction, PPFA, Glen Ellyn, IL,
1985.
[27] "Fire Test Investigation of Through-Penetration Firestop Systems in Gypsum Wallboard/Steel Stud
Wall Assemblies," UL R9700, Project 85 NK17181, 3M Company, Underwriters Laboratories,
Inc., Northbrook, IL, 18 Sept. 1985.
[28] "Report on Through-Penetration Fire Stop Devices," UL RIOI00-l,2, Project 83 NK19794, Orion
Industries, Inc., Kansas City, KS, Underwriters Laboratories, Inc., Northbrook, IL, 28 Nov. 1984.
[29] "Report on Wall Opening Protective-Devices-in Concrete or Masonry Walls Assemblies," UL
R9269-1,2 and R9700, Projects 81 NK4314 and 81 NK7951, 18 Dec. 1981; Letters, "Report Types
PSS 7902 and PSS 7904 Through-Penetration Firestop Devices in Concrete Floor Assemblies,"
UL R9269, Project 82 NK2200, 6 May 1983, and Letter Report R9269, 31 March 1983, Underwriters
Laboratories, Inc., Northbrook, IL.
[30] Meyer, G. and Brittain, W., "Two Hour Fire Resistance and Hose Stream Tests of R & G Sloane
Fire Yalve," WHI File 495.0621 and 495.0622, Warnock Hersey International (WHI), Pittsburg,
CA. 1984.
[31] "Report on Through-Penetration Fire Stop Devices," UL RIOIOOO-l,2, Project 83 NK19794, Orion
Industries, Inc., Kansas City, KS, Underwriters Laboratories, Inc., Northbrook, IL, 28 Nov. 1984.
[32] "ASTM E 814-83 Standard Method of Fire Tests of Through-Penetration Fire Stops: Polybutylene
Pipe Penetrations of a I-Hour Rated Fire-Resistive Wall Assembly," Project 01-8305-028, SRI,
Menlo Park, CA, May 1985.
[33] "Fire Test Evaluation on Polybutylene (PB) Plastic Piping in a Steel and Wood Studded Gypsum
Board Partitions," Design WP 735 for Shell Chemical Company, Houston, TX, File J.r. OK9Q6.AC
(4510), Factory Mutual Research, Norwood, MA, 13 Nov. 1984.
[34] "Report on Through-Penetration Fire Stop Devices in Concrete Floor Assemblies," UL R10338-
1,2,3, Project 83 NR4820 for Proset Systems, Atlanta, GA, Underwriters Laboratories, Inc.,
Northbrook, IL, 25 Nov. 1983.
Fire Resistance Directory 1990, System 16ll, 167, and 3m, Underwriters Laboratories, Inc .. North-
brook, IL 1990, pp. 1264, 1265, and 13H2.
Bletzacker, R. W. and Birle, J G., "Standard ASTM Fire Endurance Test on a Floor and Ceiling
Assembly," Report 5539, Ohio State University, Plastic Pipe Institute, Building Research Labo-
ratory, Columbus, OK May 1974.
Council of American Building Officials (CABO), "Fire Resistive Noncombustible Partition As-
sembly Containing Electrical Non-Metallic Tubing and Rigid Non-Metallic Conduit," Report NER-
290, Carlon, National Evaluation Service Committee, Country Club Hills, IL, May 19H7.
"Fact-Finding Report on Metallic and Nonmetallic Tubing, Conduits and Boxes in the Concealed
Space of Floor-Ceiling and Roof-Ceiling Assemblies with Suspended Ceiling," UL NC 546-5,
Project 7H NK27319, NEMA/UL Underwriters Laboratories, Inc., Northbrook, IL, Attention B.
Swytnyk, March 19H9.
"Nonmetallic Electrical Outlet Boxes for Use in Floor-Ceiling Assemblies Consisting of Wood
Joists, Wood Flooring and Gypsum Wallboard," UL RR326, Project HI NK 24419, Carlon, Un-
derwriters Laboratories, Inc., Northbrook, IL, 22 Dec. 19HI.
"Report of the Pilot Scale Fire Endurance Test of an Unrestrained 6' by go by 4'!!' Thick, Reinforced
Concrete Slab withollt an Applied Load, and with 1" Diameter, PVc, Flexible, ENT Conduit
Runs," File WHI 495-PSH-1065, 1FT Technical Services, Inc., Warnock Hersey InternationaL
Inc., Pittsburg, CA, 7 May 1990.
"Report of the Pilot Scale Fire Endurance Test of 1" Diameter, PVc, Flexible. ENT Conduit
Runs in an Unrestrained 6' by R' by 5 'I;' Thick, Reinforced Concrete/Corrugated Steel Deck Floor/
Ceiling Assembly withollt an Applied Load." File WHI-495-PSH-llI69, 1FT Technical Services,
Inc., Warnock Hersey International, Inc., Pittsburg, CA, 16 July 1991l.
"Report of the Pilot Scale Fire Endurance Test of an Unrestrained 6' by go by 4'!!' Thick. Reinforced
Concrete Slab without an Applied Load, and with I" Diameter, Steel, EMT Conduit Runs," File
WHI-495-PSH-ll17ll, 1FT Technical Services. Inc., Warnock Hersey International, Inc., Pittsburg,
CA, 17 July 1990.
Packham, S. C, and Crawford, M. P .. "An Evaluation of Smoke Toxicity and Toxic Hazard of
Electrical Non-Metallic Tubing Combustion Products," Journal of Fire Sciences, Vo\. 2, Jan.-Feb.
19R4, pp. 37-59.
Kahn, M. M., "Electrical Failure of Wires Inside I-inch Conduit Under Simulated Fire Condi-
tions," Technical Report fMRC J.I., OH4R4.Rc' Carlon, Factory Mutual Research Corporation,
Norwood, MA, Oct. 1984.
Fuller, J. F., Hanna, W. J.. and Kallenbach, G. A., "Arcing Faults in Metallic Conduit at 120
and 240 v," IEEE Transactions on Industry Applications, VoL IA-2l, No.4. May/June 19R5.
Benjamin/Clarke Associates, Inc., "Operation San Francisco: Smoke/Sprinkler Test. Technical
Report," distributed by International Association of Fire Chiefs (IAFC) , Marriot Corporation,
San Francisco Fire Dept., and FEMA/USFA, Washington, DC, Apri119R4.
Wilging, R. c" "Plastic Fire Sprinkler Piping 1967-19R7," The Building Official and Code Ad-
ministrator, July/Aug. 19R8.
"Report on CPVC Pipe and Fittings for Sprinkler Systems," UL Ex3754, Project R5 NKH37,
Underwriters Laboratories. Inc., Northbrook, IL, 17 July 19R5.
Atwood, P., "Penetration of Fire Partitions by Plastic Pipe," Fire Technology, Vo\. 16, No. I,
July 1980, pp. 37-62.
McGuire, J. H., "Penetrations of Fire Partitions by Plastic DWV Pipe," Fire Technology. Vo\. 9,
No. I, Feb. 1973, p. 5.
Burn, L. S. and Martin, K. G., "Intumescent Fire Stoppers For UPVC Pipes Penetrating Concrete
Slabs," Commonwealth Scientific and Industrial Research Organization, Division of Building
Research, Highett, Victoria, Australia, 19RO.
Brown, S. K. and Martin, K. G., "Model Fire Resistance Tests of UPVC Pipes Penetrating
Concrete Slabs," Commonwealth Scientific and Industrial Research Organization, Division of
Building Research, Highett, Victoria, Australia, 1979.
Burn, L. S. and Martin, K. G .. "Early Fire Hazard Assessment of UPVC Pipe Formulations,"
Commonwealth Scientific and Industrial Research Organization, Division of Building Research,
Highett, Victoria, Australia, 1981.
Brown, S. K., "Review of Actual and Simulated Fires Involving Plastics and Pipes and Fitting,"
Commonwealth Scientific and Industrial Research Organization, Division of Building Research,
Highett, Victoria, Australia, 1979.
Fire Hazard Assessment
Philip J. DiNenno1 and Craig L. Beyler1
Fire Hazard Assessment of Composite

Materials: The Use and Limitations of
Current Hazard Analysis Methodology
REFERENCE: DiNenno, P. J. and Seyler, C. L., "Fire Hazard Assessment of Composite
Materials: The Use and Limitations of Current Hazard Analysis Methodology," Fire Hazard
and Fire Risk Assessment, ASTM STP 1150, Marcelo M. Hirschler, Ed., American Society
for Testing and Materials, Philadelphia, 1992, pp. 87-99.
ABSTRACT: This paper reviews the application of current fire hazard analysis methods to
the evaluation of composite materials. Problems and limitations of hazard assessment pro-
cedures are identified, and alternative techniques and strategies are presented. The application
of the HAZARD I fire hazard assessment procedure to the use of composites is addressed
explicitly. Problem areas include: development of acceptance criteria, scenario development,
and hazard calculation. Specifically, scaling problems, flame spread, exposure quantification,
and mechanical response of composites to fire exposure are addressed. It is suggested that
significant limitations in both the approach and technical capability of current hazard assess-
ment procedures are significant. These limitations however do not preclude hazard analyses
per se.
KEY WORDS: hazard analysis, fire safety of composite materials, HAZARD I, exposure
fires, scaling, fire modeling
The accelerating use of composite materials and the development of current fire hazard
analysis methodologies present a unique opportunity in the area of fire safety engineering,
In effect, a set of new technological materials is proposed for use in existing applications.
Many of these applications have been previously subjected to minimal fire hazard assessment.
On the other hand, tremendous progress has been made in the area of fire hazard assessment
methods, Complete hazard analysis packages are readily available in an inexpensive user
friendly format [1],
The composite applications provide an opportunity to exploit the hazard analysis methods,
In fact, the rational use of these materials demands the use of non-traditional approaches.
Traditional approaches to regulating the fire properties of materials would prescribe
arbitrary pass/fail requirements using standardized fire test procedures, It has been widely
accepted that these test results and requirements do not directly relate to expected full-scale
performance, Two developments in the past 15 years have enabled the fire protection
community to proceed beyond this approach, They are: 1) the development of mathematical
descriptions of fire growth, and 2) the development of small scale test procedures (such as
small scale calorimeters (2,3] and LIIT flame spread apparatus [4J.
These developments led to the ability, in principle, to predict the full scale performance
of materials, as indicated by the development of hazardous conditions in a room or building,
through the use of computer based fire models and the results of improved small scale test
IVice president and technical director, respectively, Hughes Associates, Inc., 6770 Oak Hall Lane,
Suite 125, Columbia. MD 21045.
87
methods. The basic principle is to use the fire model to predict the environment of the
burning material, use the small scale test results to estimate the response of the material to
this exposure, and to use the model again to calculate the impact of this burning on room
temperature, smoke layer depth, gas concentration, and so forth. In general, the model
predicts nothing relative to the burning behavior of the material. Information on mass loss
and energy release rates and smoke and gas yields is obtained from small scale calorimeter
data. This enables one to estimate the effect of material/product flammability properties on
the onset of hazardous conditions; which is the primary objective of the regulation of fire
properties of materials. This basic approach has been used for over 10 years. Pioneering
efforts in the field have been performed by Smith [5,6] and others [7-9].
The use of current hazard assessment methods in evaluating the application of composite
materials has identified significant limitations and uncertainties in the hazard assessment
process. The purpose of this paper is to identify some of these problem areas and to propose
alternative solutions where possible.
A point of departure for this paper will be the generic hazard analysis scheme outlined
in Ref 10, which is the prototype use of HAZARD I. The basic steps of the procedure
include:
1. development of acceptance criteria;

2. scenario development;
3. calculation of hazard/outcome; and
4. evaluation of consequences.
The results of the early stages of fire hazard assessment relative to application of composites
in submarines and surface ships, form interesting lessons relative to the efficacy of this
prototype hazard assessment scheme.
Acceptance Criteria
The development of acceptance criteria is an aspect of hazard analysis which is generally
given minimal attention in formalized hazard analysis methodologies. The process of defining
acceptance criteria is a joint effort by the analyst and the hazard manager/decision maker.
In some very simple situations, the criteria may simply be that life safety hazards should be
no greater than existing installations or no greater than a code complying design. In other
situations, the decision-making process may be significantly more complex.
In the applications discussed in this paper, there are no legal/regulatory based objectives
to be met, and simple fire performance equivalency is unachievable. The benefits of using
composites are largely in non-fire areas with the trade-offs involving life safety, economic
costs, and a wide range of mission oriented objectives. Such wide ranging trade-offs cannot
readily be reduced to a single unit of measure to allow the formulation of a quantitative
hazard acceptance criterion as is required by the prototype hazard assessment scheme.
Further, the decision makers are not generally able to articulate their fire safety objectives
in terms which can readily be adopted in the prototype hazard assessment scheme. Typical
decision maker objectives include:
• small fires should not grow quickly;
• fire/smoke should not generate untenable conditions quickly;
• structural performance should be maintained for a period of one hour under load; and
• fire should be extinguishable with typical agents.
Given the complex and diverse costs and benefits associated with the use of composites,
and the difficulty of reducing even fire related objectives to a single unit of measure with
DINENNO AND SEYLER ON HAZARD ANALYSIS METHODOLOGY 89
a quantitative acceptance criterion, the whole notion of requiring a quantifiable acceptance

criterion at the outset appears to be counter productive.
A lesson from this experience may be that the hazard analysis process may proceed without
specification of acceptance criteria. This enables the analysis to define the expected per-
formance under a range of scenarios or test conditions and the decision maker is given an
opportunity to weigh this impact against a range of use/exposure conditions, while not forcing
acceptability decisions up front. It has the additional benefit in that complex risk/benefit
decisions do not lend themselves to single variable quantification, and setting quantified
acceptance conditions beforehand may not make sense.
Another way of viewing the hazard modeling process is that it forms the analytical back-
ground (through small scale testing and modeling) which enables one to screen the material/
application. A barrage of moderate to full scale performance tests can be viewed as the
"validation" of the analysis process and the decision maker is given a "real scale" context
for his decision making. Thus, small scale testing and modeling is used as a preliminary
evaluation tool to establish if full scale testing is a reasonable use of resources. Full scale
experimentation is no substitute for the hazard analysis process. Given the significant lim-
itations of the state-of-the-art fire models, full scale testing is likely to be needed to contribute
to the knowledge base required for hazard analysis for the foreseeable future.
Scenario Development
The goal of this task is to set the conditions under which the hazard of the material is to
be evaluated. It includes the product/material in its end use condition and a well characterized
structure and contents. By some definitions it also includes "well characterized" occupants
[10].
Two possible applications of composite materials as representative of the range of possible
applications should be considered. At one extreme, one might wish to replace steel pipe
hangers with composite materials. In most applications, the quantity of material is quite
small. At the other extreme, one might wish to replace a steel deckhouse structure with
composites. This involves switching all exterior and interior bulkheads from steel to com-
bustible composites. The key differences in scenario development between these two ap-
plications are related to the magnitude of the potential hazard, the range of scenarios
possible, and the analytical tools needed to properly assess their impact on fire hazard.
For the case of limited, relatively isolated use of a material, for example, a pipe hanger,
it may be possible to lump all the material in a compartment to be used as pipe hangers,
and assume it all burns at the same time. Issues such as heat release rate, toxicity, smoke
production, and so forth, may be insignificant if a) the quantity (surface area and mass) is
limited, and b) the material properties from small scale tests are similar or better than those
for the majority of the combustibles in the compartment. A simple way of evaluating this
type of composite use is given in Ref 11. A simple way to estimate the impact of a small
amount of material burning simultaneously in a closed compartment is given.
In this approach one "scales" the results of cone calorimeter tests by assuming the com-
bustion products mix uniformly throughout the volume of a closed compartment. An example
result of this type of analysis is given in Figure 1. This is a plot of the allowable combination
of extinction area (CTm), mass loss rate m and surface area of the composite A versus time
that yields a visibility of 1 m, for 3 separate compartment volumes. Similar plots have been
developed for CO-dose (ppm-min) and compartment temperature, as well as for other
visibility levels. Note that the extinction area and mass loss rate of the sample are obtained
from the cone calorimeter at a selected heat flux exposure. It assumes all of the surface
area material is burning simultaneously.
This procedure is clearly quite simple and economical to carry out and can be used in a
conservative manner to identify applications which pose little or no additional hazard above
and beyond the existing base case. In this sense, simple economical methods can be used
as a screening tool to determine if a full blown hazard analysis is required or justified.
Similarly, simple methods can also be used to demonstrate applications which are so haz-
ardous that they can be eliminated without recourse to a full blown hazard analysis. Such
incremental approaches to the hazard analysis process are not included in the prototype
hazard assessment scheme.
The other example of replacing steel bulkheads with composite panels poses a much more
challenging hazard analysis problem. There are several important observations relative to
this problem.
1. Current hazard assessment methods are not capable of addressing the range of im-
portant physical phenomenon, (for example, flame spread, fire resistance, fire duration,
detailed burning/spread descriptions, and so forth.)
2. The range of exposure/use scenarios is very large.
3. Scaling problems in the context of small scale material performance vs. full scale
assembly performance requires resolution.
Bear in mind that one is essentially replacing a non-combustible wall/ceiling/floor material

with an organic combustible. This implies that for most known composite materials under
some expected exposure conditions, the bulkhead materials will form some increased hazard
and risk. They do not have the potential for "equivalent" behavior.
A major problem encountered in the assessment of a material in this application is the
definition of fire exposures under which the material should be evaluated. Assuming that
calorimeter data in the range of 10 to 100 kW/m2 and LIFf flame spread data is available,
the important question becomes what exposure levels should be used in the evaluation.
Since in this application the material is subject to a wide range of possible exposures, it is
useful to categorize the possible scenarios. The range of exposure heat fluxes can be arbi-
trarily divided into three categories. This is done to facilitate discussion on developing
ignition/exposure scenarios.
Small Exposures
The objective of these exposures is to evaluate the ignition properties of the material in
the absence of an exposure fire, that is, the ability of the material to resist ignition from
small sources approximating small open flames, localized single cable jacket ignition, and
so forth. Because thick composite materials cannot be ignited by such ignition sources, this
is not an issue for the composites discussed in this paper and is included for sake of com-
pleteness. There is an unresolved problem of relating these small ignition sources to small
scale ignition data from the cone or LIFT apparatus. It is obvious that a composite in these
uses should not ignite from contact with a match. Perhaps this can be stated in terms of a
modest flux (=25 kW/m2) for a limited time (say 60 to 90 s). This issue has not been addressed
in hazard analysis per se, but is indirectly associated with material performance in standard
fire tests (ASTM Method for Surface Burning Characteristics of Building Materials, E 84
and ASTM Method for Surface Flammability of Materials Using a Radiant Heat Energy
Source, E 162, and so forth).
Moderate Scale Exposures
This class of exposures describes relatively modest fires (s250 kW) expected from trash
containers, sea bags, isolated cable fires, and so forth, with a duration on the order of
several minutes. In general, these fires in and of themselves pose no great hazard. The goal
here is to evaluate the incremental increase in hazard associated with the combination of
the moderate scale exposure fire and the composite. The increased hazard may be associated
with enhanced fire growth or the generation of toxic species. Over a certain range of exposure
fire size, the composite may cause flashover in scenarios which, without the composite,
would not lead to full room involvement.
The objective of evaluating the material under small to moderate exposures is to evaluate
the contribution of the material given a fire which in and of itself is not a major threat. One
seeks to define the range of exposure fires which would not involve the composite bulkhead
and the range of exposure fires in which the composite wall would allow significant fire
growth beyond the initial exposure fire.
Severe Exposure
The third category of exposure level involves exposure times which alone, are capable of
causing room flashover. In these scenarios, incident heat fluxes expected to the composite
are those from post-flashover fires. Non-contribution of the material is not an issue at this
point. Rather, the rate at which the material contributes fuel vapor, smoke, and so forth,
to the effluent gas stream from the compartment is relevant. Within the compartment, it is
unlikely that the conditions will be worsened by the composites, but increases in the duration
of the post-flashover stage are likely. Outside the compartment, the volume of un burnt
pyrolyzates flowing to other spaces may be increased.
Scenario Selection
The wide range of possible exposure conditions, as well as an almost infinite combination
of ship/personnel conditions, severely constrains the utility of hazard methodologies like
HAZARD I. While exact descriptions and location of personnel, the distribution, quantity,
type, location, and geometry of other combustibles is possible, the use of one, two, or ten
of these "well described" compartments is largely arbitrary and does not add to the under-
standing of the performance of the material. The vast number of variables and the potential
range of these variables is so broad that hundreds of simulations would be required to
represent the hazards. While this may, in isolated instances, be economically feasible, more
often this level of effort is not plausible. It is simply not feasible to represent all of the
possible or even significant combinations of conditions. Resorting to statistics in an attempt
to develop scenarios is also unfulfilling due to classic limitations in the quantity and quality
of fire experience data available. Further, the absence of composites in the historical and
statistical base limits the utility of this approach.
Fortunately, it is also unnecessary. It is not required that one have detailed descriptions
of the outcome of any particular scenario. It is much more useful to develop a description
of the performance limits of the material, that is, the conditions under which the material
will contribute (substantially) to the hazard and the nature of the contribution to the hazard.
In summary, the task of scenario development description in the HAZARD I sense is
not necessary, and certainly does not add to the understanding of the hazard changes due
to the composite material. This does not obviate the value of considering the range of
possible scenarios. It simply indicates the impossibility of enumerating all possible scenarios
and that sufficient insights into the consequences can be developed even in the absence of
specific scenario descriptions.
Cakulating Hazard
This step in the hazard analysis process poses significant technological challenges. Pre-
viously discussed issues of scenario development and acceptance criteria are arguably issues
of style and approach. The task of calculating hazard however, forms some of the most
difficult challenges in the assessment of fire hazard.
There are well known limitations of hazard modeling [12-14]. It is not the intent of this
paper to catalog these limitations, but to focus on problem areas that preclude even the
simplest hazard assessment. For the sake of brevity, we will limit all discussion to processes
occurring within the room of origin, so as not to complicate the problem with downstream
smoke transport problems. Furthermore, we do not include human behavior aspects of fire
hazard.
The discussion is cast into five generic areas: exposure fires, flame spread, scaling prob-
lems, post-flashover, and mechanical performance under fire conditions.
These can be indirectly related to issues associated with the performance of a material at
various steps in a typical fire development process; from a small steady fire exposure through
an early growth stage (flame spread), a later growth stage where compartment/scale inter-
actions occur, and into post-flashover.
Exposure Fires
In the context of HAZARD I, either a prescribed fire exposes a product, or the product
evaluated is a prescribed fire. For the shipboard applications discussed in this paper, the
product is not a prescribed fire, and, in general, for very early stages of the fire, a prescribed
fire exposes the composite. The exposure to the composite can be via direct flame impinge-
ment or by remote radiative heating.
The calculation of the radiative exposure to a remote target can be done reliably for a
wide range of exposure fires if one assumes the radiation transfer occurs as if the source
and target were in the open, with no compartment effects [15,16]. If compartment effects
begin to become important (for example, hot layer temperature/emissivity, lower layer
absorptivity, or wall/cover/ceiling flames) the calculation becomes more difficult and its
reliability is compromised.
Limiting attention for the moment to early fire times where these effects are not important,
the problem still remains of calculating the exposure due to direct flame impingement. One
must define the flux levels seen by a composite directly exposed to the flame of an exposure
fire. In this case we limit our concern to moderate exposure fires in the range of 50 to 500
kW in total energy release rate.
Due to a dearth of data available on this type of exposure, a series of experiments was
conducted to quantify the spatial distribution of total heat flux to isolated targets, walls,
and corners for steady fire sizes of 50, 100, 200, and 500 kW [17]. The fire sources were
propane burners, with varying flame height/source diameter aspect ratios. Figures 2 and 3
give some results of these experiments.
The most striking result of Figure 2, which is a plot of maximum wall heat flux versus
source fire energy release rate, is the relatively high fluxes that are expected even for small
(50 to 100 kW) fires. Figure 3 illustrates that these high heat fluxes are realized over 40%
of the flame height. These peak fluxes and areas of exposure imply that performance based
on material response characteristics measured in the cone calorimeter at less than 50 kW/
m2, may not be representative of the material's response to realistic exposure fires. Further
and perhaps more importantly, if one wishes to limit or eliminate the contribution of a
composite to a growing fire, ignition thresholds must be increased substantially over those
currently available.
It should be remembered that there is no current description of this direct flame exposure
process in any zone model. However, by independent calculation a source/composite fire
could be prescribed and the hazard model would perform the necessary bookkeeping func-
tions for smoke/gas concentration and heat loss calculations to estimate temperatures, en-
abling one to estimate the impact of the "contribution" of the composite.
Flame Spread
Having determined that the composites of interest will contribute to a relatively small
steady fire, the subsequent crucial issue is, will the fire spread via flame spread processes
beyond the area directly exposed, and, if so, how quickly? At the time these composite fire
hazards were being evaluated, there were not any vertical flame spread prediction models
available. There are now at least four [18-21] in varying stages of development. The problem,
however, is still unresolved.
As a first approximation, it may be productive to pursue a model describing if flame
spread will occur beyond the boundaries of the source fire flame area, instead of pursuing
detailed descriptions of how fast and where the flame will spread. Saito, et al [22], and
Mowrer and Williamson [20] have suggested criteria to distinguish between acceleratory and
deceleratory spread. These concepts have not been pursued nor have they been applied to
determine the conditions where no vertical spread is expected.
Scaling Problems
Problems associated with scaling up (or modeling) the behavior of a material from a small
scale test method to full scale can be divided into two categories:
I. Scaling problems associated with test methods

2. Modeling problems related to predicting full scale performance.
The current status of hazard analysis requires the assumption that the behavior of a
material to any fire exposure can be related to its behavior in several selected small scale
tests. The most critical of these is the small scale calorimeter, the current instrument of
choice being the cone calorimeter (ASTM Method for Heat and Visible Smoke Release
Rates for Materials and Products Using an Oxygen Consumption Calorimeter, E 1354).
Several problems are encountered in the use of data extrapolated from these tests. These
include the following:
1. Unsteady behavior-The heat release, mass loss, and species production rates do not
approximate quasi steady conditions. This implies that assumptions of a single rep-
resentative value of heat release rate at a given exposure are not possible. This is an
implicit assumption in current hazard modeling.
2. Edge effects-Composite materials often demonstrate the behavior of pressure buildup
internal to the species. This pressure buildup sometimes causes gases to escape via the
edges of the relatively small sample. This behavior would not be expected generally
in full scale applications.
3. Exposure-There is an implicit assumption that independent of the size of the exposure
fire, the only relevant parameter is radiative heat flux. It is unclear whether the cone
calorimeter data can be reliably used as involved area increases significantly beyond
0.025 m2•
4. Vitiation effects- Assuming the concept of a gas yield is valid as the scale increases,
the issue of gas yields in decreased oxygen environments must be further addressed.
Approaches given by Beyler [23 ,24] and Zukoski et al. [25], apparently demonstrate
the feasibility of calculating gas yields in vitiated two layer environments. Test pro-
cedures and associated modeling calculations must be developed.
The second category of modeling/scaling problems is associated with the predictive ability
of hazard models to integrate small scale test results with compartment and environmental
effects to estimate hazard variables such as temperature, gas concentrations, layer height,
and so forth.
In the simplest case, a closed room (ASET-like) no vitiation, no radiative feedback,
"model" is quite simple, and effectively "integrates" the small scale test results through
simple bookkeeping of gas/scale variables, while estimating heat loss and layer depth. No
environmental effects are considered.
Compartment induced environmental effects which form the primary reason for the use
of room fire models include:
1. Increased radiation from hot layer and/or other flames;

2. Two layer vitiation effects;

3. Room geometry/size, and so forth, (heat loss terms);
4. Ventilation (natural and forced); and
s. Other room details.
There is no current available model which can treat, in any general sense, any of these
effects except heat loss and natural vents. The impact of increased radiation on burning rate
(through some calorimeter-like assumptions) is not possible. Calculations of two layer vi-
tiation effects are feasible for some fuels, but certainly not composites, and forced ventilation
is feasible under certain assumptions regarding the distribution of mass (by layer) [26].
There are many subtle geometry effects such as the impact of deep beam/stiffener pockets
[27], flame properties in wall/corner geometries, or impact of local forced ventilation. All
are potentially important in shipboard applications and may not be treatable in the context
of a zone model.
Other scaling problems that are associated with the performance of an assembly versus
a material include: the behavior of joints, corners, edges, small openings, and so forth; all
are potentially crucial in evaluating the performance of a bulkhead system and are well
beyond the range of current hazard models. The applicability of the current generation of
hazard models is quite limited in the context of composites. For the simplest cases, a detailed
room model is not required, and for the cases of real interest there is virtually no predictive
capability.
Post-Flashover Issues
Once a typical compartment fire has grown to a size sufficient to cause upper layer
temperatures on the order of 550°C, and all combustible materials have ignited, this threshold
termed flashover marks the point in the fire development where material flammability
properties related to fire growth, flame spread, and so forth, are no longer important. The
energy release rate of a post-flashover fire will generally be limited to the amount of ven-
tilation available. Hence, increases in the amount of fuel available will not substantially
alter the temperature course of a fire.
For small quantities of material in the compartment there will be no noticeable increase
in the fire hazard after flashover. Additional quantities of composites will result in longer
post-flashover fire durations proportional to the increase in total combustible load. As long
as the surface area of these additional combustibles is minor, this will be the only expected
effect.
In general, however, one expects additional quantities of combustibles to form an addi-
tional hazard, which is one caused by the production of excess fuel vapor or excess pyrolyzate.
This results from the fact that all surfaces in the compartment are exposed to relatively high
(>100 kW/m2) heat fluxes, causing pyrolysis. Since the amount of combustible vapor which
can be burned in the compartment is limited by available ventilation, this excess fuel vapor
exits the compartment through natural or forced exhaust points. The fire hazard associated
with the vapor is that if sufficient quantities are generated, ignition of the exhaust gases will
occur. If the exhaust point is an opening to a passageway or corridor, very rapid fire spread
down the corridor will be expected to occur. In addition, this energy release occurring in a
corridor may result in much more rapid smoke propagation.
This potentially important problem has not been effectively studied or quantified. A first
order approximation is to allow all excess fuel vapor to ignite and burn in the adjacent
corridor and to ensure that the quantities of fuel vapor produced, and hence energy released,
are minimized. There is no current method for evaluating what limits should be placed on
this excess pyrolyzate production.
Mechanical Performance Under Fire Exposure

The traditional approach to assessing the residual strength of a structural assembly is by
exposure to the ASTM Method for Fire Tests of Building Construction and Materials (E
119) test procedure. This test simulates the exposure to a single post-flashover time-
temperature curve.
As such, the test does not consider the possibility of structural damage by direct flame
impingement in the pre-flashover stage. Further, even for post-flashover fires, the test results
cannot be related to the wide range of possible post-flashover compartment fires. For
composites onboard ships, difficulties arise in both the pre and post-flashover stage. Damage
to composite structures can occur due to pre-flashover direct flame impingement. As Fig.
2 indicates, fluxes in excess of 100 kW/m2 can be expected. In post-flashover situations the
time-temperature curves can differ markedly from the E 119 curve. Far more severe ex-
posures have been observed and documented in shipboard compartments [28].
Because of these diverse exposures, a modeling approach has been adopted which includes
three major components:
1. Heat transfer/thermal decomposition of the composite.

2. Determination of high temperature composite structural properties.
3. Finite element modeling of the deformation of the composite structure due to heat
and load.
These modeling/experimental tasks are being pursued initially using polyphenylene sulfide
(PPS)/glass composites.
The heat transfer/thermal decomposition modeling effort is being performed by Professor
Arvind Atreya (Michigan State University), building on his prior models for wood [29].
Single step Arhenius reactions for melting and decomposition are included as are temperature
dependent thermal properties. High temperature structural properties of the composites are
being measured by Professor Vizzini and James Milke (University of Maryland). Ply prop-
erties will be generalized to arbitrary layups using tensor rotation methods. Finite element
modeling of the structural response of the composite makes use of the thermal/decomposition
model and the high temperature structural properties. The model being used is a general-
ization of that developed by Kim and Lee [30]. These models are being validated by com-
parison with experimental results using a range of tensile loading conditions and heat fluxes.
Progress to date has been encouraging.
While this type of detailed structural modeling is very flexibl~ and widely useful, it does
not preclude the need for full scale testing. Full scale verification testing to assure that the
models include all the relevant phenomena will be required for the foreseeable future. The
effects of edges, joints, and other structural details are not included in current models and
need some full scale evaluation.
The combination of detailed structural modeling and full scale validation is expected to
yield the best results.
Summary
The evaluation of the hazards posed by the use of composite materials presents severe
challenges to current hazard analysis methodologies. These challenges arise both out of the
state-of-the-art of modeling as well as the highly structured nature of current hazard analysis
methodologies.
Fortunately, these issues do not preclude the development of hazard analyses. Progress
can be made without prior quantification of acceptable hazard conditions and exhaustive
representation of specific scenarios. The structure and content of the hazard analysis process
needs to be flexible and responsive to the technical limitations of current models as well as
the nature of the decision making environment. Issues of cost, benefit, and trade-offs are
often sufficiently complex so as to frustrate attempts at quantification in terms of single
variable acceptance criteria. In addition, decision makers are not often asked for such
quantifiable acceptance criteria and cannot be expected to produce such criteria at the start
of a project.
Useful progress has been made in evaluating the hazards of composites. Preliminary
evaluation tools have been developed, and insights into exposure fires and residual strength
have also been developed.
It is clear that no single approach to hazard analysis can be expected to perform well for
all problems. Variations in the technical nature of problems will require different approaches
and variations and the decision making environment will dictate further modifications to
the hazard analysis process. Both these factors require that we remain flexible and creative
in our approach to hazard analysis.
References
(1] Bukowski, R. B., Peacock, R. D., Jones, W. W., and Forney, C. L., "Technical Reference Guide
for the HAZARD I Fire Hazard Assessment," in NIST Handbook 146. Vol. II. National Institute
of Standards and Technology, Gaithersburg, MD, 1989.
[2] Smith, E. E., "Heat Release of Building Materials," Ignition, Heat Release. and Noncombustibility
of Materials, ASTM STP 502, American Society for Testing and Materials, Philadelphia, 1972,
pp. 119-134.
[3] Babrauskas, V., "Comparative Rates of Heat Release from Five Different Types of Test Appa-
ratuses," Journal of Fire Sciences, Vol. 4, March/April 1986, pp. 148-159.
[4) Harkleroad, M., Quintiere, J., and Walton, W., "Radiative Ignition and Opposed Flow Flame
Spread Measurements on Materials," DOT/FAA/TC-83/28. U.S. Department of Transportation,
Federal Aviation Administration, Technical Center, Atlantic City Airport, NJ, Aug. 1983.
[5] Smith, E. E., "Release Rate Tests and Their Applications." Journal of Fire and Flammability,
Vol. 8, July 1977, p. 309.
[6] Smith, E. E., "Transit Vehicle Material Specification Using Release Rate Tests for Flammability
and Smoke," Report IH-5-76-1, American Public Transit Association, Washington, DC. Oct.
1976.
[7] Clarke, F. and DiNenno, P. J., "Computer Based Analysis of the Fire Hazard of Furniture
Materials," in Proceedings of SPI 28th Annual Technical Conference, San Antonio. TX. 1984.
[8] Clarke, F. and DiNenno, P. J., "Fire Safety of Wire and Cable Materials, Part III," in Proceedings
of International Wire and Cable Symposium. Reno. NV, Nov. 1984.
[9] DiNenno, P. J., "Mathematical Fire Modeling-Toward the Rational Integration of Fire Safety
Features," in BVD/SPI Conference on Fire Protection Concepts, Zurich, Switzerland, March 1984.
[10] Peacock, R. D. and Bukowski, R. W., "A Prototype Methodology for Fire Hazard Analysis,"
Fire Technology, Vol. 26, No.1, Feb. 1990, p. 15.
[11] "Managers Guide to Submarine Composite Material System Fire and Toxicity Testing and Qual-
ification Procedure," Non Metalic Materials and Packaging Branch. Naval Sea Systems Command,
Washington, DC, Sept. 1990.
[12) Emmons, H. W., "Fire Safety Science- The Promise of a Better Future," Fire Technology. Vol.
26, No.1, P 5.
[13] Jones, W., "A Review of Compartment Fire Models," NBSIR 83-2684, National Institute of
Standards and Technology (NIST), Washington, DC, 1983.
[14] Beyler, C. L. and Barnett, J., "Hazard I-A Short Course," Course notes, Worcester Polytechnic
Institute, Worcester, MA, 1990.
[15] Mudan, K. and Croce, P., "Fire Hazard Calculations for Large Open Hydrocarbon Fires," in
SFPE Handbook of Fire Protection Engineering, P. J. DiNenno et aI., Eds., National Fire Pro-
tection Association (NFPA), Quincy, MA, 1989.
[16) Shokri, M. and Beyler, C. L., "Radiation from Large Pool Fires," Journal of Fire Protection
Engineering, Vol. 1, No.4, 1989, pp. 141-149.
[17] Back, G. and Beyler, C. L., "Direct Flame Impingement Heat Flux Distribution from Exposure
Fires," Hughes Associates, Wheaton, MD, 1990.
(18] Cleary, T. G. and Quintiere, J. Q" "A Framework for Utilizing Fire Property Tests," Third
International Symposium on Fire Safety Sciencc, Hemisphere Publishing Company, Washington.
DC. 1991, pp. 647-656.
[19] Mitler, H. E .• "Predicting the Spread Rates of Fires on Vertical Surfaces," 23rd Symposium
(International) on Comhustion. Combustion Institute. Pittsburgh. 1991, pp. 171-172.
[20] Mowrer, F. and Williamson, B., "Flame Spread Evaluation for Thin Interior Finish Materials,"
Third International Symposium on Fire Safety Science, Hemisphere Publishing Company, Wash-
ington, DC. 1991, pp. 689-698.
[21] Delichatsias, M., Saito. K., "Upward Flame Spread: Key Flammability Properties, Similarity
Solutions, and Flammability Indices," Third International Symposium on Fire Safety Science, Hem-
isphere Publishing Company. Washington. DC. 1991. pp. 217-226.
[22] Saito, K., Quintiere, J., and Williams, F., "Upward Turbulent Flame Spread," First International
Symposium on Fire Safety Science. Hemisphere Publishing Co" Washington. DC. 1986. pp. 71-
86.
[23] Beyler, C. L., "Major Species Production by Diffusion Flames in a Two Layer Environment,"
Fire Safety Journal, Vol. 10, No. 47, 1986, pp. 47-56.
[24] Beyler, C. L., "Major Species Production by Solid Fuels in a Two Layer Compartment Fire
Environment," First International Symposium on Fire Safety Science, Hcmisphere Publishing Com-
pany, Washington. DC, 1986, p. 431-440.
[25] Zukoski, E., Morehart, J., Kubota, T., and Toner, S., "Combustion Processes in Two-Layered
Configurations," in Second International Symposium on Fire Safety Science, 1988.
[26J Mitler, H., "Zone Modeling of Forced Ventilation Fires," Combustion Science and TechnolofO'·
Vol. 39, 1984. pp. 83-106.
[27] Dinenno, P. J., et al.. "Large Scale Compositc Module Fire Testing," Report 1087-90-1. Hughes
Associates, Inc., Wheaton, MD. Sept. 1990.
[28] Leonard, J. T" Fulper, C. R .. Darwin, R. L., Back, G. G" Scheffey. J. L., Willard. R. L.,
DiNenno, P. J., Steel, J. S., Ouellette, R. J .. and Beyler, C. L., "Post-Flashover Fires in Simulated
Shipboard Compartments: PHASE I-Small Scale Studies," NRL Memorandum Report 6886.
Naval Research Laboratory, Washington. DC. 3 Sept. 1991.
[291 Atreya. A., "Fire Growth on Horizontal Surfaces of Wood," Combustion Science and Technology.
Vol. 39. 1984. pp. 163-194.
[30] Kim, Y. and Lee. S., "A Solid Element Formulation for Large Deflection Analysis of Composite
Shell Structures," Computers and Structures, Vol. 30, 1988, pp. 269-274.
REFERENCE: Sumathipala, U. K., Hadjisophocleous, G., Aydemir, N. U., Yu, C. M., Sousa,
A. C. M., Steward, F. R., and Venart, J. E. S., "Fire Engulfment of Pressure-Liquefied Gas
Tanks: Experiments and Modeling," Fire Hazard and Fire Risk Assessment, ASTM STP 1150,
Marcelo M. Hirschler, Ed., American Society for Testing and Materials, Philadelphia, 1992,
pp. 100-115.
ABSTRACT: The in-vessel transient thermohydraulics, under emergency relief, of tanks of

heated pressure liquefied gases, dctermines whether vessels will safely relieve their contents,
rupture and cause a pool or flashing jet fire, or catastrophically fail and boiling liquid expanding
vapor explosions (BLEVE) resulting in a significant fire ball and rocketing debris.
This paper reviews the basic thermohydraulic phenomena and their interaction. The mech-
anisms are illustrated with reference to direct visual observation in small vessels and the
interpretation of additional data obtained in intermediate scale experiments.
The paper additionally describes a two-dimensional zone and a computational fluid dynamics
(CFD) field technique to describe the tank and fluid behavior under fire attack conditions.
KEY WORDS: boiling liquid expanding vapor explosions (BLEVE), computer simulation,
experiments, fire engulfment, pressure liquefied gases, safety, thermohydraulics, LPG
Over the last two decades, there has been a considerable increase in the use of pressure
liquefied fuels (for example, butane and propane). Commensurate with this, there has been
an increase in the quantities transported and stored. It is not unusual for transport cylinders
to be 30 to 600 ton vessels. Individual storage containers range from 100 ton bullets to 1200
ton spheres ..
A number of recent accidents involving storage tanks and transport containers has resulted
in many deaths and extensive property damage, the 1984 San Juan Ixhautepic incident being
perhaps the greatest fire disaster involving liquified petroleum gas (LPG) [1]. This accident
resulted in more than 500 deaths and 7 000 severely injured persons; the facility was also
almost totally destroyed.
The Fire Science Centre of the University of New Brunswick, Canada has been using
data obtained from moderate [2,3] and small scale [4] test facilities to improve our under-
standing of the complex thermohydraulics of tanks of pressure liquefied gases exposed to
'Research associate, professor, professor, and professor, respectively, University of New Brunswick,
Fire Science Centre, Fredericton, New Brunswick, Canada E3B 5A3.
'Research officer, National Research Council of Canada, National Fire Laboratory, Ottawa, Ontario,
Canada KIA OR6.
'Research engineer, Atomic Energy of Canada, Ltd., Whiteshell Nuclear Research Establishment,
Pinawa, Manitoba, Canada ROE ILO.
·Professor, University of Science and Technology, Department of Thermal Engineering, Beijing
100083, People's Republic of China.
100
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 101
accidental fire engulfment. This work has resulted in the development of computer models
to accurately simulate the phenomena [5]. Analysis of the data has additionally provided
information on the influence of liquid and vapor entrainment on lading thermohydraulics
and pressure relief [6]. This work has primarily been undertaken with a view to understand
how to best prevent boiling liquid expanding vapor explosions (BLEVE).
This paper will briefly describe the transient thermohydraulics of accidentally-heated
pressure liquefied gas tanks based upon our understanding and interpretation of the existing
experimental data base. Additionally, two developed computer codes, PLGS-l and PLGS-
20, used to simulate experimental field studies and accident conditions will be outlined,
and their results compared to the experimental data.
Experimental Studies
In order to develop a basic thermohydraulic understanding, small scale studies have been
carried out employing an externally electrically heated 40 L Refrigerant 11 filled vessel
equipped with a pressure relief device. A description of these experiments, the facility, and
some data analysis has been made previously [4,7]. Similarly, but on a much larger scale,
measurements have been conducted by the United Kingdom Health and Safety Executive
(HSE) [3].
Small Scale Experiments

