ASSESSING EXPLOSION HAZARDS OF LARGE HYDROCARBON CLOUDS FORMED IN CALM CONDITIONS: ARE WE DOING IT WRONG? Jerry Havens Department of Chemical Engineering University of Arkansas, USA 55th UKELG Meeting on “Dispersion and Consequences of LNG Releases” April 26, 2016 - HSE Laboratory, Buxton Derbyshire LNG IMPORT / EXPORT HISTORY (UNITED STATES) First LNG Cargo (27,400 m3 ), 1964 – Lake Charles, Louisiana to Canvey Island. Federal regulation 49 CFR 193 (Department of Transportation) Liquefied Natural Gas Facilities: Federal Safety Standards, promulgated in 1980, applicable to land portion of terminals. Approving authority for operations beyond the ship-to-shore boundary is Coast Guard. 49 CFR 193 is primarily consequence based, requiring the applicant to ensure public safety by enforcement of exclusion zones that do not extend beyond property lines for two specific hazard categories: Unignited vapor cloud travel - distance to the ½ LFL concentration level for prescribed “Design” spills. Prescribed fire heat flux limits – distance from ignited “Design” spills. To now, UVCE hazard has been neglected based on assumption of handling pure LNG (treated as methane). U. S. NOW APPROVING LNG EXPORT TERMINALS, WITH HAZARDS OF ACCIDENTAL RELEASE OF HEAVIER-THAN-METHANE HC’s. THIS PAPER IDENTIFIES POTENTIAL DEFICIENCIES IN DISPERSION MODELING AND EXPLOSION PRESSURE-LOADING (OVERPRESSURE) METHODS IN CURRENT USE. IMPROVED REGULATORY PROCEDURES ARE REQUIRED TO ADDRESS EXPORT TERMINAL HAZARDS Large design spills presently being considered for both LNG and heavy hydrocarbons make it difficult for applicants to meet the required vapor exclusion distances (distances to the ½ lfl concentration level cannot extend beyond the property line). The “new” way-around this problem is confinement of the gas cloud within property boundaries with gas-impermeable vapor fences --- introducing uninvestigated potential for increases in damaging explosion loading (overpressure). Confinement of vapor clouds by fences, in addition to increasing explosion potential, can severely limit the utility of newly approved CFD dispersion computer methods for estimating cloud concentrations and explosion overpressures. CFD dispersion models recently approved have not been evaluated adequately for predicting gas-cloud concentration fields confined by fences – introducing additional uncertainty in the accurate prediction of concentration profiles and prediction of flamerunup overpressures – most importantly with vanishingly low wind, stable, conditions. Recently approved predictions of overpressures for flammable HC clouds formed from Design Spills in low or no- wind conditions for several export terminal applications in the U.S. are using methodology for estimating the overpressures (explosive loading) that have been shown by industry evaluations to be seriously non-conservative. JORDAN COVE EXPORT TERMINAL (JCET) Proposed for Coos Bay, Oregon VERY LARGE, HIGH-COST, PLANT FULL CONTAINMENT TANK DIAMETER ~ 80 M FOUR LIQUEFACTION TRAINS SHIP OVERALL LENGTH ~ 300 M CONSTRUCTION COST ~ $6+ BILLION This paper focuses on possible errors and unacceptable uncertainties In the EIS for the JCET. The primary concern is the methodology used in the EIS to evaluate the possible consequences of unconfined vapor cloud explosions (UVCEs). Artist’s Rendition Predicting Explosion Overpressures (from documents supporting the JCET EIS) The analysis of overpressure hazards from a VCE requires two primary simulation steps. • The dispersion of the flammable cloud from a given release; • The flame front and overpressure propagation from ignition of the flammable cloud. Each of the above steps is a complex function of several parameters, including the following: • Release characteristics, location and orientation • Wind speed and direction; • Ignition timing and location. The process to identify a worst-case vapor cloud explosion is a complex task typically requiring numerous simulations. The following considerations were made (for JCET) to reduce the simulation matrix: • Releases from liquefaction trains 2 and 3 were considered; • The