Journal of Loss Prevention in the Process Industries 22 (29) 34 38 Contents lists available at ScienceDirect Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp Validation of CFD-model for hydrogen dispersion Prankul Middha, Olav R. Hansen *, Idar E. Storvik GexCon AS, P.O. Box 65, Postterminalen, Bergen N-5892, Norway article info abstract Article history: Received 3 August 27 Received in revised form 29 July 29 Accepted 3 July 29 Keywords: CFD FLACS Validation Hydrogen Dispersion To be able to perform proper consequence modelling as a part of a risk assessment, it is essential to be able to model the physical processes well. Simplified tools for dispersion and explosion predictions are generally not very useful. CFD tools have the potential to model the relevant physics and predict well, but without proper user guidelines based on extensive validation work, very mixed prediction capability can be expected. In this article, recent dispersion validation effort for the CFD tool FLACS HYDROGEN is presented. A range of different experiments is simulated, including low-momentum releases in a garage, subsonic jets in a garage with stratification effects and subsequent slow diffusion, low momentum and subsonic horizontal jets influenced by buoyancy, and free jets from high-pressure vessels. LH 2 releases are also considered. Some of the simulations are performed as blind predictions. Ó 29 Elsevier Ltd. All rights reserved.. Introduction Computational Fluid Dynamics (CFD) tools have increasingly begun to play an important role in risk assessments for the process industry. While CFD-based consequence analyses have largely been limited to the oil and gas industry in the past, it is expected that these kinds of calculations will be used more and more for safety investigations for hydrogen applications. This is driven by the large interest in the possibility of using hydrogen as an energy carrier, especially in the light of the growth in global concern about the impact of greenhouse gases and the finite nature of fossil fuel reserves. Further, simplified tools for dispersion and explosion calculations for hydrogen systems are not very useful as they lack the ability to model the physical processes well which is a prerequisite for performing proper consequence modelling. CFD tools have the potential to model the relevant physics and predict the effects of a certain incident. However, the tool needs to be well validated against a range of relevant experiments (with studies on variations of various important parameters that may affect explosion loads and hence risk). Without proper user guidelines based on extensive validation work, very mixed prediction capability can be expected. The CFD tool FLACS has been developed since 98 (more details on FLACS can be found on GexCon s website http://www.gexcon.com). The inherent capability of FLACS has been performing explosion and dispersion calculations to help in the improvement of oil and gas platform safety with initial focus on the North Sea. Significant * Corresponding author. E-mail address: olav@gexcon.com (O.R. Hansen). experimental validation activity has contributed to the wide acceptance of FLACS as a reliable tool for prediction of natural gas explosions in petrochemical process areas offshore and onshore. The development and use of FLACS as a dispersion tool was mainly focused on modelling the spreading of natural gas on offshore platforms until late 99s. In 23 a model development and validation exercise with simulations of s of onshore large-scale experiments was carried out (Hanna, Hansen, & Dharmavaram, 24). This was motivated by the importance of reliable dispersion predictions in QRA studies. An accurate description of ventilation characteristics also plays a very important role in the correct estimation of the shape, size, and location of the flammable gas cloud. Through the 99s several ventilation validation studies were performed on real offshore oil installations and large-scale tests sites, including Oseberg-C (Norsk Hydro), Beryl-B (Mobil), Nelson (Enterprise Oil), Spadeadam test site, etc. At the Nelson platform GexCon performed onsite wind measurements for all major platform areas for a range of different wind conditions, and compared these to simulations. Due to the increasing significance of hydrogen, a large part of the recent effort has been focused on improving FLACS for carrying out hydrogen dispersion calculations. The current article discusses the extensive validation activity in the area of dispersion of H 2 and presents some relevant results. Different kinds of systems, involving subsonic (low momentum), sonic (high momentum), or impinging releases are considered. 2. Description of work This section describes the effort to improve the validation basis of hydrogen dispersion simulations carried out using FLACS. This 95-423/$ see front matter Ó 29 Elsevier Ltd. All rights reserved. doi:.6/j.jlp.29.7.2