The understanding of the thermohydraulic phenomenon presented here has been ob-
tained primarily from the visual and experimental data obtained with the small scale facility
(Fig. 1) [4].
The tank is constructed of a 6-mm thick stainless steel inner cylinder, flanged on either
end. The tank is 380 mm in internal diameter and 330-mm long and can contain 40 L Q(
fluid simulating the pressure liquefied gas (Refrigerant 11).
External accidental fire exposure is simulated by radiant electric heaters strapped around
the test cylinder which is placed inside another exterior pressure cylinder. During experi-
ments the annulus between these two cylindrical vessels is pressurized above the test tank
pressure to prevent possible pressure damage to the inner cylinder. The heaters are arranged
so that it is possible to vary the intensity and distribution.
The double-walled pressure vessel is fitted with 25-mm-thick sheet acrylic windows for
visual observation, and a pressure relief device (PRD). This PRD may be oriented at any
angle to simulate a variety of valve locations under accident conditions. The pneumatically-
operated computer-controlled valve simulates the behavior of a spring-loaded relief valve.
Measurements, automatically recorded by a computer-controlled data acquisition system,
include vessel wall temperatures, tank mass, thrust exerted by mass exiting through the
PRD, a variety of vapor and liquid space temperatures, liquid void fraction, and the pressure
in the test vessel as well as the annulus.
Time synchronized video/audio recordings are taken during each experiment. Void frac-
tion measurements can be recorded using a high speed (2500 rps) voltmeter with capacitance
and radio frequency type probes.
HSE Moderate Scale Experiments

The Health and Safety Executive in association with British Gas and Shell Research in
the United Kingdom have conducted trials to experimentally assess the fire engulfment
behavior of a five ton LPG tank. These tests were conducted at the Research and Laboratory
Services Division (RLSD), Explosion and Fire Laboratory, at Buxton, England in 1985
to 86 [3].
Five tests were conducted using an extensively-instrumented tank filled with commercial
propane from 22 to 72% (by volume fill). The 10 000 L tank was subjected to kerosene
pool fires for periods ranging from 9 to 31 min.
Experimental data were collected at a frequency of 1 Hz from 128 analog instrumentation
channels. Measurements included lading, wall and fire temperatures, liquid and vapor pres-
sures, water tube calorimeters, wind speed and direction, thermal radiometers, tank mass,
and photography.
Thermohydraulic Behavior
The thermo hydraulic behavior based upon the 40 L experimental and visual evidence
[4,7,8] can be described as follows.
From initiation of uniform heating, the heat transfer to the liquid and vapor is initially
by free convection. In the vapor space, this convection leads to significant vertical temper-
ature gradients due to low fluid motion in the bulk region.
The wall heat flux produces a localized recirculation from the center of the tank bottom
and a thermal boundary layer over the remainder of the liquid-wetted wall. The boundary
layer on either side of the tank turns horizontally at the liquid/vapor interface, thus forming
a thermally-stratified layer at the liquid surface. The surface temperature of the liquid
determines the vapor pressure and thus the pressure within the tank. The temperature of
this layer increases with time and consequently also the pressure.
When the container pressure reaches the relief valve opening set point, the valve opens.
Material is discharged and the pressure is now influenced by both heat transfer and material
ejection. With a vertically-upright valve, the initial discharge(s) is(are) superheated vapor.
Depending on the level of liquid/vapor interface, subsequent valve discharges may be single-
phase vapor or a mixture of liquid and two-phase fluid. Large fills and valve sizes cause
significant swelling of the liquid by void formation due to internal flashing, resulting in
liquid entrainment of the valve discharge.
If the valve position is below the liquid/vapor interface, the valve discharge will initially
be single-phase liquid. Such a discharge causes only a minor disturbance in the vapor space,
leading to a rapid increase in its temperature. Though mass afflux is higher for single-phase
liquid discharge than for single-phase vapor discharge, the rate of removal of thermal energy
(enthalpy) from the tank contents and its volumetric flow rate is significantly less. Continued
liquid venting will rapidly decrease the interface level close to the valve entry level.
Vapor pull-through occurs at an appropriate interface level above the valve opening. This
height is dependent upon mass afflux velocity and the densities of liquid and vapor. With
further decrease in interface level, the mass flow changes from vapor pull-through to liquid
entrainment. Two-phase swelling may now occur and the tank contents depressurize.
As material continues to exit, the two-phase level decreases until single-phase vapor flow
results. With decreasing fluid levels, exposed walls rapidly increase in temperature due to
the poor heat transfer from the wall to the vapor. The valve may finally reseat when sufficient
material has been discharged.
In summary, the interactive transient thermodynamic and fluid dynamic processes oc-
curring between the fire, tank contents, and the relieving process trigger, successively, liquid
and vapor stratification, pressurization, pressure relief, frothing and liquid level swell, plus
perhaps entrainment and/or vapor pull-through and thus two-phase (critical flow) discharge,
and can result in increasing pressure response followed by pressure stabilization and decrease.
Since vapor space heat transfer is poor, even during relief, wall temperatures and vapor
superheat increase substantially, providing conditions suitable for creep rupture of the vessel.
Vessel failure at this time can cause massive two-phase releases, flares, fire balls, and perhaps
explosions.
Liquid Entrainment/Vapor Pull-Through

The ability of a PRD to maintain a safe pressure level in the tank is influenced by its size
and the possibility of liquid entrainment and vapor pull-through. In the former, liquid is
picked up off the liquid or two-phase surface in the vicinity of the valve. In vapor pull-
through, a vapor exit stream is pulled down into the valve through the liquid by a vortex
formed upon discharge. In both cases, the pressure relieving capacity of the valve changes
and the ability of the contents to absorb energy is altered.
A common feature with some of the Health and Safety Executive (HSE) and University
of New Brunswick (UNB) data sets is that at high fills and/or large valve openings, liquid
entrainment into the valve afflux appears to be significant. This phenomena exerts a con-
siderable influence on the thermofluid behavior of the contents and its pressure relief as
analysis has shown [6-8] and as indicated here (Table 1).
Experimental data from HSE Test 5 (72% volume fill) is given in Figs. 2 through 4. The
variation of tank pressure with time for this HSE Test 5 is given in Fig. 2. The spring-loaded
PRD cycles once before going into a vent which continues until the tank is almost empty
and the valve reseats. During the continuous venting period after the second PRD opening,
the pressure trace reaches a maximum at 700 s.
Figure 3 shows the time variation of tank mass and mass flowrate through the PRD. The
tank mass data is noisy and needs to be filtered. First the tank mass data was fitted to a
polynomial over the duration of continuous venting. The best least squares fit as judged by
TABLE 1-HSE field test parameters and initial conditions;

HSE test 5 [3].
Tank diameter, m 1.68
Tank length, m 4.06
Tank capacity, L 10 090
Wall thickness, mm 11.9
PRO set pressure, barg 14.3
Nominal PRO diameter; mm 33.6
PRO position, degrees 0"
PRO orifice area, mm2 887
PRO discharge coefficient 0.85
Initial temperature, DC 5.7
Propane fill, volume % 75
Power input, kW/m2 77b
Fire temperature, DC, minimax 600 to 1000
Mean fire temperature, DC 754
·Vertically upright.
bAs determined by water calorimeters.
the standard deviation, and the mean error was chosen. This equation was then differentiated
with respect to time to obtain the experimental mass flowrate of the lading discharge.
Assuming the vapor properties to be given by the temperature from the uppermost vapor
space thermocouple (#9 which is 5 mm below the top wall) and measured tank pressure, a
single-phase vapor mass flowrate through the PRD was then computed. This computation
(curve A) assumes the PRD to be fully open from 349 to 1800 s. It can be seen that curve
A and the experimental mass flow rate curve intersect at 465 and 1040 s. The theoretical
single-phase vapor flowrate curve (curve A) from 349 to 465 s and from 1040 to 1800 s,
show larger values than the experiment, indicating that the PRD was not fully open before
465 s and after 1040 s.
A simple linear valve model was used to account for the reduction in flow area due to
partial PRD closing. At 1799 s, just before PRD closure, the experimental mass flow rate
is 57.5% of the computed single-phase vapor flow rate for a fully open valve. It was,
therefore, assumed that the available flow area due to partial PRD closure is also 57.5%
of a fully open valve, at 1799 s. It was also assumed that the PRD was fully open between
465 and 1040 s. Between 1040 and 1799 s, the percentage of PRD flow area is taken to be
a linear function of tank pressure (Fig. 2).
In other words, the spring loaded valve (PRD) was assumed to behave in such a way that
the available flow area at the valve was directly proportional to the tank pressure. This is
applicable when the tank pressure was lower than the minimum pressure for a fully open
valve, as is the case between 349 to 465 s, and also between 1040 to 1800 s. The valve flow
area was assumed to be 100% at 1040 sand 57.5% at 1799 s.
An evaluated single-phase vapor flowrate, corrected for partial PRD closure, is shown in
the same figure as curve B. This curve conforms to the data not only between 1040 and
1799 s but also from 363 to 465 s, thus providing support to the assumed valve area behavior.
It also provides a basis for the conclusion that entrainment of interface material (liquid or
two-phase) occurs between 465 and 1040 s.
Estimates of the liquid/vapor interface level are given in Fig. 4. Two-phase swelling would
not be expected from the start of the test up to the first PRD opening because the tank is
pressurizing and only subcooled boiling occurs. The change in interface level is, therefore,
only due to thermal expansion of the liquid lading up to this point with some slight void
caused by heterogeneous boiling at the walls.
The unswelled height is theoretically computed using the instantaneous tank fluid mass
and liquid temperature data. The swelled height is obtained directly from internal experi-
mental temperature data. A sudden temperature rise in a previously wetted thermocouple
is taken as an indication of thermoelement uncovering. The height of the thermocouple at
the time when such a discontinuity occurs is taken as a swelled interface level at that time.
The swelled and unswelled levels allow for the calculation of an average void fraction within
the liquid lading, as also given in the figure.
Experimental Discussion
The complex interactive transient thermohydraulic behavior of fire engulfed pressure
liquefied gas tanks has been described with reference to small and moderate scale experi-
ments which indicate that the important features of the thermodynamic and fluid dynamic
behavior are influenced by the following:
(1) Thermal stratification in the liquid determines pressure response and initial valve
behavior.
(2) Dependent upon fill, heating, and valve characteristics, liquid entrainment can sig-
nificantly affect subsequent behavior of the contents and vessel.
(3) Vapor space heat transfer is extremely poor despite even a vertically upright PRD
discharge resulting in extreme vapor thermal stratification and wall temperatures.
PRD vapor discharge is strongly superheated.
Computer Simulation
Validated computer models of this phenomena are desirable for several reasons: (a) full-
scale testing is extremely costly and dangerous; (b) computer models permit extensive
variation of the initial and boundary conditions; and (c) the codes may be utilized in incident
simulation and evaluation, filling limit studies, pressure relief device sizing, and the design
of protection systems.
In developing such codes, the thermohydraulic behavior of the fluid as described earlier,
must first be fully understood in order that the effects of scale and heating be correctly
interpreted. A variety of simple models have been developed [9-12].
A simple four fluid zone model (PLGS-l) [12] and a two-dimensional free convection
model (PLGS-2D) [13] have been developed at UNB. Both models simulate a horizontal
cylindrical tank engulfed in either a uniform or distributed fire. The zone model fire, used
by these programs, has both radiant and convective heat transfer components. The structure
and use of these codes will now be briefly described.
PLGS-1
The two-dimensional wall conduction equation is solved numerically using the fire con-
ditions, and inside wall heat transfers to determine the wall temperature distribution.
The four fluid zones modeled by this code are shown in Fig. 5. These consist of thermally
stratified vapor and liquid zones, where interface temperature sets the pressure, a bulk liquid
and bottom-heated zone, and two free-convective boiling boundary layer zones; these bound-
ary regions are considered to have no mass because their size and extent are small.
The vapor zone receives heat by free convection from the fire-heated walls. There is
radiant exchange from the vapor region to the liquid interface at the top of the stratified
region. The stratified region is heated by wall heat transfer and energy transport provided
by wall boundary layers which are initiated at the angle <l> below which there is an unstable
bottom heated region which heats the bulk fluid [14,15]. Heating is also provided to the
bulk by conduction from the upper stratified layer and condensation at the edges of the
boiling boundary layers.
The characteristics of the flows and the heat transfer depend very much on the temperature
differences between the tank walls and the adjacent liquid. The two major modes of heat
transfer in the liquid space are natural convection and nucleate boiling. Film boiling may
occur due to sudden wetting of dry tank walls during venting and depressurization or by
vapor blanketing. Film boiling is, however, unlikely to occur at a location which is already
wetted unless the critical heat flux is exceeded (>400 kW/m2, for propane [16]) or the
amount of vapor generated and concentrated on downward facing walls causes vapor
blanketing.
The program considers criteria for the onset and termination of natural convection, sub-
cooled and saturated pool boiling, nucleate to film boiling, and free convective and boiling
boundary layer flows. When the pressure inside the vessel reaches the valve set pressure,
the valve opens and discharge commences with the valve discharge behavior linearly related
to pressure, similar to the behavior for spring-loaded relief valves.
The discharge, which may be either vapor, liquid, or two-phase material, is dependent
upon fire severity and distribution, valve geometry and position, as well as lading level and
thermodynamic state.
The program utilizes the wall heat fluxes to determine the state of the system by solution
of the fundamental energy and mass conservation equations for each zone and proceeds in
a time-marching solution. Details of the original program are given in Ref 3, and modifi-
cations and improvements are currently being implemented.
PLGS-2D
The two-dimensional wall conduction equation is solved numerically to determine the
wall temperature distribution. The boundary conditions used for the solution ofthis equation
are the fire heat flux and the heat transfer to the fluid from the wall at the outside surface
and the inside surface, respectively. The fire model yields the circumferential heat flux on
the tank surface due to an engulfing or partial fire. Heat is transferred from the flame by
radiation and convection using a technique similar to that of Ref 11. In this model, the
combustion gases are assumed to be rectangular and have a uniform temperature (chosen
to be representative of the appropriate experiment) with a constant flame absorption coef-
ficient. The walls of the fire rectangle are taken as a single surface zone and the hot
combustion gases as a single volume zone. The engulfed cylindrical tank surface is subdivided
into a number of surface zones corresponding to those representing the discretized interior
wall segments. Surface configuration factors are then evaluated which thus permits zonal
energy balances to be completed utilizing interrelationships between the radiative and emis-
sive powers of the relevant zones. Convection from the fire gases represents between 5 and
15% of the heat transfer to the tank. Details of both treatments are given in Ref 13. The
heat transfers from the vapor-wetted wall are a combination of conduction/convection to
the vapor and radiation to the vapor and liquid interface. The radiation heat transfer is
determined assuming that the radiative and thermal properties of the gas are uniform throughout
the vapor region.
Unlike PLGS-l, which is a zone model, the flow and temperature fields of PLGS-2D are
obtained from the solution of the discretized partial differential equations governing free
convection, continuity, and energy. These equations are simplified using the Boussinesq
approximation, and normalized using the liquid properties and tank diameter as reference
parameters. The equations are then transferred to a boundary fitted coordinate system. An
eddy viscosity model is utilized for the turbulent viscosity. The governing equations are
discretized using the control volume approach and solved following the SIMPLEC procedure
[17]. In its present form, PLGS-2D is capable of pressure and temperature predictions during
the transient period from initial fire engulfment up to the first PRD opening. Details of the
solution method and verification tests have been given in Ref 13.
Simulations Results
Computer simulations of a variety of field tests have been conducted and compared to
PLGS-l. Comparisons of PLSG-2D also show good agreement to the HSE data referred to
earlier [13].
A comparison to the 75%-fill experiment of both codes will be made here. Table 1 states
die field test and determined parameters for the HSE Test 5. During fire engulfment only
one PRD operated. The tabulated value is the effective area of one fully-open valve.
PLGS-l
Figures 6 and 7 illustrate the comparison of the results. In Fig. 6, the predicted pressure
response is compared with that measured in the field. The pressure response is closely
matched by the simulation. The number of valve cycle operations prior to the valve remaining
open also coincide.
Figure 7 shows the lading mass predicted and obtained. There is good agreement between
the simulation and the experimental data. Vapor temperatures (Fig. 8) differ significantly,
perhaps due to the influence of liquid droplet carryover caused by surface boiling which is
not accounted for in the model. Vapor wall temperature measurements and simulations are
shown in Fig. 9. The measured vapor wall temperature is lower than predicted. This is likely
as a result of the flames during the trial being blown away from this region by wind and not
considered in the simulation.
PLGS-2D
The developed model has been used to determine the effect of the filling level on the
response of the tank to fire conditions in addition to that considered here [13]. The simu-
lations here have been performed for the 72%-filled HSE tank with a variety of heating
distributions in addition to that for full engulfment, as shown in Fig. 10.
As mentioned earlier, the simulations can only be taken up to the time of initial valve
opening due to program restrictions. The tank pressure predicted by the model is in good
agreement with that recorded during the experiment. The numerical model simulation with
full fire engulfment predicts PRD to open at 305 s; during the field trial, the PRD opened
at 312 s. Comparison of the liquid, vapor, and vapor wall temperatures is satisfactory,
indicating satisfactory modeling of the basic phenomena.
Effect of Fire Exposure on Tank Behavior-The degree of fire exposure is one parameter
which is difficult to conveniently vary experimentally. Fire exposure is, however, anticipated
to exert a major influence on the free convective flows and heat transfer in the tank, and
hence on tank response. In order to investigate this effect, simulations were carried out for
fire exposures of: (a) full engulfment, discussed earlier; (b) 180 top engulfment; (c) 180
0 0
side engulfment; and (d) 90 top side engulfment [18].

0
Figure 10 illustrates these conditions. The pressure responses obtained are shown in Fig.
11. It is interesting to see that the rate of pressure rise with full engulfment is very close to
that with only the top half of the tank wall exposed to the fire. Despite the fact that the
liquid receives much more heat and the mean liquid temperature is much higher when the
tank is fully engulfed in flames than with the top 50% engulfment, the time needed for the
valve to open is only 13 siess (311 s versus 324 s).
The rate of pressure rise for the other two simulations, namely 50% side engulfment and
25% top quarter engulfment, is similar to in response, but slower than the full and 50%
top engulfment. The time for the valve to open is slightly higher, requiring 434 and 417 s,
respectively. This can be explained by the fact that there is no mixing of the stratified and
bulk liquid layers for the simulations with 50% and 25% top fire engulfments, which allow
the temperatures of the stratified layers, and therefore the interfaces, to rise more quickly.
Figure 12 shows the vertical temperature profiles for all cases and indicates that the bulk
liquid temperature remains at its initial value, for the cases of the fire being applied to the
top of the tank.
Maximum Tank Wall Temperature-The maximum tank wall temperature history for the
four cases considered is shown in Fig. 13. The figure illustrates that the development of the
maximum wall temperature for the fully engulfed simulation is practically the same as the
one for the 50% top engulfment, and that of the 50% side engulfment is the same as the
one with 25% engulfment. These results indicate that the most important factor influencing
the maximum tank wall temperature is the local thermal boundary condition on the vapor
wetted walls.
This clearly demonstrates that a fire covering only the top half of the tank wall is as
severe, in terms of causing tank pressure rise, as a fire completely engulfing the tank.
Similarly, a fire exposing half the tank wall causes the same pressure rise as one impinging
on the top quarter wall only.
Conclusions
The experiments and computer codes utilized by the UNB Fire Science Centre to study
and describe the transient response of pressure liquefied gas vessels to fire engulfment have
been outlined. The pressure liquefied gas zone model code, PLGS-l, and a boundary fitted
coordinate CFD code, PLGS-2D, have been used to successfully simulate the behavior of
several small and moderate scale experiments in which cylinders of LPG were subjected to
engulfing pool fires. The codes may be useful for safety analysis, incident simulation, filling
limit studies, and the design of protection systems.
The important experimental features observed are:
(1) Thermal stratification in the liquid determines pressure response and initial valve
behavior.
(2) Dependent upon fill, heating, and PRD valve characteristics, liquid entrainment can
significantly affect subsequent behavior of the contents and vessel.
(3) Vapor space heat transfer is extremely poor despite PRD discharge resulting in ex-
treme vapor thermal stratification and wall temperatures. Vapor discharge is extremely
superheated.
The comparisons indicate that the developed computer models (PLGS-1 and PLGS-2D)
may be used to accurately predict the complex thermohydraulic events in an externally
heated pressure liquefied gas vessel for a wide range of scales.
It may also be concluded from the simulations on partial engulfment that:
(a) tank pressure and fluid behavior is significantly affected by the degree and extent of
fire exposure;
(b) the distinct hot stratified liquid layer which is responsible for setting vessel pressure
is significantly influenced by the vapor space heat transfer; and
(c) the maximum tank wall temperatures, which determine tank integrity, are primarily
influenced by the degree of fire engulfment of the vapor wall.
Acknowledgments
Many individuals have contributed to the understanding and progress of this work. Drs.
S. Hinata, as well as K. F. Sollows, C. A. McDevitt, and other graduate students in Chemical
and Mechanical Engineering have assisted in the design, development, operation, and inter-
pretation of many aspects of the experimental work. The cooperation of the Health and
Safety Executive, Research and Laboratory Services Division, Buxton, United Kingdom.
and many comprehensive discussions with their researchers, in particular, Mr. K. Moodie
and Dr. S. Jagger, is greatly appreciated. Financial support for the University of New
Brunswick work has been provided by the Natural Sciences and Engineering Research
Council of Canada's Strategic and Operating grants program as well as the Transportation
Development Centre (Transport Canada).
References
[1] Pietersen, C. M., "Analysis of the LPG-Disaster in Mexico City." Journal of Hazardous Materials.
Vol. 20, 1988, pp. 85-107.
[2] Droste. B. and Schoen. W .• "Full Scale Fire Tests With Unprotected and Thermal Insulated LPG
Storage Tanks," Journal of Hazardous Materials. Vol. 20.1988, pp. 41-53.
[3] Moodie, K., Cowley, L. T., Denny, R. B., Small, L. M., and Williams, I.. "Fire Engulfment
Tests on a 5 Tonne LPG Tank," Journal of Hazardous Materials, Vol. 20, 1988, pp. 55-71.
[4] Venart, J. E. S., Sumathipala, U. K., Steward. F. R., and Sousa, A. C. M., "Experiments on
the Thermohydraulic Response of Pressure Liquefied Gases in Externally Heated Tanks with
Pressure Relief," Plant/Operations Progress, Vol. 7.1988. pp. 139-144.
[5] Aydemir. N. U., Hadjisophocleous, G. V., Chang-ming, Y, Sousa, A. C. M.• and Venart, J. E.
S., "Pressure Relief of Accidentally-Heated Pressure Liquefied Gas Vessels: Two-Dimensional
Models," Presented at Eurotherm 14, Heat Transfer and Major Technological Hazards, Universite
Catholique de Louvain, Belgium, 1990.
[6] Sumathipala, K., Venart. J. E. S., and Steward. F. R., "Two-Phase Swelling and Entrainment
During Pressure Relief Valve Discharges," Journal of Hazardous Materials, Amsterdam. Vol. 25.
1990, pp. 219-236.
[7] Sumathipala, U. K., Steward, F. R., and Venart, J. E. S., "The Experimental Thermohydraulics
of Small Break Loss of Coolant Accidents," Proceedings, International ENS/ANS Conference on
Thermal Reactor Safety (NUCSAFE '88), Avignon, France, 1988, pp. 1103-1112.
[8] Doyle, R. A., "Pressure Relief of Partially Filled Horizontal Cylinders," Master of Science in
Engineering Thesis, University of New Brunswick, Fredericton, 1985.
[9] Ramskill, P. K., "A Description ofthe "ENGULF" Computer Codes-Codes to Model the Thermal
Response of an LPG Tank Either Fully or Partially Engulfed by Fire," Journal of Hazardous
Materials, Vol. 20,1988, pp. 177-196.
[10] Birk, A. M., "Modeling the Response of Tankers Exposed to External Fire Impingement," Journal
of Hazardous Materials, Vol. 20, 1988, pp. 197-225.
[11] Beynon, G. V., Cowley, L. T., Small, L. M., and Williams, I., "Fire Engulfment of LPG Tanks:
HEATUP, a Predictive Model," Journal of Hazardous Materials, Vol. 20, 1988, pp. 227-239.
[/2] Aydemir, N. U., Magapu, V. K., Sousa, A. C. M., and Venart, J. E. S., 'Thermal Response
Analysis of LPG Tanks Exposed to Fire," Journal of Hazardous Materials, Vol. 20, 1988, pp.
239-262.
[/3] Hadjisophocleous, G. V., "Numerical Modelling of LPG Horizontal Cylindrical Tanks Subjected
to Heating Environments," Ph.D. Dissertation, University of New Brunswick, Fredericton, Can-
ada, 1989.
[/4] Aydemir, N. U., Sousa, A. C. M., and Venart, J. E. S., "Transient Thermal Stratification in
Heated Partially Filled Horizontal Cylindrical Tanks," ASME Transactions, Paper 84-HT-60, 1984.
[15] Aydemir, N. U., Sousa, A. C. M., and Venart, J. E. S., "Transient Laminar Free Convection in
Horizontal Cylinders," Wiirme-und Stoffubertragung, Vol. 20, 1986, pp. 59-67.
[/6] Cichelli, M. T. and Bonilla, C. F., "Heat Transfer to Liquids Boiling Under Pressure," Transactions
of the AIChE, Vol. 41, 1970, pp. 755-787.
[17] Van Doormaal, J. P. and Raithby, G. D., "Enhancements of the SIMPLE Method for Predicting
Incompressible Fluid Flow," Numerical Heat Transfer, Vol. 7,1984, pp. 147-163.
[18] Hadjisophocleous, G. V., Sousa, A. C. M., and Venart, J. E. S., "Natural Convection Heat
Transfer in Partially Filled Horizontal Cylinders," International Journal of Numerical Methods in
Engineering, Vol. 30, 1990, pp. 629-646.
Richard G. Hill,1 Constantine P. Sarkos,1 and Timothy R. Marker1
Development of a Benefit Analysis for an

Onboard Aircraft Cabin Water Spray System
REFERENCE: Hill, R. G., Sarkos, C. P., and Marker, T. R., "Development of a Benefit
Analysis for an Onboard Aircraft Cabin Water Spray System," Fire Hazard and Fire Risk
Assessment, ASTM STP 1150, Marcelo M. Hirschler, Ed., American Society for Testing and
Materials, Philadelphia, 1992, pp. 116-127.
ABSTRACT: This paper outlines a program to ultimately develop design standards for an
onboard aircraft cabin water spray system. Main emphasis is on the acquisition of technical
data to be used as input into a benefit analysis program. A disbenefit study being conducted
by an airframe manufacturer is outlined. Full-scale tests of an onboard cabin water spray
system are detailed, and how the data obtained would be used in a benefit analysis program
is explained. Three different scenarios were tested in a 707 narrow body fuselage and com-
parison data with and without water spray are presented. The benefit analysis program to be
used is explained and an example is given.
KEY WORDS: fire tests, water spray, aircraft fire, benefit analysis, disbenefit study
Full-scale fire tests, conducted in the United Kingdom (U.K.) under the auspices of the
Civil Aviation Authority (CAA), have demonstrated the feasibility of an onboard cabin
water spray system for providing a marked improvement in survivability during a postcrash
fuel fire [1]. Developed and evaluated by Safety Aircraft and Vehicles Equipment, Ltd.
(SA VE), the system produces a fine water spray or mist consisting of a "range of water
droplet diameters," with a mean droplet diameter of about 100 microns. A fine water spray
system, such as developed by SAVE, is capable of providing fire protection with relatively
low flow rates of water. The spray system that has been tested is a "breadboard" design
for the purpose of demonstrating concept feasibility. Recently, the United States (U.S.)
Federal Aviation Administration (FAA) has joined with the CAA and other regulatory
authorities to further develop the cabin water spray concept for practical implementation
in commercial airplanes.
A workflow diagram of the FAA program to develop a cabin water spray system for safe
and effective installation in a commercial transport airplane is shown in Fig. 1, depicting
major projects and order of accomplishment. Controlled full-scale aircraft fire tests are being
conducted to document the additional time available for escape as provided by the current
SA VE system. Concurrently, studies are underway to address the various problems asso-
ciated with an inadvertent discharge of the water spray system while an airplane is in flight
or on the ground. The results of these initial studies will be factored into a benefit analysis
to determine the potential for lives saved (similar to analyses conducted by the FAA and
CAA for passenger protective breathing equipment) [2]. Presuming that the benefits will
outweigh the disbenefits, the next steps will be to optimize the spray system for installation
IProgram manager, ACD-240; manager, Fire Safety Branch, ACD-240; and project manager, ACD-
240, respectively, FAA Technical Center, Building 275, 201, and 275, respectively, Atlantic City In-
ternational Airport, NJ 08405.
116
in an airplane and to develop design requirements and specifications. Additional full-scale
tests would follow to verify the additional time available for escape provided by the optimized
system. Another benefits analysis would determine potential lives saved for the optimized
system. Finally, a decision would be made as to whether to proceed with a regulation
requiring an onboard spray system for the commercial airplane fleet.
The remainder of this paper will deal with the top three boxes of the flow diagram: full-
scale tests, disbenefit studies, and the benefit analysis.
Service Considerations and Disbenefit Study

Purpose
Determine the consequences of an accidental discharge of an onboard cabin water spray

system both on the ground and in flight.
Method
Contractual studies were awarded to two major airframe manufacturers. The studies
include, but are not limited to, the following:
(a) Effect of accidental water spray activation on safe operation of aircraft in flight.
(b) Effect of water spray on rapid passenger evacuation and hypothermia during cold
weather.
(c) Impact of accidental water spray on aircraft airworthiness and extensiveness of repair
work necessary to restore aircraft to service.
Status
The study by Airbus Industries was recently completed and a technical report is available.
The study by Boeing commenced in January 1991.
Expected Outcome and Utilization

Input into benefit analysis computer program and identification of problem areas that
need to be considered during optimization project.
Full-scale Effectiveness Tests

Purpose
Evaluate and determine the additional time available for escape provided by the SAVE
water spray system under controlled full-scale test conditions for several postcrash fire
scenarios.
Method
Full-scale tests wiIl use both a narrow body fuselage and a wide body fuselage for several
poster ash fire scenarios. One test will be conducted with the SAVE installed and one test
without the SAVE system for each fire scenario in order to determine the additional time
available for escape. A section of the test article will be completely furnished with materials
marginally compliant with the FAA seat cushion and low heat release fire test standards.
Status
Narrow Body Tests-Full-scale tests were conducted using a 707 fuselage. The fuselage
was outfitted with the SAVE system, consisting of 120 nozzles along the ceiling, as shown
in Fig. 2. The fuselage was also fitted with thermocouples, smoke meters, gas analyzers,
gas sampling equipment, calorimeters, and photo and video cameras as shown in Fig. 2.
A number of preliminary tests were conducted with small fuel pan fires in the fuselage,
seats doused with gasoline, or a large external fuel fire with a bare fuselage. Table 1 lists
the various tests conducted.
Three different scenarios were selected for study because of their relevance to past sur-
vivable aircraft accidents involving fire deaths. The first scenario consisted of an external
fuel fire entering the fuselage through a break under zero wind conditions. The second
scenario was also with a fuel fire adjacent to a fuselage break; however, the effect of wind
was simulated by employing a fan to cause the fuel fire flames to penetrate further into the
cabin than in the zero wind case. The third scenario studied was that of a pool fire burning
through the fuselage skin in the lower quadrant and entering the cabin.
For all tests using the water spray, the system was charged with 272 L (72 gal) of water
and had a total discharge time of approximately 3 min. A constant rate of discharge was
maintained by a continously regulated air pressure in the water storage tank. The water
discharge rate was monitored by measuring the weight loss from the tank during discharge.
Zero wind scenario-Two tests were conducted under zero wind conditions, with and
without water spray. For both tests, a fuel fire pan 2.438 by 3.048 m (8 by 10 ft) was located
outside the fuselage adjacent to a 1.067 m wide by 1.930 m high (42 in. wide by 76 in. high)
opening in the fuselage. Five rows of seats were placed inside the cabin, with the third row
centered in the opening. Each row consisted of a triple seat on the fire opening side of the
cabin aisle and a double seat on the other side. Each seat was fire blocked in compliance
with the present FAA standard. The seats were placed in a 7.33 m long (24 ft long) section
of the cabin 3.05 m (10 ft) on either side of the opening, which was outfitted with ceiling
panels, storage bins, sidewalls, and carpet. During both tests, the forward left entry door
lliaS left ajar.
In each test, 208 L (55 gal) of jet fuel was ignited in the fuel pan. For the test with water
spray, the water was discharged at the same time the fuel was ignited. However, there was
approximately a 15 s delay from the electrical activation of the system until all of the nozzles
.'ere discharging water.
The results of the zero wind tests are shown in Figs. 3 and 4. The curves in the figures
show the range in measurements at a particular fuselage station. In all cases, the highest
readings were at the highest location, whereas the readings decreased the closer the mea-
surement location was to the floor. Temperature was measured at 0.305 m (1 ft) increments
from a location 2.1 m (7 ft) high (near the ceiling) to a location 0.305 m (1 ft) off the floor.
Smoke was measured at three heights; 1.676 m (5 ft, 6 in.); 1.067 m (3 ft, 6 in.); and 0.457
m (1 ft, 6 in.). All gas measurements were at 1.676 m (5 ft, 6 in.) and 1.067 m (3 ft, 6 in.).
Note the rapid rise in temperature and toxic gas production and the decrease in oxygen
and visibility at approximately 5 min in the test without the water spray. This behavior
indicates the development of a flashover condition. However, when water spray was used,
survivable conditions prevailed for the entire 7-min test duration. The time interval of actual
water spray discharge was from 15 s until approximately 195 to 200 s into the test. Therefore,
not only did the water spray improve conditions during the time of discharge, but,there
were notable improvements for some time afterwards. A negative effect of water spray can
be seen in the smoke (light transmission) curves. Early into the test, the water spray drives
the ceiling smoke layer to the lower cabin levels.
Wind scenario- Three material tests were conducted simulating external wind conditions
that would draw an external fuel fire into the cabin through an opening. The tests were
configured identical to the zero wind scenario except that a large fan was placed adjacent
to the forward left exit in such a way as to create the desired flow through the cabin.
entered the cabin by burning through the floor and the sidewall area. The top two pane.
of Fig. 9 show the resultant temperature profiles for the two tests. The use of water spra,
vastly increased survival time.
Narrow body tests summation-In order to use the data generated by these tests in a
benefit analysis, it must be in a useful form. Basic information is required; for example,
what is the average increase in survivability by the use of water spray and for what accident
conditions. Since it was virtually impossible to try to equate various wind and door opening
conditions in real accidents to the conditions studied in these tests, the approach taken was
to employ the data to generalize the effect of water spray. Thus, the tests represent accidents
producing flashover conditions in either 15-30 s, 2.5-3 min, or 5 min due to a large fuel
fire entering an aircraft from an opening or rupture, or accidents in which flashover occurred
in approximately 2 min due to fuselage burnthrough.
To determine the average survival time from the temperature and toxic gas measurements,
a calculation was made of the Fractional Effective Dose (FED) [3]. The lower panels of
Fig. 9 show the FED calculation for the burnthrough scenario. Because of the large increase
in hazards generated by flashover in each of the fire scenarios evaluated, any of the hazards
may be employed individually in the FED calculation (for example, temperature) to ap-
proximate the improvement in survival time furnished by water spray.
Wide Body Tests-Tests are underway to evaluate the effectiveness of water spray in a
wide body (modified DC-lO) aircraft and should be completed by early 1991. The wide body
test results will also be factored into the benefit analysis in the same fashion as the narrow
body test results.
Benefit Analysis Study
Purpose
Calculate the potential lives saved from the mandatory requirement for an onboard water
spray system (SAVE design) based on an analysis of worldwide fire accidents in transport
aircraft.
Method
The benefit analysis computer program developed by the FAA Technical Center, using
the full-scale effectiveness tests data, will be employed to determine potential savings in
lives.
A detailed description of the model and its use is contained in reference 2. The model
consists of two profiles (Fig. 10); a thermal and toxic hazard profile and a mobility rate
profile. The profiles are defined as follows:
Thermal and Toxic Hazard Profile-P(t) versus time after aircraft stopped. As can be
seen from Fig. 10, if P(t) = 1, no toxic gas or thermal hazards exist. IfP(t) = 0, the hazards
result in the incapacitation of all passengers. If all survivors were to successfully evacuate
with time to spare, P(t) would exceed 0 at the time the last passenger exited the aircraft.
Hazards that cause incapacitation accounted for in the model include elevated temperatures,
various toxic gases (for example, CO, HCN, HF, and so forth), and oxygen depletion.
Mobility Rate Profile-ERM(t) versus time after aircraft stopped. ERM(t) describes the
loss in passenger mobility due to physical effects such as poor visibility caused by smoke
and inadequate lighting, passengers carrying hand luggage, blockage of aisles by debris,
other passengers, and fire. The number of exits, type of exits, and evacuation slide condition
influence this profile. The highest mobility rate for an accident (with no improvements) is
assigned the value 1.0. When ERM(t) = 0, there is no movement of occupants out of the
aircraft.
The number of survivors is the area under the curve of the product of the two profiles
(Fig. 10).
To calculate the possible benefit of a safety improvement, past accidents are used to
develop the two profiles (from information provided by the accident investigation). New
profiles must also be developed, preferably from full-scale test data, that reflect the reduction
in hazards (P(t)) and/or increase in mobility (ERM(t)) provided by the safety improvements.
This approach was successfully employed for 20 accidents in a study on passenger protective
breathing equipment [2]. An example from this study is shown in Fig. 11. For that particular
accident, there were only 30 survivors of the 81 passengers (the 51 fatalities were all fire
related; if there were trauma deaths, they would be subtracted out). The two curves were
developed from the accident data. The top curve represents the hazards from the fire to
passenger survivability (thermal and toxic), and the bottom curve is the mobility curve which
includes reduced visibility that may effect the ability of a passenger to egress. Note that for
the first 20 s the hazard curve showed a safe atmosphere; however, no occupants exited the
aircraft because the doors were still closed. The evacuation rate for the product of the hazard
and mobility rates equal to one was calculated (A = 0.682). Therefore, if we multiply the
hazard by the mobility rate by 0.682 for each s of the accident, the result for the original
accident profiles will be 30 survivors. The next step is to modify the profiles based on
expected hazard or mobility changes from the implementation of a safety improvement.
The profiles in Fig. 11 represent the inclusion of seat fire blocking layers (FB), floor proximity
lighting (FPL), and passenger protective breathing equipment (PBE) (both with and without
a time delay due to donning).
The effects of water spray will be factored into the analysis, which will necessitate the
development of new hazard and mobility profiles. These profiles will be based on known
information about the accident and the full-scale test data currently being gathered.
Status
The CAA is presently collecting data to update the existing data base and developing
hazard and mobility profiles for as many accidents as possible. Information not needed in
past analyses, such as fuselage breakup, must now be obtained and factored in since it may
indicate whether the water spray system would have operated properly.
Expected Outcome and Utilization

