Printer Friendly

CFD study of smoke movement during the early stage of tunnel fires: comparison with field tests.

INTRODUCTION

In a tunnel environment, development of fire and smoke spread are affected by the fire set-up and ventilation conditions in the tunnel. During normal traffic operation, smoke can be diluted or pushed away from the detection system by the normal ventilation system, which is designed to maintain acceptable levels of contaminants in the tunnel (Beard and Carvel 2005). It can create conditions that may challenge the ability of detectors to detect and locate the fire in the early stage if the fire is enclosed in a vehicle or located behind an obstruction. In order to achieve early detection of fires in a tunnel, it is essential to understand how fire develops and smoke spreads during the initial stage of fire under various conditions.

An extensive Computational Fluid Dynamics (CFD) study was carried out as part of the International Road Tunnel Fire Detection Research Project (Liu et al. 2006a), which aimed at investigating the detection performance of current fire detection technologies. The CFD study included simulations of full-scale tests conducted by the National Research Council of Canada (NRCC) in the Carleton University laboratory tunnel and a series of simulations to examine effects of various fire scenarios and different ventilation schemes. Findings of this CFD study were (Kashef et al. 2008 and Ko et al. 2008);

* Simulated results exhibited relatively good agreement with laboratory test results.

* Temperature development inside the tunnel was considerably affected by fire scenarios, such that temperature rise near the ceiling was less significant for fires enclosed by a vehicle body than that for open fires.

* The simulations agreed with the laboratory test results in that the longitudinal airflow affected the burning behaviour of the fire and smoke spread in the tunnel. Moreover, the impact depended on the relative size of fire to the airflow velocity, as well as the fire scenario. In general, the ceiling temperature decreased with an increase of airflow.

* The development of temperature depended on the ventilation scheme (longitudinal, semi-transverse and fully-transverse ventilation systems) inside the tunnel.

* The length of the tunnel did not have a significant impact on the temperature development near the ceiling close to the fire location. Thus, the results found in the laboratory tunnel scale can reasonably be extrapolated to longer tunnels.

In order to further investigate and verify these results, field tests were conducted in an operating tunnel environment. This paper reports the CFD study carried out to simulate the field tests conducted in Tube A of the Carre-Viger Tunnel in Montreal. As well, the paper presents the results of the study and comparisons between model predictions and experimental data.

Numerical Simulations of Field tests

The current research employs the Fire Dynamic Simulator (FDS) version 4.07 (McGrattan and Forney 2006), developed by the National Institute for Standard and Technology, to study the fire growth and smoke movement in road tunnels. FDS is based on the Large Eddy Simulation (LES) approach and solves a form of high-speed filtered Navier-Stokes equations, valid for low-speed buoyancy driven flow. These equations are discretized in space using second order central differences and in time using an explicit, second order, predictor-corrector scheme. Turbulence parameters used in simulations were 0.2, 0.5, and 0.5 for Smagorinsky constant, turbulent Prandtl, and Schmidt number, respectively. For combustion, FDS uses a mixture fraction method based on equilibrium chemistry. Fire is modelled as the ejection of pyrolyzed fuel from the fuel surface that burns when mixed with oxygen (McGrattan and Forney 2006). Fire modelling and smoke generation are modelled based on specified stoichiometric parameters and yields for soot.

Three CFD simulations were carried out to simulate field tests conducted in the Carre-Viger Tunnel in Montreal. The aim of this study is to compare the simulation results of smoke movement with actual test results. Detailed description of the field tests can be found in Liu et al. (2008b). Comparisons were made to temperature and smoke optical density measurements.

