Validation and verification of Phast

CFD: a comparative analysis with

Phast, KFX and tests
DNV

Date: March 2024

Reference to part of this report which may lead to misinterpretation is not permissible.

No. Date Reason for Issue Prepared by Verified by Approved by

0 March 2024 Phast CFD Yongfu Xu David Alvarez Boedo Trixie Secillano

Michael Harper
1 Feb 2024 Phast CFD 9.10 Yongfu Xu Michael Harper David Alvarez Boedo

Date: March 2024

Prepared by: Digital Solutions at DNV

© DNV AS. All rights reserved

This publication or parts thereof may not be reproduced or transmitted in any form or by any means, including copying
or recording, without the prior written consent of DNV AS.
Summary

Phast 9.0 provides users with the capability to simulate pool fires, jet fires and vapour dispersion using the Phast
software and the Phast CFD capabilities within Phast (i.e. running KFX via the Phast interface) for hazard assessment.
Some users may also have access to the KFX software (i.e. running cases via KFX interface directly).

This document provides validation and verification of the results produced by Phast CFD, powered by KFX, by
comparing them with those from Phast and KFX. It summarises the work conducted by DNV to analyse a range of test
cases for dispersions, jet fires and pool fires using Phast, Phast CFD and KFX, and to highlight any significant
discrepancies in results between these tools and their underlying reasons.

The analysis reveals the following findings:

- Dispersion predictions: Predictions of dispersion for vapour releases by Phast CFD are satisfactory.
However, the predictions are sensitive to surface roughness. Thus, appropriate surface roughness values
are crucial for releases near the ground, rather than relying solely on the default setting in Phast. Finer grid
resolution (i.e. more than 500,000 cells) will generally produce better dispersion results and is recommended
if computing resources permit.

- Two-phase ammonia test cases: For the two-phase ammonia test cases, Phast CFD tends to produce
under-predictions for dispersion at locations distant from the cloud centre line. While the predictions follow a
similar trend as Phast, they exhibit broader scattering. However, Phast CFD does slightly overpredict
arcwise maximum concentrations for all FLADIS test cases. This behaviour in Phast CFD results can be
primarily attributed to the fact that Phast CFD results are predicted with fixed wind direction, which does not
account for fluctuations in ambient conditions, such as wind meandering. Additionally, droplet
characteristics, including droplet size distribution and spray angle, may have also contributed to the
observed under-prediction and broader scattering. For scenarios with stable ambient conditions or scenarios
with fixed directions, Phast CFD should give good results. However, for cases with large variations in
ambient condition, Phast CFD simulations for multiple wind speeds and directions may be necessary to
achieve a comprehensive dispersion result. Users with access to the KFX software can use the KFX Large
Eddy Simulation (LES) model to get more accurate dispersion results, but the computing time will be
significantly longer.

- Jet fire predictions: While predictions of jet fires by Phast CFD are satisfactory, potential under-predictions
are observed for jet fire scenarios with large release areas, i.e. scenarios with large, expanded diameters.

- Pool fire predictions: Predictions of pool fires by Phast CFD are also satisfactory. The results are sensitive to
surface roughness; therefore, appropriate surface roughness is essential, rather than relying solely on the
default setting in Phast.

It is important to note that the test cases analysed in this work did not include complex geometries, as these cases
can be effectively solved by simple models implemented in Phast. Phast CFD and KFX employ advanced numerical
methods and algorithms to solve and analyse problems of fluid flows and excel in scenarios involving complex
terrains and geometries. For cases with complex geometries or requiring detailed results, Phast CFD and KFX offer
superior capabilities.

Validation and Verification | Phast CFD (Powered by KFX) Page i

Table of contents
1 INTRODUCTION ......................................................................................................................................... 4

2 METHODOLOGY ........................................................................................................................................ 5
2.1 Terminology 5
2.2 The approach 5

3 DISPERSION RESULTS ............................................................................................................................. 6

3.1 Vapour release of methane 6
3.1.1 The base case of sensitivity analysis 6
3.1.2 Result of the base case 6
3.2 Horizontal vapour releases of hydrogen 10
3.2.1 The test case 10
3.2.2 Results and Observations 10
3.3 Vertical releases of natural gas and hydrogen 13
3.4 Two-phase release of ammonia 14
3.4.1 Test cases 14
3.4.2 Sensitivity study using the FLADIS09 test case 15
3.4.3 Results and Observations of the FLADIS test cases 18

4 RADIATION OF JET FIRES....................................................................................................................... 23

4.1 Horizontal jet fires 23
4.1.1 The test cases 23
4.1.2 Results and observations 23
4.2 Vertical jet fires 27
4.2.1 Input data 27
4.2.2 Results and observations 27
4.3 Horizontal two-phase jet fires 31
4.3.1 Test cases 31
4.3.2 Results and observations 31
4.4 Hydrogen jet fires 34
4.4.1 The test cases 34
4.4.2 Results and observations 34
4.5 Jet fire with a large release area 36
4.5.1 Test case 36
4.5.2 Results and observations 36
4.6 Summary of Jet fire simulations 37

5 RADIATION OF POOL FIRES ................................................................................................................... 38

5.1 Test cases 38
5.2 Results & observations 38

6 KNOWN ISSUES AND WORKAROUNDS.................................................................................................. 41

6.1 Pool fires 41

7 CONCLUSIONS AND RECOMMENDATIONS ........................................................................................... 42

8 REFERENCES.......................................................................................................................................... 43