A benefit/cost ratio for the mandatory requirement of an onboard cabin water spray
system will be calculated and this will provide a basis for a decision as to whether or not to
proceed to optimization and development of design requirements/specifications.
Conclusion
This work is still in progress and conclusions will be forthcoming when the work is
completed and data analyzed.
References
[II Whitfield, R. T., Whitfield, Q. A., and Steel, J., "Aircraft Cabin Fire Suppression by Means of
an Interior Water Spray System," CAA Paper 88014, Civil Aviation Authority, Gatwick, United
Kingdom, July 1988.
121 Hill, R. G., and Speitel, L., "Study of Benefits of Passenger Protective Breathing Equipment from
Analysis of Past Accidents," Report DOT/FAA/CT-88/03, Federal Aviation Administration Tech-
nical Center, Atlantic City, NJ, March 1988.
PI Hill, R. G., Brown, L. J., Speitel, L., Johnson, G. R., and Sarkos, C. P., "Aircraft Seat Fire
Blocking Layers: Effectiveness and Benefits Under Various Scenarios," Report DOT/FAA/CT-83/
43, Federal Aviation Administration, Atlantic City, NJ, Feb. 1984.
Fire Risk Assessment
John R. Hall, Jr.'
How to Tell Whether What You Have is a

Fire Risk Analysis Model
REFERENCE: Hall, J. R., Jr., "How to Tell Whether What You Have is a Fire Risk Analysis
Model," Fire Hazard and Fire Risk Assessment, ASTM STP 1150, Marcelo M. Hirschler, Ed.,
American Society for Testing and Materials, Philadelphia. 1992, pp. 131-135.
ABSTRACT: The growing interest in fire risk analysis in the past decade has not eliminated
a substantial degree of confusion and disagreement over the meaning of key terms like fire
risk and fire hazard and over the bases for choosing an analytical approach that is right for
each problem. This paper describes the essential elements of fire risk analysis models, some
common misconceptions about fire risk analysis, and some key principles in deciding when
and how to use it.
KEY WORDS: fire risk, fire risk analysis, fire hazard analysis, probability, scenario
There are a great many risk-related terms in circulation in the world of fire safety these
days: fire risk analysis, hazard analysis, fire risk management, to name a few, and each is
used to mean different things by different people. What I propose to do here is to sketch
out the essential elements of any type of analysis that may be validly put forth as fire risk
analysis. This will lead to a discussion of some methods that are not risk analysis at all.
The need for this kind of overview stems from the past decade's proliferation of models
and analysis methods that have been implicitly or explicitly proposed as risk analysis tools
for fire protection problems. There remains a frustratingly large degree of disagreement and
confusion over basic purposes and terminology in these models, and unlike other areas of
fire science, these disagreements are not just over the validity of alternative means to achieve
a common modeling objective. In fire risk analysis, disagreements often involve the basic
objectives of modeling. That is, what problem are we trying to solve, or what decision are
we trying to make, and what information is most appropriate and relevant to those questions?
What is "Risk"?
In 1987, the Committee on Definitions of the Society for Risk Analysis announced that
it was considering 13 possible definitions of "risk [1]." However, these and the other def-
initions found in the technical literature may be organized in terms of a couple basic concepts.
Risk refers to a, a type and degree of danger or peril or loss; and b, the relative likelihood
or degree of probability that that type and degree of danger or peril or loss will occur. The
aspects of severity and frequency are both part of risk, however it is considered. In this
sense, which underlies all varieties of risk analysis, risk is a characteristic, or set of char-
acteristics, of a person, a building, a product, a city, or any other well-defined object.
The term risk is also used more colloquially to describe not a basis for characterizing an
lAssistant vice president, Fire Analysis and Research Division, National Fire Protection Association
(NFPA), 1 Batterymarch Park, PO Box 9101, Quincy, MA 02269-9101.
131
object but the object itself. In the insurance industry, for example, it is not unusual for the
subject of an insurance policy to be called a risk, which in that context means a source of
potential losses with certain probabilities that are treated separately in terms of one form
of risk management, namely insurance. Also, the term risk may be used in common parlance
to refer to a person or object to indicate that the perils associated with him, her, or it are
so severe or so likely that they warrant special attention. For example, "Keep an eye on
Private Klutz. That foul-up is a risk to us all."
"Risk analysis" can refer to any syste:natic approach to characterize the probability and
severity of losses associated with any suitable object of analysis.
To be useful, the analysis should be able to distinguish significant changes in probability
or severity that would be associated with managed changes in the object. If you modify this
product or building, how does its risk change? If you train this person, how does his or her
risk change? If you position the fire stations of the city in these locations, how will the city's
risk change? Whether the item in question is large or small, it can be the subject of a fire
risk analysis, and if it (or he or she) can be changed in a managed way, then fire risk analysis
can provide useful information to design the programs that produce that change.
Definitions of "hazard" also may be stated in terms of the object that is the source of
danger [2] or some measure of the degree of danger (Terminology Relating to Fire Standards,
ASTM E 176-91d). In practice, definitions of hazard of the latter type have tended to be
vague when they were not duplicative of definitions of risk. For example, the American
Society for Testing and Materials (ASTM) definition of "fire hazard" is "potential for harm
associated with fire, (E 176-91d)." In this definition, it is not clear whether "potential" is
an all-or-nothing designation, in which case everything is to some degree a fire hazard, or
refers to an undefined quantitative measuring scale. The former elaboration of the definition
further suggests that fire hazard is meant to connote an unacceptable level of fire risk, which
introduces a high degree of subjectivity and variation among observers into the character-
ization of the fire hazard of an object (In fairness, it should be emphasized that ASTM has
now withdrawn the discussion linking fire hazard and acceptable levels of fire risk).
In recent years, however, fire modelers in the United States have proposed using the two
terms, risk and hazard, to distinguish two types of analysis. Under this convention, "hazard"
refers to the severity of loss under specified conditions, so "fire hazard analysis" refers to
analysis of perils without reference to probabilities [3]. In other places, this distinction has
been labelled risk analysis vs. consequences analysis [4]. Whether these terms are used or
not" this distinction is important.
What is "Risk Analysis"?

Returning to the definition of fire risk analysis, you always need one or more measures
of severity and a probability distribution for each [5,6]. The measures of severity, however,
can be quite varied. The measure might be deaths, injuries, or monetary damages. Or, to
focus on major events, the measure of severity might be a qualitative measure based on a
basic fire loss measure, such as whether or not ten people died or whether or not damages
exceeded $50 million. The measure of severity might address solely the damage to certain
critical pieces of equipment that underlie a company's ability to continue operations or a
nuclear facility's ability to avoid meltdown. The measure of severity might be spatial, such
as the number of floors, rooms, or square feet damaged by fire. The measure of severity
might be temporal, such as the time required to reach flashover or the time from ignition
to extinguishment. The measure of severity could even be a point score with no simple
physical interpretation.
This diversity of possible measures of severity should dispel one common myth about fire
risk analysis, which is that it is all about averages. There is nothing in fire risk analysis to
HALL ON DETERMINING A FIRE RISK ANALYSIS MODEL 133
preclude a focus on very severe events. However, it is also important to understand what
is and is not involved in using an average. If the measure of severity is deaths, the fire risk
measure will be an average number of deaths in fire, but that may not be the same as the
number of deaths in a typical or average fire. Most probability distributions in fire are
skewed, and so the averages are pulled upward by the influence of major events. For
example, the average death toll in hotel and motel fires in the 1980s is 10 to 15% higher
because of the MGM Grand Hotel fire.
Most fire risk analysis models and methods do not develop a probability distribution
directly on the measure of severity, because such information is not available directly, but
must be built up from information on different fires having the same or different severities.
Instead, modelers work through a scenario structure. In mathematical terms, the analyst
describes the universe of possible fires and divides them into a manageable number of non-
overlapping groups called scenarios. Each scenario consists of a set of very similar fires, so
similar that one can validly choose one representative fire and use the severity of that
representative fire to characterize all the fires in that scenario.
The analyst then develops a probability distribution, either directly for the scenarios, or
indirectly, by first developing a probability distribution function for all fires. Either way,
the result is a probability value for each scenario. Note that this probability distribution
need not be based entirely or even primarily on historical fire experience. If the past is not
an adequate guide to the future, then other sources, including expert judgment, may be
needed to produce the best estimates for the probability distributions. And the probabilities
themselves will be uncertain, so it is best to construct probability distributions on the prob-
abilities themselves, in a Bayesian analysis of the sensitivity of the probability estimates.
Any analysis that does not incorporate probabilities is not a fire risk analysis. If it focuses
on one fire or one scenario, then it fits the definition given earlier for a fire hazard analysis.
Looked at another way, a fire risk analysis is a series of that kind of fire hazard analyses
whose results are combined by use of a probability distribution.
lbe Dangers of Mishandling Probabilities

The most common mistakes today in proposing models as fire risk analysis methods involve
Ibe handling of probability distributions. Consider how often discussions of engineering
problems or code questions are framed in terms of the worst fire that is considered "likely"
or "possible." By calling a single fire likely or possible, an attempt is made to convert a
classic fire risk problem into a fire hazard problem that depends on the identification of the
one right fire to use in design.
Even ASTM has sometimes, fallen into this trap. ASTM's definition of fire risk is "the
probability that a fire will occur, and the potential for harm to life and damage to property
resulting from its occurrence" (E 176-91d). That definition identifies fire risk with one
possible definition and not a very helpful or appropriate one at that. It suggests that fire
risk should always be defined in terms of the probability of "a" fire, presumably a particular
fire scenario. There are several problems with this approach.
First, if enough characteristics are applied to the definition of a fire, its probability may
be made arbitrarily small, no matter how common or typical each characteristic is individ-
ually. Second, it is rare that a question or decision is posed in terms of the risk created by
a particular object in one specific type of fire, as this definition does. It is much more likely
that a question or decision will be posed in terms of the effect of a particular object on risk
in the full range of fires in which it might be involved. That is the natural way to phrase
the question, because you do not have the luxury of choosing the fire you will have. More-
over, the object's effect on risk in one type of fire is unlikely to be a sure guide to its effect
on risk in other types of fires, or at least, such representativeness will need to be proven,
DOtjust asserted.
This brings me to the third common misconception about fire risk analysis. The first was
the failure to realize the enormous diversity of possible measures of severity. The second
was the failure to be explicit and thorough in handling probabilities. The third is the failure
to match the scenarios, the measure of severity, and probability distribution to the needs
of the problem.
There are always a multitude of reasons for wanting to narrow the focus of an analysis
to a few fire scenarios. One reason is the high cost of analysis. The second reason is the
desire to stick to the scenarios we understand best, in terms of the state of the art of physics,
chemistry, biochemistry, engineering, and other relevant sciences. But this impulse is really
no different from the person who looks for his lost car keys, not where he dropped them,
but under a lamp where the light is better. If the problem to be solved involves estimating
the risk of a particular object of analysis in some overall sense, as it almost always does,
then only a comprehensive scenario structure will suffice.
The Importance of a Comprehensive Scope

In fire risk analysis, everything is relevant and must be included unless you can prove
otherwise. Fires of low severity and fires of low probability must be included. Unless you
can show that a certain class of fires has such low probability and low severity that its
behavior could not affect the overall calculations, you have no valid basis for excluding it
from explicit consideration.
Fires we understand well and fires we understand poorly must be included. Physical and
behavioral variables must be addressed. A method that excludes anything you can think of
is presumptively invalid unless the modeler can show that the excluded element is either
(a) inherently irrelevant to the subject of the analysis; or (b) clearly incapable of affecting
the calculated results. Fortunately for fire risk analysts and the people who pay them, many
elements often can be excluded on one of these valid bases, but the burden of proof is
always on the modeler.
In addition to this formal burden of proof, there are more subtle standards involved in
choosing the right fire risk analysis model for the question at hand. For example, the model
may depend upon who the client is and how fire risk manifests itself to that client. If I am
a client who manufactures products that can be involved in fires, you would expect me to
be interested in an analysis of all the fires my product might contribute to. However, if I
believe that fire losses involving my product will not affect me except through product
liability actions, I might be interested only in an analysis limited to the larger fires that
produce such lawsuits. These larger fires might involve a very different pattern of product
involvement than the universe of fires involving my product. Nevertheless, that might be
the right analytical approach for me if it best captures the measure of severe consequences
to me. What would not be right or defensible would be to use a narrowly focused analysis
like this for a public interest group or consumer protection agency whose valid interests
extend to all the fire losses affected by a product.
Conclusions
Everyone in the fire community is involved in some type of fire hazard or fire risk analysis
or management, whether implicit or explicit, qualitative or quantitative, scientific or im-
pressionistic, broad scope or narrow scope. The challenge to all of us, then, is to recognize
the nature of the decisions we are called upon to make and to steadily improve the quality.
relevance, and understandability of information available to us to make those decisions. It
is easy to be overwhelmed or intimidated by the number and technical complexity of models
HALL ON DETERMINING A FIRE RISK ANALYSIS MODEL 135
and data bases now available or in development to help in these areas, and the discussions
of assumptions, validity and sensitivity can become quite arcane, seemingly accessible only
to a small scientific elite. In the end, however, data gaps and technical complexity cannot
be used to justify narrowing the scope of analysis, which should be dictated by the nature
of the problem, but only to illuminate choices in the approach to the analysis.
In so doing, it is essential that each of us check every analysis and model for the essential
elements necessary for valid use of that method. If the issue involves fire losses, it probably
is a fire risk question. If the method is to be fire risk analysis, then it must deal with measures
of severity, scenarios, and probability distributions. And, it must be completely compre-
hensive with solid reasons for all exclusions. These reasons should related to the issue being
addressed, not the strengths and weaknesses of the model or analyst.
As you assess the value of candidate models or analysis methods, also consider these
common themes. Does the model take proper account of the human element in all its
aspects? Are the field data sources well-designed and demonstrably representative? Is the
potential value of prevention properly reflected? Are the systems aspects, the interrela-
tionships of elements, captured in the model? Are the laboratory data sources well-designed?
Do you understand the essential logic flow of the model: what it calculates, what the
calculations are intended to represent, why they are fitted together as they are, and what
the outputs mean? How sensitive is the model, or better yet, your decisions, to the as-
sumptions or the accuracy of the input data? How uncertain are the outputs? Where and
how does the model improve most on the assumptions implicit in the way you do things
now?
Ask yourself and the "experts" you trust these questions. If the answers make you com-
fortable, then use the emerging fire risk and fire hazard analysis techniques to improve your
decisions. If not, ask more questions. Don't be steamrollered by methods you don't under-
stand, but don't miss the right opportunity to move into the future of scientific fire safety,
either. Look to places like the National Fire Protection Association (NFPA), ASTM, the
Society of Fire Protection Engineers (SFPE), the National Institute of Standards and Tech-
nology (NIST) Building and Fire Research Laboratory, the National Research Council of
Canada (NRCC), and the International Association of Fire Safety Science (IAFSS) for
indications of which methods are sound and which are still experimental. (There is also a
large general literature on risk analysis outside the field of fire. That literature has not been
extensively referenced here because most of it focuses on issues that are not critical to the
issues in this paper, such as procedures for evaluating risk once it has been estimated and
estimation of long-term risks from chronic conditions or long-term exposure.)
Together, we can manage fire risk more intelligently and effectively, to the benefit of
everyone.
References
[i] "Defining 'Risk'" RiSK Newsletter, Vol. 7, No.3, Sept. 1987, p. 5.
[2] Accident Prevention Manual for industrial Operations, 8th Edition, Frank E. McElroy, Ed., National
Safety Council, Chicago, 1981, p. 60.
[3] Clarke, F. B., "Fire Hazard Assessment," Fire Protection Handbook, 16th Edition, Arthur E. Cote,
Ed., National Fire Protection Association, Quincy, MA, 1986.
[4] Roux, H. J., "A Discussion of Fire Risk Assessment," Fire Risk Assessment, ASTM STP 762, G.
T. Castino and T. Z. Harmathy, Ed., American Society for Testing and Materials, Philadelphia,
1982.
[5] Hall, J. R., Jr., "Fire Risk Analysis," Fire Protection Handbook, 16th Edition, Arthur E. Cote,
Ed., National Fire Protection Association, Quincy, MA, 1986.
[6] Hall, J. R., Jr. and Sekizawa, A., "Fire Risk Analysis: General Conceptual Framework for De-
scribing Models," Fire Technology, Vol. 27, No.1, Feb. 1991, pp. 33-53.
Richard W. Bukowski, 1 S. Wayne Stiefel, 1 Fredrick B. Clarke, III, 2
and John R. Hall, Jr. 3
Predicting Product Fire Risk: A Review of

Four Case Studies
REFERENCE: Bukowski, R. W., Stiefel, S. W., Clarke, F. B., III. and Hall, J. R .. Jr..
"Predicting Product Fire Risk: A Review of Four Case Studies," Fire Hazard and Fire Risk
Assessment, ASTM STP Jl50. Marcelo M. Hirschler. Ed .. American Society for Testing and
Materials, Philadelphia, 1992. pp. 136-160.
ABSTRACT: Traditional methods of assessing fire risk are based on probabilistic treatment
of fire incident data. Recent advances in the ability to make deterministic predictions of the
consequences of specific fire scenarios present an opportunity to reduce this dependency on
incident data. and greatly improve the ability to assess the risk associated with new products
for which such data does not exist. This paper outlines a risk assessment method developed
for such a purpose. A detailed report describing the methodology is available. This paper also
reviews the results of four case studies performed as part of the development of this risk
assessment method. The four reports, providing full dctails of these trial applications, are also
available.
KEY WORDS: computer models. fire statistics, hazard assessment, ignitability, probability,
risk assessmen t
Introduction to the Risk Method

The methodology employed to predict product fire risk is the result of a project funded
by a broadly based consortium of manufacturers and trade associations through the National
Fire Protection Research Foundation (NFPRF) [1]. It builds on recent advances in deter-
ministic fire modeling to provide a physics-based method of estimating the severity of
specified fires, while retaining the probabilistic framework of traditional risk analysis for
estimation of the relative probabilities of various types of fires. This fire risk assessment
method is designed to permit the quantitative estimation of the change in expected fire
fatalities· per year attributable to changes in fire performance properties of combustible
products (for example, building contents and furnishings) in the context of their end use in
specific occupancies.
The method is designed to calculate the expected severity (in deaths per fire) and the
relative likelihood (as in fire probability) of a large number of fire scenarios that may involve
INational Institute of Standards and Technology, Center for Fire Research, Gaithersburg, MD 20899.
2Benjamin/Ciarke Associates, Inc., Kensington, MD 20895.
'National Fire Protection Association, Quincy, MA 02269.
4The limit to fatalities is not inherent in the Risk Assessment Methodology described herein, but
rather is a limitation of the current state-of-the-art embodied in the hazard assessment method upon
which it is based. As the hazard assessment method is improved to consider injuries and property
damage properly, the risks associated with these end points will be calculable.
136
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 137
the product as the first item ignited or as a secondary contributor. Expected severity is
estimated through the use of a computer-based hazard assessment method, which requires
very specific information on the physical properties of burning items; the thermal properties
of rooms in which fire occurs; the sizes and layout of rooms and their associated openings;
the locations and conditions of occupants; and the status of built in detection systems.
Relative likelihood is modeled using fire incident data, data from other sources, and many
assumptions, since such information can be obtained only in terms of classes or ranges, for
example, the class of burning items, classes of rooms, and ranges of occupant ages.
The use of scenarios is the technique by which these two modeling components are joined
together. The universe of possible fires and situations is divided into scenarios, with each
scenario being represented by a single, well-defined case. Selection is of the most average
or typical case of the set of cases represented by the class.
The execution of this technique produces results somewhere between the extreme of
(1) running the hazard assessment method once on the allegedly most typical case, a gross
oversimplification that could not be reasonably expected to estimate overall risk and (2) the
use of classes so precisely and narrowly defined that every class is completely homogeneous
with respect to the properties of interest and the issue of typical or average becomes moot,
an impractical approach which would require nearly infinite time and cost to compute and
would have to use probabilities that were nearly all pure guesswork.
The selection of a best compromise between these two extremes requires many difficult
and uncomfortable choices. Class definitions will inevitably be wide enough that they will
encompass some fires or situations with significant differences, thereby placing importance
on the process of designating an average case to represent them. At the same time, class
definitions will inevitably be narrow enough that the best sources of information on relative
likelihoods will not be sufficient, and one will need to use less satisfactory sources, including
expert judgments.
Much of the parameter estimation for the method requires use of multiple data sources,
because of limitations of even the best data sources. For example, national fire incident
data are best on representativeness but tend to be quite limited on detail, while laboratory
experiments are excellent on detail but tend to have uncertain representativeness.
Throughout the cases used to test and develop the method, one will note many judgments
on best current data sources on particular points. In most instances, there is potential for
improvement in the quality of these data sources. In a few instances, there may be no
acceptable current data source, that is, even an expert panel cannot give acceptable answers.
This means that there are data gathering and data system upgrade projects implied by the
method. At the same time, it is recognized that one cannot wait for complete data on all
issues collected by only the best method.
The method is somewhat modular in that it can readily accept better quality data than
that which is now available. Changes in the type of information directly available, however,
would require restructuring of the method to fit the new data structures.
Risk Method Approach

The approach for this risk assessment method is shown below. It begins with the con-
struction of a model (Steps 1 to 5) that reflects real world fire experience as a basis for both
the probability estimates and the inputs for the calculation of scenario probabilities. Since
we can alter this model at will, it becomes a way to predict risk changes in the real world
in response to a change in any independent variable (for example, a measure of the fire
performance of the product) in the model. The model can effect changes instantly, and can
be used to examine the impact of changes in the fire performance of products without
incurring any true losses. It is also easier to observe the model world, and thus to interpolate
to data not directly available from the real world.
Summary of the Approach
1. Select product and occupancy pair.

2. Specify "representative" building characteristics.
3. Develop scenario structure.
4. Adapt fire model to meet needs of product and occupancy pair.
5. Specify occupant sets and associated model inputs.
6. Perform risk calculation for "base" case and compare results with available fatality
statistics, by scenario.
7. Perform risk calculation for the "new" product and compare results with "base" case.
8. Interpret the outcome.
Next, the consequences of each scenario are calculated for the "base" and "new product"
cases (Steps 6 and 7). Typically, a number of iterations between the initial steps and the
base case calculation (Step 6), may be needed to adjust model inputs to obtain suitable
agreement between the base case calculated probabilities and the available fire death sta-
tistics. The emerging technology of deterministic fire modeling has provided the fundamental
predictive capability that has enabled this advance. The risk assessment method encompasses
detailed procedures for formulating such a model, calibrating it so that it behaves like the
real world, and then using it to obtain the desired information.
Finally, the results of both calculations are compared to obtain a relative risk for the new
product relative to the base case (Steps 7 and 8). The risk method assumes that the full
range of fire scenarios that might involve the product in the real world can be represented
by a limited number of scenarios in the model world. In this approach, a full range of
scenarios that might involve the product is created, and probabilities and other input pa-
rameters are specified. Calibration and validation include checking that recent fire losses
are accurately predicted (for example, recent fire death rates) when recent patterns of usage
of the product are provided as input (for example, input parameters representative of the
fire-related properties of upholstered furniture now in use). This is called the base case or
baseline case. The change in the risk measure, produced by changing the assumed properties
of the product, can then be used as the measure of the fire risk impact of changing the
product to have those properties. This second calculation is called the new-product case. In
this way, a multifaceted product change (for example, universal use of a new style of
upholstered furniture that is more difficult to ignite and has a lower peak heat release rate,
but produces faster fire growth and more toxic smoke when it burns) can have its net effect
on fire risk determined.
The subject for analysis is a particular class of mutually substitutable products (for ex-
ample, upholstered furniture or carpeting) in a particular type of occupancy (for example,
detached dwelling or office building). The method calls for a scenario structure that will
represent both the most typical and the most hazardous fires that may involve that product
in that occupancy. Each scenario is defined primarily by characteristics that can be derived
from what is known about the product, the occupancy, and the pattern of use of the product
in the occupancy.
For example, specification of an occupancy typically will imply a certain range of building
sizes. Typical internal layouts and construction methods also may be implied. Databases
often exist to indicate the relative likelihood of building characteristics among the members
of this occupancy class (for example, how many dwellings are two stories). Fire data analysis
may be able to determine whether the relative likelihood of these building characteristics
differs among buildings that experience fires.
By following the defined process, one constructs a range of scenarios, each defined with
enough physical detail to permit the estimation of fire severity for that scenario, using one
of the new deterministic hazard calculation methods, such as HAZARD I [2]. At the same
time, traditional methods are used to estimate probabilities for each scenario. These methods
rely particularly on representative national databases, including the national estimates method
for estimating patterns in U.S. fire experience from the U.S. Fire Administration's (USFA)
National Fire Incident Reporting System (NFIRS) [3] and the National Fire Protection
Association's (NFPA) annual stratified random-sample survey of U.S. fire departments [4].
It may be useful to think of this modeling process as a particular form of abstracting from
the real world to a model world that can be manipulated by analysis. In the real world,
every fire can appear to be unique. However, from the point of view of estimating fire
losses, certain factors are more important than others, and they can be used as a basis for
modeling the real world. Similarly, in the real world, the possible variation (for example,
number of losses) on just one characteristic may be enormous. Again, from the point of
view of estimating fire losses, it often is possible to select only a few of the variety of
possibilities to be representative of the others and to show that such a simplified model
world produces estimates of fire loss that are not significantly different from those experi-
enced in the real world. Demonstrating that the model world accurately reproduces results
from the real world is the step of model validation.
Purpose for the Case Studies

A case consists of the pairing of a product and an occupancy. Four case studies were used
in the development of this methodology. Upholstered furniture in single family detached
dwellings was selected as the initial case study. The other cases studied were carpet in offices,
concealed combustibles in hotels, and wall coverings in restaurants. Each of the product/
occupancy pairs required the project team to address additional challenges which served to
sharpen the methodology, improve the applicability of the hazard modeling, and point out
the current limitations imposed by the state of modeling and the availability of data sources.
The methodology evolved as refinements based on the experiences of application which
were incorporated. This report summarizes the "final" structure which, in itself, will continue
to evolve with the capabilities of fire modeling and data collection. The four case studies
are documented in detail in separate reports [5-8].
It is important to emphasize that the four case study reports do not represent a complete
fire risk assessment of the selected product and occupancy. Rather they serve as:
• Examples of the application of the risk method to a problem of interest;
• a focus for expanding the technology into areas of special interest to the project team
and sponsors;
• a test of the capabilities of the method to deal adequately with the specific issues related
to these products and occupancies; and
• a demonstration of the different analytical techniques to be applied under different,
case-specific circumstances.
Upholstered furniture in residences, Case 1, represents the best possible combination of
breadth of incident data and technology squarely within the scope of the original method.
That is, HAZARD I as the core of the risk method is particularly well suited to the analysis
of fires involving the contents of residences. Thus, no extensions of the technology were
needed for the first case.
Upholstered furniture represented a product which is susceptible to both smoldering and

flaming ignitions, and is (voluntarily) regulated as to its ignitability by cigarettes. This
regulation went into effect recently enough that an apportioning of current stock was nec-
essary, and thus was demonstrated.
The fire incident data on residential fires is by far the most detailed, and non-fire data
(for example, Census data) provided a rich data base for the definition of scenarios. Thus,
Case 1 demonstrated how an abundance of data should be grouped to maintain a reasonable
number of computer runs.
Finally, the rich data base and large number of incidents (and fatalities) provide an
observed fatality rate for every scenario. This gave data against which to calibrate and
validate the method, and allowed the report to demonstrate the best way of presenting the
copious amount of data obtained (that is, stacked bar graphs organized around the usual
demographic categories used to analyze the incident data).
In many ways Case 2 (carpet in offices) represented the antithesis of Case 1. The incident
data showed no fatalities, and the product seldom burns by itself. The occupancy involves
a large building occupied by a sufficiently large number of people, so that congestion
problems during evacuation are expected.
The combination of no fatalities and only secondary product involvement meant that there
was no incident data against which to calibrate the method. The product burning charac-
teristics required the development of a means to quantify the degree of involvement as a
function of measured properties, and the occupancy required a different approach to evac-
uation simulation.
The report demonstrated how some degree of confidence in the results could be obtained
with a total lack of calibration data. In addition, it showed how the method can be used to
analyze the potential impact of a reduction in regulation.
Case 3 (concealed combustibles in hotels) involved a product which represented both an
ignition source and a fuel load. The major challenge was that it was located within the
structure rather than in a traditional room. The particularly small volume of the stud cavity
made ventilation a major determinant of how the product burned, and thus a technical
challenge for the team to address. And the occupants represented a mixture of the family
groups of Case 1 (in the guest rooms) and the larger groups of Case 2 (in the function
rooms).
The incident data showed fatalities in some but not all of the identified scenarios. Thus,
the report needed to show yet a third way of analyzing and presenting the results. Here we
chose to group the scenario classes by whether there were fatalities predicted (by the method)
and observed (in the incident data), no fatalities either predicted or observed, and fatalities
predicted but not observed and vice versa. Also, since the number of scenarios were much
fewer than in Case 1 and to present the numbers with which the user of the method must
work, the data were presented in tabular form in contrast to the histograms of Case 1.
Case 4 (interior finish in restaurants) brought additional challenges in dealing with the
subject of flame spread on interior finish in an occupancy which was harder to define than
any other due to a large variability. Here, the incident data showed a small number of
fatalities (more than offices and fewer than hotels) and a product which can range from
only secondarily involved (like carpet) to fast burning like furniture.
Case 4 exhibited similarities to the product of Case 2 (carpet) and the occupancy of Case
3 (for the function room fires). The fact that the method predicted no fatalities required a
presentation of results similar to Case 2 but much more sensitive to the assumed burning
properties of the product and to the assumed characteristics of the occupants. Thus, while
Case 2 might indicate a significant safety factor extant with carpet meeting current regulation,
Case 4 shows that good finish materials are safe while poorly performing materials can be
hazardous under certain circumstances.
Case I-Upholstered Furniture in Residences

Upholstered furniture (sofas and chairs) in detached single family homes was selected as
the initial product/occupancy pair. This choice was made for several reasons. First, uphol-
stered furniture plays a prominent role in fatal fire scenarios. In 1983 to 1987, there were
an estimated annual 1070 deaths, 2450 injuries, and $249M in direct losses attributable to
U.S. structure fires reported to fire departments where upholstered furniture was listed as
the first item ignited [9]. Second, a large volume of fire data has been accumulated for these
incidents. Validation of the risk methodology and its predictions is enhanced when sufficient
data are available. Third, because oftheir role in residential fatalities, upholstered furnishings
have been extensively studied in the laboratory and test methods are available which measure
their burning characteristics [10]. Finally, the fire model which is an integral part of the fire
risk assessment methodology was developed for use on this type of building [11]. This
combination of national fire experience data, laboratory studies, and fire development,
made upholstered furniture in detached single family dwellings a good choice to test the
fire risk assessment methodology.
Upholstered furniture includes all types of chairs, sofas, and ottomans which have cover
fabrics over filling materials. Since (as shown in Table 1) national fire experience indicates
that nearly 80 percent of the fires involving upholstered furniture as the first item ignited
occur in living rooms (66.7% in the living room, den, and lounges) and bedrooms (11.2%),
it was decided to restrict this demonstration to furniture in these two types of rooms. A
more complete analysis would also include kitchens, dining rooms, and storage areas, where
upholstered furniture's major contribution may be as a fuel source for fires originating in
other items.
We also narrowed the analysis to single-family detached houses. This was done primarily
in consideration of modeling limitations. Unlike apartments, detached dwelling fires can be
modeled without concern for barrier breaching between housing units (for example, burn-
through of doors, walls, ceilings) and complex smoke movement through elevator shafts,
heating ventilation, and air conditioning (HV AC) systems and interstitials (void spaces within
walls or above ceilings) - where both modeling and data are weak. U.S. Census data
indicates that roughly 70% of single family houses are one-story, with the remainder either
split level or two or more stories. The selection of a single house type of one-story limited
the computational burden without seriously compromising the value of the case study.
Even with this restriction to two rooms and a single house type, this yields a base case
involving 64 specified fire scenarios; 16 fire scenarios when upholstered furniture was the
first item ignited and 48 fire scenarios when items other than upholstered furniture were
first ignited. The evacuation simulation was run twice for each occupant set, once with and
once without operational smoke detectors. Since 220 occupant sets were derived from pop-
ulation and activity analyses, 440 evacuation model runs were run per fire scenario (all
together over 28 000 runs). Each additional variation causes the total to grow exponentially.
In addition, for the cases where upholstered furniture was the first item ignited, numerous
sensitivity analyses were performed, involving many more runs.
Case 1 Results
The base case consisted of living room and bedroom fires. The base case results are
divided into two groups of fire scenarios. The first group consists of those scenarios where
upholstered furniture was the first item ignited. We will be comparing the total deaths
predicted by the risk method with the national fire database estimates for the identical
scenarios. We will also be looking at how well our results compare when we examine the
finer structure (smoldering versus flaming ignitions and the distribution of deaths by time
of day).
Figure 1 indicates that the risk model and fire database estimates are in good agreement
for the base case when upholstered furniture was the first item ignited. When we compare
upholstered furniture fires by type of ignition in Figure 2, this agreement is maintained with
the percentage of deaths from smoldering equaling 77% (statistics) versus 74% (predicted
by the method).
The risk model prediction of deaths by time of day (Figure 3), however, does not do
nearly as well. The statistical values indicate 59% of deaths occur at night while the method
predicts 88%. This discrepancy might be explained by the assignment of occupant location
and behavior (sleep status by time of day). All occupants are assumed awake during the
day, all of the elderly are asleep during the evening, and all occupants are assumed to be
asleep at night. Clearly, vulnerability to fire increases for those individuals asleep (partic-
ularly without smoke detectors). To adjust this imbalance, the distribution for sleeping status
could be shifted, such that some people are awake at night and so;ne are asleep during the
day. However, there are no data known on which to base such an improvement. If decisions
are to be based on changes in predicted fire risk that are sensitive to this type of variable,
then special studies will need to be conducted first to establish the proper distribution.
In the second group are those living room and bedroom scenarios where items other than
upholstered furniture are first ignited and the secondary ignition of upholstered furniture is
deduced from the methodology. Figure 4 summarizes the results from the scenarios when
upholstered furniture was not the first item ignited. We can see that the method overpredicts
the total deaths by almost 50%, with the bulk of these excess deaths in the smoldering fire
scenarIOs.
When we examine time of day predictions, Figure 5 indicates that the night time fire
deaths are over-predicted by 100%. A more detailed analysis reveals much better agreement
by an aggregate of 141 of the 144 scenarios represented and a gross overprediction by the
other three scenarios. Three flaming, night time, living room scenarios (represented by the
slope/peak heat release rate combinations: fast/medium, medium/high, and medium/low)
add 509 deaths over and above the number expected. Therefore, without the overprediction
by these three scenarios, the total deaths would be 624 (predicted) vs 770 (statistics). Also,
the flaming model results would be 544 rather than 1053. This means the other scenarios
underpredict the total by only about 20 percent, well within the limit of the method. This
is indicative of the power of this type of methodology in helping us understand the reasons
for the results and where improvements can be most effectively made.
New Product Risk
For this example, we have selected new upholstered furniture with the same fire properties
as the base case. However, the materials used in the new product result in smoke with a
ten-fold increase in toxic potency (90 mg-min/L) over the base case. This means that the
lethal level of smoke is reached earlier in the fire with less fuel burned and with a lower
upper layer temperature. Such a ten-fold change is generally regarded as the smallest dif-
ference in toxic potency which should be considered significant when measured in smalI-
scale toxicity tests. This example will be limited to primary ignition of the furniture in living
rooms and bedrooms.
Figure 6 compares the results, indicating a 46% predicted increase from 624 to 909 deaths,
with the cause of death dramatically changed. Smoldering fire scenarios contribute three-
fourths of the increase while flaming scenarios contribute the remaining one-fourth. For the
base case, the cause of death was always convected heat. However, as Figure 5 indicates,
toxicity Ct is now the causal factor in 96% of the deaths.
It is evident from these results that the risk methodology will respond to gross changes
in toxicity both in terms of the number of deaths and the cause of death. However, the risk
method is limited to the use of the Ct variable as the only practical indicator of the toxic
potency of the product. While the N-Gas model is more accurate, it requires data on the
combustion chemistry of the item(s) burning-data which are not applicable to the generic
products with which these risk analyses deal. For a more thorough discussion of this point
the reader is referred to section C.2.e of Reference 1.
Case 2-Carpet in Offices

