1. San Sebastian, September 11-13, 2007 2nd International Conference on Hydrogen Safety CFD SIMULATION STUDY TO INVESTIGATE THE RISK FROM HYDROGEN VEHICLES IN TUNNELS by Olav R Hansen and Prankul Middha GexCon, Norway

2. BACKGROUND: As a part of a HySafe Internal Project, HyTunnel, GexCon has simulated 200-300 hydrogen PRD release/explosion scenarios to study.  Is tunnel design important for risk?  How does tunnel ventilation rate influence risk?  Can a comparison to natural gas fuelled vehicles be carried out? Similarities to previous work by Venetsanos et al.

3. SCENARIOS STUDIED: The CFD software FLACS was applied (commercially available from GexCon). The following dispersion/explosion scenario variations were investigated:  Two tunnel configurations  4 longitudal ventilation rates (0, 2, 3, 5 m/s)  9 PRD-release scenarios from buses or cars (assuming full tank) Rectangular tunnel Horse-shoe tunnel Amount of gas is not 100% accurate for all cases, this has marginal impact on estimated risk (initial leak rate is more important)

4. RISK ASSESSMENT APPROACH: Different approaches were applied in risk-study  worst-case  ”realistic” worst-case (+ mitigation)  probabilistic QRA approach The probabilities indicated below was used for probabilistic QRA approach [Assumptions discussed in paper]

5. LEAK PROFILES Transient leak profiles representative for PRD releases were applied LH2 leak rate (11 g/s) is based on input from producer of car Worst-case PRD release from buses assume relief from 4 cylinders at the same time

6. EQUIVALENT STOICHIOMETRIC CLOUD During release calculations FLACS estimates: nFlammable volume (m 3 ): nEquivalent stoichiometric volume (Q9, m 3 ) nNew volume exposed to flammable gas (m 3 /s) Explosion in smaller stoichiometric cloud Q9 is assumed to give similar consequences as real non- homogeneous cloud. Assumption needed to limit number of scenarios Without this assumption we could end up with 1 million ignition scenarios in an extensive study e.g. 100 leak-cases x 1000 times of ignition x 10 ignition locations. Q9 is an attempt to classify the reactivity (hazard) in a realistic released cloud [LFL distance or flammable volume is not a measure of consequence]

7. IGNITION INTENSITIES FOR PROB. QRA Three ignition intensities are assumed: Spontaneous ignition nignition very soon after release starts (50% of leaks assumed ignited first 5 seconds) ncaused by shock ignition, charging of particles/equipment, fire initiating PRD, engines or more Continuous/constant ignition sources nignition by fixed ignition sources present all the time nproportional to volume exposed for the first time to flammable volume last second Intermittent (time-varying) ignition sources nignition by time-varying ignition sources nproportional to flammable volume and exposure time. Total ignition intensity for one leak scenario should be less or equal to ONE The following assumptions were used in this study

8. WORST-CASE APPROACH What if all gas could mix to a stoichiometric gas cloud Car LH2 (400m 3 ) Bus CGH2 & Car CGH2 (200m 3 ) Car & Bus NG (400m 3 ) Bus CGH2 (800m 3 ) Bus NG (1600m 3 ) Quiescent Pre-ignition turbulence

9. WORST-CASE APPROACH What if all gas could mix to a stoichiometric gas cloud 25 kg hydrogen (1000 m 3 cloud) [slightly larger than inventory].

10. WORST-CASE APPROACH What if all gas could mix to a stoichiometric gas cloud 6.2 kg hydrogen (250 m 3 cloud) [slightly larger than inventory of car or 1 cylinder bus].

11. ARE PREDICTED PRESSURES REALISTIC? Ignited gas clouds Overpressures < 0.1m 0.5m1.0m2.5m5m10m20m 2:2:1Long2:2:1Long2:2:1Long H 2 incident C Horseshoe E 0.05 0.03 0.11 0.07 0.17 0.11 0.29 0.19 0.50 0.52 1.01 0.38 1.29 0.77 3.22 2.96 2.66 4.28 6.53 9.97 NG incident C Horseshoe E 0.01.005 0.01 0.02 0.05 0.04 0.05 0.03 0.14 0.08 0.09 0.06 0.54 0.41 H 2 incident C Rectangular E 0.05 0.03 0.11 0.08 0.19 0.11 0.37 0.74 2.45 3.48 1.59 2.08 2.90 4.28 3.28 4.79 9.46 7.99 5.57 11.7 NG incident C Rectangular E.005 0.01 0.01 0.02 0.01 0.03 0.06 0.04 0.05 0.04 0.21 0.08 0.14 0.12 0.67 0.39 0.60 0.41 NG-tunnel 18m cloud closed end ignition FLACS Blind predictions 2006 0.6 to 0.8 barg FLAME Facility H 2 30m cloud FZK InsHyde experiments Clouds < 0.5m 3 Typical P = 0.05 barg FLACS validated against several tests with similar dimensions and pressures  Sandia FLAME Facility hydrogen tests 30m, ignition in closed end  NIOSH Lake Lynn Experiments methane (blind predictions)  Small scale tests with methane or hydrogen

