Document 3e6XnbKyVvQ0vMXjy7GQV5E3J

PLG-0356 COMPARATIVE TRANSFORMER FIRE RISK STUDY Prspsrsdtof MONSANTO COMPANY St. Louis. Missouri August 1904 Pickard, Lowe andGarrick,Inc. Engineers Applied Scientists Management Consultants Newport Beach, CA Washington, DC MON5 020601 PLG-0356 COMPARATIVE TRANSFORMER FIRE RISK STUDY PfOfrCl DtfCtOf B. John Garncfe Principal inmttgatar Mardyro* KAzarian* Othar invMflgatofa Taong-Lun Chu Vicki M. Star Praparad for MONSANTO COMPANY St. Louis, Mtaaourl August 19B4 Pickard, Lowe and Garrick,Inc. Engineers Applied Scientists Management Consultants Newport Beach. CA Washington, DC HONS ACKNOWLEDGMENT The authors wish to express their appreciation to Mr. Raymond F. Boykin of Monsanto Company for his Interest and support In this project. Thanks ire due to Or. Raymond A. Freeman and Mr. Jerry H. Schroy of Monsanto Company for their advice and for contributing Appendices H and I. Thanks are also due to Mr. Willard C. Gekler of Pickard. Lowe and Garrick, Inc.; and Dr. George Apostolakls and Mr. Nathan 0. Slu of the University of California, Los Angeles, for their advice and painstaking review of the report. Special thanks are due to Ms. Ann M. Kallnowskl for collecting and analysing some of the data. Mr. Curtis L. Moore for helping us understand the Intricacies of different transformers, and Mr. John E. Silver for helping us establish the transformer Installation characteristics. Thanks are extended to the reviewers of this report, Or. Howard k. kunreutMr. Or. John L. Bocclo. Dr. Curtis Travis, Or. V. R. R. Uppulurl, and Or. Myron Miller, whose coamts were found to be particularly helpful. 0027M072S84 HI HONS 020603 EXECUTIVE SUWABV This study evaluate* the impact on fire risk of the decision to replace mineral oil with Askarel as the cooling and Insulating fluid for electrical equipment. It compares the frequencies of acute Injuries and fatalities arising from operating a mineral oil transformer to those from operating an Askarel transformer. The term Askarel refers to a broad class of fire resistant synthetic chlorinated hydrocarbons used as dielectric fluids. For the purposes of this study, Askarel is defined as a mixture containing 501 Aroclor 1254, a polychlorinated biphenyl (PCB), and SOI 1.2,4 trlchlorobenzene. In the evaluation of the fire risk for the two types of transformer fluid, two separate cases are considered. Case 1 describes a 1-hour fire rated transformer room In a general-use area of a public-use building. Case 2 describes a 3-hour fire rated transformer room In the nonpubllc-use basement of a building. These two cases describe typical Installations of Askarel and mineral oil transformers. The results show a significant decrease In fire risk when a mineral oil transformer Is replaced by an Askarel transformer. In case 1, the mean frequency of transformer fires that spread beyond the room of origin Is a factor of 300 smeller for Askarel compared with mineral oil transformers. For case 2, the reduction is even larger--a factor of 1.000. This demonstrates that the risk from mineral oil transformers is much greater than that from Askarel transformers, since a fire that spreads bayond the transformer room is of more concern than one contained within the room. Also, historical data show that there Is a graatar chance of multiple fatalities when a fire spreads beyond Its room of origin. An Important factor contributing to Askarel transformer fires is the presence of trichlorobenzene. If an Askarel transformer did not contain trl chlorobenzene. Its fire frequency would be significantly smaller. This is because trl chlorobenzene has a flash point of 99'C, whereas PCB (Aroclor 12541 does not have a flash point up to Its boiling point (greater then 350*0. In other words, the use of trl chlorobenzene as a blending fluid reduces the fire resistance of the resulting Askarel mixture. This study ostlmetes that about 1 In i.QOG of all fires confined to the transformer room would Involve at least one fetalIty. For fires spreading beyond the transformer room, about 1 in 100 was estlmiteo to result In fatalities. Based on these values and the frequencies of wlthln-roem and beyond room fire* estimated in the study, the fatality risks from Askarel end mineral oil transformers are compered. For case 1, the mean frequency of fires Involving one or more fatalities Is found to be a factor of 6 lower for Askarel transformers than for mineral oil. The reduction In ces* 2 Is comoarable--a factor of 4. in both casas, room fires were the dominant contributors to the fatality risk from Askarel transformers. Fires that spread beyond the transformer room were the dominant contributors to the risk from mineral oil transformers. In all cases, transformer rupture was the dominant cause of the Injury or fatality flras. 002711080B84 i v HONS 020604 In summary, this study shows that there Is a significant decrease in fire risk when Irtneral oil transforntrs are replaced by Askarel tram farmers. Conversely, replacing Askarel with mineral oil would involve a significant Increase In transforsier fire risk. The risk of conversion from Askarel to mineral oil Is most significant for nonvault-transforaier Installations. CO27 MO60484 v HONS 020605 TABLE OF CONTENTS Stctlon 1 INTRODUCTION 2 TECHNICAL SUWARY 2.1 Case 1 - A Transformer without a Vault 2.2 Cast 2 - A Transformer with a Vault 3 01SCUSSI0N CONCLUSIONS 5 OETAILS Of THE ANALYSIS 5.1 Objective. Definitions, and Scope 5.1.1 Objective 5.1.2 Risk Attributes 5.1.3 Equipment Type 5.1.4 Era of the Study 5.1.5 Site 5.2 Approach 5.3 Case 1 - Transformer in a Regular Room 5.3.1 General Assumptions 5.3.2a Case 1: Base Case - Askarel Transformer S.3.2b Cate 1: Alternate Case - Mineral Oil Transformer 5.3.3 Case 1 Risk Cmparlson 5.3.4 Sensitivity of the Results 5.4 Case 2 - Transformer Installation with Mineral Oil Characterization 5.4.1 General Assumptions 5.4.2a Case 2: Base Case 5.4.2b Case 2: Alternate Case 5.4.3 Case 2 Risk Comparison 5.4.4 Senslttvlty of the Results 5.5 Discussion 6 REFERENCES APPENDIX A: FREQUENCIES OF TRANSFORMER PROBLEMS APPENDIX B: FIRE PROPAGATION FROM AN ASKAREL TRANSFORMERRUPTURE APPENDIX C: CONDITIONAL FREQUENCIES OF INJURIES ANO FATALITIES APPENDIX 0: ROOM PRESSURE RISE FROM MINERAL OIL TRANSFORMER RUPTURE APPENOIX E: ROOM BOUNOARY FAILURE FROM SEVERE HEAT FLUX Paqf i 3 3 6 3 10 11 11 n n 11 n 12 12 16 16 20 20 34 40 42 42 44 44 53 58 58 60 A-l 8-1 C-l D-1 -1 vl 0026HU8Q384 MOMS 020606 TABLE OF CONTENTS (contlnued) APPENOIX F: FIRE PROPAGATION TO AN AOJACENT ROOM FROM AN ABOVE-LIQUID-LEVEL RUPTURE OF THE CASE l MINERAL OIL TRANSFORMER APPENOIX G: UNCERTAINTY PROPAGATION USING OISCRETE PROBABILITY DISTRIBUTIONS APPENOIX H: FUUftABILITY OF CHLORINATEO ORGANIC VAPORS APPENOIX I: FLASH POINT OF TAICHLOROBENZENE-AROCLOR MIXTURES Pjqg F-l S-L K-l 1-1 0Q26M0608B4 vll MONS 020607 UST C *A8LE5 WO FIGURES Tables 5-1 5-2 5-3 5-4 S-S 5-6 5-7 5-8 5-9 5-10 5-11 5-12 5-13 5-14 Annuil Occurrence Rite of Room and Beyond Room Fires From t Mineral Oil Transformer; Alternite Cue of Cise 1 Annuil Occurrence Rite of Room ind Beyond Room Fires fro* in Askarel Transformer; Bise Cise of Cise 1 Annuil Occurrence Rite of Injury end Fitillty Fires from in Askirel Transformer; Bise Cise of Cise 1 Annuil Occurrence Rite of Injury ind Fitillty Fires from i Mlneril Oil Transformer; Alternite Cise of Cise 1 Annuil Occurrence Rites of Injury Fires fro* Two Types of Transformers; Cise 1 Annuel Occurrence R''*i of Fitillty Fires from Two Types of Trmsformers; Ca ' Annul. Occurrence R of Fires Propagating Be> 1 the Room of Origin from Types of Trmsformers; Case 1 Annual Occurrence R . of Room and Beyond Room Fires fro* a Mineral Oil Transformer; Base Case of Case 2 Annual Occurrence Rate of Room and Beyond Room Fires fro* an Askirel Transformer; Alternate Case of Case 2 Annual Occurrence Rate of Injury and Fatality Fires from a Mineral Oil Transformer; Base Case of Case 1 Annual Occurrence Rate of Injury and Fatality Fires fro* an Askirel Transformer; Alternate Case of Case 2 Annuil Occurrence Rites of Injury Fires from Two Types of Transformers; Case 2 Annual Occurrence Rates of Fatality Fires fro* Two Types of Transformers; Case 2 Annual Occurrence Rates of Fires Propagating Beyond the Room of Origin fro* Two Types of Trmsformers; Case 2 F1qures 3! 31 33 33 35 36 3 50 SO 51 51 54 55 57 5-1 5-2 S-3 5-4 S-5 5-6 5-7 5-8 5-9 5-10 5-11 A Family of Risk Curves and the Cut Curve Pictorial Representation of the Risk Quantification Structure An Example Event Tree General Layout for Case 1 Transformer Room Event Tree for the Base Case of Case 1. the Askarel Transformer ... Event Tree for the Alternite Case, the Mlneril Oil Transformer of Case1 Summary of Case 1 Results Plan View of the Transformer Vault for Casa 2 Event Tret for the Base Case, the Mineral Oil T'"-sformer of Case2 E - : Tree for the Alternate Casa of Case 2, an A v -el Transformer in the vault Summary of Case 2 Results 13 15 l7 13 21 21 37 43 45 45 56 0026M060S84 VIII HONS 020608 1. INTRODUCTION in the 1930s, polychlorinated biphenyls (PCB) were Introduced by the electrical Industry as nonflaanable electrical Insulating and cooling fluids called Askarels. The primary goal was to replace the mineral oil in liquid-filled transformers In installations where a fire could lead to public Injury or excessive property damage. Since Askarel transformers were typically more expensive than mineral oil transformers, they were used mainly for special applications, such as Inside office buildings. In underground mines, etc. The trade name Askartl denoted a mixture of chlorobiphenyls, chlorobenzenes, and stabilizers, {Askarel properties and mixtures are discussed In documents such as References 1 and 2.) The combined flame resistant and excellent dielectric properties of Askarels led to their widespread use In other electrical equipment as well. They were the primary fluid In all capacitors and were used in special-use motors, magnets, switches, clreult breakers, underground cables, and voltaga regulators. Thpy were also used In nonelectrical equipment where the risk of fire was significant; e.g., as a flame resistant hydraulic fluid and heat exchanger fluid. In spite of the diverse and widespread applications of Askarel, its manufacture was banned In Nay 1979 because of PCB's environmental persistence. As they fall or reach the end of their useful lift, equipment Items using Askarel are decreasing In number. Also, many Askaral transformers have been systematically emptied of Askarel and rtf11 led with mineral oil. New types of dielectric fluids that may replace Askarels and mineral oil are being tested for their flwnablllty (References 3 and 4), dielectric, cooling, and health and environmental characteristics. Equipment design Is also being changed to accommodate other media, such as gates, for cooling and Insulating. In the 1930s, the decision to replace mineral oil with Askarels was based on a qualitative understanding of risk and risk reduction. In this study, we investigate the actual change In fire risk. That Is, we evaluate the difference In fire risk between Askarel-filled and mineral oxl-fillad transformers. The selected damage Indices are acute fatalities and acute Injuries caused directly by transformer fires. Risk Is expressed In terms Of the likelihood (frequency 1 per transformer year of a fir# causing at Itast one fatality or Injury. For example, the frequency per year of fires causing at least one fatality from an operating mineral oil-filled transformer for 1 year Is compared with the corresponding frequency for operating an Askerel-fWed transformer. Because of the sparsity of the available information, it became dear that the likelihood of Injury or fatality cannot be established by statistical analysis alone. Therefore, two site specific case studies were performed. One site has the cheracterlstlcs of a typical Askarei-filied transformer installation. An Askarel-f11 led transformer and a mineral oil transformer at that installation are compared. The second ease study Involves an Installation typical of mineral oil-filled transformers. As In case 1, an Askarel-filled transformer and a mineral oil transformer at the same installation are compared. --27-080384 1 HONS 020609 In this study, the subject of uncertainty and Its propagation through to tn# final rosults is given special attention. The Bayesian (subjectivist) school of thought Is adopted (References 5 and 6). Uncertainties are always expressed by probability distributions (either discrete probability distributions or density functions). To Minimize the confusion arising fron seMntlcs, we have adopted the definitions suggested in Reference $. The word "probability" is used at a maasure of our state of confidence, for rates of occurrence and fractions, the word "frequency" Is used. ::27`1080334 2 HONS 020610 2- TECHNICAL SUWttRY The approach to performing this comparative risk study Involves modeling events and the likelihood that they Mill lead to a given level of risk. Two initiating events resulting In a fire are chosen: (1) transformer fluid leakage, and (2) transformer rupture. (Note that not all leaks and ruptures lead to fires.) The barriers that limit the effects of these Initiating events are the boundaries of the room where the transformer is installed, the fire protection system In the room, etc. These barriers are modeled by an event tree. An event tree Is simply a graphic representation of potential sequences of events. For example, one possible sequence is a leak that leads to a fire during which the room boundary stays Intact. In that sequence, fire propagation to an adjacent room Is impossible because the room boundary is not breached. To link event sequences with consequences, the sequences are categorized by what we call "damage states.* Those sequences not leading to fire propagation into an adjacent room are categorlied as "room fires' (l.e., fires confined to the room of origin). Those sequences leading to fire propagation into adjacent rooms are categorized as "beyond room fires," l.e., fires propagated beyond the room of origin. This categorization is convenient because the statistical data on injuries and fatalities can be identically categorized. The conditional frequency of observing one or more injuries or fatalities given a fire in each category can therefore be easily established. 2.1 CASE 1 - A TRANSFORMER WITHOUT A VAULT The base case for case 1 is an Askarel-filled transformer installed in a medium-sized, public-use building. The installation characteristics for this case are typical of Askarel units, which means that the transformer Is not in a vault and Is located on a general-use floor. The alternate case 1$ a mineral oil-filled transformer with Identical specifications Installed In place of the Askarel transformer without any other changes to the installation or the room. The transformer Is assumed to be in a dedicated room on a general-use floor. The combustible loading and the use of the floor are similar to an office building. For example, the corridor outside the transformer room Is assumed to be carpeted and may contain combustibles such as paper and plastics. The walls of the transformer room are made of plaster, have no windows, and are rated as 1-hour fire walls. There is only one door (of the same fire rating) that opens Into the adjacent corridor. The walls art covered by mineral tiles on the Inside to reduce the transmission of transformar noise into the corridor. Ventilation is provided by fans and openings that lead dlractly to the outside air. 3 C027M080J84 HONS 020611 5pecial precautions are taken to prevent the storage of combustibles near the transformer. To control any transformer fluid leaks, a curb ts provided around the transformer with at least a 4-1nch lip and 3 feet of clearance on all sides. A wet-type sprinkler system is Installed above the transformer. The temperature setting of the heads Is of ordinary classification, which ranges from 135*F to 17Q*F (S7*C to 77*c). Under normal conditions, no one enters this room except for occasional visits by maintenance personnel. Ooors are normally locked and only maintenance and security personnel have the keys. For the base case, the liquid of the transformer Is assumed to be Askarel. A typical mixture Is SOI (by weight) Aroclor 1254 (a PCS) and 50% 1,2,4 trlchlorobenzene. Askarel leaks would generally be noticed by maintenance personnel. Leaked PCS and trichlorobenzene mixture would not Ignite because external heat fluxes of several watts per cm? are needed to sustain combustion In PC8, and the flash point and auto Ignition point of 1,2.4 trichlorobenzene are about 99*C and 57l*c, respectively. The flash point of the Askarel mixture Is UO*C. Ignition sources, heat sources, and hot surfaces are not present near leaking transformers or inside the curbs containing the letting fluid. Thus, severe accident sequences Initiated from leaks are taken to have a very low frequency and, therefore, make only negligible contributions to risk. violent rupture of the transformer casing may occur due to rapid faults (an arc) In the transformer windings that are lenersed In the Askarel. The cause Is typically deterioration of the insulation between windings coupled with a rapid increase In the line voltage. In all four Askarel transformer ruptures known to the authors, dense smoke has been reported. Therefore, It is assumed that at every rupture Incident, some combustion of the fl liable Askarel component, trl chlorobenzene, and nearby transient combustibles, will take place. For rupture of an Askarel transformer In an unvaulted room, the frequencies of Injuries and fatalities are dominated by room fires. For Injuries, 99% of the mean frequency is from room fires. For fatalities, the corresponding figure Is 98%. The dominant contributor to room fires in this cast Is fires that do not fall the roam boundaries (rather than fires that do fall the room boundaries but fall to propagate beyond tne room). The reeson for this Is that room boundary failure from Askarel transformer fires Is highly unlikely. boom boundary failure depends on the Impact of pressure waves or missiles on the walls and door, or heat Impingement on the walls fron burning Askarel. Oetonatlon (an explosion where the flame front propagetes supersonically and can create a strong pressure wave) is not judged to be possible, because PC8 vapor does not burn easily and 1,2,4 trl chlorobenzene has a narrow fl amiability range (2.5% to 6.6%) OeMigration (an explosion where the flame front propagates subsomcai'y and the resulting pressure wave Is wetter than that of a detonation) -av take place In the 1,2,4 trichlorobenzene vapor If the vaoor temperature Is well above 99'C and an Iglnlilon source Is present, but the pressure increase during deflagration Is generally snail. Thus, only under 0027-1080334 4 MONS 020612 extrema conditions (such as very h1gh arc energy, the source of transformer rupture) Is the combustion of the trlchlorobenzene expected to pose a threat to the integrity of the room boundaries. If deflagration does not occur, the sudden release of pressurized gas from the transformer may lead to a pressure pulse. The room volume is assumed to be 20 x 20 x 8 * 3,200 ft3, and the air space above the transformer fluid can only range from 0.5 to 10 ft3 depending on the size of the transformer. Thus, an increase of 1 pslg at the walls requires a 320 to 6,250-pslg Increase In the tank at the time of rupture. Such a pressure rise Is very unlikely. Also, the transformer tank would rupture well below that pressure, thereby mitigating the pressure rise. Ceiling failure from the rupture (s deemed to be very unlikely because the celling Is assumed to be made of reinforced concrete and to be 6 Inches thick. The celling can therefore withstand pressures well over 2 pslg. As discussed above, the likelihood of such pressures Is very small. Missiles may also be generated In some transformer ruptures. The most likely missiles ere the tank cover thrown In a vertical direction and slugs of the transformer fluid. The celling may be damaged by the transformer cover, but the impact Is not expected to breach the celling. Finally, the prolonged Impingement of combustion heat on the walls and the celling may also lead to room boundary failure. However, even If the sprinklers do not work properly, the walls and the celling (which are 1-hour and 3-hour rated, respectively) would contain the fire long enough to allow for manual fire extinguishing efforts. Fire propagation to an adjacent room given room boundary failure is not a likely event either. Prolonged Impingement of the heat from transformer fire mey Ignite the combustibles In the adjacent rooms If the fire Is not suppressed In about a half-hour. Overall, the smell likelihood of room boundary failure ana fire propagation due to Askarel transformer ruptures In unveulted rooms minimizes the likelihood of Askarel fires propagating beyond the transformer room, and therefore also minimizes the likelihood of Injuries and fatal 1 tits. The alternate case assumes a mineral oil-filled transformer In place of the Askarel transformer. The main differences are that (1) leakage of mineral oil could lead to severe consequences, and (2) splattered mineral oil could catch fire. For fires Involving mineral oil transformers without vaults, the main contributors to the frequencies of injuries and fatalities (89t of the mean frequency of Injuries and 911 of the mean frequency of fatalities) ere beyond room fires. The dominant contributors to this damage state are fires originating from transformer ruptures, even though Ignition of leaks Is also possible. Fire propagation to an adjacent room given transformer rupture is Judged to b* certain if splattered mineral oil escapes the transformer room and comes to rest on combustibles such as carpeting. The splattered fluid is likely to be on fire because the temperature difference between the bulk of the oil (about 100'C) and the oil fire point (150*c) Is rather small, and because the explosion may set some of the splattered fluid on fire. The explosion Itself Is not deemed to be capable of Igniting 002711080384 5 MON S 020613 combustibles in adjacent rooms, and the sprinkler system above the transformer is not expected to be of any benefit because the events take place at a rapid pace. It should be noted that the assumption of a carpeted hall does not necessarily Introduce a bias In favor of Askarel transformers, because burning mineral oil would Ignite the many of the materials typically found In a general-use area of a building, whereas the temperature of splattered Askarel would be rather low and very few materials would ignite at that temperature. if mineral oil does not splatter after the transformer ruptures, the primary fire propagation mode Is anticipated to be radiation from flames inside the transformer room. Since this process Is not rapid, fire suppression equipment such as the sprinkler system could extinguish the fire prior to propagation. However, the likelihood of fire propagation in this case is still Judged to be quite high. The change In risk when an Askarel transformer In an unvaulted -:oro is replaced by a mineral oil transformer Is significant. The mean frequency of fire related Injuries Increases from once every 300,000 transformer years with Askarel to once every 70,000 transformer years with mineral oil (over a factor of 4 difference), similarly, for fire related fatalities, the mean frequency Increases from once every 4,000,000 transformer years for Askarel to once every 700,000 transformer years for mineral oil (almost a factor of 6 difference). The use of a mineral oil transformer In an unvaulted room is allowed by the national Electrical Code If a sprinkler system Is Installed in the area. Thus, the results of this analysis carry an Important messageunder equal conditions, fire risk decreases significantly when a mineral oil transformer Is replaced by an Askaral transformer. 