MINIMISING UNCERTAINTY IN VAPOUR CLOUD EXPLOSION MODELLING

Raghu Raman, Kellogg Brown & Root, Sydney, Australia Paolo Grillo, Kellogg Brown & Root Asia Pacific, Singapore

ABSTRACT

There have been significant advances in consequence modelling of releases of hazardous materials in the last decade, but the use of simple models for predicting overpressures from vapour cloud explosions (VCEs) continues to elude safety practitioners. The TNO multi-energy model has generally been regarded as the best model available to date, for a rapid assessment of explosion overpressures and positive phase durations. This model requires two major assumptions to be made - the level of congestion in the plant that provides obstacles for flame front acceleration, and the explosion efficiency. The former enables the selection of an appropriate charge strength from the family of parametric curves in the model. While there has been some guidance available for making relevant assumptions, there is still a high level of uncertainty in the model results, and the drag load obtained cannot be confidently applied as the basis for structural design. Flame acceleration models based on Computational Fluid Dynamics (CFD) have been developed for gas explosion modelling in offshore oil and gas facilities. Sensitivity analysis is conducted on these models, using various cloud sizes and ignition point locations, and the results are processed into an exceedence curve based on event frequencies, in order to obtain the best estimate of peak overpressures and drag loads for design purposes. In this paper, the TNO multi-energy model has been applied to two different configurations of topsides of offshore installations. In these applications the assumptions regarding explosion efficiency and charge strength selections have been calibrated against the results obtained through the CFD modelling, by obtaining a match of the explosion over-pressures. The paper comments on the findings from this exercise in order to give guidance on the application of the TNO multi-energy method and on the selection of model parameters with relation to the equipment lay-out and level of equipment congestion. 1. INTRODUCTION Vapour cloud explosions are one of the most devastating events which can occur in the process industries. It was recognised since the time of the Flixborough incident that a facility design should include limiting explosion damage (Lawrence and Johnson 1974). Since Piper Alpha, it is become common practice to design blast walls on offshore oil and gas facilities, to separate process areas from utility areas and living quarters. The determination of peak overpressures from gas explosions and development of design criteria for structural support become more complex due to high pressure inventories in congested areas. There are four key factors in an explosion. These are related to the overpressure which is the pressure rise above normal atmospheric pressure, the positive phase duration which is the time during which the pressure is above atmospheric pressure, the degree of confinement of the flammable mixture which causes turbulence and

2

acceleration of the flame front and influences the overpressure, and the impulse (area under the pressure-time profile). It is now well established that it is not the size of the vapour cloud that matters when it comes to blast strength, but the degree of confinement of the vapour cloud and congestion in the path of the flame front. In offshore modules, there is additional confinement in the form of parallel planes formed by deck plating. The energy of ignition source (e.g. naked flame) plays a dominant role in determining the blast strength, although a well designed facility with strict implementation of hazardous area classification requirements in terms of hardware and safety management system can reduce the strength of a potential ignition source significantly. The Multi-Energy Model (MEM) for rapid assessment of explosion overpressure has been developed by TNO (1997). It is based on the concept that significant overpressures can be generated by the ignition of a vapour cloud only in the presence of partial confinement or obstacles in the path of the flame front. This model, however, requires assumptions on the initial blast strength, which significantly influences the predictions. CFD models used in offshore modules have shown that rapid assessment models can underestimate the blast overpressures. CFD models give reasonably good accuracies, and yield solutions that are approximately correct. They are considered adequate for explosion assessment, given the uncertainties on the other stages of the overall design process (UKOOA 2003). CFD modelling of explosion in an offshore facility is generally undertaken only when piping design is well advanced. Since it is the cluster of small diameter piping, valves and other fittings that generates turbulence in the accelerating flame front, it is essential that the 3-D model of the topsides includes as much of small diameter piping as possible. On the other hand, very often information is required on the drag loads on equipment and structures from explosions, early in the detailed engineering phase. If increased confidence can be developed in the rapid assessment tools in providing as conservative a design blast load as CFD models, it would greatly assist in the design phase. This can be further verified by CFD models. In this paper, the TNO multi-energy model has been applied to two different configurations of topsides of offshore installations. In these applications the assumptions regarding explosion efficiency and charge strength selections have been calibrated against the results obtained through the CFD modelling, by obtaining a match of the explosion overpressures. The paper comments on the findings from this exercise in order to give guidance on the application of the TNO multi-energy method and on the selection of model parameters with relation to the equipment layout and level of congestion. 2. VAPOUR CLOUD EXPLOSION MODELS A brief outline of the rapid assessment model and CFD models is provided below, with comments on their limitations and strengths. Rapid assessment models prior to the TNO Multi-Energy Model (MEM) are not discussed as these do not account for

