Multidimensional Simulation of Hydrogen
Distribution and Turbulent Combustion in Severe
U. Bielert, W. Breitung, A. Kotchourko, P. Royl, W. Scholtyssek, A. Veser
A. Beccantini, F. Dabbene, H. Paillere
, E. Studer
T. Huld, H. Wilkening
, B. Edlinger, C. Poruba
, M. Mohaved
Forschungszentrum Karlsruhe, DE
Commissariat a l'Energie Atomique, Saclay, FR
Institut de Protection et Surete Nucleaire, FR
JRC Ispra, IT
Technical Univ. Munich, DE
Siemens AG, DE
A joint research project was carried out in the EU 4
Framework Programme with
the goal to develop verified and commonly agreed physical and numerical models
for the analysis of hydrogen distribution, turbulent combustion and mitigation,
which are suited for the multidimensional CFD codes that exist at the institutions
of the participating partners. The work programme and the activities of the
partners are described. Significant progress has been made in the areas of
experiments, model development, model verification and model application to
nuclear power plants. Joint benchmark tests were defined and analysed. In
addition to the regular partners meetings, topical workshops were conducted to
harmonize the experimental and theoretical work. The paper presents major
experimental results, gives examples for the achieved multi-dimensional
modelling capabilities, describes implementation into and validation of codes, and
presents results of plant analysis activities.
Various risk studies have shown that, in the case of a severe accident in a nuclear power plant,
hydrogen combustion could lead to early containment failure and release of radioactive
material to the environment. It is therefore proposed to install mitigating systems like
recombiners or igniters in reactor containments in order to keep hydrogen concentrations
below critical levels. The design and assessment of such systems needs the detailed
simulation with high spatial resolution of the major physical processes involved including
hydrogen source terms, distribution, ignition, and combustion, which can range from slow
deflagration to fully developed detonations. Examples are H
-stratification, plume behaviour
of released gases, local steam condensation, flame acceleration, quenching, and local
detonation risk. High spatial resolution also is needed for the design of efficient hydrogen
mitigation systems, because their function depends on local gas conditions. The present
project deals with the development and verification of physical and numerical models that can
be used in the multidimensional CFD codes currently under development at partner
organizations. In these codes the general equations of reactive fluid dynamics are solved,
including all important terms and using accepted methods for closing the equations. The
objective of the joint research programme, conducted under EU 4
, was to develop CFD-models for hydrogen distribution, turbulent combustion, and
hydrogen mitigation techniques, which can be applied to risk reduction in current and future
nuclear power plants. The work programme consisted, firstly, of an experimental part with
tests suitable to provide a data base, secondly, of modeling and validation work, and thirdly,
of application of the validated numerical tools to full-scale demonstration cases.
To provide a data basis for code development numerous experiments have been carried out on
different scaled test facilities within this project. The experiments were designed to provide a
maximum of local and global information on turbulent combustion parameters. One of the
main goals was to reach highly prototypical conditions. This was achieved by selecting
appropriate geometries, by working on different length scales up to reactor scale, and by
choosing accident relevant gas mixtures and conditions.
In two small scale facilities ("Glastube" and "PhD-Tube") at the Lehrstuhl A für
Thermodynamik, Technical University of Munich (TUM), experiments provided data over a
wide range of flame regime from the laminar reaction phase after ignition, the interaction of
the flame front with a single obstacle and the acceleration of the flame by different multiple
obstacle configurations up to detonation velocities. The possible combustion regimes together
with example Schlieren images are depicted in Figure 1. The two detonation tubes were
equipped with conventional probes for pressure and light, as well as with advanced laser
optical instruments, which allow a deeper insight into the physical phenomena of flame
acceleration and flame propagation.
A medium scale test facility ("12m Tube") at the Forschungzentrum Karlsruhe (FZK)
provided data on turbulent combustion in the range of medium fast turbulent deflagrations up
to detonation velocity. Light and pressure measurements were done at different locations in
the tube enabling to quantify wave and flame speed and flame acceleration from slow to fast
fast turbulent combustion regimes. In inert tests, the conditions after a propagating shock
wave through an obstacle path were also investigated. In addition, results of large-scale tests
performed in the RUT test facility at the Kurchatov Institute (KI) have been used for model
In the experiments the composition of the H
-steam-air mixtures was varied. In agreement
with the needs in reactor applications the experiments were focused on lean hydrogen-air
mixtures with and without steam. The variation of obstacle configurations allowed control of
the turbulence parameters. Blockage ratios between 0.1 and 0.9 were employed. The smallest
obstacles can be considered as well defined wall surface roughness. Vortices are shed from
the edges of the obstacles but the mean flow is not strongly modified by the obstacles. In this
case flow and combustion are controlled by turbulence generation. Very large obstacles (or
high blockage ratios) primarily limit the mass flow rates through the tube. In this case, high
velocities and turbulence levels are produced near the orifices and large recirculation areas
exist in the corners behind the obstacles. A flame propagates in this situation very fast in a
centre region, and pockets of unburned gas remain behind the leading flame where they burn
rather slowly. Here the combustion process is dominated by the fluid dynamics at the orifices.
