1、CH-06-3-2 An 0,verview of Fire Hazard and Fire Risk Assessment in Regulation Richard W. Bukowski, PE ABSTRACT Fire hazard and jre risk assessment has gainedpopu- larity in assessing the pevformance of alternative approaches to prescriptive regulations and in justijcation of proposed changes to regul
2、ations and referenced standards. While risk is the preferred methodology, ofen the probabilities needed are not available and cannot be estimated, resulting in a default to hazard assessment. In other cases society is hazard averse, and hazard assessment is the preferable approach. This paper will p
3、rovide an overview ofjre hazard andjre risk assessment methodologies used in regulatory systems and the tools avail- able for conducting them. Examples of regulatory applications drawn from buildings, transportation, and nuclear safety will be provided. INTRODUCTION Nearly every developed country ha
4、s, or is in the process of implementing, performance-based building regulations as a means to rationalize their regulatory system and to encourage development. Most have expressed interest in the use of fire risk assessment as the means to judge performance against the explicit objectives at the cor
5、e of such systems. The fact that risk can never be eliminated may lead to the public perception that officials feel a few deaths are somehow acceptable, which is generally unpalatable as a matter of public policy. Risk of financial loss is easier to understand but is difficult to apply to life safet
6、y concerns without becoming embroiled in the value of life controversy. Since a rigorous risk assessment is computationally intense and requires a large amount of historical data that are frequently not collected, most analyses conducted in support of performance evaluation are hazard assessments. T
7、hese measure performance under a specified set of design condi- tions that are presumed to represent the principal threats. Since experience has shown that the worst fires are the result of many things going wrong together, it is desirable to account for situ- ations characterized by multiple failur
8、es in providing for the safety of the public. Further, since September 1 1,200 1, regu- lators are interested in understanding the risk of extreme events that are increasingly influencing security and insurance concerns. ASSESSING HAZARD AND RISK The goal of a fire hazard assessment (FHA) is to dete
9、r- mine the consequences of a specific set of conditions called a scenario. The scenario includes details of the room dimen- sions, contents, and materials of construction; arrangement of rooms in the building; sources of combustion air; position of doors; numbers, locations, and characteristics of
10、occupants; and any other details that will have an effect on the outcome. The trend today is to use computer models wherever possible, supplemented where necessary by expert judgment to deter- mine the outcome. While probabilistic methods are widely used in risk assessment, they find little applicat
11、ion in modem hazard assessments. Hazard assessment can be thought of as a subset of risk assessment. That is, a risk assessment is a series of hazard assessments that have been weighted for their likelihood of occurrence. The value of risk over hazard is its ability to iden- tify scenarios that cont
12、ribute significantly to the risk but that may not be obvious a priori. In the insurance and industrial sectors, risk assessments generally use monetary losses as a measure of risk since these dictate insurance rates or provide the incentive for expenditures on protection. In the nuclear Richard W. B
13、ukowski is a senior engineer at the Building and Fire Research Laboratory, National Institute of Standards and Technology, Gaithersburg, Md. 02006 ASHRAE. 387 power industry, probabilistic risk assessment has been the primary basis for safety regulation worldwide. Here the risk of a release of radio
14、active material to the environment from anything ranging from a leak of contaminated water to a core meltdown is examined. Fire hazard assessments performed in support of regula- tory actions generally look at hazards to life, although other outcomes can be examined as long as the condition can be q
15、uantified. For example, in a museum or historical structure, the purpose of an FHA might be to avoid damage to valuable or irreplaceable objects or to the structure itself. It would then be necessary to determine the maximum exposure to heat and combustion products that can be tolerated by these ite
16、ms before unacceptable damage occurs. Areas of Application In the last decade deterministic fire hazard and fire risk assessment has increasingly been used in building regulatory applications, first as substantiation for alternative materials and designs and later for performance-based buildings. Th
17、ese techniques have also been strongly embraced by the historical and cultural preservation communities as a means to raise the level of protection to nearly full compliance with current prac- tice without sacrificing the significant aspects of the building (NFPA 2001). Another area where fire hazar