12. REALISTIC WORST-CASE DISPERSION Dispersion calculation (Bus, 4 cylinders, horseshoe tunnel, no wind) Other scenarios seem much less dangerous (smaller leak rates) High momentum & buoyancy quickly removes gas

13. DOES VENTILATION REALLY MATTER? Worst-case scenario with initial leak rate 0.94 kg/s (and 20 kg being released). 0 m/s 2 m/s 5 m/s  No windMax flammable 1800 m 3 Max equivalent 27 m 3  2 m/sMax flammable 1500 m 3 Max equivalent 30 m 3  5 m/sMax flammable 1000 m 3 Max equivalent 25 m 3

14. REALISTIC WORST-CASE APPROACH Can dispersion simulations reduce the gas cloud size? Explosion with 25m 3 cloud shown (less than 1 kg hydrogen)

15. DISPERSION: RESULTS Worst-case volume of flammable cloud and equivalent stoichiometric cloud Notice: worst-case scenario for NG: Much larger cloud for rectangular tunnel than horseshoe shape!

16. Dispersion (and explosion of dispersed cloud) nBasic tests including numerical schemes nSMEDIS evaluation project n1998 GexCon 50m 3 nPhase 3B (GexCon 50m 3, Advantica 2600m 3 ) nKit Fox (52 CO 2 releases) nPrairie Grass (37 SO 2 tracer release) nMUST (42 C 3 H 6 tracer releases) nNYC Urban dispersion project nLNG (Burro, Coyote and Maplin Sands) Manhattan geometry model Basic tests: Standard numerical schemes will not give symmetrical impinging jet FLACS-99 FLACS-98 Kit Fox: 75 large and 6600 small obstacles MUST: 120 shipping containers Phase 3B geometry FLACS flammable volume versus experiments in 20 tests 1998 study comparing observed and simulated concentration and explosion pressure Coyote 5 LNG release simulation FLACS DISPERSION VALIDATION

17. INERIS 6; DISPERSION BENCHMARK HySafe BLIND benchmark  November 2005; 10-15 modelers delivered blind predictions  Jan-March 2006; INERIS performed experiment Scenario:  Room 7.4m x 3.8m x 2.7m  4 minute leak 1 g/s  2 hour waiting time after leak  Both GexCon and DNV used FLACS (good consistency)  Diffusion phase well predicted

18. FZK IGNITED IMPINGING JETS  9 different vertical leaks (rate/momentum) studied in two geometry configurations  April 2006, 300 FLACS blind simulations  April-June 2006 test performed by FZK (report available to HySafe January 2007)  Presented at ICHS2, San Sebastian, September 2007

19. EXPLOSION OF IDEALISED GAS CLOUDS WORST-CASE REALISTIC WORST-CASE  Realistic Worst-case ~ ear drum rupture level (+ windows will likely break)  Most explosion scenarios may not be noticed by people inside the cars

20. MAX PRESSURE 0.1-0.3 BARG: - Is that realistic? Are the consequences underestimated? Will the Q9 precision influence the results ?  Sensitivity studies igniting the worst-case scenario confirmed pressure level (ignition at 2.5s)

21. PRESSURE 0.1-0.3 BARG REALISTIC? 5 s after release 15 s after release Low volume at concentration above 15%, and this is quickly reduced with time Concentrations < 15% give limited contribution to overpressures (due to no congestion and low degree of confinement)

22. PROBABILISTIC STUDY => EXPECTED CONSEQUENCES FURTHER REDUCED Maximum cloud size is only there for a fraction of the time  likelihood for ignition before reaching maximum cloud may be high  very transient release rates makes duration of worst-case cloud size limited  This study has assumed worst-case fuel inventory, in real life there will be less fuel Flammable cloud Equivalent stoichiometric cloud

23. OUTPUT FROM QRA-STUDY NG rectangular tunnel not included: Worst-case cloud => 0.3 barg Mitigation (limit bus PRD rate to 0.23 kg/s) had significant risk reduction effect

24. CONCLUSIONS Interesting simulation study performed as part of HySafe IP02 (HyTunnel):  Buoyancy / high momentum of leak => very strong dilution (even inside tunnel)  Reactive gas only near leak at high momentum=> ventilation not important  Worst-case effects comparable for NG and H2 Conclusions depend on input assumptions (mostly conservative): Less dilution and potentially larger clouds can be seen if:  PRD-release gets trapped and loses momentum  Tunnel ceiling geometry / other release modes Further reasons for caution:  If releases may fill up volumes of cars, strong ”ignition” may be seen  Light armature or fans may work as ”dense” congestion, may cause turbulence Work is performed with a well validated CFD-Software FLACS  Observations are generally expected to have validity  Good precision of consequence tools is important to understand risk

25. FINAL COMMENTS Thanks to  European Commission  Norwegian Research Council  HySafe-partners  Commercial FLACS software For more information send an e-mail to olav@gexcon.com or prankul@gexcon.com ACKNOWLEDGMENTS QRA-method needs further development and improvements. Work to develop a better framework is being started in  HySafe NoE => HyQRA IP03  IEA Task 19 experts group