E. Gallego, E. Migoya, J.M. Martín-Valdepeñas, A. Crespo, J. García (UPM) ,

1. An Intercomparison Exercise on the Capabilities of CFD Models to Predict Distribution and Mixing of H2 in a Closed Vessel E. Gallego, E. Migoya, J.M. Martín-Valdepeñas, A. Crespo, J. García (UPM), A.Venetsanos, E. Papanikolaou (NCSRD), S. Kumar (BRE), E. Studer (CEA), O.R. Hansen (GexCon), Y. Dagba (INERIS), T. Jordan (FZK), W. Jahn (FZJ), S. Høiset (N-H), D. Makarov (UU), J.Piechna (WUT)

2. Standard Benchmark Exercise Problems  SBEPs Objectives: establishing a framework for validation of codes and models for simulation of problems relevant to hydrogen safety, identifying the main priority areas for the further development of the codes/models. SBEPs in HySafe /1

3. SBEPs are enriched from the variety of codes, models, approaches, user experience and points of view from industry and research agents participating in the network Selection criteria for SBEPS: Relevance to hydrogen safety of the phenomena explored in the tests Availability of the experimental data Feasibility and their possibility to be used for validating codes and models A first experiment on hydrogen release, mixing and distribution was selected and identified as SBEP-V1 SBEPs in HySafe /2

4. Based on experiments performed by Shebeko et al. on H2 distribution for a subsonic release of hydrogen in a closed vessel (“Regularities of formation and combustion of local hydrogen-air mixtures in a large volume”, Chemical Industry, 21, 1988) Vessel dimensions: height 5.5 m diameter 2.2 m volume 20.046 m3 Initial conditions  vessel filled with quiescent air: temperature 20C pressure 760 mm Hg (101325 Pa). Hydrogen released vertically upward at a rate of 4.5 litres per second during 60 seconds (0.27 m3 total): injection tube diameter 10 mm hydrogen release velocity 57.3 m/s release orifice located on the vessel axis, at 1.4 m under the top of the vessel, connected to a supply vessel, whose pressure was about 150 atm After the release, the sensors were measuring during 250 min Description of the experiment

5. Numerical visualisation of the experiment /1 • Hydrogen released vertically upward during 60 seconds(0.27 m3 total): • injection tube diameter 10 mm • hydrogen release velocity 57.3 m/s • release orifice located on the vessel axis, at 1.4 m under the top of the vessel, connected to a supply vessel • After the release, the sensors were measuring during 250 min • Sensors located along the central axis of the vessel Video: GexCon

6. Numerical visualisation of the experiment /2 • Hydrogen released vertically upward during 60 seconds(0.27 m3 total): • injection tube diameter 10 mm • hydrogen release velocity 57.3 m/s • release orifice located on the vessel axis, at 1.4 m under the top of the vessel, connected to a supply vessel • After the release, the sensors were measuring during 250 min • Sensors located along the central axis of the vessel Video: UU

7. Organisations and codes participating

8. Important remark: all details of the experimental results were known to the modellers before the submission. Further, some results were received after the deadline, with full access to the results predicted in time by other modellers! Little can be said about prediction capabilities from the simulation performed! Being a first exercise, the main interest was in learning about the strength and limitations of the models to simulate the phenomena. The predictive power of each team will be tested against blind simulations,  a new SBEP currently under development. RESULTS…

9. Absolute velocity along the vessel axis at 30s after the beginning of release

10. FZK modelled a source orifice with 10 x the original diameter and, in order to conserve the hydrogen release, a velocity 1/100 the original. This explains the different velocity pattern. UU oscillating behaviour of vertical velocity is due to visualisation peculiarity. Unstructured grid, with centres of the control volumes positioned at different distances from the vertical axis, with lightly different vertical component of velocity. Brought all together on the vertical axis, they make impression of "wiggles“. Absolute velocity along the vessel axis at 30s after the beginning of release

11. Highly convective region associated to the hydrogen jet, where the ambient gas is entrained and mixes with the hydrogen. Recirculation flow due to the impingement of the jet on the ceiling, that generates wall jets and also produces entrainment and mixing with ambient gas. Natural convection due to non-uniform density distribution because of variable H2 concentration, and maybe also due to non-adiabatic walls. Because of the variable density and stratifications, there is also a possibility of wave-like motions that have been detected by some models. Mass diffusion, which will be turbulent in the first stages, and maybe laminar in the last ones. Relevant phenomena for the modelling and interpretation of results

12. Comparison between numerical results and experimental data should only be performed once a grid-convergence study has been made: Important to model turbulence  to demonstrate that the computed results are driven by turbulence and not by numerical diffusion due to a coarse grid. In general, if the grid is not too coarse, the evolution of gas concentration with different grids is very similar. General tendency of calculated concentrations  higher than measured in the region above the source and lower below the source. Overall findings /1

13. Volume fractions along the vessel centreline (2 min after the end of release)

14. Volume fractions along the vessel centreline (50 min after the end of release)

15. Volume fractions along the vessel centreline (100 min after the end of release)

16. Volume fractions along the vessel centreline (250 min after the end of release)

17. 2 min after the end of the release: calculated concentration almost double than the measurement in the top gauge For some models, no hydrogen was numerically found in the region below the source, whereas some H2 was registered at the first gauge below the source Possible reasons: Too coarse grid for the source region Experimental asymmetry of the exit flow? Description of the jet not sufficiently accurate No information about the measurement equipment and other objects influencing the jet inside the vessel Overall findings /2

