Anda di halaman 1dari 18

GCPS 2013 __________________________________________________________________________

Effect of Flammable Cloud Treatment on Predicted Blast Loads


J. Kelly Thomas, Ph.D. Baker Engineering and Risk Consultants, Inc. (BakerRisk) 3330 Oakwell Court, Suite 100 San Antonio, TX 78218 KThomas@BakerRisk.com Jihui Geng, Ph.D. Baker Engineering and Risk Consultants, Inc. (BakerRisk) 3330 Oakwell Court, Suite 100 San Antonio, TX 78218 JGeng@BakerRisk.com

Prepared for Presentation at American Institute of Chemical Engineers 2013 Spring Meeting 9th Global Congress on Process Safety San Antonio, Texas April 28 May 1, 2013 UNPUBLISHED AIChE shall not be responsible for statements or opinions contained in papers or printed in its publications

GCPS 2013 __________________________________________________________________________

Effect of Flammable Cloud Treatment on Predicted Blast Loads


J. Kelly Thomas, Ph.D. Baker Engineering and Risk Consultants, Inc. (BakerRisk) 3330 Oakwell Court, Suite 100 San Antonio, TX 78218 KThomas@BakerRisk.com Jihui Geng, Ph.D. Baker Engineering and Risk Consultants, Inc. (BakerRisk) 3330 Oakwell Court, Suite 100 San Antonio, TX 78218 JGeng@BakerRisk.com Keywords: VCE, FLACS, Blast Load, Dispersion, Equivalent Stoichiometric Cloud, Q9

Abstract
Computational fluid dynamic (CFD) analysis of flammable gas dispersion can provide a realistic evaluation of the resulting flammable gas cloud. CFD dispersion analysis can provide a more realistic assessment of the resulting gas cloud in heavily congested and/or confined environments relative to simplified dispersion analysis methods. In contrast to simplified methods, the gas cloud predicted using CFD analysis in congested/confined environments is typically complex, both in terms of shape and concentration gradients. The blast loads predicted using CFD vapor cloud explosion (VCE) analysis for the resulting flammable gas cloud are dependent upon the representation chosen for the gas cloud. Directly utilizing the predicted dispersed gas cloud, with the corresponding complexities in shape and concentration gradients, requires an extended series of analyses, since both ignition location within the cloud and the cloud location can have a strong influence on the predicted blast load. A common approach is to replace the dispersed gas cloud with a simplified representation for the purposes of performing a CFD VCE blast load assessment, such as a cubic shaped cloud with a uniform near-stoichiometric fuel concentration. This approach is very attractive since it greatly simplifies the blast load assessment. However, the relationship between the blast loads predicted using a simplified cloud representation versus those predicted for the actual dispersed cloud is not well defined. This paper presents a comparison of the blast loads predicted using the dispersed flammable gas cloud resulting from CFD dispersion analysis versus a simplified representation of the gas cloud. The FLACS CFD code was employed for this purpose. The simplified flammable gas cloud representations used included the equivalent stoichiometric gas cloud parameters developed by GexCon (i.e., the Q9 approach) and utilizing the entire flammable gas volume (i.e., from the lower flammability limit up to the upper flammability limit). A congested volume typical of an on-shore process module was employed for this purpose. Blast loads were assessed at locations both within and external to the flammable gas cloud. A range of potential ignition locations were evaluated in order to define the impact of ignition location on the predicted blast

GCPS 2013 __________________________________________________________________________

load relative to that obtained with the simplified gas cloud representations. It was found that, in general, the blast loads predicted using the equivalent stoichiometric cloud are conservative relative to those predicted for the dispersed cloud. The maximum blast loads predicted using the equivalent stoichiometric clouds were much larger than those for the dispersed clouds for the cases evaluated.

