机器学习在煤粉着火敏感性预测中的应用

摘 要：为探究机器学习模型预测煤粉云最低着火温度（Minimum Ignition Temperature,MITc）的可行性,收集Godbert-Greenwal炉测试得到的煤粉云最小着火温度和影响因子数据,并分析了影响因子的关联性。利用AUC/ROC、Kappa系数、敏感性、特异性、MAE、RMSE等指标对三种机器学习模型在煤粉云最小着火温度和着火概率两方面的预测效果进行评价。结果发现：RSM模型的预测效果最差;RF模型在预测煤粉云MITc和着火概率时具有较好的精度和稳定性;Bagging模型在预测着火概率时AUC值均大于0.85,但预测MITc时效果较差。结果为煤粉云着火敏感性预测提供了一种新的研究思路。     

Sprinkler protection of manufactured homes with sloped ceilings using prototype limited water supply sprinklers

Three full-scale fire tests were conducted in a commercial manufactured home 4.3 m wide by 18.3 m long with 10° sloped ceilings. The purpose of the tests was to determine the performance of limited water supply (LWS) sprinklers under a sloped ceiling and to determine whether the minimum distance allowed between sprinklers could be reduced from 2.4 m to 1.8 m. Six prototype LWS sprinklers were installed at a spacing of 1.8 m by 2.4 m in the living room/kitchen area, which measured 4.0 m by 7.2 m. The ceiling and walls were combustible in all three tests, and the room was ventilated by three open windows. The fuel package for all three tests consisted of a vinyl-covered chair with polyurethane padding, double-layered curtains (100% polyester and 50% cotton/50% polyester), and a wooden end table. In the first test, the fuel package was placed in a corner; in the second test, along the wall under the apex of the ceiling between two sprinklers; and in the third test, along the longitudinal wall between two sprinklers. In all tests, no more than two sprinklers actuated. The flow rate with one and two sprinklers operating was 38 pm and 49 pm, respectively. Flow after the first sprinkler actuation was maintained for 10 minutes when one sprinkler actuated and for 7 minutes 40 seconds when two sprinklers actuated. The total flow in each of the three tests was 380. In all three tests, the sprinklers prevented the curtains and wall paneling from becoming involved in the fire during sprinkler operation. Tenability based on carbon monoxide concentration and gas temperature at the 1.5 m level was maintained in all three tests.     

Proceedings: Ad Hoc Mathematical Fire Modeling Working Group

The Ad Hoc Mathematical Fire Modeling Working Group was organized about seven years ago to facilitate voluntary cooperation and coordination in developing mathematical fire modeling capability. The group has a steering committee of representatives of agencies that support fire modeling and related research. These include the National Bureau of Standards' Center for Fire Research, Factory Mutual Research Corp., the Naval Research Laboratory, NASA, and the Federal Aviation Agency. The Group holds plenary meetings when it seems desirable to do so (about once each 1 1/2 years), and workshops on topical subjects. Three workshops were held in 1983. Normally, Group meetings are held before or after other meetings at which a number of interested personnel would likely be present. Minutes are mailed to those who attended and to others (including European and Japanese personnel) who have indicated interest. This workshop was arranged by Ron Alpert of Factory Mutual Research Corp. and held at their facilities at Norwood, MA, in November 1983. National Bureau of Standards REFERENCE: Levine, Robert S., “Proceedings: Ad Hoc Mathematical Fire Modeling Working Group”, Fire Technology Vol. 20, No. 2 May 1984, p. 47.     

Field test and evaluation of residential sprinkler systems: Part III

A series of full-scale fire tests utilizing prototype “quick-response” sprinklers was conducted in a two-story residence in Los Angeles, CA, and a mobile home in Charlotte, NC. A prime objective of the program was to test the performance of alternative sprinkler designs to “control” the development of fire in single-family dwellings and mobile homes. Part of the mobile home test series focused on a study of the comparative activation times of sprinklers and smoke detectors installed in the same areas under a variety of test conditions. The results of that phase of the test program are given here. National Fire Protection Association REFERENCE: Cote, arthur E., “Field Test and Evaluation of Residential Sprinkler Systems”, Fire Technology, Vol. 20, No. 2, May 1984, p. 41. Mr. Cote is Assistant Vice President (Engineering & Technical Services) for the NFPA. He served as project manager for the test series insofar as the NFPA had overall management responsibilities for the program.     

