Applying the FDS pyrolysis model to predict heat release rate in small-scale forced ventilation tunnel experiments

Journal Pre-proof Applying the FDS pyrolysis model to predict heat release rate in small-scale forced ventilation tunnel experiments Xiaoyun Wang, Charles Fleischmann, Michael Spearpoint PII:

S0379-7112(18)30491-0

DOI:

https://doi.org/10.1016/j.firesaf.2020.102946

Reference:

FISJ 102946

To appear in:

Fire Safety Journal

Received Date: 11 November 2018 Revised Date:

23 December 2019

Accepted Date: 8 January 2020

Please cite this article as: X. Wang, C. Fleischmann, M. Spearpoint, Applying the FDS pyrolysis model to predict heat release rate in small-scale forced ventilation tunnel experiments, Fire Safety Journal (2020), doi: https://doi.org/10.1016/j.firesaf.2020.102946. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier Ltd.

Xiaoyun Wang: Conceptualization, Methodology, Writing- Original draft preparation, Visualization. Charles Fleischmann: Conceptualization, Supervision, Reviewing and Editing. Michael Spearpoint: Conceptualization, Supervision, Reviewing and Editing.

Applying the FDS pyrolysis model to predict heat release rate in small-scale forced ventilation tunnel experiments Xiaoyun Wang1, Charles Fleischmann2 & Michael Spearpoint2,3 1 Arup Fire, Leeds, UK 2 Civil and Natural Resources Engineering, University of Canterbury Christchurch, New Zealand 3 OFR Consultants, Manchester, UK E-mail: [email protected]

ABSTRACT The pyrolysis model in FDS6 is applied to simulate a series of small-scale tunnel experiments under varied forced ventilation velocities, where cribs made of medium density fireboard (MDF) are used as the fuel source. Prior to the simulations the material properties of MDF are investigated in order to effectively describe the decomposition behaviour in FDS6 and to develop optimised values. In the tunnel simulations, the effect of the burning duration of the ignition source, the heat of combustion of the MDF in the presence of the forced ventilation, the assumed fuel mass and the modelling domain set-up on the prediction of heat release rate are examined. It is found that the cribs take longer to ignite than that in the experiment; the available burning mass is affected by the assumed thickness of the crib sticks and the available surface area; and a limited domain size can result in not all of the fuel burning. Efforts are made to obtain an equivalent burning mass to the experiments and to account for the influence of the forced ventilation on the burning efficiency by manually specifying the heat of combustion to improve the predictions of heat release rate and consumption of fuel mass.

KEYWORDS: FDS pyrolysis model, small-scale tunnel experiments, heat release rate

NOMENCLATURE

time (s) -1

pre-exponential factor (s )

temperature (K or ˚C)

component mass fraction (-)

yield of residue in solid phase reaction

-1

c

-1

specific heat (J g K ) ∗

mass conversion fraction (-)

characteristic fire diameter (m) activation energy (J mol-1)

∆

Greek symbols

effective heat of combustion (MJ kg-1)

heating rate (K min-1)

thermal conductivity (W m-1 K-1)

density (kg m-3)

thermal conductivity along the panel thickness at 30 ˚C (W m-1 K-1) ∆

correction for

-1

-1

(W m K )

Subscripts end of reactions

reaction order

ith component

number of components

jth second

heat release rate (kW)

peak -1

decomposition reaction rate (% K ) universal gas constant, 8.314 (J K-1 mol-1)

INTRODUCTION One of the most important parameters in the selection of design fire scenarios for tunnels is the heat release rate (HRR). This input provides information for the evaluation of fire hazard severity, the tenability conditions for occupants and tunnel ventilation design parameters, etc. Various tunnel fire safety standards and guidance provide design fire HRR values. Cheong [1] compared NFPA 502 (2004 and 2008), BD78/99, PIARC and CETU in which the values are generally obtained from different large-scale fire experiments. The peak fire size in these documents for passenger cars, buses and vans usually ranges from 5 to 30 MW and can exceed 200 MW for goods vehicles. However, the use of these recommended design fires has limitations when representing different tunnel fire scenarios where a range of vehicles may be involved and the influences on the fire from the tunnel size and ventilation conditions may be different to those in the original experiments. Even though valuable results can be obtained through large-scale tunnel experiments using them to measure the fire sizes for different circumstances is not a practical approach due to the complexity and high costs involved. In order to have a cost-effective approach to obtain the HRR for tunnel fires in different scenarios Cheong et al. [2] proposed the use of the Fire Dynamics Simulator (FDS) as a predictive tool. One of the experiments in the Runehamar heavy goods vehicle (HGV) programme [3] was modelled by dividing up the surface into individual elements each with a defined ignition temperature and burning rate history in order to predict the HRR. A peak value similar to the experimental result was obtained where the spread of fire was fully dependent on the pre-described surface element properties. Another potential approach to predict the HRR of tunnel fires using FDS is to apply its kinetic pyrolysis model. This approach uses the decomposition reactions of materials and one-dimensional heat transfer to predict the mass loss rate for solid fuels and further to predict the HRR of the fire using the FDS combustion model. The influences on the decomposition reactions from environmental conditions such as ventilation effects, tunnel geometry and suppression systems can be assessed so

