Professional Documents
Culture Documents
(3)
P
Di
x; y; z
1
2p
1=2
_
Y
i
x;y;z5
N
exp
_
u
2
2
_
du (4)
3.2. Frequency analysis
The containment failure frequencies used in this study were
based on the bibliography and experience from previous
researches. According to I, Tsai, and Her (1999), the general rupture
frequency for a petrochemical vessel is around 2 10
6
110
5
/yr.
Due to the lack of relevant information as well as to facilitate the
assessment process, the incident frequencies used in this study
were roughly divided into three categories: which are possible
(frequency 1 10
5
/yr), less likely (frequency 1 10
6
/yr) and
very unlikely (frequency 1 10
7
/yr). Since all the incidents
discussed may cause disastrous accidents, their frequencies were
categorized as very unlikely to occur, which is 1 10
7
/yr.
3.3. Risk quantication
This paper focused on the damage caused by res and explo-
sions. Therefore, the risk value was assumed to be a function of
containment failure frequency, gas cloud ignition probability, wind
direction probability, personnel appearance probability, and
personnel death probability. The calculation algorithm for indi-
vidual risk was a revised form from Considine (1984) as was
expressed in Eq. (5). Here IR(x, y, z) stands for the individual risk at
a specic location; i and n stand for the index and the number of the
hazardous physical effects (for overpressure, pressure impulse, and
thermal radiation; n equals 3); F
I
stands for incident frequency
(1 10
7
/yr was chosen in this study); P
I
denotes for ignition
probability of the released cloud (1 was chosen to represent 100%
ignition); P
WIND
expresses probability of wind directions, which can
be found from the local meteorological data (set at 0.0015 for
southeast wind in this study); P
Zi
(x, y, z) and P
Di
(x, y, z) represent
employee appearance probability and death percentage, respec-
tively, both of them belong to the function of the coordinate. The
total individual risk was the cumulative summation of risk values
under different hazardous physical effects fromcertain enumerated
incidents. The nal result was displayed in a 3D iso-surface form
and superimposed with the 3D plant facilities layout to enable
better understanding by the relevant personnel.
IRx; y; z
n
i 1
F
I
P
I
P
WIND
P
Zi
x; y; zP
Di
x; y; z (5)
4. Results and discussion
4.1. FLACS simulation results
Wind can directly affect the gas cloud dispersion and lead to
a different hazard scope. After the wind direction was deliberately
chosen, a prior 120 s wind eld simulation was performed to
stabilize the wind before the explosion simulation (see Fig. 6).
Figures 79 show the simulation results of Case 1. In Fig. 7 we see
that most of the gas cloud was moving to the right of the graph and
the affected sites were far to the right of the target process (the red
square) since the southeast wind was blowing from the left of the
graph. The hazard impact zone may become very large if the gas
cloud were ignited after 60 s of dispersion (the gure shown here
was a pure dispersion simulation and the gas cloud was not
Table 2
Simulation cases and their related parameters.
Case no. Scenario Gas volume (m
3
) Gas concentration (%) Ignition time (s) Ignition coordinates (m) Mitigation measure
1 Distillation column collapse 28,731 100 11 (59, 25, 1) No
2 Reux tank rupture 1767 100 0.5 (16, 12, 3) No
3 Reux tank rupture 1767 100 0.5 (16, 12, 3) Water spray
Fig. 3. Position and ignition point of a ammable gas cloud released from a distillation column.
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 509
ignited). The 3D temperature diagram of Fig. 8 indicates the
deagration ame in the air (22 s after ignition) is slightly shifted to
the northwest due to the wind effect. The high temperature at the
bottomis distributed widely beyond the target area and it may have
a very serious effect on the nearby process sites. Fig. 9 shows that
the overpressure is also distributed widely; however, its value is
rather small since the maximum value at the explosion center is
only 0.0227 barg (see Table 4).
Cases 2 and 3 represent the same scenario (see Table 2) without
and with water spray. The gas cloud was ignited immediately after
the reux tank was ruptured. The macroscopic views of the simu-
lation results for the two cases do not showa signicant difference
(the related gures are not shown here due to space limitations).