18、d and fire risk assessment is being applied is in transportation, particularly in the rail and maritime areas. The U.S. Federal Railroad Administration recently adopted new rules for passenger rail that require fire risk assessment of current and proposed rolling stock, and NFPA 130 requires hazard
19、assessment of rail stations and terminals (NFPA 2003). Beyond code compliance assessment, fire hazard assess- ment techniques are used in substantiation of proposals for changes to codes and standards. Most often this involves use of the analysis to justi thresholds contained in the require- ment. T
20、he American Society of Mechanical Engineers (ASME) has a formal hazard assessment procedure used in their code development process. The hazard assessment conducted by the code committee utilizes a template that then becomes a permanent record of the considerations and assumptions of the committee in
21、 establishing the requirements of their code. Available Tools Fire hazard assessments are routinely performed with one of the several zone models and engineering software pack- ages available in the world. In English-speaking countries, FPEtool (Deal 1995), FASTLite (Portier et al. 1996), CFAST (Pea
22、cock et al. 1993), and HAZARD I (Bukowski et al. 1989), all from NIST, FIRECALC (CSIRO 1991) from Australia, and ARGOS (DIFT 1992) from Denmark are the most frequently cited. The Japanese prefer BRI2 (Tanaka et al. 1987) and the French use MAGIC (EDF no date), as these are locally produced and use t
23、he local language for the software and manuals. Increases in computing power and the potential of parallel processing are leading to more use of field models such as NISTs Fire Dynamics Simulator (McGrattan 2005) for multi-scenario fire hazard analysis. Several nations have or are developing enginee
24、ring codes of practice, e.g., Japan (MOC 1988), UK (Barnfield et al. 1995), - Australia (ABCB 2001), and New Zealand (Buchananl994). The SFPE Handbook of Fire Protection Engineering (SFPE 2002) is a universal reference work for the underlying science, although Japan has its own version of a comprehe
25、nsive engi- neering handbook. Recently, the US, Canada, Australia, and New Zealand collaborated on an international version of the Fire Safety Engineering Guidelines (International Fire Engi- neering Guidelines 2005), consisting of a country-specific Part O (describing the countrys performance-based
26、 regulatory system) and a common methodology acceptable for building regulation in any country. Since all fire hazard assessments involve a small number of scenarios or design fires, no special software arrangements are needed. This is not true for risk assessments, which typi- cally involve hundred
27、s to thousands of scenarios. Here, special software packages, which run the cases and summarize the results, have been developed. These include - FRAMEworks (Hall et al. 1992) in the US, FiRECAM (Beck et al. 1996) in Canada, and CRISP2 (Fraser-Mitchell 1994) in the UK. Data Needs for Risk While tool
28、s exist to do both hazard and risk assessments, the greatest difficulty faced by those applying either is the availability of data. All of these analytical methods need appropriate data, but risk assessments also need statistical distributions for many parameters in order to incorporate the variabil
29、ities that underlie the desire for risk-based regulation. Take, for example, FiRECAM (Beck and Yung 1994), developed by the National Research Council of Canada and Victoria University of Technology in Australia. As with the other risk methods cited above, this product is made up of a series of submo
30、dels (names in italics) that provide the needed functionality. The following discussion outlines the complex- ity of the problem to be analyzed and the types of data and distributions necessary for these risk assessment methods. Many of these data are also needed for hazard methods. The design$re mo
31、del considers six design fires: srnoul- dering, flaming nonflashover, and flashover fires, each with the door to the room of origin open and closed. While the heat release rate curves are fixed, the statistical incidence of each 388 ASHRAE Transactions: Symposia of these fire types in the target occ
32、upancy along with the prob- ability of the door being open in each must be specified. Few countries maintainnational fire incident databases from which these data can be obtained. Australia recently initiated a fire incident data system to provide the information. The fire growth model (a single-zon
33、e model for rapid processing of large numbers of calculations) then calculates burning rate, temperature, and smokelgas concentrations. This requires heats of combustion and yield fractions that are avail- able for common homogeneous fuels but rarely for end-use products. Fuel loads (energy content
