Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
41 Information from numerous test reports has shown that road tunnel ï¬re tests are expensive, require precise measurements, and are difficult to reproduce. Numerical modeling can be repeated and allows for easy change of control parameters. Numerical modeling helps researchers to understand the physical processes and inï¬uences from design parameters. Validation of numerical models against the ï¬re tests helps to expand the models to new projects. However, validation of numerical ï¬re models is complicated and in most cases not successful, although some parts of the numerical model can be veriï¬ed against tests and ï¬eld measurements, considering that appropriate measurements have been made during the ï¬re tests. Theoretical models, especially computer-based models, can be very valuable in assisting tunnel ï¬re safety decision making. However, such models are also capable of being misleading. Nowadays, computer-based simulation models are widely used to calculate propagation of smoke and hot gases to assess the means to improve tunnel safety. They enable the simulation of the interaction of various ï¬re parameters. CFD software can model emergency ï¬re operating con- ditions in tunnels and predict the resulting contaminant con- centration levels. In areas of geometrical complexity, CFD is the appropriate tool for predicting 3D patterns of airï¬ow, temperature, and other ï¬ow variables, including concentration of species, which may vary with time and space. CFD was developed as a scientiï¬c tool for the investigation of aero- dynamic and thermodynamic processes. Nowadays, CFD software is considered as the design tool of choice for obtain- ing an optimum design, because experimental methods are costly, complex, and yield limited information. However, it requires in-depth knowledge of physical processes and numerical models and, preferably, testing experience from the numerical modeler. Many commercial CFD packages have been developed in recent years. A Fire Dynamics Simulator (FDS) is a CFD model of buoyancy-driven ï¬uid ï¬ow from a ï¬re. A separate code called Smokeview (OpenGL graphics program) is used to visualize data output from an FDS. These applications can also be conï¬gured to model pollutant levels outside the portals and around the exhaust stacks of tunnels. Both of these public domain programs are under active development and can be obtained from the National Institute of Standards and Technology (NIST). FDS uses the Large Eddy Simulation to solve the large scales of motion and model the small scales that are assumed to be universal. The Large Eddy Simulation results in a transient solution to the actual NavierâStockes equations valid for a low-speed (low Mach number) buoyancy- driven ï¬ow. A speciï¬c CFD model called SOLVENT was developed as part of the Memorial Tunnel Fire Ventilation Test Program for simulating road tunnel ï¬uid ï¬ow (ventilation), heat transfer, and smoke transport. SOLVENT can be applied to all venti- lation systems used in road tunnels, including those based on natural airï¬ow. SOLVENT has not been used for modeling ï¬xed ï¬re suppression systems and has some other limitations. Other CFD programs, both commercially available and in the public domain, have been used to model fire scenarios in road tunnels, the list of which is too numerous to include here. The most common and powerful for tunnel applications are ANSYS (Fluent CFD) and CFX veriï¬ed against testing results. Initially, the strengths and weaknesses of each program are investigated. Validation of the results against experimental data or another equivalent program is encouraged. Some pro- grams have limitations and are unable to model the required processes, including water-based ï¬re protection, moving traf- ï¬c (sliding mesh), wall roughness, and so forth. Validation is challenging for tunnel ï¬re modeling because the experimental data are far from absolute, given the complexity of the phys- ical process. Good experimental data are required. It becomes difficult to check the CFD model and results. In many cases it is up to the artistic, inventive ability of the engineer who created the model. The knowledge and experi- ence of the user becomes crucial. Users may employ different inputs in applying the same models or use different deter- ministic models to the same case, both of which produce dif- ferent results. Some studies showed signiï¬cant differences when the same user applied two different CFD-based models to the same case. The user must be knowledgeable about tunnel ï¬re science, as well as the modelâs limitations and applicable conditions. Assessment of models and their results are important and must be conducted by experienced people. It is important to establish a procedure for producing comprehensive, iterative assessment of ï¬re models. CHAPTER SEVEN ANALYTICAL FIRE MODELINGâLITERATURE REVIEW
