Influence of different make-up air configurations on the fire-induced conditions in an atrium

Building and Environment 45 (2010) 2458e2472

Contents lists available at ScienceDirect

Building and Environment journal homepage: www.elsevier.com/locate/buildenv

Influence of different make-up air configurations on the fire-induced conditions in an atrium Cándido Gutiérrez-Montes a, *, Enrique Sanmiguel-Rojas a, Antonio Viedma b a b

Fluid Dynamics Division of the Department of Mining and Mechanical Engineering, University of Jaen, Jaen, Spain Department of Thermal and Fluid Engineering, Technical University of Cartagena, Murcia, Spain

a r t i c l e i n f o

a b s t r a c t

Article history: Received 23 December 2009 Received in revised form 4 May 2010 Accepted 6 May 2010

In this paper, the influence of the make-up air velocity as well as the position and area of the vents in an atrium is assessed both experimentally and numerically. In the experiments, the effect of different makeup air supply positions and inlet area on the fire-induced inner conditions and smoke-layer descent was studied by means of three full-scale fire tests conducted in a 20 m cubic atrium. Detailed transient measurements of gas and wall temperatures, as well as pressure drop through the exhaust fans and airflow at the inlets were recorded. These data could be used as benchmark for future numerical validation studies. Later computational fluid dynamics (CFD) simulations of these tests were performed with the code Fire Dynamics Simulator (FDSv4). In the experiments, the lack of symmetry in make-up air vents and the large inlet area turn the flame and plume more sensitive to outer effects. However, no significant difference has been observed between the make-up air topologies assessed. Even make-up velocities higher than 1 m/s, with symmetric venting topology, have not induced important flame or plume perturbations. In the numerical simulations, the predictions agree well with the experiments for the cases with larger make-up air openings. Poor agreement has been found for the case with the smallest inlet openings. Ó 2010 Elsevier Ltd. All rights reserved.

Keywords: Make-up air Mechanical exhaust CFD simulations Atrium Full-scale fire tests

1. Introduction The rapid smoke spread through an atrium in case of fire is a major concern. At this point, mechanical exhaust systems are commonly installed in big atria for smoke control. It is believed that these systems have more reliable performance than others, such as vents for natural ventilation, as it does not depend strongly on ambient conditions. However, the performance of these systems can be influenced by various factors like the temperature of the smoke, the formation of a pre-stratification layer in the atrium, the exhaust rate, the outer wind or the make-up air among others. All these factors have to be studied and taken into account when designing a smoke exhaust system [1]. When using a mechanical exhaust system it is necessary to supply make-up air to conserve mass [2]. Without it, the atrium smoke exhaust component cannot be expected to perform as desired. In general, some of the goals for the make-up air to be

* Corresponding author. Despacho 022/023, Área de Mecánica de Fluidos, Departamento de Ingeniería Mecánica y Minera, Edificio A3, Campus Las Lagunillas, 23009 Jaén, Spain. Tel.: þ34 953212903; fax: þ34 953212870. E-mail address: [email protected] (C. Gutiérrez-Montes). 0360-1323/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.buildenv.2010.05.006

accomplished are to be uncontaminated outdoor air, to be supplied from positions below the smoke layer [3] or to be supplied at low velocity not to disturb the flame and plume. An extensive explanation of these goals, as well as the main methods for supplying the make-up air, was presented by Duda [4]. Failure to achieve these goals could lead to ineffective designs of smoke exhaust systems. At this point, there is a lack of practical guidance for supplying makeup air. In addition, nowadays, there is greater design flexibility [5], which puts more responsibility on the designer. Fire engineers can give architects and building designers greater scope for achieving their architectural vision in an atrium. This is a fundamental idea and gets fire scientists involved in the creative process. Make-up air vents are one of the most significant concerns for architects since they are at lower levels and are more visible, whereas mechanical venting is often on the roof and rarely as important aesthetically. In relation to make-up air, some standards and codes [1,6] suggest that make-up air should not exceed 1 m/s in areas, unless a higher velocity is supported by engineering analysis. This limit prevents significant deflection of the plume and disruption of the smoke-layer. This type of engineering analysis could be based on comparisons developed with full-scale, scale, or CFD modelling. At

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

2459

Fig. 1. Photo of the fire tests facility in (a), and sketch and main dimensions in (b).

this point, CFD modelling can be the basis for exceptions for the requirements to smoke-layer depth, the 1 m/s limitation on makeup air and plugholing, among others. Nowadays, with the possibility of implementing engineering performance-based fire codes, CFD fire models [7] are commonly used in hazard assessment in atria. However, these computational models need of further validation and verification studies of atrium fire modelling [8], especially full-scale data verification, in order to limit their range of applicability and their reliability in case of different boundary conditions, including the case of different air supply and smoke

a

exhaust conditions. Whether these models are suitable for use is queried, leading to challenges [9]. So far, few studies on the effect of make-up air have been conducted. Some authors like Hadjisophocleous and Lougheed [10] performed studies on mechanical exhaust with different make-up air positions. Yi et al. [11,12] carried out full-scale experiments at the PolyU/USTC Atrium, of 22.4 m  11.9 m  27 m, and numerical simulations to study the effects of make-up air inlet vents location. They concluded that, if the minimum smoke-layer interface height was above the safe level then air inlets lower than that should be

b

14.3 m 9.4 m

13

14

0

1

2

15

16

17

3

4

5

18

19

6

7

20 C2 23 (T) 54 (v)

C1 22 (T) 53 (v)

15.15 m

12

10.15 m 5.15 m 1.25 m

1.25 m

15.15 m 10.15 m 5.15 m

14.3 m 9.4 m 4.3 m

4.3 m

A3 9 (T) 50 (v)

A2 10 (T) 51 (v)

8 A1 11 (T) 52 (v)

1.9 m

2.9 m

6.9 m 16.9 m

17.9 m

12.75 m 9.75 m 6.75 m 3.75 m

28

29

30

34

31

27

32

9.25 m 5.25 m

48 45 49 57 (A.P.), 58 (D.P.), 60 (T)

55 (A.P.), 56 (D.P.), 59 (T) 46 44 47 35

24

25

2.6 m

33

1.75 m

d

13.25 m

c

26

Pool-fire Wall A

Fig. 2. Layout of the main sensors considered. Wall A, from the outside in (a), wall C, from the outside in (b), central section, from the wall A in (c), and roof and fans, from the top, in (d). Temperature sensors in squares, velocity and temperature sensors in triangles and pressure and temperature sensors in pentagons.

