Search:

# Knowledge gaps and recent progress

• Hydrogen Dispersion

### Contributing Authors

 Contributing author Main contributions Organisation A.G.Venetsanos Chapter coordinator NCSRD K. Verfondern Dispersion in the open atmosphere FZJ A. Crespo, E. Migoya, J. García, E. Gallego Dispersion in obstructed environment UPM J/M Martin-Valdepenas and M.A. Jimenez Dispersion in confined environment UPM A.G. Venetsanos, K. Verfondern and P. Benard Numerical simulations NCSRD, FZJ, UQTR A.G. Venetsanos and P. Benard Knowledge gaps and recent progress NCSRD
 Contributing reviewer Organisation H.S. Ledin HSL O.R. Hansen GEXCON

### Dispersion in the Open Atmosphere

Many different accident situations are conceivable, which can give rise to the inadvertent emission of a flammable substance and which have great influence on the evolution of a vapor cloud. It can be released as a liquid or a gas or a two-phase mixture. The component, from which the substance is released, may be a tank, a pump, a valve, pipe work or other equipment. The orifice, through which it is leaking, can vary over different shapes and sizes. The leaking fluid can flow into different geometries. And finally it is the thermodynamic conditions of the fluid, which determine its release behaviour. Four major categories for the release of liquid or gaseous hydrogen can be identified:

1. small-scale, moderate hydrogen release from permeation or boil off;
2. vaporization of a liquid hydrogen pool on a solid or liquid surface;
3. two-phase jet release of hydrogen after opening a system under pressure;
4. rapid escape of hydrogen to all sides after the catastrophic failure of a pressure.

## Phenomena

The generation of a gas cloud in the atmosphere is principally caused by forces resulting from the internal energy of the gas and/or from energy inside the system, from which the gas has escaped, or from a relative excess energy in the environment. Those opposed are dissipative forces, among which atmospheric turbulence is the most important one.

In case there is no early ignition, the vapour cloud shape is further determined by density differences, atmospheric conditions, and topography. Several phases of a gas cloud formation can be distinguished: In the early phase, the gas cloud is still unmixed and usually heavier than the ambient air. Its spreading is influenced by gravitational force resulting in a near-ground, flat cloud. The following phase is characterized by a gradual entrainment of air from outside into the gas cloud enlarging its volume, thus lowering gas concentration, and changing its temperature. In the final phase, due to atmospheric dispersion, density differences between cloud and ambient air will be leveled out, where concentrations eventually fall below flammability limits. Thus density of the gas mixture vapor cloud varies with time.

The turbulence structure of the atmosphere is composed of large-scale turbulence described by the large-scale wind field, and of isotropic turbulence, which is a rapid variable superimposed to the medium wind field. The latter is generated due to the fact that "roughness elements" withdraw from the medium wind field kinetic energy, which is transferred to turbulence energy. It is this energy and of particular importance the small eddies, which finally determine the spreading of the gas cloud; the larger eddies are responsible for its meandering. Further factors influencing the turbulence structure within a gas cloud, apart from the atmospheric turbulence of the wind and temperature field inside the turbulent boundary layer (5 mm < z < 1500 m), are:

1. velocity gradient (sheer force between wind field and gas cloud);
2. current created by buoyancy forces;
3. heat transfer from ground into cold gas (thermally induced turbulence);
4. rapid expansion from vaporization of cryogens.

Fluctuations in the concentration as a consequence of the atmospheric turbulence are typically in the order of a factor of 10 above the statistical average.

The spreading of a gas cloud in the atmosphere is strongly influenced by the wind conditions which change with height. Vertical wind profiles can be determined as a function of the so-called stability categories depending on the temperature conditions. As an example, Pasquill suggested the categories A, B, C for unstable, D for neutral, E, F for stable conditions. The spreading mechanism of a gas in the atmosphere is mainly by mixing with the ambient air. The boundary layer between gas and air governs momentum and mass exchange, which is much stronger than molecular diffusion. Horizontal dispersion perpendicular to wind direction is about the same for all stability categories; it is different for vertical dispersion. Under stable conditions, vertical exchange is small leading to a long-stretched downwind gas cloud. In contrast, a temperature decrease with height, which is stronger than the adiabatic gradient (-0.98 K/100 m), results in an effective turbulent diffusion and rapid exchange. This is particularly true for a hydrogen gas cloud, which behaves in a neutral atmosphere as if it were in an unstable condition. Worst-case scenario would be the existence of a large hydrogen gas cloud generated with minimal internal turbulence, on a cold, humid day with high wind velocity and strong atmospheric stability.

The jet release of a liquefied cryogen under pressure is connected with the formation of aerosols. The two (or three)-phase mixture developed exhibits an inhomogeneous concentration distribution. There will be a rapid vaporization, which may create locally high H2 concentrations. It was observed that the larger the liquid fraction of the two-phase jet, the larger was the evolving flammable vapor cloud (KneeboneA:1974). Another effect observed for vertical jet-like gas release under certain conditions is a bifurcation of the plume into two differently rotating vortices. After a short acceleration phase, a double vortex is developing which eventually splits up. This effect may reduce the height of the gas cloud and lead to a stronger horizontal spreading (ZhangX:1993).

With respect to just vaporized LH2, the lifetime as a heavier-than-air cloud (1.3 kg/m3) is relatively short. It needs only a temperature increase of the hydrogen gas from 20 K to 22 K to reach the same density of the ambient air (1.18 kg/m3). This short time span of negative buoyancy is slightly prolonged by the admixed heavier air, before the buoyancy becomes positive and enhances with further temperature increase. Unlike pool vaporization leading to only weak vapor cloud formation, instantaneous release of LH2 or high release rates usually result in intensive turbulences with violent cloud formation and mixing with the ambient air. If LH2 is released onto water, rapid phase transitions occur, which are connected with very high vaporization rates. The exiting vaporized gas also carries water droplets into the atmosphere increasing the density of the vapor cloud and thus influencing its spreading characteristics.

The spreading behaviour of a large gas cloud is different from a small one meaning that the effects in a small cloud cannot necessarily be applied to a large one. For small releases, the dynamics of the atmosphere are dominant and mainly covering gravitational effects due to the rapid dilution. For large amounts released, the evolving gas cloud can influence itself the atmospheric wind conditions changing wind and diffusion profiles in the atmosphere. This so-called "vapor blanket" effect could be observed particularly at low wind velocities, where the atmospheric wind field was lifted by the gas cloud and the wind velocity inside the cloud dropped to practically zero.

The near-ground release of cryogenic hydrogen resulting in a stable stratification has, in the initial phase, a damping influence on the isotropic turbulence in the boundary layer to the ambient air, thus leading to a stabilization of the buffer layer (so-called "cold sink effect"). For small wind speeds, additional effects such as further heating of the gas cloud due to energy supply from diffusion, convection, or absorption of solar radiation, as well as radiation from the ground will play a certain role, since they reduce gas density and enhance positive buoyancy.

