VALIDATION

DISC MODEL
DATE: December 2023
ABSTRACT
This report describes the validation of the Phast discharge model DISC.

The DISC theory is described by an accompanying report. The DISC model includes four sub-models covering the following scenarios:
an orifice leak from a vessel, a leak from a short pipe attached to a vessel, catastrophic rupture of a vessel and vent from the vapour
space of a vessel.

RELEASE NOTES
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

1 2009 PHAST 6.53
2 Oct 2017 Safeti 8.0 Stene

3 May 2021 Apply new template D. Vatier

4 July 2022 Version update David Worthington

Date: December 2023

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.

RELEASE NOTES
Table of contents
ABSTRACT ................................................................................................................................................................. 2

1 INTRODUCTION ......................................................................................................................................... 1

2 LITERATURE SURVEY............................................................................................................................... 2
2.1 Previous validation of the DISC discharge model 2
2.2 DIERS documentation 7
2.3 CCPS publications 7
2.4 Lees 8
2.5 Other literature 10

3 VALIDATION OF DISC ORIFICE AND SHORT PIPELINE MODELS .......................................................... 11

3.1 Introduction 11
3.2 Water pipe and orifice releases 11
3.3 Propane orifice and pipe releases 14
3.4 Butane orifice releases 15
3.5 Further validation 16
3.6 Figures with validation results 17
3.7 Conclusions and recommendations 28

REFERENCES .......................................................................................................................................................... 30

| Validation | DISC Model | Page i

1 INTRODUCTION
A full description of the theory underlying DISC is described in the accompanying DISC theory manual. The DISC verification
manual describes the verification of the model, whereas this report is concerned with the validation of the DISC model. To this
end discharge predictions are compared with measurements from a selection of the available experimental data.

The validation work presented in Chapters 2 and 3 was carried out as part of Phase III of the droplet-modelling joint
industry project. As such it was originally documented in the first part of the Phase III report, C1, and it is now included
here as a starting point of the DISC validation manual.

More specifically, Chapter 2 provides a review of previous DISC validation work as well as a literature overview of relevant
available experimental results for further DISC validation. Actual validation results are then presented in Chapter 3 and include
water, propane and butane releases.

| Validation | DISC Model | Page 1

2 LITERATURE SURVEY

2.1 Previous validation of the DISC discharge model

2.1.1 DISC validation against sub-cooled and saturated water experiments
John Woodward provides an overview of discharge modelling in Perry’s handbook1.
Figure 1 (Fig. 26.-63,26-64 in Perry) includes mass flux data for water (Uchida and Nariai, 1966) from an orifice and pipe
(orifice/pipe diameter of 4 mm) with a range of initial pressure P st and pipe length L. Both sub-cooled water (20C) and
saturated water is considered. He also includes results from HEM model by Leung (this would be verification against our
models).

Figure 2 (Figs. 26-70 and 26.71) shows predictions of an analytical HEM by Leung and Epstein (1990) and the NEM model
by Henry and Fauske (1976) against data by Sozzi and Sutherland (1975) for slightly subcooled water (Fig. 26-70) and
saturated water (Fig. 26-71).
Figure 3 includes PHAST4.2 predictions2 against these experiments.

In the Modelling Conference presentation he also quotes plans for PHAST discharge model validation against benchmarks
against a large range of DIERS validation tests (but these never materialised). They include both a large range of water
tests (liquid water downward, saturated/flashing water, low/high viscous flashing flow), i.e.

• DIERS Round robin validation tests:

o Benchmark 1- Water flow downward

o Benchmark 2- Saturated water vessel drain
o Benchmark 3- Sat. water blowdown overhead
o Benchmark 4- Flashing water blowdown from bottom-mounted pipes
o Benchmark 5- Flashing water nozzle flow
o Benchmark 6- Nielsen flashing water case
o Benchmark 7- Viscous water flow (same as Test 2 except 5000 cp viscosity)
o Benchmarks 8 & 9- Same as above with T-dependant viscosity
o ISPRA tests with DRACULA system

• DIERS Luviskol tests

• Case 1 - Vapour flow

• Case 2 - Low viscosity flashing flow
• Case 3 - Low viscosity flashing w/ 5 psi N2
• Case 4 - High viscosity flashing flow
• Case 5 - High viscosity flashing w/ 5 psi N2
• Case 6-7 - Churn-Turbulent & Homogeneous disengagement

| Validation | DISC Model | Page 2

Figure 1. Experimental data for sub-cooled/saturated water data with varying L and Pst

| Validation | DISC Model | Page 3

Figure 2. HEM/NEM model validation against sub-cooled/saturated water data with varying L

| Validation | DISC Model | Page 4

Figure 3. PHAST4.2 validation against sub-cooled/saturated water data with varying L

Figure 1 shows PHAST4.2 over-prediction of orifice flow 40% for flashing water, and under-predicts
orifice flow for slightly sub-cooled water by about 50%. As the pipe length increases the error reduces
to near zero, with an over-prediction of about 10% for longer pipe discharge rates. These data seem to
imply that about 1 meter is needed to reach HEM (rather than 0.1m quoted in the literature).

| Validation | DISC Model | Page 5

2.1.2 DISC validation against hydrocarbon experiments
Duree3 (Shell Research) carried out PHAST 4.2 validation against experiments of by Shell and Imperial College. This
included:

- full-bore and orifice releases of liquid propane – Shell experiments (Fig. 2.1)
- full-bore and orifice releases of liquid butane – Shell experiments (Fig. 2.2)

Note there is overall very good agreement.

