CHEMICAL KINETICS THROUGH INTRINSIC
LOWDIMENSIONAL MANIFOLDS
FOR GASEOUS HMX
by
John D. Hedengren
A thesis submitted to the faculty of
Brigham Young University
in partial fulfillment of the requirements for the degree of
Master of Science
Department of Chemical Engineering
Brigham Young University
December 2002
ii
BRIGHAM YOUNG UNIVERSITY
GRADUATE COMMITTEE APPROVAL
of a thesis submitted by
John D. Hedengren
This thesis has been read by each member of the following graduate committee and by a
majority vote has been found satisfactory.
__________________________ _________________________________________
Date Merrill W. Beckstead, Chair
__________________________ _________________________________________
Date Jennifer P. Spinti
__________________________ _________________________________________
Date Larry L. Baxter
iii
BRIGHAM YOUNG UNIVERSITY
As chair of the candidate’s graduate committee, I have read the dissertation of John D.
Hedengren in its final form and have found that (1) its format, citations, and
bibliographical style are consistent and acceptable and fulfill university requirements; (2)
its illustrative materials including figures, tables, and charts are in place; and (3) the final
manuscript is satisfactory to the graduate committee and is ready for submission to the
university library.
______________________________ ____________________________________
Date Merrill W. Beckstead
Chair, Graduate Committee
Accepted for the Department
____________________________________
Thomas H. Fletcher
Graduate Coordinator
Accepted for the College
____________________________________
Douglas M. Chabries
Dean, College of Engineering and Technology
iv
ABSTRACT
IMPLEMENTATION OF AUTOMATICALLY SIMPLIFIED
CHEMICAL KINETICS THROUGH INTRINSIC
LOWDIMENSIONAL MANIFOLDS
FOR GASEOUS HMX
John D. Hedengren
Department of Chemical Engineering
Master of Science
An automated method to generate, validate, and implement an Intrinsic Low
Dimensional Manifold (ILDM) has been developed. This method has been applied to a
detailed gaseous HMX (Octahydro1,3,5,7Tetranitro1,3,5,7Tetrazocine) mechanism
that contains 44 species and 232 reactions. A threedimensional manifold tracked
detailed chemistry based on enthalpy, pressure, and mass fraction of N
2
. A four
dimensional manifold adds the mass fraction of NO. A onedimensional (in space) BYU
combustion program has been used to compare the three and fourdimensional manifolds
with a full kinetic mechanism. Simulations show that the threedimensional manifold is
an adequate representation of the full kinetic mechanism away from the reacting surface
(> 60 µm at 20 atm). The fourdimensional manifold is an adequate representation of the
v
full kinetic mechanism closer to the reacting surface (>19 µm at 20 atm). An advantage
of the ILDM method is that computational time is reduced by an order of magnitude.
However, this advantage can be offset by the development time required to create and
implement the ILDM method. Another portion of this project is twodimensional (in
space) simulations that model high explosives in a container. Even though this portion of
the project does not implement an ILDM, it gives insight to meaningful simulations that
can be performed with an ILDM. The simulations explore heat feedback of equilibrium
gases in a heated container filled with solid HMX. Heat feedback is calculated to
investigate flame propagation and burning characteristics of the solid HMX. Further
work is necessary to implement the ILDM approach in similar container simulations.
The methodology used to implement the fourdimensional ILDM into the one
dimensional (in space) combustion code is available to be used in other applications.
One application of interest is the University of Utah’s threedimensional (in space) CFD
code, ARCHES.
vi
ACKNOWLEDGEMENTS
I would like to thank all the people who have helped me with this project,
especially my thesis advisor, Dr. Merrill W. Beckstead. Dr. Beckstead provided helpful
comments and guidance from the start. I have thoroughly enjoyed the autonomy that he
has given me while working on this project. In addition, I appreciate opportunities he has
given me to present my work at a conference and poster session.
Ephraim Washburn and Karl Meredith have both contributed by helping me
understand the computer codes available to our research group. Also, Ephraim’s
suggestions were invaluable during the project’s development.
Ryan Stevens, Ben Hardy, and Matt Gross have each helped at various points in
the development of this project by generating plots, exploring tangent ideas, and by
supplying stress relief for the research group.
I would especially like to thank the people at the University of Utah’s CSAFE
program for funding through this project. Dr. Phillip J. Smith contributed seminal ideas
and helpful comments throughout this project. In addition, I would like to thank Dr.
Jennifer P. Spinti for the practical suggestions in developing the manifold code.