3

confinement or congestion that play such a significant role in determining the magnitude of overpressures. 2.1 TNO Multi-Energy Model The Netherlands Organization for Applied Scientific Research (TNO) has conducted extensive research into blast models (van den Berg et al. 1991; Mercx et al. 1998, 2000). The TNO multi-energy model allows for blast strength to be incorporated. The method considers the total cloud as a series of sub-explosions corresponding to the various confined or unconfined regions. The confined regions might be parts of the cloud located under equipment, hemmed in by other structures and pipework or parallel planes. The key factor in this model is the ‘blast strength’ which ranges from 1 (insignificant) to 10 (detonative strength). Figure 1 shows the overpressure versus an energy scaled distance. Separate curves relate to the blast strength representing the congestion parameter.

Figure 1: Scaled peak side overpressure for MEM (TNO 1997)

4

The following parameters describe the MEM:

Scaled overpressure: oP = (Po-Pa)/Pa Scaled distance: R = R (Pa/E)1/3

where: Po = side-on absolute blast overpressure (Pa) R = fuel-air charge radius (m)

oP = scaled blast overpressure (-) R = scaled distance (-) Pa = ambient pressure (Pa) E = combustion energy in fuel-air mixture (J) The model also estimates the positive phase duration. Relevant equations may be found in Mercx et al. (1998).

To apply this method the following steps need to be carried out.

a) Assign portions of the cloud to different areas (e.g. between buildings, under vessels, in open air)

b) Determine the fuel Vo present in each zone, i.e. amount of gas-air mixture (m3) and total cloud energy, E (≡ VoEc), based on the heat of combustion of the fuel. A typical energy density Ec is 3.5 × 106 J/m3.

c) Assign initial strengths to these vapour cloud charges, based on the degree of confinement.

d) Calculate scaled distances for each charge at nominated distances, R. e) From figures for oP versus R , estimate overpressure for each charge. If blast

zones are located close to each other and the ignition is essentially simultaneous, then overpressures can be superimposed at target distances.

The blast overpressure is often expressed in the units of bars gauge for convenience. The biggest challenge in the use of the multi energy method is the selection of the charge strength curve. This depends on a number of factors that include:

(i) the level of obstruction within the gas cloud.

(ii) the ignition strength, ‘high’ representing a vented explosion with ‘low’ being a spark or flame.

(iii) the level of confinement being either an unconfined volume or confined between surfaces.

A decision table can be constructed (TNO 1997) that considers all factors and relates these to the blast strength. Mercx et al. (2000) have provided a correlation for the charge strength Po. These were used to predict overpressures and were found to vary by a factor of 2 compared with CFD methods.

5

For offshore oil and gas facilities, Kinsella (1992) has suggested an approach for selecting the charge strength, accounting for three factors: • Congestion: Congestion may be defined as the fractional area in the path of the

flame front occupied by equipment, piping, fittings and other structures such as buildings and supporting columns. If congestion is more than the threshold of 30%, it is considered ‘high’.

• Strength of ignition source: Low (spark, hot surfaces), high (naked flames, welding)

• Parallel Confinement: Low for grated decks, high for plated decks. Kinsella provides an initial blast strength matrix based on a combination of the above parameters. For offshore modules, which have been analysed in this paper, the rule set is summarised in Table 1 (Kinsella 1992). The ignition source strength is assumed to be low based on design standards and safety work methods.

Table 1: Initial Blast Strength Assessment

Category High Congestion

Low Congestion

Parallel plane confinement

Unconfined (grated deck)

MEM Initial blast strength

1 X X 5-7 2 X X 4-5 3 X X 3-5 4 X X 2-3

The main advantages of the MEM are that a rapid estimate of peak overpressures can be obtained for determining adequate separation distances between plant units in onshore facilities, and for obtaining the initial estimate of drag load on equipment. The degree of congestion can be assessed by gas dispersion calculations to determine the flammable gas isopleths for various meteorological conditions, superimposed on the facility layout plan and elevation. The main limitations of the model are: • The model predicts the maximum overpressure at the point of ignition, thereafter

decaying with distance away from the source. In reality, if there are obstacles in the flame front before sufficient decay occurs, the overpressure can peak again, sometimes even exceeding the initial overpressure. This phenomenon cannot be modeled easily, although combining the energy sources has been suggested.