MODEL DESCRIPTION

Three simulations were carried out with variations of fire set-up and location of the fire. Table 1 lists simulations and conditions used for each of the three simulations.
Table 1. List of Simulations

Simulation    NRC     Fire      Gasoline     Peak HRR    Location
ID          TEST #  Scenario    Pan Size       (kW)       of Fire

Tun4VF3     Test 3   UV (1)   0.6 m x 0.6 m  550~650       90 m
                                                       downstream of
                                                          the fan

Tun4VF6     Test 6   BV (2)   0.6 m x 0.6 m  550~650       90 m
                                                       downstream of
                                                         the fan

Tun4VF8     Test 8   BV (2)   0.6 m x 0.6 m  550~650       60 m
                                                       downstream of
                                                          the fan

Simulation    Ambient
ID          Temperature
            ([degrees]C)

Tun4VF3          17

Tun4VF6          17

Tun4VF8          17

(1) Pool fire located under a simulated vehicle body.
(2) Pool fire located behind a simulated vehicle.


Model Geometry

The simulated tunnel section was a 4-lane, 420 m long, 5 m high and 16.8 m wide, as shown in Figure 1. The fire was placed in the first lane, 4.2 m away from the north wall, as marked in Figure 1. The initial and boundary conditions of each simulation were set to mimic the conditions of the corresponding test. The boundary condition for walls, ceiling and floor was concrete. The west end of the tunnel was open. Airflow into the tunnel was specified through the east end corresponding to measured airflow velocities (about 1.3 m/s) of each experiment. The combination of this boundary condition and the air flow of a jet fan resulted in a flow of 1.4 m/s near the fire, which is close to the averaged velocity measured during tests of the fire location.

[FIGURE 1 OMITTED]

Grid Size

Grid convergence tests were conducted, in which 650 kW fire source (0.6 m x 0.6 m) was simulated under natural ventilation condition. Two different grid sizes were used; 0.3 m and 0.4 m. In addition, one grid setting using two overlapping meshes was also tested. In the setting, 0.3 m grid size was used for the tunnel, and 0.1 m grid size was used for fire area. Temperature variances over time were compared for different grid settings. Although the high resolution in the combustion volume predicts better in the combustion area, it does not improve the temperature prediction at the far field (Hadjisophocleous and McCartney 2005). Since the interest of this study is temperature variance over time at some distance from the fire, the optimal spatial size 0.3 m was selected to save computation times. The grid size of the middle of the model tunnel was 0.3 m (D) x 0.3 m (W) x 0.3 m (H) (Figure 1). To save cells for the rest of the long tunnel, the grid stretching technique (McGrattan and Forney 2006) was used. For the rest of the tunnel, a non-isometric grid of 0.3 m (D) x 2 m (W) x 0.3 m (H) was used since it was found from grid tests that this grid did not affect the temperature results in the middle section where comparisons were made with the experiments.

Ventilation Condition

The tunnel ventilation in the tests was maintained under normal operating conditions. The tunnel has a longitudinal ventilation system that is equipped with four ceiling jet fans (one in each lane spaced at 3.45 m). Detailed description of the jet fan can be found in Liu et al. (2008b). The fans are located in a recess in the tunnel ceiling, at which the maximum height is approximately 9 m. The base of the fans is at a height of 6 m. In these simulations, only one jet fan located on the ceiling of the third lane was activated. To model the jet fan in the model tunnel whose ceiling height was 5 m, the horizontal jet fan capacity was calculated taking into account the angle of the jet flow. As shown in Figure 1, the ceiling jet fan was mounted on the ceiling of the third lane. In the model a horizontal flow of 21 [m.sup.3]/s (45,000 cfm) over an area of 1.5 m x 0.5 m was defined.

During the field tests, air velocity measurements were conducted at a number of cross sections of the tunnel, including the section where the fire was located. Using a hand-held velocity meter, measurements were taken prior to and during the tests. Measurements showed that air velocities were relatively uniform at the section where the fire was located, and the average air velocity was 1.4 m/s. In the field tests, air velocities were found to be more uniform and stable with an increase in distance from the fans (Liu et al. 2008b). In the simulations, the same longitudinal airflow with a velocity of approximately 1.4 m/s was achieved at the section where the fire was located.