Figure 4. PHAST 4.2 validation against two-phase propane full-bore and orifice tests

Figure 5. PHAST4.2 validation against two-phase butane orifice tests

| Validation | DISC Model | Page 6

2.2 DIERS documentationi
Section 3.2 in DIERS project manual4 includes comparison of various nozzle models (e.g. HEM model, frozen liquid); 10
cm is stated for reaching equilibrium (in DIERS project manual referring to Fauske – 1983). It is also stated that non-
equilibrium effects are of little concern above 10% upstream quality (i.e. less than 90% storage liquid fraction). Also
included is a discussion on discharge coefficients. Recommendation is close to 1 for fully turbulent flow, which is
considerably larger than liquid flow (e.g. 0.61 for sharp-edged orifice). The HEM model is stated to be conservative for
pipe flow.

2.3 CCPS publications

CCPS QRA guidelines5 recommend that the design philosophy of the DIERS models is to select the minimum discharge
rate at the design pressure, and to maximise the relief area via a selection of a minimum mass flux model. It is noted that
use of these models to represent source term models, will under-predict the discharge rate and therefore they are un-
conservative. It also includes reference to a model of Fauske and Epstein (which is similar to the logic in the DISC model)
and compares this to experimental dataii.

Figure 6. Experimental data and model predictions for mass flux with varying P st and D

This publication also contains worked-out examplesiii and an overview of available computer codes. Also the CCPS book
on flammable-mass estimation by John Woodward6 includes worked-out examplesiv. Furthermore it includes validation
data (including model comparisons) for flashing discharge (Figure 4.8, and Figure 4.9 – Example 6)v.

i
The DIERS website includes a detailed list of references relating to two-phase releases from orifice and pipes, etc., and some of these references could be of
interest.
ii
Future. Consider adding validation against this experiment?
iii
Future. Consider DISC verification against example 2.1 for liquid-liquid discharge and example 2.4 for a gas-gas discharge.
iv
Future. Consider DISC verification in Section 1 (examples 1,2 for liquid and gas discharges, example 4 for liquid to two-phase discharge, and example 5 for
horizontal pipe.
v
Future. Consider validation for liquid to two-phase experiments corresponding to Figs. 4.8, 4.9.

| Validation | DISC Model | Page 7

Figure 7. ERM/HEM model validation for mass flux with varying Pst and L/D

Figure 8. ERM model validation for water with varying sub-cooling (L=0.2, 2m)

2.4 Lees

| Validation | DISC Model | Page 8

Lees 7 provides a detailed overview of a range of discharge models. This includes analytical derivations for discharge
coefficients (Section 15.1.4,). Table 15-3 in Section 15.2.1 includes an overview of experimental studies, largely referring to
two-phase water experiments,

Note that in Section 15.2.19 he states that the distance over which equilibrium is established is an absolute distance (Lees
mentions 75 mm, in line with 10 cm mentioned above), and not (as initially thought) a function of L/D ratio. Model
comparison as function of pipe length L against data by Fletcher and Johnson (1984) are given by Figure 15.13.

| Validation | DISC Model | Page 9

Figure 9. Refrigerant-11 and saturated water data with varying L (correlations for discharge)

2.5 Other literature

Giot8 provides results for choked data against experimental data (by Flinta, 1984 and Leung, 1987) for liquid to two-phase
releases. He also includes data from Sozzi and Sutherland. He discusses both orifice and line ruptures. He includes
several more interesting experimental data.

| Validation | DISC Model | Page 10

3 VALIDATION OF DISC ORIFICE AND SHORT PIPELINE MODELS

3.1 Introduction
This chapter describes the validation of the short pipeline rupture and orifice discharge models included in version 6.53 of
PHAST. It includes the validation of the DISC discharge models against data for sub-cooled and saturated releases of
water jets and against data for sub-cooled releases of propane and butane.

3.2 Water pipe and orifice releases

This section reports validation of the DISC discharge model against data reported by Uchida and Nariai (1966)9 (See
Figure 1) and Sozzi and Sutherland (1975)10. These data sets relate to sub-cooled and saturated releases of water.

In the Uchida and Nariai tests, sub-cooled and saturated water at stagnation pressures ranging from 2-8 bara were
released through .004m (ID) copper pipes. The pipes were of varying lengths ranging from 0 (i.e. orifice) to 2.5m.
Discharge rate data for each release scenario was logged, but no information was provided on the surface roughness of
the copper pipes. As such, for the DISC simulations, a value of 3x10-6m has been assumed6. Table 1 presents a summary
of key input data for each test case7.

Description Data
Orifice diameter (m) .004
Pipeline internal diameter (m) .004
Pipe roughness (m) 8 3x10-6
Pipe length (m) 0 – 2.5
Stagnation temperature (sub-cooled releases) (oC) 20
Stagnation pressure(s) (saturated/sub-cooled) (bara) 2, 3, 4, 5, 6, 7, 8

Table 1 Summary of key input data employed in simulating data reported by Uchida and Nariai (1966).

Sozzi and Sutherland studied the effect of fluid state (i.e. two-phase or sub-cooled liquid), flow geometry, aperture size
and flow length on release rate measurements involving water. Of the seven release scenarios studied, only data related
to the first two have been selected for this validation exercise9. The first scenario (i.e. Scenario 1) involves the release of
pressurised two-phase/sub-cooled water from a tank through a .0127m orifice with a well-rounded inlet. The second
scenario (i.e. Scenario 2) is concerned with the release of two-phase/sub-cooled water through .0127m (ID) pipes of
varying lengths ranging from 0 - 1.78m. These pipes, as in Scenario 1, possess well rounded inlets and were attached to
the same tank. No data was presented for the characteristic roughness of these pipes, hence, for the DISC simulations,
a value of 1.5x10-6m has been assumed. Table 2 presents a summary of key input data for each test case.

