This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
 Author's personal copy

Journal of Loss Prevention in the Process Industries 26 (2013) 453e461

Contents lists available at SciVerse ScienceDirect

Journal of Loss Prevention in the Process Industries

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

Two-phase jet releases, droplet dispersion and rainout I. Overview

and model validation
Henk W.M. Witlox*, Mike Harper
DNV Software, Palace House, 3 Cathedral Street, London SE1 9DE, UK

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

Article history: Many accidents involve two-phase releases of hazardous chemicals into the atmosphere. This paper
Received 28 February 2012 describes the results of the fourth phase of a Joint Industry Project (JIP) on liquid jets and two-phase
Received in revised form droplet dispersion. The objective of Phase IV of the JIP was to generate experimental rainout data for
2 June 2012
non-flashing experiments, and to develop recommendations for the best methodology to predict rainout
Accepted 4 July 2012
[total rainout mass and its spatial distribution (‘distributed’ rainout)].
Phase IV of the JIP first included rainout experiments by the UK Health and Safety Laboratory (HSL) for
Keywords:
sub-cooled releases of water and xylene with a range of orifice sizes and stagnation pressures. See the
Liquid jets
Droplet size
companion paper II for further details. Secondly model validation was carried out by DNV Software for
Droplet dispersion these experiments using different correlations for the initial droplet size (Sauter Mean Diameter, SMD),
Rainout i.e. the CCPS SMD correlation and the Phase III JIP SMD correlation. The validation includes flow rates,
Experiments droplet size, distributed rainout and cloud temperature drop. Subsequently validation was considered for
Model validation a wider range of experiments from the literature (sub-cooled and superheated releases) for both SMD
and total rainout. Adopted rainout methods comprised both methods including explicit modelling of the
droplets (using an extended version of Phast dispersion model UDM), as well as more simple methods
based on rainout correlations without droplet modelling. Recommendations are made for the most
accurate droplet size and rainout modelling. A modified CCPS UDM droplet size correlation has been
shown to agree best against experimental rainout data.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction rather than rainout at a single downwind distance. Fig. 1 postulates

Many accidents involve two-phase releases of hazardous
chemicals into the atmosphere. Rainout results in reduced
concentrations in the remaining cloud, but can also lead to
extended cloud duration because of re-evaporation of the rained-
out liquid. For accurate hazard assessment one must accurately
predict both the amount of rainout and re-evaporation of the pool.
This paper describes the results of a fourth phase of a Joint
Industry Project (JIP) on liquid jets and two-phase droplet disper-
sion. The aim of the project is to increase the understanding of the
behaviour of sub-cooled (non-flashing) and superheated (flashing)
liquid jets, and to improve the prediction of release rate, atomiza-
tion (initial droplet size after expansion to ambient pressure and
before air entrainment), droplet dispersion and rainout. Fig. 1
illustrates the consecutive phases in discharge and dispersion. It
also shows that droplets of different initial size may rainout at
different downwind distances resulting in ‘distributed’ rainout
* Corresponding author. Tel.: þ44(0)207 716 6711.
E-mail address: henk.witlox@dnv.com (H.W.M. Witlox).
larger droplets rainout earlier due to vertical momentum effects.
However, as discussed later on in the paper, experimental data
demonstrate the opposite e that larger droplets rainout later.
Phase I of the JIP (Witlox & Bowen, 2002) comprised a literature
review on flashing liquid jets and two-phase dispersion. Phase II
(Cleary, Bowen, & Witlox, 2007; Witlox, Harper, Bowen, & Cleary,
2007) and Phase III (Kay, Bowen, & Witlox, 2010; Witlox, Harper,
Oke, Kay, & Bowen, 2010) included experimental work limited to
measurements of the flow rate and the initial droplet size distri-
bution. The work included:

Scaled experiments by Cardiff University for a wide range of
volatilities (water, cyclohexane, butane, propane and gasoline)
with measurements of flow rate and initial droplet size across
the full relevant range of superheats. This also included deri-
vation of new correlations for initial droplet size distribution
and Sauter Mean Diameter (SMD).

Large-scale butane experiments by INERIS (France) to ensure
that for more realistic scenarios the derived droplet size
correlations are accurate.