2.2 CASE 2 - A TRANSFORMER WITH A VAULT The base case of caw 2 Is a mlnarel oil transformer In a vault in the basement of a large building. The alternate case is an Askarel transformer with Identical specifications. The main difference between case 1 and case 2 Is that in case 2 the walls and calling are built of fi-lnch concrete or 8-1neh brick. The following assumptions describe :ne vault and tha transformer In detail and are applicable to both the mineral oil and the Askarel transformers. A single transformer Is assumed to be located In a vault approximate'/ 20 feet by 20 feet by 8 feet. The floor of the vault is sloped par foot to a sump 3 feet square and 2 faet 6 Inches deep. The vaui: -> assumed to have a dedicated ventilation system. Tha air inlet is quipped with a fire damper, and the air outlet is equipped with lcuvets to prevent the entrance of rain. The prnietratlons of the room mc'..? air ducts, conduits for electrical cables, and a self-closing, 3-no-'- rated fire door (S feet by 8 feet), which Is normally locked close? vault Is assumed to be equipped with an automatic sprinklar system -> temperature setting of the heads Is taken to be of ordinary classification, which ranges satween 57*C and 77C (13SF and l73 r Q027M06QSB4 6 HONS 020614 For this case, the frequencies of injuries and fatalities are dominated by beyond room fires. However, the contribution of room fires is noticeable. For injuries, room fires contribute 31* of the mean frequency; for fatalities, they contribute 25*. The contribution of transformer rupture is large, contributing 92* of the mean Injury frequency and 93* of the mean fatality frequency. The main reasons for this are (1) the greater frequency of ruptures involving fires compared with teaks involving fires, and (2) the much greater likelihood of vault boundary failure given a rupture as contrasted with given a leak-induced fire. If the mineral oil In the transformer of the base case Is assumed to be replaced by Askarel while keeping other conditions unchanged, any fluid leaking out of the tank can no longer be ignited. Therefore, Askarel leaks do not require further evaluation. The mechanisms for vault boundary failure from rupture of an Askarel transformer are the same as those for room boundary failure In case 1. In this cast, however, the likelihood of failure Is much smaller because of the strength of the vault boundary. The pressure pulse from a deflagration has to be very strong (above 7 pslg at the vault boundary) to fall a wall. Similarly, the release of pressurized gas from the tank would have to Involve a very large initial pressure. This has been shown to be almost Impossible. Finally, exposure of the walls to heat flux from combustion of transformer fluid has to last much longer to fall the walls In this case than in case l because the walls are much thicker. From this comparison, we assess the tan frequency of vault boundary failure after rupture of an Askarel transformer to be only about 1 x 10-2. Because of the relatively small chance of vault boundary failure, the frequencies of Injuries and fatalities for vaulted Askarel transformers are totally dominated by room fires that do not fall the vault boundaries. Overall, we can confidently claim that the risk from fires would decrease If a mineral oil transformer In a vault were replaced by an Askarel transformer. The mean frequency of fire related Injuries would decrease from once every 100,000 years for mineral oil transformers to once every 300,000 years for Askarel--a factor of 3 difference. Similarly, the mean frequency of fire related fatalities would decrease from once every 1.000,000 years (mineral oil) to once every 4,000,000 years (Askarel)--a factor of a difference. 0027M060S34 7 020615 HONS 3. D1SCUSSI0H Many assumptions and Judgmentally evaluated parameters are used in :m$ study. In this section, the sensitivity of the final results to these assumptions and parameters Is investigated. The frequency of transformer fires was obtained from statistical data for all transformers. The uncertainties about this frequency are rather small because the number of transformers Is very large. The conditional likelihood of a rupture given a fire was evaluated Judgmentally na was deemed to be greater than the conditional likelihood of a leak given a fire. Two Important assumptions were made at this stage. It was assumed that (1) fires from Askarel transformers are almost solely from ruptures ang (2) their total fir# frequency is equal to the fire frequency from rupture of mineral oil transformers. A more realistic risk comparison could be made if the fire frequency for Askarel transformers was evaluated separately from relevant statistical evidence; however, tne evidence required for this evaluation Is not readily available. Room boundary failure frequencies for Askarel and mineral oil transformers were evaluated primarily using Judgment. For an Askarel transformer, an explosion has to occur for room boundary failure to result. An explosion Is possible because the trichlorobenzene In the Askarel can be vaporized (boiling point 214 C) and Ignited. PCB would not contribute to the explosion because Its boiling point is hijner (greater than 350 C), and It does not have a flash point before boiling. The likelihood of boundary failure for Askarel transformer; represents the likelihood of an explosion severe enough to damage the walls. An Important element of both the fire frequency and the boundary failu--e frequency evaluation Is the presence of trichlorobenzene. If an Askarel that does not contain trlchlorobenzene (e.g., pure Aroclor 1254) had been chosen for this case study, room boundary failure would have been very unlikely. In addition, the frequency of transformer fires specific to this fluid would have had to be quantified explicitly because of its highly flame resistant characteristics (high boiling point and flash point). This would have reduced the risk assessed for Askarel transformers. Room boundary failure in the case of mineral oil transformer rupture was Judged to be very likely. This is because mineral oil is a relatively flammable material, and it was judged that an explosion would occur jq:>transformer rupture. This may be a conservative conclusion though probably not an extreme one. A decrease in the likelihood of an explosion would decrease the estimated risk of mineral oil transformer; In the case of mineral oil transformer rupture, the Importance of roon fires to the final results Increases as the likelihood of vault bouioar failures decreases. If the likelihood of vault failure is further reduced, the mineral oil risk results would move closer to the result> for Askarel. 0027M0605B4 3 HONS 020646 fh* uncertainties In th fire propagation frequencies are not highly sensitive to the underlying assumptions. For example, in the case of oropagatlon to an adjacent room when room boundaries are breached, the important parameters are the heat flux and the Ignition temperatures of the materials In the adjacent room, tn view of the large uncertainties aoout the fire propagation frequencies, any reasonable changes in the values of the ignition temperatures and heat flux would not have a major impact on the final results. The conditional frequencies of Injuries and fatalities given a room or beyond room fire were evaluated using statistical evidence. The most important assumption at this stage of the analysis was the compatibility of the collected evidence with the alms of our case study. For beyond room fires caused by unvaulted transformers, the evidence was judged to be highly applicable because once a fire propagates beyond the room of origin It does not matter how It originated. For room fires caused by unvaulttd transformers, the applicability of the evidence for injuries was judged to be good because one of the two date sources on injuries was specific to Indoor transformers. For fatalities from room fires, the data may not be as good as those of the other damage Indices, such as Injuries from beyond room fires. This was reflected by conservative estimations and wide uncertainties. The data on conditional Injury and fatality frequencies are less applicable to vaulted transformers, because vaults are not typically found In buildings. Also, the vault was not assumed to be located In a general-use area, so the occupants of the building would not be highly exposed to the transformer fire consequences. This Is especially true for transformer fires that are confined to the vault. Olrect comparison of the results of the two case studies Is not warranted for two reasons. First, the same conditional frequencies of Injuries ana fatalities were used In the two cases, even though In case 1 the rooms adjacent to the transformer room art used by the general public, whereas In case 2 only the building maintenance and building management personnel are typically present in the vicinity of the transformer vault. Second, the models used to estimate the conditional frequencies of boundary failure and fire propagation Include many conservative assumptions. The two case studies may Involve differing degrees of conservatism, so that a direct comparison may be misleading. Extension of the final results to other situations is also not warranted. For axampla. the Injury and fatality frequencies used here are not applicable to transformers on poles near suburban homes or transformers Installed outside buildings. Of the Intermediate results, only the Initiating event frequencies and conditional Injury and fatality frequencies are applicable outside the two case studies. 302711060584 9 HONS 020617 . CONCLUSIONS Having compared th* fir* risk* from mineral oil and Askarel transformsr* In two case studies, let us now review the major conclusions of this study. The risk attributes chosen for the study were acute Injuries and fatalities. The two case studies both Involved sites in public-use buildings. Case 1 compares an Askarel and a mineral oil transformer it an Installation specifically designed for Askarel. For case 2, the site (a vault In a basement) Is specific to a mineral oil transformer. The major conclusions of the two case studies are as follows: e For both cases, there Is a significant decrease In fire risk when a mineral oil transformer Is repleced by an Askarel transformer in the same Instillation. Conversely, there Is a significant increase in fire risk If Askarel were to be replaced by mineral oil. e The meen frequency of trensformer fires that spread beyond the room of origin Is much less for Askarel transformers than for mineral 011 transformers. The difference Is a factor of 300 for the unvaulted transformers of case I, and a factor of 1,000 for the vaulted transformers of case 2. This demonstrates that the risk from mineral oil transformers is much greater then that from Askarel transformers, since e fire that spreads beyond the transformer room Is of more concern then one contained within the room. In particular, historical data show that there Is a greater probability of multiple fatalities when e fire spreads beyond its room of origin. e The risk of fires Involving at least one fatality Is significantly smaller for Askarel transformers than for mineral oil transformers. Th# difference Is a factor of 6 for case 1, and a factor of * for case 2. The risk of fires Involving at least one Injury is als' jntflcant'/ smeller for Askarel then for mineral oil. The frequency -f fire related injuries Is a factor of 4 smeller In case 1, and a factor of 3 smeller in case 2. e Finally, the presence of trlchiorobenzene is an Important factor contributing to the risk of Askarel transformer fires. The frequent of fires would be much smaller for Askarel transformers not containing trlchlorobenzene than for those that do. This is because trl chi orobenzene has a relatively low flash point (only 99 C), whereas PCBs do not have a flash point up to their boiling point. The SO/SO mixture of trlchiorobenzene and PCB leads to an Askarel with flash point of 110*C. 0027H060584 10 HONS 020618 S. DETAILS OF THE ANALYSIS 5.1 OBJECTIVE. DEFINITIONS, AMD SCOPE 5.1.1 OBJECTIVE The goal of this study Is to evaluate the impact on fire risk of the decision to replace mineral oil by Askarels as the cooling and Insularig fluid for electrical equipment. 5.1.2 RISK ATTRIBUTES In a risk analysis, the damage Indices (or risk attributes) can be numerous. In this case, fire can lead to property damage, fatalities. Injuries, and loss of production. Of these Indices, we evaluate only tne fatalities and Injuries caused directly by a transformer fire. Latent effects (that Is, fatalities or injuries that may take several years after the fire Incident to materialize) are not Included in this study. 5.1.3 EQUIPMENT TYPE Only mineral oil and Askarel-filled transformers will be considered because the goal of the study is to evaluate the fire risk implications of the decision to replace nlneral oil with Askarel. The term Askarel refers to a generic name for a broad class of fire-resistant synthetic chlorinated hydrocarbons (such as PCSs) used as dielectric fluids. Their properties and mixtures are discussed In documents such as References : and 2. Of the different transformer types, we concentrate on distribution transformers because other types are mainly used In power plants, switching stations, or other special Industrial uses where there are normally few personnel present. Other equipment such as capacitors, circuit breakers, switches, motors, magnets, and heat exchangers will be addressed. Except for the transformers and capacitors, all the otner Askarel-filled equipment types are used primarily for Industrial applications and art not typically installed close to publle areas. 5.1.4 ERA Of THE STUDY Ideally, to achieve the goal of this study one should analyze the transformer fire risk In the social and Industrial environments of the 1930s when the decision to uso PC8s was made. However, specific Information regarding tha charactarlsties of the transformers and the" Installations Is not easily available from that era. Also unavailable are statistical data regarding the number of transformers operating at that time as well as the number of transformer fires, and the injuries and fatalities resulting from those fires. However, good statistical evidence has been available since the mid-1970s. Several reports hav been published estimating the number of transformers and the frequenc* of fires, injuries, and fatalities. Therefore, the most convenient *for our analysis Is the past few years. It is Judged that the choice this era is appropriate because of the available data and the fact t"" transformer design and Installation characteristics have not change^ 0027*1080384 a HONS 020619 drastically in th past few decades. Transformer design improvements have been in efficiency (less heat generation) and In tolerance to overloads and shorts. There have also been Improvements in the strength of the tank that holds the insulating fluid. 5.1.5 SITE From the available statistical data, one cannot reliably distinguish between Askarel and mineral oil transformer fires. The Impact of this deficiency can be reduced If the risk comparison Is performed for some specific sites. Two sites are chosen for this study. The fire risk comparison Is done separately for each site by assuming that at the same site there Is an Askarel transformer and a mineral oil transformer *itn Identical design and Installation characteristics. The first site, which Is referred to as case l, has the Installation characteristics of an Ajkarel-fllled transformer. It is a transformer room In a public-use building, with regular room boundaries of a 1-hour fire rating; the area outside the room Is for general use by the building occupants. The second site, which Is referred to as case 2, has the Installation characteristics of a mineral oil-filled transformer. It is a transformer vault In the basement of a public use building. However, outside the vault, the floor is used for building service equipment and shops. S.2 APPROACH The approach to performing this comparative risk study Involves the modeling of events and their likelihood, thus leading to a quantitative statement of risk. The centrtl feature of the approach Is to display tne uncertainties associated with the figures of merit. The figures of men; are the frequencies of occurrence of different levels of damage. In this study, "damage* Is deaths and Injuries. Probability Is employed to express the confidence In those frequencies. This concept Is known as the "probability of frequency" formet, or the "probability and consequences" epproech to risk assessment (Reference 6). This approacn is extensively used In the very comprehensive risk assessments performed for the nuclear Industry. For each of the two fluids (Askarel and mineral olll, risk curves can oe developed for each of the risk attributes (acute fatalities and acute injuries). Risk curves are actually families of curves, eacn with its own probability. An example of such a family is shown In Figure 5-i. This "risk diagram" (Reference 6) Is In the above-mentioned "probability of frequency" format. It consists of a family of curves, p(*i), with the parameter P being the cumulative probability. To use this diagram, we would, for example, enter with a specific value, x), and choose, say, the curve P 0.95. The ordinate of this curve, e.9S(xf), is then the 95th percentile frequency of x<. That Is, we are 955 confident that the frequency with which damage level x, or greater will occur Is not larger than e.95tx()< We can now draw a vertical line tnat cuts the family of curves at a specific damage level, xj, that may be of special interes*. The 0027H060584 12 HONS 020620 DAMAGE le v e l 13 HONS 020621 intersection of this line with the family leads to the curve of ? versus as shown In Figure 5-1. This curve, which we call a "cut curve,* expresses our state of knowledge about the frequency with which events of level or greater occur. These uncertainties are due to uncertainties (n the data and the models used. In this study only, one cut curve is established for each damage index. This cut curve is for one or more fatalities or injuries. To establish complete risk curves, we had statistical data on the exact number of injuries and fatalities per fire incident. Our data sources indicate only single fatality fires and a few Multiple injury fires. Therefore, the uncertainties in the multiple injury or fatality parts of the risk curves will be very large and the differences among the different fatality or Injury levels will be dominated by the uncertainties. Because of this and because of our fire propagation model (which is described below and considers only two cases--fires that are confined to the room of origin, and fires that propagate beyond the room of origin), the final results would not be very sensitive to the level of damage, that Is, the number of fatalities or Injuries. Thus, the first-cut curve would yield sufficient Information for the risk comparison and the evaluation of any additional cut curves Is judged to be of little benefit. To assess the effect on risk of the decision to change from mineral oil to Askerel means to compare the two curves for each risk Index. There are several ways In which such a comparison can be done. The simplest way Is to compare the average frequency of one or more fatalities or Injuries. A much more detailed way Is to compare the cut curves, for the latter, based on the position, shape, and sprtad of the distributions, we can determine whether there Is a significant difference In the frequency of e certain damage level of e certain risk attribute. Both approaches are used In this study* To establish the cut curves for the two site specific cases, a three-point method graphically displayed by Figure 5-2 Is used, in the first part, the "Initleting events" (l.e.. events that create a disturbance to an otherwise normal system) are Identified and their frequencies are quantified. In the second part, "barriers" which tend to limit the effects of the initiating event are Identified and analyzed. Finally, tha consequences (the extent of damage) of the accident scenarios are assessed. Two Initiating evants are chosen: (11 transformer fluid leakage that leads to a fire, and (2) transformer rupture that leads to a fire. '>5 typms of disturbances that a transformer may experience are numerous (Reference 7) and are mostly of an electrical nature. A large portio-i )' these disturbances may not lead to a leak or rupture; those that may leal to either a leak or a rupture may also lead to a fire. It should be noted that our definition of the initiating events is very specific because not all leaks or ruptures lead to a fire. The frequencies of initiating events are evaluated In Appendix A. The barriers that limit the effects of the transformer disturbances jr the transfomar tank Itsalf, the room boundaries where ft is ins tilled, the fire protection system in the room, etc. The first barrier (t"it ' 002711060534 11 HONS 020622 CONSEQUENCES --------------- ) BARRIERS | U(/Ii UI C rs si K UI 12 S BE < o s < wa/c> o & M UI 3 > ! S S S ,NC ~ +i is HONS 020623 the tank) ts already considered part of the initiating eve-t. The remaining barriers are modeled by an event tree. Figure : snows n example from one of the cases. An event tree Is simply a grapnic presentation of potential sequences of events. The events are define? : the top of the tree and there are four of them In Figure 5-3. for example, one sequence of events Is sequence 1 of Figure 5-3 where tne initiating event is a leak that leads to a fire and the room boundary stays Intact. In that sequence, the rupture point and propagation to an adjacent room Is Imnaterlal because the Initiating event Is not a ruptur? and the room boundary is intact. The second sequence on that tree also starts with a leak and fire but the room boundary Is failed. Again, tie rupture point is immaterial but propagation to an adjacent room u certal n. To link the event sequences with the consequences, the sequences ar categorized by what we call the "damage states." The sequences coat do not lead to fire propaget on Into an adjacent room are categorized as 'room fire" (confined to 'e room of origin). The sequences that lead to fire propagation Into the djacent rooms are categorized as beyond room fire (propagated beyond t. . room of origin). This categorization s convenient because the statistical data on injuries and fatalities can 55 categorized Identically and the conditional frequency of one or more injuries or fatalities given a fire of certain category can be established. 5.3 CASE I - TRANSFORMER IM A REGULAR ROOM The base case for case l Is an Askarel-filled transformer Installed n a public-use building. The Installation characteristics are specific to Askarel type units, which means that the transformer room is not a vault and Is on a floor for general use. The alternate case Is a mineral oil-filled transformer, with specifications Identical to the Askarel unit. Installed in place of the Askarel transformer without any ot^er changes to the Installation or the room. The risks from these two installation types are evaluated separately. The general assumpt'cns me definitions pertinent to both alternatives are given first. 