same scenarios selected for vapor dispersion were used; • Ignition locations selected to maximize flame acceleration to property lines. In order to determine which scenarios will produce the largest overpressure hazards, a mapping methodology is needed to convert the inhomogeneous vapor clouds into “equivalent 5 stoichiometric” clouds. The methodology used the Q9 criterion developed by Gexcon. Focus on Selected Hydrocarbon Design Spills Specified for JCET Storage Volume, gallons LNG Tanks (2) 89,662,000 Impoundment Volume, gallons 112,338,200 Design Spill Volume, gallons 0 (Full Containment*) Propane (1) 15,670 43,935 (shared) 15,670 Isopentane (1) 31,030 43,935 (shared) 31,030 Amine (1) 17,205 17,245 17,205 Ethylene (1) 14,000 43,935 (shared) 14,000 LNG line 36 inch** NA 833,400 (shared) 827,740 MR*** line 6 inch NA 833,400 (shared) 61,060 *ASSUMES NEGLIGIBLE PROBABILITY OF ANY LIQUID RELEASE **SHIP UNLOADING LINE (LARGEST LINE) ***MIXED REFRIGERANT Vapor Cloud Exclusion Zone Issues Vast Range of Design Spills for Tank and Largest Operated Line specified by Applicants “Ship Unloading Line” design spills: range 28,900 gallons to 812,000 gallons; 2004-2009 (Left to Right) Jordan Cove Ship-Unloading Line Spill appears to be the largest to date. 900000 spill size, gallons 800000 700000 600000 500000 400000 300000 200000 100000 0 The Large Spills at Jordan Cove are confined to the Property by Vapor Fences Jordan Cove Export Ship Unloading Line 827,740 gallons Into Concrete Impoundment 833,400 gallons width length depth (ft) 85 85 ~15 VAPOR FENCES 900 m For Scaling LNG TANK DIAMETER ~ 80 M SHIP LENGTH ~300 M Overall Plant Dimensions (fence location extremes) 800 m SHIP UNLOADING LINE DESIGN SPILL 827,740 Gallons Estimated Cloud Area (> ½ lfl) 720 m by 400 m = 288,000 m2 Estimated Cloud Volume assuming 2 – 5 m depth 576,000 – 1,440,000 m3 Containment by Dikes and Fences? JETTING AND FLASHING LNG ASSUMED PARTLY CAPTURED IN CONCRETE SPILL IMPOUNDMENT, FROM WHICH EVAPORATION IS PREDICTED A SHIP AT THE DOCK IS ENGULFED. THE CONCENTRATION DISTRIBUTION OF THE CLOUD IS NOT REPORTEDNo concentration field specified; preventing independent evaluation of explosion overpressure method used. ETHYLENE DESIGN SPILL 14,000 Gallons Estimated Cloud Area (> ½ lfl) 320 m by 400 m = 128,000 m2 Estimated Cloud Volume assuming 2 – 4 m depth 256,000 – 512,000 m3 JETTING AND FLASHING ETHYLENE ASSUMED PARTLY CAPTURED IN CONCRETE SPILL IMPOUNDMENT, FROM WHICH EVAPORATION IS PREDICTED PLANT IS LARGELY ENVELOPED CONCENTRATION DISTRIBUTION OF THE CLOUD IS NOT REPORTED No concentration field specified; preventing independent evaluation of explosion overpressure method used. MIXED REFRIGERANT (MR) DESIGN SPILL 61,060 Gallons Estimated Cloud Area (> ½ lfl) 720 m by 400 m = 288,000 m2 Estimated Cloud Volume assuming 2 – 4 m depth 576,000 – 1,152,000 m3 JETTING AND FLASHING REFRIGERANT ASSUMED PARTLY CAPTURED IN CONCRETE SPILL IMPOUNDMENT, FROM WHICH EVAPORATION IS PREDICTED THE PLANT AND SHIP ARE ALMOST ENGULFED. THE CONCENTRATION DISTRIBUTION OF THE CLOUD IS NOT REPORTED. No concentration field specified; preventing independent evaluation of explosion overpressure method used. THE MR IS PROPRIETARY AND IS NOT IDENTIFIED TO THE PUBLIC – THIS CREATES ADDITIONAL UNCERTAINTY The Dispersion Phase 1. The first step in predicting overpressures for a UVCE is to estimate the concentration field throughout the cloud as a function of time. In cases considered here (U.S.) this is now done almost exclusively with the FLACS model. 2. Since the gas concentration field is not uniform, the overpressures experienced will depend critically on the time and location of ignition; for each vapor cloud predicted to result from a given “Design Spill”, numerous simulations must be made to determine the importance of ignition time and location. Additional variables that must be considered are wind direction and speed – for Jordan Cove wind speeds of 1 m/s and 2 m/s were specified for (stationary) wind directions. For each “Design Spill” numerous executions of FLACS were necessary. 