1. Introduction
The equivalent stoichiometric (i.e., Q9) approach developed by GexCon for use with the FLACS CFD code allows a real flammable gas cloud (i.e., that resulting from a dispersion analysis) to be treated in a straightforward manner for the purposes of performing a FLACS VCE blast load assessment. Hansen et al. [1] discuss the development and basis of the equivalent stoichiometric cloud approach as implemented in FLACS and provide a range of comparisons with test data and alternative gas cloud representations (i.e., dispersed cloud, replacement of entire flammable volume with worst-case concentration, etc.). The basic concept of the equivalent stoichiometric cloud approach is that the non-homogenous cloud resulting from a FLACS dispersion analysis is replaced with a cloud at a uniform concentration which will give a blast that is representative of, and generally conservative with respect to, that for the dispersed cloud. The intent of the Q9 approach is to provide a realistic worst-case blast load assessment, rather than a completely bounding result. That is, it is recognized that replacing the entire flammable gas cloud resulting from dispersion analysis with a uniform cloud of the same volume at the worst-case concentration will provide a larger blast load, and that the blast load developed in this manner will bound that for the dispersed cloud. However, since the portions of the dispersed cloud which are well away from the stoichiometric concentration (i.e., significantly lean or rich) will release less energy and will do so at a lower rate (i.e., slow flame speed), such an approach may lead to an overly conservative result. BakerRisk utilizes the Q9 approach in most FLACS blast load analyses, although a more conservative approach (i.e., cloud volume equal to the entire flammable cloud volume with the concentration set to the worst-case) is used in some cases. The dispersed clouds resulting from dispersion analyses are not normally directly ignited to determine VCE blast loads. Hansen et al. [1] noted that this was also the normal practice for GexCon, although occasionally studies are performed where the dispersed clouds are directly ignited and an example was provided of such a study (albeit without providing actual numerical results due to confidentiality restrictions). The purpose of the evaluation described in this paper was to provide an additional comparison for conditions representative of a typical on-shore process module. Section 2 describes the module utilized for this study, the corresponding FLACS model and dispersion cases evaluated. The results of the dispersion analyses are provided in Section 3. Section 4 discusses the model used for the blast load analyses. The results of the blast load analyses are presented in Section 5. Conclusions from this evaluation are provided in Section 6.

GCPS 2013 __________________________________________________________________________

2. Process Module
A typical on-shore process module was employed for this evaluation. A plan (top) view of the module is given as Figure 1. The model dimensions are 36.5 m long, 17.7 m wide and 11.5 m high (volume of 7,430 m3). The module is fully enclosed with relatively light-weight panels on all sides and the roof. A module ventilation (HVAC) system with an air flow rate of 5.4 air changes per hour (ACH) was included in the model for the dispersion analyses.

Figure 1. Module Geometry (Plan View) and Target Distribution

3. Dispersion Analysis
The background air flow induced by the ventilation system within an enclosed module strongly influences fuel dispersion and the resulting cloud size, location and concentration distribution. The ventilation system simulations indicated that the air flow patterns within the module take approximately 100 seconds to reach steady-state. The ventilation system was therefore run for 100 seconds prior to flammable gas release in each dispersion simulation. The ventilation system supply and exhaust fans were assumed to run for the entire duration of each release simulation (i.e., mitigation actions related to the ventilation systems were not considered). The averaged turbulence levels in the module from the ventilation system were used to set preignition turbulence levels for the blast load analysis. Releases of natural gas (97% methane), propane and ethylene were evaluated. Release rates in the range of 1 to 3 kg/s were found to produce the largest flammable and equivalent stoichiometric gas clouds. The results of the dispersion analyses are summarized in Table 1. As shown in this table, the flammable cloud volumes range roughly 80% to 85% of the module

GCPS 2013 __________________________________________________________________________

volume. The equivalent stoichiometric cloud volumes range from roughly 60% of the module volume (for methane and propane) up to 74% of the module volume (for ethylene); most past analyses performed by BakerRisk for enclosed process modules have shown worst-case equivalent stoichiometric cloud volumes of 50% to 60% of the enclosure volume. The larger value obtained with ethylene (74%) indicates a more uniform concentration nearer to the stoichiometric concentration. This is mainly reflective of the near neutral buoyancy of ethylene (i.e., a molecular weight of 28). This is also reflected in the ratio of the flammable cloud volume to that of the equivalent stoichiometric Q9 cloud, which ranges from 1.15 (ethylene) to 1.41 (propane). Table 1. Dispersion Analysis Results Summary Fuel Natural Gas Propane Ethylene Volume (m3) LFL-UFL 6,040 5,860 6,300 Q9 4,500 4,160 5,480 Volume (%) LFL-UFL 82% 79% 85% Q9 61% 56% 74% LFL-UFL to Q9 Volume Ratio 1.34 1.41 1.15

