1、CH-06-3-4 Predicting Flashover and Tenability Conditions in Train Fires-A CFD Approach J. Greg Sanchez, PE Member ASHRAE ABSTRACT This paper presents an approach using commercially available computational fluid dynamics (CFD) software combined with combustion and radiation models to predict the flow
2、, temperature, and smoke concentrationjelds in the event of afire-applied to a train fire scenario. This paper is not a validation but an outline ofkey aspects that need to be included in the modeling. The approach presented here can be used to predictfire growth and determine how long afire would t
3、ake to reach flashover. Although the results presented are for the first 120 seconds only, this short time is suficient to illustrate thefire dynamics predicted using the approach presented here. A comparison with available experimental data for gasoline spills was made. Although used to predict a t
4、rain fire, the approach can be used in any other space such as commercial buildings, hotel atria, and more. INTRODUCTION This paper presents an approach using commercially available computational fluid dynamics (CFD) software to predict fire spread and growth in the event of a train fire-with the us
5、e of combustion and radiation models. This is not a vali- dation paper in which extensive comparison is made with experimental data because very few detailed data are available for train fires. Nevertheless, a comparison with gasoline spill fires was made to serve as a sanity check. Currently, engin
6、eers are trying to learn how to best model fires so that their effects can be accounted for in modern designs in order to protect human life and reduce structural losses. This paper focuses on train fires but the approach can be applied to other areas. This paper is significant in that it is the fir
7、st time this approach has been presented with respect to design applica- tions and train fires. Because ofthe reality ofproject deadlines, designers cannot wait for generally established methods to be developed for each application. Design analysis has to be based on a combination of fundamental phy
8、sics, research, and assumptions. This paper follows in this time-honored tradi- tion. The approach presented accounts for radiation, mass loss, mass available to burn, flammability limits, surface ignition temperature, and fuel vapor auto-ignition temperature, thus the fuel vapors will be able to bu
9、rn when all conditions indi- cate that combustion is possible. The end result is a robust model that is able to predict fire spread and, if allowed to run long enough, can be used to predict whether or not it is possi- ble to reach flashover without prescribing any fire growth profile or heat releas
10、e rate. MOTIVATION FOR THIS PAPER The need to establish a means of predicting fire spread, including flashover, in trains (with complex geometries) has long been needed. ASHRAE Technical Committee 5.9 has been trying to promote research to develop a model with this capability. This is the main reaso
11、n why this paper has focused on train fires-to outline an approach to model the fire dynam- ics in a train fire scenario without prescribing the fire heat release rate. As part of the Memorial Tunnel fire tests (MHD/ FHA 1999), a CFD code was developed with the idea of help- ing designers model fire
12、s and design tunnel ventilation systems more realistically. However, the model developed relied on the user prescribing the fire heat release rate and did not account for combustion, so the need is still there. In Febru- ary 2000, the Society of Fire Protection Engineers (SFPE) held a meeting on fir
13、e protection research needs (SFPE 2000). The results of a survey conducted by SFPE revealed heat J. Greg Sanchez is principal mechanical engineer with New York City Transit. 02006 ASHRAE. 401 release rate, fire growth, and fully developed fires were ranked as the topics having the highest impact in
14、the fire protection industry. SFPE listed the benefits of studying these three topics as better prediction of fire protection performance, stronger underpinning of fire protection designs, better predictions of the effects of fire, and improved protection of people and property. The approach present
15、ed in this paper applies general combustion fundamentals, available in the general engineer- ing literature, to CFD in order to better resolve our engineering needs. The approach, though developed using one commer- cially available CFD software, can be applied to any other soft- ware. Thus, the appr
16、oach is not limited but open to any knowledgeable engineer who understands the physics of fires and CFD modeling theory. Several software packages are commercially available in the market; however, this paper does not endorse any one in particular. FLASHOVER Flashover is not fully understood by curr
17、ent practitioners, perhaps because its definition is very broad. Kennedy (2004) offers a review on how flashover is used in practice. Therefore, it is important to concur on what flashover may be. In this paper, flashover follows Karlsson and Quintiere?s (2000) defi- nition: flashover is determined