3. Only after these concentration fields are specified can the calculations for overpressures proceed to attempt identification of the worst cases resulting from the various times and locations of ignition. 12 Focusing on Modeling Dispersion in Calm Conditions Revisit Buncefield Parking Lot Assumed 2D Flow Area – Parking Lot Concentration Decreasing Right to Left 110 m If we can understand and predict this approximately 2-D section’s concentration dependence with time, can proceed confidently with estimation of explosion overpressures. 13 Considering Gas Gravity Current Dispersion Predictions No-wind, sufficiently dense, gas gravity currents can be turbulence-free. This has been demonstrated in laboratory (wind tunnel) experiments, justifying a realistic prediction of the dispersion in the parking lot at Buncefield. Laminarization Criteria (from published literature) Authors Criterion Expt/Buncefied McQuaid (1976) Um /(g x go)1/3 <3 Stretch (1986) (g x g ’)/U3 > 0.005 Britter (1989) (g ’ x Qo)/(U3 x W) > 0.1 Hall & Waters (1989) (g ’ x Qo1/2)/U2.5 > 1000 0.79/0.73 2.0/2.5 2.0/2.3 16329/87885 where g = gravitational acceleration, m/s2 g ’ = reduced gravitational acceleration, m/s2 Qo= volumetric flow rate, m3/s U = fluid velocity, m/s U* = friction velocity, m/s W = bund perimeter, m 1. We verified a model of the concentration field in the laboratory experiment simulating the gravity flow across the Parking Lot at Buncefield. 2. With this model we predicted concentrations in the Buncefield flow (2D section of parking lot) assuming air mixing into the cloud limited to molecular diffusion. Model Verification STEADY FEED - • Reynolds Number • Froude Number ν = kinematic viscosity, m2/s g ’ = g((ρ – ρair)/ ρair), m/s2 Model Verification Density 1 = 1.77 kg/m3 “CO2” (g’ = 4.87 m/s2) - For illustration, checks for consistency, verify model Density 2 = 1.35 kg/m3 “Buncefield” (g’ = 1.40 m/s2) - Video records to measure timing - Concentration measurements to define time-varying cloud concentrations - Measure gas concentrations with FID at specific heights and down-channel locations 16 Model Verification Synchronized video capture: side-on views of flow (smoke visualized) at 5 stations Use timed video to measure gravity current height and velocity as function of down-current time-varying position Starting cloud depth 7 cm Vertical concentration measured at downwind distances 1 m, 2 m, 3 m, 4 m at heights of 0.1, 1, 2, 3, 4, 5, 6, 7, 8 cm 3 2 Source box Cover for filling (removable) 1 5 4 Slaved Video Cameras 17 100 Concentration (mole%) Model Verification 80 8.0 cm 6.0 cm 4.0 cm 2.0 cm 0.1 cm 60 40 20 0 0 10 20 30 40 50 Time (sec) 60 70 80 90 20 30 40 50 Time (sec) 60 70 80 90 20 30 40 60 70 80 90 20 30 40 60 70 80 90 100 Concentration (mol %) 80 Experiment Data 8.0 cm 6.0 cm 4.0 cm 2.0 cm 0.1 cm 60 40 20 0 0 10 Concentration (mole%) 100 80 8.0 cm 6.0 cm 4.0 cm 2.0 cm 0.1 cm 60 40 Model Predictions 20 0 0 10 50 Time (sec) Concentration (mole%) 100 80 8.0 cm 6.0 cm 4.0 cm 2.0 cm 0.1 cm 60 40 20 0 0 10 50 Time (sec) 18 Implications for or heavy gas clouds in no wind Diffusion can act to increase the explosive volume! 110 m (300 s) 5m These predictions of Buncefield used dike-overflow concentrations estimated by HSL as starting concentrations for the cloud that advanced across the parking lot. Pink color indicates within flammable limits. 19 Predicting Explosion Overpressures The Combustion/Explosion Phase The Jordan Cove DEIS reported overpressure calculations for the mixed refrigeration and ethylene spills which resulted in the flammable clouds illustrated previously (dispersion results taken directly from the DEIS). The highest overpressures reported were slightly greater than 2 psi in the congested equipment areas that corresponded to the flammable cloud location, with maximum overpressures less than ½ psi at the property boundaries. FERC stated that approximately 2.3 psi overpressures could impact the surface of the LNG tanks, but that such overpressures were not expected to pose a threat to the integrity of the tank. Havens and Venart requested FERC to supply information substantiating the maximum overpressure