Fire Safety Journal xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Fire Safety Journal

journal homepage: www.elsevier.com/locate/firesaf

Fire-induced pressure and smoke spreading in mechanically ventilated

buildings with air-tight envelopes
⁎
Simo Hostikkaa, , Rahul Kallada Janardhana, Umar Riaza, Topi Sikanenb
a
Aalto University, Espoo, Finland
b
VTT Technical Research Centre of Finland Ltd., Espoo, Finland

A R T I C L E I N F O A BS T RAC T

Keywords: Fire-induced pressures have not been considered dangerous in building fires, but the situation may be changing
Pressure as building envelopes become increasingly air-tight. In this study, we investigate whether this can change the
Modelling fire development and pose new risks for structural and evacuation safety. We used experiments to validate the
CFD numerical models, and models for simulating the fire development in buildings with different air-tightness
Smoke
levels. The simulations considered air permeability values typical for traditional, modern and Near-Zero
Airtightness
buildings. Three different smoke damper configurations were studied, and the fire growth rates were varied
Near-zero buildings
High-rise buildings from medium to ultra-fast. The results showed that transitioning from traditional and modern buildings to
Near-Zero buildings can sufficiently increase the peak overpressures from fast-growing fires to cause structural
damage. Conditions were identified for avoiding excessively high overpressures, while preventing smoke from
spreading through the ventilation system.

1. Introduction in a relatively air-tight compartment were recently conducted in the

OECD PRISME programme [1]. The influence of air tightness of
Fire-induced pressures have gained increasing attention in fire enclosures have been investiagted using experiments and theoretical
science mainly their capability to drive flows between compart- analysis by Prétrel et al. [2,3]. They found that closing the ventila-
ments. Pressure has been identified as a potential risk for structural tion paths, and particularly exhaust ducts, during the fire signifi-
damage mainly in industrial scenarios though these have been cantly increased the pressure. This increased pressure was identified
limited to explosions or gas deflagrations. Adverse effects of pressure by Forneau et al. [4] as one of the consequences of better energy
on evacuation and structural safety in residential buildings has not efficiency, concluding that the high pressure can lead to a reverse
been studied. Our understanding of the potential threats to fire flow in the supply ventilation system. However, they did not
compartmentation has been dominated by the thermal impact of a recognize this pressure rise as posing a risk for escape or structural
fire and the load-bearing capacity of structures. integrity.
Evacuation-related risks is of greater significance in residential Calculating compartment fire pressure requires knowledge of the
fire scenarios. Recently, a group of Finnish firefighters reported that gas temperature development and leakage flows [5]. This usually
they were unable to open the inward-opening door of a fire requires numerical integration as the leakage flows depend on
apartment during the growth stage of the fire, thereby indicating pressure. The situation is even more complex if ventilation is
the overpressure to be well above 100 Pa. This makes it impossible mechanically controlled, as is the case in most modern, energy
for the occupants to use the door for escape. If we combine this efficient buildings. Pressure calculation has been included in most
observation with the rapid shift in construction requirements and numerical fire models capable of solving the flows between several
practices are rapidly moving towards more air-tight building envel- compartments. Of these models, the Fire Dynamics Simulator (FDS)
opes, as demanded for the energy efficiency and high-rise construc- [6], has been further developed by Floyd [7] to include a dedicated
tions, we can expect the pressure-related risks may become more HVAC module. The current FDS validation database for fire-induced
significant, unless the preventive measures can be found and pressures consists of three experimental campaigns, and the peak
justified. over-pressures ranging from a few Pa to 1300 Pa [8]. A detailed
Various studies have analyzed the effects of pressure on building validation of the of FDS-HVAC module was recently reported by
performance. High-quality measurements of fire-induced pressure Wahlqvist and van Hees [9].