5.3.1 GENERAL ASSUMPTIONS The transformer Is of distribution type and Is liquid-filled usi'-j natural convection for dissipating the heat to the room environne-: .n power rating Is taken to be 167 kVA. Sulldlng transformers can r,--? between 100 and 5,000 kVA depending on the size of the building - range can be deduced from Reference 8 where the number of Askare transformers (primarily used in populous areas) is tabulated by :'? r power rating. From Reference 8. it can be deduced that 66% of a transformers are in the 100 to 5,000 kVA range. The average nunoe-- 'r gallons of dleloctrlc fluid for transformers In the 101 to 500 .a------e Is 183 for Askarel transformers and 117 for mineral oil transfor-e" (Reference 8). The power rating of the transformer considered - - study Is near the lower end of the above range and, therefore. ' volume is expected to be lower than the average; It <s taken to c? 80 gallons. The .ransformer is box-shaped, measures 2 feet by 2 - ' and Is 3 feet tail. hflPTMOROSAA 16 MOMS 020624 lil cc I c s > * K t M n vu 10 oOz I z si tfl cjc <zc ! $ 5x 5 < 17 HONS 02062^ The building 1* similar to en office building (n term* of the c 0-icu >- o' loading and use. Each floor has more than 10,000 square feet of c';?r' area. All the floors are divided Into corridors, rooms of different slaes, elevator shafts, ventilation ducts, etc. The transformer room ; located in a general-use area. For example, the corridor outside t"e transformer room Is assumed to be carpeted, occupied like any otnecorrldor In the building, and may contain combustibles such as paoer plastics. The transformer room measures 20 feet by 20 feet and is 8 feet high (see Figure 5-4), The walls ire of plaster without windows and are rated as I-hour fire walls. Thare is only one door (of the same fire rating) that opens Into the adjacent corridor. The walls are covered By mineral tiles on the Inside to reduce the transmission of transformer noise into -n corridor. The ventilation for this room Is provided by fans and c_i-js that directly coonunlcate with the outside air. Special precautions are taken to prevent storage of combustibles near tne transformer. To control any transformer fluid leaks, a curb area is provided under the transformer with at least a 4-1nch lip and 3 feet of clearance on all sides. Thus, the curb area dimensions are 8 feet by 8.5 feet. A wet-type sprinkler system Is installed above the transformer and -ne curbs around it. Th temperature setting of the heads Is of ordinary classification (see Reference 9 for definition) which ranges from :353 c to 170*F (57*C to 77#C|. The room also contains other electrical equipment such as switches, breakers, and Junctions In special electrical cabinets. The eiectr-col cables to and from the transformers or other electrical equipment i tie room are generally inside conduits (metal tubes) installed dose to t-e celling. Under normal conditions, no one enters this room except for occasional visits by maintenance personnel, Qoors are normally locked and only maintenance and security personnel have the keys. The Inspection frequency Is judged to range between once a week and twice a day. In the following sections, It will be seen that the final results i e highly sensitive to only a few of these assumptions. Because of i .-`-eyr rated plaster wall. In the casa of mineral oil transformer ruptyre, f'-? propagation to an adjacent room is very likely. Because of 6-mcr -net reinforced concrete construction, propagation of transformer fire "cugn the celling to the floor above is deemed to be very unlikely fzr i scenarios considered In this case study. The sprinkler system in room allows for rapid suppression of fires and reduces the iikei'-c:o of their propagation to the adjacent rooms. The curb dimensions is important to mineral oil fires from a leaking transformer. It ''.r-ces the severity of potential fires involving oil collected m the The details of risk calculations for the two transformer types i-In Sections 5.3.2a and 5.3.2b. To facilitate their comparison. format of this report Is altered and the two sections are prese-by side. 0027)1060584 18 MONS 020626 general USE ROOM ELECTRICAL CAElNETS ETE WALL vr ELECTRICAL CAtINETS GENERAL r use ROOM w> yI-HOUR RATED WALL w GENERAL USE CORRIDOR 1-HOUR HATED WALL FIGURE 5-4. GENERAL LAYOUT FOR CASE 1 TRANSFORMER RCCM 19 HONS 02062?  . -i 1^: 1 -:s y * o w w I ijsr'tx-- e if 5*S:"i -iHJiSi-'*! nr.h*"i`. im s !:;>th 3 Sij;^=|1 # - ** c O 41 b;*s *-*ce*sWi m y _`S laSSC=*TtJf l- 4 5 oS 3 i * ;"rSsS 33"*lii X s< a?iJal^-5s80*e*I*I. < b fe V _ b 3 :=!!5-`'3j U'i'LSfg4 --* * ftf A u *!]< C'' tJaZ Ab c|Ji|y 1*2MyiOta>wV i ll%Z 9 9. & * i -2J ; e -- -- i s* I 3S 3 5? 3. sail k t<r:t :*** I:'!jM-. S- 3 & =4 I 1 .i: =:j:5!<.| i'O 52 vv v aast o 9 *< ji !s:!% si-a5s: ji*5:i{ . ^rfn*hm .1 -fi^z 3{:J3t!5l)JI-T3 9 ? C4/ 3 3 5" 3*S3S =V.*J* <o * 1 ]*3h W * `'I ai h|fi J4 *Z lb b*bf-e 9C *y** Iv 20 HONS 020628 I GO* *Q AH Mt LI* 0UNQA4T AOMCIMI icnino* t*au*i *aom C*r , t0 I rt*o U*OMti*fNCI oA|TWAATG(I 1 Ith^iuM F [ NOW| M R It (tMTACn ft# I HO coMtauinei l RAOttRlftt n.to"4 LiflUW t* RLIAvRtrvM > room Run t ie 1 lAiu/M -KfA*4< IM f-i oonr*i*nlomtiiortrooroeatOioMiiniMinr IIVONQ Tor room or NA OMfOttQTAw^>tIlCOIA*tttV<Mt O iMOPAftinORI til ir* M*O*0*OO 9*<Oi Rl FIGURE 5-5. EVENT TREE FOR THE BASE CASE OF CASE 1, THE ASKAREL TRANSFORMER kluRvmM u* oumoaat MIWI M**r tlioo MroQAPRAoGoAjoftcMai uau Urn agon OMMIITAtl 1 rofO*fuiocr TAAMrMuOo* r*rr --AH -- lorraen iir 4I'AUJ IJ>H -A04 I ' ML It ffAIU I 011 JII Mil AO0ri* iiiio'yi 1 1 URVRf oiu#Rt-aetttAM<oi nr- QRROtlf10 CCMPIH1HT ML RomiMMiflIRUOWO LtVli MV OURTUOf MOV* LOU* R. L|VHMORMATVM MVOMO TMR HOOK Of ORIGIN ICSMARIQl HA - MOT A#OUCMLl Mv- r--'-4----- imo MORMAriOW "* 4 OOWR1M >1* >0 *> l aavonoROOMfiM n.n^i FIGURE S-6. event tree for the alternate case, THE MINERAL OIL TRANSFORMER OF CASE 1 21 MONS 020629 i 2 * u i 2u I I & t i *t9-#(t|#ow^*>S-lxI^LJLe>U-1* .9e <-]=-4^-9/ W OW --***^**-*-*.***0 C9 fa-**:,*':: -,*r jf"i ii:"23''r-v w<9 |ekah#- i ae* ^S||9^9*4gE -c-2*-5eV5^ * *--b"'S'5--w' %<a-. ** it- ': = : - = rlf3: s 2 3*=**2 5w * 9* u ki a ,, w. I r these conili Hons ^l5aa*k ia*S.^--*x*i""-- ofoa_-*l ~vsve>a**4*;*ea2- --e*-1 "2s. *=. .1 sis" "=25 .-.Cl au * a > w w -wlimMbs w;.eo;e:# :go*-i HONS 020630 I 5~| * lf-f=? tni riS* #- m I <* 3 a t 1 s *4 Oh. rw * -- e 2*" * 5 ic-h* Co U If mm r'i-if <* IJ5|i?l ? _ Ca o m9* f9 ** O* I 0# lft3*s* w=_m * -1- JS-;" s S.==5J- & 1 'S-f :a- 2JSo=ja 5 S*3l! =i=rr ii "ctr^sst Stilus n i c * ? I i CM 4* S 23 HONS 020631 I I bo ISl* O>jw *5c liftat l *> * Sai k *C4 **"w S ?48 u i-<s: ^ st. 3 5 J'l! 1 .b ** pr: 3 fw W k i In: a - sail < ?L *" m * ~ P2^ k * <- M -i 55 m*. ih:.rr w* w 3 8b- b i mb s w b | Ic |i s uV3_ <| KV-**1 -*2 Hill!!2 1 a ! *O S?#s -- k 1 O* _owf^a> -o_o- I Ju* .w M i e : 51 aw 2 5 ^ if E -- *. V - j * V ri~usu *< iCt_v k?*r ~M * w) V*3V '* w *" C 2 .33-5?-- S*,SJSS oO'" Vk"w -- o oL J *> * w o b p>o2 $ 9*% a5 *kMo r *k v*-C cw ihtinsl - * o5vv -- -- 3 S S s^>*>- 24 kons *632 si r s J 9 - t m. * bKE)-b-. %' * C Jf e * 9 *. c I s + ** l at t. e a. i * HONS 020t>33 I HONS QZObl1* CASE I - AUAUL HUKSTOOftl lC*SI - HIKCRAL OU nUUnfORMIt --i*y x c*t* l o M - O 1* -I52 ISi M>* 1fei--^!1*u0!W S C* k -M.S.2 2i* *t * -- c *- % *" Oc ty ** b 9 9 u * & *" 3 i 5S2SE ii.it1 .li.ii st i-- C a#*. ."rti nil:-; s--= c--^ 2*SS3= } ** ut i M i1 Si * aa 2 1 2 s. e l3 3 3; fla ^f!-ll25c! s bii 4aJI. * s si .-rises* oi *fcS _ O _ -- !| bm l%# #k .1 ] l Is HONS 020635 t r 28 HQNS 02063t> CASt I ASlUAtl TMUSFOUEI ICISC MlNtML OIL TIM iyO N C I HONS 020637 H i d u n c ttrls tlc irt: i rS it h rc M lIlt. 2.2 < 4 HONS 020636 t stqw acci M l tM lr aug* lU U l rt I N I c ib f *lt*H *l Qll HONS 020639 U H u in . H> | I Ik 3 tel 5 S<t 33 HONS 0206<rl 5.3.3 CASE I RISK COMPARISON The change In risk when an Askarel transformer fs replaced by a mineral oil transformer Is evident from the frequencies of Table* 5-3 and 5-4. For direct comparison between base case and alternate case results, these frequencies arc repeated In Tables S-S (for Injuries! and 5-6 (for fatalities). More percentiles are given in these tables than in Tables 5-3 and 5*4. Another way of showing the differences between the risks from the two transformer types is given In Figure 5-7. The mean frequencies of injuries and fatalities can be put together in a 2 by 2 matrU form litem C of Figure 5-7) where the rows represent transformer type and columns represent risk attributes {Injuries end fatalities), it can ae shown that this matrix can be obtained by multiplying together two 2 by 2 matrices. Matrix multiplication Is described below. In the first one, the damage state frequencies ere given where rows represent transformer type and columns represent damage states (room and beyond room). In the second matrix, the conditional frequencies or injuries and fatalities are given where the rows represent damage states and columns represent injuries and fatalities. To explain matrix multiplication, start with Item C of Figure 5-7. The four elements of this table (or matrix in this case! are tne results of the four equations given In Sections 5.3.2a.2.2 and 5.3.2b.2.2. Matrix multiplication follows the same equations. For axamplt, tne mean frequency of Injuries from Askarel transformers Is obtained front 2.0 x IQ"4 x 1.6 x 10"2 4.6 x 1<T7 x 8.8 x ID*2 - 3.2 x IQ*5 Another example can be fatalities from mineral oil transformers. The mean frequency Is obtained from 1.0 X 10'* x 1.2 x 10'3 1.5 x 10'* x 8.8 x 10'3 - 1.4 x 10*6 This matrix representation of our risk calculations, is seen below, facilitates the process of Identifying the Important contributors. Also, these matrices facilitate a sensitivity study of different input parameters. Table 5-5 shows that for injuries, only a small portion of the two probability distributions overlap and tne 5th percentile of the mineral oil distribution Is graeter than the 95th percentile of the Askarel distribution. Thus, It can be claimed that the frequency of Injuries would almost certainly be Increased if an Askarel transformer is replaced by a mineral oil-filled one in a setup such as case 1 of this study. For fatalities, the overlap of the two probability distributions (see Table 5-61 is significant. The Sth percentile and median of the mineral oil distribution are at the two sides of the median and 95tn percentile of the Askarel distribution. The two distributions do not intersect. ` establish the degree of overlap between the two distributions, the Dam contributions to the uncertainties and dependencies among them must imm be established. 0027M06Q534 34 MOMS 020642 TABLE 5-5. ANNUAL OCCURRENCE RATES OF INJURY FIRES FROM TWO TYPES OF TRANSFORMERS. CASE 1 Percentile 5th 10th 20th 30 th 40th 50 th 60 th 70th 80 th 90th 9Sth Mean Value Frequency of Injury Fire* Per Transformer rear Askarel Mineral Oil 1.7 x lO-6 2.0 x 10'6 2.3 x 10*6 2.6 x IQ-6 2.8 x 10'6 3.1 x 10-6 3.4 x lO-6 3.7 x lO-6 4.1 x lO-6 4.7 x 10*6 5.2 x lO-6 6.9 x ID-6 7.9 x I0'6 9.5 x 10-6 1.1 x 10-5 1.2 x 10-5 1.3 x 10-5 1.5 x 10-5 1.7 x 10-5 J | 1.9 x 10-5 2.3 x 10-5 2.6 x 10-5 3.2 x 10'6 1.4 x 10-5 ____________________ i Q032iK>60484 35 02Q61*3 HONS TABLE 5-6. ANNUAL OCCURRENCE RATES OF FATALITY FIRES FROM TWO TYPES OF TRANSFORMERS; CASE 1 Percentile 5th 10th 20th 30th 40 th SOth 60 th 70th BOth 90th 9Sth Mean Value Frequency of Fatality Fire* Per Tramforwer Year __ ___________ __ Askarel Mineral Oil --I H **o 1 8.6 x 10*10 9.3 x 10'9 9.6 x lO-8 1.2 x 10-7 1.7 x 10*7 2.3 x 10*7 2.7 x 10'7 3.3 x 10*7 3.9 x W7 4.8 x 10'7 5.9 x ur7 2.5 x IQ-7 o ---4 H O 5.6 x lO-7 7.5 x 10-7 1.0 x lO"6 1.2 x lO*6 1.7 x lO"6 i1 | | 2.2 x lO'6 3.4 x lO-6 4.3 x lO'6 2.4 x 10'7 1.4 x 10-6 ___________________ 003211080384 36 MONS 020644 {Stoo MMtoi WM to to w > > 2 uj Ml < 3 I !!M Mm ji-lo m m 1 m w i 1m ai- C X< Sto tro ii 2I* MOM FIGURE 5 -7 . SUMMRY OF CASE I RESULTS si gii Ml si<mt t*o b m 2 % m*m tM 3SC3w* iilll8* si 5" 1 3c 3 t a 3 3 3 8iSt5n"i 53 ; K^ tal O ! !i X *5 MM S ii ! 1CtM M* Ml 3 M MS iiMi m25a | s"i s1 ;I m* 3 i: ME i*s<_ rsi * h 1Mml 1 2Ml ---# 37 HONS 0206*5 Fran Figure 5-7, It Is deduced that the frequency of fatalities for Askarel transformers Is dominated by room fires and for mineral oil trans-ormers the frequency Is dominated by beyond room fires. The conditional frequency of fatalities from a room fire has a wider uncertainty range than the frequency of fatalities from a beyond room fire. This partly explains the wider spread for the Askarel distribution. The dependency between these two conditional frequencies is judged to be weak because the only relationship between the two frequencies that can be Identified Is that the conditional frequency of fatalities from a room fire Is always less than the conditional frequency from beyond room fir# (fp rm fp bnq). This relationship holds for the bulk of the distributions ror these two frequencies. Moving back In the computation process, the uncertainties In the damage state frequencies are Investigated, tn Section S.3.2a.2.1, ft is found that for Askarel-filled transformers ROOM a *R and In Section S.3.2b.2.1, It <s found that for mineral oil-filled transformers, 96% of the beyond room fire frequency is from sequences originating by a rupture. Thus, there Is a strong dependency between the important parts of the two fatality curvas. However, the uncertainties In rupture frequency Ur) are much narrower than the uncertainties m the two conditional frequencies of fatalities (fptRN end fp^Q). In summary, to compare the two fatality distributions, three Important points are made: (1) the two cumulative probability distributions do not Inters act; (2) there Is a weak dependency between the conditional frequencies of fatalities given room and btyond room flrts (the two Important parameters contributing to the uncertainties of these cumulative curves); and (3) there Is a strong dependency among the other parameters [rupture frequency, Ar) contributing to the uncertainties. From these, we can claim with a high level of confidence that the frequency of fatalities would decrease significantly If an Askarel transformer replaces a transformer in a setup similar to case 1 of tills study. In the discussions above It Is observed that the main difference between Askarel and mineral oil transformer fires Is In the likelihood of propagation beyond the transformer room. The two frequencies of beyond room fires are compared In Table 5-7 for the two types of transformers. From that table, we conclude that fires from mineral oil transformers a*-* (more than) 100 times more likely than fires from Askarel transformers to have a large Impact on the building (beyond the transformer room). In spite of this large difference between th* two types of transformers, the Injury and fatality frequencies (as can be seen in Figure 5-7) are not as widely apart. The reason for this reduction In the difference lies in the differences between the conditional frequencies of Injuries and fatuities for room and beyond room fires (see Item 8 of Figure 5-7). Although the transformer rooms In buildings are generally unattended a^o the general users (except for the maintenance crew and the management of the building) do not have access to these types of rooms, the conditional QQ27M060S84 38 MOMS 020646 TABLE 5-7. ANNUAL OCCURRENCE RATES OF FIRES PROPAGATING BEYOND THE ROOM OF ORIGIN FROM TVQ TYPES OF TRANSFORMERS; CASE 1 Percentile 5 th iotn 20 th 30th 40th 50 th 60 th 70th 80th 90 th 95th Mean value Frequency of Beyond Room Fires Per Transformer Year Ask ire! Mineral on 1.1 * 10'9 4.2 * ID*9 1.5 x 10-8 3.1 x 10'8 5.9 x IQ"8 1.0 x 10-7 1.6 x 10"7 2.7 x 10-7 5.0 x lO*7 9.6 x lO-7 1,8 x 10`8 7.4 x 10*5 9.0 x lO'5 1.0 x 10*4 I I 1.2 x lO-4 1.3 x 10-4 1.3 x 10'4 1.5 x lO'4 1.7 x 10*4 1.8 x 10'4 2.2 x 10-4 2.2 x 10-4 4.6 * lO-7 1.5 x 10-4 0032:1060484 39 MOMS 020647 frequencies of Injuries and fatalities for room fires are significant when compared to the sane conditional frequencies for beyond room The applicability of these conditional frequencies to our case studies discussed tn Section 5.3.4, 5.3.4 SENSITIVITY OF THE RESULTS Many assumptions and judgmental1y evaluated parameters are used throughout this case study. In this section, the sensitivity of the final results to these assumptions and parameters Is envisioned. Far this. It Is convenient to follow the sequences of the event trees of Figures 5-5 and 5-6. The Initiating event (l.e., leak leading to fire or rupture leading to fire) frequencies are obtained from (see Appendix A) aR ` *flre^R *L ' * fire (1"fR> where Afira Is the frequency of transformer fires. It was obtained from statistical data for mineral oil transformers. fR is the conditional frequency of a rupture causing a fire that is obtained fro-i judgment. The uncertainties In transformer fire frequency (Afire1 are rather small because It Is assumed that all transformers behave Identically and thus the sampling set Is very large (strong statistical evidence leads to a narrow uncertainty range). In evaluating the conditional frequency of a rupture causing a fire (fp), It is concluded that the likelihood of a rupture given a fire Is greater than the likelihood of a leak leading to a fire. Two Important assumptions have been made at this stage. It Is assumed that (1) fires from Askarel transformers are almost solely from ruptures, and (2) their fire frequency Is equal to the fire frequency from rupture of mineral oil transformers. If these assumptions are true but fq is ovarestimated, It can be shown that for smaller fp the risk from Askarel transformers (that is, Injury and fatality frequencies) would decrease and move further away from the mineral oil related frequencies. Essentially, lowering of fp Is the same as reducing \fir* of Askarel transformers. It should be noted that If the Askarel does not contain trlchlorobentene, the second assumption (that Is, the frequencies of rupture and fire of two transformer types being equal) would become very conservative, because the PCBs In general (and especially Aroclor 1254) have vary high Ignition temperature (their flash point is greater than 350*0 whan compared to trlchlorobenzene. Obviously, a more realistic risk comparison would be made If Afire for Askarel transformers is evaluated separately from relevant statistical evidence. This evidence Is not readily available. Room boundary failure frequencies for Askarel and mineral oil transformers are evaluated primarily from judgment. For an Askarel transformer, an explosion has to occur first. An explosion is possible because 1,2,4 trichlorobenzene may be vaporized (Its boiling point 's 214*0 and ignited. The PCS of the Askarel would not contribute to 002711080384 10 HONS its boil in? point Is higher (greeter than 350" C) and does tft1 ^*c4U%ijch point up to boiling, Tht 1 Ik til hood of boundary failure not h**e * -p0B| judgment with a mean value of 0,084, which represents dti eval"^'"d 0f explosion occurring with sufficient severity to damage * ne '1k* the element of the boundary failure evaluation Is the presence an inrooenrene. If an Askarel that does not contain tini`obenatne <e.g.. Aroclor 1254) was chosen for this case study, C''1C K,ndarv failure would have been very unlikely. However, as f^ foo* own ^ fPequency of room boundary failure given an Ask are l (cfl" emr ruoture) becomes smaller, It can be shown that the final tr,nfr^ouid not change much. On the other hand, if the Askarel is r*,UJd to be lOOt FCBs (such as Aroclor 1254), the frequency of fnrner fires specific to this fluid would have to be quantified trliiclt>y because of Its flame resistant characteristics (high boiling * int and flash point). As shown earlier, this would have a direct affect on the risk curves and It Is anticipated to move them toward lower frequencies* roe likelihood of room boundary failure In the case of mineral oil transformer rupture Is Judged to be a cartalnty. This Is bacaus* mineral ait is a flacmbla material and It 1$ Judged that an explosion may occur uoon transformer rupture. This may ba a conservative conclusion, but a more realistic likelihood would not be much less than unity. A decrease m this likelihood would Increase the frequency of room fire damage state. The overall effect Is Judged to be a slight decrease In the final risk curves* The remaining frequencies, reflected on the event trees of Figures 5-5 and 5-6, are from heat propagation analysis (except for the conditional frequencies of below or above-liquid-level rupture). The uncertainties In these frequencies are not highly sensitive to the underlying assumptions. For example, in the case of fire propagation to an adjacent room when room boundaries are breached, the Important parameters are the heat flux from the flames and Ignition tamperature of the materials in the adjacent room. In view of the large uncertainties In the propagation frequencies, any reasonable changes In the values of ignition temperature or heat flux would not have a significant Impact on the final results. in important factor in calculating tha fire propagation likelihood is tie fire brigade response time (see Appendix B). The evaluation of this roponse tlma is based prlmlarlly on fire detection time and travel time for the fire brigade. Little consideration is glvan to special conditions arising from transformer fires. In two of the three Askarel transformer fires involving buildings, the fire brigade delayed their attack by 50 minutes and 3 hours after their arrival. These ware the times when the electric power to the transformers were disconnected, A sensitivity study was performed which Indicated that large Increases m fire brigade response time did not significantly affect the final results; that is, did not Increase frequencies of Injuries and fatalities -27'!080334 41 HONS 020649 The conditional frequencies of Injuries and fatalities, given a damage state, art evaluated using statistical evidence (see Appendix Cl. Tne most important assumption at this stage of the analysis is the compatibility of the collected evidence to our case study. For beyond room fires, it is judged that the compatibility 1$ good, because once the fire propagates out of the room of origin, It does not matter how It origfnated. For room fires, the compatibility of the evidence Is good for Injurfej because one of the two data sources Is specialized to Indoor transformers (see Appendix C) and the frequency estimates from Individual sources are very close to each other. For fatalities in a room fire, the compatibility may not be as good as the other frequencies. This is reflected by wide uncertainties. 5.* CASE 2 - TRANSFORMER INSTALLATION WITH MINERAL OIL CHARACTERIZAT`,1'1 The base case of case 2 is a mineral oil transformer in a vault in the basement of a large building, and tha alternata case Is an Askars 1 transformer with identical specifications. Its main differences from case I are that the walls and celling are required to be built of 6-incn concrete or 8-1nch brick, and that tha vault Is dedicated to the transformer, with no otner electrical equipment Inside. The following general assumptions describe the vault and the transformer in detail and are applicable to both alternatives of this case. 5.4.1 GENERAL ASSUMPTIONS A single transformer is located In a vault In tha basement of a large building (see Figure 5-8 for a plan view). The transformer converts a 4,800V power source to 120/240V. The rating of the transformer is 167 ItVA. It measures 2 feet by 2.5 feet by 3 feet tall and contains 80 gallonn of fluid. The vault Is approximately 20 feet by 20 feet and 8 feet tall. Its floor Is sloped l/16-1nch per foot to a sump 3 feet square and 2 feet 6 Inches deep. The vault nas a dedicated ventilation system. The air Inlat Is equipped with a fire damper. The air outlet Is equipped with louvers to prevent the entrance of rain. The capacity of the blower Is 2.000 ft3/inute The penetrations of th room Include air ducts, conduits for electrical cables, and a self-closing, 3-hour rated fire door (5 feet by 8 feet) which Is normally locked closed. All walls, the floor, and the ceiling are constructed of 6-Inch reinforced concrete or 8-Inch brick. The vault If assumed to be equipped with an automatic sprinkler system. The temperature settings of the heads are taken to be of ordinary classification (see Reference 9 for definition) that range between 57*: and 77*C (I35*F and 170*F). The details of risk calculations for the two transformer types are give'' In Sections 5.4-2a and 5.4.2b. To facilitate their comparison, the format of this report Is altered and the two sections are presented sii? by side. 0027M060534 42 HONS 020650 FIGURE 5-8, PLAN VIEW OF THE TRANSFORMER VAULT FOR CASE 2 43 HONS 020651 :s? .2 at -- * * * *1 Jk | aH.