2460

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

Fig. 3. Venting topology of the tests. Test #1: vents open in walls D (D2) and A (A1 and A2) in (a), test #2: vents in wall A (A1 and A3 open, A2 closed) in (b), and test #3: vents in wall C (C1 and C2 partially open) in (c). Views from inside.

installed whereas, on the contrary, extraction with higher air inlets could be advisable, as it would reduce the smoke temperature. There are other important factors such as the distribution of air inlets, distance of the air inlet from the fire, and airflow velocity through the air inlet, that should also be considered. Recently, Kerber and Milke [13] performed a numerical study on the possible effects of various make-up air supply arrangements (symmetrically located vents placed low in the spaced, an array of vents distributed from the floor to the ceiling and asymmetrically located vents) and velocities (from 0.5 to 3.0 m/s) in an atrium smoke management system within a 30.5 m high cubical atrium. It was found that the best layout of vents was that in which make-up air was supplied to the fire symmetrically to avoid plume perturbations. In addition, it was found that even velocities lower than 1 m/s could cause the smoke layer to descend below the design criterion. As a consequence, it was stated that the inlet velocities should be diffused so that they were very low when they reached the fire and had no effect on it. Lately, an ASHRAE research project was conducted to investigate the maximum velocity of make-up air in atrium smoke control applications. Zhou and Hadjisophocleous [14] performed a numerical study on the influence of different parameters such as the outer wind or the locations of the make-up air vents on fire plumes. It was noticed that, placing openings at the bottom and the top of the atrium caused the least disruption to the plume. However, locating the opening at the top caused significant mixing of smoke with the air of the lower layer. Besides, Hadjisophocleous and Zhou [15]

performed a numerical study on the influence of make-up air velocities, considering different fire sizes in various size atria (heights from 10 to 60 m, heat release rates (HRR) from 1 to 5 MW with different fire locations and inlet velocities from 0.5 to 1.5 m/s). It was concluded that the increased make-up air velocity lowered the interface height in the atrium, that the make-up air velocity restriction of 1 m/s was not conservative, as it caused plume disturbances that resulted in lower interface heights, and that the impact of the make-up air velocity was more pronounced in atria with height less than 20 m. The main aim of the present work is to provide a set of full-scale experimental data of atrium fires. These data could be used as benchmark for future numerical validation studies. The tests were carried out as part of the Murcia Atrium Fire Tests, performed in Murcia, Spain (see [16,17] for details). In these tests, the make-up air inlet area and vents distribution have been varied. In the first test, all the vents were completely open with asymmetric layout. In the second test, some of the vents were completely open with symmetric layout. Finally, in the third test, the vents open in the previous test were only partially open. Detailed transient measurements of gas and wall temperatures at different heights, as well as the pressure drop through the exhaust fans and airflow at the inlets, were taken. Later CFD simulations of the experiments have been performed using FDSv4 [18] in order to check its capability to properly predict the fire-induced conditions for the different make-up air configurations.

Table 1 Summary of laboratory and ambient conditions during the fire tests. Fire test

Volume of heptane (l)

Burning time (s)

Open vents

Exhaust fans on

Ambient temp ( C)

Pressure (mbar)

HRR (MW)

Test #1 Test #2 Test #3

44 52 52

837 883 1094

All A1, A3, C1, C2 100% A1, A3, C1, C2 22%

All All All

16.7 28.9 27.5

1018 1008 1007

1.35 1.51 1.22

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

2461

Table 3 Number of cells needed in each direction for different grid resolutions. Fire test (MW)

Cells number for R* ¼ 1/5

Cells number for R* ¼ 1/10

1.35 1.51 1.22

90 86 94

181 173 188

conditions have been measured with a meteorological station monitoring the wind velocity, temperature, humidity and pressure outside the facility. An uncertainty analysis for the measurements conducted in [16,17] showed a total experimental uncertainty of 1.5% for the thermocouples, of 0.4% for the thermistors, of 4% for the velocity probes and of 1% for the mass loss rate of fuel.

2.2. Fire tests Fig. 4. HRR vs time of test #2.

2. Murcia test facility and fire tests 2.1. Test facility The fire tests reported here have been carried out in the atrium of the Centro Tecnológico del Metal, Spain (Fig. 1a) [16,17]. This facility, unique in Spain and one of the few full-scale facilities in Europe for these kinds of tests, has global dimensions of 19.5 m  19.5 m  19.5 m. The walls and roof are made of 6 mm thick steel and the floor is made of concrete. There are four exhaust fans installed on the roof, each with a diameter of 0.56 m and a nominal flow rate of 3.8 m3/s. There are eight grilled vents arranged at the lower parts of the walls. Each vent has dimensions of 4.88 m  2.5 m. A drawing of the rig with dimensions is shown in Fig. 1b. The atrium contains up to 61 sensors distributed in 4 different regions: the front wall (wall A), the back wall (wall C), a central section parallel to walls A and C and the roof and fans region. A sketch of the location of the main sensors considered for this work as well as their nomenclature is shown in Fig. 2. The variables reported here are the air temperature next to the walls and the make-up air temperature and velocity, Fig. 2a and b, 30 cm inside the atrium, the temperature at the central section, Fig. 2c, and the smoke through the fans temperature, Fig. 2d. For the air temperature, class B bare Pt100 thermistor probes have been used at the fans, walls and vents. For the gas temperature at the central section, sheathed type K thermocouples have been used. For the make-up air velocity, hot wire anemometers have been used. A Modicom TSX Premium automaton connected to a PC has been used to register the data with a frequency of 1 Hz. A digital camera has been also used to take graphic records of the flame. Finally, the weather

Results from three full-scale fire tests with different make-up air venting conditions are reported. The burning fuel was heptane contained in a pool-fire, of 0.92 m diameter and 0.25 m deep, placed at the centre of the atrium floor. For the first test, all the make-up vents were completely open, with a total inlet area of 97.50 m2, Fig. 3a. For the second test, the vents were completely open with symmetric layout. In this case, the two vents of the wall C (back wall) and the two vents at the corners of the wall A (front wall) were completely open, with a total inlet area of 48.75 m2, Fig. 3b. Finally, for the third test, the same vents as in the test #2 were partially open, at 22% of their area, with a total inlet area of 10.83 m2, Fig. 3c. In all tests, a layer of 2 cm of water was added to the pan before the heptane was poured to insulate the metal from the burning pool heat, thus providing a more stable steady burning regime. At the end of each test, the volume of water was measured again to confirm that no water had been lost. A summary of the laboratory and ambient conditions during the tests is presented in Table 1. The heat release rate (HRR), Q_ , of each test has been calculated as,

_ ceff DHc ; Q_ ðtÞ ¼ mðtÞ

(1)

_ where mðtÞ is the mass loss rate of the fuel, DHc is the heat of combustion and ceff the combustion efficiency. The heat of reaction of heptane for complete combustion is 44.6 MJ/kg [19]. The combustion efficiency for a well ventilated heptane pool-fire of 0.92 m diameter is 0.85  0.12 [8,20,21]. Fig. 4 shows the HRR of the test #2. The uncertainty associated with the HRR is estimated to be around 15%. A detailed explanation of these measurements is presented in Ref. [17]. The complete set of measurements from the experiments is shown and discussed in Section 4 of the paper, after the description of the fire simulations.

