2018_澳洲_FRNSW_住宅烟雾报警器离子式研究_火灾技术报告.pdf

8

Australian Building Codes

Board's Domestic Smoke Alarm Study: Ionisation versus Photoelectric

Prepared for the Australian Building Codes Board By

Vasily Novozhilov PhD, Long Shi PhD, Maurice Guerrieri PhD, Khalid Moinuddin PhD, Dorothy Bruck PhD & Paul Joseph PhD

Centre for Environmental and Risk Engineering, Victoria University, Melbourne, Australia

July 2015

〔图〕VICTORIAUNIVERSITY\n MELBOURNE AUSTRALIA

Executive Summary

The present report is developed in response to the Australian Building Codes Board’s (ABCB) request to undertake a study on the comparative performance of domestic ionisation and photoelectric smoke alarms.

SCOPE AND FRAMEWORK

The National Construction Code (NCC) permits the use of both photoelectric and ionisation smoke detectors as fire alarms in residential premises.

It appears that there is a growing support within the industry for the use of photoelectric detectors over ionisation detectors. The recent study conducted by Fire & Rescue New South Wales (FRNSW) favours photoelectric alarms.

It is not clear whether there are strong objective grounds to promote the use of one type of smoke alarm over the other.

The present report provides critical analysis of available data, including the FRNSW report, in order to reveal the performance of both types of detectors and their relevance for various fire scenarios. The report also reviews historical fire data in order to evaluate probabilities of major fire causes and scenarios, and to relate this to criterion used to estimate efficiencies of various types of smoke alarms. Data from Australia, New Zealand and the United States is used in the present study.

No solid justification for the preference of photoelectric smoke alarms over ionisation smoke alarms has been found. An extended test program is proposed to conduct a robust comparative study on the performance of both types of alarms.

ALARM TIME ACTIVATION DATA

Table 1 summarises the available data on activation of photoelectric, ionisation alarms as well as dual smoke alarms under two major types of fire conditions, namely flaming and smouldering.

Table 1: Summary of average activation times of different types of smoke alarms in flaming and smouldering conditions

Consistent with various previous reports, it is found that photoelectric alarms typically respond faster to smouldering fires, while ionisation alarms respond faster to flaming conditions.

These findings, however, are to be considered in the context of probabilities of different fire scenarios. It was found that the study conducted by FRNSW involved disproportionally large number of smouldering scenarios.

Based on the review of historical fire data, undertaken in the course of the present study, an extended testing program is proposed. Table 2 summarises the proposed testing program

Table 2: Summary of proposed testing program

Table of Contents

Executive Summary ................................................................................................. ii

1. Background ............................................................................................................ 7

1. 1 Home fires and smoke alarms .......................................................................... 7

1. 2 Aims and objectives .......................................................................................... 9

1. 3 Methodology ................................................................................................... 10

1. 31 Literature review ........................................................................................ 10

1. 32 Analysis of historical fire data .................................................................... 10

1. 33 Development of additional testing scenarios.............................................. 10

1. 4 Scope of the work ........................................................................................... 11

1. Literature Review: Part 1: FRNSW Report, Smoke DETECTORS & Historical Fire

Data .......................................................................................................................... 12

1. 1 Working principles of smoke detectors ........................................................... 12

1. 1.1 Ionisation smoke alarms ........................................................................... 12

1. 1.2 Photoelectric smoke alarms ...................................................................... 13

1. 2 Fire & Rescue NSW Report ............................................................................ 14

1. 3 Analysis of historical fire data .......................................................................... 21

1. Performances of smoke alarms ............................................................................ 33

1. 1 Previous tests on smoke alarms ..................................................................... 33

1. 2 Activation time ................................................................................................. 37

1. 3 Nuisance alarm ............................................................................................... 48

1. Development of additional testing scenarios ........................................................ 54

1. 1 Principles of performance testing .................................................................... 54

1. 2 The design of additional testing scenarios ...................................................... 55

1. Conclusions .......................................................................................................... 59

References ............................................................................................................... 61

Table of Figures

FIGURE 1: A SCHEMATIC DIAGRAM OF THE IONIZATION DETECTOR .............................................................. 12

FIGURE 2: A SCHEMATIC DIAGRAM OF THE PHOTOELECTRIC DETECTOR ...................................................... 13

FIGURE 3: PERCENTAGE OF TOTAL NUMBER FIRES UNDER DIFFERENT SITUATIONS FOR CLASS 1A & 2 BUILDING

DURING 2000-2014 [6] .................................................................................................................... 15

FIGURE 4: PERCENTAGE OF FATAL FIRES UNDER DIFFERENT SITUATIONS FOR CLASS 1A & 2 BUILDING DURING

2000-2014 [6] ................................................................................................................................ 16

FIGURE 5: PERCENTAGE OF FATAL FIRES CAUSED BY HEAT OF IGNITION IN DIFFERENT AREAS OF ORIGIN FOR

CLASS 1A & 2 BUILDING DURING 2000-2014 [6] ................................................................................ 16

FIGURE 6: RESIDENCE DESIGN AND EQUIPMENT LOCATIONS IN THE FRNSW TESTS [6] ............................... 17

FIGURE 7: FRNSW: PERCENTAGES BY AREA OF ORIGIN, TOTAL FIRES VS FATAL FIRE, 2000-2014, CLASS

1A & 2 BUILDINGS [6] ....................................................................................................................... 23

FIGURE 8: FRNSW: PERCENTAGES BY FORM OF HEAT OF IGNITION, TOTAL FIRES VS FATAL FIRES, 2000-

2014, CLASS LA & 2 BUILDINGS. MODIFIED FROM [6]......................................................................... 23

FIGURE 9: FRNSW: PERCENTAGES BY FORM OF MATERIAL FIRST IGNITED, TOTAL FIRES VS FATAL FIRES,

2000-2014, CLASS LA & 2 BUILDINGS [6] ......................................................................................... 24

FIGURE 10: LEADING CAUSES OF HOME FIRES, UNITED STATES, 2003-2007 [13] ....................................... 29

FIGURE 11: CAUSES OF FATAL RESIDENTIAL BUILDING FI RES IN U.S. IN 2012 [38] ...................................... 32

FIGURE 12: REPORTED FIRES AND FALSE ALARMS IN UNITED STATES: 1980-2003 [46] .............................. 49

FIGURE 13: REASON SMOKE ALARM FAILED TO OPERATE IN REPORTED NON-CONFINED HOME STRUCTURE

FIRES: 2003 TO 2006 (%) [16, 23, 52, 53] ........................................................................................ 50

FIGURE 14: REASONS FOR INEFFECTIVENESS OF SMOKE ALARM/DETECTORS IN NEW ZEALAND DURING 1999

AND 2006 ....................................................................................................................................... 51

FIGURE 15: REASONS FOR INEFFECTIVENESS OF SMOKE ALARM/DETECTORS IN AUSTRALIA DURING 1999 AND

2006 .............................................................................................................................................. 52

Table of Tables

TABLE 1: SUMMARY OF AVERAGE ACTIVATION TIMES OF DIFFERENT TYPES OF SMOKE ALARMS IN FLAMING AND

SMOULDERING CONDITIONS ................................................................................................................III

TABLE 2: SUMMARY OF PROPOSED TESTING PROGRAM ............................................................................... IV

TABLE 3: ESTIMATES OF REQUIRED ESCAPE TIMES FOR BEST AND WORST CASE SCENARIO [4] ....................... 8

TABLE 4: OPTICAL DENSITY ALARM THRESHOLDS FOR DIFFERENT TYPES OF SMOKE ALARMS [29] ................ 14

TABLE 5: A SUMMARY OF EXPERIMENTAL TESTS BY FRNSW [6] ................................................................ 18

TABLE 6: SMOKE ALARM ACTIVATION RESULTS IN THE ROOM OF ORIGIN [6] ................................................. 18

TABLE 7: SMOKE ALARM ACTIVATION RESULTS IN THE HALL [6] ................................................................... 19

TABLE 8: NUMBER OF RESIDENTIAL FIRE DEATHS PER 100,000 IN AUSTRALIA AND NEW ZEALAND [10] ........ 22

TABLE 9: FIRE INJURIES IN NEW ZEALAND [8] ............................................................................................ 22

TABLE 10: PERCENTAGES OF SURVIVED VERSUS FATAL ACCIDENTAL RESIDENTIAL FIRES BY TYPE OF

MATERIALS IGNITED [11] ................................................................................................................... 25

TABLE 11: PERCENTAGES OF SURVIVED VERSUS FATAL ACCIDENTAL RESIDENTIAL FIRES BY FORM OF

MATERIALS IGNITED [11] ................................................................................................................... 25

TABLE 12: PERCENTAGES OF SURVIVED VERSUS FATAL ACCIDENTAL RESIDENTIAL FIRES BY FORM OF HEAT

IGNITION [11] ................................................................................................................................... 26

TABLE 13: PERCENTAGES OF SURVIVED VERSUS FATAL ACCIDENTAL RESIDENTIAL FIRES BY IGNITION FACTOR

[11] ................................................................................................................................................ 26

TABLE 14: NON-CONFINED HOME STRUCTURE FIRE DEATHS BY LEADING ITEMS FIRST IGNITED AND SMOKE

ALARM STATUS: 2003 TO 2006 ANNUAL AVERAGES [16]..................................................................... 28

TABLE 15: MAJOR CAUSES OF RESIDENTIAL FIRES, CIVILIAN DEATHS AND INJURIES IN THE UNITED STATES

FROM 1999-2002 [37] ..................................................................................................................... 29

TABLE 16: LEADING CAUSES OF RESIDENTIAL BUILDING FIRE DEATHS IN U.S. DURING 2003-2012 [9] .......... 30

TABLE 17: TREND OF RESIDENTIAL BUILDING SMOKING FIRES IN U.S. DURING 2003-2012 [9] ...................... 30

TABLE 18: TREND OF RESIDENTIAL BUILDING ELECTRICAL MALFUNCTION FIRES IN U.S. DURING 2003-2012 [9]

...................................................................................................................................................... 31

TABLE 19: TREND OF RESIDENTIAL BUILDING HEATING FIRES IN U.S. DURING 2003-2012 [9] ....................... 31

TABLE 20: TREND OF RESIDENTIAL BUILDING OPEN FLAME FIRES IN U.S. DURING 2003-2012 [9] ................. 31

TABLE 21: TOP FIRE SCENARIOS RANKED BY FREQUENCY OF OCCURRENCE, 1992-1996 [4] ....................... 32

TABLE 22: A SUMMARY OF PREVIOUS EXPERIMENTAL TESTS ON THE PERFORMANCE OF SMOKE ALARMS ...... 33

TABLE 23: A SUMMARY OF THE PERFORMANCE OF SMOKE ALARMS UNDER FLAMING CONDITION ................... 38

TABLE 24: A SUMMARY OF THE PERFORMANCE OF SMOKE ALARMS UNDER SMOLDERING CONDITION ............ 38

TABLE 25: SMOKE ALARM ACTIVATION TIMES FROM SOURCE INITIATION [3] ................................................. 40

TABLE 26: PROPORTION OF SMOKE ALARMS OF EACH TYPE AND MANUFACTURER THAT DID NOT ACTIVATE [39]

...................................................................................................................................................... 40

TABLE 27: EXPERIMENTS IN THE DWELLING AND THE RESULTS [2] ............................................................. 41