For intermediate blockage ratios both effects are equally important.
From experimental work it is concluded:
Incomplete combustion is not possible for dry mixtures with more than 10.5 vol-%
, independent of scale, obstacle configuration and initial pressure.
concentrations lower than 10.5 vol-% a lower initial pressure increases the
probability of complete combustion (flame reaches the end of the tube)
Results depend weakly on obstacle spacing.
For the mixtures under investigation (H
) three phases in the combustion process can be distinguished. A slow acceleration phase,
characterised by a final flame velocity v
< 250m/s, is followed by a fast acceleration phase
> 600m/s. Later the flame propagates with a constant flame velocity in the tubes with
III MODEL DEVELOPMENT AND CODE IMPLEMENTATION
The numerical codes involved in this project are: COM3D  and GASFLOW  at FZK,
REACFLOW at Joint Research Centre Ispra (JRC), TONUS at Commissariat a l'Energie
Atomique (CEA) and models implemented in CFX4.2 at TUM. The only pure distribution
code in this program is GASFLOW. The others are mainly dealing with combustion, which
got the focus of comparison activities on model development and code implementation.
The codes have fully compressible flow solver (FCFS). TONUS has in addition a dedicated
low Mach number flow solver (LMNFS) for slow combustion. The conservation equations
used in the models are classical Favre-averaged Navier-Stokes equations for compressible and
reactive mixtures. Small differences exist in the hydrodynamic equations, e.g. pressure work
and turbulent dissipation in the energy equation. In TONUS low Mach number flow solver, a
Poisson equation for the fluctuating pressure replaces the mass equation. Classical k -
turbulence models have been implemented with buoyancy effect (CFX and TONUS) and
mean pressure gradient effect (COM3D).
For turbulent combustion, COM3D and TONUS use the Eddy-Break-Up (EBU) model,
REACFLOW uses the Eddy Dissipation (EDC) model. EBU or EDC "constants" have been
calculated via Said and Borghi correlations , and transition regimes are often expressed in
terms of Damköhler number, which is the ratio of chemical to turbulence time scale.
A more advanced combustion model, a presumed Probability Density Function (PDF) method
, has been implemented in the code developed by TUM. A clipped gaussian function has
been chosen for the PDF, and transport equations have been solved for the mean progress
variable c and the second central moment of the fluctuating c". To reduce computational
effort, tables have been built to calculate the mean reaction rate.
Different strategies have been chosen for model implementation and numerical schemes.
Structured (COM3D), Block Structured body fitted (CFX) or unstructured (TONUS and
REACFLOW) grids are implemented. Unstructured grids allow better geometrical
representation of the object, but higher numerical diffusion and calculation expenses have to
be taken into account. Mesh refinement has been developed in the REACFLOW code. In
order to perform 3D calculation with large geometrical length scale, low diffusive schemes
appear to be essential. Convective terms usually represent the main difficulty. As spurious
oscillations can occur if space central second order discretisation is used, upwinding is
necessary. To limit artificial viscosity, several techniques can be applied. A pressure-based
approach is used in TONUS LMNFS (finite element) and CFX (finite volume). On the other
hand, methods coming from the resolution of compressible inviscid Navier-Stokes equations
(Euler equations) are used in COM3D, REACFLOW and TONUS (FCFS) codes.
V CODE AND MODEL VERIFICATION
The codes were verified in a two-step approach. Firstly, each code was tested against standard
test cases and against different experiments. Secondly, the codes were used to calculate a
common set of experiments. These benchmark calculations allowed a direct comparison of the
different numerical models and implementations.
The low Mach number flow solver of TONUS was tested on inert problems including sine-
shaped bump, lid-driven cavity and a heated cavity. For faster flows different compressible
flow solvers were evaluated on several shock reflection problems. The solvers were also
applied to a supersonic flow at a forward facing step. Calculations for reactive flow were
performed for a square box and for experiments in the TUM glass tube.
The verification process of COM3D from FZK included a comparison of the thermodynamic
data with STANJAN  results, calculation of a forward facing step problem, comparison
with Schlieren pictures from a combustion test carried out in a small scale tube at KI, and
calculations of explosion tests in the large scale RUT facility. Also, many tests carried out in
the 12m tube at FZK were simulated during the verification and calibration process.
Verification of REACFLOW from JRC included the forward facing step problem, a shock
diffraction problem over a backward facing step, and calculations of RUT experiments.
