Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

international journal of hydrogen energy xxx (xxxx) xxx

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/he

Large Eddy Simulation of low Reynolds number


turbulent hydrogen jets - Modelling considerations
and comparison with detailed experiments

I.C. Tolias a,*, A.A. Kanaev b, N. Koutsourakis a, V.Y. Glotov b,


A.G. Venetsanos a
a
Environmental Research Laboratory, INRASTES, National Centre for Scientific Research Demokritos, Patriarchou
Grigoriou E & 27 Neapoleos Str., 15341, Agia Paraskevi, Greece
b
Nuclear Safety Institute (IBRAE), Russian Academy of Sciences, Bol’shaya Tul’skaya Ul. 52, Moscow, 115191,
Russia

highlights

 LES of hydrogen jets at 885, 1360 and 2384 Reynolds number.


 Mean values and RMS fluctuations are compared against experiments.
 Smagorinsky model, RNG-LES and Implicit LES are evaluated.
 The effect of grid density, Smagorinsky constant and release modelling is discussed.
 Overall, good agreement with the experiments is achieved.

article info abstract

Article history: The main objective of this work is to apply Large Eddy Simulation (LES) on hydrogen
Received 14 July 2020 subsonic round jets in order to evaluate modelling strategies and to provide guidelines for
Received in revised form similar applications. The ADREA-HF code and the experiments conducted by Sandia Na-
28 September 2020 tional Laboratories are used for that purpose. These experiments are very suitable for LES
Accepted 1 October 2020 studies because turbulent fluctuations have been measured which is something rare in
Available online xxx hydrogen experiments. Hydrogen is released vertically from a small orifice of 1.91 mm
diameter into stagnant environment. Three experimental cases are simulated with
Keywords: different Reynolds number at the release area, namely 885, 1360 and 2384. Hydrogen mass
LES fraction and velocity mean values and fluctuations are compared against the experimental
CFD data. Several grid resolutions are used to assess the effect on the results, using mainly the
Simulation Smagorinsky subgrid scale model. In the lowest Reynolds number case, an Implicit LES
Modelling code, used independently from a different scientific group, is also tested. In this case, the
Hydrogen jet performance of the RNG-LES subgrid scale model of the ADREA-HF code is also examined.
ADREA-HF Additionally, the effect of Smagorinsky constant and of the Van Driest correction is eval-
uated. The amount of the resolved turbulence and of the velocity spectra are presented.
Finally, the effect of the release modelling is discussed. The analysis shows that even a
coarse discretization of the release area can give acceptable results for hydrogen safety
engineering applications. However, dense grids are required for the more accurate

* Corresponding author.
E-mail addresses: tolias@ipta.demokritos.gr (I.C. Tolias), kanaev@ibrae.ac.ru (A.A. Kanaev).
https://doi.org/10.1016/j.ijhydene.2020.10.008
0360-3199/© 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
2 international journal of hydrogen energy xxx (xxxx) xxx

prediction of the turbulent characteristics. The two LES codes gave similar results and the
overall agreement with the experiment was satisfactory.
© 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

performance of LES in which part of turbulence is resolved. The


Introduction increased number of LES studies (e.g. Refs. [14e23]) renders the
need of proper modelling strategies and model validation. A
Hydrogen use is expected to increase in the near future and its thorough evaluation of LES for hydrogen safety purposes which
nature brings up some new safety issues. In the case of an compares also velocity fluctuations was performed by Bernard-
accident, hydrogen mixes with air and a flammable cloud can Michel et al. (2018) [24,25]. They evaluated Smagorinsky LES
be created. An accidental release in a confined space can have against helium release experiment in a small scale enclosure
severe consequences in the case of an explosion. However, the with two vents. In their simulations, LES over-predicted the
strong buoyant nature of hydrogen assists to its dispersion, concentrations and under-predicted the turbulent fluctuations
especially in open spaces. In the past years, the increase of concluding that more efforts should be made in order to explain
computational power has rendered Computational Fluid Dy- the observed discrepancies.
namics (CFD) as a very attractive methodology for risk The hydrogen release and dispersion experiments conducted
assessment. With its high numerical accuracy, it can evaluate by Sandia National Laboratories [26,27] are used in this study.
regulations and standards and give deeper insight into the Three inlet velocity cases were examined in the experiments that
physical phenomenon. correspond to Reynolds numbers equal to 885, 1360 and 2384.
CFD modelling of hydrogen dispersion is a difficult task due These experiments are very suitable for LES studies because tur-
to the turbulence that is developed. Most fluid flows that occur bulent fluctuations have been measured which is something rare
in nature and in engineering applications are turbulent. in hydrogen experiments. These experiments have only partially
However, turbulence is not fully understood yet despite the been used in the past for CFD model validation comparing only
major research effort in the study of turbulent phenomena. To the mean values against experimental data. The case with the
simulate turbulent flows, several approaches have been highest exit velocity was simulated by Hourri et al. (2011) [28] using
developed. Reynolds-averaged Navier-Stokes (RANS) and the realizable k-ε model with buoyancy effects. Good agreement
Large Eddy Simulation (LES) are the two main approaches that with the experiment (average absolute deviation within 5.2%) of
are used today. RANS approach is the most widely used. This hydrogen concentration decay along jet centerline was found.
approach includes several well-known turbulence models The same case was simulated by Li et al. (2017) [29] and Zhang et al.
such as k-ε [1] and k-u [2], which have been proved to work (2018) [22] evaluating LES, Detached Eddy Simulation (DES) and
successfully in many problems. However, there is no univer- RANS with k-ε model. A single grid with 1,215,000 cells was used.
sal RANS turbulence model capable of predicting all types of Good agreement with the experiment was found for the time-
flow due the complexity of turbulence. On the other hand, LES averaged values used for the comparison.
[3] is considered as a more accurate approach because most CFD simulation of non-high Reynolds number cases is a
part of the turbulence is resolved. Therefore, it can provide challenge because most models have been calibrated for
reliable results in a wider range of flows and applications. Its fully-turbulent regimes. Thus, in this work, even though all
drawback is the high computational resources that are three experimental cases are examined, special focus is given
required. However, the growth in computing power over the to the lowest release velocity case (49.7 m/s) which corre-
last decades renders LES a possible choice for real scale en- sponds to Reynolds number equal to 885. The Smagorinsky
gineering applications. LES of ADREA-HF code [7,30] is evaluated by comparing mean
Many studies have been conducted with the aim to evaluate values and RMS (root-mean-square) fluctuations. Three
CFD methods and codes for hydrogen safety application (e.g. different grids are used in order to assess grid independence.
Refs. [4e11]). In the experiments used to evaluate the models in The amount of the resolved turbulence and of the velocity
these works, however, only the mean concentration values were spectra are presented. RNG-LES, which is known to adjust to
measured. Even though this is the most critical variable for regions of different Reynolds numbers [7,31], is also examined
assessing hydrogen safety, turbulent fluctuations are also very in order to evaluate the effect of the subgrid-scale model. The
important. The study of fluctuations and extreme values are effect of the Van Driest correction near solid boundaries is
significant because, for example, the mean concentration value also studied. Moreover, a different LES code is used indepen-
can be below the flammability limit but the instantaneous one dently from a different scientific group. This code utilizes the
may exceed the flammability limit. At a previous numerical Implicit LES (ILES) methodology using the CABARET (Compact
study [7], instantaneous concentration at some sensors could Accurately Boundary-Adjusting high-Resolution Technique)
get 50% higher than the mean value. High fluctuations are also method [32]. In ILES no subgrid scale model is used and the
noticed in the current work. The study of extreme values is an unresolved turbulence is handled through the numerical
active field of research [12,13]. Turbulent fluctuations are also scheme [3]. Finally, the effect of the release modelling is
important in order to gain a deeper insight into the performance examined and discussed.
of CFD simulations. This is more vital when assessing the

