You are on page 1of 8

Predicting Blast Overpressures Caused by Vapor Cloud Explosions i the Vicinity of Control Rooms n

Kees van Wingerden, Olav Roald Hansen and Pierre Foisselon Christian Michelsen Research, P.O. Box 3, Fantoft, Bergen, Norway
When considering the consequences of gas explosions in petro-chemical installutions simple methods are used such us W-equivalenc~~ methods and the Multi-Energy method. fiese methods< howewr, do not take into account directional effects, focusing effects,.factors related to the source of the explosion (initial strength, shape) and reflections. FLACS, a 3-0 CFD tool, allowsfor allfactors mentioned including focusing on effects and wJections neur buildings. A comparison of the CFD-method including the two traditionalprediction methodsfor a practical case demonstrates the upplication ureas o the FLACS method as w l as the limitaf el tions o the traditional methods. f
INTRODUCTION

Two types o f models are used to predict the potential consequences of vapor cloud explosions: TNT-equivalency methods [51 and the Multi-Energy method (CCPS and Van den Berg [S,81). Both methods predict the strength of blast waves generated by the vapor cloud explosion as a function of distance to the center of the explosion. TNT-equivalency methods use the blast generated by an equivalent amount of TNT to describe the strength of the vapor cloud explosion and the decay of the blast as a function of distance. The Multi-Energy method utilizes the blast decay and initial strength from an idealized gas explosion generated by computation. Both models assume that the blast generated is similar in all directions with no directional effects. Only an undisturbed, incoming hlast is described and the description of the explosion itself is inadequate. Comparative studies show that the Multi-Energ)r method is capable of describing far field blast effects of strong explosions (Van Wingerden et al., @I). In addition, the far field blast predicted by a TNT-equivalency method compares reasonably well with estimated pressures made from damage due to vapor cloud explosions provided the TNTequivalency or yield factor is chosen correctly. Gugan [6] demonstrates that there is little correlation between the quantiProcess Safety Progress (Vol. 18, No. 1)

ty of fuel involved in the explosion and the equivalent-charge weight of TNT required to model its blast effects. It should also be mentioned that the decay of a TNT-blast is faster than a blast generated by idealized gas [51. The applicability of these methods is limited because of the following: the choice of location o a control room and the necf essary structural strength of this facility plus the applicability of these methods. This paper investigates the limitations of such models by using a CFD-model (FLACS) to predict some aspects of the behavior of vapor cloud explosions. Simultaneously we present this CFD-model as an alternative to predict the potential consequences of vapor cloud explosions. The paper will briefly describe the FLACS code and some recent extensions which are directly related to blast prediction in both the near- and far field consisting of three parts: a multi-block concept, dividing the computational region in various contiguous blocks a special solver for describing the blast propagation away from the explosion a boundary condition to reduce reflections of blast waves from calculation boundaries In addition, the Multi-Energy method and one of the TNTequivalency methods is introduced.
SHORT DESCRIPTIONOF FIACS

FLACS is a fluid dynamic code developed for description of gas dispersion and explosion processes in complex geometries such as offshore modules [91. FLACS calculates explosion pressure and other flow parameters as a function of time and space for different geometries and explosion scenarios. It takes account of the interaction between flame, vent areas and obstacles such as equipment and pipe work. The FLACS code solves the full gas dynamic partial dfferential equations for a set of control volumes. The effects of turbulence and chemical reactions are included in the differential equations. The equations are discretized using a finite-volume technique and a weighted upwind/ central differing scheme for the convection terms.

