2017 experimental and numerical study of lng pool spreading and vaporization on water

Journal of Hazardous Materials 334 (2017) 244–255
Contents lists available at ScienceDirect
Journal of Hazardous Materials
journal homepage: www.elsevier.com/locate/jhazmat
Research paper
Experimental and numerical study of liquefied natural gas (LNG) pool
spreading and vaporization on water
Nirupama Gopalaswamia
, Konstantinos Kakosimosb
, Bin Zhanga
, Yi Liua
, R. Mentzera
,
M. Sam Mannana,∗
a
Mary Kay O’Connor Process Safety Center, Artie McFerrin Department of Chemical Engineering, Texas A&M University System, College Station, TX,
77843-3122, USA
b
Mary Kay O’Connor Process Safety Center - Qatar, Texas A&M University at Qatar, PO Box 23874, Education City, Doha, Qatar
h i g h l i g h t s
• Experimental determination of pool spreading and vaporization parameters for LNG flowing on a narrow channel of water.
• CFD methodology developed for LNG pool spreading and vaporization.
• Both pool spreading and vaporization is influenced by wind.
a r t i c l e i n f o
Article history:
Received 12 May 2016
Received in revised form 3 March 2017
Accepted 7 April 2017
Available online 8 April 2017
Keywords:
CFD
Liquefied natural gas
LNG
Pool spreading
Vaporization
a b s t r a c t
The investigation of pool spreading and vaporization phenomenon is an essential part of consequence
analysis to determine the severity of LNG spills on water. In this study, release of LNG on water during
marine operations is studied through experimental and numerical methods The study involves emulation
of an LNG leak from transfer arms during side by side loading operations. The experimental part involves
flow of LNG in a narrow trench filled with water and subsequent measurement of pool spreading and
vaporization parameters. The numerical part involves CFD simulation using a three dimensional hybrid
homogenous Eulerian multiphase solver to model the pool spreading and vaporization phenomenon. In
this method, LNG is modeled as dispersed phase droplets which can interact with continuous phases
− water and air through interphase models. The numerical study also employs a novel user-defined
routine for capturing the LNG vaporization process. The CFD solver was capable of capturing the salient
features of LNG pool spreading and vaporization phenomena. It was observed from experiment and CFD
simulation that wind influenced both pool spreading and vaporization phenomenon through entrainment
and convection.
© 2017 Elsevier B.V. All rights reserved.
1. Introduction
The development of large natural gas fields in remote locations
is currently being exploited through advancement in technology. A
marked increase in LNG utilization is forecast with an increase in
LNG marine operations like transportation, bunkering and loading.
A release of LNG on water during these marine operations can occur
due to several factors like collisions, grounding, loss of mechan-
ical integrity of flanged connections in loading arms, inadvertent
disconnection of hoses, brittle fracture of materials lining the load-
ing arms, overfilling or over-pressurizing of fuel tanks, undesirable
∗ Corresponding author.
E-mail address: mannan@tamu.edu (M.S. Mannan).
ship maneuvering due to human factor issues and extreme weather
scenarios like hurricanes [1]. Upon release, a spreading liquid can
form a pool with rapid vaporization, leading to the formation of
a flammable vapor cloud. One of the key steps in safety analyses
involves the determination of maximum extent of the pool and the
rate at which flammable vapor is being produced.
Given the relevance of this research, a number of researchers
have already studied LNG releases on water by both experimen-
tal and numerical means [2,3]. The pool spreading parameters like
pool area, height and spreading rate were measured dynamically by
recording the distance covered by the pool in small scale apparatus
[4]. In small scale experiments, LNG was impeded by ice formation,
which led to under-prediction of pool spreading parameters. Con-
trarily, it was also observed that these parameters were difficult
to measure in medium and large scale experiments due to vapor
http://dx.doi.org/10.1016/j.jhazmat.2017.04.025
0304-3894/© 2017 Elsevier B.V. All rights reserved.