Workshop on flame radiation and soot

The Ad Hoc Mathematical Fire Modeling Working Group was organized about eight years ago to facilitate voluntary cooperation and coordination in developing mathematical fire modeling capability. The group has a steering committee of representatives of agencies that support fire modeling and related research. These include the National Bureau of Standards' Center for Fire Research, Factory Mutual Research Corp., the Naval Research Laboratory, NASA, and the Federal Aviation Agency.The Group holds plenary meetings when it seems desirable to do so (about once each 1 1/2 years), and workshops on topical subjects. Normally, Group meetings are held before or after other meetings at which a number of interested personnel would likely be present. Minutes are mailed to those who attended and to others (including European and Japanese personnel) who have indicated interest. This workshop was held at the National Bureau of Standards Center for Fire Research, Gaithersburg, MD, in May 1984. Reference: Robert S. Levine, Workshop on Flame Radiation and Soot, Proceedings: Ad Hoc Mathematical Fire Modeling Working Group,Fire Technology, Vol. 21, No. 1, February 1985. p. 41. Note: The author accepts responsibility for any pertinent omissions or distortions. If you have any questions concerning statements made by an individual group member, please contact that member directly.     

Field test and evaluation of residential sprinkler systems: Part I

A series of full-scale fire tests involving sprinkler installations was conducted in 1979 and 1980 in a two-story residence in Los Angeles, CA, and in a mobile home in Charlotte, NC. Previous laboratory tests, including full-scale room fire tests, had indicated that a prototype “quick-response” sprinkler was effective in controlling residential home fires. More full-scale fire tests under less controlled and more operational conditions were considered necessary to confirm or deny these scientific and engineering judgments. The prime objective of the program was to test the performance of alternative sprinkler designs to “control” the development of fire in single-family dwellings and mobile homes. National Fire Protection Association Editor's Note: This paper is the first of three papers reporting on full-scale residential sprinker fire tests conducted under Grant 79027 for the U.S. Fire Administration by the National Fire Protection Association. In this paper, the objectives of the test program are outlined; the test facilities, Prototype sprinkler, water supplies, sprinkler system design, and instrumentation are described; and the scenarios for the total of seventy-six tests conducted are summarized. The second paper (February 1984 issue) will focus on the results of the test and the conclusions drawn from them and the third (May 1984 issue) will cover a study of the comparative activation times of sprinklers and smoke detectors in the mobile home tests. Mr. Cote is Assistant Vice President (Engineering & Technical Services) for the NFPA. He served as project manager for the test series insofar as the NFPA had overall management responsibilities for the program.     

Uncertainty in Estimating the Fire Control Effectiveness of Sprinklers from New Zealand Fire Incident Reports

The current state of fire sprinkler effectiveness information has been found to be a limiting factor when comparing the fire risk for alternative building designs in New Zealand (Determination 2005/109: single means of escape from a high-rise apartment building. Department of Building and Housing, Wellington, 7). Data on the past performance of systems in real fires is one of the best sources of information to estimate future performance, but there has not been a detailed study on sprinkler effectiveness data from fire incidents in New Zealand published since Marryatt’s work (Fire: a century of automatic sprinkler protection-revised. Australian Fire Protection Association, Melbourne, 13), which was last updated in 1986 and included data from Australia. The current research looks at the quality and quantity of data available on sprinkler effectiveness from New Zealand Fire Service (NZFS) incident reports over the period of 2001 to 2010 to evaluate the data’s usefulness for risk-informed building fire safety design. A comparison is made between the number of sprinklers reported activated in the NZFS dataset, Marryatt’s study, guidance from PD 7974-7:2003 (PD 7974-7:2003: the application of fire safety engineering principles to fire safety design of buildings. Probabilistic Risk Assessment, London, 3), and NFPA data (U.S. experience with sprinklers and other automatic fire extinguishing equipment. National Fire Protection Association, Quincy, 12). Proposals to improve the collection and reporting process to increase the informative value of future NZFS data for risk-informed fire safety design are presented.     

Estimating effectiveness of state-of-the-art detectors and automatic sprinklers on life safety in residential occupancies

The report provides a qualitative assessment of the life safety impact of early warning fire detection and automatic sprinkler technology in residential occupancies. This assessment is based on the results of full-scale studies and statistics on residential fire fatalities from the NFIRS data base. Estimates of the impact of three alternatives, smoke detectors, standard automatic sprinklers, and residential sprinklers, are provided for major fire hazard scenarios in residential occupancies. The results of this study indicate that significant life safety benefits can be derived from broad application of detectors and sprinklers in all residential buildings. National Bureau of Standards Note: This is a condensed version of the more detailed document NBSIR 84-2819 bearing the same title as above. The document is available in limited quantities from the National Bureau of Standards or it may be purchased through National Technical Information Services (NTIS). Reference: Edward K. Budnick, “Estimating Effectiveness of State-of-the-Art Detectors and Automatic Sprinklers on Life Safety in Residential Occupancies.” Fire Technology, Vol. 20, No. 3, August 1984, p. 5.     