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 3

 
vr v
Description of experiments þ ~j ¼ 0
ru (1)
vt vxj

The experiments were conducted by Sandia National Labo-     


~i
vru v vp v vu ~j
~ i vu
ratories in order to study small-scale releases of hydrogen at þ ru ~i ¼ 
~j u þ meff þ þ rgi (2)
vt vxj vxi vxj vxj vxi
low flow rates where buoyancy effects are important. The
details of the experiments are presented in Refs. [26,27,33].   
vrY~ v ~
Pure hydrogen is released vertically upwards from a þ ru ~ ¼ v Deff vY
~j Y (3)
squared-off-end straight tube with inner diameter equal to vt vxj vxj vxj
d ¼ 1.91 mm and outside diameter equal to 3.34 mm. This where r is the density, ui the velocity components, t the time,
tube is extended approximately 4 mm above a larger tube of xi the distance, p the pressure, gi the gravity acceleration, Y the
63.5 mm diameter. A honeycomb plane is located at the end hydrogen mass fraction, meff the effective dynamic viscosity
of the larger tube which normally provides a co-flowing air and Deff the effective diffusivity. The bar over the symbols
which in our cases is equal to zero. The accuracy of the fuel
represents space-averaged instantaneous values, while the
flow rate was better than 1%. The ambient room tempera-
tilde denotes density-weighted Favre-averaging. The ideal gas
ture was equal to 21 ± 1  C and the pressure equal to
equation of state is also utilized to relate density with pres-
100 ± 0.5 kPa. Three cases were examined, with average exit
sure. In the present work, subgrid-scale stresses are mainly
velocity equal to 49.7, 76.0 and 133.9 m/s. These velocities
modelled using the Smagorinsky model [38] with Van Driest
were estimated based on the volumetric flow rate and the
damping [39]. Thus:
release area (density equal to 0.0838 kg/m3). The release
Reynolds numbers are equal to 885, 1360 and 2384 which   2 qffiffiffiffiffiffiffiffiffiffiffiffiffi
2S~ij S~ij
þ
msgs ¼ r Cs V1=3 1  ey =25 (4)
corresponds to transitional and fully turbulent flow condi-
tion at the exit. The velocity field was measured using Par-
ticle Image Velocimetry (PIV) with a spatial resolution of meff ¼ m þ msgs (5)
1.4 mm. The concentration field was measured using Laser-
based Rayleigh scattering. The uncertainty in the mean and Deff ¼ rD þ msgs Scsgs (6)
RMS mass fraction values was estimated equal to ±3% and
where Cs is the Smagorinsky constant (equal to 0.1 in our
±6% respectively.
study), V the computational cell volume, yþ the non-
Due to the low Reynolds number, a laminar flow would  
dimensional distance from the wall, S~ij ¼ 12 v~ui v~
u
be expected inside the pipe. However, experimentalists vxj
þvxij the rate
states [27] that a fully developed turbulent velocity profile
of strain tensor, msgs the subgrid-scale viscocity, m the mixture
was assured at the jet exit in the 2384 Reynolds number
viscosity, D the mass diffusivity of hydrogen to air and Scsgs the
case. In addition, the 885 and 1360 cases are characterized
turbulent subgrid-scale Schmidt number (equal to 0.72 in our
as transitional [26]. Besides that, the downstream jet flow is þ
case). The Van Driest damping factor, 1  ey =25 , accounts for
usually turbulent [34,35] which is confirmed by the Sandia
the reduced growth of the small scales near the wall. Since the
experimental results for all three examined cases. In a
case examined is not a classical flow around a solid obstacle,
smaller Reynolds number (equal to 520) release case con-
the Van Driest correction might not be necessary to apply and
ducted in the same series of experiments, the flow outside
thus its effect is examined in the 49.7 m/s case. The RNG
the release area remained laminar for a notable distance.
model [40] is also tested. More details about the imple-
Even in this case, transition to turbulence occurred after the
mentation in the ADREA-HF and validations can be found in
distance of 60 mm from the pipe exit. Buoyancy affects the
Refs. [7,8,41].
flow field to a greater degree in such low Reynolds number
ADREA-HF uses the finite volume method on a staggered
case. For example in the experimental work of O’Neill et al.
Cartesian grid. The pressure and velocity equations are
[36], a non-buoyant jet with Reynold number equal to 1030
decoupled using a modification of the SIMPLER algorithm [42].
transited to turbulence whereas a jet with Reynolds number
For the discretization of the convective terms at current sim-
of 680 did not. It is also interesting to note that acoustic
ulations the second order central differences scheme is used
perturbations can affect low-Reynolds number jets. For
in the momentum equations and the MUSCL in the mass
example in Ref. [37] acoustic disturbances by clapping of
conservation equation of hydrogen. For time advancement,
hands in a few meters distance caused (temporarily) the
the second order accurate Crank-Nicolson scheme was
break-down of a non-buoyant jet (of Reynold number equal
chosen.
to 700) leading to substantial reduction of its laminar
The second code that is utilized in this study, is being
portion.
developed at Nuclear Safety Institute (IBRAE RAS) of Russia
and it is based on the CABARET method [32,43,44]. The method
is second-order accurate on non-uniform grids in space and
Numerical modelling time, has a very small dispersion error and computational
stencil defined within one spaceetime cell. For shock-
The filtered continuity equation, Navier-Stokes equations and
capturing, the scheme uses a conservative non-linear
conservation equation of hydrogen that are solved in the
correction procedure which is directly based on the
ADREA-HF code are:

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
4 international journal of hydrogen energy xxx (xxxx) xxx

maximum principle and implicitly provides the drainage of No-slip boundary conditions were used in all solid
energy from the resolved scales to the subgrid scales. For boundaries and standard wall functions were applied. In the
turbulence, the Implicit LES, formally known as Monotonically lateral planes of the domain solid boundaries were considered
Integrated Large Eddy Simulation (MILES), approach [45] is in order to increase the stability of the solution. In the top and
used. Similar to other ILES techniques, this approach does not bottom boundaries non-reflecting type of boundary condition
include any explicit subgrid-scale closure. The use of a was utilized [46].
subgrid-scale-free approach is an advantage of the method In ADREA-HF simulation three computational grids were
because it reduces the number of parameters in the modelling used in order to assess grid independence, a coarse one with
of turbulent flows. 2 cells along the diameter of the release area and total number
As initial conditions, a stagnant flow field with no turbu- of cells approximately equal to 1,600,000, a medium one with
lence is specified. Release is modelled by fixing the velocity at 6 cells along the release diameter and total number of cell
the release area to be the same with the mean experimental equal to 4,190,000 and a fine one with 12 cells along the release
one (estimated based on the volumetric flow rate and the jet diameter and total number of cell equal to 12,300,000. The
exit cross-sectional area [26]). No synthetic turbulence was control volumes size was uniform around the release, in a
imposed at the inlet except in one particular ILES case that is region with horizontal length equal to 4 times the diameter of
explicitly mentioned in the results section. the release area and with vertical length equal to 25 times the
Domain size is equal to 0.38  0.38  0.76 m (x, y, z di- diameter. The expansion ratio outside this uniform region
mensions). The domain extents vertically from z ¼ 0.06 to was set equal to 1.1 for the horizontal directions and from 1.05
z ¼ 0.7 m. In the ILES code a similar domain was used with to 1.12 in the vertical. The computational grid for the 12-cells
smaller vertical extension (top boundary at z ¼ 0.3 m). The discretization case is shown in Fig. 1.
release point was positioned at z ¼ 0 m and at the centre of the In the ILES case, a similar strategy in the grid density was
horizontal directions. Both tubes (with 1.91 and 63.5 mm used as in the ADREA-HF case in order grids to be as close as
diameter) were included in the geometry model (Fig. 1). The possible. The particular ILES code utilizes unstructured hex-
rectangular structure that surrounds the tubes (see Fig. 1 in ahedral grids. Only the cases with the 6 and 12 cells along the
Ref. [26]) was included only in the ADREA-HF simulations release diameter were examined. The grid of the 12-cells case
because it was laid inside the domain boundaries. is shown in Fig. 1.