N. Gopalaswami et al. / Journal of Hazardous Materials 334 (2017) 244–255 245
Nomenclature
A Interfacial area density (1/m)
Cp Specific heat capacity (J/kg K)
D Droplet diameter (m)
E Internal energy (J/kg)
f Pool spreading or vaporization parameter
g Acceleration due to gravity (m/s2)
h Heat transfer coefficient (W/(m2K))
Hv Latent heat of vaporization (J/kg)
k Thermal conductivity (W/(mK))
L Characteristic length of the pool (m)
n Thermocouple tolerance value (K)
m Mass flux between phases (kg/(m2s)
Nu Nusselt number
p Pressure (Pa)
Pr Prandtl number
Re Reynolds number
r Volume fraction
q Heat flux (W/m2)
˙Q Heat source (W/m3)
t Time (s)
T Temperature (K)
U Velocity (m/s)
x,y,z Cartesian coordinates
Greek Letters
Viscosity (kg/ms)
Density (kg/m3)
Ä Turbulent kinetic energy (m2/s2)
ε Eddy dissipation rate (m2/s3)
Surface tension (N/m)
Stress tensor
Subscripts
air Air
d Data retrieval time
e Experimentally determined value
LNG LNG liquid phase
lam Laminar regime
n Tolerance of thermocouple
r Response time
v LNG vapor
water Water
blocking [5]. Similar to pool spreading parameters, the vaporization
rate in LNG experiments was determined through three different
methods. The first method involved measurement of mass loss due
to vaporization and subsequent determination of vaporization rate
using the slope of mass loss data [6]. The second method involved
measurement of water temperature and application of empirical
correlations on heat transfer mechanisms like convective boiling
or conduction to obtain the vaporization rate [7]. The third method
involved measurement of vapor flow through an array of gas sen-
sors and integration of measured concentration data across the area
covered by gas sensors [8].
Over the years, efforts were also directed to develop numerical
models to simulate these physics. The approaches can be classified
as phenomenological, integral, shallow layer and Computational
Fluid Dynamics (CFD) models. The phenomenological models pro-
vide hypothesized relationship for pool spreading and vaporization
based on the forces and heat sources acting on the pool. (e.g.
[9,10]). Complex physics like hydraulic jump and flashing of LNG
droplets could not be modeled using phenomenological models.
The integral model calculates the dynamic behavior of pool spread-
ing and vaporization through mass and energy balances (e.g. PHAST
[11]). The integral models vary based on the assumptions made for
spreading rate (uniform pool thickness), heat transfer models (con-
duction, convection), release types (continuous vs instantaneous)
and the numerical scheme used for solving the equations (second
order, first order discretization). The integral models have been
validated against many LNG experiments [12]. Scenarios involving
obstacles like process equipment or transport systems cannot be
modeled using integral models. The Shallow layer model assumes
that the transport properties of the fluid vary along the lateral direc-
tions, and is constant along elevation (e.g. [13]). The shallow layer
model fails when there is significant front resistance due to partial
submergence of LNG on water, as water is modeled as a wall in these
models rather than a fluid. Inclusion of turbulence due to bubbly
vaporization also poses significant difficulty as three dimensional
solvers are required for turbulence models. A recent improvement
in consequence modeling is the use of CFD solvers. Although, CFD
simulations are computationally expensive and analytically chal-
lenging, they can replicate complex multiphase physics in obstacle
laden environments.
It is the aim of this research to address these gaps in experi-
mental and numerical methods for determining the pool spreading
and vaporization parameters of LNG released on water. Particular
attention has been paid to specific release scenario like a LNG leak
between two ships during side by side loading operation for which
data is currently lacking. An experimental study was conducted by
allowing LNG to flow through a narrow channel of water. Addi-
tionally, a numerical study was performed to simulate the pool
spreading and vaporization behavior. Finally, the CFD results are
validated against the experimental results.
2. Materials and methods
2.1. Experimental methodology
The experiment was performed in an L-shaped trench present
in Brayton Fire Training Field (BFTF), College Station. Each leg (leg
1 and leg 2) of the trench is 8.2 m long, 1.22 m wide and 1.05 m
deep (see Fig. 1). Prior to the experiment, the trench was filled
with water up to a height of 0.8m. A weather station was installed
near the trench to obtain the local conditions such as wind speed,
wind direction, atmospheric pressure, temperature, and relative
humidity. The trench was equipped with N-type thermocouples in
a linear arrangement to measure the water temperature and LNG
temperature during the spill. The tip of the thermocouples were
above water for measuring LNG temperature and below water for
measuring water temperature. Table 1 provides the locations of
thermocouples with respect to origin. The bottom left corner of
leg 1 was considered as origin. The experiment was recorded with
three video cameras and one Infra-Red (IR) camera. Cameras 1 and
2 were placed to track the pool spreading of LNG in leg 1 and leg
2 respectively and camera 3 provided an aerial view of the experi-
ment. An ultrasonic level transmitter (Echopod DL-24) was placed
in leg 2 to track the dynamic height of LNG during the vaporiza-
tion process. The thermocouples’ and level sensors’ output were
recorded every second (1 Hz) by a Data Acquisition system (DAQ).
A summary of the release conditions and atmospheric conditions
present during the experiment is provided in Table 2.
LNG was continuously discharged from a tanker of 11000 gal-
lon capacity through a cryogenic hose (internal diameter-0.076m).
The discharge point is present at the surface of Leg 1 above ori-
gin (see Fig. 1). The LNG height was monitored in the tanker and
this value was used to determine the volume lost in tanker with
respect to time and subsequently the flow rate of LNG. As LNG