Spring 1099 17

Velocities are calcukated on staggered grids. The effect of turbulence is included through the eddy viscosity concept by solving equations for turbulent kinetic energy (K) and its rate of decay (E). The tool has been validated extensively against experiments the majority of those being experiments carried out in partially confined geometries such as geometries representing confinement conditions prevailing in offshore modules [lo]. The main application area is the prediction of overpressures inside gas clouds and gas and oil production facilities. The latest version of FLACS, FLACS-96, has a possibility- to make direct predictions of the strength of blast waves generated by gas explosions. The extension of FLACS allows for describing both the explosion process inside a congested area and the propagation of blast waves generated by the explosion and their interaction with structures.The extension involves: a multi-block concept, dividing the computational region in various contiguous blocks a special solver for describing the blast propagation a~vay from the explosion a boundary condition to reduce reflections of blast waves from calculation boundaries

3
I

FIGURE 1

h example of a multi-block grid used in FLACS for

blast calculations

i2lulti-block Concept The multi-block concept allows for dividing the computational region into several contipous blocks. In this manner one can use different resolutions in different blocks. The advantage of using such an approach is that one can use a special purpose solver for the simulation of propagating blast waves in the region outside the area where the explosion takes place. This can save both CPU-time and memory in comparison t o using FLACSin a single block modus. The main changes which are necessary to allow for using different blocks in FLACS are: dormation exchange between blocks handling o f flow acros internal boundaries An example of how a multi-block grid can look like is shown in Figure 1. In order to ensure continuity across the block boundaries it is necessary for individual blocks to obtain inforimtion from their neighboring blocks. In the new multi-block extension the variable and flux variables on the boundary of the blocks are found from interpolation from adjacent block values. For flows crossing the block boundaries the variables to be obtained in the next block are found by interpolation. The multi-block feature allows for using different solvers in different blocks. For the present application a special solver was implemented to describe blast wave propagation. Hence the multi-block concept allows for using FLACS in a centrally k d t e d block for calculating the gas explosion parameters and the special S ~ I W for describing blast propagation in neighboring blocks.

scheme [4J. The method adds to a first-order monotone solution a limited amount of the difference between second-order and first-order fluxes. The difference is called an anti-diffusive flux contribution. This contribution allows for removing first-order numerical diffusion without creating unwanted overshoots or oscillations. As such the scheme allows for a gaxl description of shock waves without smearing them out over many computational cells. The blast solver as such has been validated. An example is shown in Figures 2 and 3. Figure 2 shows an experimental setup described by Absil et al. [I], consisting of two obstacles (boxes) placed in a shock tube. The obstacles were provided with pressure transducers at the positions 1-6 as indicated. The obstacles are 5 cni high and 5 cm wide and are placed 7.5 cm from each other. The obstacles are as long as the width of the shock tube. Pressure-time hstories were calculated at each monitoring position. Figure 3 shows measured and calculated pressure-time histories at the positions 3 and 4. The results show that the blast solver reproduces the measured blast pressure-time histories reasonably well. The main features o f the pressure-time histories are reproduced. Each "step" and "peak is either due to the initial shock wave or due to reflections in hemeen the two

Special Soher for Blast Propagation


The solver used for describing the propagation of blast waves is only utilized to solve the Euler equations where we assume that viscosity effects can be neglected in the uncongested regicjn. The implemented solver uses the Flux Corrected Transport

9
Incoming bla

FIGURE 2

la\,orator).~scaleset-up used for studying blast-object


interaction.

18 Spring 1999

FIGURE 3. Calculated and measured pressure-time histories at positions 3 and 4 in the laboratory scaie set-up shown in Figure 2

9 14 ---. 8 49 784 7.18 - . 6.53 - 5.88 '-.-. 5.23 '-... 4.57 --.3.92 3.27 ----

boxes. If a finer grid would have been used in the simulations the resolution ofeach pressure spike in the simulations would have been better resulting in an even better resemblance of measured and simulated pressure-time histories.

--

2.61 -... 1.96 . '..'

1.31
0653

8.6
7.93 7.27 6.61 5.95 5.28 4.62 3.96 3.29 2.63 1.97 1.31 - -

0.644 - -

8.77 ---p. , A .

.....

Boundary conditions In order to avoid reflections and rarefaction waves occurring at the outer boundaries of the blast blocks a special boundary condition was developed. The boundary condition is based on linear acoustics. The new boundary condition was validated extensively as well. An example of such a validation exercise is shown in Figure 4. Several moments of the propagation o a 2-D shock wave f at a 45" angle in a 70 m x 30 m computational domain are illustrated in the figure. The strength of the shock wave is 10 kPa. The shock waves are represented by isobars. The figure shows that no reflections occur at the boundaries and we conclude therefore that the boundary conditions work satisfactorily.
TNT-EQUIVALENCY METHODS AND MULTI-ENERGY METHOD

3.12 2.49 --...