0950-4230/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jlp.2012.07.005
 Author's personal copy

454 H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461

vapour-plume centre-line
(flashing) two-phase discharge
from pipe or vessel

SMD droplet trajectory

Distributed rainout; spreading evaporating liquid pool

Fig. 1. Discharge, droplet dispersion, distributed rainout and re-evaporation.

Validation of droplet correlations. New correlations for droplet
size distribution and Sauter Mean Diameter (SMD) were
implemented into the Phast discharge model. Along with
a range of other correlations from the literature, this was
validated against JIP experiments and other published datasets.
It was shown that the new Phase III JIP droplet size correlation
agrees better against experimental data than the existing Phast
6.54 correlation.

Distributed rainout modelling. The Phast dispersion model
(UDM) was also extended to allow simultaneous modelling of
a range of droplet sizes. These droplets of different sizes travel
along different trajectories and thus predict longitudinal
distribution of rainout (rather than at a single point).
The current Phase IV of the project is a follow-up of Phase III. Its
objective is to generate experimental data for non-flashing rainout
experiments to validate the new methodology for distributed
rainout, and to make model refinements where deemed to be
necessary.
Phase IV of the JIP first included indoor water and xylene rainout
experiments carried out by the Health and Safety Laboratory (HSL).
These experiments involved horizontal sub-cooled jet releases with
a range of nozzle diameters and source pressures. Measurements
were carried out of release rate, initial droplet size distribution,
distributed rainout, concentrations and temperatures. The reader is
referred to a separate companion paper (Part II; Bettis, Jagger,
Witlox, & Harper, 2013) for full details of these results.
Secondly, Phase IV of the JIP included model validation and
model refinements by DNV Software. This validation was carried
out using extended versions of the discharge and dispersion
models in the Phast hazard assessment package. This validation
included validation against the HSL experiments. It also includes
validation for a wider range of experiments for both SMD and total
rainout. A range of initial droplet size correlations and rainout
methods was applied, and recommendations for the best method of
modelling are formulated. Section 2 of the current paper summa-
rizes the results of this validation. Section 3 summarizes the main
conclusions and includes recommendations for further work.
2. Model validation
2.1. Introduction
The model validation and model refinements were carried out
by DNV Software. This included validation of the Phast discharge
model DISC [flow rate; post-expansion droplet size (Sauter Mean
Diameter, SMD)] and the Phast dispersion model UDM [distributed
rainout, temperatures and concentrations].
Prior to the start of the HSL experiments, pre-runs were carried
out corresponding to sub-cooled water, octane and xylene releases
for a range of orifice sizes and stagnation pressures. This included
model verification for the discharge results (flow rate and SMD)
and an explanation of the differences in model predictions between
the different releases. This analytical model verification was carried
out successfully without errors identified in the Phast software.
Thus prior to the validation, confidence was obtained that Phast
carried out the correct discharge and dispersion (rainout)
calculations.
Subsequently validation against the above HSL water and xylene
experiments was carried out. In this chapter we will report vali-
dation against HSL measurements for flow rate (Section 2.2), initial
droplet size (Section 2.3), rainout (Section 2.4), and temperature

a 600 b 600

500 500 5 mm
Flowrate (g/s)

5 mm
Flow rate (g/s)

400 400
Modelled Modelled (average)
Experimental Experimental (average)
300 300

200 200
2.5 mm 2.5 mm

100 100
2 mm

0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 18
Pressure (barg) Pressure (barg)

Fig. 2. Validation of flow rate for HSL water and xylene experiments. (a) Water (b) xylene.
 Author's personal copy

H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461 455

100

90

80

Cumulative volume (%) 70 1000 diameters

500 diameters
60 JIP Phase III

50

40

30

20

10

0
0 1000 2000 3000 4000 5000 6000
Droplet diameter (microns)

Fig. 3. Validation of droplet size distribution (water experiment, 2.5 mm nozzle, 10 barg).