⁎
Corresponding author.

http://dx.doi.org/10.1016/j.firesaf.2017.04.006
Received 15 February 2017; Received in revised form 27 March 2017; Accepted 3 April 2017
0379-7112/ © 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: Hostikka, S., Fire Safety Journal (2017), http://dx.doi.org/10.1016/j.firesaf.2017.04.006
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

Nomenclature
S compartment surface area (m2)
Roman Symbols T temperature (K)
u duct flow speed (ms−1)
A crossectional area of the duct (m2) Vleak leakage flow (m3s−1)
AL leakage area (m2) V50̇ volumetric leakage flow at ΔP =50 Pa (m3s−1)
cp spesific heat capacity (kJ kg−1K−1)
Cd discharge coefficient (-) Greek
Δp pressure difference (Pa)
h enthalpy of fluid (W m −2K−1) ρ density (kgm−3)
K loss coefficient (-)
P pressure (Pa) subscripts
Q heat release rate (HRR) (kW)
tg growth time (s) i, k nodes
q50 leakage flow at ΔP =50 Pa (m3h−1m −2 ) j duct segments

The aim of the current research was to examine the influence of 2. Methods
building envelope air tightness on fire pressure and its consequences
in residential buildings. In particular, we intend to determine 2.1 Numerical method
whether new risks can be expected for structural and evacuation
safety and explore the possibility of using the building ventilation FDS is a Large Eddy Simulation (LES) based Computational Fluid
network for reducing pressure without increasing the risk of smoke Dynamics software which solves the low-Mach number combustion
spreading between apartments. First, we use existing [10,11] and equations on a rectilinear grid over time. A dedicated module for
new [12] experiments to validate the modelling capability of fire- modelling Heating, Ventilation and Air-conditioning (HVAC) systems
induced pressures in residential buildings. Next, the validated model is coupled with the gas phase solver. The ventilation network is
is used to perform numerical experiments in hypothetical apartment described as a series of ducts and nodes. The nodes are placed at
buildings with different levels of air-tightness. The severity of the points where ducts intersect each other or the CFD computational
predicted conditions is evaluated in light of two model outcomes: domain. The ducts represent uninterrupted domains of fluid flow which
pressure inside the apartment and smoke spread to neighbouring encompass elbows, expansion/contraction fittings and other fittings.
apartments through the ventilation network. The losses due to friction and duct fittings are assigned as dimension-
less loss numbers to the ducts. The node losses are also attached to the

Fig. 1. Plan drawings and measurements of the FOA (a) and Aalto [12] (b) validation experiments.