TABLE 28: AVERAGE ALARM TIMES FOR NRC CANADA TEST SERIES [44, 45] ............................................. 42

TABLE 29: AVERAGE ALARM TIMES FOR THE NIST TEST SERIES [4, 44] ...................................................... 43

TABLE 30: ALARM TIME STATISTICS FOR THE NIST TEST SERIES OF INITIALLY FLAMING FIRES [4, 44] ........... 43

TABLE 31: ALARM TIME STATISTICS FOR THE NIST TEST SERIES OF INITIALLY SMOLDERING FIRES [4, 44] ..... 44

TABLE 32: ALARM TIME STATISTICS FOR THE NIST TEST SERIES OF KITCHEN FIRES [4, 44] .......................... 44

TABLE 33: AVERAGE ALARM TIMES FOR EACH CONFIGURATION FOR LIVING ROOM TESTS [1] ........................ 45

TABLE 34: RELATIVE TIME STATISTICS FOR THE PERFORMANCE OF SMOKE ALARMS [20] .............................. 45

TABLE 35: AVERAGE TIME TO ALARM FOR SEVERAL SMOKE ALARMS AND FIRE SCENARIOS IN A MANUFACTURED

HOME [4] ......................................................................................................................................... 46

TABLE 36: AVERAGE TIME TO ALARM FOR SEVERAL SMOKE ALARMS AND FIRE SCENARIOS IN A TWO-STORY

HOME .............................................................................................................................................. 47

TABLE 37: A SUMMARY OF STUDIES ON THE NUISANCE ALARMS .................................................................. 52

TABLE 38: SMOKE ALARMS TO DIFFERENT NUISANCE SCENARIOS [5] .......................................................... 53

TABLE 39: TOTAL PROBABILITY OF ACTIVATION FOR DIFFERENT TYPES OF SMOKE ALARM [51] ..................... 53

TABLE 40: A SUMMARY OF ADDITIONAL TESTS FOR FRNSW STUDY [6] ...................................................... 55

TABLE 41: TOP FIRE SCENARIOS RANKED BY FREQUENCY OF OCCURRENCE, 1992-1996 [4] ....................... 57

1. BACKGROUND

1. 1 Home fires and smoke alarms

Home fires are still a problem in our daily lives. From 2007 to 2011 United States fire departments responded to an average of 1,000 home structure fires every day, and home fires killed an average of seven people per day and caused roughly $28 in damage every second [7]. According to the statistics by Australasian Fire and Emergency Service Authorities Council [8], the residential fire deaths per 100,000 is between 0.07 and 0.41 during 1996-2004, and the figure for New Zealand is a little higher during the same period, that is between 0.10 and 0.70. According to The United States Fire Administration [9], the estimation of residential building fire deaths in United States is between 2385 and 3050 (yearly) from 2003 to 2012. According to Australasian Fire Authorities Council [10], it is known that the time period when most fire fatalities occurred was between the general sleeping times of 8pm-8am (72%) in Australia, and the figure increases to 78% in New Zealand with a peak occurring between midnight-4am (42%). The study [11] indicates that one out of four survived occupants (24.2%) were asleep at the time of ignition, while four out of five fatalities (80.5%) were asleep.

There are many reasons that cause home fires. A statistics [10] shows that in Australia the majority of fires (74%) were caused either by electrical faults, smoking materials or heaters/open fires/lamps, and in New Zealand 56% of fires were caused either by open fires/heaters, kitchen fire/cooking materials or electrical faults. Further, 2972 (40.6%) people were injured as result of fires that originated in the kitchen, 1600 (21.9%) people were injured in fires that started in the bedroom [12]. According to the statistics of home fires [12], in New Zealand, 46.7% of people were injured in fires that started in the kitchen,. A statistics [13] shows that 40% of homes fire were caused by cooking equipment and 18% by heating equipment in United States from 2003 to 2007. A study shows that the cooking equipment is responsible for 31.0% of the whole number of residential fires and 9% of civilian deaths [8]. It is also known that flaming cooking materials are involved in fires more than five times more frequently than any other material [4]. These scenarios are amongst the top five ranked by number of deaths, and amongst the top ten ranked by frequency of occurrence.

Home smoke alarm technology is credited as the greatest success story in fire safety in the last part of the 20th century, as it alone represented a highly effective fire safety technology that come into nearly universal usage in a remarkably short time [4]. A working smoke alarm has been reported to reduce the risk of death from residential fires by from between 50% to 70% [14, 15]. According to estimates by the National Fire Protection Association and the U.S. Fire Administration, U.S. home usage of smoke alarms rose from less than 10% in 1975 to at least 95% in 2000, while the number of home fire deaths was cut nearly in half [4]. The US Fire Administration reports that more than 88% of the homes in United States have at least 1 smoke alarm installed, but 60% of the residential fire deaths occur in homes without an operational alarm [14]. However, the coverage of Australian homes by smoke alarms is still of concern. For example, it is known from the statistics of home fires [12] that the presence of a smoke alarm/detector was determined in 2441 (33.3%) of the cases. Where it was identified that a smoke alarm/detector was present, it was found that they operated in 1811 (74.2%) of the cases and alerted 1406 of the occupants or fire injury victims that there was a fire in their property. Analysis of data from the United States Fire Administration’s National Fire Incident Reporting System and the National

Fire Protection Association's fire department survey showed that from 2003 to 2006, no smoke alarms were present in 31% of reported home fires and 40% of home fire deaths [16].

There are basically 3 different types of residential smoke alarm: the ionisation alarm, the photoelectric alarm, and the dual alarm [14]. Ionization and photoelectric alarms operate via different mechanisms, detecting visible, and invisible/fine, products of combustion, respectively [17]. Photoelectric alarms use optical sensors and are more likely to respond to slow, smouldering conditions. Working principle of ionisation detectors is based on a modified theory which includes soot particle charge fraction functionality in addition to the generally accepted particle size and number density dependence [18]. Smoke alarms of either the ionisation type or the photoelectric type consistently provide time for occupants to escape from most residential fires, although in some cases the escape time provided can be short [4]. Table 3 [4] shows the estimates of required escape time for the best and worst case scenarios.

Table 3: Estimates of required escape times for best and worst case scenario [4]

Note: a – Best case scenarios neglect any pre-movement activity; actual escape times are likely to be longer than best estimates;

b-Manufactured home; and

c- Two-story home.

Many studies have been conducted to analyse the performance of different type of smoke alarms. Consumers Union [19] tested ionisation and photoelectric alarms in 1994. It found that in a smouldering, smoky fire, the ionization alarms responded in 25 to 35 minutes, whereas the photoelectric models reacted in half that time. A statistical study was conducted to compare the performance of different residential smoke detector technologies when exposed to different fire types by Milarcik et al. [20]. The results showed that ionisation detectors, on average, respond faster to flaming fires, while photoelectric detectors, on average, respond faster to smouldering fires. Bukowski et al. [4] have performed comprehensive real-scale tests on the performance of different type of smoke alarms. They arrived at similar conclusion that ionisation type alarms provide somewhat better response to flaming fires than photoelectrical alarms, and photoelectric alarms provide (often) considerably faster response to smouldering fires than ionization type alarms. Su and Crampton [2] conducted a series of experimental studies in a residential dwelling as well as in a laboratory room to examine the effect of "dead air space" on smoke response. The results showed that smoke can reach the "dead air space" under the experimental conditions and the smoke alarms installed in the "dead air space" can respond to the fire at times comparable to, and in many cases even earlier than, the smoke alarms installed at conventional locations.

Nuisance alarms are also a problem in homes. According to a 1994 study of United States residential smoke alarm use, the leading cause of smoke alarm disconnection

was nuisance alarms [14, 21]. The Smoke Detector Operability Survey conducted by the United States Consumer Product Safety Commission [22] reported that about one half of the 1012 respondents indicated they experienced nuisance alarms, with 80% of those attributed to cooking activities, and an additional 6% citing steam from bathrooms. The fact that ionisation alarms produce more false alarms but are slower to respond to smoky fires is not the enigma that it seems [14]. A study [23] showed that in 54% of the non-confined home structure fires and 75% of the home fire deaths in which smoke alarms were present but failed to operate, smoke alarm batteries were missing or had been disconnected.

Currently, Australian National Construction Code’s (NCC) performance requirements allows both types of smoke detectors (i.e. ionisation and photoelectric), as well dual (combination) alarms to be used.

There is an apparent drive within the Fire Protection Industry to promote the use of photoelectric alarms over ionisation alarms. Fire Authorities seem to support phasing out ionisation alarms on the basis of their belief that most fire fatalities result from smouldering fires.

There is a concern expressed by the Australian Building Codes Board (ABCB) as to whether promotion of exclusive use of photoelectric alarms is soundly based on available alarm performance and fire statistics data. In particular, with response to the recent study by the Fire & Rescue New South Wales [6], ABCB expressed concerns that

• Research methodology is not robust and is biased towards a favourable outcome

for photoelectric alarms, which is reflected in the report; and

• Key deficiencies with the research methodology: 1) Comprehensive literature

review needs to be undertaken to inform research methodology; 2) Fire scenario

selection needs to reflect the probability of fire starts for incident data, in particular

kitchen fires need to be included (which represent the largest proportions of fire

incident); and 3) Energy source used for smouldering scenarios is extreme and not

representative.

1. 2 Aims and objectives

The aims of this study is to conduct a critical review of available data on comparative performance of photoelectric versus ionisation smoke alarms, and develop an

objective testing scenarios in order to obtain accurate data on performance of these two types of detectors. This analysis will provide support for undertaking additional

testing.

The present study stems from the necessity for ABCB to assess whether the

apparent current industry trend towards favouring photoelectric alarms, and

associated proposals to eliminate future use of ionisation detectors have sound

technical basis.

Recent findings [6] obtained by FRNSW appears to contradict other publicly available reports and research publications. Initial screening of the report [6] by ABCB also pointed out to potential methodology deficiencies in the methodology [6].

This present study aims to answer the following questions:

What is the effectiveness of the two types of smoke alarms (e.g. photoelectric and ionisation) in responding to different fire conditions, based on available statistical data.

What are the key causes and scenarios of fire accidents in residential premises?

What kinds of experiments have been done to test the effectiveness of smoke alarms? What are the advantages and disadvantages of these smoke alarms in their applications based on their working principles?

An ultimate objective of the study is to propose additional testing using the FRNSW experimental rig to ensure robust evaluation of the effectiveness of ionisation and photoelectric smoke alarms in domestic settings.

1. 3 Methodology

Methodology of the present study consists of three major tasks, namely 1) literature review, including FRNSW report, 2) analysis of historical fire data, and 3)

development of objective testing scenarios.

1. 31 Literature review

An extensive literature review is conducted on working principles and performance of photoelectric and ionisation detectors, key fire incident data and effects of different types of fire on activation of both types of smoke alarms.

Ionisation and photoelectric alarms have different working principles, resulting in the different response time for the same amount and type of smoke. The working

principle of these two types of alarms will be analysed to evaluate their application

areas, for example, different flammable conditions and types of buildings.

Based on these conclusions, a comprehensive literature review is conducted to

analyse the historic testing results on smoke alarms. For the literature review, one of the directions is to address the application areas of these alarms to the flammable

conditions (e.g. flaming and smouldering combustion). Another direction is to

undertake a statistic analysis on typical fire accidents in homes. It aims to evaluate data on typical fire accidents in homes, which will benefit proposing new

experimental designs. Previous experimental designs and methods are reviewed.

1. 32 Analysis of historical fire data

The study conducted by FRNSW [6] provides some detailed fire incident data.

Additional relevant data will be revealed upon conducting an extensive literature

search. We aim to develop classification criteria for fire incidents in such a way that such that these criteria are directly related to expected performance of smoke

alarms.