(Section 2.5). The validation for droplet size and rainout (Section
2.3 and 2.4) will also be summarized for a more comprehensive set
of experiments available in the literature. The validation against the
measured concentrations for the HSL experiments has been
omitted in the current paper because of perceived insufficient
accuracy of the measured concentrations; see the Part II companion
paper for further details.
xylene experiments. It also demonstrates the increase of flow rate
with increasing pressure and increasing nozzle diameter.
2.3. Droplet size
The validation was carried out using the following four different
correlations for the initial droplet size (SMD):

2.2. Flow rate
In the discharge calculations the liquid was modelled as an
incompressible liquid for both the water and xylene experiments.
In accordance with the experiments, the discharge calculations
presume discharge from a rounded sharp-edge orifice. Fig. 2 shows
that very close agreement of the flow rate observed in the HSL
experiments was obtained with the highest accuracy for the
2.5 mm water experiments and the lowest accuracy for the 5 mm

The original CCPS correlation (Johnson & Woodward, 1999),
where the SMD (m) is taken as the minimum of the value dda
derived from a mechanical break-up criterion (based on critical
Weber number) and the value ddf derived from a flashing
break-up criterion (reducing with increasing partial expansion
energy).

A modified CCPS correlation with the mechanical break-up
value dda applied for sub-cooled jets and the flashing break-
up value ddf for superheated releases.

100

90

80

70
Cumulative volume (%)

60

50

40 4 barg (expt)
8 barg (expt)
12 barg (expt)
30
16 barg (expt)
4 barg (model)
20
8 barg (model)
12 barg (model)
10 16 barg (model)

0
0 1000 2000 3000 4000 5000 6000
Droplet diameter (microns)

Fig. 4. Validation of droplet size distribution (xylene experiment, 2.5 mm nozzle).

 Author's personal copy

456 H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461

a 10000

1000
Predicted SMD [microns]

JIP Phase IV - Xylene

JIP Phase IV - Water

Cardiff_Cyclo-Hexane

Cardiff_Water

100 Cardiff_Gasoline

+30%-deviation

-30%-deviation

10
10 100 1000 10000
Measured SMD [microns]

b
10000
Predicted SMD [microns]

1000
JIP Phase IV - Xylene

JIP Phase IV - Water

Cardiff_Cyclo-Hexane

Cardiff_Water
100
Cardiff_Gasoline

+30%-deviation

-30%-deviation

10
10 100 1000 10000
c Measured SMD [microns]

10000

1000
Predicted SMD [microns]

JIP Phase IV - Xylene

JIP Phase IV - Water

Cardiff_Cyclo-Hexane

100 Cardiff_Water

Cardiff_Gasoline

+30%-deviation

-30%-deviation

10
10 100 1000 10000
Measured SMD [microns]

Fig. 5. Validation of SMD droplet size correlation against sub-cooled experiments. (a) Phase III JIP SMD correlation, (b) original CCPS correlation (Phast 6.6 default), (c) modified
CCPS correlation (Phast 6.7 default).

The Melhem correlation based on a Weber number criterion by mechanical breakup, an intermediate transition regime, and
generalized for superheated liquids a final regime governed by fully flashing break-up. This corre-

The Phase III JIP correlation defined the SMD as a reducing tri- lation also includes a correlation for the droplet size distribu-
linear function of superheat with the initial regime governed tion based on the RosineRammler droplet size distribution.
 Author's personal copy

H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461 457

a Phase III JIP SMD correlation

10000

Predicted SMD [microns]

1000

100

10

1
10 100 1000
Measured SMD [microns]

b Original CCPS correlation (Phast 6.6 default)

10000
Predicted SMD [microns]

1000

100

10

1
10 100 1000
Measured SMD [microns]

c Modified CCPS correlation (Phast 6.7 default

10000
Predicted SMD [microns]

1000

100

10

1
10 100 1000

Measured SMD [microns]