----

FIGURE 4

Se\ era1 moments of 2-D shock wave propagation at a


ii

The TNT-equivalency method is originally used for prediction of far field effects of vapor cloud explosions. The principle is to convert the available combustion energy in a vapor cloud into an equivalent charge weight of TNT using the following relationship:
w In.-,
=

.in&

a r W , H , /ITw

Process Safely Progress (Vol. 18, No. 1)

Spring 1999 19

10'

with W, W , H f

b n

s 2
t

loo

10-1

10-2 100

10'

102

"scaleddistam" =

actual distance.

rnkg-'"

FIGURE 5

Peak side-on overpressure due to a surface TNT explosion accordng to Marshall PI.

of the fuel involved of TNT or yield (kgl = heat of combustion of the fuel in question Wkgl = "NTblast energy Q/k@ Hm = TXT equivalency based on energy-(-) a, If the equivalent weight of TNT is known the blast characteristics as a function of distance can be derived from a chart containing a scaled representation of experimental data. The distance in the chart is scaled using the weight of TNT as a scaling parameter. The TNT equivalency in the equation presented varies strongly when reproducing blast effects from damage due to vapor cloud explosions [b]. spite of these variations the L In K Health and Safety Executive recommends that a TNT-equivalency of 3 % of the total combustion energy available in the cloud should be mice the theoretical flash of the amount of material released to account for spray-and aerosol formation. If this amount exceeds the total amount of fuel available the recommended amount should of course be used. The blast data according to Marshall [71 are used to estimate the blast effects (See Figure 5) The Multi-Energy method uses blast from an idealized gas explosion generated by computation to predict the consequences of vapor cloud explosions. In contradiction to conventional TNT-equiKilency methods the Multi-Energy method recognizes that blast is only generated in congested areas: areas where an interaction of combustion and conibustion generated
= the weight

= equivalent weight

10

10
5

t 0

1 1

0.5
0.1

0.01

0.1

RO

combustion energy-scaleddistance (R)

0.00

.........I.........

.-------.
FIGURE 6
Blast characteristics accordmg to the Multi-Energy method 181.

Po =atmosphericpressure co = atmosphericsoond speed E = amount o combustion energy f


R
= charge radius

20

Spring 1c)C)c)

A control room is located at a distance of 40 m from the center of the facility. A gas cloud of a stoichiometric mixture of propylene and air was assumed engulfing the entire facility reaching a height of 8 m. The effects as predicted by the TNTequivalency method presented above and the Multi-Energy methcd are calculated.

FG R 7. IUE

Representation of a 25m wide, 5 m long and 5m high 0 facility used in the present investigation to investigate some aspects of modelling of near-field blast effects clue to vapor cloud explosions.

turbulence can arise. In fact the Multi-Energy method regards a vapor cloud explosion as a number of explosions correspondmg to the various parts of the clouds where congested areas have been engulfed. Therefore only these parts of the combustion energy are used to scale blast decay away from each of these explosions. The model uses a chart representing the blast characteristics of a hemispherical fuel-air cloud. The blast characteristics are given as a function of distance from the center of the cloud. The main parameter used for scaling of the distance to the blast center is the combustion energy in each of the congested parts of the cloud (Figure 6). In contradiction to the TNTequivalency method the Multi-Energy method uses a set of curves depending on the initial strength of the explosion.
BLAST EFFECTS ON A CONTROL ROOM DUE TO NEARBY VAPOR CLOUD EXPLOSIONS

The strength of blast waves generated by vapor cloud explcsions occurring in a realistic plant were investigated using the three methods introduced above. The blast wave strength was investigated at a control room located in the vicinity o the plant. f The geometry that has been chosen as a basis for the present study is shown in Figure 7. It is 50 m long, 25 m wide and 5 m high and consists of several vessels of varying sizes located at the ground and interconnected by piping. A pipe-rack covers the vessels on the east side of the facility whereas a grated floor with some equipment located on it, covers s d a r equipment on the west side of the facility.
Table 1. Estimate of Loading on a Control Room Located 40 m From a Facility in Which a Vapor Cloud Explosion Occurs*. - -