18、by detecting when the upper layer temperature reaches 600C. CURRENT FIRE MODELING APPROACHES A large series of experiments have been conducted to learn about fires in tunnels. Among them are the Memorial Fire test program (MHNFHA 1999) and the EUREKA (1 995) project. Some researchers have been learn
19、ing how to model fires and, in particular, what happens after flashover. Knowl- edge of pre-flashover is very important in order to determine means of egress, while knowledge of post-flashover is very important to determine structural integrity. Models have been created to represent fires using one-
20、 dimensional, two-dimensional, and three-dimensional math- ematical formulations. A one-dimensional model is computa- tionally very quick, but it has many assumptions built in that can be very easily violated, and this would lead to overhnder- predictions depending on the quality ofthe assumptions m
21、ade. Therefore, increasing the dimension of the algorithm should improve the predictions made, One of the most popular two- dimensional computer models is the zone model CFAST (Jones et al. 2004). COMPF-2 (Babrauskas 1979) is another computer model developed to estimate post-flashover condi- tions i
22、n a single compartment. Both of these models rely heavily on empirical formulations to predict the flow in the two-dimensional space, trying to represent a three-dimen- sional space. A full three-dimensional approach involves the use of CFD, also known asjeld models. This is the most accu- rate appr
23、oach because the flow field is solved in three-dimen- sional space. The drawback of increasing the spatial dimension is that time and cost increase exponentially as the number of the dimension increases. Current fire modeling approaches prescribe fires with either constant or time-varying volumetric
24、 heat release rates. However, most of them do not account for fundamental phys- ics present in fire dynamics. Such physics are: pyrolysis mass loss mass availability ignition temperatures for surfaces auto-ignition temperatures for fuel vapors flammability limits radiation optical mass density to es
25、timate visibility The approach presented in this paper is an adaptation of commercially available CFD software via user subroutines/ defined functions. Other programs, such as Fire Dynamics Simulator (FDS) (McGrattan 2004), account for some ofthese physics using large eddy simulation (LES) technique
26、s and probability density function (PDF) combustion formulation. However, FDS has two major limitations: on the one hand, the computer software is limited to structured Cartesian coordi- nates. This limits the ability to model complex, non-Cartesian geometries encountered in today?s designs. On the
27、other hand, in order to obtain reasonable results for large structural geom- etries using LES, the computational domain must be large enough so that the large eddies are computed and the smallest eddies modeled. This requires- that the Reynolds number be maintained high, otherwise the solution could
28、 yield errors. There is no doubt that near the fire zone the Reynolds number can be high, but away from the fire the Reynolds number would fall below the LES characteristics, which may lead to errors in the predictions. This especially can be the case in large spaces. A simplified rule of thumb to d
29、etermine the computational number of cells required is given by Equation 1 (Fluent 2004), where 1 is the eddy characteristic length scale (eddy size). As can be seen, the computational domain needs to be rather large, and the computational demand increases with it. For instance, if the domain were 1
30、82 x 12 x 12 m and the length scale were 1.5 m, the recommended number of cells needed in order to resolve the turbulence properly would be approximately 62,122,667. Using commercially available CFD software offers great advantages. Complex geometry, different turbulence models, and heat transfer an
31、d fluid dynamics boundary conditions can be modeled with great ease. The motion of trains, via sliding mesh techniques, and the swirls ofjet fans, among many other 402 ASHRAE Transactions: Symposia features, can be taken into account as well. These features could not be accounted for using FDS. The
32、approach presented in this paper applies fundamental combustion and fire protection engineering principles to predict the propagation of fire in a train cabin. Shahcheraghi et al. (2003) modeled a train fire by subdi- viding the interior of a train car into ten subvolumes, and based on a at2 heat re
33、lease rate growth profile, the number of volumes involved increased after a certain maximum heat release rate was achieved in the subvolumes. But this formu- lation is not based on physics; rather, it is based on engineering judgement. Shahcheraghi did not model radiation. Tien et. al. (2002) establ
