You are on page 1of 11

Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries


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

Numerical simulation of evaporation of volatile liquids


A.D. Galeev*, A.A. Salin, S.I. Ponikarov
Kazan National Research Technological University, Department of Mechanical Engineering, 68 Karl Marx Str., 420015 Kazan, Russian Federation

a r t i c l e i n f o a b s t r a c t

Article history: The paper presents the results of the validation of the developed pool evaporation model using literature
Received 21 November 2014 and our own experimental data. The proposed model was used to examine the effect of wind velocity
Received in revised form and pool sizes on the evaporation rate of volatile liquid (hexane). Contrary to the semi-empirical
27 June 2015
evaporation model widely used in hazard assessment, stronger dependence of evaporation rate on
Accepted 28 August 2015
pool size at low wind speeds is obtained.
Available online 1 September 2015
© 2015 Elsevier Ltd. All rights reserved.
Keywords:
Liquid spill
Pool evaporation
Model validation
Numerical simulation

1. Introduction effect of high vapor pressure on the mass transfer process.


Khajehnajafi and Pourdarvish (2011) used Churchill's equation
The correct prediction of the consequences of hypothetical (Churchill, 1976) where the Nusselt number is determined for the
accidental scenarios is an essential component for developing and complete range of Re and Pr covering laminar, transition and tur-
implementing appropriate protective capabilities and plan miti- bulent regimes. This equation is based on experimental data for
gation measures. This requires the use of the reliable mathematical forced convection from isothermal flat plates. The Churchill's
models describing how toxic chemicals or flammable substances equation is applied to mass transfer calculation substituting the Sh
are released and dispersed in the air. A complete model for an event and Sc for the Nu and Pr. The mathematical model, represented by
can roughly be divided into three parts: source modeling, disper- Habib et al. (2009), based on numerical calculation of differential
sion modeling and effect modeling (Vik and Pettersson Reif, 2011). transient equations for boundary layer together with algebraic
There is a lot of work concerning the modeling of hazardous sub- turbulence model of Cebeci-Chang (Cebeci and Chang, 1978),
stances dispersion in the environment, whereas little attention has including correction for surface roughness. The above models have
been given to source term modeling. In emergency spills of stable been developed on the assumption that the vapor behaves as a
liquids, the input of hazardous substances into the environment is passive contaminant, i.e. it does not affect the flow field. This
the result of evaporation from the pool surface. assumption may be invalid when the molecular weight of evapo-
Generally, there are two main types of mathematical models to rating component differs from the molecular weight of environ-
evaluate the evaporation rate from a pool: semi-empirical and ment air (Desoutter et al., 2009).
computational fluid dynamics (CFD) models. In the area of hazard By using the CFD, it is possible to overcome the limitations of
analysis of accidental chemical spills, many investigators and haz- existing semi-empirical models. This approach by solving three-
ard reference books have applied Mackay and Matsugu mass dimensional conservation equations for mass, momentum and
transfer equation (Kawamura and Mackay, 1987; Mackay and energy makes it possible to take into account the complex inter-
Matsugu, 1973). In the model (Kapias and Griffiths, 1999), the action between the evaporation process and the vapor dispersion.
mass transfer coefficient is calculated using Brighton's theory Many works have used CFD to examine the evaporation rate of
(Brighton, 1987). This analytical model takes into account the ef- water (Raimundo et al., 2014) or dilute solutions of ammonia (Rong
fects of surface roughness, friction velocity of the airflow, and the et al., 2010; Rong et al., 2011; Saha et al., 2011). In the work (Rong
et al., 2010) the effects of airflow and aqueous ammonia solution
temperature on ammonia transfer are investigated and the
* Corresponding author. numerical sensitivity studies are performed by Rong et al. (2011) to
E-mail address: galeev_ainur@mail.ru (A.D. Galeev).

http://dx.doi.org/10.1016/j.jlp.2015.08.011
0950-4230/© 2015 Elsevier Ltd. All rights reserved.
40 A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

examine the effects of computational geometry and inlet turbu- 1973; Kawamura and Mackay, 1987) it is assumed, that evaporation
lence intensity on ammonia emissions. In these studies, the rate Jg,s depends on pool size L as Jg,s ~ L0.11. The exponent on the pool
ammonia was a passive contaminant and has no impacts on airflow. dimension L was obtained by evaporating water from pools of
Raimundo et al. (2014) used CFD model to investigate the mass different sizes (Mackay and Matsugu, 1973). This exponent does not
transfer at the free surface of the water tank. In the above- reflect the buoyancy effects because the evaporation rate of water is
mentioned works, the liquideair interface is assumed to be low. The buoyancy effects are caused by the density difference
isothermal due to the small liquid volatility. When volatile liquids, between the evaporating component and ambient air and they may
such as hexane and acetone, evaporate, the liquid temperature may be significant when a pool of volatile liquid has large sizes and wind
significantly change due to heat loss. Furthermore, the evaporation speed is low. It is evident that further studies are necessary to bring a
rate of volatile liquids may be significantly influenced by Stefan better understanding to the role of pool dimensions and buoyancy
flux, which is not taken into account in above CFD models. effects in evaporation of volatile liquid.
Liu et al. (2011) developed a methodology using CFD to simulate
the boiling process of liquid nitrogen. The vaporization rate of a 2. Pool evaporation model and its validation
boiling liquid is governed by the heat transfer phenomena
including conduction, convection and thermal radiation mecha- The mass flow of vapor from the pool surface Jg,s due to evap-
nisms (Ve chot et al., 2013), while the evaporation rate of non- oration was calculated on the basis of the standard wall functions
boiling liquids is controlled by the removal of vapor from the (Fluent, 2006) taking into account the correction for Stefan flow:
pool surface by airflow (van den Bosch, 1997). Thus, for boiling and
 
non-boiling liquids the using of CFD code is reasonable to improve Yg;s  Yg;P rCm0:25 k0:5
P
the prediction of convective heat and mass transfer, respectively. Jg;s ¼K (1)