Validation and Verification | Phast CFD (Powered by KFX) Page 1

List of figures
Figure 1 Footprint contours of 0.5LFL (25000ppm) at release height of the base case（default settings on parameters）
.................................................................................................................................................................................. 7
Figure 2 Sideview contours of 0.5LFL (25000ppm) crossing the release point of the base case (default settings on other
parameters)................................................................................................................................................................ 7
Figure 3 Effect of grid sensitivity on side views of 0.5LFL of Phast CFD predictions ...................................................... 8
Figure 4 Effect of release height on side views of 0.5LFL of Phast CFD predictions (2 million cells, 5mm surface
roughness) ................................................................................................................................................................. 8
Figure 5 Effect of surface roughness on side views of 0.5LFL of Phast CFD predictions (2 million cells, release height at
50m) .......................................................................................................................................................................... 9
Figure 6 Comparing effect distances predicted by Phast and Phast CFD ...................................................................... 9
Figure 7 Phast CFD predictions of cloud side view (grey contour corresponds to 0.5LFL concentration) .......................11
Figure 8 Phast predictions of cloud side views (red contour corresponds to 0.5LFL concentration) ...............................11
Figure 9 Comparing measured arc-wise max concentrations against predictions by Phast CFD and Phast ...................12
Figure 10 Statistical assessment of measured arc-wise max concentrations against predictions by Phast CFD and Phast
.................................................................................................................................................................................12
Figure 11 Locations of concentration sensors in the FLADIS experiments ...................................................................15
Figure 12 Concentration footprint of 750ppm (ERPG3) at the release height ...............................................................16
Figure 13 Concentration sideview of 750ppm (ERPG3) crossing the release point .......................................................16
Figure 14 Results sensitivity to grid and domain size (Blue: 2 million grid cells and user defined domain size; Orange: 2
million grid cells and default domain size; Red: 500,000 grid cells and used defined domain size) ................................17
Figure 15 Results sensitivity to spray specification of the FLADIS09 test case (Blue: uniform droplet size and a spray
angle of 60 degrees; Orange: uniform droplet size and default spray angle of 30 degrees; Red: a log-normal distribution
of droplets and default spray angle of 30 degrees) ......................................................................................................17
Figure 16 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS09 test case (Blue: maximum
concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD) ..................................19
Figure 17 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS16 test case (Blue: maximum
concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD) ..................................19
Figure 18 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS24 test case (Blue: maximum
concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD) ..................................20
Figure 19 Positions of location sensors with significant under-predictions in results predicted Phast CFD for FLADIS09
test case (values in the graph indicate ratios between predicted and observed concentrations at locations of the nearby
dots) .........................................................................................................................................................................20
Figure 20 Comparing predictions of arcwise maximum concentration by Phast CFD, Phast and measurements for the
FLADIS test cases .....................................................................................................................................................21
Figure 21 Statistical assessment of Phast CFD predictions of arcwise maximum concentration for the FLADIS test cases
.................................................................................................................................................................................21
Figure 22 Comparing the results of Phast CFD and KFX LES against measurements for the FLADIS24 test case ........22
Figure 23 Comparing measured incident radiation (kW/m 2) against Phast predictions for the Johnson jet fire test cases
.................................................................................................................................................................................24
Figure 24 Statistical assessment of Phast predictions of incident radiation for the Johnson jet fire test cases................24
Figure 25 Comparing measured incident radiation (kW/m 2) against Phast CFD predictions for the Johnson jet fire test
cases ........................................................................................................................................................................25
Figure 26 Statistical assessment of Phast CFD predictions of incident radiation for the Johnson jet fire test cases ........25
Figure 27 Comparing predictions of incident radiation by Phast, Phast CFD and KFX with point and planer observer
types (kW/m2) for the Johnson 1033 test case (point means point observer type, planer means planer observer type) ..26
Figure 28 Statistical assessment of predictions of incident radiation by Phast, Phast CFD and KFX with point and planer
observer types for the Johnson 1033 test case (point means point observer type, planer means planer observer type) .26
Figure 29 Comparing measured incident radiation (kW/m 2) against Phast predictions for the Chamberlain test cases ...28
Figure 30 Statistical assessment of Phast predictions of incident radiation for the Chamberlain test cases ...................28
Figure 31 Comparing predictions of incident radiation by Phast with point and planer observer types (kW/m 2) for the
Chamberlain test case (point means point observer type, planer means planer observer type) .....................................29
Figure 32 Comparing measured incident radiation (kW/m 2) against Phast CFD predictions for the Chamberlain test
cases ........................................................................................................................................................................29
Figure 33 Statistic assessment of Phast CFD predictions of incident radiation for the Chamberlain test cases ..............30
Figure 34 Comparing measured incident radiation (kW/m 2) against Phast predictions for the Bennett test cases of two-
phase jet fires ............................................................................................................................................................32
Figure 35 Statistical assessment of Phast predictions of incident radiation for the Bennett test cases of two-phase jet
fires...........................................................................................................................................................................32
Figure 36 Comparing measured incident radiation (kW/m 2) against Phast CFD predictions for the Bennett test cases of
2-phase jet fires .........................................................................................................................................................33

Validation and Verification | Phast CFD (Powered by KFX) Page 2

Figure 37 Statistical assessment of Phast CFD predictions of incident radiation for the Bennett test cases of two-phase
jet fires ......................................................................................................................................................................33
Figure 38 Comparing measured incident radiation (kW/m 2) against Phast CFD predictions of hydrogen jet fires ...........35
Figure 39 Statistical assessment of Phast CFD predictions of incident radiation for hydrogen jet fires...........................35
Figure 40 Phast CFD prediction of radiation distribution (kW/m 2) on a vertical plane (displayed using KFX View) ..........37
Figure 41 KFX prediction with the same input data and refined grid as Phast CFD on the same vertical plane as shown
above........................................................................................................................................................................37
Figure 42 Sensitivity analysis of Phast CFD predictions of incident radiation (kW/m 2) with ground roughness for pool fire
simulations ................................................................................................................................................................39
Figure 43 Comparing measured incident radiation(kW/m 2) against Phast CFD predictions of the Johnson trials of LNG
pool fires ...................................................................................................................................................................39
Figure 44 Statistical assessment of Phast CFD predictions of incident radiation for Johnson trials of LNG pool fires .....40

List of tables
Table 1 Release and ambient conditions of the base case ........................................................................................... 6
Table 2 Release and ambient conditions of Shell HSL09 ............................................................................................10
Table 3 Release and ambient conditions for the FLADIS experiments .........................................................................14
Table 4 Release and ambient conditions of the Johnson jet fire test cases ..................................................................23
Table 5 Release and ambient conditions for Chamberlain test cases ...........................................................................27
Table 6 Release & ambient conditions of the Cook model ...........................................................................................31
Table 7 Release and ambient conditions of the test cases for hydrogen test cases ......................................................34
Table 8 Release and ambient conditions for the Fishburne test ...................................................................................36
Table 9 Pool and ambient conditions of the Johnson field trials ...................................................................................38

Validation and Verification | Phast CFD (Powered by KFX) Page 3

1 INTRODUCTION

In our continuous effort to advance our consequence modeling capabilities, we have incorporated computational fluid
dynamics (CFD) capabilities in the Phast software to enable users to conduct comprehensive modeling of jet fires, pool
fires, and dispersion scenarios. We've progressively introduced these CFD capabilities: pool fire modeling in version 8.6,
jet fire modeling in version 8.7, and dispersion modeling in version 9.0, through Phast CFD. Phast CFD is powered by
the KFX software, our advanced CFD simulation software for fires and dispersion. Some users may also have access to
KFX (i.e. running cases via KFX interface directly). Naturally, variations in results can occur between these tools for the
same case.