The code developed at TUM uses a more elaborated combustion model than the previously
mentioned computer codes. Because of the high computational demands of this model only a
small number of calculations of tests, which were carried out in the PHD tube at TUM and in
the 12m tube at FZK, were performed. The results from this code are very promising.
The following tests were used in benchmark calculations.
Experiment Description Participants
FZK 12m tube R1096-03 Inert shock test, BR = 0,3 FZK, JRC
TUM PHD-tube i16
TUM PHD-tube i15
12,94 % H
, BR = 0,6
14,89 % H
, BR = 0,6
FZK 12m tube R0498-07
FZK 12m tube R0498-02
FZK 12m tube R0498-05
10 % H
, BR = 0,3
11 % H
, BR = 0,3
12 % H
, BR = 0,3
FZK, JRC, TUM
The first test case was the propagation and reflection of an inert shock wave. REACFLOW
and COM3D results for different turbulence models were compared to experimental data.
Both codes reproduced the incident and the reflected shock wave very well. The influence of
the different turbulence models was found to be small. Therefore it can be concluded that the
influence of turbulence is small in this configuration and that for such situations the applied k-
-models were sufficiently accurate.
PHD tube experiments for two different mixtures were calculated by REACFLOW, COM3D
and using the PDF model from TUM. REACFLOW and COM3D both reproduced the major
pressure waves in the combustion tube and the correct flame propagation (Figure 2).
Regarding TUM calculations, the achievable grid resolution seems not to be sufficient.
Results from FZK tube experiments for three different mixtures, resulting in a slow flame for
the leanest mixture and fast flames for the others, were calculated by all partners. The slow
burning 10 % hydrogen flame was calculated by TONUS and REACFLOW. The COM3D
code experienced problems with the combustion model at such lean conditions. The other two
experiments were too fast for the TONUS code. They were calculated without problems by
the other codes. For the 11 % mixture a detailed comparison of the flame shapes and the
turbulence data was performed. Again the differences between COM3D and REACFLOW
were small. Both codes produced similar flame shapes and reaction zone thickness.
The results of the benchmark calculations can be summarized as follows:
The TONUS code from CEA is specifically designed for low Mach number flow.
Therefore it is preferably applied in cases with dilute mixtures and low hydrogen
concentrations where only slow flames can develop.
The REACFLOW code from JRC and the COM3D code from FZK cover a wide range
of combustion regimes. The COM3D code is the only fully 3D code in the tests, but has some
limitations for H
-concentrations below 10 vol-%. While REACFLOW and COM3D have
quite different grid representations, their basic models are similar and thus calculated results
are in good agreement.
The code developed by TUM uses the most advanced combustion model. However, at
present the achievable grid resolution seems not to be sufficient when compared to the other
CPU expenses are still important for large-scale CFD applications. Therefore, they are
presently limited to analysis of selected, typical accident scenarios. However, development of
faster numerical procedures and use of fast vector processors and massive parallel computers
will lead to wider application fields in the foreseeable future.
V MODEL APPLICATIONS
The model applications for simulating H
-steam distribution and combustion, documented in
the final report , focus on the results from TONUS, GASFLOW, and COM3D-code
applications. GASFLOW was mainly applied for distribution processes and removal of H
recombiners. TONUS results are dealing with distribution and combustion processes.
COM3D was used for a combustion analysis of a snap shot of steam hydrogen distribution
predicted by GASFLOW. The following applications are reported and demonstrate the
achieved development levels of the three codes:
TONUS simulation of H
-steam distribution and combustion in a four-compartment
GASFLOW simulation of the Battelle Helium injection test Hyjet Jx7.
GASFLOW simulation of the Battelle recombiner tests GX6 and GX7.
GASFLOW simulation of H
-steam distribution with mitigation by recombiners
during a large break LOCA in two spherical PWR containments of German design .
Full reactor scale turbulent combustion simulation with COM3D for an evolutionary
reactor containment design .
TONUS includes an unstructured mesh grid generation with flexibility for the simulation of
complex geometries. An implicit scheme allows distribution calculations with large time steps
and a switch to explicit combustion simulation with small time steps. The analysis of H
steam injection with TONUS shows that during transient phase the information provided from
a lumped parameter simulation is unable to resolve distribution of the gases with sufficient
accuracy. This demonstrates the need for using a multi-dimensional CFD model.
The GASFLOW analysis of H
-steam distribution in a large break LOCA is one of the first
comprehensive 3D containment simulations for a realistic containment geometry and source
term. Comparisons with different lumped parameter simulations show the inability of lumped
parameter models to conservatively predict the intermediate states during the transition to
mixed configurations. The successful GASFLOW simulation of mitigation by catalytic
recombiners in the Battelle GX tests and the reasonable prediction of the buoyant jet in the
Helium injection test JX7 give a validation basis for the simulation of the key mechanisms
that control the H
-steam distribution in the large break LOCA case.