Fire Scenarios

The field fire tests in an operating tunnel were carried out using fire scenarios developed in Task 1 of the International Road Tunnel Detection Project for evaluating performance of road tunnel detectors. Two types of fire scenario were simulated as in the field tests: a pool fire located under a mock-up vehicle body (UV) and an open pool fire located behind a vehicle (BV). The fire size was approximately 650 KW. These fire scenarios are encountered in the majority of tunnel fire incidents and presented a challenge to the fire detection systems (Liu et al 2006b). The same fire scenarios were tested in the previous laboratory tunnel study as well as in this field tests. Detailed descriptions on these scenarios as well as geometry of mock-ups are provided in Liu et al. (2006b). As in the field tests, the effect of changing fire location was also simulated. The fire source was placed at two different locations in the tunnel. For simulation Tun4VF3 and Tun4VF6, the fire was placed, in the first lane, 90 m downstream of the fan, whereas the fire was placed, in the first lane, 60 m downstream of the fan for Simulation Tun4VF8, as shown in Figure 2.

[FIGURE 2 OMITTED]

Figure 3-(a) shows the vehicle body mock-up used in the field tests for a fire under a vehicle scenario in which a vehicle crashed, and the fuel leaks formed a pool under the vehicle body. In the simulations, a steel plate vehicle mock-up was built over the pool pan. The size of the plate was 1.5 m wide by 2.4 m long and was located 0.5 m above the ground. A 0.6 m x 0.6 m gasoline pool fire was placed under this obstruction.

[FIGURE 3 OMITTED]

For an open gasoline pool fire located behind a vehicle, which is a more general tunnel pool fire scenario (Liu et al. 2006b), a metal plate (2.5 m wide by 4.2 m high) obstacle simulating the front portion of a crashed truck was placed at a distance of 6 m in front of the pool fire and 0.3 m above the ground, as in the field tests [Figure 3-(b)].

Fire and Smoke Modelling

In FDS, the burning rate of gasoline can be prescribed by specifying the heat release rate of the fire, or alternatively the burning rate can be predicted based on the energy fed back from the fire. It was found, from preliminary simulations, that simulating the actual burning was sensitive to the thermophysical properties of the fuel and boundary conditions, particularly the vent conditions. Therefore, the fire source in this study was modelled by prescribing the heat release rate, eliminating some complications, and avoiding potential errors due to a combination of factors, such as insufficient grid resolution and uncertainty in the absorption coefficient and flame temperature.

The growth rate was modified to correspond to the test results so that the burning rates reflected the environment. The ramp-up time parameter was used to control the burning rate such that the fire developed within 60 seconds as in the tests. Satisfactory results were obtained in modeling these field tests.

Instrumentation

The predicted temperature and smoke optical density were compared against the field data. Temperatures were monitored at three different locations corresponding to the thermocouple trees used in the tests. Figure 2 shows schematic of thermocouple and smoke meter locations. Drop-1 and Drop-2 were placed at the middle of the tunnel with five thermocouples spaced at 1.0 m intervals starting 1 m above the tunnel floor. Drop-F was placed above the fire placed in the first lane with four thermocouples spaced at 1.0 m intervals staring 2 m above the pool. The spacing between drops was 15 m for Tun4VF3 and Tun4VF6. For Tun4VF8, Drops 1 and 2 were placed at 30 m and 60 m downstream of the fire, respectively. Smoke optical density values were monitored at two heights, 4.4 m and 2.7 m above the tunnel floor, for both Drop 1 and 2.

RESULTS

Flow Behaviour

Figure 4 shows a cross sectional view of the simulated temperature contour obtained from Simulation Tun4VF3 (at about 120 s from ignition). The fire plume and smoke were pushed downstream of the fire by the longitudinal airflow. The plume from 0.6 m x 0.6 m gasoline pan fire under the vehicle mixed with more air due to the vehicle obstruction.

[FIGURE 4 OMITTED]

When the same size of fire was placed behind an obstruction representing a vehicle front body, the smoke from the open fire was less diluted, as shown in Figure 5. When the open fire was placed close to the fan in Simulation Tun4VF8 in Figure 6, the plume was highly turbulent and mixed with more air.

[FIGURE 5 OMITTED]

[FIGURE 6 OMITTED]

For both fire scenarios, the fire was tilted by the longitudinal airflow. The obstructions used in each scenario contained the flame or shield the flame, which might challenge detectors particularly flame detectors.

