- Email: [email protected]
Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases
Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Contents lists available at ScienceDirect
Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp
Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases Henk W.M. Witlox*, Mike Harper, Adeyemi Oke, Jan Stene DNV Software, Palace House, 3 Cathedral Street, London SE1 9DE, UK
a r t i c l e i n f o
a b s t r a c t
Article history: Received 12 March 2013 Received in revised form 4 October 2013 Accepted 9 October 2013
The consequence modelling package Phast examines the progress of a potential incident from the initial release to the far-field dispersion including the modelling of rainout and subsequent vaporisation. The original Phast discharge and dispersion models allow the released substance to occur only in the vapour and liquid phases. The latest versions of Phast include extended models which also allow for the occurrence of fluid to solid transition for carbon dioxide (CO2) releases. As part of two projects funded by BP and Shell (made publicly available via CO2PIPETRANS JIP), experimental work on CO2 releases was carried out at the Spadeadam site (UK) by GL Noble Denton. These experiments included both high-pressure steady-state and time-varying cold releases (liquid storage) and high-pressure time-varying supercritical hot releases (vapour storage). The CO2 was stored in a vessel with attached pipework. At the end of the pipework a nozzle was attached, where the nozzle diameter was varied. This paper discusses the validation of Phast against the above experiments. The flow rate was predicted accurately by the Phast discharge models (within 10%; considered within the accuracy at which the BP experimental data were measured), and the concentrations were found to be predicted accurately (well within a factor of two) by the Phast dispersion model (UDM). This validation was carried out with no fitting whatsoever of the Phast extended discharge and dispersion models. Ó 2013 Elsevier Ltd. All rights reserved.
Keywords: Carbon dioxide Consequence modelling Model validation Discharge Atmospheric dispersion Thermodynamics
1. Introduction This paper discusses the validation of discharge and subsequent atmospheric dispersion for pressurised carbon dioxide releases using the consequence modelling package Phast based on experimental data shared by the CO2PIPETRANS JIP (Holt, Helle, & Brown, 2012). Phast examines the progress of a potential incident from the initial release to the far-field dispersion including the modelling of rainout and subsequent vaporisation. The original Phast discharge and dispersion models allow the released chemical to occur only in the vapour and liquid phases. The models in the latest versions 6.6 and 6.7 of Phast were extended by Witlox, Harper, and Oke (2009) to also allow for the occurrence of fluid to solid transition for CO2 releases. This applies both for the post-expansion state in the discharge model, as well as for the thermodynamic calculations by the dispersion model. The extended dispersion formulation was
* Corresponding author. Tel.: þ44 (0)207 716 6711. E-mail addresses: [email protected], (H.W.M. Witlox).
[email protected]
tested extensively by means of a sensitivity analysis for a comprehensive range of base cases (Witlox, Stene, Harper, & Nilsen, 2011). The Phast dispersion model (UDM) was previously validated for unpressurised releases of CO2, i.e. against the McQuaid wind-tunnel experiments for isothermal heavy-gas-dispersion from a groundlevel CO2 line source (Witlox & Holt, 1999), the Kit Fox experiments for heavy-gas-dispersion from a ground-level areas source (Witlox & Holt, 2001), and wind-tunnel experiments carried out by CHRC (Chemical Hazards Research Centre at University of Arkansas) for a CO2 ground-level vapour pool source (Witlox, Harper, & Pitblado, 2012). The focus of the current paper is validation of Phast against pressurised CO2 experiments. As part of BP’s engineering project DF1, experimental work on CO2 releases was carried out at the Spadeadam site (UK) by Advantica (now part of GL Noble Denton) for BP. These experiments included both high-pressure steady-state cold releases (liquid storage) and high-pressure supercritical time-varying releases (vapour storage). As part of a project funded by Shell, the same two types of experiments were carried out and in addition time-varying cold releases were considered. The CO2 was stored in a vessel with attached pipework. At the end of the pipework a nozzle was attached, where the nozzle diameter was varied. For the steady-
0950-4230/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jlp.2013.10.006
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
2
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
state cold releases the pressure was kept constant. BP and Shell, when joining the DNV led CO2PIPETRANS Phase 2 Joint Industry Project (JIP), transferred the CO2 experimental work to the JIP. As part of this JIP’s goal to reduce uncertainty associated with CO2 pipeline design and operation, the majority of the experimental data has been made available in the public domain. The current paper discusses the validation of Phast against the above experiments. In Section 2 first a brief overview is provided for Phast modelling of discharge and dispersion for CO2 releases. Section 3 subsequently describes the experiments. Section 4 describes the validation of the Phast steady-state discharge model DISC and the Phast time-varying discharge model TVDI. Section 5 outlines the validation of Phast dispersion model UDM adopting the source-term data derived from DISC and TVDI. The reader is referred to the detailed data review reports by Witlox (2012a, 2012b) for further detailed results not included in the current paper.
2. Overview of Phast modelling of discharge and dispersion for CO2 releases Fig. 1 includes a schematic phase diagram for CO2; CO2 has a critical temperature of 31.06 C (304.2 K) above which Phast presumes it to be always vapour and a triple point of 5.1 atm and 56.55 C (216.6 K) below which all non-vapour CO2 will be solid. Phast examines the progress of a CO2 release from the initial release to far-field dispersion including the modelling of solid rainout and subsequent sublimation to vapour. The main areas for modelling of CO2 in Phast as shown in Fig. 2 are as follows: Discharge modelling of CO2 which includes atmospheric expansion of CO2 (depressurisation to ambient pressure) during which liquid to solid/vapour expansion occurs. In case of initial supercritical temperature (above 31 C), vapour to vapour, or vapour to solid/vapour expansion occurs. The applied Phast discharge models are DISC (steady-state releases) and TVDI (time-varying releases). Starting from the specified vessel stagnation conditions, the discharge model DISC/TVDI is used for modelling the discharge of the CO2. This includes expansion from storage conditions to orifice conditions, and the expansion from orifice to ambient conditions. For the latter expansion the DISC/TVDI sub-model ATEX is used. Dispersion modelling involving the possible presence of solid CO2 in addition to vapour CO2. The ATEX post-expansion
Critical point
Pressure (atm)
fusion line vaporisation line
SOLID VAPOUR 5.1
sublimation line
Triple point
1 -78.4
-56.6
The UDM model invokes a thermodynamics sub-model for mixing of the released material and the ambient air. This model calculates the phase composition and temperature of the mixture at the cloud centre-line. For the BP and Shell CO2 experiments the stagnation pressures are very large and therefore the initial solid particle following depressurisation to ambient pressure is expected to be very small (initial fine solid mist of CO2). Furthermore the atmospheric sublimation temperature is very low (78.4 C) and therefore the solid particles are expected to sublime very quickly. As a result for the mixing of solid/vapour CO2 with air, the UDM thermodynamics sub-model assumes homogeneous equilibrium without deposition of the solid CO2 onto the substrate. Thus trajectories of solid particles are not modelled. The latter assumption was further verified by a detailed sensitivity analysis by Witlox, Stene, et al. (2011). The reader is referred to Witlox et al. (2009) for further details of the modelling. 3. BP and Shell experiments Experiments involving pressurised CO2 releases were carried out at Spadeadam by GL Noble Denton (previously Advantica) for BP in 2006 and for Shell in 2010. The data from these experiments along with other material was transferred into the DNV led CO2PIPETRANS JIP. DNV Software was commissioned by the JIP to undertake a critical review of the data to assess its suitability for public release for model validation, i.e. those corresponding to horizontal non-impinging releases, before the data from these tests were approved for external release. The data review for the BP experiments (Witlox, 2012a) was carried out based on the Advantica report by Evans and Graham (2007) and the overview report by Holt (2012) as well as some supplementary information provided by the original 2006/2007 BP model validation exercise. The data review for the Shell experiments (Witlox, 2012b) was based on the GL Noble Denton report by Allason and Armstrong (2011). In the experiments the CO2 was stored under high pressure in a horizontal cylindrical vessel. The modelled experiments include three sets of experiments: - Cold steady-state releases (liquid storage; BP tests 1, 2, 3, 5, 6, 11 and Shell tests 3,5,11). - Time-varying cold releases (liquid storage; Shell tests 1,2,4) - Hot supercritical releases (dense vapour storage; BP tests 8, 8R, 9 and Shell tests 14,16).
LIQUID
72.9
conditions are used as the source term (starting condition) for the UDM dispersion model. The UDM calculates the CO2 dispersion further downwind ignoring possible deposition on the ground and re-sublimation. The UDM assumes that the release direction is in the same vertical plane as the wind direction.
31.1
Temperature (oC) Fig. 1. Schematic phase diagram for CO2 (not on scale).
For the steady-state tests nitrogen padding gas was used to maintain the pressure in the vessel and to keep the test vessel full of liquid CO2. It should be noted that care was taken to ensure the CO2 being released was not contaminated by the nitrogen. For the timevarying tests the vessel was first filled with CO2 at the required test pressure and test temperature, where in case of hot releases the CO2 was heated using heating pads. Subsequently the CO2 was released through the nozzle driven only by the pressure in the vessel with the vessel pressure decaying as the release progressed. Fig. 3 depicts the release system. Downstream of the vessel a 3 m horizontal flexible hose was attached (200 inner diameter), connected to a 2 m 200 metering spool (BP) or a 300 coriolis meter (Shell) and a 0.5 m 200 nozzle with an orifice plate bolted on the
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
leak orifice
vessel (stagnation)
pipe
3
atmosphere
expansion zone
CO2 plume
vapour-plume centre-line
flow
SUBSTRATE Fig. 2. Discharge modelling (DISC/TVDI/ATEX) and dispersion modelling (UDM).
nozzle. Thus the total length of attached pipe is 5.5 m, with no external insulation applied to the pipe. A wide range of experiments was carried out including variation of orifice diameters (approximately ¼00 ,½00 or 100 ), storage pressures (82e158 barg) and temperatures (0 Ce150 C). Pressure and temperature measurements were made at a range of locations including vessel inlet, vessel outlet and nozzle spool; see references given above for further details. Table 1 summarises the key BP experimental data required as input to the Phast models. In this table the values of the storage pressure and the storage temperature are taken at the discharge end of the vessel (upstream of the pipework), with mean values during the release applied for the steady-state liquid releases and with initial values applied for the transient vapour releases. The ambient data were measured upwind of the release and mean values are adopted for these data during the release. This is with the exception of the wind-speed measurement of 1.65 m above the pad, which was taken 40 m downwind of the release. Since this measurement was disturbed by the CO2 jet, the value listed in Table 1 corresponds to the mean value prior to the release. Likewise Table 2 summarises the key Shell experimental data required as input to the Phast models. In this table the values of the storage pressure and the storage temperature are taken at the discharge end of the vessel (upstream of the pipework) and the nozzle pressure/temperature are taken along the nozzle, with mean values during the release applied for the steady-state releases and with initial values applied for the transient releases. The wind
Fig. 3. Photographic view of release system. Figure corresponds to BP tests; the release system was similar for the Shell tests with a 300 coriolis flow meter fitted upstream of valve V2; see Evans and Graham (2007) and Allason and Armstrong (2011) for further details of the release system.