Fig. 1 e Part of the computational domain and of the fine computational grid in the ADREA-HF (top) and ILES (bottom) case.
Some of the sensor positions are shown in the top left figure (spheres). The grid around the release area is shown in the
right figures for the 12-cells discretization case.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 5

In ADREA-HF case, the selected time step corresponds to


maximum CouranteFriedrichseLewy (CFL) number equal to Results and discussion
about 0.4. In CABATER the CFL number was set equal to 0.3.
However, the sound speed is also accounted for in the esti- Reynolds number 885 case
mation of the CFL number. As a result, for the 49.7 m/s release
velocity case and 12-cells grid, the time-step was equal to 3.5e- In Fig. 2, the mean values of hydrogen mass fraction, Y, and
008 s in ILES case whereas equal to 7.5e-007 s in the ADREA-HF the inversed mean axial (vertical) velocity, w, are presented
case. A time-step dependency of the results in the ADREA-HF for the three grids of the ADREA-HF code along with ILES 12-
case was conducted using half time-step and the impact on cells simulation and the experiment. Values have been
the results was minor. normalized with the release velocity, W, in the case of axial
The simulation time was different among the simulations velocity profile and with the mean mass fraction value at the
due to high cpu-time demand of LES, especially when dense centerline, YCL, in the case of mass fraction radial profiles. In
grids are used. In ADREA-HF, the simulation of each grid was the X-axis of the diagrams, vertical distance z has been
initialized with the results achieved with the coarser one for normalized with the release diameter d whereas radial dis-
faster convergence of the statistics. For example, when the tance y with L1/2 which is the distance at which the mean
Reynolds number is equal to 885, the simulation time was mass fraction is equal to half of the mass fraction at the
equal to 2.35 s, 0.72 s and 0.34 s for 2-cells, 6-cells and 12-cells centerline. Normalization without L1/2 is also presented at the
discretization case respectively. The initial period of 0.25 s, insets.
0.05 s and 0.01 s respectively were excluded from statistics Regarding the centerline profiles of ADREA-HF results, we
calculations in order to account only for statistically steady observe that the coarse grid does not predict well the mixing
state conditions. A time period around 0.3 s was also used in with the surrounding air, overestimating hydrogen and axial
ILES case in order to estimate the statistics. In the next Sec- velocity at the centerline. The two denser grids have a much
tion, only the results at the points where a satisfactory sta- better agreement with the experiment. Concerning the radial
tistics convergence was achieved are presented. In general, as profiles, there is satisfactory agreement with the theoretical
the distance from the release point gets higher, the required values (the theoretical curve presented in the figures agrees
simulation time gets also higher because velocities decrease very well with the measurements [26]) for all grids, with the
and the time-scales increase. Thus, some vertical profiles in only exception being the tail of the z/d ¼ 25 profile for the 2-
section Results and discussion are shorter than the others. cells case. Regarding the L1/2 distance, L1/2 is estimated based

Fig. 2 e Mean values for the three examined grids of ADREA-HF along with ILES 12-cells grid and with the experiment. Top:
Hydrogen mass fraction and normalized inversed axial velocity at the centerline. Bottom: Radial profiles of normalized
hydrogen mass fraction versus normalized radial distance with L1/2 (and with d at the insets).

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
6 international journal of hydrogen energy xxx (xxxx) xxx

on the experimental results [26,27] equal to 3.2 d at z/d ¼ 25 the resolved part of turbulence (calculated based on time se-
distance and 6.5 d at z/d ¼ 50. In the simulations, L1/2 at z/ ries produced by LES) which is the dominant one compared to
d ¼ 25 is equal to 2.8 d, 3.0 d and 3.3 d for the coarse, medium the subgrid one (even in the 2-cells case, the subgrid scale
and fine grid respectively, and at z/d ¼ 50 equal to 5.2 d, 5.9 d viscosity estimated by the Smagorinsky model is generally
and 5.0 d. We observe that at distance z/d ¼ 25 the predicted very small and comparable to the laminar viscosity). The non-
values seems to converge to the experimental one as the grid smooth predicted profiles and the scatter of the experimental
is refined. On the other hand this does not happen at the z/ results is a sign of possible unsteadiness and of large scale
d ¼ 50 distance, probably due to the coarse non-uniform grid turbulent features of the flow field. ADREA-HF results without
at this area. the Van Driest correction are also presented for mass fraction
ILES 12-cells results are in general in good agreement with and axial velocity fluctuations in the 12-cells case.
ADREA-HF results and with the experiments. Some differ- In ADREA-HF simulations, we observe that the coarse grid
ences exist in the centerline mass fraction near the release predicts approximately half values of mass fraction RMS
where ILES overestimates the experiment in a higher degree fluctuations at the centerline and approximately two-thirds of
compared to ADREA-HF results. On the other hand, ILES the values at the radial profiles. Denser grids are in much
achieves similar performance regarding the axial velocity better agreement with the experiment. Near the release, the 6-
being slighter closer to the experiment. We should mention cells grid achieves good results whereas the 12-cells grid
here, however, that the high density grid is vital for the good overestimates mass fraction and w-velocity fluctuations.
performance of the ILES code. Simulation with a coarser grid Mass fraction fluctuations are satisfactorily predicted in
(6-cells discretization of the release diameter) failed to predict moderate distances (z/d ¼ 10e30) and are underpredicted at
a turbulent flow immediately downstream the release. The higher points failing to reproduce the increase with distance.
flow transited to turbulence only after the distance of Axial velocity fluctuations are also underpredicted (but to a
approximately 15 diameters. Enforcement of synthetic tur- smaller degree), at distances greater than z/d ¼ 25. The u-ve-
bulence at the release area was tested in that case in order to locity fluctuations are in better agreement with the experi-
improve the results and it was found that the transitional ment. One possible reason for the underprediction of mass
distance decreases to 5 diameters, and that in the far field the fraction and axial velocity fluctuations far from the release is
results are close to 12-cells case. More discussion about inlet the fact that the grid gets coarser after the z/d ¼ 25 height.
treatment is made in section Effect of release modelling. In the mass fraction fluctuations a peak is observed at 12-
In Fig. 3, the root mean square (RMS) of mass fraction and cells case at z/d ¼ 5, overestimating significantly the
velocity components fluctuations are presented normalized experiment. In 12-cells case the flow is better resolved and
with the centerline mean mass fraction YCL and centerline thus a near-laminar region is predicted after the release
mean velocity wCL respectively. These values correspond to until the distance of approximately z/d ¼ 1.0. We assume

Fig. 3 e RMS fluctuations for the three examined grids of ADREA-HF along with ILES 12-cells grid and with the experiment.
Top: Normalized hydrogen mass fraction and normalized velocities at the centerline. Bottom: Radial profiles of normalized
hydrogen mass fraction versus normalized radial distance with L1/2 (and with d at the insets).