Table 2 Centreline averaged plume temperatures at different heights as a function of the grid size for a 1.3 MW fire. Grids are expressed as number of cells per atrium side (19.5 m). Height

Exhaust fan Plume at 13 m Plume at 9 m Plume at 5 m

Temperature predictions ( C)

Relative error respect to finest grid (%)

40 cells

60 cells

90 cells

120 cells

150 cells

180 cells

40 cells

60 cells

90 cells

120 cells

150 cells

49 64 80 116

66 110 152 226

64 99 173 333

53 78 129 293

58 81 160 503

56 74 136 487

13 14 41 76

18 49 12 54

14 34 27 32

5 5 5 40

4 9 18 3

2462

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

a T (ºC)

100

50

0

b

0

100

200

300

400

500

600

700

800

900

1000

0

100

200

300

400

500

600

700

800

900

1000

0

100

200

300

400

500

600

700

800

900

1000

T (ºC)

80 60 40 20 0

c T (ºC)

60 40 20 0

t (s) Fig. 5. Comparison of temperature measurements and predictions for different grid sizes for a 1.3 MW test [17] at the plume 12.55 m high in (a), fans in (b), and 10 m high near wall A in (c). Experiments in solid line, 120 cells per side in triangle, 150 cells per side in circle, and 180 cells per side in square.

3. Description of fire simulations 3.1. Mathematical model The CFD code FDSv4 [18] has been used for the simulation of the three fire tests to check its capability to predict properly the fire-induced conditions. The turbulence has been modelled using a large-eddy simulation (LES) approach [22], and the combustion model is based on the mixture fraction approach which assumes a mixing-controlled combustion, appropriate for the simulation of turbulent diffusion flames with fast chemistry such as pool-fires. The HRR is prescribed as an input, function of time, at the pool, Fig. 4. The radiative heat transfer is computed by solving the radiation transport

equation for a non-scattering grey gas. In calculations in which the grid cells are on the order of a centimetre and larger, such as the ones presented in this paper, the temperature near the flame surface cannot be relied upon when computing the source term in the radiation transport equation [18]. To compensate for the computation of this source term in the neighbourhood of the flame, the radiative fraction of a heptane pool-fire, 0.35 [23], has been set, which is the default value in FDSv4. The fans have been set a constant exhaust velocity across their area providing the nominal flow rate of 3.8 m3/s each. The grilled vents have been simulated as openings to the atmosphere at ambient pressure. Finally, the walls and roof have been modelled as 6 mm thick steel sheets [24] and the floor as a thick layer of concrete [24].

Table 4 Timeemeasurements for different sensors in test #1. Temperature values in  C, sensors 25, 27, 29, 60, 1, 4, 7, 12, 16 and 19, and velocity values in m/s, sensors 50 and 53. See Fig. 2 for sensor labels. Time (s)

0.0 54.6 111.8 167.3 222.5 277.7 331.5 386.8 443.6 497.4 552.7 607.5 663.7 720.4 776.8

Sensor number 25

27

29

60

1

4

7

12

16

19

50

53

14.9 29.6 101.2 108.9 104.0 99.1 99.5 99.9 103.6 104.8 113.8 123.9 142.7 162.2 159.3

16.1 21.8 51.5 55.2 56.8 59.7 64.1 69.0 72.7 77.6 82.9 88.1 89.0 92.2 93.8

16.7 20.6 37.0 43.6 48.1 50.5 56.3 61.4 65.1 70.4 73.1 76.8 80.0 80.6 82.1

15.9 24.6 34.5 41.3 48.1 52.1 57.8 62.0 66.3 70.5 73.3 75.7 77.2 77.5 78.5

16.7 19.7 26.3 32.8 38.5 44.0 48.8 52.8 57.4 60.8 63.8 66.1 67.9 69.3 70.8

16.8 17.6 19.0 23.2 30.2 36.5 42.7 47.5 52.0 56.1 58.9 61.8 63.9 65.8 66.9

16.7 17.9 18.6 19.5 20.2 22.8 25.0 28.3 32.1 35.7 39.6 42.4 43.5 44.4 44.4

17.0 19.5 24.8 32.1 38.4 44.9 49.8 54.4 59.2 62.7 66.0 68.1 69.5 70.9 72.0

16.8 17.9 20.3 24.4 31.1 37.0 42.5 46.7 50.7 55.1 58.7 61.9 63.5 65.1 66.7

16.8 18.0 19.3 21.0 22.9 24.9 27.0 29.2 32.4 35.6 38.4 41.1 42.0 42.3 43.2

0.3 0.3 0.4 0.3 0.3 0.2 0.3 0.3 0.6 0.5 0.4 0.4 0.5 0.3 0.3

0.1 0.1 0.2 0.3 0.3 0.3 0.4 0.0 0.0 0.4 0.3 0.1 0.2 0.1 0.2

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

2463

Table 5 Timeemeasurements for different sensors in test #2. Temperature values in  C, sensors 25, 27, 29, 60, 1, 4, 7, 12, 16 and 19, and velocity values in m/s, sensors 50 and 53. See Fig. 2 for sensor labels. Time (s)

0.0 59.0 118.9 177.1 236.1 294.9 354.3 413.3 472.4 532.7 592.8 651.7 707.7 766.1 825.7

Sensor number 25

27

29

60

1

4

7

12

16

19

50

53

29.2 67.0 160.1 144.3 136.6 142.2 172.8 180.9 178.9 156.5 143.1 174.4 165.8 171.1 209.8