Description Data
Orifice diameter [Scenario 1] (m) .0127
Pipeline internal diameter [Scenario 2] (m) .0127
Pipe roughness (m) 10 1.5x10-6
Pipe length (m) 0 – 1.78
Stagnation temperature (sub-cooled releases) (oC) Various
Stagnation pressure(s) (saturated/sub-cooled) (bara) Various

Table 2 Summary of key input data employed in simulating data reported by Sozzi and Sutherland (1975)

DISC results for water pipe releases (Uchida and Nariai, 1966)

Figure 10 to Figure 16 show the variation of simulated and observed discharged mass flux with pipe length for sub-cooled
water initially at stagnation pressures of 8, 7, 6, 5, 4, 3 and 2 bara respectively. Each figure compares simulated results

6
The adopted surface roughness is twice the recommended/typical value for drawn tubing but less than the value for commercial carbon steel pipes (see Table 6-1,
pp 6-10, Perry’s Chemical Engineers Handbook, 7th edition). Due to the relatively small aperture sizes of the pipes employed in the study, the pipe surface
roughness is expected to approach the recommended value for drawn tubing.
7
Further details related to specific input can be found in the results processing spreadsheet DISC_Validation_Summary_of_Simulated_Results.xls
8
Assumed
9
These scenarios represent simple flow geometries which typify releases through orifices and rupture of short pipelines as is currently modelled by PHAST.
10
Assumed
| RELEASE NOTES | Product version XX |
using PHAST 6.5311 and 4.2 with experimental data logged by Uchida and Nariai (1966). From these figures, within limits
of uncertainty in measurements12, the following observations can be made:

• Simulated discharge rate results obtained using PHAST6.53 and 4.2 are marginally conservative but compare
very well with experimental data. The maximum and average absolute deviations of simulated results using
PHAST6.53 from experimental data are ca 20% and 7% respectively.
• PHAST6.53 generally overestimates the discharge rate for near zero-flow-length (i.e. “orifice-like”) releases by
ca 13.9%. This is likely as a result of fluid property calculation errors identified during the verification exercise
(see the DISC verification document).

Figure 17 to Figure 23 show the variation of simulated and observed discharged mass flux with pipe length for saturated
water initially at stagnation pressures of 8, 7, 6, 5, 4, 3 and 2 bara respectively. Each figure compares simulated results
using PHAST 6.5311 and 4.2 with experimental data logged by Uchida and Nariai (1966). From these figures, the following
observations can be made:

• With the exception of near zero-flow-length (i.e. pipe length ≤ 0.01m) discharge rate results, simulated data using
PHAST6.53 and 4.2 are in close agreement. Observed discrepancies in simulated results between both versions
have been traced to differences in adopted modelling assumptions. In PHAST4.2, it appears that near zero-flow-
length releases have been modelled as meta-stable releases across an orifice. This constraint has been relaxed
in PHAST6.53.13
• Discharge rate results obtained from both versions of PHAST compare less good with experimental data and are
generally under conservative. However, the slope of the discharge rate versus pipe length profiles for the
simulated and experimental data are very similar. This suggests the presence of a constant source of error in the
simulated as compared with the measured discharge rates for two-phase flashing releases.14 The maximum and
average absolute deviations of simulated results using PHAST6.53 from experimental data are ca 82% and 54%
respectively.

DISC results for water pipe releases (Sozzi and Sutherland, 1975)

Figure 24 shows the variation of Gpred/Gobs (i.e. the ratio of simulated to measured discharge rate data) with pipe length
for sub-cooled15 water based on Sozzi and Sutherland’s experimental data. Unlike the Uchida and Nariai tests, in most
cases, the sub-cooled water in the Sozzi and Sutherland tests is observed to flash along the length of the pipeline.
Simulated results obtained using PHAST4.2 and 6.5311 are compared. From Figure 24, within limits of uncertainty in
measurements, 12 the following observations can be made:

• Simulated discharge rates using PHAST4.2 and 6.53 generally approach experimental data (i.e. Gpred/Gobs ≈
1) with increasing pipe length. These results are unlike the discharge rate results for saturated water in the Uchida
and Nariai tests: here, a constant offset between simulated and measured data was observed.
• PHAST4.2 generally underestimates release rate data for pipe lengths ≤ 0.5m.
• However, PHAST6.53 generally predicts conservative release rates when compared with measured data.
Release rate data are generally underestimated in PHAST6.53 when flashing is predicted to occur within pipe
lengths ≤ 0.2m.
• The above observations suggest for flashing flows that the shorter the flow/pipe length, the less valid is the use
of the homogenous equilibrium modelling assumption.
• On the whole, simulated results using PHAST4.2 and 6.53 can be said to compare well with Sozzi and
Sutherland’s data. The respective maximum/average percentage deviations of simulated results using
PHAST4.2 and 6.53 from experimental data are ca 52%/20% and 74%/17%.

Figure 25 shows the variation of Gpred/Gobs with pipe length for saturated/two-phase16 water based on Sozzi and
Sutherland’s experimental data. Simulated results obtained using PHAST4.2 and 6.53 are compared. From Figure 25, the
following observations can be made:

11
“Capping-disallow flashing” flow option. This implies that simulated release rates from the pipeline model are subject to a maximum value. The adopted maximum
value corresponds to the simulated release rate from an “orifice” with the same diameter as the pipeline based on the meta-stable flow assumption.
12
No value for this has been quoted by the authors
13
It appears relaxing this constraint in PHAST6.53 results in poorer performance of the discharge model for near zero-flow-length release scenarios.
14
Thiscouldbeduetoamodellinglimitation/error,thermodynamicpropertycalculationerrororinstrumentation(i.e.experimental)error.Itwouldbeexpectedthatas
flow length increases, simulated results based on PHAST’s homogenous-equilibrium
discharge model should approach experimental data (see Figure 2,
Figure 3 and Figure 7). While this appears not to be the case with the Uchida and Nariai data, the expected trend is generally observed with the Sozzi and Sutherland
data.
15
Stagnation condition
16
Stagnation condition
| Validation | DISC Model | Page 12
 • As previously observed, simulated discharge rates using PHAST4.2 and 6.53 generally approach experimental
data (i.e. Gpred/Gobs ≈ 1) with increasing pipe length. These results are unlike the discharge rate results for
saturated water in the Uchida and Nariai tests.
• While PHAST4.2 is seen to overestimate, PHAST6.53 generally underestimates the discharge rate for near zero-
flow-length (i.e. pipe length ≤ 0.01m) releases. The observed behaviour in PHAST4.2 is due to the modelling of
near zero-flow-length releases as meta-stable releases across an orifice.
• On the whole, simulated results using PHAST4.2 and 6.53 can be said to compare well with Sozzi and
Sutherland’s data. The respective maximum/average percentage deviations of simulated results using
PHAST4.2 and 6.53 from experimental data are ca 53%/13% and 60%/11%.