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 7

that this may cause a more violent transition to turbulence exists in these distances may also contribute to the divergence
downstream the release and to the creation of the fluctua- from the experimental results. Nevertheless, these assump-
tion peak. We expect that the non-ideal release conditions tions need to be investigated.
that likely exist at the experiment (e.g. non-flat velocity ILES results are very close to ADREA-HF results, especially
profile, velocity perturbations etc.) produce a smoother in the case in which the Van Driest correction is not applied.
transition to turbulence closer to the release reducing the This is explained by the fact that no such correction is
maximum fluctuation value. imposed in the case of ILES. The peak of the mass fraction and
Regarding radial mass fraction fluctuations, we observe velocity fluctuations around z/d ¼ 6 height is in very good
that the maximum does not occur at the centerline but at agreement between the two codes. The impact of the Van
some distance from it. This is the region where the highest Driest correction on the results can be explained by the fact
gradient of hydrogen mass fraction exists and thus turbulence that, when Van Driest correction is not used, a near-laminar
level and mixing with the surrounding air are strongest. This flow is predicted in a bigger area (until approximately z/
is common feature of momentum-dominated jets [26,34,47]. d ¼ 1.9e2.8) due to the increased subgrid-scale turbulence
At z/d ¼ 25 distance, where momentum dominates, the which hinders resolved fluctuations, resulting in a more rapid
agreement of the 6-cells is excellent whereas in the 12-cells transition to turbulence after this distance and thus to the
case a small underprediction exists. The fact that 12-cells fluctuations peak at z/d ¼ 6. This is obvious is the mass frac-
case predicts a sharp peak at the centerline fluctuations at tion fluctuations figure in which zero values are predicted in a
about z/d ¼ 5 overestimating the experiment, may correlate larger area of the vicinity of the release in the “no Van Driest”
with the underprediction at the radial profile. At z/d ¼ 50 case. The impact of the Van Driest correction on the u-velocity
distance, all grids underestimate the radial fluctuations. The fluctuations is very small. The same is true for the mean
coarse grid at this area may be responsible for the deviations values of mass fraction and axial velocity (not shown).
from the experiment. Moreover, after the distance of z/d ¼ 40, Fig. 4 presents the instantaneous and averaged volume
where buoyant effects get stronger, the increase in the concentrations in a yz plane passing through the middle of the
measured mass fraction fluctuations at the centerline is not release until the height of 80 mm (z/d ¼ 42). A colored video of
reproduced in the simulations. These fluctuations perhaps are the predicted hydrogen contours in which concentration
related with very low-frequency oscillations of the buoyant values and turbulent structures are more easily seen is pre-
plume. As a result much higher simulation time is needed in sented for the 12-cells case of ADREA-HF simulation in the
order to be reproduced. The coarser and non-uniform grid that Supplementary materials. Fine scale structures can be seen in

Fig. 4 e Instantaneous (top) and averaged (bottom) volume fraction contours for various grids of ADREA-HF simulations and
of 12-cells ILES case. Experimental photos [26] are also shown on the right in which the initial part of the jet is missing
(photos are aligned to the rest of the figures).

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
8 international journal of hydrogen energy xxx (xxxx) xxx

usual limit of 80% for properly resolved LES. This is in accor-


dance with the underprediction of the RMS velocity fluctua-
tions that is observed in the profiles presented in Fig. 3. Away
from the release, the resolved ratio increases because the SGS
stresses decrease due to the fact that velocity gradients
decrease. Even though the resolved ratio is high, the agree-
ment with the experiment is not improved (Figs. 2 and 3),
probably due to the influence of the inaccurate predictions
near the release. The medium and fine grids (6 and 12-cells
discretization of the release respectively) exhibit very good
resolved ratio, with the minimum value being greater than
90% and 95% respectively.
Fig. 5 e Resolved ratio of turbulent kinetic energy in The normalized energy density spectra of the axial ve-
ADREA-HF LES for various grids. locity component are presented in Fig. 6 at the centerline in
three distances from the release (f is the frequency, Sww the
energy density spectrum, s2w the axial velocity variance, z the
vertical distance from the source and wav the mean axial
the instantaneous hydrogen contours, especially in the 12-
velocity at the given distance). We observe that the coarse
cells case, being in general in good qualitative agreement
grid gives an odd spectrum near the release (z/d ¼ 5) due to
with the experimental photo. ADREA-HF and ILES 12-cells
the very low resolution. At higher points the spectrum is
predict similar results. However, the laminar region very
improved. In ADREA-HF simulations, fine grids give physical
near the release that ILES predicts is also exhibited in these
spectra even at the z/d ¼ 5 distance. These spectra follow the
contours. It is interested to note that even in the two 2-cells
expected 2/3 law which means that part of the inertial
case where fine scale structures cannot be reproduced, the
subrange is resolved. The spectra of the two fine grids differ
qualitative agreement with the experiment of the overall
mainly at the end of the inertial subrange, with the densest
mixing is satisfactory. As a result even such a coarse mesh can
grid resolving greater spectrum of frequencies as expected.
give good results in real scale applications as for example is
Regarding ILES, an excellent agreement with the 12-cells case
demonstrated in Ref. [48]. Fig. 4 is provided just to show an
of the ADREA-HF simulation is achieved at z/d ¼ 10 and a
indicative visual representation of the case examined and
good agreement at z/d ¼ 21. A smaller part of the inertial
should by no means be used for quantitative comparison with
subrange is resolved from ILES at z/d ¼ 21 distance though.
the experiment, due to the different visualization methodol-
Close to the release, at z/d ¼ 5, ILES does not predict a fully
ogies used, the different brightness of the pictures and the not
turbulent flow and as a result the spectrum deviates from
exactly the same contour levels.
ADREA-HF results.
In Fig. 5 the resolved ratio is presented. Resolved ratio is
In Fig. 7, results of the RNG-LES are presented and com-
estimated as kres =ðkres þksgs Þ where kres is the resolved part of
parison with the Smagorinsky LES and the experiment is
turbulent kinetic energy equal to kres ¼ 0:5ðu0 2 þv0 2 þw0 2 Þ and performed for the two fine grids. We observe that in the 6-cells
ksgs the subgrid part estimated from the relation case the RNG-LES underpredicts significantly the turbulent
m2sgs =ðr2 C2s V2=3 Þ where msgs is the subgrid scale turbulent vis- fluctuations which results in overprediction of the centerline
cosity estimated by the Smagorinsky model, r the mixture hydrogen mean mass fraction. It is known that the RNG-LES
density, Cs the Smagorinsky constant (equal to 0.1 in our case) adjusts the subgrid turbulence to the local levels of turbu-
and V the volume of the computational cell. We observe that lence: in case of close-to-laminar flow, it eliminates the sub-
with the coarse grid (2 cells discretization of the release) the grid turbulent contribution and usually provides better results
resolved ratio near and above the source is low, lower than the in cases where both turbulent and laminar parts of the flow

Fig. 6 e Normalized energy density spectra of the axial velocity at various distances from the release for the three examined
grids of ADREA-HF simulations and of 12-cells ILES case.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 9

Fig. 7 e Comparison of the results of the Smagorinsky subgrid model with those of the RNG-LES subgrid model for the
medium and the dense grid.