32.0 45.4 58.0 58.4 67.4 73.1 80.0 83.3 88.1 88.5 89.0 95.1 97.9 100.8 108.5

32.3 41.9 50.1 50.7 59.3 62.2 72.4 75.1 79.8 79.6 79.4 84.8 87.6 91.3 95.6

32.1 38.8 47.3 53.9 62.4 66.8 74.8 78.6 82.5 83.1 83.3 85.9 90.0 93.9 92.4

31.6 34.1 41.9 47.1 52.0 56.7 63.9 69.6 73.3 75.6 77.0 78.9 81.1 82.6 83.8

31.0 31.1 34.1 38.9 43.4 48.2 55.5 61.1 65.6 68.0 70.6 73.1 75.1 76.5 78.1

30.5 30.7 31.8 33.1 35.0 36.7 39.2 42.7 45.2 45.7 45.9 47.8 50.1 50.1 49.7

30.6 32.3 38.5 44.8 50.4 56.5 63.8 68.4 72.7 75.7 77.2 79.6 81.5 83.4 84.0

31.9 32.0 34.6 40.3 44.4 51.8 58.5 64.0 68.6 72.7 75.0 77.1 79.4 81.2 82.0

31.4 31.7 32.5 34.0 35.6 37.2 40.1 43.2 45.7 46.9 46.9 47.9 48.6 48.0 46.9

0.1 0.2 0.4 0.2 0.2 0.2 0.3 0.5 0.3 0.5 0.4 0.8 0.4 0.7 0.6

0.6 0.5 0.6 0.4 0.4 0.6 0.4 0.6 0.6 0.5 0.5 0.5 0.4 0.6 0.4

3.2. Grid sensitivity study

1 1 < R* < 10 5

The computational domain includes the atrium space, the walls and the roof. The grid has been systematically refined until no significant difference is noticed with a cell size reduction and a compromise solution between numerical accuracy and computational cost is achieved. Six different grids have been assessed, 40, 60, 90, 120, 150 and 180 cells per side of the cubic atrium. For this study, constant HRR fires of 1.3 MW have been simulated and averaged quasi-steady conditions at different locations have been considered. The results have been compared between them, and with the finest grid results, to quantify grid independence. In LES it is not possible to archive perfect grid independence although little variations can be theoretically expected between grids if they are fine enough [22,25]. Table 2 shows the temperature predictions in the atrium for each grid. It can be observed that the temperature values and the relative errors for the three coarser grids are quite large. The temperature differences are lower than 10% at the upper parts, above 13 m high, from 120 cells per side. Thus, it could be concluded that any of the three finer grids (120, 150 and 180 cells per side) could be valid for simulating the fire tests. This grid sensitivity study would be enough for a conventional simulation. However, as an LES technique is used, the grid spatial resolution, R*, is needed to be between,

R* ¼

Dx z*

(2)

:

(3)

for a proper simulation [8,18]. The grid spatial resolution, R*, is defined in terms of the characteristic diameter of a plume [8,19], z*,

z* ¼




2=5 Q_ : pffiffiffi rN cp;N TN g

(4)

In the above expressions Dx is the characteristic length of a cell for a given grid, g is the gravity acceleration modulus and rN , cp;N and TN are the ambient density, specific heat and temperature, respectively. Table 3 shows the necessary grid cells per side regarding the above resolutions. Regarding this criterion, the results from Table 2 and in order to assure grid independency, grids of 180 cells per side (z5.8 million cells) have been used, that is, cells smaller than 0.11 m in length. Finally, Fig. 5 shows a comparison with the experimental results from [17] for a 1.3 MW heptane fire test conducted at the Fire Atrium of the Centro Tecnológico del Metal. The agreement between experiments and the grids of 150 cells and 180 cells per

Table 6 Timeemeasurements for different sensors in test #3. Temperature values in  C, sensors 25, 27, 29, 60, 1, 4, 7, 12, 16 and 19, and velocity values in m/s, sensors 50 and 53. See Fig. 2 for sensor labels. Time (s)

0.0 73.4 148.4 222.2 293.2 366.8 440.9 515.2 590.1 661.4 735.5 807.7 879.8 952.6 1024.1

Sensor number 25

27

29

60

1

4

7

12

16

19

50

53

28.3 59.6 86.1 101.2 124.8 118.2 123.9 145.9 130.4 129.6 130.9 130.4 127.6 128.4 132.5

32.0 44.9 54.0 60.1 67.0 69.4 73.9 82.4 82.0 83.7 86.5 87.7 87.7 87.7 89.8

32.5 41.3 49.7 55.6 60.2 64.3 66.7 73.5 75.3 76.3 79.4 79.0 78.6 79.0 81.1

32.5 43.2 52.8 60.3 66.0 68.7 74.8 79.0 78.4 82.4 84.6 83.9 84.9 86.2 87.3

31.9 33.7 39.4 46.3 52.4 57.4 61.6 66.4 69.3 71.3 73.4 74.7 75.3 76.8 77.5

30.3 30.3 32.3 37.6 42.9 49.8 54.7 59.1 63.4 66.3 68.6 70.3 71.2 72.5 73.3

28.5 28.9 29.9 31.3 33.1 34.9 36.8 38.5 40.2 41.8 42.4 42.6 42.6 43.2 43.6

32.6 34.9 41.4 48.2 54.0 59.6 64.3 68.2 70.9 73.8 76.5 77.4 78.5 79.9 80.5

30.8 31.0 32.7 37.5 42.7 50.8 57.3 62.0 67.3 71.0 73.2 74.9 76.7 77.8 78.6

29.1 29.6 30.9 32.4 33.8 35.5 37.1 38.6 40.1 41.2 41.8 42.0 42.1 42.4 42.9

0.7 0.8 0.9 0.8 0.3 0.7 1.1 1.2 1.0 0.9 1.3 1.0 1.1 1.0 1.2

0.9 1.0 1.4 1.0 1.2 0.9 1.0 0.9 1.0 1.7 1.5 1.3 1.1 0.9 1.6

2464

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

a

25 FDS

700

b

600

27 FDS

300 250

T (ºC)

T (ºC)

500 400 300

200 150 100

200

50

100 0

0

200

400

600

0

800

c

d

120

200

400

600

800

80 70 60

80

50

T (ºC)

100

T (ºC)

0

60

40 30

40 20 20 0

e

29 FDS 0

200

400

600

0

800

f

70

200

400

600

800

60 50

T (ºC)

50

T (ºC)

0

70

60

40 30

40 30 20

20

1 12 FDS

10 0

59 60 FDS

10

0

200

400

600

0

800

g

4 16 FDS

10

h

0

200

400

600

800

20 Predicted smoke layer height N−percent 10% N−percent 20% N−percent 30%

40 15

H (m)

T (ºC)

30

20

0

i

0

200

400

600

50 FDS

0.6

j

0.4 0.2

0

200

400

600

800

1 53 FDS

0.8

v (m/s)

v (m/s)

0.8

0

800

1

0

5

7 19 FDS

10

10

0.6 0.4 0.2

0

200

400

t (s)

600

800

0

0

200

400

600

800

t (s)

Fig. 6. Test #1 measurements and predictions. Temperature at the plume at 5.25 m high in (a), 9.25 m high in (b), and 13.25 m high in (c), at the exhaust fans in (d), near the walls at 15 m high in (e), 10 m high in (f), and 5 m high in (g). Smoke-layer height in (h). Make-up air velocities at the vent A1 in (i), and C2 in (j). Measurements identified by sensor number according to Fig. 2.