-a-2 a Islj =w 5 - -- M; 2 -15 32 2 . !* u v--c Iy 3 'p pC i* %S 2 * --e oj ^c j n n e k wv_ It-5 ns? ^ <9,^4(i| #i* K _ f k ***a4m 4 #* -- j4as - **2S-**? > a vj j* - *< w I *-- i=2= l-5i i Kin*! ISJt si* a .*slsa~f w . r Ipmw tk k * -- (N 3 sjItJUi*. -<{>.s! S i-a- j. - *t':arT i2s2*ss-cs =1-2 -, e ** i* * kb I --* *s=^ rj* i). t ** a - . __ 1 * J^ k _ < '. "^Vt *r*`tSu"* -| v3p- -x:r n . 2.!|3. ;4i! - i :|c3e i~ u. ,23312*.. ti't a *t * **. -1 |4* I * s "*<a }*.3 3 2 fiji - * J*5p! * + 9w seal m M * iillMl a^S s-s -IT 3*2 - r?s l;b sui* VSI -j - -s. i C2 a is: bis t ki* Uk l - - nm -s 4 iisi -I *> - p r t* W 4*^ w* t iwfiirt iil-d i a *5-kSl*--= s lift. js-rb <s_I .aki* wd a ** 4a HONS 020652 tlAfcfnum** L tvOautflfOt OV 'AHUM *9*0 aumum 'Owr MUM 0**tOVfl kVQUlOUVtL MOMOAttOai TOAMAOJAClMt *OC* H0ut*ci UUMMM OAMAGfr*T ''*O!*ut SC* IllMI ^iH1 H frAill li i|`4 Tl> II "It U l * klC A AurrvMil vfJ vM.r MUMMY 'AMLUM miMIKO - I So* CDMniMINTO* III ML MlMLIQMUVIl *urmM Mv nmuTvTiMu*awg lwil - pwe*A*newMvo^ tot NOTMUCMi * -- M # " MV -- <MLOM -- .... 'H > noon 'i*t <0 * I NTOMOOQHIlH )l*t * 1 MOM'iAC i MTQNMOM'HlI 1 f if * I nOOMMftt K.i* 1 MTOM ROOM r1*1 41 A >0 * FIGURE 5-9. EVENT TREE FOR TVE BASE CASE. THE MINERAL OIL TRANSFORMER OF CASE 2 LiMumvnMt. 1 FIGURE 5-10. EVENT TREE FOR THE ALTERNATE CASE OF CASE 2. AN ASKAREL TRANSFORMER IN THE VAULT 45 MONS 020653 1 * 2 t m 1 t * ^ 2 | Sjl 3 tTfc~! S'" . | *5 ,, S a i ^ (i e 3* 46 HUNS 02065* U U 2 . tU LU IL liu u ia w i ft* j t ~. 2X .t b3 iiim;j J fsjs Inn ^ O ft* > Pal* IS i !*t-i lpb C SI'*ljlr.f 1 ?|H! * m. o 1w> buz aS >-Ss= Si i;iss C A U 2 - M iNLAAL O K I M a V O M I 47 NQNS 0*0655 W ; Jf 1 *u* ^a S' u V nw l3|i* =-sr*--a2 * A i -wvj ^ Hi taco* m i" a * =I*~5Ijs- I isJ;*" 3* 3 - Ji rs H 22 3 u 3 a 3 48 HONS 020656 50 HONS 0*0658 I "I2 M ; r!5lse 1 1*^ 1 : :H " * 4 <!> k J S*3 .* ~ *s 2 '! *u *s2 2# * *il I- 0 iZz m. 5:&i ** '*bV4 *oW. !I*s:Oi 1hj-i1 r s ^u * * "* if:-. O *W w iis --9^ i i ujfc I fjS. wi N )l|U L O il H M V O M .I -1? *Ji*'l -iis -E*, e:si2 !i||! iUllir*3 l<u 51 HONS 020659 t* 52 HONS 020660 4 3 CAS 2 RISK COMPARISON change in rislt when * "tn#r*1 0<1 transformer Is replaced by an T kirel transformer is evident from the frequencies listed In Tables 5-9 **** s-ii. For direct comparison between base case and alternate case 4,,tuns these frequencies are repeated In Tables S-12 (for injuries) Ifld 5-13 (for fatalities). These tables 9lve more percentiles than fables 5-9 and 5-11. The results are also sunmarlzed In the form of Ltrices (Figure S-llh The mean frequencies of Injury and fatality fires *r* put together in the form of a 2 by 2 matrix. As shown in Section 5.3.3, this matrix can be obtained by multiplying two 2 by 2 matrices (matrix multiplication Is described in Section 5.3.3). the frequencies of room or beyond room fires and the conditional frequencies of Injuries or fatalities given a room or beyond room fire. Based on the following observations. It Is judged that with a high level of confidence, the frequencies of Injuries and fatalities would decrease if a mineral oil transformer Is replaced by an Askarel transformer,1n a setup similar to case 2 of this study. Table 5-12 shows that for Injuries, the probability distributions overlap to some extent but do not Intersect each other. The Sth percentile of th* mineral oil distribution Is between the median and 95th percentile of the Askarel distribution. Th* latter (l.e., th* Askarel distribution) Is mainly from th* product of two terns frequency of ruptures Up) and conditional frequency of an Injury fir* given a room fire (f; rj,|) . This can be easily verified by Inspecting Figure 5-11. The mfneral oil distribution, however, cennot be expressed as simply; 69% of Its mean value Is from beyond room fires. This means that an important part of th* uncertainties Is from the conditional frequency of an Injury fire given a beyond roe* fire (f{ gug) end a small portion of It from fj RM. It should be noted tftat th* dependency between fj gun and m'rm Judged to be very weak because for almost all prbbaole values fI,RM`< fI,M0 which Is Intuitively expected to hold. Another important pert of injury frequencies Is the initiating avent frequency. For Askarel-filled transformers, 100% of It Is from transformer rupture and for mineral oil, 92% of the frequency Is from rupture. Thus, there Is a strong correlaton between the two Injury curves at the Initiating event level. For the fatality distributions th* same type of th* dependencies among th* Input parameter exists ts for the Injury distributions because the same type of contributors ere used. Tnus, th* seme conclusions can be reached. Similar to cast 1. from tha above discussions we conclude that an Important difference between Askarel and mineral oil transformer fires is in the likelihood of propagating beyond th* transformer room. The two frequencies of beyond room fires are compared (n Table S-14 for the two C027HQ80334 53 MOWS 020661 TABLE 5-12. ANNUAL OCCURRENCE RATES OF INJURY FIRES FROM TWO TYPES OF TRANSFORMERS; CASE 2 Percentile 5 th 10th 20th 30th 40 th SOth 60th 70th SOth 90th 95th Kean Value Frequency of Injury fires p*r Transformer year Askarel Mineral Oil 1.7 * 10-* 1.8 x 10`6 2.3 x 10*6 2.6 x 10'6 2.8 x 10-6 3.0 x 10'6 3.3 x 10*6 3.6 x 10`6 4.0 x 10`6 4.7 x 10'6 5.1 x IQ*6 3.7 x 10'6 4.3 x lO'6 5.4 x 10'6 6.3 x 1Q`6 | 7.2 x IQ*6 0.3 x 10*6 9.2 x 10*6 1.1 x 10'5 1.2 x lO*5 1.5 x lO*5 1.8 x lO*5 3.2 x IQ'* 9.2 x 10*6 003211052964 54 MQNS 020662 TABLE 5-13. ANNUAL OCCURRENCE RATES OF FATAL!IT FIRES FROM TWO TYPES OF TRANSFORMERS; CASE 2 Percentile 5 th 10th 20th 30th 40th 50th 60th 70th 80 th 90th 95th Mean Value Frequency of Fetality Fire* Per Trensfomer Year i Askerel Mineral Oil 2.0 * 10'9 4.9 * lO'9 9.3 * 10*fi 1.3 * 10*7 1.9 * 10*7 2.1 * 10*7 2.6 x 10-7 3.1 x 10-7 3.9 x 10-7 5.2 x 10*7 5.9 x IQ*7 1.3 x IQ*7 1.9 x ia*7 2.9 x 10*7 3.9 x 10*7 | 4.9 x IQ*7 6.0 x 10*7 7.5 x 10*7 9.4 x 10*7 1.3 x 10*6 1.9 x 10*s 2.7 x 10'fi 2.5 x 10'7 8.5 x IQ*7 1Q12MQ529B4 55 MOMS 020663 tii> n FIGURE 5 -1 1 . SUMMARY OF CASE 2 RESUITS tofc t i>s" VI <5 f*b 2 Zhim g(CS 3<< 32 2S * v ;s sMl 1 m a 1 #4 38 1a< !* 1 3 2. li le 8 ;e| i:S U* < 2*3 is;*> lit *5; 8 < ait Ik isuBiMn ISl ssi is ssi mt M 2 3I - i5 gif all wax m is am > S> m 1s1 Is-i 2 m m IM h < C r 2m M i "i 3 3H M a 23 hIM 9 X l kI* k V*M M m ll 3 3 M .j Ml 8<3 < MIORLA | OIL 1 56 HONS 02Q66* TA31E 5-1*. ANNUAL OCCURRENCE RATES OF FIRES PROPAGATING BEYONO THE ROOM OF ORIGIN FROM TWO TYPES OF TRANSFORMERS, CASE 2 Percentile 5 th 10th 20th 30th 40th 50 th 60 th 70 th 80 th 90th 95th Mean Value Frequency of Beyond (too* FT res Per Transformer Yeer Askarel Mineral Oil 1.4 x 10'10 4,7 x IQ'10 1.6 x 10-9 3.5 x 10-9 7.0 x IQ*9 1.2 x 10* 1.9 x 10*8 3.4 x 10*8 6.4 x 10-8 1.4 x 10*7 3.1 x 10*7 1.9 x 10-5 2.2 x 10*5 3.1 x 10-5 4.3 x 10-5 5.3 x 10*5 6.4 x 10-5 7.7 x 10-5 8.9 x 10'5 1.1 x lO-^ 1.4 x IQ'4 1.6 x 10'4 7.3 x 10*8 7.3 x IQ'5 0032ll0b04fl4 57 MONS 020665 transforms types. From that table, It Is concluded that fires frCn mineral oil transformers are (more thanl 60 times more likely man fir#? fro* Askarel transformers to have a large Impact on the building. 5.4.4 SENSITIVITY OF THE RESULTS The discussions on the sensitivity of the results to the underlying assumptions and judgmentally evaluated parameters for case 1 (Section 5.3.4) are almost directly applicable for this case Smite- to case 1, Askarel transformer fire frequency and the likelihood of vault boundary failure are dependent on the presence of trichlorobenzene m ?, Askarel. Also, similar to case 1, changes In Askarel transformer fir# frequency would have a direct effect on the final results, for nore realistic results, the frequency of Askarel transformer fire would nav to be evaluated from relevant data and Is anticipated to be much s-iall#than the value that has been used. The likelihood of a vault boundary failure in the case of mineral oil transformer rupture Is taken to bo small (YB-Z on Figure 5-9). This nas increased the importance of room fires to the final results. It can t>e shown that if this paramettr is further reduced (that is, a smaller llkallhood for vault failure), the mineral oil risk curves would move closer to Askarel risk curves (given that evarythlng also Is kept as is). Injury and fatality conditional frequencies are less relevant here because (1) those conditional frequencies were based on data from all types of office buildings, and (2) vaults are not typically found in buildings and are not accessible for the general building occupants. Also, outside the vault In this case Is not a general use area; thus, ft general occupants of the building are less exposed to the transformer fire consequences In this case than In case 1. 5.5 DISCUSSION Having compared the fire risk from mineral oil and Askarel transformer? for two caso studies, has the goal of this study been achieved? To answer this question, let us examine some Important facets of this st.-., The risk attributes chosen are acute Injuries and fatalities. In the 1930s, the decision to roplaco mineral oil with PCBs was mada based f i qualitative understanding of risk. It Is difficult to assess today whether public health protection or property damage was the main motivation for the decision. Certainly at that time, the envlronne-t i persistence of the PCBs was unknown to the experts and the fire res--.* characteristic of the PCBs was tha focal issua of the decision, "hj?. the Injuries and fatalitlas caused by transformer fires were chosen j, the two risk attributes for this study. Property damage and loss oproduction because of transformer fires are not addressed. The two sltos chosen for this study are In publicise buildings. ># comparison Is made In case 1 between an Askarel transformer at an Installation specifically designed for It and a mineral oil transro--=' replacing tha Askarel transformer. Because of the sprinkler system the transformer, the mineral oil transformer in this setup Is alio-*' 002711060504 53 HONS 020666 per the national electrical code (Reference 16), Thus, the case 1 results carry an important message: under equal conditions, the fire risk decreases significantly when a mineral oil transformer Is replaced by an Askarel transformer. The results of case 2 can be interpreted in a similar way. The setup (a *ault in the basement) is specific to a mineral oil transformer. Askarel transformers above 35,000 volts are also put In vaults according to the electrical codes (Reference 161, The risk comparison in this case also represents a realistic situation. The primary side of the transformer considered 1$ 4,800 volts, which is much less than 36,000 volts. However, our analysis is nor# sensitive to the size of the transformer [dictated by its power level) than the primary voltage. For larger transformers, the amount and surface area of the combustibles (tnchlorobenzene in the case of Askarel transformers) are larger than those considered In case 2. Our Judgment is that for larger transformers, the injury and fatality frequencies would be farther apart than those obtained in this study, because the mineral oil transformer fire would be more severe. Comparison between the base cases of the two case studies is not warranted for two reasons. First, the same conditional frequencies of injuries and fatalities are used. These frequencies were obtained from general building fire occurrence data. In both cases, the room of origin is not readily accessible for the general building users (the door is locked closod). In case 1, the rooms adjacent to tha transformer room are usod by the general building users, whereas in case 2, outside the vault, only the building maintenance or management personnel are typically present. Second, the models used to estimate the conditional frequency of boundary failure and fire propagation Include many conservative assumptions so that a direct comparison may be misleading. The extension of the final results to other situations Is also not warranted. For example, the Injury and fatality frequences ere not applicable to distribution transformers on poles at the backyard of suburban homes or transformers Installed outside tha building. However, transformer rooms In Urge buildings are caammplace and the base cases of our case studies represent real situations. Of the Intermediate results, only the initiating events' frequencies and Injury and fatality conditional frequenclat are applicable outside the two case studies. In sumaary, although only two specific cases are examined, the goal of this study was achieved. The comparison of probability distributions for Injuries and fatalities for both cases Indicate that there it a significant decrease in the fire risk when an mineral oil transformer is replaced by an Askarel transformer. 0027M06Q584 59 MQNS 020667 6. REFERENCES I. Monsanto Company, "Aroclors - Physical Properties and Suggested Applications," Application Data Bulletin No. O-P-115, St. Louts, Missouri. 2 . Monsanto Company, "Transformer Askarel Inspection and Maintenance Guide," Bulletin No, IC/FF-38R-2, St. Louis, Missouri, August 1975. 3. Gann, R. G., "Deuel opment of FI amiability Criteria for Transformer Dielectric Fluids," NBSIR 80-1992, U.S- National Bureau of Standards, February 1980. 4. Llpowltx, J., "Fire Safety Properties of Some Transformer Dielectric Liquids," Journal of Fire and Flanaablllty, Vol. IS, pp. 39-53, January 1982. 5. Winkler, R. L., and u. L. Hays, Statistics. Second Edition, Holt, Rinehart and Winston, 1975. 6. Kaplan, S., and B. J. Garrick, "A Quantitative Definition of Risk," Risk Analysis. Volume 1, No. 1, pp. 11-27, 1981. 7. Caldwell, J., and J. Olmsted, 'The Real World of Transformers * Risk Evaluation,` American Risk Management Loss Control Conference, Louisville, Kentucky, April 25, 1979. 8. Resource Planning Corporation, `Coments and Studlas on the Use of Polychlorinated Biphenyls In Response to an Order of the united States Court of Appeals for tha 01 strict of Columbia Circuit," submitted to U.S. Environmental Protection Agency, February 12, 1332. g. Flra Protection Handbook, Fifteenth Edition, National Fire Protection Association, 1976. 10. Hemstreet, R. A.. "Flaimaeblllty Tests of Askerel Replacement Transformer Fluids,* Factory Mutual Research Corporation, RC78-T-42, August 1978. II. "Fire Huerd Properties of Flamble Liquids, Gases, Volatile Solids, 1977," NFPA 325M, National Fire Protection Association, Inc,, Quincy, Massachusetts, 1977. U. K1rk-0thmer, "Encyclopedia of Chemical Technology," Vol. 5, Third Edition, 13. Barkan, P., B. L. Damsky, L. F. Ettllnger, and E. J. Kotskl. "Overprassurt Phenomena In Distribution Transformers with Low lipedance Faults: Experiment and History," IEEE Transactions on Power Apparatus and System, Vol. PAS-95, No'. 1, January/ February 1976. 0027M060584 60 020668 kons 14. Zabetakls, M. G., "FlanmabllIty Characteristics of CombustiDle Gases and Vapors," Bulletin 627, Bureau of (Hnes, U.$. Department of Interior, 1965. 15. Kaplan, S., "On the Methods of Discrete Probability Distributions in Risk and Reliability Calculations * Application to Seismic Risk Assessment," Risk Analysis. Vol. 1, p. 189, 1981. 16. "The National Electrical Code Handbook," National Fire Protection Association, Quincy, Massachusetts, 1984. 17. Levlnthal, D., "Application of Decision Analysis to a Regulatory Problem: Fire Safety Standards for Liquid Insulated Transformers." NBS-CGR-B0-L98, Center for Fire Research, National Bureau of Standards, December 1979. 18- Slu, N. 0., "COMPBRN Computer Code Users Hanual," Pickard, Lowe and Garrick, Inc., PLG-0275, July 1983. 19. Thomas, P. H., "Some Problem Aspects of Fully-Developed Fires," Fire Standards and Safety. ASTM STP614, A. F. Robertson, Editor, American Society ^or Testing and Materials, pp. 112-130, 1977. 20. Slu, N. 0., "THEAT Computer Code Users Manual," Pickard, Lowe and Garrick, Inc., PLG-0272, July 1983. 21. Lie, T. T.. Fire and Building. Applled Science Publishers, Ltd., London, 197z. 22. Gibson, T. 0., "Learning Value from Recent Loss," Plant/Operations Progress, Vol. 3, No. 1, p. 18, January 1984. 23. Bartknecht, w., Explosions. Course Prevention Protection, Sprlnger-Verlag, Berlin, Heidelberg, New York, 1981. -02711080384 61 MONS 020669 APPENDIX A FREQUENCIES OF TRANSFORMER PROBLEMS APPENDIX A FREQUENCIES OF TRANSFORMER PROBLEMS the frequencies of transformer problems are assessed In this appendix using Bayesian methods. In Section 5, It Is found that the transformer problems of Interest are: (II Ignition of a leaking mineral oil transformer, (2) ignition of a ruptured mineral oil transformer, and (J) rupture of an Askarel transformer. To establish these frequencies, statistical evidence on a sample of transfonsers and the number of fires (categorized by rupture or leak) that were observed Is needed. The data from several sources were reviewed. Individually, none of the sources could provide sufficient Information for establishing the number of occurrences and the size of the sampling set. Data used in this appendix are primarily from sources covering all of the United States; however, some important data are obtained from sources specific to the State of California. Also, from the available Information, the fire Incidence could not be categorized by rupture or leak. Therefore, the combined frequency of fire Is first established and then two conditional fractions are introduced. One depicts the conditional frequency of a leak given a fire and the other depicts the conditional frequency of a fire given a rupture. The frequencies for mlnqral oil transformers are evaluated in Section A.l. The approach to arrive at these frequencies Is also discussed In that section. The results of Section A.l are extended In Section A.2 to assess the frequencies of problems for an Askarel transformer. Section A.3 discusses all the available Information, including Information Indirectly related to the frequencies of Interest. This type of Information is used to better establish some of the parameters using judgment. A.l FREQUENCY Of FIRE IN MINERAL OIL TRANSFORMERS A.1.1 GENERAL FRAMEWORK In Section A.3, It is found that from the available information sources the fire Incidence data cannot be categorized by rupture or leak. Therefore, an overall frequency of fire, Afire. "HI be established first. This frequency can be written as the sum of two frequencies Afire * A rupture and fire * Meak and fire where (A-; Aflr* frequency of transformer fire. ^rupture and fire " frequency of fire caused by transformer rupture. Aleak and fire " frequency of fire caused by the Ignition Of the leaking mineral oil. -032*1060484 A-1 MDNS 020671 further, we define ffi fraction of transformer fires that are caused oy ruptures. fL 1 - fp * fraction of transformer fir** that are caused by the Ignition of the leaking mineral oil. and we can write k rupture and fire * * fire fR and Mean and fire * Arire W 11 Probability distributions are first assessed for Affrt and fg, and the frequencies of fire and leak and fire and rupture are obtained fr:~ Equations (A.2) and (A.3). A. 1.2 BAYES' THEOREM The frequency of transformer fires, Xfire> 1* assessed using Bayes' theorem (References A1 and A-2). The frequency of transformer fires per transformer year can be considered as the parameter of a binomial distribution. Let M be the nuiber of transformers and K be the number of transformer fires. Given the frequency of transformer fires, Afjre (or X for short), the likelihood that 1C transformer fires occur in a population of N transformers is L(N,K|k) A.X The probability distribution t0(A) represents our state-ofknowledge uncertainty about x prior to collecting the evidence, as the subscript o denotes. Bayes' theorem Incorporates the new evidence that K transformer fires occur In a population of M transformers Into the prior distribution to obtain the posterior distribution v(x); l.e., v (A) ; `u (X ; which represents our new state of knowledge aboutx. [f assume tna: the prior distribution is a beta distribution with density function (Reference A-2) f (A) (n-1)! . k-1,, . ,n-k-l 0 < A < 1 (k.l)Hn-k-l)! ' 3Q30MQ72584 A-2 M3NS 020672 where n and k are the parameters of the distribution, the posterior distribution (s also a beta distribution whose parameters are n+N and liK. Since K and N are very large numbers (a few hundred and more than a million, respectively), the precise form of the prior distribution does not matter, as long as It represents the relatively vague state of Knowledge that we have prior to receiving the evidence. The nonmformative prior with ntto Is used, which means that the prior mstnbutjon nas no effect on the parameters of posterior distribution. Sayes' theorem can be directly applied only If N and K are known exactly. In Section A.3. It can be seen that there are uncertainties about the exact number of transformers and transformer fires; therefore, these uncertainties must also be assessed and propagated. A.1.3 NUMBER OF FIRES. K The number of transformer fires In California per Reference A-3 are: Number of Fires Year 648 1980 599 1981 500 1982 There Is a significant variation In the number of events experienced m different years. This temporal variation Is not considered explicitly in our models, because It Is judged that the conclusions of a time-dependent analysis would be of little use to the goals of this study. An average value Is used and the variations are Incorporated; l.e., the assessed state-of-knowledge uncertainty for It Is wide enough to accommodate such temporal variation In the uncertainty analysis. A possible explanation for this variation Is that state business activity declined In those years and resulted In a decreased load on the transformers and this drop in load Ms ltd to a sharp decrease In fire frequency. There may be some underlying phenomenon. For example, soma transformers may be operated at loadings slightly above their rated design and consequently experience i higher rate of fire occurrence. A slight drop In the load may thus lead to a substantial decrease In the fire frequency. It should be added that Urga variations In statistics of this sort are commonplace. For example, airplane crashes, auto accident rates, and drunk drlvar arrests experience large variations at a state level. An average number for transformer fires In California Is established by 53 * * 500 m 5022 Incidents per year This number Is assumed to correspond only to mineral oil transformers for several reasons: (1) Tht national ratio of Askarel transformers to mineral oil units Is 1 to 1M (see Section A.3I; (2) fire frequency of Askarel transformers is expected to be less than fire frequency of -330H060484 A-3 MDNS 020673 mineral oil transformers; (3) Askarel leaks are not expected to eaten fire whereas mineral oil leaks have the potential of doing so, and (4) the cause of transformer rupture Is mainly due to the failure of insulation of the windings which does not depend on transformer type. The standard deviation of the observations Is obtained by 1648-582.3)2 * (S99-582.3)2 (500-582.3>2 L 3-1 75.0 The