speed data were taken from tower A [20 m west (behind) and 5 m south of release point; see Fig. 5] at 10 m height (averaged prior to release) given anomalies observed at other measurement locations. Furthermore, based on an analysis of the experimentally observed vertical wind-speed profiles a surface roughness of 0.1 m and a stability class of D was assumed for all (BP and Shell) tests. Finally with respect to the wind direction it is noted that the release direction corresponds to 270 . 4. Validation of Phast discharge models AGAInst CO2 experiments For the Shell experiments, the flow rate and density were measured accurately using a coriolis flow meter. In addition for both the Shell and BP experiments, the vessel weight was measured using load cells. Thus for the BP supercritical vapour releases, the flow rate was derived from the measured vessel weight using load cells. For the BP cold steady-state liquid releases, the flow rate was estimated by Advantica (Evans & Graham, 2007) from the load cells by assuming that the total flow rate dM/dt (kg/s; derived from the vessel mass M as measured by the load cells) satisfies dM/ dt ¼ rCO2VCO2 rN2VN2. Here rCO2 is the CO2 density (kg/m3), VCO2 the CO2 volume rate (m3/s), rN2 the nitrogen density (kg/m3), and VN2 the nitrogen volume flow rate (m3/s). Pressure and temperature were measured at a range of locations upstream of the vessel, inside the vessel, and downstream of the vessel along the pipe and the release valve. The Phast discharge models either assume the release to be directly from an orifice from a vessel (‘Leak’ scenario), or from a short pipe attached to a vessel (with orifice diameter ¼ pipe diameter, i.e. full-bore rupture). Except for the 100 orifice tests (BP test 5 and Shell tests 2,5), the observed pressure at the discharge end was seen to be very close to the observed pressure at the vessel inlet and vessel outlet. Thus the Phast ‘Leak’ scenario was applied, while neglecting the pressure loss from the stagnation conditions to the nozzle conditions. Also the 100 orifice tests can be modelled using the ‘Leak’ scenario, provided that measured nozzle pressures/ temperature are specified as model input instead of storage pressure/temperature (at vessel outlet). The Phast discharge model DISC was used to simulate the steady-state liquid releases, while the Phast discharge model TVDI was used to model the time-varying releases. Default Phast parameters were applied with two exceptions (for two parameters specified in Phast 6.7 and 7.01 in the ‘Discharge parameter’ tab). First the metastable assumption (non-equilibrium with liquid ‘frozen’) was not applied, but flashing was allowed at the orifice (equilibrium at the orifice) to account for the pipework upstream of the orifice. Secondly conservation of momentum was applied for the expansion from orifice to post-expansion conditions, since this
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
4
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Table 1 Experimental conditions for BP CO2 tests. Input Discharge data Steady-state/transient Storage phase Storage pressure (barg) Storage temperature ( C) Vessel volume (m3) Orifice diameter (mm) Orifice length (mm) Release duration (s) Ambient data Ambient temperature ( C) Ambient pressure (mbara) Relative humidity (%) Wind direction (degrees) Wind speed (m/s)
Test1
Test2
Test3
Test5
Test6
Test11
Test8
Test8R
Test9
Input for models
Steady Liquid 103.4 5 e 11.94 46.78 60
Steady Liquid 155.5 7.84 e 11.94 46.78 59
Steady Liquid 133.5 11.02 e 11.94 46.78 60
Steady Liquid 157.68 9.12 e 25.62 72.41 40
Steady Liquid 156.7 9.48 e 6.46 47.79 120
Steady Liquid 82.03 17.44 e 11.94 46.78 58
Trans. Vapour 157.76 147.12 6.3 11.94 46.78 120
Trans. Vapour 148.7 149.37 6.3 11.94 46.78 132
Trans. Vapour 154.16 69.17 6.3 11.94 46.78 179
e DISC,TVDI DISC,TVDI DISC,TVDI TVDI DISC,TVDI e e
14.2 999.4 74.4 322.4 4
7.5 958.2 96 265.6 3.44
10.6 972.5 95.8 288.8 3.37
5.8 985.4 96.7 278.6 5.13
6.1 938.4 1 299 2.20
11.6 960.2 94 270.8 5.99
11.19 957.99 100 269.3 4.71
11.1 957.1 100 270 0.76
8.2 958.9 99.9 270.7 4.04
DISC,TVDI,UDM DISC,TVDI,UDM DISC,TVDI,UDM UDM uses 270 UDM
assumption was previously found to provide the most accurate concentration predictions [e.g. against the SMEDIS experiments; see the UDM validation manual (Witlox, Harper, & Holt, 2011) for details]. 4.1. Time-varying releases Table 3 shows that the measured initial CO2 density derived from the coriolis flow meter for the time-varying Shell tests is very close to values predicted by the SpaneWagner Equation of State (SW EOS) in line with recommendations of a report by EON (2011). Also accurate predictions are obtained of the liquid density in Phast using the (non-default) PengeRobinson Equation Of State (PR EOS). The default method in Phast presumes the liquid density equal to the saturated liquid density at the given temperature, i.e. independent of the pressure. This was shown to result in accurate predictions of the liquid density after depressurisation to saturated conditions, but to lead to a significant under-prediction of the initial liquid density (i.e. total initial vessel mass). Table 3 also demonstrates that improved predictions are obtained of the initial vapour density for the hot tests 14 and 16 using the PR EOS. Fig. 4 includes TVDI-predicted (default Phast density) and observed values for flow rate (kg/s) versus time (s) for the timevarying releases. The thick solid lines represent the experimental data. The other lines represent TVDI predictions, while allowing flashing at the orifice. For the liquid releases, Fig. 4a includes TVDI predictions using both the default EOS and (using a new more robust development
version of TVDI) the PR EOS. The default EOS predicts both the initial flow rate and the flow rate after depressurisation to saturated conditions quite accurately. However less accurate results are obtained during the regime of depressurisation to saturated, with a too rapid decrease of flow rate caused by the usage of the too low saturated liquid CO2 density. The PR EOS using a more accurate CO2 density provides more accurate predictions of the flow rate for all regimes. For the hot vapour releases, Fig. 4b includes predictions of the current TVDI model (version 6.7) using both the default EOS and the PR EOS. For the BP tests, the observed values for the flow rates are averaged over a period over 8 s to reduce oscillations caused by inaccuracies of the load-cell measurements. This was not necessary for the coriolis flow meter measurements. The following can be concluded from Fig. 4b: - For the very hot BP tests 8, 8R (storage temperature about 150 C), the vapour remains vapour upon depressurisation (condensation not relevant), and it is seen that very close agreement is obtained between TVDI predictions and observed data using both the default and PR EOS, since they both provide very accurate predictions of CO2 vapour density. - For the BP test 9 and the Shell test 14 (temperature about 70 C), PR EOS is seen to produce most accurate results. Furthermore the default-EOS TVDI runs are seen to terminate prematurely, which was due to convergence problems apparently caused by the release temperature being lower and closer to the critical temperature. An under-prediction of the flow rate is seen at larger times.
Table 2 Experimental conditions for Shell CO2 tests. Input Discharge data Steady-state/transient Storage phase Storage pressure (barg) Nozzle pressure (barg) Storage temperature ( C) Nozzle temperature ( C) Vessel volume (m3) Orifice diameter (mm) Orifice length (mm) Release duration (s) Ambient data Ambient temperature ( C) Ambient pressure (mbara) Relative humidity (%) Wind direction (degrees) Wind speed (m/s)
Test3
Test5
Test11
Test1
Test2
Test4
Test14
Test16
Input for models
Steady Liquid 147.3 144.8 9.8 8.2 e 12.7 47.78 120
Steady Liquid 148.8 126.4 17.8 13.7 e 25.4 46.84 40
Steady Liquid 81.9 80.3 0.2 1.4 e 12.7 47.78 120
Trans. Liquid 148.3 143 26.7 23 6.3 12.7 47.78 90
Trans. Liquid 147.1 118 24.6 18 6.3 25.4 46.84 145
Trans. Liquid 148.2 148.2 20.1 20.1 6.3 6.3 47.79 >700
Trans. Vapour 151.6 147.7 71 65.0 6.3 12.7 47.78 315
Trans. Vapour 150.6 146.0 36.7 31.7 6.3 12.7 47.78 370
e DISC,TVDI DISC,TVDI (either storage or nozzle pressure is input) DISC,TVDI (either storage or nozzle temperature is input) TVDI DISC,TVDI e e
11.2 1017 66 267 4.05
9 905 91 213 1.30
3.6 995 78 261 2.76
14.7 1006 83 263 3.93
10.3 1005 77 250 5.43
13.8 975.5 77 215 1.98
0 1005 88 303 1.34
2.9 997 88 292 1.48
DISC,TVDI,UDM DISC,TVDI,UDM DISC,TVDI,UDM UDM uses 270 UDM
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 Table 3 Predicted versus observed initial CO2 density for time-varying Shell tests. Input
Test1 Test2 Test4 Test14 Test16 Comments
Storage phase Storage pressure (bara) Storage temperature ( C) Predicted and measured density Phast default density (kg/m3) Phast PR density
Liquid Liquid Liquid Vapour Vapour 149.3 148.1 149.2 152.6 151.6 ¼ gauge þ ambient pressure 26.7 24.6 20.1 71 36.7
686
719
773
445
704
854
868
901
476
779
SpaneWagner density (kg/m3)
866
878
903
507
803
measured density (kg/m3)
890
919
907
493
826
Phast (default method) Phast (modified method) derived via interpolation of data in EON report measured by coriolis flow meter
- For Shell test 16 the above effects are seen to be even more pronounced, since the initial storage temperature is only a few degrees above the critical temperature. At larger times the vessel fluid may become liquid, but the transition from vapour to liquid is not modelled by TVDI resulting in under-prediction.
4.2. Initial rate for steady-state and time-varying releases 4.2.1. BP tests Table 4 summarises the overall results of the discharge rates for all BP tests. For the steady-state tests only the DISC initial release rate is given, while for the time-varying releases also the TVDIpredicted averaged release rate over the first 20 s is indicated. It is noted that the difference between the averaged and initial rate is relatively small. From the table it is seen that the timevarying Phast predictions align well with the observed discharge rate for the hot tests 8, 8R and 9. The predicted flow rate for the cold releases, with the exception of test 5 (100 release), is also very close to that of the experiments. For test 5 (100 release) the flow rate is over-predicted with 23% (50.74 kg/s predicted versus 41.17 kg/s experimental) using the ‘Leak’ scenario, while using the pipe (‘Line Rupture’) scenario it is under-predicted with 34.5% (26.95 kg/s predicted versus 41.17 kg/ s). The over-prediction for the orifice scenario is believed to be caused by the fact that pressure loss is ignored along the pipework (hose/spool/nozzle). Test 5 has the largest orifice diameter (100 ) and therefore will be most susceptible to upstream pressure loss and reduced flow rate. Indeed if a more accurate pressure would be applied of 128.6 barg (corresponding to averaged observed pressure close to the orifice) a release rate of 45.34 kg/s is predicted using the ‘Leak’ scenario corresponding to a much smaller overprediction of 10.1%. For the other ¼” and ½” tests very little difference between vessel-outlet and nozzle pressures was observed, and this is confirmed by Table 4 which shows accurate flow-rate predictions for these tests (within experimental accuracy). The DISC output data reported in Table 4 are derived in Phast from the ‘Discharge report’ (where ‘liquid mass fraction’ should be interpreted for CO2 as ‘solid mass fraction’). The key difference in discharge model input between tests 8/8R, test 9, and test 2 is the storage temperature (around 148C versus 69C versus 8C). It is seen from the post-expansion data reported in Table 4 that the lower temperature results in increasing two-phase solid mass fraction (0 versus 15% versus 40%), a larger density, a larger flow rate (around 4 versus 6 versus 11 kg/s) and a smaller release velocity (around 470 versus 290 versus 190 m/s).