Predictions o Pressure L a s at the Control Room Using f od the 71vT-EquivalencyMethod and Multi-Energy Method Table 1shows the results of predictions of the pressure loading at the control room due to a vapor cloud explosion in the installation described above using the TNTequivalency method and the Multi-Energy method. For calculating the TNTequivalency the size of the full cloud was used: 25 m x 50 m x 8 m = 10 000 m3 involving 774 kg of propylene. This corresponds to a TNT-equivalencyof 258 kg. Using Figure 5 one can find that the side-on overpressure is 300 mbar. As the control room walls will reflect the incoming shock wave the pressure loading on the control room will be larger. This can be estimated using data presented in Baker et al. [31. Thus the loading on a control room wall at an angle of incidence of 0" with respect to the dlrection of propagation of the incoming blast wave can be estimated to be 6% mbar. When estimating the strength of the incoming blast wave by the Multi-Energy method the initial strength of the gas explosion must be estimated. Regarding the dfiiculty in estimating the initial strength a range of initial blast strengths varying from 6 (m. overpressure in the cloud is 0.5 bar) to 9 (initial strength 5 bar). The size of the facility and the reactivity of the released material have been the main reasons for choosing thts range of values. A more accurate estimate is only possible by either performing scaled experiments for this particular plant or performing CFDsimulations using a recognized model such as FLACS [lo].The combustion energy used for estimating the strength of the blast
Table 2. Results of Predictions for the Pressure loading at a Control Room Located 40 m from the Center o a f Facility in Which a Vapor Cloud Explosion Occurs*
- _____ -

Control room locabon

central --0.66 0.31 1.7 0.37

Igrxhon source locahon north east south -~


~~~~~

west
0.77
0.76 1.2 0.37

North East south West Maximum pressure inside plant

0.12 0.3 3.3 0.23

0.29 0.21 0.59 0.53

2.7 0.32 0.22 0.27

Bkast prediction method

Side-on overpressure (bar) 0.3


0 43-1 2
-

Reflected overpressure (bar) ~~0.67


10-3 4
~-

method Multi-Fnergy method

1.0
~ ~ ~

2.2

0.36
~- ~~ ~~~~~

1.7
~~ ~ ~~ ~~

0.54
~~ ~ ~

_.__

'The e\tinidte has k e n made usmg n TNT-equivalency method

.ind the Multi-Energy method The reflected overpressure at the control r(x)ni w s estimated using data presented by Raker et a a1 , [31

T h e overpressure predictions were inade for varic )us ignition source locations and locations of the control room. The overpressures are given in bar.

Spring I999

21

FIGURE 8

Moment of vapor cloud explosion in a 25m wide, join long and im high installation. The figure shows the pressure distribution at moment 1.065s after ignition. To limit the number of simulations, control rooms have been erected south, west. north and east of the facility.

tion. This coincides with a hlast energy of 3.51010J. Using the blast curves of Figure 6 it can be estimated that the sideon overpressure is equal to 0.43 -1.2 bar. [:sing the data presented in Raker et al. [31 again it can be estimated that the reflected pressure of the blast a w e at a wall of the control room at 40 ni from the center of the plant is 1.0 - 3.4 bar overpressure. Both the TNT-equivalency method and the Multi-Energy method cannot describe directional effects, the effect of ignition source location and the effect of location of the control room other than the distance to the center of the facility-. How important these latter factors are will be demonstrated by using the FLACS code.

FIGURE 9

Similar to Figure 8 but at 1.09s after ignition.

FIGURE 10

Similar to Figure 8 hut at 1.1% after ignition. The Figure shows the reflection of the incoming shock wave n-it11the control room.

wave at the location of the control room is calculated from the size o the plant. The volume of the cloud occupied by the plant f having the dimensions 25 m x 50 111 x 5 m.representing a volume of 10 000 mi, will only contribute to hlast genera-

Predictions at the control room using FLACS Table 2 shows predictions using the FLACS code. The FLACS code can take the effect of gas explosion scenario into account and calculates the reflected blast loading on the building directly. In the calculations the multi-block facility and associated blast code was not applied as the distance between blast source and target is not large. The calculations were performed assuming the location of the control room to he varying as far as orientation is concerned. Keeping the distance to the center of the plant constant, the control room was assumed to be west, north, east and south of the plant respectively. The loading on the control room was then investigated for several ignition source locations. An example of a result is shown in Figures 8 through 10. The result concerns a simulation of a cloud ignited at the south end of the facility. The figures show the pressure development at several moments in time. The flame propagates and accelerates towards the south generating gradually higher pressures during this propagation. The highest overpressure that is generated in the plant occurs at the southernmost end of the plant and reaches a value of 1.9 bar. In the other direction the pressure is considerably lower and high pressures are generated only if the control room is located towards the south of the plant: ; control room at the I south end 40 in from the center of the plant will experience pressures up t o 2.7 bar, a control room at the same distance but at the opposite direction (i.e. south of the plant) will experience a pressure o f only 0.22 bar. If the control room is on the west or east side of the plant reflected pressures of 0.27 bar and 0.32 bar respectively is experienced. This strong directional effect is also seen in Figures 8 through 10. Similar effects are seen when ignition is effected on the north end o f the plant: a maximum pressure of 2.2 bar in

Table 3. Blast Decay Away from a Propylene-air Explosion in a 50 rn x 25 rn x 5 rn Facility Where ignition of the Cloud was Effected in the North End of the Facility.
~~~ ~ ~~~ ~~~ ~~ ~~ ~ ~~~~ ~