Carpet in offices was selected as the second product occupancy pair for several reasons.
First, unlike the initial case study on upholstered furniture in residences, carpet in offices
does not playa major role in fatalities as a first item ignited. In fact, a review of national
fire experience for the period from 1980 to 1984 revealed an average of only 6 fatalities per
year reported for offices, none involving carpet as a first item ignited. However, its prev-
alence in offices means that carpet could be a contributor to the fire hazard if ignited by
other items.
Second, modeling a fire involving carpet required developing an additional computational
procedure to supplement the HAZARD I fire model FAST. The burning characteristics of
carpet have been studied extensively at both small- and full-scale levels [12], and a standard
test method is available which measures the critical radiant flux needed for carpet to spread
flame (ASTM Method for Critical Radiant Flux of Floor-Covering Systems Using a Radiant
Energy Heat Source, E 648). The fire model [11], which is an integral part of the fire risk
assessment methodology computes the flux to the floor, and could be easily modified to
estimate the area of involvement of the carpet as a function of time.
The third reason for selecting this product involved the challenge to the method offered
by office buildings, which require an evacuation model capable of simulating as many as
several hundred persons evacuating a large building. The EXITT model which is part of
HAZARD I, was designed for a family evacuating a residence. Thus, we were challenged
to develop an alternate evacuation calculation which is more appropriate for non-residential
occupancies where there are a larger number of occupants and where behavioral interactions
are not reported in actual fire incidents.
This combination of national fire experience data showing no fatalities, available labo-
ratory studies and relevant data, and the opportunity to make an appropriate modification
to the fire and evacuation models, made carpet in offices a good choice to extend the fire
risk assessment methodology relative to the first case.
Case 2 Results
All occupants of the office are predicted to evacuate safely for all cases examined. In
most cases, the people are in the stairway before the carpet becomes involved. Even where
this is not the case, the people move faster than the carpet spreads flame, and conditions
in the occupied spaces remain survivable until the occupants have left (see Table 2).
The fire induced conditions in the fire room immediately after the occupants vacated the
space are tabulated below. Since the evacuation times were sufficiently short to preclude
any effect from toxic gas exposure (as these are time-integrated values), the only possible
limiting condition WliS the temperature of the upper layer.
Based on the procequres in TENAB (13], occupants are exposed to the upper layer only
when the layer interface is below 1 m or when it is below 1.5 m and the upper layer
temperature is below 50°C. At all other times, the occupant is exposed to the lower layer.
In the present case, the lower layer temperatures are at most only a few degrees above
ambient during evacuation. Using the TENAB criteria, none of the occupants in these cases
were exposed to the upp~r layer.
The statistics show that tn~re are few fatalities in office fires, most of which do not occur
in the daytime and no fatal fin~soriginating in carpet at any time. The risk method predictions
show no fatal fires at all. Nearly ~Il of the fatalities contained in the statistics involve rapidly
developing fatal injuries to people close to the point of ignition, a scenario the method does
not include, and one whic/1 was of decidedly less importance in Case 1. However, this is
not a sufficient basis for amibuting a degree of quality to the calculation.
New Product Risk

For this study, it was (jecided (with input from the Advisory Committee) that the new
product case would consist of identifying the values of critical radiant flux and flame spread
rate beyond which we would begin to record fatalities in daytime office fires. This was fairly
straightforward, in thpt it required only that we determine:
• a lower limit for the critical radiant flux for spread such that the ON TARGET value
was sufficient to allow flame spread during evacuation; and
• a flame spread rate sufficiently high that it could overtake evacuating occupants.
With respect to th~ critical radiant flux, the value (for spread) must be low enough that the
carpet will spread flame without incapacitating the occupants by the flux criterion (the
Derkson Curve [14]). This criterion is a time-integrated rather than a single value, so for
the evacuation times obtained in this case, it would be in the range of 0.1 to 0.2 W/cm2•
This also brings the carpet into the range where it might be susceptible to primary ignitions,
resulting in the need to analyze all of the "product first ignited" scenarios not currently
included.
The minimum flame spread rate required to obtain fatalities would be equal to the assumed
walking speed of the occupants-here assumed to be 235 ft/min. Thus, any new carpet
which exhibited a flame spread rate higher than this (at the observed flux levels) might
cause fatalities. We know from data on old carpet that critical flux values in this range were
once observed. While it is not likely that usual carpet materials would exhibit flame spread
rates this high, unusual materials (for example, woven mats), orientations (for example,
carpet on walls), or circumstances (for example, carpet soaked with a flammable liquid)
could exhibit high flame spread rates.
Case 3-Concealed Combustibles in Hotels

Concealed combustibles in hotels were selected as the third product/occupancy pair to be
analyzed using the prototype risk assessment method. Considerable controversy surrounds
questions of the risk posed by combustible materials located within wall and ceiling spaces
of buildings. One side argues that these materials represent a concentrated fuel load located
in a space from which combustion products can spread through unseen and unplanned
passages. The other side counters that ignition sources within the concealed spaces are rare,
and the material is protected from exposure to the effects of room fires by finish materials
which are often part of a fire-rated assembly. With respect to the objective of challenging
the prototype risk method, concealed combustibles in hotels presents several benefits. First,
like the dwelling furniture case (Case 1), the statistics show that there are fatalities from
fires beginning in concealed spaces in hotels. While the rate is comparatively low, fatalities
are recorded for each time of day category and for fires both starting in, and spreading to,
these spaces. Thus, unlike the office carpet case (Case 2), there is nonzero fire fatality
experience against which we can calibrate the risk prediction.
Second, as in the carpet case, modeling a fire in a confined space like a stud cavity within
a wall, involved an additional computational challenge to the HAZARD I fire model, FAST.
When the case was selected, the FAST model could not automatically account for the effect
of reduced oxygen on the combustion process. While this was not an unsurmountable
problem in the first two cases, the very limited volume of the concealed space leads to rapid
vitiation with an attendant reduction in burning rate and production of unburned fuel. Thus,
this modeling limitation had to be addressed in order to handle this case properly.
Third was the new challenge to the evacuation modeling approach. In the first case, the
method had to deal with family groups of five or fewer persons, awake or asleep, who
exhibited a significant degree of interaction. The second case involved a large group of
constantly awake persons who did not interact at all. Hotels are mixed occupancies with
the potential for both. In the guest room area we find individuals or small groups who again
might be asleep, but with little interaction. Additionally, there are function rooms (for
example, ballrooms, meeting rooms, or other assembly occupancies) in which you find
persons like the office occupants. Thus, we need to combine the two evacuation modeling
approaches in some way.
This combination of national fire experience data showing relevant fire fatalities, the
opportunity to make additional enhancements to the fire and evacuation models, and the
strong interest in questions of the risk of concealed combustibles, made this a good choice
to extend the fire risk assessment methodology relative to the first two cases.
Case 3 Results
Since all fires which began in rooms and then spread to the concealed space, (involving
the product) were assumed to do so at or near flashover, the impact of the cable was negligible
on both the temperature and gas levels and fatalities. This is because of the largeness of
the fires when the cable is first exposed. If, however, the cable is exposed (for example,
surface mounted or above a ceiling with tiles missing) the result might be different. But
since we could not develop any reasonable estimates of how often this happens or how
extensively the cable is exposed, we discounted such scenarios.
For the fires which begin within the concealed space, we found that the assumed quantities
of cable and other combustibles could spread fire to adjacent spaces with fatal results. Again,
however, the majority of the fire's effects, and all of the occupant exposure, occur after the
fire spreads to the room; therefore the relative impact of the cable is minimal. This can be
quantified by comparing the burning (mass loss) rate of the cable to that of the rest of the
fuel, at the point of flashover.
Over the range of fires examined there were scenarios where the method predicted deaths,
and those where it predicted no deaths. The statistics also predict deaths on only some of
the scenario groups examined. Thus, the comparisons will be presented in groups where
the method and statistics agree that there were deaths, and those where one predicts deaths
and the other does not. To say they agree says nothing about the quality of that agreement,
but only that deaths did or did not occur in both. We should also keep in mind that the
risk method cannot reproduce individual events, but only general trends.
The data taken from the fire statistics are reported in Tables 3 through 6 to two decimal
places to give the reader a feel for the numbers which were obtained in the calculation.
This most likely represents more significant figures than the data uncertainties justify. Like-
wise for the method, numbers were rounded to one or two significant figures, although one
is the most that can be justified.
The output of the method, and that which is compared to the statistics, is the death rate
for fires involving the product. This death rate (and the number of deaths per year) can be
fractional numbers because they represent annual averages; that is, the statistical value is
the average number of deaths per year for the years 1981 through 1985. Likewise, the
number of deaths per year will be less than the death rate (deaths per 100 fires) whenever
there are fewer than 100 fires per year in a given scenario class.
Scenarios Where the Method and Statistics Predict No Deaths

For guest room fires originating in the wall (product is first ignited), there were no deaths
predicted or observed where the fire was confined to the wall space or where it was confined
to the first object ignited in the room. For the function room fires, there were no deaths
predicted or observed for scenarios where the fire started in and remained in the ceiling
space and spread to the room but remained there, or where the function room went to
flashover.
Scenarios Where the Method and Statistics Predict Deaths

Those scenarios where deaths are both predicted and observed include those beginning
in the guest room and spreading beyond the room of origin, and those beginning in the
function room and spreading beyond the room of origin. Note that in none of these scenarios
is the product the first item ignited. The results of these cases is presented in Tables 3 and 4.
In Table 3 we see that the method underpredicts annual fatalities by a factor of 2 in the
day, to a factor of 100 at night. By our criteria this ranges from good to poor agreement
with the total annual deaths in the good agreement category. We find this generally the case
throughout the results from this case study.
Note that the only fatalities in these scenarios are incapacitated, and all fully-capable
adults escape. This is not likely, and is not borne out by the statistics. This observation leads
us to believe that the differences with statistics may be related to the alerting criteria and
the response to fire queues assumed in the evacuation analysis.
Table 4 shows a similar level of agreement with statistics when smoke detectors are present
as was evidenced in Table 3. However, without detectors the method grossly overpredicts
the death rate compared to what is observed. This result seems to derive from the high
occupant load, conservative criterion for thermal fatality, and the zone modeling assumption
of laterally uniform layers. In the corridors, this leads to exposure of occupants to lethal
conditions at a distance down the corridor from the fire room where the temperature would
likely be lower than predicted. This gross overprediction also led to observations on
sensitivity.
Scenarios Where the Method and Statistics Predict Opposite Outcomes

In the guest room scenarios where the fire begins in the cable within the wall and the
extent of spread is 1) confined to area, 2) confined to room, or 3) beyond room, the risk
method predicts a small number of deaths where the statistics show none. These results are
presented in Tables 5 and 6.
In Tables 5 and 6 we see very small numbers of deaths predicted where none are observed.
These are scenarios of low frequency (probabilities on the order of 10% of that for other
scenarios discussed), so they represent a small number of fire incidents. Thus, this result
may mean that we have been lucky in the incidents to have not had fatalities.
New Product Risk

Using the new cable properties, the function room scenarios were repeated, and new
results obtained. These results are presented in Table 7.
Case 4-Interior Finish in Restaurants

Interior finish in restaurants was selected as the fourth and final product occupancy pair
for several reasons. First, combustible interior finish materials have long been considered
(justifiably or not) as critical contributors to fire disasters. Of particular note in this area
are low density fiberboard (cellulosic) ceilings which were implicated in a number of fires
ignited by overheated fluorescent ballasts, and thin plywood paneling which has a tendency
to delaminate and burn vigorously when exposed to fire. The flammability of interior finish
materials used in selected areas (for example, exit access) of high risk occupancies (for
example, assembly and health care) has long been regulated.
Notes: (1) These results are for fires starting in the new cable within the ceiling void, without detectors.
The base case prediction was for no deaths so these are all new fatalities associated with
the higher flammability cable. The method predicts no deaths with detectors for both the
base case and the new product case.
(2) All victims start in rooms remote from the fire room and die from temperature in the
corridor at 661 s. They begin their evacuation at 536 s and require 135 s to escape. This
means that they need an additional 10 s.
(3) The conditions in the rooms from which the victims came remain survivable, so they would
have survived if they had not evacuated.
Second, modeling a fire involving flames spreading over vertical surfaces required de-
veloping an additional computational procedure to supplement the HAZARD I fire model
FAST. The flame spread process has been studied extensively at both small- and full-scale
levels (15], and a standard test method is available which measures important properties of
materials and assemblies relative to how they spread flame (ASTM Test for Determining
Material Ignition and Flame Spread Properties, E 1321).
A number of flame spread or burning rate models and subroutines appropriate to walls,
are currently under development. But significant effort would be required to incorporate
these into the FAST model on which the risk software depends. A model (called HEMF AST)
which predicts fire development on upholstered furniture (16] has been under development
by a CFR grantee for some years, and operates in conjunction with FAST. Since the walls
of a room can be thought of as a scale-up of the back and arms of a sofa, we felt that
HEMF AST could be utilized to estimate the involvement of the interior finish as a function
of time. Also important is the fact that HEMFAST is designed to require only data from
the Cone Calorimeter (E 1354) [17] and the LIFT apparatus (ASTM E 1321) to characterize
a material's fire performance.
The third reason for selecting this product/occupancy combination was that it allowed the
expansion of the occupancies addressed by the method to Assembly Occupancies. While
Assembly Occupancies do not offer any substantial differences in evacuation simulation
over offices or the function areas of hotels, a number of the most famous major fires have
occurred in assembly properties. Examples range from the Iroquois Theater fire of 1903
(602 dead) to the Cocanut Grove fire of 1942 (492 dead) and the Beverly Hills Supper Club
fire of 1977 (165 dead)-with each having at least some involvement of interior finish.
It was these considerations, and the particularly strong interest on the part of many of
the project sponsors, to see a demonstration of the capabilities of the risk method as applied
to interior finish, which resulted in the selection of this product-occupancy pair as the fourth
case study.
Differences from the Other Cases

This case is distinct from the prior three, in that the project team encountered a lack of
the data required to apply the method to fully assess the risk of this product. Specifically,
the team was unable to ascertain:
• the distribution of finish materials used in restaurants, or even on classes of such

materials;
• the burning behavior of any but a small set of interior finish materials as measured in
the apparatus required by the fire model;
• the physical size or construction characteristics of restaurants;
• physical and mental characteristics of restaurant patrons; and
• only limited information on ignition sources.
Therefore, a very different approach was required. Since this lack of data impairs the
ability to make an overall risk prediction, the risk method was adapted to estimation of
product risk through an examination of hazards in key scenarios identified from the incident
data. In fact, it is likely that many of the initial attempts to apply this risk method to product
occupancy pairs will encounter such limits in the data available. Thus, the lessons learned
from this case study will serve as an example of how to proceed in these cases.
To summarize the approach followed in this case, we
1. Identified the properties of interior finish materials which can serve as a benchmark
rather than representing the performance of the products in current use. These were
a set of textile wallcoverings selected by the American Textile Manufacturers Asso-
ciation (A TMI) as representing the range of fire performance typical of these materials.
2. Specified the physical characteristics of two restaurants which are considered typical,
but where we cannot necessarily quantify their relation to the distribution of properties
III use.
3. Specified a generic set of adult occupants for which an egress time can be computed.
By examining the difference between this value and the time-to-untenable-conditions
in the occupied spaces, the potential impact of variations from the generic occupant
set can be determined.
What results is not a true risk analysis, since the data do not exist to weigh the results by
the degree to which the product properties, occupancy characteristics, or occupants represent
their counterparts in the real world. Instead, what can be done is a range of hazard analyses,
which serve to bound the result that could be obtained if probabilities were available to
combine them. Various measures on this range can be used as benchmarks against which
the performance of other products, buildings, or occupants can be judged in terms of whether
they are safer or less safe than the benchmark.
Case 4 Results
In this section we will present the performance of the wall materials predicted by HEM-
FAST. Since HEMFAST requires significant computer time, simulations were run for only
three min because this was enough time for complete evacuation. Figures 7 to 9 show burn
patterns predicted by HEMFAST. The contours show the predicted flame front at regular
intervals. Thus, closely spaced lines represent slow flame spread rates, and widely spaced
lines indicate rapid spread.
Fire Performance of Textile Wallcoverings
The two materials selected for examination were a polypropylene (ATMI Code PP-PF)
and a polyester (A TMI Code Q). These were the two worst performing materials in the
full-scale tests, but Material Q was substantially worse than PP-PF. Material Q burned
completely and flashed over the test room. Material PP-PF burned vigorously, but self-
extinguished near the corners of the room and did not flash over the room.
The HEMFAST model predicted behavior in the restaurant dining rooms in this case
study, similar to that observed in the full-scale tests despite the difference in room size. As
shown in Figure 7, Material PP-PF ignited and spread flame partially along the long wall
of the dining room. It did not spread around the corners to the side wall but rather self-
extinguished before reaching the corners of the room. Remember that the burning behavior
and resulting burn pattern were not user specified, but rather were predicted by the model
from the small-scale data on the materials. For this wallcovering material, no lethal or
incapacitating conditions were predicted in either restaurant within the first three min (the
maximum time that we ran the calculations) of the time that the wall first becomes involved.
Material Q was predicted to perform similarly to PP-PF, but burned more vigorously,
releasing more energy and mass. Figure 8 shows it spreading flame just into the corners of
the room and stopped with minor involvement of the side walls. Its burning was sufficiently
intense to create untenable conditions in the dining room from temperature just at three
minutes, allowing sufficient time for egress of the fully-capable patrons assumed to be
present. Flashover of the dining room was predicted not to occur.
The flame spread behavior of these materials is driven by the radiant flux to the wall.
The restaurant dining rooms are larger than the room in which the full-scale tests were
conducted, and the flux will decrease as the room size increases, for the same fire size.
Thus, we examined the sensitivity of this result to room size.
Since all other textile materials reported [18], performed better than PP-PF. they would
be predicted to perform no worse when used in the prototype restaurants and would result
in no predicted fatalities.
Since we do not have data on which to quantify the representativeness of the wallcoverings,
buildings, or occupants, we are unable to establish the appropriate weightings and thus the
weighted fatality rate. This means that we can only make qualitative comparisons to the
fire statistics. These statistics show that there is a very low fatality rate for restaurant fires
which may involve interior finish in the dining rooms. While this analysis predicted no
fatalities, the results for Material Q were marginal, and with materials whose results were
only slightly worse, or when there are other extenuating circumstances as discussed above,
the result will be different.
This means that the fire performance properties of material Q provide a reference against
which other wallcovering materials can be compared. As measurements are made on other
materials and assemblies in the LIFT Apparatus and Cone Calorimeter, those whose per-
formance is worse (higher release rates or faster spn;ad) should result in fatalities in these
scenarios, and those whose performance is better, should not.
For this case, the project team decided to examine the impact of just such a diminution
in properties, in terms of both the rate and quantity of energy and mass released by the
wallcovering. Since there is a significant amount of input data required to run the HEMF AST
simulation, and since data files for upholstered furniture assemblies were available (HEM-
FAST is an upholstered furniture model) we decided to use them. This might be similar to
a restaurant where several layers of wallcovering had been applied over one another.
The HEMF AST model predicted very rapid flame spread and vigorous burning until all
of the wall materials were consumed (See Fig. 9). Maximum tenability times were of the
order of only 20 s. With a 6 s reaction time, this leaves 14 s for safe evacuation. For fires
originating in the kitchen, most patrons of either restaurant were killed as they queued at
the doors. For fires originating in the dining rooms, more patrons successfully escaped
because the fire was smaller when evacuation began, but most patrons died in all scenarios;
the number depending on the assumed properties of the material.
Conclusions
In this report, we have described the initial application of a quantitative method for the
estimation of the fire risk, associated with a specified product class in a specified occupancy.
These initial case studies provided a benchmark against which the prototype risk prediction
method's capabilities could be measured against the goals of the project as originally con-
ceived. In some cases, the performance of the method was better than expected for the first
application. In others, differences were attributable to shortcomings in the method arising
from a lack of technical knowledge of fire phenomenology or a lack of detail in required
data. Thus, this effort was beneficial in identifying key needs for improved fire science or
data.
In particular, the first case demonstrated the ability of the risk method to provide a
reasonable agreement with the national estimates for fire losses, including detailed demo-
graphic breakdowns for death rates by type of housing, time of day, age group, detector
presence, and type of ignition. The method provides a structure for developing detailed fire
scenario descriptions which include not only primary ignitions as documented in current
incident databases, but also secondary product involvements which have never before been
addressable. The method also succeeded in providing a mechanism by which nearly all of
the important properties of a product are accounted for in the context of its end use.
Properties of products such as ignitability, flame spread, burning rate, smoke production,
toxic potency, critical flux for ignition and spread, and total combustible mass are explicitly
addressed as independent variables. Although not a focus of the project, the effect on
societal risk of such factors as market penetration or sales demographics could be examined
using the method.
Sensitivity studies performed as part of this case study showed that, for this case, the
method was insensitive to:
• occupant inputs relating to rescue by the fire department;

• the smoke height assumed to alert sleeping occupants where there is no working smoke
deteCtor;
• fire inputs relating to the assumed length of smoldering prior to initiating of flaming
combustion in the upholstered furniture; and
• the addition of additional oxygen at flashover through the breaking of windows by the
fire effects.
The methodology proved sensitive to changes in the assumed volume of the house and
to order of magnitude changes in toxicity of smoke. As a result of the work described herein,
improvements to the risk method were identified and implemented.
The exercise of the method on this case study revealed few areas where the state-of-the-
art in fire science was lacking. This was expected since this first case was selected to be
compatible with the scope of HAZARD 1. However, the case did identify many areas where
current data collection systems were lacking. In some instances, it may be possible to
supplement the data collected to fill these gaps. In others, special studies may be necessary
to attempt to capture the needed information. In still others, we may never be able to satisfy
the needs of the system. But, the identification of needs coupled with the potential value
of the method, should provide incentive for advances in these areas.
The second case demonstrated how the risk method can be used to evaluate a product,
which at present, does not contribute to life loss. The risk method provided a means to
examine a product whose involvement is almost entirely secondary, as secondary product
involvements have previously been a particular weakness of risk analysis methods. The fire
model succeeded in providing a mechanism by which the important properties of a product
are accounted for in the context of its end use. Properties of the product such as ignitability,
flame spread, burning rate, smoke production, toxic potency, critical flux for ignition and
spread, and total combustible mass are explicitly addressed as independent variables. In the
specific case of office carpet, the model was able to quantify the impact of critical flux and
flame spread rate in terms of the quantity of carpet ultimately involved and the consequences
of relaxing existing regulations on the fire risk of the product.
The deaths reported in the statistics for office occupancies rarely occur during business
hours, and carpet is never the first item ignited. These are rare occurrences involving
individuals performing cleaning, maintenance, or in the act of committing arson. This analysis
did include the scenarios which resulted in the maximum exposure to occupants in office
fires. The results obtained in this case are consistent with fire experience in that office fires
in general, and in particular, (fires which involve floor coverings with a reasonable resistance
to ignition and spread), almost never result in fatalities.
The exercise of the method on this case study revealed a few areas where some enhance-
ments to the hazard method were necessary. This was expected since this case was outside
the scope of HAZARD 1. The case identified areas where current data collection systems
were lacking (for example, information on office arrangements, construction, and worker
characteristics, particularly handicapped workers). In some instances, it may be possible to
to capture the needed information. In still others, we may never be able to satisfy the needs
of the system. But the identification of needs coupled with the potential value of the method
should provide incentive for advances in these areas.
The third case demonstrated how the risk method might be modified to handle a product
located within a structural element of the building. The risk method provided a means to
examine a product whose involvement can be both as ignitor and as fuel, either as the first
item ignited or secondarily ignited. And the fire model was enhanced to provide a mechanism
by which an important interaction with the product's environment (oxygen vitiation effects
on burning rate and species yields) were accounted for. Properties of the product such as
ignitability, flame spread, burning rate, smoke production, toxic potency, and total com-
bustible mass are explicitly addressed as independent variables. In addition, the case dem-
onstrated that the method was capable of producing a reasonable increase in deaths for a
doubling of the assumed flame spread rate for the cable. A factor of two difference in flame
spread rate is within the range of values observed for materials found in cable jacketing
materials in use today.
The evacuation simulation provided a realistic treatment of two very different occupant
groups. The guest room occupants had the demographic diversity of those in the residential
case, while the function room sets were generic like the office occupants. With both sets,
queuing was an issue that was addressed by the calculation.
The exercise of the method on this case study revealed a few areas where some enhance-
ments to the hazard method were necessary. This was expected since this case was outside
the scope of HAZARD 1. The case identified areas where current data collection systems
were lacking (for example, information on hotel arrangements, construction, and guest
characteristics, particularly handicapped workers). In some instances, it may be possible to
to attempt to capture the needed information. In still others, we may never be able to satisfy
the needs of the system. But the identification of needs, coupled with the potential value
of the method, should provide incentive for advances in these areas.
The fourth case represented a very different application of a quantitative method for the
estimation of the fire risk associated with a specified product class in a specified occupancy.
This case study tested the limits of the method with respect to addressing cases where data
on representative products, buildings, or occupants are not available. With all of this lacking,
the method collapses to a benchmark hazard analysis against which product changes can be
compared.
Like Case 2 (carpet), this case demonstrated how the method can be used to evaluate a
product, which at present, does not contribute significantly to life loss. The calculation
provided a means to examine a product whose involvement is largely secondary; as secondary
product involvements have never before been addressable in risk analysis of any type.
Through enhancements to the fire model, we succeeded in providing a prediction of flame
spread and fire development on a wall, which was qualitatively the same as was observed
in full-scale experiments conducted with these materials.
Because of the lack of required data, any risk assessment for interior finish in restaurants
must be made in the traditional way from the fire incidence data. What we can derive from
the hazard results obtained on the textile wallcoverings examined, is a quantification of the
flammability properties that represent the margin at which life safety is threatened in the
restaurant arrangements and patron characteristics considered. While there were no fatalities
predicted in the examined restaurant scenarios, fairly small additional delays in exiting,
faster fire development, or increased occupant susceptibility could result in a much different
outcome. The interrelationship of the product, building, and occupant, demonstrated in this
analysis, is something that cannot be derived from the incident data and has a value of its
own in better understanding fires of these types.
This case pointed out the crucial role of demographic and product use data in the success
of the method. In some cases, special studies may be necessary to attempt to capture the
needed information. In still others, we may never be able to satisfy the needs of the system.
But the identification of needs, coupled with the potential value of the method, should
provide incentive for advances in these areas.
Taken as a whole, we feel that this series of case studies has demonstrated a significant
potential for the risk method to provide highly detailed analyses of the risk (and cost) impact
of regulation and of the fire performance of products in our society. The potential benefits
to both the provision of public safety and the costs of such safety are enormous. But further
investments, particularly in the area of data collection, will be required. We hope that these
reports are sufficiently convincing to influence this investment.
References
[1] Clarke, F. B., Bukowski, R. W., Stiefel, S.W., Hall J. R., Jr., and Steele, S.A., "A Method to
Predict Fire Risk: The Report of the National Fire Protection Research Foundation Risk Project,"
NFPRF, Quincy, MA, 1990.
[2] Bukowski, R. W., Peacock, R. D., Jones, W. W., and Forney, C. L., HAZARD I Fire Hazard
Assessment Method (three volumes), NlST Handbook 146, National Institute of Standards and
Technology, Gaithersburg, MD, 1989.
[3] NFIRS Data Tapes, National Technical Information Service, Springfield, VA, 1981-1985.
[4] Karter M. J., Jr., "Fire Loss in the United States," annual publication in the Sept. issue of Fire
Journal, National Fire Protection Association, Quincy, MA.
[5] Stiefel, S. W., Bukowski, R. W., Hall, J. R., Jr., and Clarke, F. B., "Fire Risk Assessment
Method: Case Study 1, Upholstered Furniture in Residences," NISTIR 90-4243, National Fire
Protection Research Foundation, Quincy, MA, 1990.
[6] Stiefel, S. W., Bukowski, R. W., Hall, J. R., Jr., and Clarke, F. B., "Fire Risk Assessment
Method: Case Study 2, Carpet in Offices," NISTIR 90-4244, National Fire Protection Research
Foundation, Quincy, MA, 1990.
[7] Stiefel, S. W., Bukowski, R. W., Hall J. R., Jr., and Clarke, F. B., "Fire Risk Assessment
Method: Case Study 3, Concealed Combustibles in Hotels," NISTIR 90-4245, National Fire Pro-
tection Research Foundation, Quincy, MA, 1990.
[8] Bukowski, R. W., Jones, W. W., Hall, J.R., Jr., and Clarke, F. B., "Fire Risk Assessment
Method: Case Study 4, Interior Finish in Restaurants," NISTIR 90-4246, National Fire Protection
Research Foundation, Quincy, MA, 1990.
[9] Fire International, "The Role of Upholstered Furniture in U.S. Fires," Vol. 63, 1979, pp. 32-33.
(10] Babrauskas, V. and Krasny, J., "Fire Behavior of Upholstered Furniture," NBS Monograph 173,
National Institute of Standards and Technology, Gaithersburg, MD, 1985.
[11] Jones, W. W. and Peacock, R. D., Technical Reference Guidefor FAST Version 18, NISTTechnical
Note 1262, National Institute of Standards and Technology, Gaithersburg, MD, 1989.
[12] Grand, A. F., Kaplan, H. L., Beitel, J. J., Switzer, W. G., and Hartzell, G. E., "A Research
Study on the Potential Contribution of Carpets and Rugs to Toxic Emission Hazards in Building
Fires," Final Report 01-7369, Southwest Research Institute, San Antonio, TX, 1984.
[13] Bukowski, R. W., Peacock, R. D., Jones, W. W., and Forney, C. L., Technical Reference Guide
for the HAZARD I Fire Hazard Assessment Method, NIST Handbook 146, Volume II, National
Institute of Standards and Technolgy, Gaithersburg, MD, 1989, Chpt. 7.
[14] Bukowski, R. W., Peacock, R. D., Jones, W. W., and Forney, C. L., Technical Reference Guide
for the HAZARD I Fire Hazard Assessment Method, NIST Handbook 146, Volume II, National
Institute of Standards and Technology, Gaithersburg, MD, 1989, Chpt. 8.
[15] Kulkarni, A. K., "Upward Flame Spread on Vertical Walls," NIST-GCR-89-565, National Institute
of Standards and Technology, Gaithersburg, MD, 1989.
[16] Dietenberger, M. A., "A Validated Furniture Fire Model with FAST (HEMFAST)," NIST-GCR-
89-564, National Institute of Standards and Technology, Gaithersburg MD, 1989.
[17] Babrauskas, V., "Development of the Cone Calorimeter- A Bench Scale Heat Release Apparatus
Based on Oxygen Consumption," Fire and Materials, Vol. 8, No.2, 1984, pp. 81-85.
[18] Harkleroad, M. F., "Fire Properties Database for Textile Wall Coverings," NISTIR 89-4065,
National Institute of Standards and Technology, Gaithersburg, MD, 1989.
Judith Steciak1 and Robert G. Zalosh2
A Reliability Methodology Applied to Halon

1301 Extinguishing Systems in Computer
Rooms
REFERENCE: Steciak, J. and Zalosh, R. G., "A Reliability Methodology Applied to Halon
l30t Extinguishing Systems in Computer Rooms," Fire Hazard and Fire Risk Assessment,
ABSTRACT: This fault tree analysis of a total flooding Halon 1301 extinguishing system in a
computer room illustrates a methodology developed to estimate the probability that vaporizing
liquid suppression systems may fail to protect against fire or smoke damage. The methodology
follows four steps: 1) specify a prototype system; 2) identify failure scenarios; 3) determine
the probability of occurrence of the failure scenarios; and 4) perform uncertainty calculations.
The prototype used here is a total flooding Halon 1301 system installed in a computer room
in 1977. Three fire scenarios are analyzed: (1) a fire of electrical origin inside an equipment
cabinet; (2) a paper trash fire; and (3) smoke entering from a fire outside the room (this last
scenario is a measure of the detection, control system, and smoke damper effectiveness).
The results show that, of the three scenarios, the halon system is most effective against the
fire of electrical origin providing the electrical ignition source is de-energized before the halon
supply is depleted. The results also show that the system provides reliable protection against
smoke damage from a fire outside the room, but is less effective against the paper trash fire
because of the reignition hazard if the supply is depleted. Other analyses performed include
the effect of different inspection intervals, human intervention, and acceptance tests on overall
system reliability.
KEY WORDS: reliability of fire protection systems, Halon 1301, computer room fire protec-
tion, probabilistic methods
Objective
The objective of this research was to develop a methodology for quantifying the reliability
of site-specific vaporizing liquid agent suppression systems (1]. The approach is illustrated
by its application to a total flooding Halon 1301 system installed in a specific computer room
that is challenged by three different fire situations (an electrical cable fire within a computer
cabinet, a paper trash fire, and smoke entering through ventilation ducts from a fire outside
the room).
The methodology systematically identifies contributors to system failure, quantifies a
specific system failure frequency, and suggests methods of reducing the failure frequency.
System failure is defined as failure to protect against significant fire or near-term smoke
damage. The specific quantitative damage criteria depend on the fire scenario.
The analysis pivots around the site and fire scenario details, a fault tree logic model,
failure frequency data, and engineering calculations.
'Advanced research scientist, Factory Mutual Research Corporation, 1151 Boston-Providence Turn-
pike, Norwood, MA 02062.
2Professor, Worcester Polytechnic Institute, Center for Fire Safety Studies, Worcester, MA 01609.
161
Methodology
The methodology consists of four basic steps: (1) specify and describe a prototype central
supply halon system installed in a computer room; (2) identify failure scenarios; (3) determine
the probability of occurrence of the failure scenarios; and (4) construct sensitivity calculations
to determine how the results vary with key assumptions and with alternative operating
configurations.
The first step involved reviewing installation standards, visiting halon protected computer
rooms, and selecting a prototype installation with well-defined equipment and operating
procedures (Fig. 1). Discussions with system designers and installers, computer manufac-
turers, and operators helped to clarify the halon system operation for this application.
The second step involved a review of computer room loss incident data [2-4] and an
analysis of equipment damage thresholds considering thermal and nonthermal damage modes.
Three fire scenarios were selected as a result of this step.
The third step was accomplished by using fault trees to represent the failure logic of the
system. A general fault tree was constructed by identifying basic events that cause the top
event, failure of the system to protect against fire damage, to occur. Considerable data
STECIAK AND ZALOSH ON HALON 1301 EXTINGUISHING SYSTEMS 163
analysis and some engineering calculations were used to quantify the failure probabilities
of the basic events. The general fault tree (Figs. 2-8) was customized for each scenario by
changing the probabilities of appropriate basic events.
The data used for the quantification of the tree included test data for extinguishment
concentrations, full discharge acceptance test statistics, smoke detector response times,
equipment damage thresholds, human error statistics, and published hardware failure prob-
abilities. References are reported here under Quantitative Analysis.
Quantification of the tree gives the probability of the system to fail to protect against fire
damage and the minimal cut sets3 of bottom events and system components that are major
contributors to system failure.
The fourth step was to repeat the fault tree calculations with different probabilities of the
basic events so as to account for different operating conditions and uncertainties in the
component failure rate data.
3Minimal cut sets are groups of basic events wherein each event must occur for the top event to
occur.
System Description
The prototype system chosen for quantification is a central storage total flooding system.
It was installed in the 758 m3 computer room (volume of room, 522 m3, subfloor, 48 m3,
and ceiling plenum, 188 m3) in 1977 and was subjected to a full discharge test at that time.
The volume of the ceiling plenum was not included in the design calculations to determine
the amount of halon needed. The relevant standard at the time of installation was the 1977
edition of the National Fire Protection Association (NFPA) Standard on Halon Fire Extin-
guishing Systems (12A).
The ionization-type point smoke detectors are cross-zoned in accordance with the NFP A
Standard on Automatic Fire Detectors (72E). A detector in two different zones must alarm
for the system to actuate. All alarms are relayed to a constantly monitored off-site central
station. No delay is programmed between the second alarm and agent release. An abort
switch is mounted near the main exit.
The system is inspected annually. The inspection procedure includes cleaning the ionization-
type point smoke detectors, weighing the storage cylinders, exercising each logic option of
the control panel, and testing the backup batteries. Since installation, the computer room
has undergone several renovations. New cable runs have been made to accommodate more
users on the time shared system; this necessitated enlarging existing cable penetrations and
creating new ones.
The room ventilation system circulates air at a rate of 50 air exchanges per h. Air is drawn
into the ceiling plenum, passes through the air conditioning units, and is released into the
subfloor .
Computer operators are present for three 8-h shifts each day.
Scenarios
The fire scenarios chosen for quantification are: (1) a fire of electrical origin involving
flame spread along vertically oriented electrical cables in a computer cabinet; (2) a paper
trash fire started by careless disposal of smoking material; and (3) smoke entering the
computer room through ventilation ducts from a fire outside the room.
The cable in the computer cabinet is assumed to be electrically ignited by a power supply
fault so that the fire propagates up the cable and cable harness as in an actual fire. These
are discussed in Ref 4 and the tests described in Ref 5. Depending on the cable flammability,
self-sustained flame propagation may require sustained or repeated electrical shorting/arcing
for up to ten minutes. We assume that the electrical ignition source is removed after flame
propagation is established. No detection or suppression systems are installed in the cabinet.
The trash fire is assumed to be ignited by a cigarette and subsequently become deep
seated as occurred in fire tests described in Ref 6.
Although the halon system is not designed to provide suppression for the last scenario,
it tests the reliability of the ventilation smoke dampers, control system, and smoke detectors.
Halon systems are designed to isolate the protected room upon smoke detection and this
scenario is relevant to these detection and control features.
All three scenarios have, in fact, occurred in reported computer room fires. Table 1
provides a compilation of the relative frequency of these and other types of fires. Typical
designs of computer rooms can be found in Ref 5 or in Ref 6.
An important consideration in the fire scenarios is the level of damage at which halon
system failure is decreed, and at what point after ignition this damage will occur. In the
electrical fire, the time to damage was defined as incipient fire and smoke spread to adjacent
equipment from the cabinet of fire origin. We estimated that this would occur between 3
and 10 min after ignition and used these values as the 5th and 95th percentiles of a lognormal
distribution of time to damage for this scenario. This estimate was supported by fire test
data [5] that showed the cabinet of fire origin could be damaged in two min. Adjacent
equipment would be subject to smoke and fire shortly thereafter as suggested by the fire
tests in Ref 6. These tests showed that, within 2 to 3 min, the temperature of surfaces near
the cabinet of fire origin approached or exceeded the NFP A Standard on Electronic Com-
puter/Data Processing Equipment (75) levels at which damage to equipment, disk media,
and so forth, may begin (the fires in Ref 6 were suppressed before the damaging temperatures
were sustained).
In the paper trash fire, the time to unprotected hazardous conditions was defined as the
time for the fire to become deep seated and hence difficult to extinguish. The hazard in this
situation is that the fire may reignite after the halon supply is depleted, and the reignited
fire could cause significant thermal damage before it is extinguished manually. Preburn times
reported during extinguishment tests of paper fires [6,10-]4] were used to establish a
distribution of time for the fire to become deep seated (lognormally distributed with 5th
and 95th percentiles of 43 and 579 s, respectively).
The damage times were used in a risk analysis of the competition between the time to
system actuation versus the time to damage []5].
In the last scenario, damage is characterized by the probability that the smoke concen-
tration exceeds the smoke damage threshold of the electronic equipment before the detectors
can actuate. We judged that the smoke damage threshold varied uniformly between 0.0005
and 0.5 g/m3• The lower bound of this estimate is supported by particulate limits recom-
mended by a computer manufacturer for an industrial computer in normal operation []6].
The Underwriter's Laboratories Standard for Smoke Detectors, UL 268, lists sensitivity
thresholds between 0.016 and 0.345 m -I (turbidity or optical density per meter on a natural
logarithmic scale). The optical density was expressed as smoke particulate concentration
using the following relationship from Newman and Steciak [17].
The first two scenarios also include the probability of smoke damage before detection,
that is, the likelihood of damage is the probability that either thermal or smoke damage
occur before automatic detection and suppression.
Quantitative Analysis
The quantification of the basic events, including uncertainty bounds, is summarized in
Table 2.
Extinguishment Data
Extinguishment data were collected for combustibles found in computer rooms (for ex-
ample, surface and deep seated paper fires [6,10-14], polyester magnetic tape [13,6], and
PE/PYC coated electrical cables [6,12,22,23]). Although sparse, the data indicates trends
that were quantified with Bayesian analysis [24] to obtain cumulative distribution functions
for failure to extinguish as a function of Halon 1301 concentration (Fig. 9). These distribution
functions are representative of materials used in references through 1986. They mayor may
not be appropriate for newer materials used in computer rooms.
Ful! Discharge Test Data