34、ished that typical smoke layers are generally at temperatures ranging between 825C and 1225C. They also stated that a typical flame volumetric heat output is on the order of 1200 kW/m3. Sanchez (2003) presented the argument that the maxi- mum size of train fires is limited by air availability; furth
35、er- more, train fires should not be prescribed to follow a growth curve alone. Sanchez compared COMPF-2 results (a two- dimensional zone model, fully mixed, ventilation-controlled with radiation) with CFD results (three-dimensional, fully mixed, ventilation-controlled without radiation) to predict p
36、ost-flashover conditions. Karlsson and Quintiere (2000) use a general rule of thumb to determine flashover by detecting when the upper layer temperature reaches 600C. Following their rule of thumb, the peak heat release rate predicted by CFD was 6.25MW, about 1.5 MW higher than the COMPF-2 predictio
37、ns. But the problem at hand is more complex. In fire protec- tion engineering, we are dealing with fuel-rich diffusion flames and radiation. If we are trying to predict fire spread, radiation is essential. One way some practitioners have tackled the problem is by using PDF combustion and eddy dissip
38、ation concept formulation. In order to use PDF, a third program that develops the PDF tables is required. This technique works well in problems with a single felstream, as in premixed gas combustion. For real-life fires such as train fires, where there are a large number of types of fuels and the ma
39、terial properties are estimated as a weighted average, the use of PDF becomes a challenge. In general, CFD practitioners conduct fire modeling not accounting for oxygen availability or radiation. Some account for radiation by subtracting the radiation component from the total heat generation rate, c
40、laiming that this radiation energy is lost to the environment without affecting the solution. This approach is not valid in closed environments such as tunnels. Zones models (CFAST and COMPF-2) account for radiation in their formulation. Radiation is a key factor in heating the train cabin walls and
41、 the tunnel walls (which act as a heat source) in the vicinity of the train fire. Removing the radiation component would drop the temperature predictions and buoy- ancy, especially when some reduce it by up to 30%. Some practitioners model fires by applying the heat release rate to a fixed space, bu
42、t especially in train fires, oxygen becomes limited at one point or another. This will lead to overprediction of temperatures if the heat release rate is too large for the space allocated. In the 1970s, an approach was developed to estimate the fire heat release rate for train fires (USDOT 2002). Th
43、e meth- odology consisted of quantifying the heat of combustion of the materials and establishing a predetermined allowable time for burning (based on experience or observation of other fires); thus, the fire heat release rate was calculated by multiplying the total mass by the average heat of combu
44、stion and dividing it by the established time for burning. An example of such an approach can be found in the Subway Environment Simulation (SES) Computer Program handbook (USDOT 2002). However, all this was developed without considering whether or not there was enough ventilation to burn the fuel v
45、apors. Sanchez (2003) has made a first attempt to rectify this (for train fires) by arguing that in determining evacuation means, we not only need to determine how the fire grows but monitor whether or not there is enough oxygen to bum the fuel vapors as the fire grows. Sanchez also points out that
46、based on his own research, and in accordance with Peacock et al. (2002), passenger rail car materials and components that comply with the current fire tests and performance criteria cited by the Federal Railway Administration (FRA) exhibit fire growth rates below the medium t-squared level. This ass
47、essment was based on AMTRAK rail cars. However, if AMTRAK rail cars were found to be below medium t-squared level, this would be also valid (although on the conservative side) for mass transit systems such as New York and Hong Kong, which use fire-hardened cars with fire- retardant materials and in
48、some cases stainless steel seats. Sanchez concluded that mass transit systems complying with the FRA-cited test criteria, such as New York and Hong Kong, should be at a medium t-squared level or lower. Currently, New York City Transit considers medium t-squared fire growth rate appropriate for trans
49、ient analysis in stations. Additionally, Peacock et al. (1994) report that in the 1970s Prof. E.E. Smith from Ohio State University, based on a results comparison with real-scale fires, notably concluded that real-scale fire tests were neither reliable nor useful for evaluating individual materials used in rail transit systems. Real-scale tests were seen as mainly useful for checking results predicted using relevant bench-scale data. This leads to the conclusion that if a bench-scale sample test determines that the heat release rate given off by the sample (say 0.1 m x 0.1 m) is 1 O0 kW,