• The model is often used by treating the VCE as an omni-directional incident, i.e.

pressure waves radiating from the ignition location. This is possible only in the unconfined and uncongested case. In fact, depending on the wind direction, if there are no obstacles within the isopleth, no overpressure may be generated. Conversely, if there as significant obstacles, a higher overpressure is generated. Thus, the size of the vapour cloud, the location of the vapour cloud depending on wind direction, and the location of ignition source influence the peak overpressure significantly. This the rapid assessment model often fails to predict to required level of accuracy for design purposes.

6

2.2 CFD Models

The growing interest in the use of more fundamental approaches to explosions based on Computational Fluid Dynamics (CFD) has led to the development of a number of models, now routinely used for off-shore explosion assessments and situations with complex geometries. The MERGE project in Europe during the early 1990s sought to understand factors that would lead to improvements in CFD code predictions (Popat et al. 1996). These models deal with such issues as ignition and laminar flame propagation, turbulent flame propagation and combustion.

The current status of CFD explosion modelling is given by Bull (2004) and Lea and Ledin (2002) who cover many issues related to the modelling, solution and validation of these codes. Of particular significance to CFD approaches are the following:

1. Use of crude approximations to complex geometries 2. Considerable uncertainty in combustion sub-models 3. Importance of considering pre-existing turbulence, such as high pressure gas

releases 4. Simple treatments of turbulence 5. The need for more large-scale experimentation 6. The use of adaptive grid refinement and improved numerical solution schemes 7. Incorporation of flame distortion phenomena and flame interactions 8. Incorporating the interaction of blast-structure effects such that the movement

of structures is considered on the propagation of the blast wave.

Despite the many challenges in developing and applying CFD explosion models, predictions from these codes in complex offshore and onshore facility geometries appear to lie within a factor of 2 of the experimental data (Bull 2004). The CFD explosion model does not yet incorporate a fully realistic combustion model and may not adequately to represent all important obstacles in real complex geometries. The major finding in CFD modelling is that, because of turbulence causing flame acceleration, it is possible to have higher overpressures much further away from the ignition location, whereas MEM predicts gradual decay of the pressure wave away from incident location. The MEM is an empirical model based on correlations with experimental data used to predict far field blast effects outside the gas cloud combustion region. Therefore it may not address the scenarios typical of offshore modules where the filling ratio by the gas cloud can be much higher than in onshore process plants, which are open to atmosphere. One limitation in the application of CFD models for blast analysis in offshore oil and gas facilities is that the volume of module filled by flammable gas cloud is unknown, and hence a sensitivity analysis would be required, using various module fill fractions. Assumption of full module fill may give high blast overpressures and drag, introducing over-conservatism in structural design. A front end 3-D dispersion model to obtain the fraction of volume of module filled by flammable gas cloud, as a precursor to flame acceleration studies, has been found to be useful.

7

2.3 The Challenge CFD models are complex, require significant modeling efforts, and are expensive to run. At the initial stages of a project, pending CFD analysis, the challenge is: Can we refine the assumptions required for the rapid assessment model, based on information generated from gas explosion simulations using 3-D CFD models, in order to have a similar level of conservatism in the blast design load selected? The following is an attempt to partially address this challenge. 3. CASE STUDIES ON CFD MODELS The case studies are based on studies undertaken by KBR for detailed engineering of two offshore oil and gas platforms. Two contrasting designs, differing in confinement and ventilation, have been selected to test the versatility of the MEM. 3.1 Platform A - Conventional Design The platform is consists of an integrated deck structure and a four-legged steel jacket. There are three principal deck levels. An inclined flare tower is located on the east side of the platform. A crane is located on the west side of the platform. The platform is linked by walkway bridge to an adjacent platform. The deck dimensions are approximately 45 m long by 36 m wide. Approximately 40% of deck area at the north is occupied by utilities. The process areas are located to the south. A blast wall separates the process and utility areas on the two lower levels. Process areas on the south side of the blast wall on Levels 1 and 2 have the potential for gas buildup should a loss of containment occur. In addition to the blast wall to the north, the modules on Levels 1 and 2 are further confined by solid deck plating above and below. The modules are also reasonably congested with large items of equipment and pipework. 3.2 Platform B - Unconfined Design Platform B selected for this study differs significantly from Platform A in that it has grated decks and no parallel confinement, and no blast wall. The design has focused on elimination of serious explosion consequences rather than structural design to cope with high explosion overpressures. The platform is approximately 60m long by 40m wide and consists of an integrated deck structure and a twelve-legged steel jacket. There are five deck levels, including mezzanine deck above Cellar Deck level. The platform is bridge-linked to two adjacent platforms. A crane is located on the north side of the platform. The platform has a largely open structure, with the decks open on all sides and exposed to winds approaching from all directions. Some wind-shielding is provided at two of the deck levels by switch room and equipment room walls. Most of the deck on each level is grated, allowing air and gas movement between the decks. The