246 N. Gopalaswami et al. / Journal of Hazardous Materials 334 (2017) 244–255
Fig. 1. Experimental setup.
Table 1
Thermocouple Locations.
Origin (0,0,0)
Leg 1 Leg 2
Width from Origin-0.89 m Width from Origin-0.91m
TC No. Length of TC from Origin (m) Height of TC from Origin (m) TC No. Length of TC from Origin (m) Height of TC from Origin (m)
TC-1 0.89 0.9 TC-14 9.73 0.88
TC-2 1.19 0.76 TC-15 10.03 0.75
TC-3 1.50 0.9 TC-16 10.34 0.84
TC-4 2.09 0.76 TC-17 10.64 0.76
TC-5 2.72 0.9 TC-18 10.97 0.83
TC-6 3.35 0.76 TC-19 11.26 0.76
TC-7 3.61 0.91 TC-20 11.55 0.86
TC-8 4.04 0.79 TC-21 11.86 0.76
TC-9 4.34 0.89 TC-22 12.47 0.85
TC-10 4.66 0.76 TC-23 12.77 0.76
TC-11 5.23 0.93 TC-24 13.08 0.85
TC-12 5.84 0.78 TC-25 13.38 0.76
TC-13 7.06 0.98 TC-26 13.69 0.85
TC-27 13.98 0.76
TC-28 14.3 0.85
TC-29 14.6 0.78
TC-30 14.91 0.85
TC-31 15.21 0.76
TC-32 15.52 0.88
was released into the trench, the contact area between LNG and
water increases steadily as it spreads on water. This area covered
by LNG was determined by recording the time at which the ther-
mocouples record LNG boiling temperature. As the locations of the
thermocouples were known prior to the experiment, the area was
then determined by multiplying the length and the width from
the origin. It is assumed that LNG covers the entire width of the
trench when it touches the thermocouples. The pool area during
regression was obtained experimentally by recording the time at
which water temperature returns to atmospheric temperature dur-
ing the experiment. The pool spreading and regression of LNG on
water during experiment is shown in Fig. 2. The spreading rate is
determined in a similar manner by calculating the time taken by
LNG to reach selected thermocouples present in the experiment.
The spreading rate is then determined as distance by time. This
method gives an estimate of spreading rate when LNG touches the
thermocouples for the ﬁrst time. It was also observed from previ-
ous experiments that pool height strongly affected the vaporization
rate of LNG [14]. Hence pool height was measured using three dif-
ferent methods. In the ﬁrst method, the pool height was measured
manually using a dip stick at regular intervals in the intersection
of leg 1 and leg 2 (Fig. 3(a)). The second method involved a lin-
ear array of thermocouples in the intersection of leg 1 and leg 2
to determine the pool height (see Fig. 3 (b)). As LNG level started
to increase, the thermocouples array measured the LNG temper-
ature. Since the positions of thermocouples are known prior to
experiment, pool height is calculated by recording the thermocou-
ples which registered LNG temperature. As the pool reached the
end of leg 2, the pool started to regress. The pool height was cap-
tured dynamically using ultrasonic level sensor (Fig. 3(c)). In this

Fig. 2. Snapshots of LNG pool (a) traversing leg 1 of trench (b) entering the leg 2 of trench (c) covering 1/8th of leg 2 (d) covering 1/4th of leg 2 (e) covering ½ of leg 2 (f)
covering ¾ th of leg 2 (g) entire trench (f) regressing due to vaporization.
Fig. 3. Pool height measurement through (a) dipstick (b) thermocouples (c) ultrasonic level sensor.
Table 2
Summary of release conditions and atmospheric conditions.
Release Conditions
LNG flow rate 9.5 ± 0.08 kg/s
Spill duration 1200 s
Water temperature 20 ◦
C
Spill area 17.8 m2
LNG Composition 100% Methane
Height of discharge 0.1 m
Diameter of discharge pipe 0.076 m
Atmospheric Conditions
Wind velocity 6.2 ± 1.3 m/s
Wind direction SE Global
Atmospheric temperature 25.6 ± 0.03 ◦
C
Relative humidity, 52.6%
Dew point temperature 15.1 ◦
C
Stability class C
Solar radiation 4.34 kW/m2
/day
Sensible heat flux 18.06 kW/m2
Ground Roughness length 3 mm [28]
Water surface roughness 1 mm [28]
experiment, both mass loss measurement using weighing balance
and gas detectors were not employed and hence, the vaporization
mass flux was determined using temperature values. The temper-
ature of water and LNG was measured and empirical correlations
(Eqs. (19)–(25)) on convective heat transfer and boiling was applied
with water temperature to obtain the vaporization mass flux. This
experiment will be referred to as “MKOPSC Trench experiment” in
the rest of the manuscript.
2.2. Uncertainty analysis
The pool spreading and vaporization parameters were primarily
obtained from thermocouples. The uncertainty associated with the
thermocouples can be due to several factors like the type of thermo-
couple, response time and data retrieval interval. The uncertainty
can be quantified by quantifying the error in these parameters. The
tolerance value of N-type thermocouples is ±1.7 K (iconel alloy −
class 3 type- working range −270 to 1300 ◦C). The response time
for this type of thermocouple is 0.25s( tr). The thermocouples are
connected to Data Acquisition System which can record data at the
speed of 1 s ( td).
Additionally, the uncertainty propagates when the tempera-
tures values are used to determine pool spreading and vaporization
parameters. The uncertainty in any experimentally derived vari-
able (fe)(pool spreading/vaporization parameter) is then a function
of tolerance value, response time, data retrieval time and temper-
ature. The error propagation is determined using Kline McClintock
Method [15], expressed as
fe
fe
=
∂fe
∂n
( n)
2
+
∂fe
∂tr
( tr )
2
+
∂fe
∂td
( td)
2
+
∂fe
∂ T
( T)
2
(1)
The propagated error is expressed as error bars in experimen-
tally determined parameters (Figs. 8, 10–12).