This document provides validation and verification of the results predicted by Phast CFD by comparing them with those
from Phast and KFX. Through a series of test cases, we assess the accuracy of Phast CFD, identify any significant
discrepancies, and investigate their causes. Such validation and verification are crucial to ensure our software's
reliability and effectiveness in supporting critical safety decisions.

Each test case analysis begins with building a Phast case, then running it with both Phast and Phast CFD. If differences
are observed, KFX, which offers more model control, is used for further investigations.

While this analysis currently covers only a selection of test cases, not all Phast test cases are included in this work and
ongoing updates to this document will include additional scenarios as more results become available.

Validation and Verification | Phast CFD (Powered by KFX) Page 4

2 METHODOLOGY

2.1 Terminology

Four types of results are compared in this document:

- Test results, also refer to as observed data, were obtained through experiments and tests and are used to
validate model predictions.

- Phast results were predictions generated by models implemented within Phast.

- Phast CFD results were predictions produced by running KFX through the Phast interface. Assumptions
have been made by Phast CFD to enable accurate CFD results with minimal efforts. It is important to note
that users have limited control over the modelling and the default model settings may lead to less accurate
results for some cases.

- KFX results were predictions obtained using the KFX graphical user interface (GUI) of the KFX software.
While most KFX runs in this work started with JSON files generated by Phast as inputs, some changes were
made manually based on complexity of the geometry and release scenarios. These changes include grid
distribution, parameter settings or additional outputs.

2.2 The approach

The majority of the cases analysed in this work are test cases of the Phast models. The following steps were taken for
each case to generate results for comparison:

1. Phast results. Phast results were primarily obtained from existing study files of the test cases, with a few
cases generated to test specific scenarios.

2. Phast CFD results. For dispersion cases, Phast CFD, powered by KFX, was executed from Phast via CFD
dispersion cases created within Phast. Jet fire and pool fire cases were run from the Phast scenarios directly.

3. KFX results. For cases needing more control over the model, such as grid, parameter settings or additional
outputs, KFX was run using JSON files created by Phast as inputs.

4. Comparing results. Results from Phast CFD are compared with Phast predictions and test results if available.
For dispersion scenarios, the comparative analysis includes contours of specified concentrations and
concentrations at specific points or along arcs. In cases of jet fires and pool fires, the focus is on comparing
radiation levels at measurement points. In instances where significant discrepancies are noted, KFX results are
applied to understand the differences.

Validation and Verification | Phast CFD (Powered by KFX) Page 5

3 DISPERSION RESULTS

3.1 Vapour release of methane

3.1.1 The base case of sensitivity analysis

This is a simple test specially created to assess sensitivity of Phast CFD predictions to grid distributions and parameter
settings. The results are only compared against Phast predictions. Table 1 below lists the release and ambient
conditions of the case. High jet velocity significantly increases CFD running time. In this sensitivity analysis, the vessel
pressure was intentionally set low to ensure a reasonable CFD running time for a large number of cases. The discharge
velocity of this methane case is 165 m/s, which may be lower than in many release scenarios; however, the findings
should still be applicable to a wide range of vapor release situations.

Table 1 Release and ambient conditions of the base case

Value
Material methane
Vessel pressure (barg) 0.1
Vessel temperature (degC) 10
Vessel inventory (kg) 10000
Release height (m) 1
Hole size (mm) 100
Ambient air temperature (degC) 9.85
Ambient air pressure (bar) 1
Ambient humidity (%) 70
Wind Speed (m/s) 5
Pasquill stability D
Averaging time (s) 18.75 (Phast results)
Surface roughness (mm) 5
183.15 (Phast default)

3.1.2 Result of the base case

- Differences are observed between dispersion results by Phast and Phast CFD for the base case at default
setting for surface roughness in Phast, i.e. 183.156mm, as shown in Figure 1 & Figure 2. The differences
can be attributed to several factors:

 Phast CFD imposes a limit on surface roughness that may lead to a smaller roughness value being used
in simulations for cases with large roughness, such as the default roughness value in Phast. When surface
roughness is larger than the height of the control volumes used in a CFD simulation, it implies that there is
geometry considered by Phast CFD, so that downwind obstructions (i.e. 3D geometry or terrain data)
resemble the surface roughness.

 Phast CFD directly simulates the ground effects on vapour dispersions.

- Dispersion results from Phast CFD are shown to be sensitive to grid resolution, release height and surface
roughness as shown in Figure 3, Figure 4 & Figure 5, respectively.

- Finer grid resolution (i.e. more than 500,000 cells) will generally produce better dispersion results and is
recommended if computing resources permit.

Validation and Verification | Phast CFD (Powered by KFX) Page 6

 - Surface roughness has a large impact on dispersion results for releases near to the ground. So, it is
important to select appropriate surface roughness for dispersion simulations.

- Comparison of the effect distances predicted by Phast and Phast CFD is presented in Figure 6. Large
differences are revealed between Phast & Phast CFD predictions for releases near to the ground. Phast
CFD gives more conservative predictions of the effect distance for releases near to the ground.

Figure 1 Footprint contours of 0.5LFL (25000ppm) at release height of the base case（default settings on
parameters）

Figure 2 Sideview contours of 0.5LFL (25000ppm) crossing the release point of the base case (default settings
on other parameters)

Validation and Verification | Phast CFD (Powered by KFX) Page 7

Figure 3 Effect of grid sensitivity on side views of 0.5LFL of Phast CFD predictions

Figure 4 Effect of release height on side views of 0.5LFL of Phast CFD predictions (2 million cells, 5mm surface
roughness)

Validation and Verification | Phast CFD (Powered by KFX) Page 8

Figure 5 Effect of surface roughness on side views of 0.5LFL of Phast CFD predictions (2 million cells, release
height at 50m)

Figure 6 Comparing effect distances predicted by Phast and Phast CFD

Validation and Verification | Phast CFD (Powered by KFX) Page 9

3.2 Horizontal vapour releases of hydrogen

3.2.1 The test case

The Shell HSL tests (Roberts, 2006) were used to validate dispersion results of UDM (Unified Dispersion Model) in
Phast. Measurements of arc-wide max concentrations were used in the validation. Table 2 below shows input data of
the Shell HSL09 test case for Phast and Phast CFD simulations.