5
As indicated above the flow rate changes little for the timevarying tests 8, 8R, 9 within the first 20 s, and it is believed that within 20 s the maximum concentrations will be achieved within the first 80 m (given relatively large initial jet momentum and relatively large values of wind speed). Therefore in the next section the UDM dispersion calculations are modelled as steady-state using the averaged flow rate over the first 20 s for tests 8, 8R and 9, while for the other tests the overall averaged observed value is adopted. All other UDM source-term input data (temperature, solid fraction, velocity) are chosen as predicted above by the discharge model DISC. 4.2.2. Shell tests Table 5 summarises the overall DISC predictions of the initial discharge rates for all Shell tests. In this table a range of model assumptions is applied: - Time-varying releases are calculated either based on measured initial storage (vessel outlet) pressure/temperature or measured initial nozzle pressure/temperature; as expected, particularly for the largest 100 orifice size (test 2), usage of nozzle data significantly improves the predictions given the significant pressure decay between storage and nozzle conditions; for the smallest ¼00 orifice size identical results are obtained because of negligible pressure decay. The DISC flow-rate predictions for the transient hot releases based on the initial values of temperature and pressure at the nozzle are seen to be larger than those based on the higher values of temperature and pressure at the vessel discharge end. This is because a lower DISC input temperature results in a larger solid fraction and a larger flow rate prediction, and this effect appears to be dominant to the reduction of flow rate due to the lower DISC input pressure. - Phast liquid density either based on default (saturated density) or more accurate Peng Robinson density, with more accurate results obtained using the PR EOS - Flashing (non-default Phast) or non-flashing (default Phast; metastable liquid assumption). Using PengeRobinson (PR) density, this is seen to affect results very little. Using the saturated density, the default non-flashing option provides conservative results while the non-default flashing assumption produces significantly more accurate results. Table 6 includes results and UDM source terms based on those model assumptions which would provide the best agreement against the experimental data, i.e. based on the PR EOS, the flashing assumption, and nozzle data for liquid releases and storage data for hot releases. Note that unlike the BP experiments, the UDM dispersion data are chosen to depend on the initial rate rather than on the averaged rate during the first 20 s. This choice was made, since as discussed later the concentration sensors only measured accurately the initial concentration, and not the concentration at subsequent times. 5. Validation of Phast dispersion model against CO2 experiments The CO2 concentration was largely measured via O2 cells for both BP and Shell experiments; see Fig. 5 [taken from Allason and Armstrong (2011)] for the location of the O2 concentration sensors. Thus a total of 43 sensors was applied at downstream distances of 5 m (sensor OC01), 10 m (OC02), 15 m (OC03), 20 m (OC04-OC08), 40 m (OC9-OC21), 60 m (OC22-OC28) and 80 m (OC29-OC43), with sensors position at a range of different heights (0.3, 1 or 3 m) and
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
6
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
(a)
BP Test 8 - experiment (load cell - 8s averaged)
(b)
BP Test8 - TVDI (default EOS) BP Test8 - TVDI (PR) BP Test 8R - experiment (load cell - 8s averaged) BP Test8R - TVDI (default EOS) BP Test8R - TVDI (PR) BP Test 9 - experiment (load cell - 8s averaged) BP Test9 - TVDI (default EOS, flashing) BP Test9 - TVDI (PR, flashing) Shell Test14 - experiment (coriolis flow meter) Shell Test14 - TVDI (default EOS, flashing) Shell Test14 - TVDI (PR, flashing) Shell Test16 - experiment (coriolis flow meter) Shell Test16 - TVDI (default EOS, flashing) Shell Test16 (TVDI PR, flashing)
Fig. 4. TVDI validation of flow rate for time-varying releases. (a) Cold liquid releases (Shell tests 1,2,4). (b) Hot vapour releases (BP tests 8, 8R, 9 and Shell tests 14, 16).
Table 4 Predicted versus observed flow rates and UDM source-term data (BP tests). Test1 Discharge rate DISC initial discharge rate (kg/s) DISC/TVDI discharge rate (kg/s) (averaged over first 20 s for tests 8.8R,9) Observed discharge rate (kg/s) (averaged over first 20 s for tests 8.8R,9) Deviation predicted from observed Final (post-expansion) state (UDM input) Discharge rate (kg/s) (from experiments) Temperature (K) (DISC output) Solid fraction () (DISC output) Velocity (m/s) (DISC output)
8.84 8.84 e 7.8% 8.2 194.6 0.397 156.7
Test 2 10.98 10.98 11.41 3.9% 11.41 194.1 0.403 189.8
Test3 9.988 9.988 9.972 0.16% 9.988 194.26 0.384 179.2
Test 5 50.75 50.75 41.17 þ23% 41.17 194.4 0.399 191.7
Test6
Test 11
Test 8
Test 8R
Test 9
3.21 3.21 3.50 8.2%
7.03 7.03 7.12 1.1%
4.19 4.01 4.07 1.5%
3.90 3.73 3.80 1.8%
6.86 6.25 6.05 þ3.4%
3.50 193.8 0.397 191.3
7.12 194.1 0.330 154.2
4.07 198.2 0 466.5
3.80 204.8 0 472.8
6.05 194.1 0.154 289.0
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
7
Table 5 Predicted versus observed flow rates e vary Phast assumptions (Shell tests). Prescribed pressure/temperature
Mean nozzle
Initial nozzle
Release type
Steady-state liquid
Transient liquid
Shell CO2 test number
Test 3
Predicted flow rate (kg/s) - Default density, flashing - Default density, no flash. - PR density, flashing - PR density, no flashing Observed flow rate Ratio predicted/observed - Default density, flashing - Default density, no flash. - PR density, flashing - PR density, no flashing
Test5
Test11
Test1
Test2
Initial storage (vessel discharge end)
Test4
Transient hot
Transient liquid
Test14
Test1
11.93 13.58 12.37 12.16 12.4
43.38 50.88 44.36 43.92 44.7
8.89 10.21 9.10 9.29 8.9
11.23 13.61 11.38 11.12 10.55
40.90 49.48 41.26 41.15 38
2.85 3.39 2.92 2.85 3.17
7.82 7.46 7.67 7.71 7.37
96.2% 109.5% 99.8% 98.0%
97.1% 113.8% 99.2% 98.3%
99.9% 114.7% 102.2% 104.4%
106.4% 129.0% 107.9% 105.4%
107.6% 130.2% 108.6% 108.3%
90.0% 106.9% 92.0% 89.8%
106.1% 101.2% 104.1% 104.7%
Test16 10.92 Error 10.88 Error 10.5 104.0% Error 103.6% Error
Test2
Transient hot Test4
Test14
11.29 13.95 11.40 11.06 10.55
45.36 55.30 45.94 44.71 38
2.85 3.39 2.92 2.85 3.17
7.33 7.22 7.45 7.45 7.37
107.0% 132.2% 108.0% 104.8%
119.4% 145.5% 120.9% 117.7%
90.0% 106.9% 92.0% 89.8%
99.5% 97.9% 101.1% 101.1%
Test16 10.87 Error 10.74 Error 10.5 103.6% Error 102.3% Error
Table 6 Predicted versus observed flow rates and UDM source-term data (Shell tests). Prescribed pressure/temperature
Mean nozzle
Initial nozzle
Release type
Steady-state liquid
Shell CO2 test number
Test 3
Discharge rate DISC initial discharge rate e PR,fl. (kg/s) Observed rate (kg/s) (initial rate for transient tests) Deviation predicted from observed Final (Post Expanded) State (UDM input) Discharge rate (kg/s) (from experiments) Temperature (K) (DISC output) Solid fraction () (DISC output) Velocity (m/s) (DISC output)
12.37 12.4 0.24% 12.4 194.82 0.40 176.25
Test5 44.36 44.7 0.77% 44.7 193.37 0.37 170.94
Initial storage
Transient liquid Test11 9.10 8.9 2.25% 8.9 194.55 0.42 132.23
Test1 11.38 10.55 7.86% 10.55 194.69 0.34 187.93
Test2 41.26 38 8.59% 38 194.67 0.35 170.80
Transient hot Test4 2.92 3.17 8.03% 3.17 194.30 0.36 187.48
Test14
Test16
7.45 7.37 1.08% 7.37 194.67 0.15 292.91
10.74 10.5 2.26% 10.5 194.57 0.29 208.16
Fig. 5. Field detector array for concentration measurements. Figure corresponds to Shell tests, but concentration sensor location is also applicable to BP tests.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
8
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Fig. 6. BP Test 11 e observed raw and time-averaged concentration data. (a) Sensor OC16 (40 m downstream). (b) Sensor OC03 (15 m downstream). (a) Sensor OC01 (5 m downstream).
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
9
Fig. 7. BP Test 11 e UDM validation for maximum contraction versus distance.
cross-stream distances (between 20 and þ20 from the release direction) with two additional Servomex CO2 analysers. Phast assumes that the release direction is the same as the wind direction, while for some of the experiments (see Tables 1 and 2) there is a significant deviation from the wind direction. This may lead to less accuracy of the predictions in the far-field but will not significantly affect the prediction for the momentum-driven dispersion in the near-field.
raw concentrations for sensors OC01, OC03 and OC16 locations at 5, 15 and 40 m downstream distances along the release axis and at 1 m height. In addition it includes observed concentrations timeaveraged over 11 s, 21 s and 59 s. Here 59 s approximately corresponds to the release duration (reported as 60 s). Furthermore for a given averaging period tav, the time-averaged observed concentration cav ðtÞ is evaluated as function of time t from the raw obs observed concentration cav ðtÞ as follows: obs
5.1. BP tests (270.8 )
For the steady-state test 11 the averaged wind direction is very close to the release direction (270 ). Fig. 6 includes observed
cav obs ðtÞ
1 ¼ tav
tþt Z av =2
craw obs ðtÞdt ttav =2
Fig. 8. BP Test 9 e UDM validation for maximum concentration versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
10
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 20
Concentration (%CO2)
15 Draeger sensor (5m downstream, 1m high) O2 sensor OC01 (5m downstream, 1m high)
10
5
0
-5 0
20
40
60
80
100
120
140
160
180
200
220
240
Time (s)
Fig. 9. Shell Test 11 e Draeger versus O2 sensor predictions.