1. 33 Development of additional testing scenarios

A new set of experiments is proposed based on the FRNSW existing test rig [6]. The new testing program is proposed based on the above analysis results, including fire scenario design, experimental procedures and data processing. The new

methodology will result in a robust evaluation of the effectiveness of ionisation,

photoelectric and dual/multi smoke alarms in homes.

Scenarios are tailored to the capabilities of the FRNSW testing facility [6]. Different types of proposed tests reflect the classification of fire accidents developed under the above task.

1. 4 Scope of the work

The present report includes a critical literature review of available data on the historical fire data, performances of different types of smoke alarms, and a development of additional testing scenarios which would be used to obtain accurate data on the performance of different types of detectors. The content of the present report is as Scopes of this report is as follows:

Chapter 1: Background. This chapter introduces the background of home fires, causes of home fires and briefly discusses the needs for smoke alarms to be installed in residential premises. Further, it overviews basic operating principles of photoelectric and ionisation smoke alarms, and provides general discussion of previous studies on the efficiency of smoke detectors.

Chapter 2: Literature review. Part I. First part of the literature review makes the analysis of the methodology and key conclusions of the recent FRNSW report. Further, it discusses in more detail the principles of photoelectric and ionization smoke alarm technology. The last section of this Chapter is devoted to the analysis of available fire statistics, with the emphasis on the information that is relevant for the present study. Of particular interest here are major causes of home fires, distribution of fatalities between fires of different types, and analysis of combustible materials involved in fires. Chapter 3: Literature review. Part II. Second part of the Literature review provides analysis of performances of smoke alarms. Previous experimental tests on the performance of smoke alarms (in particular, ionisation and photoelectric) are reviewed, as well as major conclusions of these studies. Activation times of three types of smoke alarms, namely photoelectric, ionization and combination are directly compared. Discussion of nuisance alarms is also provided in order to understand their reasons, which will allow the probability of nuisance alarms to drop in the future.

Chapter 4: Development of additional testing scenarios. Based on the literature review (including the FRNSW report) on the historical home fire data and previous smoke alarms tests, new additional testing scenarios are proposed to obtain accurate data on the performance of smoke alarms of different types. Quantitative statistical methodology is also proposed for processing the results.

Chapter 5: Conclusion. Major conclusions of the study are presented in this section.

2. LITERATURE REVIEW: PART 1: FRNSW REPORT, SMOKE DETECTORS & HISTORICAL FIRE DATA

2.1 Working principles of smoke detectors

2.1.1 Ionisation smoke alarms

In an ionisation chamber, an electric field is maintained by applying a suitable voltage between two electrodes. A radioactive source situated inside the chamber, usually an alpha-emitter, furnishes the ionization radiation. Under an applied electric field, oppositely-charged ions move resulting in a small current flow between the electrodes. Smoke or aerosol particles, when admitted into the chamber, will cling on to the ions and reduce their mobility. This will in turn result in an overall altered electrical current and an associated change in the output voltage. Figure 1 shows a schematic diagram of the ionisation detector.

〔图〕SMOKECLEAR AIRSMOKEPARTICLESRADIOACTIVEMATERIAL\n

Figure 1: A schematic diagram of the ionization detector

Instead of voltage, generally, a dimensionless variable Y is used which is a linear function of the degree of obscuration [24]:

$$Y=\left(\Delta V/V_0\right)\frac{2-\left(\Delta V/V_0\right)}{1-\left(\Delta V/V_0\right)}$$

Equation 1

where $V$ is the alarm output voltage, and is proportional to the ion chamber current; $V_0$ is the current of the ionisation chamber filled with clean air; and $\Delta V$ is the change in the voltage, i.e. ($V_0$-V).

The obscuration can be expressed by the following relation [4]:

$$O=\left[1-\left(\frac{I}{I_0}\right)^{\frac{1}{L_f}}\right]\times100$$

Equation 2

where $I$ is the intensity of the emerging light; $I_0$ is the initial intensity of the incident light; and $L_f$ is the path length, expressed in feet.

Ionisation detectors generally provide faster response to flaming fire conditions as they are best suited to detect fine particles, from 0.01 to 0.3 microns.

2.1.2 Photoelectric smoke alarms

Photoelectric smoke detectors, on the other hand, operate on a totally different principle as compared to the ionization smoke detectors, i.e. based on light scattering by the interference due to smoke particles. [25]. Photoelectric smoke detectors constitute of a light source, typically a light-emitting diode (LED), and a light receiver such as a photocell, and the latter is arranged at such an angle impeding the reception of any light from the LED. The scattering volume element is defined by the intersection of the viewing angles of the light beam from the LED and the photocell [26]. As smoke particle enter the scattering volume, light from the LED will get scattered and will reach the photosensitive element (i.e. photocell), and this in turn will generate a current. As the smoke concentration in the scattering volume increases, the luminous flux received by the photocell also will increase proportionally. An alarm is triggered as the amount of scattered light reaching the photocell exceeds a pre-set threshold. The signal produced by photoelectric detectors is also sensitive to a number factors, including the physical characteristics of both the detector and smoke including the number concentration, size distribution, shape, and refractive index of the smoke particles as well as the scattering volume and wavelength of light used in the detector [27]. Figure 2 provides a schematic diagram of the photoelectric detector.

〔图〕CLEAR AIR3PHOTOSENSITIVE DEVICELIGHTLENSSOURCESMOKE PARTICLES\n

Figure 2: A schematic diagram of the photoelectric detector

Photoelectric alarms generally provide faster response to smouldering fire conditions as they are best suited to detect large smoke particles, from 0.3 to 10 microns, resulting from incomplete combustion.

There are usually two basic methodologies in use for estimating the response of smoke detectors – the Temperature Rise Method and the Optical Density Method (ODM) [28, 29]. The basic principle of the Temperature Rise Method is that the ratio of the optical density (OD) to temperature rise is approximately constant for a given fuel and combustion model. The Optical Density Method consists of directly calculating the smoke concentration at the detector and comparing the smoke level to an alarm threshold for the detector. A summary of previous testing results about optical density

alarm thresholds for different types of smoke alarms is provided in [29]. Table 4 shows optical density alarm thresholds for different types of smoke alarms.

Table 4: Optical density alarm thresholds for different types of smoke alarms [29]

2.2 Fire & Rescue NSW Report

The FRNSW report [6] introduces first the background of smoke alarms, such as the principles of photoelectric and ionisation smoke alarms, smouldering and flaming fires and tenability criteria of incapacitation and death. Then the report analyses the fire historical data attended by FRNSW between 2000 and 2014 for class 1a and 2 buildings. Key elements of the FRNSW study, its results and conclusions are overviewed in this section. Further, deficiencies of research methodology employed by FRNSW are discussed.

Figure 3 (aggregated data from the Report) demonstrates that residential fires happened most frequently in kitchen (25%), and then in the bedroom (18%) and lounge room (13%). The four most probable modes ignition were electrical distribution & appliance (48%), open flame (13%), lighter/match (11%) and smouldering materials (7%). The first ignited materials are usually cooking materials (15%), followed by electrical wire & cable insulation (14%), mattress or bedding (9%), and upholstered furniture (5%).

〔图〕插图

Figure 3: Percentage of total number fires under different situations for class 1a & 2 building during 2000-2014 [6]

Figure 4 shows the percentage of fatal fires under different conditions. It is apparent that fatal fires mostly happened at bedroom (36%), then the lounge room (30%) and kitchen (9%). The percentages by the heat of ignition, from higher to lower, are electrical distribution & appliance (37%), smoking materials (23%), open flame (13%) and lighter/match (12%). Most frequent first ignited material is mattress or bedding (21%), followed by upholstered furniture (15%), electrical wire & cable insulation (6%), and cooking materials (3%).

〔图〕插图

Figure 4: Percentage of fatal fires under different situations for class 1a & 2 building during 2000-2014 [6]

Figure 5 shows the percentage of fatal fires, classified by heat of ignition, in different rooms of origin. Most frequent ignition source in the kitchen is the electrical distribution & appliance (80%), and then lighter/match and open flame have the same percentage of 7%. It is quite interesting that there is no fire caused by smoking materials in the kitchen. In the lounge room, the most frequent ignition sources are smoking materials (43%), followed by electrical distribution & appliance (31%), open flame (10%) and lighter/match (5%). Compared to the situation in the lounge room, the electrical distribution & appliance (38%) and smoking materials (31%) occupy the first two places as ignition sources in the bedroom, and the percentages for lighter/match (15%) are higher than those for the open flame (8%).

〔图〕插图

Figure 5: Percentage of fatal fires caused by heat of ignition in different areas of origin for class 1a & 2 building during 2000-2014 [6]

FRNSW used a test rig, simulating residential premises, to perform test burns. The setup was used to test the performance of different types of smoke alarms, such as ionization, photoelectric and combination. Figure 6 shows the layout of the residential test burn setup and the locations of testing equipment. Building structure of the residence was made of radiata pine with 16 mm fire resistant plasterboard lining the internal walls and ceiling. The outside of the building was lined with extra heavy duty premium wall warp and clad with 7.5 mm plywood.

〔图〕Charlie12000G2G43101835Sa3TaTbSa2Sa2Tc13100TTTSpL'drySa1HalTjTk3900Sa3Sa3G1Tn30005000000S078tBravoSa1BathroomDeltaTSpSp0 Sa1 - ceiling level smoke alarm location

Sa2 - wall mounted smoke alarm location

Sa3 - dead zone smoke alarm location

$G$ = gas monitoring locations

Sp - sprinkler head

Thermocouple location labelling with corresponding thermocouple numbering.

Numbering goes bottom up. For thermocouples Ta-Ti, thermocouples located

at floor, 1m, 1.5m and ceiling height. For thermocouples Tj-Tn, thermocouples located at floor, 0.75m, 1.5m and ceiling height.

Ta - 1,2,3,4 Tf - 21,22,23,24 Tk - 41,42,43,44

$$\text{Tb} - 5,6,7,8\qquad\text{Tg} - 25,26,27,28\qquad\text{Tl} - 45,46,47,48$$

Tc - 9,10,11,12 Th - 29,30,31,32 Tm - 49,50,51,52

Td - 13,14,15,16 Ti - 33,34,35,36 Tn - 53,54,55,56

Te - 17,18,19,20 Tj - 37,38,39,40

Figure 6: Residence design and equipment locations in the FRNSW tests [6]

A summary of experimental runs, performed by FRNSW, are listed in Table 5. It is important to note that 8 out of 10 runs were carried out under smouldering fire conditions. Two ignition methods were used, including cartridge heater and LPG gas flame. Other conditions included either closed or opened doors 1 and 2.

The focus of the test burns was to analyse smoke alarms in a real home fire environment. Therefore, although testing conditions as per AS3786:2015 were not followed (exact repeatability, all tests starting at 23°C ± 5 °C, test fire room design and structure, fire ignition sources, etc.), all tests simulated real home fires. The tests were undertaken in replica home residence with available bulk furniture from a national furniture chain, and under ambient Sydney conditions.

Table 5: A summary of experimental tests by FRNSW [6]

Smoke alarm activation results in the room of origin and the hall are presented in Table 6 and Table 7, respectively. As the fire under flaming condition grew fast in the FRNSW test, there is no big difference among the activation time of the three types of smoke alarms. For example, in the burn 10, activation time for the three types of smoke alarms are 0.87 min in both the places of SA1 and SA2, and all of them did not active in the place of SA3. The only difference for the two flaming test is that the photoelectric smoke alarm activated at 0.17 min, a little early than those of ionization and combination smoke alarms, namely 0.18 min.