Table 3
Grid sensitivity analysis.
Parameters Grid 1 Grid 2 Grid 3
Smallest element size 0.075 m 0.05 m 0.05 m
Largest element size 0.15 m 0.1 m 0.075 m
Size difference between smallest and largest element 0.075 m 0.05 m 0.025 m
Total Nodes 176154 573256 1346905
Total Elements 969218 3245728 7706835
Total CPU Time 68080 s 262200 s 691200 s
Water temperature at 600 s 290.8 K 291.4 K 291.4 K
2.3. Numerical methodology
The computational domain consists of the L-shaped trench and
a rectangular region to simulate the atmosphere above the trench
(see Fig. 4).
2.3.1. Grid sensitivity analysis
The computational domain was divided into tetrahedral cells. A
grid sensitivity analysis was performed by considering three differ-
ent grids (see Table 3). All the other simulation parameters were
kept identical to ensure that the results were changing as a result of
the grid characteristics alone. Fig. 5 shows the water temperature
with respect to the distance from the LNG discharge. The water
temperature is influenced by the discharge location and reduced
more near the discharge when compared to regions remote from
the discharge location. As can be seen in Fig. 5, the results with
grids 2 and 3 are almost identical, whereas grid 1 is different in
magnitude and trend. A grid-independent solution was achieved
with grid 2 consisting of 3.2 × 106 elements. The element size in
this grid was restricted to 0.05 m in the trench region and 0.1 m
in the atmospheric region to allow for the optimum resolution of
interface between water-LNG and LNG-air phases.
2.3.2. Governing equations
To model the LNG flow on water the homogeneous Eule-
rian multiphase model was adopted. This is a limiting case of
Eulerian–Eulerian multiphase flow where all fluids share the same
velocity, pressure, temperature and turbulence fields. The trans-
port properties are solved in terms of phasic quantities. Alternate
approach involves the multifluid Volume of Fluid (VOF) model [16],
but it faces scale-up issues. The release of LNG on water involves
a fragmented jet where small droplets vaporize and large droplets
aggregate to form a pool [17]. To address this phenomenon, three
components were defined- LNG, water and air. LNG was modeled
as dispersed phase droplet with a mean droplet size of 0.017m. Air
and water was defined as continuous phases. Currently there is lack
of information regarding LNG droplet sizes. The droplet formation
is part of discharge modeling which requires detailed study and is
out of scope of this paper. The mean diameter was then based on
Fig. 4. Computational domain.
the size range of bubbles (0.01-0.018m) observed in pure methane
[18] and light LNG [19] during the vaporization process. Sensitivity
analysis was performed on the droplet sizes 0.01 m and 0.02m. Any
value within this range provides similar results. Values lower than
0.01 m increase the computational time, whereas values greater
than 0.02 m are too large to be discretized with the current grid
configurations. LNG was modeled as 100% methane.
2.3.2.1. Interphase model.
The three interfaces LNG-air, LNG-water and air-water are modeled
using interphase models. The Particle model is selected for interfa-
cial transfer between LNG-air and LNG-water. These two interfaces
are constantly evolving with respect to time. The free surface model
is selected to model the water-air interface. The water–air inter-
phase is distinct and effects of water movement like ripples and
waves are captured well using the free surface model. The interfa-
cial area density (surface area per unit volume) is determined based
on the assumption that LNG is present as spherical particle of mean
diameter. The interfacial area density ‘A’ of each interphase is given
by
Particle Model : ALNG|Air = 6
rLNG
DLNG
(2)
Particle Model : AWater|LNG = 6
rLNG
DLNG
(3)
Free Surface Model : AWater|Air =
2|∇rwater||∇rair|
|∇rwater| + |∇rair|
(4)
2.3.2.2. Mass transport.
The mass transfer of LNG between different phases is governed by
the continuity equation [20] expressed as
∂
∂t
+ ∇. ( U) = mA (5)
The transport properties in terms of phasic fraction is expressed
as
= rLNG LNG + rair air + rwater water (6)
mA = mLNG|airALNG|air + mLNG|waterALNG|water + mair|waterAair|water (7)
mLNG|Water = mAir|Water = 0 as there is no mass transfer from LNG to
water or air to water phase. However, there is transfer from LNG to
air phase through vaporization. Eq. (5) now becomes,
∂
∂t
+ ∇. ( U) = mLNG|airALNG|air (8)
The mass transfer from LNG to air phase mLNG|air is modeled using
CFX expression language using Eq. (25). Here, the mass conserva-
tion equation for the mixture is obtained by summing the mass of
three components alone (LNG, water and air).The non-zero source
term in RHS of Eq. (8) denotes the production of LNG vapor. The
production of LNG vapor is assumed to be rapid and it escapes the
computational domain immediately upon production. Hence LNG
vapor is not included in the mixture mass fraction nor as a separate
phase.