Fig. 6a shows that the concentration fluctuations are relatively small with respect to the mean concentration, and therefore a relatively accurate measurement of the concentration can be provided. This is also because the jet centre-line will pass sensor OC1 very closely. In theory (so close to the release point) the concentrations should be approximately constant over a period of 60 s (roughly between 75 s and 135 s). There is a relative small spread between the maximum value for 11-s averaged concentration (21.15 mol %) and the 59-s averaged concentration (18.79%). The subsequent figures Fig. 6b and c however show that the relative differences increase with increasing distance from the source. This is partly because the plume centre-line is more likely to miss the sensors at distances further downstream (because of fluctuating wind direction, as confirmed by wind direction variation observed by Advantica), and also because the sensor readings become relatively less accurate further downstream. For small concentrations, the sensor seems to have an accuracy of no better than 0.5%. Thus measurements with average concentrations less than 1% are considered to be of less value (except to confirm that the concentrations are small). Fig. 6c shows that the
concentration away from the plume is erroneously ‘negative’, and one may therefore consider to re-calibrate the observed concentrations (i.e. increase all measured values with the negative minimum value). However this has not been carried out as part of the current work. Fig. 7 plots for test 11 the maximum values over time of the measured concentration along with the Phast predicted concentrations as a function of downstream distance. The measured data include the maximum concentration of the raw data over all times, 11-s, 20-s and 59-s averaged concentrations. For the measured data at a given downstream distance the maximum value of all sensors at that distance is taken, Sensor 14 (located at 40 m downstream, 3 m height) has been excluded since it appeared to give erroneous too high readings (higher than sensors at 1 m height and sensors further upstream). Furthermore no further analysis has been carried out (e.g. via spline fitting of the measured values to obtain a better fit of the crosswind concentration profile and a better estimate of the maximum concentration) to further refine this maximum value. The Phast predictions were found not to be affected by time-averaging effects due to plume meander (transition to passive dispersion occurring downwind of 80 m). In the near field (<20 m) the 59-s averaged concentration predicted by Phast is close to the measured concentrations. This is also in line with UDM validation against previous experiments, where very close agreement was obtained in the near-field, jet-momentum dominated regime. Further downstream (at 20 m and 40 m) it is seen that the spread in the measured concentrations becomes larger with a larger effect of averaging. This is because of (a) larger relative inaccuracy of the sensors, and (b) the CO2 plume centre-line more likely to be further away from the sensor (also because of plume meander). Thus for this case, as is clearly illustrated by Fig. 7, the maximum value would lead to ‘too’ large (rather random) value of the maximum concentration (it would increase with the release duration), while on the other hand the 59-s averaged concentration may lead to too small values. Fig. 8 includes results of UDM validation for maximum concentration versus downstream distance for the time-varying test 9 (vapour release). It is again seen that good agreement with the processed averaged experimental data is obtained. For this test,
Fig. 10. Shell Test 11 e UDM validation for maximum contraction versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
11
Fig. 11. Shell Test 16 e UDM validation for maximum concentration versus distance.
sensors 17 and 14 were considered to give possible incorrect readings for similar reasons to sensor 14 in test 11. 5.2. Shell tests For the Shell tests, a limited number of 3rd party commercial CO2 detectors including instruments from Draeger as well as two Servomex gas analysers were used in addition to the O2 sensors in order to verify the accuracy of the O2 sensors. From the results of this it was deduced that the O2 sensors did reasonably predict the initial (maximum) value of the concentration, but did subsequently show an erroneous decay with time which could have
been caused by significant cooling of the sensors; see Fig. 9. For unknown reason this was observed for the Shell tests only and not the BP tests. See the data review by Witlox (2012b) for further details. Figs. 10e12 plot for tests 11 (steady-state cold release), 16 (transient hot release) and 1 (transient liquid release) the maximum values over time of the measured concentration along with the Phast predicted concentrations as a function of downstream distance. The measured data include maximum values and (for steady-state test 11 only) averaged values (over release duration) for the O2 sensors, Servomex sensors, and (if present) Draeger sensors.
Fig. 12. Shell Test 1 e UDM validation for maximum concentration versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
12
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Overall it is seen that the maximum O2 values agree well with the UDM predictions, where negligible difference was observed between UDM predictions at 1 m release height and centre-line (C/ L) height. Because of erroneous decay with time as illustrated in Fig. 9, the averaged O2 values result in too low observed values for Test 11. The maximum concentration derived from the Draeger sensors is reasonably aligned with that derived from O2 sensors, but it is particularly less accurate in the far-field because of an insufficient number of sensors (and thus the Draeger sensors may miss the centre-line of the plume).
For the Shell experiments, maximum concentrations for the O2 sensors were used (as discussed in Section 5.2), and none of the O2 sensors was ignored even though they may provide a less accurate reading. This may have caused the overall under-prediction for the Shell experiments. Furthermore for the steady-state releases a higher accuracy is obtained than for the BP experiments, because of (a) input of more accurate measured flow rate and (b) use of conservative 11-s average estimate (maximum over all times during release duration) for the BP experiments. 6. Conclusions and future work
5.3. Comparison statistics between predicted and observed concentrations For a given experimental dataset, it is common practice (Hanna, Strimaitis, & Chang, 1991) to calculate the geometric mean bias MG (averaged ratio of observed to predicted concentrations; MG < 1 over-prediction and MG > 1 under-prediction) and the geometric variance VG (variation from mean; minimum value ¼ 1). Ideally, MG and VG would both equal 1.0. Geometric mean bias (MG) values of 0.5 and 2.0 can be thought of as a factor of 2 in over-predicting and under-predicting the mean, respectively. Likewise, a geometric variance (VG) of about 1.6 indicates scatter from observed data to predicted data by a factor of 2. Table 7 includes the predictions of MG and VG for the BP and Shell experiments. It is noted that all MG values are well within the range of [0.5, 2], and all variances less than 1.6 which is normally considered to be excellent agreement with the experimental data. In Table 7 the observed maximum concentration at a downstream distance is taken as the maximum value of all sensors at that downstream distance. For the BP experiments, the maximum value over all times of the 11-s averaged concentrations has been applied and sensors of apparent false readings have been ignored as discussed in Section 5.1. Therefore conservative estimates are obtained of the averaged observed concentrations for the cold releases (1, 2, 3, 5, 6, 11), which may (partly) explain the under-prediction of the concentrations for the experiments 2, 3, 5, 6. For tests 1, 3, 6 there was a significant difference between the wind direction (averaged over the entire release duration) and the release direction. However the above results show that the plume centre-line did not significantly miss the sensors. Further downstream this may have been caused because we adopt 11-s averaged concentrations (maximum overall all times) rather than concentrations averaged over the entire release duration. Furthermore it must be noted that for tests 3 and 6 a 200 1.44 m extension tube was attached downstream to the ½00 (test 3) and ¼00 (test 6) nozzle, which is not expected to affect the discharge flow rate but is likely to have affected the dispersion. This may explain the largest under-prediction of the concentrations (largest MG values) for tests 3 and test 6.
This paper described the validation of the Phast discharge and dispersion models against BP and Shell experimental data made available via the CO2PIPETRANS JIP. The experiments involve pressurised CO2 releases involving steady-state and transient cold liquid releases, and time-varying supercritical hot vapour releases. The steady-state cold releases were modelled by the Phast discharge model DISC as steady-state orifice releases, while the Phast discharge model TVDI was used to model the time-varying orifice releases. The flow rate was accurately predicted for both the BP and Shell tests (within about 10%), which was deemed to be within the accuracy at which the BP experimental data were measured. Particularly for temperatures closer to the critical temperature, more accurate and robust behaviour was obtained using the PR EOS instead of the default EOS. The releases were all modelled by the Phast dispersion model UDM as steady-state releases, with 20-s averaged flow rates applied for the time-varying BP releases and with the initial maximum flow rate applied for the time-varying Shell releases. For all cases the solid carbon dioxide was found to sublime rapidly and no fallout was predicted, which was fully in line with the experiments. The concentrations were found to be predicted accurately (well within a factor of two). No fitting of the extended Phast models has been carried out whatsoever as part of the above validation. More CO2 experiments are being planned (as part of the joint industry projects CO2PIPETRANS and COSHER) involving pipeline releases instead of orifice releases, and these experiments could be considered for future model validation. Acknowledgements BP is acknowledged for providing funding for the initial validation of the BP DF1 experiments in 2007. BP and Shell are acknowledged for making the data available via the CO2PIPETRANS JIP. The help of Hamish Holt (CO2PIPETRANS JIP project manager) is acknowledged for providing input regarding the BP and Shell experiments. References
Table 7 UDM values of mean MG and variance VG for CO2 experiments. (a) BP tests
(b) Shell tests.
Test
Release type
MG
VG
Test
Type of release
MG
VG
1 2 3 5 6 11 8 8R 9
Steady-state cold Steady-state cold Steady-state cold Steady-state cold Steady-state cold Steady-state cold Transient hot Transient hot Transient hot
0.89 1.44 1.60 1.59 1.63 0.99 1.25 1.25 1.40
1.18 1.15 1.30 1.24 1.31 1.20 1.07 1.12 1.13
3 5 11 1 2 4 14 16
Steady-state cold Steady-state cold Steady-state cold Transient cold Transient cold Transient cold Transient hot Transient hot
1.39 0.98 1.17 1.08 1.06 1.06 1.21 1.35
1.12 1.03 1.17 1.02 1.03 1.01 1.07 1.28
Allason, D., & Armstrong, K. (2011). Liquid and supercritical carbon dioxide release and dispersion experiments on behalf of Shell International Exploration and Production BV, 19th September 2011, Report nr. 10793. Cumbria, UK: GL Noble Denton. Equation of state prediction of carbon dioxide properties. Report KCP-GNS-FAS-DRP0001, Rev.02, Kingsnorth Carbon Capture & Storage Project. UK: EON (2011). Evans, J. A., & Graham, I. (2007). DNV CO2PIPETRANS JIP e Data release 1-Advantica overview. June 2012, Extract from a confidential report by Advantica for BP, Cumbria, UK. Hanna, S. R., Strimaitis, D. G., & Chang, J. C. (1991). Hazard response modelling uncertainty (A quantitative method). Report for the API. Westford, USA: Sigma Research Corp. Holt, H. (June 2012). DNV CO2PIPETRANS JIP e Data release 1-DNV overview report. DNV. Holt, H., Helle, K., & Brown, J. (20e21 June 2012). Capturing and sharing CO2 release data: the CO2PIPETRANS JIP. Newcastle, UK (2012). In 3rd International forum on transportation of CO2 by pipeline.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 Witlox, H. W. M. (2012a). “Data review and Phast analysis (discharge and atmospheric dispersion) for BP DF1 CO2 experiments”, Contract 96000056 for DNV Energy (CO2PIPETRANS Phase 2 JIP WP1). London: DNV Software. Witlox, H. W. M. (2012b). “Data review and Phast analysis (discharge and atmospheric dispersion) for Shell CO2 experiments”, Contract 984C0004 for DNV Energy (CO2PIPETRANS Phase 2 JIP WP1). London: DNV Software. Witlox, H. W. M., Harper, M., & Holt, A. (2011). “UDM validation manual”, Phast 6.7 technical documentation. London: DNV Software. Witlox, H. W. M., Harper, M., & Oke, A. (2009). Modelling of discharge and atmospheric dispersion for carbon dioxide releases. Journal of Loss Prevention, 22(6), 795e802. Witlox, H. W. M., Harper, M., & Pitblado, R. (April 2012). Validation of Phast dispersion model as required for USA LNG Siting Applications. LNG plant safety symposium, 12th topical conference on gas utilisation, AICHE Spring Meeting, Houston, USA.