DISC results for all sub-cooled water pipe releases

Figure 26 shows the variation of Gpred/Gobs with pipe length for all the sub-cooled water pipe releases by Sozzi and
Sutherland (1975) and Uchida and Nariai (1966). The predicted flow rates are based on Phast 8.0.

DISC results for water orifice releases (Uchida and Nariai, 1966; Sozzi and Sutherland, 1975)

Figure 27 compares the PHAST6.53 predicted mass flux for discharge across an orifice for sub-cooled 17 water to
measured data reported by Uchida and Nariai (1966) and Sozzi and Sutherland (1975). Simulated results with the “prevent
flashing across the orifice” option enabled and disabled [i.e. “PHAST6.53 (Flashing)” and “PHAST6.53 (old method-No
Flashing)”] in addition to the “PHAST6.53 (Bernoulli)” modelling option are presented. The simulated scenarios were
extended to include all cases involving very short pipes (i.e. pipe length ≤ 0.1m) for which it is assumed that the overall
effect of pipe-wall friction on discharge rate is negligible. Discharge rates corresponding to 30 and 50% from measured
data are also presented. From Figure 27, the following observations can be made:

• Where flashing is predicted to occur, simulated discharge flux results with “Flashing” enabled generally
underestimate measured data, while the converse is observed for the same cases with the “No Flashing” option
enabled.18
• Both the “Flashing” and “No Flashing” option yield the same results for purely sub-cooled releases (i.e. the Uchida
and Nariai tests). For these, simulated discharge flux results slightly overestimate measured data.
• When compared with measured data, accuracy in simulated results using the Bernoulli model is observed to
increase with decrease in stagnation pressures19.
• For stagnation pressures below 57barg, the Bernoulli model is seen to be generally conservative and to predict
release rates to within ±10% of measured data. The converse behaviour is generally observed above ca 57barg.
Interestingly, above this pressure (i.e. 57barg), simulated results using the “No Flashing” option is observed to
perform better when compared against measured data. It is suspected that at higher pressures (e.g. above ca
57barg for water), the assumption of liquid incompressibility is less valid. Thus when compared with measured
data in this pressure range, the Bernoulli model is observed to diminish in accuracy while the converse is
observed for the “No Flashing” model20.
• In all, simulated results using the “Bernoulli”, “Flashing” and “No Flashing” options generally lie within 50% of
measurements. In comparison with measured data, the “Bernoulli” model performs best on average followed by
the “No Flashing” and lastly the “Flashing” models. The respective maximum/average percentage deviations of
simulated results using the “Bernoulli”, “Flashing” and “No Flashing” options from experimental data are ca
33%/11%, 67%/30% and 61%/18%.

Figure 28 compares PHAST6.53’s predicted mass flux for discharge across an orifice of saturated/two-phase17 water to
measured data reported by Uchida and Nariai (1966) and Sozzi and Sutherland (1975). As in Figure 27, simulated results
with the “Flashing” option enabled and disabled in addition to the “PHAST6.53 (Bernoulli)” 21 modelling option are
presented. Simulated scenarios were extended to include all cases involving short pipes (i.e. ≤ 0.1m). Discharge rates
corresponding to 30 and 50% from measured data are also presented. From Figure 28, the following observations can
be made:

• Where the stored fluid at stagnation point exists as a two-phase mixture, both the “Flashing” and “No Flashing”
option yield the same under-conservative results. 22 The “Bernoulli” model, for these cases, yields over-
conservative release rate estimates with inaccuracies tending to increase with increase in upstream/stagnation
pressure and/or vapour quality.
• For 100% liquid saturated water, 17 simulated discharge flux results with the “Flashing” option enabled generally
underestimate measured data by more than 50%. With the “No Flashing” and “Bernoulli” options, a converse,

17
Stagnation condition
18
The “No Flashing” option is likely to be more prone to errors related to inaccuracies in simulated liquid enthalpies using PHAST’s fluid property system. These
errors may overstate the observed differences.
19
This observation is in line with the basic assumption on which the Bernoulli model is built. The assumption of liquid incompressibility tends to approach reality at
lower stagnation pressures.
20
This model attempts to model the variation of liquid compressibility with pressure using the cubic equation of state (CEOS).
21
For the “Bernoulli” modelling option, all two-phase fluids are assumed to possess a liquid fraction of 1 (i.e. pure saturated liquid). Thus all non-zero vapour quality
data recorded in the experiments are ignored and set by default to 0 for the Bernoulli based results. This is known as the forced-phase flow assumption.
22
For these cases, the “No Flashing” option would ideally be expected to collapse to the “frozen” fluid assumption. This however appears not to be the case.
| Validation | DISC Model | Page 13
 but less dramatic behaviour, is generally observed for the same cases 18. Overestimates generally lie within 30%
of measured data, while the “Bernoulli” option is observed to provide more accurate estimates of measured data.
• On the whole, the respective maximum/average percentage deviations of simulated results using the “Bernoulli”,
“Flashing” and “No Flashing” options from experimental data are ca 110%/50%, 92%/42% and 63%/37%.

3.3 Propane orifice and pipe releases

This section reports validation of the DISC discharge model against data reported by Bennett et al. 11 (1991) (see
Figure 4), which relates to sub-cooled releases of liquid propane.