A still deep-cold hydrogen gas cloud exhibits a reduced heat and mass exchange on the top due to the stable stratification. A stronger mixing will take place from the bottom side after the liftoff of the cloud resulting from buoyancy and heating from the ground. The dilution is slightly delayed because of the somewhat higher heat capacity of hydrogen compared to air. In case of a conversion of para to ortho hydrogen, a heat consuming effect (708.8 kJ/kg), reduces the positive buoyancy. This process, however, is short compared to dispersion.

Another effect determining a cold hydrogen cloud behaviour is the condensation and solidification, respectively, of moisture which is always present in the atmosphere. The phase change is connected with the liberation of heat. Therefore density is decreased and thus buoyancy is enhanced. The higher the moisture in the atmosphere, the sooner is the phase of gravitation-induced spreading of the vapour cloud terminated. The effect of condensation also results in a visible cloud, where at its contour lines, the temperature has just gone below the dew point. For high moisture contents, the flammable part of the cloud is inside the visible cloud. For a low moisture content, flammable portions can also be encountered outside the visible cloud. Visible and flammable boundaries coincide at conditions around 270-300 K ambient temperature and humidities of 50-57%.

According to the "model of adiabatic mixing" of ambient air and hydrogen gas, assuming there is no net heat loss or gain for the mixture, there is a direct correlation between mixture temperature and hydrogen concentration, if air temperature and pressure and relative humidity be known. This means on the other hand that thermocouples could be used as hydrogen detectors. The model was found to be in good agreement with measured concentrations. Taking the conditions of the NASA LH2 spill trials as an example, the cloud boundaries were assessed of having had a hydrogen concentration of around 8-9%.

The topography has also a strong influence on the atmospheric wind field and thus on the spreading of the gas cloud. Obstacles such as buildings or other barriers increase the degree of turbulence such that the atmospheric stability categories and their empirical basis are loosing their meaning locally. This situation requires the application of pure transport equations which may become very complex due to the generation of vortices or channeling effects (PerdikarisGA:1993). A gas cloud intersecting a building will be deflected upwards reducing the near-ground concentration in comparison to unobstructed dispersion. On the other hand, if the source is near the building in upwind direction, a vortex is created with a downwards directed velocity component, which may increase the near-ground concentration. This effect, however, may be more important for heavy gases than for the lighter gases.

## Experimental Activities

The first hydrogen release experiments conducted with LH2 date back to the late 1950's (CassutLH:1960), (ZabetakisMG:1961). They included, however, only little information on concentrations and were basically limited to visual recordings. The experimental series with LH2 release conducted by Arthur D. Little were dedicated to the observation of the dispersion behavior showing that still cold hydrogen gas does not rise immediately upwards, but has the tendency to also spread horizontally. The initial column-like cloud shape later transforms into a hemispherical shape. Measurements of the translucence reveal large variations in the concentrations indicating incomplete mixing (see Fig. 1). The continuous release at a rate of 2 l/s over 16 min and of 16 l/s over 1 min and for wind speeds between 1.8-7.6 m/s, the developing visible vapor cloud had an extension of up to 200 m before fading away. Gusty winds had the effect of splitting up the gas cloud.

Fig 1: Shape of H2-air cloud, (ZabetakisMG:1961)

The first and up to now most relevant test series to study hydrogen dispersion behaviour was conducted by NASA in 1980 with the near-ground release of LH2. In five tests, a volume of 5.7 m3 was released within 35-85 s; in two more tests the released volumes were 2.8 m3 in 18 s and 3.2 m3 in 120 s, respectively (WitcofskyRD:1984) and (ChirivellaJE:1986). Eight times the concentration was measured at a total of 27 positions. Temperature measurements were also indicators for H2 concentration. These trials have shown that drifting of the H2 vapour cloud can go up to several hundred m, particularly if the ground is able to sufficiently cool down. The tests also demonstrated that the vaporization rate of LH2, much more than that of other cryogens, is strongly dependent on the type of release.

In 1994, the German BAM has conducted LH2 release experiments with the main goal to demonstrate the safety characteristic of a rapidly decaying hydrogen vapour cloud in the open atmosphere in contrast to the behaviour of vaporizing LPG. Six LH2 spill tests were conducted with amounts released of 0.5-1 m3 (total 260 kg) at rates of around 0.6-0.8 kg/s. The tests were also to show the influence of adjacent buildings on the dispersion behaviour (SchmidtchenU:1994), (DienhartB:1995).

## References

Invalid BibTex Entry!

### Dispersion in Obstructed Environment

When studying the hydrogen dispersion in obstructed environment it should be taken into account that the dispersing cloud behaviour completely differs for the cases of gaseous and liquefied hydrogen spills. Actually, hydrogen disperses as a heavier-than-air gas when escapes to the atmosphere from the liquid state and is characterized by horizontal movement and relatively long dilution times, whereas in gaseous form hydrogen is a buoyant gas. In this sense some of the results presented for the dispersion of other flashing liquids could be applicable to the hydrogen dispersion. For example Chan (ChanST:1992) performed calculations for the numerical simulations of LNG vapour dispersion from a fenced storage area, and found that vapour fence can significantly reduce the downwind distance and hazardous area of the flammable vapour clouds. However, a vapour fence could also prolong the cloud persistence time in the source area, thus increasing the potential for ignition and combustion within the vapour fence and the area nearby. Also Sklavounos and Rigas (SklavounosS:2004) performed a validation of turbulence models in heavy gas dispersion over obstacles, which could also be applied for the earlier stages of the spill.

In the presence of buildings or other obstacles, the wind direction is also expected to play an important role for the cloud dispersion, due to the shielding effects of these obstacles.

The obstacle effect is twofold, in one way it inhibits gas convection, but on the other hand creates turbulence, thereby increasing gas dilution, extending the flammable region, and even accelerating the flame. Hydrogen may cause a series of accidental events (jet fire, flash fire, detonation, fireball, confined vapor cloud explosion), depending on the time of ignition and the space confinement. Unless an immediate ignition occurs, it becomes evident that dispersion calculation is required, in order to determine the lower flammable limit zones in the greater area of hydrogen facilities and hence preventing, via appropriate measures, flash fires and confined vapor cloud explosions corresponding to delayed ignition, see for example the work of Rigas and Sklavounos (RigasF:2005).

Though accidents related to storage and use of hydrogen will certainly occur, there is not much data available in the literature about what happens when liquid hydrogen is accidentally released near the ground between buildings of a residential area. Only few numerical codes used for dispersion estimation can be applied to hydrogen, which means that further developmental work is necessary in this field.

