You are on page 1of 13

G Model

NED-8343; No. of Pages 13 ARTICLE IN PRESS


Nuclear Engineering and Design xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Nuclear Engineering and Design


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

The role of CFD combustion modeling in hydrogen safety


management – IV: Validation based on non-homogeneous
hydrogen–air experiments
Pratap Sathiah a,∗ , Ed Komen a , Dirk Roekaerts b
a
Nuclear Research and Consultancy Group (NRG), Westerduinweg 3, 1755 ZG Petten, The Netherlands
b
Delft University of Technology, Department of Process and Energy, Section Fluid Mechanics, Mekelweg 2, 2628 CD Delft, The Netherlands

h i g h l i g h t s

• TFC combustion model is further extended to simulate flame propagation in non-homogeneous hydrogen–air mixtures.
• TFC combustion model results are in good agreement with large-scale non-homogeneous hydrogen–air experiments.
• The model is further extended to account for the non-uniform hydrogen–air–steam mixture for the presence of PARs on hydrogen deflagration.

a r t i c l e i n f o a b s t r a c t

Article history: The control of hydrogen in the containment is an important safety issue in NPPs during a loss of coolant
Received 20 December 2013 accident, because the dynamic pressure loads from hydrogen combustion can be detrimental to the
Received in revised form 12 May 2015 structural integrity of the reactor safety systems and the reactor containment. In Sathiah et al. (2012b),
Accepted 26 May 2015
we presented a computational fluid dynamics based method to assess the consequence of the combus-
tion of uniform hydrogen–air mixtures. In the present article, the extension of this method to and its
validation for non-uniform hydrogen–air mixture is described. The method is implemented in the CFD
software ANSYS FLUENT using user defined functions. The extended code is validated against non-uniform
hydrogen–air experiments in the ENACCEF facility.
It is concluded that the maximum pressure and intermediate peak pressure were predicted within
12% and 18% accuracy. The eigen frequencies of the residual pressure wave phenomena were predicted
within 4%. It is overall concluded that the current model predicts the considered ENACCEF experiments
well.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction residual risk of possible hydrogen deflagrations and for the opti-
mal design of hydrogen mitigation systems in order to reduce this
The risk of hydrogen release and combustion during a severe risk as far as possible. The computer modeling based on CFD sim-
accident in a light water reactor (LWR) has received considerable ulates the physical phenomena governing hydrogen combustion,
attention after the three mile island (TMI, 2011) accident in 1979 that is, turbulence, combustion, turbulence-chemistry interaction,
and the Fukushima accident, Japan (Fukushima, 2011) in 2011. The heat-transfer, and buoyancy.
recent accident in Fukushima showed the destructive power of a In Sathiah et al. (2012b), we presented a hydrogen risk
hydrogen explosion, and therewith the importance of hydrogen assessment methodology which is based on a combined lumped
control measures. Hydrogen mitigation systems like e.g. passive parameter/CFD code modeling approach. It is worth mentioning,
autocatalytic recombiners (PARs) and igniters can be designed to that the considered CFD modeling approach can be applied to slow
reduce the risk of a possible hydrogen deflagration. Reliable com- and fast deflagration and cannot be used for deflgration to detona-
puter modeling is needed for the assessment of the associated tion transition (DDT) and detonation. Before such a methodology
can be adopted for the assessment of hydrogen risks in real scale
containments, validation of the CFD modeling approach is neces-
∗ Corresponding author. Tel.: +31 75809477607. sary. To reach this objective, we take a step by step approach. In
E-mail addresses: pratap.sathiah78@gmail.com (P. Sathiah), komen@nrg.eu the first step, the model was validated against respectively small
(E. Komen), d.j.e.m.roekaerts@tudelft.nl (D. Roekaerts). scale uniform methane-air and larger scale uniform hydrogen–air

http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
0029-5493/© 2015 Elsevier B.V. All rights reserved.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
2 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

considered model predicted the considered ENACCEF experiments


Nomenclature very well. As a next step, we extended our modeling approach to
account for the effect of dilution in hydrogen–air mixtures and
TFC turbulent flame closure validated this extension against uniform hydrogen–air–CO2 –He
AMR adaptive mesh refinement experiments. The results of this validation are presented in Sathiah
ISP International Standard Problem et al. (2013). From these validation analyses, it was concluded
CFD computational fluid dynamics that the maximum value of the mean pressures and the interme-
UDF User Defined Functions diate peak pressures were predicted respectively within 12 and
LDV Laser Doppler Velocimetry 29% accuracy, while the rate of pressure rise dp/dt was typically
PDF probability density function underpredicted by about 90%. The eigen frequencies of the residual
JPDF joint probability density function pressure wave phenomena were predicted within 6%. It was overall
TMI three mile island concluded that the Turbulent Flame Closure (TFC) model predicted
LWR light water reactor the considered ENACCEF experiments well.
PAR passive autocatalytic recombiner In order to apply the combustion model to real scale con-
PMT photomultiplier tubes tainments, the model must be further extended and validated to
BR blockage ratio simulate combustion of non-uniform hydrogen–air mixtures.
CNRS Centre National de la Recherche Scientifique The objectives of the current paper are therefore to:
D diameter of pipe (m)
d diameter of baffle (m)
• describe the extension of the TFC combustion model for applica-
Df,t turbulent diffusivity of mixture fraction (m2 /s)
Ds,t turbulent diffusivity of steam mixture fraction tion to non-uniform hydrogen–air mixtures;
• validate the extended model against non-uniform hydrogen–air
(m2 /s)
Df laminar diffusion coefficient of mixture fraction deflagration experiments performed in the ENACCEF facility;
(m2 /s)
Ds laminar diffusion coefficient of steam mixture frac- The paper is structured as follows: Firstly, the ENACCEF experi-
tion (m2 /s) ments used for the validation of the model are described in Section
f mixture fraction 2. Secondly, Section 3 presents the extension of the combus-
fst stoichiometric mixture fraction tion model to non-uniform hydrogen–air mixtures. In addition,
g steam mixture fraction it presents the description of the applied mesh, initial condi-
h nitrogen mixture fraction tions, physical properties, and post-processing method. Thirdly, the
k turbulent kinetic energy (m2 /s2 ) simulation results are presented and discussed for two different
p pressure (Pa) experiments in Section 4. Fourthly, the summary and conclu-
t turbulent Schmidt number sion are presented in Section 5. Finally, future steps for further
 s,t turbulent Schmidt number validation of NRG’s CFD combustion modeling for hydrogen defla-
t time (s) gration are presented in Section 6. Furthermore, a priori analysis is
u velocity (s) described in Appendix A. In addition, the model extension to simu-
 turbulent dissipation rate (m2 /s3 ) late flame propagation in non-homogeneous hydrogen–air–steam
ω turbulent source term (kg/m3 s) mixtures is presented in Appendix B. The extension to flame
t turbulent viscosity (kg/m s) propagation in hydrogen–air–steam mixtures in the presence of
t turbulent dynamic viscosity (m2 /s) passive autocatyltic recombiners (PARs) is described in Appendix
 density (kg/m3 ) C (Fig. C.1).
 equivalence ratio
u unburnt gas density (kg/m3 ) 2. Selected facility and experiments for code validation
u unburnt gas thermal diffusivity (m2 /s)
b burnt gas density (kg/m3 ) 2.1. ENACCEF facility
Sl laminar flame speed (m/s)
¯ Reynolds-averaged mean quantities A large number of hydrogen deflagration experiments have been
˜ Favre-averaged mean quantities executed in the ENACCEF test facility (Chaumeix and Bentaib, 2011,
 fluctuating quantities 2010; Bleyer et al., 2012; ISP-49, 2011). The measurements have
Cd constant been performed by the Centre National de la Recherche Scientifique
Cg constant (CNRS). The test facility consists of a dome and an acceleration tube
a0 constant (see Fig. C.3). The acceleration tube is 3.2 m long and has an inter-
a1 constant nal diameter of 0.154 m. It contains 9 annular baffles with varying
a2 constant blockage ratios BR = [1 − (d/D)2 ], where d and D are the baffle and
a3 constant pipe diameter, respectively. The first baffle is located at a distance
a4 constant of 0.776 m and the distance between subsequent baffles is 0.154 m.
a5 constant The thickness of the baffles is 2 mm. The dome is 1.7 m long and has
a6 constant an internal diameter of 0.738 m.
In the ENACCEF experiments, the flame position has been
measured using 16 UV-sensitive photomultiplier tubes (PMT)
experiments, Sathiah et al. (2012a,b). From these validation analy- (Hamamatsu, 1P28). The presence of the flame was detected based
ses, it was concluded that the maximum pressures were predicted on the total emission of the flame recorded using PMT. Using
within 13% accuracy, while the rate of pressure rise dp/dt was pre- this method, OH radicals are measured because of their very high
dicted within about 30%. The eigen frequencies of the residual concentration in the flame front and in the burnt gases. The uncer-
pressure wave phenomena after complete combustion were pre- tainty in the flame position measurements is 8 mm. The pressure
dicted within a few %. Therefore, it was overall concluded that the signal has been measured using high speed pressure transducers

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 3

Fig. C.3. The variation of unburnt gas density u with hydrogen mass fraction.