Fig. 8 e Effect of the Smagorinsky constant on mass fraction and velocity fluctuations for the 6-cells grid.

exist (see also [7]). In the particular case, RNG-LES predicts a Effect of Smagorinsky constant
negligible amount of SGS viscosity along the centerline. This The effect of Smagorinsky constant was examined in the 6-
behavior seems to hinder unexpectedly the development of cells grid. This grid was chosen over the 12-cells grid
resolved turbulence fluctuations away from the release lead- because 6-cells grid is more realistic for large scale hydrogen
ing to the deviation from the results of the Smagorinsky safety applications and additionally because the effect of
model. In the 12-cells case though, the effect of the subgrid the Smagorinsky constant is more pronounced in coarser
scale model is smaller because its contribution is reduced due grids. In Fig. 8 the effect of different values of the Smagor-
to the very fine resolution. Thus the results of RNG-LES and insky constant on the fluctuations’ time-series are pre-
Smagorinsky are close. sented for the 6-cells grid. We observe that in a short
distance from the release (approximately z/d ¼ 2e5), the
higher the value of the Smagorinsky constant is the lower
the fluctuations are. The reason for this is the increase of
the subgrid-scale viscosity as the Smagorinsky constant
increases. As a result the resolved fluctuations are
smoothed out due to the increased viscosity. However, it is
interesting that after some distance from the release, this
trend is inverted and the higher Smagorinsky constant ex-
hibits higher fluctuations. For example at z/d ¼ 10, Cs ¼ 0.15
achieves the highest fluctuations value, followed by the Cs ¼
0.10, Cs ¼ 0.05 and Cs ¼ 0.00. It is assumed that a possible
reason for this behavior is the fact that as the Smagorinsky
constant increases, the transition to fully turbulent flow
occurs at a greater distance from the release. In fact, the
study of contour lines diagrams (not shown) reveals that in
the extreme case of Cs ¼ 0.2, the flow is practically laminar
approximately until the distance of z/d ¼ 25, exhibiting low-
Fig. 9 e Effect of the Smagorinsky constant on the axial frequency and high-amplitude oscillations due to the un-
velocity energy spectrum at z/d ¼ 10 for the 6-cells grid. steady transitional flow. As a result, a high peak is created

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
10 international journal of hydrogen energy xxx (xxxx) xxx

for mass fraction and axial velocity fluctuations. In Cs ¼ 0.15


case the transition to turbulence seems to occur around z/
d ¼ 6 and in the other cases at smaller distances.
The axial velocity spectra at the area of z/d ¼ 10 is pre-
sented in Fig. 9. We observe that for lower values of the
Smagorinsky constant (0.00 and 0.05) the peak of the spectra
has been moved to higher frequencies. This may be correlated
with the underprediction of fluctuations at this area. More-
over, the resolved inertial subrange is very small in these
cases. The Cs ¼ 0.15 spectrum is similar to the one of Cs ¼ 0.1
with the exception of the inertial subrange which is smaller.
Finally, the spectrum in Cs ¼ 0.20 case is moved to lower fre-
quencies due to the unsteady laminar flow that develops in
this area.
In terms of mean mass fraction concentrations and mean
axial velocity (not shown), the constant Cs ¼ 0.1 achieves in
overall the best agreement with the experiment. Only the
small values of Cs (0.00 and 0.05) achieve slighter better results Fig. 10 e Hydrogen volume fraction (left) and axial velocity
in mean mass fraction concentrations at small distances from (right) for the case where the flow inside the pipe is solved
the release (lower than z/d ¼ 7). However, the deviation at for the Reynolds number equal to 885.
medium distances is significant, overestimating the experi-
mental value due to the underestimation of fluctuations. The
value of Cs ¼ 0.15 overestimates the experiment near the
release more than the Cs ¼ 0.1 case, whereas at higher dis- case where a constant velocity profile is imposed. As a result,
tances exhibits similar results. Finally, the value of Cs ¼ 0.2 the flow does not transit to turbulence immediately after the
achieved by far the worst agreement with the experiment, exit from the pipe as it is the case where no flow inside the
overestimating significantly both the mass fraction and the pipe is solved. Moreover, in the case of a parabolic-like profile,
axial velocity due to the predicted large laminar region. denser grid around the edge of the release area may be
required to predict the transition correctly. This is indicated
Effect of release modelling by the fact that ILES case with the 6 computational cells along
In the present simulations, the release was modelled using a the diameter did not predict an immediate transition to tur-
constant inlet velocity, which is equal to the average experi- bulence (even if constant inlet profile is used) whereas the 12-
mental one, at all points of the release area. No inlet turbu- cells simulation predicted it.
lence was imposed. This is an easy and approximate way to Another explanation for the deviation from the experi-
model the release. In reality, a parabolic-like velocity profile is mental behavior in the pipe case is the fact that experimental
expected to exist at the release area due to the hydrogen flow imperfections may have disrupted the laminar characteris-
that has developed inside the pipe. Moreover, disturbances in tics of the flow inside the pipe. Indeed, experimentalists
the profile are likely to exist depending on the actual experi- characterize this case as transitional [26]. As a result, a ve-
mental conditions. For example in Ref. [49] a top-hat velocity locity profile closer to a top-hat profile probably exists instead
profile was used and in Ref. [50] a laminar Blasius profile, both of the predicted parabolic one. An approach that could lead
with synthetic perturbations at the inlet. An additional issue is to better agreement with the experiment is to impose such a
that in LES some space is needed from the entrance of the flow profile at the inlet including synthetic perturbations, as for
in the domain in order the flow to adjust its turbulent example is made in Ref. [51] for a jet of Reynolds number
characteristics. equal to 1273. Synthetic perturbations at the inlets are very
In order to investigate the effect of the release modelling, common in LES because they minimize the distance that is
an additional simulation was conducted in which part of the needed for the flow to adjust to its turbulent characteristics.
flow inside the pipe is also solved. The pipe length was chosen In fact, synthetic perturbations in the case of the ILES 6-cells
equal to 0.05 m and a constant inlet velocity equal to 49.7 m/s simulation improved the predictions significantly. The con-
was imposed at its bottom. No fluctuations were imposed, stant velocity profile that we used in this work seems to
similarly to our previous simulations. The 12-cells grid was correct to some extent the absence of turbulence-generating
used for better resolution of the flow inside the pipe. physical parameters that probably existed in the experiment,
In Fig. 10 the hydrogen mass fraction and the axial velocity leading to the satisfactory agreement with the experimental
are presented. We observe that the results have changed results.
significantly. The most important difference is the fact that
after the exit of hydrogen the flow remains laminar and Reynold number 1360 and 2384 cases
transits to turbulence away from the release (z/d ¼ 16). This
behavior does not agree with the experiment. The reason for In Fig. 11, the mean values of hydrogen mass fraction and the
this behavior probably is the parabolic velocity profile that has inversed axial velocity in the centerline along with the radial
been formed at the pipe exit. This profile results in lower ve- profile of the mean mass fraction are presented for the 1360
locity gradients when hydrogen exits the pipe compared to the and the 2384 Reynolds number cases, for three grids each. We