distribution of K is assumed to be a normal distribution with its mean equal to 582.3 and standard deviation equal to 75.0. It is represented by an equivalent discretized probability distribution. K Probability. ^ <18 3.99 x 10-2 498 0.1866 582 0.547 666 0.1866 746 3.99 x 10-2 A.1.4 NUMBER OF TRANSFORMERS, N In Section A.3, It can be found that information on the number of transformers Is available only at a national level. Our fire incidence data are from California. From Reference A-4, it Is concluded that the total number of distribution transformers owned by Pacific Gas and Electric Company and San Olego Gas and Electric Company (two of several large electric utilities In California) Is 892,197, while data are not available from other utility companies and other transformer owners m California. Therefore, we estimate It Indirectly, using population! ratios. It is assumed that the populatlonel characteristics of transformers In California are almost the same as In the united States and the total mmAer of transformers Is proportional to the number of people In the area. The only variations from this that are anticipate*, are that California has a smaller population of heavy Industries (wmen typically use a lot more energy than the eastern states such as Pennsylvania, Ohio, Michigan, and Illinois) and large California cities have milder climates (which put less demand on area heating or cooling Thus, we can write Nuaber of mineral oil transformers In CalIfornla California population /Number of mlnera' U.S. population \transformers m California population 23.7 million (Reference A-5) 0030M060484 A-4 M0NS 020674 U,$. population 227.2 nil lion (Reference A-5) Number of mineral oil transformers in California 23.7 x 106 * 23.1 x 10 227.2 x 105 2.4 x 10S The statistical data on the population are reasonably accurate. The error rate Is probably less than u. The number of transformer* in the united States estimated by Reference A-l may nave larger uncertainty. Besides the estimate of 23.1 x 10 transformers. Reference A-4 also gives an estimate of 22,870,000 based on the methodology that Reference A-6 had developed. This corresponds to approximately 1* variation. Also, the assumption that the number of transformers is proportional to the population introduces some error. The distribution depicting these uncertainties Is derived from Jud^Mnt. A 201 variation about the point estimate of 2,4 x 10 Is chosen in assessing the following distribution: Number of Transformers In California, M Probability, PH 1.9 x 10* 2.4 x 10* 2.9 x 10* 0.3 0.4 0.3 A.1.5 FREQUENCY OF FIRE From the two discrete probability distributions for n and K. 15 pairs of <M,K) can be formed with probability Pm Ppp*. for each pair, a beta posterior distribution *m(A) can be obtained with parameters n and K. The distribution wU) that depicts our state-of-knowledge uncertainty about the frequency of mineral oil transformer fires can be written as v(l) Each of the 15 (fill's Is a narrow distribution and can ba approximated by abnormal distribution. Table A-l lists the 15 pelrs of parameters, along with their probability and mean values. It can be shown that some of these distributions overlap near tne tails or the distributions. Therefore, v(xl Is a multimodal distribution. Instaed of discretizing w(A) using the exact expression, the first column and the last column of Table A-l Is chosen as an approximate discretized probability distribution for Afir#. This approximation would be exact If the distributions nine's did not overlap. Tha resulting distribution Is collapsed to the discretized probability distribution listed in Table A-2. 003011060484 A-5 MCNS 020675 TABLE A-l. DISTRIBUTION PARAMETERS FOR *NKU ) PNR-PK*PN K N 1.2 x 10-2 5.6 x 10'2 0.164 5.6 x 10*2 1.2 x 10'2 1.6 x 10-2 7.5 x 10`2 0.219 7.5 x 10*2 1.6 X lO-2 1.2 x lO"2 5.6 x 10"2 0.164 5.6 x lO'2 1.2 x 10"2 418 1.9 x 10 498 1.9 x 10 582 1.9 x 10 666 1.9 x 10 746 1.9 x 10 418 2.4 x 10 498 2.4 x 10 582 2.4 x 10 666 2.4 x 10 746 2.4 x 10 418 2.9 x 10 498 2.9 x 10 582 2.9 x 10 666 2.9 x 10 746 2.9 x 10 Mean 2.2 x 10*4 2.6 x IQ*4 3.1 x IQ'4 3.5 x 10`4 3.9 x lO'4 j' j X e--O* 1 2.1 x lO*4 2.4 x 10-4 2.8 x 10-4 3.1 x lO*4 1.4 x IQ'4 1.7 x lO*4 2.0 x IQ-4 2.3 x lO-4 2.6 x lO'4 | | 1 l i ! ; ' A-6 001311033134 MONS 020676 TABLE A-Z. DISCRETIZED PROBABILITY DISTRIBUTION Of Afire ``fire (per transformer year) ProbabllIty Cumulative Probability 1.7 x 10'* 2.0 x 10-* 2.4 x 10-* 3.0 x 10'* 3.6 x 10-* 8.4 x 10*2 0.251 0.343 0.255 6.B 10*2 8.4 x 10*2 3.35 x lO'l i 0.678 0.933 ! ____1 NOTE: Nun 2.48 x 10** per transformer year. oounoans4 A-7 MCSIS 020677 A.1.6 FREQUENT OF MINERAL OIL FIRES CAUSEO BY LEAKS ANO RUPTURES Tha statistic*! data used In assessing the frequency of fire do not provide information on the relative frequency of fires caused by leans (fi ) and ruptures (fq). Other data have to be used to assess the relative frequency. Table A-3 lists some estimates of transformer leakage rate by transformer type. Reference A-7 estimates that 15 out of 143 failures of Indoor transformers resulted In ruptures and fires. Reference A-3 estimates the failure rate for liquid filled transformers to be 4.1 x 10"3 per transformer year. Therefore, a point estimate of Arupture and fire 4.1 x 10'3 x A|- 4.3 x 10** per transformer year This estimate Is at the extreme high tall of the distribution for Afire *n<! 1* Judged to be inappropriate for direct application. Thu discrepancy Is not surprising because the exact definitions of the two parameters used In the multiplication are not known to us. This result Is Interpreted as depicting that the mejorlty of transformer fires result from ruptures. The frequency of a fire from leakage can be obtained fora the estimated leakage rates for the moderate leaks that are listed In Table A-3. They represent the type of leak that can potentially lead to Ignition and fire. Reference A-7 estimates the frequency of an Ignition of leaked ail as ID*3. Therefore, e point estimate of Xie|( lnd fire can be 7.7 x IQ"3 x 10"3 7.7 x ltr/transformer year This estimate also lies at an extreme tall of the distribution for Afire. this Is interpreted as Indicating that leaks constitute the smaller portion of Xf{r(. The above estimates of the frequencies of fires caused by ruptures srp leeks Indicate that the relative frequency of rupture-caused fires is much higher than that of leak-caused fires. The following distribution Is developed judgmentally. It reprtsents our beliefs about the reUtw? frequency of the two phenomena. Probability 0.05 0.80 0.15 fR 0.5 0.80 0.95 \ * U,R 0.5 0.2 0.05 Mean 0.81 0.19 C030M060484 A-8 MQNS 020678 TABLE A-3. ESTIMATES OF LEAKAGE FREQUENCY (PER TRANSFORMER YEAR) Trantfonaar Typa Mineral Oil PC8J fl9 1 C* CTILf 1. Seal 1 Leak 8.5 x I0*z , Moderate Leak 7.7 x IQ'3 2. Active Leak 6.0 x 10-3 3. 1.6 x 10`4 A-4 Active Laak 3.2 x 10~2 A-9 1.6 x 10*4 u_~_J JJ3h033ig4 A-9 MDNS 020679 Th* frequency of fire* from leeks and ruptures can b* obtained from Equation* (A.2) and (A.3). Table A-4 give* their characteristic values. A.2 FREQUENCY Of RUPTURES OP ASKAREL TRANSFORMERS Trantforaer ruptures are caused by very large surges In the primary voltage or rapid breakdown of fnsulatfon between the windings, they cause arcing between the windings which generates gases and produces pressure spikes. Voltage surges and insulation failures do not depend on the type of fluid used. The gas generated per kJ of arc energy is approximately 100 cc for both mineral oil and Askarel (Reference 4-10). Therefore, the conditions leading to ruptures are ttia same for mineral oil and Askarel, and It Is expectad that the frequancias of ruptures per transformer year will be the same for both types of transformers. Thus, we will use the frequency of mlntrtl oil transformer rupture that was assessed fn Section A.l for the frequency of Askarel transformer rupture. To put this estimate In a batter perspective, using the mean frequency of rupture of 2.0 x 10'4 per transformer year, the expected number of Askarel transformer ruptures nationwide Is calculated 2.0 x 10*4 x 140,000 28.0 ruptures per year This seems to be a large number because Askarel transformer troubles generally draw new* media atttntlon. The frequency of citations of Askarel transformer fires Is far lass than 28 per yaar. A.3 SOURCES Of DATA In this part of the appendix, th# sources of data used In Section 4.1 are described. Also, some supplemental Information art discussad such as sources of statistical data and breakdown of the estimates for the nunoer of transformers, th# frequencies of transformer leak, failure, rupture, and fire. A.3.1 NUMBER OF TRANSFORMERS tn Reference A-4, the utilities throughout the United States ware surveyed regarding thm numbers of equipment they own that contain mineral oil and polychlorinated biphenyls (PCS) and the amounts of fluid they contain. It Is estimated that tho number of mineral oil transformers ownod by the utility industry Is 20,227,428 and tha number of Askarel transformers Is 39,640, Also given fs breakdown of tho number of transformers according to their power rating and the average amount of fluid that the transformers in a range of power ratings contain, it is also ostlmated that tha total number of mineral of 1 transformers owned 3* the utilities end other users fn tn* United States Is 23,137.493 and the 0030M060484 A-1Q MONS 020680 TABLE A-4. CHARACTERISTIC VALUES OF THE FREQUENCIES OF MINERAL OIL TRANSFORMER FIRES CAUSED BT LEAKS ANO RUPTURES Parcwitll* 5tn parctntll* Median 9Stt> Parctntll* AL*afc tnd Fir* * Rupture and Fire (D*r transformer yiir] (p*r transformer year! 1.2 x 10-5 4.8 x 10- 7.4 x 10's 1.4 x io-* 2.0 x 10-4 2.6 x 10-4 ; H*an 4.7 x 10-5 2.0 x 10-4 -OUMO32084 A-11 HONS 020681 total number of Askarel transformers is 140.000. The following list -5 , breakdown of tne mineral oil transformers accoraing to their type: Distribution Transformers: Overhead and Pole-Mounted Pad Mounted Small Power Transformers Secondary unit Substation Transformers Large Power Transformers 19.274.B19 3,565,842 192,748 52,042 5;,04; 23.137,493 The breakdown of the PCS transformers according to users (Reference A-'.. < Is: Category of user Humber of units Utilities Industrie) end Commerclel Railroad 42,000 97,000 1,000 140,000 In Section A.l, the number of transformers In California (s used for frequency evaluation. The only available Information Is the number of distribution transformers owned by two utility companies in California (Reference A-4>: e Pacific Gas end Electric 779,764 e Sen 01090 Gas 4 Electric Co, 112,433 Thus, the nwbtr of transfomers in California must be much greater tjn 900,000 units since there are several ether large utilities such as Southern California Edison end Los Angeles Department of Water and Power. A.3.2 TRANSFORMER FIRE At the national level, one primary source of fire date Is the National Fire Incident Reporting System (NFIRS) (Reference A-12). In e survey of MFIRS during 1976, 1977, end early 1978 (Reference A-13), one million fires were found to hive occurred in eight states and 209 indoor transformer fires were identified. Seven transformer fires were Identified in the more recent survey of the 1978 MFIRS data. These transformer fires could be caused by a rupture or the ignition of tne leaking oil, and It Is not known which type of transformer is involved. At the state level, the primary source of fire data for California 1; California Flrt Incident Reporting System (CFIRS) (Reference 4-141. computer starch of the date base for fires caused by transformer and associated overcurrtnt and disconnect equipment shows the following result (Reference A-3): Year Number of Fires 1980 648 1981 599 1982 500 0330M060484 A-12 MO NS 020682 Similar to HFIRS, no information 1$ available to deterimne the types of the transformers Involved and the type of abnormality leading to the fire. Only a few fires involving Askarel transformers are recorded. They are listed In Table A-S. There is insufficient Information to determine whether Askaret was burning during these fires, and whether the Askarel transformer was the source of Ignition. A.3.1 TRANSFORMER LEAKS Three sources have estimated the frequency of transformer leaks. Table A-3 summarizes their results and each source fs discussed separately. A.3.3.1 Electric Utility Owned Transformers ;n Reference A-4, data are collected for transformers owned by the electric utilities. Two levels of leak severity are defined. Greasy or oily spots on the transformer body are considered as small leaks and noticeable spills are considered as moderate leaks. Reference A-4 shows that the frequency of a leak Is greater for larger transformers and they are less likely to have a moderate leak than a small leak. For mineral oil transformers with power levels below LOO kVA, point estimate for small leak frequency is 0.0077 per transformer year and for moderate leaks it Is less than 0.0077. For mineral ofl transformers with power levels between 101 and 500 kVA the small leak frequency fs 0.085 and for a moderate leak the frequency Is 0.0077 per transformer year. A.3.3.2 Survey of Chtmlcal Manufacturers Members of the Chemical Manufacturers Association were surveyed to Identify Incidents of transformer laaks (Reference A-91. Sixty-eight firms representing about 33S of all U.S. chemical Industry sales In Standard Industrial Classification Code 29 were surveyed. Information based on the Environmental Protection Agency (EPA) Interim measures regulatory program and firm records Indicates 351 leakage Incidents per year occurred In a population of 5,231 transformers, including both mineral oil transformers and Askarel transformers. Therefore, the leakage rate is (.7 x 10"`/transformer year. If the nondocumented sources such as the Jud^wit or opinion of equipment maintenance or service personnel are Included, there are 1,005 leakage incidents In 15,752 transformers. This produces a leakage rate of 6.4 x i0`` per transformer yeer. The rates listed in Table A-3 are the estimated rates for 'active leaks" which include ruptures and moderate leaks but not minor leeks where there is no evidence for the gathering or dripping of fluid on or about the transformers, using all sources of ^formation collected from the survey. A-3.3.3 National Bureau of Standards In Reference A-7, a Uestlnghouse service engineer is quoted as estimating the leakage rate to be 10" per transformer year and the frequency that * ne transformer experiences a leak to be 1 out of 25 durfng the first C010M060584 A-13 020683 HONS TABLE A-S. TRANSFORMER FIRES INVOLVING ASKAAEL TRANSFORMERS Oate of Occurrence 2/S/81 1/19/82 S/lS/83 9/26/93 Piece Location Transformer Type Blnghanton, New Tort Sequoyah Nuclear Power Plant, Tennessee San Francisco, California Chicago, Illinois Baseaent of Office Building 01 strlbutlon Outside, Above Grade, Next to Turbine Building Power Transformer Vault Distribution in Sub-Baseaent of Office Building Utility Vault Three Stories Below the Street Distribution Reference A-15 A-16 A-17 0013H033084 A-14 HONS 02068** 2 week* of service. TBu*. during the expected ltr span of J5 years, ;*.e iveragt leikije riti is l * 7T 2 week* 1 * 10 1.6 x 10 per trn*Conner year 25 years also given in Reference 4-7 Is the estimated failure rate of a transformer of 4.1 x 10"3 per year. This i* based on the iEEE-sponsortd reliability survey of industrial users of transformers (defartnee *-191. where failure meins loss of power for the transformer. The frequency of a leak being ignited is estimated to be 10*3 In Reference 4-7. A.3.4 TRANSFORMER RUPTURES Reference 4-7 states that statistical date gathered by researchers at Factory Mutual Indicate that out of 143 failures of indoor mineral oil transformers recorded by Factory Mutual, 18 resultad in rupturt of the transformer tank; In 15 out of the 18 ruptures, the mineral oil was ignited, according to Reference 4-8, the failures of transformers are censed by flashover or erclng (6111, electrical defect (lZtl, and mechanical defect (101). The failure rate* for different types of transformers ere estimated to be; Liquid Filled, 411 Types *.l x lO'^Transformer veer 601 - 15,000V Above 15,000V 300 - 750 kVA 3 x 10-3 1.3 x 10*z 3.7 x 10"3 4.4 REFERENCES 4-1. Apostolakl*, 6.. 'Data Analysis In Risk Assessments,* Nuclear Engineering and Oeslgn, Vol. 71, pp. 375-381, 1982. A-2. Winkler, #. U. end V. L. Hays, Statistics, Holt, Rinehart and Winston, 1975, A-3. Letter from J. Stark of California State Fire Marshal to m. Kazarians of Pickard, Low# and Oarrlck, Inc,, dated August 23, 1963. A-4. Resource Planning Corporation, `Cements and Studies on the use of Polychlorinated diphenyls In Response to an Order of the United States Court of Appoals for tne District of Columbia Circuit,* submitted to u.S. Environmental Protection Agency, February 12, 1982. A*$. U.S. Department of Comerce, Bureau of Census, 'Statistical Abstracts of tn* Uni tad States,* 1981. 0030/1060484 MONS 020685 A-S. Versar, le., "PCB Manufacturing. Processing, Distribution n Commerce and Use Ban Regulations: Economic Impact Analysis.' April 1979. a-7. Lavinthal, D., "Application of Decision Analysis to a Regulatory Problem: Fir* Safety Standards for Liquid Insulated Transformers,* N8S-GCR-80-198, Center for Fire Research, National Bureau cf Standards. December 1979. A-B. "IEEE Reeoanended Practice for the Oestgn of Reliable Industrial and Commercial Power Systems,' IEEE Std. 493-1960. A-9. Chemical Manufacturers Association, 'Comments in Response to an ANPR Relating to Characterizing Equipment as Totally Enclosed {46FED REG 16096),* EPA Docket Xo. OPTS-62015. A-10. Askarel Inspection and Maintenance Guide, Monsanto Company, St, Louis, Missouri. A-ll. versar, Inc., 'Assessment of the Use of Selected Replacement Fluids for PCBs in Electric Equipment,* U.S. Environmental Protection Agency, EPA $6076-77-006, March 1979. A-12, `National Fire Incident Reporting System,' U.S. Fire AiPsini strati on, currently part of the Federal Emergency Management Administration, Washington, D.C., 1978. A-13. Gann, R. G., "Development of Flaneabillty Criteria for Transfor-er oialactric Fluids,* Center for Fire Research, National Engineering Laboratory, National Bureau of Standards, NBSIR 80-1992. February 1, i960. A-14. `California Fire Incident Reporting System,' Annuel Report. California State FIra Marshal, 1962. A-15. Schecter, A., 'Contamination of an office Building m Binghamton, New York by PCBs, Dioxins, Furens, and Blphenylenes After an Electrical Panel and Electrical Transformer Incident, Chamosphert. fol. 12, No. 415, pp. 669-660, 1983. A-16. 'Quick Response Controls PCB Spill at TKA Plant,* field note. Power Engineering. April 1963. A-17. Firm Incident Report. Incident Number 0146S. San Francisco nTM Department, State of California, Office of the State Fire `'arsnsi. A-18. 'Toxic Smoka Forces EvKuetion of Bank,* Chicago Tribune. September 29, 1963. A-19. 'Hallability Survey.' IEEE Transactions of Industrial Applications. Vol. 1A-1Q. no. 2, March - April, 1914, 003QH060484 A-16 HONS 020*86 APPENOIX 8 FIRE PROPAGATION FROM AN ASKAREl TRANSFORMER RUPTURE HONS 020687 APPENDIX B FIRE PROPAGATION FROM AN ASKAREL TRANSFORMER RUPTURE in tfii$ appendix, for the base case of Case I, the potential of fire propagation to an adjacent room given an Askarel transformer rupture leading to room boundary failure Is examined. For the purposes of this study. Asxarel is defined as a mixture containing 50*. (by weignt) PCBs (Aroclor 1254) and 50X 1,2,4 trlchlorobeniene. In Section 5.3,2a.1.6 of trie main report, four modes of fire propagation are Identified. Three of tne four involve radiative heat transfer to the target combustibles from the flames above the transformer tank, the hot gas and smoke layer under tne celling, and hot transformer parts. This appendix concentrates on these three modes of fire propagation. The fourth mode Involves heat conduction from the splattered fluid and Is discussed In Section 5.3,2a.1.6. The target combustibles for fire propagation are carpets, wallpaper, paper, wood, plastics, and similar materials typically found In an office building environment. Reference fl-I Indicates that the Ignition temperatures of fabrics range from 400'C {752'F) for cotton to 600*C (1,112F1 for wool. Synthetic fibers have Ignition temperatures Between these two points. Fire retardant fabrics have Ignition temperatures well above 1,000*C (1,832'F). The Ignition temperature for wood ranges from 313`C (595'F) to 393*C (740*F), depending on the specific gravity of the wood. For this study, carpeting will be used as representative of the combustibles In the room. Its Ignition temperature Is chosen to be 600'C (1,112'F). To calculate the heatup of these materials to their Ignition teeiperature, it 1$ assumed that they are' In the form of a semi-infinite slab with in initial uniform tasperature T.. The heat flux 4o (w/m2) Impinges on Its surface for t* seconds, the time that the surface temperature reaches the ignition point. From Reference 8-2. It can be written (B.l) Mere k * thermal conductivity (wfm *k). a thermal dlffuslvlty (m2/*). T* * Ignition temperature t'K). "'om Reference 8-3, it Is found that cellulose acetate (a form of synthetic fiber) has the following thermal properties: k 0.17 to 0.3 w/m *K i * 7.6 x IQ"8 to 2.02 x 10*7 m2/S 32:'06048d 3-1 hons 0206S6 For our calculations. we will use 0.Z5 for k and 1.4 x 10-7 n2's a. Using T* 60Q'C and T/j 20#C, the following relationship between :* and q Is obtained: We must now establish i. For this we need the burning characteristies of the Askarel; l.e., the mess burning rate, the heat of combustion, etc. Heat of combustion for trlchlorobeniene Is 1.5 x 10* U/kg (Reference 6*4). The other parameters have not been established 'or Askarel, except that Reference B-5 Indicates that a 4-Inch diameter Dool of Askarel can burn and generate heat at the rate of about 5 kcal per minute only when exposed continuously to a pilot flame or Ignition source with greeter then 30 kU/ir heat flux. These values compare with 30 keel per minute or greater heat generation rates for mineral oll-t/oe fluids that, unlike the Askarel, do not require a continuous exposure to the pilot. Also, the heat of combustion 1.5 x 104 kJ/kg is about one-third of that of mineral oil. In an Askarel transformer rupture, the heat source may be present temporarily. The windings may become red hot from lack of cooling or abnormal currents. This situation Is temporary because either the protective or the maintenance personnel devices would Isolate the transformer. The Isolation time may be as long as 1 hour, depending jn how soon the transformer troubles are noticed. It is deemed that tie local fire brigade would certainly be called In and that they would request the shutoff of the electric power to the building. Based on the results of Reference 8-5, mentioned above. It Is concluded that Askarel! have poor burning characteristics. This Is Interpreted is Implying shorter flame heights and smeller mass burning rates than for mineral oil. The latter means less overall heat generation for the burning surface. Thus, because of shorter flames and less heat generation rate, the heat flux leaving the flames is Judged to be muen smaller. In Appendix E, the heat generation rate for a pool fire of mineral oil Is estimated to be 4.0 x 103 kW, and the heat flux from fire Impinging on the walls Is found to be about 34 kw/m*. it is conservatively assumed that the heat flux impinging on the wells fron -Askarel Is one-fourth that of a mineral oil fire; that is, 8.5 k'x/-* If this heat Is totally absorbed by the carpeting, the ignition t>me - is about t* 1.2 * 10-- . (8.500P sec a 28 minutes Within this time frame, the sprinkler system (If operable! is very to suppress the fire. If the sprinklers are not available. the loci' 003011060484 3-2 WONS 020689 Mr* brigtdt eey 4rr1ve. The condition*} frequency of fir# propagition, fp_l, given that a transformer rupture and rooe boundary failure have occurred, can then be written as fP*l* tfSPR * fr{t* * *SPR,<1 * fSPR^ Vb1 (6*3) where fjpp the unavailability of the sprinkler system. fr(x) frequency of event > tpB time of fire suppression by the fire brigade. tsPR * time of fire suppression by the sprinkler system. The teres In the brackets represent the frequency of the sprinkler system falling to suppress the fir* before the carpeting Ignites. The first term Inside the brackets Is the unavailability of the sprinkler system and the second tern Is the frequency of failure to suppress before Ignition given that the system Is available. The term outside the brackets Is the frequency of the fire brigade falling to suppress the fire before Ignition. It should be noted that It Is Judged that the two suppression processes are Independent because the fire brigade would be notified upon detection of the fire and they would bring along tht appropriate equipment Irrespective of tho sprinkler system effectiveness. The unavailability of the sprinkler system can be established by pooling Information from several sources. Reference 8-5 lists sprinkler performance records for several typos of occupancies* The nimdser of fire Incidents in areas with sprinklers Is more than 81,000 and tht numbar of unsuccessful operations Is about 3,100. Tho percentagq of satisfactory sprinkler operation ranges from 88.21 (for Industrial facilities handling bnverages and essential oils) to 98.21 (for industrial facilities aamifecturlng textiles). For offlco buildings, this porcontage is 97.4 (Mart were 13 unsetlsfKtory performances recorded In 494 offlet building fires where there were sprinklers). Thus, based on those records, the unavailability of the system rangos from 0.018 to 0.118. However, the uncertainties about the very sprinkler system of the transformer roem under consideration ere Judged to be greater because the sprinkler system hat to operate after surviving the impact of a pressure pulse and the transformer parts or fluid after tho transformer rupture. Th* following distribution Is dsrlvod from Judpmnt and daplets our uncertainties about tlw unavailability fSPR: fSPRProbability 0.02 0.10 0.05 0.70 0.20 0.20 Its wean value is 0.077. 