Statharas et. al. (StatharasJC:2000) describe the modelling of liquid hydrogen release experiments using the ADREA-HF 3-D time dependent finite volume code for cloud dispersion and compare with experiments performed by Batelle Ingenieurtechnik for BAM Bundesanstalt fur Materialforschung und Prufung , Berlin, in the frame of the Euro-Quebec-Hydro-Hydrogen-Pilot-Project. They mainly deal with LH2 near ground releases between buildings. The simulations illustrated the complex behaviour of LH2 dispersion in presence of buildings, characterized by complicated wind patterns, plume back flow near the source, dense gas behaviour at near range and significant buoyant behaviour at the far range. The simulations showed the strong effect of ground heating in the LH2 dispersion, as can be observed comparing figures Fig. 1 and Fig. 2. The model also revealed major features of the dispersion that had to do with the "dense" behaviour of the cold hydrogen and the buoyant behaviour of the "warming-up" gas as well as the interaction of the building and the release wake.

Fig. 1: Predicted 4% iso-surface at t=100 s, without ground heating effects (StatharasJC:2000)

Fig. 2: Predicted 4% iso-surface at t=100 s, with ground heating effects (StatharasJC:2000)

Schmidt et. al. (SchmidtD:1999) performed a numerical simulation of hydrogen gas releases between buildings. Gas cloud shape and size were predicted using the Computational Fluid Dynamics Code FLUENT 3. The modelling was made as close as possible to the pattern of the liquid hydrogen release experiments performed by BAM in the framework of the EQHHPP. Four main results were found.

The release of hydrogen at high velocities (up to the critical velocity) results in a much more hazardous situation than a release at low exit velocities. At high velocities, high concentrations of H2 near the ground and a considerable enlargement of the range where explosive mixtures occur, have to be expected. See figures Fig. 3 and Fig. 4.

Fig. 3: Fast release in z direction, 4 Vol. % iso-surface (SchmidtD:1999)

Fig. 4: Slow release in z direction, 4 Vol. % iso-surface (SchmidtD:1999)

Venetsanos et. al. (VenetsanosAG:2003) performed a study of the source, dispersion and combustion modelling of an accidental release of hydrogen in an urban environment. The paper illustrates an application of CFD methods to the simulation of an actual hydrogen explosion. Results from the dispersion calculations together with the official accident report were used to identify a possible ignition source and estimate the time at which ignition could have occurred, see Fig. 5. The combustion simulation shows that an initially slow (laminar) flame, that accelerates due to the turbulence generated by the geometrical obstacles in the vicinity (primarily the pressure tanks). Since the hydrogen cloud is very concentrated, with a large region with more than 15% hydrogen by volume, there is ample scope for flame acceleration. However, the general geometrical configuration is rather open, and beyond the bundles of pressure tanks there are few obstacles. This will tend to restrain the acceleration of the flame and prevent the flame from accelerating to very high speeds as seen in the simulations. These results are to a large extent compatible with the reported accident consequences, both in terms of near-field damage to building walls and persons, and in terms of far-field damage to windows. Their results demonstrate that hydrogen explosions in practically unconfined geometries will not necessarily result in fast deflagration or detonation events, even when the hydrogen concentration is in the range where such events could occur in more confined situations

Fig. 5: Predicted velocity and volume concentration field on a vertical cross-canyon plane at 9 m downwind from the source and time 10 s after start of the release (VenetsanosAG:2003)

## References

Invalid BibTex Entry!

• Dispersion in Confined Environment

### Hydrogen Behaviour

The accidental release of hydrogen in confined environment differs from the open atmosphere and semi-confined cases in the fact that the leakage is located in a room. Then, the released hydrogen gets mixed with the room atmosphere, building up there or dispersing outwards through venting holes.

Depending on the storage system, hydrogen leaks as liquid or gaseous phase. For leaks involving LH2, vaporization of cold hydrogen vapour towards the atmosphere may provide a warning sign because moisture condenses and forms a fog. This vaporization process usually occurs rapidly, forming a flammable mixture. On the other hand, for GH2 leaks, gas diffuses rapidly within the air.

The hydrogen gas released or vaporised will disperse through the environment by both diffusive and buoyant forces. Being more diffusive and more buoyant than gasoline, methane, and propane, hydrogen tends to disperse more rapidly. For low-momentum, gaseous hydrogen leaks, buoyancy affects gas motion more significantly than diffusivity. For high-momentum leaks, which are more likely in high-pressure systems, buoyancy effects are less significant, and the direction of the release will determine the gas motion; on the other hand, a jet is established, which reduces its inertia while it mixes.

Conversely, saturated hydrogen is heavier than air at those temperatures existing after evaporation. However, it quickly becomes lighter than air, making the cloud positively buoyant. At the end, localized air streams due to ventilation will also affect gas movement. Therefore, in all cases, a light gas cloud is developed near to the leak. It is rich in hydrogen, which is less dense than air in the room. This density difference induces a vertical buoyant force, making the hydrogen-rich cloud rise up and the heavier atmosphere air drop down. A region which is richer in hydrogen is developed and a buoyant plume is established. This plume mixes with the surrounding atmosphere but in a non-homogeneous way. When the plume impinges the top of the enclosure, it spreads throughout the ceiling and stagnates there. Depending on the release location and the geometrical aspect ratio (slenderness) of the building, the inertia forces would be able to drive the atmosphere to either well-mixed or stratified conditions.

In the medium and long-term, other mixing phenomena could appear and change the atmosphere conditions. Other releases of hydrogen could push the hydrogen-rich cloud downwards favouring homogeneous conditions. Heat transfer (mainly by convection) between room atmosphere and walls could induce secondary circulation loops, thereby enhancing the mixing processes.

All of these phenomena yield the final distribution of hydrogen within the confined environment: well-mixed, stratified, locally accumulated, etc. Moreover, the presence of some systems could change these conditions, normally helping the mixing process. They are: venting systems, connections to other rooms, fan coolers, sprays, rupture disks, etc. In order to deal with accident events, some mitigation systems have been developed as dilution systems (by injection of inert gas), igniters (which burn flammable mixtures) and recombiners (which oxidise hydrogen in a controlled way) etc. Valuable devices are the Passive Autocatalytic Recombiners (PAR), which reduce the hydrogen mass by inducing the reaction with oxygen at low hydrogen concentrations, using palladium or platinum as catalysts.

In summary, six stages of a release in a confined environment can be identified the phenomena after a release in a confined environment can be grouped together as:

(1) leakage;
(2) jet;
(3) buoyant plume;
(4) homogeneous/stratified dispersion;
(5) convective and venting phenomena; and
(6) mitigation systems (if any).

## References

Invalid BibTex Entry!

### Molecular versus Turbulent Mixing

The relative importance of advection and diffusion in the distribution or mixing of a chemical species like hydrogen may be derived through non-dimensioning the general advection-diffusion equation of transport. This dominance is a function of flow velocity u, species diffusivity D and time t, and may be expressed in terms of the non-dimensional Péclet number:

 $$Pe = {D \over {u^2 t}} = {D \over {uL}}$$ (1)

where D is the diffusion coefficient, u is the velocity, t is time and L is a characteristic length scale. Diffusion is the dominant mechanism when Whereby diffusion is dominant for Pe>>1 and transport by advection transport dominates for Pe<<1. It is important to note that, whenever large times, t, or characteristictravelled lengthscale, L = ut, are considered, the advection transport would always dominate. The Péclet number expresses the ratio between the characteristic times of advection and diffusion. The length travelled by a particle is proportional to t for advection and to t1/2 for diffusion.