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 11

Fig. 11 e Mean values comparison with the experiment for the three examined grids for the 1360 (top) and 2384 (bottom)
Reynolds number.

observe that in both cases, the results are qualitatively the coarse grid underestimating the fluctuations and the fine
similar. The results of the coarse grid deviate from the ex- grids being in a much better agreement with the experiment
periments the most. In general, the denser the grid is, the in both Reynolds number cases. The peak in the mass fraction
better the results are. Similarly to the previous case though fluctuations around z/d ¼ 5 in the 12-cells grid is also created
(885 Reynolds number), the 6-cells grid achieves better here. However its magnitude is much smaller than in the 885
agreement with the experiment regarding the centerline mass Reynold number case. Regarding the axial velocity fluctua-
fraction near the release. The laminar region that is predicted tions, the 12-cells case predicts the higher values, being in the
in the first diameter distance from the release in the 12-cells best agreement with the experiment until approximately z/
case is responsible for this. Regarding the centerline axial d ¼ 30. However in the 2384 Reynolds number case, even the
velocity in the 12-cells grid, an excellent agreement with the 12-cells grid underpredicts the fluctuations around the z/
experiment approximately until z/d ¼ 30 is achieved in the d ¼ 10 distance. This probably causes the overprediction in the
1360 Reynolds number case whereas an overestimation exists centerline mean velocity value that was discussed above
in the 2384 case. The reason for this probably is the fully indicating the need of imposing synthetic fluctuations at the
developed turbulent velocity profile at the jet exit that was inlet.
formed in this experiment [27]. Consequently, turbulent fluc- An interesting observation, which is a common feature in
tuations are probably required to be imposed at the inlet in both mass fraction and axial velocity fluctuations, is that
order to match the experimental conditions. This would in- smaller values near the release lead to bigger values after
crease the mixing with the surrounding air, lowering the ve- some distance whereas larger values causes smaller values
locity at the centerline (e.g. see Ref. [52]). As far as the radial downstream. A similar behavior is seen in section Effect of
profiles is concerned, both 6-cells and 12-cells grids achieves Smagorinsky constant. As a result, away from the release,
very good results in both Reynolds number cases with the 2- there are regions where the 6-cells case predicts higher (bet-
cells grid failing to reproduce the correct profile especially at ter) values than the 12-cells case, or even the 2-cells case
bigger radial distances. Even if the normalization with L1/2 is achieves similar or higher values compared to the denser
removed, the agreement remains similar. In the 1360 Rey- grids. The coarser grid after the distance of z/d ¼ 25 may
nolds number case, L1/2 distance at z/d ¼ 25 is predicted equal contribute to this behavior in these cases.
to 2.9 d, 2.9 d and 3.1 d for the coarse medium and fine grid Regarding u-velocity fluctuations, in the 1360 Reynolds
respectively, whereas for the 2384 Reynolds number case, number case, the 12-cells grid overestimate the experiment in
equal to 2.8 d, 2.8 d and 3.0 d. These values are very close to the a significant portion of the flow whereas the 6-cells grid agrees
experimental value of 3.2 d [26,27]. well with the measurements. In the 2384 Reynolds number
In Fig. 12, the root mean square (RMS) values are presented. case the 12-cells grid is in excellent agreement with the
Qualitatively, the behavior is similar to the previous case with experiment while the 6-cells grid underestimates it.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
12 international journal of hydrogen energy xxx (xxxx) xxx

Fig. 12 e Centerline RMS values comparison with the experiment for the three examined grids for the 1360 (top) and 2384
(bottom) Reynolds number.

Fig. 13 e Radial RMS values comparison with the experiment for the three examined grids for the 1360 (left) and 2384 (right)
Reynolds number.

In Fig. 13, the radial distributions of mass fraction fluctu- turbulent hydrogen jets. The ADREA-HF code was mainly
ations are presented at z/d ¼ 25. In 1360 Reynolds number case utilized for that purpose. Three grids were used with the
both 6 and 12-cells grids achieve an excellent agreement with intention of assessing the effect of grid resolution on the re-
the experiments. In the 2384 Reynolds number case the 6-cells sults. It was seen that even the coarser grid with the 2-cells
grid achieves the best agreement whereas the 12-cells grid discretization of the release diameter is enough to reveal the
overpredicts slightly the measurements. Finally, 2-cells grid basic qualitative aspects of the phenomenon. However, the
underpredicts the experiment in both Reynolds number cases. grid that uses 2-cells discretization fails to accurately repro-
duce the turbulent characteristics of the flow whereas the
denser grids with 6 and 12-cells discretization reproduce
Conclusions satisfactorily the experiment. The study of the resolved ratios,
velocity spectra and instantaneous hydrogen volume fraction
Three experimental cases were used in order to evaluate contours indicates that LES properly resolves the turbulent
different modelling strategies in LES of low Reynolds number flow field. However, various discrepancies between the