Distmce from plant centre (m)

FLACS

TNT equk method


NE 0 14 0 11
0 07

Multi-Energ) method

sw
23.4 37 5 58 7
22 Spring 1999

11

0 92 0 24

NW 0 14 0 10 0 06

SE 12

0 58
0 32

09 0 37

0 5-48
0 45-1 0 0 4-0 5

17

Irocexs Safety frogress (Vo1.18. No.1)

the plant (south end) and a pressure of 3.3 bar at a control room towards the south of the plant and only 0.12 bar towards the north. The difference in maximum pressure both inside the plant and at the control room for the two ignition sources reflects the differences in geometry details in the plant. Ignition in the center results in a maximum pressure of 1 bar in the plant whereas reflected pressures of 0.66 bar, 0.31 bar, 2.7 bar and 0.34 bar are found at control rooms located at 40 m from the center of the plant in the north, east, south and west respectively. The lower maximum overpressure inside the plant in comparison to the two end ignition cases reflects the shorter distance of flame travel. A summary of the results using the FLACS code is presented in Table 2 . Another interesting feature to investigate is the blast decay away from the facility according to the two traditional methods and FLACS. The comparison has been made for 1 single case, viz. ignition in the north end of the plant (Table 3). Blast pressures have been measured in four directions W, NW, NE and SE. The traditional methods cannot predict the effect of direction and they therefore predict only 1 overpressure for each distance to the centre of the plant irrespective the direction. The results show that blast decay away from the facility is stronger than predicted by the Multi-Energy method. This reflects the assumption of an exploding hemi-spherical cloud in the Multi-Energy method whereas in reality the cloud is strongly asymmetrical. Such asymmetrical blast decay behavior has also been noticed experimentally [ill. Further, it should be noted that the Multi-Energy method predicts very conservatively near the location where ignition occurs whereas an underestimate in the near field and an overestimate beyond the location of the control room occurs at the opposite direction. Also the TNT-equivalency method predictions overestimate in the direction where ignition occurs and underestimates pressure loads in the other direction. On the high pressure side blast decay appears to be stronger initially but further down FLACS and the TNT-equivalency method come closer. It is. of course, possible to investigate several other aspects, such as, the influence of installations located in between the installation in which the explosion occurs, and the control room, aspects related to the cloud (including realistic clouds generated from a leak). aspects related to the installation (including the effects of distance between two separate installations: what should the distance be to assure that an explosion in one installation does not cause a much stronger explosion in the second), pressure distribution on the control room walls, etc. It should be pointed out, however, that the use of a code such as FLACS for predicting the potential consequences of vapor cloud explosions necessitates the investigation of several explosion scenarios. In this way one can find the scenario which gives the strongest loading on the control room or one can find the most optimum location for the control room. This approach is now common in the offshore industry a n d is very often performed as part of a risk analysis
Process Safety frogress (Vol. IS, No. 1)

approach where several scenarios are investigated to build up a relationship between the consequence of an explosion. This is usually described by the peak overpressure generated, and the frequency at which these overpressures are predicted to be exceeded (a so-called exceedance curve. [21).
CONCLUSIONS