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

2465

Fig. 7. Flame slightly leaned because of the non-symmetric venting conditions, t ¼ 56 s in (a), t ¼ 270 s in (b), and t ¼ 532 s in (c). View from wall B.

side is good, confirming the results from the grid sensitivity study and spatial resolution requirements. 4. Results and discussion Next, the fire-induced transient conditions within the Fire Atrium are studied. In this section, the main results from the fire tests and the simulations as well as from the comparisons between them are reported and discussed. Results in four key regions are considered: the centreline temperature (ideally the plume), the smoke through the fans temperature, the air close to walls

temperature and the make-up air inlet velocity at the vents. Measurementetime evolutions at different locations are presented for test #1, in Table 4, test #2, in Table 5, and test #3, in Table 6. The temperature evolution near the walls shows the built up of the smoke layer. The smoke-layer height has been calculated in the experiments by means of the N-percent method [26] and compared with the one predicted by FDS. In the experiments, temperature increases from ambient temperature of 10e20% [26] and of 30% [12] of the highest temperature rise have been assessed to locate the smoke-layer interface. In general, good agreement has been achieved between FDS and the value of N ¼ 30%.

Fig. 8. Predicted flame inclinations due to the ventilation asymmetry, t ¼ 669.6 s in (a), and t ¼ 837.0 s in (b). Predicted velocity contours at h ¼ 1.25 m, t ¼ 669.6 s in (c), and t ¼ 837.0 s in (d). View from wall C.

2466

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

Fig. 9. Detail of predicted velocity vectors near the flame, h ¼ 1.25 m, t ¼ 669.6 s in (a), and t ¼ 823.6 s in (c). Predicted velocity contours at the central section, t ¼ 669.6 s in (b), and t ¼ 837.0 s in (d). View from wall C.

4.1. Test #1 This test was with the four exhaust fans activated and all the vents completely open (total inlet area of 97.5 m2), therefore, nonsymmetric venting conditions were tested, Fig. 3a. Fig. 6 shows the measurements and predictions vs time at different locations. During the test, the flame and plume presented small but persistent deviations due to the non-symmetric vents layout, Fig. 7. This can be observed from the temperature registered at the centreline close to the flame, Fig. 6. At h ¼ 5.25 m, Fig. 6a, the temperature rises fast at the first instants due to the growing combustion phase, which lasts 200 s. Then, it stabilizes at 100  C, below the temperature obtained from the correlation from Heskestad [27] for axis-symmetric plumes, T ¼ 210  C. At the final stages, from t ¼ 590 s, the flame becomes nearly vertical and the temperature rises up to 170  C. The same trend has been observed at h ¼ 9.25 m, Fig. 6b, where the temperature increases up to T ¼ 80  C, at t ¼ 470 s, near the temperature obtained from Ref. [27], T ¼ 88  C, and keeps on increasing slowly to 95  C at the end. At higher locations, i.e. h ¼ 13.25 m, these flame and plume deviations hardly affect the temperature evolution as the flame deviations are not very strong and the temperature increases continuously due to the smoke accumulation at the upper parts, Fig. 6c. At the far field, there is no influence of the flame inclinations and plume deviations. At the exhaust fans, Fig. 6d, the temperature increases reaching steady conditions at t ¼ 590 s, when T ¼ 75  C, remaining constant until the end. At the higher locations near the walls, h ¼ 15 and 10 m, the temperature rises continuously reaching quasi-steady conditions at the end, with T ¼ 72 and 68  C, respectively, Fig. 6e and f. At h ¼ 5 m, Fig. 6g, the temperature rises fast to

42  C, at t ¼ 600 s, then it remains almost constant. These trends observed at the near the walls region indicate that the inner fireinduced conditions were near to reach steady values within the whole facility. At the vents, Fig. 6i and j, averaged make-up air inlet velocities ranging from 0.2 to 0.4 m/s were measured, below the recommended value of 1 m/s. In the numerical simulation, no perturbation affecting the flame or the plume was observed in the beginning. However, from t ¼ 630 s, the flame started to lean provoking subsequent plume deviations, Fig. 8a and b. These are small perturbations that affect mainly the region close to the flame. This can be appreciated at h ¼ 5.25 m, Fig. 6a, where sudden temperature drops are observed at the end of the simulation. These effects have little influence at higher locations, above h ¼ 13 m, and the far field, where no significant disturbance is observed in the temperature evolutions. These little flame inclinations are provoked by the nonsymmetric make-up air vents layout. Fig. 8c and d shows the formation of a circular air stream around the fire, which increases its intensity with time. At the final moments, Fig. 9, azimuthal velocity values higher than 1.5 m/s are observed around the fire, which disturb it and make it to rotate and lean. All these phenomena affecting both the experiment and the simulation generate differences between the measurements and the predictions. Comparison between them shows that FDS overpredicts the temperature near the flame at the centreline, Fig. 6. The bigger discrepancies are found in the beginning, when FDS predicts faster temperature rises. These are enhanced by the perturbations affecting both the experiment and the simulation, which generate relative errors of 300%, at h ¼ 5.25 m, of 100%, at h ¼ 9.25 m, and lower than 50%, at h ¼ 13.25 m. At the quasi-steady combustion regime, the differences reduce and remain fairly

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

a

800

2467

b

700

200

500

T (ºC)

T (ºC)

600

400 300

150

100

200

50 25 FDS

100

27 FDS

0

0 0

200

400

600

800

0

c 120

d 100

100

80

T (ºC)

T (ºC)

80 60

200

400

600

800

60 40

40 20 0

0 0

200

400

600

800

0

f

80

70

60

60

50

50

40

20

600

800

40

20

1 12 FDS

10

4 16 FDS

10

0

0 0

200

400

600

800

0

200

400

600

800

20

hh

50 40

Predicted smoke layer height N−percent 10% N−percent 20% N−percent 30%

15

30

H (m)

T (ºC)

400

30

30

g

200

80

70

T (ºC)

T (ºC)

e

59 60 FDS

20

29 FDS

10

20 5

7 19 FDS

10 0

0 0

400

600

800

1

0

j

0.8

0.6

0.6

0.4 0.2

50 FDS

0

200

400

600

800

1

0.8

v (m/s)

v (m/s)

i

200

0.4 0.2

54 FDS

0 0

200

400

t (s)

600

800

0

200

400

600

800

t (s)

Fig. 10. Test #2 measurements and predictions. Temperature at the plume at 5.25 m high in (a), 9.25 m high in (b), and 13.25 m high in (c), at the exhaust fans in (d), near the walls at 15 m high in (e), 10 m high in (f), and 5 m high in (g). Smoke-layer height in (h). Make-up air velocities at the vent A1 in (i), and C2 in (j). Measurements identified by sensor number according to Fig. 2.