34、per floor area in wood equivalent) have been surveyed for a few occupancies, but smoke and gas yields can vary substantially across fuels (e.g., soot yield fractions vary by two orders of magnitude from wood to plastics). Since the smoldering and nonflashover fires remain fuel controlled, the values
35、 used for these inputs have a significant effect on the results. For the flashover fires, venti- lation is the controlling factor. Window breakage and other sources of combustion airplay a critical role, especially for the cases with the door closed. The smoke movement model then calculates the dist
36、ribu- tion of energy and mass and associated tenability times for all spaces. Compartment dimensions and connections as well as heat transfer properties of surfaces must be entered for the target building, but they will be known. The distribution of smoke is dominated by the probabilities of interio
37、r doors being open, which will be difficult to assess for many buildings. The $re detection model calculates the probability of detector or sprinkler activation, which depends on the proba- bility that they are present and the probability that they are working. The former is usually available based
38、on code requirements and common practice, but the latter is usually not. Quantification of the operational reliability of fire protec- tion systems is the subject ofcurrent studies in the US andUK. The occupant warning and response model depends on the fire detection model to initiate the evacuation
39、 of occupants where detectors or sprinklers are present and working. Other- wise the fire growth model predicts a “fire cue time” (presum- ably for the room of origin only) when the fire would be sufficiently threatening to initiate action. Where occupants are in remote spaces and especially when th
40、ey are asleep, it is unclear when (or if) evacuation would begin when the only stimulus is cues from the fire. Thefire brigade action model evaluates the effectiveness of the fire brigade in both suppression and evacuation assis- tance. This usually assumes that the fire brigade is successful in sup
41、pression if they arrive before flashover, but four of the six design fires do not reach flashover by definition. No differ- entiation is made for fire brigade staffing, equipment, training, or other variables, although these issues were addressed in the original Australian work. Fire brigade respons
42、e times are often reported by the brigades, but the time needed after arrival to begin operations (either suppression or rescue) generally is not. The smoke hazard model, evacuation duration model, and egress model all deal with the time needed for occupant evac- uation and the probability that some
43、 or all successfully escape. The abiliy to react, speed of movement, and sensitivity to smoke and gas are all dependent on the assumed physical char- acteristics of the occupants. The distribution of age, physical and mental impairments, drug or alcohol use, etc., within the mix of people in a given
44、 occupancy is sometimes available but is uncertain. Most evacuation models suggest a large safety factor (at least two to three) to account for these uncertainties. A boundary element model is used to assess the probabil- ity that the fire will spread to other spaces by failure of a boundary element
45、 or a closed door. If such failures occur, the fire spread model calculates the extent of such spread. Deter- ministic models of the failure of structural assemblies when exposed to an arbitrary fire are in their infancy. All such approaches rely on properties of materials at elevated temper- atures
46、 that are generally unavailable. The performance of rated assemblies to the standard time-temperature exposure must be extrapolated or statistical data from past incidents must be used. Such statistical data are rare (the author is only aware of such data being collected in the United Kingdom). The
47、life loss model uses toxicology data from animals to estimate the effect on people. Animal data for lethality are well documented but, for incapacitation, are highly uncertain since assessing incapacitation in animals is difficult. In either case the extrapolation from animals to humans is controver
48、sial. When assessing economic losses, theproperty loss model must integrate the replacement costs of contents and structure with the damage expected given the exposure and whether or not the items can be cleaned (along with these costs). The economic model andfire cost expectation model need the add
49、i- tional inputs of the capital and maintenance costs of all fire protection features. While the capital costs are available in construction cost manuals, maintenance costs are not. The data needs and availability for CRISP2, FRAME- WORKS, or any risk assessment are the same as for Fim- CAM and represent the greatest barrier to the widespread application of these techniques. Data unavailability leads to the use of estimates, which adds to the uncertainty of the results. Of course, there are other approaches to risk, includ- ing probabilistic risk analysis (PRA), used extensively in the nu