The results from nearly 1900 full discharge tests were compiled [25-29]. Specific system
failure modes and their frequency of occurrence were quantified (for example, inadequate
halon supply, leaks, obstructed nozzle, unsealed penetrations, and incomplete trip). These
data were used to assess the impact of different acceptance tests, full discharge and/or leakage
(room pressurization) tests, on system reliability.
Smoke Detector Response Data

Estimates of smoke detector response times from tests in highly ventilated compartments
[6,30-35] were used to develop probabilistic detector response times (lognormally distrib-
uted with 5th and 95th percentiles of 7.6 and 140 s, respectively). Detection time was used
in the competing risk analysis of time to actuation versus time to damage.
Human Response
Human response data are needed for events that describe manual detection, activation,
suppression, and system abort. The Technique for Human Error Rate Prediction (THERP)
and a Time Reliability Correlation (TRC) were used to estimate human response [18]. The
method used is noted in Table 2.
The THERP method was used to quantify events that were failures to follow established
policies or procedures; the estimate is an occurrence frequency.
The TRC method returns the probability of success (or failure) as a function of response
time. The response times were used in a risk analysis of the competition between the time
to suppression versus the time to damage.
Hardware Failures
Published reliability data were used to quantify the failure probabilities of smoke detectors,
the control panel, backup and main power supplies, and wiring. The data sources are noted
in Table 2. The failure rates and probabilities are based on data for unsupervised control
and detection systems. Supervision was not required in the NFPA 12A standard in 1977
when this system was installed.
Results
The fault tree was quantified for each fire scenario. The impact of inspection frequency
(quarterly, annually, or once every five years) was calculated. The fault tree was used as a
tool to ascertain the impact of different types of acceptance tests (full discharge versus room
integrity tests after renovations or a combination of the two). The fault tree was also used
to quantify the impact of human actions on system reliability.
The probability of failure of the system to protect against fire damage for these different
circumstances is summarized in Table 3. The results of the analysis reflect the uncertainty
of the data available for quantification, as indicated in the confidence bounds cited below.
The mean probability of failure of the halon system to protect against fire damage for
the three scenarios is 0.05 (with 5th and 95th percentiles of 0.04 and 0.08) for the electrical
fire, 0.13 (with 5th and 95th percentiles of 0.08 and 0.24) for the paper trash fire, and 0.08
(with 5th and 95th percentiles of 0.05 and 0.13) for the fire outside the room, based on the
damage criteria defined above. These failure probabilities are for a system inspected annually
that underwent a full discharge acceptance test when it was installed in 1977, but was not
retested after renovations.
As indicated in Table 3, the mean probability of system failure is reduced slightly by
conducting quarterly inspections, since covert hardware failures are detected and repaired
more frequently.
The mean probability of system failure increases significantly when inspections are con-
ducted only once every five years. Hardware failures have the largest contribution to system
failure for all three scenarios because of the increased equipment failure probability with a
five year inspection interval.
If an older system did not undergo a full discharge acceptance test, but did undergo room
leakage tests after renovations, then, (1) for fires within the room, improvement in room
integrity has a positive impact on the likelihood of protection against the paper trash fire
and a less dramatic one against the electrical cable fire (this is due to the relative ease of
suppressing electrical cable fires at below design halon concentrations, if the electrical
ignition source is not sustained); and (2) for fires outside the room, improvements in room
integrity improves the mean likelihood of protection due to the improved reliability of
dampers preventing smoke from entering.
In this analysis, the actions of people contribute significantly to the system's success. The
mean probability of system failure increases substantially when no human actions contribute
to detection, activation, or suppression. These results also underscore the importance of
alarm activation at a constantly attended central station in another building, and the prompt
investigation of all such alarms.
The probability of system failure could be reduced by changing the characteristics of the
important contributors to system failure, including, (1) improving the reliability of wiring
and the control panel (for example, by using supervised circuits as is required in the current
NFPA 12A Standard); (2) increasing the speed of automatic detection (while not sacrificing
accurate detection); (3) affirming room integrity after renovations; (4) removing the abort
switch or reducing the likelihood of people using it inappropriately; and (5) improving the
reliability of dampers and the ventilation system interlock to prevent smoke from entering
via this path from a fire outside the room. The relative contribution of these events to
system failure can be estimated from the minimal cut sets listed in Tables 4 to 6. These cut
sets are for a system inspected annually that underwent a full discharge acceptance test when
it was installed in 1977, but was not retested after renovations.
Other vaporizing liquid agent suppression systems (for example, carbon dioxide or can-
didate replacement agents for the halons) can be analyzed with this approach, if scenario
specific extinguishment concentration data are available and if the control and agent delivery
subsystems are similar to those for a halon central supply system. Quantification of other
suppression system installations and fire scenarios may have very different results from the
examples presented here, but the basic methodology that we described above still holds.
Acknowledgments
The authors appreciate the helpful suggestions, data, and comments received from Walter
F. Maybee (retired from the U.S. Department of Energy); Nathan O. Siu, Massachusetts
Institute of Technology; and Dimitrios M. Karydas, Robert C. Merritt, and David O.
Richardson of Factory Mutual Research Corporation.
This research was conducted while Professor Zalosh was Assistant Vice President and
Manager of the Applied Research Department at Factory Mutual Research Corporation.
References
[1] Steciak, J. and Zalosh, R. G., "Fault Tree Analysis of Halon 1301 Extinguishing Systems in
Computer Rooms," Factory Mutual Research Corporation Job Identification ONIJ7.RU/OQ4E5.RU,
Factory Mutual Research Corporation, Norwood, MA, Nov. 1989.
[2] Taylor, K. T., "Special Report on Computer and Data Processing Centers: 1981-1985 Fire Ex-
perience," National Fire Protection Association, Fire Analysis Division, Batterymarch Park, Quincy,
MA, May 1988.
[3] U.S. Department of Energy, "Standard for the Protection of DoE Electronic ComputerlData
Processing Systems, DOE/EP-0108 (formerly Wash 1245-1)," Office of Operational Safety, Wash-
ington, D.C., Jan. 1984.
[4] U.S. Department of Commerce, "Standard Practice for the Fire Protection of Essential Electronic
Equipment Operations," RP-l, National Fire Prevention and Control Administration, Washington,
D.C., Aug. 1978.
[5] Kainz, C. and HofIing, B., "Fire and Extinguishing Tests on Computer Equipment," Allianz Loss
Control Service, Munich, Germany, 1988.
[6] Reichelt, E. F., Walker, J., Vickers, R. N., and Kwan, A. J., "Report of Test Results: Halon
1301 vs. Water Sprinkler Fire Protection for Essential Electronic Equipment," Air Force Engi-
neering & Services Center, Engineering & Services Laboratory, Tyndall Air Force Base, FL, July
1982.
[7] Loss Report 87-1061, Factory Mutual Research Corporation, Norwood, MA, 1987.
[8] Loss Report 88-026, Factory Mutual Research Corporation. Norwood, MA, 1988.
[9] Reference Standard for Electrical Wires, Cables and Flexible Cords., UL 1581 1st Edition. Section
1160. Vertical Tray Flame Test (periodically updated) Aug. 1983. Underwriters' Laboratories,
Inc., Northbrook, IL.
[10] Abbott, J. L., "Model A-1301 Series Total Flooding Fixed Extinguishing Systems Using Ansul
1301." Factory Mutual Research Corporation Serial Number 19612, Factory Mutual Research
Corporation. Norwood, MA. Prepared for the Ansul Company. Marinette. WI. March. 1971.
[11] Carey, W. M., "Fact Finding Report on Extinguishment of Class A and B Fires in Electronic
Computer Rooms with Halon 1301," File NC535, Underwriters' Laboratories, Inc., Chicago, IL,
Jan. 1972.
(12] Cholin, R. R., "How Deep is Deep? Use of Halon BOlon Deep-Seated Fires," Fire Journal,
Vol. 66, No.2, March 1972.
[13] Ford, C. L., "Halon 1301 Computer Fire Test Program, Interim Report," Contributions from the
Ansul Company, Card ox (Division of Chemetron Corporation), E. I. du Pont de Nemours &
Company, Inc., and Fenwal, Inc., prepared by E. I. du Pont de Nemours & Company, Inc.,
'Freon' Products Division, Wilmington, DE, Jan. 1972.
[14] Miller, M. J., "Extinguishment of Charcoal, Wood and Paper Fires by Total Flooding with 'Freon'
FE 130l/Air and Carbon Dioxide/Air Mixtures," Factory Mutual Research Corporation Serial
Number 16234.1, Presented for E. I. du Pont de Nemours & Company, Inc., Factory Mutual
Research Corporation, Norwood, MA, July 1967.
[15] Siu, N. O. and Apostolakis, G., "A Methodology for Analyzing the Detection and Suppression
Fires in Nuclear Power Plants," Nuclear Science and Engineering, Vol. 9A, 1986, pp. 213-226.
[16] "IBM 7532 Industrial Computer," 6523293-1, International Business Machines, Manufacturing
Systems Products, Boca Raton, FL, 1986.
[17] Newman, J. S. and Steciak, J., "Characterization of Particulates from Diffusion Flames," Com-
bustion and Flame, Vol. 67, 1987, pp. 55-64.
(18] Dougherty, E. M. and Fragola, J. R., "Human Reliability Analysis: A Systems Engineering Ap-
proach with Nuclear Plant Applications," John Wiley and Sons, New York, 1988.
[19] Karydas, D. M., "Probabilistic Analysis for Low Water Level Conditions in Boilers," Factory
Mutual Research Corporation Job Identification ON1E3.RU, Factory Mutual Research Corpo-
ration, Norwood, MA, Dec. 1987.
[20] U.S. Atomic Energy Commission, "Reactor Safety Study. An Assessment of Accidental Risks in
U.S. Commercial Nuclear Power Plants. Appendix III, Failure Data," WASH 1400, Washington,
D.C. 1974.
[21] Rickers, H. c., "Residential Smoke Detector Reliability Handbook, Section 2: System Reliability
Evaluation Techniques; Section 3: Component Screening Test Techniques," NBS-GCR-79-162,
Reliability Analysis Center, Griffiss Air Force Base, New York, March, 1979.
[22] Chavez, J. M. and Lambert, L. D., "Evaluation of Suppression Methods for Electrical Cable
Fires," NUREG/CR-3656, Sandia National Laboratories, Albuquerque, NM, Oct. 1986.
[23] "Extinguishment of Distribution Frame Fires," PSR-375, Fenwal Inc., Ashland, MA, Jan. 1970.
[24] Martz, H. F. and Waller, R. A., Bayesian Reliability Analysis, John Wiley and Sons, New York,
1982.
[25] "ANI Tabulation of Total Flooding Gaseous Suppression Acceptance Test Results," American
Nuclear Insurers, Farmington, CT, 1980.
[26] "ANI Reliability Survey of Fire Suppression Systems in Nuclear Power Facilities," American
Nuclear Insurers, Farmington, CT, 1979 and 1981.
[27] "Halon, CO2 Tests," Interoffice Correspondence, Factory Mutual International, Melbourne, Aus-
tralia, Feb. 1986.
[28] "New Installation Test Survey Results," Fire Suppression Systems Association, New Update,
Baltimore, MD, April 1987.
[29] "Conducted Your Halon 1301 Acceptance Test Yet?," The Sentinel, Industrial Risk Insurers,
Hartford, CT, Third Quarter, 1985.
[30] "The Report of the Performance of the FIRETEK 2160 Detector in the Computer Room Fire
Tests Conducted in Conjunction with the Ford Motor Company on July 3 and 4, 1983," Technical
Report 10, FIRETEK Design Group, FIRETEK Corporation, Hawthorne, NJ, Nov. 1983.
[31] Healey, W. E., "Products of Combustion Smoke Testing; Ford Motor Company Research &
Engineering Center Tests Conducted July 4, 1983," FIle 12569A, W. E. Healey & Associates,
Inc., Bloomfield Hills, MI, Sept. 1983.
[32] "Report of Tests Conducted at Ford Motor Company, 4 July 1983," Pyrotronics, Livonia, MI,
1983.
[33] "Detector Response Tests," Advanced Safety Systems, Inc., Woburn, MA, May 1975.
[34] Costello, N. F. and Wickham, R. T., "Halon 1301 Computer Fire Test Program (Cellulosic
Materials)," The Ansul Company, Marinette, WI, Aug. 1971.
[35] Newman, J., "Fire Tests in Ventilated Rooms: Detection of Cable Tray and Exposure Fires,"
EPRI NP-2751, Electrical Power Research Institute, Palo Alto, CA, Feb. 1983.
Thomas F. Barry I
Application of Fire and Explosion Risk

Assessment to an LPG Bulk Storage Facility
REFERENCE: Barry, T. F., "Application of Fire and Explosion Risk Assessment to an LPG
Bulk Storage Facility," Fire Hazard and Fire Risk Assessment, ASTM STP ll50, Marcelo M.
Hirschler, Ed., American Society for Testing and Materials, Philadelphia, 1992, pp. 183- 206.
ABSTRACT: This paper contains an analytical examination of risk to human life beyond the
boundaries of a hypothetical plant site, labeled the XYZ Plant, which stores and handles
liquefied petroleum gas (LPG). The focus is the integration of available risk assessment re-
search and techniques into a structured decision support tool. Procedures presented in this
paper can be applied to a wide array of hazardous storage versus off-site risk exposure projects.
Quantitative risk assessment procedures are applied in determining fire and explosion effect
zones and fatality probabilities in the area directly north of the XYZ Plant. The risk is assessed
assuming that all four 113 562 L LPG tanks are updated with specified high reliability emer-
gency shutdown systems and fire protection systems.
Event tree methodology is used in developing the fire and explosion scenarios which could
result from the accidental release of LPG from containment.
Risk assessment tasks include the selection of significant LPG failure modes, calculation of
release rates, and consequence modeling of human fatality effect zones. Boiling liquid ex-
panding vapor explosion (BLEVE), flash fire radiant heat exposure, and explosion overpres-
sures from unconfined vapor cloud explosions are modeled. Probability modeling is conducted
by structuring event trees and assigning initiating event frequencies and conditional proba-
bilities to the defined loss events.
Results are presented in an individual risk profile which depicts the probability of human
death versus distance from the LPG storage tanks. The individual risk profile is suitable for
use by the local Planning Commission as one source of information for making informed
decisions concerning future development in the area north of the XYZ Plant.
KEY WORDS: liquefied petroleum gas (LPG), fires, explosions, boiling liquid expanding
vapor explosion (BLEVE), failure modes, fire prevention, fire protection, risk assessment,
individual risk
Risk assessment consists of both identifying risks and estimating the probabilistic likelihood
and magnitude of potential loss. During the past ten years, there has been an increase in
the use of formal risk assessment methods to support decisions on industrial fire safety
issues. Predictive methods which integrate statistical data, deterministic models, and expert
opinion are being increasingly used in industries involved in the processing, storage, and
liquification transportation of hazardous materials.
The XYZ Plant is an air liquefication facility producing liquid oxygen, nitrogen, and
argon. Both bulk storage product tanks and cylinder filling operations are located at this
site. This plant operates continuously. There is a constantly manned control room and, at
a minimum, one instrumentation technician and one maintenance person always on site.
The XYZ Plant is also involved in the bulk storage of liquefied petroleum gas (LPG) in
four 113 562 L pressurized tanks and owns two LPG tank trucks. Trucks transport LPG to
customer locations where small LPG tanks and cylinders are filled from the trucks.
'Senior fire protection engineer, Professional Loss Control, Inc., PO Box 446, Oak Ridge, TN 37831.
183
Presently, there are no high level automatic alarms or automatic shutdown features for
the XYZ Plant LPG tanks. There are only visual level and pressure gauges on each tank
and manual valving.
There is no gas detection or flame detection in the LPG tank storage area. However, the
XYZ Plant utilizes both a continuous gas detection system and UV-IR flame detection in
areas storing and handling hydrogen and in the cylinder filling area. All detection alarms
are monitored at the constantly attended control room.
There is no fireproofing or fixed automatic water spray system on the LPG tanks. Present
procedure during a fire emergency would be notification of the primarily volunteer public
fire department by phone or actuation of a manual pull box which is located near the entrance
of the plant.
The local Planning Commission is concerned with LPG storage tanks in areas which are
becoming heavily populated. The plant owners are concerned with future liability, public
image, and safe operation; therefore a risk assessment study was funded.
XYZ Plant Risk Assessment Approach

The goal of the hypothetical XYZ plant site risk assessment is to apply state-of-the-art
probabilistic and deterministic modeling techniques to assess the risk of off-site human
fatality. Event tree methodology is applied to generate the probabilities of representative
fire and explosion scenarios. The representative set of incidents selected for this study
includes catastrophic failures and consequences, and does not include localized incidents,
which would cause only on-site property damage and casualties.
This risk assessment process includes:
• Hazard identification
• Representative incident selection and event tree structuring
• Fire and explosion modeling
• Probability estimating
• Individual risk calculation
The results of this assessment are presented in terms of an individual risk profile. The
developed risk profile graphically depicts the estimated probability of human death versus
distance from the XYZ Plant.
Because of the unavailability of plant specific failure probability data or actuarial failure
data for identical equipment, event probabilities are addressed by researching data sources
and compiling and extracting historical failure probability data which represent the equip-
ment design and operating environment of the XYZ Plant's LPG tank operation as closely
as possible. The failure probabilities for tanks, piping, and hoses for flammable liquid and
gas storage operations which were referenced in the Rijmond study [1], Canvey Study [2],
and American Institute of Chemical Engineers (AIChE) data sources [3], were selected as
relevant data for this project.
The majority of data in these references is generic data accumulated and aggregated from
a wide variety of plants and industries (U.S. and U.K.) and nuclear power plants (U.S. and
foreign). The data include equipment of numerous manufacturers with varied operating
environments and maintenance practices. Therefore, the risk analyst must have an under-
standing of the origin and limitation of generic actuarial-based failure data for proper ap-
plication. The uncertainties of data selection can be reduced by learning as much as possible
about data origins; including the type, design, construction of the equipment, the process
medium, and plant inspection and maintenance programs. These precautions in the use and
BARRY ON RISK ASSESSMENT TO AN LPG BULK STORAGE FACILITY 185
selection of failure probabilities are managed by establish-ing a systematic methodology for

direct selection of historical event probabilities or estimation of event probabilities from
within generic failure ranges.
The methodology used in addressing probability estimation in this project involved the
following methods.
Reviewing Data Sources

Relevant data sources were researched and compiled.
Checking Data Applicability

Careful review of source data was conducted to determine specific applicability to LPG
operations.
Estimating Event Probability Directly

If data are judged consistent with the XYZ Plant LPG tank operation, event probabilities
were applied directly from the historical data. This method was applied to LPG piping and
aggregate failure probabilities, and for the majority of Emergency Shutdown System (ESS)
component reliabilities.
Estimating Event Probability Using A Risk Scoring Model

A risk scoring model provides an explicit, rational, and consistent method for selecting
a probability within data ranges which have been established for similar systems. An example
of this is the LPG pressurized tank failure probabilities where upper and lower bounds were
established from data sources for similar pressurized vessels.
The risk scoring model allows engineering examination of factors which contribute to the
upper and lower frequency bounds. Evaluation and selection are based on knowledge of
where the data is extracted from, and the contributing quality factors which include design,
inspection, maintenance, operation, and safety standards.
Estimating Event Probability Based on Engineering Judgement

Upper and lower probability bounds were selected from generic failure data and con-
servative engineering judgement was exercised in the selection process. An example of this
is the case of flexible hose failures, where a worst case conservative approach was applied
in selecting an upper boundary failure probability.
Another example is ignition probabilities. Because of the unavailability of actuarial-based
ignition frequencies, evaluation of ignition probabilities requires application of subjective
engineering judgement based upon systematic evaluation of:
• Potential ignition source type and distribution around the XYZ Plant's LPG tanks.
• The type and pathway of the LPG release which occurs.
• Estimation of the cumulative ignition probability based on a flammable LPG vapor
cloud engulfing individual ignition source potentials.
Engineering judgements needed to assign equipment and operational failures to occur-

rence frequencies or performance reliability follow the categorical descriptions in Table 1.
Engineering decisions were based upon failure for similar components and operations from
applicable data sources.
LPG Hazards
Off-site fire and explosion exposure to the public from large quantity LPG storage pri-
marily includes the effects of radiant heat from an LPG fireball resulting from a BLEVE
or LPG vapor cloud formation and delayed ignition resulting in explosion blast overpressures,
termed Unconfined Vapor Cloud Explosion (UVCE). Both of these events can be considered
large severity, rare occurrence events. However, past loss incidences of this nature have
been widely publicized, increasing the awareness and concern of both the public and land
use planning groups.
The major hazard presented by the BLEVE of an LPG tank is the fireball created by
combustion of the mixture of vapor and liquid that is explosively dispersed by the sudden
rupture of the tank. The heat created by the burning of the dispersed LPG causes a powerful
thermal updraft, which interacts with the burning LPG to create a constantly rising fireball.
The effect of radiant heat from the fireball on objects at grade level is difficult to predict
with accuracy; although the diameter and duration of the fireball can be calculated, the
distance between the fire and the object is constantly increasing. Therefore, the heat flux
to the object is not constant but decreases with time. In most cases, the area directly beneath
the fireball will be subjected to sufficient flame contact with the fireball in its early stages
(ground flash) to cause ignition of ordinary combustibles. Structural damage to noncom-
bustible objects would be expected to be minimal because the fireball lasts for only a few
seconds.
A continuous release of vapor with dispersal and delayed ignition can result in an UVCE.
Historical data have shown that for an explosion to occur, the cloud must be large (at least
4536 kg of hydrocarbon must be released), the release rate of vapor must be large (15 kg/
s or more) and a significant delay before ignition is required (greater than 30 s) [2].
Proposed Emergency Shutdown System and Fire Protection Improvements

The capability to isolate or shut down an accidental release of LPG affects the size of the
release and the degree of off-site exposure. The size of the release depends on the inventory
of material, and the design effectiveness and reliability of the emergency shutdown system
(ESS).
The system design objectives for the XYZ Plant are the ability to shut down upon detection
of an LPG gas release prior to the release of 4.54 metric tons of LPG and the ability to
shutdown within 3 min upon detection of flame at the tank, tank piping, or associated
loading equipment. Reports indicate that the shortest times to LPG tank BLEVE have
generally been in the 10 min range [4].
Based on the evaluation of significant failure modes, the primary design emphasis is to
isolate the tank and quickly shut down operations upon detection of an abnormal condition
(high level alarm) or gas leakage and/or flame.
The XYZ Plant design will follow the intent of NFP A 58, Standard for the Storage and
Handling of Liquefied Petroleum Gases [5], concerning the installation of emergency safety
shutoff valving. Emergency valve locations include the following:
• liquid fill line

• liquid unloading (discharge) line
• vapor return line
• pullaway valves at hose connections
Activation of the ESS system will perform the following functions:
1. Close all emergency shutoff valves (ESV's)

2. Shut down all pumps
3. Initiate the XYZ Plant local alarm system
Detection devices for automatically actuating the ESS system will include gas detectors
and UV-IR flame detectors similar to those currently in use at the plant. In addition, there
will be two manual emergency shutdown stations (one at the control room area and one in
the vicinity of the tanks). The conceptual ESS system design is shown in Fig. 1.
Estimating Probability of Successful Isolation/Shutdown

Equipment-related failure data summarized in Table 2 are primarily extracted from the
AIChE publication, Process Equipment Reliability Data Book [3]. This reliability data is
from failure data collected and screened by the Gas Research Institute. This data was
collected primarily from Liquefied Natural Gas plant operations which are similar to LPG
operations.
Probability of successful ESS actuation is based on the design expectation graph in Fig.
2, for detection and shutdown versus the time following release. The curves represent the
various detection methods which include gas detection, flame detection, and detection by
XYZ Plant personnel. The detection probabilities versus time are multiplied by equipment
shutdown success probabilities to estimate the overall probability of successful ESS actuation
versus time following an LPG release. For example, the curve labeled C is the probability
versus time profile for personnel (operator) detection and response to an accidental LPG
release. There is a manual pull box in the vicinity of the tank and the LPG is very visible
when released. Based on engineering judgment, it would be expected that 50% of the time
an operator (XYZ employee with average experience) would be able to initiate ESS within
3 min, 75% of the time within 6 min, and 90% of the time within 10 min.
The probability of successful ESS system operation from manual initiation by XYZ Plant
personnel is:
where
Pp = Probability of personnel detection of LPG release.
Px = Probability of successful pull-box (x) operation.
Pz = Probability of successful controller (z) operation.
Pv = Probability of successful emergency shutdown valve (v) operation.
Ps = Probability of system shutdown at the designated time.
It should be noted that the operator manual activation of the ESS is viewed as a backup
mechanism if the automatic detection system malfunctions or is impaired.
In addition to the ESS, proper design of the fire protection system is important. The
approach taken for the fire protection of the XYZ Plant LPG tanks is fireproofing ia
accordance with API Standard Design and Construction of Liquefied Petroleum Gas (LPG)
Installation (2510) and API Standard on Fire Protection Considerations for the Design and
Operation of Liquefied Petroleum Gas (LPG) Storage Facilities (251OA) [7,8]. A fixed,
manually operated monitor system will be installed as backup. This approach was based on:
1. Increased performance effectiveness of fireproofing in reducing the likelihood of a

BLEVE situation from a torch fire exposure. A torch fire exposure from severed piping
or hoses is a high probability event.
2. The cost of installing and maintaining fireproofing and monitors was e,stimated to be
less than the cost of installing a fixed automatic water spray system.
If there is a major LPG liquid release from ruptured piping, immediate ignition, and the
LPG system is not shut down, then there is a probability of a high velocity torch fire.
impinging on the LPG source tank or adjacent tank. The reliability of the fireproofing to
withstand direct flame impingement for 10 to 20 min until hose streams can be applied is
assigned a probability of 0.90, based on engineering judgement. This 0.90 probability is
supported by the Factory Mutual Research approval tests [9] specifically for LPG tank
insulations for two h rating under extremely high temperatures and hose stream integrity
tests. The 0.90 probability is further based on the proper installation of the fireproofing.
adequate inspection-maintenance, and activation of the water monitor system within 20 min.
For a torch-pool fire exposure where there is not direct flame impingement on LPG tanks,
fireproofing would prevent a BLEVE. A probability of 0.99 will be assigned to the success
of fireproofing to eliminate BLEVE conditions during this type of scenario. In addition.
reinforced concrete barrier walls will be installed perpendicular to the longitudinal axis at
the end of each LPG tank.
Consequence Modeling
The objectives of the consequence modeling approach applied in this section are:
• Selecting representative failure modes leading to critical fire/explosion events.

• Quantifying LPG release rates for the selected failure modes and assessing LPG vapor
dispersion characteristics.
• Calculating radiant heat and explosion overpressure intensities versus human fatality
threshold distances.
Based on the evaluation of historical loss incidents and failure modes utilized in past risk
studies involving liquefied gas storage operations [1,2], the following representative LPG
release modes were selected:
Major Tank Failure

This would involve major rupturing of the tank from over-pressure or mechanical damage
with near instantaneous release of the entire contents. This failure mode will be modeled
using a 203.2 mm release opening.
Fractured Fill or Discharge Piping

This would involve a complete rupture of the 152.4 mm fill or discharge piping with a
continuous release of LPG liquid which would start flashing into vapor following release
from containment.
Flexible Hose Rupture

This would involve a complete rupture of the flexible fill or discharge hose. This failure
will be modeled using a 101.6 mm release opening with a continuous release of LPG liquid
which would start flashing into vapor following release from containment.
Fractured LPG Vapor Balance Line

This would involve a complete rupture of the 50.8 mm vapor lines with a limited continuous
release of LPG vapor.
Aggregate 25.4-mm Continuous Release Failures

This category of release involves partial failures at various LPG system locations, which
include:
• Pipe connections (primarily flanges and valves)

• Instrument connections, relief valves (tank overfilling)
• Drain-sampling lines
• Small cracks on tank or piping
• Pump seals
This aggregate failure mode category will be modeled assuming an average 25.4 mm
release opening of LPG liquid which would start flashing into vapor following release from
containment.
This category of failure is evaluated because it represents a torch fire potential which, if
not controlled, could cause a tank BLEVE involving the entire tank contents.
The CHEMS-PLUS computer model (10] was used to expedite laborious LPG release
rate and dense cloud dispersion calculations. Because LPG does not present a quickly
dispersing toxic cloud exposure to off-site population, there is no need to run numerous
detailed calculations or cloud dispersion variables. Calculations were limited to a single
typical wind speed! stability class.
The results in Table 3 identify potential downwind travel distance for LPG vapor cloud
at approximately 10 min following the release.
Calculations are based on a steady 2.23 m!s wind speed, slightly unstable weather con-
ditions (C stability factor), and a flat unobstructed terrain.
Radiant Heat and Explosion Overpressure Effects
Potential fatality thresholds from radiant heat flux as a result of a BLEVE fireball incident
have been modeled by numerous researchers [11,12].
Table 4 summarizes the results of calculating the BLEVE fireball, radiation, and human
threshold distance limits at 1%, 50%, and 99% fatality levels. The empirical equations
described by Zalosh in the SFPE Handbook [13] were applied in modeling BLEVE effects
in terms of radiant heat versus distance.
The problem of quantifying injury effects and potential fatality levels from vapor cloud
explosion overpressures is complex. There is a considerable amount of literature which
suggests varying levels of injury and fatality probabilities associated with unconfined vapor
cloud explosions [2]. Sources such as AIChE [14] and the World Bank [15] suggest a value
of 1.034 bars as a point where lung damage may cause death. Based on review of this
literature, and the specific focus of this project, human threshold limits of 1.034 and 0.483
bars were both evaluated in calculating explosion exposure distances. A 0.483 bar limit was
used in assessing fatality threshold limits in the event trees.
An accepted unconfined vapor cloud explosion (UVCE) model (the TNT equivalency
method) as described by Zalosh in the SFPE Handbook [13] was applied in modeling UVCE
effects. A 5% explosion yield was used for estimating the explosion efficiency [10]. These
results are summarized in Table 5.
LPG vapor clouds may be ignited from a number of ignition sources, which are numerous
in industrial areas. Vapor clouds are generally ignited at the edges. The outer envelope of
the cloud mixes with air and reaches the lower flammability limit. As the cloud travels, the
likelihood of ignition keeps increasing. Therefore, based on evaluation of ignition potential,
engineering judgments must be applied in establishing credible ignition zones, and flash fire
limits, as ignition would prevent vapor clouds from reaching their full extent.
Based on LPG dispersion characteristics, release size, and engineering judgment about
credible ignition limits (distance and time), the following flash fire parameters will be applied:
For less than a 9072-kg release, a 198-m distance from the LPG tanks is considered a credible
flash fire limit with a wind blowing to the north. Without wind, flash fire limits will be
assumed at 22.9 m from the tanks, which is within the property lines of the XYZ Plant. For
greater than a 9072-kg release, a 91.4-m credible flash fire limit will be set with wind blowing
to the north, and without wind, a 91.4 m flash fire limit will be assumed. For all types of
flash fire it is assumed that any person within the region exposed by the flammable cloud
at the moment of ignition would be killed.
Fig. 3 is a graphical presentation of credible fire and explosion hazard contours or zones
from a BLEVE fireball, an unconfined vapor cloud explosion, and a flash fire. It should be
noted that tank fragmentation exposure to people located north of the plant site boundaries,
and the exposure from tank fragments initiating damage to adjacent tanks (domino effect),
are not included in the hazard zone analysis. This is based on the proposed installation of
reinforced concrete barrier walls at the LPG tank which would greatly minimize the off-site
risk exposure of tank fragments.
Probability Estimation
LPG Tank Failure Probability
Data from the Rijmond Study (1], Canvey Study [2], and AIChE data [3] on pressurized
vessel failures were used to establish upper and lower frequency bounds. A risk scoring
model (Fig. 4) was developed to provide an explicit, rational, and credible method for
selecting a probability value for the XYZ Plant LPG tanks.
The importance weightings and implied quality ratings are used to relate a score (weight
X quality rating) to one of the probability ranges. The importance weightings are assigned
via a Delphi engineering group. They are based on a review of historical causes of pressure
vessel failures, and engineering review of the contributing quality control factors associated
with tank failures.
Based on review of the failure range data, worksheet evaluation of contributing factors,
and XYZ Plant quality rating, a frequency of 8.3 x 10-6 major tank failure events, per
unit-year, was selected. The XYZ Plant has four tanks which equates to a major tank failure
probability of 3.3 x 10-5 per year.
Pipe Failure Probability

For piping failures of the I52A-mm diameter LPG liquid fill, discharge piping, and the
50.8-mm LPG vapor balance lines, probability data was applied directly from the Rijmond
pipe failure rates for liquefied propylene storage [1] which are failure probabilities per foot
of pipe. This failure data, used in the calculation of the XYZ Plant piping lengths, was
compared to pipe failure data compiled in the World Bank's Manual of Industrial Hazard
Techniques [15] and showed agreement at the upper frequency bounds, for failures of 50.8-
mm and I52A-mm pipe diameters.
Based on the XYZ Plant LPG tank arrangement, a I8.3-m section length for fill/discharge
piping and 9.I-m section length for vapor line piping were applied in the calculation of
failure probabilities. Results are listed in Table 6.
Flexible Hose Rupture Probability

The failure rate data for the rupture of flexible hoses are dependent upon the age of the
hoses, how the hoses are stored and handled, frequency of inspection and testing, and
operator training. From review of historical incidents, the primary factors contributing to
accidental release of LPG from hose failures are bursting of hose due to improper connec-
tions, rupture of hose from failure of the operator to disconnect hose prior to pulling away,
and leakage from the defective hose.
For flexible hose failure, the upper and lower bounds data from the Rijmond [1], Canvey
[2], and AIChE data sources [3] were evaluated. Both the Rijmond and Canvey data are
extracted from U.K. refinery data. The AIChE data is extracted from U.S. EPA data
compiled from hazardous material spills. Because there is a large human error element
implicit in this data, conservative engineering judgment was applied in selecting a worst
case upper boundary failure probability of 1 x 10-3 failure/hose-yr (based on 250 usage h
per year).
Aggregate 25.4 mm Failure Probability

Unless it results in an uncontrolled torch fire which may cause a BLEVE, this category
has a minor impact on off-site risk. To expedite assessment, historical data on small leaks
were applied and an aggregate small leakage failure probability was established in Table 7.
• Automobile Electric System - P = 0.06

• Continuous Operating Traffic Signals - P = 0.24
• Flashing Intermittent Signal Lights - P = 0.08
Research was conducted to evaluate previous methods used to assess ignition probabilities
[11,17,18]. However, as a practical matter, flammable LPG vapor/air mixtures may be ignited
fairly easily. In the vicinity of an LPG spill, ignition sources might include pilot flames,
welding torches, very hot exhaust gases from some engines, etc. Another high-temperature
source which may ignite a flammable mixture is electrical arcs or sparks. Based on minimum
ignition energy of propane vapor, a very tiny spark of a few tenths of a millijoule energy
can initiate ignition. Thus, arcs from ordinary light switches, electric motor brushes, opening
and closing of relays and circuit breakers, sparks from a broken electrical wire, electros-
tatically generated discharges (for example, generated by the rapid flow of LPG or a spray
onto an ungrounded object), and so forth, are potential ignition sources.
Considerations in evaluating ignition potential at the XYZ Plant included:
• Type of Release
• Release Pathway (vapor cloud dispersion direction)
• Number of ignition sources within the pathway area
• Cumulative ignition potential
Each ignition source is characterized by its location, distance, energy, and an estimated
likelihood of ignition. The probabilistic likelihood value is the proportion of times that the
ignition source would cause ignition when engulfed in a flammable vapor cloud at concen-
trations above the Lower Flammability Limit (LFL). Estimated ignition likelihoods are
summarized in Table 8.
Event Trees
For the XYZ Plant, event tree loss scenarios were developed based upon research and
evaluation of past LPG storage tank fires and explosions. Research also involved National
TABLE 8-Ignition likelihood, XYZ Plant.
Likelihood of Intermittent
Ignition Based on Source; Estimated
Ignition Sources Time (per year) for Likelihood of
Being Engulfed which Intermittent Ignition, Given
with Flammable Source is Present Release and
Potential Ignition Source(s) Propane Vapor and Active Flammable Cloud
LPG tank truck 0.06 10% 0.006

LPG vaporizer 0.9 15% 0.135
XYZ plant vehicles 0.06 30% 0.018
Public traffic 0.06 100% 0.06
Public traffic light and
switchgear 0.24 100% 0.24
XYZ plant automatic gate
switchgear 0.08 10% 0.008
XYZ plant equipment
within 38.1 m 0.134 100% 0.134
Human, cigarettes 0.9 1% 0.009
Human, welding torch I 1% 0.01
Lightning 1 0.1% 0.001
Fire Protection Association (NFPA) and American Petroleum Institute (API) protection
standards which describe potential loss scenarios and evaluation of XYZ Plant loss prevention
and control features. It is very important in the development of event tree logic to provide
a complete presentation of significant loss propagation factors which are relevant to the
scope of the assessment. To start this process, a complete categorization of initiating failure
events must be developed. Omitting important initiating events from the analysis can bias
the results by underestimating the overall risk.
Under each category there is an infinite number of release modes, from a pinhole leak
to complete rupture. The focus of this risk assessment is the off-site exposure to humans.
Therefore, emphasis is on catastrophic failure incidents which represent maximum credible
releases of LPG resulting in large fireball radiant heat effects or vapor cloud explosion blast
pressure effects. Five representative failure modes were selected as significant initiating
events which could affect public off-site risk:
1. Rupture of an LPG tank

2. Complete fracture of LPG fill and discharge piping
3. Complete fracture of LPG vapor line piping
4. Rupture of the flexible hose
5. Aggregate 1 in. failures
Based on research and review of historical loss incidents and specific case studies con-
cerning similar type LPG tank storage operations, these failures are considered complete
in terms of off-site risk exposure.
An event tree was structured for each of the selected LPG release cases previously
described. Due to space limitations, all of the event trees are not presented in·this paper.
The event tree for fractured LPG piping is presented in Fig. 5. Main branches of this
event tree include:
(a) LPG release probability.