18. Volume fractions along the vessel centreline (2 min after the end of release)

19. Ratio between calculated (Cp) and measured (C0) concentrations (2 min after the end of release)

20. For longer times: The gradual increase with time in the measured concentrations in the lower part of the vessel is seen by all the models; the agreement between experiments and models can even be improved by choosing an appropriate Prandtl number However, the fact that the measured concentrations are identical for the three lower measuring positions is not reproduced in the calculation results Approximate mass-balance estimates, made from the calculated concentrations at the gauge points, indicate a loss-of-mass in some models Overall findings /3

21. 250 min 100 min 50 min

22. Ratio between calculated (Cp) and measured (C0) concentrations (250 min after the end of release)

23. Using standard k- modelwith adiabatic walls as thermal boundary condition, and different CFD codes, a first group of four partners (BRE-A, FZJ, NCSRDa, and UPM) got very similar results: Predicted concentration levels overestimated near the top of the vessel and underestimated near the bottom In one calculation (BRE-A) with flow initially turbulent, once the hydrogen release finished, the flow eventually became laminar, and was a diffusion dominated problem: This raised the issue of whether a single turbulence model, e.g. the 'standard k- model', is suitable for both the turbulent initial release and later diffusion dominated phases (WUT proposes to use k- model for the first stages and a k- for the diffusion stage) Results by model types /1

24. GexCon, FZK and DNV applied standard k- turbulence modelsgetting better predictions with different boundary conditions: FZK used adiabatic walls but assumed different conditions at the source artificially: same mass flow rate but 100 times lower velocity. Under these conditions hydrogen has more time to mix with the surrounding air before it reaches the vessel top GexCon simulated wall heat transfer with a cold draft downwards established along the walls (in test simulations ignoring the wall temperature very little gas migrated to the lower parts of the vessel with the laminar diffusivity) Any kind of non-symmetry in temperature or external heat source/sink will contribute to better mixing, as observed in the experiments Results by model types /2

25. Using a 1D transient pure diffusion model, CEA obtained general trends similar to the previous group. Also, underestimation of the time-evolution of hydrogen enrichment in the lower part of the vessel. Some flattening of the profiles close to the injection level, attributed to the presence of the injection pipe in the grid. The results of the pure 1D diffusion model show that some other phenomena have an effect on the experimental distribution of hydrogen. Results by model types /3

26. With LVEL turbulence model NCSRDb and INERIS got very different results INERIS results underestimate the total hydrogen mass inside the tank: A filling tube was modelled from the bottom of the tank The PHOENICS version used, does not allow the setting of the inlet condition as a volume source concentration data not at the axis but at a radius close to the tube wall The LVEL model, combined with the friction on the tube wall accelerates the hydrogen diffusion at the top of the tank Better results could be obtained by using volume source with laminar inlet velocity profile as the NCSRDb results show. Results by model types /4

27. RNG-LES, unstructured grid, and adiabatic wall boundary conditions was used by UU (previous paper) The model reproduced more realistic transport of hydrogen to the bottom compared to the most of the RANS models applied Still differences with experimental data  possibly due to non-uniformity of the vessel wall temperature (the vessel was located at open air) Convective transport dominates over “turbulent” diffusion transport even at times long after the release was completed  assuming transport of hydrogen mainly by diffusion in such kind of problems seems not valid. Residual chaotic velocities are as high as about 0.10 m/s at 50 min after the release, 0.08 m/s at 100 min and 0.05 m/s at 250 min Results by model types /5

28. It is difficult to compare the combined effects of turbulence model (LES RNG, RANS k-e standard), grid (structured vs. unstructured), size of the grid, time steps... An appropriate choice of turbulence model must be made: turbulent flow becomes laminar in a relative short time Using different Prandtl turbulent numbers during the diffusion can improve the results Grid-convergence study  important to demonstrate that the computed results are driven by physical phenomena and not by numerical diffusion or inadequate grid resolution Mass balance problems occurred when the time steps were too high. Shorter time steps and stricter convergence criteria could probably guarantee the mass conservation. Conclusions about the models

29. Experiments were not ideal  Reproducibility was not reported. Temperature at release exit was not reported. Temperature at the walls was not monitored. Information on the uncertainty of the measured data not reported. Concentration values at the three lowest sensors suspiciously identical. Sensors above the source were hit by the jet and maybe not calibrated for such conditions… These issues certainly provide recommendations to experimentalists and future SBEPs. A better control of the boundary conditions is a necessary aspect in order to produce experimental data for benchmark exercises. This has to be a requirement for further SBEP exercises. Conclusions about the experiments

30. In summary, this SBEP has provided a very useful comparison of the performance of different models, which could hardly be possible to conduct by any single partner alone The reasons for hydrogen transport down to the bottom of the vessel remain a gap of knowledge. To improve our understanding of slow hydrogen movement in a closed vessel, further research on flow decay during long periods of time is needed Final Conclusions

31. Many thanks • To other colleagues who contributed to the calculations and the discussions of the results: • S. Miles (BRE), • T. Elvehøy (DNV), • J. Travis (FZK), • W. Jahn (FZJ), • V. Molkov (UU) • A. Teodorczyk (WUT) • To the HySafe Network of Excellence of the VI Framework Research Programme of the European Commission.