2
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

ducts as loss terms only appear in the duct Eq. (3). However, since the 31–33 a supply network. Schematic diagrams of the ventilation
module does not presently store any mass, mass flux into a duct is networks, simplified for modelling, are shown in Fig. 2. Test 23 had
equal to the mass flux out of the duct. The nodal conservation a slightly different exhaust ventilation network, as explained in [11].
equations for mass, energy and momentum equations are as follows: The ventilation networks consisted of a main duct connecting a fan to
the fire room, and additional branches connecting the main duct to
∑ ρj uj Aj =0
ambient through a damper. The purpose of the branches was to
j (1)
simulate additional rooms of actual ventilation systems. The dampers
∑ ρj uj Aj hj = 0 were used to limit the flow rate to these virtual rooms.
j (2) A detailed description of the Aalto experiments is given in [12], and
only a brief summary is presented here. The experiments are summar-
du ized in Table 2. They were conducted in a single apartment of a 1970's
ρj L j = (Pi − Pk ) + (ρgΔz )j + ΔPj + 0.5Kj ρj |uj |uj
dt (3) apartment building. The apartment (see Fig. 1b) included a living
room, bedroom, kitchen, bathroom, aisle and a closet, having a total
2.2 Leakage modelling floor area of 58.6 m2 and a ceiling height of 2.57 m. It was located on
the first floor of the building. The apartment had two exhaust
Leakage refers to the amount of air flowing through the building ventilation ducts leading to the roof. Three different ventilation
envelope due to the pressure difference between the inside and outside. configurations were created by keeping the original dampers
A standardized method for measuring the leakage (e.g. SFS EN: 13829) (Normal), removing them completely (Open) or closing the ducts
makes use of a variable-speed fan mounted to a door in the envelope tightly (Closed). All the other ventilation paths were closed. Two
being tested [13]. All the ventilation paths are sealed during the test. different fires were used: heptane pools with peak HRR of 800 kW
The building interior is then pressurized or depressurized using the fan were used in experiments 1–3 and flexible polyurethane foam with
and the leakage flows are determined by monitoring the flow through peak HRR of about 1 MW in test 4. Experiments 1–3 were repeated
the fan. Energy efficiency studies commonly measure leakages at an several times.
underpressure of 50 Pa. Overpressures would be more suitable for fire
analyses, though both directions can appear. The leakage can be 2.3 Simulation models
reported as a volumetric flow V50 ̇ , air permeability q50 or air exchange
rate n50. The simulation model of the FOA experiments consists of a single
From the viewpoint of fire CFD, since leakage is a sub grid -scale room with a leakage area connected to an additional duct. The main
phenomenon, a leak path boundary condition cannot be directly components of the HVAC models, shown in Fig. 2, are nodes, ducts,
specified. In FDS simulations, the leaks are modelled as a finite area and a fan. The other components such as dampers and expansion/
using vents on the domain boundaries. FDS utilizes the HVAC module contraction fittings are accounted for in the loss terms. Multiple ducts
to solve the leakage flow have been combined into a single duct with appropriate loss coeffi-
cients for further simplification. Dampers do not completely block the
⎛ ⎞1/2
̇ = AL sign(Δp ) ⎜ 2Δp ⎟
Vleak smoke flow, they only limit the volume flow to 25 l/s.
⎝ ρ ⎠ (4) The exhaust network model consists of 8 nodes, including a room
node connecting the main duct to the CFD domain, four ambient
nodes, and three internal nodes for connecting the branches to the
2.3 Validation studies
main duct. The total length of the main duct is 7.5 m. The exhaust fan
that drives the flow is placed in the duct connecting the main duct to
The FDS validation is performed by simulating two series of
the ambient. The fan curve was defined by specifying the flow rates (0,
compartment fire experiments that were conducted by the Swedish
60 and 120 l/s) at three static pressures (310, 190 and 18 Pa),
FOA Defence Research Establishment and Aalto University. The FOA
respectively. The layout of the supply network is similar to the exhaust
experiments were carried out in a room with dimensions of
network. The fan is obviously operating in different direction, and it is
4.0 m×5.5 m×2.6 m (height), shown in Fig. 1a. The fire room was
connected directly to the ambient without any duct segment.
divided in two parts with a wall, and the wall had a 1.9 m wide opening
In theory, it would be possible to determine the loss coefficients for
from floor to ceiling. The fire source was a heptane pan of 0.73 m×1 m.
A t2-type HRR curve was achieved using a lid that was moved over the
Table 1
pan at a given rate thereby increasing the heptane burning area. FOA experiment configurations, following the original test numbering in [10,11]. Test 15
Assuming a value of 1600 kW/m2 for the HRR per unit area, the had constant pool area of 0.5 m2.
movement speed of the lid was adjusted to yield three different growth
rates: Test Fire type Opening D (m) Exhaust network Supply network

FOA series 1
Type 1: Q̇ = 0.035t 2 kW (between fast and medium). FOA 1 1 0.2 m – –
Type 2: Q̇ = 0.075t 2 kW (between ultra-fast and fast). FOA 2 2 0.2 m – –
FOA 3 3 0.2 m – –
Type 2: Q̇ = 0.085t 2 kW (between ultra-fast and fast).
FOA series 2
The FOA experimental parameters are summarized in Table 1. Expe- FOA 11 1 0.2 – –
riments 1–3 [10] did not include an actual ventilation network, but the FOA 12 1 0.15 – –
fire room had a circular opening (D=0.2 m) at height 0.6 m from the FOA 13 2 0.2 – –
FOA 14 2 0.15 – –
floor, connected to a 2.2 m long and 0.2 m diameter tube. Temperature FOA 15 – 0.2 – –
and flow speed were measured at the end of the tube. FOA 21 2 0.2 Exhaust network 1 –
FOA experiments 11–33 [11] included three different ventilation FOA 22 2 0.15 Exhaust network 1 –
configurations. The leakage openings of different diameters were FOA 23 2 0.15 Exhaust network 2 –
FOA 31 1 0.10 Exhaust network 1 Supply network 1
connected to a 0.32 m diameter and 3.2 m long tube connecting to
FOA 32 2 0.15 Exhaust network 1 Supply network 1
the ambient. The openings were placed at height 0.6 m from the floor. FOA 33 2 0.2 Exhaust network 1 Supply network 1
Experiments 21–33 included an exhaust network, and experiments

3
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

all the duct sections from the individual losses of the components and
the friction. As all the details cannot be determined at this stage,
effective loss coefficients were estimated first by defining a constant
roughness of 1 mm for the ducts, and then adjusting the loss of the first
duct to match the known volume flow of 25 l/s. The losses from the
damper and the 90 degree bend were thus combined into a single loss
coefficient. Identical losses are used for forward and reverse flow.
For leakages, the bulk leakage method was used over all wall
surfaces. For other boundary conditions, the models of both FOA and
Aalto experiments included only the interiors of the enclosures,
assuming that the exterior sides of the walls are exposed to the
ambient. For the concrete boundaries, thermal conductivity, specific
heat and density were set to 0.7 W/(m. K), 0.75 kJ/(kg. K) and
2200 kg/m3, respectively. Fire boundary conditions were specified as
time dependent heat release rates per unit area (HRRPUA) of the fuel
pan surfaces. Technically, they were converted to time dependent mass
Fig. 2. The ventilation networks of FOA experiments as modelled in FDS. Top: Exhaust flux and species concentration boundary condition of the CFD solver.
network 1. Bottom: Supply network 1. The simulations were carried out using the FDS version 6.3.2 with a
single mesh and 6 OpenMP threads on a personal computer with a 3.2-
GHz Intel Xeon processor and 32-GB ram. The CPU time for a single
Table 2 simulation of 300 s was approximately 8 h.
Parameters of the Aalto experiments. Different numbering from [12] is used here for
simplicity.
3. Validation results
Test Fuel Ducts Roof fan
The experimental and simulated fire room pressures in FOA
Aalto 1 Heptane 3.0 L Open OFF experiments 1–3 are shown in Fig. 3. The positive pressures during
Aalto 2 Heptane 3.0 L Normal ON
the fire growth stage are reproduced by the simulation model with good
Aalto 3 Heptane 3.0 L Closed ON
Aalto 4 Polyurethane foam 3.82 kg Normal ON accuracy. The negative pressures after the fire suppression, in turn, are
not captured as well. The same behaviour was observed in all the
validations tests of this work.
A summary of the gas temperature and peak overpressure predic-

Fig. 3. Measured and simulated gas pressures in FOA experiments 1–3.

4
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

of the building, nor will the corridor pressure conditions be studied

here. The ceiling height is 2.5 m. The room walls and ceiling are made
of 15-cm-thick concrete. The fire is assumed to ignite in one of the
smaller apartments. Within this apartment, the structures dividing the
apartment into rooms are included, but the doors are assumed to be
open. Fig. 5 shows the geometry.

4.2 Modern ventilation systems

The traditional ventilation systems in residential buildings have

been based on mechanical or buouyancy-driven exhaust, with the
supply air provided through the building envelope either as an
uncontrolled leakage or through valves. Modern HVAC systems usually
provide separate networks for supply and exhaust air. Both networks
are typically equipped with a fan unit to control the flow rate and to
implement the heating/cooling, as well as heat recovery and air
filtration for indoor air quality control. Single fan unit may serve the
entire building or a single floor of a multi-storey apartment building.
The coils and filters introduce drag to the flow. As a result, the fan unit
will cause pressure losses even when the fans are turned off. For fire
compartmentation and smoke control, fire and/or smoke dampers are
typically installed to the ducts entering and leaving the apartments. In
addition, the modern ventilation fans are often equipped with dampers
that automatically close the ambient connection when the fan is turned
off. In a fire situation, turning off the fan can, therefore, lead to
complete closing of the ventilation system.

4.3 Simulation model

Each room in the model has a designated mesh. The discretization

is 0.10 m for the fire room and 0.50 m elsewhere. Each room is
considered its own pressure zone with an individual solution for the
background pressure. There is no heat transfer through the walls
between the rooms and all leakages are to the ambient. The ventilation
system consists of independent inlet and exhaust networks, both
equipped with a fan with stalling pressure of Pmax=550 Pa and a
zero-pressure flow rate of Vmax=650 l/s. The fan unit model parameters
were tuned to produce 150 Pa pressure loss to the flow. Each network
consists of a central duct (∅=0.25 m) and smaller (∅=0.125 m) ducts
for each individual apartment. Two inlet and exhaust connections are
used for the fire apartment, but only one for all the other apartments.
Fig. 4. Uncertainty of peak temperature (top figure) and overpressure (bottom figure) Loss coefficients K for the inlet and exhaust ducts are adjusted in non-
predictions. Red solid lines indicate the average over- or underprediction of the model.
fire conditions to achieve a ventilation rate of 40 l/s and slightly
The dotted lines on either side of the diagonal lines indicate the experimental and model
standard deviations.
negative pressure for the apartment.
In total, 34 simulations were performed, varying the damper
configuration, envelope air-tightness level and the fire growth rate.
tions is shown in Fig. 4. Gas temperature comparisons were made for
all the individual temperature measurements, not for the averaged
layer temperatures as in [8]. The temperatures were overpredicted by
11% on average. Overall, the temperature uncertainties are satisfac-
tory, and increase our confidence in the accuracy of the estimated
model inputs for HRR. The peak overpressures are underpredicted by
7% in average. The model relative standard deviation is dominated by
the combined experimental uncertainty.

4. Residential case study description

4.1 Building description

The influence of the envelope air-tightness on the fire-induced

pressures and smoke spreading in the ventilation network was studied
through the fire simulations in a hypothetical residential building. The
simulation geometry for the case study consists of a single floor of a
multi storey apartment building. The floor contained 11 apartments Fig. 5. Geometry for the apartment case study.
with floor areas of either 50 m2 or 100 m2 and a corridor. The model
does not include a staircase connecting the domain to the other floors

5
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

The influence of the fire/smoke dampers installed in the ventilation The predicted peak pressures are summarized in Fig. 7. These
ducts was studied by analyzing three different compartmentation results were not corrected for the estimated model bias of - 7%. The
damper configurations: trends in the results are clear and consistent. All three parameters - fire
growth rate, damper configuration and air-tightness - are found to be
Damper=Off Both inlet and outlet remain open during the fire. important for the expected peak pressure. Interestingly, the sensitivity
Damper=Inlet The inlet duct of the fire apartment is closed by a da- of the pressure to the parameter values seems to increase when moving
mper 10 s after the ignition. towards a scenario with higher pressure. For instance, the damper
Damper=Both Both inlet and outlet are closed by dampers 10 s from configuration is not very important in traditional or normal buildings,
the ignition. but can become crucially important in very air-tight buildings. The
results in Fig. 7 were found to be independent of the fan operation (on
Additionally the effect of the dampers located at the inlet and outlet or off) and position of the fan unit damper. Therefore, the leakages
fans was investigated. The ventilation network parameters are shown through the other apartments can compensate for the fan pressure
in Table 3. differences and complete closing of the ventilation system. Of course,
Three different levels of the building envelope air-tightness were this leads to smoke spreading to the other apartments through the
examined. These levels were defined using the air permeability values network.
q50, listed in Table 4. The class “Traditional” represents an average of The pressure results can be compared against the simulations of
the required and reference air-tightness value (for heat loss calcula- Fourneau [4] who assumed a fire with a growth rate between medium
tions) described in the current Finnish building code (Part D3: Energy and fast, more airtight envelope (n50 ≈ 0.6 h−1), and two open paths
efficiency, 2012, Ministry of Environment). The “Modern”, corresponds through the mechanical ventilation system. The peak pressure in their
to the measured air-tightness in the concrete element multistorey simulation was about 550 Pa, i.e. between our results obtained for the
buildings [13] and the “Near-zero” represents the current, technically Damper=Off configuration.
achievable target level. In order to evaluate of the occupants’ possibility of escaping from
The volumetric leakage flow rate V50̇ at 50 Pa can be calculated from the fire apartment, we compare the overpressure against an appro-
the air permeability values using Eq. (5) priate threshold. Here, we assume that opening the door would not be
possible if the overpressure were above 100 Pa. For design purposes
q50
V50̇ = S the difference in pressure is usually limited to 50 Pa, but we choose a
3600 (5)
less conservative value for the risk analysis. Based on the predicted
and the air exchange rate as n50 = V50 / V , where V is the building peak pressures, it should be possible to open the door in traditional
volume. The leakage areas AL , through which these air-tightness levels buildings at a medium fire growth rate. For more air-tight buildings
are specified in the FDS models, are then calculated from the and faster fires, opening the door would be challenging. However, it is
volumetric flow rates as important to note that these correspond to momentary peak pressures,
and reveal little about the time duration of the peak pressure.
V50̇
AL = Challenges for the structural integrity can be expected in fast or
⎛ 2Δp ⎞1/2
Cd ⎜ ρ ⎟ ultra-fast fires. If the criterion of failure is chosen based on our own
⎝ ∞⎠ (6) experimental observation (1500 Pa), the fast fires could pose a risk
when the dampers are closed and the envelope is very airtight. This
In the models, these leakage areas are distributed equally at the
could occur in near-zero or high-rise buildings. In ultra-fast fires,
locations of doors and windows.
problems could be expected in all modern buildings using dampers. In
Three different fire scenarios were used in the simulations. Two of
near-zero buildings, the capacity of the ventilation network is insuffi-
them were defined as typical t2 fires with prescribed maximum HRR,
cient to relieve the pressure of an ultra-fast fire even without dampers.
and the third one was based on the experimental heptane pool burning
In such a scenario, alternate means for relieving pressure should be
rate in the Aalto experiments.
investigated.

Medium: Q̇ = Q0 (t / tg )2 , tg=300 s, Qmaẋ =4 MW.

Fast: ̇ =4 MW.
Q̇ = Q0 (t / tg )2 , tg=150 s, Qmax 5.2 Smoke spreading through the ventilation network
Ultra-fast: Q̇ experimental, tg ≈ 70 s, Qmax ̇ =1 MW.
The above pressure results indicate that open ventilation ducts
In the equations above, Q0=1 MW. The targeted peak HRR of the t2 could be used as a potential path for pressure relief, at least in the
fires was set to 4 MW which was considered as a realistic value for buildings constructed according to current air-tightness norms, as well
apartment fires [14]. This is also sufficiently high to consume the O2 in as in near-zero constructions if the fire is slower than ultra-fast. The
the compartment and hence yield a physically relevant duration of fire possibility of smoke spreading to the other apartments and the
and yield of combustion products. resulting loss of compartmentation becomes an issue when evacuation
safety is undermined. Figure 8 visualizes the smoke concentration in
5. Residential case study results and discussion
Table 3
HVAC model inputs [12].
5.1 Pressure in the fire apartment
Ventilation L1 L2 L3 A1, A2 A3 K1 K2 K3
Fig. 6 presents the simulation results for compartment fire pressure in configuration (m) (m) (m) (m2) (m2)
the most severe conditions, i.e. when the building envelope is very air-tight
and the dampers are closed on both the supply and exhaust ducts. The peak Bathroom duct
overpressure is shown to depend very strongly on the fire growth rate. The Open 0.4 10 1.0 0.01227 0.049 0 3 1
pressure in the medium fire remains below 1000 Pa, while the fast and Normal 0.4 10 1.0 0.01227 0.049 0 38 1
ultra-fast fires reach a pressure of 3000 Pa and 7000 Pa respectively. It is
Closet duct
obvious, that the structure would not withstand such high pressures.
Open 0.4 10 0.01227 0 2
Although the absolute values of the underpressures are lower than those of Normal 0.4 10 0.01227 0 70
the overpressures, they are associated with higher uncertainty.

6
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

Table 4 A more quantitative presentation of the same results is given in

Envelope specifications. Fig. 9 showing the minimum visibility over the entire fire duration in
the different damper-fan combinations. The red horizontal bars denote
Envelope Type q50 (m3m−2h−1) V50 (m3/s) n50 (h−1) AL (m2)
the median values from all the neighbouring apartments, and the boxes
Traditional 3 0.146 4.2 0.02690 indicate a “typical range”. In all the cases wihout any dampers, the
Modern 1.5 0.073 2.1 0.01345 visibility values get down to a few meters whereas closing the fan
Near-zero 0.75 0.036 1.05 0.006725
reduces the visibility to about two meters. Making the building more
air-tight increases the amount of soot in the neighbouring apartments,
thus reducing the visibility. In the simulations with fan operating and
damper only on the inlet side, smoke is not observed in the neighbour-
ing apartments. If the inlet side is open, then operating the fan worsens
the situation as the fan pressure head prevents smoke from escaping
through the inlet branch.

6. Conclusions

We used previously published experimental results to validate the

FDS fire modelling of the pressure development in fires taking place
within closed and mechanically ventilated compartments. This in-
cluded the modelling of HVAC systems and building envelope leakages.
The validated model was used to investigate the influence of damper
configuration, envelope air-tightness and fire growth rate on the
overpressures and the spreading of smoke to the neighbouring apart-
ments in a mechanically ventilated building with many apartments.
Fig. 6. Effect of the fire type on pressures.
The simulated over-pressures were compared against the two experi-
mentally observed [12] criteria: A 100 Pa over-pressure was found to
the neighbouring apartments 170 s from the ignition. If no dampers be sufficient to prevent apartment occupants from opening an inwards-
are used, the smoke spreads to the neighbours regardless of fan opening door, thus preventing them from escaping. Secondly, the light-
operation. The third figure corresponds to a scenario in which the weight structures were found to fail at 1450–1600 Pa overpressure.
damper is used only on the inlet side and the exhaust ventilation fan is Numerical simulations of the fire scenarios showed that the peak
kept running. Interestingly, the smoke does not spread to other overpressure is sensitive to damper configuration, envelope air-tight-
apartments in this case. Obviously, the cases with both dampers ness and fire growth rate, but practically independent of the fan
operating can be expected to be safe in this respect. operation. The pressure was found to increase with improving air-

Fig. 7. Simulations Peak Pressures Bar Plots.

7
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

Fig. 8. Visualization of smoke concentration 170 s after ignition.

Fig. 9. Minimum visibility in neighbouring apartments.

tightness (reduced leakage), increased use of fire or smoke dampers, exist at fast and ultra-fast fire growth rates.
and the increased rate of fire growth. The 100 Pa limit was reached in Simulations of different damper configurations and fan operation
all the simulations except those made for the traditional buildings with modes showed that the spreading of smoke to neighbouring apart-
medium fire growth rate. In a Near-Zero building, the limit was ments can be avoided if only the inlet ventilation branch is closed with
reached in less than 20 s for ultra-fast fire and in about 100 s for a a damper and exhaust fan is operating. This mode of operation was
medium fire. The duration varied with respect to the fire type, being in found to be the only combination for preventing smoke spread, while
between 100 and 200 s. The risk of structural damage was found to simultaneously maintaining the pressure at an acceptable level.

8
S. Hostikka et al. Fire Safety Journal xxx (xxxx) xxx–xxx

Acknowledgements (2012) 349–361.

[7] J. Floyd, Coupling a Network HVAC Model to a Computational Fluid Dynamics
Model Using Large Eddy Simulation. Fire Safety Science, in: Proceedings of the
The research project was funded by the Finnish Fire Protection Tenth International Symposium, International Association for Fire Safety Science,
Fund (PSR), Ministry of Environment, Hagab AB, and the Criminal 2011, pp. 459-470.
[8] K. McGrattan, S. Hostikka, R. McDermott, J. Floyd, C. Weinschenk, K. Overholt,
Sactions Agency of Finland. The work was also partially supported by Fire Dynamics Simulator, Validation Guide, Volume 3: Validation, National
the Academy of Finland under grant no. 289037. Institute of Standards and Technology, Gaithersburg, Maryland, USA, and VTT
Technical Research Centre of Finland, Espoo, Finland, sixth edition, November
2015.
References [9] J. Wahlqvist, P. van Hees, Validation of FDS for large-scale well-confined
mechanically ventilated fire scenarios with emphasis on predicting ventilation
[1] L. Audouin, L. Rigollet, H. Prétrel, W. le Saux, M. Röwekamp, OECD PRISME system behavior, Fire Saf. J. 62 B (2013) 102–114.
project: Fires in confined and ventilated nuclear-type multi-compartments - [10] B. Hägglund, K. Nireus, P. Werling, Pressure rise due to fire growth in a closed
Overview and main experimental results, Fire Saf. J. 62 B (2013) 80–101. room. Description of three full-scale tests, FOA Defence Research Establishment,
[2] H. Prétrel, J.M. Such, Effect of ventilation procedures on the behaviour of a fire FOA-R-96-00347-2.4-SE, 1996, p. 27.
compartment scenario, Nucl. Eng. Des. 235 (2005) 2155–2169. [11] B. Hägglund, K. Nireus, P. Werling, Pressure rise due to fire growth in a closed
[3] H. Prétrel, W. le Saux, L. Audouin, Pressure variations induced by a pool fire in a room. An experimentalstudy of the smoke spread via ventilation ducts, FOA
well-confined and force-ventilated compartment, Fire Saf. J. 52 (2012) 11–24. Defence Research Establishment,FOA-R-98-00870-311-SE, 1998, p. 97.
[4] C. Forneau, C. Delvosalle, H. Breulet, S. Desmet, S. Brohez, Comparison of fire [12] R.K. Janardhan, S. Hostikka, Experiments and numerical simulations of pressure
hazards in passive and conventional houses, Chem. Eng. Trans. 26 (2012) effects in apartment fires, Fire Technol. 53 (2017) 1353–1377. http://dx.doi.org/
375–380. http://dx.doi.org/10.3303/CET1226063. 10.1007/s10694-016-0641-z.
[5] F.W. Mowrer, Enclosure Smoke Filling and Fire-Generated Environmental [13] J. Vinha, E. Manelius, M. Korpi, K. Salminen, J. Kurnitski, M. Kiviste,
Conditions, The SFPE Handbook of Fire Protection Engineering, in: M.J. Hurley A. Laukkarinen, Airtightness of residential buildings in Finland, Build. Env. 93
(Ed.)Society of Fire Protection EngineersFifth ed. (2016), 2016, pp. 1066–1101. (2015) 128–140. http://dx.doi.org/10.1016/j.buildenv.2015.06.011.
http://dx.doi.org/10.1007/978-1-4939-2565-0. [14] P. Blomqvist, L. Rosell, M. Simonson, Emissions from fires part II: Simulated room
[6] K.B. McGrattan, R. McDermott, J. Floyd, S. Hostikka, G. Forney, H. Baum, fires, Fire Technol. 40 (2004) 59–73. http://dx.doi.org/10.1023/
Computational fluid dynamics modelling of fire, Int. J. Comput. Fluid Dyn. 26 (6-8) B:FIRE.0000003316.63475.16.