2.2. Selected experiments

Institut de Radioprotection et de Sureté Nucléaire (IRSN) has


released the data of nine ENACCEF experiments for code bench-
marking. That is, the data of the tests RUN 153, 158 and 160
(see Table C.1) have been released within the SARNET EU project,
and the data of three tests with CO2 –He dilution have been used
for code benchmarking within the SARNET-2 project. A 60 vol.%
CO2 –40 vol.% He gas mixture was added in these tests to mimic the
effect of steam dilution. The data of three tests with non-uniform
mixtures (RUN 733, 736 and 765) have been released within the
OECD International Standard Problem (ISP-49) (see Fig. C.2). In
these experiments, the initial temperature is 296 K and the initial
pressure is 100 kPa.
NRG has used the ENACCEF tests RUN 153, 158 and 160 for the
validation of its CFD modeling approach for hydrogen deflagration
in Sathiah et al. (2012b). Furthermore, NRG has used ENACCEF test
RUN 153-10, RUN 153-20 and RUN 153-30 for the validation of the
Fig. C.1. Schematic of ENACCEF geometry.
NRG’s CFD approach in Sathiah et al. (2013). As already mentioned
in the introduction, this modeling approach predicted these ENAC-
CEF experiments well. For the objectives of this paper, the two tests
(7 Chimie Metal and 2 Kistler), which are mounted on the inner RUN 765 and 736 with non-uniform hydrogen concentration have
surface of the tube. been considered.
Ignition of the gas mixtures in ENACCEF is achieved using two
thin tungsten electrodes (2 mm in diameter), which are linked to 3. CFD model
a high voltage source. The ignition source is located 0.138 m from
the bottom of the facility. Further details about the experiments The commercial CFD code ANSYS FLUENT is used as the basis for
are available in the report of Chaumeix and Bentaib (2011, 2010), the development of our hydrogen deflagration model. The unsteady
Bleyer et al. (2012) and ISP-49 (2011). It is worth stressing here Favre (density-weighted) averaged Navier–Stokes equations are
that in the considered experiments, initial turbulence levels were solved using the standard k −  turbulence model. Within the CFD
not measured, and heat losses from the domain boundary are not code, the equations for the conservation of mass, momentum,
specified. energy and the progress variable are solved. For completeness the
combustion model with submodels for preferential diffusion and
flame stretch effects for uniform hydrogen–air mixture is presented
here.

3.1. Combustion model

The TFC combustion model proposed by Zimont (1979), Zimont


and Lipatnikov (1993, 1995) and Zimont (2000) is used in our calcu-
lations. This model solves the following progress variable equation
   
∂c̃
¯ ∂¯ ũi c̃ ∂ ∂c̃ ∂ ∂c̃
+ = ¯ D̄ + D
¯ t + u Ut |∇ c̃| (1)
∂t ∂xi ∂xj ∂xj ∂xj ∂xj
where, c̃ is the Favre-averaged progress variable, Dt is the turbulent
diffusivity, Ut is the turbulent flame speed and u is the unburned
gas density. The turbulent diffusivity is evaluated as
t
Fig. C.2. Initial hydrogen concentration gradient for ENACCEF tests RUN 765, 736 Dt = . (2)
and 733. Sc t

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
4 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

Table C.1
ENACCEF experiments released for code benchmarking.

Run [H2 ] (vol.%) Blockage ratio Gas mixture

153 13 0.63 Uniform hydrogen–air mixture


158 13 0.33 Uniform hydrogen–air mixture
160 13 0 Uniform hydrogen–air mixture
153–10 13 0.63 Uniform hydrogen–air mixture with 10 vol.% CO2 –He
153–20 13 0.63 Uniform hydrogen–air mixture with 20 vol.% CO2 –He
153–30 13 0.63 Uniform hydrogen–air mixture with 30 vol.% CO2 –He
765 11.6–8.1 0.63 Hydrogen–air mixture with negative concentration gradient
736 11.4–5.8 0.63 Hydrogen–air mixture with negative concentration gradient
733 5.7–12 0.63 Hydrogen–air mixture with positive concentration gradient

Here, t is turbulent kinematic viscosity and Sct is turbulent this concept, the turbulent flame speed is controlled by the flamelet
Schmidt number. Turbulent kinematic viscosity is calculated using which is advanced farthest in the unburnt mixture. This concept is
standard k −  turbulence model (Launder and Spalding, 1972). This also supported by the analysis of Kolmogorov–Petrovski–Piskunov
is already described in our previous article (Sathiah et al., 2012a). (KPP) (Lipatnikov and Chomiak, 2002) of premixed turbulent com-
To close Eq. (1), Zimont (1979), Zimont and Lipatnikov (1993, 1995) bustion, which shows that the burning velocity is controlled by the
and Zimont (2000) have theoretically derived the following expres- leading edge.
sion for turbulent flame speed Strongly perturbed laminar flamelets. In this approach, a crit-
 l 1/4  1/4 ically curved laminar flamelet is proposed as the structure of
Ut = Au Da1/4 = Au = Au the leading edge flamelet (Karpov and Lipatnikov, 1995; Karpov
t t
(3)
u c c et al., 1997). Based upon this assumption, (Lipatnikov and Chomiak,
where A is a constant, Da is the Damkohler number, u and lt are 2005) defined a critical chemical time scale given as follows
the turbulent intensity and integral length scale, and c = u /Sl 2  T 3/2  
−1 b (Tb − Tr )
is the chemical time scale of a corresponding laminar flame, Sl cr = c Le exp (5)
Tr 2Tb Tr
is the unstretched laminar flame speed and u is the molecular
heat diffusivity of the unburned mixture. This model was later for spherical flames. Here, Tb is the adiabatic flame temperature,
extended by Lipatnikov and Chomiak (2002) to account for local is the activation temperature, and Tr is the inner layer temperature,
combustion quenching by strong turbulent stretching, local varia-
Tb − Tu
tions in temperature and mixture composition due to the difference Tr = Tu + . (6)
in molecular heat and mass diffusivities. This extended turbulent Le
flame speed model is given as The above time scale was derived by Zeldovich et al. (1985)
 1/4 under the assumption of single step chemistry and » Tb . We have
Ut = Au
t used this approach for our simulations.
F(Le)G(gcr ). (4)
c
Here, the functions F(Le), G(gcr ) model preferential diffusion 3.1.2. Stretch effects
effects and quenching, respectively. The Lewis number Le = /D is As mentioned above, the turbulent premixed flame can be
the ratio of thermal diffusivity to mass diffusivity of the deficient quenched if subjected to stretch and hence this must be considered
reactant. In case of lean hydrogen–air mixtures, hydrogen is the when modeling turbulent premixed flames. By invoking model by
deficient reactant. The expression for F and G is described in the Bray (1987), the stretch factor G(gcr ) has been closed as follows
following subsection.
 
1 −2 ln(gcr ) +  2 /2
3.1.1. Preferential diffusion effects G(gcr ) = erfc √
2 2
Numerous experiments (Karpov and Severin, 1980; Kido et al.,


1989, 1994; Wu et al., 1991; Kwon et al., 1992; Koroll et al., 1993;
1 1
  

cr
Betev et al., 1995; Karpov et al., 1997) show that the turbulent = erfc − ln + , (7)
2 2  2
burning velocity Ut depends substantially on the Lewis number.
The local composition, combustion rate, and temperature inside 
the flamelets perturbed by turbulent eddies can vary substantially where “erfc” is the complementary error function, = 15u /
due to the differences between heat losses by the molecular ther- is the Kolmogorov time scale, and  is the standard deviation of the
mal conductivity and the reactant supply by molecular diffusion. distribution of the turbulent dissipation rate 
Lipatnikov and Chomiak (1997) have shown that the increase in l 
t
turbulent flame speed with increase in turbulent intensity is much  = str ln , (8)
t
larger for hydrogen–air mixture than for hydrocarbon mixtures,
which is due to the Lewis number effects. where str is a constant with a value of 0.26, t is the Kolmogorov
Recently, preferential diffusion and Lewis number effects are length scale and cr is the turbulent dissipation rate at the critical
reviewed in detail by Lipatnikov and Chomiak (2005). According rate of strain, given by
to them, there are mainly two approaches to model these effects
cr = 15u (gcr )2 . (9)
which are based on (a) weakly perturbed laminar flamelets and (b)
strongly perturbed laminar flamelets. The latter approach is based Here, u is the unburned gas viscosity, gcr is the critical rate of strain.
on the leading point concept proposed by Zeldovich et al. (1985) The critical rate of strain can assumed to be constant or calculated
and Kuznetsov and Sabelnikov (1990). The leading point concept using an algebraic expression described here in Zimont et al. (1998).
is based on the assumption that the turbulent combustion process The model has only one model constant, i.e. the turbulent flame
is controlled by the processes localized to thin laminar flamelets speed constant, which takes a value of 0.6 for hydrogen–air flames,
which are characterized by large spatial gradients. According to as recomended in Lipatnikov and Chomiak (2002).