Table 2 Release and ambient conditions of Shell HSL09

Shell HSL09
Material Hydrogen
Mass flowrate (kg/s) 0.0728
Release temperature (deg C) -130
Release velocity (m/s) 2036
Release height (m) 1.5
Hole size (mm) 3
Ambient air temperature (deg C) 13.5
Ambient air pressure (bar) 1
Ambient humidity (%) 70
Wind Speed (m/s) 3
Pasquill stability D
Averaging time (s) 18.75 (Phast results)
Surface roughness (mm) 10

3.2.2 Results and Observations

- The contours of LFL & 0.5LFL predicted by Phast CFD for this case do not touch the ground (as shown in
Figure 7) and are similar to the Phast predictions (as shown in Figure 8).

- Phast CFD predicts longer 0.25LFL contours compared to Phast predictions. The predicted 0.25LFL contour
extends to the downwind boundary of the computational domain (as shown in Figure 7), which suggests that
a larger computational domain is necessary for Phast CFD if 0.25LFL concentration is of concern.

- Predictions from both Phast & Phast CFD show good agreement with measurements of max arc-wide
concentrations, as illustrated in Figure 9 & Figure 10. The geometrical mean bias, MG, and geometric
variance, VG, are two statistical values in commonly used to assess quality of datasets. These
measurements are all within 11m from the release point at the release height and so fall within the contour of
predicted 0.5LFL concentration. The relatively stable ambient conditions may have led to the max arc-wide
concentrations occurring near to cloud centre line, potentially contributing to the good agreement between
the predictions and measurements.

Validation and Verification | Phast CFD (Powered by KFX) Page 10

Figure 7 Phast CFD predictions of cloud side view (grey contour corresponds to 0.5LFL concentration)

Figure 8 Phast predictions of cloud side views (red contour corresponds to 0.5LFL concentration)

Validation and Verification | Phast CFD (Powered by KFX) Page 11

Figure 9 Comparing measured arc-wise max concentrations against predictions by Phast CFD and Phast

Figure 10 Statistical assessment of measured arc-wise max concentrations against predictions by Phast CFD
and Phast

Validation and Verification | Phast CFD (Powered by KFX) Page 12

3.3 Vertical releases of natural gas and hydrogen

DNV consultants have performed CFD simulations of vertical venting of natural gas and hydrogen using FLACS. The
work focused on determining the effect ranges of hydrogen and natural gas releases (i.e. the downwind extents and
heights of the cloud) at specified concentrations, particularly the 0.5LFL concentration. While details of these simulations
are confidential, the following observations have been made:

- Both Phast CFD and FLACS generate very similar predictions in trends and magnitudes of the effect ranges
for the simulated scenarios of hydrogen and natural gas releases.

- Phast produces results that are comparable to Phast CFD and FLACS in terms of the heights of 0.5LFL
contours for both natural gas and hydrogen release scenarios. However, Phast results tend to be more
conservative in predicting the extent of 0.5LFL contours at high wind speeds and less conservative at low
wind speeds.

Validation and Verification | Phast CFD (Powered by KFX) Page 13

3.4 Two-phase release of ammonia

3.4.1 Test cases

FLADIS (Nielsen, 1996) is a project designed to investigate dispersion of ammonia aerosols. Liquefied ammonia was
released under pressure through a nozzle positioned at a height of 1.5m. The experiments, characterised by low release
rates, were aimed at investigating the far-field passive effects. Ammonia was released in two phases, but no liquid pool
was observed during the tests. Table 3 below shows the input data for Phast & Phast CFD simulations. Figure 11 shows
the locations where concentrations were measured in the test. Most of the concentration sensors were arranged on
three arcs at distances of 20m, 70m and 238m. Further details of the tests can be found in the Phast validation
document for the Unified Dispersion Model (DNV, Validation: Unified Dispersion Model, 2023).

Table 3 Release and ambient conditions for the FLADIS experiments

Validation and Verification | Phast CFD (Powered by KFX) Page 14

Figure 11 Locations of concentration sensors in the FLADIS experiments

3.4.2 Sensitivity study using the FLADIS09 test case

- Figure 12 & Figure 13 compare the contours of ERPG3 concentration of ammonia, i.e. 750ppm, predicted by
Phast and Phast CFD. Phast predictions appear to be more conservative in this case.

- Figure 14 compares dispersion results with varied grid resolutions and user-defined domain sizes. While the
default number of grid cells (500,000) produces similar results to 2 million cells, the default domain size
based on Phast results seems large, particularly in the vertical direction. A reduced domain size via user-
defined domain setting has resulted in slightly improved predictions

- Figure 15 shows the sensitivity of predicted concentrations to spray characteristics for liquid ammonia in the
releases. Three results are as follows:

 Default specification: uniform droplet size and a spray angle of 30 degrees

 60 deg: uniform droplet size and a spray angle of 60 degrees

 DropletDist: a log-normal distribution of droplet sizes and a spray angle of 30 degrees

Spray characteristics do have some impacts on dispersion results. The release with an increased spray angle
of 60 degrees yields marginally better results than the other two spray specifications in this test case.

- Based on this sensitivity analysis, the results presented below for the FLADIS experiments are modelled
with the following settings: reduced domain size, 2 million grid cells and default spray specification. Domain
size and cell number can be adjusted in Phast CFD via the CFD grid parameters.