The present paper aims to validate the developed pool evapo-
ration CFD model using literature and our own experimental data (  
and to study the effect of the pool sizes on the evaporation rate Sc$yþ yþ < yþ
C
þ    ;
Y ¼ (2)
from volatile liquid spill using numerical simulation. In the previ- Sct uþ þ PC yþ > yþ
C
ous work (Galeev et al., 2013a) the pool evaporation model was
validated using only experimental data on liquefied butane evap-
oration. However, before to use a mathematical model in risk rCm0:25 k0:5
P yP
yþ ¼ ; (3)
assessment, the model must be carefully tested against experi- m
mental data from different sources. Therefore, in the present work
we also compared simulation results both with the experimental 1  þ
data on evaporation of ethanol and cyclohexane (Habib et al., 2009) uþ ¼ ln Ey  DB; (4)
k
and with our own experimental data on evaporation of acetone.

8  
>
> 0 Ksþ < 2:25
>
>
>
> !
>
> n  o
<1 Ksþ  2:25
DB ¼ ln þ Cs $Ks $sin 0:4258$ ln Ksþ  0:811
þ
ð2:25 < Ksþ < 90Þ ; (5)
>
> k 87:75
>
>
>
>1    
>
>
: ln 1 þ Cs $K þ Ksþ > 90
s
k

The pool evaporation rate sensitivity due to pool length has not
been fully addressed in the literature. In the work (Khajehnajafi and
Pourdarvish, 2011) the pool size determines the different flow re-
gimes over the flat surface. Depending on the intensity of turbu-
lence in the main stream, the flow near the surface becomes r$Ks $Cm0:25 k0:5
turbulent at some downstream distance (Schlichting, 1968). The Ksþ ¼ P
; (6)
m
point of transition from laminar to turbulent boundary layer flow
may be estimated in terms of the length Reynolds number, " 3=4 #
ReL ¼ ULn1, where U is the main stream air velocity (ms1), L is the PC ¼ 9:24
Sc
 1 ½1 þ 0:28expð  0:007Sc=Sct Þ; (7)
pool length in the direction of the airflow (m) and n is the kinematic Sct
viscosity of air (Nielsen et al., 1995). For the length Reynolds
number, the critical value of 5  105 is often assumed (Incropera   
ln 1  Yg;P 1  Yg;s
et al., 2006). The critical values for the length Reynolds numbers K¼ : (8)
from 1  105 to 3  105 have been reported (Pasquill, 1943; Bird Yg;s  Yg;P
et al., 1960; Schlichting, 1968; Coulson and Richardson, 1993; It is assumed that the transition between the fully turbulent
Nielsen et al., 1995). When considering accidental spills, the pool region and the viscous layer near the wall occurs at a value yþ
C of 11,
sizes are large and ReL exceeds the critical value and the prevalence independent of the concentration of the evaporating component.
of turbulent mechanism of the transport of vapor away from the In Eqs. (1)e(6) the coefficient Cm is assumed to be a constant:
evaporating pool is expected. In the model (Mackay and Matsugu, Cm ¼ 0.09. In the gas dispersion model, this coefficient is
A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49 41

determined by relating Cm to the mean flow deformation as adopted The mass flux Jg,s is used as a boundary condition for the
in the Realizable keε turbulence model (Shih et al., 1995). dispersion model in the pool area. The atmospheric dispersion of
The change of local liquid temperature is calculated from the vapor is calculated by solving the three-dimensional Reynolds-
equation averaged NaviereStokes equations and the energy and species
equations. The governing equations are closed using the Realizable
vTliq qa þ qgrd þ qs  qp þ qar  Jg;s $DHg keε equations for turbulence.
¼ ; (9)
vt CP;liq $mliq To calculate the dispersion of vapor in the atmosphere, it is
important to define the correct profiles for the mean velocity, tur-
where qa is the heat flux from the atmosphere, W/m2; qgrd is the bulence kinetic energy and its dissipation rate on the inlet to the
heat flux from the ground to the liquid, W/m2; qs is the heat flux computational domain. These inlet profiles should be maintained
from solar radiation, W/m2; qp is the heat flux emitted by the pool throughout a calculation domain. The fully developed profiles
due to long-wave radiation, W/m2; qar is the heat flux absorbed by provided by Richards and Hoxey (1993) are mathematically
the pool due to long-wave radiation from the atmosphere, W/m2 consistent, i.e. they are a solution of the mathematical models
The heat flux from the ground to the liquid phase, describing a homogeneous atmospheric boundary layer (Parente
qgrd ¼ lgrd(vTgrd/vy)y¼0 is found from the numerical solution of et al., 2011). However, the assumption of constant kinetic energy,
three-dimensional nonstationary heat conduction equation for the k is not consistent with wind-tunnel measurements (Robins et al.,
substrate as was done in the work (Galeev et al., 2013a, b). 2001; Yang et al., 2009) where a variation of k with height is
The heat flux from the atmosphere qa is calculated using wall generally observed. Yang et al. (2009) proposed a new set of inlet
functions (Fluent, 2006). conditions where the k profile is a function of height. Gorle, et al.
The qs, qp, qar were determined from the equations given in the (2009) proposed formulations for the turbulence model co-
paper (Kawamura and Mackay, 1987). efficients to ensure stream-wise homogeneity while using the k
The change in liquid mass is calculated from the equation: profile proposed by Yang et al. (2009). In the present work the inlet
profiles were determined through repeated calculations when the
vmliq velocity and turbulence parameters profiles deduced at the outlet
¼ Jg;s : (10) section of the computational domain were set at its inlet boundary
vt
and the computation was repeated until the profiles for outlet and
The Equations (9) and (10) were incorporated into the CFD code
inlet boundaries were close. The plots in Figs. 1e3 show the com-
Fluent as user-defined scalar (UDS) transport equations without
parison between calculated profiles and profiles by Yang et al.
convection and diffusion terms.