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 5

In the following subsections, the extension of the above com-


bustion model to non-uniform hydrogen–air mixtures is presented.   ¯ 
¯
For the actual plant application, the model must be further ∂  2 ∂ 2 ∂ ∂f  2 ∂ ∂f̃
(¯ f  ) + (¯ ũi f  ) = Df + 2f  Df
extended to simulate flame propagation in non-homogeneous ∂t ∂xi ∂xi ∂xi ∂xi ∂xi
hydrogen–air–steam mixtures. Furthermore the presence of PARs
   2
∂f ∂
should be taken into account. The corresponding model extensions  2
∂ f  2
are presented respectively in Appendices B and C. + D
¯ f,t + Cg t − Cd ¯ f  . (13)
∂xi ∂xi ∂xi k
3.2. Extension of TFC combustion model to non-uniform
hydrogen–air mixtures Here, ũi and f̃ are the Favre-averaged velocity and mixture fraction,
Df is the laminar diffusion coefficient, Df,t = t / t is the turbulent
During a severe accident in a nuclear power plant, the relatively diffusion coefficient, t is the turbulent kinematic viscosity,  t is
light hydrogen gas released into the containment may accumulate the turbulent Schmidt number, t is the time, and ¯ is the mean
in the upper part of containment compartments, thereby forming density. The default values of the constants  t , Cg , and Cd are 0.85,
a stable stratified hydrogen-rich layer. Considerable efforts have 2.86, and 2.0, respectively.
been spent within, for example, the OECD/THAI project (Kanzleiter In practice, the PDF is often assumed to be a double delta func-
et al., 2010), to understand the processes leading to the formation tion or a ˇ-function (Abramowitz and Stegun, 1972) in most of the
and dissolution of such stratified layers, and further research is still commercial CFD codes (ANSYS FLUENT, 2008; ANSYS CFX, 2007).
on-going in for example the European ERCOSAM FP7 project. Due The ˇ-PDF is indeed a function of the mean and variance of the
to the stratification of hydrogen, the hydrogen–air mixture which is mixture fraction. Then, the laminar flame speed can be calculated
formed is not uniform. This essentially means that the containment as follows:
contains regions with hydrogen-lean and hydrogen-rich mixtures. ˜2
S̃l,0 = Sl,0 (f̃ , f  ). (14)
This non-uniformity of the hydrogen–air mixture may have a sig-
nificant impact on the flame propagation and pressure dynamics. In the simplest approach, the fluctuations are neglected, and the
In the past, measurements have been performed by Whitehouse PDF is assumed to be a single delta function. In this case, the equa-
et al. (1996) in order to investigate the flame propagation in non- tion for the variance does not have to be solved, and the laminar
uniform hydrogen–air mixtures. In their experiments, much higher flame speed is simply given by:
turbulent flame speeds were observed in non-uniform mixtures
compared to uniform mixtures. Hence, this effect is important and S̃l,0 = Sl,0 (f̃ ). (15)
must be taken into account in order to adequately predict the flame
and pressure dynamics during accident scenarios in real contain- This simple approach is used here as a first step. The assessment of
ments. the validity of this approach is checked as a first step by performing
In order to extend the combustion model to non-uniform a priori mathematical analysis which is presented in Appendix A.
mixtures, concentration gradients are taken into account in the cal- In the second step, CFD simulations can be performed by solving
culation of the laminar flame speed Sl,0 . This is done by describing for mixture fraction fluctuations which is planned for future work.
the local instantaneous equivalence ratio  of the hydrogen–air A non-homogeneous distribution of the hydrogen concentration in
mixture by the mixture fraction f, as follows: a hydrogen–air mixture also affects the unburned gas density u,0 ,
the thermal diffusivity u,0 , the Lewis number Le and the molecular
f 1 − fst diffusivity D. As described in the subsequent subsections, the mean
= (10)
1 − f fst values of these quantities can be obtained in a similar way.
where fst denotes the stoichiometric mixture fraction. For a It is worth mentioning that the model presented here assumes
hydrogen–air mixture, fst takes a value of 0.028. The fluctuations statistical independence of the progress variable and the mixture
in the mixture composition, can be taken into account in the cal- fractions. This approach is used by several other researchers, for
culation of the local mean flame speed Sl,0 by using the probability example Polifke et al. (2002), Durand (2007) and ANSYS FLUENT
density function of the mixture fraction PDFf (f) as follows: (2008), ANSYS CFX (2007). A general method is based on the solu-
 tions of the joint PDF of the progress variable and the mixture
1
fractions (see for example Fox and Raman (2004)). However, this
S̃l,0 = Sl,0 (f )PDF f (f )df. (11)
general approach is beyond the scope of this paper.
0

It is worth mentioning that the laminar flame speed is used in the


3.3. Mesh, initial and boundary conditions
estimation of the chemical time scale which is required in the tur-
bulent flame speed according to expression Eq. (4) in Sathiah et al.
Adaptive mesh refinement (AMR) is an efficient method for
(2012b). To close the model, an algorithm is needed to obtain the
computing flame propagation in large and complex geometries,
PDF of the mixture fraction PDFf (f). The standard approach is to
since very high grid resolution is required only in the flame front
obtain this PDF from an assumed mathematical expression charac-
region, while away from the flame, coarser cells can be used with-
terized by the mean and the variance of the mixture fraction. Then,
out loss of accuracy. Hence, depending upon the gradients of the
the PDF is a function of the statistical variable f, parameterized by
progress variable, the mesh will be further refined or coarsened.
its Favre-averaged mean value f̃ and its variance f  . The latter two
2
Therefore, dynamic grid refinement leads to a much smaller num-
quantities are obtained from additional transport equations which ber of computational cells, which ultimately means savings in
take the following form (Poinsot and Veynante, 2001): computational time. In the past, this strategy was used by other
Transport equation for the mean mixture fraction  f: groups to simulate flame propagation, see for example Catlin et al.
    (1995) and Huld and Wilkening (2001).
∂  ∂ ∂ ∂
f ∂ ∂
f
(¯ f ) + (¯ ũi
f) = ¯ D̄f + D
¯ f,t . (12) We have used this approach for our simulations reported in this
∂t ∂xi ∂xi ∂xi ∂xi ∂xi work. Mesh refinement is done based on iso-values (0.01 and 0.99)
of the progress variable. This essentially means that the mesh is
Transport equation for variance of the mixture fraction f  :
2
refined when the progress variable is between 0.01 and 0.99. When

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
6 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

the progress variable is less 0.01 or greater than 0.99 then the mesh
is coarsened.
2D axisymmetric simulations were performed using AMR in
order to resolve the flame brush thickness. In our previous paper
(Sathiah et al., 2012b), we have compared a fully 2D axisymmetric
simulation with a full 3D simulations. From the results, we con-
cluded that the 2D axisymmetric domain is a representative for the
full 3D case and can therefore be used for the present analyses. A
uniform mesh in the x and y directions is used in the base grid. This
base grid is subsequently refined using AMR. The mesh dimensions
in x and y directions before and after the mesh refinements are
0.02 m and 0.0025 m, respectively. Further details of the applied
mesh and mesh refinement are described and validated in Sathiah
et al. (2012b).
The walls of the domain are assumed to be adiabatic, and no
Fig. C.4. The variation of unburnt gas thermal diffusivity u with hydrogen mass
flame-wall interaction was taken into account. For each experi- fraction.
ment, the initial distribution of the hydrogen concentration was
measured. These measured profiles of hydrogen concentration are
diffusivity, and Lewis number are plotted as a function of the mean
fitted by a fourth order polynomial and specified as an initial hydro-
mixture fraction f̃ in Figs. C.3–C.5.
gen distribution in the simulation.
Grid and time step sensitivity analyses have been performed for
3.6. Applied numerical schemes
each case in order to guarantee that the numerical errors can be
practically neglected in the present validation analyses.
The density based coupled solver of ANSYS FLUENT (2008) is
Stagnant flow conditions have been applied as initial conditions
used in our simulations. It solves the conservation equations for
for velocity. Since the initial turbulence parameters, for exam-
mass, momentum, energy, and the progress variable (Eqs. (1)–(4)
ple, turbulent length scale and turbulent kinetic energy are not
in Sathiah et al. (2012a)) simultaneously. The equations for the tur-
measured in the ENACCEF experiments, the values of turbulent
bulent kinetic energy and the turbulent dissipation rate (Eqs. (7)
kinetic energy, and the turbulent dissipate rate are estimated to be
and (8) in reference Sathiah et al. (2012a)) are solved sequentially.
1.5e−04 m2 s−2 and 4.8e−05 m2 s−3 , respectively. The correspond-
The code employs a 4-stage Runge–Kutta scheme which is fourth
ing values of the turbulent length scale and turbulent intensities
order accurate for time integration of unsteady flows. The time
are 0.007 m and 0.01 m/s, respectively. These values of initial tur-
step in this method is determined by the Courant–Friedrichs–Lewy
bulence were assumed for all runs. It is worth stressing that the
(CFL) condition. This time-step is used in each control volume of the
values for initial turbulence are not tuned to match the experimen-
domain. A CFL number of 0.8 is used in the simulations. A complete
tal results and these values were also used in our previous (Sathiah
description is provided in Blazek (2005).
et al., 2012b, 2013) validation work.
The spatial discretization of flow, species and turbulence equa-
tions is performed using the second-order upwind numerical
3.4. Models and post-processing method scheme. The convective flux is calculated using the flux-difference
splitting (FDS) scheme by Roe (1981) and Roe (1986).
Since ENACCEF is a vertical facility, buoyancy effects can be
important. Hence, effects of buoyancy on turbulent kinetic energy 4. Results and discussion
and turbulent dissipation rate were considered in the simulation.
A simple ignition method is used with an ignition radius of 1 cm in As mentioned in the introduction, our current objective is to
the present analyses. As demonstrated in Sathiah et al. (2012a), a present the validation of the TFC combustion model combined
small increase in ignition radius results only in a small time shift with the AMR procedure against non-uniform hydrogen–air exper-
of the results. In addition, it is assumed in our simulations that iments. We selected the cases RUN 765 and 736 described in the
the counter-gradient transport has negligible effects on the results. previous section for this purpose.
Effects of stretching on flame propagation were not considered, In our previous papers (Sathiah et al., 2012a,b, 2013), the impor-
which means that the function G in Eq. (4) from Sathiah et al. tance of grid resolution on flame propagation was emphasized.
(2012b) is set to unity. However, preferential diffusion and com- Hence, as a first step, a grid sensitivity study has been performed.
pression effects (see Sathiah et al., 2012b) were considered in the
simulations.
The flame position obtained from the simulations was extracted
from the axial-coordinate corresponding to an iso-surface value of
c̃ = 0.5, as done by Gubba et al. (2009). This method is also used
in our previous Sathiah et al. (2012b) and Sathiah et al. (2013)
validation work.