ANALYTICAL (NUMERICAL) FIRE MODELING TECHNIQUE The CFD simulations of tunnel fires driven by buoyancy forces with signiï¬cant energy release require a solution of the NavierâStokes equations with appropriate boundary condi- tions. The physics of ï¬re modeling is complicated by many uncertainties. A number of assumptions need to be made for numerical modeling. The number of unknown variables and the calculation duration vary according to the hypotheses and assumptions made. There are at least eight equations to solve in 3D simulations, the unknown variables being Ï, p, T, ux, uy, uz, k, and ε, and seven equations to solve in two-dimensional (2D) simulations. Additional equations may be required to take into account the radiative heat transfer, the combustion process, or the heat transfer by conduction inside the walls. Different ways to model fire have been discussed. Airflow in tunnels is usually turbulent and the user has to make an assumption on the type of turbulence modeling to apply. One of the most common turbulent models is the k-ε model and its variations. The user is required to select turbulent length scale along with k and ε coefficients. There is insufficient information from a full-scale test to provide recommendations on the coeffi- cients to use. With this lack of information, the users apply the default numbers or follow some recommendations that may not be applicable to the road tunnel ï¬re test modeling. The better choice is to calculate the k and ε coefficients of the model based on length scale. In recent years the development of computers (i.e., speed and memory) has allowed for the development of larger and more complicated numerical models. However, considering the length of road tunnels, a model may require millions of cells. Even with todayâs computer power, the transient simu- lations of this size of a model may take months of computer time. Usually the user has to examine the grid by performing sensitivity analysis and ï¬nd the grid scale that will allow for reasonably accurate simulation results (32, 33). The better (ï¬ne) grid quality is usually in the ï¬re inï¬uence zone, whereas a coarser grid is created in other parts of the tunnel. A road tunnel fire is a combustion process with many unknowns, such as the substance that is burning, the method of the burning, and when it is burning. Some conservative assumptions can be made based on previous experience and full-scale ï¬re tests. Those assumptions may include ï¬re growth and decay rates, ignition location, and ï¬re size. However, how the vehicle burns may be one of the most complicated questions for the numerical modeler, especially if modeling a ï¬re as a chemical reaction, providing soot particles and combustion products at high temperatures. One approach is to represent a vehicle as a blocked volume, consider vehicle windows as inlet boundaries, and have hot combustion gases emerge as the ï¬ame temperature. 42 For combustion process modeling, the Eddy-Break-Up model is generally used. This method may be helpful if the information required from the simulation concerns the ï¬re zone. The limitation concerns the ï¬re load. It is not always possible to provide equivalence in terms of fuel consumption. In December 2005, NIST performed CFD modeling of the 1982 Caldecott Tunnel Fire (34). They used the FDS code and a combustion model. They concluded that fire consumed roughly 70% of the available oxygen, with an HRR of about 400 MW (1,365 MBtu/hr). However, the authors accepted that this was probably an overestimate because the model uses a simple âmixed is burntâ combustion model in combination with an empirical local extinction algorithm. The actual combustion processes are far more complicated and potentially much less efficient in the tunnel environment. The model overly predicted the combustion efficiency of the ï¬re, in which case most of the fuel was consumed somewhere in the tunnel or never consumed at all. Another possibility was that the observed ï¬ames at the east portal were a result of unsteady evaporation of the gasoline. It was assumed that the gasoline evaporated at a constant rate for 40 min (about 10 kg/s or 22 lb/s). However, had there been periods of greater evaporation this would explain the discrepancy between the observations and the simulation. The maximum predicted gas temperature near the ceiling was just below 1100°C (2012°F) with ablation and 1150°C (2102°F) without. This high temperature region was located roughly 40 to 120 m (131 to 394 ft) east of the overturned truck, near the ceiling and along the tunnel centerline. However, the tunnel inspec- tion report suggested that the maximum gas temperature could not have exceeded the melting temperature of copper [1065°C (1949°F)], because copper wiring in the upper wall light fixtures was not melted. The peak wall surface temperatures were approximately 950°C (1742°F). A more simpliï¬ed approach is to consider ï¬re as a volume source of energy at a given changing FHRR and a source of smoke and soot as a function of HRR. The last approach does not require combustion and chemical reaction modeling, but does require knowledge of the heat, smoke, and soot release rates. ⢠Fixed HRR in a volume: In this model, the ï¬re source is represented by an HRR ï¬xed inside a given volume. This value is not inï¬uenced by ventilation. This method leads to a more accurate energy distribution inside the tunnel volume, and experience has shown that it can lead to quite realistic temperatures except very near a ï¬re. ⢠Fixed heat ï¬ux through a horizontal surface: This tech- nique imposes a heat ï¬ux or a mass ï¬ow rate at a ï¬xed temperature to get the design HRR. The latter method leads to the mass flow rate, which is not always in agreement with the combustionâs production of burned gases (it must then be combined with a sink of mass). The volume energy distribution is not as good as in the previous case, and the results are not as reliable.