Temperature Results

Figure 7 shows the ceiling temperature variation over time obtained at 30 m downstream of the fire for all simulations. Field test results were also plotted in Figure 7. The predicted temperatures are slightly higher than test data. Both simulated results and experimental results show that the ceiling temperature from a fire under the vehicle is lower than that from an open fire because the open fire grows quickly and burns freely. This result is consistent with those found in the laboratory tunnel tests, where more tests and detailed analyses of the impact of fire types were made (Liu et al 2008a).

[FIGURE 7 OMITTED]

When the fire is placed close to the fan (Tun4VF8), the air flow from the fan directly impacts on the fire and plume generation, resulting in relatively low ceiling temperature rise, comparable to the temperature from the fire under the vehicle. At the early stage of the fire, the ceiling temperature rise is substantially low particularly for the fire under the vehicle and open fire close to the fan, which might lead to detection delay for thermal detectors at the ceiling.

From the CFD modeling, time-averaged mean values of temperature between 60 s and 150 s were obtained. In Figures 8, 9 and 10, the time-averaged temperature variations along the central axis for the tunnel length were plotted for Tun4VF3, Tun4VF6, and TunVF8, respectively. Comparisons with test data, which were also averaged over the same period of time, were also made at locations of Drops 1 and 2 in the figures. The figures show that the ceiling temperatures predicted in model results are slightly higher than the test results. As with experimental measurements, the temperatures decreased with an increase in height.

[FIGURE 8 OMITTED]

[FIGURE 9 OMITTED]

[FIGURE 10 OMITTED]

The largest simulated temperature gradients are found near the fire source, however the highest ceiling temperature was not at the fire location but was observed at about 3~5 m downstream of the fire. This is attributed to the tilting of the plume due to the longitudinal airflow in the tunnel.

The ceiling temperatures downstream of the fire gradually decreased with an increase in distance from the fire source. The ceiling temperature upstream of the fire abruptly dropped down to the ambient temperature for Simulation Tun4VF3, while for simulation Tun4VF6 the ceiling temperature upstream of the fire slowly dropped down to the ambient temperature (smoke moved back about 30 m against the longitudinal ventilation).

Comparing the results from simulation Tun4VF6 with Tun4VF8, it can be seen that the change of the fire location had an impact on temperature development near the ceiling close to the fire. The ceiling temperature estimated in Tun4VF8, 30 m downstream of the fire was lower than that estimated in Tun4VF6, which indicates that the detection range can also be affected by the fire location. Results from Tun4VF3 showed the least ceiling temperature variation along the central cross section of the tunnel.

Smoke Optical Density

Generation of smoke and smoke transport were modelled by FDS. Every time step, the mixture fraction at each grid cell is calculated so that smoke concentration ([C.sub.S]) is tracked along with other combustion products. Smoke concentration is calculated based on soot yield and mass burning rate. The value of soot yield used was 0.1, which is suggested for gasoline pool fires (Evans et al. 1988 and Notarianni et al. 1993). Using smoke concentrations ([C.sub.S]), the smoke optical density (OD) is calculated, which indicates the level of smoke obscuration. The smoke optical density was calculated by Equation

(1) (Mulholland 2002) using the extinction coefficient (K ) calculated by FDS [Equation (2)]. The value used for mass extinction coefficient ([K.sub.m]) was 7600 [m.sup.2]/kg which is suggested for flaming combustion (Mulholland 2002). The visibility (S) may be calculated from the optical density (OD) as Equation (3) (Mulholland 2002). In order to achieve visibility of 10 m, the smoke optical density should be lower than 0.13 [OD/m] for light-reflecting sign.