Table 6: Smoke alarm activation results in the room of origin [6]

It can be also inferred from Table 6 and Table 7 that in the room of origin the combination smoke alarm activated fast in most of the cases that is in 19 out of 24

cases. Photoelectric smoke alarm activated faster than the other two types of alarms in 8 out of 24 cases. No case was found where the first activated smoke alarm was ionisation. The results for the hall area are very much similar to those for the rooms of origin. In 15 out of 24 cases, combination smoke alarm activated faster than the other two types, then followed the photoelectric smoke alarm, namely in 4 out of 24 cases.

Table 7: Smoke alarm activation results in the hall [6]

The FRNSW report also provides mean values of activation times for the two types of detectors in various rooms of fire origin. Only smouldering fire scenarios are considered in the statistical analysis. These mean activation times (in minutes) are 17.03 for photoelectric alarm, and 44.19 for ionisation alarm. Similar results for the Hall location are 32.19 (photoelectric) and 57.55 (ionisation). These are average figures across the number of tests. There was no possibility to evaluate Standard Deviations at each fixed location as each type of the test was not repeated. There were no statistical analysis on flaming conditions provided in the view of extremely limited number (two only) tests at such condition.

Some key conclusions can be extracted / quoted from the FRNSW report:

● The investigators admit that "the area of research is too broad to draw definite conclusions based on a single test study".

● Nevertheless, they conclude (based primarily on average response times) that, across the range of burns and placement locations, ionisation alarms were significantly inferior to their competitors (photoelectric and dual detectors)

● Photoelectric and dual photoelectric/ionization alarms still fell short of the arbitrarily applied yet reasonable expectation of 90% activation in the home fire scenarios tested.

● The focus of the study was on hallway alarm activation. The findings, although preliminary, clearly demonstrated that smoke alarms did not provide effective notification for safe egress when located in hallway only. When room of fire origin alarms were included, the results still fall short of acceptable levels.

● “Aerosol density measurements were not taken, thereby limiting data comparison to the AS3786 test standard. However, the fact that all alarms used met the standard, yet such low percentages of smoke alarms provided safe egress, the question must be raised as to whether smoke alarm test standards should require conditions to be maintained with regard to heat and toxic gases, rather than focusing on aerosol density. Or, if aerosol density has been derived around tenability limits, then perhaps this threshold needs to be reviewed. This would be a

difficult challenge to meet, as different home furnishing materials provide different

time to loss of tenability, dependent also on the type of fire occurring”.

• The results suggest that alarms operate in fires where fatalities still occur, due to

them not alerting the occupant. There are a range of possibilities as to why such

fatalities have occurred, including the possibility that the occupant had been already

incapacitated due to toxic smoke inhalation.

• “The failure of smoke alarms to respond within tenability and safe egress limits

raises the important question of whether or not smoke alarms are performing as

desired in the modern home. Although much research has been done to show

activation rates and that alarm positions and types are comparable and statistically

equivalent, relatively less work has focused on the need for smoke alarms to

provide safe egress. Without providing safe egress, activation time becomes

irrelevant”.

The present study focuses on objective quantitative procedure to compare performance of ionisation and photoelectric detectors. For this point of view, several major drawbacks of research methodology employed by FRNSW have been identified:

• Number of tests is too limited for a meaningful statistical analysis of a very complex

phenomenon with multiple factors and uncertainties.

• Repeatability. Experimental runs were not repeated for each of the scenarios, which

may have had an influence on the final outcome on the performance of different

types of smoke alarms. No Standard Deviations could be established for the results

of each of the tests.

• Literature review with respect to various fire initiation conditions and scenarios is

too limited. The research methodology must have been informed by the

comprehensive literature review.

• There is a severe imbalance in the testing program, even at a glance, between

smouldering and flaming fire conditions. The proportion of tests (8:2 ratio between

smouldering and flaming scenarios) is not realistic. This is further confirmed by the

analysis of the present study conducted in the Section 2.3.

An example of important fire conditions that have been excluded by such a

methodology are kitchen fires. The FRNSW report itself states that 25% of total fire

happened in the kitchen (Page 26) and that 15% of all the fires start from cooking

materials (Page 27).

• With only two tests under flaming conditions, no statistical analysis could be

performed on these scenarios. With a negligible number of tests, flaming conditions,

effectively, have been excluded from the consideration altogether. This inevitable

predetermined the final outcome being in favour of photoelectric alarms. Therefore,

the major conclusion of the FRNSW study on the comparative performance of the

two types of detectors cannot be considered as soundly justified.

• Energy supply provided to initiate smouldering fire scenarios (cartridge heater)

seems excessive compared with a large number of smouldering fires starting from

a burning cigarette. This potentially produced another skew in the data towards

faster responses of photoelectric alarms.

• It seems that for flaming conditions fire heat release rate increased too fast,

resulting in a very close results of performance for the three types of smoke alarm.

The details can be seen in Table 6 and Table 7 of the Report. There is a concern

that under more modest fire growth rates ionisation detectors could respond faster,

which was not picked up by the FRNSW study. The energy output of the LPG burner

used to initiate flaming scenarios should be reduced.

1. 3 Analysis of historical fire data

In the view of the working principles of smoke detectors, in order to evaluate their statistical performance in real fires. A detailed analysis of flaming fire frequencies versus smouldering fires frequencies is required. A very important factor which also need be considered is fire fatality, i.e. whether fatalities occur in one type of fire conditions more often than in the other.

In principle, the criteria which is needed to classify fires with regard to photoelectric/ionization alarms response may be proposed in the following manner: weighted frequency of flaming fires = (number of flaming fires without fatalities + number of flaming fires with fatalities x fatality weight) / (total number of fires) (on per year basis)

weighted frequency of smouldering fires = (number of smouldering fires without fatalities + number of smouldering fires with fatalities x fatality weight) / (total number of fires) (on per year basis)

Fatality weight is a factor attached to the importance of fire loss in a fire, i.e. 10.100, etc. (The methodology to determine a precise reasonable value for this factor is outside of the scope of the present study).

This approach would provide fatality-weighted frequency distribution

( )sf ,

1

0

,

1

0

≤

≤

≤

≤

s

f

;

1

=

• f

s

Equation 3

where f refer to flaming fire frequency, and s refer to smouldering fire frequency.

It also possible to further separate fires resulting in non-fatal injuries, it assign it yet different weighting.

In principle, smoke alarm testing should be arranged in such a way that the test scenarios reflect the frequency distribution (1)

Unfortunately, available fire statistics data do not allow the distribution (1) to be derived confidently.

Typically, fire statistics reveals the following major data:

• total number of fires

• number of deaths

• number of injuries

• causes of fires (by the type of materials first ignited, types of materials involved, area of fire origin, etc.)

Although Australian fire data is obviously of the paramount importance, and extended set of data may be obtained when international statistics is taken into account. Data from the United States and New Zealand is also considered in the present study. There is strong evidence [6, 35] that Australian data is comparable to the international statistics. Therefore, analysis of extended (international) data set is justified in the context of the present study.

In general, fire statistics data, especially obtained from the number of different countries, is rather non-uniform, and consistent statistical analysis is difficult to perform until the data is in some way harmonised in a way suitable for algorithmic processing.

The key findings from various literature sources are summarised, however, below

Australia-wide report on residential fire fatalities by Australasian Fire and Emergency Service Authorities Council [10] provides data on the rate of fatalities for the financial years from mid-1996 to mid-2004. It demonstrates (Table 8) the rates for Australia varying overall from 0.1 to 0.7 per 100,000 persons per year. No consistent trend is evident over time.

Table 8: Number of residential fire deaths per 100,000 in Australia and New Zealand [10]

Similar data (Table 9) on injuries in residential fires in Australia [8] showed that injuries per year in the period from mid-1999 to mid-2006 ranged from 4.37 to 6.03 per 100,000 persons per year.

Table 9: Fire injuries in New Zealand [8]

More detailed data on residential fires in Australia are available from several sources.

Fire and Rescue NSW (FRNSW) in their recent report [6] provides analysis for the fires in classes 1a and 2 buildings which it attended to between 2000 and 2014.

For the purpose of the present study the following detailed breakdowns [6] are important to consider:

· Percentages of fires by the area of origin (fatal fires vs total).

· Percentages of fires by the form of heat of ignition (fatal fires vs total)

· Percentages of fires by the form of material first ignited (fatal fires vs total)

These are presented in Figure 7, Figure 8 and Figure 9 respectively.

〔图〕插图

Figure 7: FRNSW: Percentages by Area of Origin, Total Fires vs Fatal Fire, 2000-2014, Class 1a & 2 Buildings [6]

〔图〕插图

Figure 8: FRNSW: Percentages by Form of Heat of Ignition, Total Fires vs Fatal Fires, 2000-2014, Class Ia & 2 Buildings. Modified from [6]

〔图〕插图

Figure 9: FRNSW: Percentages by Form of Material First Ignited, Total Fires vs Fatal Fires, 2000-2014, Class Ia & 2 Buildings [6]

Major findings of FRNSW informed design of their testing program and consequently final conclusions of their report. They found (Figure 7) that majority of home fires resulting in fatalities originated in either lounge room (30%) or bedroom (36%). Much less (9%) of fatal fires originated in the kitchen area. It should be noted that data on 25% of fatal fires is missing from Figure 7, i.e. it cannot be established with confidence where these fires originated.

With respect to the Form of heat of Ignition, the major scenarios that need be considered are electrical distribution / appliance malfunction, and the smoking materials Figure 8.

Further, FRNSW considered materials first ignited (Figure 9). The proper figure to look at, from the fatality point of view, will be percentage of fatal fires started from either mattresses, bedding or upholstered furniture. It is clear from Figure 9 that this figure is actually 36%. Such materials are considered by FRNSW as burning most likely in smouldering mode, with majority of ignitions occurring from smoking materials.

Even based on assumption that any fire involving mattresses, bedding or upholstered furniture would be a smouldering fire (which is obviously overestimation), the percentage of fatalities occurring from smouldering fires cannot exceed 36% (within the statistical FRNSW data set). This is consistent with further reference on p. 29 [6] which puts the figure to 25%. Although we agree with FRNSW that this shows that “reasonable percentage of these fatal fires may have begun by means of smouldering combustion”, “reasonable” actually means only around quarter of fatal fires.

This finding cannot be seen as a justification for severe bias in the FRNSW testing program towards smouldering fires (8 smouldering scenarios out of 10 tests).

It is important to analyse alternative statistical data to derive some further conclusions on the likely ratio between flaming and smouldering fires.

Various forms of fire statistical data are available from [11] (Australian data), [36] (Australia), [35] (Australia, NZ, US), [4] (US), [23] (US), [16] (US), [9], [13], [37], [38].

Further Australian fire statistics is provided by [11].

Percentages of survived versus fatal residential fires by type of materials ignited, form of materials ignited, form of heat ignition, ignition factor, adopted from this report, are presented in Table 10, Table 11, Table 12, and Table 13.

Table 10: Percentages of survived versus fatal accidental residential fires by type of materials ignited [11]

Table 11: Percentages of survived versus fatal accidental residential fires by form of materials ignited [11]

Table 12: Percentages of survived versus fatal accidental residential fires by form of heat ignition [11]

Table 13: Percentages of survived versus fatal accidental residential fires by ignition factor [11]

These data provides important information potentially shading the light on percentage

of fatalities resulting from smouldering vs flaming fires. Although exact figures cannot

be obtained explicitly from the tables, estimations based on some reasonable

assumptions are possible.