Characteristic length and time scales for advection and diffusion transport may be expressed by

 $$L_{advection} = ut$$ (2) $$t_{advection} = {L \over u}$$ (3) $$L_{diffusion} = \sqrt {Dt}$$ (4) $$t_{diffusion} = {{L^2 } \over D}$$ (5)

These expressions are useful as rules of thumb.

## References

Invalid BibTex Entry!

### Potential for Accumulation Depending on Leaking

When the jet or buoyant plume is established within the enclosure, the medium-term atmosphere conditions would be either well-mixed or inhomogeneous. The relevant phenomena are:

1. local accumulation on dead-end regions
2. stratification-homogenisation in ceilings
3. homogenisation by convective motions
4. venting phenomena
5. mitigation systems

Local accumulation usually happens in regions near the release point or in the way of circulation loops. There are regions in the enclosure with dead-end enclosures, badly-vented or ceiling, which make difficult the ascending dispersive motions.

Stratification-homogenisation in ceilings is a more complex phenomenon. The stratification consists on forming stable layers of fluid which do not mix each other, because of the lack of atmosphere gradients apart from jets, plumes or boundary layers. When stratification happens the fluid is not stagnated, but the motion does not allow mixing between separate layers.

The mixing patterns established in the enclosure are induced by jets, plumes and convective heat transfer. These phenomena induce moments in the fluid, which produce the competition between two forces: inertia and buoyancy. When the inertia forces are dominant, the enclosure atmosphere will get mixed. When buoyancy is prevailing the stratification remains, and in this case the vertical gradient is established as a balance of ((WoodcockJ:2001)):

1. thermo-hydrodynamic stability (by temperature gradients)
2. horizontal fluxes (by air entrainment)
3. ceiling plumes (by Rayleigh-Bénard convective motions)

The thermo-hydrodynamic stability is diagnosed characterised by the Richardson number, Ri,

 $$Ri_{e,H} = {{(\rho _\infty - \rho _0 )gH} \over {\rho _\infty u_e ^2 }}$$ (1)

and the Reynolds number

 $${\mathop{\rm Re}\nolimits} _{e,H} = {{\rho _\infty u_e H} \over {\mu _0 }}$$ (2)

where being H the enclosure height, ue the entrainment velocity and μ0 the dynamic viscosity at the break point. When Richardson number is higher greater than unity and greaterhigher than the inverse of the Reynolds number, buoyancy forces dominate inertia forces and the density gradients at the horizontal direction are negligible. From this condition, (PetersonPF:1994) established criteria for stratification in a confined environment (bounded at the upper part, but open in transversal directions to avoid accumulation).

In the case of a round jet, stratification happens occurs when the following criterion is satisfied:

 $$\left( {{H \over {d_0 }}} \right)Fr^{ - 1/6} \left( {1 + {{d_0 } \over {0.2\sqrt 2 H}}} \right)^{2/3} > > 1$$ (3)

as demonstrated by Lee et al. (LeeJH:1974) and Jain and Balasubramanian (BalasubramanianV:1979) (Fig. 1).

Fig. 1: Stratification criteria for round jets (PetersonPF:1994)

In case of a round buoyant plume, the criterion is the following

 $${{4.17} \over {0.35}}\left( {{H \over {d_0 }}} \right)^{5/3} Fr_{} ^{ - 1/6} > > 1$$ (4)

These criteria are not conditions sufficient to quantify the buoyancy gradients. It is necessary to analyse other effects in order to establish mixing or stratified conditions in enclosures: horizontal fluxes and ceiling plumes. The Rayleigh number, Ra, which is defined as

 $$Ra_H = {{g(\rho _\infty - \rho _w )H^3 } \over {\rho _\infty \nu ^2 }}\Pr$$ (5)

is used in this case to determinate these conditions. When this dimensionlessthe Rayleigh number is greaterhigher than 109, the turbulent fluxes generate density gradients which reduce the stability of the stratified layers, and poses initiates mixing patterns (WoodcockJ:2001).

Finally, the location of the release point could reduce established inhomogeneous conditions in a closed room, by the fact that the inertia forces are not enough to impulse the hydrogen below the release point. Then hydrogen will accumulate at the upper part of the room and the mixing process will be led by slower phenomena: molecular diffusion or convective heat. Some experiments as NUPEC M-8-1 (CSNI:1999) and Shebeko (ShebekoYN:1988) have illustrated this phenomenon.

## References

Invalid BibTex Entry!

### Influence of Natural Ventilation of Structures

In general, the accidental release of hydrogen in confined environments will be affected by ventilation streams coming from other rooms or from the atmosphere. As a general safety rule: “Any structure containing hydrogen system components shall be adequately ventilated when hydrogen is in the system” (NASA:NSS:1740:16:1997). The amount of ventilation required will vary in each case depending on the total supply of hydrogen, the rate of generation, and the venting arrangement from the process or hydrogen system. The goal of any hydrogen ventilation system is to keep the concentration below the lower flammability limit (4% at normal conditions).

Ventilation systems use vents, ducts, heat exchangers, fans and other components. They are based on the principles of natural or forced convection as described above in this chapter. Under most common conditions, hydrogen has a density lighter density than air and tends to rise upwards when in contact with air. On the other hand, the air temperature affects the ventilation behaviour: warmer air is less dense than cooler air, and therefore warm air tends to push upwards when in contact with cooler air.

The use of a fan, as a forced ventilation system, has to supply approximately 25 times the volume of hydrogen to maintain a safe concentration of hydrogen. The reliability of this system depends on the eventual event of a mechanical or electrical failure.

In order to avoid these reliability problems, passive systems are usually established on hydrogen applications in enclosures. A typical configuration is based on using high and low vents on walls. Under most common conditions, hydrogen has a lighter density than air and tends to rise upward when in contact with air. Moreover, warmer air is less dense than cooler air, and so warm air tends to push upwards when in contact with cooler air. In general, a ventilation system is driven more by air temperature differences than by hydrogen concentration, and can be affected by the difference of temperature between the enclosure and the external environment.

On a cool day, when the inside temperature is hotter than outside, the lighter warm air mixed with the lighter hydrogen in the enclosure will rise together out the high vent, drawing fresh cool air in through the low vent. Both the temperature of the warm air and the presence of hydrogen will drive the ventilation rate. Under these conditions, the hydrogen amount in the enclosure will decrease. On a warm day, the direction of air flowing through the vents can reverse. When this happens, warm air trying to enter the top vent pushes back the hydrogen trying to rise out the same vent, causing the hydrogen to stagnate and build up inside the enclosure. On this condition the hydrogen is trapped within the enclosure and the molecular diffusion is the only mechanism to mix the hydrogen. Therefore, the explosive conditions could be not avoided.