3.5. Physical properties

The values of specific heat capacity, Lewis number, adiabatic


flame temperature, thermal diffusivity, density of unburned and
burned gas, and pressure and temperature coefficients for thermal
and molecular diffusivities were calculated using GASEQ (2011) and
CANTERA (2011). The overall thermo-kinetic index ˛ is extracted
from Babkin (2003) and HySafe (2007). The unburnt gas density, Fig. C.5. The variation of Lewis number Le with hydrogen mass fraction.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 7

turbulent flame acceleration, (c) turbulent flame deceleration and


(d) turbulent jet flame, similar as observed in Sathiah et al. (2013).
The times at which the flame reaches the first and last baffles and
dome entrance are also shown here. In the quasi-laminar flame
propagation phase (i.e. t = 0–0.03 s), the flame propagates slowly
because of low levels of turbulence. The flame velocity (defined as
the ratio of axial flame positions with time i.e. ds/dt) predicted by
the simulation in this phase is higher than in the experiments. The
possible causes for the overprediction could be (a) even lower ini-
tial turbulence levels in the experiments than we have assumed, (b)
overprediction of the turbulence levels by the standard k −  turbu-
lence model in the quasi-laminar regime, and (c) overprediction of
the turbulence source term by the applied TFC combustion model in
the quasi-laminar regime. This overprediction in the flame velocity
introduces a time shift between the simulation and the experi-
ments. Two additional possible reasons for the time shift are (a)
the application of a larger ignition radius in the computations, and
(b) a possible ignition delay time in the experiments. As explained
in Sathiah et al. (2012a) and Sathiah et al. (2012b), the application
of a bigger ignition radius results in a time shift. It will be shown
later that the considered time shift and the overprediction of the
turbulent flame speed do not affect the subsequent turbulent flame
acceleration phase and the corresponding pressure dynamics.
The turbulent flame acceleration phase starts when the flame
reaches the first baffle (t = 0.03 s) and ends when the flame arrives
at the last baffle in the acceleration tube (t = 0.038 s). The slope
of flame position versus time in the turbulent flame acceleration
phase increases exponentially because of the turbulence gener-
ated by the baffles. The mechanism causing the increased burning
rate is the wrinkling of the flame front by the turbulent eddies.
Thereby, the flame surface area is increased, which increases the
total reaction rate. The simulation very well predicts this phase,
and the slope of the flame position versus time compares well with
the experiments.
The third phase is the flame deceleration phase (time
t = 0.038–0.043 s). It starts when the flame reaches the last obstacle
and ends when the flame arrives at the dome entrance. The slope of
the axial flame position versus time decreases due to the absence
of the turbulence generation by the baffles. The slope of the axial
flame position versus time predicted by the simulation is in good
agreement with the experiments. The final phase is the turbulent
jet flame phase (time t = 0.043–0.085 s) where the flame expands
into the dome until complete combustion.
During the quasi-laminar and turbulent flame acceleration
phases, the rise in the pressure is negligible (see Fig. C.6b). Dur-
ing the flame acceleration, compression pressure waves detach
from the flame and travels to the dome entrance. At this dome
Fig. C.6. Results for RUN 765 (blockage ratio = 0.63). The variation of (a) the axial entrance the compression waves are reflected as expansion waves.
flame front position, (b) the pressure with time and the variation of (c) amplitude The reflected expansion waves create the sudden decrease in the
with frequency calculated using three different levels of AMR. intermediate peak pressure during the deceleration phase. The
value of the intermediate peak pressure is slightly overpredicted in
Next, the numerical results were compared against experiments the simulation (see Fig. C.6b). Moreover, the first peak in the pres-
for each case. sure is observed at 0.04 s in the simulation, while in the experiments
this is observed at 0.09 s. This time shift is due to the overpredic-
4.1. RUN 765 tion of the flame velocity in the quasi-laminar flame propagation
phase. As explained, two possible additional reasons for the time
Fig. C.6a illustrates the variation of the axial flame front position shift could be the application of the bigger ignition radius and a
with time calculated using three different levels of AMR. Practically possible ignition time delay in the experiments.
grid independent solutions were obtained using 2 levels of AMR. Next, the pressure increases quickly because of burning of the
The variation of the pressure with time results also shows simi- unburnt gases in the dome. The increase in the pressure with time
lar grid independence (see Fig. C.6b) for 2 levels of AMR. This grid i.e. the slope of pressure dp/dt, is captured very well in the simu-
independence was also confirmed by comparing the volume inte- lations. However, the maximum value of the mean peak pressure
gral source term of the progress variable equation. For brevity, the is slightly overpredicted, because heat loss effects are neglected as
corresponding results are not shown here. adiabatic boundary conditions were used in the simulations. Once
The following four different phases of flame propagation (see the maximum value of the mean pressure is reached, the pressure
Fig. C.6a) can be identified: (a) quasi-laminar flame propagation, (b) decreases with time in the experiment because of heat losses.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
8 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

Table C.2
Quantitative comparison of TFC for two different ENACCEF experiments.

Experiment TFC Difference

RUN 765
First peak pressure 1.783e05 Pa 2.094e05 Pa +17.4%
Maximum pressure 3.26e05 Pa 3.679e05 Pa +12.8%
First rate of pressure rise dp/dt 3.595e09 Pa/s 8.97e08 Pa/s −75%
Second rate of pressure rise 6.36e06 Pa/s 6.052e06 Pa/s −5.3%
dp/dt
First eigen-frequency 206.7 Hz 200.0 Hz −3.2%
Amplitude corresponding to 933.7 Pa 735.6 Pa −21.2%
first eigen freq.
Second eigen-frequency 410 Hz 400 Hz −2.4%
Amplitude corresponding to 195.4 Pa 1076 Pa −450.6%
second eigen freq.

RUN 736
First peak pressure 1.631e05 Pa 1.758e05 Pa +7.7%
First rate of pressure rise dp/dt 6.89e07 Pa/s 1.107e09 Pa/s +1506%
First eigen-frequency 113.3 Hz 112.5 Hz −0.7 %
Amplitude corresponding to 742 Pa 1396 Pa +88.1%
first eigen freq.
Second eigen-frequency 220 Hz 212.5 Hz −3.4%
Amplitude corresponding to 912.6 Pa 699.2 Pa −23.3%
second eigen freq.