Fig. 6. Validation of SMD droplet size correlation against super-heated experiments. (a) Phase III JIP SMD correlation, (b) original CCPS correlation (Phast 6.6 default), (c) modified
CCPS correlation (Phast 6.7 default).
 Author's personal copy
458 H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461
See Witlox et al. (2010) for full details of the above correlations.
Fig. 3 includes validation against the HSL water experiment
(2.5 mm nozzle, 10 barg) for which images are shown in the
companion paper (Bettis et al., 2013). It is seen that the Phase III JIP
initial droplet size distribution correlation closely matches the
droplet size distribution observed at 500 nozzle diameters down-
stream of the release point. It is also apparent that further down-
stream (at 1000 nozzle diameters) droplets are smaller because of
secondary break-up of non-spherical droplets. The same qualitative
behaviour can be seen in the images shown in the companion
paper. The above can be explained since the Phase III JIP correlation
was determined by a best fit against droplet-size measurements
close to the release point, i.e. approximately at the end of the jet
break-up length after initial droplet-size atomisation and before
secondary breakup. The predicted SMD for this experiment equals
1736 mm, while the measured SMD data are 1245 mm (500 diam-
eters downstream) and 673 mm (1000 diameters downstream).
Fig. 4 includes validation against the xylene 2.5 mm experiment,
with droplet size measurements taken at 800 nozzle diameters
(2 m) downstream. The model predictions clearly show the
reduction of the droplet size with increasing pressure. However
there is an unclear trend for measured droplet size against pres-
sure, especially for larger volume fractions. This could be caused by
insufficient number of measured droplets causing inaccuracies in
the droplet size distribution for large droplet sizes (particularly for
the case of 4 barg). See the companion paper for a further detailed
discussion. The predicted SMD values for this experiment for 4, 8,
12, 16 barg equal 1455, 938, 725 and 605 mm respectively, while the
measured SMD values are 707, 744, 805 and 522 mm.
Figs. 5 and 6 include validation results for initial droplet size
(SMD) for a comprehensive set of sub-cooled and superheated
experiments respectively, i.e. the Phase IV JIP HSL experiments
(water and xylene), Phase III JIP Cardiff and INERIS experiments
(cyclo-hexane, water, gasoline, butane and propane) as well as other
experiments available from the literature: the STEP propane
(Hervieu & Veneau, 1996), the VKI R134-A (Yildiz, Rambaud, & van
Beeck, 2004), the HSL propane (Allen, 1998), Ecole de Mines water
and INERIS butane (Touil, Bigot, Bonnet, Lacome, & Duplantier, 2004,
pp. 3201e3209) experiments. See Witlox et al. (2007, 2010) for
further details on input data (hole size, stagnation pressure, stag-
nation temperature, etc.) for these experiments. The original CCPS
correlation generally under-predicts since it takes the minimum of
the mechanical and flashing droplet size correlation regardless of
the degree of superheat, while the modified CCPS correlation shows
improved results for super-heated releases. The Phase III JIP corre-
lation was based on a best fit of the Phase III experimental data for
water and cyclohexane, and it is shown by Figs. 5 and 6 to perform
best against the overall set of experimental data.
2.4. Rainout
The UDM outdoor dispersion model in Phast allows for two-
phase dispersion including droplet modelling, rainout of the
droplets to form a pool, pool evaporation and subsequent addition
of vapour back to the cloud; see Fig. 1. The JIP III correlation predicts
a distribution of droplet sizes (see Fig. 3). The UDM model was
extended to divide the distribution into a number of equal mass
‘parcels’ and for each parcel a representative droplet size was
assigned. Each parcel therefore follows its own trajectory in the
cloud (with large parcels falling more rapidly due to reduced drag
effects) and rainout is distributed in the downwind direction; see
Witlox et al. (2010) for full details of the model.
In addition to the above UDM rainout method (including explicit
modelling of the droplets), also more simple rainout methods were
applied based on rainout correlations without droplet modelling.
100
90
80
Cumulative rainout (%)
70
60
50
40
30 Experimental
JIP III + parcels
20
CCPS
10 Melhem
0
0 2 4 6 8 10
Downstream distance (m)
Fig. 7. Validation of rainout model against HSL xylene experiment (2.5 mm nozzle,
8.0 barg).
This included rainout correlations by Kletz (1977), De Vaull and
King (1992) and Lautkaski (2008).
Fig. 7 depicts the validation of the UDM results against
a selected xylene experiment for a range of droplet size correla-
tions. A single droplet size (SMD) with rainout at a single point was
modelled for the CCPS (there was no difference in this case between
modified and original CCPS) and Melhem correlations. For the
Phase III JIP droplet size correlation the initial droplet size distri-
bution was modelled by twenty droplet parcels as described above.
It is seen that the Phase III JIP correlation predicts too much rainout
over too narrow a rainout zone. The other two correlations predict
all rainout at a single point, and is seen that the CCPS correlation
most accurately predicts the total amount of rainout.
The above conclusions were found to be valid for the overall set
of HSL experiments. Table 1 includes measured and predicted
rained out mass (percentage mass fractions of released material)
for the entire set of HSL experiments. The yellow cells indicate
those experiments where a significant amount of rainout was not
captured. Thus it is seen from the table that overall the CCPS
correlation most accurately predicts the rainout. As expected, the
rainout is seen to reduce with increasing pressure for the 2.5 mm
xylene experiments which were all carried out at the same
temperature (11  C). In addition to possibly inaccuracy of the
prescribed experimental conditions and limitations of the rainout
measurements, the rainout for the 5 mm 8.5 barg 2  C xylene
experiment is expected to be higher than the 5 mm 7.7 barg 11  C
xylene experiment because of the significantly colder temperature.
Overall the Phase III correlation significantly underestimates rain-
out distance, as shown in Fig. 7.
Validation was also carried out for other two-phase elevated
releases from the existing UDM validation dataset. This included
the EEC propane, FLADIS ammonia, and Goldfish HF experiments
for which no rainout was reported. It also included the Desert
Tortoise ammonia experiments for which a maximum of 20e40%
was identified as missing mass, which may potentially have
rained out. Table 2 illustrates again that if most of the missing mass
indeed rained out the modified CCPS correlation produces
improved predictions compared to the original CCPS correlation.
In Table 2 exit pressures are taken from the SMEDIS project
(FLADIS, EEC and Desert Tortoise), and from the MDA database of
Hanna, Strimaitis, and Chang (1991) for Goldfish. For Desert
Tortoise experiments rainout (assumed equal to missing mass) is
given by Ichard, Hansen, and Melheim (2010); for DT4 it was also
given by Goldwire (1986).
 Author's personal copy