2468

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

Fig. 11. Predicted velocity contours at h ¼ 1.25 m. t ¼ 44.2 s in (a), t ¼ 353.2 s in (b), t ¼ 618.1 s in (c) and t ¼ 883.0 s in (d). View from wall A.

constant. These differences reduce with height, being the agreement good at h ¼ 13.25 m, from t ¼ 400 s. At the final stages, there is good agreement at the centreline caused by the predicted flame inclinations due to the make-up air effect. The far field conditions are not influenced by the flame inclinations. At the exhaust fans, Fig. 6d, the predicted temperatures agree well with the measurements during the tests. From t ¼ 700 s, FDS slightly under-predicts the temperature with differences lower than 9%. At the upper parts of the near the walls region, h ¼ 15 and 10 m, the agreement is also good, with relative errors of 8 and 6% at the end, Fig. 6e and f. At h ¼ 5 m near the walls, Fig. 6g, in the experiment the temperature rises continuously to 42  C, at t ¼ 600 s, remaining constant until the end. However, in the numerical simulation no temperature rise was predicted. The differences at this location are of 50% at the end. As a consequence, the predicted final smoke-layer height is higher than the experimental one, Fig. 6h. At the vents, Fig. 6i and j, FDS predicts well the make-up inlet velocities. 4.2. Test #2 This test was with all the fans activated and the vents completely open with symmetric layout (total inlet area of 48.75 m2), that is, the two vents of the wall C (vents C1 and C2) and the vents closest to the corners of wall A (vents A1 and A3), Fig. 3b. Fig. 10 shows the measurements and predictions vs time at different locations. During the experiment, small flame inclinations were observed. These effects can be noticed at the locations closest to the flame, e.g. h ¼ 5.25 m, where sudden temperature drops and rises are registered, Fig. 10a. These perturbations weaken with height, being negligible at the far field.

Regarding the numerical simulation, it evolved normally without any noticeable perturbation affecting the flame or the plume. Fig. 11, shows the velocity contours, at h ¼ 1.25 m for four different times, in which it can be observed make-up air velocities lower than 1 m/s and that no air stream formed around the flame, therefore, the flame and plume remained vertical most of the time. Comparison with the experimental data shows that FDS overpredicts the temperature near the flame, Fig. 10aec. For the quasisteady combustion regime, the relative errors are between 50 and 100%, at h ¼ 5.25 m, and lower than 50%, at h ¼ 9.25 m, where good agreement is found at the end. The differences reduce with height being the agreement really good at h ¼ 13.25 m, where no difference is found at the end. At the exhaust fans, Fig. 10d, the agreement is good too, with differences lower than 5% during the whole fire. At h ¼ 15 and 10 m near the walls, Fig. 10e and f, FDS predicts faster temperature rises at the fire growing period. Then, the differences reduce being the agreement good until the end, when the differences are lower than 7 and 8%, respectively. These discrepancies are larger at the lower regions of the far field. At h ¼ 5 m high, Fig. 10g, in the experiment the temperature increases to 50  C, at t ¼ 700 s, remaining constant until the end whereas, in the numerical simulation no temperature increase is predicted. This provokes that the smoke layer arrival to the upper locations near the wall is well predicted but the final smoke-layer height is over-estimated, as in the previous test, Fig. 10h. At the vents, FDS predicts well the make-up air velocity, which is ranged between 0.3 and 0.5 m/s, approximately. In the experiment, the influence of outer effects such as the outer wind cause larger velocity variations than the predicted ones. However, the average velocity values agree well, above all from the middle stages of the fire.

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

25 FDS

b

120

300

100

250

80

T (ºC)

T (ºC)

a 350

200 150

2469

27 FDS

60 40

100 20

50 0

0 0

200

400

600

800

1000

200

400

600

800

80

60

60

40

40

59 60 FDS

20

20 29 FDS 0

0 0

200

400

600

800

1000

0

f

80 70

70

60

60

50

50

40

400

20

4 16 FDS

0 0

200

400

600

800

1000

70

0

h

200

400

H (m)

40 30 20

1000

10

5

7 19 FDS

10

800

Predicted smoke layer height N−percent 10% N−percent 20% N−percent 30%

15

50

600

20

60

T (ºC)

1000

40

10

0

0

0 0

200

400

600

800

0

1000

j

2.5

2

1.5

1.5

1 0.5

50 FDS

0

200

400

600

800

1000

2.5

2

v (m/s)

v (m/s)

800

20

1 12 FDS

10

i

600

30

30

g

200

80

T (ºC)

T (ºC)

e

1000

d 80

T (ºC)

T (ºC)

c

0

1 0.5

53 FDS

0 0

200

400

600

t (s)

800

1000

0

200

400

600

800

1000

t (s)

Fig. 12. Test #3 measurements and predictions. Temperature at the plume at 5.25 m high in (a), 9.25 m high in (b), and 13.25 m high in (c), at the exhaust fans in (d), near the walls at 15 m high in (e), 10 m high in (f), and 5 m high in (g). Smoke-layer height in (h). Make-up air velocities at the vent A1 in (i), and C2 in (j). Measurements identified by sensor number according to Fig. 2.

2470

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

Fig. 13. Detail of predicted velocity vectors near the flame, h ¼ 1.25 m, t ¼ 218.8 s in (a), and t ¼ 853.3 s in (c). Predicted velocity contours at the central section, t ¼ 218.8 s in (b), and t ¼ 853.3 s in (d). View from wall A.

4.3. Test #3 This test was with all the fans on and the vents partially open (total inlet area of 10.83 m2) with symmetric layout, as in the test

a

500 plume 4.55 m 150 cells plume 4.55 m 180 cells

300

b 120 100

T (ºC)

T (ºC)

400

#2 but at 22% of their inlet area, Fig. 3c. In this case, large discrepancies between the experiment and the simulation have been observed. Fig. 12 shows the measurements and predictions vs time at different locations.

200 100 0

c

80 60 40

0

200

400

600

800

20

1000

d

80 70

plume 12.55 m 150 cells plume 12.55 m 180 cells 0

200

400

600

800

1000

70 near the wall 5 m high 150 cells near the wall 5 m high 180 cells

60

T (ºC)

T (ºC)

60 50

50 40

40 30 20

fans 150 cells fans 180 cells 0

200

400

600

t (s)

800

1000

30 20

0

200

400

600

800

1000

t (s)

Fig. 14. Comparison of temperature predictions from a 150 cells grid and a 180 cells grid. Plume at h ¼ 4.55 m high in (a), plume at h ¼ 12.55 m in (b), exhaust fans in (c), and near the walls at h ¼ 5 m high in (d).