Other systems use tubes as a small chimney that catches the hydrogen at some elevation. The use of these tubes in combination with low or high vents, being set at the same height on the wall, prevents some of the unwanted thermal convection described above. However, the combination of these systems has to be studied in detail, considering the effect of the external conditions in order to avoid failures in the ventilation behaviour under certain circumstances.

## References

Invalid BibTex Entry!

### Experiment

Tests most representative of hydrogen behaviour within enclosures are the following:

• The Russian tests by Shebeko (ShebekoYN:1988): hydrogen distribution experiments for a subsonic release of hydrogen in a closed vessel.
• GEXCON, NH and STATOIL (GexCon:2003) performed hydrogen dispersion experiments in a confined compartmented space.
• CEA has performed slow release helium dispersion tests in their MISTRA facility, Caron-Charles and Blumenfeld (CaronCharlesM:2001), also Blumenfeld and Caron-Charles (BlumenfeldLCEA).
• Tests in the gallery facility of INERIS (LacomeJM:2007)

Aiming at the development and validation of predictive numerical tools for simulating the hydrogen and mixing processes, a valuable knowledge and experimental database has been compiledachieved in the field of nuclear fission safety through an extensive program of tests for hydrogen distribution and mixing within confined geometries, with the aim to develop and validate numerical tools for modelling hydrogen relases and mixing processes. Geometrical conditions are typical of nuclear reactor containments (large and multi-compartment), and test conditions are very nuclear-specific (high contents of steam, no venting). As an example they are worth mentioning the tests performed in the HDR and BMC facilities in Germany or the NUPEC one in Japan (CSNI:1999), as well as those planned or performed at the ThAI (KanzleiterT:2005) and PANDA (AubanO:2007) facilities.

## References

Invalid BibTex Entry!

### Numerical Simulations

The object of dispersion modeling of hydrogen releases is the calculation of the concentration distribution of hydrogen in their vicinity. From this distribution, envelopes of constant concentrations encompassing higher concentration levels can be determined, from which clearance distances to limit the consequences resulting form accidental ignition. The shape of these envelopes can be complex, and will depend on the emission problem, which determines the nature of the flow and its rate of release, the obstacle configurations and the environmental conditions. Computer fluid dynamics simulations are thus often used to perform such calculations, as they can take into account in principle any level of details.

### Calculation Models for the Simulation of Atmospheric Dispersion of Gas Clouds

There are several classes of calculation models to simulate the atmospheric dispersion of gas clouds:

1. Gaussian model
2. Jet model
3. Box or slab model
4. Particle simulation model
5. k-ε model representing CFD models

The Gaussian plume model is the classical approach for the simulation of the spreading of neutral (sufficiently diluted) gases incl. pollutants or radioactivity. It is a simple model describing the concentration profile as a solution of the diffusion/advection differential equation with empirical coefficients depending on the atmospheric conditions. This model, however, is inappropriate for treating the buoyant behavior of light or heavy gases.

Dispersion models are often accompanied by a jet release model to calculate the dispersion of a released gas with significant momentum flux, which is the dominant parameter for jets. The jet can be classified into two main zones, a region of adjustment from storage conditions to atmospheric pressure and a region of ¡§conventional¡¨ jet dispersion at ambient conditions. If storage conditions are pressurized, the initial zone of adjustment will possibly include flashing for a liquid or choked two-phase jet. The conventional dispersion region begins with a so-called region of flow establishment, in which similarity profiles for the concentration and axial velocity evolve; following this the jet evolves with self-similar profiles. The main features distinguishing the various jet models are the treatment of the air entrainment and the choice of the similarity profile (e.g., top-hat, Gaussian).

The macro or two-zones mixture model developed by BASF (GiesbrechtH:1980) regards the bursting of a pressure vessel, where the high exit velocity results in a fully turbulent propagation of the vessel contents. Two zones are distinguished: a core zone of the vapor cloud where (cold) liquid droplets are still existing, and a boundary zone. In the core zone, ideal mixture with spatially constant and temporally decreasing concentration is assumed, while in the boundary zone, a spatially constant and temporally decreasing turbulent diffusion coefficient is assumed.

In box or slab models, the released gas cloud is assumed to be of cylindrical shape. The processes of advection (transport by the mean wind field), air entrainment, and gravitational spreading are implemented in empirical correlations which were derived from experiments. Box models were basically developed to simulate heavier-than-air vapor clouds with averaged temperature and concentration. In extended versions, vertical profiles of temperature and concentration can be assumed. Acknowledged box models are the US code DEGADIS (HavensJA:1990) or the British code HEGADAS (McFarlaneK:1990).

Particle simulation models are based on the stochastic nature of the movement of particles in the atmospheric wind field. In a simulation, numerous (typically 5000-15,000) particles are being emitted and their trajectories traced making a statistical analysis of the velocity fluctuations. The turbulent velocity is considered to undergo changes only after a certain time defined as the Lagrange correlation time. The distribution of the particles in a given calculation grid is then a measure for the concentration distribution. An improvement of the model is given by assuming a so-called Markov process for a particle meaning that the fluctuation part is further subdivided into a component representing the capability of remembering, and a random component. The velocity at time t is then composed of a fraction proportional to the ¡§old¡¨ velocity at time t-dt and a remainder produced in a random number generator. One representative particle simulation model is the German code LASAT (MartensR:1991).

State-of-the-art modeling of the transient behavior of gases with either positive or negative buoyancy in the atmosphere is provided by Computer Fluid Dynamics (CFD) models, which simulate complex flow processes by solving the Navier-Stokes equations in an Eulerian 3D (or 2D) calculation grid structure. This approach comprises the conservation equations of mass, momentum, and energy. Apart from being (in most cases) immensely time-consuming, these models require a detailed input of initial and boundary conditions.

In the two-equation k-ε turbulence model, special partial differential equations are solved to describe the transport of turbulence as well as its generation and dissipation. Of all the approaches, the k-ε model offers the highest relative independence of empirical relations. It appears to be the only one to allow a proper simulation of hydrogen dispersion, because it meets the requirements of describing effects such as turbulence energy in the gas cloud, interaction with the atmospheric wind field, the characteristic positive buoyancy, transient sources with initial momentum, and last but not least, gas flow in a complex geometry (buildings, terrain). K-ε modeling has been realized in a variety of computer codes distinguished by the choice of the numerical solution method, which was found to have a significant effect on the calculation procedure.

## References

Invalid BibTex Entry!

## Simulation of dispersion in the Atmosphere

Only a few efforts have been made in the past to simulate the dispersion of hydrogen gas mostly due to the poor experimental data base available. Early efforts were made in the late 1970s by the Los Alamos National Laboratories on a box model for hydrogen taking into account heat transport from the ground into the cloud Invalid BibTex Entry! , and then applying the Gaussian model of neutral and buoyant dispersion as part of the WHAZAN software package (StewartWF:1990). The NASA has developed the code AFGASDM applicable to LH2 and other aviation fuels. The model is something between a Gaussian model and an Eulerian grid model solving the conservation equations following a gas ¡§parcel¡¨ released as a puff until it has diluted below the flammability limits. Effects of air entrainment and phase changes are also taken into account (BrewerGD:1981).