Fig. 1. Profiles of (a) velocity U, (b) turbulence kinetic energy k, (c) turbulence kinetic energy dissipation rate ε and (d) turbulent viscosity mt from numerical calculation (solid line)
and from Yang et al. (2009) (dashed line) at the wind velocity of 1 m/s at the height of 10 m.
42 A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

Fig. 2. Profiles of (a) velocity U, (b) turbulence kinetic energy k, (c) turbulence kinetic energy dissipation rate ε and (d) turbulent viscosity mt from numerical calculation (solid line)
and from Yang et al. (2009) (dashed line) at the wind velocity of 2.5 m/s at the height of 10 m.

(2009). These profiles were obtained at wind velocities of 1, 2.5 and ground. The experimental data for cyclohexane are presented in the
5 m/s and under neutral atmospheric stability. The plots show that work (Khajehnajafi and Pourdarvish, 2011).
there is no significant difference in the velocity and ε at the wall The difference between the pool evaporation model used in the
while there is the difference in k. The similar discrepancy between present study and the one used in our previous works (Galeev et al.,
experimental and simulated profiles of k is observed in the work 2013a; Galeev et al., 2013b; Galeev and Ponikarov, 2014) is that the
(Mokhtarzadeh-Dehghan et al., 2012). At close velocity profiles, the model in the present study takes into account roughness for very
impact of using different turbulence parameters profiles on the rough area when the roughness height larger than the height of the
results of vapor dispersion calculation will be governed by centre point of the wall-adjacent cell.
the parameter mt/Sct. Due to the fact that coefficient Cm ¼ 0.028 in The two parameters of the rough wall model, i.e. the roughness
correlations of Yang et al. (2009) is much lower than the same height Ks and roughness constant Cs, were determined according to
coefficient in the used Realizable keε model (is equal to about the relationship between Ks, Cs and z0 (Blocken et al., 2007):
0.09e0.14 depending on height from wall), the turbulent viscosities
at the wall practically coincide. By optimizing the value of turbulent
Schmidt number, it is possible to reach a good agreement between KS ¼ 9:793$z0 =Cs : (11)
the predicted and measured parameters for both approaches. In the
work (Galeev et al., 2013b), the results of calculation of neutral gas In the calculation, it is assumed that the aerodynamic rough-
dispersion with Sct ¼ 0.5 and calculated profiles, agreed better with ness of flat terrain z0 is equal to pool surface roughness
experiment data than with Sct ¼ 0.7, whereas in the work (Galeev z0 ¼ 0.0002 m (Ks ¼ 0.00195 m) and Cs ¼ 1. For the very rough area
and Ponikarov, 2014) the good agreement between simulation the aerodynamic roughness is set equal to 0.04 m (Ks ¼ 0.39 m).
and experiment on heavy gas dispersion was obtained at Sct ¼ 0.7. Although in the work (Troen and Petersen, 1989) for area with
In the present work numerical simulations are performed with many trees and buildings it is recommended to use the value of
calculated profiles for mean velocity and turbulence parameters z0 ¼ 0.1 m we assumed a lower value to avoid the divergence
and with Sct ¼ 0.7. problems during iterative process. Ks should be smaller than the
To validate the CFD evaporation model, the predicted results height of the centre point of the wall-adjacent cell. The height of
were compared with the experimental data obtained by Habib et al. wall-adjacent cells was 0.1 m, therefore, in the simulation of a very
(2009). They carried out experiments in the open air for evapora- rough area, the roughness height, Ks, was set to the value of 0.05 m
tion of ethanol and cyclohexane at constant temperatures. One and from Eq. (11), the corresponding value of constant Cs was 7.83.
facility comprised a flat terrain without buildings whilst the other Fluent Manual states that constant Cs should be smaller or equal to
was situated in a very rough area. The pool consisted of a circular 1 and it is not possible to set a higher value via a user interface.
basin with a diameter of 0.74 m which was insulated from the Therefore, a higher value is defined through a User Defined
A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49 43

Fig. 3. Profiles of (a) velocity U, (b) turbulence kinetic energy k, (c) turbulence kinetic energy dissipation rate ε and (d) turbulent viscosity mt from numerical calculation (solid line)
and from Yang et al. (2009) (dashed line) at the wind velocity of 5 m/s at the height of 10 m.