Validation and Verification | Phast CFD (Powered by KFX) Page 15

Figure 12 Concentration footprint of 750ppm (ERPG3) at the release height

Figure 13 Concentration sideview of 750ppm (ERPG3) crossing the release point

Validation and Verification | Phast CFD (Powered by KFX) Page 16

Figure 14 Results sensitivity to grid and domain size (Blue: 2 million grid cells and user defined domain size;
Orange: 2 million grid cells and default domain size; Red: 500,000 grid cells and used defined domain size)

Figure 15 Results sensitivity to spray specification of the FLADIS09 test case (Blue: uniform droplet size and a
spray angle of 60 degrees; Orange: uniform droplet size and default spray angle of 30 degrees; Red: a log-
normal distribution of droplets and default spray angle of 30 degrees)

Validation and Verification | Phast CFD (Powered by KFX) Page 17

3.4.3 Results and Observations of the FLADIS test cases

- Phast CFD predictions for the three FLADIS test cases are shown in Figure 16, Figure 17 and Figure 18.
The predictions exhibit deviations from test measurements, with broader scattering compared to Phast
predictions.

- Figure 19 illustrates contours of ERPG2 & ERPG3 concentrations predicted by Phast CFD, along with the
locations of sensors exhibiting significant under-predictions. The majority of the under-predictions occur
outside the ERPG2 contours. The under-predictions can be primarily attributed to the fact that Phast CFD
results are calculated for a fixed wind direction and do not account for fluctuation in ambient conditions, such
as wind meandering. Consequently, under-predictions occur at locations far from the cloud centre line (as
shown in Figure 19), owing to wind meandering’s propensity to induce wider dispersion compared to a fixed
wind direction. Wind meandering is addressed in Phast by applying an averaging time of 600s for ammonia.
Additionally, droplet characteristics, including droplet size distribution and spray angle, may have also
contributed to the observed under-prediction and broader scattering.

- To counter the influence of wind meander, a method commonly applied to evaluate dispersion predictions
involves comparing arcwise maximum concentrations as used in UDM validation (DNV, Validation: Unified
Dispersion Model, 2023). Figure 20 presents a comparison of arcwise maximum concentrations for the
FLADIS test cases. Phast CFD tends to produce a bit overpredictions with results near top of the factor-of-
two lines. Statistical MG/VG of the arcwise maximum concentrations of Phast CFD predictions are shown in
Figure 21. Notably, FLADIS09 & 24 fall just outside the vertical dotted lines and confirm slight
overpredictions as illustrated in Figure 20. This slight overpredictions of arcwise maximum concentration
suggests that the under-predictions shown in Figure 19 are primarily due to fluctuations in ambient
conditions.

- The Large Eddy Simulation (LES) model in KFX solves the largest eddies in a turbulent flow by a Navier-
Stokes solver, and then models the smaller eddies by an eddy viscosity, and therefore yields more realistic
results for gas dispersion and fires. However, this model is only available in KFX (i.e. not via Phast CFD) and
requires significantly more computing power. The LES model in KFX was applied here to investigate the
under-predictions observed in Phast CFD predictions and the results are shown in Figure 22 for the
FLADIS24 test case. Even though the simulation has been stopped a bit too soon due to significant
computing time, the KFX LES model does give improved predictions at some locations. The Large Eddy
Simulation (LES) model may present an alternative for users with access to KFX to get more accurate
dispersion results.

- For scenarios with stable ambient conditions or QRA scenarios with fixed directions, Phast CFD should give
good dispersion results. However, for cases with large variations in ambient condition, Phast CFD
simulations with fixed directions may significantly under-predict off-centreline concentrations. Hence Phast
CFD simulations for multiple wind speeds and directions may be necessary to achieve a comprehensive
dispersion result of a case. Users with access to the KFX software can use the KFX Large Eddy Simulation
(LES) model to get more accurate dispersion results, but the computing time will be significantly longer.

Validation and Verification | Phast CFD (Powered by KFX) Page 18

Figure 16 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS09 test case (Blue:
maximum concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD)

Figure 17 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS16 test case (Blue:
maximum concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD)

Validation and Verification | Phast CFD (Powered by KFX) Page 19

Figure 18 Comparing predictions by Phast CFD, Phast and measurements for the FLADIS24 test case (Blue:
maximum concentrations predicted by Phast; Orange: maximum concentrations predicted by Phast CFD)

Figure 19 Positions of location sensors with significant under-predictions in results predicted Phast CFD for
FLADIS09 test case (values in the graph indicate ratios between predicted and observed concentrations at
locations of the nearby dots)

Validation and Verification | Phast CFD (Powered by KFX) Page 20

Figure 20 Comparing predictions of arcwise maximum concentration by Phast CFD, Phast and measurements
for the FLADIS test cases

Figure 21 Statistical assessment of Phast CFD predictions of arcwise maximum concentration for the FLADIS
test cases

Validation and Verification | Phast CFD (Powered by KFX) Page 21

Figure 22 Comparing the results of Phast CFD and KFX LES against measurements for the FLADIS24 test case

Validation and Verification | Phast CFD (Powered by KFX) Page 22

4 RADIATION OF JET FIRES

4.1 Horizontal jet fires

4.1.1 The test cases

Seven tests (1042, 1040, 1083, 1037, 1033, 1036 and 1089) that relate to free jet flames by Johnson et al (Johnson
A.D., 1994) were used to validate the Johnson Cone jet fire model implemented in Phast for horizontal vapour releases
(DNV, Validation: Jet fire, 2023). Detailed information about these test cases can be found in the validation document of
Phast for the jet fire models. In this analysis, three of the test cases were analysed to compare jet fire predictions
between Phast and Phast CFD. The release and ambient conditions are listed in Table 4.

Table 4 Release and ambient conditions of the Johnson jet fire test cases
1033 1083 1089
Material Natural gas Natural gas Natural gas
Mass flowrate (kg/s) 8.2 8.6 3.7
Expanded temperature (K) 172.8 203.7 143.9
Expanded diameter (m) 0.119 0.1472 0.07279
Release height (m) 3 3 3
Hole size (mm) 75 152 20
Ambient air temperature (K) 282.35 281.25 285.85
Ambient air pressure (bar) 1 0.984 0.970
Ambient humidity (%) 81 80 91
Wind Speed (m/s) 4 0.3 9
Wind direction from north (o) 255 340 260
Pasquill stability F F F
Surface roughness (mm) 183.156 183.156 183.156

4.1.2 Results and observations

- These test cases were integral to the data used to develop the Johnson Jet fire model implemented in
Phast. It is unsurprising that Phast predictions are in good agreement with the measurements, as shown in
Figure 23 & Figure 24. In this comparison, both measurements and Phast predictions are directional
radiation, i.e. radiation measured by radiometers with fixed directions.