1011060484 HONS Q2Q4>9 Since the frequency of Ignition time t' 1* lets then the suppression ting by the sprinkler* or the fir* brigade, *: 1* evaluated using the methods of Reference B-7. Th stat*-of-knowledge uncertalntle* In t*. tepe, and tn are established first and the statistical distributions of these time periods are convoluted next to establish fr{t* < t^p*) and fr(t* < tfg). The uncertainties In t* are primarily from Ignoring target heat losses and the augmented heat flux from the flames, the hot gas layer and the hot transformer parts, and using conservative flame surface area, conservative source to target distance, and conservative Ignition temperatures. In Reference B-7, an error factor, E0 Is Introduced, which Is multiplied by the point estimate resulting from the thermal analysis. Also, Reference B-7 interprets the point estimate as representing the mean value of the statistical uncertainty about the Ignition time because these uncertainties are overwhelmed by the state-of-knowlege uncertainties. The latter is depicted by the uncertainties m Eg, for which a lognormal distribution it used with 5th and 95th percentiles at 0.8 and 4.0. for this case. It Is Judged that our uncertainties are wider because wt have used judpwnt in deriving t*. Ey Is Introduced as the error factor with a lognormal distribution ana Its 5th and 95th percentiles at 0.5 and 4.0. Thus, the mean tin* to Ignition can be expressed as t*Ey IB.4} and the distribution for t* has the following characteristic values: 5th Percentile: 14.0 Minutes Median: 39.5 Minutes 95th Percentile: 112.0 Minutes Mean: 74.5 Minutes The fire suppression time by the sprinklers, tcea. can be envisioned as the sum of two time periods: the time of sprinkler head opening and the time of fir* suppression given water Is pouring from the sprinklers. The statistical uncertainties of suppression time are derived to bo significant (Reference B-7) and can be expressed by an exponential distribution. He can writ* frit* < t$pR> txp 1- t*/t$pr) <6,5) where tp# 1 the mean time of suppression (the parameter of the exponential distribution) and Is established from Judgment. Reference 8-: has used the following distribution for the mean time to fire suppression 0030M060484 3-4 HONS 0206^1 for I coopirtment In * nuclear power plant that contilns n automatic CO2 fir* suppression systan: Mean Tin to Suppress Ifflinutesi Probability $ 0.4 15 0.3 30 0.2 60 0.1 it should b* notad that this tstlmt* includes th* addad capability provided by the specialized fir* brigade that is typically found in a nuclear power plant. This brigade would attack a fir* if the automatic systMt fail t control it. Sine* ttw suppression system m our transfomr room usas watar, ttia distribution for r$pp Is chosen (in iplta of tilt prtsanc* of a fir* brigade In a nucl aar plant! to bo skewed further to the longer tins than tht above distribution. This is supported by Itoforanc* 1*4 whart th* effectiveness of tho suppressIon system Is compared. [t shows that a CO; tystaai nay not properly control the fire for mra than 30S of the donnds. The distribution for r$pp is chosen as: t (ailnutts) SH________ Probability 5 O.S 15 0.4 30 0.1 Tht wi of this distribution Is ll.S wingtes. Tho suppression tin by tho flro brlgadn tfg can b* envisioned as tht siaa of throa tin periods: detection tin, fir* brigade arrival ttaw. and fir* suppression tin, Tho detection tin Is judgnd to bo loss thin 5 ninutas because tho nolsa or saw** from a transfomr mptur* and room boundary failures would bo noticed by the occupants of tho bssonat, or tha abnomlitles in tho electric circuits would bo noticed by the building ntntemncd portonnl and tho general occupants. Tht fir* angads arrival tin Is onpactod to range fm 1 to 10 ninutas. This is taken fro* Ktferonce 1-9 whore data an collected for fin angina travol tins in Manhattan, how York. Finally, tho fin suppression tin (given the arrival of brlgadn) Is judged to vary fm S to 30 minutes. This rstlwata does not take Into account any serious 0*1 ays In taking action for raisons such at delay In disconnecting tna electrical agulpnent In the trensfamr room fm electric power sources. Slullar to the suppression tin by tho sprinklers, tha statistical uncertainties in the suppression tin by the fire brigade are taken to be asponantlal. Thetis, fr(t* < tpg) * a*p(- r*/TFt> (-61 XUMKMsa MOMS 020692 where tfb 1* tht **" suppression tine and 1$ derived from judgment. Based on the above discussions, the following distribution is chosen ro TFB-` TFB (minutes) Probability 15.0 30.0 45.0 0.5 0.4 0.1 The mean of this distribution Is 24.0 minutes. The egression for fp , can than be rewritten as fP-l CfSPR * (1 - fsppl exp(- t*/tSpn)] exp (-r*/tfB) (3.7) The distribution for fp,,i can be obtained by using discretized probability distribution arithmetic (see Appendix G and Reference 3-10). Its characteristics values art: 5th Percentile: 2.2 x 10-* Median: 1.3 x lO-2 95th Percentile: 1.1 x 10-1 Mean: 2.8 x 10-2 B.l REFERENCES 8-1. Fire Protection Handbook, Fifteenth Edition, National Fire Protection Association, 1981. B-2. Slu, N. O., "Physical Models for Compartment Fires," RellabD^v Englneerlno. Vol. 3. pp. 229-252, 1982. 8-3. Edwards, 0. K., V. E. Denny, and A. F. Mills, "Transfer Processes,* Hemisphere Publishing Corporation, Washington, 1973. B-4. Personal Communications between Raymond F. Boykin of Monsanto Corporation, and M. Kazarians of Pickard, Lowe and Garrick. : -- . March 1984, B-5. Hemstreet, R. A., "Flammability Tests of Askarel Replacement Transformer Fluids," Factory Mutual Research Corporation, RC78-T-42. August 1978. 3-8. Fire Protection Handbook, Fourteenth Edition, National FW Protection Association, 1978. 8-7. Slu. N. 0., and G. Apostolakis, 'Probabilistic Models for cao > ray Fire," Reliability Engineering, vol. 3, pp. 213-227, 1982 0030H060484 3-6 MOWS 020693 g-8. Miller, H. J., "Risk Management and Reliability," presented at Third International System Safety Conference, Washington, o.C. October 17-21, 1977. ' 8-9. Kolesar, P., If. Walker, and J. Hausner, "Determining the Relation between Fire Engine Travel Times and Travel Olstances In New York City," Operations Research, Vol. 23, No. 4, July-August 1975. 8-10. Kaplan, $. "On the Method of Olscrete Probability Distributions in Risk and Reliability Calculations - Application to Seismic Risk Assessment," Risk Analysis. Vol. 1, p. 189, 1981. Q03CM060484 B-7 020^ HOnS APPENDIX C CONDITIONAL FREQUENCIES OF INJURIES ANO FATALITIES ^ONS 020695 APPEMOIX C COKOITIOMAL FREQUENCIES OF INJURIES AND FATALITIES The conditional frequencies of injuries or fatalities given a fire are evaluated In this appendix for the two damage states (see Section 3 of tne man text): (1) "room fire," fires tnat are confined to the room of origin; and (2) "beyond room fire," fires that propagate beyond the room of their origin. Two sources are found to contain relevant Information: (1J Tables 3-18 and 3-19 of Reference C-l, and (2) the 1982 fire incidents as collected by the California Fire Incident Reporting System (CFIRS) (Reference C-2). The applicability of each data source is examined first. Bayesian methods (References C-3 and C-4) are then used to combine the pieces of evidence and evaluate the four conditional frequencies. C.l GENERAL FRAMEWORK As mentioned earlier, the statistical analysis Is based on a Bayesian framework, In the next section, we will see that for "room fires," two sources of data are available and for 'beyond room fires," only one source Is used. For the latter (that Is, only one source of data), Bayes' theorem Is used In Its slnpiest form. Bayes' theorem Is written as *(f) 1 IT L(E|f) (C.l) where (f) "the posterior distribution for f. * (f) the prior distribution for f (prior to obtaining 0 evidence E). L(E|f) the likelihood of observing evidence E given f is the true velue. N a normal 1 cation factor to make <i(f) a probability dlstrlbutlon. The evidence E can be, for example, r Injury fires out of n fires or it can be a set of experts' opinions about the valua of f. for the "beyond the room" damage state, the evidence is in the form of r occurrences out of n observations. Therefore, the likelihood function can be binomial (Reference C-5). That Is, L(r, nIf) n! fr(l- f)n ' r forO<r<n r! (n - r)T ~ (C.2) C017M040284 C-i HONS 020696 Front Reference C-S, it is also found that the beta distribution conjugate* with a binomial likelihood function. That n, if the prior distribution is beta, the posterior also belongs to the beta family of distributions. The density function of the beta distribution is n ( f1 (no-ll! /o <r0 * l>Hn7- r#:||! 1n (i - f) 0 ro (C.3) where r0 and n,, are the two parameters of the beta distribution. If this density is used as the prior distribution for an evidence of r events In n observations, the posterior beta distribution would have the parameters rl * r0 + r 1^-4, nj n f C.:) From these equations. It can be seen that ona can reduce the influence of the prior distribution by choosing very small values for r0 and na. uhen r>o, one can choose r0 n0 0, a nonlnfomatlve prior distribution. For "room fire," the two pieces of evidence art also In the form of r events In n observations. However, since they ere obtained differently and are from Independent sources, they are treated as statements from two Independent experts about the value of f (conditional frequency of an Injury or a fatality). To eosfclne the two pieces of evidence, a method given in Reference C-6 is employed. For M experts, Reference C-6 writes Bayes' theorem [Equation C.11J as n(f|ff.........fjj) - J Uff............fjj|f r0(T) <--' where f*1i the 1th experts' estimates for the value of f and L Is the likelihood that the experts' estimate will be (f;,.. fj) when the true value is f. For independent experts, it can be written Uf*......... fjj|f) <f*|f) i n 1.1 1 1 C where Li Is the likelihood that the 1th experts' estimate will be f" when the true value Is f. Li can be Interpreted as a measure of credibility, assigned by the analyst, to the ith expert for estimating the true value of f. If the error of an expert Is considered as an additive quantity, the likelihood can be represented by a normal distribution. The posterior distribution can be obtained analytically (similar to tTM beta and binomial distribution discussed earlier) if the prior C019M04Q284 C-2 HONS 02069? distribution (In the case of a normal likelihood) is taken to be norma), The posterior distribution Is alio normal and for two experts Its mean (f) and variance (u2) are given by f w0f0+ w^ w2f2 1 *1*1 T TT 0 1 *1 T J2 (c.8) (C.9) where f0 anda02 are the mean and variance of the prior distribution and f, and o<2 (for 1 l or 2) are tne mean and variance of the likelihood function of the 1th expert. The weights w0, wlt and wj are obtained from w o2 0T o (C.1Q) w (C.11) 1 w (C.L2) 2 Note that the Influence of the prior distribution can be reduced by choosing a very large p2. A nonlnformetlve normal distribution can thus be one with j0approachlng infinity. For the type of evidence that we have collected, the Man and variance of the likelihood are taken to be the Man and variance estlMtes of each evidence; that Is, r/n and r/ir, respectively, where r Is the nwsber of Injury or fitallty fires and n Is the total nuatter of fires. By using these estlMtes, the assumption Is made that the two pieces of evidence presented by the sources are equally good and the difference In our confidence levels in the two sources Is solely dependent on the number of observations that they have Mde. C.2 SOURCES Of DATA C.2.1 DATA FROM REFERENCE C-i The statistics given in Reference C-l are extracted from the National Fire incident Reporting System (NFIRS) (Reference C-7) for 1977 and 1978, based on tna National Fire Protection Association (NFPA) Standard no. 901, "uniform coding for Fire Protection* (see Reference C-8). nfiss 0019(090284 C-3 HONS 020698 is a voluntary system. In 1977, 19 states ware providing data to NFIRS (Reference C-9) and In 1978, 3 more states wart added. Within those 22 states, the extent of reporting varied widely, from nearly complete reporting (most of the fire departments In a given state providing data on most or all of their incidents! to minimal reporting (only a few Jurisdictions In a given state reporting partial data). The data base queried to obtain transformer fire losses for Reference c-i covered the entire year of 1977 plus one or two quarters from 1978. The data ease contained 530,040 records (fire incidents), of whlcn 209 fire Incidents Involved Indoor transformers. The records extracted were tnose that were coded, per Reference C-6, "Equipment Involved In Ignition Code 42 - Transformer and Associated Overcurrent or Disconnect Equipment." Of these 209 fires, 200 were coded as confined to the room of origin or materia) of origin. Table 3-18 of Reference C-l gives four Incidents leaning to one injury out of 200 transformer fires that were confined to the room of origin, and Table 3-19 gives zero fatalities for the same 200 fires. For fires extending beyond the room of origin, the data given In Reference C-l are deeamd to be significantly less relevant to the two case studies of this report (that Is, a transformer Installed in an office-type building) than the data from CFIRS which <s discussed below. This (s because the data for fires propagating beyond tne room of origin were extracted from the fires in manufacturing properties which typically have fire nazard and occupancy characteristics much different from office buildings. C.2.2 OATA FROM CFIRS The CFIRS data are collected In a manner similar to NFIRS. The CFIRS data are actually transferred to the NFIRS to be combined with the records from other states. This may raise a question of dependency between the two sources. Since the data from these sources are collected from different years, they are Judged to be Independent, There are 165,470 records (fire Incidents) In the CFIRS for the year 1982. Out of the 165,470 fires, 500 Involved "transformers and associated overcurrent and disconnect equipment" as the source of heat causing Ignition (Reference C-101. These 500 Incidents Included all types of locations Including exterior and Interior fires. No Injuries or fatalities were reported for any of these fires. The Investigation of transformer fire Incidents was not pursued any further because It was Judged that better evidence could be collected for the conditions specific to the case studies Since our two case studies Involve transformer rooms in an office building environment, the 1982 CFIRS records were searched for similar conditions. Computar searches were conducted for the following "fixed property use classification" (Reference C-2) codes: Code Description 441 Hotels, Inns, and Lodges Used Year-Round S91 General Business Office G019MQ33184 C-4 MONS 020699 and for "fire damage" codes: Code Description 1 Confined to Material of Origin 2 Confined to Area of Origin 3 Confined to floor of Origin 4 Sulldlng of Origin 5 Spread Beyond Ares Fire diMge codes 1 and 2 are Interpreted as fires confined to the room of origin; damage codes 3, 4, and 5 are interpreted as fires spreading beyond tne room of origin; and the numbers of injury and fatality fires are taoulated. Table C-l summarizes the findings from CP IAs and NFIAs. it should be noted that all fatality fires Involved only one death and almost all Injury ftres Involved only one Injury. C-J STATISTICAL ANALYSIS The approaches for statistical analysis an described In Section C-l. for the case of fires confined to the room of origin, the two pieces of evidence are combined using the approach delineated In Reference C-6. for the fires spreading beyond the room of origin, Bayes' theorem is used In Its general fort. C-J.I INJURIES FROM FIRES COSFIMED TO THE ROOM OF ORIGIN Tha conditional frequency of an Injury, given a fire that Is confined to its room of origin, fj au, 1 evaluated using the method given in Reference C-6 ji described In Seifton C.l. The two pieces of evidence are 4 Injury fires In 200 Incidents (NFIR5 date) and 16 injuries In 1,048 Incidents (CFIRS data). The two sources of data are treated as two experts estimating Tj rk nth their point estimates being ' The likelihood function of each expart Is taken to be normal with means equal to these point estimates. Tha variances of their likelihood are taken to be equal to an astlmate of the variance based on the collected W9H033ia4 HONS 02070 TABLE C-l. STATISTICS OH INJURY ANO FATALITY FIRES Parameter Fire Spread Limits Fire Confined to Room of Origin From Reference C-L From CFIRS Fire Spread Jeycrrj Roost of Origin : ``zm CFIRS) Number of Injury Fires Number of Fatality Fires Total Number of Fires 4 0 200 16 l 1,048 10 1 114 0017M033184 C-6 HONS 020 701 dti. A" dstiisata for Che variance can be r/n* where V* is the number 0f Injury fires and "n* ft the total number of fires, it should be noted that Reference C-ll shows that for strong evidence, the posterior distribution of the Bayesian updating Is approximately normal with mean r/n and variance r/n2. Thus, the two variances are a,Z" var 1.00 x 10*4 dj,2* Var 1.46 x 10`5 1,048* For the prior distribution, a nonlnfomatlve normal distribution can be used to eliminate Its Influence on the posterior distribution. The variance of a nonlnformatlve prior can be viewed as very large compared to oi? *nd 0 ? 2 such that eg?* >. The variance of the posterior normal distribution can be computed as The relative weights of the two pieces of evidence wi and w? are needed to estimate the mean of the posterior distribution ,,2 *1 --2 0.127 al2 (C.14) 2 w2 2 - 0.873 2 (C.15I The mean of the posterior distribution can bt written as fl,W,S0 " "l fl,I,RM * *2 f2,I,RM ' 0,0159 (C'16i vlth this mean and variance, all the other parameters of the normal distribution can be found. The characteristic values of fj^H are: 5th Percentile: 0.0100 Median: 0.0159 3030(1072484 7 020702 HONS 95th Percentile: 0.0217 Mean: 0.0159 C.3.2 FATALITIES FROM FIRES CONFINED TO THE ROOM OF ORIGIN The conditional frequency of a fatality, given a fire confined to its room of origin, fp RH> is evaluated using the sane method as for fI,RM* Tl1t ovidente from NFIRS 1* no fatality events in zoo fires jnc from CFIRS is I fatality fire in 1,048 fires. If r/n and r/n2 are used for estimating the mean and variance for NFIRS data, zeros will oe obtained, which Is contrary to our expectation. Therefore. Equations (C.8) through (C.12) cannot be used directly. However, -e ran try different assumptions about the NFIRS data such as 1 fatality fi-e m 200 fires or 1 fatality fire In 1,000 fires and see how sensitive tne final results are to these assinptlons. A reasonable distribution for fp mt based on these results can be chosen. Let us begin by assuming that a fatality fire could have been observed n NFIRS If one more observation took place. That Is, we modify the evidence to be 1 fatality fire In 200 Incidents from the first source dnd 1 fatality fire In 1,048 Incidents from the CFIRS. For this evidence, the mean and variance of tne first source are fl,P.RN 1 0.005 TOT o.2 l_- 2.5 a 10*5 1 TUlr and of the second source are f2,F,HH 1__ 9.5 x 10`4 1,048 ' rW" 9.1 x 10 -7 The mean end variance of the posterior distribution using Equations iC.ai through (C.12> are obtained as 1.1 * 10*3 and 8.8 x 10**. respectively. Let us nou assume that the fatality fires In NFIRS occur at a freaue-c/ much loss than 1 In 1,000 fires. This means that the likelihood of observing more than 1,000 fires without any fatalities is signifies: Let us assume that the mean of tne first likelihood Is fl.F.RM ' I- * I*3 001911031184 C-8 HONS 020703 je keep the variance the sane; that is. Oj2* 2.5 * 10*5 because it ripresents our confidence in the first source of data and it ts tne best 200 observations can provide. Using the same CFIR5 data, the mean and variance of the posterior becomes 9.6 x 10'* and ' 8.8 * 10*'. Similarly, If we assume fi r m 5.0 x 10** and the (ana variance, the mean and variance or the posterior becomes 9.4 x 10** and 8.8 x IQ*'. TjO notes are In order here. First, It can be shown that the Influence of a prior distribution with wide uncertainties is very small and can be ignored In the computations. Second, the nonael distributions given above cover negative numbers as well as numbers above tne unity. The probability of a number being negative Is 0.15 or less for the abova distributions and the probability of being greater than unity Is insignificant. To circumvent these problami, the final distribution wilt pe truncated and renormalized between zero and one. This Is judged to be of little slgnlflcsnce to the flnel results because the bulls of the uncertainties Is not sensitive to these approximations. The above three exercises show that the final distributions are rather insensitive to the choice of fi y jw and are dictated primarily by tna vidence from CFIRS. Thus, a njnial distribution Is chosen with a naan and variance of 1.0 x 10** and 8.8 x 10*'. respectively. To truncate and renormalize this distribution. It Is first discretized. Table C-2 gives this distribution and shows that the probability of negative values is 0.1055 and the largest value is less than 1. To truncate, the first four values are deleted, and to renormalize, each remaining probability is divided by 1,0 - 0.1055 0.6945, The resulting distribution Is given by Table CO. The mean value of this distribution Is 1.2 x 10*3. C.3.3 INJURIES FROM FIRES PROPAGATING 8CYONO THE ROOM OF ORIGIN The conditional frequency 0f (n injupy given a fire that has propagated beyond the room of origin, fi in, Is evaluated using 8yes' theorem (Equation C.l). The evldence'fs (r)10 Injury fires In (na)U4 fires that propagated beyond the room of origin. Since the number of injury fires is greater than zero, the influence of the prior distribution on the posterior distribution ctn be nlnlmlzed by choosing a nonlnformatlve prior; l.e., rn ro 0. Thus, from Equations (C.41 and (C.5). It *s found that the parameters of the posterior distribution for fj oq becomes r 10 and n 1X4, Table C-4 gfves a discretization of tne beta distribution with these parameters. Its characteristics values are; 5th Percentile; 4.9 x 10*2 Median; 8.5 x 10*2 95th Percentile: 1.3 x 10*1 'Indian: a.8 x 10*2 -1*013184 HONS 02070-1 TABLE C-2. OISCRETIZED NORMAL DISTRIBUTION WITH MEAN - 1.0 x TO'* and VARIANCE - 8.8 x 10-7 Value Probabl11t/ Cuwlatl ve Probaoi 1 i ty -2.0-3 -l.S-3 -9.9-4 -5.0-4 3.0-6 5.0-4 1.0-3 1.5-3 2.0-3 2.5-3 3.0-3 3.5-3 4.0-3 0.0030 0.0091 0.0278 0.0656 0.1210 0.1750 0.1970 0.1750 0.1210 0.0656 0.0278 0.0091 0.0030 0.0030 0.0121 0.0399 0.1055 0.2265 0.4015 0.S98S 0.7735 O.B945 0.9601 0.9879 0.9970 1.0000 ! NOTE: Exponential notation Is Indicated In abbreviated Torn; l.e., -2.0-3 -2.0 x 10-J. C-10 0017H033184 HONS 020705 TABLE C-3. PROBABILITY distribution for Ff.rm fF,HH 3.0-6 5.0-4 1.0-3 1.5-3 2.0-3 2.5-3 3.0-3 3.5-3 4.0-3 Probability emulative Probability 0.13S 0.195 0.220 0.196 0.135 0.073 0.031 0.010 0.004 0.135 0.331 0.551 0.747 0.882 0.9SS 0.986 0.996 1.000 NOTES: 1, Mam 1.2 x 10*3. 