One can only nd the appearing time of the highest overpressure in
Case 3 (2.5 s after ignition) was about 0.5 s earlier than that of
Case 2, which might be the result of turbulent effects caused by the
water spray activation. A turbulent ow can accelerate the mixing
effect of the released gas and the surrounding air and push the gas
cloud to reach its ammable limit earlier. A similar phenomenon
can also be observed in Table 4, where the maximum temperature
value of Case 3 does not decrease as expected. In addition, the
maximum overpressure and pressure impulses are ever higher
than those of Case 2. It was postulated that the turbulent ow can
accelerate the ame speed, and thus simultaneously increase the
maximum value of overpressure and pressure impulse.
To further investigate whether the water spray has a positive
mitigation effect, two monitoring points (P9 and P18, see Fig. 4)
were observed and compared at the same time (where P9 was
Fig. 4. Position and ignition point of a ammable gas cloud released from a reux tank.
Fig. 5. Layout of 20 water spray zones (blue: double layer, black: single layer). (For interpretation of the references to color in this gure legend, the reader is referred to the web
version of this article.)
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 510
located inside of both the gas cloud and the water spray zones,
while P18 was located outside of the gas cloud but within the
water spray zones). Table 5 shows that after water spray is acti-
vated, the overpressure values of these two monitoring points are
still larger than those without mitigation. However, it was
observed that both temperature values of these two points with
water spray were actually lower than those without mitigation
(P9 decreased 300 K, while P18 decreased 650 K). These results
indicated that although water spray cannot reduce the maximum
temperature near the explosion center, nevertheless, it can still
alleviate the thermal radiation hazard of the surrounding area. In
this study, the hazard impact zones of Cases 2 and 3 were obvi-
ously much smaller than that of Case 1 and mainly concentrated
at the surrounding of the target area. However, the impact can
become more severe should the ignition time be delayed until the
concentration of most gas cloud is within the ammable limit. An
additional simulation case, Case 4, was implemented to prove
such speculation. Case 4 is similar to Case 2 except the gas cloud
was located at the center of the target area and its concentration
equaled 1.5 times of stoichiometric concentration of the complete
combustion reaction. It was observed that the overpressure of
Case 4 (0.0144 barg, not shown here) was about 3 times as large
as that of Case 2. Case 2 behaved much more like a reball
phenomenon instead of a deagration since its gas cloud
concentration was 100%; therefore, the ammable cloud center
Table 3
Effect models of different hazard categories.
Hazard category Effect model
Thermal radiation Y
1
x; y; z 14:9 2:56 ln
_
teIx; y; z
4=3
10
4
_
Pressure impulse Y
2
x; y; z 46:1 4:82 ln Jx; y; z
Overpressure Y
3
x; y; z 77:1 6:91 ln Px; y; z
Fig. 6. Side view of wind eld simulation result, where VVEC stands for velocity vector and V denotes directional velocity projected on y-axis.
Fig. 7. Gas cloud dispersion result of Case 1 (60 s after released, without ignition).
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 511
Fig. 8. 3D temperature diagram of Case 1 (22 s after ignition).
Fig. 9. 3D cross-sectional overpressure diagram of Case 1 (3.5 s after ignition).
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 512
could not have been burnt unless its concentration was diluted
below the upper explosion limit (UEL).
4.2. 3D risk analysis results
The risk analysis module adopted the FLACS output value for
risk calculation and presented the calculation results in a 3D
format. From the 3D risk diagram (gure not shown here), Struc-
ture R1 had the highest risk value once the Case 1 accident
occurred. The mortality rate of the target area could be as high as
98.6%, which means almost no one could have survived under such
high temperature circumstance. In addition, a domino effect could
also have been triggered since there were many different processes
surrounding the target area. Figs. 10 and 11 show the results of
thermal radiation effect and its 3D risk value for Cases 2 and 3,
where different graphs represent (a) temperature iso-surfaces, (b)
mortality rate iso-surfaces, and (c) 3D risk iso-surfaces (the
temperature and the mortality rate iso-surfaces of Case 3 are not
Table 4
Maximum values of different simulation hazards.
Case Temperature (K) Overpressure (barg) Pressure impulse (Pa s)
1 2351 0.0227 3233
2 2277 0.0046 2055
3 2277 0.0051 2061
Table 5
Comparison of water spray effect at different monitoring points.