## Simulation of dispersion in Semi-Confined Environments

A subsonic horizontal hydrogen jet experiment and subsequent CFD simulation was reported by Swain Invalid BibTex Entry! (Fig. 1 and Table 1). Hydrogen was released at 25 scf/min through a 1 cm diameter orifice. The predicted concentrations showed good agreement with measurements.

Fig. 1: Concentration contour of a 25 scf/min release. The numbers show locations where the concentration of hydrogen was measured experimentally

Table 1: Comparison between experimental and simulated H2 conc.

 Sensor position Experimental H2 conc. (%) Simulation H2 conc. (%) Deviation (%) 1 5.0-5.9 5.04 -8.13 2 5.6-7.0 6.96 10.48 3 9.4-10.8 13.99 38.50 4 8.1-9.4 8.25 -5.70 5 5.6-6.6 5.29 -13 6 3.5-4.6 5.37 32.60

## Simulation of dispersion in Confined Environments

The use of hydrogen applications (especially automotive) in confined environments like private garages, public parking, maintenance shops, electrolysers, compressor buildings, tunnels, etc. requires detailed evaluation of the risk related to potential hydrogen leaks and if necessary identification of measures to be taken in order to avoid buildup of flammable/explosive hydrogen atmospheres. In this respect the CFD methodology has been widely used as shown in the review below, since it is generally able to realistically account for the various geometrical configurations and complex release conditions.

## References

Invalid BibTex Entry!

### Open Environment

The k-å atmospheric dispersion model POLLUT was used in a DLR study (EichertH:1992) to investigate hydrogen dispersion from accidental release of LH2 from vehicle tanks both in open terrain.

The computer code CHAMPAGNE was successfully applied to the NASA LH2 spill tests from 1980 (ChitoseK:1994).

## References

Invalid BibTex Entry!

### Urban Environment

The BAM 5 and 6 LH2 dispersion experiments in the presence of buildings Invalid BibTex Entry! were simulated by (StatharasJC:2000) using the ADREA-HF code. The source was modelled using a series of two phase flow jets. Liquid phase evaporation during the dispersion was taken nto account using the homegeneous equilibrium model. The calculations were performed using a one-equation turbulence model. Reasonable agreement with measured concentrations was reported when ground heating was taken into account.

The Battelle k-å model BASSIM originally designed for hydrogen combustion in nuclear containments has been applied to predicting the BAM LH2 release trials in 1994, providing reasonable qualitative results for 3D effects of hydrogen dispersion behavior (RastogiAK:1994).

The 1983 Stockholm (Sweden) hydrogen accident was simulated by Venetsanos et al. (VenetsanosAG:2003) using the ADREA-HF code for dispersion and the REACFLOW code for the combustion. An intergal tool was applied to simulate the release of 4kg of hydrogen from a network of 18 interconnected cylinders of 15 l volume each containing hydrogen at 200 bar. Figure 1 shows the predicted lower flammability (4%) hydrogen-air cloud at 10 seconds after start of release.

Fig. 1: Simulation of the 1983 Stockholm hydrogen accident (VenetsanosAG:2003), Left: Modelled site and lorry carrying 4 kg of hydrogen in 18x50 l, 200 bar bottles (red circle), Right: Predicted lower flammability hydrogen-air cloud at 10 seconds after start of release

Simulations of hydrogen releases from CGH2 commercial vehicles (busses) in urban environment were reported by (VenetsanosAG:2008). The work was performed during the EIHP2 project (EIHP:url), using the ADREA-HF code for dispersion and the REACFLOW code for combustion. Three different storage pressures were considered for the CGH2 releases 200, 350 and 700 bar. Compartive simulations were performed for a 200bar CNG bus.

## References

Invalid BibTex Entry!

### Tunnels

The k-å atmospheric dispersion model POLLUT was used in a DLR study (EichertH:1992) to investigate hydrogen dispersion from accidental release of LH2 from vehicle tanks in a road tunnel.

Simulations of hydrogen releases from LH2 and CGH2 private vehicles (cars) in a naturally ventilated tunnel were reported by Wilkening et al. (WilkeningH:2000). The work was performed within the framework of the EIHP project (EIHP:url). The ADREA-HF code was used for the dispersion calculations. The REACFLOW code was used for the combustion calculations. Two LH2 release scenarios were considered. A flow restrictor installed and a two-phase release of 8.3 g/s was assumed in the first scenario. A shut-off valve activated 5 seconds after the release and a gaseous release of 60g/s was considered in the second. For the CGH2 scenario sonic release from a 200bar storage tank was assumed. The reported overpressure results indicated that for the scenarios considered there is no major difference in using liquid hydrogen or compressed hydrogen fuel.

Tunnel accidents with an LH2 powered vehicle were simulated by Breitung et al. (BreitungW:2000). The investigated scenarios assume damage of the LH2 system, release of gaseous hydrogen, mixing with air, ignition and finally combustion. Calculations showed that gaseous hydrogen rises to the tunnel ceiling forming a strongly stratified mixture. Shape, size, inner structure and temperature of the evolving H2-air cloud were calculated. Using new developed criteria, the time and space regions with potential for fast combustion modes were identified. For the given hydrogen sources the combustion regime is governed by the ignition time. For late ignition a slow and incomplete combustion of the partly premixed H2-air cloud along the tunnel ceiling was predicted. For early ignition a standing diffusion flame develops with dimensions and heat fluxes determined by the hydrogen release rate. Temperature, oxygen and flow velocity fields during the combustion were computed. In both cases only minor pressures were generated. The highest damage potential appeared to exist for intermediate ignition times.

Simulations of hydrogen releases from CGH2 commercial vehicles (busses) in tunnel environment were reported by (VenetsanosAG:2008). The work was performed during the EIHP2 project (EIHP:url), using the ADREA-HF code for dispersion and the REACFLOW code for combustion. Three different storage pressures were considered for the CGH2 releases 200, 350 and 700 bar. Compartive simulations were performed for a 200bar CNG bus.

CFD simulations of hydrogen dispersion in tunnels was performed by (MukaiS:2005), using the STAR-CD code and standard k-ε model. The amount of hydrogen leaked was 60 m3 (approximately 5.08 kg), which corresponds to the amount necessary for future fuel cell vehicles to achieve their desired running distance. The study considered the typical longitudinal and lateral areas of tunnels, the undergound ventilation facilities and the electrostatic dust collectors.

## References

Invalid BibTex Entry!