Fig. 5. Grid sensitivity analysis showing water temperature with respect to distance from discharge at t-600s.
The volume fraction is solved using the volume conservation
equation. The volume conservation equation is the obtained by
dividing the mass conservation equation by phasic density and
summing for all phases on both side of the equation,
3
r=1
1 ∂
∂t
+ ∇. ( U) =
3
r=1
1
mLNG|airALNG|air (9)
The volume fraction of water and air were specified as initial con-
ditions. The volume fraction of LNG and change in volume fraction
of all water and air are solved using Eq. (9).
2.3.2.3. Momentum transport.
The momentum transfer between the phases is expressed as −
∂ ( U)
∂t
+ ∇ ( UU) = −∇p + ∇. + g (10)
= rLNG LNG + rair air + rwater water (11)
p = rLNGpLNG + rairpair + rwaterpwater (12)
For Homogeneous model, velocity fields are shared between the
components
ULNG = Uair = Uwater = U (13)
The interfacial terms cancel out resulting in a single transport
equation with variable density.
2.3.2.4. Turbulence model.
The initial growth of vapor cloud is characterized by a large magni-
tude of turbulence due to vaporization. This turbulence was caused
due to convection currents called ‘thermals’ which rise from the
water-LNG interface and break in the LNG-air interface [14]. The
turbulence was modeled using the standard k-␧ turbulence model
[21] in Ansys CFX with default values for constants. The values of
turbulence kinetic energy and eddy dissipation rate for cryogenic
liquid boiling on water were obtained from a high speed flow visu-
Fig. 6. Comparison of flashing phenomenon from (a) normal camera (b) IR camera (c) CFD simulation.

Fig. 7. Water temperature profiles.
alization experiment [14]. The results provided values of turbulent
kinetic energy (0.005m2/s2) and eddy dissipation rate (0.004 m2/s3)
which were provided as inputs to the turbulence models.
2.3.2.5. Energy transport.
The energy transport is given by
∂
∂t
( E) + ∇. ( UE) = ∇. (k∇T) + : ∇U + ˙Q (14)
E = rLNGELNG + rairEair + rwaterEwater (15)
k∇T = kLNG∇TLNG + kair∇Tair + kwater∇Twater (16)
˙Q = ALNG|airqLNG|air + Awater|airqwater|air + ALNG|waterqLNG|water (17)
The heat transfer between water to air is negligible when com-
pared to heat transferred between LNG and water and LNG and air
phases. Hence,
qwater|air∼0 (18)
When heat is being transferred from water to LNG, the water
temperature declines with time whereas the LNG temperature
stays at its boiling point. From the previous experiments, it was
observed that LNG stays in film boiling when released on water
[22]. Due to this reason, the Berenson model for film boiling was
selected for determining the heat transfer coefficient [23]. Accord-
ing to Berenson’s model [23], the heat flux is a combination of
temperature difference and heat transfer coefficient expressed as
qLNG|water = h. T = 0.425 (Twater − TLNG) .
⎡
⎢
⎣
k3
V
HV V ( LNG − V )
V T
g( LNG− V )
LNG
1
2
⎤
⎥
⎦
1
4
(19)
Similarly the heat transfer from air to LNG is convective in
nature, provided by
qLNG|air = hair. (Tair − TLNG) (20)
The heat transfer coefficient, hair is given by
hair =
Nukair
L
(21)
Where the Nusselt number, Nu is given by
Nu = 0.037Re0.8
Pr0.33
(22)
Reynolds number, Re is given by
Re =
Lvair air
air
(23)
And Prandtl number, Pr is given by
Pr =
Cpair air
kair
(24)
The mass vaporization flux of LNG to air phase (mLNG|air) is then
obtained from energy balance as
mLNG|air =
˙Q
HV
(25)
Eqs. (19)–(25) were previously validated with experimental
data involving continuous releases of cryogenic liquid [22].
2.3.3. Simulation parameters
The walls of the trench were provided with an adiabatic no slip
boundary condition. The LNG pipe of diameter 0.076 m was pro-
vided as inlet by specifying the bulk mass flow rate. The head space
above the trench was open to atmosphere where LNG and air could
flow into and out of the domain. A wind velocity was provided at the
wind inlet to simulate the wind effects on the pool (see Fig. 4(a)).
The pool was found to be influenced only by the wind above the
pool. Hence a uniform wind velocity was used in this study. The ini-
tial height and volume fraction of water and air was defined using
the step function in CFX.

The simulation was solved in Linux HPC cluster with 16 parallel
processors and 16 CPU cores. The criterion of convergence of Navier
Stokes equation was set to 1E−4. The time-stepping was adaptive
providing an automatic control of time-step to achieve convergence
in transient run. A list of salient inputs of the simulation is given in
Table 4. The total computational time was 80 h. This CFD methodol-
ogy will be referred to as ‘LSPREAD’ simulation (LNG pool spreading
and vaporization), in the rest of the manuscript.
3. Results
The results from LSPREAD simulation were validated against the
experimental results. Experimental observations are provided in
Sections 3.1 and 3.2. Validation of key parameters like pool area,
spreading rate, pool height and vaporization mass flux are demon-
strated in Sections 3.3–3.6.
3.1. Flashing of LNG
When LNG is released from a pipe, it flashes initially as LNG gets
heated by the walls of the pipe before it contacts the atmosphere.
Fig. 6 compares the vapor emanating from the LNG pipe during the
initial stages of the spill using (a) normal camera, (b) IR camera and
(c) CFD simulation. LNG was found to be flashing for duration of
30 s producing vapor and LNG droplets. The concentration of vapor
during flashing was measured previously in field experiments and
was typically low of the order of 2–3% v/v [24]. The visible boundary
of vapor was also observed from IR camera and the temperature of
vapor was approaching the atmospheric temperature during flash-
ing. While the LNG vapor cloud has a distinct white appearance due
to condensation of water droplets in air, the LNG vapor cloud during
flashing was relatively translucent providing a distinct variation in
the visible appearance between flashing and vapor cloud forma-
tion. In MKOPSC trench experiment and LSPREAD simulation, the
height of the methane vapor extended up to the fence surrounding
Table 4
Salient inputs of LSPREAD methodology.
Boundary Conditions Specification
Trench wall No-slip
LNG mass flow inlet 9.5 kg/s at 111 K
Wind Inlet 6.5 m/s at 293 K
Atmosphere Opening at T=293 K and P=1 atm
Operating Conditions Specification
Reference values Reference pressure −
1 atm
Reference density −
1.225 kg/m3
Acceleration due to
gravity- 9.8 m/s2
Fluids LNG, water and air
Models Specification
Turbulence Model k-␧ [21]
Interphase Model LNG-Air and
LNG-Water- Particle
model
Air-Water - Free
surface model
Heat Transfer Model User-Defined Routine
Mass Transfer Model User-Defined Routine
Solution Methods Specification
Time-Stepping Adaptive
Minimum timestep -
0.001 s
Maximum timestep -
1 s
Coefficient loops - 10
Advection Second-order scheme
Time derivative Second-order
backwards Euler
Volume Fraction Coupled solver
Fig. 8. Comparison of pool area between MKOPSC Trench experiment and LSPREAD simulation.