Function. This approach to set roughness parameters in Fluent To validate the evaporation model, our own experimental data
was used in the gas dispersion calculation by Labovsky and for acetone evaporation were also used. The liquid was poured into
Jelemensky (2011). the pan with sizes of 0.6 m  0.4 m  0.035 m. The air velocity was
The computational domain with dimensions 100 m  500 m measured at the height of 2 m by an acoustical anemometer with an
 100 m in the x, y, z directions was used. A structured non-uniform error ± 0.2 m/s. An averaging time of the wind speed was 2 min. To
grid consisting of parallelepiped cells was generated. The grid cells measure the change in mass of the pan with the liquid, the Mettler-
were clustered in the region of the pool. The number of grid cells was Toledo XP8002S balance was used with an accuracy of 0.01 g. To
approximately 96,000. measure the temperature of the liquid the thermocouple chromel-
The second-order scheme was used for the discretization of the copel with an accuracy of 0.1  C was used.
advective terms of the governing equations. The PRESTO! algorithm The measurements were carried out at the air temperature of
(Fluent, 2006) to calculate the pressure and the SIMPLE algorithm 23  C and humidity of 40%. The initial temperature of acetone was
(Patankar, 1980) to correlate the velocity field and pressure were equal to 23.5  C. Wind speed was varied during the experiment
used. The convergence criterion for dispersed material concentra- (Fig. 4). In the calculation the average wind speed for the consid-
tion was 1  104, for other variables was 1  103. The maximum ered time period was used, which was equal to 1.05 m/s. The solar
time step of 2 s was used. flux was determined using the equation given in the paper
Accuracy of the models was estimated by the value of percent (Kawamura and Mackay, 1987), with taking into account the lati-
error (%): tude and longitude of terrain and cloud cover, and its value was
430 W/m2. The heat fluxes due to the long-wave radiation of the
.
atmosphere and the pool surface have been taken into account
d ¼ 100$ Jg;s;calc  Jg;s;expt Jg;s;expt : (12) through the use of formulae from the same source. The heat flux
from the ground is assumed to be equal to zero because the pan
Tables 1 and 2 show experimental data and prediction values for with liquid was placed on the balance. The aerodynamic roughness
evaporation of ethanol and cyclohexane at different temperatures of the area around the pool is set equal to 0.01 m (Ks ¼ 0.098 m),
and wind speeds. The computed values of the Mackay-Matsugu that corresponds to the area with short grass (Troen and Petersen,
equation are obtained with a difference d to the measured ones 1989).
between 27.5 and 61.1%, with an average value of 47.4%. The Fig. 5 shows that the model somewhat underestimates the
computed values of the numerical model are obtained with a dif- evaporated mass. The difference between the measured and
ference d to the measured ones between 0.6 and 47.4%, with an calculated evaporated mass is 17.5%. The difference between
average value of 12.7%.
44 A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

Table 1 3. Sensitivity study and discussion


Experimental data and predictions for ethanol evaporation.

