1. 1/6/2016 1 Comparison of NFPA and ISO Approaches for Developing Separation Distances Jeffrey L. LaChance, Bobby Middleton, & Katrina Groth Sandia National Laboratories Albuquerque, NM Presented at the 4 th International Conference on Hydrogen Safety San Francisco, CA September 12-14, 2011

2. Presentation Outline Harmonization of NFPA and ISO Separation Distance Approaches Separation Distance Table Format Representative Facilities Approaches for Developing Separation Distances Criteria, Models, and Data Utilized Comparison of Leak Sizes Used to Determine Separation Distances 1/6/2016 2

3. Separation Distances Separation distances for small leaks – not major ruptures –Desirable to cover events that may occur during facility life time –Risk from larger events not covered by separation distances should be acceptable Quantitative Risk Analysis (QRA) was used to help establish many of NFPA and ISO separation distances QRA requires information for possible accidents: –Component leak frequencies (e.g., hoses, valves, and joints) –Ignition probabilities –Consequence models –Harm and risk criteria Under U.S. DOE sponsorship, Sandia provided methods, data, models, and manpower to support both efforts 1/6/2016 3

4. Harmonization Desirable to harmonize NFPA and ISO approaches and separation distances Commonalities in approaches: –Both use same QRA approach (limited scope QRA) –Same consequence models and component leak data Differences that challenge harmonization of separation distances: –Evaluated for different types of facilities: Bulk storage (NFPA) versus refueling facility (ISO) –Different separation table format –Different risk criteria –Application of data is different 1/6/2016 4

5. Separation Distance Table Format 1/6/2016 5 Pressure > 15 to ≤ 250 psig >103.4 to  1724 kPa > 250 to ≤ 3000 psig >1724 to  20,684 kPa > 3000 to ≤ 7500 psig >20,684 to  51,711 kPa > 7500 to ≤ 15000 psig >51,711 to  103,421 kPa Internal Pipe Diameter (ID) d mm d = 52.5 mm d = 18.97 mm d = 7.31 mm d = 7.16 mm NFPA bulk storage: Typical bulk storage facility defined for each pressure range All facility components/modules assumed to be co-located Gas volume not a variable in table format (also not a factor in QRA) Category 1Category 2Category 3 (<= 55 Mpa &<= 100kg)(> 55 Mpa & <= 3000L)(>100 kg & >3000L VSSCA SCASC ISO refueling station: Six different subsystems ranging from very simple and limited volume to complex and high volume Risk criteria applied to each subsystem (2.5 m separation between systems is required) Gas volume included in system categorization but not a factor in QRA Both standards have methods to modify separation distances in tables to account for differences in pressures and maximum component diameter.

6. ISO System Classification 1/6/2016 6

7. Comparison of Representative Facilities 1/6/2016 7 First number is limit for LPI for the system. Value in parenthesis is actual LPI for example system used to evaluate separation distance in table.

8. Comparison of Approaches NFPA approach: –Most separation distances based primarily on expected frequency of leakage events –Cumulative risk from larger leaks reviewed Risk to person at facility lot line used to establish leak size for four facility configurations (3% leak sizes chosen for all) –Other factors considered, safety margin added to address uncertainties and limited scope of analysis ISO approach: –Separation distances for six systems based only on limited risk evaluation or subjective judgment Risk to a person used to establish leak size for four Category 1 and 2 systems (variable leak sizes chosen) Leak sizes for two Category 3 systems based on subjective judgment 1/6/2016 8

9. Risk Evaluation Model Used in Both Approaches 1/6/2016 9 Risk curve is discretized to evaluate separation distances

10. Harm and Risk Criteria Both NFPA and ISO assumed exposure to hydrogen flame would result in fatality NFPA used single fatality risk criteria of 2E-5/yr to maximum-exposed individual based on: –Fatality risk at gasoline stations –10% of risk from other accidental causes –Risk criteria used in several countries ISO used two risk criteria: –Normal exposures – 1E-5/yr (International Energy Agency Task 19 (Hydrogen Safety) recommended value for fatality risk) –Critical exposures (propagation potential, potential for multiple people being harmed) – 4E-6/yr 1/6/2016 10