Fig. 9. Pool spreading at different time intervals from LSPREAD simulation.
the trench (∼1m). With further continuation of LNG spill, a dense
vapor cloud was formed with significant increase in vapor cloud
volume. The restriction of the spill to the boundaries of the trench
was well captured by the LSPREAD simulation.
3.2. Temperature profiles
Fig. 7. shows water temperature profiles of leg 1 and leg 2
thermocouples respectively. The thermocouples that were placed
closer to the water surface underwent a large temperature gradi-
ent. An overall reduction in bulk water temperature of 4–14 ◦C was
observed in the experiment. This is an accordance with a similar
study where liquid nitrogen was found to reduce the temperature
of water to about 5–10 ◦C during a continuous spill [25]. Similar
reduction of temperature up to 8 ◦C was obtained in the Falcon
LNG tests [26]. It was also observed that the water temperature in
leg 1 reduced more than leg 2. This is due higher contact time of
LNG with water in leg 1 when compared to leg 2. Moreover, higher
spreading rate in leg 1 contributed to steep temperature reduction
in leg 1. Higher spreading rate in leg 1 lead to increased contact area
Fig. 10. Comparison of spreading rate between MKOPSC Trench experiment and LSPREAD simulation. Contour from LSPREAD simulation shows the velocity near the discharge
area.

Fig. 11. Comparison of pool height between MKOPSC Trench experiment and LSPREAD simulation. Contour from LSPREAD simulation shows pool height in leg 1.
and heat transfer. The temperature profiles from experiment were
also compared to temperature profiles from CFD (Fig. 7). As the
LNG spread on water, significant amount of heat transfer occurred
at the water surface. The water temperature from CFD was found
to match well up to duration of 600 s in leg 1. After 600 s, an under-
prediction was obtained by CFD. This is due to formation of ice layer,
which hinders the temperature measurement of water below ice.
The water temperature of leg 2 from CFD was similar in trend com-
pared to experimental value with a small difference. This implied
that the heat transfer was higher in CFD when compared to exper-
iment.
3.3. Pool area
A comparison of pool area between MKOPSC trench experiment
and LSPREAD simulation is provided in Fig. 8. The spreading of LNG
pool on water with subsequent water temperature reduction from
CFD is also expressed as contours in Fig. 9(a)–(d). The contours are
obtained from LSPREAD simulation. As the pool spreads, the pool
covers the first leg of the trench (t∼130s). The CFD results predict
the pool area very well up to this duration. As the pool spreads, the
pool area reaches a maximum value when the LNG pool contacts the
end of leg 2 of the trench (t∼400s). The CFD results over-predict the
pool area up to this duration. A maximum pool area is maintained
for a few minutes where addition of LNG into the pool increases the
height, rather than the area. At this stage, the mass inflow to the
pool is equalized by the LNG vaporized. It was also observed that
due to the high wind speed blowing in the opposite direction of
pool spreading in leg 2, rapid vaporization was found to take place
instead of pool spreading. As a result of this, the pool was vaporizing
steadily in experiment and CFD during the interval 400–700s. Once
the pool touches the end of leg 2, it starts regressing. During the
time duration of 900–1200s, the pool was spreading and regress-
ing constantly. The time at which regression and pool spreading
occurred in the experiment during the time period 900–1200 s was
not matched well with the CFD. As a result of this, a difference is
observed between experiment and CFD results. With further vapor-
ization, the pool area starts reducing, from the end of the second leg
of trench. The relative error between LSPREAD and MKOPSC trench
experiment was determined to be 21% for pool area.
3.4. Spreading rate
A comparison of spreading rate from experiment and simula-
tion is provided in Fig. 10. The spreading rate increases initially
up to a value of 0.06 m/s till 130 s after which, the velocity starts
to reduce to a minimum value of 0.02 m/s. However, in CFD, the
maximum spreading rate in leg 1 was around 0.08 m/s. During the
experiment, wind entrainment was from all direction. However,
wind entrainment was restricted to pre-dominant wind direction
in CFD. After 130 s the pool reaches the boundary of leg 1 of the
trench, where it takes a turn to leg 2 causing a reduction in speed.