43 ⢠Fixed temperature in a volume: The advantage of this method is its ability to control the maximum temperature reached inside the fire. A disadvantage of this method is that the HRR will strongly depend on ventilation conditions. The ï¬xed HRR in a volume method is generally preferred because it is less expensive (central processing unit time) than modeling the combustion process and presents fewer disadvantages than the other methods. The design HRRs and ï¬re curves can be directly used with this method. It is always suggested that sensitivity studies be performed before ï¬nal design simulations. This may take more time and effort than the design, but leads to a better understanding of the results. There are many other boundary conditions that may affect the end results. ⢠Initial air movementâAir in the tunnel is never still. There is always some airï¬ow caused either by the piston effect of traffic, by normal tunnel ventilation, or by winds and other natural factors. For example, in uni-directional tunnels, the assumption is made that in a ï¬re emergency traffic will be trapped behind the ï¬re, whereas traffic downstream of the ï¬re will leave the tunnel. The depart- ing traffic will cause a residual piston effect, driving smoke and airflow in the direction of travel. Adverse winds may also have a signiï¬cant impact on the airï¬ow. Residual air movement, caused by approaching the ï¬re location traffic, may also drive the airï¬ow. ⢠Trapped traffic behind the fire incidenceâTrapped traffic creates a signiï¬cant obstruction to the airï¬ow. This results in substantial resistance to the airï¬ow and obstructions to the air jets developed by the tunnel venti- lation system. The last phenomena could be modeled by CFD; however, it would require complicated geometric modeling and many additional computational grid cells. ⢠Wall boundary conditionsâIt is usually considered that approximately 30% of the total heat is transferred to the tunnel walls by radiation and 70% by convective heat. There are radiation models available in commercial CFD products; however, radiation models are complicated and require the use of absorption coefficients and other empirical information. Often users consider convection portions only. Temperatures inside the ï¬re may reach 1300°C (2372°F). The heat transport is locally more radiative than convective. Calculations performed with- out radiative models have led to the prediction of higher temperatures, even a 100 m from the ï¬re zone. Several techniques can be used to take radiation into account. ⢠Radiative heat transfer model coupled with the conser- vation equation of energyâThis technique solves an additional equation. The greatest difficulty comes from inadequate knowledge of the radiative properties of smoke, which explains the need for additional research on this topic. ⢠Control of heat ï¬uxes at the walls without modeling the radiative heat transferâThis solution entails combining the radiative and convective heat transfer coefficient to form a local empirical transfer coefficient. No additional equation is required. In France, this method has been applied to simulate the heptane ï¬re test H32, which is carried out during the EUREKA 499 experiments. Results obtained with this method are claimed to be within reason. ⢠Reduction of the HRR at the ï¬re sourceâThis considers reducing the actual heat release source injected in the model by deducing the radiative part. This technique has been shown in several publications and the percent- age of energy lost by radiation at the ï¬re source is esti- mated to be in the range of 20% to 50% of the total heat energy released by combustion. The major problem with this method is caused by not taking into account the loss in radiative energy from the hot gases to the walls farther from the ï¬re. Smooth wall surfaces are generally the default CFD conditions. However, the user may generally deï¬ne rough surfaces by modifying the layer parameters to represent the zone very near the walls. The use of rough surfaces depends on the objectives of the simulations; if they concern the analysis of the forceâs balance or the propagation speed of the smoke front, the assumption made on surfaces will inï¬u- ence the results. Heat transfer boundary conditions may also affect the end results. ⢠Fixed temperature or fixed heat fluxesâIn this case, the temperatures at the walls or the heat ï¬uxes through walls are ï¬xed to constant values. ⢠Combination of ï¬xed temperature with heat ï¬uxesâ This technique may be used to roughly model the heat conduction process in the rock (soil). ⢠Heat conduction inside the rock (soil)âThis method appears as the best physical interpretation of the prob- lem. The heat transfer to the walls may have noticeable effects, especially in the case of extended ï¬res. However, it leads to larger meshes and longer calculations. Boundary conditions at the portals may seriously inï¬uence the ï¬nal results. ⢠Fixed pressures at both portalsâThis directly represents the atmospheric effects. The critical size of the out- side zone to be modeled is between 3 and 5 hydraulic diameters. However, acceptable results can be obtained without an outside domain, provided that some pre- cautions or even corrections are used. ⢠Specifying ï¬uid properties ï¬xed at one end and ï¬xed pressure at the other endâThis may be justiï¬ed for the analysis of the conditions inside the tunnel with known