Pool temperature, К Wind velocity at Mass flow, g/s This section presents the numerical analysis of the effect of pool
2 m height, m/s
Experiment Mackay e CFD model
sizes on evaporation of hexane. Hexane has high volatility (at a
Matsugu Eq. liquid temperature of 303 К the hexane vapor pressure is equal to
25 kPa) and hexane vapor is 3 times heavier than the ambient air.
Flat terrain
310.15 1.4 0.456 0.681 0.393 These factors affect pool evaporation and vapor cloud dispersion
309.65 1.9 0.533 0.855 0.523 processes. The pools with sizes of 10 m  10 m, 30 m  30 m,
325.15 1.7 1.02 1.59 1.06 60 m  60 m and 100 m  100 m are considered. The initial liquid
Very rough terrain
layer height was equal to 0.05 m. The initial temperature of ambient
305.65 1.4 0.488 0.663 0.536
310.15 1.8 0.679 0.99 0.9 air and pool substrate was 303 K. The numerical study included a
324.65 1.8 1.32 1.95 1.96 series of calculations at wind speeds 1, 2.5 and 5 m/s at the height of
10 m. The solar flux had a range of 0e900 W/m2 (Khajehnajafi and
Pourdarvish, 2011), therefore in calculation the average value
qs ¼ 450 W/m2 was used.
Table 2
Experimental data and predictions for cyclohexane evaporation. The geometry of the computational domain is shown in Fig. 6.
The calculation was performed for half of the pool because of
Pool temperature, К Wind velocity at Mass flow, g/s
symmetry of the problem. The computational domain with di-
2 m height, m/s
Experiment Mackay e CFD model mensions 2000 m  500 m  500 m in the x-, y-, z-directions was
Matsugu Eq.
used. A structured nonuniform grid consisting of parallelepiped
Flat terrain cells was generated. The number of grid cells was approximately
303 2.71 1.08 1.74 1.11 370,000. The height of wall-adjacent cells was equal to 0.2 m. In
310 3.05 1.63 2.49 1.66
317 3.49 2.33 3.63 2.55
downwind (x-axis) and crosswind (z-axis) directions, the grid cell
Very rough terrain size was 1 m over the pool surface. Additionally to the above
303 1.71 0.967 1.39 1.15 domain, a domain under the pool containing the ground of 2 m
310 1.62 1.35 1.80 1.46 depth was also included for in the calculations. The profiles of wind
317 1.4 1.67 2.13 1.68
velocity, k and ε on the inlet to the computational domain (plane
ABCD) were obtained through calculation and are shown in
Figs. 1e3. On the lateral and upper boundaries (ABFE, BCGF and
experimental and predicted dependencies of liquid temperature DCGH) of the domain the symmetry condition was specified, i.e.
is explained by the difference in evaporation rates. It should be velocity component normal to the boundary and normal de-
noted, that up to about 10 min, the experimental and calculated rivatives of the other variables assumed to be equal to zero. The
values of the mass and temperature are close, although the wind bottom boundary (ADHE) was defined as a no-slip wall. At the
velocity in calculation (1.05 m/s) was about 2.5 times greater outlet boundary (EFGH) the gauge pressure was fixed at zero. The
than the one in the experiment. This could be explained by the view of used computational grid near the pool is shown in Fig. 7.
fact that at very low wind speeds and at the small pool size the The roughness coefficient value Cs ¼ 1 was applied in this study.
molecular mechanism of vapor removal from the evaporation The roughness height Ks for the pool surface was given as
surface could dominate, when the sensitivity of evaporation to 0.00195 m and for the adjacent territory was 0.098 m. The liquid
wind velocity is not as high as in the turbulent regime. temperature was imposed as a thermal boundary condition on the
Conventionally, the Sherwood number, Sh, (the dimensionless wall under the pool while the wall around the pool was considered
parameter describing the mass transfer) is proportional to Re0.5 adiabatic.
for laminar flow over a flat surface, while Sh ~ Re0.8 in turbulent In the simulation, it was assumed that physical properties of
flow (Incropera et al., 2006). liquid, gas and solid layer did not depend on temperature. The
following values have been used: Dm,g ¼ 6.9$106 m2/sec;
mg ¼ 6.6$106 kg/(m sec); ma ¼ 1.78$105 kg/(m sec); CP,g ¼ 1680 J/
(kg К); CP,a ¼ 1006 J/(kg К); CP,liq ¼ 2300 J/(kg К); lg ¼ 0.013 W/
(m К); la ¼ 0.0242 W/(m К); DН g ¼ 370,000 J/kg; lgrd ¼ 1.28 W/
(m К); С P,grd ¼ 1130 J/(kg К); rgrd ¼ 2300 kg/m3. Here, the Dm, m, CP,
l, r, denote molecular diffusion coefficient, molecular viscosity,
specific heat, thermal conductivity and density, respectively. The a,
g and grd subscripts indicate air, evaporating gas and ground value,
respectively.
The vapor pressure Pg of hexane as a function of liquid tem-
perature Tliq was determined using the formula (Green and Perry,
2008):

   .   
E
Pg Tliq ¼ exp A þ B Tliq þ C$ln Tliq þ D$Tliq ; (13)

where A ¼ 104.65, B ¼ 6995.5, C ¼ 12.702, D ¼ 0.000012381,


E ¼ 2.
In Fig. 8 the plots of specific evaporated mass mvap and volume-
averaged liquid temperature Tliq against time are presented. The
plots show that the specific evaporated mass is higher when the
Fig. 4. Measured wind velocity. pool sizes are larger, in other conditions being equal. The influence
A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49 45

Fig. 5. Measurements (dashed line) and predictions (solid line) of liquid mass and temperature.

evaporation rate. In addition to the buoyancy effect, the increasing


of vapor concentration above the pool surface reduces the driving
force of the evaporation. The graphs of temperature change are
consistent with the time dependence of mvap. A higher evaporation
rate results in a lower pool temperature, since the heat losses in-
crease with the increasing of evaporation rate. At the wind velocity
of 1 m/s, the temperature rises throughout the considered period of
time because heat flux from the sun to pool prevails over the heat
losses due to evaporation.
The regression analysis of numerical results has shown that at
the wind speed of 1 m/s the average specific evaporated mass de-
pends on a pool size as mvap ~ L0.553 with accuracy of approxi-
mation of R2 ¼ 0.9981, at wind speed of 2.5 m/s as mvap ~ L0.318
(R2 ¼ 0.9153) and at wind speed of 5 m/s as mvap ~ L0.0578
(R2 ¼ 0.9755). Thus, at low wind velocities the stronger dependence
of evaporation rate on pool size is observed, than the one
Fig. 6. Schematic representation of the computational domain.
(mvap ~ L0.11) adopted in the Mackay-Matsugu equation commonly
used in hazard assessment (Kawamura and Mackay, 1987; Mackay
and Matsugu, 1973).
Table 3 presents the values of the physical properties of gases
and liquids for the upper and lower limits of the variation range of
liquid temperature. At these temperatures the physical properties
do not differ significantly, that confirms the validity of the
assumption of constant physical properties. The molecular diffu-
sion coefficient was determined using Gilliland's correlation
(Gilliland, 1934) and other values were taken from the ChemCAD
program database.
In Fig. 9 the time dependences of the relative turbulent viscosity
and the driving force (DYg ¼ Yg,s-Yg,P) are shown. The relative tur-
bulent viscosity is defined as the dynamic coefficient of average
turbulent viscosity above the pool surface (y ¼ 0.1 m) divided by
the turbulent viscosity in the absence of the disturbing effect of the
source. The relative turbulent viscosity for an undisturbed flow
(when there is no hexane vapor) is equal to 1.0 for each of the wind
velocities. In the initial period of evaporation, a sharp decay of
turbulent viscosity occurs due to the formation of the negative
density gradient above pool surface. The importance of buoyancy
Fig. 7. The computational grid near the pool.
effect rises as the air velocity decreases and pool sizes increase. The
slight growth of the eddy viscosity at wind speeds of 2.5 m/s and
of pool sizes on the evaporation decreases when the wind velocity 5 m/s is attributed to the reduction of liquid temperature, and, as
increases. As the pool sizes increase the vapor concentration above consequence, the reduction of mass flux of vapor from the pool
the surface of the evaporating liquid and, consequently, the vapor- surface.
air cloud density increases. The negative vertical density gradient At low wind velocities, the driving force of evaporation signifi-
appearing above pool surface causes the suppression of turbulence cantly drops as pool sizes increase due to the increasing of the
(stable stratification) that creates a large resistance to the transport concentration above the pool surface. Since in the model the
of vapor away from the evaporating pool and results in a slow driving force is both in the numerator and the denominator of the
46 A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