Bennett et al. (1991) report discharge rate data for 19 tests involving the release of pressurised LPG. Nine tests relate to
releases through .010m and .020m orifices, while the rest are concerned with releases following the full-bore-rupture (FBR)
of 2.58m and 11.63m long, .052m (ID) carbon-steel pipes. The authors, for these tests, quote a value of 0.0015m 23 for
pipe roughness, while no information was provided for the fluid’s stagnation pressure (Pstag) or temperature (Tstag).
Nevertheless, based on available data, it is assumed that the ambient (Tamb) and stagnation temperatures are equal. The
same assumption is applied to the drive (upstream) [Pdrive] and stagnation pressures. Table 3 presents a summary of
pertinent input data and measured data for each test case, while Table 4 reports the measured composition (mole %) for
the LPG mixture employed in the tests. For the DISC simulation of each release scenario, the LPG mixture is assumed to
be composed of 100% propane.

Test Name Release Pipe Orifice/ Pdrive/ Tstag Ambient Tamb Relative Measured
Type Length Pipe Pstag Pressure Humidity Flowrate
Diameter [Pamb] [Gobs]
(m) (m) (barg) (K) (mmHg) (K) (fraction) (kg/s)
Test_3007 Leak 0 0.01 9.7 288.85 744.5 288.85 0.69 1.8
Test_3006 Leak 0 0.01 9.7 286.35 746.3 286.35 0.79 1.5
Test_3028 Leak 0 0.02 7.7 278.45 721.3 278.45 0.92 5.7
Test_3047 Leak 0 0.02 7 285.05 737.2 285.05 0.77 4.7
Test_3048 Leak 0 0.02 7.1 283.95 736.5 283.95 0.87 4.9
Test_3039 Leak 0 0.01 7 279.15 750 279.15 0.58 1.5
Test_3080 Leak 0 0.02 7.6 288.55 75024 288.55 0.85 3.4
24
Test_3012 Leak 0 0.02 9.5 286.55 750 286.55 0.9 5.8
Test_3090 Leak 0 0.02 7.6 282.85 75024 282.85 0.88 3.9
24
Test_3026 FBR 2.58 0.052 6.5 286.85 750 286.85 0.59 16.1
Test_3029 FBR 2.58 0.052 6.3 281.15 75024 281.15 0.82 18
Test_3019 FBR 2.58 0.052 7.1 285.05 730.2 285.05 0.85 19.4
Test_3063 FBR 11.63 0.052 7.4 279.75 730 279.75 0.86 11.6
Test_3064 FBR 11.63 0.052 7.4 278.55 722 278.55 0.85 11.5
Test_3084 FBR 11.63 0.052 7.1 285.35 741.3 285.35 0.96 10.4
Test_3020 FBR 2.58 0.052 6.9 283.75 740 283.75 0.9 15.3
Test_3087 FBR 11.63 0.052 6.5 286.65 742.4 286.65 0.92 5.5
24
Test_3054 FBR 11.63 0.052 8 284.75 750 284.75 0.48 11.8
Test_3077 FBR 11.63 0.052 7.1 286.35 75024 286.35 0.92 10.1

Table 3 Summary of measured data for discharge experiments involving the release of pressurised
propane reported by Bennett et al. (1991).

Component Mole %
Methane 0.0
Ethane 0.2
Propane + 97.4
Iso-butane 1.6
N-butane 0.8

23
This value appears to be rather high for carbon-steel pipes. Perry’s Chemical Engineers handbook quotes a value of 0.0000457m as typical surface roughness for
commercial steel pipes. The higher a pipe’s surface roughness, the greater its wall resistance to flow (i.e. exit flow-rate decreases with increasing surface
roughness).
24
Assumed
| Validation | DISC Model | Page 14
 Table 4 Typical LPG composition for the Bennett et al. (1991) tests

DISC results for propane orifice releases

Figure 29 compares the PHAST6.53 predicted flow-rate for sub-cooled 17 propane flowing across an orifice to measured
data reported by Bennett et al. (1991). Simulated results with the “Flashing” option enabled and disabled in addition to the
“PHAST6.53 (Bernoulli)” modelling option are presented. Discharge rates corresponding to 30 and 50% from measured
data are also presented. From Figure 29, within limits of uncertainty in measurements, 12 the following observations can
be made:

• As previously observed, simulated discharge flux results with “Flashing” enabled generally underestimate
measured data, while predictions based on the “Bernoulli” and “No Flashing” options tend to be generally
conservative.
• In all, simulated results using the “Bernoulli”, “Flashing” and “No Flashing” option generally lie within 30% of
measurements. In comparison with measured data, the “Bernoulli” model performs best on average followed by
the “No Flashing” and lastly the “Flashing” models. 18 The respective maximum/average percentage deviations
of simulated results using the “Bernoulli”, “Flashing” and “No Flashing” options from experimental data are ca
54%/16%, 32%/19% and 60%/18%.25

DISC results for propane pipe releases

Figure 30 compares the PHAST6.53 predicted flow-rate for steady-state discharge following FBR of 2.68 and 11.63m
pipelines conveying sub-cooled17 propane to measured data reported by Bennett et al. (1991). Discharge rates
corresponding to 30 and 50% from measured data are also presented. From Figure 30, the following observations can
be made:

• Simulated discharge rate results obtained from PHAST6.53 generally underestimate26 but largely compare well
with measured data. 90% of these results lie within 30% of measurements.
• The respective maximum and average percentage deviations of PHAST6.53 simulated results from experimental
data are ca 36% and 19%.

3.4 Butane orifice releases

This section reports validation of the DISC discharge model against data reported by Duree et al. (1995) for sub-cooled
releases of liquid butane from 0.01m orifices.

Although the original report contained information on 6 tests, it has only been possible to source sufficient data to simulate
two of these tests (i.e. Test_5045 and Test_5035).27 Table 5 presents a summary of pertinent input data and measured
data for tests 5045 and 5035.