Figure 2 shows, as an example, the natural gas concentration contours for a 1 kg/s release at the time when the dispersed cloud attains the largest equivalent stoichiometric cloud volume. The contour plots are sliced elevation and plan views near the module center. The contour values shown are in terms of equivalence ration, and range from the LFL to the UFL. As can be seen, due to the positive buoyancy of the natural gas, the cloud is fuel-rich near the top of the module fuel-lean near the bottom. The opposite behavior is observed for the propane due to its negative buoyancy, while a more homogeneous mixing behavior is seen with ethylene releases.

4. FLACS Model for Blast Load Analysis


The solid model used in the FLACS blast load analyses was shown in Figure 1. A total of 333 targets were evenly distributed within the module to record blast pressure. Eight additional targets were placed outside the module. The four closest external targets are located 20 m from the module corresponding module wall. A computational mesh size of 0.65 m was used for all of the FLACS blast load analyses. This size was selected to ensure that there would be at least 10 computational cells through the cloud for the Q9 clouds. To represent pre-ignition turbulence due to the ventilation system, a length scale of 0.6 m and a relative turbulence intensity of 0.09 (characteristic velocity of 1 m/s) were employed. The array of potential ignition source locations shown in Figure 2 and Figure 3 was evaluated in the blast load analysis. The ignition source locations evaluated are at three elevations (z = 7 m, 10 m and 13 m) with five ignition locations along the length of the module (5 m spacing) at each elevation. Ignition source locations off the module centerline, which are shown in these figures, were also evaluated but found to give generally lower blast loads. The same ignition locations were used for both the real (i.e., dispersed) and equivalent stoichiometric clouds.

GCPS 2013 __________________________________________________________________________

Elevation View (y= 47 m)

x-y Sliced Top View (z= 10 m) Figure 2. Real (Dispersed) Flammable Gas Cloud for Natural Gas

GCPS 2013 __________________________________________________________________________

Elevation View (y= 47 m)

x-y Sliced Top View (z= 10 m) Figure 3. Equivalent Stoichiometric (Q9) Cloud for Natural Gas

GCPS 2013 __________________________________________________________________________

5. Results and Discussion


5.1 Blast Loads from Real Clouds (Natural Gas) The average of the peak overpressures for the real (i.e., dispersed) natural gas cloud internal and external to the module for all ignition source locations considered are shown in Figure 4 and Figure 5, respectively. For the internal pressures, the values shown are average of the peak pressures recorded at all monitor locations within the module. For the external pressures, the values shown are the average of the peak pressures recorded at the 4 closest monitor locations (i.e., 20 meters outside each of the module walls). There is roughly a factor of two between the largest average internal pressure (ignition location near the center of the module at the upper elevation) and the smallest; this difference is roughly 50% for the average external overpressure.

Figure 4. Impact of Ignition Location on Internal Overpressure (Real Cloud, Natural Gas)

Figure 5. Impact of Ignition Location on External Overpressure (Real Cloud, Natural Gas)