### Residential Garages

Hydrogen leaks inside a residential garage compared against gasoline, natural gas and propane leaks in the same environment were simulated by Swain (SwainMR:1998) using the FLUENT code. Calculations were based on the GEOMET Invalid BibTex Entry! garage geometry (2.52x6.59x2.74 m) with a single vent placed in the center of one wall of an otherwise sealed garage. The leak rates for fuels other than hydrogen were adjusted for equal size holes and equal energy flow rates in the fuel lines considering both laminar and turbulent flow where applicable. The results of the simulations show that for the lower leakage rate (1000 l/hr) and typical garage air exchange (2.92 ACH representing natural ventilation), hydrogen and methane did not create dangerous conditions while propane and gasoline did produce dangerous conditions in similar accident scenario. For the larger fuel leakage rate (7200 l/hr) and minimal air exchange rate (0.2 ACH) all four fuels produced very large combustible clouds after 2 hours of leakage. However, the energy content of the combustible clouds was different, with hydrogen being at most % that of the other fuels. Both natural gas and hydrogen filled the entire garage with a flammable mixture after two hours, while propane and gasoline filled just over half of the garage volume with a flammable mixture. At the higher air exchange rate (2.92 ACH), the hydrogen still filled the garage with a flammable mixture, which reached about 6.6% hydrogen concentration after two hours.

Boil-off from the cryogenic hydrogen tank of a car in a private garage was simulated by Breitung et al. (BreitungW:2000) using the GASFLOW code to calculate the temporal and spatial distribution of hydrogen and criteria to evaluate the flame acceleration and detonation potential. Boil-off was assumed occurring at a rate of 170 g day-1 and the boil-off release to occur intermittently in five pulses per day of 100 or 10 s time period each, which gave 0.34 or 3.4 g s-1 respectively.

CFD modelling (FLUENT) was used in Invalid BibTex Entry! to analyze H2 leak scenarios inside residential two vehicle garages. The study was based on parking a 5-passenger sedan with compressed H2 gas reservoir carrying capacity of 6 kg at 10,000 psi (689.5 bar) pressure. The HFCV was designed to comply with SAE J2578 and J2579 standards for H2 and fuel cells, which include provisions for safety systems onboard the vehicle. Such assumed mechanisms include the implementation of a hydrogen detector in each wheel well. Each detector was designed to signal a shut down and isolation H2 procedure upon detecting 1% H2. Another assumed mechanism includes the use of an on-board computer that is capable of shutting down H2 flow upon detecting a larger than 20 CFM leak (9.4 lt s-1) when the vehicle is dormant. In addition the HFCV was equipped with a valve that isolates H2 in the tank upon engine (fuel cell) shut down. This assumed isolation mechanism was designed to monitor and test for leaks upon vehicle shutdown and prior to start up by the on-board computer. In the study most of the modelling scenarios were based on a 20 CFM leak from beneath the vehicle. This leak rate corresponded to a fuel cell power output of about 50 kW. It was found that for the considered garage layout and scenarios no modifications to the baseline structures would appear to be necessary for vehicles equipped with safety systems that detect hydrogen leaks according to the chosen scenarios. CFD modelling of one configuration showed that a 1% hydrogen concentration would reach the wheel well within 28 seconds where detectors could initiate a shut off of the fuel supply. Since not all vehicles will be equipped with hydrogen detectors, or be configured like the chosen vehicle design, additional modelling provided data on the time for a 5% hydrogen concentration to accumulate at the garage ceiling. This information may be used by carmakers to develop strategies to limit the amount of time that the vehicle operates at zero speed before shutting off the fuel supply.

Helium dispersion experiments in a private garage including a mockup of a car and performed by Swain Invalid BibTex Entry! were simulated by Papanikolaou and Venetsanos (PapanikolaouEA:2005) using the ADREA-HF code and the standard k-å model. Helium was released below the car at a flow rate of 7200 l/hr. The predicted results were found generally in acceptable agreement with the experiment. For the case with the lowest vent size the vertical concentration gradient was found underestimated compared to the experiment. This was attributed to the turbulence model overestimating mixing under the given low flow conditions.

## References

Invalid BibTex Entry!

### Maintenance Shops

CFD modelling (FLUENT) was used in Invalid BibTex Entry! to analyze H2 leak scenarios inside a commercial maintenance/repair/service station. The study was based on a 5-passenger sedan with compressed H2 gas reservoir carrying capacity of 6 kg at 10,000 psi (689.5 bar) pressure. The HFCV was designed to comply with SAE J2578 and J2579 standards for H2 and fuel cells, which include provisions for safety systems onboard the vehicle. Such assumed mechanisms include the implementation of a hydrogen detector in each wheel well. Each detector was designed to signal a shut down and isolation H2 procedure upon detecting 1% H2. Another assumed mechanism includes the use of an on-board computer that is capable of shutting down H2 flow upon detecting a larger than 20 CFM leak (9.4 lt s-1) when the vehicle is dormant. In addition the HFCV was equipped with a valve that isolates H2 in the tank upon engine (fuel cell) shut down. This assumed isolation mechanism was designed to monitor and test for leaks upon vehicle shut down and prior to start up by the on-board computer. In the study most of the modelling scenarios were based on a 20 CFM leak from beneath the vehicle. This leak rate corresponded to a fuel cell power output of about 50 kW. For the considered maintenance facility and scenarios no modifications to the baseline structures were recommended. The high rates of ventilation would dilute the assumed 20-CFM leak and result in a flammable mixture only in close proximity to the vehicle. The potential for flammable mixtures forming at the ceiling of the facility was assessed. The time required for a hydrogen leak to result in 20% of LFL at the ceiling was determined for different vehicle leak rates. Options for improving the ventilation in the building were presented.

## References

Invalid BibTex Entry!

### Public Parking

CFD modelling (FLUENT) was used in Invalid BibTex Entry! to analyze H2 leak scenarios inside a commercial multi-story above-ground parking and a commercial multi-story below-ground parking. The study was based on parking a 5-passenger sedan with compressed H2 gas reservoir carrying capacity of 6 kg at 10,000 psi (689.5 bar) pressure. The HFCV was designed to comply with SAE J2578 and J2579 standards for H2 and fuel cells, which include provisions for safety systems onboard the vehicle. Such assumed mechanisms include the implementation of a hydrogen detector in each wheel well. Each detector was designed to signal a shut down and isolation H2 procedure upon detecting 1% H2. Another assumed mechanism includes the use of an on-board computer that is capable of shutting down H2 flow upon detecting a larger than 20 CFM leak (9.4 lt s-1) when the vehicle is dormant. In addition the HFCV was equipped with a valve that isolates H2 in the tank upon engine (fuel cell) shut down. This assumed isolation mechanism was designed to monitor and test for leaks upon vehicle shut down and prior to start up by the on-board computer. In the study most of the modelling scenarios were based on a 20 CFM leak from beneath the vehicle. This leak rate corresponded to a fuel cell power output of about 50 kW. For the considered below and above ground parking facilities and scenarios no modifications to the baseline structures were recommended. Existing ventilation in the below ground structure would dilute a 20-CFM hydrogen leak so that a flammable mixture would only exist in close proximity to the vehicle. Similarly, natural ventilation would dilute hydrogen leaks for the above ground parking facility.

## References

Invalid BibTex Entry!