that theoretically a more realistic fire can be obtained by using the pyrolysis model approach. Currently the main applications of the pyrolysis model are for material- and bench-scale experiments where the heat transfer can be simply modelled in one-dimension [4, 5]. The use of the pyrolysis model to simulate large-scale fires has not been widely studied although Li [6] notes that the accuracy of the HRR prediction from the FDS pyrolysis model is limited due to the assumptions it contains. However, it is still useful to investigate the predictive capability of the pyrolysis model for both smallscale and large-scale fire scenarios to investigate where its limits may be. In this paper the pyrolysis model in FDS version 6.3.1 (the latest version at the time of writing) is adopted to simulate a series of small-scale tunnel experiments and compare the predicted HRR with measured values. The purpose of this comparison is to evaluate whether the FDS pyrolysis model is able to effectively predict the HRR for a tunnel fire with the influences from forced ventilation velocities. Wood-based cribs are chosen to represent the solid fuels in this study since the majority of the fuel sources used in large-scale tunnel experiments comprise of wood pallets [2]. In addition, wood cribs are commonly used as the fuel source in many small-scale tunnel experiments since they give consistent results for a given geometry and stick configuration [7]. This paper briefly describes a series of small-scale tunnel experiments in which cribs constructed of medium density fireboard (MDF) where burnt under a range of forced ventilation velocities.The paper describes the derivation of the MDF material properties which have been evaluated with a series of cone calorimeter experiments. The modelling of the source used to ignite the crib is described and finally the simulations of the tunnel experiments are presented from which HRR results have been obtained. SMALL-SCALE TUNNEL EXPERIMENTS The small-scale tunnel experiments were conducted in the medium-scale fire laboratory at the University of Canterbury. The tunnel was 0.365 m (W) × 0.26 m (H) × 11.9 m (L). The downstream end of the tunnel was connected to a circular duct for the measurement of the flue gases to obtain the HRR using oxygen consumption calorimetry. Longitudinal ventilation at different velocities was provided by a speed controlled fan that was installed 2.58 m upstream of the fire location. The main tunnel body was constructed of 0.9 mm thick SS304 sheets with 5 mm thick insulation blanket. The section in which the fuel source was located was 1.22 m in length and had fire resistant glazing along the front side for observation purposes. An insulated platform raised up 50 mm above tunnel floor was used to locate the fuel source and the platform was connected to a load cell so that the mass loss could be measured. Units: mm

(a)

Front side

Figure 1.

Left side (b)

(a) Tunnel dimensions, (b) crib geometry.

Photograph of crib

Cribs using sticks of MDF were used as fuel. The cribs were constructed with five layers of 15 mm thick sticks comprised of three 375 mm long-sticks and six 100 mm short-sticks equally spaced, with a resulting crib porosity of 0.8 mm. The average weight of the cribs was 1.44 ± 0.05 kg. Figure 1 shows the overall view of the small-scale tunnel geometry and the crib geometry. A more detailed description of the small-scale tunnel and fuel arrangement is given by Wang et al. [7]. Results and analysis of the experiments [7] found that the forced ventilation affected the fire spread as well as burning efficiency of the crib, and further affected the HRR. Figure 2 presents the measured effective heat of combustion for the MDF cribs at different forced ventilation velocities. When the air velocity was less than ~ 0.6 m/s the effective heat of combustion is found to be 12 MJ/kg which is the same as that obtained from the cone calorimeter experiments discussed later. Thereafter an increase in air velocity up to 1.2 m/s gradually increases the burning efficiency; however the burning efficiency then falls when the velocity exceeds 1.2 m/s. Effective heat of combustion (MJ/kg)

18

16

14

12 Cone calorimeter value for MDF: 12 MJ/kg 10 0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

Forced ventilation velocity (m/s)

Figure 2.

Effective heat of combustion at different forced ventilation velocities [7].

DETERMINATION OF MDF PROPERTIES Thermal properties The thermal properties of MDF required in the pyrolysis model are density, specific heat and thermal conductivity. Li et al. [8] have carried out a series of studies on the same MDF adopted in the smallscale tunnel experiments. In Li et al.’s [8] study, the specific heat and thermal conductivity for both virgin and charred MDF have been investigated as functions of temperature (T) and moisture content (MC) with the results shown in Table 1.

Table 1.

Thermal properties for virgin MDF and char MDF [8]

MDF, specific heat (J/kg/K) MDF, thermal conductivity (W/m/K)

cp ( dry ) = 25 . T + 1080

k030 = 4.86 ×10−8 ρ 2 + 4.63 ×10−5 ρ + 4.38 ×10−2 ∆k = 4.9 ×10−3 MC ×100 + (1.1×10−4 ρ + 4.3 ×10−5 MC ×100)(T − 30) k = k030 + ∆k

Char, specific heat (J/kg/K) Char, thermal conductivity (W/m/K)

cpchar (dry) = 3.7T + 547.86

0.09 W/m/K at ambient temperature and 7% increase per 10 K

Note: All symbols shown in equations can be found in the Nomenclature

Kinetic properties In the work of Li et al. [9], the kinetic properties (activation energy ( ), pre-exponential factor ( ) and reaction order ( )) of MDF have been analysed through an advanced computational searching method. They used four components to correspond to resin, hemicellulose, cellulose and lignin to represent MDF. The kinetic properties of each component were derived inversely from three differentiated thermogravimetric (DTG) curves at heating rates of 5, 20 and 60 K/min. However, the suitability of these kinetic properties to model decomposition behaviour in FDS was not part of the original research. In this study, the kinetic properties of the MDF are re-analysed by using a hand calculation method developed by Wang et al. [10] using the DTG experimental curves for the application of these properties to FDS modelling. This analysis consists of two steps: the first step adopts the Kissinger analysis method [11] to obtain a linear relationship based on Eq. (1) to derive and .  β ln 2  Ti , p 

  = − Ei + ln Ai R   E   RTi , p  i  

(1)

The same four-component scheme proposed by Li et al. [9] is adopted for this analysis. According to the relationships of ln(β/ #,% ) and 1/ #, for each component in Figure 3, the value of and for each component can be calculated based on the slope and the intercept of each line.