11. Sandia hydrogen leak models were used to evaluate safety distances in both NFPA and ISO standards Objects exposed to a hydrogen plume can encounter Heating from radiation (ignited jet) Flame impingement (ignited jet) Combustible cloud contact (unignited jet) Flame impingement and presence in 4% combustible cloud after ignition assumed to result in high probability of fatality Experimental measurements Flame shape and flame impingement distances for different flow rates Hydrogen flame radiation values Lean ignition limit for hydrogen/air mixtures Computational models with validation Jet flame radiation model Unignited jet flammability limit contour model Predictions outside the range of available data Models and experiments published in peer reviewed journal articles Consequence Evaluation Nighttime photograph of 413 bar (6000 psig) large-scale H2 jet-flame test (d j = 5.08mm, L vis = 10.6 m) from Sandia/SRI tests. 11.3 m Reference: Houf and Schefer, “Predicting Radiative Heat Fluxes and Flammability Envelopes from Unintended Releases of Hydrogen,” IJHE Paper GI-353

12. Component Leak Frequencies Currently there is insufficient hydrogen data to generate hydrogen leak frequencies using traditional statistical methods Thus, a Bayesian approach was used by SNL to generate hydrogen component leak data –Multiple sources of generic data (non-hydrogen) used to generate a “guess” for each hydrogen component leak frequency (prior distribution) –Uncertainty in assignment of generic data to specific leak sizes –Available hydrogen data used to update the prior distribution for a component to obtain a hydrogen-specific leak frequency estimate (posterior distribution) –In some cases, hydrogen data did not always match the prior distribution shape or magnitude Reference: “Handbook of Parameter Estimation for Probabilistic Risk Assessment,” NUREG/CR-6823, U.S. Nuclear Regulatory Commission, Washington, D.C. (2003).

13. Example Results - Joints 1/6/2016 13 Amount of hydrogen data (number of failures and component years of operation) is large. Generic data has little influence on shape and magnitude of hydrogen leak frequency curve. Available data suggests leak frequencies are similar over a large range of leak sizes

14. Example Results - Valves 1/6/2016 14 Hydrogen data provides similar frequencies as generic data. Generic data influences shape and magnitude of hydrogen leak frequencies.

15. ISO Leak Frequencies SNL hydrogen component leak frequencies were modified for use in ISO QRA: –Linearized (on log-log scale) Steep slopes selected for all components(not justified by SNL data results) to facilitate selection of risk-based safety distances - can result in under shorter separation distances Similar slopes for each component allows establishing “Leak Probability Indicator “(LPI) which allows modification of tabular safety distances for plant-specific configurations –Shifted an order of magnitude lower based on selected rebinning of a fraction of the generic leak frequencies into alternate bins No hydrogen data was reviewed Bayesian analysis was not performed Shifted curves provides safety distances that are a factor of 2 to 3 shorter when leak frequencies are not shifted ISO leak frequencies results in shorter safety distances than if SNL leak frequencies were used directly

16. Example of Modification of Leak Frequencies for Use in ISO QRA 1/6/2016 16 ISO curve is conservative over a large range compared to hydrogen mean from Bayesian analysis

17. Ignition Probabilities 1/6/2016 Values used in NFPA QRA Hydrogen Release Rate (kg/s) Immediate Ignition Probability Delayed Ignition Probability Total Ignition Probability <0.1250.0080.0040.012 0.125 – 6.25 0.0530.0270.08 >6.250.230.120.45 17 ISO QRA used probability of 0.04 for all leak sizes and did not differentiate between immediate and delayed ignition

18. Comparison of NFPA and ISO Leak Sizes

19. Sensitivity Results- Joints 1/6/2016 19 Shifting generic data order of magnitude has little effect on hydrogen frequencies. No justification for shifting frequencies based on this prior distribution.