Fig. 8. Dependencies of specific evaporated mass (left) and average liquid temperature (right) against time at wind velocities of 1 m/s (a), 2.5 m/s (b) and 5 m/s (c).

formula for the diffusion flux, its effect on the evaporation rate
Table 3 will be determined by the factor ln((1Yg,P)/(1Yg,s)). At the wind
Physical properties of gases and liquid.
speed of 1 m/s when the pool sizes increase from 10 m  10 m to
Physical property Temperature, K Variation, % 100 m  100 m the average factor reduces by 11%, at the wind speed
286 308 of 2.5 m/s e by 6.4% and at the wind speed of 5 М/с e by 3.5%, while
average specific evaporated mass mvap reduces by 71.5%, 53% and
Dm,g, m2/sec 6.48$106 7.24E-06 11.7
mg, kg/(m sec) 6.5$106 6.7$106 3.1 12.8%, respectively. Thus, changes in the evaporation rate with
ma, kg/(m sec) 1.78$105 1.88$105 5.6 increasing of pool sizes are determined mainly by the effect of
CP,g, J/(kg К) 1590 1690 6.3 turbulence suppression over the pool surface, and not by changes in
CP,a, J/(kg К) 1005 1008 0.3 the driving force.
CP,liq, J/(kg К) 2215 2310 4.3
lg, W/(m К) 0.0115 0.0135 17.4
In the paper (Britter, 1989), a criterion that indicates when the
la, W/(m К) 0.024 0.026 8.3 plume may be considered passive, i.e. when the influence of the
DН g, J/kg 380,000 362,500 4.6 density difference on the inertia small and may be neglected, is
given. For continuous source this criterion is
A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49 47

Fig. 9. Dependencies of average relative turbulent viscosity (left) and driving force (right) at wind speeds of 1 m/s (a), 2.5 m/s (b) and 5 М/с (c).

In Table 4 the values of B и Ric for each of considered cases are


  1=3 .
B ¼ g00 $q0 L U  0:15; (14) given. In the calculation of these criteria, the average values of
the gas flow rate have been substituted in expressions (14)
and (15).
where g00 ¼ gðrg  ra Þ=ra ; q0 is the volume flow rate; L is the source
The values of Britter's criterion are less than 0.15 only when the
dimension and U is the ambient velocity.
pool sizes are 10 m  10 m and the wind speed is 5 m/s. This result
For determining the relative importance of the flow regimes
is confirmed by the graphs (see Fig. 9), which show that the change
(buoyancy-dominated, stably stratified and passive dispersion) the
of turbulence due to buoyancy effects is small at wind speed of 5 m/
Richardson number, Ric, is also used (Havens, 1992):
s and at the pool sizes of 10 m  10 m, while in the rest cases the
. appreciable suppression of turbulence (stable stratification) above
Ric ¼ g00 $H u2* ; (15) pool surface takes place. Generally, at the wind speed of 5 m/s the
Britter's criterion is less, or slightly greater than the critical value,
where H ¼ q0/UL is the characteristic cloud height and u* is the whereas at low wind speeds and extended pools it significantly
friction velocity exceeds the critical value. The values of Richardson number are also
If Ric  30 the flow is negative buoyancy dominated, if consistent with the numerical calculations. The Ric in most cases is
1Ric  30, the shear flow is stably stratified and if Ric  1 the in the range from 1 to 30 indicating the prevalence of stably
passive dispersion occurs (Havens, 1992). stratified shear flow. The Ric < 1 is obtained only for the case where
48 A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49