When the pressure reaches the maximum value, significant


oscillations in the pressure are observed. Such oscillations can
be detrimental to the structural integrity of the safety systems
or the containment when the frequencies of the oscillation cor-
respond to the eigen frequencies of such structure. To analyze
the frequency spectrum of the pressure, Fast Fourier Transform
(FFT) analyses were performed. Fig. C.6c shows the variation of
the amplitude with the frequency obtained using the simulation
and experimental data. The first eigen frequency predicted by the
simulation compares very well with the experiment, although the
corresponding amplitude is underpredicted. The quantitative com-
parison of the first intermediate peak pressure, maximum value of
the mean pressure, first and second rate of pressure rise dp/dt, first
eigen frequency and its corresponding amplitude are presented in
Table C.2.
As mentioned in the previous section, much higher flame speeds
were obtained in non-uniform hydrogen–air mixture than in uni-
form hydrogen–air mixture (Whitehouse et al., 1996). This means
that hydrogen concentration gradient substantially effects the
flame dynamics. In order to study the affects of the hydrogen con-
centration gradient on the flame and pressure dynamics, the results
of uniform and non-uniform hydrogen–air mixtures are compared.
This comparative study is presented below.
Fig. C.7. Results for RUN 765 and RUN 153 (blockage ratio = 0.63). The variation of
4.1.1. Effects of the hydrogen concentration gradient on the flame (a) the axial flame front position and (b) the pressure as a function as a function of
and pressure dynamics time and (c) variation of amplitude with frequency calculated using two levels of
AMR.
The flame dynamics for uniform (RUN 153) and non-uniform
hydrogen–air mixtures (RUN 765) are shown in Fig. C.7a. Both
the simulation and experimental results are compared. As men- in the turbulent flame acceleration phase is higher in RUN 153.
tioned, RUN 153 corresponds to a uniform mixture with hydrogen We have no explanation of the trends observed in the experi-
concentration of 13 vol.%, while RUN 765 correspond to a non- ments.
uniform mixture with decreasing hydrogen concentration from Because of the higher flame velocity in RUN 153, the interme-
11.6 to 8.1 vol.%. It is well known from literature that an increase diate peak pressure is higher for RUN 153. This trend is correctly
in the hydrogen concentration leads to an increase in the laminar captured in the simulations and experiments (see Fig. C.7b).
flame speed. Since, the hydrogen concentration close to ignition In the dome, the hydrogen concentration in RUN 765 is 8.1 vol.%,
location is somewhat higher in RUN 153, the laminar flame speed while in RUN 153 the hydrogen concentration is 13 vol.%. As is well
is also higher by 5% in RUN 153. This essentially means that one known in literature, a decrease in hydrogen concentration leads to
can expect that the flame velocity (the slope of axial flame dis- a decrease in the adiabatic flame temperature and Adiabatic Iso-
tance versus time) in RUN 153 is higher than the flame velocity in choric Complete Combustion (AICC) pressure. This means that the
RUN 765. This trends is consistently observed in CFD simulations. maximum mean pressure reached in RUN 765 is lower than in RUN
In contrast, the flame velocity in the quasi laminar flame prop- 153. This trend can be observed both in the simulations and in the
agation phase is higher in RUN 765, whereas the flame velocity experiments.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 9

reasons for the overprediction are already explained in the previous


section.
The slope of the axial flame position versus time during the
turbulent flame acceleration phase is well predicted by the sim-
ulation in comparison to the experiments. The compression and
expansion pressure wave phenomena observed during the flame
deceleration phase create the intermediate peak pressure. The
maximum value of the intermediate peak pressure is slightly over-
predicted in the simulation. The second increase in the pressure
with time observed in the simulation is a result of the burning of
the unburnt gas present in the dome. Finally, the maximum value
of the mean pressure is reached in the simulation. This value cor-
responds to the AICC pressure for the considered gas mixture. In
contrast, this increase in the pressure is not observed in the exper-
iments. This is because of flame quenching which was observed in
the experiments when the flame reaches the dome. The simula-
tion clearly does not reproduce this behavior. This is due to lack
of a sub-model in the current CFD model to simulate this flame
quenching.
To simulate the flame quenching and to reproduce the flame
dynamics and pressure dynamics observed in the experiments, the
turbulent flame speed constant used in the TFC closure (see Eq. (3)
in Sathiah et al., 2012b) is artificially changed from 0.52 to 0 when
the flame reaches close to the dome i.e. at an axial location of 3.8 m.
Additional simulations were performed with this changed turbu-
lent flame speed constant in order to mimic the flame quenching.
So, the flame is artificially quenched when it reaches an axial
distance of 3.8 m. The results of the computation with this artificial
quenching are shown in Fig. C.9a–c.
In the experiments, pressure oscillations can be observed after
the intermediate peak pressure is reached. This behavior is also
Fig. C.8. Results for RUN 736 (blockage ratio = 0.63). The variation of (a) the axial
reproduced in the simulations. An FFT analyses is performed to
flame front position and (b) the pressure with time calculated using three different obtain the frequencies and amplitudes corresponding to the con-
levels of AMR. sidered pressure oscillations. The results of these FFT analyses are
shown in Fig. C.9c. The first, second and third eigen frequencies
are predicted well in the simulation. The values of the interme-
The pressure spectrum obtained using the simulations and
diate peak pressure, first rate of pressure rise, first and second
experiments for RUN 153 and RUN 765 are shown in Fig. C.7c. A
eigen frequencies and corresponding amplitudes are summarized
decrease in the hydrogen concentration leads to decrease in the
in Table C.2.
speed of sound. In turn, this leads to a decrease in the eigen fre-
quencies and acoustic impedance (see Eqs. (5) and (6) in Sathiah
4.3. Quantitative comparison
et al. (2013)). It can be observed both in the simulations as well as
the experiments that the value of the first eigen frequency is indeed In order to conclude about the predictive capability of our
smaller in RUN 765. Moreover, the corresponding amplitude of the combustion model for the present experiments, it is necessary to
pressure oscillation is also smaller in RUN 765 than in RUN 153. The make quantitative comparisons of the simulation and experimental
trend in the amplitude is observed both in the experiments and in results. For that purpose, we made a quantitative comparison of the
the simulations. following parameters (a) the first peak pressure, (b) the maximum
To summarize, the simulations consistently predict the trends of mean pressure, (c) the first rate of pressure rise (dp/dt), (d) the sec-
the decrease in the flame velocity, the intermediate peak pressure, ond rate of pressure rise (dp/dt), (e) the first eigen frequency, (f) the
and the maximum value of the mean pressure with decrease in second eigen frequency and (g) the amplitudes corresponding to
hydrogen concentration. In addition, the trends of decrease in the these frequencies. These parameters have been selected because of
values of the first eigen frequency and its amplitude with decrease their possible importance for structural integrity analyses. Table C.2
in hydrogen concentration is captured by the simulations. shows a detailed comparison of these parameters for the three dif-
ferent cases RUN 765 and 736. From this table, it can be concluded
4.2. RUN 736 that the intermediate peak pressure and maximum mean pressures
are predicted within an accuracy of 17 and 13%. The rate of pressure
Fig. C.8a and b shows the flame development and the pres- rise dp/dt is predicted in the within the range of 75%, and the first
sure development for three different levels of AMR. Similar as in and second eigen frequency of the pressure wave phenomena are
the case of RUN 765, practically grid independent solutions are predicted within 4%, while their corresponding amplitudes were
obtained using 2 levels of AMR. Similar to RUN 765, four dif- predicted in the range of 20–450%. Overall, it can be concluded that
ferent phases of flame propagation can be identified i.e. (a) a the current model predicts the first peak pressure, maximum mean
quasi-laminar flame propagation phase, (b) a turbulent flame accel- pressure and the first and second eigen frequencies very well. The
eration, (c) a turbulent flame deceleration and (d) a turbulent jet rate of the first and second rates of pressure rise and the amplitudes
flame in the simulation. The quasi-laminar flame propagation phase of the pressure oscillations are predicted less well. Further work on
is significantly overpredicted in the simulation in comparison to modeling flame quenching due to heat loss could be considered as
the experiments, similar as in the case of RUN 765. The possible a next step in the model development.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
10 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

The main conclusion are as follows:

• A quasi-laminar flame propagation phase takes place until the


flame reaches the first baffle. In this initial phase, the simula-
tions tend to overpredict the flame velocity in comparison to
the experiments. As explained in this paper, the possible rea-
sons for this could be (a) even lower initial turbulence levels in
the experiments that we have assumed, (b) overprediction of the
turbulence levels by the standard k −  turbulence model in the
quasi-laminar regime, and (c) overpredictions of the turbulence
source term by the applied TFC combustion model in the quasi-
laminar regime. Since the overprediction in this initial phase is
deemed to be of very little importance of the subsequent phase,
the effect of these possible reasons has therefore not been deter-
mined in this paper.
• In the acceleration phase, the flame speed increases rapidly with
time because of the generation of turbulence due to the inter-
action of the unburned gases with the baffles. This increase in
turbulence enhances the flame speed which results in signif-
icantly faster flame propagation. In RUN 765, the decrease in
hydrogen concentration that takes place during the flame prop-
agation results in a decrease in the flame speed compared to the
cases with uniform hydrogen concentration. This trend is cap-
tured well in the simulations.
• The simulations consistently predict the trend of a decrease in the
flame velocity, intermediate peak pressure, and the maximum
value of the mean pressure with the decrease in the hydrogen
concentration. This is due to a decrease in he laminar flame speed
with decrease in hydrogen concentration. Moreover, the simula-
tions consistently predict the trend of a decrease in the value of
eigen frequency and amplitude of the residual pressure waves
with a decrease in the hydrogen concentration. This is because
the decrease in hydrogen concentration leads to a decrease in
the acoustic impedance and speed of sound.
• For the considered tests, the maximum value of the mean
pressures and the intermediate peak pressures were predicted
respectively within 12 and 18% accuracy. The rate of pres-
sure rise dp/dt was typically underpredicted in the range of
75–90% for RUN 765. The first and second eigen frequencies of
the residual pressure wave phenomena were predicted within
4%. Therefore, it could be overall concluded that the presented
CFD modeling predicts the considered ENACCEF experiments
well.