ventilation effects. The upstream condition appears to be quite limited, because it forces the ï¬ow in one direction, especially if the velocityâs components are imposed. For modeling, it may be more reasonable to specify differ- ent pressure boundary conditions at the portals, such as wind effects. Simplifying the model can also have an effect on the results. Another challenge is in setting the time step for transient modeling. Sensitivity analyses are suggested to select the appropriate time step. The use of simpliï¬ed models may be applicable for long tunnels for which the boundary conditions are difficult to describe. A simpliï¬ed model, such as a one- dimensional (1D) model, can be used to estimate globally the ï¬ow in the complete tunnel and derive the boundary conditions to impose at both ends of the mesh. Simpliï¬ed 1D models can evaluate the critical fire locations, depending on the tunnel geometry and ventilation scheme. As a ï¬rst approach a simpliï¬ed model, such as a 1D model, is used to estimate globally the ï¬ow in the complete tunnel and derive the boundary conditions to impose at both ends of the mesh for 2D and 3D models. Simpliï¬ed 1D models are used to evaluate the critical ï¬re locations depending on the tunnel geometry and ventilation scheme, evaluate ventilation requirements, and provide ample sensitivity information. Although 3D simulations require long calculation times, 2D simulations may appear as attractive alternatives. This simpliï¬cation of the problem by utilizing a ï¬ow between two planes requires some precautions, such as: ⢠To take into account the reduction of the friction forces, and ⢠To perform modeling using similarities based on the Reynolds and Froude numbers and energy dimensionless parameters. The energy dimensionless parameters represent the energy released by the ï¬re or the quantity of fuel injected per square meter (square foot) of tunnel cross-sectional area. Therefore, where: Qï¬re is the heat release rate, and Stunnel is the tunnel cross-sectional area. These constraints inï¬uence the geometrical and gravita- tional terms. A validation assessment has been performed on the basis of Ofenegg Tunnel experiments. Some important limitations must be mentioned. For exam- ple, this technique does not correctly describe the stratiï¬cation in the instance when the longitudinal velocity is lower than Q Sfire tunnel is the name in 2D or 3D ( ),21 44 the critical value. Consequently, 2D simulations require spe- ciï¬c precautions and can be used in speciï¬c situations only. FINDINGS ON NUMERICAL MODELING BASED ON LITERATURE REVIEW The main advantage of the CFD models is to allow the study of cases for which no experimental data are available. After a preliminary validation has been made from full-scale tests, simulations are used to study many other situations. This tech- nique provides a general description of the various phenomena. This is the only method that offers such possibilities, even if the results must be considered as orders of magnitude. The major restriction is the time needed for the calculation and the complexity of the model. Lengthy preliminary vali- dations and skillful users are necessary; otherwise the obtained results may be misleading. The CFD models are therefore adapted principally to certain speciï¬c uses: ⢠To set up general design rules by simulating typical cases. ⢠To investigate new or especially complex situations. ⢠To obtain a thorough understanding of actual ï¬res and to analyze them. Moreover, there are many advantages that can be drawn from the use of CFD models in conjunction with other study methods. For instance, full-size testing will beneï¬t from some preliminary computational simulations (if necessary, very approximate) to assess the expected phenomena and their orders of magnitude. Also, a test program on scale models could advantageously be prepared by calculations aimed at evaluating the quality of similarities and orders of magnitude. The reduced-scale model will then allow for studies by varying the useful parameters. Last, new calculations can be made to calibrate the computational code in a ï¬rst step, and in a second step to understand, even extrapolate the model measurements. To simulate ï¬res in long tunnels or complex underground networks it may be useful to couple a 3D simulation with simplified ones, such as 2D or 1D models, to determine boundary conditions. This technique appears as a potential development of the numerical simulation. The CFD models include several physical models that have been validated against fundamental experiences, but where few global validations have been made on full-scale ï¬re tests. Therefore, the databases drawn from the EUREKA 499 and the Memorial Tunnel experiments are very useful for validat- ing CFD models (35). A plan to set up a CFD model that has been calibrated and ï¬tted with the numerous parameters that can easily be used by an unskilled user is probably not realistic. A preliminary val- idation work must compare qualitatively and quantitatively