13
Witlox, H. W. M., & Holt, A. (1999). A unified model for jet, heavy and passive dispersion including droplet rainout and re-evaporation. In International conference and workshop on modelling the consequences of accidental releases of hazardous materials (pp. 315e344). San Francisco, California: CCPS. September 28 e October 1. Witlox, H. W. M., & Holt, A. (2001). “Validation of the unified dispersion model against Kit Fox field data”, Contract 44003900 for Exxon Mobil. London (: DNV Software. Witlox, H. W. M., Stene, J., Harper, M., & Nilsen, S. (2011). Modelling of discharge and atmospheric dispersion for carbon dioxide releases including sensitivity analysis for wide range of scenarios. In 10th Int. Conf. On Greenhouse gas Control Technologies, 19-23 September 2010, Amsterdam, The Netherlands. Made available online via Energy Procedia (Vol. 4; pp. 2253e2260). Elsevier.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
Contents lists available at ScienceDirect
Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp
Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases Henk W.M. Witlox*, Mike Harper, Adeyemi Oke, Jan Stene DNV Software, Palace House, 3 Cathedral Street, London SE1 9DE, UK
a r t i c l e i n f o
a b s t r a c t
Article history: Received 12 March 2013 Received in revised form 4 October 2013 Accepted 9 October 2013
The consequence modelling package Phast examines the progress of a potential incident from the initial release to the far-field dispersion including the modelling of rainout and subsequent vaporisation. The original Phast discharge and dispersion models allow the released substance to occur only in the vapour and liquid phases. The latest versions of Phast include extended models which also allow for the occurrence of fluid to solid transition for carbon dioxide (CO2) releases. As part of two projects funded by BP and Shell (made publicly available via CO2PIPETRANS JIP), experimental work on CO2 releases was carried out at the Spadeadam site (UK) by GL Noble Denton. These experiments included both high-pressure steady-state and time-varying cold releases (liquid storage) and high-pressure time-varying supercritical hot releases (vapour storage). The CO2 was stored in a vessel with attached pipework. At the end of the pipework a nozzle was attached, where the nozzle diameter was varied. This paper discusses the validation of Phast against the above experiments. The flow rate was predicted accurately by the Phast discharge models (within 10%; considered within the accuracy at which the BP experimental data were measured), and the concentrations were found to be predicted accurately (well within a factor of two) by the Phast dispersion model (UDM). This validation was carried out with no fitting whatsoever of the Phast extended discharge and dispersion models. Ó 2013 Elsevier Ltd. All rights reserved.
Keywords: Carbon dioxide Consequence modelling Model validation Discharge Atmospheric dispersion Thermodynamics
1. Introduction This paper discusses the validation of discharge and subsequent atmospheric dispersion for pressurised carbon dioxide releases using the consequence modelling package Phast based on experimental data shared by the CO2PIPETRANS JIP (Holt, Helle, & Brown, 2012). Phast examines the progress of a potential incident from the initial release to the far-field dispersion including the modelling of rainout and subsequent vaporisation. The original Phast discharge and dispersion models allow the released chemical to occur only in the vapour and liquid phases. The models in the latest versions 6.6 and 6.7 of Phast were extended by Witlox, Harper, and Oke (2009) to also allow for the occurrence of fluid to solid transition for CO2 releases. This applies both for the post-expansion state in the discharge model, as well as for the thermodynamic calculations by the dispersion model. The extended dispersion formulation was
* Corresponding author. Tel.: þ44 (0)207 716 6711. E-mail addresses: [email protected], (H.W.M. Witlox).
[email protected]
tested extensively by means of a sensitivity analysis for a comprehensive range of base cases (Witlox, Stene, Harper, & Nilsen, 2011). The Phast dispersion model (UDM) was previously validated for unpressurised releases of CO2, i.e. against the McQuaid wind-tunnel experiments for isothermal heavy-gas-dispersion from a groundlevel CO2 line source (Witlox & Holt, 1999), the Kit Fox experiments for heavy-gas-dispersion from a ground-level areas source (Witlox & Holt, 2001), and wind-tunnel experiments carried out by CHRC (Chemical Hazards Research Centre at University of Arkansas) for a CO2 ground-level vapour pool source (Witlox, Harper, & Pitblado, 2012). The focus of the current paper is validation of Phast against pressurised CO2 experiments. As part of BP’s engineering project DF1, experimental work on CO2 releases was carried out at the Spadeadam site (UK) by Advantica (now part of GL Noble Denton) for BP. These experiments included both high-pressure steady-state cold releases (liquid storage) and high-pressure supercritical time-varying releases (vapour storage). As part of a project funded by Shell, the same two types of experiments were carried out and in addition time-varying cold releases were considered. The CO2 was stored in a vessel with attached pipework. At the end of the pipework a nozzle was attached, where the nozzle diameter was varied. For the steady-
0950-4230/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jlp.2013.10.006
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
2
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
state cold releases the pressure was kept constant. BP and Shell, when joining the DNV led CO2PIPETRANS Phase 2 Joint Industry Project (JIP), transferred the CO2 experimental work to the JIP. As part of this JIP’s goal to reduce uncertainty associated with CO2 pipeline design and operation, the majority of the experimental data has been made available in the public domain. The current paper discusses the validation of Phast against the above experiments. In Section 2 first a brief overview is provided for Phast modelling of discharge and dispersion for CO2 releases. Section 3 subsequently describes the experiments. Section 4 describes the validation of the Phast steady-state discharge model DISC and the Phast time-varying discharge model TVDI. Section 5 outlines the validation of Phast dispersion model UDM adopting the source-term data derived from DISC and TVDI. The reader is referred to the detailed data review reports by Witlox (2012a, 2012b) for further detailed results not included in the current paper.
2. Overview of Phast modelling of discharge and dispersion for CO2 releases Fig. 1 includes a schematic phase diagram for CO2; CO2 has a critical temperature of 31.06 C (304.2 K) above which Phast presumes it to be always vapour and a triple point of 5.1 atm and 56.55 C (216.6 K) below which all non-vapour CO2 will be solid. Phast examines the progress of a CO2 release from the initial release to far-field dispersion including the modelling of solid rainout and subsequent sublimation to vapour. The main areas for modelling of CO2 in Phast as shown in Fig. 2 are as follows: Discharge modelling of CO2 which includes atmospheric expansion of CO2 (depressurisation to ambient pressure) during which liquid to solid/vapour expansion occurs. In case of initial supercritical temperature (above 31 C), vapour to vapour, or vapour to solid/vapour expansion occurs. The applied Phast discharge models are DISC (steady-state releases) and TVDI (time-varying releases). Starting from the specified vessel stagnation conditions, the discharge model DISC/TVDI is used for modelling the discharge of the CO2. This includes expansion from storage conditions to orifice conditions, and the expansion from orifice to ambient conditions. For the latter expansion the DISC/TVDI sub-model ATEX is used. Dispersion modelling involving the possible presence of solid CO2 in addition to vapour CO2. The ATEX post-expansion
Critical point
Pressure (atm)
fusion line vaporisation line
SOLID VAPOUR 5.1
sublimation line
Triple point
1 -78.4
-56.6
The UDM model invokes a thermodynamics sub-model for mixing of the released material and the ambient air. This model calculates the phase composition and temperature of the mixture at the cloud centre-line. For the BP and Shell CO2 experiments the stagnation pressures are very large and therefore the initial solid particle following depressurisation to ambient pressure is expected to be very small (initial fine solid mist of CO2). Furthermore the atmospheric sublimation temperature is very low (78.4 C) and therefore the solid particles are expected to sublime very quickly. As a result for the mixing of solid/vapour CO2 with air, the UDM thermodynamics sub-model assumes homogeneous equilibrium without deposition of the solid CO2 onto the substrate. Thus trajectories of solid particles are not modelled. The latter assumption was further verified by a detailed sensitivity analysis by Witlox, Stene, et al. (2011). The reader is referred to Witlox et al. (2009) for further details of the modelling. 3. BP and Shell experiments Experiments involving pressurised CO2 releases were carried out at Spadeadam by GL Noble Denton (previously Advantica) for BP in 2006 and for Shell in 2010. The data from these experiments along with other material was transferred into the DNV led CO2PIPETRANS JIP. DNV Software was commissioned by the JIP to undertake a critical review of the data to assess its suitability for public release for model validation, i.e. those corresponding to horizontal non-impinging releases, before the data from these tests were approved for external release. The data review for the BP experiments (Witlox, 2012a) was carried out based on the Advantica report by Evans and Graham (2007) and the overview report by Holt (2012) as well as some supplementary information provided by the original 2006/2007 BP model validation exercise. The data review for the Shell experiments (Witlox, 2012b) was based on the GL Noble Denton report by Allason and Armstrong (2011). In the experiments the CO2 was stored under high pressure in a horizontal cylindrical vessel. The modelled experiments include three sets of experiments: - Cold steady-state releases (liquid storage; BP tests 1, 2, 3, 5, 6, 11 and Shell tests 3,5,11). - Time-varying cold releases (liquid storage; Shell tests 1,2,4) - Hot supercritical releases (dense vapour storage; BP tests 8, 8R, 9 and Shell tests 14,16).
LIQUID
72.9
conditions are used as the source term (starting condition) for the UDM dispersion model. The UDM calculates the CO2 dispersion further downwind ignoring possible deposition on the ground and re-sublimation. The UDM assumes that the release direction is in the same vertical plane as the wind direction.
31.1
Temperature (oC) Fig. 1. Schematic phase diagram for CO2 (not on scale).
For the steady-state tests nitrogen padding gas was used to maintain the pressure in the vessel and to keep the test vessel full of liquid CO2. It should be noted that care was taken to ensure the CO2 being released was not contaminated by the nitrogen. For the timevarying tests the vessel was first filled with CO2 at the required test pressure and test temperature, where in case of hot releases the CO2 was heated using heating pads. Subsequently the CO2 was released through the nozzle driven only by the pressure in the vessel with the vessel pressure decaying as the release progressed. Fig. 3 depicts the release system. Downstream of the vessel a 3 m horizontal flexible hose was attached (200 inner diameter), connected to a 2 m 200 metering spool (BP) or a 300 coriolis meter (Shell) and a 0.5 m 200 nozzle with an orifice plate bolted on the
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
leak orifice
vessel (stagnation)
pipe
3
atmosphere
expansion zone
CO2 plume
vapour-plume centre-line
flow
SUBSTRATE Fig. 2. Discharge modelling (DISC/TVDI/ATEX) and dispersion modelling (UDM).
nozzle. Thus the total length of attached pipe is 5.5 m, with no external insulation applied to the pipe. A wide range of experiments was carried out including variation of orifice diameters (approximately ¼00 ,½00 or 100 ), storage pressures (82e158 barg) and temperatures (0 Ce150 C). Pressure and temperature measurements were made at a range of locations including vessel inlet, vessel outlet and nozzle spool; see references given above for further details. Table 1 summarises the key BP experimental data required as input to the Phast models. In this table the values of the storage pressure and the storage temperature are taken at the discharge end of the vessel (upstream of the pipework), with mean values during the release applied for the steady-state liquid releases and with initial values applied for the transient vapour releases. The ambient data were measured upwind of the release and mean values are adopted for these data during the release. This is with the exception of the wind-speed measurement of 1.65 m above the pad, which was taken 40 m downwind of the release. Since this measurement was disturbed by the CO2 jet, the value listed in Table 1 corresponds to the mean value prior to the release. Likewise Table 2 summarises the key Shell experimental data required as input to the Phast models. In this table the values of the storage pressure and the storage temperature are taken at the discharge end of the vessel (upstream of the pipework) and the nozzle pressure/temperature are taken along the nozzle, with mean values during the release applied for the steady-state releases and with initial values applied for the transient releases. The wind
Fig. 3. Photographic view of release system. Figure corresponds to BP tests; the release system was similar for the Shell tests with a 300 coriolis flow meter fitted upstream of valve V2; see Evans and Graham (2007) and Allason and Armstrong (2011) for further details of the release system.