Figure 3.

% #,

Linear relationships of ln( /

) and 1/

#,

for each component in MDF [9].

The second step is to develop a mathematical model according to the decomposition rate presented in Eq. (2) in order to depict the reaction rate curves at different heating rates.

ri , j =

N , te

∑ (1 − v

s ,i

i =1 j =0

 E )ci Ai exp  − i  RT j 

 ni  (Yi , j ) 

(2)

The values of # and # can be determined through visual comparisons between the model and corresponding experimental results. In this analysis, an average of 20% residue for ( based on the TG experiment data in Li et al. [9] is considered. Due to the different mathematical expressions for the decomposition rate between the hand calculation and the FDS pyrolysis model, a further modification on the pre-exponential factor obtained from the hand calculation method is required in order to apply it in FDS. The details of the correction process are introduced in Wang et al. [10]. Table 2 summarises the final results of the kinetic properties for the four components of MDF for the application in the FDS6 pyrolysis model. Table 2

Kinetic properties for MDF.

Components Resin Hemicellulose Cellulose Lignin

(J/mol ) 1.30)105 1.57)105 1.92)105 1.96)105 #

(s-1) 6.24)1015 7.64)1012 6.78)1013 3.90)1019 #

#

5.0 3.0 0.9 8.0

#

0.10 0.42 0.36 0.12

FDS6 simulations of the TG experiments at the heating rates of 5, 20 and 60 K/min have been conducted using the kinetic properties in Table 2 by applying the methods described in Wang et al. [10]. Figure 4 is the comparison of DTG curves obtained from the simulations and from the experiments. The predictions for the DTG curves indicate that the kinetic properties derived here can satisfactorily represent the decomposition behaviour of the MDF.

Figure 4.

Comparison of DTG curves between simulations and experiments.

Evaluation through FDS cone calorimeter simulations In the TG simulations the heat transfer within solids is not included. However, when using the pyrolysis model to simulate a fire, the thermal properties have a significant influence on the heat transfer results and consequently the HRR predictions will be affected. Therefore the thermal properties and the kinetic properties of the MDF are evaluated and optimised using FDS simulations of the cone calorimeter experiments. Cone calorimeter ignition, mass loss and HRR results at incident heat fluxes of 25, 35 and 50 kW/m2 are adopted in this analysis. In the experiments, conditioned samples with dimensions of 100 mm ) 100 mm ) 18 mm thick were used. The MDF samples were dried at 60 ˚C for 12 h similar to the conditions used for the cribs in the small-scale tunnel experiments (the moisture content is be less than 3%). Experimental procedures were based on AS/NZ 3837:1998 [12] and each experiment was repeated three times. The average initial bulk density of the MDF samples was 710 kg/m3 and the remaining mass from the experiments gave an average bulk density of 180 kg/m3 for the char, i.e. 25% of the original mass remained similar to the 20% in Li et al. [9]. According to the measured HRR and mass loss data, the average effective of heat of combustion is 11 to 12 MJ/kg which is consistent with the value obtained from Li et al. [13].

300

Experiment 1

HRR (kW/m2)

(a) 25 kW/m2

Experiment 2 Experiment 3 Simulation

200

100

0 0

240

480

720 Time (s)

960

1200

1440

Experiment 1

300

Experiment 2

(b) 35 kW/m2

Experiment 3

HRR (kW/m2)

Simulation 200

100

0 0

360

720

1080

Time (s) Experiment 1

400

Experiment 2

HRR (kW/m2)

(c) 50 kW/m2

Experiment 3

300

Simulation

200

100

0 0

240

480

720

960

1200

Time (s)

Figure 5.

Simulations for cone calorimeter experiments at incident heat fluxes of (a) 25 kW/m2, (b) 35 kW/m2, (c) 50 kW/m2.

To evaluate the MDF properties for the FDS predictions of HRR, simulations for the three cone calorimeter experiments at each incident heat flux have been conducted. The sample was represented by the top surface of a solid block (100 mm ) 100 mm). The thermal conductivity and specific heat for the MDF and char were defined according to the values listed in Table 1 and the density as discussed previously. The back face was defined according to the properties of the insulation board used in the cone calorimeter experiments (density 336 kg/m3, thermal conductivity 0.07 W/m/K and specific heat 1.08 kJ/kg/K) and the side surfaces were defined as inert. The incident heat flux from the cone heater was specified as a constant external heat flux to the surface.

The corresponding predicted HRR curves from cone calorimeter simulation at the three incident heat fluxes are plotted in Figure 5 and are compared with the experimental results. The ignition delay has not been captured in the simulations with the first peak in the HRR occurring prior to the experiments for all three incident heat fluxes. Experimentally the peak HRR values following ignition increase from 200 kW/m2 to 280 kW/m2 as the incident heat flux increases from 25 kW/m2 to 50 kW/m2. The simulations obtain a lower peak HRR value of 175 kW/m2 at the 25 kW/m2 incident heat flux whereas the simulated peak HRR values in excess of the experiments for the other two incident fluxes. At 50 kW/m2 incident flux the simulated peak HRR is almost 400 kW/m2. All three predictions show comparable decay curves to the experimental curves. The values of the second peak HRR are consistently lower in the simulations than the experiments, while the predictions for the burning duration are less successful. The total energy predictions for the simulations under different heat fluxes are approximately 1.18 ± 0.02 kW/m2 ,which are consistent with the averaged experimental values at the different heat fluxes (0.99 ± 0.12 MJ for 25 kW/m2, 1.02 ± 0.07 MJ for 35 kW/m2 and 1.11 ± 0.03 MJ for 50 kW/m2). Although the simulation results cannot precisely match the experimental HRR curves the FDS results shown in Figure 5 demonstrate that FDS can capture the overall burning characteristics of the MDF samples in the cone calorimeter experiments. Simulations show an initial peak in the HRR, a decay curves that follows the initial peak with a corresponding growth and second lower peak in the HRR as the thermal wave reaches the back surface of the sample before the material is finally consumed. Lower incident heat fluxes result in a lower initial peak HRR and a longer burning duration similar to that observed in the experiments. SMALL-SCALE TUNNEL EXPERIMENT SIMULATIONS Basic settings The ‘Deardorff’ turbulence model and the simple chemistry, mixing-controlled combustion model in FDS have been used in this study. After a series of sensitivity analyses using cell sizes of 15 mm, 7.5 mm and 3.75 mm, it was found that numerical instability occurred at ~100 s when the 7.5 mm and 3.75 mm cell sizes were applied. In order to achieve a numerically stable calculation within a reasonable computational time as well as to give sufficiently accurate predictions of the small-scale tunnel fire simulations, a cell size of 15 mm is adopted in this study. For the solid phase, a stretch factor of one and cell size factor of 0.5 are applied to have a more uniform and smaller cell size for the solid phase calculations based on the studies in the previous work [4, 17]. According to the FDS user’s guide [14], the parameter ∗ can be obtained from Eq. (3), where the heat release rate, and , , , are the properties of ambient air.

  Q& D* =    ρ 0 c p ,0T0 g   

is

2/5

(3)

The values of interest in the small-scale tunnel experiments under different ventilation conditions were from 20 kW to around a peak value of 100 kW. The corresponding ∗ at the different heat release rates range from 0.20 m to 0.38 m. As suggested by Li and Ingason [15], a cell size of 20 cm is a reasonable value for the simulation of full-scale tunnel fires and the number of cells spanning the characteristic fire diameter in their simulations was 13. Zhang et al. [16] adopted 20 mm cell size to simulate the behaviour of a wood crib fire in a confined space and the spanning-cell number was 20. The uniform cell size of 15 mm adopted for the simulations in this work results in 13 to 25 cells spanning the characteristic diameter of the fire. Since the dimensions of the small-scale tunnel is 360 mm (W) ) 260 mm (H) ) 11900 mm (L), a domain with dimensions of 420 mm (W) ) 300 mm (H) ) 12645 mm (L) was used in the simulations to ensure sufficient volume to represent the entire tunnel and to accommodate the 15 mm cell size set-

up. In the simulations the insulated platform was represent as a solid block adjusted to a dimension of 300 mm (W) ) 495 mm (L) ) 45 mm (H) and the surfaces assigned the insulation material thermal properties . The tunnel walls were given the thermal properties of the insulation material used in the tunnel experiments and the thin steel sheets were omitted. The observation window glass was not specifically simulated due to the assumed minor thermal influence on the results. The ventilation fan was represented by a supply air vent at 2585 mm upstream away from the fuel location. The circular duct for the collection of flue gases was not modelled, while the downstream end of the tunnel was initially modelled as being directly open to ambient conditions. A Smokeview image of the simulated tunnel is shown in Figure 6 (a).

Ventilation fan

Crib

(a)

Tunnel end open to ambient

(b) Figure 6.

Simulation geometry set-ups for (a) tunnel, (b) crib geometry.

In the simulations, the obstruction function in FDS was adopted to construct the crib. The dimensions of the crib were defined as those used in the experiments except that the length of the short stick was modified to 105 mm so it corresponded to the 15 mm cell size. The representation of the crib geometry in FDS is shown in Figure 6 (b). For the application of the pyrolysis method, the decomposition reactions of the MDF had to be defined. A surface line in the FDS input file was defined to prescribe the boundary conditions for the obstructions that corresponded to the crib. The four different components (resin, hemicellulose, cellulose and lignin) were used to represent the fuel with the corresponding mass fraction of each component described in order to specify the kinetic properties, thermal properties and heat of combustion for each component. In FDS6, the burning efficiency can be controlled though the heat of combustion parameter. In order to investigate the influence of burning efficiency on HRR predictions, the heat of combustion of 12 MJ/kg based on the cone calorimeter results and the calculated heat of combustion obtained from the small-scale tunnel experiments [7] were applied to conduct corresponding simulations. Another important parameter is the thickness for heat conduction and in the version of FDS6 used in this study only one-dimensional heat transfer in solids is available. Therefore, the actual solid thickness could not be simply adopted to represent the thickness for the heat conduction calculation. In order to effectively characterise the heating conditions over four exposed faces of the sticks, an approximation has been to use ¼ thickness of wood stick (3.75 mm) which represents a scenario in which the wood stick is heated evenly over the primary surfaces excluding the ends. Ignition source In the experiments 20 ml of methylated spirits was placed in an 80 mm diameter circular pan as the ignition source for the cribs. The burning of this fuel lasted for approximately 120 s. By using a density of 789 kg/m3 and heat of combustion 26.8 MJ/kg [18] to 28.9 MJ/kg [19] for the fuel the corresponding steady-state heat release rate is calculated as 3.5 kW to 3.8 kW and the total energy content as 423 to 456 kJ.

In the simulations, the ignition source was simplified to a rectangular area with a dimension of 60 mm ) 90 mm with a 3.8 kW maximum HRR obtained from a 700 kW/m2 HRRPUA. To represent the burning of the methylated spirits the ignition source was set to linearly growth to 3.8 kW over the first 10 s and the value was kept constant for a further 110 s. By using this specification, it was found that the ignition source would not ignite the crib over the 120 s duration. In order to investigate the ability to ignite the crib, different burning times of 120 s, 240 s, 360 s and 1500 s (the full simulation time) were used to simulate the ignition source for the tunnel scenario with a 0.23 m/s forced ventilation velocity. The ignition source in this experiment lasted for about 120 s and the experiment was stopped at 1500 s when the crib residue was at a smouldering stage. The corresponding simulation results are plotted in Figure 7. Time: 1500 s Time: 500 s Time: 360 s Time: 240 s Time: 120 s Experiment 0.23 m/s

30 (a) 20

Time: 1500 s Time: 360 s Time: 500 s Experiment 0.23 m/s

1500 1200 Mass (g)

Heat Release Rate (kW)

40

(b)

900 600

10 300 0 0

Figure 7.

240

480

720 960 Time (s)

1200

1440

0 0

240

480

720

960

1200 1440

Time (s)

Predicted crib ignition and burning using different ignition source burning durations: (a) HRR, (b) mass.

From these results it can be seen that when the 120 s duration is used the HRR curve has an average value of 3.8 kW which lasts about 120 s and then it drops to zero, which suggests that the crib has not been ignited. When the ignition duration time is extended to 240 s, 360 s, 500 s and 1500 s, the crib is ignited so that the HRR values increase after the ignition source burns out. As seen in Figure 7 (a), the HRR for the 240 s duration ignition source is lower than the predicted results when longer ignition source durations are applied which indicates that the crib has not fully ignited in this case. For the cases of the 360 s, 500 s and 1500 s durations the predicted peak HRR values are similar while the burning period increases with the increase of ignition duration. Figure 7 (b) plots the mass consumption for the cases of the 360 s, 500 s and 1500 s ignition durations where the longer ignition duration, the more crib that is consumed. In order to represent the results from the small-scale tunnel experiments the influence of the ignition source on the crib needs to be minimised. Simulations found that a 360 s ignition source duration can effectively ignite the crib under different forced ventilation conditions from 0.23 m/s to 1.2 m/s when an effective heat of combustion (∆ ) of 12 MJ/kg is used for the MDF, while the duration of ignition source needs to increase to 480 s for the 1.6 m/s scenario. However as discussed previously, different values of ∆ can be obtained when different forced ventilation velocities are applied. Simulations found that a 240 s ignition source duration is sufficient to ignite the crib when the correspondingly higher values of ∆ are applied. Table 3 gives the ignition source duration times for the different forced ventilation conditions when a heat of combustion of 12 MJ/kg (referred to as the ‘Fixed HoC’ condition) is used and the revised durations (referred to as the ‘Modified HoC’ condition) when the modified heat of combustion values are applied.

Table 3

Ignition source duration and assumed effective heats of combustion for different forced

ventilation velocity simulations. Velocity (m/s) 0.23 0.40 0.68 0.90 1.20 1.60

Fixed HoC ∆

(MJ/kg)

12

Duration (s) 360 360 360 360 360 480

Modified HoC ∆

(MJ/kg) 12 12 13 14 17 15

Duration (s) 360 360 240 240 240 240

RESULTS AND DISCUSSION HRR Predictions The HRR predictions for the Fixed HoC and the Modified HoC groups and the corresponding experimental curves at different forced ventilation velocities are plotted in Figure 8 where the HRR generated from the ignition source has been subtracted from the HRR curves. The heat of combustion values calculated from the predicted values of HRR and mass loss rate are consistent with the set-up values in the FDS data file. The consistent heat of combustion results indicate that all of the available fuel is burned in the domain. For the simulations using the fixed heat of combustion conditions, as the forced ventilation velocity changes from 0.23 m/s to 0.4 m/s, the burning duration reduces from 800 s to about 650 s and the peak HRR increases from 18 kW to 27 kW. However, the predicted HRR curves demonstrate similar burning behaviour in terms of burning duration and peak HRR when the 0.4 m/s, 0.68 m/s, 0.9 m/s and 1.2 m/s forced ventilation velocities are examined. When the velocity increases to 1.6 m/s, the burning duration and the peak HRR are both less than those predictions at the 0.4 m/s to 1.2 m/s forced ventilation velocities. In general the predicted peak HRR value for each scenario are noticeably less than the results from the experiments. As shown in Figure 8, when the values for the heat of combustion are modified, the predicted HRR values largely improve when compared with the fixed value simulations. The predicted peak HRR values are similar to the experimental values at the 0.68 m/s, 0.9 m/s and 1.2 m/s forced ventilation velocities. However, the predictions of the fire growth and the entire burning duration at each forced ventilation velocity are unsatisfactory when compared to the experimental results. When the total energy release results are considered, there are significant differences between the experiments and simulations. As shown in Figure 9, less than half of the energy is predicted in the simulations for each forced ventilation scenario compared with the energy released in the corresponding experiment.

Simulation Fixed & Modified HoC 12 MJ/kg Experiment 0.23 m/s

80 60 40 20 0

Heat Release Rate (kW)

100

240

480 Time (s)

720

Simulation Fixed HoC 12 MJ/kg Simulation Modified HoC 13 MJ/kg Experiment 0.68 m/s

80 60 40 20

60 40 20

0

100

240

480 Time (s)

720

Simulation Fixed HoC 12 MJ/kg Simulation Modified HoC 14 MJ/kg Experiment 0.9 m/s

80 60 40 20 0

0 240

120

480 Time (s)

Simulation Fixed HoC 12 MJ/kg Simulation Modified HoC 17 MJ/kg Experiment 1.2 m/s

100

0

720

80 60 40 20 0

240

480 Time (s)

720

Simulation Fixed HoC 12 MJ/kg Simulation Modified HoC 15 MJ/kg Experiment 1.6 m/s

100 Heat Release Rate (kW)

0

Heat Release Rate (kW)

Experiment 0.4 m/s 80

0

960

Heat Release Rate (kW)

0

Simulation Fixed & Modified HoC 12 MJ/kg

100 Heat Release Rate (kW)

Heat Release Rate (kW)

100

80 60 40 20 0

0

240

480 Time (s)

Figure 8.

720

0

240

480 Time (s)

720

Predictions of HRR curves at different velocities using different values of heat of combustion.

25

Experimental total energy Fixed HoC Modified HoC

20

20.3

Revised fuel mass & extended domain 18 16

18.6

18.1 16.3

16

15 11.3

12.5

11.7

10.9

13.2

9.9

9.5

10 7.9 7.9

8.4

7.6 7.6

7.9

7.2 6.2 4.9

5

3.8

0 0.23 m/s

Figure 9.

0.4 m/s

0.68 m/s

0.9 m/s

1.2 m/s

1.6 m/s

Measured and predicted total energy released at different forced ventilation velocities using the different simulation set-ups.

Improvements to predictions In order to improve the predictions from FDS an investigation into the mass and energy consumption is carried out for the 0.68 m/s forced ventilation velocity scenario. Figure 10 shows MDF burning at different times along with the corresponding HRR and mass loss curves. The fire in the experiment reached its peak heat release rate at around 260 s when the crib was partially engulfed by the fire the some parts of the fuel surface was noticeably charred. At 420 s the crib has started to collapse and the burning showed in the figure involving the burning of both wood and char. The fuel was almost burnt away when time reached 480 s. In the experiment, the initial mass of the crib was measured as 1.4 kg and the remaining mass of the residue material (char and ash mixture) after burning was measured as ~ 0.2 kg, which was about 14% of the original fuel mass.

60

120 s

Mass (g)

1050 40 700 20

350

0

Heat release rate (kW)

1400

260 s

0 0

120 240 360 480 600 720 Time (s)

420 s

480 s

Figure 10. Experimental mass loss and HRR curves and crib burning at different times under 0.68 m/s forced ventilation velocity conditions. When the crib with the geometrical form was applied to the simulations, the available fuel surface for

burning could not be as large as in the experiments due to the overlapping sections of the sticks. FDS uses the surface properties of one side obstruction only when two obstructions overlap each other [14]. As a result of the available surface area (0.51 m2), thickness (3.75 mm) and density (710 kg/m3) applied in the simulations, the available mass was 1.36 kg rather than 1.40 kg. In addition, 20 % of MDF was set to convert to char in the simulations, which means that no combustion reaction occurs for this component proportion. Therefore the available burnable fuel mass in the simulations was less than that in the experiments. In order to obtain a comparable fuel mass between FDS and the experiments some modifications were made to the 0.68 m/s forced ventilation simulation case to re-assess the results. The thickness of sticks was increased from 3.75 mm to 3.90 mm to compensate for the ‘missing’ fuel mass due to the overlapping area and also in order to maintain the same fuel density and crib geometrical shape as the experiments. No residue was considered in this case, which means all of the exposed fuel was available to be consumed in the simulation. Thus a total of 1.4 kg of fuel was available as a result of these modifications. The revised simulation results are plotted in Figure 10 along with the experimental data and the previous simulation results using the 3.75 mm stick thickness. 100

Experiment 0.68 m/s

Experiment 0.68 m/s

1400

Revised fuel mass

80

Revised fuel mass & extended domain

Modified HoC

1200

Revised fuel mass

46% 1000

60

Mass (g)

Heat Release Rate (kW)

Modified HoC

40

Revised fue massl & extended domain

800 65%

600 400

20 200 0

0 0

120

240

360 Time (s)

480

600

720

0

120

240

360 480 Time (s)

600

720

Figure 11. Simulation results with the modification of the available fuel mass for (a) HRR curves, (b) mass loss curves. As shown in Figure 11 (a), the predicted shapes of the fire growth curve are similar for the Modified HoC and the revised fuel mass cases, albeit with a delayed time shift when compared to the experiment. The increase in fuel mass improves the prediction of the peak HRR, where a value of 62 kW is obtained for the revised fuel mass case (~60 kW was obtained in the experiment) compared with 47 kW with the previous Modified HoC case. The consumption of the fuel has improved from 46 % to 65 % in the mass loss curves shown in Figure 11(b). However, the simulation results still do not fully represent the experimental results in terms of the peak values and HRR curve shapes. There is around 35 % of the fuel remaining even though the revised fuel mass case simulation had been set up to fully consume it.

16

Revised fuel mass Revised fuel mass and extended domain

Heat of Combustion (MJ/kg)

15

14

13

12

13 MJ/kg

11

10 0

120

240

360

480

600

720

Time (s)

Figure 12.

Results for heat of combustion from the simulation for the 0.68 m/s forced ventilation velocity condition.

Further investigation of the revised fuel mass case simulation shows that the predicted heat of combustion values (the predicted HRR / the predicted mass loss rate) are not fully consistent with the set-up value of 13 MJ/kg, as shown in Figure 12. The heat of combustion obtained from the simulation reduces to about 11.5 to 12 MJ/kg between 280 s and 360 s. This reduction in the heat of combustion suggests the mass loss rates in the simulation were higher during that period and that some fuel is lost without contributing any energy to the burning. In order to investigate this, an extra mesh with 15 mm cell size was used at the end of the original mesh with sufficient height and length to allow the unburned fuel to burn. The predicted heat of combustion values based on the extended domain is plotted in Figure 12 and the results of mass loss rate and HRR are plotted in Figure 11. The heat of combustion values over 280 s to 360 s are improved after the changes in the domain as shown in Figure 12. The HRR curve for the revised domain shown in Figure 11 (a) also demonstrates a higher peak HRR values over this time period, while no change is shown on the mass loss curve. Based on the improvements obtained above for the 0.68 m/s forced ventilation velocity condition, the modifications to the fuel mass and domain were applied to the other ventilation scenarios. The corresponding results are shown in Figure 13 which illustrates the changes in the total energy predictions after the increase of the available fuel mass, however, the predicted HRR curves at different velocities are still not ideal. The initial fire growth phase at each forced ventilation velocity is not improved, where significant ignition delays and slower fire growth rate are still found at 0.23 m/s, 0.4 m/s and 0.68 m/s and earlier ignition and faster fire growth rates are obtained for the higher 1.2 m/s and 1.6 m/s forced ventilation velocities. It is noted that the cell size adopted in the simulations is relatively coarse in terms of the fuel dimensions. Due to the limits of resources, the simulations with a finer cell size are not carried out. However, the ignition delay might be improved with the use of a finer cell size. The predicted total energy, shown in Figure 9, is still less than the experimental value at each velocity as the areas under the curves from the simulation results are less than experimental results.

Heat Release Rate (kW)

Modified HoC 12 MJ/kg

80

Revised fuel mass & extended domain 12 MJ/kg

60 40 20

240

100

480 720 Time (s)

Revised fuel mass & extended domain 12 MJ/kg

60 40 20

Experiment 0.68 m/s Modified HoC 13 MJ/kg

80

Resived fuel mass & extended domain 13 MJ/kg

60

0

960

40 20

240

480 Time (s)

720

960

Experiment 0.90 m/s

100 Heat Release Rate (kW)

0

Heat Release Rate (kW)

Modified HoC 12 MJ/kg 80

0

0

Modified HoC 14 MJ/kg 80

Revised fuel mass & extended domain 14 MJ/kg

60 40 20 0

0 240

480 Time (s)

720

240

480

720

Time (s)

160

Experiment 1.2 m/s

140

Modified HoC 17 MJ/kg Revised fuel mass & extended domain 17 MJ/kg

120

0

100 80 60 40

Experiment 1.6 m/s

100

Modified HoC 15 MJ/kg Heat Release Rate (kW)

0

Heat Release Rate (kW)

Experiment 0.4 m/s

100 Heat Release Rate (kW)

Experiment 0.23 m/s

100

80

Revised fuel mass & extended domain 15 MJ/kg

60 40 20

20 0

0 0

Figure 13.

240

480 Time (s)

720

0

240

480 Time (s)

720

HRR predictions using different available fuel mass set-ups.

In terms of the predictions of peak HRR, values at different forced velocities are within 20% of the experimental data as shown in Figure 14 except for the result obtained from the ‘revised fuel mass & extended domain’ at 1.2 m/s case.

Peak HRR (kW)

160

Experimental data

140

Revised fuel mass & extended domain

120

Modified HoC

100 80 60 40 20 0 0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

Velocity (m/s)

Figure 14.

Peak HRR prediction comparison.

Based on the above discussion, the predictions of peak HRR are improved by the inclusion of the impact of forced ventilation velocity on HoC and the extra fuel mass in the simulations. However, the overall predictions of the HRR curves in terms of total energy and the fire growth stage for the smallscale tunnel fires at different forced ventilation velocities through the application of the FDS6 pyrolysis model are limited.The possible reasons for the limitations in the FDS results are that firstly, the assumption of one-dimensional heat conduction in the solid may limit the heat transfer and the increase in temperatures therefore hindering the pyrolysis reactions, slowing down the ignition of the fuel and so reducing the growth of the fire. Secondly, the specified pyrolysis reactions are not able to represent the decomposition reactions in the presence of air. The pyrolysis rate adopted in this study was under nitrogen environment, which represents the scenarios that decomposition reactions occur under flame without the presence of oxygen. When forced ventilation conditions are present the heat may be imposed on the downstream side of the fuel. Therefore, the fuel surface may be heated first without being covered with flame and the pyrolysates may mix with air before reacting.

CONCLUSIONS This work applies the pyrolysis model in FDS to simulate a series of small-scale tunnel experiments under different forced ventilation velocities in order to examine a practical approach to predict HRR in tunnel fires that includes the impact of the ventilation conditions. Modelling of this scenario is complex and the results presented in this paper show that there is a range of challenges involved that suggest the approach is not ready for routine use in a design environment. To apply the pyrolysis model, the MDF material properties are investigated and evaluated through simulations of TG and cone calorimeter experiments. It is found that the kinetic properties for MDF obtained in this work can satisfactorily represent the decomposition behaviour in FDS by reproducing differentiated thermogravimetric (DTG) curves obtained from experiments at various heating rates. The results from simulations of the cone calorimeter are not wholly comparable to the experimental results, however, predictions of peak HRR, burning periods and total energy release are able to reflect the overall burning behaviour of the MDF. Based on the simulation results for the tunnel experiments, some factors are found to have a significant influence on the predictions: the effect of the forced ventilation on the burning efficiency needs to be accounted for when using the pyrolysis model; the available fuel mass for burning is

affected by the thickness of the crib sticks and the available surface area; the use of an appropriate domain is important in order to allow the unburned fuel to be completely consumed. However, even with these factors included, the match between the FDS predictions of HRR and the small-scale tunnel experiments has its limitations. However, there are still many factors that have not been investigated in this work such as the decomposition of char and burning at the late stage, collapse of the fuel (increasing burning area), the influence of air velocity on decomposition reactions and applying bulk density to reduce overlapping errors. The associated work by Wang et al. [7] demonstrates how the impacts of ventilation velocities on burning efficiency can be generalised. In order to assess the importance of the existing differences it would be useful to conduct a parameter study using the findings from this paper. The future study could assess the simulation predictions of a selected full-scale experiment, similar to the earlier work of Cheong et al. [2]. It would also be advantageous to use the latest version of FDS and to apply a finer cell size than used in this study. ACKNOWLEDGEMENTS The primary author gratefully acknowledges the University of Canterbury (UC), New Zealand for the UC Doctoral Scholarship award; The continued support of the UC fire engineering programme by the New Zealand Fire Service Commission is gratefully acknowledged. REFERENCES 1.

Cheong, M.K., Assessment of Vehicle Fire Development in Road Tunnels for Smoke Control Ventilation Design, PhD thesis. 2009, University of Canterbury.

2.

Cheong, M.K., Spearpoint, M.J., and Fleischmann, C.M. Design fires for vehicles in road tunnels. in 7th International Conference on Performance-Based Codes and Fire Saftey Design Methods. 2008. Auckland, New Zealand.

3.

Ingason, H. and Lönnermark, A., Heat release rates from heavy goods vehicle trailer fires in tunnels. Fire Safety Journal, 2005. 40(7): p. 646-668.

4.

Pau, D.S.W., A comparative study on combustion behaviours of polyurethane foams with numerical simulations using pyrolysis models, PhD thesis, in Civil and Natural Resources Engineering 2013, University of Canterbury: Christchurch, New Zealand.

5.

Matala, A. and Hostikka, S. Pyrolysis modelling of PVC cable materials. in 10th International Association for Fire Safety Science (IAFSS). 2011. Maryland, USA.

6.

Li, Y.Z., CFD modelling of fire development in metro carriages under different ventilation conditions. 2015, SP Technical Research Institue of Sweden.

7.

Wang, X.Y., Spearpoint, M.J., and Fleischmann, C.M., Investigation of the effect of tunnel ventilation on wood crib fires through small-scale experiments. Fire Safety Journal 88 (2017) 45-55.

8.

Li, K.Y., Fleischmann, C.M., and Spearpoint, M.J., Determining thermal physical properties of pyrolyzing New Zealand medium density fibreboard (MDF). Chemical Engineering Science, 2013. 95(0): p. 211-220.

9.

Li, K.Y., Huang, X.Y., Fleischmann, C., Rein, G., and Ji, J., Pyrolysis of medium-density fiberboard: optimized search for kinetics scheme and parameters via a genetic algorithm driven by Kissinger's method. Energy & Fuels, 2014. 28(9): p. 6130-6139.

10.

Wang, X.Y., Fleischmann, C., Spearpoint, M.J., and Li, K.Y. A simple hand calculation method to estimate the pyrolysis kientics of plastic and wood materials. in 10th Asia-Oceania Symposium on Fire Science and Technology. 2015. Japan.

11.

Kissinger, H.E., Reaction Kinetics in Differential Thermal Analysis. Analytical Chemistry, 1957. 29(11): p. 1702-1706.

12.

ASNZS3837-1998: Method of test for heat and smoke release rates for materials and products using an oxygen consumption calorimeter. 1998, SAI Global Limited and Standards New Zealand. p. 54.

13.

Li, K., Pau, D.S.W., Wang, J., and Ji, J., Modelling pyrolysis of charring materials: determining flame heat flux using bench-scale experiments of medium density fibreboard (MDF). Chemical Engineering Science, 2015. 123(0): p. 39-48.

14.

McGrattan, K., Hostikka, S., McDermott, R., Floyd, J., Weinschenk, C., and K.Overholt, NIST special publication 1019 Fire Dynamics Simulator user’s guide version 6. 2015: National Institute of Standards and Technology.

15.

Li, Y.Z. and Ingason, H., Numerical Simulation of Runehamar Tunnel Fire Tests, in 6th International Conference 'Tunnel Safety and Ventilation'. 2012: Graz. p. 203-210.

16.

Zhang, S., Ni, X., Zhao, M., Feng, J., and Zhang, R., Numerical simulation of wood crib fire behavior in a confined space using cone calorimeter data. Journal of Thermal Analysis and Calorimetry, 2015. 119(3): p. 2291-2303.

17.

Wang, X., Fleischmann, C.M., and Spearpoint, M.J., Parameterising study ot tunnel experiment materials for application to the Fire Dynamics Simulator pyrolysis model. Journal of Fire Sciences, 2016.

18.

Drysdale, D., An introduction to fire dynamics Second Edition 2011, Hoboken, N.J: Wiley.

19.

Thermal-FluidsCentral. Heat of Combustion [cited 2016 13th Sep]; Available from: http://www.thermalfluidscentral.org/encyclopedia/index.php/Heat_of_Combustion.

The authors disclose that there is no potential conflict of interest in this research.