GCPS 2013 __________________________________________________________________________

5.2 Blast Load Comparison with Simplified Clouds (Natural Gas) The maximum blast overpressures for the real (i.e., dispersed) and equivalent stoichiometric natural gas clouds internal and external to the module for the worst-case ignition elevations are shown in Figure 6 and Figure 7, respectively. The values shown are for the monitor location giving the highest peak overpressure. The maximum overpressures are much larger for the equivalent stoichiometric cloud than for the real cloud. The maximum internal overpressure with the equivalent stoichiometric cloud is more than an order of magnitude larger than that for the real cloud. However, the maximum external overpressure with the equivalent stoichiometric cloud is only 3 times larger than that for the real cloud. The maximum internal overpressure is therefore not a very good indicator of the overall blast loads, and particularly not a very good indicator of the external blast loads, as the maximum may be highly localized. The average internal overpressure and impulse for the real cloud, the equivalent stoichiometric (i.e., Q9) cloud, and for the LFL-UFL cloud (i.e., a cloud with the same total flammable volume as that of the real cloud, but at the worst-case concentration) are shown in Figure 8 and Figure 9, respectively. The worst-case average internal overpressure for both the Q9 and LFL-UFL clouds are approximately 3 times that of the real cloud. The worst-case average internal impulse for the real cloud is slightly (less than 10%) higher than that for the simplified cloud representations, with the worst-case average internal impulse for the LFL-UFL cloud being slightly higher than that for the Q9 cloud. The average external overpressure and impulse for the real cloud, Q9 cloud and for the LFL-UFL cloud are shown in Figure 10 and Figure 11, respectively. The worst-case average external overpressure for both the Q9 cloud is 2 times that of the real cloud, with the LFL-UFL cloud giving about 10% higher overpressures than the Q9 cloud. The worst-case average external impulse values are similar for the three cloud representations, with the worst-case LFL-UFL cloud giving slightly higher impulse values. 5.2 Blast Load Comparison with Simplified Clouds (Propane) The average internal overpressure and impulse for the real and Q9 propane clouds are shown in Figure 12 and Figure 13, respectively. The worst-case average internal overpressure for the Q9 cloud is almost 4 times that of the real cloud. The worst-case average internal impulse for the Q9 cloud is 50% higher than that for the real cloud. The average external overpressure and impulse for the real and Q9 propane clouds are shown in Figure 14 and Figure 15, respectively. The worst-case average external overpressure for the Q9 cloud is 3 times larger than that of the real cloud. The worst-case average external impulse for the Q9 cloud is about 20% larger than that of the real cloud.

GCPS 2013 __________________________________________________________________________

Figure 6. Internal Maximum Overpressure Comparison (Natural Gas)

Figure 7. External Maximum Overpressure Comparison (Natural Gas)

GCPS 2013 __________________________________________________________________________

Figure 8. Internal Average Overpressure Comparison (Natural Gas)

Figure 9. Internal Average Impulse Comparison (Natural Gas)

GCPS 2013 __________________________________________________________________________

Figure 10. External Average Overpressure Comparison (Natural Gas)

Figure 11. External Average Impulse Comparison (Natural Gas)

GCPS 2013 __________________________________________________________________________

Figure 12. Internal Average Overpressure Comparison (Propane)

Figure 13. Internal Average Impulse Comparison (Propane)

GCPS 2013 __________________________________________________________________________

Figure 14. External Average Overpressure Comparison (Propane)

Figure 15. External Average Impulse Comparison (Propane)

GCPS 2013 __________________________________________________________________________

5.3 Blast Load Comparison with Simplified Clouds (Ethylene) The average internal overpressure and impulse for the real cloud, Q9 cloud and for the LFL-UFL cloud are shown in Figure 16 and Figure 17, respectively. The worst-case average internal overpressure for the Q9 cloud is 30% higher than that of the real cloud, with the LFL-UFL cloud giving about 20% higher overpressures than the Q9 cloud. The worst-case average internal impulse values are similar for the three cloud representations, with the worst-case LFL-UFL cloud giving slightly higher impulse values.. The average external overpressure and impulse for the real cloud, Q9 cloud and for the LFL-UFL cloud are shown in Figure 18 and Figure 19, respectively. The worst-case average external overpressure for the Q9 and real clouds are approximately equal, with the LFL-UFL cloud giving about 20% higher overpressures than either the Q9 or real clouds. The worst-case average external impulse values are similar for the three cloud representations, with the worst-case real cloud giving the largest maximum external impulse.