### Hydrogen Refuelling Stations

High pressure hydrogen release experiments inside the storage room of a full scale model of a hydrogen refueling station were reported by Tanaka et al. (TanakaT:2005). Storage pressure was 40 MPa while nozzle diameters in the range 0.8-8 mm. The storage room with dimensions 6x5x4 m included 35 cylinders of 250 l capacity each. Ventilation openings of 1m height existed on all 4 sidewalls and where either 50% or 100% open. The time history of the average hydrogen concentration in the room was modelled usimg a simple gas accumulation model (CleaverRP:1994) and compared against the experimental data. It was found that the model is able to predict well the experimental concentrations in the experiments involving more slowly varying pressure (lower nozzle diameters), but tends to overpredict the concentrations for the higher nozzle diameter.

## References

Invalid BibTex Entry!

### Other Applications

Hydrogen dispersion experiments in a half-scaled hallway and subsequent CFD validation using the FLUENT code were performed by Swain (SwainMR:1999). Hydrogen leaked at a rate of 2 SCFM (0.94 lt/s) from the floor at the left end. At the right end of the hallway, there were a roof vent and a lower door vent for the gas ventilation. Four sensors were used to record the local hydrogen concentration variations with time. Predicted hydrogen concentrations time series were found in good agreement with the experimental data. The same experiments were simulated by Agranat et al. {[DB, AgranatV:2004]], using the PHOENICS code. Predicted results were found similar to the ones obtained using the FLUENT code with maximum concentration differences between the two models of about 20%.

Catastrophic hydrogen releases inside the Alpha H2BPS (H2 Backup Power System by Stewart Energy) generator room were simulated by Agranat et al. (AgranatV:2004) under real industrial working scenarios and real geometry, using the PHOENICS code and the LVEL turbulence model. Two scenarios were considered in the simulations: a vertical fast release from a high-pressure line and a horizontal fast release from a medium-pressure line. The CFD simulations showed that the two installed sensors are capable of detecting 10% LFL cloud (0.41%) separately at 8.8 and 9.7 seconds for the high-pressure vertical leak, but only one sensor which is closer to the leak orifice can detect the same concentration cloud within 20 seconds for the medium-pressure horizontal release. The numerical simulation confirmed that the current sensor installation can promptly report the potential catastrophic hydrogen leak under the above scenarios. However, the fact that 10% LFL hydrogen cloud cannot reach one sensor during the horizontal release indicates that the sensor location can be further optimized and more sensors are required for the systems.

The method for determination of maximum ventilation described in the standard IEC EN 60079-10 was validated by Nilsen et al. (NilsenS:2004) for a small hydrogen production (by water electrolysis) unit located inside two different enclosure geometries, using the FLACS code for dispersion and the PHAST code for release. It was found that the model suggested in standard IEC EN 60079-10 is not a conservative approach when deciding the ventilation capacity large enough to keep flammable gas clouds at a negligible size and therefore must be used with care.

## References

Invalid BibTex Entry!

### Knowledge Gaps and Recent Progress

Simulations of hydrogen dispersion using the CFD methodology have increasingly grown in number during the last 10 years and are expected to grow even more in the near future. Prediction of the time and space distribution of the flammable hydrogen clouds evolving after accidental hydrogen leaks of various types in widely different environments is the main output necessary for subsequent risk assessment estimation of the various hydrogen applications. In this process, simulations have been performed using different CFD codes (commercial or research tools) and different modelling strategies (turbulence models,source treatment, discretization options, etc.).

To ensure the quality and trust in industrial CFD applications best practice guidelines have been developed in the past either of a general character like ERCOFTAC 2000 (CaseyM:2000) or more related to particular applications like HSL 2003 Invalid BibTex Entry!. No CFD guidelines specific to hydrogen dispersion applications have been proposed.

Taking the above in consideration a significant effort has been concentrated within the European Network of Excellence HYSAFE with aim to perform a systematic evaluation of the various CFD approaches (codes and models) in predicting hydrogen dispersion, based on a series of benchmark exercise problems, using existing and new state of the art experimental data.

The results of the first such hydrogen dispersion benchmark exercise (SBEP-V1) were reported by Gallego et al. (GallegoE:2005). The experiment simulated was that of Shebeko et al. (ShebekoYN:1988), who investigated the dispersion of hydrogen in an hermetically closed cylinder (20m3 volume) by measuring axial hydrogen concentrations (6 locations) at times from 2 to 250 minutes following an initial 60 s vertical subsonic jet release at a rate of 4.5 l/s from a 10mm nozzle. Large variations in predictions were monitored during this first benchmark (as expected), which could be attributed to variations in turbulence models, boundary conditions as well as discretization options.

The aim of the second hydrogen dispersion benchmark exercise (SBEP-V3) was three fold a) to further investigate on the ability of the models to predict the long term stratification/diffusion problem in a confined space, b) to test the ability of models to predict the concentration field of a vertical subsonic hydrogen jet release and c) to attempt to minimize or justify large variations between model predictions. Newly performed hydrogen dispersion experiments by INERIS at their gallery facility (garage like enclosure with dimensions 7.2x3.8x2.9 m) were used for this benchmark. The release was vertical upwards at a rate of 1 g/s from an orifice of 20 mm diameter and lasted for a period of 240 s. The total simulation time was 5400 s. The benchmark took part in two phases a blind pre-test phase and a post–test one.

Further benchmarks focus on testing the ability to predict free choked hydrogen flows, obstacle effects on hydrogen dispersion within confined spaces as well as hydrogen dispersion from LH2 releases.

Computer fluid dynamics simulations of chocked flows are difficult to tackle due to the presence of the shock waves. The simulations may require, for commercial solvers, resolving the Mach cone and the shock wave patterns to some degree of details. Since the extent of the flammability envelopes resulting from chocked releases from apertures of about 1 cm may reach 10 to 100 m depending on the storage pressure, length scales of up to five orders of magnitude must be covered by the mesh. In addition, convergence will usually be problematic.

The difficulties faced by direct CFD simulations of chocked releases may be alleviated by using effective diameter approaches/ The applicability of effective diameter approaches to horizontal releases of hydrogen should be investigated further, particularly for the large hydrogen releases resulting from high pressure flows, where the effects of buoyancy on the shape of the release remain an issue.

For choked hydrogen releases the fact that the molar concentration is proportional to the inverse distance has been observed experimentally, but given that significantly different proportionality constants have been reported, a systematic investigation both experimental and computational is still required to cover a wider range of storage pressures and orifice diameters.

Regarding obstacle effects on hydrogen dispersion it should be mentioned that steady-state flow rates can lead to unsteady behaviour of the dispersion pattern in some cases, particularly when impingement flows or external flows (over a surface) are considered, due to significant vortex shedding. Such situations may require a statistical definition of the constant (flammable) concentration envelope, based on the probability distribution of finding a given concentration of hydrogen at a specific location at a given time.

Finally as far as LH2 release and dispersion are concerned it seems that more experimental information is needed to trigger further physical understanding and model development/improvement. From the past experience it seems that these proposed tests should focus on better control over the experimental conditiosn (less uncertainty).

## References

Invalid BibTex Entry!