Table 4 yP normal distance from the pool surface to the neighboring


Values of B and Ric depending on pool sizes and wind velocity. node of the computational grid
Pool sizes Britter's criterion B (Richardson number Ric) z0 aerodynamic roughness, m
1 m/s 2.5 m/s 5 m/s
Greek symbols
10 m  10 m 0.35 (12.4) 0.20 (2.2) 0.12 (0.5)
30 m  30 m 0.42 (20.3) 0.27 (5.5) 0.17 (1.3)
ε turbulence kinetic energy dissipation rate, m2/s3
60 m  60 m 0.46 (26.7) 0.31 (8.4) 0.21 (2.6) k von Karman constant equal to 0.41
100 m  100 m 0.51 (35.9) 0.33 (10.3) 0.25 (4.1) l coefficient of thermal conductivity, W/(m K)
m coefficient of molecular viscosity, kg/(m s)
mt coefficient of turbulent viscosity, kg/(m s)
the pool sizes are 10 m  10 m and the wind speed is 5 m/s and r density, kg/m3
Ric > 30 is obtained when the pool sizes are 100 m  100 m and the
wind speed is 1 m/s. Subscripts
no index vapor-air mixture
a air
4. Conclusion g evaporating component
grd ground
The validation of the developed pool evaporation CFD model liq pool
against both literature and our own experimental data is per- P centroid of the wall-adjacent cell
formed. A good agreement is obtained between the measured and s surface of pool
the predicted evaporation rates. Based on the proposed model a
sensitivity analysis is conducted to determine the effect of pool References
sizes on evaporation of a volatile liquid (hexane). The numerical
analysis has shown that the evaporation rate falls as the pool sizes Bird, R.B., Stewart, W.E., Lightfoot, E.N., 1960. Transport Phenomena. Wiley & Sons,
New York.
increase due to the enhancement of buoyancy effects above the Blocken, B., Stathopoulos, T., Carmeliet, J., 2007. CFD simulation of the atmospheric
pool surface. The importance of the pool sizes effect decreases as boundary layer: wall function problems. Atmos. Environ. 41, 238e252.
the air velocity increases. It was found that in the dependence of Brighton, P.W.M., 1987. Evaporation from a Plane Liquid Surface into a Turbulent
Boundary Layer. UK Atomic Energy Authority, Safety and Reliability Directorate.
evaporation rate on pool size Jg,s ~ Ln the exponent on pool size SRD/HSE-report R375.
varies between 0.553 and 0.0578 for wind speeds between 1 Britter, R.E., 1989. Atmospheric dispersion of dense gases. Annu. Rev. Fluid Mech. 21,
and 5 m/s, while in risk analysis handbooks it is generally accepted 317e344.
that Jg,s is proportional to L0.11 regardless of wind speed.
Cebeci, T., Chang, K.C., 1978. Calculation of incompressible rough-wall boundary-
layer flows. AIAA J. 16, 730.
Churchill, S.W., 1976. A comprehensive correlating equation for forced convection
from flat plates. AIChE J. 22, 264e268.
Nomenclature Coulson, J.M., Richardson, J.F., 1993. Chemical Engineering (Fluid Flow, Heat Transfer
and Mass Transjk), fourth ed. Pergamon Press, Oxford.
Desoutter, G., Habchi, C., Cuenot, B., Poinsot, T., 2009. DNS and modeling of the
DB roughness function turbulent boundary layer over an evaporating liquid film. Int. J. Heat Mass
CP specific heat, J/(kg K) Transf. 52, 6028e6041.
Fluent, 2006. Fluent 6.3 User's Guide. Fluent Inc, Lebanon, New Hampshire, USA.
Cs roughness constant Galeev, A.D., Starovoytova, E.V., Ponikarov, S.I., 2013a. Numerical simulation of the
Cm coefficient in the turbulence model consequences of liquefied ammonia instantaneous release using FLUENT soft-
Dm molecular diffusion coefficient, m2/s ware. Process Saf. Environ. Prot. 91, 191e201.
Galeev, A.D., Salin, A.A., Ponikarov, S.I., 2013b. Consequence analysis of aqueous
E empirical constant equal to 9.1
ammonia spill using computational fluid dynamics. J. Loss Prev. Process Ind. 26,
DH heat of vaporization, J/kg 628e638.
Jg,s mass flux of vapor from the pool surface due to Galeev, A.D., Ponikarov, S.I., 2014. Numerical analysis of toxic cloud generation and
dispersion: a case study of the ethylene oxide spill. Process Saf. Environ. Prot.
evaporation, kg/(m2 s)
92, 702e713.
K correction factor for Stefan flow Gilliland, E.R., 1934. Diffusion coefficients in gaseous systems. Ind. Eng. Chem. 26,
Ks roughness height, m 681e685.
Kþs non-dimensional roughness height Gorle, C., van Beeck, J., Rambaud, P., van Tendeloo, G., 2009. CFD modeling of small
particle dispersion: the influence of the turbulence kinetic energy in the at-
k turbulent kinetic energy, m2/s2 mospheric boundary layer. Atmos. Environ. 43, 673e681.
L characteristic pool size, m Green, D.W., Perry, R.H., 2008. Chemical Engineer's Handbook, eighth ed. McGraw-
mliq mass of the liquid per unit surface area of the pool, kg/m2 Hill.
Habib, A., Schalau, B., Acikalin, A., Steinbach, J., 2009. Transient calculation of the
Pg(Tliq) vapor pressure of evaporating component at liquid boundary layer flow over spills. Chem. Eng. Technol. 32, 306e311.
temperature Tliq Havens, J.A., 1992. Review of dense gas dispersion field experiments. J. Loss Prev.
PC function which takes into account the resistance of the Process Ind. 5, 28e41.
Incropera, F.P., DeWitt, D.P., Bergman, T.L., Lavine, A.S., 2006. Fundamentals of Heat
diffusion sublayer to mass transfer and Mass Transfer. Wiley.
Sc and Sct molecular and turbulent Schmidt numbers, respectively Kapias, T., Griffiths, R.F., 1999. Modelling the Behaviour of Spillages of Sulfur
T absolute temperature, K Trioxide and Oleum. Health and Safety Executive. Contract Research Report 217/