Case Monitoring point Temperature (K) Overpressure (barg)
2 P9 1700 0.0025
2 P18 1900 0.0025
3 P9 1400 0.0034
3 P18 1250 0.0031
Fig. 10. (a) Maximum temperature iso-surfaces of Case 2. (b) Mortality rate iso-surfaces of Case 2. (c) 3D risk iso-surfaces of Case 2.
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 513
shown here due to space limitations). They showed that the related
hazardous impact contours of Figs. 10 and 11 are similar, no matter
whether the water spray activated or not; nevertheless, the risk
values of Structure R (indicated by red squares) for the two cases
were obviously different as can be seen by their colors. The risk
value of Structure R for Case 3 was 1.0 10
12
person/yr, which is
somewhat smaller (has lighter color) than that of Case 2 (1.4
10
12
person/yr). Such phenomena can also be seen by conse-
quence analysis from the mortality rate of employees at the target
area, where Case 2 was 53.7% but Case 3 decreased to 51.2%.
Therefore, the water spray still had a positive effect on relief of
thermal radiation hazard even during a disastrous accident.
The individual risk values of three simulation cases at different
oors and regions of the target area are shown in Table 6. Because
of the largest gas cloud, Case 1 had the highest risk value (from
8.0 10
20
to 3.4 10
12
person/yr) all over the regions except for
Structures P and R since some of the ammable gas had been
dispersed to the other processes via wind effect (see Figs. 7 and 8).
For Cases 2 and 3, Structures P and R had the largest risk values
among all the regions. The risk values of Case 3 at these two
structures (6.0 10
13
and 1.0 10
12
person/yr) were smaller
than those of Case 2 (7.0 10
13
and 1.4 10
12
person/yr), the
reason for which had already been analyzed in the previous para-
graph. One can notice that Structure R1 has zero risk values for
Cases 2 and 3 because their gas clouds were relatively smaller and
were ignited immediately. Therefore, the corresponding thermal
hazards were relatively concentrated around the leak point and did
not spread so widely as that of Case 1. One can also observe that the
risk values on the high oors were all larger than those on the
ground areas, since thermal radiation played the major role in
the mortality rate in all these cases and the reball became larger
and larger once it rose up to the sky (see Fig. 8).
5. Conclusions
A CFD model, FLACS, was employed in this study to simulate the
re andexplosionconsequences of three WCScases withina complex
naphtha-crackingprocess area. Witha self-developedriskcalculation
module, the FLACS simulation results can be used to analyze the 3D
risk value inside the target area. The 3D risk analysis technique can
overcome the limitations of the traditional methods that can only
predict the risk value on the ground. Unlike the traditional methods
that usually neglect the inuence of terrain and buildings, the 3D
method can differentiate the risk difference at different heights with
the help of the CFD algorithm. The WCS simulation results showed
that inmost places employees at highlevels wouldexperience higher
risks (1.0 10
12
5.7 10
13
person/yr), while the risk values onthe
ground areas and vacant lots would be the lowest (6.0 10
20
8.0 10
20
person/yr). Therefore, an apparent risk difference existed
between different heights at the same location.
In addition, this study also investigated whether a re and
explosion hazard can be mitigated by installing a water spray
system. According to the simulation results, water spray activation
alone has very limited potential to directly reduce the hazardous
consequences of a disastrous incident. Besides, the turbulent effect
caused by water droplets can also shorten the initiation time of an
explosion and slightly increase the overpressure. All these subtle
differences can be observed via the 3Drisk values. One believes that
the 3Drisk analysis technique not merely can be used in the re and
explosion simulation in this study, but can also be employed as
a test platform in the future for validating the effectiveness of any
novel mitigating measure for chemical release (Gupta, 2005) before
it can be formally implemented in a process plant. Such application
is being studied by the authors currently. It is foreseeable there are
more and more engineering projects that related to the process
safety design can be evaluated via using this technique.
Acknowledgments
The authors are grateful to GexCon AS, Norway for providing the
FLACS software and partial nancial aid (NSC 96-2628-E-224-001-
MY3) from the National Science Council of Taiwan to support this
study.
References
Berg, J.T., Bakke, J.R., Fearnley, P., Brewerton, R.B. (2000). A CFD layout sensitivity
study to identify optimum safe design of a FPSO. OTC paper number 12159.