8

modules themselves are congested, with areas of significant blockage to airflow, caused by large pieces of equipment, structural components, and the accumulative effect of pipework. 3.3 Blast Simulation Study Blast simulation studies of the topsides of both Platforms A and B were undertaken using AUTOREAGAS software. The software has been developed by TNO in the Netherlands, and has been calibrated and validated against data obtained in the full scale tests of the Spadeadam test rig. These tests provided important reference data for the explosion predictions tools (TNO 2004). The AUTOREAGAS model requires the volume filled by gas as an input and hence a front end gas dispersion analysis was used to obtain the flammable gas volume in the modules, for various release scenarios postulated. For Platform A, it was found that the percentage of module filled by gas varied from 10% to 45%, depending on the size of the release. Stoichiometric composition of cloud was assumed in the explosion modelling. Larger releases of the rupture type may give flammable cloud volumes > 50%, but these were not modelled as the frequency of the ignited release was low, and the event would be screened out in the frequency exceedence analysis. For Platform B, very interesting results were obtained in the dispersion analysis. Owing to the buoyant gas properties and the fact that the decks are considerably porous, gas clouds spanning several deck levels were obtained. Also, for most releases, the largest percentage volume fill was not recorded on the source of release deck, but on a higher deck level. In general, significant gas cloud sizes were found to occur for large releases and rupture cases only. For these scenarios, maximum gas clouds of the order 15% to 20% module fill were estimated. These larger gas clouds formed on the two upper decks. Cloud volume fractions on the two lower decks were significantly less than 5% of module fill. The dispersion calculations indicated that small releases such as flange leaks result in gas clouds that are insignificant in size compared to the volume of the module (much less than 1%). Pressure-time history at nominated gauge points was obtained by blast simulation for various module fills, various locations of the cloud on deck, and various ignition locations. The time step used in the CFD models was 5 ms, and the mesh step was 1m. More than 50 simulations were undertaken accounting for the release size, the release location, the wind speed, the wind direction and the locations of the ignition source within the flammable cloud. 4. CALIBRATION OF MEM BASED ON CFD MODEL RESULTS 4.1 Blast Scenarios Selected for Platform A Only a limited number of scenarios were selected for the present exercise. The MEM was applied to the same cloud size and location as the detailed blast study. The

9

explosion efficiency (% cloud occupied by equipment and obstacles) was estimated using the layout plan and elevation diagrams. Once calculated, this value was fixed as one parameter for MEM. The blast strength curve number that best matched the detailed blast simulation result was then fitted for each scenario selected. Tables 2 and 3 provide a summary of the scenarios analysed on Levels 1 and 2. Figures 2 and 3 show the deck layout for Levels 1 and 2, and the gauge points, G1 to G4 (different for each level). Figure 2: Platform A - Level 1 Layout with Gauge Points

Table 2: Summary of Blast Scenarios Analysed - Platform A, Level 1 No. Source Gauge points for overpressure estimation Comments

G3

G1

G2 G4

HP Separator

(G1)

HP separator gas outlet

(G2)

Flare header (G3)

HP separator oil outlet

(G4)

Distance from source to gauge points, m S1 HP Separator - 7 25 20 S2 Flare header 25 28 - 16

Scaled Energy Distance, R S1 HP Separator - 0.27 0.95 0.76 20% module fill S2 Flare header 0.79 0.89 - 0.51 30% module fill

MEM - peak overpressure (barg) S1 HP Separator 1.0 1.0 0.46 0.63 MEM Curve 7,

Congestion 33%

S2 Flare header 0.4 0.35 0.53 0.53 MEM Curve 6, Congestion 38%

CFD Model - peak overpressure (barg)