Test number 5045 5035

Reference report TNER.92.038 TNER.92.012
Fuel28 100% butane 100% butane
Stagnation pressure (barg) 7.7 (same as pressure 4.2 (different to pressure
immediately upstream of orifice) immediately upstream of the orifice)
Stagnation temperature29 (oC) 16 ± 1 13.3
Orifice diameter (m) 0.01 0.01
Ambient pressure (bar) 0.990 0.979
Ambient temperature (oC) 16 ± 1 14
Ambient humidity (%) 46.5 76
Flow rate (kg/s) 1.7 ± 0.2 1.1 (<10% accuracy)

25
There is cause to doubt the authenticity of two reported data points lying between Gobs = 3 and 4 kg/s. While simulated results based on the “Flashing” option always
underestimate measured data, for these two data points the converse is observed. This deviation from expected trend points to the likelihood of data
logging/instrumentation error. If the “Flashing” option is established as always yielding underestimates of measured data, the logged data for these two data
points are expected to be higher than reported. Coincidentally, the maximum percentage deviation for simulated results using the “No Flashing” option is observed
to result from a comparison with one of these data points. If these points are excluded from the analysed data set, the respective maximum/average percentage
deviations of simulated results using the “Bernoulli”, “Flashing” and “No Flashing” options from experimental data become ca 18.0%/7.6%, 31.9%/23% and
14.6%/8.1%.
26
The observed underestimates may stem from the use of the reported, but likely exaggerated, wall surface roughness.
27
These data were obtained following personal communication with Chamberlain G. (Shell) 30-05-2007.
28
100% butane is assumed; the actual composition is close enough to 100% in order not to significantly affect the results (the precise composition is included in
TNER.92.054)
29
For test 5045 a temperature of 24.6C is mentioned in report, but discarded because of fire effects. For test 5035 there are also fire effects resulting in heating of the
pipe.
| Validation | DISC Model | Page 15
 Table 5 Summary of measured data for discharge experiments involving the release of pressurised n-
Butane reported by Duree et al. (1995)27

Table 6 compares the PHAST6.53 predicted mass flux for discharge across an orifice for sub-cooled n-Butane to
measured data reported by Duree et al. (1995). Simulated results with the “Flashing” option enabled and disabled in
addition to the “PHAST6.53 (Bernoulli)” modelling option are presented.

Predicted Flow-rate % deviation from Measured data

Test Measured
Number Flow-rate
“PHAST6.53 (old
“PHAST6.53 “PHAST6.53 “PHAST6.53 “PHAST6.53 (old “PHAST6.53
method-No
(Bernoulli)” (Flashing)” (Bernoulli)” method-No Flashing)” (Flashing)”
Flashing)”
kg/s kg/s kg/s kg/s % % %
5045 1.7 1.412988 1.465898 1.386298 -16.883044 -13.770687 -18.453055
5035 1.1 1.046337 1.085385 1.010692 -4.878496 -1.328681 -8.118909

Table 6 Comparison of observed versus PHAST6.53 predicted discharge rate data for release across an
orifice involving sub-cooled n-butane at orifice inlet (Data from Duree et al., 1995, SHELL TNER
92.038 and 92.012)

From Table 6 the following observations can be made:

• Simulated results from all PHAST6.53 discharge models compare very well with measured data. Each model,
for both test cases, marginally underestimates measured data.
• As observed previously, the “No Flashing” model predicts the highest release rates followed by the “Bernoulli”
and lastly the “Flashing” model.

In all, simulated results using the “Bernoulli”, “Flashing” and “No Flashing” option generally lie within -30% of
measurements. In comparison with measured data, the “No Flashing” model performs best 30 on average followed by the
“Bernoulli” and lastly the “Flashing” models. 18 The respective maximum/average percentage deviations of simulated
results using the “Bernoulli”, “Flashing” and “No Flashing” options from experimental data are ca 17%/11%, 18%/13% and
14%/8%.

3.5 Further validation

The DISC model has also been validated in the context of validating other linked models in Phast. This validation work
can be found in Chapter 6 of the documentation for the atmospheric expansion (ATEX) model12 and in Chapter 4 of the
validation document for the unified dispersion model13 (UDM).

30
The choice as to the best model for simulating these data is quite subjective. Within limits of uncertainty either the “Bernoulli” or the “No Flashing” model may be
selected as being best suited. For conservative results, the “No Flashing” model is recommended.
| Validation | DISC Model | Page 16
3.6 Figures with validation results

30000

25000
Experimental data
PHAST6.53
PHAST4.2
20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 10 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 8bara] (Data by Uchida
and Nariai, 1966)
25000

Experimental data
PHAST6.53
PHAST4.2
20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 11 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 7bara] (Data by Uchida
and Nariai, 1966)

| Validation | DISC Model | Page 17

 25000

Experimental data
PHAST6.53
PHAST4.2
20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 12 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 6bara] (Data by Uchida
and Nariai, 1966)
25000
Experimental data
PHAST6.53
PHAST4.2

20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 13 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 5bara] (Data by Uchida
and Nariai, 1966)

| Validation | DISC Model | Page 18

 18000

Experimental data
PHAST6.53
16000
PHAST4.2

14000

12000
Mass Flux (kg/m2/s)

10000

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 14 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 4bara] (Data by Uchida
and Nariai, 1966)
16000

14000
Experimental data
PHAST6.53
PHAST4.2
12000

10000
Mass Flux (kg/m2/s)

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 15 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 3bara] (Data by Uchida
and Nariai, 1966)

| Validation | DISC Model | Page 19

 12000
Experimental data
PHAST6.53
PHAST4.2
10000

8000
Mass Flux (kg/m2/s)

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 16 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for sub-cooled water [stagnation pressure = 2bara] (Data by Uchida
and Nariai, 1966)
25000

Experimental data
PHAST6.53
PHAST4.2
20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 17 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 8bara] (Data by Uchida
and Nariai, 1966)

| Validation | DISC Model | Page 20

 25000

Experimental data
PHAST6.53
PHAST4.2
20000
Mass Flux (kg/m2/s)