(b) Probability of immediate ignition. (The probability of immediate ignition for all re-
leases is estimated at 0.24).
(c) Probability of successful LPG system shutdown prior to the release of 4536 kg (from
reliability analysis of emergency shutdown capabilities).
(d) Probability of wind toward populated area based on local wind data. The weather
conditions are estimated in this assessment using a Pasquill atmospheric stability factor
C and a wind speed of 2.23 m/s. It is estimated that 85% of the time there are winds
toward the north.
(e) Probability of delayed ignition, using the cumulative probabilities previously devel-
oped.
(f) Probability of Unconfined Vapor Cloud Explosion (UVCE) versus a flash fire. Given
a large release and delayed ignition, the result is either an explosion which produces
explosion overpressures or a large flash fire incident which produces a radiant heat
exposure. The probabilities assigned to this event are from the compilation of relative
consequences, concerning 80 large LPG losses, by Galatula [19], which indicated that
when a large vapor cloud forms and there is delayed ignition, a relative probability
of 32% UVCE potential versus 68% flash fire potential exists.
For cases where the wind is directed north, and delayed ignition occurs, the center of the
modeled vapor cloud explosion is approximated at 45.7 m north of the LPG Tanks. For
release of LPG of less than 4536 kg vapor, the likelihood of having an unconfined vapor
cloud explosion has been assigned an event probability of 0.10 (regardless of wind conditions)
based on engineering review of available data [2,17].
(g) Probability of torch fire impinging on a tank. The probability of a torch type flame
being directed at the shell of the source tank or adjacent tank is difficult to define.
A conservative engineering judgment of a 50% chance or 0.5 probability will be
assigned, due to lack of a valid analytical technique to quantify this event.
(h) Probability that LPG tank fails from heat exposure (BLEVE). This probability is
related to the effectiveness and reliability of the LPG tank fireproofing, fixed water
monitor system, and fire department response. These probabilities were previously
developed.
Risk Assessment Results

Individual Risk Profile
Individual risk is the frequency at which a given individual may be expected to sustain a
given level of harm from the realization of specified hazards. The individual risk profile.
Fig. 6, graphically depicts the estimated probability of human fatality versus distance north
from the XYZ Plant. This graph is the combined results of the event tree fire and explosion
probabilities. The probabilities for all discrete consequential effects (99% human fatali~'
potential) are added together so that the total individual risk at each point in the graph is
the sum of the individual risk, derived from the event tree analysis.
Acceptable Risk Criteria

The development of individual risk profiles allows the estimated risk of human fatality at
various distances from the XYZ LPG tanks to be compared directly with other risks such
as being killed by lightning or vehicle accident. Therefore, the individual risk profile is very
suitable for use by Planning Commissions as one source of information for making informed
decisions concerning future land development issues.
Research on individual risk categories in use in Europe [20], suggests an acceptable
individual risk range of 10-4 to 10-5 per year for people living outside the boundaries of
an industrial facility involved in the storage and handling of hazardous materials. According
to Rasbash [21], other countries have established a similar range for acceptable individual
risk in relation to industrial plant siting and safety levels.
Land Use Guidance-North of XYZ Plant

If there is an acceptable individual risk level of 10-5, compliance with the proposed
emergency shutdown system, and fire protection, the following guidance for future devel-
opments in the vicinity north of the XYZ Plant is suggested (see Fig. 7):
1. At distances greater than 200 m from the LPG storage tanks, there need be no con-
straints on land use or population density. This distance corresponds to the 1 per
100000 years (10-5) individual risk range.
2. At distances between 100 to 200 m from the LPG storage tanks, industrial development
may be permitted, but neither residential development nor industries requiring a high
population density should be allowed.
3. At distances within 100 m from the LPG storage tanks, development should be re-
stricted. Only developments requiring an extremely low density of staff should be
considered. Ideally, all developments should be located beyond this distance.
The XYZ Plant individual risk profile plot in Fig. 6 consists of single point values, which
resulted from combining the discrete probabilities calculated in the event trees. In further
expanding this type of event tree methodology, it will be important to consider the sensitivity
of the results obtained to the various data inputs. This would require the development and
application of a computer program to handle extensive calculations and integration of prob-
abilistic and deterministic models with expert system capabilities to support and refine the
results. An excellent discussion of sensitivity analysis and uncertainty is provided in Ref 14.
Where risk estimates are to be compared in an absolute sense to risk acceptability criteria,
concern should increase over the issue of absolute accuracy of these estimates which are
typically one or two orders of magnitude. Uncertainty in conducting risk assessments will
decrease in the future, as models become standardized, and failure rate data bases become
more fully developed and available.
References
[1] Rijmond Public Authority, Risk Analysi~' of Six Potentially Hazardous Industrial Objects In the
Rijmond Area, A Pilot Study, D. Reidel Publishing Company. London, United Kingdom, 1982.
[2] Lees, F. P., Loss Prevention in the Process Industries. Butterworths. London, 1980.
\31 Center for Chemical Process Safety, Guidelines For Process Equipment Reliability Data With Data
Tables, AIChE. New York. 1989.
[4] Walls, Wilbur, L., Ed. Liquefied Petroleum Gases Handbook, 1st edition, National Fire Protection
Association, Quincy, MA, 1986.
[5] Standard for the Storage and Handling of Liquefied Petroleum Gases, NFPA 58, National Fire
Protection Association (NFPA). Quincy, MA, 1989.
[6] Kletz, Trevor A., An Engineer's View 'of Human Error, The Institution of Chemical Engineers,
Warwickshire, United Kingdom, 1983.
[7] Design and Construction of Liquefied Petroleum Gas (LPG) Installation, Sixth Edition, Publication
25/0, American Petroleum Institute (API), Washington, DC, April 1989.
[8] Fire-Protection Considerations for the Design and Operation of Liquefied Petroleum Gas (LPG)
Storage Facilities, Publication 25/OA, American Petroleum Institute (API), Washington, DC, April
1989.
[9] "Approved Test Results for Carboline Pyrocrete 241, Fire Protection Coating For Liquefied
Petroleum Gas Steel Storage Tanks," Contract 11OEOA9.AC, Factory Mutual System, Norwood,
MA, July 1980.
[/0] Arthur D. Little, Inc., Reference Manual for Chems-Plus Computer Model, Version 1.0, Cam-
bridge, MA, July 1988.
[1/] Considine, M., Grint, G. c., and Holden, P. L., "Bulk Storage of LPG-Factors Affecting Offsite
Risk," in The Assessment of Major Hazards, Institution of Chemical Engineers, Pergamon Press,
New York, 1982.
[12] Ramsey, C. G., Evans, R., and English, M. A .. "Siting and Layout of Major Hazardous Instal-
lation," The Assessment of Major Hazards, Institution of Chemical Engineers, Pergamon Press,
New York, 1982, pp. 335-351.
[/3] Zalosh, R. G., "Explosion Protection," in SFPE Handbook of Fire Protection Engineering, Na-
tional Fire Protection Association (NFPA), Quincy, MA, 1988.
[14] Cente~ for Chemical Process Safety, Guidelines For Chemical Process Quantitative Risk Analysis,
Amencan Institute of Chemical Engineers (AIChE). New York, 1989.
[15] Technical, Ltd., Techniques for Assessing Industrial Hazards, World Bank Technical Paper 55,
The World Bank, Washington, DC, 1988.
[16] Braustowski, T. A., "Risk Assessment in Large Chemical Energy Projects," in Technological Risk.
Proceedings of a Symposium on Risk in New Technologies, University of Waterloo, Dec. 1981,
University of Waterloo Press, Waterloo, Ontario, Canada, 1982.
[17] Withers, John, "Major Industrial Hazard: Their Appraisal and Control" Halsted Press, New York,
1988.
[18] Mudan, K. and Richard Gustafson, "Ignition Potential Distribution For Heavy Gas Plumes," in
International Conference on Vapor Cloud Modeling, American Institute of Chemical Engineers
(AIChE), New York, 1987.
[19] Galatola, E. and Colombari, V., SONATA: An Accident Data Base for Industrial Activities and
Transport. Scope. Organization. and Main Results, Proceedings of the 5th Eurodata Conference,
Heidelsburg, Germany, Springer-Verlog Press, NY, April 1986.
[20] Shortreed, John H. and Stewart, Angela, "Risk Assessment and Legislation," Journal of Haz-
ardous Materials, Vol. 20, 1988, pp. 315-334.
[2/] Rasbash, D. J., "Criteria for Acceptability for Use with Quantitative Approaches to Fire Safety,"
Fire Safety Journal, Vol. 8, 1984-85, pp. 141-158.
Fire Risk Assessment and
Building Codes
Vaughan R. Beck, 1 Ian R. Thomas,2 G. Caird Ramsay, 3
Hamish MacLennan,4 Raymond Lacey,5 Peter Johnson,6 and
Claude Eaton 7
Risk Assessment and the Design of Fire

Safety Systems in Buildings
REFERENCE: Beck, V. R., Thomas, I. R., Ramsay, G. C, MacLennan, H., Lacey, R.,
Johnson, P., and Eaton, C., "Risk Assessment and the Design of Fire Safety Systems in
Buildings," Fire Hazard and Fire Risk Assessment, ASTM STP 1150, Marcelo M. Hirschler,
Ed., American Society for Testing and Materials, Philadelphia, 1992, pp. 209-223.
ABSTRACT: In 1989, a research project entitled, "Fire Safety and Engineering" was conducted
at the Warren Centre for Advanced Engineering, the University of Sydney, and involved 70
leading researchers and engineers from around Australia. It was found that fire safety and
protection facilities constructed in accordance with current Australian regulatory requirements
are a significant component of the cost of many buildings in this country. The project team
concluded that with the development of systematic approaches to building fire safety and
protection design, based on risk assessment models and fire-engineering design techniques,
substantial cost savings can be made while achieving satisfactory levels of life safety. There is
an urgent need for a greater understanding by designers and regulatory officials of fire phe-
nomena, human behavior, and fire engineering techniques.
KEY WORDS: risk assessment, systems, cost-effective, life safety, design, fire protection
Fire Risks And Costs In Australia

Fire safety and protection facilities to satisfy current Australian regulatory requirements
are a significant component of the cost of many buildings in Australia. With the development
of systematic approaches to building fire safety and protection design, based on risk-
assessment models and fire-engineering design techniques, substantial cost savings can be
made while achieving satisfactory levels of life safety.
The Risks
The community faces a wide range of risks every day. Some of these risks are accepted
voluntarily, while others are imposed on people without much choice. Involuntary risk levels
'Professor, Victoria University of Technology, Centre for Environment Safety and Risk Engineering,
MMO, Melbourne, Australia, 3000.
2Manager, Engineering Research, BHP Research and New Technology, PO Box 264, Clayton, Australia
3168.
'Leader, Fire Initiation and Development, CS.I.R.O., Division of Building, Construction, and
Engineering, PO Box 56, Highett, Australia 3190.
"Associate professor, University of Technology, Sydney, Department of Building Science, PO Box
123, Broadway, Australia 2007.
'Director, Lincolne Scott Australia Pty, Ltd., Consulting Engineers, 121 King Street, Melbourne,
Australia 3000.
6Principal material scientist, Australian Construction Services, Scientific Services Laboratory, 177
Salmon Street, Port Melbourne, Australia 3207.
7Consultant, Building Owners and Managers Association, Ltd., 14 Martin Place, 10th Floor, Sydney,
Australia 2000.
are expected to be much lower than those entered into voluntarily, or they will arouse
opposition.
The risk of death from fire and flames in Australia is some eight deaths per 1 000 000
people a year. While the risk of death from fire and flames is small, it is still significant
(although it has fallen by half during the last 20 years). It would appear from available
statistics, that approximately half of the fire and flame deaths occurred in buildings. Thus.
the number of fatalities per 1 000 000 people a year from building fires in Australia is about
four. This compares very favorably with the American figure of over 20, and the Britisb
figure of approximately 15.
Analysis of available fire statistics reveals that 47% of building fires were in one and two
family dwellings. These fires accounted for 69% of all fire deaths, and 53% of fire-related
injuries in buildings. Private accommodations, other than one and two family dwellings.
accounted for 10% of all fires, 7% of deaths, and 16% of injuries. Commercial accom-
modation (hotels and boarding houses) had 2% of all fires, 3% of deaths, and 3.2% of
injuries. Thus, some 80% of all fire deaths occur in residential-type buildings.
The Costs
Reliable data on the cost associated with building fires in Australia is difficult to obtain.
For countries with a similar technological development to Australia, the following are typical
figures: direct fire loss claims (0.23% Gross Domestic Product (GDP)), indirect fire loss
claims (0.03% GDP), expenditure on fire fighting (0.20% GDP), and expenditure on fire
protection (0.26% GDP). Hence, the total cost of fire in buildings is some 0.72% GDP.
which is equivalent to $2000 million per year for Australia. The expenditure of fire safety
and protection in buildings can be attributed to the costs of materials and labor, as well as
costs for design and construction time. Delays associated with obtaining regulatory approvals
can be very expensive.
Implications
Currently, most fire deaths (69%) occur in one and two family dwellings where there is
little or no expenditure on fire safety. Conversely, in high-rise and other buildings, a relatively
small proportion of deaths occur, yet there is substantial expenditure on fire safety and
protection facilities. For example, a 10% reduction in the costs associated with fire in
buildings would represent a saving of 200 000 000 dollars per year.
Current Situation And Need For Change

Society has responded to the threat of fire in buildings in many ways. Among those are
included: fire brigades, insurance, education on fire hazards, controls on the use of materials,
products in buildings, and the design of buildings to resist the effects of fire. The level of
safety reflects the general economic, social, and cultural features of society. The control
mechanisms and organizational arrangements used in Australia have obviously been suc-
cessful as Australia has one of the lowest death rates from building fires in the world.
Currently, building regulations are an important component in design for fire safety in
buildings. The design requirements of building regulations were introduced many years ago
and are applicable to the technology and practices then in vogue. Many of these provisions
were empirically derived but they have assumed great authority with the passage of time,
although possibly lacking technical substantiation. The prescriptive requirements in building
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 211
regulations are a reflection of the low level of technology previously available for the design
of fire protection and safety in buildings.
There has been substantial recent progress in the development of building regulations in
Australia, with the Building Code of Australia being introduced progressively throughout
the country. However, the current design for fire safety and protection is unscientific, being
largely based on empirical rules. The building regulations do not contain any explicit state-
ments of the fire-safety design objectives to be achieved.
It is acknowledged that the current design approach has resulted in the achievement of
safety levels which the community appears to accept. Accordingly, it is appropriate to require
that in the development of alternative designs for building fire safety and protection, existing
safety levels should be maintained.
However, the current design approach is unlikely to result in the most cost-effective design
solutions, nor in designs which maintain a consistent level of safety. The complexity of the
regulations, the number of disciplines involved (none of which is primarily responsible for
fire safety), and the fragmentation of the design process is a central problem for the design
of fire safety in buildings. It appears that the situation is essentially no different in most
other countries; that is, the procedure involves observance of regulations which are consid-
ered inflexible and inefficient.
Furthermore, the current regulatory approach restricts the range of choices available,
inhibiting and restricting innovation. The recycling of existing buildings, often historic,
demands that there be a more systematic approach to the design for fire safety and protection
in buildings.
The fields of fire dynamics and fire protection engineering are rapidly advancing. This
brings opportunities for reducing the overall cost of safety and protection measures while
maintaining current safety levels.
Warren Centre Project

In 1989, a research project, "Fire Safety and Engineering," was conducted at the Warren
Centre for Advanced Engineering, the University of Sydney [1] and involved 70 leading
researchers and engineers from around Australia. The Warren Centre project represented
a major development in Australia, in that wide spread consensus was reached on the basis
for undertaking the design of fire safety systems in buildings.
The project found that fire safety and protection facilities constructed in accordance with
current Australian regulatory requirements, are a significant component of the cost of many
buildings in this country. The project team concluded that with the development of systematic
approaches to building fire safety and protection design, based on risk assessment models
and fire-engineering design techniques, substantial cost savings can be made while achieving
satisfactory levels of life safety. There is an urgent need for a greater understanding by
designers and regulatory officials of fire phenomena, human behaviour, and fire engineering
techniques.
Presented herein is a brief description of some of the issues addressed by the Warren
Centre project team [1] which includes the following list:
(a) Design objectives and criteria.
(b) Risk assessment model. The risk assessment model comprises a number of submodels
which are used to represent the performance of various fire safety subsystems. A brief
description is given of each of the subsystems. Also presented are the results obtained
from risk assessment model studies conducted prior to, and including, the Warren
Centre project.
Design Objectives
The major objective of design for the effects of fire in buildings is to achieve satisfactory
levels of life safety for:
(a) occupants of the building of fire origin;

(b) occupants of adjoining buildings; and
(c) fire brigade personnel.
The level of fire safety in buildings is a reflection of the complex interaction between fire
growth and spread, and human behavior. This depends on many features of the building
including active and passive protection facilities, provision for egress, occupant mobility and
familiarity, and building management. Thus, fire safety in buildings is a system consisting
of many interacting subsystems.
Alternative building designs must obtain satisfactory levels of life safety. The level of
property protection in buildings should not be subject to community regulation, but a matter
for building owners and their insurers.
Risk Assessment Models (RAM)

Risk assessment models (RAM) can be used to help identify those combinations of building
subsystems which provide the requisite level of safety in a cost-effective manner. Once a
building fire safety system has been chosen, it is then appropriate to use deterministic-type
models to undertake the detailed design and specification of the individual fire safety and
protection subsystems. Deterministic fire-engineering design methods cannot be used to
identify those combinations of subsystems which provide the requisite level of safety in a
cost-effective manner.
In the use of risk assessment models, deterministic models are employed to estimate the
effects of particular fires within the enclosure of fire origin, as well as the spread of fire to
a few adjoining enclosures. Deterministic models and risk assessment models are comple-
mentary to one another in the design process. However, deterministic models alone are not
suited to the identification of cost-effective designs for building fire safety and protection
systems.
To model the risk from fire in buildings, it is necessary to estimate the likelihood of
harmful consequences to people and property, by taking a global, or systems view, of fire
safety and protection. This is a radical change from the traditional approach. In developing
a systematic approach to building fire safety and protection, it is appropriate to consider a
number of subsystems, namely, (a) nature of occupancy, (b) fire growth and development,
(c) active subsystems, (d) passive subsystems, (e)occupant avoidance, and (f) fire fighting.
A description of these subsystems is given subsequently.
Design Criteria
In design, it is appropriate that consideration be given to the level of safety afforded to
occupants of buildings, and to the costs associated with such provision. When applied to
alternative designs, such an approach enables designers to select the most appropriate cost-
effective solution. Accordingly, risk assessment models that have been developed charac-
terize building design performance in terms of two parameters, namely: (a) expected risk
to life and (b) fire cost expectation.
No attempt was made to assign monetary values to either the loss of life or the value of
lives saved. This avoids serious moral, ethical, and economic difficulties which arise when
attempting to assign monetary value to human life or suffering. The risk-to-life parameter
provides an explicit estimate of the expected number of deaths resulting from fire in buildings.
To identify alternative designs which are considered equivalent to, and more cost-effective
than designs conforming with current regulatory provisions, the decision criterion was, "For
an alternative design to be considered acceptable, the expected risk-to-life value shall be
equal to, or less than, the risk-to-Iife value of a building conforming with the regulations,and
the fire-cost expectation for the alternative design shall be less than the value for the
conforming building."
With such a comparative approach, it is not required to directly compare estimated risk-
to-life values, derived from a risk assessment model, with an acceptable level of risk derived
from independent sources. The expected risk-to-life values for designs conforming with
current regulatory requirements provide an estimate of current levels of risk-to-life safety.
These risk levels are assumed to be acceptable to the community. However, reduced or
lligher levels of risk may be justified or sought for a variety of reasons. The expected risk-
ao-life values for designs conforming with current regulatory requirements provide a con-
.enient reference, or benchmark, to compare the performance of other designs.
A Description of the Subsystems

The Risk Assessment Model developed during the Warren Centre project comprises
several submodels to represent the performance of various fire safety subsystems. Given
below is a brief description of the subsystems.
Nature of Occupancy
Modern building codes categorize all occupancies into a limited number of generic clas-
sifications, which in turn, form a primary determinant for the type of construction and the
extent of fire and safety facilities to be provided. The assumption that similar hazard char-
acteristics apply to all occupancies within each occupancy classification is often not valid.
A systematic approach to fire safety in buildings requires the specific needs and characteristics
of particular occupancies to be quantified so that related construction and fire and safety
facilities can be designed appropriately. The following factors were identified as the fun-
damental characteristics of various types of building use.
• combustible contents
• elements of construction
• energy services
• ignition sources
• installed equipment
• premises-fire cover
• premises-management
• occupant characteristics
A different approach to the same classification involved four groups of characteristics:

population, management, premises, and contents. However, more work is necessary to
quantify data for the fire safety risk assessments.
The limited statistics available were analyzed to provide quantitative input to the risk
assessment model, particularly in regard to the frequency of fire starts in the various oc-
cupancies and the proportion of those fires which corresponded to the design fires used in
the demonstration risk assessments undertaken in the project.
Fire Initiation and Development
Clearly, fire initiation and development plays a central role in fire safety in buildings.
The design of the building in terms of the fire resistance of the structure, the inclusion of
various means of escape, the provision of detection, suppression subsystems and firefighting
equipment, and the design of air-handling subsystems, are all required to mitigate and control
fire development, heat, smoke, and toxic gases which threaten life. Much less attention is
paid to preventing ignition and controlling fuel.
Control of ignition sources and fuel could, in principle, provide a general means of
improving the fire safety of buildings. Such an approach could reduce the "design require-
ments" discussed above, leading to decreased costs and perhaps even greater levels of safety
because experience has shown that the "design systems" sometimes fail to protect life and
property.
In simplistic terms, fires will not start without ignition sources or fuel to ignite, and they
will not develop without sufficient fuel. The elimination of all ignition sources and fuel is
generally not a feasible proposition, and limitation of sources and fuel according to their
fire hazard and limitation of their interaction is an appropriate approach. For rational and
economic decisions to be made on the appropriate degree of limitation, it is necessary to
balance this control with the design requirements discussed above. RAMs, such as those
developed in the Warren Centre project, provide the means to do this.
The RAM was directed at two case studies: a multi-storey apartment building and a multi-
storey office building. Specifically, information was required on fire conditions for other
subsystems and on event times for direct input to the RAM.
It is not possible, with the present state of fire science and mathematical modelling of
fires, to predict the initiation and development of a fire and the environment produced from
first principles. In the present work, "design fires," appropriate to fire scenarios typical of
each of the two case studies, were used to provide input to mathematical models for predicting
fire conditions. The design fires chosen were based on experimental data for rate of heat
release available in the literature. Three design fires (smoldering, flashover, and nonflash-
over) were selected for the apartment, and two design fires (partial flashover and nonflash-
over) were selected for the office.
For the apartment, the smoldering fire was based on experiments [2] carried out on a
smoldering chair as analyzed by Quintiere et al. [3]. The flashover fire was based on data
presented by Rockett [4] from experiments performed by Lee [5], and corresponds to a
simplified rate of heat release curve for Lee's "free burn" Number 6. The fuel involved in
this fire is essentially a made up double bed with sidetable and wastepaper basket. For the
nonflashover fire, rates of heat release equal to approximately two-thirds the rates of heat
release for the flashover fire were used. This represents a reduced fire load approximately
equivalent to removing the headboard and using a lighter (lower quality) mattress.
For the office building, the partial flashover fire was based on Nelson's simulation of fire
that had occurred in a Los Angeles bank [6]. The primary fuel load for this fire was a floor-
wide array of computer workstations. For the non-flashover fire, the rate of heat release
curve used by Nelson [6] to represent the burning of a single workstation was used. This
curve was derived from experimental data collected by Walton and Budnick [7].
Two mathematical models were used to predict fire conditions and provide the necessary
input data for the RAM. The smoldering fire in the apartment was simulated using a program
called SMOLDER® [8]. This program is based on the zone model proposed by Quintiere
et al. [3], using Zukoski's [9] computations. The program maps the movement of heat,
smoke and carbon monoxide in up to two adjoining rooms on one level. The input required
is floor plan data, ceiling height, door dimensions, height of fire source, and ambient
temperature (other necessary parameters are fixed). The outputs include, for each enclosure
being studied, temperature and volume of upper gas layer, carbon monoxide concentrations,
and accumulated dose of carbon monoxide at a set height. Flaming fires, flashover and non-
flashover in the apartment, and partial flashover and nonflashover in the office were sim-
ulated using a zone model called FAST®, Version 17.3, 3 Aug. 1987 [10]. This model maps
the movement of heat, smoke, and gases in up to five adjoining rooms on the one level.
The inputs required include floor plan data, connections between enclosures, thermophysical
properties of the enclosing surfaces, and fire specifications (fire position and fuel charac-
terized; namely, either heat release rate or mass loss rate). The outputs include, for each
enclosure being studied, temperature and volume of upper gas layer, surface temperature,
and smoke and gas concentrations.
Active Subsystems
Active fire safety subsystems may be defined as those subsystems which react to a fire
situation. They may also be characterised by having time-dependent performance. They
may playa role in building fire safety by
(a) directly influencing the development of fire (by providing or removing the provision
of fuel, by controlling ventilation or by acting to suppress or extinguish the fire);
(b) inhibiting the spread of fire or spread of the products of fire;
(c) assisting people affected or potentially affected by a fire to become aware of the
existence of fire, assisting such people to remove themselves from a hazardous lo-
cation, lessening the likelihood of hazard in a particular location, or to assist in gaining
access to fight a fire; and
(d) providing means of actively fighting a fire for the occupants or trained fire fighters.
These active subsystems include:
• smoke detection equipment

• thermal detection equipment including sprinkler heads
• smoke management subsystems (for both maintaining egress routes tenable and con-
tainment of smoke spread)
• fire suppression subsystems (automatic, manual, or purpose provided for fire brigade
use)
• alarm, information, and communication subsystems (for alerting occupants, relaying
alarms to the fire brigade, and other active systems)
• emergency support subsystems (egress route illumination, emergency power supplies,
public utilities serving the building)
While active subsystems have been used extensively in the recent past, little is known in
detail about their performance under fire conditions, except for broad statistical data on
the performance of automatic sprinklers. Many subsystem components such as thermal
detectors, smoke detectors, and hose nozzles have been individually tested to provide known
component performance. However, the role and performance of the subsystems, of which
these components form a part, require much investigation to provide the required perform-
ance data from the RAM.
Active sub-system components can playa role at the various stages of fire growth. The
Warren Centre study assessed, for a range of active sub-system components, the performance
characteristic likely to be achieved at the various stages of fire growth.
Passive Subsystems
Passive fire safety subsystems may be defined as those systems which do not react to a
fire situation. They may play three major roles in building fire safety by
(a) directly influencing the development of fire (by providing fuel or by controlling ven-
tilation) ;
(b) inhibiting the growth or spread of fire itself, or the spread of the products of fire;
and
(c) assisting the people affected or potentially affected by the fire to remove themselves
from a hazardous location, or to gain access to fight the fire.
Passive subsystems were considered to include
(a) vertical separating elements (interior and exterior walls);

(b) horizontal separating elements (room and floors below and above fire);
(c) designed openings and their treatment (doors, windows, pipe, cable or duct penetra-
tions, construction gaps and interstitial spaces in separating elements);
(d) load bearing structure (vertical and horizontal elements, structural system); and
(e) architectural configuration (floor area, architectural layout, storey height intercon-
necting spaces, such as the atria and shafts).
Much is known about passive sub-system behavior during post-flashover fires due to
extensive testing using the standard fire-resistance test (rather than in real fire situations).
A review of this information and the translation of its applicability to other stages of fire
growth, the consideration of real fire scenarios, and aspects associated with the movement
of people was undertaken. Techniques for the design of passive subsystems were assessed.
and recommendations for research presented.
The role of a particular passive subsystem in acting as a barrier to the spread of fire or
its effects is, of course, dependent on the particular fire scenario. The relationship between
various fire scenarios and characteristics of passive subsystems that relate to performance
is based on time; the time period over which the passive subsystem effectiveness degrades
in the particular fire.
The presence of openings is the dominant performance characteristic for most passive
subsystems. The possible presence of openings in a barrier has a significant effect on its
likely effectiveness. Opening treatments such as doors or penetration sealants may restore
effectiveness, provided that their performance is adequate, and provided they are closed.
or in place when the fire occurs. Maintenance of passive subsystems thus requires close
attention.
Windows are an important feature in determining the effectiveness of exterior wall sub-
systems in preventing fire spread to other building spaces and to adjacent buildings. Windows
can form effective barriers to smoke spread. Nonetheless, they are a major mechanism for
flame spread up building facades and radiative heat transfer/fire spread to adjacent buildings.
Occupant A voidance
Occupant avoidance in fire-related emergencies is defined as the responses and likely
actions available to occupants during a fire to ensure their safety. These actions generally
involve responding to a cue (smoke, heat, unfamiliar sounds or alarms), making a decision
to evacuate, preparing for evacuation, and then travelling to and through an exit route to
a place of safety.
Occupant avoidance (or means of egress) was previously modelled assuming that upon
activation of an alarm or detection of fire by other means, occupants of a building would
immediately proceed (by the most direct means) to travel the exit route to an open space
or a safe haven. Another generally accepted view was that people would tend to panic in
the event of a fire, and that steps should be taken to alleviate this problem. Further,
"hydraulic-based" egress models have often been used as the best available approach to
predict the evacuation times for building occupants. It has been demonstrated that this
approach is quite inadequate and fails to understand, or account for, human behavior.
The time taken for occupants to decide that they are going to evacuate can exceed the
time to travel to and through the exit route. This time is generally taken up with occupants
investigating the fire, and may mean that occupants move towards the fire. Occupants may
also seek information from other people. The design and installation of emergency warning
systems is therefore critical to ensure that people receive the appropriate level of information.
The importance of education and training in improving the preparedness of occupants to
respond quickly, and take effective avoidance action, should be recognized.
The current and emerging technologies in human behavior, egress movement (including
way finding), and cue response were reviewed, and a qualitative assessment of the control
factors that would affect occupant avoidance was prepared. A further issue examined was
the choice of exit; this being a function of an occupant's familiarity with the building.
For the RAM input, a quantitative assessment of the number of persons exposed to
untenable conditions in the eight-storey apartment building and the twelve-storey office
building was undertaken. Probabilities were assigned to estimate the likelihood of response
under certain given cue/alarm conditions.
Fire Fighting
The fire brigade is an emergency service whose role is to save lives and property in the
case of fire. This is accomplished by rescue, prevention of the spread of fire, and the timely
extinguishment of the fire.
Fire services attempt to provide a level of service commensurate with the presumed hazard.
This level of service reflects the deployment of stations -and staff in order that rescue and
fire fighting can be provided. Fire brigade operations are dependent on the receipt of an
alarm. Thereafter, the appropriateness of the response and the time for response, setup,
control, and extinguishment are the main determinants of fire brigade effectiveness.
Fire brigades have adopted advanced technology to enhance fire suppression activities.
However, it is recognized that major influence on the fire-loss performance is the ability of
the building fire safety and protection system to limit fire growth and spread, to limit the
number of people exposed, and to allow better accessibility for fire suppression resources.
Successful fire fighting may be achieved with preplanning, early alarm, rapid response, and
the provision of sufficient equipment and labor to respond to the hazard presented by the
occupancy.
The performance of the fire brigade was incorporated into the RAM by comparing the
time at which the fire brigade could expect to start extinguishment operations with the time
associated with fire growth. Provided operations commenced prior to flashover, the fire
brigade was deemed to be successful in extinguishing the fire.
Risk Assessment Results

Previous Results
The application of risk assessment models to quantify the levels of life safety, arising from
the effects of fire in buildings, have been under development for some time [11]. Risk
assessment models have been developed and applied to office buildings [12] and apartment
buildings [13]. While a discussion is given below of the results obtained, it must be recognized
that these results have not been calibrated, and further work is required to improve their
reliability.
Results obtained from research into office buildings [12] have shown that the fire-resistance
ratings (FRR) specified in Australian building codes for office buildings can be reduced
slightly (typically 1/2 h) without introducing any appreciable change to the level of risk-to-
life safety. These reductions in FRR can represent significant savings (typically 10% for
steel-framed buildings) in the capital cost required for fire safety and fire protection in office
buildings.
It was further found that increasing the code-specified fire-resistance ratings do not im-
prove the level of risk-to-life safety, but rather substantially increase the capital cost required
for fire safety and protection. Conversely, significant reductions (but not elimination) in the
code-specified fire-resistance ratings will result in a slight increase in the level of risk-to-life
safety.
Results from a sensitivity analysis revealed that while fire-resistance ratings have, in some
cases, important effects on the level of risk-to-life safety, other input variables have a more
significant impact on risk-to-life safety values. Facilities designed to control the movement
of smoke were found to have a significant effect on the risk-to-life safety values. Examples
of such facilities include stairway doors, smoke exhaust subsystem, and stair pressurization
subsystem. The input variables associated with the performance of these facilities is char-
acterized by the reliability of these facilities. It may then be inferred that both the inclusion
and improvements in the reliability of these facilities will have a significant beneficial effect
on the risk-to-life safety for the occupants of office buildings. Improvements in reliability
are also contingent upon the design, operation, maintenance, and use of these facilities.
From an extended sensitivity analysis, it was found that despite some significant changes
in the absolute values for the output decision-making parameters, the same possible ac-
ceptable alternative design was identified, as was the case for the standard data set. This
provides greater confidence in adopting the suggested changes for fire-resistance ratings for
elements of construction in office buildings.
To further demonstrate the potential application of the assessment of cost effectiveness
of fire-safety provisions, a simple case of a three-storey apartment building was considered
[13]. The three storey apartment building is assumed to have six apartments on each floor
and two stairwells. For such a building, the present National Building Code of Canada
requires the installation of a central alarm system, but not sprinklers or stairwell pressuri-
zation. To determine the cost effectiveness of these various fire protection measures, 12
cases were considered, encompassing all possible combinations of these three fire protection
options, plus the option of a higher reliability central alarm system (higher detection and
warning reliabilities). The results show that, first of all, stairwell pressurization in such a
low-rise building does not reduce the risk, but will increase the protection cost. This is
probably because in such a low-rise building with a low evacuation time, the chance of being
trapped in the stairwell is very low. The effect of stairwell pressurization may be more
pronounced in taller buildings.
The reference option, as required by the Building Code of Canada, consists of a central
alarm and no stairwell pressurization. Compared with the reference code option, it was
found that the option of solely sprinklers and the option of a central alarm and sprinklers
will reduce the risk, but will incur higher protection costs. The option of a higher reliability
alarm only is the most cost effective. This option will reduce the risk but will not increase
the cost compared with the reference option (in the model, the cost for a higher reliability
central alarm was assumed to be about the same as that of a regular central alarm). The
option of both a higher reliability central alarm and sprinklers will reduce the risk further,
when compared with the option of a higher reliability central alarm only. The option of a
higher reliability central alarm system and sprinklers has little effect on the risk to life
compared with the option of a central alarm system and sprinklers. This is because the risk
has already been reduced substantially in buildings containing regular central alarms and
sprinklers.
It is interesting to note from the results that the effectiveness of fire protection measures in
reducing fire risk is not as dramatic in apartment buildings [13] as in the office building [12].
The results [13] show that various active fire protection measures can reduce the risk from
no active protection in a three-storey apartment building by a factor of about 1.7 to 3.2.
Previous research [12] shows that in an office environment, similar fire protection measures
can reduce the risk by one to two orders of magnitude. This apparent difference is the result
of the fact that in apartment buildings, there is inherently more compartmentation and fewer
people exposed to the fire in the zone of fire origin.
Current Results
In the Warren Centre project [1], and following a critical appraisal of existing risk as-
sessment models (including Refs 12 and 13), it was decided to give much greater emphasis
to modelling time effects in terms of the interactive processes associated with building fires,
namely, fire growth and spread, occupant avoidance, and the intervention of engineering
subsystems and the fire brigade.
Two demonstration risk assessment models (DRAM) were developed during the Warren
Centre Project [1]. One DRAM was applied to an 8 storey apartment building, and the
other DRAM was applied to a 12 storey office building. The DRAMs were so structured
that various subsystems could be included or excluded in any combination. Substantial input
data is required for the DRAMs. While the development of this data represents a significant
task, once established and validated for a generic type of occupancy, much of the data will
remain unchanged, and appropriate for use and future applications.
During the Warren Centre project, improvements were made in modelling the time of
fire growth and spread. However, because of time constraints during the project, insufficient
time was available to reliably model the probabilities of smoke spread and fire spread.
Further, a deliberate policy was adopted of invoking conservative assumptions for the initial
application of the DRAMs. This resulted in estimates of risk-to-life safety which are clearly
very conservative (that is, higher than expected in reality). This situation is ameliorated
somewhat by using the results from the DRAM on a comparative, rather than absolute
basis. However, this does not relieve the need to further improve the reliability of the
subsystem models and thereby improve the accuracy of the results from the DRAMs.
Results for expected risk-to-life and fire-cost expectation were plotted for various building
fire safety design configurations. Typical results obtained from the DRAM for the apartment
building are presented in Fig.!. The results are applicable to an 8 storey apartment building,
containing 6 apartments per level, a central core consisting of two lifts and two stairways.
The apartment building is constructed of reinforced concrete.
When using the DRAMs during the Warren Centre project, no attempt was made to
identify alternative cost-effective design solutions. Rather, the purpose was to investigate
whether the DRAMs could quantify and discriminate the effect of various subsystems. From
the results presented in Fig. 1, it is seen that the DRAMs achieve this objective.
Conclusions
During the Warren Centre project, the following conclusions were noted:
Current Situation
(a) Australia has achieved an excellent fire safety record compared with other countries.
It is essential that this record be maintained.
(b) The vast majority of fire deaths in buildings occur in dwellings. A very small percentage
of fire deaths occur in high rise and other nonresidential buildings, yet there is very
significant expenditure of fire safety and protection.
(c) There is evidence that substantial cost savings are possible while maintaining our
current fire safety record.
(d) The design for fire safety in buildings is controlled and administered in a highly
legalistic regulatory environment.
(e) There is a need to introduce design flexibility to consider a wider range of possible
fire-safety systems.
Design For Fire Safety
(a) During the Warren Centre project, an examination was made of the most appropriate
philosophy for a systematic approach to engineering design for fire safety in Australia.
(b) Risk assessment models were developed to identify cost effective fire safety systems
for particular buildings and occupancies.
(c) The risk assessment approach was used to model the effect of fire growth and spread
and technologically advanced procedures for the achievement of cost-effective building

designs which conform to the fire safety levels implicit in the Building Code of Australia.
It is expected that the NBFSSC will eventually be published as an Australian Standard,
and will be referenced in the Building Code of Australia as providing an equivalent but
alternative route for fire safety design. The NBFSSC will be supported, in part, by the
preparation of an explanatory manual and educational programs.
The NBFSSC will be fully developed over a three-stage program. The first (six-month)
stage has been approved and is being funded by the Building Regulation Review Task Force
(BRR). The NBFSSC, Stage 1, is being compiled by the authors of this paper, who were
the principal participants involved in the Warren Centre project. A draft code, suitable for
the design of office and apartment buildings, will be developed during Stage 2 (18 months).
During Stage 3 (18 months) the Code will be further developed and made applicable to all
occupancies. Currently no commitments have been given to fund Stages 2 and 3.
An independent but related development is the formation of the Australian Centre for
Building Fire Safety and Risk Engineering. The operations of the Centre will, in part,
support the development of the National Building Fire Safety Systems Code. Clearly, tan-
gible and significant steps are being taken to implement the initiatives developed during the
Fire Safety and Engineering Project that was conducted at the Warren Centre for Advanced
Engineering.
The successful introduction of the NBFSSC will require the active involvement and support
of all sectors of the fire safety and protection industry.
Acknowledgments
This paper is based largely on the Warren Centre "Fire Safety and Engineering" Project
Report [1]. The Fire Safety and Engineering Project involved some 70 project fellows who
are senior professionals from various disciplines and organizations associated with fire safety
and protection in Australia. The significant contribution of the project fellows is gratefully
acknowledged. The subsequent support of the Building Regulation Review Task Force for
the first stage development of a National Building Fire Safety Systems Code is also gratefully
acknowledged.
References
[1] "Fire Safety and Engineering," Project Report and Technical Papers, Books 1 and 2, The Warren
Centre for Advanced Engineering, The University of Sydney, Australia, Dec. 1989.
[2] Bukowski, R. W., "Investigation into the Effects of Heating and Air Conditioning on the Per-
formance of Smoke Detectors in Mobile Homes," NBSIR 79-1915, National Institute of Standards
and Technology (formerly National Bureau of Standards), Washington, DC, Oct. 1979.
[3] Quintiere, J. G., Birky, M., MacDonald, F., and Smith, G., "An Analysis of Smoldering Fires
in Closed Compartments and Their Hazard Due to Carbon Monoxide," Fire Materials, Vol. 6,
Nos. 3/4, 1982, pp. 99-110.
[4] Rockett, J. A., "Park Room Fire Test Simulations, Mining the Havard Level 5.2 Computer Fire
Model," NBSIR 83-2805, National Institute of Standards and Technology (formerly National
Bureau of Standards), Washington, DC, June 1984.
[5] Lee, B. T., "Effect of Wall and Room Surfaces on the Rates of Heat, Smoke and Carbon Monoxide
Production in a Park Lodging Bedroom Fire," NBSIR 85-2998, National Institute of Standards
and Technology (formerly National Bureau of Standards), Washington, DC, Feb. 1985.
[6] Nelson, A. E., "An Engineering View of the Fire of May 4, 1988 in the First Interstate Bank
Building, Los Angeles, California," NBSIR 89-406, National Institute of Standards and Technol-
ogy, (formerly National Bureau of Standards), Washington, DC, March 1989.
[7] Walton, W. R., and Budnick, E. K., "Quick Response Sprinklers in Office Configurations: Fire
Test Results," NBSIR 88-3645, National Institute of Standards and Technology (formerly National
Bureau of Standards), Washington, DC, Jan. 1988.
[8] Shestopal, V., Private Communication, Division of Building Construction and Engineering, CSIRO,
Sydney, Australia, 1989.
[9] Zukoski, E. E., "Development of Stratified Ceiling Layer in the Early Stages of a Closed-Room
Fire," Fire and Materials, Vol. 2, No.2, 1978, pp. 54-62.
[10] Jones, W. J., "A Multi-Compartment Model for the Spread of Fire, Smoke and Toxic Gases,"
Fire Safety Journal, Vol. 9, 1985, pp. 55- 79.
[11] Beck, V. R., "Outline of a Stochastic Decision-Making Model for Building Fire Safety and Pro-
tection," Fire Safety Journal, Vol. 6, No.2, 1983, pp. 105-120.
[12] Beck, V. R. and Poon, S. L., "Results from a Cost Effective Decision-Making for Building Fire
Safety and Protection," Fire Safety Journal, Vol. 13, 1988, pp. 197-210.
[13] Beck, V. R. and Yung, D., "A Cost-Effective Risk Assessment Model for Evaluation Fire Safety
Protection in Canadian Apartment Buildings," Journal of Fire Protection Engineering, Vol. 2, No.
3, 1990, pp. 65-74.
George V. Hadjisophocleous1 and David Yung1
A Fire Risk and Protection Cost Assessment