Cheng, T.L. (2005). The development of 3D risk analysis technology and its application
on hazardous substances release simulation within a petrochemical storage area.
Master Thesis, National Yunlin University of Science and Technology, Yunlin,
Taiwan.
Coirier, W. J., Kim, S., & Marella, S. (2005). Progress towards a coupled mesoscale
and microscale modeling capability. In American meteorological society seventh
symposium on the urban environment. Paper 4.1. http://www.ametsoc.org.
Considine, M. (1984). The assessment of individual and societal risks. SRD Report R-
310, the Safety and Reliability Directorate. Warrington: UK Atomic Energy
Authority.
CPR-14E. (1997). Methods for the calculation of the physical effects due to releases of
hazardous materials (TNO Yellow Book). Hague, Netherlands: Committee for the
Prevention of Disasters.
Dharmavaram, S., Hanna, S. R., & Hansen, O. R. (2005). Consequence analysis using
a CFD model for industrial sites. Process Safety Progress, 24, 316327.
Di Benedetto, A. (2009). CFD simulations for explosion phenomena. Journal of Loss
Prevention in the Process Industries, 22, 257.
Eisenberg, N. A., Lynch, C. J., & Breeding, R. J. (1975). Vulnerability model: A simu-
lation system for assessing damage resulting from marine spills. Springeld, VA,
Fig. 11. 3D risk iso-surfaces of Case 3.
Table 6
Comparison of individual risk value of three cases at different oors and regions.
Location Case 1 Case 2 Case 3
Ground areas 8.0 10
20
6.0 10
20
6.0 10
20
P-2 area 5.7 10
13
7.0 10
13
6.0 10
13
P-3 area 5.7 10
13
7.0 10
13
6.0 10
13
R-2 area 5.7 10
13
1.4 10
12
1.0 10
12
R-3 area 5.7 10
13
1.4 10
12
1.0 10
12
R1-2 area 3.4 10
12
0 0
R1-3 area 3.4 10
12
0 0
R1-4 area 3.4 10
12
0 0
M-2 area 8.0 10
20
6.0 10
20
6.0 10
20
M-3 area 8.0 10
20
6.0 10
20
6.0 10
20
M-4 area 8.0 10
20
6.0 10
20
6.0 10
20
Vacant lots 8.0 10
20
6.0 10
20
6.0 10
20
Y.-P. I et al. / Journal of Loss Prevention in the Process Industries 22 (2009) 506515 514
U.S.A.: US Coast Guard, Ofce of Research and Development. Report No. CG-D-
136-75, NTIS, AD-A015-245.
Finney, D. J. (1971). Probit analysis. Cambridge: Cambridge University Press.
Gupta, J. P. (2005). Dilution with air to minimise consequences of toxic/ammable
gas releases. Journal of Loss Prevention in the Process Industries, 18, 502505.
Hanna, S. R., Brown, M. J., Camelli, F. E., Chan, S., Coirier, W. J., Hansen, O. R., et al.
(2006). Detailed simulations of atmospheric ow and dispersion in urban
downtown areas by computational uid dynamics (CFD) models an applica-
tion of ve CFD models to Manhattan. Bulletin of the American Meteorological
Society, 87, 17131726.
Hanna, S. R., Hansen, O. R., Ichard, M., & Strimaitis, D. (2009). CFD model simulation
of dispersion from chlorine railcar releases in industrial and urban areas.
Atmospheric Environment, 43, 262270.
Hansen, O.R., Talberg, O., Bakke, J.R. (1999). CFD-based methodology for quantitative
gas explosionrisk assessment incongestedprocess areas: examples andvalidation
Status. In Proceedings of the AIChE/CCPS international conference and workshop on
modeling the consequences of accidental releases of hazardous
materials, ISBN: 0-8169-0781-1, 457-477, SanFrancisco, CA, September 28October
1, 1999.
I, Y. P., Chiu, Y. L., & Wu, S. J. (2009). The simulation of air recirculation and re/
explosion phenomena within a semiconductor factory. Journal of Hazardous
Materials, 163, 10401051.
I, Y. P., Tsai, W. T., & Her, D. C. (1999). Final QRA report for the developing plan of
Kwantong industrial park at Taoyuan County, Taiwan. Hsinchu, Taiwan: Industrial
Technology Research Institute.
Kisa, M., & Jelemensk y, L