- Phast CFD predictions exhibit a conservative trend cross all three test cases, particularly Test-1083, as
shown in Figure 25 & Figure 26. Phast CFD predicts maximum radiation at the measurement locations, i.e.
radiation for point observers, and so the predictions should be slightly higher than the measurements which
are for planer observers. Hence, a portion of the conservatism observed in Figure 25 & Figure 26 can be
attributed to the difference in observer type in Phast CFD predictions and measurements.

- However, only a small portion of the conservatism shown in Figure 25 & Figure 26 are due to differences in
observer type, as demonstrated in Figure 27 & Figure 28 for the Test-1033 case. Radiation variances
between point and planer observers are small in Test-1033 and similar behaviors are also observed in other
test cases. This indicates that radiometers were exposed to near maximum radiation at the fixed directions
during the tests and the measured radiation are close to radiation of point observers. Thus, the
measurements can be compared with Phast CFD predictions. KFX was required to obtain radiation of planer
observers at specified directions.

Validation and Verification | Phast CFD (Powered by KFX) Page 23

 - However, the majority of Phast CFD predictions fall within an acceptable range – specifically, within the
factor-of-two lines. Therefore, the predictions are considered reasonable.

Figure 23 Comparing measured incident radiation (kW/m2) against Phast predictions for the Johnson jet fire test
cases

Figure 24 Statistical assessment of Phast predictions of incident radiation for the Johnson jet fire test cases

Validation and Verification | Phast CFD (Powered by KFX) Page 24

Figure 25 Comparing measured incident radiation (kW/m2) against Phast CFD predictions for the Johnson jet
fire test cases

Figure 26 Statistical assessment of Phast CFD predictions of incident radiation for the Johnson jet fire test
cases

Validation and Verification | Phast CFD (Powered by KFX) Page 25

Figure 27 Comparing predictions of incident radiation by Phast, Phast CFD and KFX with point and planer
observer types (kW/m 2) for the Johnson 1033 test case (point means point observer type, planer means planer
observer type)

Figure 28 Statistical assessment of predictions of incident radiation by Phast, Phast CFD and KFX with point
and planer observer types for the Johnson 1033 test case (point means point observer type, planer means
planer observer type)

Validation and Verification | Phast CFD (Powered by KFX) Page 26

4.2 Vertical jet fires

4.2.1 Input data

The Chamberlain Cone jet fire model in Phast was validated against field data reported by Chamberlain (Chamberlain,
1987). This model requires specification of the fuel’s post-expansion thermodynamic properties (temperature or liquid
fraction) and dynamic properties (expanded radius or velocity) as input data. Unfortunately, the values for these release
properties were not included in the report. To address this, back calculations were performed using secondary data and
recommended discharge equations. As a result, the post-expansion temperature and velocity for each test case were
estimated, as listed in the table below.

Table 5 Release and ambient conditions for Chamberlain test cases

Test 3A Test 3D Test 4A Test 4C
Material Natural gas Natural gas Natural gas Natural gas
Mass flowrate (kg/s) 21.1 55.6 5.6 22.2
Expanded temperature (K) 289.83 326.51 284.15 198.3
Jet velocity (m/s) 26.83 74.29 238.71 544.83
Release height (m) 110 110 10 10
Ambient air temperature (K) 289.15 286.15 289.45 187.45
Ambient air pressure (bar) 1 1 1 1
Ambient humidity (%) 53 56 50 51
Wind Speed (m/s) 7.5 8 8.1 10.3
Pasquill stability D D D D

4.2.2 Results and observations

- These test cases were integral to the dataset used to develop the Chamberlain Jet fire model implemented
in Phast. Consequently, Phast predictions are in good agreement with the measurements, as shown in
Figure 29 Figure 29 & Figure 30.

- Radiation values compared here are for point observers in Test 3A & 3D. Measured radiation in Test 4A &
4C is for planer observers (i.e. radiation measured by radiometers with fixed directions). However, the
radiometers in Test 4A & 4C were exposed to near maximum radiation during the tests as demonstrated in
Figure 31. Thus, the measured radiation is close to radiation of point observers and can be compared with
Phast CFD predictions.

- Overall, Phast CFD predictions are satisfactory. All predictions for the four test cases fall within the factor-of-
two lines, as shown in Figure 32 & Figure 33. Minor variations are observed in case Test 3A.

Validation and Verification | Phast CFD (Powered by KFX) Page 27

Figure 29 Comparing measured incident radiation (kW/m2) against Phast predictions for the Chamberlain test
cases

Figure 30 Statistical assessment of Phast predictions of incident radiation for the Chamberlain test cases

Validation and Verification | Phast CFD (Powered by KFX) Page 28

Figure 31 Comparing predictions of incident radiation by Phast with point and planer observer types (kW/m 2)
for the Chamberlain test case (point means point observer type, planer means planer observer type)

Figure 32 Comparing measured incident radiation (kW/m2) against Phast CFD predictions for the Chamberlain
test cases

Validation and Verification | Phast CFD (Powered by KFX) Page 29

Figure 33 Statistic assessment of Phast CFD predictions of incident radiation for the Chamberlain test cases

Validation and Verification | Phast CFD (Powered by KFX) Page 30

4.3 Horizontal two-phase jet fires

4.3.1 Test cases

Phast CFD predictions for horizontal two-phase jet fires are assessed against radiation measurements from three test
cases conducted by Bennet et al (Bennett, 1991). These same cases were also used to validate the Cook Cone model
in Phast for two-phase jet fires. Table 6 below lists the input data required for these cases as for the standalone jet fires
in Phast. Some of the input data were back-calculated, and additional details can be found the theory manual of the jet
fire models of Phast (DNV, Jet fire, 2023).

Table 6 Release & ambient conditions of the Cook model

Bennett 3006 Bennett 3026 Bennett 3029
Material Propane Propane Propane
Mass flowrate (kg/s) 1.5464 16.1 18
Post expansion liquid fraction 0.737411 0.758563 0.77005
Jet velocity (m/s) 168.129 152.71 145.67
Release height (m) 1.5 3 1.5
Ambient air temperature (K) 286.35 286.85 281.15
Ambient air pressure (bar) 1 1 1
Ambient humidity (%) 79 59 82
Wind Speed (m/s) 5.8 3.7 2
Pasquill stability D D D
Surface roughness (mm) 183.156 183.156 183.156

4.3.2 Results and observations

- All radiation values compared here for two-phase releases are for point observers.