10

No. Source Gauge points for overpressure estimation Comments HP

Separator (G1)

HP separator gas outlet

(G2)

Flare header (G3)

HP separator oil outlet

(G4)

S1 HP Separator 1.00 0.80 0.55 0.70 Ignition on the south and flame front moving into deck

S2 Flare header 0.50 0.40 0.30 0.50 Overpressure away from source higher

Figure 3: Platform A - Level 2 Layout with Gauge Points

G3

G2

G1G4

Table 3: Summary of Blast Scenarios Analysed - Platform A, Level 2

No. Source Gauge points for overpressure estimation Comments Compressor

gas line (G1) Glycol contactor (G2)

HP sep. inlet manifold (G3)

Flare header (G4)

Distance from source to gauge points, m S3 Compressor

gas line - 13 16 17

S4 Glycol contactor

13 - 4 29

S5 HP sep. inlet manifold

16 4 - 32

Scaled Energy Distance, R S3 Compressor

gas line - 0.37 0.46 0.49 30% and 45% module

fill S4 Glycol

contactor 0.37 - 0.12 0.84 10% module fill

S5 HP sep. inlet manifold

0.46 0.12 - 0.92 20% module fill

11

No. Source Gauge points for overpressure estimation Comments Compressor

gas line (G1) Glycol contactor (G2)

HP sep. inlet manifold (G3)

Flare header (G4)

MEM - peak overpressure (barg) S3 Compressor

gas line (45% module fill)

0.2 0.2 0.2 0.2 MEM Curve 5 to match ignition at cloud edge south. Congestion 41%

S3 Compressor gas line (45% module fill)

1 1 1 1.0 MEM Curve 7 to match ignition at cloud edge north

S3 Compressor gas line (30% module fill)

0.2 0.2 0.2 0.2 MEM Curve 5, congestion 50%

S4 Glycol contactor

0.20 0.20 0.20 0.11 MEM Curve 5, 10% module fill, congestion 26%

S5 HP sep. inlet manifold

0.10 0.10 0.10 0.07 Curve 4, 20% module fill, congestion 19%

CFD Model - peak overpressure (barg) S3 Compressor

gas line 0.2-0.3 0.1-0.2 0 0-0.3 45% module fill,

ignition at cloud edge south

S3 Compressor gas line

0.7-1.1 0.5 0.1 1-1.2 45% module fill, ignition at cloud edge north

S3 Compressor gas line

0.2 0.1 0 0.1 30% module fill

S4 Glycol contactor

0.2-0.3 0.2 0 0.1-0.2 10% module fill

S5 HP sep. inlet manifold

0.5- 0.8 0.5-0.8 0.1-0.3 0.1 20% module fill

Observations on Tables 2 and 3 are provided in Section 5. The CFD model, being a 3-D model, is capable of estimating overpressures at gauge points at any level, for a given ignition of gas cloud at a specified level. This feature is not possible with MEM. 4.2 Blast Scenarios Selected for Platform B For a release on Level 1 (the lowest deck), the maximum cloud sizes of 10 and 15% module fill were on Levels 3 and 4 and not on Level 1 and 2. Similarly, for each release source, the clouds on decks above the release source deck were more significant. Therefore, for blast calculations, it is not the release location that matters, but rather the location of ignition. Tables 4 and 5 provide a summary of the scenarios analysed on Platform B. Figures 4 and 5 show the deck layout for Levels 2 and 4 gauge points.

12

Figure 4: Platform B - Level 2 Layout with Gauge Points

G1G4 G3

G2

Table 4: Summary of Blast Scenarios Analysed - Platform B, Level 2 No. Cloud

Location Gauge points for overpressure estimation Comments

HP separator (G1)

Oil export pumps (G2)

Surge vessel (G3)

Bridge landing

(G4)

Distance from cloud centre to gauge points, m S1 Level 2 4 8 5 32 5% fill. Release on

Level 1 (Gas lift) S2 Level 2 2 6 6 30 3% fill. Release on

Level 2 (HP separator) Scaled Energy Distance, R

S1 Level 2 0.18 0.41 0.26 1.65 S2 Level 2

0.09 0.28 0.31 1.38

MEM - peak overpressure (barg) S1 Level 2 0.2 0.2 0.2 0.07 MEM Curve 5,

congestion 40% S2 Level 2 0.1 0.1 0.1 0.05 MEM Curve 4,

congestion 35% CFD Model - peak overpressure (barg)