20. Sensitivity Results- Valves 1/6/2016 20 ISO shifted curve is below the revised (new) hydrogen mean curve. Shifting generic frequencies had minor effect on hydrogen frequencies. Shifting hydrogen curve an order of magnitude is not justified.

21. Risk Results Using ISO Systems and NFPA Data 1/6/2016 21 Risk profile is flatter when NFPA data is utilized due primarily to variable ignition probability. Risk is acceptable but not as low as predicted with ISO data. 3% of flow area

22. Summary NFPA and ISO approaches for determining separation distances are very similar –Both use QRA, but with different levels of emphasis and complexity –Selected leak frequency distributions and ignition probabilities can significantly affect separation distances Differences between reference systems used in QRA evaluations result in differences in separation distances 1/6/2016 22

23. Additional Slides 1/6/2016 23

24. 1/6/2016 Mean Component Leakage Frequencies from Bayesian Analysis

25. 1/6/2016 System Leak Frequency Results From NFPA Analysis Expert opinion used to select 3% of system flow area captures >95% percent of the leaks covers leaks expected during facility life time the resulting separation distances protect up to the 3% leak size QRA performed to determine if associated risk from leaks greater than this is acceptable

26. Risk Results From NFPA QRA 1/6/2016 Risk close to the “guideline” of 2E-5 fatalities/yr selected by NFPA Task Group 6 Risk from leaks greater than 3% of flow area were deemed acceptable Total Risk 20.7 MPa (3000 psig) System Total Risk 103.4 MPa (15000 psig) System J. LaChance et al., “Analyses to Support Development of Risk-Informed Separation Distances for Hydrogen Codes and Standards”, SANDIA REPORT, SAND2009-0874, Printed March 2009

27. ISO QRA Results 1/6/2016 27 Risk Criteria

28. Effect of ISO Leak Frequency Modification 1/6/2016 28

29. Data Sensitivity Studies Modification of SNL leak frequency data was not based on rigorous statistical methods –A change in generic frequencies does not necessarily result in an equivalent change in hydrogen frequencies To evaluate the potential effect of generic leak frequency-size assignments, sensitivity evaluations have recently been performed –Generic leak frequencies and hydrogen information re-binned into 0.01%-0.1%, 0.1%- 1%, 1%-10%, and 10%-100% (fraction of flow area) leak size bins 1/6/2016 29

30. Alternative Prior - Joints 1/6/2016 30 Generic leak frequencies for flanges were used as an alternative prior distribution.

31. Sensitivity Results- Valves 1/6/2016 31 Shifting generic frequencies changed magnitude and shape of curves

32. Sensitivity Results- Hoses 1/6/2016 32 Shifted ISO curve provides reasonable fit if data is re-binned.

33. Sensitivity Results- Compressors 1/6/2016 33

34. Sensitivity Results- Compressors 1/6/2016 34 ISO shifted curve is below revised (new) hydrogen mean curve. Moving ISO shifted curve upwards would provide better fit.

35. Summary of Data Sensitivity Stu dy Shifted ISO leak frequencies for valves and compressors are not consistent with results of sensitivity studies where generic and hydrogen data was re-binned to lower leak sizes (i.e., leak intervals) There is justification for the shifted ISO leak frequencies for hoses and joints if generic leak frequencies are modified 1/6/2016 35

36. Impact on Separation Distances Based on results of sensitivity studies, use of shifted ISO leak frequencies for hoses and joints and non-shifted frequencies for valves and compressors results in following increase in ISO separation distances: 1/6/2016 36

37. Ignition Probability Sensitivity Study on ISO Separation Distances 1/6/2016 Use of constant ignition probability does not necessarily result in conservative separation distances in a risk- based approach 37