speed data were taken from tower A [20 m west (behind) and 5 m south of release point; see Fig. 5] at 10 m height (averaged prior to release) given anomalies observed at other measurement locations. Furthermore, based on an analysis of the experimentally observed vertical wind-speed profiles a surface roughness of 0.1 m and a stability class of D was assumed for all (BP and Shell) tests. Finally with respect to the wind direction it is noted that the release direction corresponds to 270 . 4. Validation of Phast discharge models AGAInst CO2 experiments For the Shell experiments, the flow rate and density were measured accurately using a coriolis flow meter. In addition for both the Shell and BP experiments, the vessel weight was measured using load cells. Thus for the BP supercritical vapour releases, the flow rate was derived from the measured vessel weight using load cells. For the BP cold steady-state liquid releases, the flow rate was estimated by Advantica (Evans & Graham, 2007) from the load cells by assuming that the total flow rate dM/dt (kg/s; derived from the vessel mass M as measured by the load cells) satisfies dM/ dt ¼ rCO2VCO2 rN2VN2. Here rCO2 is the CO2 density (kg/m3), VCO2 the CO2 volume rate (m3/s), rN2 the nitrogen density (kg/m3), and VN2 the nitrogen volume flow rate (m3/s). Pressure and temperature were measured at a range of locations upstream of the vessel, inside the vessel, and downstream of the vessel along the pipe and the release valve. The Phast discharge models either assume the release to be directly from an orifice from a vessel (‘Leak’ scenario), or from a short pipe attached to a vessel (with orifice diameter ¼ pipe diameter, i.e. full-bore rupture). Except for the 100 orifice tests (BP test 5 and Shell tests 2,5), the observed pressure at the discharge end was seen to be very close to the observed pressure at the vessel inlet and vessel outlet. Thus the Phast ‘Leak’ scenario was applied, while neglecting the pressure loss from the stagnation conditions to the nozzle conditions. Also the 100 orifice tests can be modelled using the ‘Leak’ scenario, provided that measured nozzle pressures/ temperature are specified as model input instead of storage pressure/temperature (at vessel outlet). The Phast discharge model DISC was used to simulate the steady-state liquid releases, while the Phast discharge model TVDI was used to model the time-varying releases. Default Phast parameters were applied with two exceptions (for two parameters specified in Phast 6.7 and 7.01 in the ‘Discharge parameter’ tab). First the metastable assumption (non-equilibrium with liquid ‘frozen’) was not applied, but flashing was allowed at the orifice (equilibrium at the orifice) to account for the pipework upstream of the orifice. Secondly conservation of momentum was applied for the expansion from orifice to post-expansion conditions, since this
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
4
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Table 1 Experimental conditions for BP CO2 tests. Input Discharge data Steady-state/transient Storage phase Storage pressure (barg) Storage temperature ( C) Vessel volume (m3) Orifice diameter (mm) Orifice length (mm) Release duration (s) Ambient data Ambient temperature ( C) Ambient pressure (mbara) Relative humidity (%) Wind direction (degrees) Wind speed (m/s)
Test1
Test2
Test3
Test5
Test6
Test11
Test8
Test8R
Test9
Input for models
Steady Liquid 103.4 5 e 11.94 46.78 60
Steady Liquid 155.5 7.84 e 11.94 46.78 59
Steady Liquid 133.5 11.02 e 11.94 46.78 60
Steady Liquid 157.68 9.12 e 25.62 72.41 40
Steady Liquid 156.7 9.48 e 6.46 47.79 120
Steady Liquid 82.03 17.44 e 11.94 46.78 58
Trans. Vapour 157.76 147.12 6.3 11.94 46.78 120
Trans. Vapour 148.7 149.37 6.3 11.94 46.78 132
Trans. Vapour 154.16 69.17 6.3 11.94 46.78 179
e DISC,TVDI DISC,TVDI DISC,TVDI TVDI DISC,TVDI e e
14.2 999.4 74.4 322.4 4
7.5 958.2 96 265.6 3.44
10.6 972.5 95.8 288.8 3.37
5.8 985.4 96.7 278.6 5.13
6.1 938.4 1 299 2.20
11.6 960.2 94 270.8 5.99
11.19 957.99 100 269.3 4.71
11.1 957.1 100 270 0.76
8.2 958.9 99.9 270.7 4.04
DISC,TVDI,UDM DISC,TVDI,UDM DISC,TVDI,UDM UDM uses 270 UDM
assumption was previously found to provide the most accurate concentration predictions [e.g. against the SMEDIS experiments; see the UDM validation manual (Witlox, Harper, & Holt, 2011) for details]. 4.1. Time-varying releases Table 3 shows that the measured initial CO2 density derived from the coriolis flow meter for the time-varying Shell tests is very close to values predicted by the SpaneWagner Equation of State (SW EOS) in line with recommendations of a report by EON (2011). Also accurate predictions are obtained of the liquid density in Phast using the (non-default) PengeRobinson Equation Of State (PR EOS). The default method in Phast presumes the liquid density equal to the saturated liquid density at the given temperature, i.e. independent of the pressure. This was shown to result in accurate predictions of the liquid density after depressurisation to saturated conditions, but to lead to a significant under-prediction of the initial liquid density (i.e. total initial vessel mass). Table 3 also demonstrates that improved predictions are obtained of the initial vapour density for the hot tests 14 and 16 using the PR EOS. Fig. 4 includes TVDI-predicted (default Phast density) and observed values for flow rate (kg/s) versus time (s) for the timevarying releases. The thick solid lines represent the experimental data. The other lines represent TVDI predictions, while allowing flashing at the orifice. For the liquid releases, Fig. 4a includes TVDI predictions using both the default EOS and (using a new more robust development
version of TVDI) the PR EOS. The default EOS predicts both the initial flow rate and the flow rate after depressurisation to saturated conditions quite accurately. However less accurate results are obtained during the regime of depressurisation to saturated, with a too rapid decrease of flow rate caused by the usage of the too low saturated liquid CO2 density. The PR EOS using a more accurate CO2 density provides more accurate predictions of the flow rate for all regimes. For the hot vapour releases, Fig. 4b includes predictions of the current TVDI model (version 6.7) using both the default EOS and the PR EOS. For the BP tests, the observed values for the flow rates are averaged over a period over 8 s to reduce oscillations caused by inaccuracies of the load-cell measurements. This was not necessary for the coriolis flow meter measurements. The following can be concluded from Fig. 4b: - For the very hot BP tests 8, 8R (storage temperature about 150 C), the vapour remains vapour upon depressurisation (condensation not relevant), and it is seen that very close agreement is obtained between TVDI predictions and observed data using both the default and PR EOS, since they both provide very accurate predictions of CO2 vapour density. - For the BP test 9 and the Shell test 14 (temperature about 70 C), PR EOS is seen to produce most accurate results. Furthermore the default-EOS TVDI runs are seen to terminate prematurely, which was due to convergence problems apparently caused by the release temperature being lower and closer to the critical temperature. An under-prediction of the flow rate is seen at larger times.
Table 2 Experimental conditions for Shell CO2 tests. Input Discharge data Steady-state/transient Storage phase Storage pressure (barg) Nozzle pressure (barg) Storage temperature ( C) Nozzle temperature ( C) Vessel volume (m3) Orifice diameter (mm) Orifice length (mm) Release duration (s) Ambient data Ambient temperature ( C) Ambient pressure (mbara) Relative humidity (%) Wind direction (degrees) Wind speed (m/s)
Test3
Test5
Test11
Test1
Test2
Test4
Test14
Test16
Input for models
Steady Liquid 147.3 144.8 9.8 8.2 e 12.7 47.78 120
Steady Liquid 148.8 126.4 17.8 13.7 e 25.4 46.84 40
Steady Liquid 81.9 80.3 0.2 1.4 e 12.7 47.78 120
Trans. Liquid 148.3 143 26.7 23 6.3 12.7 47.78 90
Trans. Liquid 147.1 118 24.6 18 6.3 25.4 46.84 145
Trans. Liquid 148.2 148.2 20.1 20.1 6.3 6.3 47.79 >700
Trans. Vapour 151.6 147.7 71 65.0 6.3 12.7 47.78 315
Trans. Vapour 150.6 146.0 36.7 31.7 6.3 12.7 47.78 370
e DISC,TVDI DISC,TVDI (either storage or nozzle pressure is input) DISC,TVDI (either storage or nozzle temperature is input) TVDI DISC,TVDI e e
11.2 1017 66 267 4.05
9 905 91 213 1.30
3.6 995 78 261 2.76
14.7 1006 83 263 3.93
10.3 1005 77 250 5.43
13.8 975.5 77 215 1.98
0 1005 88 303 1.34
2.9 997 88 292 1.48
DISC,TVDI,UDM DISC,TVDI,UDM DISC,TVDI,UDM UDM uses 270 UDM
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 Table 3 Predicted versus observed initial CO2 density for time-varying Shell tests. Input
Test1 Test2 Test4 Test14 Test16 Comments
Storage phase Storage pressure (bara) Storage temperature ( C) Predicted and measured density Phast default density (kg/m3) Phast PR density
Liquid Liquid Liquid Vapour Vapour 149.3 148.1 149.2 152.6 151.6 ¼ gauge þ ambient pressure 26.7 24.6 20.1 71 36.7
686
719
773
445
704
854
868
901
476
779
SpaneWagner density (kg/m3)
866
878
903
507
803
measured density (kg/m3)
890
919
907
493
826
Phast (default method) Phast (modified method) derived via interpolation of data in EON report measured by coriolis flow meter
- For Shell test 16 the above effects are seen to be even more pronounced, since the initial storage temperature is only a few degrees above the critical temperature. At larger times the vessel fluid may become liquid, but the transition from vapour to liquid is not modelled by TVDI resulting in under-prediction.
4.2. Initial rate for steady-state and time-varying releases 4.2.1. BP tests Table 4 summarises the overall results of the discharge rates for all BP tests. For the steady-state tests only the DISC initial release rate is given, while for the time-varying releases also the TVDIpredicted averaged release rate over the first 20 s is indicated. It is noted that the difference between the averaged and initial rate is relatively small. From the table it is seen that the timevarying Phast predictions align well with the observed discharge rate for the hot tests 8, 8R and 9. The predicted flow rate for the cold releases, with the exception of test 5 (100 release), is also very close to that of the experiments. For test 5 (100 release) the flow rate is over-predicted with 23% (50.74 kg/s predicted versus 41.17 kg/s experimental) using the ‘Leak’ scenario, while using the pipe (‘Line Rupture’) scenario it is under-predicted with 34.5% (26.95 kg/s predicted versus 41.17 kg/ s). The over-prediction for the orifice scenario is believed to be caused by the fact that pressure loss is ignored along the pipework (hose/spool/nozzle). Test 5 has the largest orifice diameter (100 ) and therefore will be most susceptible to upstream pressure loss and reduced flow rate. Indeed if a more accurate pressure would be applied of 128.6 barg (corresponding to averaged observed pressure close to the orifice) a release rate of 45.34 kg/s is predicted using the ‘Leak’ scenario corresponding to a much smaller overprediction of 10.1%. For the other ¼” and ½” tests very little difference between vessel-outlet and nozzle pressures was observed, and this is confirmed by Table 4 which shows accurate flow-rate predictions for these tests (within experimental accuracy). The DISC output data reported in Table 4 are derived in Phast from the ‘Discharge report’ (where ‘liquid mass fraction’ should be interpreted for CO2 as ‘solid mass fraction’). The key difference in discharge model input between tests 8/8R, test 9, and test 2 is the storage temperature (around 148C versus 69C versus 8C). It is seen from the post-expansion data reported in Table 4 that the lower temperature results in increasing two-phase solid mass fraction (0 versus 15% versus 40%), a larger density, a larger flow rate (around 4 versus 6 versus 11 kg/s) and a smaller release velocity (around 470 versus 290 versus 190 m/s).