Analysis of materials leading to fatal accidents (Table 11) allows those with the

potential to support smouldering fires to be identified. These are upholstered items,

bedding, mattresses, and some floor coverings (such as carpets). Estimation based

on the data from Table 11 shows that maximum number of fatalities resulting from

smouldering fires is about 78, out of 161 total fatalities. Therefore, smouldering fires

contribute to about 46% of total fatalities number.

Similar estimation based on the US data from Table 14 (number of fatalities by leading

items first ignited) provide the figure of about 34%.

Table 14: Non-confined home structure fire deaths by leading items first ignited and smoke alarm status: 2003 to 2006 annual averages [16]

US data on statistics on the causes of home fires (Figure 10) suggests that cooking and heating equipment combined result in more deaths than smoking materials (39% and 25%, respectively).

〔图〕插图

Figure 10: Leading causes of home fires, United States, 2003-2007 [13]

Similar set of data available online [37] (Table 15) indicates a ratio between deaths resulting from flaming fires to deaths resulting from smouldering fires of about 40/23.

Table 15: Major causes of residential fires, civilian deaths and injuries in the United States from 1999-2002 [37]

Similar assessments are possible from the data presented in Table 16 [9].

Table 16: Leading causes of residential building fire deaths in U.S. during 2003-2012 [9]

Death number resulting from “other unintentional” fires is quite comparable with the number of smoking fire deaths and has outnumbered them in some recent years. When combined with deaths resulting from “electrical malfunction”, the figure becomes higher that supposed number of smouldering fire deaths.

When considering detailed breakdown of fire deaths causes [9] (Table 17, Table 18, Table 19 and Table 20)

Table 17: Trend of residential building smoking fires in U.S. during 2003-2012

Table 18: Trend of residential building electrical malfunction fires in U.S. during 2003-2012 [9]

Table 19: Trend of residential building heating fires in U.S. during 2003-2012 [9]

Table 20: Trend of residential building open flame fires in U.S. during 2003-2012 [9]

It appears that, for example, in 2012 580 deaths had flaming fire as a likely cause (electrical malfunction, building heating, open flame), while fires that likely went through smouldering stage resulted in 330 deaths, by comparison.

Further data from the US Fire Administration [38] is presented below in Figure 11. Again, it is clearly seen that electrical fires, combined with other causes result in more deaths compared to fires started by smoking.

〔图〕插图

Figure 11: Causes of fatal residential building fi res in U.S. in 2012 [38]

Study by NIST, Table 21 [4] allows direct comparison between number of deaths in smouldering and flaming fires to be made.

Table 21: Top fire scenarios ranked by frequency of occurrence, 1992-1996 [4]

It is seen from Table 21 that smouldering fires caused 623 fatalities while flaming fires caused 844 fatalities.

It should also be taken into account that cooking fires are top ranked as causes of fire injuries [9, 35, 38].

In the absence of the ways to produce more precise estimation, it is proposed that flaming and smouldering fires are considered as resulting, statistically, in approximately equal number of fatalities.

This balance is reflected in the extended testing program proposed in the Section 4.2

3. PERFORMANCES OF SMOKE ALARMS

3.1 Previous tests on smoke alarms

In the recent years, questions were raised about the efficacy of some smoke alarm technologies, whether their numbers and locations in homes still represents the optimum configuration, and if multi-sensor designs (as are becoming popular for commercial fire alarm systems) might perform better or produce fewer nuisance alarms [4]. Table 22 shows a summary of previous experimental tests on the performance of smoke alarms.

Table 22: A summary of previous experimental tests on the performance of smoke alarms

Almost all the previous tests in Table 22 are real-scale houses. While some tests used laboratory testing room, Su and Crampton [2] used both real-scale and laboratory tests, and Thomas and Bruck [39] utilized the constructed house in a large scale fire testing facility. The houses used for tests included bedroom, kitchen and living room. For example, Cleary [1] used a 15.8 m long and 4.9 wide house, which was wood- framed with interior walls and ceiling covered with gypsum wall board. The tests were taken under different fire location (living room or bedroom) and ventilation (bedroom door

open or closed). Thomas and Bruck [39] constructed the house used cardboard with the doors and windows cut out, and opened or shut as required for each case tested. Su and Crampton [2] performed the tests in a bungalow (93.0 m2), which had two bedrooms, one bathroom, a kitchen, a living and dining room on the main floor. Mealy et al. [3, 40] used an apartment-style enclosure (41.8 m2) comprised of four interconnected rooms to test the performance of smoke alarms.

Some multiple storey buildings were also selected to conduct the experiments. Two story brick structures were used to test the performance of smoke alarms by Harpe et al. [43]. Meland and Lonvik [41] carried out the tests in a room on the ground floor of a three story building. The ground floor is about 17 m2 with a ceiling height of about

1. 5 m. Bukowski et al. [4] conducted their tests in the actual homes with representative sizes and floorplans, utilized actual furnishings and household items for fire sources, and tested actual smoke alarms sold in retain stores. Two different geometries of residential structures were used for the tests: manufactured home geometry (84.7 m2) representing an apartment, condominium or small ranch house, and two-story home with three brick-clad bedrooms (139 m2). In the experiments performed by Gottuk et al. [5], a 49 m3 test room was used, with natural ventilation provided through a 38 cm × 30 cm duct.

All the tests have considered both flaming and smouldering conditions. A two-level screening experimental design was developed to examine the sensitivity to fabric flammability (a slow burning cotton or a fast burning polyester), polyurethane foam density (low density, 21 kg/m3 or high density, 29 kg/m3), considering flaming or initially smouldering fires [1]. Different smoke source were used in Thomas and Bruck’s experiments [39], such as burning ethanol, braided wick, heptane, decalin, smouldering wood, wood cribs, smouldering towel and polyurethane foam. The fuel packages used in Su and Crampton’s experiments [2] included woo, paper, polyurethane foam, cotton flannel and upholstered furniture. Bukowski et al. [4] tested three types of combustion conditions, including smouldering, flaming and cooking conditions.

In some experiments [42], a white and highly light-scattering artificial smoke produced by smoke machine, specifically a paraffin oil mist, was used considered to be comparable to smoke from smouldering materials. The artificial smoke was produced at three widely different rates chosen to be representative of those observed in laboratory tests of materials: low smoke produced by one machine at its lowest rate, considered qualitatively similar to that produced from polyurethane foam upholstered furniture smouldering after contact with a lighted cigarette; medium smoke produced by one machine at a rate comparable to a large smouldering source such as a polyurethane foam upholstered lounge chair; and high smoke produced by two machines at their highest settings, considered comparable to the rate of smoke production from highly smoke producing wall linings after flaming ignition.

For the same type of fuel, both smouldering and flaming fires may occur under different ignition scenarios. In the tests performed by Mealy et al. [3, 40], of the seven tests conducted with smoke alarm in place, there were three flaming ignition scenarios and four smouldering ignition scenarios, including cotton batting, sofa and wooden cabinet. In the tests, two types of sofa were successfully in developing self-sustaining smoulder and flaming fires. And from the tested taken by Gottuk et al. [5], the polyurethane and cotton fabric showed both flaming and smouldering combustion.

These tests included several types of smoke alarm. In the Cleary’s test [1] photoelectric, ionization, and dual photoelectric/ionization alarms were co-located at multiple locations to facilitate comparisons of each alarm type, and di

ff

erent designs for the same type of alarm. In Thomas and Bruck’s [39] tests, each room was fitted with two ionization alarms, two photoelectric alarms and one duel (ionization and photoelectric) alarm. In the tests by Mealy et al. [40], smoke alarm clusters consisted of 3 ionization, 3 photoelectric and 2 dual sensor alarms from three manufactures. Besides the above mentioned smoke alarms, Bukowski et al. [4] also tested the CO alarms under smouldering fire scenarios and the closed-door flaming mattress. Gottuk et al. [5] tested the performance of CO smoke alarm and showed that improved firedetection capabilities can be achieved over standard smoke detectors by combining smoke measurements with CO measurements.

Some tests also considered the transition from smouldering to flaming combustion. For example, Cleary [1] included fires with smouldering to flaming transition times varying from 81 to 182 min. Smouldering chairs were allowed to transition to flaming with no artificial inducement in 11 out of 12 smouldering tests. It was noticed that all smoke alarms activated prior to transition to flaming in all tests, consequently disturbances from an open door may affect the smoke transport prior to alarm activation. Su and Crampton [2] considered fire sources such as pine sticks, cotton, paper, foam, chairs, which transitioned from smouldering conditions to flaming conditions in a range of 68-854 s.

The repeatability was considered in some of the previous tests. Cleary [1] replicated each experimental configuration three times. Bukowski et al. [4] considered the test matrix to perform sufficient replicates to allow estimates of experimental uncertainty and repeatability.

1. 2 Activation time

Most home smoke alarms use ionisation or photoelectric sensor technology, alone or in combination, to detect a fire. Ionisation smoke alarms activate when smoke reduces the flow of current between charged electrodes, whereas photoelectric alarms activate when smoke reflects light beams. Both technologies are effective in most fire scenarios; using their combination offers the best protection. Ionisation alarms are less expensive than photoelectric alarms, which contributes to ionisation alarms being the most widely used [13].

It is largely known from previous studies that ionisation smoke alarm is more sensitive to flaming condition while photoelectric smoke alarm is more sensitive to smouldering condition [4, 30, 41, 43]. Slow, smouldering fires, such as those from cigarettes igniting a mattress or couch, are the types quite often associated with residential fire fatalities, yet the ionisation alarms that are found in 87% of US homes are more sensitive to flaming-type fires [14].

Although fire growth rate has generally increased in modern homes, both photoelectric and ionisation smoke alarms seem capable of giving warnings early enough to provide the necessary escape time in most scenarios. However, some concerns have been separately raised about whether ionization smoke alarms actually do operate early

enough in smouldering fires and whether photoelectric do operate early enough in flaming fires [23].

Table 23 shows a summary of the performance of smoke alarms under flaming condition. The average activation time of ionisation smoke alarms is 249.1 s, which is considerably smaller when comparing to that of photoelectric smoke alarms, namely 337.2 s. The performance of dual smoke alarms is between the two, namely at 270 s.

Table 24 shows a summary of the performance of smoke alarms under smouldering condition. As expected, the photoelectric smoke alarm (average activation time of 1254.3 s) is more sensitive to the smouldering fire, comparing to the ionisation smoke alarm (1746.1 s). The performance of dual smoke alarm is between these two types, which show an average activation time of 1569.2 s.

Table 23: A summary of the performance of smoke alarms under flaming condition

Note: $^{a}$ Combustion transitioned from smouldering to flaming conditions. The condition was considered as smouldering combustion if all the smoke alarms activated before the transition;

$^{b}$ In this reference, only the smoke alarms positioned in the fire of origin are included;

$^{c}$ In this reference, only the activation time of smoke alarms in the closest room to the room of fire origin are shown;

$^{d}$ The dual smoke alarms (dual 2) are from different manufactures with the previous one (dual 1);

e A configuration of every level plus bedroom is selected in this table. It is following the National Fire Alarm Code released in 1993, requiring smoke alarms in every bedroom for new construction in addition to the every level location.

Table 24: A summary of the performance of smoke alarms under smoldering condition

Note: $^{a}$ Combustion transitioned from smouldering to flaming conditions. The condition was considered as smouldering combustion if all the smoke alarms activated before the transition;

$^{b}$ In this reference, only the smoke alarms positioned in the fire of origin are included;

$^{c}$ In this reference, only the activation time of smoke alarms in the closed room to the fire origin are shown;

$^{d}$ The dual smoke alarms (dual 2) are from different manufactures with the previous one (dual 1);

^e A configuration of every level plus bedroom is selected in this table. It is following the National Fire Alarm Code released in 1993, requiring smoke alarms in every bedroom for new construction in addition to the every level location.