6. Future steps
Fig. C.9. Results for RUN 736 (blockage ratio = 0.63). The variation of (a) the axial
flame front position and (b) the pressure as a function of time, and (c) the variation In order to apply our combustion modeling to real scale contain-
of the amplitude with frequency calculated using two levels AMR. ment scenarios, further validation must be performed. Within the
framework of the ISP-49 benchmark, hydrogen deflagration exper-
5. Summary and conclusions iments have been performed also in the THAI experimental facility
of Becker Technologies (Germany) (Kanzleiter and Langer, 2008;
The CFD based method described and validated in our previous ISP-49, 2011). The THAI facility is a cylinder with a height of 9.2 m
articles (Sathiah et al., 2012a,b, 2013) is further extended to sim- and a diameter of 3.2 m with no baffles. Hence, the flame is not
ulate turbulent flame propagation in non-uniform hydrogen–air accelerated by the presence of baffles as in the case of the ENACCEF
and hydrogen–air–steam mixtures. The method consist of solv- facility. As a result, experimental data for relatively slow deflagra-
ing the density-averaged Navier–Stokes equations together with tions have been obtained from the THAI facility. These data will be
the standard k −  model and an extended TFC combustion model. used for further validation of our combustion model and the results
A dynamic grid adaptation method is used to capture the flame will be presented in a follow-up paper.
front propagation. ENACCEF tests 765 and 736 with two different
hydrogen concentration gradients have been used for the validation
purposes in this paper. For each test, mesh and time step sensitiv- Acknowledgement
ity analyses have been performed in order to demonstrate that the
numerical errors can be practically neglected. Grid independence The work described in this paper is funded by the Dutch Ministry
is observed for two levels of AMR for the two considered tests. of Economic Affairs.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 11

Appendix A. A Priori Analysis of Effect of Mixture Fraction


Fluctuations on Flame Propagation

In order to mathematically analyze the effect of the mixture frac-


tion fluctuations on the flame propagation, we need an approach
to estimate the mixture fraction fluctuations. The approach that
we used is based on the assumption of equilibrium between the
production and dissipation of the mixture fraction variance (see
Poinsot and Veynante, 2001). For that situation, Eq. (13) can be
simplified to:
 2
∂
f  2
Cg t = Cd ¯ f  . (A.1)
∂xi k

The mixture fraction variance can be evaluated as follows:


 2
2 Cg k  ∂
f Fig. C.10. % relative fluctuation in the laminar flame speed Sl versus % relative
f  = t . (A.2) fluctuation in the mixture fraction f . The lines correspond to the different values of
Cd ¯  ∂xi the mean mixture fraction f̃ specified in the legend.

Using Eq. (A.2), we can estimate the mixture fraction variance


2
f  from the mixture fraction gradient. The hydrogen concentra-
The figure demonstrates that a 10% relative fluctuation in the mix-
ture fraction leads to a 10% relative fluctuation in the laminar flame
tion gradients for RUN 765, 736 and 733 are in shown in Fig. C.2.
speed. This means that the laminar flame speed is fairly sensitive
For RUN 765, for example, a relative mixture fraction fluctuation
 to mixture fraction fluctuations. Similarly, sensitivity analyses of
˜2
f  /f̃ of 6.13% is obtained. The values of turbulent kinetic energy mixture fraction fluctuations on unburnt gas density and unburnt
and turbulent dissipation rate required for a priori estimation of thermal diffusivity have been performed (not shown here). In order
mixture fraction fluctuations i.e. Eq. (A.2) were taken to be initial to investigate the effect of mixture fraction fluctuations on the
values as mentioned in Section 3.3. It is worth mentioning that the source term of the progress variable, the source term can be written
values of turbulent kinetic energy and turbulent dissipation rate as follows:
are much higher than their initial values because of generation of ω(f ) ∝ u (f )Sl (f )
0.5
u (f )
−0.25
. (A.6)
turbulence due the interaction of flame with obstacles.
Now in order to investigate the effects of the mixture fraction Fig. C.11 shows the relative difference in the source term ω
fluctuations on the source term in the progress variable equation, (defined in Eq. (A.7)) as a function of the % relative difference in
we need to investigate its effects on the laminar flame speed Sl , the mixture fraction.
the unburnt gas density u , and the unburnt thermal diffusivity ω(f ) − ω(f̃ )
u . For this purpose, these properties are fitted into a 6th order ω = (A.7)
ω(f̃ )
polynomial as a function of the mass fraction of hydrogen using
Matlab (MATLAB, 2007). For example, the 6th order polynomial of The figure clearly demonstrates that a 10% relative fluctuation
the laminar flame speed is given by: in the mixture fraction leads to a 0.1% relative fluctuation in the
source term, while a 30% relative fluctuation in mixture fraction
Sl = a0 + a1 f + a2 f 2 + a3 f 3 + a4 f 4 + a5 f 5 + a6 f 6 (A.3)
leads to a 0.2% fluctuations in the source term. This essentially
where, a0 , a1 , a2 , a3 , a4 , a5 and a6 are the polynomial coefficients. means that the mixture fraction fluctuations have no significant
Similarly, we can describe the unburnt gas density and the thermal effect on the source terms, while its effect on the laminar flame
diffusivity using 6th order polynomials. To fit the above polynomi- speed was significant.
als, the laminar flame speed is obtained using a correlation e.g. Liu Hence, it can be concluded that the mixture fraction fluctua-
and MacFarlane (1983). The data for fitting the unburnt gas density tions have no significant effect in the ENACCEF simulations, and
and thermal diffusivity were calculated using the CANTERA (2011) the properties based on the Favre-averaged mean mixture fraction
code. can therefore be used in the simulations. This means that mixture
Now, let us assume a fluctuation f in the mixture fraction. Then fraction fluctuations can be practically neglected, and the PDF can
the laminar flame speed at f = f̃ + f  is given as follows (using
Taylor series expansion, Abramowitz and Stegun, 1972):
2
dS l f  d2 Sl
Sl (f ) = Sl (f̃ ) + f  + + ··· (A.4)
df 2 df 2

where, dSl /df and d2 Sl /df2 are the first and second derivatives of the
laminar flame speed with respect to the mixture fraction. For sim-
plicity, the higher order terms are neglected. The first and second
derivatives of the laminar flame speed required in Eq. (A.4) can be
obtained by differentiating Eq. (A.3) with respect to f.
Fig. C.10 shows the variation of the % relative fluctuation in the
laminar flame speed Sl (defined in Eq. (A.5)) with relative fluctu-
ation in the mixture fraction f (defined in Eq. (A.5)). The results
correspond to different values of the mean mixture fraction f̃ .
1/2 1/2
Sl (f ) − Sl (f̃ ) f − f̃ f  Fig. C.11. % relative fluctuation in source term ω versus relative fluctuation in the
Sl = 1/2
and f = = (A.5) mixture fraction f . The lines correspond to different values of mean mixture fraction
Sl (f̃ ) f̃ f̃
f̃ as specified in the legend.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
12 P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx

assumed to be a single delta function. It is worth reminding that To determine the averaged laminar flame speed equations for the
this conclusion holds for the considered ENACCEF cases. For other Favre averaged steam mixture fraction 
g and variance of the steam
cases, similar a priori analysis should be performed in order to esti- mixture fraction, g 2 , subsequently Eqs. (B.4) and (B.5), are added to
mate the effect of the mixture fraction fluctuations on the source the model in addition to Eqs. (12) and (13) for the hydrogen mixture
term. Alternatively, the code can be further developed to take the fraction. The Eqs. (B.4) and (B.5) are given as follows:
fluctuations in the mixture fraction into account.    
∂ ∂ ∂ ∂
g ∂ ∂
g
(¯ 
g) + (¯ ũi
g) = D
¯ s + D
¯ s,t (B.4)
Appendix B. Extension of the TFC combustion model to ∂t ∂xi ∂xi ∂xi ∂xi ∂xi
hydrogen–air–steam mixtures
   