Fig. 12. Comparison of vaporization mass flux between LNG experiments and LSPREAD simulation.
The wall present at end of leg 1 of the trench exerts a drag force
on the moving pool that reduces the momentum of the pool. As
a result of this, a marked difference in spreading rate is observed
between leg 1 and leg 2 of the trench. Fig. 10 also shows the veloc-
ity contour of LNG from LSPREAD simulation near the discharge
area. The momentum of LNG pool near the discharge is very high.
As LNG moves away from the source, the momentum of the LNG
pool reduces and becomes stable. The behavior of spreading rate
between experiment and LSPREAD simulation was similar, how-
ever, LSPREAD was found to over-predict the velocity in leg 1. The
relative error between LSPREAD and MKOPSC Trench experiment
was determined to be 22% for spreading rate.
3.5. Pool height
Fig. 11 provides a comparison of pool height between different
experimental measurements and LSPREAD. A difference in various
measurements of pool height is observed due to the different loca-
tion in which it was measured. From Fig. 11, we can also observe
that the height of the pool tends to increase only after duration of
400s. This is the time taken by the pool to reach the end of leg 2
of the trench. However, the pool height starts to increase earlier
in CFD due to higher LNG spreading rate in CFD. The pool height
predicted by simulation was representative of the values recorded
by manual measurement and array of thermocouples. The ultra-
sonic level sensor showed the dynamic increase and decrease in
pool height due to pool spreading and vaporization. The average
relative error between MKOPSC Trench experiment and LSPREAD
was found to be 25% for pool height.
3.6. Vaporization
Fig. 12 provides a comparison of vaporization mass flux between
LNG experiments and LSPREAD. The average vaporization mass flux
determined from experiment is found to be around 0.2 kg/(m2s)
in the experiment. The vaporization mass flux includes heat input
from wind and water. The wind had a significant effect on vapor-
ization in two different ways. Wind provides heat to LNG in the
form of convection. Additionally, wind also provides advection in
the form of entrainment. The advection process makes the air above
LNG unsaturated, which leads to an increase in vaporization. The
advection process dominates over the convection when high wind
speeds are observed. However, it was difficult to measure the level
of entrainment in both experiment and CFD. The high initial vapor-
ization mass flux was due to large temperature difference that
is present during the initial stages of the continuous spill. Once
the pool covered the entire water surface, the vaporization mass
flux was found to remain constant at 0.195 kg/(m2s) after a time
period of 500s. The shift from 0.2 to 0.195 kg/(m2s) shows a minor
change in vaporization mass flux, after 500s. During the time period
500–800s, the pool area was maintained constant. The vaporiza-
tion mass flux further reduces to 0.19 kg/(m2s) once the pool starts
regressing. The overall relative error between LSPREAD and exper-
iment was determined to be 4% for vaporization mass flux. The
values of the current tests are compared with similar LNG experi-
ments. The vaporization rates obtained in this test were comparable
to values obtained in Esso Test [8] and Bureau of Mines [7], but
over-predicted when compared to Maplin Sands [27] and Shell Test
[6].
4. Conclusion
In this study, a comprehensive experimental and numerical
study of LNG pool spreading and vaporization was performed. Mea-
surements of pool spreading and vaporization parameters were
made using novel methods. A transient three-dimensional homo-
geneous multiphase model called LSPREAD was developed in CFX
for LNG pool spreading and vaporization on water. A user defined
routine for vaporization was implemented in LSPREAD with adap-
tive time-stepping technique. While modeling, it was noted that