predictions of approximately 2 psi. FERC replied to our comments in the FEIS, but not helpfully - provide no additional information that we could use to make an independent evaluation of the overpressure predictions that led them to approve the terminal application. Professor Venart died about the time the FEIS came out. 20 Predicting Explosion Overpressures The Combustion/Explosion Phase This following statements are based on the published paper identified immediately below. The paper, published in 2008, describes an Industry/Academia-led evaluation of current methods for determining flammable gas volumes to be considered in explosion modeling : SIMPLIFIED FLAMMABLE GAS VOLUME METHODS FOR GAS EXPLOSION MODELING FROM PRESSURIZED GAS RELEASES: A COMPARISON WITH LARGE SCALE EXPERIMENTAL DATA V.H.Y. Tam1, M. Wang1, C.N. Savvides1, E. Tunc2, S. Ferraris2, and J.X. Wen2 1EPTG, BP Exploration, Chertsey Road, Sunbury-on-Thames, TW16 7LN, UK 2Faculty of Engineering, Kingston University, Friars Avenue, London, SW15 3DW, UK IChemE Symposium Series No. 154, 2008 Quoting from the Abstract Gas explosion modelling is used widely to assess explosion loading due to overpressure or drag forces. These loadings depend upon the conditions when ignition is assumed to occur. We consider one of these conditions, namely the volume of a flammable gas cloud, specifically from a release of pressurized gas: and the application to the most commonly used commercial explosion code, FLACS. There is a number of ways this volume is defined. The three main methods are (i) volume enclosed by the LFL contour surface, (II) volume bounded by LFL and UFL contour surfaces and (iii) burning velocity weighted volume (Q9). These methods give vastly different flammable volumes …, this situation is not satisfactory. In this paper, we compared the prediction using different methods against data from large scale explosion experiments (Phase 3b.) Our results show that the burning velocity weighted volume Q9 significantly underpredicted loads for explosion higher than 0.1 bar compared with the other two measures 21 which gave neutral to slightly over-conservative prediction of explosion loads. Predicting Explosion Overpressures The Combustion/Explosion Phase Additional Quotes from the Paper Q9: Weighted Equivalent Stoichiometric Volume. Q9 is a volume measure which accounts for the effects of gas concentration by weighting the volume with the effect of burning velocity and expansion ratio. Experiments showed that burning velocity varies with concentration … For hydrocarbons, burning velocity is maximum at or near stoichiometric concentration of 1 and dropping off rapidly as gas concentration is rich … or lean …, reaching zero at UFL and LFL. Further, the expansion ratio follows a similar pattern … General Behavior of the (three) Methods. These measures give different flammable gas cloud volumes. … >LFL gives the largest volume, followed by ∆FL, then Q9. Magnitude of explosion loading would follow the same pattern. We found that Q9 measures are being used increasingly by consultants. We are concerned that there has not been work to verify that this approach is indeed correct. Our observation is that there appears to be little fundamental understanding of the Q9 measures by consultants we encountered. Its application is based on a belief that since there is a varying gas concentration in a gas cloud formed from pressurized gas release, assuming a uniform gas cloud concentration is thus ‘overconservative’, and using Q9 would remove this perceived ‘over-conservatism’. As we see … this is not necessarily so. 22 Predicting Explosion Overpressures The Combustion/Explosion Phase And: Superficially, Q9 seems to be the most accurate measure out of the three as it accounts for the well known effect of gas concentration on flame speed and expansion ratio. It may be a surprise that our results showed that the Q9 measure performs poorly. However, one should not confuse complexity and accuracy. In reality, the three measures described in this paper are no more than a much simplified and idealized representation of a