- Most Phast predictions fall within the factor-of-two lines, except for two measurements in the Bennett 3006
test case, as shown in Figure 34. Overall, Phast has produced slight overpredictions, as indicated by the
statistical assessment shown in Figure 35.

- Phast CFD predictions are generally satisfactory, with the majority of predictions falling within the factor-of-
two lines. However, there is one measurement in the Bennett 3006 case and two measurements in the
Bennett 3029 case that deviate, as shown in Figure 36. The predictions are conservative in all three cases,
particularly for the Bennett 3029 case.

Validation and Verification | Phast CFD (Powered by KFX) Page 31

Figure 34 Comparing measured incident radiation (kW/m2) against Phast predictions for the Bennett test cases
of two-phase jet fires

Figure 35 Statistical assessment of Phast predictions of incident radiation for the Bennett test cases of two-
phase jet fires

Validation and Verification | Phast CFD (Powered by KFX) Page 32

Figure 36 Comparing measured incident radiation (kW/m2) against Phast CFD predictions for the Bennett test
cases of 2-phase jet fires

Figure 37 Statistical assessment of Phast CFD predictions of incident radiation for the Bennett test cases of
two-phase jet fires

Validation and Verification | Phast CFD (Powered by KFX) Page 33

4.4 Hydrogen jet fires

4.4.1 The test cases

Tests of hydrogen jet fires were conducted by DNV at the Spadeadam test site on behalf of Air Products & Chemicals
Inc. in 2009 (Advantica, 2009). These tests were also used to develop the Miller jet fire model for hydrogen releases,
which was implemented in Phast 8.6. Table 7 below lists the input data in the Phast studies for the test cases.

Table 7 Release and ambient conditions of the test cases for hydrogen test cases
13_AP&DNVGL 14_AP&DNVGL 15_AP&DNVGL
Material Hydrogen Hydrogen Hydrogen
Mass flowrate (kg/s) 1 7.5 6.6
Expanded temperature (K) 142.303 141.385 142.065
Jet velocity (m/s) 1969.52 1978.97 1971.90
Orifice diameter (mm) 20.9 52.5 20.9
Release height (m) 3.25 3.25 3.25
Ambient air temperature (K) 280 287 287
Ambient air pressure (bar) 1 1 1
Ambient humidity (%) 94.3 94.2 94.3
Wind Speed (m/s) 3.4 2.6 2.6
Pasquill stability D D D
wind direction relating the release 178.5 -151.7 -142.7
direction (o)
Surface roughness (mm) 183.156 183.156 183.156

4.4.2 Results and observations

- All radiation values compared here for hydrogen jet fires are for point observers

- Most Phast CFD predictions fall within the factor-of-two lines, except for one measurement in the
14_AP_DNVGL and 15_AP_DNVGL test cases, as shown in Figure 38.

- Minor variations are observed in these test cases, as shown in Figure 39 from the statistical assessment.

- Overall, Phast CFD predictions for hydrogen jet fires are satisfactory.

Validation and Verification | Phast CFD (Powered by KFX) Page 34

Figure 38 Comparing measured incident radiation (kW/m2) against Phast CFD predictions of hydrogen jet fires

Figure 39 Statistical assessment of Phast CFD predictions of incident radiation for hydrogen jet fires

Validation and Verification | Phast CFD (Powered by KFX) Page 35

4.5 Jet fire with a large release area

4.5.1 Test case

This test case involves a hydrogen jet fire for the development the Miller jet fire model (Fishburne, 1979). Table 8
provides the input data used in the Phast study for this specific case. It is worth noting that the expanded diameter for
this case is 1.4415 meters, which is relatively large compared to most other test cases. The significance of this lies in
the fact that the expanded diameter of the input determines the size of the release cell employed in Phast CFD
simulations. Therefore, this case has been specifically examined to evaluate the impact of the release area on the
predicted results.

Table 8 Release and ambient conditions for the Fishburne test

Value
Material Hydrogen
Mass flowrate (kg/s) 22
Release temperature (deg C) 0
Release velocity (m/s) 150
Release height (m) 127
Ambient air temperature (deg C) 9.85
Ambient air pressure (bar) 1
Ambient humidity (%) 58
Wind Speed (m/s) 1.5
Pasquill stability C/D
Surface roughness (mm) 183.156

4.5.2 Results and observations

- There is no radiation measurement available for this case. Therefore, Phast CFD results are only compared
with KFX predictions for the same release with refined grids at the release point.

- Notable differences between the two sets of results can be observed in Figure 40 & Figure 41 for radiation
distribution. KFX results, obtained with refined release cells, show higher radiation distances, i.e. larger
contours for the specified radiation intensities. The differences are likely to be caused by the large release
cells automatically generated in the Phast CFD simulations.

- Therefore, for scenarios with large release areas, Phast CFD may produce under-predictions of the radiation
extent. Users are advised to simulate such cases using KFX with finer grid resolutions at the release point.

Validation and Verification | Phast CFD (Powered by KFX) Page 36

Figure 40 Phast CFD prediction of radiation distribution (kW/m 2) on a vertical plane (displayed using KFX View)

Figure 41 KFX prediction with the same input data and refined grid as Phast CFD on the same vertical plane as
shown above

4.6 Summary of Jet fire simulations

- Phast CFD predictions for jet fires are generally satisfactory based on the results of the test cases shown
above.
- For scenarios with large release areas, Phast CFD may produce under-predictions of the radiation extent.

Validation and Verification | Phast CFD (Powered by KFX) Page 37

5 RADIATION OF POOL FIRES

5.1 Test cases

Johnson conducted three tests related to LNG pool flames with diameters of 1.8, 6.1, and 10.6 metres (i.e. Field Trials
1, 6 and 7, respectively) (Johnson, 1992). These tests were carried out in shallow bunds with thermally insulated
concrete floors to minimise heat transfer to the pool from the substrate. Detailed information about the validation
process can be found in the theory manual for the pool fire models implemented in Phast (DNV, Pool fire model, 2023).
Table 9 below shows the input data for the test cases.