The FLACS code as a tool for prediction of blast due to vapor cloud explosions has been introduced and compared to traditional methods being a TNT-equivalency method and the MultiEnergy method. Due to simplifications in the two traditional methodologies the effects of vapor cloud explosions in the near field can be both overestimated and underestimated. The main reasons for this anomaly are: limited possibilities to vary accident scenarios no representation of asymmetry in the geometry in which the explosion occurs no representation o asymmetry in the explosion event f limited possibilities to represent the strength of the gas explosion Additional advantages of the use of FLACS are: investigation of the influence of installations located in between the installation in which the explosion occurs and a control room investigation of explosion effects due to realistic releases investigation of aspects related to the installation in which the explc )sion occurs possibility to describe the pressure distribution on the control room walls An obvious disadvantage is that an analysis using FLACS is more comprehensive and therefore more resources are needed.
LITERATURE CITED

1. Absil, LHJ., Van den Berg, A.C. and W e h e i j m , J. ( 19931, Blast interaction w t multiple obstacles. Paper presented at ih the 13th Int. Symp. On Military Appl. Of Blast Simulation, The Hague, The Netherlands (193). 2. Bailey, E. and Gregory, C.A.J. (1976) The application of a meta-model to the analysis of explosion risk. International Symposium on Hazards, Prevention and Mitigation of Industrial Fxplosions,pp. 552-5.73,June 1796, Bergen. 3. Baker, W.E., Cox, P.A., Westine, P.S., Kulesz,J.J. and Strehlow, R.A. Explosion Hazards and Eualuation, Fundamental Studies in Engineering, 5. Elsevier Scientific Publishing Company, Amsterdam, (1783). 4. Boris,J.P. and Book, D L Solution o continuity equations . f by the method of Flu-Corrected Trampofl,Meth. Computat. Phys., ~01.16. Academic Press, New York (1976) 5. Center for Chemical Process Safety. C u ~ l i n e s ~ ~ ~ E v a l ~ t i n ~ the Charactdtics of VaporCloud Exposiom, Flash Fires, and BLEW3 AIChE: New York (1794) 6. Gugan, K. Unconjked vapor cloud qlosiois IChemE. London (1978) 7. Marshall, V.C. he siting and construction o f control huilding.5 - a strategic approach, IChemE Symp. Series. no. 47 (1976) 8. Van den Bag, A.C. B e Multi-Enmgy method - Aframework

for vapor cloud explosion blastprediction,J.Haz. Mat., vol. 12, pp. 1-10 (1985). 9. Van Wingerden, C.J.M., Opschmr, G. and Pasman, HJ. Analysis of vapor cloud incidents, Les accidents industriels: quelles lecons en tirer?, Sociktk Alpine de Publications, Grenoble (190). 10.Van Whgerden, K., Storvik, I., Arntzen, B., Teigland, R, W e , J.R, Sand, 1.0.and SBrheim, H.R ,FUCS-93 - A new qlosion simulator, Proceedings of the 2nd International Conference on Offshore Structural Design against Extreme

Loads, London (1993).


11.Zeeuwen,J.P., Van Wingerden, C.J.M. and Dauwe, R.M.

Ehperimental investigation into the blast effect produced by unconfined vapor cloud explosions, Proceedings of the Fourth International Symposium on Loss Prevention and Safety Promotion in the Process Industries, pp. D20-D29 (1983). nispaper (4b) waspresented at the AIChE Spring National Meeting in New Orleans, Louisiana on March 10, 1998.

24

Spring 1999