H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461 459

Table 1
Measured and predicted rainout for HSL experiments.

Experiment e input data (release height ¼ 1 m) % Rainout % Rainout predicted (UDM rainout methods) % Rainout predicted
measured (simple correlations)

Chemical Nozzle Stagnation Temperature CCPS original/ JIPIII þ parcels Melhem Kletz Lautkaski De Vaull
(mm) pressure ( C) modified
(barg)
Water 2.5 5.0 7 98.4 98.5/99.5 99.9 e 100 60 81.7
9.5 7 96.3 98.1/98.7 99.8 e 100 60 81.7
5 4.8 7 99.0 98.7/99.5 99.9 e 100 60 81.7
Xylene 2.5 4.2 11 96.8 96.7 99.5 99.3 100 60 29.2
8.0 11 91.4 90.8 98.9 98.0 100 60 29.2
10.3 11 90.4 86.7 98.6 97.1 100 60 29.2
15.8 11 82.1 75.1 97.7 94.0 100 60 29.2

5 4.0 11 94.3 97.2 99.7 99.3 100 60 29.2

7.7 11 89.9 93.1 99.5 98.3 100 60 29.2

8.5 2 95.3 95.2 99.7 98.8 100 60 25.6

13.1 5 85.2 88.8 99.4 97.2 100 60 26.7

Table 1 and Table 2 illustrate that the UDM rainout methods
produce overall significantly more accurate results than the simple
rainout correlations, with the Kletz and Lautkaski correlations
producing worse results than the De Vaull correlation. This is likely
to be caused by the fact that the UDM rainout methods include
explicit modelling of the droplet dispersion and therefore include
more relevant governing physics.
The extended Phast discharge and dispersion models (DISC,
UDM) have also been validated against a subset of the CCPS
rainout experiments (two tests randomly chosen for each of
the five chemicals (water, CFC-11, chlorine, cyclohexane and
monomethylamine)). See Johnson and Woodward (1999) for
full details on the CCPS and MMA Rohm and Haas rainout
experiments.
The solid lines in Fig. 8a include the measured (uncorrected)
rainout results as well as measured rainout ‘corrected’ to account
for evaporation. The rainout data are given as function of increasing
superheat. It is seen that the experimental data show a trend to
decrease with superheat; the experimental data points for water
may need to be ignored because of possible rainout of water from
ambient air with high humidity.
In Fig. 8a UDM results are provided for the JIP III correlation
both without parcels (model SMD droplet only) and with
droplet parcels, as well as for the CCPS correlation (original and
modified). Fig. 8a demonstrates the over-prediction of the
rainout at low superheats by the Phase III JIP correlation
(because of too large initial droplet size), and the under-
prediction of rainout by the original CCPS correlation (because
it erroneously picks up the initial mechanical breakup droplet
size instead of the flashing break-up size). The best predictions
are provided by the modified CCPS correlation, but it must be
noted that it is based on the flashing CCPS droplet-size criterion
derived from a best fit against the corrected rainout for the CCPS
experiments.
Fig. 8b includes results of the simple rainout correlations against
the CCPS experiments. Since the De Vaull and King correlation
appears to have been fitted against the uncorrected rainout data, it
matches most closely these data. Likewise the Lautkaski correlation
has been fitted against the corrected rainout data, and therefore it
matches most closely these data.
Ramsdale and Tickle (2000) indicate it may be more appropriate
to estimate rainout using a simple rainout correlation rather than
a more sophisticated method including droplet modelling.
However it has been shown in the current section that the UDM
rainout method produces significantly more reliable results for
a wider range of scenarios. The simple correlations particularly
perform poorly if they are used to predict scenarios against which
they have not been fitted.