A set of full-scale apartment fires with both flaming and smouldering fire scenarios were used to assess the performance of various smoke alarm technologies by Mealy et al. [3]. Individual smoke alarm response times are presented in Table 25 for each test and location within the enclosure. It is evident that in general, for the smouldering fire scenarios, the combination alarms responded the earliest, with photoelectric alarms providing a slightly slower response, and ionisation alarms responding the slowest. On average, combination alarms responded 270 s sooner than photoelectric alarms and 822 s faster than ionisation alarms for the smouldering fire scenarios. In the flaming fires, the ionisation alarms were generally the quickest to respond, with the combination alarms lagging only slightly behind and the photoelectric alarms responding the slowest. In these scenarios, the ionisation alarm responded on average 13 s sooner than combination alarms and 67 s faster than photoelectric alarms.

Table 25: Smoke alarm activation times from source initiation [3]

Note: DNA – Did not activate;

N/P – Alarm not present at this location during test;

N/D – Activation could not be determined due to malfunction.

Thomas and Bruck [39] tested the time of activation of smoke alarms in four houses. Table 26 shows the proportion of alarms that did not activate at rall in the tests. While there was a minor difference between the manufacturers, the major difference was observed between the ionisation and photoelectric alarms, with the photoelectric alarms not activating much more frequently. The other major difference was between the two storey house and the other houses which were all single storey. This difference is principally due to the smoke alarms in the lower storey not activating when the room of fire origin was on the second floor. There was a weak trend to a greater proportion of non-activations in the larger houses compared with the smaller houses and in the more compartmented houses compared with the more open plan houses.

Table 26: Proportion of smoke alarms of each type and manufacturer that did not activate [39]

-40-

Su and Crampton [2] conducted a series of experimental studies in a residential dwelling as well as in a laboratory room to examine the effect of “dead air space” on smoke-alarm response. Table 27 shows the length of the smouldering phase, the total length of smouldering plus flaming phase, and the response times of the smoke alarms at various locations for each experiment. Except for the Test 3 where the cotton flannel fire quickly changed from smouldering to flaming, all other test fires had a relatively long period of smouldering before changing into flaming. Tests 6 and 7 had the same experimental set up except that the amount of air available for combustion was controlled in Test 7 by sealing some of the holes around the metal bucket to lengthen the smouldering period. Since the door of Bedroom 1 was closed initially in Tests 2, 6 and 7, the smoke alarms in the living room did not respond to the fire that was located in Bedroom 1 until the door was opened and, vice versa, most smoke alarms in Bedroom 1 did not respond to the fire that was located in the living room until the door was opened.

Table 27: Experiments in the dwelling and the results [2]

*Bedroom 1 door opened at 1353 s in Test 2 and 1080 s in Tests 6 and 7. Notes, d fpr dual, i for ionization and p for pholelectric smoke alarms. NA for not accurate smoke alarms.

Cleary [44] analysed the data from two full-scale residential smoke alarm fire test series to estimate the performance of dual sensor photoelectric/ionisation alarms as compared to co-located individual photoelectric and ionisation alarms. The NRC Canada tests [45] that used solid combustible sources all started out as smouldering fires with most transitioning to flaming at some time during each test. Likewise, the cooking oil fire produced smoke from the heated oil before igniting. Alarm sensitivities were not measured prior to testing, thus there is no information on relative sensitivity between ionisation, photoelectric or dual alarms. There were 54 instances where a set of alarms were co-located during the 13 individual tests. The average alarm time and standard deviation (SD) for each type of alarm are given in Table 28. The dual alarm responded 616 s faster on average than the ionisation alarm, and 72 s faster on average than the photoelectric alarm.

Table 28: Average alarm times for NRC Canada Test Series [44, 45]

The NIST test series [4] were also analysed by Cleary [44]. There were 92 instances where a set of alarms were co-located during the 30 fire tests. The average alarm times and standard deviations for the ionisation, photoelectric, and dual alarm configuration are shown in Table 29. On average, all three dual alarm configurations provide faster average alarm times compared to the photoelectric, or ionisation alarms at any one of the three ionisation sensor sensitivities. Considering the instances when the ionisation alarm (at a sensitivity setting of 4.3%/m) responded first, the dual alarm

on configurations (from high to low sensitivities) activated 89 s, 67 s, and 47 s faster on average than the photoelectric alarms. Conversely, considering the instances when the photoelectric alarms responded first, the dual alarm configurations (from high to low sensitivities) activated 535 s, 523 s, and 518 s faster on average than the ionization alarms.

Table 29: Average alarm times for the NIST test series [4, 44]

Note: Italicized entries highlight sensitivity settings used in the NIST report analysis

The situation for flaming and smouldering fires in the NIST test series [4] were also analysed by Cleary [44]. There were 36 instances of co-located alarms during initially flaming fires. Table 30 gives the mean, median and standard deviation of the alarm times for initially flaming fires with the bedroom door opened. The dual alarm configurations yielded faster average alarm times than the photoelectric alarm and average alarm times nearly equivalent to the ionisation alarms. Table 31 gives the mean, median and standard deviation of the alarm times for initially smouldering fires with the bedroom door opened. The dual alarm configurations yielded much faster average alarm times than the ionisation alarms and average alarm times nearly equivalent to the photoelectric alarm. Table 32 gives the mean, median and standard deviation of the alarm times for the cooking fires. The dual alarm configurations yielded faster average alarm times than the photoelectric alarm.

Table 30: Alarm time statistics for the NIST test series of initially flaming fires [4, 44]

Note: Italicized entries highlight sensitivity settings used in the NIST report analysis

Table 31: Alarm time statistics for the NIST test series of initially smoldering fires [4, 44]

Note: Italicized entries highlight sensitivity settings used in the NIST report analysis.

Table 32: Alarm time statistics for the NIST test series of kitchen fires [4, 44]

Note: Italicized entries highlight sensitivity settings used in the NIST report analysis

A series of 24 full-scale experiments was conducted to examine the effects of alarm type, alarm location, fabric type, polyurethane foam density, ignition scenarios, and room configuration, on smoke alarm performance by Cleary [1] in a two-level, fractional factorial design of eight experimental configurations. Alarm times for only two of the three configuration 7 trials were averaged in Table 33. It is evident that photoelectric alarm responded quicker on average than ionization alarm in two of four smouldering fire configurations. And ionization alarm responded quicker on average than photoelectric alarm in all four flaming fire configurations. It is also indicated that dual alarm D2 had the fastest average alarm time for all four smouldering fire configurations, and responded first in 11 of the 12 trials. And dual alarm D2 yielded faster average alarm times than dual alarm D1 in seven of eight configurations, and was the first dual alarm to respond in 22 out of 23 trials where dual alarms were present.

Table 33: Average alarm times for each configuration for living room tests [1]

Note: The number in parenthesis is the standard deviation.

a Average of two tests

Milarcik et al. [20] conducted a statistical study to compare the performance of different residential smoke detector technologies when exposed to different fire types, such as flaming, smouldering and kitchen fire. Three detector studies were included, such as Dunes 1970s [30, 43], Kemano [45] and Dunes 2000 [4]. Table 34 shows the relative time statistics for the performance of smoke alarms under flaming, smouldering and kitchen conditions. The statistical results indicate that, on average, one detector will react faster to smoke from a flaming fire than the other two detectors under different conditions. For example, under flaming condition, the Common language Effect Size results indicate that there is only a 72% chance that an ionisation detector will activate before a photoelectric detector, and only a 71% chance that a dual detector will activate before a photoelectric detector. Furthermore, there is only a 51% chance that an ionisation detector will activate before a dual detector. The odds represent the ratio for one detector activate faster than the other detector, which is calculated from the Common language Effect Size results. It is demonstrated that ionisation, photoelectric, and dual detectors provide statistically equivalent warning to different types of fires. It was confirmed, as, on average, ionisation detectors performed better for flaming fires, and photoelectric detectors performed better for smouldering fires. Dual detectors were actually better, on average, for smouldering fires than photoelectric detectors and were superior to single-type technologies, when all tests were considered as a whole.

Table 34: Relative time statistics for the performance of smoke alarms [20]

Bukowski et al. [4] tested a range of residential smoke alarm technologies in a controlled laboratory test and in a series of real-scale tests conducted in two different residential structures: manufactured home geometry and a two-storey brick home. Table 35 and Table 36 show the average time to alarm for several smoke alarms and fire scenarios in a manufactured home and two-story home, respectively. Three installation criteria were used in the tests: every level; every level plus bedrooms; and every room. The "every level" installation represents the minimum arrangement allowed by code, in which smoke alarms are required outside the sleeping rooms and on each additional story of the home. In 1993 the National Fire Alarm Code was revised to require smoke alarms in every bedroom for new construction in addition to the every level location, which is named as "every level plus bedrooms". The greatest escape time would be only guaranteed if smoke alarms were required in every room, called "every room", an arrangement that has never been required in any code. It is indicated that, consistent with prior findings, ionisation type alarms provided somewhat better response to flaming fires than photoelectric alarms, and photoelectric alarms provided (often) considerably faster response to smouldering fires than ionisation type alarms. Smoke alarms of either type installed on every level generally provided positive escape times for different fire types and locations. Also the results demonstrated that adding smoke alarms in bedrooms increased the escape time provided, especially for smouldering fires.

Table 35: Average time to alarm for several smoke alarms and fire scenarios in a manufactured home [4]

Table 36: Average time to alarm for several smoke alarms and fire scenarios in a two-story home

Overall, the variety of studies is consistent in pointing out to the fact that photoelectric alarms are more efficient to alert to smouldering fires while ionisation alarms are generally more efficient to respond to flaming conditions. This is, of course, expected given the operating principles of the two technologies.

There is no solid evidence to prefer one technology over the other. In essence, Table 23 and Table 24 providing the direct comparisons, are most informative. They demonstrate that photoelectric detectors respond to smouldering fires approximately 39% faster (in terms of average activation times), and ionisation detectors respond to flaming fires approximately 35% faster. The two figures are statistically indistinguishable.

The analysis in Section 2.3 demonstrates that available fire data, although not entirely complete satisfactory, seems to indicate that both scenarios are equally important taking into account fire frequencies, as well as death and injuries rates.

If this is indeed the case, then it must be concluded (in line with some of the studies discussed above) that ionisation, photoelectric, and dual detectors provide statistically equivalent warning to different types of fires.It is also worth noting that some studies conclude that only in extreme cases the difference between photoelectric and ionisation detectors would be critical in saving lives. There are also studies indicating that photoelectric alarms may fail to activate much more frequently than ionisation alarms.

3.3 Nuisance alarm

Nuisance activations interrupt other activities and may lead people to ignore the early warning of a smoke alarm. They are the leading reason for deliberately disabling smoke alarms [46]. In 2003, U.S. fire departments responded to 2,189,500 false alarms and only 7% were to fires [47]. In New Zealand, false alarms accounted for almost 40% of the fire call responses during 1998 and about half of all false alarms came from fire alarm systems [48]. In the United Kingdom, false alarms accounted for about 47% of the fire call responses in 2001, with 27% of all fire calls being false calls "due to apparatus" [49]. In the National Smoke Detector Project, 97% of the devices tested for involvement in nuisance alarm were ionisation-type devices, although they comprised only 87% of all devices in the study [50].