Model for Highrise Apartment Buildings
REFERENCE: Hadjisophocleous, G. V. and Yung, D., "A Fire Risk and Protection Cost
Assessment Model for Highrise Apartment Buildings," Fire Hazard and Fire Risk Assessment,
ASTM STP lI50, Marcelo M. Hirschler, Ed., American Society for Testing and Materials,
ABSTRACT: This paper describes a stochastic system model that was developed to represent
the dynamic interaction between human behavior and fire growth and spread in high rise
apartment buildings. The effects of the smoke and fire spread are calculated in terms of two
performance parameters: the expected risk-to-life (ERL) and the fire-cost expectation (FCE).
These performance parameters are used to assess the cost-effectiveness of fire safety and
protection provisions in highrise apartment buildings. The system model can be used as a tool
to assess whether alternative designs provide equivalent safety and whether these alternative
designs are cost effective. As an example, the model is applied to a 28-storey apartment
building where the cost-effectiveness of 9 different combinations of alarm and sprinkler systems
is assessed.
KEY WORDS: risk, cost, assessment, model, apartment, building
Nomenclature
CO Carbon monoxide concentration in ppm
CO2 Carbon dioxide concentration in percentage
dt Time increment in min
ERL Expected risk-to-life
FCE Fire-cost expectation
FID Fraction of incapacitating dose
FTD Fraction of incapacitating thermal dose
KID A constant defined in Eq 1
P" Probability of death as a result of smoke spread
To Ambient temperature in DC
Ts Hot smoke temperature in DC
Introduction
In response to the growing need and interest in economic assessment of building code
provisions, the Institute for Research in Construction of the National Research Council of
Canada (NRCC) is currently developing analytical tools that can be used for assessing the
cost effectiveness of fire safety and protection provisions in buildings. These analytical tools
1 Research officer and group leader, respectively, National Research Council of Canada, Institute for
Research in Construction/National Fire Laboratory, M59, Ottawa, Ontario, Canada KIA OR6.
224
HADJISOPHOCLEOUS AND YUNG ON HIGHRISE APARTMENT BUILDINGS 225
are being developed in collaboration with the Victoria University of Technology, FIT campus
in Australia, where in 1987 the first such risk-cost assessment model for application to
Australian office buildings was developed [/,2]. The first step by NRCC has been the
development of a similar risk-cost assessment model for Canadian apartment buildings, with
a longer term program of developing similar models for other types of buildings.
This paper describes the NRCC Canadian apartment model. It is based on a similar
approach as was used in the Australian office model [3,4]. The apartment model incorpo-
rates, among other things, the construction and occupancy features applicable to apartment
buildings. In this paper, a brief description of the apartment model is given and some
preliminary results concerning the cost-effectiveness of various alarm and sprinkler systems
are presented. It should be noted that many of the submodels are still in the developmental
stage or are being improved to provide a more accurate representation of fire growth and
human behavior. The present paper, therefore, gives only a general description of the risk-
cost assessment model developed and some preliminary results of its use.
System Model
The basic system model has been described in detail in previous papers [/-4]; only a brief
description of the system model, shown in Fig. 1, is given in this paper. The system model
consists of a number of stochastic, state-transition submodels and interrelated deterministic
submodels that were developed to represent the dynamic interaction between fire growth
and human behavior. These submodels can be classified into the following four major groups:
(1) Fire Growth and Spread Submodels- The development of a fire in the apartment of
fire origin is characterized in terms of a number of distinct fire states. The probability of
the fire reaching each of these states is decreased if automatic fire suppression systems are
installed in the apartment. For post-flashover fires, the probabilities of fire spread to other
parts of the building are calculated, based on an analysis of the possible paths and the
probabilities of failure of compartment boundaries when subjected to fully developed fires.
Similarly, the probabilities of smoke spread to other areas of the building are calculated
using a smoke spread model. The smoke spread model is described in more detail later.
(2) Occupant Response and Egress Submodels- These sub models calculate the proba-
bilities of fire detection, warning, and egress of the occupants. Upon receipt of the warning
signals, the occupants are assumed to commence evacuation immediately. The time available
for evacuation is from the time when the occupants decide to leave until the time when the
conditions in the stair shafts become critical. Critical is defined as the condition when the
toxic gases and the temperature in the stair shafts are such that occupants trying to pass
through will be incapacitated. The present system model is based on the supposition that
prior to the occurrence of the critical condition in the stair shafts, the time-dependent
evacuation process occurs unimpeded by the effects of fire and smoke. At the occurrence
of the critical condition, it is assumed that the time-dependent evacuation process ceases,
and that the residual population in the building is then subjected to the effects of smoke
spread, and possible fire spread, throughout the building. The movement of the occupants
through the building is based on an empirically derived evacuation model [5]. The residual
population in the building and its distribution at the occurrence of the critical condition are
calculated and used to determine the number of expected deaths in the building.
(3) Expected Risk-to-Life Submodels- These submodels calculate the expected risk-to-
life based on the probabilities of death and safety in the building as a result of fire and
smoke spread in the building, the residual population in the building at the occurrence of
the critical condition, and the probability of fire starts over the design life of the building.
The probability of fire starts over the design life is assumed to follow the Poisson distribution
with the mean rate of annual fire starts based on statistical data. The expected risk-to-life
(ERL) is defined as the expected number of deaths over the design life of the building
divided by the building population and by the design life of the building.
(4) Fire-Cost Expectation Submodels- The expected fire cost is characterized in terms
of another parameter called fire-cost expectation (FCE), which is the expected direct costs
and losses divided by the total cost of the building. The direct costs and losses include the
capital cost for the provision of active and passive fire safety and protection facilities, the
maintenance and inspection costs associated with active fire safety and protection facilities,
and the expected monetary losses resulting from fire spread in the building.
Generic Apartment Buildings
The present apartment model uses a simplified layout of apartment buildings with two
categories of floor plans: (a) level of fire origin and (b) level, other than level of fire origin.
On the level of fire origin, the model considers two types of apartments: (a) the apartment
of fire origin and (b) the remaining apartments on that level which are aggregated into a
single compartment. On other levels, all apartments are combined into a single compartment.
The assumed floor plan layouts for the generic apartment building are shown in Fig. 2.
The apartment model is used to define the cost-effectiveness of various design features
that are present in generic apartment buildings. Design features that can be specified include
the building architectural layout, passive fire protection systems, active fire protection sys-
tems, fire alarm and detection systems, and the number of occupants. Details of the design
features that can be included were described in a previous paper [3].
Model Limitations
The modelling of the complex interaction between fire growth and spread, occupant
response and egress, life risks and fire costs in the present system model has been briefly
described in the previous section. As in many risk assessment models, certain assumptions
and approximations were made in the mathematical modelling due to the lack of sufficient
data. Because of this, the results obtained from the present system model can be regarded
only as estimates of reality. Under such circumstances, the proper use of the system model
is for comparisons of alternative fire protection designs. The process of such comparisons
is described later.
Until such time as the model is further developed, it is not suitable for risk-cost assessments
where the design criterion is based on achieving an acceptable level of risk on an absolute
basis. The present structure of the system model is such that it permits easy updating of the
sub models by simply changing individual submodels should additional information or data
become available. It is anticipated that the system model will follow a course of continuous
development as undoubtedly more information becomes available for improving individual
submodels. Eventually, as many of the uncertainties are removed, the model can be used
to assess fire risks and costs on an absolute basis.
Code Equivalency Comparisons

The performance of any building design is characterized by the system model in terms of
two decision-making parameters: expected risk-to-Iife (ERL) and fire-cost expectation (FCE)
which have been defined previously. The present approach of separating the life risks and
fire costs avoids the difficulties of assigning a monetary value to human life. It also allows
the comparison of various designs, in terms of relative risks and relative costs.
In the assessment of alternative design strategies, the following criterion can be used to
identify those that are considered equivalent to, or more cost effective than, the current
building code provisions: For an alternative design to be considered cost-effective, the fire-
cost expectation should be minimized subject to the constraint that the expected risk-to-Iife
value is similar to, and preferably lower than, the expected value based on current code
requirements for a particular apartment building under consideration.
It is recognized that procedural and nontechnical considerations can, and do, have a
significant role in the decision-making process. Accordingly, it is intended that while the
results developed by the present research will assist in the identification of cost-effective
alternative design strategies which are considered equivalent to the code provisions, the
results must be regarded only as an aid to the decision-making process.
Present Development
The present system model for apartment buildings has been described in the System Model
section. Although the basic framework of the apartment model has been developed, some
of the submodels are still being revised to provide a better representation of the fire and
smoke spread characteristics and human behavior in apartment buildings. The basic approach
in the development of these submodels is to adopt stochastic models that are based, where
possible, on either physical laws or test data. Only when there are no physical laws that can
be used or when there are no test data available, would the stochastic models be based on
input data from a panel of experts. As an example of this continuous development, the
smoke spread submodel, which was based on expert input data, was revised recently to be
based on time-independent experimental data [4]. For this study, the smoke spread model
was improved further to be based on a time-dependent mathematical model.
Smoke Spread Model

The smoke spread model is a simple, time-dependent model that calculates the spread of
the heat and the toxic gases (CO and CO2) from the apartment of fire origin to the corridor,
up the stair and elevator shafts, and then to the upper corridors and apartments. The model
cannot be described here in detail because of space limitations, but a paper has been
submitted for journal publication. The model is a simpler model when compared with other
more comprehensive models such as the Tanaka model [6], but the present model is faster
and is more suitable for highrise buildings with a significantly large number of floors and
an equally large number of compartments. In this model, each corridor, shaft, or apartment
is considered as a single compartment. In each compartment, conservation of energy and

mass is maintained. The force driving the hot gases from one compartment to the next is
mainly the buoyancy force. At the door to the stair or elevator shaft, the driving force is
corrected by the stack effect. On the floor of fire origin, the model allows the smoke to fill
the upper portion of the corridor above the door height first before it spreads to the stair
or elevator shafts.
Fire Growth Models

In the assessment of fire risks in a building, all possible fire scenarios should be considered.
The fires that can occur in a building, however, can vary in size in a wide spectrum, ranging
from small fires to fires that reach flashover and become fully developed. Since it is not
possible to consider all the possible fires in the assessment, these possible fires are divided
into two characteristic groups, called design fires. Group 1 consists of fires that reach
flashover and beyond, and Group 2 consists of all other fires that do not flashover. The
probabilities of occurrence of these two fire groups are obtained from statistical data. The
fire growth models developed for the risk-cost assessment model are used to predict the
behavior of typical fires in these two groups. These simple fire growth models characterize
the behavior of the two design fires as a function of the major design factors such as the
compartment size, the fuel load and whether the door is open or closed. Preliminary results
of these models were presented previously [7], and full details have been submitted for
journal publication. The predicted fire growth characteristics include the mass burning rate,
the temperature and the concentration of toxic gases (mainly CO and CO2) as a function
of time. As the fire growth calculations extend beyond flashover, the models account for
the difference in the production of CO between pre flashover and postflashover fires [7].
The effect of sprinklers on fire growth is modelled based on the recommendations of a
recent study conducted at the Warren Centre in Australia [8]. Based on this study, the effect
of sprinklers on large fires that without sprinklers will flashover, is to render them to small
nonflashover fires. That is, sprinklers will increase the probability of the occurrence of small
fires and decrease the probability of the occurrence of flashover fires. For nonflashover
fires, the sprinklers have no effect since whether the sprinklers operate or not, these fires
are still considered nonflashover fires.
In the case of a partial sprinkler system, the sprinklers are assumed to be installed in the
corridors only (usually a sprinkler is installed outside each apartment door). With such a
system, the effect on the fire is mainly on the fire plume extending from the apartment
door. If there is a flashover fire, the fire plume leaving the door would cause the sprinkler
to activate. Energy and mass balance calculations in the corridor yield the resulting tem-
perature and concentration of toxic gases as a result of the water spray. The amount of
water that will evaporate is obtained from experimental data {rom a room {ire test [9J. In
this test, the cooling effect as a result of an activated sprinkler located at the front window
was measured. If the design fire is a nonflashover fire, it is assumed that there would be
no fire plume leaving the door and the sprinkler will not activate.
Life Risk From Smoke Spread

The life risk to a person in the building as a result of smoke spread is measured by the
incapacitating effect of both the toxic gases (CO and CO") and the temperature of the hot
gases. The toxic effect of CO and CO" is measured by the fraction of incapacitating dose
of CO, multiplied by the hyperventilation effect of CO". The fraction of incapacitating dose
where KID is a constant, CO is carbon monoxide concentration in ppm, CO2 is carbon
dioxide concentration in percentage, and dt is the time increment in min. Occupants who
are awake will try to leave the building and will receive the FlD in the apartment, the
corridor, and the stair shaft. In the stair shafts, the time duration is assumed, conservatively,
to be the time required by a person to travel from the top floor to the fire floor. For all
other areas, the time duration is the time from the beginning of the fire. Occupants who
are asleep in their apartments will receive the FID in their apartments.
In addition to the incapacitating effect of the toxic gases, there is also an incapacitating
effect from the temperature of the hot gases. For this study, the critical untenable temper-
ature is assumed to be 100°C [11], and the fraction of incapacitating thermal dose (FTD) is
calculated using the following linear relationship:
where Ts is the hot smoke temperature in °C and To is the ambient temperature in 0c. In
this study, the FID and FTD values at the critical time are considered as equivalent to the
probability of incapacitation from toxic gases and from temperature, respectively. The com-
bined probability of incapacitation due to smoke is
where Pss is the probability of death as a result of smoke spread. The probability of death
in each compartment is used in the expected risk-to-life submodels to determine the number
of deaths in the building at the critical time when the conditions in the stair shafts become
untenable.
Results
The present risk-cost assessment model is still in the developmental stage. As an example
to show its potential application in the assessment of the cost-effectiveness of alternative
fire protection designs, the apartment model was applied to an existing highrise building
under renovation. The building is a 28-storey apartment building with 16 units on each floor
and two stairwells. (In the present system model, the apartments and stairwells are grouped
into the simple layout of Fig. 2.) The building has 1064 occupants, an average of 38 per
floor. At the present time, the building has a smoke detector in each apartment and a central
alarm system, but no sprinklers. The present design does not meet the code requirements
and the owner wants to improve the fire safety facilities as part of the renovation project.
He is considering the installation of a full sprinkler system, a partial sprinkler system (in
corridors only), and/or a higher reliability central alarm system (which has a detection and
warning reliability of 95% rather than the 75% used for ordinary systems). The risk-
cost model was used to assess all the possible combinations and the results are shown in
Table 1.
In Table 1, the expected risk-to-life (ERL), as defined earlier, is the expected number
of deaths over the lifetime of the building divided by the total population of the building
and by the design life of the building. The fire-cost expectation (FCE) is the expected fire
cost divided by the total cost of the building. The FCE values are based on new installation
costs and not retrofit costs, because the latter costs are very difficult to estimate. Never-
theless, the values shown in Table 1 can be used as a guide for considering retrofit options.
In Table 1, the ERL and FCE values have been normalized by dividing by those computed
for the reference option (existing option with no sprinklers but with central alarm). Normally,
the numbers are normalized by the computed results for the building code option to establish
equivalency. For this particular exercise, however, the intent is to search for a cost-effective
design to improve the safety level. As shown in Table 1, the ERL and FCE values for the
reference option are both equal to 1. A value greater than 1 indicates an increase in risk
or cost relative to the reference option whereas a value less than 1 indicates the opposite.
Table 1 shows that all the sprinkler options will reduce the risk but will increase the cost.
The full sprinkler option will reduce the risk much more than the partial sprinkler option,
but at a higher cost. In the present study, the installation cost of a partial sprinkler system
is assumed to be 60% of that of a full sprinkler system. This number is derived from an
assumption that the basic system cost for the pumps and the risers is about the same in a
partial sprinkler system as in a full sprinkler system. The only savings in a partial sprinkler
system is in the reduction of the number of sprinkler heads.
Table 1 also shows that for all the sprinkler options, the installation of a central alarm
system will reduce the risk, but will increase the cost. The option of a higher reliability
central alarm will reduce the risk further, but will not increase the cost (in the present
system model, the cost for a higher reliability central alarm is assumed to be about the same
as that of a regular central alarm). The most cost effective option shown in Table 1 is the
one with no sprinklers but with a higher reliability alarm. This option will reduce the risk
to a level comparable to those of the full sprinkler options but at a much lower cost. It
should be noted that the model assumes that the occupants in the building can move without
major difficulties. However, in this particular building, there are a number of older people,
disabled people, and young children who cannot evacuate at the speed assumed by the
model once they receive the warning signal. Therefore, it may be worthwhile to consider
the sprinkler options as they provide greater protection from fire and smoke spread in the
building. In a retrofit situation, the cost difference between converting to a higher reliability
central alarm system or adding a sprinkler system to an existing regular central alarm system
may be not as great as shown in Table 1. The conversion to a higher reliability central alarm
system may mean the full cost of a new installation; whereas the addition of a sprinkler
system to an existing regular central alarm system may mean only the incremental cost of
the sprinkler system.
Conclusions
The complex interactive process between human behavior and fire dynamics in apartment
buildings is modelled in a series of stochastic state-transition submodels, and interrelated
deterministic submodels. These submodels were integrated into a system model which is
used to calculate the effects of smoke and fire spread in highrise apartment buildings. The
effects of fire and smoke spread are calculated in terms of two performance parameters,
the expected risk-to-life (ERL), and the fire-cost expectation (FCE). The present approach
of separating the life risks and fire costs avoids the difficulties of assigning a monetary value
to human life. Furthermore, the code equivalency approach which has been adopted as the
basis for decision-making criterion allows the comparison of alternative designs with building
code requirements in terms of relative risks and relative costs. This was demonstrated by
applying the model to a 28-storey apartment building where the cost-effectiveness of 9
different combinations of alarm and sprinkler systems were assessed.
The system model can be used to:
• Identify alternative design configurations that give equivalent performance to the ex-
isting code requirements (in terms of ERL values), but at a lower net cost (FCE value);
that is, the alternative designs are more cost effective.
• Provide a performance-based approach to the design for fire. For example, designers
could be permitted to use any design configuration, provided it can be shown to give
an ERL value which is not greater than that based on the code-specified design re-
quirements.
• Assess both existing code requirements and proposals to change code requirements,
and to investigate whether consistent cost-effective performance is provided by the
various code requirements.
• Guide future research efforts into those areas which are identified as having a significant
impact on the cost-effective provision of fire safety and protection.
Acknowledgments
The authors would like to thank their colleague, Dr. H. Takeda, for performing the fire
growth calculations, and Mr. S. Amara for carrying out some of the computer runs.
References
[1] Beck, V. R., "A Cost-Effective Decision-Making Model for Building Fire Safety and Protection:"
Fire Safety Journal, Vol. 12, No.2, 1987, pp. 121-138.
[2] Beck, V. R. and Poon, S. L., "Results From A Cost-Effective Decision-Making Model for Building
Fire Safety and Protection," Fire Safety Journal, Vol. 13, No. 2-3, 1988, pp. 197-210.
[3] Beck, V. R. and Yung, D., "A Cost-Effective Risk-Assessment Model for Evaluating Fire Safety
and Protection in Canadian Apartment Buildings," Journal of Fire Protection Engineering, Vol.
2, No.3, 1990, pp. 65-74.
[4] Yung, D. and Beck, V. R., "A Risk-Cost Assessment Model for Evaluating Fire Risks and
Protection Costs in Apartment Buildings," Proceedings of the International Symposium on Fi~
Engineering for Building Structures and Safety, Institution of Engineers, Barton, Australia, 1989,
pp. 15-19.
[5] Paul, J. L., "Some Problems with the Movement of People in Buildings and Design Solutions for
Means of Egress," Proceedings of the 9th National Conference, Australian Fire Protection Asso-
ciation, Australian Fire Protection Association, Randwick, Australia, 1983.
[6] Tanaka, T. and Nakamura, K., "A Model for Predicting Smoke Transport in Buildings," Report
of the Building Research Institute, 123, Tsukuba, Japan, 1989.
[7] Takeda, H. and Yung, D., "Simplified Fire Growth Models for Risk-Cost Analysis," Fourth CIB
Workshop on Fire Modelling, National Institute of Standards and Technology, Gaithersburg, MD,
12-14 Feb., 1990.
[8] Task Group 1, "A Systematic Approach of Fire Safety Design of Buildings" and Task Group 4,
"Fire Initiation and Development," Fire Safety and Engineering, Technical Papers, Book 1, The
Warren Centre for Advanced Engineering, The University of Sydney, Sydney, Australia, Dec.
1989.
[9] Kim, A. K. and Lougheed, G. D., "The Protection of Glazing Systems with Dedicated Sprinklers,"
Journal of Fire Protection Engineering, Vol. 2, No.2, 1990, pp. 49-59.
[10] Purser, D. A., "Toxicity Assessment of Combustion Products," SFP E Handbook of Fire Protection
Engineering, National Fire Protection Association, Quincy, MA, 1988, pp. 1.200-1.245.
[11] Budnick, E. K., "Mobile Home Living Room Fire Studies: The Role of Interior Finish," NBSIR
78-1530, National Institute of Standards and Technology, Gaithersburg, MD, Sept. 1978.
Harold Katzin,l Marwan Khoury,l and AU Arlam"2
Fire Risk Analysis and Assessment for the

Canadian Building Code Assessment
Framework
REFERENCE: Katzin, H., Khoury, M., and Arlani, A., "Fire Risk Analysis and Assessment
for the Canadian Building Code Assessment Framework," Fire Hazard and Fire Risk Assessment
ABSTRACT: The Building Code Assessment Framework (BCAF) is an analytical tool which
systematically evaluates changes to the National and Provincial Building Codes in Canada. It
is intended to provide strategic advice to code developers. This paper discusses the fire risk
analysis and assessment approach that is part of the BCAF. This approach is a strategic one
that provides a systematic and rational method for comparing trade-offs between risk changes
and incremental costs as a result of code changes. A fire risk analysis event tree provides a
structure at a macro level to optimize the use of existing fire statistics and estimates of fire
risk event probabilities by a Delphi Panel. The use of both fire statistics and the Delphi Panel
in the analysis is described. The use of the Delphi Panel has proven viable and necessary for
applying the BCAF to more than 40 changes to the Ontario Building Code (OBC). The results
of analyzing a proposed code change to the OBC (mandating of sprinklers in dwellings) are
presented. The cost of mandating residential sprinklers per life saved is judged to be prohibitive.
The results of this assessment are compared to other studies. This approach to fire risk analysis
and assessment use in the BCAF has proven a viable one for the systematic and quantitative
assessment of fire risks because of proposed code changes and provides flexibility for incor-
porating other experimental fire risk analytical models.
KEY WORDS: fire risk, building codes, risk analysis, risk assessment, risk tree
Canada now has an analytical tool to evaluate systematically changes to the National
Building Code (NBC) and provincial building codes. This tool, called the Building Code
Assessment Framework (BCAF), was developed by the IEI Group and Trow Inc., under
the sponsorship of the Ministry of Housing of the Province of Ontario. The BCAF analyzes
any proposed code change, addition, or deletion in relation to the building code objectives
and assesses economic and social impacts, including fire and other risks. It performs these
analyses in consideration of 19 to 22 building types representative of the full range of building
construction in recent years in each of the provinces across Canada.
This paper is concerned with the fire risk analysis and assessment approach that is part
of the BCAF. The paper describes the BCAF in general terms, and then discusses in detail
the fire risk analysis and assessment approach in terms of its main characteristics: the fire
risk analysis tree, the analysis process, and the assessment method. As an application ex-
ample, the results of assessing the mandating of sprinklers in all new residential dwellings
lDirector and building science consultant, respectively, IBI Group, 230 Richmond St., West, 5th Fl.,
Toronto, Ontario, Canada, M5V IV6.
2Manager, Code Services, Ontario Ministry of Housing, 777 Bay St., 2nd Fl., Toronto, Ontario.
Canada, M5G 2E5.
234
KATZIN ET AL. ON CANADIAN BUILDING CODE ASSESSMENT FRAMEWORK 235
governed by the Ontario Building Code (OBC) are presented. In addition, a brief discussion
of possible applications of experimental fire risk computer models to supplement certain
aspects of the fire risk analysis process is provided.
There are a number of models that address the problem of estimating risk and economic
impacts of building codes. The U.S. Department of Commerce (National Institute of Stand-
ards and Technology, NIST, formerly the National Bureau of Standards, NBS) developed
a method for analyzing the cost-effectiveness of changes in building code requirements [1].
If adequate data on building construction, costs, and safety are available in a jurisdiction,
this method can be used to estimate costs and benefits of proposed code changes. This
method's approach to risk analysis is comparable to that of the BCAF.
Another model is the flexible approach to fire code compliance used in Boulder, CO
since the early 1980s [2]. Based on fire growth potential and success of extinguishment within
certain allowable building areas, the potential success (that is, acceptance by the authority
having jurisdiction) of a code requirement or equivalency can be graphically evaluated. This
model is not as detailed or comprehensive as the BCAF, but is a practical one that is
successfully applied.
Many manageable fire safety evaluation strategies and models incorporate, at least im-
plicitly, a form of hierarchial structure with Delphi generated databases. The Delphi tech-
nique [3] is used in a number of mathematical models to provide a quantification basis. A
typical model to relate fire safety evaluation and policy formulation for buildings is described
by Donegan et al. [4]. This is a hierarchical model, part of a fire safety points scheme for
dwellings. The model relates components affecting a fire-to-fire safety policy using basic
consensus data and deals with the interaction among components at various levels.
Other models include the cost-effective risk assessment model for fire safety and protection
in Canadian apartment buildings [5]. This model is based on a similar decision making model
that was developed to appraise the cost-effectiveness of provisions of fire safety and property
protection in Australian office buildings [6]. The effects of fire spread, and thus building
performance in these models, are measured in terms of expected risk-to-life and fire-cost
expectations. The models are based on a series of stochastic state-transition models and
interrelated deterministic models, integrated into a system model.
The Building Code Assessment Framework

The NBC and its equivalent provincial building codes are updated and modified through
various committees subject to a public review process. The varying make-up of these com-
mittees, and the lack of a rational and systematic approach to evaluating proposed code
changes limit the extent of objectivity that can be applied to these changes.
Amid a growing climate of regulatory reform, the Buildings Branch of the Ontario Ministry
of Housing undertook a number of projects to rationalize the building industry in the
province, which has the largest building construction dollar volume of all provinces in
Canada. Included was the development of the BCAF; a framework and procedure to evaluate
more rationally and systematically proposed changes to the NBC and the OBC. The BCAF
has been primarily designed as a strategic tool for use by experts with responsibilities for
code development at the Provincial and Federal levels. Since its development, it has already
been applied in assessing selected changes to the latest edition of the OBC. Through the
BCAF, code developers and policy makers are able to develop a greater level of confidence
that suggested changes fit within the mandate and objectives of the building code, are
worthwhile in terms of costs in relation to benefits, and ensure that risks are at acceptable
levels.
Structure of the BCAF

The BCAF has two components; the Code Analysis Component (CAC) and the Impact
Analysis Component (lAC). The CAC essentially identifies the objectives and requirements
of a proposed code change. These objectives are fundamentally related to ensuring fire
safety, structural sufficiency, and/or public health and welfare through the control of related
hazards (that is, sources of danger) and risks. The CAC contains an automated relational
database that organizes articles in the building code according to code objectives, code
requirements, and building spaces involved.
The lAC identifies the type and nature of potential cost and risk impacts of a proposed
change, and quantitatively measures and assesses these impacts. Any proposed change may
alter the level of risk associated with hazards addressed by building codes. To estimate this
change in risk, the lAC use a risk analysis and assessment procedure focusing on four major
hazards and their consequences.
• fire (deaths, injuries, property damage);

• structural failure/collapse (deaths, injuries, property damage);
• indoor air contamination (deaths, illnesses); and
• personal accidents (deaths, injuries).
The lAC contains various databases including sets of typical Reference Buildings (RB)
representative of the volume of annual new construction of all building types in each province
and representative of the share of fire losses, existing building stock and new building
construction, statistics related to hazards addressed by the building code, and expert risk
estimates by Delphi Panel members. Most databases and the lAC processes are automated
and enable modifications to be made on a modular basis as new statistics become available
and as judgments and opinions evolve. In fact, scientific and statistical information on
structural failure/collapse, indoor air contamination, and personal accidents is limited relative
to that for the incidence of fire and its consequences. It is accepted that modifications and
additions with respect to all hazards will be made as knowledge expands, thereby enhancing
the BCAF. These kinds of modifications and additions are fully anticipated and that is why
the BCAF is referred to as a "framework," that is, a structure to which updated knowledge
and techniques can be added.
Figure 1 shows the structure of the BCAF and a proposed code change evaluation process.
As indicated above, because of the more readily available fire statistics, the fire risk element
of the lAC is most developed.
A Strategic Approach to Risk Analysis and Assessment

The lAC's strategic approach to risk analysis and assessment provides a systematic and
rational approach for comparing trade-offs between fire risk increaselreduction and cost
increase/reduction. The costs and benefits resulting from this approach focus on society as
a whole, because often the benefactors of a code requirement are not necessarily those who
pay for it. The approach essentially looks at the net overall costs per benefit. Regardless
of which hazard is being considered, the approach to assessing the risk impact (including
that of fire risk) of a proposed code change consists of the following characteristics:
• An overall society viewpoint is taken .

• Regarding users of a specific building (that is, the group at risk), there may be a need
to identify the size of this group on a provincial or national basis; most causes (or group
of causes) of risks are considered, thereby enabling differentiation between as-built

causes and in-use causes, including human error. This in turn permits a focus on those
causes that can be affected by Canadian building codes .
• Data and expert judgement are used to develop base case (that is, existing code)
probabilities of exposure and to develop information on the extent of a hazard for each
building type in the RB set for each province.
• The construction volume of the reference year as a proportion of the overall building
stock is used in estimating risk impacts in new building construction relative to average
annual risk.
• Risk probabilities are developed on an annual basis based on statistics related to hazards
addressed by the building code, adjusted to relate to the volume of building construction
in the reference year. Changes in probabilities of each hazard cause and subsequent
events are estimated. Note that as a conservative measure, no allowance is made for
the relation of hazard and building age, condition, and level of design and construction
standards applied at the time of construction. Although the evaluation is done based
on new construction conditions and new code requirements, and although conditions
worsen over time, the fire risk database is continually updated and would necessarily
reflect the change in levels of risk as a result of the volume of new construction.
• For each hazard, the difference between the overall risk before the code change and
after the code change is applied to the base case incidence of death, injury, and property
damage to determine the overall impact of the code change on new building construction
in the reference year.
Fire Risk Analysis Tree

Risk for purposes of this work is defined as the possibility of occurrence of adverse
consequences for the building (that is, fire damage) or its occupants (that is, deaths and
injuries), or both. To analyze systematically fire risks, a fire risk analysis tree is provided
(see Fig. 2).
Events leading to a certain consequence are organized as branches which in turn form a
tree structure. Parallel branches on a tree represent mutually independent events, and each
branch is assigned a probability. End branches represent exposure-leading-to-consequence
events, that is, exposure-leading-to-death event or to injury, and so forth. End branches
representing no-deaths, no-injuries, or no-property damages are not shown on the tree. The
end branches dealing with the possibility of exposure leading to losses are superimposed
and therefore, the sum of probabilities of any four end branches is not necessarily equal to
one. The sum of probabilities of any other two parallel branches starting from the same
node is equal to one.
The product of a series of dependent branch probabilities results in a hazard path prob-
ability. The product of a given hazard path probability and the expected loss associated with
that path gives the path risk. The sum of risks as a result of all paths in a tree represents
the total risk caused by fire. For fire (or any other given hazard), each building type may
have a unique set of probability values comprising its tree. The following sections discuss
the rationale for selecting the risk tree structure and describe the fire events in the tree.
Fire Risk Tree as Simple Fire Event Model

The different fire events forming the fire risk analysis tree are defined for purposes of
the lAC as "mutually independent events." This does not apply in any given actual fire
scenario. For example, spread occurs only through destruction or only through convection.
The definition also does not mean that in an "actual fire" smoke spreads either to an exit
or to an adjacent compartment or, that if fire spreads, smoke does not. The nature of a
building fire will not lend itself to useful probabilistic analysis at the level required by the
lAC without some simplification. For this reason. it is assumed that fire spread is either
predominantly by destruction or predominantly by convection.
In his efforts to develop a simple logic for trading between fire safety measures. Harmathy
[7] presents a scheme for classifying fires following ignition that considers, for example. a
fully developed fire spreading by convection as a mutually independent event from a fully
developed fire spreading by destruction.
A number of factors led the fire experts on the development team of the BCAF to simplify
the relationships among the fire events described below and to consider these events as
independent. These reasons include the following:
• The need to develop risk event trees at a macro level where the structure is selected
to optimize the use of existing data and facilitate the estimation of probabilities and
parameters by members of the Delphi Panel (described below).
• The need to allow the use of results from existing and future detailed models (fire
spread models, evacuation time models) outside the basic tree and to allow for the
incorporation of these results within the macrostructure.
• The primary interest and the necessity to estimate risks associated with impacts of code
changes. The components and events of the fire risk tree had to reflect the building
features that could be affected by code changes (for example, effects on smoke spread
to means of egress only; effects of smoke spread to adjacent compartments, thus the
zonal separation of means of egress and other adjacent compartments/buildings; de-
struction of fire separations and closures).
• The requirement to limit the level of sophistication and detail that is practical in the
actual application of the fire risk tree as a result of limits on availability of historical
and other type data and the wide range of possible events and probabilities.
• The considerable effort that will be required to match the level of detail of a fire risk
tree (including the entire realm of possible fire events and their interdependencies
without any simplifications) to both the requirements of the lAC and the overall man-
date of the BCAF.
Above all, it must be remembered that the BCAF is a strategic tool. The changes in risk
probability are compared with broad estimates (on a logarithmic scale) of acceptable and
unacceptable risk probabilities and willingness to pay per year of life saved (see Fig. 3).
This permits a wide "factor of safety" in the calculations, while maintaining an acceptable
level of credibility in decision making.
Description of Fire Events

The first branch of the tree deals with the probability of ignition as the starting point of
the fire in a compartment (the basic building space addressed by the building code).
The second set of branches is concerned with the progress of fire, after ignition, within
the area of origin. Two mutually exclusive events are defined and considered for purposes
of risk analysis (as with all other branches). The fire may remain in the preflashover stage
or progress to postflashover.
Given that the fire does not progress beyond preflashover, the tree structure considers
the possibility of manual or automatic extinguishment. Occupant and/or fire fighter injuries
may result from exposure to such conditions and even death in some cases as a result of
smoke inhalation and/or burns. Property damage to building elements and components may
also result from these paths.
If the fire progresses to the postflashover stage, it can either remain confined to the
compartment of origin or spread by convection or destruction (assuming either a predom-
inantly convective or a predominantly destructive spread). Furthermore, this spread may
result in smoke only, or in fire and smoke in the areas that it occurs.
Spread was assumed to occur in either of two areas: means of egress or adjacent com-
partments/buildings. Therefore, any particular building may be schematically described as
FIG. 3-Comparative risk assessment graph for the Canadian building code assessment framework.
having three zones: compartment of origin, means of egress, and adjacent compartments
and buildings.
All paths in the fire risk analysis tree lead to four loss types (consequences) and, therefore,
fire risk analysis deals with four risks:
• risk of death to occupants, fire fighters, and others,

• risk of fire fighters' injuries (as a result of the fact that these injuries constitute on
average approximately 40% of the total reported injuries),
• risk of injury to occupants and others, and
• risk of property damage to building elements, components, and structures.
Method of Fire Risk Analysis and Assessment

This section discusses the use of statistics and expert opinion as a data source in the fire
risk analysis process, the method of application of the fire risk analysis process, the method
of application of the fire risk analysis tree, and the assessment method and use of results
in the lAC.
Use of Existing Statistics and Expert Opinion

There are two types of data that need to be obtained before the application of the fire
risk trees:
• the probability of different branch events per building type that may depend on the
design, construction, and occupancy, and
• the losses associated with a series of events (scenario); in the case of the fire tree, these
losses are deaths, fire fighters' injuries, occupant injury, and property damage.
Published and unpublished provincial statistics for the different building types providing
the probability of certain events such as ignition (all provinces) and confinement and/or
extinguishment (a few provinces) may be obtained. Similarly, available statistics may provide
the value of the total losses related to a given building type.
However, for a number of fire events along the tree, statistics are not available from
Canadian sources. The share of each fire scenario of the total losses in a building type is
also difficult to assess from statistics. To complement the statistical values available and
estimate the probabilities of events and loss shares, expert opinion is necessary. A Delphi
approach is used to estimate the needed probability values. This approach is described in
detail in the section below. Estimates from experts in the field are used as substitutes for
unavailable statistics until appropriate and reliable statistics are available for use in the lAC.
The Delphi Method

The Delphi Method can be defined as a method of "structuring a group communication
process so that the process is effective in allowing a group of individuals, as a whole, to
deal with a complex problem" [31·
This process is mainly concerned with the use of experts' opinions. The communication
process is usually structured so as to allow the following:
• a certain degree of anonymity for individual responses,

• feedback of individual contributions of information and knowledge,
• an assessment of the views and opinions contributed by experts, and
• some opportunity for individuals to revise judgment and views.
Applications
The Delphi Method was developed in the early 1950s to estimate the probable effects of
an atomic war on the United States. Subsequent applications in the 1960s were mainly
concerned with technological forecasting. Although technological forecasting remains one

of the major applications, the Delphi Method is used today in other areas in which judgmental
information is indispensable.
Applications include ascertaining values and preferences, simulated decision making, and
evaluating policy issues. This method is also used in management and operation research
in which there is a need to incorporate subjective information (risk analysis). Delphi has
also been used in solving societal problems in areas of health care, the environment, trans-
portation, and fire safety [3J.
One example of its use in fire safety evaluation has been in identifying cost-effective fire
safety retrofits in health care facilities [81. Another example is the use of Delphi input for
decisions concerning the equivalence of fire safety alternatives [9J and the use of Delphi in
the optimization of fire safety decisions [101.
Delphi Approach in Risk Analysis

The Delphi approach to the risk estimation used in the fire risk analysis of the lAC is
outlined below. Note that the base case risk for each building type is carried only once.
Also, for purpose of efficiency, panel member comments are obtained for a "batch" of
proposed code changes (that is, several changes are presented at once, each one being
evaluated independently).
1. A panel of 30 to 35 fire experts is assembled. The types of backgrounds reflected in

the participant group include fire service officials, building officials, building scientists,
building researchers, members of building industry associations, and private consultants.
Regional representation is also a consideration in the selection process.
2. A draft questionnaire is designed and pretested on a few of the participants. The final
questionnaire is developed taking into consideration the initial comments and feedback. The
structure of the final questionnaire used to obtain estimates of existing (base case) risk
probabilities is as follows:
• The overall purpose and mandate of the BCAF is described.