there is a lack of information regarding LNG droplet sizes that
are formed during the initial stages of release. Further research
is required on LNG discharge modeling to identify the range of
droplet sizes based on the LNG storage conditions. The results from
CFD were validated with an experiment to simulate the release of
LNG on water during side-by side loading operations. Photographic
images reveal distinct difference between flashing and vapor cloud
formation. Validation results reveal that LSPREAD methodology is
capable of emulating the typical behavior of LNG movement in
narrow pathways. However, differences were observed between
LSPREAD and experimental values due to the ways of determina-
tion of source term parameters. It was also observed that wind
influenced both pool spreading and vaporization phenomenon.
The wind blowing in the opposite direction of the pool tends to
impede the spreading phenomenon significantly. The wind also
affects the vaporization by providing additional heat and unsat-
uration through entrainment.
Acknowledgements
The authors like to acknowledge financial support from Mary
Kay O’Connor Process Safety Center and National Grid for this
experiment. The authors would like to acknowledge the help
extended by Mr. Kirk Richardson (BFTF) and his team of firefighters
for the LNG experiments.
References
[1] LNG Bunkering: Technical and Operational Advisory, ABS, Houston, 2014.
[2] A. Luketa-Hanlin, A review of large-scale LNG spills: experiments and
modeling, J. Hazard. Mater. 132 (May (2–3)) (2006) 119–140.
[3] D.M. Webber, S.F. Jagger, S.E. Gant, M.J. Ivings, LNG Source Term Models for
Hazard Analysis: a Review of the State-of-the-Art and an Approach to Model
Assessment, Derbyshire, UK, 2009.
[4] H.R. Chang, Boiling and Spreading of liquid nitrogen and liquid methane on
water, Int. Commun. Heat Mass Transf. 10 (1983) 253–263.
[5] G.F. Feldbauer, J.J. Heigl, W. McQueen, R.H. Whipp, W.G. May, Spills of LNG on
Water–vaporization and Downwind Drift of Combustible Mixtures, API
Report EE61E-72, 1972.
[6] G. Boyle, A. Kneebone, Laboratory investigations into the characteristics of
LNG spills on water. Evaporation, spreading and vapor dispersion, Chester,
Report Number 6Z32, 1973.
[7] D. Burgess, J. Murphy, M. Zabetakis, Hazards of LNG Spillage in Marine
Transportation, U.S. Department of Interior, Bureau of Mines, Pittsburgh,
Pennsylvania, 1970, pp. S-4105.
[8] W. May, W. McQueen, R. Whipp, Spills of LNG on water, Operating Section
Proceedings (1973), p. D-140-D-150.
[9] D.M. Webber, On models of spreading pools, J. Loss Prev. Process Ind. (May)
(2012) 3–6.
[10] F. Briscoe, P. Shaw, Spread and evaporation of liquid, Prog. Energy Combust.
Sci. 6 (1) (1980) 127–140.
[11] DNV (Det Norske Veritas), PHAST Consequence Modelling Software Version
6.7, DNV Corporate Headquarters, Hovik, Norway, 2012.
[12] H.W.M. Witlox, M. Harper, R. Pitblado, Validation of PHAST dispersion model
as required for USA LNG siting applications, Chem. Eng. Trans. 31 (2013)
49–54.
[13] D.M. Webber, M.J. Ivings, Modelling bund overtopping using shallow water
theory, J. Loss Prev. Process Ind. 23 (no. 5) (2010) 662–667.
[14] N. Gopalaswami, D.M. Laboureur, R.A. Mentzer, M.S. Mannan, Quantification
of turbulence in cryogenic liquid using high speed flow visualization, J. Loss
Prev. Process Ind. 42 (2016) 70–81.
[15] S.J. Kline, F.A. McClintock, Describing uncertainties in single-sample
experiments, Mech. Eng. 75 (1953) 3–8.
[16] R. Marcer, B. Yerly, L. Pomié, B. Lequime, M. Rivot, E. De Carvalho, F. Baillou,
CFD simulation of LNG spillage, 6th European Conference on Computational
Fluid Dynamics(ECFD VI) (2014) (p. 12).
[17] T.A. Cavanaugh, J.H. Siegell, K.W. Steinberg, Simulation of vapor emissions
from liquid spills, J. Hazard. Mater. 38 (1) (1994) 41–63.
[18] E.M. Drake, A.a. Jeje, R.C. Reid, Transient boiling of liquefied cryogens on a
water surface I. Nitrogen, Methane and Ethane, Int. J. Heat Mass Transf. 18
(December (12)) (1975) 1361–1368.
[19] E.M. Drake, A.A. Jeje, R.C. Reid, Transient boiling of liquefied cryogens on a
water surface. II light hydrocarbon mixtures, Int. J. Heat Mass Transf. 18 (12)
(1975) 1369–1375.
[20] ANSYS Inc, Ansys CFX: Release 15, Ansys Inc, Canonsburg, PA, 2016.
[21] B.E. Launder, D. Spalding, The numerical computation of turbulent flows,
Comput. Methods Appl. Mech. Eng. 3 (2) (1974) 169–289.
[22] N. Gopalaswami, R.A. Mentzer, S. Mannan, Investigation of pool spreading and
vaporization behavior in medium-scale LNG tests, J. Loss Prev. Process Ind. 35
(2015) 267–276.
[23] P.J. Berenson, Film-boiling heat transfer from a horizontal surface, J. Heat
Transfer 83 (3) (1961) 351–356.
[24] R. Qi, D. Ng, B.R. Cormier, M.S. Mannan, Numerical simulations of LNG vapor
dispersion in Brayton Fire Training Field tests with ANSYS CFX, J. Hazard.
Mater. 183 (November (1–3)) (2010) 51–61.
[25] N. Gopalaswami, T. Olewski, L.N. Véchot, M.S. Mannan, Small-scale
experimental study of vaporization flux of liquid nitrogen released on water,
J. Hazard. Mater. 297 (2015) 8–16.
[26] S.T. Chan, Numerical simulations of LNG vapor dispersion from a fenced
storage area, J. Hazard. Mater. 30 (2) (1992) 195–224.
[27] J.S. Puttock, D.R. Blackmore, G.W. Colenbrander, Field experiments on dense
gas dispersion, J. Hazard. Mater. 6 (1982) 13–41.
[28] J. Sienfield, S. Pandis, Atmospheric Chemistry and Physics From Air Pollution
to Climate Change, 2nd ed., John Wiley & Sons, Inc., New Jersey, 2006.

2017 experimental and numerical study of lng pool spreading and vaporization on water

More Related Content

What's hot (19)

Similar to 2017 experimental and numerical study of lng pool spreading and vaporization on water (20)

Recently uploaded (20)

2017 experimental and numerical study of lng pool spreading and vaporization on water