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 13

simulation and the experimental results are observed. These


differences are believed to be a result of the approximate way Appendix A. Supplementary data
that hydrogen inlet was modelled due to the lack of knowl-
edge for the exact experimental conditions at the release. A Supplementary data to this article can be found online at
more accurate velocity profile and synthetic fluctuations at https://doi.org/10.1016/j.ijhydene.2020.10.008.
the inlet might be required for improved results, especially at
the high Reynolds number case, in which the simulations references
overestimate the mean mass fraction and axial velocity.
Moreover, the non-uniform grid that was used away from the
release is believed to contribute to the discrepancies found in [1] Launder BE, Spalding DB. The numerical computation of
these distances. turbulent flows. Comput Methods Appl Mech Eng
Regarding the subgrid scale model, its impact on the re- 1974;3:269e89. https://doi.org/10.1016/0045-7825(74)90029-2.
sults is important for the grid of 6-cells discretization of the [2] Wilcox DC. Turbulence modeling for CFD. 3rd ed. La Canada,
release diameter. In that case, RNG-LES significantly under- California: DCW Industries, Inc.; 2006.
predicts the turbulent fluctuations compared to both experi- [3] Sagaut P. Large eddy simulation for incompressible flows: an
introduction. 3rd ed. 2006.
ment and Smagorinsky subgrid model. In the 12-cells
[4] Gallego E, Migoya E, Martı́n-Valdepen~ as JM, Crespo A, Garcı́a J,
discretization case the differences between models results are Venetsanos A, et al. An intercomparison exercise on the
much smaller. Smagorinsky constant has an interesting effect capabilities of CFD models to predict distribution and mixing of
on the results on the 6-cells case. In general, small values of H2 in a closed vessel. Int J Hydrogen Energy 2007;32:2235e45.
the constant achieve higher fluctuations values near the https://doi.org/10.1016/j.ijhydene.2007.04.009.
release but lower fluctuations away from the release. The [5] Papanikolaou EA, Venetsanos AG, Heitsch M, Baraldi D,
Huser A, Pujol J, et al. HySafe SBEP-V20: numerical studies of
differences in the transitional area and in the energy spectra
release experiments inside a naturally ventilated residential
might correlate with this behavior.
garage. Int J Hydrogen Energy 2010;35:4747e57. https://
An additional code, based on the ILES methodology, was doi.org/10.1016/j.ijhydene.2010.02.020.
also tested from a different modelling group. In general, the [6] Venetsanos AG, Papanikolaou E, Hansen OR, Middha P,
ADREA-HF code and the ILES code gave very similar results in Garcia J, Heitsch M, et al. HySafe standard benchmark
the 12-cells case. The fact that both LES codes provide similar Problem SBEP-V11: predictions of hydrogen release and
values and close to the experimental ones increases the con- dispersion from a CGH2 bus in an underpass. Int J Hydrogen
Energy 2010;35:3857e67. https://doi.org/10.1016/
fidence on the reliability of the results. Regarding grid sensi-
j.ijhydene.2010.01.034.
tivity, the dependency of the ADREA-HF results from the grid [7] Koutsourakis N, Venetsanos AG, Bartzis JG. LES modelling of
density was less than that of the ILES code. Even in the 2-cells hydrogen release and accumulation within a non-ventilated
discretization of the release diameter case a turbulent flow ambient pressure garage using the ADREA-HF CFD code. Int J
was predicted immediately after the release point whereas in Hydrogen Energy 2012;37:17426e35. https://doi.org/10.1016/
ILES case the 6-cells discretization case predicted turbulent j.ijhydene.2012.05.146.
flow only when synthetic turbulence was imposed at the [8] Koutsourakis N, Tolias IC, Venetsanos AG, Bartzis JG.
Evaluation of an LES code against a Hydrogen dispersion
release area. This is an asset of the ADREA-HF code in per-
experiment. CFD Lett 2012;4:225e36. http://www.
forming LES for safety assessments of large scale hydrogen akademiabaru.com/doc/CFDLV4_N4_P225_236.pdf.
applications where the total number of computational cells [9] Giannissi SG, Shentsov V, Melideo D, Cariteau B, Baraldi D,
can be a limiting factor. Venetsanos AG, et al. CFD benchmark on hydrogen release
and dispersion in confined, naturally ventilated space with
one vent. Int J Hydrogen Energy 2015;40:2415e29. https://
doi.org/10.1016/j.ijhydene.2014.12.013.
Declaration of competing interest
[10] Giannissi SG, Hoyes JR, Chernyavskiy B, Hooker P, Hall J,
Venetsanos AG, et al. CFD benchmark on hydrogen release
The authors declare that they have no known competing and dispersion in a ventilated enclosure: passive ventilation
financial interests or personal relationships that could have and the role of an external wind. Int J Hydrogen Energy
appeared to influence the work reported in this paper. 2015;40:6465e77. https://doi.org/10.1016/
j.ijhydene.2015.03.072.
[11] Molkov V, Shentsov V. Numerical and physical requirements
to simulation of gas release and dispersion in an enclosure
Acknowledgements with one vent. Int J Hydrogen Energy 2014;39:13328e45.
https://doi.org/10.1016/j.ijhydene.2014.06.154.
This Publication is supported by the program of Industrial [12] Efthimiou GC, Andronopoulos S, Tolias I, Venetsanos A.
Scholarships of Stavros Niarchos Foundation, Greece. The Prediction of the upper tail of concentration distributions of
a continuous point source release in urban environments.
authors would like also to acknowledge the Greek Research &
Environ Fluid Mech 2016;16. https://doi.org/10.1007/s10652-
Technology Network (GRNET) for the computational time 016-9455-2.
granted us in the Greek National HPC facility ARIS (http://hpc. [13] Efthimiou GC, Hertwig D, Andronopoulos S, Bartzis JG,
grnet.gr) under project HyJet (pr006029). ILES simulations were Coceal O. A statistical model for the prediction of wind-speed
carried out using the equipment of the shared research facil- probabilities in the atmospheric surface layer. Boundary-
ities of HPC computing resources at Lomonosov Moscow State Layer Meteorol 2017;163:179e201. https://doi.org/10.1007/
s10546-016-0221-2.
University [53] under project 1773.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
14 international journal of hydrogen energy xxx (xxxx) xxx