Table 9 Pool and ambient conditions of the Johnson field trials

Field trial1 Field trial 6 Field trial 7
Material LNG LNG LNG
Pool diameter (m) 1.8 6.1 10.6
Air temperature (K) 283.15 (assumed) 280.15 284.25
Air pressure (bar) 1.01325(assumed) 0.943 0.943
Relative Humidity (%) 70(assumed) 83 87
Wind speed (m/s) 2.4 6.6 4.0
Wind direction (clockwise from North) 270 250 90

5.2 Results & observations

- Phast CFD predictions of pool fires are sensitive to ground roughness, as shown in Figure 42. Surface
roughness of 5mm was used for the results shown below.

- The majority of Phast CFD predictions for the three test cases are within the factor-of-two lines, as shown
below in Figure 43. Overall, the predictions are satisfactory.

- However, the Phast CFD predictions are less conservative at some locations (i.e. the predictions are below
the factor-of-two lines), as shown in Figure 43 & Figure 44. Pool fires have low momentum and are easily
disturbed by wind variations, such as fluctuations in wind speed and direction, or effects of the local terrain
and obstructions. These variations are not included in the Phast CFD simulations and may have contributed
to the under-predictions.

Validation and Verification | Phast CFD (Powered by KFX) Page 38

Figure 42 Sensitivity analysis of Phast CFD predictions of incident radiation (kW/m2) with ground roughness for
pool fire simulations

Figure 43 Comparing measured incident radiation(kW/m2) against Phast CFD predictions of the Johnson trials
of LNG pool fires

Validation and Verification | Phast CFD (Powered by KFX) Page 39

Figure 44 Statistical assessment of Phast CFD predictions of incident radiation for Johnson trials of LNG pool
fires

Validation and Verification | Phast CFD (Powered by KFX) Page 40

6 KNOWN ISSUES AND WORKAROUNDS

6.1 Pool fires

In Phast CFD, a pool fire involving a mixture is modelled as a pseudo-component, with the fire starting at the boiling
point of the mixture. For a mixture of n-Dodecane and water, the boiling point is set at 100 °C (i.e. the boiling point of
water) in a Phast CFD simulation. Consequently, a pool fire of this mixture results only a small amount of vaporisation of
n-Dodecane and a small pool fire. To address this issue, the case should be modelled as a multi-component pool fire.
Unfortunately, the multi-component pool fire option is not available in Phast CFD; instead, KFX software must be used.

When using KFX for this case, water vaporises from the pool first due to its lower boiling point. Subsequently, n-
Dodecane vaporises and burns, produce a large pool fire. However, because water vaporises slowly in the simulation, it
may take a long computing time to reach the point where the large pool fire develops.

Validation and Verification | Phast CFD (Powered by KFX) Page 41

7 CONCLUSIONS AND RECOMMENDATIONS

The results from Phast CFD have been compared with predictions from Phast and measurements across a range of test
cases of dispersion, jet fires and pool fires. The key findings are as follows:

- Dispersion predictions: Phast CFD predictions of dispersion for vapor releases are satisfactory. However,
the predictions are sensitive to surface roughness. Thus, appropriate surface roughness values are
important for releases near the ground, instead of relying on the default setting of Phast.

- Two-phase ammonia test cases: In the case of the two-phase ammonia test cases, Phast CFD tends to
produce under-predictions for dispersion. While the predictions follow a similar trend as Phast, they exhibit
broader scattering. However Phast CFD does slightly overpredict arcwise maximum concentrations for all
FLADIS test cases. This behaviour of Phast CFD predictions can be primarily attributed to the fact that Phast
CFD results are predicted with fixed wind directions, which does not account for fluctuations in ambient
conditions, such as wind meandering. Additionally, droplet characteristics, including droplet size distribution
and spray angle, may have also contributed to the observed discrepancies. So, for scenarios with stable
ambient conditions or scenarios with fixed directions in risk assessments, Phast CFD should give good
dispersion results. However, for cases with large variations in ambient conditions, Phast CFD simulations for
multiple wind speeds and directions may be necessary to achieve a comprehensive dispersion result. Users
with access to the KFX software can use the KFX Large Eddy Simulation (LES) model to get more accurate
dispersion results.

- Jet fire predictions: While predictions of jet fires by Phast CFD are satisfactory, potential under-predictions
are observed for jet fire scenarios with large release areas, i.e. scenarios with large, expanded diameters.

- Pool fire predictions: Predictions of pool fires by Phast CFD are also satisfactory. The results are sensitive
to surface roughness; therefore, appropriate surface roughness is essential, rather than relying on the
default setting of Phast

It is important to note that the test cases analysed in this work did not include complex geometries, as these cases often
can be effectively solved by the simple models implemented in Phast. Nevertheless, for scenarios with complex
geometries or requiring detailed results, both Phast CFD and KFX offer superior capabilities.

Validation and Verification | Phast CFD (Powered by KFX) Page 42

8 REFERENCES

Advantica. (2009). Report on hydrogen jet fires carried out on behalf of Air Products & Chemicals Inc. Private report.
Allason, D. (2009). Report on hydrogen jet fires carried out on behalf of Air Products & Chemical Inc. . Report No 8653.
Bennett, J. C. (1991). Large scale narural gas and LPG jet fires. TNER 91.022.
Chamberlain, G. A. (1987). Development in design methods for predicting thermal radiation frlm flares. Chem. Eng. Res.
Des., Vol 65, pp299-309.
DNV. (2023). Jet fire. Phast theory Manual.
DNV. (2023). Pool fire model.
DNV. (2023). Validation: Jet fire.
DNV. (2023). Validation: Unified Dispersion Model.
Fishburne, E. a. (1979). The dynamics and radiant intensity of large hydrogen flame. Colloquium on Fire and Explosion.
Johnson A.D., B. H. (1994). A model for predicting the thermal radiation hazard from large scale horizontally released
natural gas jet fires. Trans IChemE, Vol 72, Part B, pp 157-166.
Johnson, A. (1992). A model for predicting thermal radiation hazards from large-scale LNG pool fires. IChemE Symp.,
Series 130, pp 507-524.
Nielsen, M. O. (1996). FLADIS field experiments. Roskilde, Denmark: Final report Tiso-R-898(EN), Riso National
Laboratory.
Roberts, P. S. (2006). Dispersion of hydrogen from high-pressure sources. Institution of Chemical Engineers
Symposium Series 151, (pp. 410-421).

DNV – www.dnv.com Page 43