Table 2
Model validation of rainout for large-scale 2-phase field experiments. [Release heights: 1.5 m for FLADIS, 0.5 m for EEC, 0.79 m for Desert Tortoise, 1.263 m for Goldfish; hole
diameter: 6.1e9.5 cm Desert Tortoise, 2.4e4.2 cm Goldfish].

Experiment Chemical Exit Super- heat ( C) % Rainout % Rainout predicted (UDM) % Rainout predicted (correlations)
pressure reportede
Original Modified CCPS JIP III JIPIII þ parcels Kletz Lautkaski De Vaull
(bar) estimated
CCPS
Fladis 9 Ammonia 6.93 47.0 None 0 0 0 4 69.4 32.4 0.0
Fladis16 Ammonia 7.98 50.4 None 0 0 0 4 67.1 30.4 0.0
Fladis 24 Ammonia 5.7 42.9 None 0 0 0 4 72.1 34.9 2.0
EEC 360 Propane 7.70 55.9 None 0 0 0 3 37.4 3.6 0.0
EEC 550 Propane 10.13 55.4 None 0 0 0 4 38.0 4.2 0.0
EEC 560 Propane 10.23 42.4 None 0 0 0 3 53.3 18.0 0.0
DT 1 Ammonia 10.12 57.1 <20 0 27 26 25 62.9 26.6 0.0
DT 2 Ammonia 11.16 55.7 <36 0 41 39 28 63.8 27.4 0.0
DT 3 Ammonia 11.374 57.8 <39 0 42 41 27 62.4 26.2 0.0
DT 4 Ammonia 11.798 59.8 <30 0 38 38 27 61.1 25.0 0.0
Goldfish 1 HF 6.89 20.4 None 0 0 0 e 71.0 33.9 0.0
Goldfish 2 HF 7.45 18.4 None 0 0 0 e 73.9 36.5 9.3
Goldfish 3 HF 7.58 19.4 None 0 0 0 e 72.4 35.2 4.7
 Author's personal copy

460 H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461

exactly with the plume centre-line and it was also visually

a UDM rainout methods (explicit droplet modelling)
confirmed by the experiments to correspond with the plume
100 centre-line. Fig. 9 indicates for the 2.5 mm and 5 mm HSL xylene
90 rainout experiments the vertical measurement heights above the
80
ground at which the temperature is measured, the height at which
the maximum temperature drop is measured, and the height at
70
Rainout (%)

which the maximum is predicted. It is seen that the predicted data

60 agree quite well with the measured data.
50 The CCPS correlation was shown to agree most closely to the
experimental data for the prediction of the temperature drop (more
40
close results for larger pressures); see Fig. 10. Smaller droplets
30
result in less rainout, more droplet evaporation and hence more
20 cooling in the cloud. This explains the significantly greater cooling
10 predicted by the CCPS correlation compared to the JIP Phase III
correlation. The adopted mechanical break-up criterion in both
0
0 10 20 30 40 50 60 70 80 90 correlations predicts the droplet size to decrease with pressure, and
Superheat (K) this explains the predicted increase of cooling with increasing
pressure. The trend for the measured data could possibly be caused
b Simple correlations (no droplet modelling) by inaccuracies in the measurements.
100