OD = K/2.3 (1)

K = [K.sub.m][C.sub.S] (2)

S = 3/K for light-reflecting sign (3)

The smoke optical density predicted by FDS was compared with values measured in the field tests. The smoke optical density values were monitored at two heights, 4.4 m and 2.7 m, at Drop 1 and 2. Figure 11 compares the smoke optical density variation over time at Drop 2 resulted form simulation Tun4VF6 and measured from the Test 6. The simulated results showed that the smoke produced by the open fire under the airflow conditions fluctuated, which was consistent with measurements in field tests. At the lower height, the smoke production was quite small both in simulation and field test, which indicates the smoke layer was maintained in the upper part of the tunnel, above 2.7 m from the floor, in the early stage of fire.

[FIGURE 11 OMITTED]

The time-averaged mean values of the smoke optical densities [OD/m] near the ceiling were obtained at Drops 1, 2 both from FDS modeling and field tests. The smoke optical density values were compared in Figure 12 with respect to the distance from the fire. In general, the smoke optical density near the ceiling decreased with an increase in distance from the fire source, which is consistent with field test results. However, simulated results of the smoke optical density near the ceiling 30 m downstream of the fire for Tun4VF3 is significantly higher than that 15 m downstream of the fire. In general, the comparisons were more favourable for the scenario with the fire behind a vehicle than those for the case with a fire under a vehicle. At 30 m downstream of the fire, the smoke optical density produced by the open fire behind a vehicle was higher than that by the fire under a vehicle under the same test conditions; however, when the fire was closed to the fan, the smoke optical density near the ceiling is comparable with that from the fire under a vehicle. Both simulation and test results shows that the smoke optical density near the ceiling maintains the same level 60 m downstream of the fire. The field measurement of smoke optical density at 60 m is equivalent to visibility of about 70 m.

[FIGURE 12 OMITTED]

CONCLUSION

A CFD study of temperature and smoke spread in the early stage of the fire was carried out to simulate the field tests conducted in an operating tunnel in Montreal, Canada. The simulation examined the ceiling temperature distribution and smoke movement over the 420 m long section of the tunnel. Different fire scenarios and locations were simulated. The simulation results were also compared to those obtained in field tests, showing reasonable agreement. Results were also consistent with those determined in the laboratory tunnel tests.

The ceiling temperature produced by the fire behind the vehicle was higher than that produced by the fire under the vehicle. This indicates that the response times of fire detection systems for the fire enclosed in a vehicle body can be longer than those for the open fire. However, if there is an obstacle such as a truck in front of the open fire, the flame might be invisible to flame detectors downstream of the fire.

In general, the ceiling temperatures predicted in model results are slightly higher than the test results, and the ceiling temperature downstream of the fire decreased with an increase in distance from the fire source, which is consistent with results of the smoke optical density.

The temperature variation along the central cross section of the tunnel located the highest ceiling temperature 3~5 m downstream of the fire since the plume was tilted by the air flow inside the tunnel. This indicates that locating the exact point of the fire might be challenging if the air flow inside tunnel is strong and unstable.

The change in fire location had a significant impact on the smoke production and its distribution in the tunnel. Depending on airflow condition at the fire location, the smoke temperature distribution along the tunnel can be altered so that the detection range can also be affected. This indicates that air velocity field inside the design tunnel must be considered.

REFERENCES

Beard, A., and R. Carvel. 2005. The Handbook of Tunnel Fire Safety

(Bendelius, A. and Brinckerhoff, P.), Thomas Telford Ltd., p.127.

Evans, D.D., Mulholland, G.W., Gross, D., Baum, H.R., and K. Saito. 1988. Burning, smoke production and smoke dispersion from oil spill combustion, Arctic and Marine Oilspill Program (AMOP) Technical Seminar 11th. Proceedings, Ontario, Canada, 41-87 pp.

Hadjisophocleous, G.V., and C. McCartney. 2005 Guidelines for the use of CFD simulations for fire and smoke modelling, ASHRAE Transactions, DE-05-7-3 pp. 583-594.

Kashef, A., Ko, Y., Hadjisophocleous, G., Liu, Z.G.,

Lougheed, G.D., and G. Crampton. 2008. International road tunnel fire detection research project--Phase II, Task3: Computer Modelling, Fire Research program, Institute for Research in Construction, National Research Council of Canada, Canada.