The COM3D application is the first full scale combustion analysis for a reactor containment
(Figure 3). It demonstrates the interfacing capability to GASFLOW to obtain H
distribution, and to continue with a detailed combustion analysis in a much finer mesh,
making use of sophisticated, validated turbulence and combustion models.
TONUS allows the mechanistic analysis of hydrogen distribution and combustion with one
computational tool and will soon reach the same level of development with more modern
computational techniques. GASFLOW also allows mechanistic combustion simulation, but
the combustion model is simpler and does not yet have the wide validation basis of COM3D.
The model applications demonstrate that codes do now allow full-scale simulations of H
steam-air distribution and combustion in the analysis of representative hydrogen specific
With data from four different facilities, the experimental database, which has been developed
within the present project, is unique in its size and completeness. In addition to provide data
for the validation of numerical codes, the experiments also provide useful insight into the
physical phenomena involved in turbulent combustion processes.
The range of applicability of combustion codes used within this project is shown in the
laminar flame transition to fully turbulent detonation
turbulent flame deflagration
The diagramme emphasizes that the codes complement each other. While no single code
covers the whole area of interesting combustion regimes, a combination of the different codes
can describe the whole combustion process from ignition over the flame acceleration regime
to fully developed detonations.
Da > 1
Da < 1
Limitations of the present combustion models and need for further validation do not allow
fully quantitative predictions of the detailed containment loads under all conditions. However,
they allow studies of the complex turbulence/chemistry interaction processes taking place in
realistic large-scale 3-D geometry configurations.
Within this project significant progress has been made in the areas of experiments, model
development, model verification and model application to nuclear power plants. Uncertainties
in the field of combustion behaviour have been reduced by experimental and numerical
investigations. Codes developed in this project allow the investigation of hydrogen specific
problems on reactor scale by experienced experts. The numerical tools give interesting
insights into distribution and combustion processes in complex 3D geometries. Performance
of large scale, multi compartment combustion tests and validation of the codes on these tests
would be suited to reduce considerably remaining uncertainties.
For practical safety applications, further numerical improvement and the use of fast computers
are desirable. This will allow more frequent application of the codes by safety bodies,
industry and research organisations. The capability of predicting local H
concentrations and combustion phenomena will support design, optimisation and reliable
assessment of hydrogen mitigation systems.
Other potential application fields of the developed tools may be safety research for fusion
systems and technical systems based on hydrogen as energy carrier.
This work was supported by the European Community within the 4
 G. van Goethem et al., “EU Research on Severe Accident Phenomenology: Results of the
Current EURATOM Framework Programme and Prospects for the Future,” Proc. SMIRT-15,
Seoul, Korea, August 15-20, 1999
 W.Breitung, A.Kotchourko, „Numerische Simulation von turbulenten Wasserstoff-
Verbrennungen bei schweren Kernreaktorunfaellen.“ Nachrichten - Forschungszentrum
Karlsruhe. Jahrgang 28 2-3/96 S. 175-191
 J.R. Travis, J.W. Spore, P. Royl, “GASFLOW: A Computational Flud Dynamics Code
for Gases, Aerosols and Combustion.” Report LA-13357-M and FZKA-5994, Vol I - III , Oct.
 H. Naji, R. Said, and R. Borghi., “Towards a general turbulent combustion model for
spark ignition engines.”, SAE 890672, SAE Technical Paper Series, 1989.
 R. Borghi, “Turbulent combustion modelling.”, Prog. Energy Combust. Sci., 14:245-292,
 WC Reynolds., “The element potential method for chemical equilibrium analysis:
Implementation in the interactive program STANJAN version 3.”, Report, Dept. of
Mechanical Engineering, Stanford University, Palo Alto, CA, USA, January 1986.
 W. Breitung et al., “Multi-Dimensional Simulation of Hydrogen Distribution and
Combustion in Severe Accident – Final Report”, 4
EURATOM Framework Programme,
Project Nr. FI4S-CT95-00001, to be published.
 P.Royl, H. Rochholz, W. Breitung, J.R. Travis, G. Necker, A. Veser., “GASFLOW-
Analysen zur Wirksamkeit eines Rekominatorkonzepts bei einem postulierten Surge-line
LOCA im Kernkraftwerk Neckarwestheim-2.“ Report FZKA 6333, Forschungszentrum
 A.Kotchourko, W. Breitung, A.Veser., „Reactive Flow Simulations in Complex 3D
Geometries using the COM3D Code.”, Jahrestagung Kerntechnik, pages 173-176, 1999.
Description of the different stages in the behaviour of a
flame without quenching. Pictures are taken from PHD-
Tube and Glastube experiments.
Benchmark calculation on PHD-tube i15 test
Fully developed turbulent combustion in the containment
with flame speed ranging from 70 to 150 m/s,
time = 0,412 s after ignition – COM3D calculation