90
3. Discussion, conclusions and future work
80

70 From the above validation (for all methods and all experiments),
Rainout (%)

it is concluded that the modified CCPS droplet size correlation

60
provides the overall best prediction of rainout (within a few
50 percent accuracy for the Phase IV rainout experiments), while the
40 Phase III JIP droplet size correlation (typically resulting in larger
droplet sizes and more rainout) provides the best overall prediction
30
of initial droplet size. As a result, the modified CCPS droplet size
20 correlation has been selected as the new default method in the
10
latest version 6.7 of Phast and Phast Risk, while the Phase III JIP
correlation has been included as an additional non-default option.
0
0 10 20 30 40 50 60 70 80 90
The UDM rainout methods were overall shown to be superior to the
simpler rainout methods.
Superheat (K)
It is apparent that despite the best agreement against measured
Fig. 8. Model validation against CCPS experiments. (a) UDM rainout methods (explicit initial droplet sizes in xylene experiments, the Phase III JIP
droplet modelling), (b) simple correlations (no droplet modelling). correlation under-predicts droplet evaporation and rainout
distances. The current UDM model presumes that droplets of all
sizes move with the same horizontal speed as the vapour plume,
2.5. Temperature with the larger droplets raining out more early and with the
smaller droplets staying more close to the cloud vapour centre-line
Model predictions were compared against temperature because of vertical drag. However it was observed in the experi-
measurements for the HSL xylene experiments. The location of the ments that the larger droplets retain their momentum longer
maximum temperature drop corresponds for the model predictions (because of reduced drag to inertia ratio in the horizontal

a 160 b 160
Measured
140 140
CCPS
JIP Phase III Measured
120 120
Measurement points CCPS
Height (cm)

JIP Phase III

Height (cm)

100 100
Measurement points

80 80

60 60

40 40

20 20

0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12
Pressure (barg) Pressure (barg)

2.5 mm xylene 5 mm xylene

Fig. 9. Validation of height of maximum temperature drop (6 m downstream). (a) 2.5 mm xylene (b) 5 mm xylene.
 Author's personal copy

H.W.M. Witlox, M. Harper / Journal of Loss Prevention in the Process Industries 26 (2013) 453e461 461

a 0 b 0

-0.5 -0.5 Measured

CCPS
Temperature change (K)

JIP Phase III

Temperature change (K)

Measured
-1 CCPS -1
JIP Phase III
-1.5 -1.5

-2 -2

-2.5 -2.5

-3 -3

-3.5 -3.5
2 4 6 8 10 12 14 16 2 3 4 5 6 7 8 9 10 11 12
Pressure (barg) Pressure (barg)

2.5 mm xylene 5 mm xylene

Fig. 10. Validation of maximum temperature drop (6 m downstream). (a) 2.5 mm xylene (b) 5 mm xylene.