complex situation… by focusing on a couple of obvious factors, many others are overlooked – here are a couple of examples: Initial Turbulence – …. A pressurized gas leak can impart a large amount of and high intensity turbulence which may be considered similar to turbulence produced by the obstacles. This is not taken into account by (any of) the simple measures discussed here. Size of the gas cloud – Limiting the flammable gas cloud to a smaller effective volume reduces the effect of flame acceleration over a larger distance and over longer period of time than that produced by larger cloud volumes and could lead to lower and the wrong distribution of overpressure…. Another reason for possible underestimation of flammable volume is that volumes with rich gas mixtures can be diluted with air or with lean gas mixtures during the course of a gas explosion, rendering the rich mixture closer to the stoichiometric ratio of 1. Applying the Q9 method blindly, it is possible to reach a conclusion that a very large leak of flammable gas would not pose a hazard. 23 Predicting Explosion Overpressures The Combustion/Explosion Phase And, finally: Following the completion of the Phase 3b exercise and publishing of the results there was no published validation of the methodology for calculating effective cloud volume measures for use with CFD codes. Any methods used should be verified against experimental data as far as possible. It should be the duty of the model developer or user of the model to verify any new methods against available data… Based on the results of this work, we recommend that the non-conservative flammable cloud volume measure ∆FL be adopted as the basis for FLACS at mid to late stage of an engineering project definition. “>LFL is a conservative measure which may be appropriate during the early stage of design of a process facility where uncertainties in the design is high. THIS WORK DOES NOT SUPPORT THE USE OF Q9 (emphasis added). 24 Conclusions – Dispersion (Cloud Formation) Modelling • Low-wind, shallow clouds, and long time scales indicate potential for important, if not controlling, molecular diffusion effects. In such conditions, even in the presence of congestion, it should be anticipated there will be little or no turbulence contribution to dispersion – raising serious questions about the prediction of cloud formation with the FLACS model. • The requirements for accurate gas dispersion prediction (cloud concentrations) under low-wind conditions, with plant equipment and/or gas confinement with fences, have not been verified for the CFD models presently used in the U.S. (FLACS). Further, the models are proprietary, so the validity of the model predictions cannot be checked independently by (or for) stakeholders. • There is important potential for molecular diffusion effects to make concentration more uniform and increase potential for explosion. 25 Conclusions - Prediction of Cloud Explosion Overpressures Vapor Cloud Explosion overpressure predictions presented in the JCET EIS are seriously deficient: • The cloud concentrations are not specified. Without such specification, it is impossible for any stakeholder to verify the predictions, which are coupled to the subsequent estimation of overpressures. • The composition of the mixed refrigerant liquid is not reported, further denying stakeholders the possibility of assessing the overpressure calculations. • The explosion hazards (overpressures) are not considered for the pure components of the MR, even though design spills for the components were selected (ethylene, propane, isopentane, and amine). • Serious questions have been raised by competent investigators regarding the application in FLACS of the Q9 method. • Approval of JCET, recently denied but requesting rehearing, should be delayed until these questions and uncertainties have been satisfactorily 26 addressed. Acknowledgments Herman, Jeremy J., “An Experimental and Computational Evaluation of the Importance of Molecular Diffusion in Gas Gravity Currents”, Doctoral Dissertation, University of Arkansas, December 2013 27
© Copyright 2026 Paperzz