∂  ¯ g  2 ∂ ¯
(¯ ũi g
∂ ∂ ∂ ∂g̃
When simulating combustion in real accident scenarios, the (¯ g  2 ) +  2 ) = Ds + 2g  Ds
model must also include the effect of the presence of steam on
∂t ∂xi ∂xi ∂xi ∂xi ∂xi
the hydrogen–air flame propagation. The addition of steam in the    2
∂g
 2 ∂
− Cd ¯ g
hydrogen–air mixture can have a significant effect on the flame ∂ g 
 2 ,
+ D
¯ s,t + Cg t (B.5)
and pressure dynamics. In particular, steam addition can influence ∂xi ∂xi ∂xi k
the lower and upper flammability limits and also affect the lam-
inar flame speed of hydrogen–air mixtures. Recently, within the where Ds is the laminar diffusion coefficient of the steam mixture
framework of the OECD-THAI project (Kanzleiter et al., 2010), the fraction and Ds,t = t / s,t is the turbulent diffusion coefficient. The
importance of the addition of steam on combustion of hydrogen–air default values of the constants  s,t , Cg and Cd are 0.85, 2.86, and 2.0
mixtures is investigated. Because of its importance, hydrogen and respectively. When the influence of fluctuations in the steam mix-
steam distribution analyses were performed in a large containment ture fraction is neglected, the PDFs (g) of the steam mixture fraction
(Jian and Xuewu, 2007; Houkema et al., 2003; Kim et al., 2004) as is assumed to be a single delta function and the equations for the
a first step prior to hydrogen combustion analysis using CFD. To variance of the hydrogen mixture fraction and steam mixture frac-
conclude, it is important to consider the effects of steam on the tion do not have to be solved. In a similar way, the mean value
hydrogen–air mixtures for real scale accident scenarios. Therefore, of thermal diffusivity, u,0 , the unburnt gas density, u,0 the Lewis
an extension of the method presented in Section 3.2 is presented number, Le and the molecular diffusivity, D can be obtained.
in the appendix for non-homogeneous hydrogen–air–steam mix- To summarize, in addition to the standard conservation equa-
tures. tions for mass, momentum, energy, and the progress variable
Steam is considered as a third component in the unburnt (Sathiah et al., 2012a), Eqs. (12) and (B.4) must be solved for non-
mixture, next to hydrogen and air. The steam mass fraction is uniform hydrogen–air–steam mixtures. It is worth mentioning that
characterized by an additional steam mixture fraction. The instan- the extended model can also be used for diluents other than steam,
taneous local composition of the unburnt mixture is given now as for example CO2 –He.
a function of the steam mixture fraction g, and the hydrogen mix-
ture fraction f. For each composition, the laminar flame speed is Appendix C. Extension of the TFC combustion model to
described as a function of the hydrogen and steam concentrations hydrogen–air–steam mixtures to account for the presence
which can be obtained from the laminar flame speed correlation of PARs
described in Eq. (10) in Sathiah et al. (2013). The local mean value of
the laminar flame speed now is given by averaging using the joint PARs are widely used as a mitigation measure to reduce the risk
probability density function (JPDF) of the steam mixture fraction of possible hydrogen deflagrations. When PARs are under opera-
and hydrogen mixture fraction: tion, hydrogen and oxygen are converted into water/steam, thereby
 1 1
reducing the local hydrogen and oxygen concentrations. Because
of local consumption of oxygen, the ratio of the mole fraction of
S̃l,0 = Sl,0 (f, g)JPDF(f, g)dfdg. (B.1)
nitrogen and the mole fraction of oxygen is locally different than
0 0
the ratio present in air. When simulating hydrogen deflagrations in
Here, JPDF(f, g) is the joint PDF of the hydrogen mixture fraction such scenarios, the model must therefore also include the effect of
and steam mixture fraction. Assuming that the fluctuations in stoi- the different ratios of nitrogen and oxygen mole fraction on the
chiometry (first mixture fraction) and fluctuations in steam fraction hydrogen–air–steam flame propagation. Therefore, an extension
are statistically independent, similar as done by Riechelmann and of the method presented in Section B is presented to account for
Uchida (2010), the joint PDF of the two mixture fractions equals the different ratios of nitrogen and oxygen mole fraction.
product of the marginal PDF’s. That is, the PDFf (f) for the hydrogen Nitrogen is considered as a fourth component in the unburnt
mixture fraction and PDFs (g) for steam mixture fraction. That is, mixture, next to hydrogen, oxygen and steam. The nitrogen mass
statistical independence between the two mixture fractions f and g fraction is characterized by an additional nitrogen mixture frac-
has been assumed for the present purposes. Then, Eq. (B.1) can be tion. The instantaneous local composition of the unburnt mixture
rewritten as follows is given now as a function of the nitrogen mixture fraction. For each
 1 1 composition, the laminar flame speed is described as a function of
S̃l,0 = Sl,0 (f, g)PDF f (f )PDF s (g)dfdg. (B.2) the hydrogen, steam, and nitrogen concentration. The local mean
0 0
value of the laminar flame speed is given now by the JPDF of the
To close the model, the shape of the PDF of the steam mixture steam mixture fraction, hydrogen mixture fraction, and nitrogen
fraction must be assumed to be of a specific mathematical form mixture fraction. That is:
(double delta function or a ˇ-function (Abramowitz and Stegun,  1 1 1
1972)) characterized by the mean  g and the variance g 2 . Then, S̃l,0 = Sl,0 (f, g, h)JPDF(f, g, h)dfdgdh. (C.1)
the mean local laminar flame speed is given as a function of the 0 0 0

mean and the variance of the hydrogen mixture fraction and steam Here JPDF(f, g, h) is the joint PDF of the hydrogen mixture fraction,
mixture fraction as follows: steam mixture fraction, and nitrogen mixture fraction. Assuming
2 that the fluctuations in stoichiometry (first mixture fraction) and
S̃l,0 = S̃l,0 ( g , g
f , f  ,   2 ). (B.3) the fluctuations in steam fraction and nitrogen mixture fraction

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030
G Model
NED-8343; No. of Pages 13 ARTICLE IN PRESS
P. Sathiah et al. / Nuclear Engineering and Design xxx (2015) xxx–xxx 13

are statistically independent, the joint PDF of the three mixture ISP-49, 2011. International Standard Problem ISP-49 on containment thermal
fractions equals the product of the marginal PDF’s, that is, PDFf (f) hydraulics. Tech. rep., Nuclear Energy Agency.
Jian, D., Xuewu, C., 2007. Hydrogen and steam distribution following a small-break
for hydrogen mixture fraction, PDFs (g) for steam mixture fraction, LOCA in large dry constainment. Nucl. Sci. Techn. 18 (3), 181–185.
and PDFn (h) for nitrogen mixture fraction. Kanzleiter, T., Gupta, S., Fischer, K., Ahrens, G., Langer, G., Kühnel, A., Poss, G., Lan-
Then Eq. (B.1) can be rewritten as follows grock, G., Funke, F., 2010. Hydrogen and fission product issues relevant for
containment safety assessment under severe accident conditions. Tech. rep.