direction) and therefore move faster in the horizontal direction References

than the smaller droplets. As a consequence they rainout at
a distance further downwind. Thus this may explain the current Allen, J. T. (1998). Laser-based measurements in two-phase flashing propane jets.
Part two: droplet size distribution. Journal of Loss Prevention in the Process
model under-prediction of rainout distance. In addition, the Industries, 11, 299e306.
underestimation in evaporation may be due to the absence from Bettis, R., Jagger, S., Witlox, H. W. M., & Harper, M. (2013). Two-phase jet releases,
the model of secondary break-up effects of large unstable droplets. droplet dispersion and rainout, II. Rainout experiments. Journal of Loss
Prevention in the Process Industries, 26(3), 462e467.
In this regard the CCPS sub-cooled correlation (Weber correlation Cleary, V. M., Bowen, P. J., & Witlox, H. W. M. (2007). Flashing liquid jets and two-
due to mechanical break-up), which predicts a maximum stable phase dispersion -I. Experiments for derivation of atomisation correlations.
droplet size, seems to perform better. Given the relatively good Journal of Hazardous Materials, 142, 786e796.
De Vaull, G. E., & King, J. A. (1992). Similarity scaling of droplet evaporation and liquid
agreement of the modified CCPS correlation against experiments rain-out following the release of superheated liquid to the environment, 85th
further complication of the model to incorporate an enhanced Annual Meeting. Kansas: Air and Waste Management Assoc.
horizontal momentum model (including horizontal drag) and Goldwire, H. C. (1986). Large-scale ammonia spill tests. Chemical Engineering
Progress, 82(4), 35e41.
secondary break-up effects is hard to justify. In this light also note
Hanna, S. R., Strimaitis, D. G., & Chang, J. C. (1991). Hazard response modelling
that the CCPS correlation was developed to give the best fit for uncertainty (a quantitative method). report. Westford, MA: Sigma Research
rainout, and therefore implicitly is meant to include both primary Corp., for the API.
Hervieu, E., & Veneau, T. (1996). Experimental determination of the droplet size and
and secondary break-up.
velocity distributions at the exit of the bottom discharge pipe of a liquefied
A literature survey, validation and possibly improved modelling propane storage tank during a sudden blowdown. Journal of Loss Prevention in
are recommended for post-expansion modelling (velocity and the Process Industries, 9(6), 413e455.
liquid mass fraction) and near-field jet dispersion modelling, Ichard, M., Hansen, O. R., Melheim, J. (2010). Releases of pressurized liquefied gases:
Simulations of the Desert Tortoise test series with the CFD model FLACS. In 16th
possibly together with some limited experimental work as neces- Conference on air pollution meteorology.
sary. This would complement previous JIP validation work associ- Johnson, D. W., & Woodward, J. L. (1999). A model with data to predict aerosol rainout
ated with flow rate, and post-expansion droplet size modelling. in accidental releases. New York: Center of Chemical Process Safety (CCPS).
Kay, P., Bowen, P. J., & Witlox, H. W. M. (2010). Sub-cooled and flashing liquid jets
Also additional rainout experiments are recommended for both and droplet dispersion, II. Scaled experiments and derivation of droplet
sub-cooled releases (at larger scale) and flashing releases atomisation correlations. Journal of Loss Prevention in the Process Industries, 23,
(including pool re-evaporation). At a later stage, additional work 849e856.
Kletz, T. A. (1977). Unconfined vapour cloud explosions. AIChE Loss Prevention, 11, 50.
could be related to multi-component releases to study the effect of Lautkaski, R. (2008). Experimental correlations for the estimation of the rainout of
release of mixtures. This could include further improvement of flashing liquid releases e Revisited. Journal of Loss Prevention in the Process
multi-component modelling (droplet modelling and pool Industries, 21, 506e511.
Ramsdale, S. A., & Tickle, G. A. (2000). Review of RELEASE rainout model and CCPS
evaporation).
data, contract by AEA technology for HSE. HSE Books, Contract Research Report
277/2000.
Touil, A., Bigot, J. P., Bonnet, P., Lacome, J. M., & Duplantier, S. (2004). Rainout prediction:
Acknowledgements Initial droplet diameter experimental determination, loss prevention and safety
promotion in the process industries. In 11th Int. Symp., Prague, 31 Maye3 June 2004.
Financial support of the work reported in this paper (Phase IV Witlox, H. W. M., & Bowen, P. (2002). Flashing liquid jets and two-phase dispersion e
A review. Contract 41003600 for HSE. Exxon-Mobil and ICI Eutech, HSE Books,
project) was provided by DNV Software, Exxon Mobil, UK Health Contract research report 403/2002.
and Safety Executive (HSE), RIVM (Dutch Government), Statoil and Witlox, H. W. M., Harper, M., Bowen, P. J., & Cleary, V. M. (2007). Flashing liquid jets
TOTAL. The contents of this paper including any opinions and/or and two-phase dispersion-II. Comparison and validation of droplet size and
rainout formulations. Journal of Hazardous Materials, 142, 797e809.
conclusions expressed, are those of the authors alone and do not Witlox, H. W. M., Harper, M., Oke, A., Kay, P., & Bowen, P. J. (2010). Sub-cooled and
necessarily reflect the policy of these organizations. Also help from flashing liquid jets and droplet dispersion, I. Overview and model imple-
Jan Stene (DNV Software) is acknowledged in producing droplet mentation/validation. Journal of Loss Prevention in the Process Industries, 23,
831e842.
size results. The authors also wish to acknowledge the assistance of Yildiz, D., Rambaud. P., van Beeck, J. (2004). Break-up, droplet size and velocity char-
Prof. Phil Bowen (Cardiff University), who acted as an external acterisations of a two-phase flashing R134A jet. In 5th International Conference on
advisor for the current project. Multiphase Flow (ICMF’04), Yokohama, Japan, May 30eJune 4, Paper No. 408.