[14] Hamzehloo A, Aleiferis PG. Large eddy simulation of highly classical similarity analysis. Int J Hydrogen Energy
turbulent under-expanded hydrogen and methane jets for 2011;36:15913e8. https://doi.org/10.1016/
gaseous-fuelled internal combustion engines. Int J Hydrogen j.ijhydene.2011.09.044.
Energy 2014;39:21275e96. https://doi.org/10.1016/ [29] Li Y, Zhanghai, Xiao J. LES simulation of buoyancy jet from
j.ijhydene.2014.10.016. unintended hydrogen release with GASFLOW-MPI. Int. Conf.
[15] Kim J, Jung E, Kang S. Large eddy simulation of hydrogen Hydrog. Safety, Hamburg, Ger. Sept., 2017:11e3.
dispersion from leakage in a nuclear containment model. Int [30] Venetsanos AG, Papanikolaou EA, Bartzis JG. The ADREA-HF
J Hydrogen Energy 2015;40:11762e70. https://doi.org/10.1016/ CFD code for consequence assessment of hydrogen
j.ijhydene.2015.04.156. applications. Int J Hydrogen Energy 2010;35:3908e18. https://
[16] Bonelli F, Viggiano A, Magi V. How does a high density ratio doi.org/10.1016/j.ijhydene.2010.01.002.
affect the near- and intermediate-field of high-Re hydrogen [31] Yakhot V, Orszag SA. Renormalization group analysis of
jets? Int J Hydrogen Energy 2016;41:15007e25. https://doi.org/ turbulence. I. Basic theory. J Sci Comput 1986;1:3e51. https://
10.1016/j.ijhydene.2016.06.174. doi.org/10.1007/BF01061452.
[17] Hamzehloo A, Aleiferis PG. Gas dynamics and flow [32] Karabasov SA, Goloviznin VM. Compact accurately
characteristics of highly turbulent under-expanded boundary-adjusting high-REsolution technique for fluid
hydrogen and methane jets under various nozzle pressure dynamics. J Comput Phys 2009;228:7426e51. https://doi.org/
ratios and ambient pressures. Int J Hydrogen Energy 10.1016/j.jcp.2009.06.037.
2016;41:6544e66. https://doi.org/10.1016/ [33] Houf W. Analytical and experimental investigation of small-
j.ijhydene.2016.02.017. scale unintended releases of hydrogen. Int J Hydrogen
[18] Li X, Wu K, Yao W, Fan X. A comparative study of highly Energy 2008;33:1435e44. https://doi.org/10.1016/
underexpanded nitrogen and hydrogen jets using large eddy j.ijhydene.2007.11.031.
simulation. Int J Hydrogen Energy 2016;41:5151e61. https:// [34] Viggiano B, Dib T, Ali N, Mastin LG, Cal RB, Solovitz SA.
doi.org/10.1016/j.ijhydene.2016.01.120. Turbulence, entrainment and low-order description of a
[19] Sarikurt FS, Hassan YA. Large eddy simulations of erosion of transitional variable-density jet. J Fluid Mech
a stratified layer by a buoyant jet. Int J Heat Mass Tran 2018;836:1009e49. https://doi.org/10.1017/jfm.2017.822.
2017;112:354e65. https://doi.org/10.1016/ [35] Landa PS, McClintock PVE. Development of turbulence in
j.ijheatmasstransfer.2017.04.134. subsonic submerged jets. Phys Rep 2004;397:1e62. https://
[20] Zhao M, Zhou T, Ye T, Zhu M, Zhang H. Large eddy doi.org/10.1016/j.physrep.2004.03.004.
simulation of reacting flow in a hydrogen jet into supersonic [36] O’Neill P, Soria J, Honnery D. The stability of low Reynolds
cross-flow combustor with an inlet compression ramp. Int J number round jets. Exp Fluid 2004;36:473e83. https://doi.org/
Hydrogen Energy 2017;42:16782e92. https://doi.org/10.1016/ 10.1007/s00348-003-0751-5.
j.ijhydene.2017.04.250. [37] Koller-Milojevie D, Schneider W. Free and confined jets at
[21] Soleimani nia M, Maxwell B, Oshkai P, Djilali N. Experimental low Reynolds numbers. Fluid Dynam Res 1993;12:307e22.
and numerical investigation of turbulent jets issuing through https://doi.org/10.1016/0169-5983(93)90033-7.
a realistic pipeline geometry: asymmetry effects for air, [38] Smagorinsky J. General circulation experiments with the
helium, and hydrogen. Int J Hydrogen Energy primitive equations. I. The basic experiment. Mon Weather
2018;43:9379e98. https://doi.org/10.1016/ Rev 1963;91:99e164.
j.ijhydene.2018.03.197. [39] Van Driest ER. On turbulent flow near a wall. J Aeronaut Sci
[22] Zhang H, Li Y, Xiao J, Jordan T. Detached Eddy Simulation of 1956;23:1007e11. https://doi.org/10.2514/8.3713.
hydrogen turbulent dispersion in nuclear containment [40] Yakhot V, Orszag SA. Renormalization group analysis of
compartment using GASFLOW-MPI. Int J Hydrogen Energy turbulence. I. Basic theory. J Sci Comput 1986;1:3e51. https://
2018;43:13659e75. https://doi.org/10.1016/ doi.org/10.1007/BF01061452.
j.ijhydene.2018.05.077. [41] Tolias IC, Koutsourakis N, Hertwig D, Efthimiou GC,
[23] Li X, Chen Q, Chen M, He Q, Christopher DM, Cheng X, et al. Venetsanos AG, Bartzis JG. Large Eddy Simulation study on
Modeling of underexpanded hydrogen jets through square and the structure of turbulent flow in a complex city. J Wind
rectangular slot nozzles. Int J Hydrogen Energy 2019;44:6353e65. Eng Ind Aerod 2018. https://doi.org/10.1016/
https://doi.org/10.1016/j.ijhydene.2019.01.079. j.jweia.2018.03.017.
[24] Bernard-Michel G, Saikali E, Sergent A, Tenaud C. [42] Kovalets IV, Andronopoulos S, Venetsanos AG,
Comparisons of experimental measurements and large eddy Bartzis JG. Optimization of the numerical algorithms of
simulations for a helium release in a two vents enclosure. Int the ADREA-I mesoscale prognostic meteorological model
J Hydrogen Energy 2019;44:8935e53. https://doi.org/10.1016/ for real-time applications. Environ Model Software
j.ijhydene.2018.07.120. 2008;23:96e108. https://doi.org/10.1016/
[25] Saikali E, Bernard-Michel G, Sergent A, Tenaud C, Salem R. j.envsoft.2007.05.004.
Highly resolved large eddy simulations of a binary mixture [43] Semiletov VA, Karabasov SA. CABARET scheme with
flow in a cavity with two vents: influence of the conservation-flux asynchronous time-stepping for nonlinear
computational domain. Int J Hydrogen Energy aeroacoustics problems. J Comput Phys 2013;253:157e65.
2019;44:8856e73. https://doi.org/10.1016/ https://doi.org/10.1016/j.jcp.2013.07.008.
j.ijhydene.2018.08.108. [44] Faranosov GA, Goloviznin VM, Karabasov SA, Kondakov VG,
[26] Schefer RW, Houf WG, Williams TC. Investigation of small- Kopiev VF, Zaitsev MA. CABARET method on unstructured
scale unintended releases of hydrogen: buoyancy effects. Int hexahedral grids for jet noise computation. Comput Fluids
J Hydrogen Energy 2008;33:4702e12. https://doi.org/10.1016/ 2013;88:165e79. https://doi.org/10.1016/
j.ijhydene.2008.05.091. j.compfluid.2013.08.011.
[27] Schefer RW, Houf WG, Williams TC. Investigation of small- [45] Fureby C, Grinstein FF. Large eddy simulation of high-
scale unintended releases of hydrogen: momentum- Reynolds-number free and wall-bounded flows. J Comput
dominated regime. Int J Hydrogen Energy 2008;33:6373e84. Phys 2002;181:68e97. https://doi.org/10.1006/
https://doi.org/10.1016/j.ijhydene.2008.05.041. jcph.2002.7119.
[28] Hourri A, Gomez F, Angers B, Be nard P. Computational study [46] Givoli D. Non-reflecting boundary conditions. J Comput Phys
of horizontal subsonic free jets of hydrogen: validation and 1991;94:1e29. https://doi.org/10.1016/0021-9991(91)90135-8.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008
international journal of hydrogen energy xxx (xxxx) xxx 15

[47] Pitts WM, Kashiwagi T. The application of laser-induced Flow 2019;79:108460. https://doi.org/10.1016/
Rayleigh light scattering to the study of turbulent mixing. J j.ijheatfluidflow.2019.108460.
Fluid Mech 1984;141:391e429. https://doi.org/10.1017/ [51] Zhou X, Luo KH, Williams JJR. Large-eddy simulation of a
S0022112084000902. turbulent forced plume. Eur J Mech B Fluid 2001;20:233e54.
[48] Giannissi SG, Tolias IC, Melideo D, Baraldi D, Shentsov V, https://doi.org/10.1016/S0997-7546(00)01117-1.
Makarov D, et al. On the CFD modelling of hydrogen [52] Salkhordeh S, Mazumdar S, Tyler Landfried D, Jana A,
dispersion at low-Reynolds number release in closed facility. Kimber ML. Les of an isothermal high Reynolds number
Int J Hydrogen Energy 2020. https://doi.org/10.1016/ turbulent round jet. In: Int. Conf. Nucl. Eng. Proceedings,
j.ijhydene.2020.09.078. In press. ICONE. vol. 4. American Society of Mechanical Engineers
[49] Dairay T, Fortune V, Lamballais E, Brizzi LE. LES of a (ASME); 2014. https://doi.org/10.1115/ICONE22-31154.
turbulent jet impinging on a heated wall using high-order [53] Voevodin VV, Antonov AS, Nikitenko DA, Shvets PA,
numerical schemes. Int J Heat Fluid Flow 2014;50:177e87. Sobolev SI, Sidorov IY, et al. Supercomputer lomonosov-2:
https://doi.org/10.1016/j.ijheatfluidflow.2014.08.001. large scale, deep monitoring and fine analytics for the user
[50] Wawrzak K, Boguslawski A, Tyliszczak A, Saczek M. LES community. Supercomput Front Innov 2019;6:4e11. https://
study of global instability in annular jets. Int J Heat Fluid doi.org/10.14529/jsfi190201.

Please cite this article as: Tolias IC et al., Large Eddy Simulation of low Reynolds number turbulent hydrogen jets - Modelling consid-
erations and comparison with detailed experiments, International Journal of Hydrogen Energy, https://doi.org/10.1016/
j.ijhydene.2020.10.008

You might also like