Performance of selected coatings applied to polystyrene block walls under simulated mine fire conditions

A test program was undertaken by the Mine Safety and Health Administration (MSHA) to determine the feasibility of coating polystyrene block walls for fire protection. Selected coatings, used in the mining industry, were tested under semilarge scale, simulated mine fire conditions to determine the appropriate thicknesses of these coatings for protection of the polystyrene foam block against fire for specified time periods. Building plasters containing gypsum and perlite and an expanded vermiculite, portland cement, and limestone coating were particularly effective in protecting the foam blocks against the heat of the simulated mine fire. Reference: Steven J. Luzik, Performance of Selected Coatings Applied to Polystyrene Block Walls Under Simulated Mine Fire Conditions,Fire Technology, Vol. 22, No. 4, November 1986, p. 311.     

The Response Time of Different Sprinkler Glass Bulbs in a Residential Room Fire Scenario

The response time of fire sprinklers is essential for their performance, especially in applications where life safety protection is desired. The earlier the sprinkler activates, the smaller the size of the fire. Most commercial residential sprinklers are fitted with 3 mm, 68°C glass bulbs. However, thinner sprinkler glass bulbs with lower operating temperatures are available. The aim of this study was to determine the response time—and the corresponding heat release rate—of different glass bulbs in a residential room fire scenario. A series of tests were conducted inside a compartment measuring 3.66 m by 3.66 m having a ceiling height of 2.5 m. The compartment was either enclosed or had two walls removed to provide a more ventilated scenario. A propane gas burner was positioned at one of the corners. The mass flow rate of the gas was controlled such that either ‘slow’, ‘medium’ or ‘fast’ fire growth rate scenarios were simulated. In each test, nine Response Time Index (RTI) and operating temperature combinations were tested. Each test was replicated three times. In addition, two commercial fire detectors were tested. The results show that the fire is considerably smaller upon activation with a combination of a low RTI and a low operating temperature, as compared to the 3 mm, 68°C glass bulb typically used for residential sprinklers. The operating temperature proved to have a larger impact on the results than the RTI. The heat from the fire was typically detected by the fire detectors prior to the activation of the sprinkler glass bulbs, especially for the ‘slow’ and ‘medium’ fire growth rate scenarios.     

A final report on fire tests involving stored plastics

The Factory Mutual Research Corporation has completed a series of fire tests of stored plastics. This paper highlights the major points of the final report. Note: The author presented this paper at the National Fire Protection Association annual meeting held in Chicago, May 1975.     

Evaluation of unsprinklered fire hazards

This paper is based on a report written as a background guide for Factory Mutual engineers. Several basic concepts are introduced as an aid to understanding a wide range of fire behavior when sprinklers are not present. In addition, several formulas are provided so that the potential for damage age to metal structures in the absence of sprinklers can be estimated easily. A substantial effort has been made to avoid detailed derivations and keep these formulas or equations simple enough for rapid computation with either SI metric or Imperial engineering units. At the same time, adequate safety factors have been incorporated into these equations to insure that a conservative estimate can be made as to the need for sprinklers in a variety of situations.     

Ablative fluids in the fire environment

Experiments carried out on high-challenge fires demonstrate that ablative water is practical and highly effective for fire extinguishment as compared to plain water. Ablative water, films are much thicker and provide insulation. Chemicals and Plastics, Union Carbide Corp. Factory Mutual Research Corporation Note: The authors presented this paper at a meeting of the Industrial Fire Protection Section held in conjunction with the 73rd Annual Meeting of the National Fire Protection Association in New York City on May 13, 1969.     

The Florida palm coast fire: An analysis of fire incidence and residence characteristics

A quantitative analysis of the relationship between residential fire incidence, fire intensity, house characteristics, and location is presented. Fire intensity (ground vs. crown fire) was shown to be the most significant factor. Brush clearance and type of soffit vent were also shown to be related to fire incidence. Logit analysis was used to analyze the joint effect of the various factors. Five factors allowed the model to correctly predict whether a house burned in 83 percent of the observations. Reference: Robert Abt, David Kelly, and Mike Kuypers, The Florida Palm Coast Fire: An Analysis of Fire Incidence and Residence Characteristics,Fire Technology, Vol. 23, No. 3, August 1987, pp. 186–197.Florida Agricultural Experiment Station Journal Series No. 8241     