A United States Consumer Product Safety Commission (CPSC) report released in 1994 [51] indicated that while 87% of smoke alarms in United States homes have ionization sensors, 97% of nuisance alarms are reported by those units. A further study in 2000 installed smoke alarms in Alaskan Eskimo homes between 10 and 15 ft from the cooking appliances. This study found that 92% of homes with ionization alarms recorded nuisance alarms while only 11% with photoelectric alarms recorded nuisance alarm. Figure 12 shows the reported fires and false alarms in United States during 1980 and 2003 [46].

〔图〕插图

Figure 12: Reported fires and false alarms in United States: 1980-2003 [46]

For United States, Figure 13 shows the reasons that smoke alarm failed to operate in reported non-confined home structure fires during 2003 and 2006 [16, 23, 52, 53]. It is known that power source issues were the leading reason smoke alarms failed to operate, with missing or disconnected batteries being the leading problem. It is also shown that in 62% of the fires in which battery-powered smoke alarms failed to sound, the batteries were missing or disconnected. Dead or discharged batteries accounted for 26% of the battery-powered smoke alarm failures. When hardwired smoke alarms with no battery backup failed to operate, the power had failed, been shut off, or disconnected in 62% of the fires. When hardwired smoke alarms with battery backup failed to operate, 31% of the failures were due to hardwired power failure, shut off, or disconnection; 23% were due to missing or disconnected batteries; and 3% were due to dead or discharged batteries.

〔图〕插图

Figure 13: Reason smoke alarm failed to operate in reported non-confined home structure fires: 2003 to 2006 (%) [16, 23, 52, 53]

In New Zealand, the presence of a smoke alarm/detector was determined in 418 (24.6%) of the cases from 1999 to 2006. Where it was identified that a smoke alarm/detector was present, it was found that they operated in 329 (78.7%) of the cases and alerted 230 of the fire injury victims that there was a fire in their property [8]. Where it was possible to determine a reason for the ineffectiveness of smoke alarm/detectors, the most frequently cited reasons are shown in Figure 14.

〔图〕插图

Figure 14: Reasons for ineffectiveness of smoke alarm/detectors in New Zealand during 1999 and 2006

In Australia [8], the presence of a smoke alarm/detector was determined in 2441 (33.3%) of the cases during 1999 and 2006. Where it was identified that a smoke alarm/detector was present, it was found that they operated in 1811 (74.2%) of the cases and alerted 1406 of the occupants or fire injury victims that there was a fire in their property. Where it was possible to determine a reason for the ineffectiveness of smoke alarms/detectors, the most frequently cited reasons were shown in Figure 15.

〔图〕插图

Figure 15: Reasons for ineffectiveness of smoke alarm/detectors in Australia during 1999 and 2006

Table 37 shows a summary of the studies on nuisance alarms. Two types of studies have been identified: direct experimental tests; and field studies. Nuisance alarms in residential settings from typical cooking activities, smoking or candle flames were tested by Bukowski et al. [4]. Fazzini et al. [14] conducted a cohort study of 4 rural Alaskan villages. It was concluded that the incidence of nuisance alarms is much higher in small dwellings using ionisation smoke alarms. The higher rates of alarm disconnection in the homes with ionisation alarms are likely related to the high rate of nuisance alarms in these homes. The use of photoelectric smoke alarms in small dwellings may lead to a lower rate of disconnection and improved survival in the event of fire. Feng and Milke [51] carried out tests to assess the sensitivity of different alarm technologies to a range of common problem scenarios.

Table 37: A summary of studies on the nuisance alarms

Gottuk et al. [5] tested the performance of smoke alarms under various nuisance sources. Table 38 shows the activation time of smoke alarms under different nuisance scenarios. It is evident that in 3 out of 4 cases the photoelectric smoke alarms activate faster than the ionisation smoke alarm. And for the exceptional case that the activation time for ionisation smoke alarm (891 s) is very close to that of photoelectric smoke alarm (818 s).

Table 38: Smoke alarms to different nuisance scenarios [5]

Feng and Milke [51] used six nuisance sources to assess the sensitivity of different alarm technologies to a range of common problem scenarios. Table 39 shows the probability of activation for each technology to given nuisance source and to the total test series.

Table 39: Total probability of activation for different types of smoke alarm [51]

1. DEVELOPMENT OF ADDITIONAL TESTING SCENARIOS

1. 1 Principles of performance testing

To test the performance of different types of smoke alarms, several aspects need be considered in order to obtain an objective outcome, not biased toward any type. It includes the selection of fire sources, ignition methods, areas of fire origin, ventilation conditions, as well as considerations of repeatability, fire growth rates, nuisance alarms and potentially other factors.

The selection of fire sources should represent flaming and smouldering and cooking conditions. A study by Ahrens [16] showed that the leading items first ignited in United States home fires from 2003 to 2006 are, upholstered furniture (22%), mattress or bedding (18%), flammable or combustible liquid or gas, or pine, hose, duct or filter (9%), clothing (8%), unclassified furniture or utensil (5%), structural member of framing (4%), cooking materials including food (4%), etc. Both smouldering and flaming fires may develop from the same type of fuel, depending on ignition scenarios. For example, in the tests performed by Gottuk et al. [5], the polyurethane and cotton fabric demonstrated both flaming and smouldering combustion.

Various ignition methods were used in previous tests. Geiman [25] and Su et al. [45] used an electronic heating element as the source of ignition energy for the tests in both flaming and smouldering ignition. Bukowski et al. [4] utilized different methods for different types of fire. An electric match was chosen to start flaming ignition, a rod made with nichrome wire enclosed in ceramic was used in a smouldering scenario, and cooking condition was produced by a 0.3 m diameter aluminium sauté pan. Butane lighter and torch were used to ignite flaming oily rags, paper, cardboard in the tests [33].

Area of fire origin is also an important influencing factor to test the alarm performance. Common locations for fires to start include kitchen, bedroom, living room and even laundry room. A statistics by Australasian Fire and Emergency Service Authorities Council [8] shows that the majority of injured person in Australia from 1999 to 2006 were in kitchen (2927), bedroom (1600), lounge (843), laundry (204), garage (183), etc. National Fire Protection Association indicated [55] that from 2007 to 2011 in United States cooking equipment represented the leading causes of total fires (43%), followed by heating equipment (16%), intentional (8%), electrical distribution and lighting equipment (6%) and smoking materials (5%). It should be noted that cooking fire is one of the main causes of home fire. It is also indicated in the FRNSW report [6] that 25% of total number of fires for class 1a & 2 building during 2000-2014 started in kitchen, 18% in bedroom and13% in the lounge room.

Ventilation condition is an important factor has influence on the smoke movement, resulting in the variation of alarm activation times. Thomas and Bruck [39] used fullscale model house to conduct the tests with the doors and windows cut out and opened or shut as required for different ventilation conditions. Cleary [1] also conducted tests under different ventilation conditions (bedroom door open or closed). The average activation time of different types of smoke alarms reduced from 1775 s to 1269 s after the bedroom door was closed. Meland and Lonvik [41] indicated that with a flaming fire and open door to the corridor the critical limit for being in the corridor

associated with heat stress and reduced visibility was typically reached after 200-240 s, depending on the size of the corridor.

Residential bathrooms and kitchens are the most common locations for nuisance alarms [51]. Bukowski et al. [4] performed nuisance tests in a manufactured home and the selection of sources were based on what are commonly thought to be causes of residential nuisance alarm to mimic expected activities. Feng and Milke [51] tested the performance of smoke alarms under six nuisance sources, such as toast, cooking onions, hamburgers, vegetable oil, steam and cement dust. It is suggested that the tests should be split into a 'pure nuisance' phase and an 'aggressive nuisance' phase. The threshold obscuration value chosen for separating of the two phases of each test was based subjectively on the average obscuration exceeding 0.15% obscuration per foot, in part accounting for the systematic errors in the optical density meter.

Repeatability was taken care of in some of the previous tests. It was shown that repeated experimental runs are necessary as many tests they showed high standard deviations. Cleary [1] replicated each experimental configuration three times. Bukowski et al. [4] considered the test matrix adequate to perform sufficient replicates to allow estimates of experimental uncertainty and repeatability.

High fire growth rate may be not good to identify the performance of different types of smoke alarms, as if fire progresses too fast the three types of smoke alarm may produce very similar results. To address this issue Su and Crampton [2] considered variety of fire sources such as pine sticks, cotton, paper, foam, chairs, which transitioned from smouldering conditions to flaming conditions over a wide range of times (68-854 s).

4.2 The design of additional testing scenarios

FRNSW study [6] considered 10 different burning scenarios conducted in the FRNSW testing rig, including 8 smouldering scenarios and 2 flaming scenarios. The ignition sources were cartridge heater and LPG gas flame.

Due to the drawbacks of the study described in Section 2.1, extended testing program is proposed. It is assumed that the same FRNSW testing facility will be used for additional tests, shown in Table 40.

Table 40: A summary of additional tests for FRNSW study [6]

The following major changes are implemented in the extended testing program: Additional clusters of smoke alarms to be installed in the Laundry room and in the Bathroom (for the purpose of testing response to nuisance alarms).

Equal number of tests is proposed for flaming and smouldering combustion. This reflects analysis on the statistics of causes of fatal fires carried out in the Section 2.3

Test repeatability is required in order to ensure that the results of each tests are statistically stable, and that the Standard Deviations in the alarm activation times can be established in each of the scenarios. Each fire scenario needs be repeated at

least three (3) times to confirm the repeatability of testing conditions, including the scenarios already considered in the FRNSW report. Methodology which is to be followed for processing the results is discussed below. It is desirable, if possible, to repeat each of the tests more than three times.

The group of the first ten tests (highlighted in the Table 40 are the same as in the FRNSW study, but the ignition source in smouldering scenarios is replaced. Energy source used to initiate smouldering scenarios (cartridge heater) is extreme and not representative of real scenarios. This should be replaced with burning cigarette, match or small electric heater with comparable size and heat output.

Energy output of the LPG gas burner, used to initiate flaming combustion, should also be reduced to consider slower fire growth rates. This will allow more clear distinction between alarms response times to be made.

Selection of materials used as a “first item to ignite” reflect available fire history statistics

Robust statistical methodology, available from some of the previous studies, is suggested to process testing results and discussed below.

· Test of smoke alarms response to naissance sources is suggested. This is an important issue to be considered in the context of evaluating performance of smoke alarms. Frequent nuisance responses are likely to lead to smoke alarms being disconnected or being ignored if activated.

Choice of room of fire origin reflects statistics provided by [4], Table 8 below.

Table 41: Top fire scenarios ranked by frequency of occurrence, 1992-1996 [4]

Suitable and robust statistical processing of smoke alarm activation times is proposed [20]. The essence of this methodology is a Common Language Effect Size (CL) statistical method. The CL calculation takes two population distributions (e.g. activation time distributions of alarms of different types) and evaluates the extent of overlap of these distributions.

A normal distribution is created with a mean equal to the difference of the two means, and a standard deviation equal to the root square of the variances of the two populations

$$Z_0=\frac{\left|mean_1-mean_2\right|}{\sqrt{\left(\sigma_1^2+\sigma_2^2\right)}}$$

Equation 4

The Z score calculated by the equation (4) is utilised in the normal distribution lookup tables to determine the probability that one type of smoke detector will activate faster than another.

If the value

$$CL = P(Z < Z_0)$$

Equation 5

is close to unity, then one of alarms will activate significantly faster, or in other words will have high Odds

$$Odds = \frac{CL}{CL - 1}$$

Equation 6

of being faster than the other detector. If the CL value is close to zero, then the two technologies have equal odds to activate first.