1999.
non-dimensional temperature Kawamura, P.I., Mackay, D., 1987. The evaporation of volatile liquids. J. Hazard.
t time, s Mater. 15, 343e364.
U wind velocity, m/s Khajehnajafi, S., Pourdarvish, R., 2011. Correlations for mass transfer from a liquid
uþ non-dimensional velocity spill: comparisons and recommendations. Process Saf. Prog. 30, 178e184.
Labovsky, J., Jelemensky, L., 2011. Verification of CFD pollution dispersion modelling
Yg mass fraction of the evaporating component in the gas based on experimental data. J. Loss Prev. Process Ind. 24, 166e177.
phase, kg/kg Liu, Y., Olewski, T., Vechot, L., Gao, X., Mannan, S., 2011. Modelling of a cryogenic
Yþ non-dimensional mass-fraction liquid pool boiling using CFD code. In: Proc. 14th Annual Symposium, Mary Kay
O'Connor Process Safety Center “Beyond Regulatory Compliance: Making Safety
yþ non-dimensional distance Second Nature”. Texas A&M University, College Station, Texas, USA, pp. 25e27.
yþC non-dimensional diffusion sublayer thickness October 2011.
A.D. Galeev et al. / Journal of Loss Prevention in the Process Industries 38 (2015) 39e49 49

Mackay, D., Matsugu, R.S., 1973. Evaporation rates of liquid hydrocarbon spills on mass transfer coefficient. J. Air Waste Manag. Assoc. 60, 419e428.
land and water. Can. J. Chem. Eng. 51, 434e439. Rong, L., Elhadidi, B., Khalifa, H.E., Nielsen, P.V., Zhang, G.Q., 2011. Validation of CFD
Mokhtarzadeh-Dehghan, M.R., Akcayoglu, A., Robins, A.G., 2012. Numerical study and simulation for ammonia emissions from an aqueous solution. Comput. Electron.
comparison with experiment of dispersion of a heavier-than-air gas in a simu- Agric. 75, 261e271.
lated neutral atmospheric boundary layer. J. Wind Eng. Ind. Aerodyn. 110, 10e24. Saha, C.K., Wu, W., Zhang, G., Bjerg, B., 2011. Assessing effect of wind tunnel sizes on
Nielsen, F., Olsen, E., Fredenslund, A., 1995. Prediction of isothermal evaporation air velocity and concentration boundary layers and on ammonia emission
rates of pure volatile organic compounds in occupational environments d A estimation using computational fluid dynamics (CFD). Comput. Electron. Agric.
theoretical approach based on laminar boundary layer theory. Ann. Occup. Hyg. 78, 49e60.
39, 497e511. Schlichting, H., 1968. Boundary-layer Theory, sixth ed. McGraw-Hill, New York.
Parente, A., Gorle, C., van Beeck, J., Benocci, C., 2011. Improved keε model and wall Shih, T.H., Liou, W.W., Shabbir, A., Zhu, J., 1995. A new keε eddy-viscosity model for
function formulation for the RANS simulation of ABL flows. J. Wind Eng. Ind. high Reynolds number turbulent flows e model development and validation.
Aerodyn. 99, 267e278. Comput. Fluids 24, 227e238.
Pasquill, F., 1943. Evaporation from a plane, free liquid surface into a turbulent air Troen, I., Petersen, E.L., 1989. European Wind Atlas. Risø National Laboratory, Ros-
stream. Proc. R. Soc. Lond. A 182, 75e95. kilde, Denmark.
Patankar, S.V., 1980. Numerical Heat Transfer and Fluid Flow. Hemisphere, Wash- van den Bosch, C.J.H., 1997. Pool evaporation. In: van den Bosch, C.J.H.,
ington, DC. Weterings, R.A.P.M. (Eds.), CPR 14E, Methods for the Calculation of Physical
Raimundo, A.M., Gaspar, A.R., Oliveira, A.V.M., Quintela, D.A., 2014. Wind tunnel Effects Due to Releases of Hazardous Materials (Liquids and Gases), Yellow
measurements and numerical simulations of water evaporation in forced Book. Committee for the Prevention of Disasters, The Hague, 3.1e3.128.
convection airflow. Int. J. Therm. Sci. 86, 28e40. chot, L., Olewski, T., Osorio, C., Basha, O., Liu, Y., Mannan, M.S., 2013. Laboratory
Ve
Richards, P., Hoxey, R., 1993. Appropriate boundary conditions for computational scale analysis of the influence of different heat transfer mechanisms on liquid
wind engineering models using the kee turbulence model. J. Wind Eng. Ind. nitrogen vaporization rate. J. Loss Prev. Process Ind. 26, 398e409.
Aerodyn. 46e47, 145e153. Vik, T., Pettersson Reif, B.A., 2011. Implementation of a New and Improved Evapo-
Robins, A., Castro, I., Hayden, P., Steggel, N., Contini, D., Hesit, D., 2001. .A wind ration Model in Fluent. Norwegian Defence Research Establishment (FFI). FFI-
tunnel study of dense gas dispersion in a neutral boundary layer over a rough rapport 2011/00116.
surface. Atmos. Environ. 35, 2243e2252. Yang, Y., Gu, M., Chen, S., Jin, X., 2009. New inflow boundary conditions for
Rong, L., Nielsen, P.V., Zhang, G.Q., 2010. Experimental and numerical study on modeling the neutral equilibrium atmospheric boundary layer in computa-
effects of airflow and aqueous ammonium solution temperature on ammonia tional wind engineering. J. Wind Eng. Ind. Aerodyn. 97, 88e95.

You might also like