15000

10000

5000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 18 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 7bara] (Data by Uchida and
Nariai, 1966)
20000

18000 Experimental data

PHAST6.53
PHAST4.2
16000

14000
Mass Flux (kg/m2/s)

12000

10000

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 19 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 6bara] (Data by Uchida and
Nariai, 1966)

| Validation | DISC Model | Page 21

 18000
Experimental data
PHAST6.53
16000 PHAST4.2

14000

12000
Mass Flux (kg/m2/s)

10000

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 20 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 5bara] (Data by Uchida and
Nariai, 1966)
16000

14000
Experimental data
PHAST6.53
PHAST4.2
12000

10000
Mass Flux (kg/m2/s

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 21 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 4bara] (Data by Uchida and
Nariai, 1966)

| Validation | DISC Model | Page 22

 14000

12000
Experimental data
PHAST6.53
PHAST4.2
10000
Mass Flux (kg/m2/s

8000

6000

4000

2000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 22 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 3bara] (Data by Uchida and
Nariai, 1966)
9000

8000 Experimental data

PHAST6.53
PHAST4.2
7000

6000
Mass Flux (kg/m2/s

5000

4000

3000

2000

1000

0
0 0.5 1 1.5 2 2.5 3
Pipe Length (m)

Figure 23 Performance of PHAST4.2 and 6.53 simulated discharged mass flux versus pipe length data
against experimental data for saturated water [stagnation pressure = 2bara] (Data by Uchida and
Nariai, 1966)

| Validation | DISC Model | Page 23

 1.8
PHAST4.2-Sozzi&Sutherland

1.6
PHAST6.53-Sozzi&Sutherland

1.4

1.2
Gpred/Gobs (-)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5
Pipe Length (m)

Figure 24 Performance of PHAST4.2 and 6.53 discharge rate versus pipe length predictions against
experimental data for slightly sub-cooled water (Data by Sozzi and Sutherland, 1975)31
1.8

PHAST4.2-Sozzi&Sutherland
1.6

1.4 PHAST6.53-Sozzi&Sutherland

1.2
Gpred/Gobs (-)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5
Pipe Length (m)

Figure 25 Performance of PHAST4.2 and 6.53 discharge rate versus pipe length predictions against
experimental data for saturated/flashing water (Data by Sozzi and Sutherland, 1975)31

| Validation | DISC Model | Page 24

 1.8
PHAST8.0-Sozzi&Sutherland
1.6
Ratio of predicted to observed flowrate (-)

PHAST8.0-Uchida&Nariai
1.4

1.2

1.0

0.8

0.6

0.4

0.2

0.0
0 0.5 1 1.5 2 2.5
Pipe length (m)
Figure 26 Performance of Phat 8.0: discharge rate versus pipe length predictions against experimental data for
slightly sub-cooled water (Data by Sozzi and Sutherland (1975) and Uchida and Nariai (1966).

80000 Experimental data

PHAST6.53(Flashing)
PHAST6.53(old method-No Flashing)
70000 PHAST6.53(Bernoulli)
+30%-deviation
-30%-deviation
60000
+50%-deviation
-50%-deviation
50000
Gpred (kg/m2/s

40000

30000

20000

10000

0
0 10000 20000 30000 40000 50000 60000 70000 80000 90000
Gobs (kg/m2/s)

Figure 27 Observed versus PHAST6.53 predicted discharge mass flux across an orifice for inlet sub-
cooled water (Data by Uchida and Nariai, 1966; Sozzi and Sutherland, 1975)

| Validation | DISC Model | Page 25

 80000 Experimental data

PHAST6.53(Flashing)
70000 PHAST6.53(old method-No Flashing)

PHAST6.53(Bernoulli)
60000 +30%-deviation

-30%-deviation
50000 +50%-deviation
Gpred (kg/m2/s

-50%-deviation
40000

30000

20000

10000

0
0 10000 20000 30000 40000 50000 60000 70000
Gobs (kg/m2/s)

Figure 28 Observed versus PHAST6.53 predicted discharge mass flux across an orifice for inlet
saturated/two-phase water (Data by Uchida and Nariai, 1966; Sozzi and Sutherland, 1975)
10
Experimental data
PHAST6.53(Flashing)
9
PHAST6.53(old method-No Flashing)
8 PHAST6.53(Bernoulli)
+30%-deviation
7
-30%-deviation
6 +50%-deviation
Gpred (kg/s)

-50%-deviation
5

0
0 1 2 3 4 5 6 7
Gobs (kg/s)

Figure 29 Observed versus PHAST6.53 predicted discharge mass flux across an orifice for sub-cooled
propane at orifice inlet (Data from Bennett et al., 1991, SHELL TNER 91.022)

| Validation | DISC Model | Page 26

 35

30
Experimental data

PHAST6.51
25

+30%-deviation

20
Gpred (kg/s)

-30%-deviation

+50%-deviation
15

-50%-deviation

10

0
0 5 10 15 20 25
Gobs (kg/s)

Figure 30 Observed versus PHAST6.53 predicted steady-state discharge mass flux following FBR of 2.68
and 11.63m pipes conveying sub-cooled propane at pipe inlet (Data from Bennett et al., 1991,
SHELL TNER 91.022

| Validation | DISC Model | Page 27

3.7 Conclusions and recommendations
From the above results, the following conclusions can be drawn:

Orifice releases

For saturated/two-phase fluids approaching an orifice, the use of the homogenous equilibrium “Flashing” flow assumption
yields poor and under-conservative discharge rate results when compared with measured data. For the same cases, the
use of the “No Flashing”/frozen flow assumption (meta-stable liquid; default option in PHAST 6.53) yields better and
generally conservative estimates. This is true for saturated liquids only, for two-phase fluids the “Flashing” and “No
Flashing” options yield the same results.

