Documente Academic
Documente Profesional
Documente Cultură
ASTM
1916 Race Street
Philadelphia, PA 19103
Library of Congress Cataloging-in-Publication Data
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.
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
How to Tell Whether What You Have is a Fire Risk Analysis Model-1. R. HALL, JR. 131
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.
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.
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.
(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.
10 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
DEBANNE ET AL. ON THE IMPORTANCE OF CARBON MONOXIDE 13
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.
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
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
DEBANNE ET AL. ON THE IMPORTANCE OF CARBON MONOXIDE 17
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.
monoxide poisoning than does the population as a whole or than does the population that
dies from CO poisoning in non-fire cases.
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.
• 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].
DEBANNE ET AL. ON THE IMPORTANCE OF CARBON MONOXIDE 21
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
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%.
22 FIRE HAZARD AND FIRE RISK ASSESSMENT
(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.
DEBANNE ET AL. ON THE IMPORTANCE OF CARBON MONOXIDE 23
(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
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.
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.
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.
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
28 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
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].
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
WATTS ON FIRE RISK RATING SCHEDULES 29
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-
WATTS ON FIRE RISK RATING SCHEDULES 31
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
32 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
WATTS ON FIRE RISK RATING SCHEDULES 33
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.
34 FIRE HAZARD AND FIRE RISK ASSESSMENT
[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
'Supervisor, senior members of technical staff, respectively, Sandia National Laboratories, Thermal
Test and Analysis Division, 2737, Albuquerque, NM 87185.
37
38 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
40 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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
42 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
• 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
46 FIRE HAZARD AND FIRE RISK ASSESSMENT
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:
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
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 47
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:
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
48 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
Commercial Cables:
IBM Type I
IBM Type II
Cable Peak Flame Height, em Total Char Length, em Combined Loss, % Pass/Fail
Cable Peak Flame Height, em Total Char Length, em Combined Loss, % Pass/Fail
Cable Peak Flame Height, em Total Char Length, em Combined Loss, (%) Pass/Fail
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCl, ppm
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
50 FIRE HAZARD AND FIRE RISK ASSESSMENT
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCl, ppm
Cable Peak CO, ppm Peak CO2, % CO/CO2 Peak HCI, ppm
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."
-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."
"Background burner heat was subtracted out from total heats reported. The TSR results have been
.ultiplied by the cross sectional area of the duct.
52 FIRE HAZARD AND FIRE RISK ASSESSMENT
TABLE lO-Heat and smoke release results from CSA FT-4 cable tray tests."
THR 15 PK RHR PK RSR
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
54 FIRE HAZARD AND FIRE RISK ASSESSMENT
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:
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).
56 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
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.
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:
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
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 59
TABLE IS-Correlation between the cone calorimeter and the ICEA T-29-520 cable tray test.
See Eq 2.
Correlation Coefficient
Property Flux R2/Adj R2 P
TABLE 16-Correlation between the cone calorimeter and the CSA FT4 cable tray test.
Correlation Coefficient
Property Flux R2/Adj R2 P
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.
Cone and as U Rate of Heat Calorimeter Test Instruments- The results of the tests carried
60 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 61
62 FIRE HAZARD AND FIRE RISK ASSESSMENT
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 63
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
HIRSCHLER ON ELECTRICAL CABLE FIRE HAZARD ASSESSMENT 65
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
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.
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.
70 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
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.
ZICHERMAN ON PLASTIC PLUMBING AND ELECTRICAL PRODUCTS 71
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.
TABLE I-Fire endurance and finish ratings of I- and 2-h walls including PVC pipe,
tube. and conduit.
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].
'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.
ZICHERMAN ON PLASTIC PLUMBING AND ELECTRICAL PRODUCTS 73
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.
ZICHERMAN ON PLASTIC PLUMBING AND ELECTRICAL PRODUCTS 75
"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
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.
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
80 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
[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.
ZICHERMAN ON PLASTIC PLUMBING AND ELECTRICAL PRODUCTS 83
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
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
88 FIRE HAZARD AND FIRE RISK ASSESSMENT
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:
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
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.
90 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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).