Ko, Y., Kashef, A., and G. Hadjisophocleous. 2008. Modeling of smoke movement during the early stage of tunnel fires under different ventilation conditions. The International Congress of Smoke Control in Buildings and Tunnels, Santander, Spain.

Liu, Z.G., Kashef, A., Lougheed, G.D., Su, J.Z., Benichou, N., and K.H. Almand. 2006. An Overview of the international road tunnel fire detection research project. 10th Fire Suppression and Detection Research Application Symposium, February 01, Orlando, FL., USA. pp. 1-7.

Liu, Z.G., Crampton, G., Kashef, A., Lougheed, G.D., Gibbs, E., Su, J.Z., and N. Benichou. 2006. International road tunnel fire detection research project--Phase II, Task1: Fire detectors, Fire scenarios and Test protocols, Fire Research program, Institute for Research in Construction, National Research Council of Canada, Canada.

Liu, Z.G., Crampton, G., Kashef, A., Lougheed, G.D., Gibbs, E., and S. Muradori. 2008. International road tunnel fire detection research project--Phase II, Task2: Full-scale fire tests in a laboratory tunnel for study of tunnel fire detection technologies, Fire Research program, Institute for Research in Construction, National Research Council of Canada, Canada.

Liu, Z.G., Crampton, G., Kashef, A., Lougheed, G.D., Gibbs, E., and S. Muradori. 2008. International road tunnel fire detection research project--Phase II, Task 4: Field fire tests on performance of fire detection systems in an operating road tunnel in Montreal, Fire Research program, Institute for Research in Construction, National Research Council of Canada, Canada.

Liu, Z.G., Crampton, G., Kashef, A., Lougheed, G.D., Gibbs, E., and S. Muradori. 2008. International road tunnel fire detection research project--Phase II, Task 7: Full-scale ventilated fire tests in a laboratory tunnel for study of tunnel fire detection technologies, Fire Research program, Institute for Research in Construction, National Research Council of Canada, Canada.

McGrattan, K.B., and G.P. Forney. 2006. Fire Dynamics Simulator (Version 4)--User Guide. NISTIR 6784, National Institute of Technology and Standards, USA.

Mulholland, G.W. 2002. SFPE Handbook of Fire Protection Engineering, Smoke Production and Properties. National Fire Protection Association, Quincy, Massachusetts, 3rd edition.

Notarianni, K.A., Evans, D.D., and W.D. Walton. 1993. Smoke production from large oil pool fires, Interflam 1993, 6th International Fire Conference, Oxford, England, pp. 111-119.

Yoon J. Ko

George V. Hadjisophocleous, PhD, PE

Member ASHRAE

Ahmed Kashef, PhD, PE

Member ASHRAE

Yoon Ko is a PhD candidate in the Fire Protection Engineering Program and George Hadjisophocleous holds the Industrial Research Chair in Fire Safety Engineering and is a professor at Carleton University, Ottawa, Ontario, Canada. Dr. Ahmed Kashef is a senior research officer at the Fire Research Program of the National Research Council of Canada (NRC), Ottawa, Ontario, Canada.
COPYRIGHT 2009 American Society of Heating, Refrigerating, and Air-Conditioning Engineers, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2009 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:computational fluid dynamics
Author:Ko, Yoon J.; Hadjisophocleous, George V.; Kashef, Ahmed
Publication:ASHRAE Transactions
Article Type:Report
Geographic Code:1CANA
Date:Jul 1, 2009
Words:4266
Previous Article:Comparisons of numerical predictions and filed tests in a road tunnel.
Next Article:Evaluation of a novel liquid-flooded Ericsson cycle cooler for vending machine applications.
Topics:


Related Articles
Analysis of drafting effects in swimming using computational fluid dynamics.
Verification of CFD modeling for smoke control using two compartment fire experiments.
Impact of various parameters on the CFD predictions of atrium smoke management systems.
Siemens and Singapore's A*STAR-IHPC to Establish Joint Computational Fluid Dynamics (CFD) Laboratory.
Comparisons of numerical predictions and filed tests in a road tunnel.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |