Sie sind auf Seite 1von 22

Modeling and Simulation of High Pressure

Water Mist Systems


Topi Sikanen*, Jukka Vaari, Simo Hostikka and Antti Paajanen,
Fire Research, VTT Technical Research Centre of Finland, Kemistintie 3,
P.O. Box 1000, 02044 Espoo, Finland
Received: 16 March 2012/Accepted: 15 March 2013
Abstract. This paper describes work done to improve and validate the capability of
re dynamics simulator (FDS) to predict the dynamics of water mist sprays. Three
single orice and ve multi-orice spray heads are modeled with FDS based on infor-
mation on the ow-rate, spray angle, operating pressure and experimentally deter-
mined particle size distribution. The capability of FDS to predict the drop size,
velocity, mist ux and number concentration proles within the spray cone is asses-
sed. The eects of turbulence modeling on the predictions of the spray dynamics is
investigated. The capability of FDS to predict the air entrainment by high-speed
water sprays is validated using experiments in rectangular channels with open ends.
Keywords: CFD, Water mist, High pressure, Spray dynamics, Turbulence, Three-way coupling
1. Introduction
Water mist systems are widely used to protect people and property from res and
are increasingly being used for projects that are signicantly larger than the scale
of experimental test. Despite the large body of experimental work, no general
design and installation rules have emerged. The demonstration of the eectiveness
of large scale water mist systems then calls for large scale experiments. These large
scale experiments require considerable amounts of time and money. It is however
expected that the demands for both can be substantially reduced by the use of
numerical simulations as part of the R&D process.
Water mist systems suppress res by three main mechanisms: Removal of heat
from the gases, displacement of oxygen by water vapor and the attenuation of
radiation by droplets Grant et al. [4]. The relative importance of these mecha-
nisms depends on the water mist system in question as well as the application. In
order to use the numerical simulation to examine the eectiveness of water mist
systems, it is critical that we can predict these eects with sucient accuracy.
The key to water mist system performance is the droplet size: water mist is
made of very ne droplets. The increased surface area of the spray means that the
droplets vaporize quickly and often do not reach the burning surfaces. However,
the large surface area to volume ratio of ne droplets leads to high cooling capac-
ity in the gas phase. The small size of the droplets indicates a high degree of
* Correspondence should be addressed to: Topi Sikanen, E-mail: topi.sikanen@vtt.
Fire Technology
2013 Springer Science+Business Media New York. Manufactured in The United States
DOI: 10.1007/s10694-013-0335-8
1
coupling between the gas phase and the water mist. This makes modeling of water
mist systems challenging.
Another and sometimes overlooked key, especially for high pressure water mist
systems, is spray dynamics. Due to aerodynamic drag, individual ne water drop-
lets will decelerate quickly in the surrounding air even if their initial velocity is
high. To eectively distribute water mist, high pressure water mist nozzles often
consist of a number of micro-nozzles, each discharging a relatively narrow jet.
When the jet entrains surrounding air, the ne droplets in the core of the jet will
not experience signicant drag, and can therefore be transported far away from
the discharge orice. Depending on the nozzle design, individual jets may or may
not interact with each other. In the former case, relatively narrow spray patterns
are produced with a good penetration far into the protected space. In the latter
case, wider patterns are produced allowing larger nozzle spacing and good spray
coverage in shallow spaces. Understanding spray dynamics is crucially important
in the design of water mist systems, and being able to accurately model spray
dynamics is therefore needed whenever it is desired that the design of water mist
systems could be aided with simulations. The focus in this paper is the near-eld
dynamics of high-pressure water mist sprays.
A number of studies have utilized computational uid dynamics (CFD) models
for water mist systems. Prasad et al. [15, 16] and Shimizu et al. [18] studied extin-
guishment of small scale res by water mist using two-uid models. Two-uid
models were also used by Prasad et al. and Nmira et al. [12] to study extinguish-
ment by water mist in large enclosures and tunnels, respectively.
Kim and Ryou [8, 9] used the re dynamics simulator (FDS) version 3.0 for
modeling re suppression of pool res by water mist in steel enclosure. They mod-
ied the FDS-droplet model to include secondary breakup but did not report on
the eects of this modication. Hart [5] used Fluent and a Lagrangian spray
model to study re suppression by water mists. An extensive study of water mist
modeling by CFD was done by Husted [6] in his PHD Thesis. He used FDS ver-
sion 4.0 and compared predictions to experimental measurements. In the present
work a development version of FDS is used. The goal is to validate the FDS par-
ticle model for use in modeling water mist systems based on a series of validation
experiments. Both single orice nozzles and multi orice spray heads are consid-
ered.
2. Description of the Validation Experiments
Here we give only a brief description of the validation experiments. More com-
plete descriptions can be found in the report [19].
The experimental data used to validate the simulation results comes from to dif-
ferent experimental campaigns. In one set of experiments, single water mist nozzle
orice were characterized. The data was obtained by the Direct Imaging (Shadow-
graphy] technique, and it consisted of size distribution, velocity, mist ux and par-
ticle concentration at several locations across the spray cone. The measurement
locations were those dened in the NFPA 750 standard and an additional
Fire Technology 2013
measurements point on the center line of the cone. The center line measurement
point was added because otherwise the dense core of the sprays would have been
missed. All the measurements were made at 70 bar operating pressure.
In the second set of experiments mist nozzles were placed in a rectangular chan-
nel and the induced gas velocities in the channel were measured. Both single and
multi-orice spray heads were used. A smaller channel was used for single orice
nozzles and a larger channel for multi-orice nozzles
For single orice nozzles, the channel was 1.5 m long and had a cross section of
0.15 m by 0.15 m. The nozzles were installed 0.6 m from the downwind end of the
channel and they were spraying along the channel axis. The air velocity in the
direction of the channel axis was measured with a hot-wire anemometer. Measure-
ments were taken 0.45 m upwind of the nozzle on the channel axis.
For multi-orice spray heads, the channel was 2.0 m long and had a cross section
of 0.6 m by 0.6 m. The spray heads were installed at the midpoint of the channel and
they were spraying along the channel axis. Air velocity measurements were taken 0.5
m upwind the nozzle on the channel axis and 0.06 m from the channel wall.
3. Model Description
The FDS (svn revision 10155) is used in this study. The governing equations and
solution methods for the gas phase are described in the FDS Technical Reference
Manual [10] and will not be repeated here. FDS uses Large Eddy Simulation to
describe the turbulent motion of the gas phase and Lagrangian description of the
dispersed phase. The sprays in this paper are evaporating. The evaporation model
used is described in the FDS Technical Reference Manual [10]. However evapora-
tion is not expected to be important since we are simulating cold sprays. Further-
more, the residence times of the particles in the simulations are very short.
Secondary breakup was not considered in this study.
For this paper, modications were made to the FDS source code. A model for
aerodynamic interactions between droplets was added and a new shape function
was implemented for the spray boundary condition. These modications are
described in detail below.
3.1. Lagrangian Particle Model
Ignoring buoyancy, lift and forces arising from uid acceleration, the motion of a
single spherical droplet is governed by the equation of motion
dm
p
v
p
dt
= m
p
g q
g
C
D
pr
2
p
v
p
v
g
_ _
v
p
v
g
_
_
_
_
(1)
Here on the left hand side m
p
is the mass of the droplet and v
p
is the velocity of
the droplet. On the right hand side, g is the gravitational acceleration, q
g
is the
density of the surrounding gas, v
g
is the velocity of the surrounding gas, r
p
is the
radius of the droplet, and C
D
is the drag coecient. The drag coecient is given
by McGrattan et al. [10]
High Pressure Water Mist Systems
C
D
=
Re=24 Re <1
Re
24
1 0:15Re
0:683
_ _
1 <Re <1000
0:44 Re >1000
_
_
_
(2)
where Re = 2 t
p
t
g
_
_
_
_
r
p
=m is the droplet Reynolds number and m is the kinematic
viscosity of air. Due to the large number of droplets in a real spray, only a frac-
tion of these droplets is tracked. Each droplet in the simulation represents a parcel
of droplets with the same properties.
To ensure that the momentum lost by the droplets is distributed correctly in the
gas, an additional CFL-like condition can be placed on the global time step. The
global time-step of the numerical solver is adjusted according to
Dt _ Cmin
v
x
Dx
;
v
y
Dy
;
v
z
Dz
_ _
; (3)
where C is a user dened constant between 1 and 0, and v
x
, v
y
, and v
z
are the
three components of the particle velocity v
p
:
3.2. Three-way Coupling Between Droplets and Gas
Typically Lagrangian particle models only consider two-way coupling of the gas
phase and the dispersed phase. This means that each particle interacts with the
carrier uid individually. Momentum lost from a particle is added to the uid and
vice versa.
If the spray is dense enough, the individual droplets start to inuence each
other through aerodynamic interactions. These eects cannot be captured by the
current EulerianLagrangian model for two reasons: First, the Lagrangian parti-
cles have no volume in the Eulerian space, preventing the model from seeing
eects that take place in the length scale of the droplet diameter. Secondly, the
eects of this length scale would be sub-grid scale in most practical simulations.
These aerodynamic interactions may start to have an eect when the average
droplet spacing is <10 [13, 14] droplet diameters. This corresponds approximately
to a droplet volume fraction a = 0.01. Volume fractions as high as this can some-
times be achieved inside water mist sprays. If the spray was even more dense, par-
ticle-particle collisions or four-way coupling would need to be considered. The
sprays considered in this paper are relatively dilute, so only the possibility of
three-way coupling eects is considered.
In a conguration where two particles with the same diameter are directly in
line, the reduction of the hydrodynamic forces to the second (trailing) sphere due
to the wake eect was studied in [17]. They developed the following analytical for-
mula for the hydrodynamic force to the second sphere. In our work, this formula
is used to compute a reduction factor for drag coecient
C
D
= C
D
0
F
F
0
(4)
Fire Technology 2013
where C
D0
is the single droplet drag coecient and F/F
0
is the hydrodynamic
force ratio of trailing droplet to single droplet:
F
F
0
= W 1
Re
16
1
L=d 1=2 ( )
2
exp
Re
16
1
L=d 1=2 ( )
_ _
_ _
(5)
where Re is the single droplet-Reynolds number, L is the distance between the
droplets and W is the non-dimensional, non-disturbed wake velocity at the center
of the trailing droplet
W = 1
C
D
0
2
1 exp
Re
16
1
L=d 1=2 ( )
_ _ _ _
(6)
This model assumes that the spheres are traveling directly in-line with each other.
As such, this provides an upper bound for the strength of the aerodynamic inter-
actions between two particles of the same size. The separation distance L/d
between droplet centers is calculated from the local droplet volume fraction and
local average droplet diameter

d
L=d =

d p=6a ( )
1
3
(7)
Here local quantities are taken as average over a single computational cell.
In reality, the spray is of course not monodisperse and the separation distance
between the interacting particles varies. However seeking for each the closest par-
ticle to interact with would be very time consuming. In the simulations, the drag
reduction factor in Equation 4 is only used, when the local droplet volume frac-
tion exceeds 10
-5
. This drag reduction model is turned on by default.
An alternative data on drag reduction was provided by Prahl et al. [13] who
studied the interaction between two solid spheres in steady or pulsating ow by
detailed numerical simulations. According to their study, the above correlation
underestimates the drag reduction signicantly at small drop-to-drop distances.
The inow pulsations were found to reduce the eect of the drag reduction. At
large distances the two results are similar, the Ram rez-Mun oz correlation show-
ing more drag reduction. This is not surprising since the velocity prole of a fully
developed axisymmetric wake behind a axisymmetric body is used in developing
the drag reduction correction in Eqs. 5 and 6. At short distances, the wake is not
fully developed and the assumption does not hold. The sprays considered in this
paper are relatively dilute and hence these short separation distances are not
expected to be important.
3.3. Spray Boundary Condition
In this work, no attempt is made to model the primary or secondary atomization
processes of the spray. Instead the spray is described as a droplet inlet boundary
condition. The droplets are injected to the simulation on a section of a spherical
High Pressure Water Mist Systems
surface at distance l from the nozzle. The section of the surface where droplets are
launched is determined by the spray angle parameter.
It is assumed that all the atomization processes are over at the position of
boundary condition. It is easy to prescribe the ow rate and speed of droplets so
that correct amount of liquid phase momentum is introduced into the simulation.
However the concentration and velocity proles on the boundary require more
thought. The concentration of particles in a turbulent round jet has been observed
to follow a self-similar prole [7]. Thus the spray boundary condition is set to
mimic this behavior.
The initial position of a particle is picked randomly within the conical section
described previously. The variable ux density within the spray is implemented by
dening a probability density function for the initial position that depends on the
latitude h but not longitude u: The joint probability of the initial latitude and lon-
gitude is
p h; u ( ) = p h ( )p u ( ) =
1
2p
sin hf h ( ): (8)
Here the longitude is assumed independent of the latitude. If the function f h ( ) is
taken to be unity, the resulting mass ux is uniform. The exponential shape of the
mass ux is implemented by using the probability density function
f h ( ) = exp b
h h
min
h
max
h
min
_ _
2
_ _
(9)
The spread parameter b = 5 was used in this work (Figure 1). This value was
chosen so that the simulations best t the nozzle characterization experiments. All
the droplets are given the same initial speed in the direction of the surface normal.
3.4. Nozzle Modeling
Nozzle characterization was made with the aid of the experimental data from sin-
gle orice nozzle characterization experiments.
The droplet size distribution was found to vary with position in the spray. The
gross cumulative volume (GRV) droplet size distribution from all the measure-
ment points was therefore used in estimating the size distribution parameters. The
GRV distribution was calculated in accordance with the NFPA 750 as
GRV
j
=

i
R
i;j
A
i
V
i
_ _

i
A
i
V
i
( )
(10)
where GRV
j
is the cumulative volume fraction (CVF) of all droplets equal or less
than d
j
, R
i,j
is the CVF of droplets equal or less than d
j
at location i, A
i
is the
cross-sectional area at location i and V
i
is the mist ux at location i.
In the simulations the CVF of droplet size is assumed to follow a mixture of
RosinRammler and log-normal distributions
Fire Technology 2013
F d ( ) =
1

2p
_
_
d
0
1
rd
/
exp
ln d
/
=d
m
( ) [ [
2
2r
2
_ _
dd
/
d _ d
m
1 exp 0:693
d
d
m
_ _
c
_ _
d _ d
m
_
_
_
(11)
Here d
m
is the volumetric median diameter, while r and c are width parameters.
The width parameters are related by r = 1.15/c to ensure continuity at d
m
. In the
simulations the droplet diameter is picked according to the cumulative number
fraction (CNF) dened as
f d ( ) =
_
d
0
F
/
(d
/
)d
/
3
dd
/
_

0
F
/
(d
/
)d
/
3
dd
/
; F
/
=
dF (d)
dd
(12)
The distribution parameters were found by least squares t of Equation 12 to the
experimentally determined cumulative number distribution. The dierence between
using Eqs. 11 and 12 for parameter estimation is that the former places more
weight on large droplets while the latter emphasizes the smaller drop sizes. This
point was discussed extensively in [2]. Since in the numerical algorithm the droplet
sizes are picked from the CNF, it was also used in the parameter estimation. In
the simulations, the particle size is bounded from below by the parameter
d
min
= 1 lm. Droplets with diameters smaller than this are assumed to vaporize
instantly. Figure 2 shows a comparison of the tted CVF and experimental GRV.
For nozzles B and C, the amount of very large droplets is underestimated. For all
nozzles the amount of very small droplets is also underestimated.

R
f()
Rsin()
Figure 1. Illustration of the spray boundary condition.
High Pressure Water Mist Systems
The median droplet size is known to depend on the operating pressure used.
Since the experimental droplet size distributions were determined at certain pres-
sure (70 bar) but the simulations were also performed at other pressures, this vari-
ation in droplet size is taken into account by scaling the median droplet size as
d
m
~ p
-1/3
.
The spray angle h for each orice were determined from spray photographs.
This method is quite crude and furthermore depends somewhat on the choice of
the oset parameter. The estimated parameters for the three single-orice nozzles
are summarized in Table 1.
The tabulated ow constants (K-factors) for the high pressure nozzles are based
on the ow measurements performed for multi-orice spray heads involving a sin-
gle orice type. The ow constant K is dened as
K =
Q

p
_ ; (13)
where Q is the ow rate of water, and p is the water pressure measured at the
spray head. Strictly, the ow constant is not a constant, but depends slightly on
pressure. However, for practical purposes, it is sucient to determine the ow
constant for a single pressure representative of the working pressure range of the
spray head.
In all the simulations initial droplet velocities were calculated from
v
0
= 0:95

2p
q
l

; (14)
0 100 200 300 400 500 600
0
20
40
60
80
100
Diameter (m)
C
u
m
u
l
a
t
i
v
e

V
o
l
u
m
e

F
r
a
c
t
i
o
n

(
%
)
A B C
Experiment
Fit
Figure 2. Experimentally observed droplet CVF distributions versus
tted distributions.
Fire Technology 2013
where p is the operating pressure of the nozzle and q
l
is the density of the liquid.
Factor 0.95 was used to account for the friction losses in the nozzle. This value is
the authors estimate.
Table 2 gives the descriptions of the multiorice nozzles considered here. The
multiorice nozzles were modeled by positioning several single orice models with
dierent orientations at one point in the computational domain. Each individual
orice on the nozzle was modeled with the parameters given in Table 1. The cen-
ter nozzle points in the axial direction and the perimeter nozzles are equally
spaced and at the same angle in relation to the center nozzle. The smaller the
perimeter angle, the more parallel are the orices in the spray head.
4. Modeling and Simulation of Validation Experiments
Grid sensitivity studies and studies concerning the eects of three-way coupling
and turbulence modeling on the sprays is done via comparisons with nozzle char-
acterization tests. Air entrainment predictions are validated against induced ow
velocity data from several entrainment tests conducted for nozzles and multi-ori-
ce spray heads.
4.1. Nozzle Characterization Experiments
The experimental measurements and the calculation of GRV distribution were in
accordance with the NFPA750 standard except that one measurement point was
added in the center of the spray, since the measured nozzles (A, B and C) were
producing a relatively narrow cone with a dense core. The NFPA750 is more
intended for sprinklers with deector plates where the spray is wide and the center
of the spray is unoccupied with droplets. If the center point was not included, a
signicant amount of the mist ux would be ignored.
Table 1
FDS Simulation Parameters for Nozzles A,B and C
Nozzle K L= minatm
1=2
_ _
h () d
m
(lm) c
A 0.2 10 84 2.9
B 0.433 12 79 2.26
C 0.767 14 116 1.98
Table 2
Description of Multiorice Spray Heads
SH1 SH2 SH3 SH4 SH5
Center nozzle A C B B B
Perimeter nozzle A B A B B
Number of perimeter nozzles 6 6 8 8 8
Perimeter angle () 60 60 45 45 30
High Pressure Water Mist Systems
The experiments were modeled using a rectangular computational area 1.5 m
high, 0.5 m wide and 0.5 m deep. The computational area was open to ow on all
sides. The nozzles were placed 0.1 m from the top of the computational domain
and the measurements were taken 1 m below the nozzle. Figure 3 shows the com-
putational model and placement of the measurement points. The simulation
results correspond to droplet properties averaged over a sphere with 1 cm radius
cent red at the measurement location. After a sensitivity study described in section
5.2, 2 cm grid was selected. The same 2 cm grid was used for the air entrainment
simulations also. Simulation time was 5 s and the droplet statistics were collected
for the last 3 s of the simulations. The rst 2 s were ignored to ensure that the
simulation had reached steady state.
Figure 3. Pictures of the nozzle characterization simulations for noz-
zle A. On the left, the measurement locations with green dots. On the
right time averaged average velocity slice (Color gure online).
Fire Technology 2013
4.2. Air Entrainment
Induced air velocity data was used to validate the air entrainment predictions. The
data was from experiments where single- and multiorice nozzles were placed in a
rectangular channel.
The single nozzle experiment was modeled simply as a rectangular computa-
tional domain, with dimensions of the experimental channel. On the channel
walls, inert solid wall boundary conditions were applied and open pressure bound-
aries were used for the ends of the channel. In the multi-orice spray head experi-
ments, a wooden block near the inlet end of the channel aected the airow and
to capture this eect, the computational domain was extended outside the channel
as shown in Figure 4. Discretization interval was 2 cm in both channels.
The oset parameter was found to have a signicant eect on the simulation
results, especially for the multi-orice spray heads. Oset value of 0.04 m was
used here instead of the 0.1 m used for the simulation of the nozzle characteriza-
tion tests. Using a too small oset value was found to give unphysical results,
where the ow on the channel axis behind (upwind) the sprinkler was in the nega-
tive x-direction, while the sprinkler was oriented in the positive x-direction. It was
concluded that the correct oset value depends on the numerical grid used: The
oset should be large enough, so that the incoming droplets are distributed
among large enough number of computational cells. This eect is related to the
transfer of momentum from dispersed phase to the gas phase. By distributing the
droplets in a sucient number of cells, we ensure that the momentum source term
in the gas phase equations is better approximated. Naturally, the grid also needs
to be ne enough to resolve the important ow features near the nozzle. In the
case of multi-orice spray heads this means that it is important to ensure that the
Figure 4. Picture of the large channel computational model.
High Pressure Water Mist Systems
grid is ne enough that each of the spray jets is resolved. This becomes a problem
when the orices in the spray have orientations that are almost parallel.
5. Results and Discussion
The simulations were run on a computer cluster with 2.92 GHz, 8-core Intel Xeon
processors and 2 GB of memory per core. Simulations were ran as serial jobs,
with one processor handling a single simulation. The nozzle characterization simu-
lations took between 7 h and 22 h for 5 s of simulation time. The entrainment
simulations run from few hours for the small channel to 25 h for the large chan-
nel simulations.
5.1. Turbulence Modeling
By default FDS version 5 uses a constant Smagorinsky coecient model of turbu-
lent viscosity. However for the current simulations the dynamic Smagorinsky
model [11, 3] was used. In newer versions of FDS the constant Smagorinsky coef-
cient model is no longer the default. Instead several alternatives are given. Fig-
ure 5 shows a comparison of results between four dierent turbulent viscosity
models. The models considered are those available in the development version of
FDS used: Deardor [1], Vreman [20] the constant Smagorinsky and the dynamic
Smagorinsky coecients.
There is almost 50% dierence in the predicted velocities and mist uxes on the
spray axis. However, the velocity and mist ux proles are extremely sensitive to
the spray boundary condition. Since there is considerable uncertainty in dening
the shape function and spray angle for the spray boundary condition, these results
alone cannot be used to determine which turbulence model is more suited for this
application. The diameter prole on the other hand was found to be very insensi-
tive to the boundary condition. The correct at shape of the diameter prole is
only reproduced with the dynamic Smagorinsky model. Furthermore, the dynamic
Smagorinsky model produces overall better t of velocity and mist ux proles
than the other models. The results are also sensitive to grid size. These simulations
were run on a 2 cm grid but on ner grids Deardor, Vreman and dynamic Sma-
gorinsky models tend to converge on the same result.
This result suggests that the turbulent mixing of the particles is important at
these distances. Near the nozzle, the entrained air will tend to pull droplets
towards the middle of the spray. Smaller droplets have shorter response times and
thus are entrained more easily. This process should result in a diameter prole
that shows mores small droplets at the center of the spray and larger droplets on
the spray boundaries. However the droplet size in these simulations is very small
and their relative velocities are relatively small at 1 m distance from the spray
nozzle, where the measurements are made. At this point the particles tend to dis-
perse towards the edges of the spray. The smallest particles are inuenced more
by the turbulent dispersion than heavier particles which might still have signicant
relative velocity. The sum of these eects is that the dierences in average particle
size are diminished.
Fire Technology 2013
It is not immediately clear how important this mixing is for predicting the e-
ciency of water mist systems. However, within the dense spray core where most of
the water in the spray is located, the droplet size predictions are similar for both
turbulence models tested.
5.2. Sensitivity Study
Sensitivity of the single orice models to model parameters and numerical param-
eters was studied. The model parameters considered were the spray angle h and
initial velocity v
0
. The numerical parameters included discretization interval Dx;
eect of the CFL parameter in Eg. 3, eect of oset parameter R and the droplet
insertion rate. The parameters for the base case are given in Table 3. The grid res-
olution and DPS-parameters were arrived at through a sensitivity study. The o-
set parameter and CFL-condition limit were set at their default values for the base
case. Table 4 presents the full experimental design for sensitivity study and the
error in each simulation. The error in the simulations is quantied as
e =
max
i
E
i
max
i
S
i
max
i
E
i
; (15)
0 5 10 15 20
0
10
20
30
40
Radial position (cm)
V
e
l
o
c
i
t
y

(
m
/
s
)
Exp
CSmag
DSmag
Deardorff
Vreman
(a)
0 5 10 15 20
0
5
10
15
20
Radial position (cm)
M
i
s
t

f
l
u
x

(
k
g
/
m
2

s
)
Exp
CSmag
DSmag
Deardorff
Vreman
(b)
0 5 10 15
40
60
80
100
120
140
160
Radial position (cm)
d
1
0

(

m
)
Exp
CSmag
DSmag
Deardorff
Vreman
(c)
Figure 5. Effect of turbulence model on average droplet velocity,
mist ux and diameter proles. Results shown for nozzle B.
High Pressure Water Mist Systems
where E
i
and S
i
are the experimental measurements at measurement point number
i and simulation results respectively.
The grid sensitivity study was conducted using a uniform Cartesian grid with
discretization intervals 1 cm, 2 cm and 4 cm. The results of the grid sensitivity
study were similar for all the single orice nozzles. Figure 6 shows the results for
nozzle B. In general, coarser grids produced atter velocity and concentration dis-
tributions. However, the diameter distribution retained its shape. It can be seen
from Table 4 that going from 2 cm grid to 1 cm grid did not improve the enough
to warrant the increased computational cost. Increasing the number of Lagrangian
particles inserted to the spray from 2 10
5
to 4 10
5
s
-1
tended to yield larger
velocities on the spray center line but did not otherwise aect the results. The
number of Lagrangian particles in the simulation at any given time was ~17000
for the base case. The eect of the time step was studied by using values 0.1, 0.5
and 1 for the parameter C in Equation 3. For the base case, the time step was
approximately 10 ms for the steady state phase of the simulation. Overall the
time-step had very little eect on the results.
Oset values from 5 cm to 15 cm were tested with little eect on the results in
the nozzle characterization test. However it should be stressed that it was discov-
ered later that the oset parameter was important in modeling multi orice noz-
zles. Varying the initial velocity by 6 m/s changed the results only by a few
percent. The spray angle parameter h had the largest eect on the results with
narrowing the angle by 5 increasing the error by as much as 120% points.
Choosing a smaller spray angle increased error more than choosing a larger angle.
The minimum diameter d
min
was set to 1 lm in most simulations. No droplets
smaller than this diameter are added to the simulation. Also, droplets smaller than
this are assumed to vaporize instantly. This parameter was sometimes increased to
alleviate problems with numerical stability. Sensitivity analysis showed that this
parameter did not have a noticeable eect on the results.
5.3. Three-way Coupling
The importance of three-way coupling eects on water mist properties was investi-
gated by running the nozzle characterization tests, with the aerodynamic interaction
Table 3
Parameters for the Base Case in Sensitivity Study
Parameter Value Description
Dx 0.02 m Discretization interval
DPS 2 10
5
How many droplets per second are inserted to the
simulation
3-way coupling On Use the three-way coupling model from Section 3.2
v
0
112 m/s Initial velocity of droplets
CFL 1 Coecient C in Equation 3
R 0.1 m Oset parameter for the spray boundary condition
h - See Table 1
Fire Technology 2013
model turned on and turned o. Figure 7 shows the eects of the drag reduction
model on the simulation results for nozzle C. Nozzle C has the highest ow rate
and densest spray, being thus the best candidate for a spray where aerodynamic
interactions are important. Results for other nozzles were very similar and are
omitted for brevity The drag reduction by aerodynamic interactions had a very
modest eect on the results. The most noticeable eect was the slight attening of
the droplet diameter prole. The droplet volume fractions in the densest parts of
the spray are just slightly over a > 0.01 for all nozzles. These results indicate that
droplet-droplet aerodynamic interactions are not important in modeling the cur-
rent water mist systems. The three-way coupling model was used in all simulations
of this paper.
5.4. Nozzle Characterization Experiments
Here we give an overview of the results for the simulation of the nozzle character-
ization tests. Note that the nozzle characterization experiments were also used in
selecting the employed turbulence model and the shape function for the spray
boundary condition.
The numerical parameters used in these simulations are listen in Table 3. These
parameters were selected after the sensitivity study described in Sect. 5.2 and same
set of numerical parameters is used for all nozzles. One important result of the
sensitivity study was that, reducing the grid size and increasing the number of
Table 4
Sensitivity Analysis Table with Global Errors (%) for Each Parameter
Combination
Test
Nozzle
A B C
Vel. Diam. Flux. Vel Diam Flux Vel Diam Flux
Dx = 0:01 m 10.3 7.20 52.93 1.99 50.42 9.73 29.51 11.57 12.84
Dx = 0:02 m 9.72 7.90 52.53 13.43 2.71 28.14 11.09 11.72 1.21
Dx = 0:04 m 24.67 13.81 79.06 34.89 8.33 63.68 14.3 6.73 42.63
3-way O 13.3 13.93 46.99 10.4 4.76 18.38 13.76 9.59 16.43
DPS = 4 10
5
8.37 8.99 54.21 12.06 2.17 23.46 11.91 13.21 0.01
v
0
= 106 m/s 6.05 8.68 52.9 7.83 1.22 13.31 5.63 12.14 1.01
v
0
= 118 m/s 11.83 6.86 54.39 3.34 6.86 13.22 14.4 12.69 4.21
CFL = 0.1 7.05 16.91 57.14 12.48 24.62 27.86 10.4 13.36 1.43
CFL = 0.5 9.45 7.57 53.09 9.78 1.36 25.17 13.88 7.81 4.15
R = 0.05 m 15.92 6.84 47.58 7.83 0.69 16.84 17.67 20.02 12.43
R = 0.08 m 9.72 7.9 52.53 13.43 2.71 28.14 11.81 5.28 4.98
R = 0.12 m 3.92 9.2 55.88 13.93 1.93 27.88 9.45 13.83 3
R = 0.15 m 0.8 9.46 61.63 16.9 5.26 32.23 8.44 10.23 7.12
h - 5 77.87 5.51 104.85 33.3 40.48 165.16 40.44 9.77 134.7
h + 5 13.53 4.35 77.36 25.5 2.27 59.57 0.92 15.89 32.93
Base case is Dx = 0:02 m
High Pressure Water Mist Systems
particles used to describe the spray doesnt always lead to better results. Better ts
than those shown in Figure 8 could be found if the numerical and model parame-
ters were selected in tandem. Where possible, we chose to nd the model parame-
ters from experimental data and without consideration of the numerical
parameters.
The comparison of the predicted and experimentally observed velocity, mist ux
and diameter proles are shown in Figure 8. Velocity on the spray axis is under
predicted for nozzle B and over predicted for nozzle A. For nozzle C the velocities
are over predicted throughout the spray.
The mist ux proles in Figure 8 are over predicted for nozzle A, signicantly
under predicted for nozzle B and well matched for nozzle C. Further away from
the spray center line, the discrepancy between the simulation and experiment
diminishes. It should be noted, however, that there is considerable measurement
uncertainty associated with the experimental mist ux values.
Qualitatively the mist ux proles are correct: There is a relatively thin dense
core with a less dense outer spray as observed during experiments. On the spray
axis, the mist ux is best predicted for nozzle C with only 1.2% error. For nozzles
A and B the errors are 52% and 28% respectively, with local mist uxes over pre-
dicted for nozzle A and under predicted for nozzle B.
The experimental data available was for the number median diameter, while
only number mean was calculated from the simulations. The experimental data on
0 5 10 15 20
0
5
10
15
20
25
30
Radial position (cm)
V
e
l
o
c
i
t
y

(
m
/
s
)
Experiment
x = 1
x =2
x 4
(a)
0 5 10 15 20
0
2
4
6
8
10
Radial position (cm)
M
i
s
t

f
l
u
x

(
k
g
/
m
2

s
)
Experiment
x = 1
x =2
x 4
(b)
0 5 10 15
40
50
60
70
80
90
100
110
Radial position (cm)
d
1
0

(

m
)
Experiment
x = 1
x =2
x 4
(c)
Figure 6. Nozzle B grid sensitivity.
Fire Technology 2013
diameter in Figure 8 is derived from the measured droplet CNFs at these points.
The almost at shape of the droplet size proles is predicted for all the nozzles.
For nozzles A and B, the mean diameter predictions are within 10% of the experi-
mental value and within 11% for nozzle C.
5.5. Air Entrainment
Figure 9 shows snapshot of particle trajectories and the average velocity eld at
the center of the channel. From the particle trajectories we can see that a consid-
erable amount of the spray hits the walls of the channel. The average velocity eld
shows the eect of the block at the channel entrance on the ow as the ow is not
symmetrical at the channel entrance.
Comparisons of the air entrainment simulations to the experimental results are
shown in Figure 10. The error bars correspond to two standard deviations of the
measured values. They do not account for the possible systematical errors in the
measurement.
Velocities on the center line of the channel are shown in Figure 10a for the multi-
orice spray head cases and in Figure 10c for the single orice spray heads. Veloci-
ties near the channel wall (for multi-orice spray heads) are shown in Figure 10b.
Of the single-orice nozzles, the entrainment for B nozzle is closest to the exper-
imental value. For nozzle C the velocities in the channel are overestimated by
0 5 10 15 20
0
5
10
15
20
25
30
Radial position (cm)
V
e
l
o
c
i
t
y

(
m
/
s
)
Experiment
Base
Drag reduction
(a)
0 5 10 15 20
0
2
4
6
8
10
12
Radial position (cm)
M
i
s
t

f
l
u
x

(
k
g
/
m
2

s
)
Experiment
Base
Drag reduction
(b)
(c)
0 5 10 15
40
50
60
70
80
90
Radial position (cm)
d
1
0

(

m
)
Experiment
Base
Drag reduction
Figure 7. Effect of three-way coupling on nozzle characterization
simulations. Results for nozzle C.
High Pressure Water Mist Systems
about 20% and for nozzle A the velocities are underestimated by similar amount.
This is the same behavior that was discovered in the nozzle characterization simu-
lations, where droplet velocities were over predicted for nozzle C and under pre-
dicted for nozzle A. The dierence between experiments and simulation of
induced air velocities is signicantly larger than the dierence in average droplet
velocity proles.
For the multi-orice spray heads, the simulation results fall within the error
bound of experimental results only for SH2 at 50 bar and 70 bar pressures. Other-
wise velocities at the channel center are under predicted for SH5, SH3 and SH1
and over predicted for SH2 at 100bar. The induced velocities near the channel
wall are within two standard deviations of the measured value for all spray heads
except SH4 and SH5. For these spray heads the velocities near the wall are over
predicted.
The largest errors are seen in the predictions for spray heads SH5 and SH4.
While the spray heads SH4 and SH5 are both constructed from the same B type
orices, the velocities in the channel axis are over-predicted for SH4 and under
predicted for SH5. The dierence between these spray heads is in the amount of
x-momentum injected in to the simulation. The SH5 spray head has the perimeter
nozzles at an 30 angle relative to the x-axis giving the most x-momentum of all
the sprays considered in this paper.
0 5 10 15
0
10
20
30
40
Radial position (cm)
V
e
l
o
c
i
t
y

(
m
/
s
)
C Exp.
B Exp.
A Exp.
C Sim.
B Sim.
A Sim.
(a)
(c)
0 5 10 15
0
5
10
15
Radial position (cm)
M
i
s
t

f
l
u
x

(
k
g
/
m
2
s
)
C Exp.
B Exp.
A Exp.
C Sim.
B Sim.
A Sim.
(b)
0 5 10 15 20
30
40
50
60
70
80
90
Radial position (cm)
M
e
a
n

d
i
a
m
e
t
e
r

(

m
)
C Exp.
B Exp.
A Exp.
C Sim.
B Sim.
A Sim.
Figure 8. Simulated versus experimental velocity, mist ux and
mean diameter proles in the NFPA tests for nozzles A,B and C.
Fire Technology 2013
Dierences in the results for dierent spray heads may have to do with interac-
tions of the spray jets. Parallel jets will be entrained in to a single jet after a dis-
tance that depends on the jet boundary conditions. For spray heads SH1 and SH2
the perimeter nozzles are at an 60 angle with the center nozzle. Therefore the
interaction between the perimeter jets and the central jet is weaker than for the
other cases. Furthermore, for these two spray heads, a large portion of the spray
hits the channel walls. With lower perimeter nozzle angles, the spacing between
the jets diminishes and the interactions grow stronger. For spray head SH5 the
interaction of the spray jets is the strongest. The results for SH3, SH4 and SH5
might be explained by wrong prediction of the aerodynamic interaction dynamics
of the individual spray-jets from the perimeter and center nozzles. However with
the available data, this hypothesis cannot be conrmed.
Resolving the individual spray jets was already found to be important when
conducting a sensitivity study for the large channel simulation cases. If a too
small oset parameter was used, the spray jets would fuse into one jet and
momentum was lost from the simulation. This would also lead to some numerical
artifacts like unphysical ow. Using ne enough grid to accurately describe, and
more importantly, discriminate between the individual jets is not enough to get
the entrainment predictions correct. This remains a topic that requires further
experimental and theoretical work.
Figure 9. Snapshots from the large channel simulations for spray
head SH1. Picture on the top shows particle trajectories in the simula-
tion. Also shown are the two gas velocity measurement points behind
the spray head. The picture on the bottom shows average gas velocity
at the center of the channel.
High Pressure Water Mist Systems
6. Conclusions
Three single orice and ve multi-orice spray heads were modeled with the FDS
based on ow-rate, spray angle, operating pressure and experimentally determined
50 60 70 80 90 100 110 120 130
4
6
8
10
12
14
16
Pressure (bar)
V
e
l
o
c
i
t
y

(
m
/
s
)
SH1 Sim.
SH1 Exp.
SH3 Sim.
SH3 Exp.
SH4 Sim.
SH4 Exp.
(a)
50 60 70 80 90 100 110 120 130
4
6
8
10
12
14
16
Pressure (bar)
V
e
l
o
c
i
t
y

(
m
/
s
)
SH2 Sim.
SH2 Exp.
SH5 Sim.
SH5 Exp.
(b)
50 60 70 80 90 100 110 120 130
2
3
4
5
6
7
8
9
10
11
Pressure (bar)
V
e
l
o
c
i
t
y

(
m
/
s
)
SH1 Sim.
SH1 Exp.
SH3 Sim.
SH3 Exp.
SH4 Sim.
SH4 Exp.
(c)
50 60 70 80 90 100 110 120 130
2
3
4
5
6
7
8
9
10
11
Pressure (bar)
V
e
l
o
c
i
t
y

(
m
/
s
)
SH2 Sim.
SH2 Exp.
SH5 Sim.
SH5 Exp.
(d)
40 50 60 70 80 90 100 110
4
6
8
10
12
14
16
18
Pressure (bar)
V
e
l
o
c
i
t
y

(
m
/
s
)
A Sim
A Exp
B Sim
B Exp
C Sim
C Exp
(e)
Figure 10. Average induced velocities in the air entrainment tests.
a, b Results on the center line and c, d at the wall of the large chan-
nel, respectively. e The results for the small channel tests. The pres-
sures are exactly 50 bar, 70 bar or 100 bar in all cases.
Fire Technology 2013
particle size distribution. The observed at droplet diameter distribution was
reproduced when the Dynamic Smagorinsky turbulence model was used. How-
ever, the importance of the turbulent mixing of droplets on the eectiveness of
water mist systems could not be determined. Three-way coupling eects were
found to be small. This suggests that droplet densities in individual water mist
sprays under investigation are not high enough for droplet-droplet aerodynamic
interactions to be important.
The predictions of spray induced gas velocities inside the rectangular channels
were found to be within 30% of the experimental values. This indicates the accu-
racy that can be expected from the simulations where the capability of the water
mist to penetrate to the vicinity of re and to mix the gas space are important.
The dynamics of the spray were insensitive to the actual droplet size distribution
or the number of particles used to describe the spray.
It was found that the numerical grid had a large eect on the simulation results.
For multi-orice spray heads it is important that each of the individual orices
discharges within a dierent computational cell. This implies that the oset
parameter and grid resolution need to be selected so that there are separate spray
jets for each orice. This is challenging to achieve if the perimeter angle of the
spray nozzles is small. In addition, the number of Lagrangian particles used to
describe the spray needs to be suciently high. Larger the number particles used
to describe the spray, the smoother the predicted droplet density eld is.
The numerical resolutions of the simulations in this paper can be unattainable
in typical re safety engineering applications. Further model development is nee-
ded to facilitate simulation of large scale water mist systems. It should be noted
however that the ne grids and high delity simulation is mostly needed in the
near eld of nozzles. Further away from the nozzles, perhaps coarser grids would
suce. The small scale validation tests considered here do not address this ques-
tion.
7. Acknowledgments
The authors would like to thank Maria Putkiranta, Riina Rajala and Pentti Saa-
renrinne from Tampere University of Technology for conducting the direct imag-
ing experiments. Technical assistance was provided by Mr. Peter Gro nberg, Mr.
Ville Heikura, Mr. Toni Neitola and Mr. Veli-Pekka Vaari of VTT. The work
was sponsored by the Finnish Funding Agency for Technology and Innovation,
Mario Corporation Oy, Rautaruukki Oyj, YIT Kiinteisto tekniikka Oy and
Insino o ritoimisto Markku Kauriala Ltd.
References
1. Deardor JW (1972) Numerical investigation of neutral and unstable planetary bound-
ary layers. J Atmos Sci 29:91115. doi:10.1175/1520-0469(1972)029<0091:NIO-
NAU>2.0.CO;2
High Pressure Water Mist Systems
2. Ditch B, Yu HZ (2008) Water mist spray characterization and its proper application
for numerical simulations. Fire Saf Sci 9:541552. doi:10.3801/IAFSS.FSS.9-541
3. Germano M, Piomelli U, Moin P, Cabot WH (1991) A dynamic subgrid-scale eddy vis-
cosity model. Phys Fluids A 3(7):17601765. doi:10.1063/1.857955
4. Grant G, Brenton J, Drysdale D (2000) Fire suppression by water sprays. Prog Energy
Combust Sci 26(2):79130. doi:10.1016/S0360-1285(99)00012-X
5. Hart R (2006) Numerical modelling of tunnel res and water mist suppression. PhD
thesis, University of Nottingham. http://etheses.nottingham.ac.uk/185/1/thesis.pdf
6. Husted B (2007) Experimental measurements of water mist systems and implications
for modelling in CFD. PhD thesis, Lund University
7. Kennedy IM, Moody MH (1998) Particle dispersion in a turbulent round jet. Exp Ther-
mal Fluid Sci 18(1):1126. doi:10.1016/S0894-1777(98)10009-2
8. Kim SC, Ryou HS (2003) An experimental and numerical study on re suppression
using a water mist in an enclosure. Build Environ 38(11):13091316. doi:10.1016/S0360-
1323(03)00134-3
9. Kim SC, Ryou HS (2004) The eect of water mist on burning rates of pool re. J Fire
Sci 22(4):305323. doi:10.1177/0734904104041796
10. McGrattan KB, Hostikka S, Floyd JE, Mell WE, McDermott R (2007) Fire dynamics
simulator, technical reference guide, vol 1: Mathematical model. NIST Special Publica-
tion 1018, National Institute of Standards and Technology, Gaithersburg
11. Moin P, Squires K, Cabot W, Lee S (1991) A dynamic subgrid-scale model for com-
pressible turbulence and scalar transport. Phys Fluids A 3(11):27462757. doi:
10.1063/1.858164
12. Nmira F, Consalvi J, Kaiss A, Fernandezpello A, Porterie B (2009) A numerical study
of water mist mitigation of tunnel res. Fire Saf J 44(2):198211. doi:10.1016/
j.resaf.2008.06.002
13. Prahl L, Holzer A, Arlov D, Revstedt J, Sommerfeld M, Fuchs L (2007) On the inter-
action between two xed spherical particles. Int J Multiph Flow 33(7):707725. doi:
10.1016/j.ijmultiphaseow.2007.02.001
14. Prahl L, Jadoon A, Revstedt J (2009) Interaction between two spheres placed in tandem
arrangement in steady and pulsating ow. Int J Multiph Flow 35(10):963969.
doi:10.1016/j.ijmultiphaseow.2009.05.001
15. Prasad K, Li C, Kailasanath K (1999) Simulation of water mist suppression of small
scale methanol liquid pool res. Fire Saf J 33(3):185212. doi:10.1016/S0379-
7112(99)00028-4
16. Prasad K, Patnaik G, Kailasanath K (2002) A numerical study of watermist suppres-
sion of large scale compartment res. Fire Saf J 37(6):569589. doi:10.1016/S0379-
7112(02)00004-8
17. Ram rez-Mun oz J, Soria A, Salinas-Rodr guez E (2007) Hydrodynamic force on inter-
active spherical particles due to the wake eect. Int J Multiph Flow 33(7):802807.
doi:10.1016/j.ijmultiphaseow.2006.12.009
18. Shimizu H, Tsuzuki M, Yamazaki Y, Koichi Hayashi A (2001) Experiments and
numerical simulation on methane ame quenching by water mist. J Loss Prev Process
Ind 14(6):603608. doi:10.1016/S0950-4230(01)00055-9
19. Vaari J, Hostikka S, Sikanen T and Paajanen A (2012) Numerical simulations on the
performance of water-based re suppression systems VTT TECHNOLOGY 54,
http://www.vtt./inf/pdf/technology/2012/T54.pdf. Accessed 12 Jan 2013
20. Vreman AW (2004) An Eddy-viscosity subgrid-scale model for turbulent shear ow:
algebraic theory and applications. Physics of uids, 16:3670. doi:10.1063/1.1785131
Fire Technology 2013

Das könnte Ihnen auch gefallen