For saturated liquids approaching an orifice, the Bernoulli model is observed to yield the best release rate estimates of
the three PHAST6.53 discharge models studied. For two-phase fluids at stagnation point, the “Bernoulli” model coupled
with the “forced-phase flow assumption” 21 can be employed in defining “worst case” release rates (i.e. the limits for
conservatism).

For sub-cooled liquids approaching an orifice, the use of the “No Flashing” option again provides better and generally
conservative estimates of measured data, while the converse is observed with the use of the “Flashing” option.

For sub-cooled liquids approaching an orifice at lower stagnation pressures (e.g. ≤ ca 57barg for water), the Bernoulli
model, like the “No Flashing” model generally yields conservative estimates of measured data. In terms of modelling
accuracy in this pressure range, the Bernoulli model is seen to generally perform better than the “No Flashing” model. For
water, this trend (i.e. better accuracy) is reversed at higher stagnation pressures with the Bernoulli model tending to be
under-conservative in predictions.

Pipe releases

For non-flashing liquid flow through pipelines, simulated discharge rate results using PHAST6.53 are conservative but
compare quite well with available field data.

For flashing (i.e. saturated/two-phase at pipe inlet) flows however, simulated discharge rate results using PHAST6.53 is
observed to agree better with measured data as the adopted pipe length increases. For short pipes (i.e., ≤ 0.1m), the
simulated discharge rate results when compared with measurements are generally under-conservative, while the use of
the “Bernoulli”/“No Flashing” orifice model for simulating these cases is likely to yield conservative but more accurate
results.

Overall recommendations

In all, the following recommendations are proffered:

• The Bernoulli flow model is generally recommended for modelling discharge across orifices and very short
pipelines (i.e. pipe-length ≤ 0.1m) involving sub-cooled and 100% saturated liquids.
• The Bernoulli flow model coupled with the forced-phase flow assumption may be applied to obtain conservative
but reasonably accurate release rate estimates for orifice and very short pipe (i.e. pipe-length ≤ 0.1m) releases
involving low-vapour quality (i.e. ca < 0.7%) two-phase fluids at stagnation point.
• For very high pressure sub-cooled liquid orifice releases, the non flashing compressible flow model [i.e.
“PHAST6.53 (old model- No Flashing)”; current default option in PHAST6.53] may provide better flow rate
estimates than the Bernoulli model. Thus, the use of this modelling option (i.e. “No Flashing”) is recommended
for release scenarios in which the assumption of liquid incompressibility is judged to be less valid (e.g. water at
stagnation pressures > 57barg). Otherwise it is expected to lead to conservative results.
• For pipelines longer than 0.1m, the use of PHAST’s homogenous equilibrium pipeline discharge model should
yield fairly accurate results and is recommended.

These recommendations have been based on the results of a representative, but not necessarily exhaustive, validation
exercise.

| Validation | DISC Model | Page 28

About DNV
We are the independent expert in risk management and quality assurance. Driven by our purpose, to safeguard life,
property and the environment, we empower our customers and their stakeholders with facts and reliable insights so that
critical decisions can be made with confidence. As a trusted voice for many of the world’s most successful
organizations, we use our knowledge to advance safety and performance, set industry benchmarks, and inspire and
invent solutions to tackle global transformations.

Digital Solutions
DNV is a world-leading provider of digital solutions and software applications with focus on the energy, maritime and
healthcare markets. Our solutions are used worldwide to manage risk and performance for wind turbines, electric grids,
pipelines, processing plants, offshore structures, ships, and more. Supported by our domain knowledge and Veracity
assurance platform, we enable companies to digitize and manage business critical activities in a sustainable,
cost-efficient, safe and secure way.

DNV | [email protected] | dnv.com/digital

REFERENCES
1
Perry, R.H, Green, D. W. and Maloney, J.D., (eds.), “Perry Chemicals Engineering Handbook”, 7 th Edition, McGrawhill
(1999), Section 26 “Process safety”
2
Woodward, J.L., “Discharge modelling in PHAST/SAFETI”, Proceedings of “Conference on consequence
modelling and risk analysis”, Witlox, H.W.M. (editor), at DNV, London, 28-29 April 1997 (confidential)
3
Duree, P., “Model evaluation of the DNV Technica PHAST package”, Draft Report TNER.95.***, Thornton Research
Centre, Chester, UK (1995)
4
Fisher, H.G. et al., “Emergency Relief System Design using DIERS technology”, DIERS project manual, ISBN No. 0-
8169-0568-1, Pub. No. X-123 (1992), AICHE, New York
5
“Guidelines for chemical process quantitative risk analysis”, Second Edition, CCPS, New York (2000), Section 2.1.1 –
discharge rate models
6
Woodward, J.L., “Estimating the flammable mass of a vapour cloud”, CCPS, New York (1998)
7
Lees, F.P., “Loss Prevention in the process industries: hazard identification, assessment and control”, Second Edition,
Butterworth-Heinemann, Oxford (1996), , Volume 1, Sections 15.1 until 15.9 (pp. 15.1-57)
8
Giot, M., “Two-phase releases”, J. Loss Prev. Process Ind. 7 (2), pp. 77-93 (1994)
9
Uchida, H., and Nariai, H., “Discharge of saturated water through pipes and orifices”, Proceedings 3day International
Heat Transfer Conference, ASME, Chicago, 5, 1-12, (1966)
10
Sozzi, G. L., and Sutherland, W. A., “Critical flow of saturated and sub-cooled water at high pressure”, General Electric
Co. Report No. NEDO-13418, July (1975)
11
Bennett, J. F., Cowley, L. T., Davenport, J. N., and Rowson, J. J., “Large scale natural gas and LPG jet fires final report
to the CEC”, TNER 91.022, (1991)
12
Witlox, H.W.M. and Stene, J., “Atmospheric expansion modelling – Literature review, Model refinement and
Validation” (2016)
13
Witlox, H.W.M., Harper, M., and Fernandez, M., “UDM validation document”, UDM Technical Reference Manual, DNV
(2017)

DNV | [email protected] | dnv.com/digital