My wife, Sarah, has been supportive when deadlines approached. The occasional
evening or Saturday spent without me is a tribute to her support.
vii
Table of Contents
CHAPTER 1 INTRODUCTION ...........................................................................1
CHAPTER 2 LITERATURE REVIEW.................................................................5
2.1 Reduced Mechanism Methods ....................................................................................................... 5
2.1.1 Sensitivity Analysis...................................................................................................................... 6
2.1.2 Species Lumping Procedures ....................................................................................................... 6
2.1.3 Computational Singular Perturbation (CSP) Technique .............................................................. 7
2.2 Intrinsic LowDimensional Manifold (ILDM).............................................................................. 7
2.2.1 The Equations .............................................................................................................................. 9
2.2.2 Chemical and Physical Process Time Scales ............................................................................... 9
2.2.3 Decoupling Fast Chemical Time Scales..................................................................................... 10
2.2.4 Manifold Dimension .................................................................................................................. 11
2.2.5 OneDimensional Manifolds...................................................................................................... 12
2.2.6 TwoDimensional Manifolds ..................................................................................................... 13
2.2.7 HigherDimensional Manifolds ................................................................................................. 14
2.2.8 Examples of Manifold Implementation...................................................................................... 15
2.2.9 Recent Improvements to the Manifold Method ......................................................................... 16
CHAPTER 3 MANIFOLD GENERATION FOR GASEOUS HMX.....................19
3.1 Manifold Generation .................................................................................................................... 20
3.1.1 Dimension of the Manifold ........................................................................................................ 20
3.1.2 Starting Points in the Composition Space .................................................................................. 21
3.1.3 Reaction Progression.................................................................................................................. 22
3.1.4 Manifold Identification .............................................................................................................. 25
3.1.4.1 Graphical Method............................................................................................................. 25
3.1.4.2 Eigenvector Analysis........................................................................................................ 26
3.1.5 Choice of Parameterization........................................................................................................ 26
3.2 Manifold Storage and Retrieval .................................................................................................. 27
CHAPTER 4 3D MANIFOLD VALIDATION WITH 1D SIMULATION............29
4.1 3D Manifold Direct Substitution................................................................................................ 31
4.2 3D Manifold Implementation in 1D Laminar Simulation ...................................................... 33
4.2.1 The Equations ............................................................................................................................ 34
4.2.2 Discretization of the Equations .................................................................................................. 34
4.2.3 Iterative Solving Routine ........................................................................................................... 35
4.2.4 Grid Independent Study ............................................................................................................. 37
viii
4.2.5 Results ........................................................................................................................................39
CHAPTER 5 4D MANIFOLD VALIDATION WITH 1D SIMULATION............43
5.1 4D Manifold Direct Substitution ................................................................................................43
5.2 4D Manifold Implementation in 1D Laminar Simulation.......................................................44
5.2.1 The Equations.............................................................................................................................44
5.2.2 Discretization of the Equations ..................................................................................................45
5.2.3 Iterative Solving Routine............................................................................................................45
5.2.4 Grid Refinement for Enthalpy Profile ........................................................................................46
5.2.5 Grid Independent Study..............................................................................................................48
5.2.6 A Model for NearSurface Kinetics ...........................................................................................49
5.2.7 Results ........................................................................................................................................53
CHAPTER 6 CONTAINER SIMULATION........................................................57
6.1 CSAFE Container Simulation ....................................................................................................57
6.2 A PostIgnition Model with Gas Products at Equilibrium........................................................58
6.2.1 Heat Transfer Analysis...............................................................................................................60
6.2.2 Open System Simulation............................................................................................................62
6.2.2.1 The Model.........................................................................................................................63
6.2.2.2 Ignition Model ..................................................................................................................63
6.2.2.3 Grid Independent Study....................................................................................................65
6.2.2.4 Simulation Results ............................................................................................................66
6.2.2.5 Gap Thickness Parametric Study......................................................................................69
6.2.2.6 Conclusions for the Open System Simulation ..................................................................70
6.2.3 Closed System Simulation..........................................................................................................71
6.2.3.1 The Model.........................................................................................................................71
6.2.3.2 Ignition Model ..................................................................................................................73