S1 Level 2 0.24 0.38 0.13 0.07 S2 Level 2 0.09 0.10 0.10 0.04

13

Figure 5: Platform B - Level 5 Layout with Gauge Points

G1

G2

G3

G4

Table 5: Summary of Blast Scenarios Analysed - Platform B, Level 4

No. Cloud Location

Gauge points for overpressure estimation Comments

Gas compressor KO drum

(G1)

Turbo-generator

(G2)

Fuel gas scrubber

(G3)

Gas export pipeline

(G4)

Distance from cloud centre to gauge points, m S3 Level 4

10 17 12 10 15% fill. Release on Level 1 (Gas lift)

S4 Level 4 10 17 12 10

8% fill. Release on Level 2 (HP separator)

S5 Level 4

10 17 12 10

5% fill. Release on Level 4 (Compressor KO drum)

Scaled Energy Distance, R S3 Level 4 0.29 0.5 0.35 0.29 S4 Level 4 0.29 0.5 0.35 0.29 S5 Level 4 0.29 0.5 0.35 0.29

MEM - peak overpressure (barg) S3 Level 4 0.2 0.20 0.2 0.2 MEM Curve 4,

14

No. Cloud Location

Gauge points for overpressure estimation Comments

Gas compressor KO drum

(G1)

Turbo-generator

(G2)

Fuel gas scrubber

(G3)

Gas export pipeline

(G4)

congestion 39% S3 Level 4

0.5 0.5 0.5 0.5 MEM Curve 5, congestion, 39%

S4 Level 4 0.2 0.20 0.2 0.2

MEM Curve 4, congestion, 39%

S5 Level 4 0.2 0.20 0.2 0.2

MEM Curve 4, congestion 35%

CFD Model - peak overpressure (barg) S3 Level 4 0.31 0.17 0.37 0.34 S4 Level 4 0.28 0.26 0.24 0.43 S5 Level 4 0.2 0.15 0.11 0.2

5. STUDY FINDINGS 5.1 Results of MEM Calibration From Table 2, it is seen that the MEM is able to predict the overpressure to within 20% of the CFD model. The curve numbers of 6 and 7 selected are consistent with the Kinsella rule set in Table 1. In Table 3, for Scenario S3, the location of ignition makes a significant difference in the overpressures generated. Ignition at the south edge of the cloud results in the flame front propagating away from congestion, thus generating a lower overpressure. For ignition at the northern edge of the cloud, the MEM prediction is significantly higher than the CFD model, at some gauge points. Once again, the selection of ignition point, the path of the flame front thereafter and the obstacles present in its path, as assessed from the layout, would influence the blast strength selection. For scenarios S4 and S5, once again most of the predictions are within 20%. For the unconfined module design, once the cloud size is established on the deck of interest by gas dispersion analysis, the MEM is able to predict blast overpressures within 20-40% of that estimated by CFD analysis (Table 4). The results in Table 5 show a wider departure between CFD model and MEM prediction in many locations. For congested, but unconfined design, the Kinsella rule set (Table 1) recommends a blast strength of 4-5. The values actually fall between those predicted using Curves 4 and 5 individually. This suggests that some form of averaging may be better than choosing a single initial blast strength. Another point of interest is that the peak overpressure at a gauge point away from the point of ignition can be higher than that at the point of ignition, as predicted by the CFD model. On other hand, MEM predicts a maximum blast strength at the point of ignition, with decay thereafter, as the flame front advances.

15

The positive pressure duration was not compared in this exercise, as this information was not readily available from the blast simulation data, in a useable form. 5.2 Selection of Blast Strength The present review, even though limited in its application, has shown that it is possible to refine the blast overpressure predictions using the rapid assessment model. Some guidance is provided below. Additional work in this area would be required to convert the guidance into an empirical rule set. 1. The rule set for blast strength selection provided by Kinsella (1972) and abridged

in Table 1 here is a good start for the application of MEM. The calibration check in this paper has shown that the rule set works well.

2. Reliable gas dispersion results must be available for the selection of blast strength.

Special gas dispersion tools are required for the congested environment of an offshore module (Savvidas et al. 1999, HSE 2001).