2. Exponential notation Is Indicated in abbreviated fora; 1.*., 3.0-6 3.0 x 10-6. C-U 0017M033184 HONS 020706 TABLE C-4, A OISCRETIZEO DISTRIBUTION for fI>B0 fi,eo Probability Cumulative Probability 3.3-2 4.44-2 5.29-2 5.87-2 6.33-2 6.72-2 7.08-2 7.42-2 7.74-2 8.22-2 B.88-2 9.40-2 9.78-2 1.02-1 1.07-1 1.12-1 1.19-1 1.28-1 1.44-1 1,71-1 1.04-2 3.96-2 5.21-2 4.79-2 5.24-2 4.80-2 5.16-2 4.81-2 4.93-2 9.89-2 1.00-1 4.93-2 5.08-2 5.05-2 5.21-2 4.99-2 4.87-2 4.99-2 4.03-2 1.0-2 0.010 0.050 0.102 0.150 0.202 0.250 0.302 0.350 0.399 0.498 0.598 0.648 0.698 0.749 0.801 0.851 0.900 0.950 0.990 1.000 NOTE: Exponential notation Is indicated In abbreviated fora; I.*., 3.3-2 - 3.3 x 10*2. 0017H072484 MONS 020707 I C.3.4 FATALITIES FROM FIRES PROPAGATING BEYOND THE ROOM OF ORIGIN The unconditional frequency of a fatality given a fire that ha* propagated beyond the room of origin, fc gg, is evaluated In this section using the sane method as described In Section c.3.3, The evidence Is 1 fatality fire in 114 fires that propagated beyond tne room of origin. The parameters of the posterior oeta distribution are then r I and n ll. A discretization of this distribution Is given in Toole C-S. Its characteristic values are; 5th Percentile: 4.5 * 10'4 Median: 1.1 i 10*3 95tn Percentile: 2.6 * 10-2 Mean: 8.8 x IQ*3 C.4 REFERENCES C'l. levlnthal, 0., "Application of Decision Analysis to a Regulatory Problen: Fire Safety Standards for liquid Insulated Transformers," NBS-GCR-80-198. Center for Fire Research. National Bureau of Standards, December 1979, C-2. "California Fire incident Reporting System," Annual Report, California State Fire Marshal, 1982. CO. Apostolakls, G., S. Kaplan, 8. J. Garrick, and R. J. Duphlly. "Data Specialization for Plant*Specif1c Risk Studies." Nuclear Engineering and Design. 56. pp. 321-329, 1980. C-4. Apostolakls, 6., "Data Analysis In Risk Assessment," Nuclear Engineering end Otslgn. 71. PP 375-381, 1982. C-5, Winkler, R. L., and W. 1. Hays, Statistics. Probability Inference and decision. Holt, Relnhtrt and umston, 1975. C-t. Moil eh. A,, end 9. Apostolakls, `Models For the use of Expert Opinions," Society for Risk Analysis, workshop on Lw-Probablllty/Hlgh-Consequence Risk Analysis, Arlington. Virginia, June 1982. C-7. "Appendices to Fire in the United States," Second Edition. Foderel Emergency Management Agency, U.S. Fire Administration, Washington, O.C.. July 1982. C-8. N.F.P.A. No. 901, "Uniform Coding for Fire Protection." 1976, National Fire Protection Association, Boston, Massachusetts. C-9- Letter from A. 1. Comberg of Firepro, Inc., to M. Kazarians of Pickard, Lowe a Garrick, Inc., February 7, 1984. J0I9MO4DJ84 C-13 MONS 020708 TABLE C-S, A DISCRETIZATION OF THE DISTRIBUTION FOR fp>80 fF,B0 Probability Cuaulative 1 Probability 1.00-4 3.49-4 7.98-4 1.2S-3 1.75*3 2.30-3 2.90-3 3.5S-3 4.20-3 4.93-3 6.17-3 8.10-3 9.93-3 1.14-2 1.32-2 1.54-2 1.84-2 2.29-2 3.14-2 4.64-2 1.12-2 4.37-2 S.20-2 3.95-2 5.60-2 4.39-2 5.75-2 4.58-2 4.97-2 5.22-2 9.64-2 1.00-1 5.20-2 5.01-2 4.88-2 5.18-2 4.95-2 4.99-2 3.99-2 1.00-2 0.011 0.055 0.102 0.146 0.202 0.246 0.304 0.350 0.399 0.452 0.548 0.648 0.700 0.750 0.799 0.851 0.900 0.950 0.990 1.000 1 j NOTE: Exponential notation Is Indicated In abbreviated form: l.e., 1.00-4 1 x 10-*. 001711072684 C-14 MONS 020709 C-10. letter from J. Stork of California State Fire Marshal to M. Kazarians of Pickard, Lowe end Garrick, Inc., doted August 23, 1983. C-ll. Llndley, 0. introduction to Propaolllty and Statistics from Bayesian Viewpoint: rart t, inference. Combndqt university Press, Ly/(L 0lW033ia4 C-15 MOWS 020710 APPENDIX 0 ROOM PRESSURE RISE FROM MINERAL OIL TRANSFORMER RUPTURE MOWS 020711 APPENOIX 0 BOOK PRESSURE RISE PROM MINERAL OIL TRANSFORMER RUPTURE in this appenal*, room pressure rise as a consequence of mineral oil transformer rupture is calculated. The pressure rise is from the eomOustion of mineral oil vapors. For a mixture of fuel and air to star* burning, a source of Ignition must exist and the fuel concentration must oe within the flamablllty limits. That fs. the flame cannot propagate if the concentration Is too low, and there will not be enough oxygen if the concentration Is too high. A table of flaenabillty llmfts for several types of fuel Is given In Reference 0-1. The lower limits typically range from 1.2t to 12.55 by volume and the upper limits range from 7.8t to 761. Two inodes of combustion m#y take place, depending on the flame velocity detonation (supersonic velocity! and deflagration (subsonic velocity). The detonation limits (in terms of concentration I for a vapor that can detonate are always within the deflagration limits, which are the flammability limits. In this study, because of lack of sufficient information It Is assumed that mineral oil vapor can detonate. Thus, in a mineral oil transformer rupture, either mode of combustion may occur. Upon rupture, mineral oil vapor would leavt the pool of fluid and would expand In the air. As it expands. It would first reach the upper flammability limit, and if deflagration does not take place, it would expand to the upper detonation limit (If It exists). The process can continue and the air and mlnaral oil vapor mixture would reach the lower detonation and deflagration limits in the absence of an Ignition source. In this appendix. It Is assumed that glvan a rupture, one of the two modes of combustion takes place. The pressure rise in the room depends on the type of explosion (detonation or deflagration). In this appendix, the relative likelihood of the two modes Is not evaluated. 5.1 DEFLAGRATION In a deflagration, the flame front propagates at subsonic speeds, and the pressure equalizes throughout the enclosure where combustion Is taking place. As will be shown below, a small amount of mlnaral oil vapor is generated In a transformer rupture. Therefore, the uniform mixture of the vapor with air Is below tha flanublllty limits. However, deflagration may occur In a smaller volume around the transformer as the vapor cloud Is axpandlng to the limits of the enclosure. Deflagration generates a pressure wave. This wave attenuates as It propagates toward the roam boundaries. Uhen it reaches the boundaries, the room it approximately at uniform pressure. Thus, an adiabatic, constant volume model (Reference 0-2), as opposed to a dynamic model, can be used for calculating the pressure rise In the room. Using the adiabatic constant voluma modal, the amount of mineral oil vapor is computed from an estimate of the energy of the arc, the cause for the rupture. All of the vapor Is assumed to burn instantaneously and the heat generated is completely absorbed by the air within the room. -037:1080684 D-l 0207 U mo ns The pressure is then determined using the ideal gas law. The details of the moael and the assumptions are given below. The thermal properties and other constants used in the computations and the results are listed In Tables 0-1 and 0-2, respectively. D.1.1 THE ARC ENERGY Reference D-8 presents the results of some experiments in which transformer tanks are ruptured by electrical arcs. Transformers were used in the experiments with ratings in the range of 10 kVA to 100 k'/A and applied arc energies ranging from 10 kJ to 80 kJ. The minimum arc energy needed for rupture increases with the transformer rating. The transformer rating considered in their case studies is 225 kVA. A range of arc energies from 10 kJ to 1,000 kJ is used in this study. Based on the experiments described In Reference 0-8, it is believed that the minimum energy for Rupturing the transformer under consideration is well above 80 kJ. D.1.2 THE AMOUNT OF VAPOR GENERATED It Is assumed that all of the arc energy is available for heating the mineral oil from Its initial temperature to its boiling point and evaporating it. This is a conservative assumption because the energy forrupturing the tank and displacing the rest of the fluid is Ignored. The initial temperature of the bulk oil Is assumed to be 100C. This is somewhat conservative because the normal operating temperature is in the range of 6Q#C to 75C (Reference D-9). Under abnormal high load or poor ventilation conditions, this temperature can be above 100C up to 150#C. The specific heat of the oil (in liquid form) is about 1.5 kJ/kg #K (Reference D-3) and the heat of vaporization Is 610 kJ/kg (Reference 0-4). The boiling point Is 360#C (Reference D-5). To obtain the mass of the vapor generated, the following heat balance equation Is used E - M Cp(Tg - Tq ) + MH where E arc energy (kJ). M mass of the vapor (kg). Cp specific heat (kJ/kg #K). Tg boiling point (K). Tq * initial temperature (#K). H * heat of vaporization (kJ/kg). 0007M080684 0-2 HONS 020713 TABLE 0-1. THERMAL PROPERTIES OF MINERAL OIL AMO OTHER CONSTANTS USED IN THE CALCULATION Parameter Value Reference Specific Heat Liquid (kJ/kg K) Heat of Vaporization (kJ/kg) Flash Point (C) Heat of Combustion (kJ/kg) Initial Oil Temperature (C) Room Oimension (ft x ft x ft) Specific Heat of Air at Constant Volume (kJ/kmole *K) Ratio of Specific Heats of Air at 1 ATM, y 1.5 610 145 4.7 x 104 100 20 x 20 x 8 21.27 1.4 0-3 ; D-4 D-5 - 0-6 See Section 0.1.2 | i ! ! Assumed . 0-2 ! 0-7 ! I -OC8M080284 0-3 MONS 0207 0 0 0 'I TABLE D-2. CALCULATION OF PRESSURE RISE USING THE ADIABATIC CONSTANT VOLUME MODEL 1 0 0 400 009 O o *" . CM --. 03 ce r\| CM aO -- CD O O-- lT r <M o ao o 'O . ^- OCS' <7> O CO . -- CM cm vO O'"" -- o CO -- 5 O * 0 4/1 * ou*l CM -- S <M W> M O CM <M --4 CM 9- vs O - C- O O cm O M0 So 3 * o 00 01 nm en 9 wo cn ro wo ^ CO CM CM 3 V r*> O mo CO CM <ni co O O r- d^ < m So D-4 HONS 020715 It should be noted that since heat losses into the bulk of the oil are not considered properly, this equation is conservative; that is, it overestimates the vapor mass. Using the above values, we can solve this equation for M, the mass of the vapor. For example, for 200 kJ arc energy, 0.200 kg of vapor is generated. The result of this equation is conservative because the energy expended for superheating the vapor and heating the rest of the pool is ignored. 0.1.3 THE ENERGY GENERATED FROM COMBUSTION Upon transformer rupture, some of the oil and most of the vapor would be released. It is assumed that all the above computed vapor mass is released into the air. This ignores some condensation that may take place on colder surfaces and while the vapor is still submerged in the liquid. It Is also assumed that ignition would occur because the vapor temperature would be above Ignition temperatures Indicated In Reference D-5 for lubricating mineral oil, and the transformer winding may be red hot and exposed. The energy generated from the vapor combustion is the product of the heat of combustion and the vapor mass. The heat of combustion Is taken to be 4.7 x 10^kJ/kg (Reference 0-6). This result Is also conservative because it assumes that all of the vapor is consumed. For example, combustion of 0.200 kg of vapor can generate 9,400 kJ of energy. We can now compute the energy generated per kllomole of air. This Is simply the total energy generated divided by the total amount of air in the room. The room dimensions are 20 feet by 20 feet and 8 feet from the floor to the ceiling. The volume of the room is, therefore, 3,200 feet-* (9.06 x IQ4 liters). Assuming 1 atmosphere pressure and 25C temperature, the number of moles is 9.06 x 104 --ZO--x 273 273"+'25 = 3.71 x 103 moles For example. In the case of 200 kJ arc energy, which leads to 9,400 kJ of energy after vapor combustion, the energy per kilo mole is 2,541 kJ/k mole. In the above calculations, the presence of the combustion products is Ignored. It can be shown that their Impact Is insignificant. In the worst case, that Is, 1,000 kJ arc energy, 1 kilogram of vapor Is generated. Reference 0-10 Indicates that burning 1 gram of hydrocarbon leads to about 2 grams of COo. Thus, 2 kg or 45.5 moles of CO2 would be generated in this case. This constitutes 1.2% of the air in the room. Therefore, the presence of CO2 after the combustion for all arc energy levels under consideration can be Ignored. 0007M080984 D-5 020716 HONS 0.1.4 POST-BURK TEMPERATURE OF THE AIR To calculate the post-burn air temperature, It is assumed that all the combustion energy is absorbed by the air and the specific heat of air at constant volume Is needed. From Reference 0-2, it is concluded that the specific heat of the air ranges from 20.9 to 21.5 kJ/kmole SK for the temperature range of 298 to 400K. Since this variation Is small for temperatures below 400K, a constant specific heat at constant volume of 21.27 kJ/kmoleK is used. For temperatures above this level, the correlations for the internal energy of oxygen and nitrogen are used (Reference D-2). That is, U(T) =* 0.8 x [31317 + (37.46 - 8.314)T - 4559.3 inT] + 0.2 x [43388 + (42.27 - 8.314)7 - 6635.4 inT] (kJ/kmole) (0.2) where U(T) is the internal energy of air per unit kllomole at temperature T. The expressions within the brackets are the correlations for the internal energy of nitrogen and oxygen, respectively. It is assumed that the air is composed of 80% nitrogen and 20% oxygen. To compute the air temperature above 400K, the following expression is used: u U(T) - U(400) + 21.2 x (400 - 298) 0.8 x [31,317 + (37.46 - 8.314JT - 4,559.3 tn(T)] + 0.2 x [43,388 + (42.27 - 8.314)T - 6,635.4 in(T)] - 0.8 x [31,317 + (37.46 - 8.314) x 400 - 4,559.3 in (400)] - 0.2 x [43,388 + (42.27 - 8.314) x 400 - 6,635.4 in (400)] +21.2 x (400 - 298) (D.3) Thus u - 30.IT - 4,974.5 An(T) + 19,923.8 (D.4) where u is the combustion energy generated per kllomole of air as computed In Section 0.1.3. For example, for the case of 200 kJ arc energy, the InternaJ energy Is 2,541 kJ/kmole. The temperature rise using the constant specific heat Is 2,541/21.2 - 119.9*K 0007M080984 0-6 MONS 020717 and the absolute temperature is 119.9 + 298 = 417.99K which is above 400k. Thus, the above equation should be used for establishing the post-burn air temperature 2,541 = 30.IT - 4,974.5 k.n(T) + 19,923.8 (D.5) Using iterative computational methods, the post-burn temperature of 422K (149C) is obtained. D.1.5 POST-BURN AIR PRESSURE Knowing the post-burn temperature as well as the amount of air and the volume of the room, the post-burn pressure can be calculated using the Ideal gas law (0.6) For our case study, n Is reasonably approximated by 3.7 kmole, V is 90.6 cubic meters, and R is 8.314 kJ/kmole K. Thus, P - 0.3395T x^- (psla) (0.7) where T Is in degrees K and 6.89 is the conversion factor to express the pressure in terms of pounds per square inch. For example, for the case of 200 kJ arc energy, the temperature is 422#K, which would result in an absolute pressure of 20.8 psla, and the pressure Increase is 20.8 - 14.7 * 6.1 psl 0.1.6 PRESSURE VENTING The pressure rise In a deflagration Is sufficiently slow to permit relieving of the pressure through openings in the transformer room. These openings for the two transformer rooms considered in this study are ventilation openings toward the outside air. The following equation from Reference D-ll can be used to establish whether sufficient vent area is available. 0007M080884 0-7 MONS 020718 where Av * necessary vent area (ft2). C = constant, depends on the burning vapor. Ll 3 smallest dimension of the transformer room (ft). 1-2 3 second smallest dimension of the transformer room. P 3 maximum pressure that the room boundaries can withstand (psi). For the type of gases that may be generated from mineral oil vaporization, Reference D-ll suggests C 3 2.6 (in English units for a 20 ft x 20 ft x 8 ft room, we obtain the relationship 416 (ft2) rr Thus, For case l, the room boundaries may withstand pressures up to 1 psig.' To prevent the burning gases from reaching this level the total vent area must be 416 ft2. For case 2, the room boundaries may withstand . pressures up to 7 psig. For this case, the vent area must be 157 ft2. For both cases, the required vent area is signflcantly larger than reasonably sized ventilation openings. Thus, for arc energy levels that lead to room pressures above l and 7 psig, the room boundaries for cases 1 and 2, respectively, are very likely to fall. D.1.7 RESULTS Using the method outlined above, the pressure Increase In the room is calculated for 11 levels of arc energy. Table D-2 shows the results. As can be seen, above 80 kJ arc energy the differential pressure is above 2 psi. The walls of the room will experience this pressure rise for only a short period because the ventilation openings will rapidly relieve the pressure. The momentary pressure rise Is deemed to be uniform throughout the room, because the deflagration process is envisioned as being an expanding pressurized gas volume in the room. 0.2 DETONATION In a detonation, the-flame front moves at supersonic velocity and there Is significant pressure drop across the flame front. The maximum pressure that can be reached at the flame front Is several times (it can be as much as seven times) the Initial pressure (Reference D-12). The Impact of the shock wave on the room boundaries Is a complex process, much different from the Impact of the pressure waves from deflagration. The forces exerted by the shock wave on the walls are dependent on the shock wave characteristics and wall shape and angle. Simple relations, such^as PV 3 constant (P Is pressure and V Is volume), cannot be used to represent the physical processes. 0007M08Q984 D-8 MONS 020719 The impact of detonation on the room boundaries is significantly more violent than the impact of the pressure waves from deflagration. It is judged that (for the amounts of vapor generated above 80 kJ of arc energy) a detonation of the mineral oil vapor would almost surely lead to an equivalent static pressure on the walls greater than 1/2 atmospheric pressure. 0.3 REFERENCES 0-1. Bartknecht, W.t Explosions, Course Prevention Protection, Springer-Verlag, (lew fork, 1981. D-2. Campbell, A. S., Thermodynamic Analysis of Combustion Engines, John Wiley and Sons, New York, 1^9. 0-3. Gann, R. G., "Development of Flammability Criteria for Transformer Olelectric Fluids", National Bureau of Standards, Center for Fire Research, MBS IR 80-1992, March 1980. 0-4. Heard, 0. B., "Study of Fire Extinguishment of a Replacement Fluid for Use in Transformers In Lieu of Askarel," Factory Mutual Research Corporation, FRA/0RD-82/12, April 1982. 0-5. "Fire Hazard Properties of Flammable Liquids, Gases, Volatile Solids, 1977," National Fire Protection Association, Inc., NFPA 325M, 1977. 0-6. Hemstreet, R. A., "Flammability Tests of Askarel Replacement Transformer Fluids," Factory Mutual Research Corporation, D0T-TSC-1703, August 1978. D-7. Perry, R. H.. C. H. Chilton, Chemical Engineering Handbook, 5th Edition, McGraw Hill, 1971! D-8. Barkan, P., B. L. Damsky, L. F. Ettllnger, and E. J. Kotski, "Overpressure Phenomena In Distribution Transformers with Low Impedance Faults Experiment and Theory," IEEE Transactions on Power Apparatus and Systems, Vol. PAS-95, No. 1, January/ February 1976. D-9. Caldwell, J., and J. Olmsted, "The Real World of Transformers Risk Evaluation," American Risk Management Loss Control Conference, Louisville, Kentucky, April 25, 1979. 0-10. Tewarson, A., J. L. Lee, and R. F. Plon, "Fire Behavior of Transformer Dielectric Insulating Fluids," Factory Mutual Research Corporation, DOT-TSC-1703, September 1979. 0-11. "Explosion Venting," National Fire Protection Association, Inc., NFPA 68, 1978. D-12. Zabetakls, Michael G., "Flammability Characteristics of Combustible Gases and Vapors," Bulletin 627, Bureau of Mines, 1965. 0007M080984 D-9 MONS 020720 APPENDIX E ROOM BOUNDARY FAILURE FROM SEVERE HEAT FLUX HONS 020 721 APPENDIX E ROOM BOUNDARY FAILURE FROM SEVERE HEAT FLUX In this appendix, for the alternate case of case l. the chance that the room boundary is breached due to a fire caused by the ignition of leaked mineral oil is evaluated. The failure of the wall due to the radiative heat flux impinging on it is the most likely mode of failure. The convective element of heat transfer from the fire to the walls is judged to be less important than radiative heat flux. Convective heat flux is important when ceiling failure is considered. The ceiling in this room has a 3-hour rated concrete construction. Therefore, its failure is less likely than the walls which are 1-hour rated. For the thermal calculations, it is assumed that there is enough mineral oil for the fire to last until boundary failure and that the fire is not affected by the sprinklers or the fire brigade. The approach that was used in Appendix B to model fire propagation from an Askarel transformer rupture is also used here. First, the time to wall failure is established and the expression fRB-2 = *-fSPR + ^ " fsPR^ CXp ^^SPR^ exp ^'T*^TFB* is used to find the conditional frequency of boundary failure given the fire. The distributions for f$pR, r$pq, and as assessed in Appendix B are applicable here, t* is the mean time at which wall failure occurs. It is determined using a one-dimensional heat conduction code (Reference E-l). The leaked mineral oil is assumed to be contained by the curb on the floor. The surface area of the mineral oil Is the area of the curb minus the area of the transformer; l.e., 8 ft x 8.5 ft - 2 ft x 2.5 ft - 63.5 ft2 - 5.9 m2 The mass burning rate depends on the amount of air available, if it is assumed that there Is sufficient air, combustion Is controlled by the surface area; that Is, the mass burning rate Is equal to the surface area times the mass burning rate per unit area, which has been determined experimentally to equal 3.3 x 10*2 kg/m2 second (Reference -2). This Is the rate at which the mineral oil Is consumed due to combustion and the formation of unburned fumes, which occurs because the combustion efficiency Is not 100%, Reference E-2 estimates the heat of combustion of mineral oil to be 4.7 x 104 kJ/kg. Assuming 100% combustion efficiency, the heat generation rate for an area of 5.9 m2 is 9.3 x lO^kJ/second. The mass burning rate is 0.196 kg/second. At this rate, the 80 gallons of mineral oil can last for about 23 minutes. Reference E-3 estimates that approximately 1 cubic foot of air is needed for every 113 Btu of energy generated in a combustion. To sustain a mass Q030M072584 E-l MONS 020722 burning rate of 0.196 kg/second for a surface area controlled fire, the air supply needed is (9.3 x 103 kJ/sec\ /60 sec\ / Btu \ \ 113 Btu/ft3 / \ n,in / \1.055lkjj 4.68 x IQ3 ft3/min which is higher than the assumed capacity of 2,000 ft3/minute for the ventilation system. Therefore, the actual burning is ventilation controlled. It can be calculated using the assumed capacity of the ventilation system; i.e., Heat Generation Rate 113 Btu/ft3 x 2,000 ft3/min * 3.97 x 103 kJ/sec and Mass Burning Rate 3.97 x 10 kJ/sec a g 45 x 10 2 kg/sec 4.7 x 10 kJ/kg At this rate, 80 gallons of mineral oil can last for 53.3 minutes. Wall failure is assumed to occur when the temperature rise at the cool side of the wall reaches 140C. This is the definition used in the fire rating of structures (Reference E-3). The time to failure t* is determined by using the THEAT computer code (Reference E-l). The wall is assumed to be a 2-1nch thick concrete slab (Reference E-3 Indicates that 2-1nch thick concrete covers are 1-hour rated fire protection for beams, joists, and girders). On one side, a constant heat flux Is used as the boundary condition; on the other side, convective cooling is assumed. The constant heat flux is estimated using a conservative model. Because of the large surface area of the base, the flame Is expected to reach the celling. It is assumed that the flame Is In the shape of an 8-foot high cylinder with its horizontal cross sectional area equal to the area enclosed by the curb. Therefore, its radius is equal to V - -xI--f - - 4.65 ft The room boundary Is assumed to be a concentric cylinder with radius equal to 10 feet and height equal to 8 feet. Assuming that radiative heat flux Is emitted from the flame radially, the heat flux decreases linearly with distance. The heat flux at the room boundary is therefore equal to the total radiative power divided by the curved surface area of the outer cylinder; I.e., 3.97 x 103 kJ/sec x 0.4 2n x 10 ft x 8 ft 34 kW/m2 where 0.4 Is the fraction of total energy emitted as radiative heat flux (Reference E-4). 