S̃l,0 = S̃l,0 (
2 g , g
f , f  ,   2 ,  2
h, h ). (C.2)
Becker Technologies GmbH, Eschborn, Germany.
Kanzleiter, T., Langer, G., 2008. Quick look report for hydrogen deflagration tests hd-
15, hd-16, hd-17, hd-18, hd-19, hd-20 and hd-24. Tech. rep. Becker Technologies
To determine the Favre averaged nitrogen mixture fraction h̃ and GmbH, Eschborn, Germany.
Karpov, V., Lipatnikov, A., 1995. Premixed turbulent combustion and some thermod-
2
variance of the nitrogen mixture fraction, h , equations similar to iffusional effects in lamianr flames. In: Combustion, Detonation, Shock Waves,
Proceedings of the Zeldovich Memoeirla Moscow 1, pp. 168–180.
Eqs. (B.4) and (B.5), are added to the model. When the influence
Karpov, V., Lipatnikov, A., Zimont, V., 1997. Flame curvature as determinant of pref-
of the fluctuations in the hydrogen mixture fraction, steam mix- erential diffusion effects in premixed turbulent combustion. Prog. Astrnaut.
ture fraction, and nitrogen mixture fractions are neglected, then Aeronaut. 173, 235–250.
the mixture fractions are assumed to be a single delta function. As Karpov, V., Severin, E., 1980. The influence of molecular transfer process on turbulent
combustion velocity. Combust. Explos. Shock Waves 16, 41.
a result, the equations for the variance of the hydrogen mixture Kido, H., Huang, S., Taoue, K., Nitta, T., 1994. Improvment of lean hydrocarbon mix-
fraction, steam mixture fraction, and nitrogen mixture fraction do tures combustion performance by hydrogen addition and its mechanism. In:
not have to be solved. Then the laminar flame speed in Eq. (C.2) is International Symposium COMODIA, vol. 95, pp. 119–124.
Kido, H., Kitagawa, T., Nakashima, K., Nato, K., 1989. An Improved Model of Turbu-
given by lent Burning Velocity, vol. 49. Memoirs of the Faculty of Engineering, Kyushu
University, pp. 229–247.
S̃l,0 = S̃l,0 ( g, 
f , h). (C.3) Kim, J., Haong, S., Kim, S., Kim, H., 2004. 3-Dimensional analysis of the steam-
hydrogen behaviour from a small break loss of the coolant accident in the
In a similar way, the mean value of thermal diffusivity, u,0 , the APR1400 containment. J. Korean Nucl. Soc. 36 (1), 24–35.
Koroll, G., Kumar, R., Bowles, E., 1993. Burning velocities of hydrogen–air mixtures.
unburnt gas density, u,0 the Lewis number, Le and the molecular
Combust. Flame 94 (3), 330–340.
diffusivity, D can be obtained. Kuznetsov, V., Sabelnikov, V., 1990. Turbulence and Combustion. Wiley.
To summarize, in addition to the standard conservation equa- Kwon, S., Wu, M., Dricoll, J.F., Faeth, G.M., 1992. Flame surface properties of premixed
flames in isotropic turbulence: measurements and numerical simulations. Com-
tions for mass, momentum, energy, and the progress variable
bust. Flames 88, 221–238.
(Sathiah et al., 2012a)), Eqs. (12) and (B.4) and a similar equation Launder, B., Spalding, D., 1972. Mathematical Models of Turbulence, vol. 1972. Lon-
for the mean nitrogen mixture fraction must be solved. don, Academic Press.
Lipatnikov, A., Chomiak, J., 1997. A simple model of unsteady turbulent flame prop-
agation. J. Eng. SAE Trans. 106 (3), 2441–2452.
References Lipatnikov, A., Chomiak, J., 2002. Turbulent burning velocity and speed of developing
curved and strained flames. Proc. Combust. Inst. 29, 2113–2121.
Abramowitz, M., Stegun, I.A. (Eds.), 1972. Handbook of Mathematical Functions: Lipatnikov, A., Chomiak, J., 2005. Molecular transport effects on turbulent flame
With Formulas, Graphs, and Mathematical Tables (Paperback). Dover Publica- propagation and structure. Prog. Energy Combust. Sci. 31, 1–73.
tions, Inc., 31 East 2nd Street, Mineola, New York, p. 11501. Liu, D., MacFarlane, R., 1983. Laminar burning velocities of hydrogen–air and
ANSYS CFX, 2007. ANSYS, 2007, ANSYS CFX-10.0 User Manual. ANSYS Inc. hydrogen–air–steam flames. Combust. Flame 49 (1-3), 59–71.
ANSYS FLUENT, 2008. Fluent 12.0. ANSYS-Fluent Inc., Lebanon, NH. MATLAB, 2007. http://www.mathworks.com/products/matlab/
Babkin, V., 2003. Private Communication. Institute of Chemical Kinetics and Com- Poinsot, T., Veynante, D., 2001. Theoretical and Numerical Combustion.
bustion, Siberian Branch, Russian Academy of Science, Novosibirsk, Russia. Polifke, W., Flohr, P., Brandt, M., 2002. Modeling of inhomogeneously premixed
Betev, A., Karpov, V., Lipatnikov, A., Vardosanidze, Z., 1995. Hydrogen combustion in combustion with an extended TFC model. J. Eng. Gas Turbines Power 124 (1),
engines and preferential diffusion effects in laminar and turbulent flames. Arch. 58–65.
Combust. 15 (3–4), 199–227. Riechelmann, D., Uchida, M., 2010. Three-stream flamelet model for industrial appli-
Blazek, J., 2005. Computational Fluid Dynamics: Principles and Applications. Elsevier cations. J. Eng. Gas Turbines Power 32 (6), 1–8.
Science Ltd. Roe, P., 1986. Characteristic based schemes for the Euler equations. Ann. Rev. Fluid
Bleyer, A., Taveau, J., Djebaïli-Chaumeix, N., Paillard, C., Bentaib, A., 2012. Com- Mech. 18, 337–365.
parison between FLACS explosion simulations and experiments conducted in Roe, P.L., 1981. Approximate Riemann solvers, parameter vectors, and difference
a PWR steam generator casemate scale down with hydrogen gradients. Nucl. schemes. J. Comput. Phys. 43, 357–372.
Eng. Design 245, 189–196. Sathiah, P., Komen, E., Roekaerts, D., 2012a. The role of CFD combustion modeling in
Bray, K., 1987. Methods of Including Realistic Chemical Reactions Mechanism in Tur- hydrogen safety management – I: Validation based on small scale experiments.
bulent Combustion Model. Complex Chemical Reaction Systems. Mathematical Nucl. Eng. Design 248, 93–107.
Modelling and Simulations. Springer, Heidelberg, pp. 356–375. Sathiah, P., vanHaren, S., Komen, E., Roekaerts, D., 2012b. The role of CFD combustion
CANTERA, 2011. Objected-Oriented Software for Reacting Flows. http://www. modeling in hydrogen safety management-II: Validation based on large scale
cantera.org/index.html/,. experiments. Nuclear Engineering and Design 252, 289–302.
Catlin, C., Fairweather, M., Ibrahim, S., 1995. Predictions of turbulent, premixed flame Sathiah, P., Komen, E., Roekaerts, D., 2013. The role of CFD combustion model-
propagation in explosion tubes. Combust. Flame 102 (1), 115–128. ing in hydrogen safety management – III: Validation based on homogeneous
Chaumeix, N., Bentaib, A., 2010. SARNET H2 combustion benchmark. Tech. rep., hydrogen–air–diluent experiments. Nucl. Eng. Design.
Bureau de Physique des Accidents Graves, IRSN. TMI, 2011. Three Mile Island Nuclear Generating Station. http://en.wikipedia.org/
Chaumeix, N., Bentaib, A., 2011. ISP 49 – Specification of ENACCEF Test Flame Prop- wiki/ThreeMileIslandNuclearGeneratingStation
agation in a Hydrogen Gradient. Whitehouse, D., Greig, D., Koroll, G., 1996. Combustion of stratified hydrogen–air
Durand, L., Ph.D. thesis 2007. Development, Implementation and Validation of LES mixtures in the combustion test facility cylinder. Nucl. Eng. Design 166,
Models for Inhomogeneously Premixed Turbulent Combustion. Technische Uni- 453–462.
versität München. Wu, M., Kwon, A., Driscoll, G., Faeth, G., 1991. Preferential diffusion effects on the
Fox, R.O., Raman, V., 2004. A multienvironment conditional probability density func- surface of turbulent premixed hydrogen/air flames. Combust. Sci. Technol. 78,
tion model for turbulent reacting flows. Phys. Fluids 16 (12), 4551–4565. 69–96.
Fukushima, 2011. Fukushima daiichi nuclear disaster. http://en.wikipedia.org/wiki/ Zeldovich, Y., Barenblatt, G., Librovich, V., Makhviladze, G., 1985. Mathematical The-
FukushimaDaiichinucleardisaster ory of Combustion and Explosions. Plenum Press, NY and London.
GASEQ, 2011. A Chemical Equilibirum Program for Windows. http://www.c.morley. Zimont, V.L., 1979. Theory of turbulent combustion of a homogenous fuel mixture
dsl.pipex.com/ at high Reynolds number. Combust. Explos. Shock Waves 15, 305–311.
Gubba, S.R., Ibrahim, S., Malalasekera, W., Masri, A., 2009. An assessment of large Zimont, V.L., 2000. Gas premixed combustion at high turbulence. turbulent flame
eddy simulations of premixed flames propagating past repeated obstacles. Com- closure combustion model. Exp. Thermal Fluid Sci. 21 (1–3), 179–186.
bust. Theory Model. 13 (3), 513–540. Zimont, V.L., Lipatnikov, A., 1993. To computations of heat release rate in turbulent
Houkema, M., Komen, E., Siccama, N., Willemsen, S., 2003. CFD analyses of steam flames. Dolady Phys. Chem. 332, 592–594.
and hydrogen distribution in a. nuclear power plant. In: The 10th International Zimont, V.L., Lipatnikov, A., 1995. A numerical model of premixed turbulent com-
Topical Meeting on Nuclear Reactor Thermal Hydraulics (NURETH-10), October bustion. Chem. Phys. Rep. 14, 993–1025.
5–9, Seoul, Korea. Zimont, V.L., Polifke, W., Bettelini, M., Weisenstein, W., 1998. An efficient compu-
Huld, T., Wilkening, H., 2001. 3D simulations of turbulent deflagrations using tational model for premixed turbulent combustion at high Reynolds numbers
dynamic grid adaptation. J. Appl. Math. Mech. 81 (S3), 537–538. based on a turbulent flame speed closure. J. Eng. Gas Turbines Power 120 (3),
HySafe, 2007. Biennial report on hydrogen safety. Tech. rep., European NoE HySafe. 526–532.

Please cite this article in press as: Sathiah, P., et al., The role of CFD combustion modeling in hydrogen safety management – IV: Validation
based on non-homogeneous hydrogen–air experiments. Nucl. Eng. Des. (2015), http://dx.doi.org/10.1016/j.nucengdes.2015.05.030

You might also like