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

a

800 10 ºC 20 ºC 30 ºC 40 ºC

700 600

t (s)

500 400 300 200 100 20

b

40

60

80

100

800 10 ºC 20 ºC 30 ºC

700 600

t (s)

500 400 300 200 100 20

c

40

60

80

100

800 10 ºC 20 ºC 30 ºC

700 600

t (s)

500 400 300 200 100 20

40

60

80

100

2

make−up inlet area (m ) Fig. 15. Time needed for a certain temperature increase vs make-up air inlet area. Exhaust fans in (a), h ¼ 15 m near the walls in (b), and h ¼ 10 m near the walls in (c). Walls A and C have been considered for these calculations.

In the experiment, small but fairly persistent flame inclinations and plume deviations towards wall C were observed during the fire. These flame inclinations can be also appreciated from the temperature measurements at the centreline at h ¼ 5.25 m, Fig. 12a, where they are lower than those obtained from [27], T ¼ 200  C. However, these flame inclinations were not very strong as, at h ¼ 9.25 m, Fig. 12b, the temperature measurements agree well with Ref. [27], T ¼ 89  C, from t ¼ 400 s. At higher locations and the far field no influence of these perturbations has been observed. At the exhaust fans, Fig. 12d, the temperature rises continuously reaching quasi-steady conditions at t ¼ 700 s and, at the near the walls region, similar temperature evolution trends are observed in both walls, showing the high symmetry of the smoke-layer growth. In the numerical simulation, different fire-induced conditions have been predicted. For this case study and grid resolution a no physical solution is obtained. The fire evolved normally during the first 200 s. From then on, strong flame inclinations and plume deviations caused by the make-up air influence are observed. As happened in test #1, a circular air stream surrounding the flame forms at lower locations and grows in intensity with time. In addition, the HRR of these fire tests are relatively low and the

2471

influence of the make-up air becomes even more important. This effect can be clearly appreciated from the velocity contours at the central section which are completely chaotic from t ¼ 300 s, Fig. 13. From that moment, azimuthal velocity values higher than 1.3 m/s appear around the flame, increasing to values of 2.5 m/s, at t ¼ 500 s, and higher than 3.5 m/s near the end, Fig. 13. As a consequence, the plume becomes unsteady and looses its verticality, leaning towards the walls randomly. These perturbations cause big discrepancies between the experimental data and the numerical predictions. At the near field region, in the numerical simulation the temperature drops suddenly from t ¼ 300 s, Fig. 12aec, and values much lower than the measurements are predicted. For this case study and grid resolution, the disturbances caused by the make-up air affect the whole facility. In this way, at the exhaust fans, Fig. 12d, and h ¼ 15 and 10 m near the walls, Fig. 12e and f, there is good agreement until t ¼ 350 s. Then, the predicted temperature drops to values lower than the experimental ones. At the end, temperature differences between 22 and 30%, at the exhaust fans, and equal to 18%, at h ¼ 15 and 10 m near the walls, are found. This effect is more important at h ¼ 5 m near the walls, Fig. 12g. At this location, no temperature rise is predicted at the beginning. Then, when the perturbations become important, the temperature rises suddenly. This indicates that the mixing rate at the smoke-layer interface increases considerably. As a consequence, the smoke dilutes and cools and the smoke-layer thickens. The mixing enhancement between the smoke layer and the fresh air provokes that the simulation over-predicts the temperature at this location by more than 55% at the end. The predicted smoke-layer height is also lower than the measured one, Fig. 12h, eventually reaching the height of h ¼ 5 m. At the vents, there is good agreement, Fig. 12i and j, where the inlet velocities are ranged from 1 to 1.5 m/s in average. These numerical perturbations have been observed not only in one simulation and neither with only one grid cell size. Up to three different simulations with a grid of 180 cells per side and an additional one with a grid of 150 cells per side have been performed with the same result, Fig. 14. It has been also observed that these predicted perturbations grow with the grid resolution, that is, when the grid becomes finer, which, on the other side, is essential for obtaining reliable predictions. The predicted smoke-layer height has been compared with the experimental results obtained by means of the N-percent method [12,26]. In general, there is good agreement between FDS predictions and the 30% temperature increase. It is observed that the simulations over-estimate the final smoke-layer height for the two first tests, Figs. 6h and 10h, in agreement with other works [12]. For the particular case of the third test, Fig. 12h, good agreement is found for the higher locations, before the numerical perturbations become important. From that moment, the predicted smoke-layer growth varies significantly, e.g. at t ¼ 400 s. At t ¼ 600 s the smoke layer reaches a quasi-steady value equal to 5.5 m, lower than the final values for tests #1 and #2 of higher HRR. Finally, the time needed to reach a certain temperature increase at the exhaust fans and at h ¼ 15 and 10 m near the walls has been calculated from the experimental results in order to roughly evaluate the influence of the make-up air inlet area, Fig. 15. From this comparison, no significant difference has been noticed at the smoke-layer growth. The times needed to reach a certain temperature increase are the same for tests #1 and #2, which indicate that there is no influence of the make-up air inlet reduction on the smoke-layer growth, in agreement with [1], as the make-up air velocities are lower than 1 m/s. For the test #3, make-up air inlet velocities higher than 1 m/s have been measured. In this case, the times needed for a certain temperature increase are similar or even larger than in previous tests, due to the HRR difference between