• The components of the BCAF are described in general terms.
• The approach to fire risk analysis and assessment is discussed.
• Definitions of all fire events that are part of the fire risk tree are provided.
• Schematic drawings and basic construction characteristics of the reference buildings are
provided.
• Tables corresponding to each fire event are provided for the reference building in
question. The expert inputs his/her estimate in these tables.
• Tables to allow the estimation of a particular scenario's (path) percentage share of fire
losses for each reference building are also provided.
3. Members of the Delphi panel are then interviewed on an individual basis. The purpose
of these interviews is to provide the experts with the questionnaire and answer any questions
related to its structure. During these meetings, the approach to risk estimation is explained
as per the questionnaire described above and the questionnaire is discussed. The question-
naire is then left with the participant to fill out. The estimates required are related to existing
(or base case) risks in different building types. Experts are asked to give their best estimates
based on knowledge, experience, and typical characteristics of building types involved.
4. A preliminary analysis of the results is carried out. This includes a statistical analysis
and qualitative assessment of participants' comments and suggestions. A consultation paper
is then prepared incorporating a statistical summary (mainly the mean and standard devia-
tion) and comments from the first round are sent back to the participants allowing for a
second round of estimation. To assure anonymity, the statistical summary does not indicate
any names, and all comments are edited so as to maintain confidentiality.
5. In a typical Delphi survey, reiteration continues until consensus is reached before
preparing a final analysis. Consensus is assumed to have been reached when a certain
percentage of votes fall within a prescribed range (for example, two units on a ten-unit
scale) [3]. In this case, a second round could be considered sufficient to reach an acceptable
consensus between panel members on base case estimates. A standard deviation of less than
15% is considered satisfactory for purposes of the BCAF (see discussion below).
6. For a proposed code change, a similar process to the one described above applies.
Experts are asked to evaluate the change in base case risk estimates (based on consensus)
as a result of a requirement(s) of the code change. The experts are provided with a ques-
tionnaire structured as follows:
• a description of the code change (that is, existing article and changes as provided by
the official code development agency);
• the reference buildings affected by a change and conceptual design and construction
possibilities of how a change can be implemented; and
• base case risk trees of the reference buildings affected, with the event and branch
probabilities for the base case (consensus values) shown on them.
The interviewer presents the questionnaire, answers any questions related to the process
and the code changes, and leaves the questionnaire with the expert. The expert changes
any branch event probabilities he/she judges affected. A similar reiterative process and
statistical analysis is carried out until consensus is reached (standard deviation of estimates
is less than 15%).
Statistical Analysis of Delphi Estimates

The statistical analysis is based on a population of 30 or more, and is, therefore, carried
out according to a large sample size.
For each estimate, the mean, standard deviation, and median are calculated. The fre-
quency and distribution of estimates is determined and plotted graphically. This allows noting
large scattering of estimates and which participants may have caused this. Scattering is
expected after the first round because of different backgrounds and views, and in this case,
because of regional differences as well.
In addition to the above, a 95% confidence interval is determined for the mean based on
maximum error of estimate. Based on this range, the percentage of estimates falling within
is determined as another indication of reaching consensus. The degree of skewness of the
results is also determined.
The above analyses make it easier to determine which estimates may not be part of the
consensus group and enable the second round Delphi to concentrate on these areas and on
comments from such participants.
Two final points should be noted. First, the estimates are given in 5% intervals. Second,
in a transportation study using the Delphi method, the standard deviations after the third
round were between 10 and 24% (16% on average) [11]. This may give an indication of an
acceptable deviation in this case and which may be below 15%. In both the base case risk
estimation process and the code change risk estimation process, standard deviations after
the second round only varied between 3 and 15% and were considered acceptable.
Method of Application of Fire Risk Analysis Tree

The following is a step-by-step description of the process used to determine a change in
risk (fire deaths in this case) as a result of a proposed code change and for one particular
building type.
Base Case Risk Estimation

Step i- Assign base case branch probabilities based on statistics and/or expert opinion.
Step 2-Calculate path probabilities for the base case; path probability is the product of
branch probabilities along that path.
Step 3-Knowing from statistics that on average there are, for example, 0.013 deaths
occurring per fire in a certain RB type, the analyst distributes this risk among those paths
ending with a death according to expert estimation.
Loss Constants Calculation

Step 4-Calculate loss constants for each death path. A loss constant equals the base case
path risk (Step 3) divided by the base case path probability (Step 2).
Code Change Risk Estimation

Step 5- Assign new branch probabilities as a result of a code change as estimated by
experts.
Step 6-Calculate new path probabilities for each death path.
Step 7-Calculate new path risks by multiplying the new path probabilities (Step 6) by
the corresponding death path loss constants (Step 4).
Step 8-Sum up all new path risks obtained from Step 7 to get the new death risk caused
by code change.
incremental Change in Risk

Step 9-Calculate incremental change in death risk. This equals the difference between
the base case risk of 0.013 (Step 3) and a new risk determined from Step 8 (assume 0.0152).
Therefore, the incremental change in death risk as a result of the code change in this
hypothetical example is an increase of 0.0022 deaths per average fire in that building type.
The changes in risks related to fire fighters' injuries, occupants' injuries, and property
damage are assessed in the same manner.
Method of Assessment and Use of Results in iAC

The incremental changes in risks determined from Step 9 is the value that is transferred
and used in the lAC. Such a value is changed to years of life gained/lost over the life cycle
of the building based on the average age of a casualty in a building fire and the individual's
average life span. It is then prorated to represent the total volume of new construction of
a certain building type.
Other risks such as injuries are changed to monetary values and, with property damage,
are prorated to total new construction.
To assess the socio-economic impacts in terms of overall societal expectations and stand-
ards, the results are applied to the Comparative Risk Assessment Graph (see Fig. 3). This
graph is based on fundamentals of risk appraisal developed by a Swiss consulting engineering

firm [12,13]. The fundamental concepts have been adapted to the Canadian context for
purposes of the BCAF. Based on the graph, the dollar costs/savings per year of life saved/
lost and the probability of individual risk can be assessed according to established societal
standards corresponding to different risk categories and degree of an individual's control
over his or her environment. This method enables a quantitative assessment of the results
of the lAC.
Applications
The BCAF has already been applied to more than 40 proposed changes to the OBC, and
the results of the analysis have been taken into consideration by code development decision
makers. This paper presents the results of the assessment of one of the proposed code
changes requiring that all residential dwellings governed by Part 9 of the OBC have fire
sprinklers designed and installed in conformance with NFP A 13D Standard for the Instal-
lations of Sprinkler Systems in One and Two Family Dwellings and Mobile Homes. The
analysis and assessment concluded the following:
• The dollar cost to society per year of life saved is around $950 922 (1989 dollars).
• Deaths are reduced by 67% in dwellings with sprinklers, which is comparable to a
reduction of 63% estimated in a NIST study on sprinklers [14].
• Dollar savings as a result of reduced number of injuries and reduced property damage
are only about 0.5% of the costs involved in mandating sprinklers.
• The cost per life saved is around $35 180 000, which is comparable to the cost per life
saved estimated independently by a Canada Mortgage and Housing and Corporation
study.
• This cost is clearly higher than what society is willing to pay as determined from the
Comparative Risk Assessment Graph.
Flexibility of Framework
The overall approach and the BCAF are both flexible enough to incorporate future
modifications and/or adjustments. Experimental fire risk analytical models such as HAZ-
ARD I (Centre for Fire Research, NIST) can be used to supplement the fire risk analysis
and assessment approach. Possible uses of such models include the following:
• validating and complementing statistical databases,

• complementing expert judgment,
• predicting risks for specific conditions,
• predicting path probabilities, and
• providing worst case/best case scenarios for use by the Delphi Panel.
The validity of HAZARD I scenarios in a Canadian context are being considered before
attempting to make use of it for tempering the results of the BCAF. The limitation of
HAZARD I modelling of the OBC and the extent to which it can be directly linked to the
BCAF are also being studied.
Further assessment of possible uses of HAZARD I and other experimental models with
the BCAF can prove vital to enhance the BCAF to suit new users and new applications.
Conclusions
The BCAF's approach to fire risk analysis and assessment provides a rational approach
at a macro level to assess the fire risk impacts of building code changes.
The approach allows an optimum use of existing statistics and relies on a Delphi Panel
to complement the statistical data and estimate fire event risk probabilities. The use of a
Delphi Panel has proven viable and necessary at this stage for applying the BCAF to more
than 40 changes to the GBe. This is supported by the results of assessment of mandating
residential sprinklers which indicate a prohibitive cost per life saved and which are com-
parable to results from other studies.
The structure of the fire risk tree and the lAC allows for future integration of more
statistical data to replace the Delphi values and is flexible enough to incorporate future
modifications and/or adjustments and integrate experimental fire risk models.
In conclusion, the fire risk analysis and assessment approach used in the Canadian BCAF
has proven to be a viable method for the systematic and quantitative assessment of fire risks
as a result of proposed code changes. The results of its application have generally been in
accordance with expectations of code policy makers and with results of other assessments.
References
[1] Rawie, C C, "Estimating Economic Impacts of Building Codes," NBSIR 81-2402, United States
Department of Commerce, National Institute of Standards and Technology (formerly National
Bureau of Standards), Gaithersburg, MD, 1981.
[2] Harvey, C S., "A Flexible Approach to Fire-Code Compliance," Architectural Record, 1988, pp.
130-135.
[3] Linestone, H. A. and Turoff, M., Eds., The Delphi Method, Techniques and Applications,
Addison-Wesley Publishing Co., Reading, MA, 1975.
[4] Donegan, H. A .. Shields, T. J., and Silcock, G. W., "A Mathematical Strategy to Relate Fire
Safety Evaluation and Fire Safety Policy Formulation for Buildings," in Proceedings of Fire Safety
Science. the 2nd International Symposium, Hemisphere Publishing Corp., 1988, pp. 433-441.
[5] Beck, V. R. and Yung. D., "A Cost-Effective Risk-Assessment Model for Evaluating Fire Safety
and Protection in Canadian Apartment Buildings," paper submitted for presentation at the In-
ternational Fire Protection Engineering Institute-V, Ottawa, 21-31 May 1989, National Research
Council of Canada. Ottawa, Jan. 1989.
[6] Beck, V. R .. "A Cost-Effective, Decision-Making Model for Building Fire Safety and Protection."
Fire Safety Journal. Vol. 12, 1987, pp. 121-138.
[7] Harmathy. T. Z .. "A Suggested Logic for Trading Between Fire-Safety Measures," Fire & Ma-
terials. Vol. 10. 1986, pp. 141-143.
[8] Chapman. R. E .. Hall. W. G .. and Chen. P. T .. "A Computerized Approach for Identifying Cost-
Effective Fire Safety Retrofits in Health Care Facilities," NBSIR 79-1923, National Institute of
Standards and Technology (formerly National Bureau of Standards). Gaithersburg. MD. 1979.
[9] Harmathy. T. Z .. "A Decision Logic for Trading Between Fire Safety Measures by Delphi Group,"
paper to be submitted to Fire & Materials. revised March 1988.
[10] Harmathy. T. Z .. 'The Delphi Method-A Complement to Research," Fire & Materials. Vol.
6. No.2. 1982. pp. 76-79.
[11] Wilson D. and Schofer J .. "Decision-Maker-Defined Cost Effectiveness Framework for Highway
Programming." Transportation and Research Record. 677, Transportation Research Board. Wash-
ington. DC, 1978.
[12] Merz. H. A. and Schneider Th .. "Who is Afraid of Risk Criteriary,'· Hazard Prevention. Jan./Feb.
1987. pp. R-13.
[13] Bohnenblust. H. and Schneider Th .. "From Legal Principles to Engineering Decisions: Where Do
We Shift to Quantifications')." paper presented at the V. International Safety Summer Symposium
on Safety Science. Leuven. Belgium. 1983.
[14] Ruegg. R. T. and Fuller. S. K.. "A Benefit-Cost Model of Residential Fire Sprinkler Systems,"
N BS Technical Note 1203. United States Department of Commerce. National Institute of Standards
and Technology (formerly National Bureau of Standards). Gaithersburg. MD. 19R4.
Clifford S. Harvey!
The Measurement of Building Fire Safety-

Example of the Application of an Existing
Method for Fire Hazard and Fire Risk
Assessment at the Community Level
REFERENCE: Harvey. C. S .. "The Measurement of Building Fire Safety-Example of the
Application of an Existing Method for Fire Hazard and Fire Risk Assessment at the Community
Level," Fire Hazard and Fire Risk Assessment. ASTM STP 1150. Marcelo M. Hirschler. Ed ..
American Society for Testing and Materials. Philadelphia. 1'i92. pp. 24H-254.
ABSTRACT: A proven method for the evaluation of community fire protection was imple-
mented to accomplish fire and life safety goals without the usual increases in cost for equipment
and personnel. The method increases relative protection. yet demonstrates an actual decrease
in the cost of new construction. Entry into the system was voluntary during the first years of
activity; participation became mandatory when the savings became clear and no significant
opposition was present. The method relies on the understanding of equivalency concepts for
code compliance.
KEY WORDS: measurement. equivalency. fire safety. life safety. goals. management. com-
munity fire protection
Those readers expecting this paper to be an engineering problem with an engineering

solution will be disappointed. As construction costs continue to rise, fire safety design will
be driven by the need for mutual respect and understanding between fire professionals and
the community they serve. Buildings designed as fire safety systems will be less expensive
to build and more user friendly. None of the adopted codes will be met to the letter, but
all will be met as to their intent. Although firmly based on a sound understanding of the
dynamics, growth, and control of fire, this paper will discuss the translation of those under-
standings into the philosophical approach necessary for acceptance by the community. Im-
plemented successfully, the method provides a solution for curbing increasing costs of manual
fire suppression. This paper will focus entirely on Boulder CO; the method, however, is
germane to communities of all sizes.
Discussion of the Problem

In 1968, Boulder had three fire stations housing four engines, one truck company, one
squad, and 42 total members on three shifts. As the city began to grow, no additional monies
were provided to increase the fire fighting staff or equipment. The need became so critical
that, in 1976, a study was undertaken to determine what Boulder actually needed for fire
protection, and further to explore innovative approaches for providing that protection. The
study group consisted of representatives from the local firefighters union, the fire chiefs
'Assistant fire chief. Fire Department. Boulder. CO.
248
HARVEY ON MEASUREMENT OF BUILDING FIRE SAFETY 249
office, and the city administration. The goal was to develop a plan to provide adequate
levels of fire protection, without the expected corresponding increase in dollars spent.
Development of the Solution
Boulder's residential areas posed one problem; they were spread out in such a manner
that acceptable response times to emergencies were many times not possible. The commercial
growth was another concern. New construction was often beyond the fire fighting capabilities
of the community even though new buildings were being built in compliance with national
codes. A single method of addressing both existing problems and future concerns was
necessary if the community was to avoid the almost certain future fire disasters. Enter the
active thought process of "designing" the acceptable fire.
In the mid-70s two members of the fire department had attended a college level course
entitled The Measurement of Building Fire Safety at Worcester Polytechnic Institute in
Massachusetts. This course had been developed by Prof. Robert Fitzgerald and Rexford
Wilson, both of whom spent a great deal of time fine tuning the concepts before offering
the first classes. The methodology had its roots in the systems concepts approach then being
evaluated and molded by the National Fire Protection Association's (NFPAs) Technical
Committee for Fire Protection in Structures. The full range of fire development and extin-
guishment, from overheat, through full building involvement, and on to complete extin-
guishment, was thoroughly discussed. Each event having an affect on how a fire grows and
what slows that growth was discussed and evaluated. At the end of the course, each par-
ticipant had the skills to evaluate, that is, place a numerical value on, all conditions con-
tributing to the growth and extinguishment of any fire. With this knowledge the user could
develop a strategy for community fire protection based on the inherent strengths and weak-
nesses it had built into its codes and manual fire fighting forces.
The measurement of building fire safety describes a series of events, each of which must
be evaluated before a determination of fire safety can be made. The events which are a
part of the evaluation process are too numerous to discuss here, so it is strongly recommended
that everyone reading this paper attend the next course session offered. A short overview
is, however, helpful and necessary. Most importantly, remember that a fire will go out in
one or more of only three ways.
Evaluating the Self-Termination Approach
All fires eventually go out by themselves. Because the fuel is dry and in close proximity
to other fuel some fires do not go out until they run out of fuel. For others, because their
fuel is densely packed, and in small amounts some distances from other fuel, fires go out
without involving other fuel. When something about the dynamics of fire is known one is
able to evaluate the probability that a certain room will or will not go to full room involve-
ment. A weak fire room will have a low value, while a strong fire room will have a high
value.
Using the methods and concepts learned at Worcester Polytechnic Institute (WPI), the
group started its focus by evaluating how easy it would be to control the potential for any
fire occurring in Boulder. It was quickly seen that this was highly improbable to any mea-
surable degree, except through the work of a strong fire code that was vigorously enforced.
Even with a strong fire code, it was impossible to control the type, amount, and configuration
of fuel in any space. A strong, vigorously enforced fire code will, however, allow a fire
department to leave a strong fire prevention message with its residents, in a manner we
usually refer to as the yearly fire safety inspection. During this interface with the community,
engine companies and other fire prevention teams are able to have some effect on the
configuration and type of fuel in the space being inspected, and can therefore have some
level of control over the fire spread potential, although we know this control is limited even
under ideal conditions.
Evaluating the Automatic Sprinkler Approach

Next, a fire can be extinguished due to the efforts of an automatic sprinkler system.
Assuming the sprinkler system has been properly designed, properly installed, and is in
service, we know the system has a very high probability of successful operation. A poorly
designed or maintained system has a lesser chance of successful operation, and would show
a lower probability of success.
The group already had a great deal of information concerning the performance of sprinkler
systems. All the information seemed to point to one thing; an adequate sprinkler system is
98 to 99% effective in controlling a fire. Further, multiple life losses in fully sprinklered
buildings are unheard of. The group was comfortable with the WPI concepts, so they
evaluated all the events which must occur to have a strong value for success of the sprinkler
system. They found the success of almost all of those events was assured during the design,
installation, and general maintenance of the system. They also recognized how valuable
sprinkler protection could be when folded into a community fire protection plan.
Evaluating the Manual Attack Approach

Lastly, a fire has some probability of being extinguished by the local fire department,
who accomplish the job manually. Although people usually see their local firefighters as
able to do any job they are asked to do regardless of its size, it is clear from studying the
events for successful manual extinguishment that local fire departments are two things; they
are crucial to ensuring that their community does not burn to the ground, but of very limited
value when it comes to extinguishing fire. Any local fire fighting force can be evaluated.
The many events leading to successful extinguishment of any fire can be evaluated, and a
numerical value indicating the relative strength or weakness of each event occurring suc-
cessfully can be assigned.
The study group began to focus on manual suppression and the realistic effect the fire
department could have on a fire they were asked to extinguish. Using current thinking and
technology concerning how much equipment and personnel it took to fight a fire, it was
determined that Boulder's capability at the time was less than 2 000 ft2 (185.8 m2).
As this area seemed rather small to all of us, we did some research into past fires and a
detailed analysis was performed using the measurement concepts learned at WPI. Not only
did our past experience verify our actual capabilities to be less than the 2 000 ft2 (185.8 m2),
but the work done with the measurement concepts showed us where we might improve our
service delivery. With those changes made, we pegged our capability at 2000 ft2 (185.8 m2).
The actual cost of providing fire suppression services was also discussed in detail. In
Boulder, the average yearly cost of providing just one firefighter on duty is about $45 000.
On a three platoon schedule, providing one person on duty means hiring four people. That
amounts to $180 000 per year for that one position on one fire engine. So the group had to
keep the total number of firefighters to the absolute minimum while expanding the services
they provided.
When one has completed evaluations and assigned values to each of the three probabilities
that a fire will be extinguished, those three main events (self-termination, automatic sprin-
klers, and manual attack) are combined for a final, cumulative limit of flame spread value.
If the people are willing to accept the levels of actual capability determined to exist, no
other work is necessary. However, if they believe those levels are not acceptable, decisions
need to be made to elevate the community to a higher level that is acceptable. This was the
task of our study group.
The Measurement Approach to Developing a City Fire Protection Plan

Now that the group had evaluated the various ways a fire could be extinguished and had
seen the relative strengths and weaknesses of each one, it set out to develop a community
fire protection delivery plan which recognized those facts, that is, one which discouraged
redundancy and encouraged equivalency through the use of sprinklers.
It was clear a fire prevention inspection program, no matter how rigorous, could only do
so much to prevent a fire. The fire fighting capability of our community was limited, and
expensive to increase to any noticeable degree. They further saw the protection provided
by sprinklers, protection which was ongoing year after year at only maintenance costs after
installation. In order to avoid as much staffing costs as possible, the group recommended
that the city fathers build two small residential type fire stations, housing two-person engine
companies. They recommended reducing the total number of people on other equipment,
from four to three. They recommended that new equipment be purchased which would do
much of the traditional human work, thus allowing the reduced levels of personnel to
concentrate on the tasks their equipment could not perform. And they recommended a
strong emphasis be placed on convincing people of the value of sprinkler protection. The
group stopped short of recommending a sprinkler ordinance, even though they felt it was
appropriate. They felt it would better serve the community to let it see for itself what the
voluntary installation of sprinkler protection, in return for a large number of tradeoffs, could
accomplish. This was a far sighted recommendation, one which we believe made the ultimate
implementation of a sprinkler ordinance virtually painless.
Implementation of the Study Group's Recommendations

In 1978, the city opened a fourth fire station, and a year later opened a fifth and sixth.
It is important to recognize these were capital expenditures, and included only limited
additional personnel costs. Two of the three stations were situated in predominantly resi-
dential areas, and housed small pieces of equipment with two people assigned to them. They
were limited in their capability, but were closer to the fire or medical call, thus better able
to handle the call with fewer people.
This recommendation has concerned our fire fighting personnel for some time, but sta-
tistically the decision has been very successful. At no time have either of the two, two-
person pumpers ever come upon a situation they were not able to deal with until a backup
engine arrived.
The saving on personnel alone currently amounts to over four million dollars. This, with
no increase in fire losses or deaths.
The other recommendation the city fathers adopted was more philosophical in nature.
They agreed that if the city was going to have codes, they were going to be enforced. After
a presentation concerning equivalency, they endorsed the concept of '"meet the codes or
meet the curve," which referred to the graphical curves used to illustrate the strengths and
weaknesses arrived at when evaluating the three ways fire could be extinguished. Both the
Building and Fire Departments were given a mandate that people needed to meet the codes
to the letter or, in lieu of that. meet the intent of those codes through equivalency.
Recognize that this is a fundamentally different approach than simply passing a sprinkler
ordinance.
The fire department, having been schooled in the art of equivalency determination. felt
very comfortable making decisions of this type. For example, if a sprinkler system is installed
where not otherwise required, it can take the place of rated walls, rated interior finishes,
wire glass, and so forth. Either one or the other can be used to limit the spread of fire. The
Building Department had a harder time with this, and since we were trading off their code
requirements as well as ours, they felt very uncomfortable with the whole issue until they
had received more training. Further, builders initially felt we were mandating sprinkler
protection where it was not required. When we explained they had a choice to either meet
the code to the letter or provide sprinklers in lieu of many code requirements, they felt
more comfortable. It was only after one builder embraced the program that its acceptance
was ensured.
Case In Point- Builders and Developers

We were presented with plans for a five story, 50 OOO-ff(1415.85 m~) office building.
The codes in place did not require the building to be sprinklered. They did, however, require
the corridors to be one-hour rated. They also required rated interior finish, enclosed stair-
ways, maximum dead end distances, secondary fire department equipment access, proper
numbers and locations of hydrants, and so forth.
We asked the developer to do a spreadsheet on the building which showed it meeting
code to the letter. That meant all the Building and Fire Codes, and all the Fire Department
requirements for hydrants, access, standpipes, and so forth, had to be met to the letter. He
prepared the spreadsheet and showed us what it was going to cost him to build the building.
We then asked him to prepare another spreadsheet showing the voluntary installation of
a full sprinkler system, but this time allowing for the list of trade-offs the City would allow
for the building being fully sprinklered.
The difference in the two spreadsheets amounted to $50 000, or $10 000 per floor. This
was the amount he would save if he sprinklered and took advantage of the trade-offs.
Needless to say he provided the sprinkler protection. Word spreao quickly, and virtually
all new construction was presented, up front, as voluntarily sprinklered property.
Case In Point-Architects and Designers

Several architectural firms approached us concerning design of Planned Unit Develop-
ments (PUDs). They knew we were encouraging equivalency to written codes, something
they felt would allow their design juices to flow freely. These firms also knew the Fire
Department requirements, and felt many of these requirements might be eliminated or
diluted if a building was self-protected with a sprinkler system. We worked with these firms,
and were able to eliminate some of the required access, the required turn around, and
several of the required fire hydrants. This meant they could design the building and its
surrounding land area to work for the users of the building, not for massive fire department
operations which would no longer be necessary. They too now felt comfortable with our
method of determining equivalency.
Case In Point-City, Land, and Urban Planners

Since becoming proactive instead of reactive, the Fire Department has become highly
visible in the community. No higher visibility was accomplished than with the Planning
Board itself. With our strict enforcement and interpretation of the fire code, we were accused
of driving the design process in the community with our requirements. When the Planning
Board wanted a building situated in a certain section of a lot, doing so invariably triggered
a fire department requirement for additional access or hydrants. Houses behind houses are
a popular concept in Boulder now, but that sort of configuration creates a situation that is
unmanageable for our fire department. With the limited personnel in the department, we
can not fight fires that are not easily accessible. With the advent of meet the code or meet
the curve, however, the Planning Board embraced the concept, because building projects
they wanted to rearrange were routinely coming through as sprinkle red projects, not trig-
gering any additional requirements of the Fire Department. They too have actively endorsed
what we are doing with the measurement concepts.
Codifying Every thing-A Good Faith Move
The system of voluntary sprinklers in return for trade-offs worked very well. but it took
a greal deal of staff time to continually reinvent the wheel for new people. After seven years
of working with a voluntary, optional system, we felt comfortable that those involved in the
building industry trusted us, trusted equivalency. and liked the way the use of equivalency
smoothed out the whole process. When we proposed to require sprinkler protection however,
there was some immediate flinching from those same people. People in all phases of the
building industry had a concern. What happens if Chief Harvey and the Building Department
personnel leave the City. Who will replace them, and will they have the same philosophy
that their predecessors had? This was a valid concern. We drew up an ordinance which not
only required sprinkler protection in most new (and some existing) construction, but one
which codified the trade-offs, so when my counterparts in the Building Department or I
should leave the City, the construction disciplines will be protected.
How is this different from what other communities have done? It is categorically different.
The usual way a sprinkler ordinance is passed is this; today we do not require more than
the national codes, tomorrow we will. This sort of mandated, brick wall approach creates
anger, frustration, and mistrust in the community. Our way was to slowly get people up to
the level of full sprinkler protection through voluntary installations with trade-offs, let them
get comfortable with it, then pass an ordinance which, in essence, says that business as usual
will continue. Our public process was the smoothest of any community with which I have
communicated. Usually there is a great clamor of complaint; "I can not afford it, and so
forth." Before we made anything mandatory, we showed the community they could not
afford not to. Our ordinance passed with unanimous approval. something unheard of these
days. Further, we had representatives from the architectural. planning, developing, and
building ownership arenas standing up and supporting what we were doing. Truly a win-
win situation.
What Should We Have Done Differently?
There are some inherent hazards to approaching community fire protection in the manner
we did, hazards that were not apparent when we first began. First and foremost, a fire
department and its members become highly visible in the building community when un-
dertaking a project such as this. Until and unless all parties affected by the process feel
comfortable with the concepts, there will be distrust and anger. Builders may feel you are
mandating installations that are not required. Your Building Department may be very
uncomfortable leaving the comfort of the written word in favor of something else. Planners
may feel you are placing their concerns for aesthetically pleasing results on the back burner
when you mandate code to the letter. The answer to calming all these fears is found in one
word, education. Hold classes and seminars for whomever will attend. Include city fathers
and people from other city departments who might not even be involved. The more people
you can expose to this concept, the less trouble you will have with people's understanding
of it.
Secondly, and almost more importantly, the question of consistency in code enforcement
will most certainly surface. In the old days, an office building was an office building, and
all office buildings had to meet the same standards, even though those requirements might
not be necessary or apply to a specific office building. Certainly, everyone can see the
differences between a three story, 30000-ftc (S49.51 mC) hotel, and a 50 story, 3000000
ftc (84 951 mC) hotel. The codes might have identical requirements for both buildings, but
the buildings should be treated differently. Although they have the same function on the
surface, one is clearly dealing with a different problem than the other, and should be dealt
with differently. Even though this is pure logic. and pure use of a systems approach, many
citizens will not take the time to try and understand it.
Conclusions
Successful fire protection and life safety design depend on all areas of the many construc-
tion disciplines in our communities getting together with each other, including planners,
building and fire officials, owners, designers, and builders. New engineering methods and
techniques will certainly smooth the implementation of, and enhance, fire protection de-
cisions made by those communities. People will have to consciously decide what level of
protection from fire they prefer, and actively pursue the design of future buildings with that
philosophy in mind. A strong understanding of the methods used to determine specifically
where any community stands from a fire protection standpoint is critical, and using the
Measurement of Building Firesafety is an excellent way to accomplish this. Only when a
community knows its actual fire protection capabilities, will it be able to determine where
it wants those capabilities to go. And only after determining that, will they be able to design
new and remodeled buildings to meet those goals. Buildings in the future will be designed
as systems rather than structures, and such design criteria will see buildings which not only
hold the line from a cost standpoint. but which are easier to use for their intended purposes.
Bibliography
Fitzgerald. R. and Wilson. R .. The Measuremelll of Building Firesafety, course workbook and study
guide, 1975 and 1976 editions.
Lucht, D. A .. "The Role of the Systems Concept in Firesafety," New Approaches to Evaluating Firesafety
in Buildings, Proceedings of National Academy of Sciences Conference, National Academy of
Sciences. Office of Publications, Washington, DC, Sept. 1978, pp. 5-19.
Roux, H. J., "Design Alternatives Analysis," New Approaches to Evaluating Firesafety in Buildings.
Proceedings of National Academy of Sciences Conference, Sept. 1978, pp. 20-24.
Walts, J. M .. Jr., "A Review of the Past, Present, and Future of Education and Training in Systems
Approaches to Evaluating Firesafety in Buildings." New Approaches to Evaluating Firesafety in
Buildings, Proceedings of National Academy of Sciences Conference, Sept. 1978, pp. 137-141.
Wilson, R., "An Introduction to the Measurement of Firesafety in Buildings," New Approaches to
Evaluating Firesafety in Buildings, Proceedings of National Academy of Sciences Conference, Sept.
1978, pp. 83-118.
Technical Committee on Systems Concepts for Fire Protection in Structllres, National Fire Protection
Association, Quincy. MA.
Standards Council, Guide to the Firesafety Concepts Tree, NFPA #550, 1986 Edition. National Fire
Protection Association, Quincy, MA. 30 Dec. 1985.
Managing Fire Services, 2nd Edition, International City Management Association, Washington. DC,
pp.118-120. /"

Risk Assessment

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Risk Assessment

Încărcat de

Drepturi de autor:

Formate disponibile

STP 1150

Fire Hazard and Fire Risk

Marcelo M. Hirschler, editor

ASTM Publication Code Number (PCN)

Fire hazard and fire risk assessment/Marcelo M. Hirschler, editor.

Copyright © 1992 AMERICAN SOCIETY FOR TESTING AND MATERIALS, Phil-

Peer Review Policy

INTRODUCTION TO FIRE HAZARD AND FIRE RISK ASSESSMENT

~ Importance of Carbon Monoxide in the Toxicity of Fire Atmospheres-

rife Risk Rating Schedules-J. M. WATTS, JR. 24

USE OF FIRE TESTS FOR FIRE HAZARD ASSESSMENT

£qluating the Hazards of Large Petrochemical Fires-N. R. KELTNER,

adncal Cable Fire Hazard Assessment with the Cone Calorimeter-

Pa-formance of Plastic Plumbing and Electrical Products in Fire Resistive

FIRE HAZARD ASSESSMENT

Fire Hazard Assessment of Composite Materials: The Use and Limitations of

Fire Engulfment of Pressure-Liquefied Gas Tanks: Experiments and Modeling-

Development of a Benefit Analysis for an Onboard Aircraft Cabin Water Spray

FIRE RISK ASSESSMENT

Predicting Product Fire Risk: A Review of Four Case Studies-R. W. BUKOWSKI,

A Reliability Methodology Applied to Halon 1301 Extinguishing Systems in

Introduction to Fire Hazard and Fire Risk Assessment

Use of Fire Tests for Fire Hazard Assessment

Fire Risk Assessment

dential dwellings governed by the Ontario Building Code.

The work has shown that:

Forensic Investigation on CO Victims Across the United States

CARBON MONOXIDE AND FATALITIES

TABLE 2- Victim variables included in the USM database.

1. Data about the Victim Itself:

TABLE 3-Details of USM database.

Fire Non-fire Unknown Total

llatistical Analysis of Forensic Investigation

Forensic Investigation on CO victims across time in Cleveland, OH

Study of Fire Retarded and Non-Fire Retarded Products

the quantity of smoke.

Upholstered Furniture Heat Release Rates

Qpeak = 210.0 [fabric factor] x [padding factor] x [mass]

x [frame factor] x [style factor] (1)

_ the constant 210.0 has units kW/kg.

Assessing the Fire Risk of Products

• Identify appropriate fire response characteristics.

• Assign weights indicating relative importance .

Assessment of Fire Risk by Occupancy Classification

1) provide a uniform procedure for assessment of fire risk; and

Decision Logic for Trading Between Fire Safety Measures

TABLE 1- Evaluation of fire response characteristics of curtains and drapes.

Fire Response Characteristic Test Method Normalizing Factor Weighting Factor

Ignitabilitylflame spread NFPA 701 [4] 0.4 0.9

spread (dollars per incident);

Generalized Characteristics of Fire Risk Rating Schedules

TABLE 2-lncremental effect of various factors on probabilities Pp and PFN•

Fires that might occur in a transportation accident involving hazardous materials or in a

Heat Transfer in Petrochemical Fires and Fire Tests

Comparing the Response of Different Items

It was hoped that an additional objective would be achieved:

OSU RSR (V * OD)/Light Path Length

Cone: RSR (V * OD * In1O)/(Sample area * Light Path Length)

TABLE 1- Test parameters for full-scale cable tests,

CSA-FT-4 UL 1581 ICEA T-529-520

Harne length, m 0,25 0,30 0,75

Tray: RSR = OD/(Light Path Length)

5. Gas concentrations (CO, CO2, HCl), in ppm or %.

Experimental Vinyl Power Cables: