effects of combustible stacking in large compartments

Effects of Combustible Stacking inLarge Compartments

by

Filippo Gentili, Luisa Giuliani and Franco Bontempi

Reprinted from

Journal of Structural FireEngineeringVolume 4 · Number 3 · September 2013

Multi-Science PublishingISSN 2040-2317

Volume 4 · Number 3 · 2013

187

Effects of Combustible Stacking inLarge Compartments

Filippo Gentili1, Luisa Giuliani2 and Franco Bontempi11School of Engineering, Sapienza Università di Roma, Via Eudossiana 18,

00186 Rome2Civil Engineering Department, Technical University of Denmark,

2800 Lyngby, Denmark E-mail: [email protected]

ABSTRACTThis paper focuses on the modelling of fire in case of various distributions ofcombustible materials in a large compartment.

Large compartments often represent a challenge for structural fire safety, because oflack of prescriptive rules to follow and difficulties of taking into account the effect of nonuniform distribution of the combustible materials and fire propagation.

These aspects are discussed in this paper with reference to an industrial steel building,taken as case study. Fires triggered by the burning of wooden pallets stored in thepremises have been investigated with respect to different stacking configurations of thepallets with the avail of a CFD code. The results in term of temperatures of the hot gassesand of the steel elements composing the structural system are compared with simplifiedanalytical model of localized and post-flashover fires, with the aim of highlightinglimitation and potentiality of different modelling approaches.

Keywords: Structural fire safety; CFD modelling; steel industrial hall; fire propagation;distribution of combustible; ventilation conditions

1. INTRODUCTIONThe design process for structural fire safety can be divided into three main steps [1]: first the fire actionhas to be modelled, then the temperature of the elements has to be computed with a heat transfer model,and finally the system response has to be evaluated with a structural model.

More or less sophisticated models are available at each design step. This paper focuses in particularon the advanced design of the fire action and heat transfer by means of Computational Fluid Dynamics(CFD) models, with the aim of highlighting limits and potentiality of this design approach.

1.1. Fire in Large CompartmentsAdvanced fire models obtained with the avail of CFD investigations become particularly important ifgreater design flexibility is desired or untraditional architectural or structural solutions are employed.In this respect, the presence of large compartments is often required in industrial halls and publicbuildings and is nowadays also a desired characteristic of many offices and residential premises.Nevertheless, prescriptive design and verification methods for structural fire safety can only be appliedto compartments not exceeding specific dimensions, while the design of larger compartments oftenrepresents a challenge for architects and structural engineers.

In particular, even if large halls are typically neither heavily cluttered nor densely furnished, thedistribution of goods or furniture may be strongly inhomogeneous, leading to possible concentration ofthe fuel load, whose effects need to be carefully investigated.

Corresponding author: Filippo Gentili, Sapienza University of Rome, School of Civil Engineering, Via Eudossiana 18, 00184 Rome, Italy;

Tel.: +39 06 4458 5224;

In order to highlight the effects of combustible stacking in large compartment, an industrial steelstructure is taken as case study and investigated with respect to a fire triggered by the burning ofwooden pallets stored in the premises. Problematic issues of the CFD modelling concerning thepresence of uncertainties, the objectivity of the solution, and the reduction of computational onus arepresented and discussed. The advantage of more realistic simulations that take into account the effectsof fire propagation and the distribution of the combustible are also stressed out.

The outcomes of the CFD investigations in terms of temperatures of the hot gasses and temperaturesof the steel elements are compared with the results obtained by using simpler analytical models such aslocalised and parametric fires. In particular, it is shown that a high spatial variability of temperaturescharacterizes some type of large compartment fires. In those cases, post-flashover fires, which assumea uniform distribution of the temperature along the compartment, can be not very representative of thereal phenomenon and can possibly lead to an underestimation of element temperatures.

1.2. State of the Art and Current Design PracticeThe use of CFD models is fire safety has been originally developed and mostly used for the predictionof smoke movements [2], but recent advances is performance-based design have led to an increasinginterest and research on the use of CFD models for structural fire design [3], [4], [5]. Nevertheless, theuse of these advanced fire models, even if contemplated by several regulations [6] and guidelines [7]and despite the increasing use of advanced numerical program in building design and the availabilityof more and more powerful computers, remains quite limited in the current design practice, which ismostly based on the use of post-flashover fire model such as parametric fires and, for a large part,nominal fire curve such as the standard ISO834 fire curve [6].

One reason of that can be found in the complexity of advanced CFD investigations, which require acareful calibration and validation of the models (as better discussed in the last part of this paper) andtrained engineers to carry out the analyses and correctly interpret the results [8]. Another reason lies inthe common belief that the assumption of a severe post-flashover such the one described by standardfire curve would necessary determine the highest temperature in the elements and therefore to aconservative structural design.

Even if true in most cases, this assumption is not valid in general and current design practice maylead to an unsafe design of structural elements, as evidenced by several cases of structural collapses [9],[10], [11] and recently pointed out by several independent studies [12], [13], [14]. In particular, properconsideration of the fire duration and the effective fuel load density seems to be fundamental for theassessment of realistic element temperatures. Both aspects present difficulties in case of a fire in a largecompartment, due to the unlikelihood of flashover and uniform fuel load density. The study of differentgrade of combustible staking seems therefore of particular interest with respect to the structural designof large compartments, such as exhibition halls [15] and hangars [16].

2. CASE STUDYA steel industrial hall has been considered as case study. The structural system considered has beentaken from a report [17], where FEM investigations of the structural response of the hall are presentedwith respect to a standard fire.

Advanced investigations of the structural response to fire [18] are not within the scope of this paper,which focuses mainly on aspects related to the modelling of the fire for structural design.

2.1. Description of the StructureThe configuration of the structural system and the properties assumed for the compartment and thecombustible are described below and summarized in Tab. 1, Tab. 2 and Tab. 3 with respect to theproperty of the compartment, of the combustible and of the structural elements respectfully.

2.1.1. Geometry of the CompartmentThe premises, whose geometry is shown in Fig. 1, consist in a large hall, 40 m long and 30 m wide andcovering a floor area of 1200 m2.

188 Effects of Combustible Stacking in Large Compartments

Journal of Structural Fire Engineering

The structural system is composed by 5 main frames, connected by 9 transversal purlins sustaininga steel-concrete deck. The main frames consist of 2 bays, spanning 20 m between 5 m height columns.The beams have a pitched configuration, so that a maximum height of 5.5 m is reached incorrespondence of the mid-span of each bay, while the average height of the hall is 5.25 m.

The total area of the enclosure (floor, ceilings and walls) results therefore to be equal to 3135 m2.

2.1.2. Amount and Properties of the CombustibleThe premises are devoted to storage of goods and are assumed to be empty at the time of fire. Only thepresence of 320 wooden pallets (Fig. 2) used to support and transport goods is considered in thepremises and the effects of different disposition and stacking of the pallets are investigated.

Filippo Gentili, Luisa Giuliani and Franco Bontempi 189


Table 2. Properties of the combustible

CombustibleFuel Proper Fire growing rate α 0.156 kJ/s3

Calorific value of combustible H 17.5 MJ/kgWeight of combustible G 4800 Kg

Fuel Load Total fuel load Q 84000 MJFuel load density (floor) qf 70 MJ/m2

Fuel load density (enclosure) q 27 MJ/m2

Table 3. Properties of the structural system

Structural PropertiesSteel Density ps 7850 kg/m3

Specific heat cps 450 J/(kg .K)Resultant emissivity εr 0.5 –

Profiles Purlin section factor Ap/Vp 192 m–1

Rafter section factor Ab/Vb 134 m–1

Central column section factor Ac/Vc 162 m–1

Table 1. Properties of the compartment

Enclsure PropertiesSize Width B 30 m

Length L 40 mHeight (average) H 5.25 m

Floor area Af 1200 m2

Enclosure area At 3135 m2

Openings Average opening height hw,av 2.8 mTotal opening area Aw 103.7 m2

Air Flow Factor AF 173.49 m5/2

Opening factor O 0.055 m0.5

Thermal Inertia Gypsum surface A1 1426 W .s0.5/(K .m2)Gypsum thermal inertia b1 762 m2

Concrete surface A2 1’920 m2

Concrete thermal inertia b2 1200 W .s0.5/(K .m2)Thermal Inertia b 1017 W .s0.5/(K .m2)

Each pallet has dimensions 1.2 m × 1.2 m × 0.15 m and a weight of 15 kg. The pallets are assumedto be made of wood with a calorific value of 17.5 MJ/kg, so that the total amount of fuel load in thepremises results to be equal to 84000 MJ, as reported in Tab. 2.

Wooden pallets are typically stored in stacks of different height, so that each pallet stack can beconsidered similar to a firewood crib [19] and a constant plateau of the heat release rate (HRR) can beseen if the stack is higher than 0.5 m (SFPE).



y

x

20 m

20 m

40 m

5 m

DoorHeight: 3.6 mWidth: 4.8 m

WindowHeight: 1.2 mWidth: 3.6 m

5.5 m

7.5 m

z 30 m

Figure 1. Industrial hall considered as case study.

1.2

1.2

1.2

1.2

0.15

hs

Pallet height

Pallet area

Pallet weight

No. pallets in the hall

Moisture content

Stack height 3.0 1.5

10

3.86

32

178

1.2

8

3.28

40

189

0.6

4

2.10 MW

MW

80

242

m

---

---

20

6.81

16

157

No. of pallets in a stack

HHRmax of a stack

No. stacks in the hall

HRRmax, tot

0.15

1.44

15

320

4

m

m2

kg

---

%

Figure 2. Geometry (top) and property (bottom) of a single pallet (left) and of a palletstack (right).

The maximum heat release rate (HRRmax) of a pallet stack can be therefore calculated with theexpression proposed by Krasner [20] and reported in Eq. 1, where the HRR of a stack of pallets is givenper unit area of the floor occupied by the stack (indicated as HRRs,max). The HRRmax calculated for differentstack height is visible on the right of Fig. 2: for a stack of 3 m, the HRRmax results to be equal to 9.81 MW,which is consistent with the value reported by La Malfa [21], who suggests a value of 6.81 MW/m2 ofoccupied floor area for a 3 m high stack of pallets; the value is also quite close to the value of 7 MWreported by Karlsson & Quintiere [22] for a 3m high pallet stack.

HRRs,max = 919.(1 + 2.14.hp).(1–0.03.M) (1)

where:hp indicates the height of a stack of palletsM represents the moisture content of the wood

2.1.3. Ventilation of the CompartmentFour doors and eight windows have been assumed to be placed with a symmetrical disposition on theexternal perimeter of the hall: in particular, a 4.8 m wide and 3.6 m high door has been placed in the centreof each external pitched bay, while a 3.6 m wide and 1.2 m high window has been placed at 3.6 m fromthe ground in the centre of each bay of the secondary frames in the transversal direction.

In case all doors and windows are assumed to be open during the fire, as assumed in the followinginvestigations, the opening factor of the premises results to be O = 0.055 m1/2 and the limit heat releaserate due to maximum oxygen income (HRRlim) is equal to 306.3 MW, which ensure a well-ventilatedcondition for the development of the fire.

2.1.4. Materials of the Enclosure and of the Structural SystemThe structural system is a framed system of steel beams and columns, with the profiles shown in Fig. 3.In particular, the purlins are realized with hot rolled steel S235, while the rafters with hot rolled steel S355.All beams are considered to be exposed to fire on three sides and insulated on the top flange by thepresence of the roof deck.



PurlinSection: HEA 160

Material: S235

y

xz

RafterSection: IPE 500

Material: S355

ColumnSection: IPE 450

Material: S355

Figure 3. Structural system and steel element profiles.

For the purpose of thermal inertia calculations, the walls are assumed to be made of gypsum whilethe floor and the ceiling are considered to be made of concrete, as shown in Tab. 1.

2.2. Simplified Fire ModelsAnalytical models for describing the temperature evolution of the hot gasses and of the elements in acompartment can be synthetically distinguished in pre- and post-flashover models.

Pre-flashover models can be used for describing localized fires in case of either a low flame [23] ora flame impinging the ceiling [24]. For the purpose of structural fire safety design however, generallyonly post-flashover conditions are assumed. Simplified models describing post-flashover fires refer toeither nominal fires, such as the standard ISO834 curve, or to parametric fire curves, characterized bya heating phase, a peak and a cooling phase.

Parametric fire curves were first introduced by the Swedish school [25] for describing post-flashovercompartment fires. The compartment was assumed to have a standard thermal inertia and an air inflowcapable of limiting the burning rate of the combustible. The method led to a graphical formulation oftemperature-time curves of the hot gasses and of the steel elements for different opening factors of thecompartment and fuel load densities of the combustible material. Further refinement of the model alsoallowed considering different values of the thermal inertia and following studies [26], [12] based on thesame approach led to parametric curves described by analytical expressions. In particular, in the Danishregulation [27] a unique expression describing both the heating and cooling phase is used (Eq. 2):

(2)

where:Tg is the temperature of the hot gasses in °Ct is the time in minb is the thermal inertia of the compartment in W .s0.5/(K .m2)O is the opening factor of the compartment in m0.5

q is the fuel load density in MJ/m2 of enclosure surface of the compartmentThe parametric curve indicated by the Eurocodes [28] instead uses two different expressions for

describing a natural fire: the heating phase is described with a monotonically increasing temperature,while the cooling phase is represented by a linear decrement of the temperature, whose gradient varieswith the duration of the heating phase.

In this model, the assumption of ventilation controlled fires is removed and a different expression ofthe fire curve can be used in case of well ventilated fires. A punctual comparison of design resultingfrom the use of the Danish or of the Eurocodes parametric fires is presented in Giuliani et al. [13]. Tothe purpose of this paper however, it seems relevant to point out that the use of the EN parametric firewith linear cooling may lead to a strong reduction of the fire severity: this can be observed on the leftof Fig. 4, where the parametric fires calculated for the considered case study are compared with thestandard fire; on the right side of Fig. 4 the temperature curve of the rafters related to the Danishparametric fire is instead reported, which results to be heated up to 400 °C during the fire.

It has to be pointed out however that the use of the EN parametric fire is recommended for fuel loaddensity calculated with respect to the enclosure are not trespassing the lower limit of 50 MJ/m2 and forcompartment not exceeding 500 m2 of floor area and 4 m of height. The same prescriptions on thecompartment size also apply to other parametric fire such as the Danish fire curve, limiting theirapplicability to small compartments. This is due to the fact that parametric fire curves assume aflashover-like fire with uniform temperature in the compartment, condition which hardly will occur inlarge hall and atria. These limitations however are often disregarded in the practice, both because noadditional information can be found in the code for a simple modelling of fire in large compartmentsand because other codes and literature references [29], [30] indicate that these limitations can be safely

T tt

t tg =

+ +( )

−( )

. log . .

. . / ma

20 345 8 1

1 0 04

10 Γ

xx

. max

/ .

/( )

=( )( )

=3 5

20 04

1160with :

OandΓ

bt 77 8 10 3. . − q

O



ignored at the expenses of a less economic but conservative design. In particular, thanks to thesymmetry of the structure considered as case study, exact results can be obtained by referring to 1/4 ofthe compartment, as indicated by Hertz [12], in case a flashover is assumed in the compartment. Byusing this method, the floor area of the compartment can be reduced to a size compliant with theprescriptive limits. Always according to the same document, conservative results are obtained forcompartment higher than 4 m, provided that all the openings above 4 m will be ignored.

The case study presented here, refers to a compartment which is symmetric about the horizontal andtransversal centreline and could be therefore reduced to 1/4 of the original floor area, which becomesequal to Af’ = 300 m2.The resulting opening factor of the reduced symmetric compartment will be0.045. The parametric curves and the corresponding element temperatures shown in Fig. 4 havetherefore been calculated with reference to 1/4 of the compartment and to this reduced opening factor,as according to the procedure described in Hertz [12].

2.3. Advanced Fire ModelsThe limitations of parametric fire curves can be overcome by using more advanced fire models. Inparticular, the temperature evolution of hot gasses and elements can be obtained by CFD investigationsas a variable of time and space, and possible non uniform distributions of the temperatures in largecompartments can be highlighted.

Advanced fire models allow taking into account the ignition of secondary objects that represents acritical event in the fire development. Radiation from the flame, which depends on the configurationfactor, assumes considerable importance for the ignition of nearby items. Many parameters such as firelocation, size and shape of flame, and material proprieties affect the phenomenon [31]. Among thepossible criteria for determining the ignition in a code, one based on flux-time product (FTP) providesresults in good agreement with experiments [32]. In this paper, for sake of simplicity, the ignitiontemperature of pallets (275 °C) was considered as a criterion of ignition.

Non uniform distributions of temperatures in a compartment stem either from non-uniformdistribution in space of the combustible or from non-uniform burning time of the combustible:i. in the first case, concentration of the fuel load in a relatively small area of the compartment could

lead to an underestimation of the flame height and of the temperatures above:ii. in the second case, a slow propagation of the fire would determine a longer fire duration than in

case of a flashover-like burning of the combustible is assumed.Both situations are likely to occur in large compartments (typically atria, auditoria, warehouses,

industrial halls, etc.), where the need of free stream of people or goods requires a low density offurniture and encumbering materials, which can be either piled up, leading to fuel load concentration(i) or placed far one from the other, leading to slow propagation of the fire (ii). Either way, the



ISO 834

DK parametric

EN parametric

800

600

400

200

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

Gas temperature

Rafter temperature

800

600

400

200

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

Figure 4. Comparison between nominal (ISO834) and parametric (DK and EN) gastemperature (left) and temperature of the main beams according to the DKparametric fire curve.

assumption of a uniform temperature in the compartment may result in an underestimation of theelements temperatures and possible structural failures.

Those aspects are better highlighted in the following, where the results of CFD investigations arepresented, which refer to fires triggered by different distribution of the combustible materials. Theinvestigations presented have been carried out with Fire Dynamic Simulator (FDS), which is a fieldCFD code released by NIST [33].

2.3.1. Fire ScenariosAn overview of the fire scenarios considered for the investigations is reported in Fig. 5 with respect to1/4 of the compartment.

In every scenario, the pallets are considered to be staked in the centre of the hall. The number of woodenpallets piled up in each stack varies from a maximum of 20 in scenario C to a minimum of 4 in scenario Dand the number and height of the stacks in the hall varies accordingly (Fig. 2), as the total number of palletsin the hall is constant: in particular, a maximum height of 3 m is reached by the stacks considered forscenario C and a minimum height of 0.6 m is reached by the stacks considered in scenario D.

This dispositions lead to different extensions of the floor area involved in the fire, but also todifferent values of the HRRmax during the fire, which, as explained above and summarized in Fig. 2,varies with the stack height. In all scenarios, the mutual distance between the stacks is constant andequal to 1.2 m and the fire is assumed to trigger always in the 4 central stacks. As a consequence, thedifferent values of HRRmax are expected to determine a different speed of the fire propagation.

In the following sections, the outcomes of the investigations in term of temperatures of the fire andof temperature on the elements are presented, with respect to different locations within the compartment.

In particular, the temperatures of the hot gasses are referred to 8 thermocouples TC-1 to TC-8 placedat 4.5 m from the ground, while the temperatures of the elements have been measured with the adiabaticsurface temperature method on the points AST-1 to AST-6. The position of the measurement device isvisible in Fig. 6 (right) and the coordinates of the points as reported as table in the same figure (left).



Scenario A10 stacks1.2 m high

Scenario D20 stacks0.6 m high

Scenario C4 stacks

3.0 m high

Scenario B8 stacks

1.5 m high

Figure 5. Fire scenarios represented on 1/4 of the model.

The outcomes of all considered scenarios are compared and three out of four scenarios are discussedin detail in the following, as representative of the most significant fire phenomena.

Scenario A: uniform distribution of temperaturesThis scenario considers that the 320 wooden pallets have been piled up in group of 8, forming 40 stacksof height 1.2 m. The pallet stacks are placed at a mutual distance of 1.2 m in a regular pattern (Fig. 5, top left), which covers a squared area of 162 m2 in the centre of the hall.

The outcomes of the investigation are presented in Fig. 7 in term of temperatures of the gas (leftcolumn) and temperatures of the rafters (right column). The upper row refers to measurements takenabove the combustible, while the bottom row refers to measurements taken farer from the flame.



TC - 2

ID

TC – 1

AST – 1

AST – 3

AST – 5

AST – 2

AST – 4

AST – 6

TC – 3

TC – 5

TC – 7

TC – 2

TC – 4

TC – 6

TC – 8

X

5.0 6.0

6.0

6.0

6.6

6.6

6.6

6.0

0.0 4.8 16.8 0.0

7.5

15.0

4.6

4.6

4.6

16.8

16.8

4.8

4.8

7.5

15.0

4.5

4.5

4.5

4.5

4.5

4.5

4.5

4.5

10.0

15.0

20.0

5.0 12.0

12.0

12.0

12.0

10.0

15.0

20.0

Y Z X Y ZID

ID X Y Z X Y ZID

Thermocouple

Adiabatic surface temperature

TC - 4 TC - 6 TC - 8

TC - 7TC - 5TC - 3TC - 1

AST - 5 AST - 6

AST - 4AST - 3

AST - 1 AST - 2

Figure 6. Coordinates of thermocouples (TC) and devices for element temperatures(AST) (left) and their graphical representation on 1/4 of the model (right).

1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1000

750

500

250

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1000

750

500

250

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1600TC – 6 AST – 4 Localised fire

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

TC – 5TC – 1

Tg (DK parametric)

AST – 1AST – 3

Ts (DK parametric)

Figure 7. Outcomes from scenario A in term of temperatures registered by thethermocouples (left) and on the element surfaces (right) inside (top) and outside(bottom) the area occupied by the combustible.

With respect to the latter, the temperatures of the gas in two different locations of the hall arecompared with the Danish parametric fire curve (Fig. 7, top left). Both the temperature registered bythe thermocouple most distant from the combustible area (TC-1) and by a closer thermocouple (TC-5)show a good accordance with the temperatures provided by the parametric fire. The comparisonbetween the two fire models has to be intended just as confrontation of the shape and temperatures ofthe fire, while a check on the starting time and initial growing rate of the two fires is hardly possible.The reason is that the parametric fire is a post-flashover model, which assumes the simultaneousburning of all combustible material present in the compartment. The fire scenario investigated insteadrefers to a fire, which triggers more realistically in few stacks and then propagates to the adjacent ones.Therefore also the initial phase of the fire is represented in the outcomes of the investigation.

The same accordance in term of shape and maximum temperatures stemming from simplifiedmodels is observable with respect to the elements outside the combustible area (Fig. 7, bottom right),where the temperatures of two rafters are compared with the steel heating calculated from the Danishparametric fire for the rafter profile. Even if the position of the two rafter with respect to the fire isdifferent (farer for AST-1 and closer for AST-3), the temperatures of the two elements are very similarand close to the steel heating curve.

The same uniformity of temperatures can be observed on all structural elements having the sameprofiles, except those just above the area occupied by the combustible (AST-6) or spanning from thatarea (AST-5). For those elements (Fig. 7, top right) and for the thermocouples above the combustiblearea (TC6 and TC8 in Fig. 7, top left) the temperatures are much higher, since the flame of the fire isimpinging the ceiling. This is due to the relatively high pile of pallets, which gives a HRRmax of 4.72 MW per stack and leads to a potential height of the flame equal to 5.53 m from the floor, accordingto the model of [23]. It seems therefore more reasonable to refer to a localized fire model for acomparison of the temperature of those elements. The Eurocodes model for a localised high flame fire[34], which is based on the above referenced model of Hasemi [24], would however lead to anoverestimation of about 25% of the element temperatures in this case, as visible in the figure.

Scenario C: non uniform distribution of combustible in spaceIn this scenario, a higher stacking grade of the combustible is considered and the pallets are assumedto be piled up in group of 20, forming 16 stacks of height 3 m. The pallet stacks are placed at a mutualdistance of 1.2 m in a regular pattern (Fig. 5, bottom right), which a covers a squared area 72 m2 in thecentre of the hall.

Contrarily to the previous case, differences in the distribution of the gas and element temperaturesalong the compartment can be observed in this case also outside the area occupied by the combustible.

In particular, two temperatures of the gas outside the combustible are reported and compared withthe Danish parametric fire (Fig. 8 bottom left). In spite of the temperature evolution of the parametriccurve is much faster than the observed temperatures, due to what explained above, a consistencybetween the temperatures of the parametric fire and of the hot gasses can be observed only withreference to a point very far from the flame (TC-1). The temperature increases by moving towards thecentre of the hall and even when the distance from the flame is still consistent (TC-5) a significantdifference (around 40%) is shown with respect to the other curve (TC-1). The same difference reflectson the temperature of the elements measured at two different distances (AST-1 on the most externalrafter and AST-3 on the adjacent one) from the combustible (Fig. 8 bottom right). The temperaturesshown in Fig. 8 refer to the same measurement points as the previous case; however, since the stackinggrade of the combustible is much higher, the area coved by the combustible is just 70.6 m2 and thedevices distance from the fire are higher, so that a lower difference on element temperatures could atfirst be expected. However, the HRR developed by the fire in this scenario (Fig. 11) is higher, since thestacking grade of the combustible also affects the fire propagation, as better explained in the followingsection.

The higher stacks also determine a higher flame length. Therefore, also in this scenario the flame isimpinging the ceiling and the temperatures above the combustible area registered by the thermocouples



TC-6 (Fig. 8 top left) and TC8, as well as by the element devices AST-5 (Fig. 8 top right) and AST-6are higher than in the previous scenario.

Scenario D: non uniform burning of combustible in timeThis scenario considers a lower stacking grade of the combustible, where 80 stacks of 4 pallets eachand an height of 0.6 m are assumed to be placed at a mutual distance of 1.2 m, covering a squared areaof 229 m2 in the centre of the hall (Fig. 5, bottom left).

The development of the fire and the consequent temperatures in the compartment and on theelements are very different in this scenario than in the previous ones. Due to the significantly lowergrade of staking, the fire propagation is very slow and one stack gets on fire when the fire on theadjacent one is about to extinguish, as visible in Fig. 9. The fire therefore moves from one stack toanother, maintaining a low HRR and lasting much longer than a normal fire. This phenomenon has beenevidenced and investigated in recent studies [35], where it is referred to as travelling fire.

Results in term of temperatures of the gas and of the elements are reported in Fig. 10 with respectto the first 30 min of fire, even if the duration of the fire is much longer in this case. With respect to theprevious cases, the temperatures obtained in this scenario are lower and spatially uniform in the area ofthe compartment not occupied by the fire. Furthermore, they also show almost a constant trend withrespect to time, as it can be seen in the left part of Fig. 10, where the temperature registered by thethermocouple TC-1 is reported. The same constant trend is shown by the temperature of the elementsoutside the combustible area, as visible in the right part of Fig. 10 with respect to the temperatureregistered for the device AST-1.

The temperatures of the thermocouples above the area occupied by the combustible show insteadpeaks of temperatures of short duration and then they drop down to temperatures not very dissimilar,even if slightly higher, to the temperatures outside the combustible area. Due to the significantly lowerHRRmax of a stack, the flame height in this scenario is lower than the ceiling and is not impinging the



1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1000

750

500

250

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1000

750

500

250

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

1600TC – 6 AST – 4 Localised fire

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

TC – 5

TC – 1

Tg (DK parametric)

AST – 1

AST – 3

Ts (DK parametric)

Figure 8. Outcomes from scenario C in term of temperatures registered by thethermocouple (left) and on the element surfaces (right) inside (top) and outside(bottom) the area occupied by the combustible.

beams, which never get very hot as in the previous scenarios. The same peaks characterize thetemperatures of the elements inside the combustible area, which occur at different times, depending ontheir distance from fire. The first peak is the one registered by the device AST-6 placed on the rafterjust above the centre of the fire, which is heated by the four central pallet stacks, assumed to get on firesimultaneously, which then propagate the fire.

The above mentioned studies on travelling fire show that, despite the low HRR, this type of fire canbe detrimental for the structure, which is heated for a significantly longer time than in case of a



Time = 5 min Time = 14 min

Time = 30 minTime = 20 min

Figure 9. Graphical representation of the HRR at different time for the fire scenario D.

600

450

300

150

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

600

450

300

150

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

(°C

)

TC – 1 TC – 8 AST – 1 AST – 6

Figure 10. Outcomes from scenario D in term of temperatures registered by thethermocouples (left) and on the element surfaces (right) inside and outside the areaoccupied by the combustible.

flashover-like fire. As well understandable, this situation can be critical especially in case of concretestructural system [36].

This doesn’t seem the case in this particular fire scenario, where, even if the temperature of theelements slowly increases with time, won’t get to very high values. This can be explained consideringthat a low fuel load has been assumed in this study, in order to have the possibility of implementingdifferent stacking level of the combustible. When the combustible is concentrated in a small area, thelocal fuel load density is high, which may lead to fire that are locally much more severe than whatexpected from a compartment fire. However, when the combustible is relatively spread along thecompartment as in this case, the fuel load density is more uniform along the compartment and has alow value.

Despite specific considerations on the temperature values however, it seems important to point outthat non uniform distribution in time and space of the temperatures, such as those generated by atravelling fire, may have negative effect on the structural behaviour, e.g. in case cold elements hinderthe thermal expansion of hot ones [37].

2.4. ComparisonThe results obtained from all the four fire scenarios considered (Fig. 5) are summarized in Fig. 11. Theleft graph at the top of the figure shows the development in time of the HRR; the adjacent graph on the right shows instead the temperature of the hot gases measured by the thermocouple TC-1, which isthe most distant from the combustible area, while the temperature measured by the thermocouple TC-7,placed just above the combustible, are shown in the graph below (bottom right) of the figure; the graph



160

120

80

40

00 5 10 15 20 25 30

Time (min)

HR

R (

MW

)

1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

AS

T –

4 (

°C) 1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

TC

– 6

(°C

)

1000

750

500

250

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

TC

– 1

(°C

)

Scenario A

Scenario B

Scenario C

Scenario D

Scenario A

Scenario B

Scenario C

Scenario D

Scenario A

Scenario B

Scenario C

Scenario D

Scenario A

Scenario B

Scenario C

Scenario D

Figure 11. Comparison of the outcomes for the different scenarios.

on the left at the bottom of the figure reports instead the temperature of the central rafter just above thecombustible, measured by the device AST-6.

By observing Fig. 11, it can be concluded that the primary effect of the grade of combustiblestacking is the propagation rate of the fire. The fire propagation is strictly related with the maximumHRR achievable by the fire and therefore with the fire duration too. By moving from high to low gradesof combustible stacking, the peak of the HRR decreases and the fire becomes lower and longer, asvisible at the top of Fig. 11. If the HRR becomes sufficiently low for a given distance of the combustiblematerials, a peak in the HRR curve cannot be evidenced anymore, as a different fire phenomenondevelops, where the fire moves throughout the compartment, as the combustible material of one areaburns out.

Among the three scenarios A, B, and C, the most severe fire in term of element temperatures seemsto be the one referring to the fire scenario C, which corresponds to a high staking grade of thecombustible. This result is reported in the figure for the temperatures of the elements above the flame,but can be also evidenced in the temperatures of the elements outside the combustible area.

2.5. Problematic Aspects of CFD ModelsThe results presented above show that simplified models are not capable of providing with asufficient grade of accuracy the element temperatures in case of fire developing in largecompartments. A more realistic modelling of the phenomenon can be obtained by means of CFDinvestigations, which can account for different distribution of the combustible and velocity of firepropagation and are capable of describing possible inhomogeneous distribution of the temperatureswithin a compartment.

2.5.1. Reduction of Computational OnusAs discussed in the previous sections, the above mentioned aspects are particularly relevant for thedesign of large compartments. In case of large compartments however, CFD models requires generallya particularly high computational onus. This is partially due to the greater extension in space and timeof the investigations: this problem can be avoided in case some symmetry lines are present in thecompartment. For example in the case study presented, the compartment was symmetric about thehorizontal and transversal central lines and a reduction of the computational onus has been obtained bycarrying out the investigations in 1/4 of the model (Fig. 12), where appropriate boundary conditions



Completemodel

Model

Complete

Partial

X

Dimensions

Y Z

40 30 6

20 15 6

Partialmodel

Figure 12. Full model and reduction to 1/4 of the model.



have been considered for simulating the two symmetry planes. The validation of the reduced modelagainst the full model is shown in Fig. 13, where a complete accordance in terms of fire and elementtemperatures can be observed.

Apart from the greater physical dimensions of the problem and possible longer durations of the fire,more significant increment of the computational onus can be ascribed to the convergence problems,which may arise in case of highly concentrated or highly spread combustible materials, bothrepresenting situations likely to occur in large compartments: in the first case, difficulties stem from avery high growing rate of the fire, while in the second case the slow propagation of the fire from oneobject to another can be an issue. As a consequence, the grid size required for this kind of investigationsis generally smaller than for investigations of more conventional structures, as better explained in thefollowing.

2.5.2. Mesh OptimizationEven if FDS is one of the most acknowledged codes for fire modelling, the use of the program requiresa certain attention in case the HRR has to be predicted rather than specified, such as in case themodelling of fire growth and spread is of interest. In these cases, the limits on the grid size are moresevere than the value recommended in the guidelines [38], [39], [40] and for an optimal representationof the buoyant plume dynamic a careful mesh sensitivity study becomes fundamental. With respect tothe case study investigated, despite the property of the compartment did not vary, different physicalphenomena stemmed from the variation of the distribution of the combustible and different values ofthe optimal grid size had to be identified for each case. In the following, the study of the meshsensitivity is reported for the case of scenario C. The most severe limitation from the Danish CFD guide[7] recommends a mesh size around 0.7 m with respect to the HRR of this scenario. A starting size of60 cm has been therefore used for the sensitivity study and then the mesh size has been decreased untilconvergence in the temperature curves s is obtained. In Fig. 14 the results of the sensitivity study arereported in term of HRR (top left), smoke height (top right), temperature of the gas registered by thethermocouple TC-6 (bottom right), and temperature of the rafter registered by the device AST-4(bottom left).

It can be seen that the mesh of 60 cm leads to inconsistent results for the HRR curve, for the smokeheight, and for the gas and element temperatures as well. A mesh of 30 cm would be sufficient for therepresentation of the smoke movements; however, the description of the HRR curve would notsufficiently accurate in this case and would lead to an underestimation of the gas and elementtemperatures, which are of main interest in this study. Finer mesh sizes of 20 cm and of 15 cm aretherefore required for reliable results. Since the sensitivity study shows that the gas and elementtemperatures don’t change whit a mesh finer that 20, this mesh size was therefore chosen for theinvestigation of the scenario C discussed here.

1600Partial model

Complete model1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

TC

– 6

(°C

) 1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

AS

T –

4 (

°C)

Partial model

Complete model

Figure 13. Validation of the reduced model which refer to 1/4 of the compartment,in case of scenario C.

3. CONCLUSIONSCurrent design methods of structural fire safety refer to post-flashover conditions, where a uniformdistribution of the temperatures can be assumed within the compartment. These methods have beenextensively tested and safely used in small compartment, which are typically dense furnished andpresent a reasonably uniform distribution of the combustibles along the floor area.

Large compartments may be instead less densely furnished and present therefore lower fuel load persquare meter of floor area, as shown by several statistic investigations conducted in the ‘70ies inSweden [41], [42] and more recently in Denmark [27]. Nevertheless, the assumption of a uniformdistribution of the combustible materials is generally unrealistic: especially in case of industrial halls,furniture and other combustibles can occupy just a part of the premises or stored materials can be piledup with different grade of stacking.

Since the flashover is unlikely to occur in large compartments, the Eurocodes [28] indicates arelatively low limit on the compartment size for the applicability of parametric fire. If this doesn’trepresent a problem for traditional residential buildings, it hinders the use of simplified fire models fora growing number of structures, given that nowadays longer span width and innovative solutions foropen spaces are made available by the constant advance in structural design. The need of a bettercomprehension and modelling of the fire which develops in large compartment seems therefore acritical aspect of the fire safety design.

In this paper a well-ventilated fire in a large compartment devoted to storage of wood pallets hasbeen investigated with respect to structural fire safety considerations. Four fire scenarios, each referringto a different staking grade of the combustible, have been considered and the results in term of gas andelement temperatures have been discussed and confronted. These temperatures have also beencompared with those obtained by simplified fire models such as a localized fire and a parametric fire,



1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

TC

– 6

(°C

)

160

120

80

40

00 5 10 15 20 25 30

Time (min)

HR

R (

MW

)

5

4

3

2

1

00 5 10 15 20 25 30

Time (min)

Hei

ght o

f sm

oke

(m)

1600

1200

800

400

00 5 10 15 20 25 30

Time (min)

Tem

pera

ture

AS

T –

4 (

°C)

Mesh 60 cm

Mesh 15 cm

Mesh 20 cm

Mesh 30 cmMesh 60 cm

Mesh 15 cm

Mesh 20 cm

Mesh 30 cm

Mesh 60 cm

Mesh 15 cm

Mesh 20 cm

Mesh 30 cmMesh 60 cm

Mesh 15 cm

Mesh 20 cm

Mesh 30 cm

Figure 14. Mesh sensitivity study for fire scenario C.

which assumes a ventilation controlled fire and a uniform distribution of the fuel load. The outcomesof the investigations show that in case of highly concentrated fuel the beam temperatures of the beamsobtained with a parametric fire model are underestimated. The use of localized fire on the other sidewould not be sufficient, as the temperatures of elements far from the combustible cannot be neglectedand show instead accordance with the temperatures obtained by using a parametric fire model. This isnot the case for the last scenario presented, where the element temperatures didn’t vary significantly inspace and were greatly lower than those obtained in the other fire scenarios or calculated from theparametric fire.

It can be concluded that simplified analytical models may not accurately describe the elementtemperatures, both in case the combustible is highly spread along the compartment and in case it isinstead highly concentrated in a small area of the floor.

With respect to the first case, a very low fuel load density deriving from spreading the combustibleover a large floor area, may result in very long fires, especially in case of a travelling fire, where thepropagation of the fire to the adjacent combustible materials occurs only after the burnout of the firstobject. This phenomenon has been recognised to have been possibly responsible of major structuralfailures both in steel [43] and concrete buildings [44], as the heating of the elements may be affectedby the fire duration more than by a higher gas temperature. Even if in the case study here presentedhigh element temperatures were not evidenced, when a spread distribution of the combustible wasconsidered, the occurrence of a travelling fire has been highlighted in this case.

With respect to the second case, a high staking grade of the combustible determines a non-uniformdistribution of the temperatures of both the gas and the elements, also outside the area occupied bythe combustible. Furthermore, due to the high local fuel load density, the temperatures of theelements above the flame may be heated up to temperatures much higher than those predicted by aflashover fire. This result may appear in contrast with recommendations for the structural fire designof large compartments [12], which show that conservative results are expected if the limits on thecompartment size for applicability of parametric fire are exceeded. It has to be stressed out however,that a value of 200 MJ/m2 of enclosing surface is recommended for domestic buildings, offices,hospitals, schools and hotels, disrespectfully from the effective fuel load present [12] n the view ofthe results presented above, in case a uniform distribution of the combustible is assumed in largecompartments, it seems essential to use a nominal high value of the fuel load, even when the amountof combustible is known to be lower, in order to take into account the effect of a possible stackingof the material.

ACKNOWLEDGMENTThe authors would like to thank Prof. Kristian Hertz for fundamental support to this study. Thecontributions of Prof. Grunde Jomaas, Prof. Luca Grossi and Dr. Francesco Petrini are also gratefullyacknowledged.

REFERENCES[1] Buchanan A.: “Structural Design for Fire Safety”, Wiley&Sons Ed., England, 2001.

[2] Hadjisophocleous G.V., McCartney C.J.: “Guidelines for the use of CFD simulations for fire andsmoke modelling”, NRCC-47740, National Research Council, Canada, 2005.

[3] Buchanan A.: “The challenges of predicting structural performances in fire”, proc. of the 9th Int.Symposium on Fire Safety Science, Germany 2009.

[4] Byström A., Cheng X., Wickström U., Veljkovic M.: “Measurement and calculation of adiabaticsurface temperature in a full-scale compartment fire experiment”, Journal of Fire Sciences,August 2012.

[5] Duthin D., Mc Grattan K., Khaskia A.: “Recent progress in fire-structure analysis”, EJSE specialissue on the 1st Int. Conference on Modern Design, Construction and Maintenance of Structures,Hanoi, Vietnam, December 2007.

[6] EN1991-1-2:2002 - General actions - Actions on structures exposed to fire, 2002.



[7] CFD best practice, Supplement to the Danish Building Regulations, Best Practice group,November 2009 (in Danish).

[8] Torero J., Steinhaus T.: “Application of computer Modelling to Fire Safety Design”, proc. of the53rd Jahresfachtagung de Vereinigung zur Foerderung des Deutschen Brandschutzes, Essen, June2004.

[9] Meacham B., Engelhardt M., Kodur V.: “Collection of Data on Fire and Collapse, Faculty ofArchitecture Building, Delft University of Technology”, Proc. of NSF Engineering Research andInnovation Conference, Honolulu, Hawaii, 2009.

[10] National Institute for Land and Infrastructure Management (NILIM): “Report on the WindsorBuilding Fire in Madrid, Spain”, Japan, Jul 2005 (in Japanese).

[11] Gann R.G.: “Final Report on the Collapse of World Trade Centre Building 7”, Federal Buildingand Fire Safety Investigation of the World Trade Centre Disaster, NIST NCSTAR 1A, November20, 2008.

[12] Hertz K., Parametric Fires for Structural Design, Journal of Fire Technology, Springer Science,2001 (DOI: 10.1007/s10694-011-0246-5).

[13] Giuliani L., Budny I.: “Different design approaches to structural fire safety”, International Journalof Lifecycle and Performance Engineering (IJLCPE), special issue on Fire Safety Design andRobustness Considerations in Structural Engineering, accepted for publication on July 2012(in press).

[14] Roben C.: “The effect of cooling and non-uniform fires on structural behaviour”, PhDdissertation, University of Edinburgh, January 2010.

[15] Petrini F.: “Numerical analyses for Performance-Based Fire Engineering (PBFE)”, 4th

International Conference on Structural Engineering, Mechanics and Computation (SEMC’10),Cape Town, South Africa, September 2010.

[16] Crosti C.: “Structural analysis of steel structures under fire loading”, Acta Polytechnica, Vol. 49,No. 1:21–28, 2009.

[17] EUR 24222: “Fire safety of industrial halls – A valorization project”, Final report, Research fundfor coal and steel publications (RFCS), Luxembourg, 2007.

[18] Gentili F.: “Advanced numerical analyses for the assessment of steel structures under fire”,International Journal of Lifecycle and Performance Engineering (IJLCPE), special issue on FireSafety Design and Robustness Considerations in Structural Engineering, accepted for publicationon August 2012 (in press).

[19] Drysdale D.: “An Introduction to Fire Dynamics, Wiley&Sons Ed., England 2008.

[20] Krasner L.M.: “Burning Characteristics of Wooden Pallets as a Test Fuel,” Serial 16437, FactoryMutual Research Corp., Norwood, MA, 1968.

[21] La Malfa A., La Malfa S.: “Approccio Ingengeristico alla Sicurezza Antincendio”, LegislazioneTecnica, 2009 (in Italian).

[22] Karlsson B., James G. Quintiere J.G.: “Enclosure Fire Dynamics”, Taylor & Francis Inc., 2009.

[23] Heskestad G.: “Fire Plumes”, SFPE Handbook of Fire Protection Engineering, pp. 2–9, NationalFire Protection Association, Quincy, MA, 1995.

[24] Hasemi Y., Tokunaga T.: “Flame Geometry Effects on Buoyant Plumes from Turbulent DiffusionFlames”, Fire Science and Technology, 4, 15–26, 1984.

[25] Pettersson O., Magnusson S.E., Thor J.: “Fire Engineering design of steel structures”, Bulletin 52,Sweden, 1976.

[26] Wickström U.: “Application of the standard fire curve for expressing natural fires for designpurposes”, Fire Safety: Science and Engineering, American Society for Testing and Materials(ASTM), STP 882, Harmathy Ed., Philadelphia, 1985.



[27] EN 1991-1-2 DK NA: 2007 – Danish National Annex to Eurocode 1: Actions on Structure, Part1-2: General Actions, Actions on structures exposed to fire, 2008.

[28] EN1991-1-2:2002 - General actions - Actions on structures exposed to fire, Annex A: Parametrictemperature-time curves, 2002.

[29] PD 6688-1-2:2007: Background paper to the UK National Annex to BS EN 1991-1-2: 2002-Actions on structures. General actions. Actions on structures exposed to fire, 30 April 2007.

[30] Hertz K.D.: “Vejledning i dimensionering af bygningskonstruktioner for fuldt udviklet brand”,(Guide for design of building structures for fully developed fires), version 2–3, National Agencyfor Enterprise and Construction, Copenhagen, 2006 (in Danish).

[31] Jahn W., Rein G., Torero J.L.: “The Effect of Model Parameters on the Simulation of FireDynamics”, Fire Safety Science 9: 1341-1352, doi:10.3801/IAFSS.FSS.9–1341.

[32] Baker G., Fleury R., Spearpoint M., Fleishmann C., Wade C.: “Ignition of Secondary Objects ina Design Fire Simulation Tool”, Fire Safety Science 10: 1359-1372, doi:1038/IAFSS.FSS.10–1359.

[33] McGrattan K., Klein B., Hostikka S., Floyd J.: “Fire Dynamics Simulator (Version 5), User’sGuide”, March 2006.

[34] EN1991-1-2:2002 - General actions - Actions on structures exposed to fire, Annex C: LocalisedFire, 2002.

[35] Stern-Gottfried J., Law A., Rein G., Gillie M., Torero J.: “A Performance Based MethodologyUsing Travelling Fires for Structural Analysis”, Conference of Fire Protection Engineering,(SPFPE’10), Lund, June 2010.

[36] Law A., Stern-Gottfried J. Gillie M., Rein G.: “The influence of travelling fires on a concreteframe”, Engineering Structures, Vol. 33, Issue 5, Pages 1635–1642, May 2011.

[37] Usmani A.S., Chung Y.C., Torero J.L.: “How Did the WTC Collapse: A New Theory”, Fire SafetyJournal, 38, 6, 501–591, 2003.

[38] NUREG 1824 - Verification and validation of selected fire models for nuclear power plantapplications, vol. 7: Fire Dynamic Simulator (FDS), Final report, US Nuclear RegulatoryCommission, 2007.

[39] Ascenzi G. Villi G., Vulpiani G.: “Ingegneria della sicurezza antincendio - Guida all’utilizzo diFDS”, Flaccovio Ed., 2010.

[40] CFD Best Practice, Best Practice gruppen, Brandteknisk Selskab, 12. November 2009.

[41] Nilsson L.: “Brandbelastning i bostadslägenheter”, (Fire Load in Appartments), Statens Institutför Byggnadsforskning, Rapport R34, 1970. Stockholm 1970 (in Swedish).

[42] Forsberg U., Thor, J.: Brandbelastningsstatistik för skolor och hotel (Fire Load Statistics forSchools and Hotels), Stålbygnadsinstitutet, Rapport 44:1, Stockholm 1971(in Swedish).

[43] Gann R.G. Hamins A., McGrattan K.B., Mulholland G.W., Nelson H.E., Ohlemiller T.J., Pitts W.M., Prasad K.R.: “Reconstruction of the Fires in the World Trade Center Towers”, NIST NCSTAR1-5, September 2005’’.

[44] Fletcher, I. et al, “Model-Based Analysis of a Concrete Building Subjected to Fire,” AdvancedResearch Workshop on Fire Computer Modelling, Santander, Spain, 2007.



effects of combustible stacking in large compartments

Design

advanced design

large compartments advanced

structural model

use of cfd models

design of larger compartments

prescriptive design

presence of large compartments

structural system