P. Middha et al. / Journal of Loss Prevention in the Process Industries 22 (29) 34 38 35 H 2 concentration.2.6.2.8.4 Experiment 5 5 2 25 3 Time (s) H 2 concentration.4.3.2. Experiment 5 5 2 25 3 Time (s). Geometric Mean Fig.. Comparison between FLACS blind predictions and experimental measurements for the INERIS gallery tests (Venetsanos et al., 29) for sensor 6 (left) and sensor 2 (middle). Overall comparison for all measurement sensors in terms of a parabola plot is presented in the right figure. Geometric Variance Parabola exercise has primarily been carried out as a part of our involvement in theeuropeanunionsponsorednetworkofexcellencehysafeandiea Task 9. Some work has also been carried out under the aegis of a dedicated research project sponsored by Norsk Hydro, Statoil, and IHI where GexCon carried out small-scale dispersion experiments and FLACS simulations. In the following, some key results are presented. 2.. Subsonic jets Subsonic releases involve low-momentum releases, which may occur from a non-pressurized hydrogen source. Various experiments involving this kind of release have been considered. FLACS has been used to simulate gas dispersion experiments performed by INERIS in their gallery facility (Venetsanos et al., 29). These simulations were carried out blind with no prior knowledge of results. The experiment is a hydrogen release ( g/s for 24 s) through a 2 mm orifice on top of a release chamber (26 cm above ground) with a 2 h dispersion time thereafter in a 78.4 m 3 rectangular room. No ventilation is provided except 2 small openings near the floor on one of the walls. Concentration sensors are provided at various locations to monitor the hydrogen concentration. A comparison between FLACS blind predictions and observations is presented in Fig. for sensor 6, which was located on the jet axis.38 m above the ground and for sensor 2, which was located at a lateral distance of.4 m from the jet axis 88 cm above the ground. It can be seen that the blind predictions correlate very well with experimental data. Similar results were seen for all other sensors. 5 other partners in the HySafe project simulated the same experiment(venetsanos et al., 29). The results obtained by another company using FLACS were very consistent with our predictions. The overall results were also evaluated in terms of a parabola plot. The X-axis of such a plot is the geometric mean and the Y-axis is the geometric variance. The parabola plot shown below in Fig. confirms the very good agreement between simulation and experiments. We have also carried out blind simulations with FLACS for similar experiments using Helium conducted by CEA in its garage facility (Gupta, Brinster, Studer, & Tkatschenko, 29). The goal was validating the results obtained in the INERIS experiments as well as generating much more detailed data on the potential risks of releases in enclosed spaces. Subsequent results have indicated good agreement between simulations and experimental observations (results are shown in Fig. 2). As a part of the hydrogen research project at GexCon, many small-scale dispersion experiments (release flow rate.5 NLs ) were carried out (Hansen, Storvik, & Renoult, 25). Test D27 has been used as a benchmark for CFD code validation. The experimental rig consists of a small rectangular vessel, divided into compartments by use of four baffle plates with a vent opening at the wall opposite the release location centrally located about cm above the floor. There was some uncertainty in the experimental Fig. 2. Comparison between predicted and observed helium volume concentrations in the CEA garage geometry (test ) from start of leak until 3 s at one sensor rod directly above the release. Fig. 3. Comparison between FLACS simulation and experimental measurements for Mach ¼. release for the H 2 release experiments (Swain et al., 27).
36 P. Middha et al. / Journal of Loss Prevention in the Process Industries 22 (29) 34 38 H 2 concentration (vol. %) 2 8 6 4 2 2 4 6 8 2 data (due to sensor response time, etc.). However, good agreement with the observations is generally seen. Recently, Swain and coworkers have conducted experiments to determine the maximum distance of an ignition source to a hydrogen leak to ignite the leak successfully (Swain, Filoso, & Swain, 27). As a part of this work, they have studied horizontal leaks through an aluminum wall (2 SCFM) at two different momentums (Mach number ¼. and.2). Fig. 3 presents FLACS simulations for the Ma ¼. leak along with corresponding experimental concentration values. It can be seen that the results agree well with the experimental data. This experiment and GexCon test D27 (described above) has further led to modification of FLACS user guidelines for lowmomentum releases of buoyant gases as better grid refinement along the jet was needed to model the buoyancy accurately. 2.2. Sonic jets Distance (m) Experiments FLACS Fig. 4. Comparison between experiments and FLACS calculations for H 2 release experiments ( bar source, 3 mm nozzle) conducted by HSL (Shirvill et al., 26). Sonic releases occur when hydrogen is leaked from a highpressure source. Increased momentum results in much better mixing in this case, which can have both positive and negative influence on the hazards posed by the resulting gas cloud. We have sought to simulate many different experiments that cover this phenomenon. Deviations between observations and predictions were seen when we simulated experiments conducted by Chaineaux at INERIS (Chaineaux, 999). The experiments involved hydrogen releases from a.5 mm nozzle through a tank pressurized to 2 bar. The concentrations predicted by FLACS were about 5% higher than measured values. However, the correlation between experimental data and simulation results is much better for recent tests conducted by HSL (Shirvill, Roberts, Roberts, Butler, & Royle, 26). In particular, Fig. 4 presents experimental and computational concentrations as a function of distance for Test 7 (H 2 release through a 3 mm nozzle from a tank pressurized to bar). It can be seen that the simulations are able to predict the measured concentration values very well. These tests pointed out how