2472

C. Gutiérrez-Montes et al. / Building and Environment 45 (2010) 2458e2472

them. From these results, make-up air velocities higher than 1 m/s do not necessarily have to influence the smoke-layer descent velocity for this case study. 5. Conclusions This paper reports results from three full-scale atrium fire tests carried out as part of the Murcia Atrium Fire Tests. At these tests, the effect of different make-up inlet positions and area on the fireinduced conditions has been studied. These tests have been conducted placing a heptane pool-fire of 0.92 m diameter at the centre of the floor and with four exhaust fans activated on the roof. The test #1 has been carried out with all the vents open (97.5 m2 of inlet area) and non-symmetric layout. The test #2 was with half of the inlet area of the previous one (48.75 m2 of inlet area) and symmetric layout. In the test #3, the same vents than in test #2 were partially open, to 22% of their area (10.83 m2 of inlet area). Later CFD simulations of the tests have been performed using FDSv4, in order to check the capability of this code to predict the fire-induced inner conditions and to show the validity of the experimental data to be used as benchmark for numerical validations. In the experiments, the tests with the larger make-up inlet areas (test #1 and #2), induce make-up air velocities much lower than 1 m/s (average values ranged from 0.2 to 0.5 m/s), within the recommended values [1]. However, large areas of make-up air supply turn the flame and plume more sensitive to outer effects. In addition, the lack of symmetry can also induce larger instabilities on the flame than when symmetric make-up air supply conditions are considered, in agreement with Ref. [13]. In the tests reported here, these perturbations have caused small flame inclinations and plume deviations, although no significant disturbance influencing the far field conditions has been noticed. In the test #3, with the smallest vents area, inlet velocities higher than 1 m/s have been measured. However, these have not influenced the flame or plume and, thus, the fire-induced conditions at the far field. Therefore, make-up air velocities higher than 1 m/s do not have to necessarily provoke a significant enhancement of smoke production. In the numerical simulations, for the grid resolution used, FDS predicts well the air inlet velocities at the vents. For the tests #1 and #2, with the larger make-up air inlet areas and make-up air velocities lower than 1 m/s, FDS behaves well at the upper parts of the far field, the most important ones for the design of smoke evacuation systems. For these tests, FDS significantly over-predicts the plume temperature near the flame with differences larger than 40%, below h ¼ 10 m. However, the discrepancies reduce with height being the agreement good above h ¼ 13 m, in agreement with other works [8,17]. At the exhausts fans and the upper parts near the walls (h  10 m), the agreement is also good, with relative errors lower than 10%, whereas, at lower locations, FDS underpredicts the temperatures, slightly over-predicting the final smokelayer height. For the test #3, with the smallest make-up air inlet area and make-up air velocities higher than 1 m/s, poor agreement has been found. FDS predicts too strong flame and plume disturbances caused by the make-up air. These numerical perturbations enhance the smoke-layer mixing with the fresh lower air, homogenizing the smoke-layer conditions and thickening it, which generate relative errors larger than 20% respect to the experimental data. A deeper study would be required to make more definitive conclusions on the behaviour of FDS for this case study.

Finally, additional fire tests and simulations should be conducted varying the fire HRR and the make-up air conditions in order to continue assessing the capacity of FDS for the simulation of atrium fires and its use for the performance-based design of smoke evacuation systems in atria, similar to the one used here.

References [1] NFPA 92B. Standard for smoke management systems in malls, atria and large spaces. Quincy, MA: National Fire Protection Association; 2005. [2] Butcher EG, Pamell AC. Smoke control systems e the provision of replacement air. Fire Safety Engineering 1997;4(3):12e3. [3] Klote JH, Milke JA. Design of smoke management system. Atlanta, GA, USA: ASHRAE Inc.; 2002. [4] Duda S. Atria smoke exhaust: 3 approaches to replacement air delivery. ASHRAE Journal 2004;46(6):21e7 (Atlanta, GA). [5] Tubbs JS. Smoke management in the 2006 IBC. ASHRAE Journal 2006;48 (10):50e5. [6] International Code Council. 2006 International building code. Country Club Hills, IL: International Code Council; 2006. [7] Olenick SM, Carpenter DJ. An updated international survey of computer models for fire and smoke. Journal of Fire Protection Engineering 2003; 13:87e110. [8] Dreisbach J, McGrattan K. Verification and validation of selected fire models for nuclear power plant applications. In: Fire dynamics simulator (FDS), NUREG-1824 final report, vol. 7. U.S. Nuclear Regulatory Commission, Office of Nuclear Regulatory Research; May 2007. [9] Chow WK, Li SS, Gao Y, Chow CL. Numerical studies on atrium smoke movement and control with validation by field tests. Building and Environment 2009;44:1150e5. [10] Hadjisophocleous GV, Lougheed G. Experimental and numerical study of smoke conditions in an atrium with mechanical exhaust. International Journal on Engineering Performance-Based Fire Codes 1999;1(3):183e7. [11] Yi L, Huo R, Li YZ, Peng L. Study on the effect of unsymmetric air supply to mechanical exhaust in large space fire. Journal of University of Science and Technology of China 2003;33:579e85. [12] Yi L, Chow WK, Li YZ, Huo R. A simple two-layer zone model on mechanical exhaust in an atrium. Building and Environment 2005;40:869e80. [13] Kerber S, Milke JA. Using FDS to simulate smoke layer interface height in a simple atrium. Fire Technology 2007;43:45e75. [14] Zhou J, Hadjisophocleous GV. Parameters affecting fire plumes. ASHRAE Transactions 2008;114(Part 1):140e6. [15] Hadjisophocleous GV, Zhou J. Evaluation of atrium smoke exhaust make-up air velocity. ASHRAE Transactions 2008;114(Part 1):147e55. [16] Gutiérrez-Montes C, Sanmiguel-Rojas E, Kaiser AS, Viedma A. Numerical model and validation experiments of atrium enclosure fire in a new fire test facility. Building and Environment 2008;43(11):1912e28. [17] Gutiérrez-Montes C, Sanmiguel-Rojas E, Viedma A, Rein G. Experimental data and numerical modelling of 1.3 and 2.3 MW fires in a 20 cubic atrium. Building and Environment 2009;44(9):1827e39. [18] McGrattan K. Fire dynamics simulator (version 4) technical reference guide. National Institute of Standards and Technology, Special Publication 1018; 2006. [19] SFPE of fire protection of engineering. 3rd ed. Quincy, MA, USA: National Fire Protection Association; 2002. [20] Hostikka S, Kokkala M, Vaari J. Experimental study of the localized room fires. NFSC2 test series (VTT building technology) VTT research notes 2104; 2001. [21] Tewarson A. Combustion efficiency and its radiative component. Fire Safety Journal 2004;39:131e41. [22] Pope SB. Computations of turbulent combustion: progress and challenges. Proceedings of the Combustion Institute 1990;23:591e612. [23] Hamins A, Kashiwagi T, Buch R. Characteristics of pool fire burning. In: Totten, Reichel, editors. Fire resistance of industrial fluids. ASTM STP 1284; 1996. p. 15e41. [24] Incropera FP, DeWitt DP. Fundamentals of heat and mass transfer. 4th ed. John Wiley & Sons; 1996. [25] McGrattan KB, Baum HR, Rehm RG. Large eddy simulations of smoke movement. Fire Safety Journal 1998;30:161e78. [26] Cooper LY, Harkleroad M, Quintiere J, Reinkinen W. An experimental study of upper hot layer stratification in full-scale multi-room fire scenarios. Journal of Heat and Mass transfer 1982;104:741e9. [27] Heskestad G. Engineering relations for fire plumes. Fire Safety Journal 1984;7:25e32.