45 calculated results and measurements to deï¬ne rules for run- ning ï¬re simulations. However, it may be able to ï¬t all the parameters without physical reasons because such calibration could not be transposed to other conï¬gurations. The international community has made large efforts and investments in research programs on ï¬re safety in tunnels during the last decade. The number of international congresses on this subject, the development of ï¬re model and large ï¬re test programs carried out in recent years in Europe and in the United States (Memorial Tunnel) conï¬rm this tendency. The CFD codes are already largely used to study ï¬re situ- ations in tunnels; however, additional research and validation works are required to ensure the validity of their results. Some research is required to improve existing models, such as turbulence or combustion models. This research is generally done by universities and laboratories, the activity of which deals with fundamental ï¬uid dynamics phenomena and development of CFD codes. With the development of the sprinkler system application for road tunnels came the need to model its performance. Much research has been published on CFD modeling of sprinkler systems and water mist systems (36â40); however, there is a need to validate the sprinkler models against full-scale tunnel ï¬re tests. Additional research is needed for numerical modeling of sprinkler system impacts on ï¬ame and ï¬re size. The next step is to undertake new small- and large-scale experiments with the primary objective of validating and calibrating physical models. It may include understanding of flow generated by fire as well as measurements of some physical smoke properties, which are critical for models (i.e., radiative smoke properties, generation of soot). SUMMARY Theoretical models, especially computer-based models, can be valuable in assisting tunnel ï¬re safety decision making. However, such models can also be misleading. Nowadays, CFD software is considered as the design tool of choice for obtaining an optimum design, because experi- mental methods are costly, complex, and yield limited infor- mation. However, it requires in-depth knowledge of physical processes and numerical models and, preferably, testing experience from the numerical modeler. The CFD simulations of tunnel ï¬res driven by buoyancy forces with significant energy release require a solution of the NavierâStokes equations with appropriate boundary conditions. Table 8 provides a summary of the objectives of analyti- cal ï¬re modeling for tunnel ï¬re safety based on the literature review. Many commercial CFD packages have been developed in recent years. Initially, the strengths, weaknesses, and limita- tions of each program are investigated. Validation of the results against experimental data or another equivalent pro- gram is necessary in order to have accurate results. Most of the commercially available CFD codes used in this synthesis report have been validated against some tests; however, at times users try to stretch the software application to areas where the applications have not been validated. For example, it is difficult to ï¬nd a CFD program that has been validated for sprinkler system application in full-scale tunnel tests. The same applies to turbulent models; radiation models applied for road tunnels. There is a need for additional tests and val- idations of the CFD models for road tunnels. Means Use for Research Use for Design Use for Operation Numerical Models (CFD) Advantages: - Possibility to study many different situations - Information on flow structures unattainable with other methods Disadvantages: - The conclusions must be correlated to existing experimental references Conclusions: - Useful method for research Advantages: - Possibility to get an optimization by the use of different assumptions Disadvantages: - The model requires qualification Conclusions: - Useful method for projects, if validated Advantages: - Possibility to describe the physical conditions in several locations of the tunnel Disadvantages: - Theoretical results lead to theoretical conclusions Conclusions: - The adaptation depends on the use of the model Source: PIARC (21). TABLE 8 OBJECTIVES OF ANALYTICAL FIRE MODELING FOR TUNNEL FIRE SAFETY