crosswind could strongly influence measured concentration values. On the other hand, some overprediction of measured concentrations is again seen in the modelling of recent high-pressure release experiments conducted by FZK, especially for leaks from a.25 mm nozzle from a 6 bar source. Therefore, it can be concluded that jet concentrations seem to be overpredicted sometimes (for small nozzles/flow rates). Possible explanations for this deviation could be the fact that the internal details of the nozzle (not described in simulations) are much more important for smaller nozzles. Also, for small jets, there are more fluctuations in the resulting plume, which implies lower averaged measurements. Differences in measurement methods for releases from a small and large nozzle can also contribute to the discrepancy. Another explanation that has been propounded in discussions with other researchers is that the pseudo source approach used has limitations for the case of very small nozzles. We believe that this issue still needs to be understood in greater detail. One more limitation of the current study is that only small to medium sized releases are considered. This is due to the unavailability of any data for larger releases e.g. pipe rupture. 2.3. Impinging jets (low and high momentum) Although the simulation of INERIS experiments also involved modelling of impinging jets, FZK has recently carried out a much more extensive study which can be used to validate this phenomenon. As a part of this study, 9 different vertical leaks (varying nozzle size, release rate, and momentum) impinging on a plate were studied in two geometrical configurations: () Square horizontal plate (dimension. m) at a distance of.5 m above the release nozzle, and (2) Same set-up as configuration but with four additional vertical sidewalls of.5 m height, forming a downward open hood with a volume of 5 L above the release nozzle. The resulting gas cloud was ignited and overpressures were recorded at various different locations. More details of the experiments, including a summary of the results is available in (Friedrich et al., 27). We carried out blind simulations of the planned experiments with FLACS before the experiments were conducted. Both release phase and ignition of jets were modelled. The predictions were evaluated against experimental data after the tests were completed (Middha, Hansen, Grune, & Kotchourko, 27). This kind of experiment is very important for developing risk assessment techniques for Fig. 5. Comparison of experimental concentration profiles and FLACS predicted concentration contours for releases from mm (left figure) and 2 mm (right figure) nozzle for plate geometry.
P. Middha et al. / Journal of Loss Prevention in the Process Industries 22 (29) 34 38 37 Fig. 6. Simulated concentration contours for a H 2 release (3 g/s) from a 2 mm nozzle for plate geometries. hydrogen applications, as it is possible to study the explosion loads resulting from a realistic gas cloud. In Fig. 5, concentration profiles from FLACS blind predictions for all three release rates for the mm and 2 mm nozzles for the plate geometry are compared with experimental results (release position.5 m from plate). The figure indicates that there is a reasonable correlation between predictions and experiments for the tests. In Fig. 6, a BOS (Background oriented Schlieren) picture with numerical measurements added (2 mm nozzle, 3 g/s) is shown and compared to a predicted concentration profile. The measured concentrations and shape of plume correspond well to the predictions. Similar agreement was seen between the blind predictions and observations for all different leak scenarios in terms of hydrogen centerline and lateral concentrations and the shape of the plume (Middha et al., 27). The gas clouds were further ignited as described above. Fig. 7 presents the overpressure as a function of height above leak for four different scenarios. Comparison with experimental data revealed that in general, FLACS was able to predict representative values of overpressures for both geometries and all release scenarios. This points to the ability of FLACS to model the combined phenomenon of dispersion and direct ignition of non-homogeneous clouds with jet induced turbulence. More details are given in (Middha et al., 27). 2.4. Liquid hydrogen releases A significant fraction of hydrogen is stored and transported as a cryogenic liquid (liquid hydrogen, or LH 2 ) as it requires much less volume compared to gaseous hydrogen. One of the most important concerns in the use of liquid hydrogen to power large vehicles is that unintended releases (loss of containment) from storage tanks may spread on the ground (pool formation), evaporate and form a potentially dense hazardous gas cloud and accurate hazard assessment is needed for this. A new pool model handling the spread and the evaporation of liquid spills on different surfaces has recently been developed in the 3D Computational Fluid Dynamics (CFD) tool FLACS. The model has been validated for LNG spills on water with the Maplin Sands, the Burro and Coyote experiments (Hansen, Melheim, & Storvik, 27; Middha, Ichard, & Arntzen, 29). In the present work, the model is extended and liquid hydrogen spill experiments carried out by NASA are simulated with the new pool model. The experiments consisted of spills of up to 5.7 m 3 of liquid Fig. 7. FZK experimental overpressures compared with FLACS simulated explosion overpressures of ignited jets for ignition location along the jet axis.8 m from release nozzle.