6.2.3.3 Simulation Results ............................................................................................................73
6.2.3.4 Conclusions for the Closed System Simulation................................................................77
CHAPTER 7 CONCLUSIONS..........................................................................79
CHAPTER 8 FUTURE WORK..........................................................................81
CHAPTER 9 REFERENCES............................................................................83
APPENDIX A......................................................................................................87
ix
List of Figures
Figure 1  Time Scales Important to Reacting Flow......................................................... 10
Figure 2  OneDimensional Manifold ............................................................................. 12
Figure 3  TwoDimensional Manifold............................................................................. 14
Figure 4  MultiDimensional Manifold ........................................................................... 15
Figure 5  Reaction Path for Manifold Generation........................................................... 22
Figure 6 – Generating a Gas Phase Chemkin Link File.................................................... 23
Figure 7 – Manifold Generation ....................................................................................... 23
Figure 8  Batch Reactor Results ...................................................................................... 24
Figure 9  Portion of HMX Manifold ............................................................................... 25
Figure 10  ILDM Implementation into 1D, SteadyState Laminar Simulation ............. 29
Figure 11 – Generating a Condensed Phase Chemkin Link File...................................... 30
Figure 12 – Generating a Transport Properties Chemkin Link File ................................. 30
Figure 13 – 1D, SteadyState Laminar Simulation.......................................................... 31
Figure 14 – Direct Substitution of 3D Manifold ............................................................. 32
Figure 15  Flow Time Scale ............................................................................................ 33
Figure 16  Control Volume Layout ................................................................................. 34
Figure 17  Iterative Solving Procedure............................................................................ 36
Figure 18  Oscillations in the Final Solution................................................................... 37
Figure 19  Grid Independent Study ................................................................................. 38
Figure 20  Grid Independent Study ................................................................................. 39
Figure 21  Results of OneDimensional, SteadyState Simulation ................................. 40
Figure 22  Results of OneDimensional, SteadyState Simulation ................................. 40
Figure 23  Results of OneDimensional, SteadyState Simulation ................................. 41
Figure 24  Results of OneDimensional, SteadyState Simulation ................................. 41
Figure 25 – Full Kinetic Mechanism Compared to Manifold Results.............................. 44
Figure 26  Iterative Solving Procedure............................................................................ 46
Figure 27  Data Flow with Enthalpy Grid Refinement ................................................... 47
Figure 28  Grid Independence Study (050 µm) ............................................................. 48
Figure 29  Grid Independence Study (>25 µm)............................................................... 49
Figure 30  NO Profile for Various Starting Points at 20 atm.......................................... 50
Figure 31  NO Profile for Various Starting Points at 5 atm............................................ 51
Figure 32  NO Profile for Various Starting Points at 60 atm.......................................... 51
Figure 33  Enthalpy Profile at 60 atm.............................................................................. 52
Figure 34  Results of OneDimensional, SteadyState Simulation ................................. 53
Figure 35  Results of OneDimensional, SteadyState Simulation ................................. 54
Figure 36  Results of OneDimensional, SteadyState Simulation ................................. 54
Figure 37  Results of OneDimensional, SteadyState Simulation ................................. 55
Figure 38  Pressurization of Container ............................................................................ 58
Figure 39  Model of Container ........................................................................................ 59
x
Figure 40  Heat Transfer Simulation............................................................................... 61
Figure 41  Heat Transfer at the SolidGas Interface........................................................ 62
Figure 42  Open System Model....................................................................................... 63
Figure 43  Ignition Data from Heating Tests................................................................... 64
Figure 44  Grid Layout .................................................................................................... 65
Figure 45  Temperature Contours.................................................................................... 67
Figure 46  Comparison of Heat Fluxes............................................................................ 68
Figure 47  Distance to Ignition for a 5 mm Gap.............................................................. 69
Figure 48  Ignition Prediction for Various Gap Widths .................................................. 70
Figure 49  Closed System Model .................................................................................... 72
Figure 50  Initial Velocity Profile.................................................................................... 74
Figure 51  Temperature Profile in the Closed System Simulation.................................. 75
Figure 52  Heat Flux Analysis for Flame Propagation.................................................... 76
xi
List of Tables
Table 1  Eigenvalue Analysis .......................................................................................... 10
xii
1
Chapter 1 Introduction
Modeling of reacting flows is a computational challenge even for the world’s
largest supercomputers (Tomlin, Turanyi, and Pilling 1997). This challenge is being
explored by DOE’s ASCI (Accelerated Strategic Computing Initiative) program. Part of
the program is delegated to the University of Utah through CSAFE (Center for the
Simulation of Accidental Fires and Explosions). In 1996, the University of Utah received
funding from the DOE ASCI (Accelerated Strategic Computational Initiative) program to
develop new tools for the simulation of accidental fires and explosions. To carry out this
program, the University of Utah created CSAFE. This initiative seeks to simulate the
rapid heating of a container filled with a plastic bonded explosive, PBX9501, in a large
jetfuel pool fire. PBX9501 is composed of 95% HMX (Octahydro1,3,5,7Tetranitro
1,3,5,7Tetrazocine) as the explosive and 5% binder. To simplify the chemical kinetics,
the high explosive is modeled as pure HMX. In the simulation, heat from the jetfuel fire
ignites the high explosive. As the high explosive combusts, the container pressurizes,
swells, and finally bursts. CSAFE’s goal is to simulate this scenario from just after the
ignition of the pool fire up to the rupture of the container.