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
92 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
DINENNO AND SEYLER ON HAZARD ANALYSIS METHODOLOGY 93
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.
94 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
DINENNO AND SEYLER ON HAZARD ANALYSIS METHODOLOGY 95
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:
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:
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.
DINENNO AND SEYLER ON HAZARD ANALYSIS METHODOLOGY 97
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
98 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
DINENNO AND SEYLER ON HAZARD ANALYSIS METHODOLOGY 99
(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.
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].
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.
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
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 103
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.
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).
106 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 107
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.
108 FIRE HAZARD AND FIRE RISK ASSESSMENT
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].
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 109
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
110 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 111
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
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
112 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
114 FIRE HAZARD AND FIRE RISK ASSESSMENT
(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.
SUMATHIPALA ET AL. ON PRESSURE-LIQUEFIED GAS TANKS 115
[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
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.
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.
118 FIRE HAZARD AND FIRE RISK ASSESSMENT
(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.
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.
120 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
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.'
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
132 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
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.
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
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
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.
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.
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
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 139
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.
Case 4 shows that good finish materials are safe while poorly performing materials can be
hazardous under certain circumstances.
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
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 143
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
144 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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,
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 145
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 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
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 147
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.
• 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.
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.
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 149
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.
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.
the data required to apply the method to fully assess the risk of this product. Specifically,
the team was unable to ascertain:
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.
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 157
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:
• 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
supplement the data collected to fill these gaps. In others, special studies may be necessary
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-
BUKOWSKI ET AL. ON PREDICTING PRODUCT FIRE RISK 159
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
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 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
160 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
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
162 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
STECIAK AND ZALOSH ON HALON 1301 EXTINGUISHING SYSTEMS 165
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.
STECIAK AND ZALOSH ON HALON 1301 EXTINGUISHING SYSTEMS 167
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.
168 FIRE HAZARD AND FIRE RISK ASSESSMENT
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
STECIAK AND ZALOSH ON HALON 1301 EXTINGUISHING SYSTEMS 169
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].
170 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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
178 FIRE HAZARD AND FIRE RISK ASSESSMENT
(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.
STECIAK AND ZALOSH ON HALON 1301 EXTINGUISHING SYSTEMS 179
180 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
182 FIRE HAZARD AND FIRE RISK ASSESSMENT
[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
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
184 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
• 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
• 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.
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].
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:
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.
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:
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:
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:
BARRY ON RISK ASSESSMENT TO AN LPG BULK STORAGE FACILITY 191
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.
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
196 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
• 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
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
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:
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:
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.
1. At distances greater than 200 m from the LPG storage tanks, there need be no con-
BARRY ON RISK ASSESSMENT TO AN LPG BULK STORAGE FACILITY 205
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.
206 FIRE HAZARD AND FIRE RISK ASSESSMENT
[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
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
'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.
210 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
Design Objectives
The major objective of design for the effects of fire in buildings is to achieve satisfactory
levels of life safety for:
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.
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
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 213
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.
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
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
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 215
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.
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.
216 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 217
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.
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
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 219
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.
(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
222 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
BECK ET AL. ON DESIGN OF FIRE SAFETY SYSTEMS IN BUILDINGS 223
[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
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.
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.
226 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
HADJISOPHOCLEOUS AND YUNG ON HIGHRISE APARTMENT BUILDINGS 227
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
228 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
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
232 FIRE HAZARD AND FIRE RISK ASSESSMENT
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~
HADJISOPHOCLEOUS AND YUNG ON HIGHRISE APARTMENT BUILDINGS 233
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
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 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.
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.
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.
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 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.
• 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.
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
KATZIN ET AL. ON CANADIAN BUILDING CODE ASSESSMENT FRAMEWORK 243
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
244 FIRE HAZARD AND FIRE RISK ASSESSMENT
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%).
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:
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.
KATZIN ET AL. ON CANADIAN BUILDING CODE ASSESSMENT FRAMEWORK 247
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!
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
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.
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.
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,
250 FIRE HAZARD AND FIRE RISK ASSESSMENT
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.
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.
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.
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.
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.
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
254 FIRE HAZARD AND FIRE RISK ASSESSMENT
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. /"