3. A thorough review of the layout is essential in the selection of the blast strength.

The gas dispersion results (% module fill) needs to be superimposed on the layout (sometimes layouts of more than one module level for unconfined design). The congestion on the path of the flame front is to be reviewed from two perspectives: (a) wind direction and (b) location of ignition source. An ignition source at the middle of the module may result in a lower overpressure as the flame front advances to the edge, compared to an ignition source at the edge of the module, with wind directing the flame front into the module. Thus, the flame front has a longer distance to traverse for module edge ignition.

4. Where there are two different release scenarios from the same source (i.e.

different hole sizes), and the rule set suggests blast strength of say curve 4 or 5, Curve 4 may be used for the smaller module fill and Curve 5 may be used for the larger module fill. The % congestion value estimated from layout would also play a role in determining the blast strength.

6. SUMMARY An attempt has been made to refine the predictions of peak overpressures in vapour cloud explosions, using the TNO multi-energy model. The approach involved ‘fitting’ a suitable blast strength curve to an explosion scenario, by matching the overpressure generated by AUTOREAGAS model on two offshore oil and gas platforms. The method was applied to two different geometries, one confined with solid deck plating and a blast wall, and the other unconfined with a porous grated deck for gas dispersion and explosion venting. Initial indications are that it is possible to improve the blast prediction to within 20-40% of the complex 3-D model, by taking into account the layout geometry and the obstacles in the path of the flame front. The location of ignition point and the strength of ignition source determine the blast pressure rather than the location of the release

16

source. The reliability of gas dispersion in assessing the cloud size plays a significant role in explosion assessment. Additional work is required in this area to develop an empirical rule set that may be used for rapid assessment using the MEM. The MEM approach, as it stands, still requires the heuristic feel of the analyst. 7. REFERENCES Bjerketvedt, D., J.R. Bakke, and K. Van Wingerden (1997):“Gas explosion

Handbook”, J. Haz. Mat., Vol. 52, no. 1, pp. 1-150. Bull, D.C. (2004): “A critical review of post Piper-Alpha developments in explosion

science for the off-shore industry”, HSE Research Report #89, HMSO, Norwich, UK.

Health & Safety Executive (UK) (2001): “A model for jet dispersion in a congested environment”, prepared by Advantica Technologies for HSE, Contact Research Report 396/2001.

Kinsella, K.G. (1992): “A Rapid Assessment Methodology for the Prediction of Vapour Cloud Explosion Overpressure”, International Conference on Safety and Loss Prevention, Singapore.

Lawrence, F.E. and E.E. Johnson (1974): “Design for limiting explosion damage", Chem Eng., 7, January.

Lea, C.J. and H.S. Ledin (2002): “A Review of the State-of-the-Art in Gas Explosion Modelling”, Health and Safety Laboratory Report HSL/2002/02, Fire and Explosion Group, Buxton, UK.

Mercx, W.P.M., van den Berg, A.C. and D. van Leeuwen (1998): “Application of correlations to quantify the source strength of vapour cloud explosions in realistic situations: Final report for the project ‘GAMES’, TNO Report, PML1998-C53, Rijswijk, The Netherlands.

Mercx, W.P.M., van den Berg, A.C., Hayhurst, C.J., Robertson, N.J. and K.C. Moran, (2000): “Developments in vapour cloud explosion blast modelling”, J. Haz. Mat., 71, pp. 301-319.

Popat, N.R., Catlin, C.A., Arntzen, B.J., Lindstedt, R.P., Hjertager, B.H., Solberg, T., Saeter, O. and A.C. van den Berg (1996): “Investigations to improve and assess the accuracy of computational fluid dynamic based explosion models”, J. Haz. Mat., 45, pp.1-25.

Savvidas, C., Tam,V., Cleaver, R.P., Darby, S., Buss, G.Y., Britter, R.E. and Connolly, S. (1999): “Gas dispersion in a congested, partially confined volume”, International Conference and Workshop in Modelling the Consequences of Accidental Releases of Hazardous Materials, San Francisco, AIChE.

TNO (1997) Methods for the Calculation of Physical Effects (Yellow Book 3rd edition) CPR14E Part 1, Director-General for Social Affairs & Employment, The Netherlands.

TNO (2004) Internet website www.tno.com.nl UKOOA (2003) Fire and explosion guidance Part 1: Avoidance and mitigation of explosions, Issue 1, October. van den Berg, A.C., van Wingerden, C.J.M. (1991): “The Vapour Cloud Explosions:

Experimental investigation of key parameters and blast modelling”, Trans. IChemE., Part B, 69, pp. 139-148.