The challenge of such a simulation lies in calculating detailed transport and
complex chemical kinetics simultaneously. Currently, it is very computationally
expensive to calculate both. Therefore, the options are to make simplified transport
calculations (i.e. Reynolds Averaged NavierStokes methods), reduce the chemical
kinetics, or both.
2
Efforts to reduce chemical kinetics include systematically reduced mechanisms,
the constrained equilibrium approach, repromodeling, computational singular
perturbation, the method of Intrinsic LowDimensional Manifolds (ILDM), and dynamic
dimension reduction (Blasenbrey, Schmidt, and Maas 1998). All of these methods fall
under one of two general categories: (1) reduced mechanism methods or (2) ILDM
methods. The appeal of the ILDM method over reduced mechanism methods is that
much of the detailed chemistry is maintained while still showing a dramatic
computational speedup for chemistry calculations (Yang and Pope 1998b). Instead of
simplifying a reaction mechanism by eliminating reactions or species, the manifold
method explores time scales for chemical reactions. For reacting flow calculations, the
progression of reactions with fast time scales (compared to flow time scales) can be
ignored. The manifold method ignores reactions with fast time scales yet is still based on
detailed kinetics.
To assist in the CSAFE effort of providing a fullphysics simulation at a
reasonable computational cost, the ILDM method for reducing chemical kinetics has been
applied to a detailed, gaseous HMX mechanism. The objectives of this project were to:
• Automate a system for generating an ILDM for gaseous HMX.
• Implement the ILDM in a 1D (in space), steadystate laminar simulation.
• Evaluate the performance of the HMX ILDM by comparison with full kinetics in
the 1D (in space), steadystate laminar simulation.
• Lay a foundation for higherdimensional (in space) simulations that apply ILDM
simplified chemistry by investigating 2D (in space) simulations with equilibrium
gas properties.
3
This thesis describes the achievement of these objectives. Specifically, Chapter 2
reviews two methods for simplifying chemical kinetics: reduced mechanism methods and
ILDM methods. The ILDM method is described mathematically, and successful
applications of this method are reviewed. Chapter 3 describes original work of this
project in generating a 3D (in pressure, enthalpy, and mass fraction of N
2
) ILDM for
HMX. The application of the 3D HMX ILDM in a 1D (in space), steadystate laminar
simulation is described in Chapter 4. Because of shortcomings of the 3D ILDM, a 4D
(in pressure, enthalpy, and mass fractions of N
2
and NO) ILDM was created. Chapter 5
describes the application of the 4D ILDM in the 1D (in space), steadystate laminar
simulation. Because of the 4D ILDM inadequacy in the fast reaction zone, a model for
nearsurface kinetics is proposed. Chapter 6 describes 2D (in space) simulations with
equilibrium HMX products. These simulations lay a foundation for higherdimensional
(in space) simulations that apply ILDMsimplified chemistry. Conclusions and general
observations about the ILDM method are given in Chapter 7. Chapter 8 describes future
work that can be done by other researchers to meet CSAFE’s objective of implementing
more accurate HMX chemistry into container simulations.
4
5
Chapter 2 Literature Review
This chapter discusses efforts to reduce chemical kinetics that fall into two
general categories: reduced mechanism methods and ILDM methods (Yang and Pope
1998b). The ILDM method, in particular, is described in theoretical detail. The final
two sections of the chapter show applications of ILDM techniques to practical
combustion simulations and recent improvements to the method.
2.1 Reduced Mechanism Methods
Traditionally, when chemical kinetic mechanisms were constructed manually, it
required the technical expertise of someone familiar with the reaction mechanism
(Tomlin, et. al. 1997). This person selected chemical species and reactions that would be
important under given conditions. In the precomputer era, the chemical expert would
apply either the steadystate or partialequilibrium approximation to various elementary
reactions in the mechanism to come up with an analytical solution. Because analytical
solutions were necessary, the size of the kinetic mechanisms was limited.