5
As indicated above the flow rate changes little for the timevarying tests 8, 8R, 9 within the first 20 s, and it is believed that within 20 s the maximum concentrations will be achieved within the first 80 m (given relatively large initial jet momentum and relatively large values of wind speed). Therefore in the next section the UDM dispersion calculations are modelled as steady-state using the averaged flow rate over the first 20 s for tests 8, 8R and 9, while for the other tests the overall averaged observed value is adopted. All other UDM source-term input data (temperature, solid fraction, velocity) are chosen as predicted above by the discharge model DISC. 4.2.2. Shell tests Table 5 summarises the overall DISC predictions of the initial discharge rates for all Shell tests. In this table a range of model assumptions is applied: - Time-varying releases are calculated either based on measured initial storage (vessel outlet) pressure/temperature or measured initial nozzle pressure/temperature; as expected, particularly for the largest 100 orifice size (test 2), usage of nozzle data significantly improves the predictions given the significant pressure decay between storage and nozzle conditions; for the smallest ¼00 orifice size identical results are obtained because of negligible pressure decay. The DISC flow-rate predictions for the transient hot releases based on the initial values of temperature and pressure at the nozzle are seen to be larger than those based on the higher values of temperature and pressure at the vessel discharge end. This is because a lower DISC input temperature results in a larger solid fraction and a larger flow rate prediction, and this effect appears to be dominant to the reduction of flow rate due to the lower DISC input pressure. - Phast liquid density either based on default (saturated density) or more accurate Peng Robinson density, with more accurate results obtained using the PR EOS - Flashing (non-default Phast) or non-flashing (default Phast; metastable liquid assumption). Using PengeRobinson (PR) density, this is seen to affect results very little. Using the saturated density, the default non-flashing option provides conservative results while the non-default flashing assumption produces significantly more accurate results. Table 6 includes results and UDM source terms based on those model assumptions which would provide the best agreement against the experimental data, i.e. based on the PR EOS, the flashing assumption, and nozzle data for liquid releases and storage data for hot releases. Note that unlike the BP experiments, the UDM dispersion data are chosen to depend on the initial rate rather than on the averaged rate during the first 20 s. This choice was made, since as discussed later the concentration sensors only measured accurately the initial concentration, and not the concentration at subsequent times. 5. Validation of Phast dispersion model against CO2 experiments The CO2 concentration was largely measured via O2 cells for both BP and Shell experiments; see Fig. 5 [taken from Allason and Armstrong (2011)] for the location of the O2 concentration sensors. Thus a total of 43 sensors was applied at downstream distances of 5 m (sensor OC01), 10 m (OC02), 15 m (OC03), 20 m (OC04-OC08), 40 m (OC9-OC21), 60 m (OC22-OC28) and 80 m (OC29-OC43), with sensors position at a range of different heights (0.3, 1 or 3 m) and
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
6
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
(a)
BP Test 8 - experiment (load cell - 8s averaged)
(b)
BP Test8 - TVDI (default EOS) BP Test8 - TVDI (PR) BP Test 8R - experiment (load cell - 8s averaged) BP Test8R - TVDI (default EOS) BP Test8R - TVDI (PR) BP Test 9 - experiment (load cell - 8s averaged) BP Test9 - TVDI (default EOS, flashing) BP Test9 - TVDI (PR, flashing) Shell Test14 - experiment (coriolis flow meter) Shell Test14 - TVDI (default EOS, flashing) Shell Test14 - TVDI (PR, flashing) Shell Test16 - experiment (coriolis flow meter) Shell Test16 - TVDI (default EOS, flashing) Shell Test16 (TVDI PR, flashing)
Fig. 4. TVDI validation of flow rate for time-varying releases. (a) Cold liquid releases (Shell tests 1,2,4). (b) Hot vapour releases (BP tests 8, 8R, 9 and Shell tests 14, 16).
Table 4 Predicted versus observed flow rates and UDM source-term data (BP tests). Test1 Discharge rate DISC initial discharge rate (kg/s) DISC/TVDI discharge rate (kg/s) (averaged over first 20 s for tests 8.8R,9) Observed discharge rate (kg/s) (averaged over first 20 s for tests 8.8R,9) Deviation predicted from observed Final (post-expansion) state (UDM input) Discharge rate (kg/s) (from experiments) Temperature (K) (DISC output) Solid fraction () (DISC output) Velocity (m/s) (DISC output)
8.84 8.84 e 7.8% 8.2 194.6 0.397 156.7
Test 2 10.98 10.98 11.41 3.9% 11.41 194.1 0.403 189.8
Test3 9.988 9.988 9.972 0.16% 9.988 194.26 0.384 179.2
Test 5 50.75 50.75 41.17 þ23% 41.17 194.4 0.399 191.7
Test6
Test 11
Test 8
Test 8R
Test 9
3.21 3.21 3.50 8.2%
7.03 7.03 7.12 1.1%
4.19 4.01 4.07 1.5%
3.90 3.73 3.80 1.8%
6.86 6.25 6.05 þ3.4%
3.50 193.8 0.397 191.3
7.12 194.1 0.330 154.2
4.07 198.2 0 466.5
3.80 204.8 0 472.8
6.05 194.1 0.154 289.0
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
7
Table 5 Predicted versus observed flow rates e vary Phast assumptions (Shell tests). Prescribed pressure/temperature
Mean nozzle
Initial nozzle
Release type
Steady-state liquid
Transient liquid
Shell CO2 test number
Test 3
Predicted flow rate (kg/s) - Default density, flashing - Default density, no flash. - PR density, flashing - PR density, no flashing Observed flow rate Ratio predicted/observed - Default density, flashing - Default density, no flash. - PR density, flashing - PR density, no flashing
Test5
Test11
Test1
Test2
Initial storage (vessel discharge end)
Test4
Transient hot
Transient liquid
Test14
Test1
11.93 13.58 12.37 12.16 12.4
43.38 50.88 44.36 43.92 44.7
8.89 10.21 9.10 9.29 8.9
11.23 13.61 11.38 11.12 10.55
40.90 49.48 41.26 41.15 38
2.85 3.39 2.92 2.85 3.17
7.82 7.46 7.67 7.71 7.37
96.2% 109.5% 99.8% 98.0%
97.1% 113.8% 99.2% 98.3%
99.9% 114.7% 102.2% 104.4%
106.4% 129.0% 107.9% 105.4%
107.6% 130.2% 108.6% 108.3%
90.0% 106.9% 92.0% 89.8%
106.1% 101.2% 104.1% 104.7%
Test16 10.92 Error 10.88 Error 10.5 104.0% Error 103.6% Error
Test2
Transient hot Test4
Test14
11.29 13.95 11.40 11.06 10.55
45.36 55.30 45.94 44.71 38
2.85 3.39 2.92 2.85 3.17
7.33 7.22 7.45 7.45 7.37
107.0% 132.2% 108.0% 104.8%
119.4% 145.5% 120.9% 117.7%
90.0% 106.9% 92.0% 89.8%
99.5% 97.9% 101.1% 101.1%
Test16 10.87 Error 10.74 Error 10.5 103.6% Error 102.3% Error
Table 6 Predicted versus observed flow rates and UDM source-term data (Shell tests). Prescribed pressure/temperature
Mean nozzle
Initial nozzle
Release type
Steady-state liquid
Shell CO2 test number
Test 3
Discharge rate DISC initial discharge rate e PR,fl. (kg/s) Observed rate (kg/s) (initial rate for transient tests) Deviation predicted from observed Final (Post Expanded) State (UDM input) Discharge rate (kg/s) (from experiments) Temperature (K) (DISC output) Solid fraction () (DISC output) Velocity (m/s) (DISC output)
12.37 12.4 0.24% 12.4 194.82 0.40 176.25
Test5 44.36 44.7 0.77% 44.7 193.37 0.37 170.94
Initial storage
Transient liquid Test11 9.10 8.9 2.25% 8.9 194.55 0.42 132.23
Test1 11.38 10.55 7.86% 10.55 194.69 0.34 187.93
Test2 41.26 38 8.59% 38 194.67 0.35 170.80
Transient hot Test4 2.92 3.17 8.03% 3.17 194.30 0.36 187.48
Test14
Test16
7.45 7.37 1.08% 7.37 194.67 0.15 292.91
10.74 10.5 2.26% 10.5 194.57 0.29 208.16
Fig. 5. Field detector array for concentration measurements. Figure corresponds to Shell tests, but concentration sensor location is also applicable to BP tests.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
8
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Fig. 6. BP Test 11 e observed raw and time-averaged concentration data. (a) Sensor OC16 (40 m downstream). (b) Sensor OC03 (15 m downstream). (a) Sensor OC01 (5 m downstream).
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
9
Fig. 7. BP Test 11 e UDM validation for maximum contraction versus distance.
cross-stream distances (between 20 and þ20 from the release direction) with two additional Servomex CO2 analysers. Phast assumes that the release direction is the same as the wind direction, while for some of the experiments (see Tables 1 and 2) there is a significant deviation from the wind direction. This may lead to less accuracy of the predictions in the far-field but will not significantly affect the prediction for the momentum-driven dispersion in the near-field.
raw concentrations for sensors OC01, OC03 and OC16 locations at 5, 15 and 40 m downstream distances along the release axis and at 1 m height. In addition it includes observed concentrations timeaveraged over 11 s, 21 s and 59 s. Here 59 s approximately corresponds to the release duration (reported as 60 s). Furthermore for a given averaging period tav, the time-averaged observed concentration cav ðtÞ is evaluated as function of time t from the raw obs observed concentration cav ðtÞ as follows: obs
5.1. BP tests (270.8 )
For the steady-state test 11 the averaged wind direction is very close to the release direction (270 ). Fig. 6 includes observed
cav obs ðtÞ
1 ¼ tav
tþt Z av =2
craw obs ðtÞdt ttav =2
Fig. 8. BP Test 9 e UDM validation for maximum concentration versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
10
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 20
Concentration (%CO2)
15 Draeger sensor (5m downstream, 1m high) O2 sensor OC01 (5m downstream, 1m high)
10
5
0
-5 0
20
40
60
80
100
120
140
160
180
200
220
240
Time (s)
Fig. 9. Shell Test 11 e Draeger versus O2 sensor predictions.