The proposed testing program represents a minimum set of tests, which seems to be realistic to implement within reasonable time frame and with reasonable resources, using the existing FRNSW testing facility. Much more extensive program can be developed should the demand for it becomes evident.

1. CONCLUSIONS

The present study addresses ABCB concerns that some of the recent studies conducted to compare performance of photoelectric and ionisation smoke alarms may not employ sound technical methodology and may be biased towards favourable outcome towards one type of smoke alarms.

The aims of this study was to conduct a critical review of available data on comparative performance of photoelectric versus ionisation smoke alarms, and propose an objective and representative testing scenarios in order to obtain accurate data on performance of these two types of detectors..

Comprehensive literature review was conducted, included recent FRNSW study [6], as well as a variety of other sources. Both Australian and international (New Zealand and the United States) data was considered.

The literature review has focused on the two major issues:

1) historical fire data for residential premises, and in particular major factors influencing performance of photoelectric and ionisation alarms, and probabilistic distributions of these factors across real fire accidents;

2) review of available studies on comparative performance of photoelectric and ionisation smoke detectors.

With respect to real fire scenarios, it has been established that the two major factors need be taken into account: First is the type of combustion that is flaming or smouldering. This factor is of paramount importance as the two types of smoke alarms under consideration respond in a significantly different way to these two types of combustion process.

The second factor is a statistical number of fatalities occurring in the fires of these two types.

It was concluded that an ensemble of representative fire scenarios for assessing performance of smoke alarms should properly reflect historical distribution of fatalities between various scenarios.

There are uncertainties in the existing fire data which prohibits exact number of fatalities in both types of fires to be calculated. This is reflected primarily by the fact that many reported scenarios cannot be confidently attributed to either flaming or smouldering type, although they can be classifies with a reasonably high degree of confidence.

Based on available data, it has been concluded that number of fatalities in smouldering and flaming fires may, in fact, be comparable.

Based on this outcome, equal importance was assigned to the two types of combustion when developing proposed testing program.

Existing data, reported in various studies, has been found to be consistent with the general perception that photoelectric smoke alarms respond faster to smouldering

fires while ionisation alarms respond faster to flaming conditions. This outcome is also consistent with operating principles of the two technologies.

Across the set of experiments, photoelectric detectors responded to smouldering fires approximately 39% faster (in terms of average activation times), and ionisation detectors responded to flaming fires approximately 35% faster. The two figures are statistically indistinguishable.

Given the identified evidence on fire fatalities in the two types of fire scenarios, it is concluded that ionisation, photoelectric, and dual detectors provide statistically equivalent warning to di

ff

erent types of fires.

In line with this conclusion, some reviewed studies conclude that only in extreme

cases the difference between photoelectric and ionisation detectors would be critical in saving lives. There are also indications that photoelectric alarms may fail to

activate much more frequently than ionisation alarms.

A number of major drawbacks have been identified in the research methodology employed in the FRNSW study [6].

Major of these drawbacks are

1) a very limited number of tests absence of tests repeatability;

2) lack of proper review of other existing studies and statistical fire data;

3) Severe imbalance in the testing program, between smouldering and flaming fire conditions, with flaming scenarios being effectively ignored. This include total absence of kitchen fire scenarios.

4) not representative energy sources used to imitate some of the burning scenarios. It is hard to see how the study [6] can be seen as a benchmark for making major regulatory decisions regarding implementation of domestic smoke alarms. It is just one of many studies on smoke alarm performance, in fact with a very limited coverage of possible fire scenarios.

To address identified deficiencies of the FRNSW research methodology, an extended testing program has been proposed. It reflects outcomes of the comprehensive literature review.

The program consists of 30 tests, with required number of replications. The program also identifies nuisance alarms as an important issue, and proposes relevant testing.

REFERENCES

Cleary T, Results from a Full-Scale Smoke Alarm Sensitivity Study. Fire Technology, 2014. 50(3): p. 775-790.

Su JZ and Crampton GP, An experimental examination of dead air space for smoke alarms. Fire Technology, 2009. 45: p. 97-115.

Mealy CL, Wolfe A, and Gottuk DT, Smoke Alarm Response: Estimation Guidelines and Tenability Issues – Part 2. 2009, National Fire Protection Association.

Bukowski RW, et al., Performance of home smoke alarms analysis of the response of several available technologies in residential fire settings. 2007, National Institute of Standards and Technology.

Gottuk DT, et al., Advanced fire detection using multi-signature alarm algorithems. Fire Safety Journal, 2002. 37: p. 381-394.

Fire and Rescue NSW, Smoke alarms in homes: an analysis. 2015, Fire & Rescue NSW.

Ahrens M, Home structure fires. 2013, National fire protection association.

Australasian Fire and Emergency Service Authorities Council, Accidental fire injuries in residential structures: who's at risk? 2009.

U.S. Fire Administration, Residential Building Fires. 2014.

1. Australasian Fire Authorities Council, Accidential fire fatalities in residential structures: who's at risk? 2005.

1. Xiong L, Ball M, and Bruck D, Comprarative investigation of 'survival' and fatality factors in

accidental residential fires. 2013, Victoria University.

1. Australasian Fire Authorities Council, Accidental fire fatalities in residential structures: who's at risk? 2009.

1. Diekman S, Ballesteros MF, and Ahrens M, Home fires in America: progress and

opportunities. American Journal of Lifestyle Medicine, 2012. 6: p. 141-151.

1. Fazzini TM, Perkins RM, and Grossman D, Ionization and phtoelectric smoke alarms in rural Alaskan homes. Western Journal of Medicine, 2000. 173(2): p. 89-92.

1. Miller TR, et al., Increasing smoke alarm operability through theory-based health education: a radomised trial. Journal of Epidemiology & Community Health, 2014. 68: p. 1168-1174.

1. Ahrens M, Smoke Alarm Presence and Performance in U.S. Home Fires. Fire Technology,

1. 47: p. 699-720.

1. Mueller BA, et al., Randomized controlled trial of ionization and photoelectric smoke alarm functionality. Injury Prevention, 2008. 14(80-86).

1. Newman JS, Modified theory for the characterization of ionization smoke detectors. Fire

Safety Science, 1994. 4: p. 785-792.

1. Consumers Union, Smoke detectors; essential for safety. Consumper Reports, 1994. 59: p.

336-339.

1. Milarcik EL, Olenick SM, and Roby RJ, A relative time analysis of the performance of

residential smoke detection technologies. Fire Technology, 2008. 44(4): p. 337-349.

1. Hall JR, U.S. Experience with smoke detectors. 1994, National Fire Protection Association. p. 36-46.

1. Smith C, Smoke Detector Operability Survey: Report on Findings. 1994, United States

Consumer Product Safety Commission.

1. Ahrens M, Home smoke alarms: the data as context for decision. Fire Technology, 2008. 44: p. 313-327.

1. Helsper C, et al., Verification of ionization chamber thoery. Fire Technology, 1983. 19(1): p. 14-21.

1. Geiman JA, Evaluation of Smoke Detector Response Estimation Methods, in Fire Protection

Engineering. 2003, University of Maryland, College Park.

1. Bukowski RW and Mulhoooland GW, Smoke detector design and smoke properties. 1978,

National Bureau of Standards.

1. Schifiliti RP, Meacham BJ, and Custer RLP, Design of detection systems, in SFPE Handbook of Fire Protection Engineering (Third Edition). 2002, National Fire Protection Association.

1. Schifiliti RP, Meacham BJ, and Custer RLP, Design of detection systems, in SFPE Handbook of Fire Protection Engineering. 1995.

1. Geiman JA and Gottuk DT. Alarm thresholds for smoke detector modeling. in The 7th International Symposium on Fire Safety Science. 2002. Worcester Polytechnic Institute, USA.

1. Bukowski RW, Waterman TE, and Christian WJ, Detector Sensitivity and Siting Requirement for Dwellings. 1975, National Bureau of Standards, USA.

1. Gottuk DT, et al., Identification of Fire Signatures for Shipboard Multi-criteria Fire Detection Systems. 1999, Novel Research Laboratory.

1. Rose-Phersson SL, et al., Multi-Criteria Fire Detection Systems Using a Probalistic Neural Network. Sensors and Actuators B: Chemical, 2000. 69(3): p. 325-335.

1. Wong JT, et al., Results of Multi-Criteria Fire Detection System Tests. 2000, Naval Research Laboratory.

1. Spearpoint MJ and Smithies JN, Practical Comparison of Domestic Smoke Alarm Sensitivity Standards. 1997, Fire Research Station, Home Office Fire Research and Development Group.

1. Xiong L, Ball M, and Bruck D, Causal and risk factors for accidental residential fires: a literature review (2004-2015). 2015, Victoria University.

1. Australian Bureau of Statistics, Australian Social Trends 2000. 2000.

1. National Fire Protection Association. [cited 2015 22nd June]; Available from: www.nfpa.org/index.asp.

1. U.S. Fire Administration. U.S. Fire Statistics. 2015; Available from: www.usfa.fema.gov/data/statistics.

1. Thomas I and Bruck D. The time of activation of smoke alarms in houses - the effect of location, smoke source, alarm type and maufacturer, and other factors. in Suppression, Detection and Signaling Research and Applications Symposium. 2010. Orlando, Florida, USA.

1. Mealy CL, Wolfe A, and Gottuk DT. Smoke alarm response and tenability. in The 14th International Conference on Automatic Fire Detection. 2009. Universitat Duisburg-Essen, Germany.

1. Meland O and Lonvik LE. Detection of smoke: full-scale tests with flaming and smouldering fires. in The 3rd International Symposium on Fire Safety Science. 1991. Edinburgh, Scholand.

1. Johnson PF and Brown SK, Smoke detection of smoldering fires in a typical Melbourne dwelling. Fire Technology, 1986. 22(4): p. 295-310.

1. Harpe SW, Waterman TE, and Christian WJ, Detector Sensitivity and Siting Requirements for Dwelling: Phase 2. 1977, National Bureau of Standards.

1. Cleary T, Performance of Dual Photoelectric/Ionization Smoke Alarms in Full-Scale Fire Tests. Fire Technology, 2014. 50(3): p. 753-773.

1. Su JZ, et al., Kemano Fire Studies - Part 1: Response of Residential Smoke Detectors. 2003, National Research Council of Canada.

1. Ahrens M, U.S. experience with smoke alarms and other fire detection/alarm equipment. 2004, National Fire Protection Association.

1. Karter MJ, Fire Loss in the United States During 2003. 2004, National Fire Protection Association.

1. New Zealand Fire Service. False alarms costs. 2004; Available from: www.fire.org.nz.

1. The Office of the Deputy Prime Minister, Fire Statistics, United Kingdom 2001. 2003.

1. Smith CL, Smoke detector operability survey: report on findings. 1993, Consumer Product Safety Commission.

1. Feng JT and Milke JA, Analysis of the Response of Smoke Detectors to Smoldering Fires and Nuisance Sources. 2012, University of Maryland, College Park.

1. United States Fire Administration National Fire Date Center, National Fire Incident Reporting System 5.0. 2008.

1. Smith CL, Smoke detector operability survey: report on findings. 1994, Consumer Product Safety Commission.

1. Hall JR, Experience with smoke detectors and other fire detectors, Who has them? How well do they work? When don't they work? 1996, National Fire Protection Association.

1. National Fire Protection Association, U.S. home structure fires fact sheet. 2014, National Fire Protection Association.