5.4 Blast Load Comparison Summary Table 2 provides a summary of the worst-case average blast loads (over pressure and impulse) internal and external to the module for the real (i.e., dispersed) and equivalent stoichiometric (i.e.., Q9) gas clouds. As noted previously, the natural gas and propane flammable gas cloud (LFL-UFL) volumes are significantly larger that the Q9 due to significant concentration gradients (i.e., significant portions of the worst-case cloud are lean or rich), whereas this is not the case for the ethylene clouds due primarily to its near neutral-buoyancy. Table 2. Summary of Q9 to Real Blast Load LFL-UFL to Q9 Volume Ratio 1.34 1.41 1.15 Ratio of Worst-Case Average Blast Load (Q9 to Real Cloud) Internal to Module Overpressure 3.0 3.7 1.3 Impulse 0.8 1.5 1.0 External to Module Overpressure 2.0 3.1 1.0 Impulse 0.9 1.2 0.9

Fuel Natural Gas Propane Ethylene

The worst-case average internal blast overpressures for the Q9 clouds are 3 to 4 times those for the real clouds with natural gas and propane, and those external to the module for the Q9 clouds are 2 to 3 times higher than those for the real clouds. Conversely, the worst-case average internal blast overpressure for the Q9 cloud is only 30% larger than that for the real cloud with ethylene, with the worst-case average internal blast overpressures being nearly equal for the Q9 and real clouds. The differences between the worst-case blast impulses are less dramatic than for the overpressures. The worst-case average internal impulse for propane is 50% larger with the Q9 cloud than for the real cloud. For the other fuel/location combinations, the difference between the Q9 and real cloud impulses are less than approximately 20%.

GCPS 2013 __________________________________________________________________________

Figure 16. Internal Average Overpressure Comparison (Ethylene)

Figure 17. Internal Average Impulse Comparison (Ethylene)

GCPS 2013 __________________________________________________________________________

Figure 18. External Average Overpressure Comparison (Ethylene)

Figure 19. External Average Impulse Comparison (Ethylene)

GCPS 2013 __________________________________________________________________________

6. Conclusion
The results of the analyses presented in this paper show that, for the conditions examined (ventilated process module, fuel release rate leading to largest flammable cloud, specified fuels, range of ignition source locations, etc.), the FLACS equivalent stoichiometric cloud (Q9) approach provides generally conservative loads with regards to those for the real (dispersed) flammable gas cloud. The maximum blast loads predicted using the equivalent stoichiometric clouds were much larger than those for the real clouds. The natural gas and propane flammable gas cloud volumes are significantly larger that the Q9 clouds due to significant concentration gradients (i.e., significant portions of the worst-case flammable gas cloud are lean or rich). This was not the case for the ethylene clouds due primarily to its near neutral-buoyancy. For the natural gas release, it was shown that the maximum overpressures are much larger for the Q9 cloud than for the real cloud. The maximum internal overpressure with the Q9 cloud was more than an order of magnitude larger than that for the real cloud, while the maximum external overpressure is 3 times larger. Comparing approaches based on the maximum internal overpressure may be misleading in terms of the overall blast loads as the maximum internal overpressure may be highly localized. The worst-case average internal blast overpressures for the Q9 clouds were 3 to 4 times those for the real clouds with natural gas and propane, and those external to the module for the Q9 clouds were 2 to 3 times higher than those for the real clouds. The differences between the two approaches are much less with ethylene (i.e., due to the similarity in the Q9 and real clouds). The differences between the worst-case blast impulses are less dramatic than for the overpressures. The worst-case average internal impulse for propane is 50% larger with the Q9 cloud than for the real cloud. The difference between the Q9 and real cloud impulses both internal and external to the module for natural gas and ethylene are less than approximately 20%.

7. References
[1] Hansen, O.R., F. Gavelli, S.G. Davis and P. Middha (2011) Equivalent Cloud Methods used for Explosion Risk and Consequence Studies, Mary Kay OConnor Process Safety Center, 14th Annual Symposium, College Station, TX, 25-27 October 2011, pp. 78-106.

Anda mungkin juga menyukai