0030M072584 E-2 MQNS 020723 Using this constant heat flux, t* is found to be 31 minutes. It relates to t* by t* = Eot* (see Appendix B for a discussion of the parameter E0), where E0 is lognormally distributed with 5th and 95th percentiles equal to 0.8 and 4.0, respectively. Therefore, r* is lognormally distributed with parameters p * 4.02 and o 0.489. Its characteristic values are: 5th Percentile: 24.8 Minutes Median: 55.5 Minutes 95th Percentile: 124.0 Minutes Mean: 62.5 Minutes The conditional frequency of breaching the boundary, fRB-2 have the following characteristic values: fund to 5th Percentile: 9.7 x 10'5 Median: 5.4 x 10"^ 95th Percentile: 5.5 x 10"2 Mean: 1.3 x 10'2 E.l REFERENCES E-l. Slu, N. 0., THEAT Computer Code Users Manual, Pickard, Lowe and Garrick, Inc., PIG-0272, July 1983. E-2. Hemstreet, R. A., "FlammabilIty Tests of Askarel Replacement Fluids," Factory Mutual Research Corporation, RC78-T-42, August 1978. E-3. National Fire Protection Association, NFPA Fire Protection Handbook, 14th Edition, Boston, 1976. E-4. Slu, N. 0., "Probabilistic Models for the Behavior of Compartment Fires," NUREG/CR-2269, August 1981. QQ30M072584 E-3 HONS 020724 APPENDIX F FIRE PROPAGATION TO AN ADJACENT ROOM FROM AN ABOVE-LIQUID-LEVEL RUPTURE OF THE CASE 1 MINERAL OIL TRANSFORMER MONS 020725 APPENDIX F FIRE PROPAGATION TO AN ADJACENT ROOM FROM AN ABOVE-LIQUID-LEVEL RUPTURE OF THE CASE I MINERAL OIL TRANSFORMER The possibility of fire propagation to an adjacent room in the case of an above-liquid-level rupture of the Case 1 mineral oil transformer is analyzed in this appendix. Since the rupture point is above the liquid level, little fluid is splattered. If any fluid escapes the tank, it is anticipated to end up within the curb boundaries around the transformer. It is certain that the oil remaining in the tank would be In flames because of the hot windings and contact with the air. Some of the hot gases and smoke generated from this fire would escape to the outside of the building through the openings, and some of It would escape to the adjacent rooms through the walls breached by an explosion following the rupture. The hot gases would also engulf the sprinkler heads above the transformer, which would eventually open and pour water on the fire. The combustibles in the adjacent room would heat up primarily from the radiation of the hot gas layer and the flames. In Appendix E, it is found that In the case of a burning oil leak, the heat flux on the room walls is expected to be about 30 kW/m2. This can conservatively be taken as the heat flux impinging on the carpeting or other combustibles In the adjacent rooms. Using the semi-infinite slab model of Appendix 8, the time of Ignition t* can be calculated (see Appendix B for its derivation) from *>. `-vu (,) where q0 Is the impinging heat flux in W/m2. Thus, t* * 133 seconds To establish the conditional frequency of fire propagation, fp.4, the approach used In Appendix B can be followed. We can write fp-4 a [fSPR * exP (t*/t$Pr) (1-fSPR)] exP (-t*/tfb) Since t* is very short Irr this case, the likelihood of the fire brigade arriving Is smell and exp (-t*/tfb) is assumed to be unity. As discussed In Appendix B, the statistical uncertainty about t* is deemed to be overwhelmed by the state-of-knowledge uncertainties. The error factor, Eq. suggested In Reference F-l can be used here. It is lognormally distributed with 5th and 95th percentiles at 0.8 and 4.0, respectively. Multiplying this error factor by t*, the mean fire propagation time Is obtained t* t* E0 3030M060484 F-l MGNS 020726 This quantity Is also lognormally distributed and has the following characteristic values: 5th Percentile: 1.8 Minutes Median: 4.0 Minutes 95th Percentile: 8.9 Minutes Mean: 4.5 Minutes Using discretized probability distribution arithmetic (see Appendix G and Reference F-2), the distribution for fp.4 can be computed. Its characteristic values are: 5th Percentile: 0.30 Median: 0.62 95th Percentile: 0.88 Mean: 0.60 F.l REFERENCES F-l. Slu, N. 0., and G. Apostolakls, "Probabilistic Models for Cable Tray Fire," Reliability Engineering, Vol. 3, pp. 213-227, 1982. F-2. Kaplan, S., "On the Method of Discrete Probability Distributions In Risk and Reliability Calculations - Application to Seismic Risk Assessment," Risk Analysis, Vol. 1, p. 189, 1981. 0030M060484 020*72? mqns APPENDIX G UNCERTAINTY PROPAGATION USING DISCRETE PROBABILITY DISTRIBUTIONS QZQ123 MQNS APPENDIX G UNCERTAINTY PROPAGATION USING DISCRETE PROBABILITY DISTRIBUTIONS The contents of this appendix are based on Reference G-l. To propagate uncertainties through a complicated equation such as (see Elation (B.7) in Appendix 8) using analytical methods can often be an impossible task. One method for uncertainty propagation is using discrete probability distribution (OPO) and replacing the multiple integrations by consecutive multiplications and summations. All probability distributions can be approximated by OPDs. Let us use the notation <p^, Xj> for the ith element of a OPO for variable X, where X-j is the ith value of X and p.j is its probability. Also, the pj of every DPD must satisfy n Z p, = l i *I Let us use the notation (<pt, X^>} to represent the OPO for variable X. If X has a continuous probability distribution, Px(X), its DPO can be obtained by first choslng a set of Intervals in the range of all possible X and obtain the Xj and pi from 1-1 where X^.i and Xi are the boundaries of the ith interval. It should be noted that several of the frequencies used in this study are assessed in terms of DPO. 0030M072484 MONS 020729 As a simple example for uncertainty propagation, suppose the variables x, Y, and Z are related by Z *X +Y and the variables X and Y have DPDs of {<p}t X}>) (total of I elements) and (<qj, Yj>} (total of J elements). The 0P0 for Z can then be written as {<Piqj, Xf + Yj>> That is, the DPD of Z has I x J elements with probability r1j * PiQj and values Z-fj * X^ + Yj. As a simple numerical example, take the DPDs for X and Y to be X * {<0.1,-1><0.5,1><0.4,2>} Y - {<0.2,5x0.8,10>} then Z - {<0.02,4x0.1,6><0.08,7x0.08,9> x <0.4,ll><0.32,12>) In the general form, when Z - f (X, Y) where X,Y are Independent DPDs, X (<Pi,X-(>>, Y - {<qj, Yj>}, Z is the DPO z* {<nj,. zid>} where r1 j * PiQj. Zij f<Xi, Yj) and 2 rij " 1,0 all 1 and j In the case of a complicated equation, the number of variables of the function f can be Increased and the probability of each combination of values becomes the product of the appropriate probabilities. For example, the above equation for fp_i has four variables: fSPR- T > repo and Tfrn. All are assessed In terms of three-element DPDs (see Appendix B) except fori*, which has a lognormal distribution. C030M06Q484 G-2 HONS 020730 A discretization of that distribution into 20 (elements) yields for its third value 13 3 8.75 with pXi3 = 0.012. The first element of fePR is <0.10, 0.02>, the second element of tspr is <0.4, 15.0> and the second element of t^b is <0.4, 30.0>. Using these values, one of the elements of the DPD for fp_j becomes ' <Pl 1 ? ? fn 1 > 3 <0.10 x 0.012 x 0.4 x 0.4, [0.02 ^ 1322 ''' + (1 - 0.02) exp (- 8.75/15.0)] exp (- 8.75/30.0)> 3 <1.9 x 10-4, 0.42> Since these variables have 3, 20, 3, and 3-element OPDs (see Appendix 8), the DPD for fp_i would have 540 elements. The number of elements Is obviously very large if it has to be used In further uncertainty propagation. The elements of a large OPD can be condensed by taking the mean of subgroups of elements x< 7 E v pv P1 all k of the subgroup and p1 a pk all k of the subgroup The above example Involved a relatively complicated function of four variables. The terms within the function [such as exp (-t*/tpr) and exp (-tVtpij)] share identical variables (In this case t*). Because of this dependency, these terms could not be separately evaluated and combined by the appropriate relationship. However, in many cases, uncertainty propagation has to be done for Independent terms. For example, Q 3 X(Y+Z) The uncertainties can be propagated through this equation In two steps. In the first step, the DPD for Y+Z can be obtained by summing the OPOs for Y and Z. In the second step, the OPD for Q can be obtained from multiplying the DPOs for X by Y+Z (obtained in the first step). This process can be repeated for complicated equations but comprised of independent variables. G.l REFERENCES G-I. Kaplan, S., "On the Methods of Discrete Probability Distributions In Risk and Reliability Calculations - Application to Seismic Risk Assessment, "Risk Analysis, Vol. 1, p. 189, 1981. C030M060484 . G-3 MONS 020731 APPENDIX H FLAMMABILITY OF CHLORINATED ORGANIC VAPORS 020?32 APPENDIX H FLAMMABILITY OF CHLORINATEO ORGANIC VAPORS Examined In this appendix are the flammability characteristics of chlorinated organic vapors. The two principal components of Askarel, polychlorinated biphenyl (PCB) and trichlorobenzene (TC8), are such organics. Flame velocity In the vapor phase for a chemical is used in this appendix to compare the flammability of chlorinated organics with other flammable chemicals such as benzene and methane. Slower velocities indicate a less flammable material, and slower energy release, by implication, yields a less violent explosion. It should be noted that the rate of energy released by a burning gas Is directly related to its flame velocity. Although the physical properties of TCB are known (References H-L through H-3), very little is published on their flammability. Currently, data pertaining to the flammability of TCB consist of data on a flash point (99C) and reports indicating that it is nonflammable or suppresses flammability (References H-l, H-4, and H-5). Since these data are inconsistent, the following analysis is intended to determine the hazard level of TCB by determining its flame velocity and thus determine which data are correct. An analysis of the flammability properties of specific chemicals is given by Exner (Reference H-6). The analysis In Reference H-6 is based, to a large extent, on data developed by Gupta and Valeiras (Reference H-7) which was published at a date later than Reference H-6. Also, work by other specialists In the field (Reference H-8) gives some Insight Into how mixtures of nonflammable and flammable gases will behave. An algorithm for determining the flame velocity of a chlorinated organic vapor is: 1. Establish the weight fraction of chlorine in the vapor phase. 2. Establish the heat of combustion from Figure H-l. 3. Establish the flame velocity from a correlation equation with heat of combustion. The heat of combustion of all chlorinated organics can be established from Figure H-l. If their molecule is structured like benzene, the curve for aromatics should be used, and if their molecule is structured like methane, the curve for alkanes should be used. TCB is an aromatic compound. The weight fraction of chlorine in TCB is 0.59. From Figure H-l the heat of combustion of TCB is 3.4 Kcal/gram. From the heat of combustion, one can establish the flame velocity for an organic vapor. This is concluded from an Inspection of the equations for burning velocities given in Reference H-6. Table H-i summarizes the data 0035M080884 H-l MONS 020733 HEAT OF COMBUSTION, kcal/gram FIGURE H-l. CORRELATION BETWEEN HEAT OF COMBUSTION AND CHLORINE WEIGHT FRACTION OF MOLECULE H-2 MONS 020734 i . i l j r r j t i o o f ch1oromethane and methane. REPORTED IN REFERENCE H-6 SUMMARY OF FLAME VELOCITY RELATED INFORMATION 0) OW</> ra* .o toj roj<tj-rccwMocoio^ ajr^tr vncjocorsrsyD >u1 ua> ^ 'p E ra o S IZ ZJ a> o OP-**p--H`i--*OOOV rN*^CoO iC-MnC\oJ n-h \CLnf)OoNVWOT> o-~; ooo <\J rn f-4 1--4 i oo I oo OOO* N* OO^O^iv^HOncD- m^c-o<c\jpHi L--O< t<nnjao^oc>njLfnviLcn\icO I2 <<%U9- ^^^OO^&h--tCi0^O>OC^CSO>CCONSJiNpf^^Pi4pn^^C*pnD^mOO CgCNJCN/pHp*CS<M<\jCVlCS<V CO _ tr o o o O <S --I II II II It ** +* ** ** m^y^ p- guu"v. xtg mu ^Um Unu^ #-- O r- *-- "v. 1XpXviom UU V m rv _ _i i 1JIC o (xJUx Uz ux uo v0 VAX ooo X <-> X O o O a a> j a; cn} c*q (cQ c4 g<U ^ (g1 a> e +J V. 4-> 4- kL +> t. *cD> ^4oLNeJ <oi4u3> *C <c4SJ xweoVU *oC 3 rT3 JI x ooo o XZr rr CaMipO<*pX- xfa> ^xo Xoo oo xo x xo TABLE H - l. H-3 MONS 020735 given in Reference H-6. A linear or regression analysis (curve fitting) of the chloromethane data yields the correlation Ub = 5.34 (AHc) - 4.15 (H.i) where aHc * heat of combustion (Kcal/g). Ub 3 flame velocity (cm/sec). The number of data points is 7 and the correlation coefficient (R2) is 0.964. Although this linear equation is not consistent with our understanding of the phenomenology, it is an acceptable correlating form. Table H-2 gives a comparison of calculated velocities (using Equation H.I) and experimentally determined flame velocities for different chemicals. The experimental data are summarized in Table H-3. The deviations between measured and calculated flame velocities in Table H-2 implies that the relationship is different for each family of organics. This is consistent with the relationship found for the chlorine content and heat of combustion. Therefore, we modify the correlation in Equation H.I to specialize it to chlorinated benzenes. The ratio of measured and calculated velocities for chlorinated benzenes, per Table H-2, ranges from 0.8 to 0.9. The slope of the curve is thus modified by multiplying 0.9 and 5.34. The correlation equation for chlorinated benzenes becomes Ub = 4.81 (AHc) - 4.15 (H.2) This correlation gives a burning velocity for trichlorobenzene (C6H3CI3) of 12 cm/sec. This Is equivalent to the burning velocity of methylchlorlde. The prediction of a positive burning velocity for trichlorobenzene Implies that the reported flash point Is a valid number. Based upon this, we conclude that TCB Is flammable. However, TC8 has poorer flammability characteristics than chemicals such as benzene and methane. This becomes evident when the flame velocities and heats of combustion are compared. Finally, it should be noted that similar calculations for PCBs (e.g., Aroclor 1254) are not possible because almost no data are available for this homologous series. Reference H-9 reports combustion of an Askarel that was composed of 70% PCB and 30% TCB. Burning was possible only when considerable external heat (3 to 6 watts/cm2) was applied to a 4-inch diameter pool of Askarel. The flames were weak and the heat release rate from the Askarel was 3 to 6 Kcal/mln. It is also reported in Reference H-9 that self-sustained combustion of the Askarel was not possible. The combustion occurred only within the range of external heat fluxes noted above. In the same test, other transformer fluids such as mineral oil and 50-centlstoke dimethyl silicone established selfsustained combustion at lower external heat fluxes than 3 watts/cm2 and generated heat at much greater rates (above 15 Kcal/mln). 0Q35MO726B4 H-4 MONS 020736 TABLE H-2. COMPARISON OF FLAME VELOCITIES f hemlra1 Methane CH4 Flame Velocity (cm/sec) Measured Calculated -------------- :----------- 70 67 Chloromethanes CH3CI ch2ci2 CHC13 11 1.7 0 13 to 15 2.6 -0.14 to 1.3 Mixtures of Methane and Chloromethanes CH3CVCH4 K* * 0.48 CH3CVCH4 K* = 1 CH3CI/CH4 K* * 2 CCI4/CH4 K* 1 30 25 20 13 34 26 21 5.6 Benzene C66 Chlorobenzene C6H5C1 40 48 to 50 27 30 *K Is the molar ratio of chloromethane and methane. 0036MG72784 H-5 MONS 020737 TABLE H -3 . FLAMMABILITY PROPERTIES OF CHLORINATED ORGANICS SUMMARIZED FROM REFERENCES H-4 AND H-7 ui CO O' O' o ii O' CMk. OO'' W> ii CM C -- x o <) J 5 < x >, >. * u O k. u. o H-6 HONS 020738 TABLE H-3 (continued) H-7 020739 HONS H. 1 REFERENCES H-l. Dreisbach, R. R., Physical Properties of Chemical Compounds, American Chemical Society,""Washington, DC, 1955. H-2. MacKay, 0., A. 8obra, 0. W. Chan, and W. Y. Shin, "Vapor Pressure Correlations for Low Volatility Environmental Chemicals," Environmental Science and Technology, 16, 10, 1982, pp 645-649. H-3. Ohe, S., Computer Aided Data Book of Vapor Pressure. Data Book Publ i shi ng Company, Tokyo*, 1976. H-4. CRC, Applied Engineering Science, 2nd Edition. The Chemical Rubber Company, Cleveland, Ohio, 1973. H-5. CRC - 63rd, Handbook of Chemistry and Physics. The Chemical Rubber Company, Cleveland, Ohio, 1982. H-6. Exner, J., Detoxication of Hazardous Waste, Ann Arbor Science, 1982. H-7. Gupta, A. K., and H. A. Valeiras, "Burning Velocities of . Chlorinated-Methane-Air Mixtures," Combustion and Flame, 55, 1984, pp 245-254. H-S. Senkan, S. M., J. M. Robinson, and A. K. Gupta, "Sooting Limits of Chlorinated Hydrocarbon-Hethane-Air Premixed," Combustion and Flame, 49. 1983, pp 305-314. H-9. Hemstreet, R. A., "Flammability Tests of Askarel Replacement Transformer Fluids," Factory Mutual Research Corporation, RC78-T-42, August 1978. 0035M080884 H-8 MONS 020740 APPENDIX I FLASH POINT OF TRICHLOROBENZENE-AROCLOR MIXTURES MONS 020741 APPENDIX I FLASH POINT OF TRICHLOROBENZSNE-AROCLOR MIXTURES The flash points of different Askarels are determined in this appendix, Askarels are mixtures of Aroclors [polychlorinated biphenyls (PCB) and trichlorobenzene (TCB)]. The flash point is the liquid temperature at which the gas phase in equilibrium with the liquid contains sufficient fuel to burn in air upon exposure to an ignition source. The fuel is provided by vaporization of the liquid. At the flash point, the vapor phase concentration of the fuel may be calculated using vapor-liquid thermodynamics and the flash,point temperature. The thermodynamic relation between the gas and liquid phases may be written as (Reference 1-1) where y^ is the vapor phase mole fraction in equilibrium with the liquid phase. The liquid phase mole fraction, xj, equals 1 for a pure substance. The activity coefficient, y, equals 1 for an ideal solution or a pure substance. The total pressure, Pt, equals 760 mm Hg at standard conditions. The vapor pressure, Pj, of the liquid may be calculated using a vapor pressure correlation such as the Antoine equation (Reference 1-2) and the flash point temperature. The lowest reported flash point of pure 1,2,4 trichlorobenzene (TC8) is 99C (Reference 1-3). The vapor pressure of TCB has been correlated by Ohe (Reference 1-4) as logio(Pl) " 7.51512 - 2110.983/(T 242.429) (1.2) where P] Is given in mm Hg and T is in C. The vapor pressure of TCB at 99C may be estimated from the correlation of Ohe as 21.5 mm Hg. The calculated vapor phase mole fraction of TCB at the flash point is 0.02828 or 2.828 volume percent (mole percent). Table 1-1 presents flash point data on various fluids. From Reference 1-5, Aroclor 1254 does not have a flash point. For a mixture of TCB and Aroclor 1254, the flash point will occur when the vapor contains 2.828 volume percent TCB. Because both the TCB and the Aroclor 1254 are chlorinated aromatic compounds, the mixture of the two chemicals should behave In an Ideal manner. The vapor pressure of Aroclor 1254 may be correlated, using an Antoine equation, as 1 n{Pg) * 37.999 - 33256/(T + 733.83) ( 1.3) where ?z is given In mm Hg and T is in aC. The estimated vapor pressure of Aroclor 1254 Is 0.40 mm Hg at 120*C. The vapor pressure of TCB is two orders of magnitude greater than that of the Aroclor 1254. Thus, the contribution of the Aroclor 1254 to the total vapor pressure of the mixture may be neglected. l-l MONS 020742 TABLE 1-1. FIRE PROPERTIES OF VARIOUS FLUIDS Fluid Aroclor 1254 Flash Point, C None Reference (with page number) 1-5 ' 1 Aroclor 1260 None 1-5 1, 2, 4 Trichlorobenzene 99 110 1-2 (799) 1-6 (1043) 1, 2, 4, 5 Tetrachlorobenzene 155 1-6 (1012) Mineral- Oil 229 1-6 (835) Transformer Oil 149 1-6 (1038) 003BM072684 1-2 MONS 020743 The flash point of a mixture of TC8 and Aroclor 1254 is calculated by following algorithm: 1. Set the liquid phase mole fraction of TC3, Xj. 2. Set the vapor phase mole percent of TC3 equal to 2.828. 3. Compute the vapor pressure of TCB using the thermodynamic relation between the liquid and gas phases (Equation 1.1). 4. Compute the flash point temperature using the vapor pressure correlation for TCB (Equation 1.2). 5. Pick a new value of the liquid phase composition and repeat the above calculation. The results of these calculations are presented in Table 1-2 and plotted in Figure I-1. The effect of the Arocior 1254 is to dilute the TC8 and raise the flash point. As the weight percent of TCB is reduced from 70* to 30% (by weight) TCB, the flash point of the mixture changes from 104C to 120C. The flash point of a 50 weight percent mixture of Aroclor 1254 and TCB Is 110*0. 1.1 REFERENCES 1-1. Smith, J. M., and H. C. Van Ness, Introduction to Chemical Engineering Thermodynamics. Second Edition, Chapter 12, pp. 378-381, McGraw-Hill, New York, 1959. I-Z. Chemical Engineer's Handbook. R. M. Perry and C. H. Chilton, Editors, Firth Edition, chapter 3, McGraw-Hill Book Company, 1973 . 1-3. Kao, C. I., "Chlorinated 8enzenes," Chapter In Klrk-Othmer's Encyclopedia of Chemical Technology. Third Edition, Volume 5, John Wiley and Sons, New Vork, 1978. 1-4. Ohe, S., Computer Aided Data Book of Vapor Pressures. Data Book Publ1shing Company, Tokyo, 1976. 1-5. Monsanto Company, "The Aroclor Polychlorinated Polyphenyls," Technical Bulletin O-FF/1. 1-6. Sax, N. I., Dangerous Properties of Industrial Materials, Fifth Edition, Van Nostrand fteinhold Company, New York, 1979. 1-3 MQNS 020744 TA8LE 1-2. CALCULATED FLASH POINTS OF MIXTURES OF TCB ANO AROCLOR 1254 1 --1 Fraction TCB Weight Percent Flash Point, C 0.0 2.8 5.8 12.2 19.2 27.0 35.7 45.5 50.0 56.5 69.0 83.3 100 None 190 165 143 131 123 116 111 110 108 104 101 99 0038M072684 1-4 MOWS 020 745 1-5 MONS 020746 FIGURE 1 -1 . ESTIMATED FLASH POINTS OF MUTUIUS OF TC(i AtID AKOCI OK