38 P. Middha et al. / Journal of Loss Prevention in the Process Industries 22 (29) 34 38 Radius (m) 8 7 6 5 4 3 2 Data Range H 2 concentration (m 3 /m 3 ).8.6.4.2 Experiments 2 3 4 5 Time (seconds) 2 4 6 8 Time (seconds) Fig. 8. (Left) Comparison of pool radius between the simulations and experimental data (Right) Simulated and experimental hydrogen concentrations (NASA tests) as a function of time at sensor M. hydrogen (z42 kg), with spill durations of approximately 35 s. The experiments did not report any information concerning atmospheric stability. Therefore, a sensitivity study was performed with respect to this parameter. Three different Pasquill stability classes were used, namely the very stable (F), neutral (D) and unstable (B) Pasquill stability classes. Fig. 8 presents the predictions for the pool radius. The observations reported a maximum pool radius ranging between 2 and 3 m and the simulations agree reasonably well with the experiments. Fig. 8 also shows the simulated hydrogen concentrations as a function of time compared with the experimental data at sensor M (located 9. m downwind at a height of m). Good agreement between observed and predicted maximum peaks of hydrogen concentration is seen. More details are given in (Middha et al., 29). 3. Conclusions This article presents the extensive validation activity for simulation of dispersion of hydrogen in recent years. Modelling results are compared to experimental data, and in general, reasonable agreement is seen for many different kinds of release conditions. Examples of subsonic and sonic gas releases (free space and impinging) as well as liquid hydrogen releases are presented. Some of the simulations are performed as blind predictions. Acknowledgements The authors acknowledge the support of the European Union under the HySafe project (Contact No SES6-CT-24-5263) and Norwegian Research Council (RENERGI program) for this work. References Chaineaux, J. (999). Leak of hydrogen from a pressurized vessel. Workshop on dissemination of goals, preliminary results and validation of methodology. Brussels, March, 999. Friedrich, A., Grune, J., Kotchourko, N., Kotchourko, A., Sempert, K., Stern, G., & Kuznetsov, M. Experimental study of jet-formed hydrogen-air mixtures and pressure loads from their deflagrations in low confined surroundings. In: 2nd International conference of hydrogen safety. San Sebastian, Spain, September 3, 27. Gupta, S., Brinster, J., Studer, E., & Tkatschenko, I. (29). Hydrogen related risks within a private garage: concentration measurements in a realistic full scale experimental facility. International Journal of Hydrogen Energy, 34(4), 592 59. Hanna, S. R., Hansen, O. R., & Dharmavaram, S. (24). FLACS CFD air quality model performance evaluation with Kit Fox, Prairie Grass, and EMU observations. Atmospheric Environment, 38(28), 4675 4687. Hansen, O. R., Melheim, J. A., Storvik, I. E. CFD-Modeling of LNG dispersion experiments. In: AIChE Spring National Meeting, 7th Topical Conference on natural gas utilization. Houston, USA, April 23 26, 27. Hansen, O. R., Storvik, I. E., & Renoult, J. Hydrogen R&D at GexCon, experiments and simulations, Fire Bridge Second International Conference. Belfast, Northern Ireland, UK, May 25. Middha, P., Hansen, O. R., Grune, J., & Kotchourko, A. Validation of CFD calculations against ignited impinging jet experiments. In: 2nd International conference of hydrogen safety. San Sebastian, Spain, September 3, 27. Middha,P.,Ichard,M.,Arntzen,B.J.ValidationofCFDmodelingofLH2spread and evaporation against large-scale spill experiments. In: 3rd International conference of hydrogen safety. Ajaccio, France, September 6 8, 29. Shirvill, L. C., Roberts, P. T., Roberts, T. A., Butler, C. J., & Royle, M. Dispersion of hydrogen from high-pressure sources. Proceedings of Hazards XIX conference. Manchester, UK, March 27 3, 26. Swain, M. R., Filoso, P. A., & Swain, M. N. (27). An experimental investigation into the ignition of leaking hydrogen. International Journal of Hydrogen Energy, 32(2), 287 295. Venetsanos, A. G., Papanikolaou, E., Delichatsios, M., Garcia, J., Hansen, O. R., Heitsch, M., et al. (29). An inter-comparison exercise on the capabilities of CFD models to predict the short and long term distribution and mixing of hydrogen in a garage. International Journal of Hydrogen Energy, 34(4), 592 5923.