Fig. 6a shows that the concentration fluctuations are relatively small with respect to the mean concentration, and therefore a relatively accurate measurement of the concentration can be provided. This is also because the jet centre-line will pass sensor OC1 very closely. In theory (so close to the release point) the concentrations should be approximately constant over a period of 60 s (roughly between 75 s and 135 s). There is a relative small spread between the maximum value for 11-s averaged concentration (21.15 mol %) and the 59-s averaged concentration (18.79%). The subsequent figures Fig. 6b and c however show that the relative differences increase with increasing distance from the source. This is partly because the plume centre-line is more likely to miss the sensors at distances further downstream (because of fluctuating wind direction, as confirmed by wind direction variation observed by Advantica), and also because the sensor readings become relatively less accurate further downstream. For small concentrations, the sensor seems to have an accuracy of no better than 0.5%. Thus measurements with average concentrations less than 1% are considered to be of less value (except to confirm that the concentrations are small). Fig. 6c shows that the
concentration away from the plume is erroneously ‘negative’, and one may therefore consider to re-calibrate the observed concentrations (i.e. increase all measured values with the negative minimum value). However this has not been carried out as part of the current work. Fig. 7 plots for test 11 the maximum values over time of the measured concentration along with the Phast predicted concentrations as a function of downstream distance. The measured data include the maximum concentration of the raw data over all times, 11-s, 20-s and 59-s averaged concentrations. For the measured data at a given downstream distance the maximum value of all sensors at that distance is taken, Sensor 14 (located at 40 m downstream, 3 m height) has been excluded since it appeared to give erroneous too high readings (higher than sensors at 1 m height and sensors further upstream). Furthermore no further analysis has been carried out (e.g. via spline fitting of the measured values to obtain a better fit of the crosswind concentration profile and a better estimate of the maximum concentration) to further refine this maximum value. The Phast predictions were found not to be affected by time-averaging effects due to plume meander (transition to passive dispersion occurring downwind of 80 m). In the near field (<20 m) the 59-s averaged concentration predicted by Phast is close to the measured concentrations. This is also in line with UDM validation against previous experiments, where very close agreement was obtained in the near-field, jet-momentum dominated regime. Further downstream (at 20 m and 40 m) it is seen that the spread in the measured concentrations becomes larger with a larger effect of averaging. This is because of (a) larger relative inaccuracy of the sensors, and (b) the CO2 plume centre-line more likely to be further away from the sensor (also because of plume meander). Thus for this case, as is clearly illustrated by Fig. 7, the maximum value would lead to ‘too’ large (rather random) value of the maximum concentration (it would increase with the release duration), while on the other hand the 59-s averaged concentration may lead to too small values. Fig. 8 includes results of UDM validation for maximum concentration versus downstream distance for the time-varying test 9 (vapour release). It is again seen that good agreement with the processed averaged experimental data is obtained. For this test,
Fig. 10. Shell Test 11 e UDM validation for maximum contraction versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
11
Fig. 11. Shell Test 16 e UDM validation for maximum concentration versus distance.
sensors 17 and 14 were considered to give possible incorrect readings for similar reasons to sensor 14 in test 11. 5.2. Shell tests For the Shell tests, a limited number of 3rd party commercial CO2 detectors including instruments from Draeger as well as two Servomex gas analysers were used in addition to the O2 sensors in order to verify the accuracy of the O2 sensors. From the results of this it was deduced that the O2 sensors did reasonably predict the initial (maximum) value of the concentration, but did subsequently show an erroneous decay with time which could have
been caused by significant cooling of the sensors; see Fig. 9. For unknown reason this was observed for the Shell tests only and not the BP tests. See the data review by Witlox (2012b) for further details. Figs. 10e12 plot for tests 11 (steady-state cold release), 16 (transient hot release) and 1 (transient liquid release) the maximum values over time of the measured concentration along with the Phast predicted concentrations as a function of downstream distance. The measured data include maximum values and (for steady-state test 11 only) averaged values (over release duration) for the O2 sensors, Servomex sensors, and (if present) Draeger sensors.
Fig. 12. Shell Test 1 e UDM validation for maximum concentration versus distance.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
12
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13
Overall it is seen that the maximum O2 values agree well with the UDM predictions, where negligible difference was observed between UDM predictions at 1 m release height and centre-line (C/ L) height. Because of erroneous decay with time as illustrated in Fig. 9, the averaged O2 values result in too low observed values for Test 11. The maximum concentration derived from the Draeger sensors is reasonably aligned with that derived from O2 sensors, but it is particularly less accurate in the far-field because of an insufficient number of sensors (and thus the Draeger sensors may miss the centre-line of the plume).
For the Shell experiments, maximum concentrations for the O2 sensors were used (as discussed in Section 5.2), and none of the O2 sensors was ignored even though they may provide a less accurate reading. This may have caused the overall under-prediction for the Shell experiments. Furthermore for the steady-state releases a higher accuracy is obtained than for the BP experiments, because of (a) input of more accurate measured flow rate and (b) use of conservative 11-s average estimate (maximum over all times during release duration) for the BP experiments. 6. Conclusions and future work
5.3. Comparison statistics between predicted and observed concentrations For a given experimental dataset, it is common practice (Hanna, Strimaitis, & Chang, 1991) to calculate the geometric mean bias MG (averaged ratio of observed to predicted concentrations; MG < 1 over-prediction and MG > 1 under-prediction) and the geometric variance VG (variation from mean; minimum value ¼ 1). Ideally, MG and VG would both equal 1.0. Geometric mean bias (MG) values of 0.5 and 2.0 can be thought of as a factor of 2 in over-predicting and under-predicting the mean, respectively. Likewise, a geometric variance (VG) of about 1.6 indicates scatter from observed data to predicted data by a factor of 2. Table 7 includes the predictions of MG and VG for the BP and Shell experiments. It is noted that all MG values are well within the range of [0.5, 2], and all variances less than 1.6 which is normally considered to be excellent agreement with the experimental data. In Table 7 the observed maximum concentration at a downstream distance is taken as the maximum value of all sensors at that downstream distance. For the BP experiments, the maximum value over all times of the 11-s averaged concentrations has been applied and sensors of apparent false readings have been ignored as discussed in Section 5.1. Therefore conservative estimates are obtained of the averaged observed concentrations for the cold releases (1, 2, 3, 5, 6, 11), which may (partly) explain the under-prediction of the concentrations for the experiments 2, 3, 5, 6. For tests 1, 3, 6 there was a significant difference between the wind direction (averaged over the entire release duration) and the release direction. However the above results show that the plume centre-line did not significantly miss the sensors. Further downstream this may have been caused because we adopt 11-s averaged concentrations (maximum overall all times) rather than concentrations averaged over the entire release duration. Furthermore it must be noted that for tests 3 and 6 a 200 1.44 m extension tube was attached downstream to the ½00 (test 3) and ¼00 (test 6) nozzle, which is not expected to affect the discharge flow rate but is likely to have affected the dispersion. This may explain the largest under-prediction of the concentrations (largest MG values) for tests 3 and test 6.
This paper described the validation of the Phast discharge and dispersion models against BP and Shell experimental data made available via the CO2PIPETRANS JIP. The experiments involve pressurised CO2 releases involving steady-state and transient cold liquid releases, and time-varying supercritical hot vapour releases. The steady-state cold releases were modelled by the Phast discharge model DISC as steady-state orifice releases, while the Phast discharge model TVDI was used to model the time-varying orifice releases. The flow rate was accurately predicted for both the BP and Shell tests (within about 10%), which was deemed to be within the accuracy at which the BP experimental data were measured. Particularly for temperatures closer to the critical temperature, more accurate and robust behaviour was obtained using the PR EOS instead of the default EOS. The releases were all modelled by the Phast dispersion model UDM as steady-state releases, with 20-s averaged flow rates applied for the time-varying BP releases and with the initial maximum flow rate applied for the time-varying Shell releases. For all cases the solid carbon dioxide was found to sublime rapidly and no fallout was predicted, which was fully in line with the experiments. The concentrations were found to be predicted accurately (well within a factor of two). No fitting of the extended Phast models has been carried out whatsoever as part of the above validation. More CO2 experiments are being planned (as part of the joint industry projects CO2PIPETRANS and COSHER) involving pipeline releases instead of orifice releases, and these experiments could be considered for future model validation. Acknowledgements BP is acknowledged for providing funding for the initial validation of the BP DF1 experiments in 2007. BP and Shell are acknowledged for making the data available via the CO2PIPETRANS JIP. The help of Hamish Holt (CO2PIPETRANS JIP project manager) is acknowledged for providing input regarding the BP and Shell experiments. References
Table 7 UDM values of mean MG and variance VG for CO2 experiments. (a) BP tests
(b) Shell tests.
Test
Release type
MG
VG
Test
Type of release
MG
VG
1 2 3 5 6 11 8 8R 9
Steady-state cold Steady-state cold Steady-state cold Steady-state cold Steady-state cold Steady-state cold Transient hot Transient hot Transient hot
0.89 1.44 1.60 1.59 1.63 0.99 1.25 1.25 1.40
1.18 1.15 1.30 1.24 1.31 1.20 1.07 1.12 1.13
3 5 11 1 2 4 14 16
Steady-state cold Steady-state cold Steady-state cold Transient cold Transient cold Transient cold Transient hot Transient hot
1.39 0.98 1.17 1.08 1.06 1.06 1.21 1.35
1.12 1.03 1.17 1.02 1.03 1.01 1.07 1.28
Allason, D., & Armstrong, K. (2011). Liquid and supercritical carbon dioxide release and dispersion experiments on behalf of Shell International Exploration and Production BV, 19th September 2011, Report nr. 10793. Cumbria, UK: GL Noble Denton. Equation of state prediction of carbon dioxide properties. Report KCP-GNS-FAS-DRP0001, Rev.02, Kingsnorth Carbon Capture & Storage Project. UK: EON (2011). Evans, J. A., & Graham, I. (2007). DNV CO2PIPETRANS JIP e Data release 1-Advantica overview. June 2012, Extract from a confidential report by Advantica for BP, Cumbria, UK. Hanna, S. R., Strimaitis, D. G., & Chang, J. C. (1991). Hazard response modelling uncertainty (A quantitative method). Report for the API. Westford, USA: Sigma Research Corp. Holt, H. (June 2012). DNV CO2PIPETRANS JIP e Data release 1-DNV overview report. DNV. Holt, H., Helle, K., & Brown, J. (20e21 June 2012). Capturing and sharing CO2 release data: the CO2PIPETRANS JIP. Newcastle, UK (2012). In 3rd International forum on transportation of CO2 by pipeline.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006
H.W.M. Witlox et al. / Journal of Loss Prevention in the Process Industries xxx (2013) 1e13 Witlox, H. W. M. (2012a). “Data review and Phast analysis (discharge and atmospheric dispersion) for BP DF1 CO2 experiments”, Contract 96000056 for DNV Energy (CO2PIPETRANS Phase 2 JIP WP1). London: DNV Software. Witlox, H. W. M. (2012b). “Data review and Phast analysis (discharge and atmospheric dispersion) for Shell CO2 experiments”, Contract 984C0004 for DNV Energy (CO2PIPETRANS Phase 2 JIP WP1). London: DNV Software. Witlox, H. W. M., Harper, M., & Holt, A. (2011). “UDM validation manual”, Phast 6.7 technical documentation. London: DNV Software. Witlox, H. W. M., Harper, M., & Oke, A. (2009). Modelling of discharge and atmospheric dispersion for carbon dioxide releases. Journal of Loss Prevention, 22(6), 795e802. Witlox, H. W. M., Harper, M., & Pitblado, R. (April 2012). Validation of Phast dispersion model as required for USA LNG Siting Applications. LNG plant safety symposium, 12th topical conference on gas utilisation, AICHE Spring Meeting, Houston, USA.
13
Witlox, H. W. M., & Holt, A. (1999). A unified model for jet, heavy and passive dispersion including droplet rainout and re-evaporation. In International conference and workshop on modelling the consequences of accidental releases of hazardous materials (pp. 315e344). San Francisco, California: CCPS. September 28 e October 1. Witlox, H. W. M., & Holt, A. (2001). “Validation of the unified dispersion model against Kit Fox field data”, Contract 44003900 for Exxon Mobil. London (: DNV Software. Witlox, H. W. M., Stene, J., Harper, M., & Nilsen, S. (2011). Modelling of discharge and atmospheric dispersion for carbon dioxide releases including sensitivity analysis for wide range of scenarios. In 10th Int. Conf. On Greenhouse gas Control Technologies, 19-23 September 2010, Amsterdam, The Netherlands. Made available online via Energy Procedia (Vol. 4; pp. 2253e2260). Elsevier.
Please cite this article in press as: Witlox, H. W. M., et al., Phast validation of discharge and atmospheric dispersion for pressurised carbon dioxide releases, Journal of Loss Prevention in the Process Industries (2013), http://dx.doi.org/10.1016/j.jlp.2013.10.006