七氟丙烷气体泡沫灭火技术试验研究

试验研究表明,空气泡沫难以彻底扑灭凝析油、环氧丙烷、正戊烷等低沸点可燃液体储罐火灾,且不能有效控制环氧丙烷等液体储罐火灾。开发了用七氟丙烷灭火剂替代空气发泡的七氟丙烷气体泡沫灭火技术,并对环氧丙烷、正戊烷等低沸点可燃液体进行了试验研究。试验表明,七氟丙烷气体泡沫可快速、有效扑灭该类液体火灾。对上述试验研究进行了归纳总结。     

Limited-water-supply sprinklers for manufactured (mobile) homes

A prototype limited-water-supply (LWS) sprinkler has been developed for manufactured (mobile) homes in a research program sponsored by the United States Fire Administration. The LWS sprinkler is designed to be installed at a 2.44 m (8-foot) spacing and to have a total water supply of 380 (100 gallons). The installation spacing was determined in a series of freeburn fire tests that indicated that the heat release rate at sprinkler actuation could be halved by reducing the sprinkler spacing from 3.66 m to 2.44 m (12 feet to 8 feet). A series of eight fullscale fire tests, including a corner living room scenario similar to that used in the Los Angeles Residential Test Program, was conducted to evaluate the performance of the prototype sprinkler. In five of the tests, room tenability was maintained during the 10-minute period following the actuation of a single sprinkler at a flow rate of 38 pm (10 gpm). In three tests, tenability was maintained with multiple sprinkler actuation (2 or 3 sprinklers) and a total system flow rate of 49 pm (13 gpm). The spray of the sprinkler was characterized in terms of its water flux distribution and drop size distribution. The thermal sensitivity requirements of the sprinkler are to be based upon RTI, C, and temperature rating, which would ensure that sprinkler actuation would occur at fire sizes comparable to those encountered using the prototype LWS sprinkler in this study.     

SMART Sprinkler Protection for Highly Challenging Fires—Part 2: Full-Scale Fire Tests in Rack Storage

An experimental study was conducted to demonstrate the concept of a new sprinkler protection system using Simultaneous Monitoring, Assessment and Response Technology (SMART). Part I of this study focuses on the system design and function evaluation at the component level. The present work is Part II of the study, focusing on full-scale suppression tests to evaluate the performance of SMART sprinkler technology in protecting rack storage fires. The selected fuel was cartoned unexpanded plastic (CUP) commodity representing an intermediate level of fire hazard. The storage height increased from 3 tiers (9.1 m) to 5 tiers (10.7 m) to 7 tiers (12.2 m) in the three tests. The sprinkler activation was initiated by a smoke detector and a ceiling temperature rise threshold. The fire location was calculated as the thermal centroid based on ceiling temperatures. A group of six sprinklers, closest to the calculated fire location, was activated simultaneously. Subsequent fire development was monitored through visual observation as well as ceiling temperature data. Test results show that the SMART sprinklers can provide adequate protection for the CUP commodities stored up to 7-tiers (12.2-m) high within a rack storage under the tested conditions. The water densities used in these tests were approximately 50% of those in existing protection recommendation. These results lay the foundation for exploring potential applications of the SMART sprinklers to Highly Challenging Fires.     

Directions for improving manual fire suppression using a physically based computer simulation

The Swedish Fire Research Board and the U.S. Federal Emergency Management Agency are sponsoring a project to further the understanding of the basic mechanisms involved, as well as to support the development of standards for and to seek ways of improving the performance of portable fire suppression systems used by fire departments.This paper describes a physically based computer model developed to simulate one aspect of the problem: the manual suppression of postflashover fires. This includes: (1) an overview of the physical basis behind the model; (2) a comparison of model predictions with available experimental data, and (3) an analysis of fire suppression effectiveness using the model.The analysis concludes that, when direct access and extinguishment of the burning fuel is not possible, improved fire control occurs with water sprays having a Rosin-Rammler distribution of droplet sizes with volume-median-drop diameters in the 0.15 to 0.35 mm range. This agrees with available experimental data. It is also shown that fire fighting venting and standoff distance requirements may lead to more severe fires requiring more water for control; although venting and water spray induced air/gas flow also serve to channel hot steam and gases away from the fire fighter adding to his safety. The analysis also shows that allowing higher gas and surface temperatures at fire control through improved fire fighter protective clothing and equipment design reduces water flow rate requirements. Additional experimental work is recommended before all these conclusions are considered definitive. Reference: L. M. Pietrzak and G. A. Johanson, Directions for Improving Manual Fire Suppression Using a Physically Based Computer Simulation,Fire Technology, Vol. 22, No. 3, August, 1986, p. 184.