With the development of faster computers, it became possible, though still very
expensive, to model complex chemical reactions with detailed kinetics. To reduce the
cost of computing, the chemical expert would systematically reduce the complex
mechanisms into skeletal mechanisms. The advantage of using skeletal mechanisms over
analytical solutions is that less development time is required. The drawbacks to this
method are that it requires substantial knowledge of a detailed mechanism to generate a
reduced mechanism. In addition, the resultant reduced mechanism only works for a
6
limited range of temperature, pressure, and composition. The range of the reduced
mechanism is limited because reactions can depend greatly on physical conditions. For
example, there is a change in reaction pathway as the combustion moves from fuelrich to
fuellean conditions.
Other techniques have also been developed to assist in reduced mechanism
development. Three of these techniques are described below.
2.1.1 Sensitivity Analysis
A sensitivity analysis or principal component analysis is one of the tools to reduce
complex chemical kinetics (Tomlin, et. al. 1997). The sensitivity analysis identifies
reactions that have little effect on the overall reaction by measuring a normalized local
sensitivity. The normalized local sensitivity represents the amount that a species
concentration changes after a small perturbation in one of the system parameters (e.g.
temperature, pressure, etc.). Reactions that contain species with low sensitivities can be
eliminated from the detailed mechanism. In this way, reactions that involve principal
components remain in the new, reduced mechanism.
2.1.2 Species Lumping Procedures
Even after a reduced mechanism is achieved, the number of reactions can still be
too great to implement the mechanism into a practical combustion code. Lumping
procedures can be used to reduce the reaction system to a lowerdimensional system of
equations (Tomlin, et. al. 1997). This technique is founded on the assumption that
chemically similar species have similar rate parameters and can therefore be treated as a
single compound. Accordingly, the dimension of the lumped mechanism is equal to the
7
number of unique classifications of chemical species. The weakness of this method is
that it is a compromise between a detailed mechanism and a global mechanism. By
lumping species together, much of the detailed reaction information is lost.
2.1.3 Computational Singular Perturbation (CSP) Technique
The CSP technique is a formal way to apply partialequilibrium approximations
on an a priori basis (Lam and Goussis 1988). The procedure begins by evaluating time
scales of reactions and ordering them from fastest to slowest. The entire chemistry
calculation proceeds with the time scale of the fastest reaction. As the reaction proceeds,
the fastest reaction becomes exhausted and its reaction rate approaches zero. Once the
fastest reaction is exhausted, it is labeled as a dead mode and is discarded from the
calculation. The chemistry calculation is then allowed to proceed at the time scale of the
next fastest reaction. As the fastest reactions are eliminated, the system of equations
becomes less stiff and therefore less computationally expensive to solve. CSP has been
used to investigate mechanisms and to reduce them rather than to provide a
computationally efficient technique for combustion modeling (Tomlin, et. al. 1997).
2.2 Intrinsic LowDimensional Manifold (ILDM)
Many of the previously mentioned methods for creating reduced mechanisms rely
on steadystate or partialequilibrium approximations. However, the reduced
mechanisms are generally limited to a range of temperature, pressure, and/or species’
concentrations (i.e. state space). Outside of this defined state space, large errors can
occur.
To overcome this deficiency, Mass and Pope proposed a new method for reducing
chemical kinetics based on Intrinsic LowDimensional Manifolds (ILDM) (Maas and
8
Pope 1992a, Maas and Pope 1992b). Mathematically, a manifold is a topological space
or surface. An ILDM is a topological space or surface defined by independent
thermochemical parameters. Maas and Pope’s original work proposed a method to
automatically apply the steadystate and partialequilibrium approximations over all
possible thermochemical states of the system (state space). A thermochemical system
can be characterized by fast and slow reactions. It is assumed that the fastest reactions
can be decoupled from the slow ones. Then, the decoupled fast reactions (those that are
faster than the flow time scales) can be ignored while those with slow time scales are
tracked using progress variables. For example, for a particular mechanism, three
variables (e.g. enthalpy, pressure, and mass fraction of N
2
) could be selected as the
progress variables. By this means, a limited number of progress variables can be used to
characterize the entire thermochemical system. Once these variables are calculated, all
statespace variables, including mass fractions of chemical species, temperature, entropy,
and reaction rates, are known. Since the progress variables completely describe the
system, only these variables must be calculated. This leads to a dramatic reduction of
CPU time for solving the chemistry in a reacting flow calculation.
The ILDM method is especially suited for nonpremixed reacting flow
calculations where mixing controls much of the chemical reaction. In premixed reacting
flow calculations, fewer of the fast time scales can be decoupled. This can lead to higher
dimensional manifolds that are more difficult to implement in realistic combustion
simulations.
9
2.2.1 The Equations
The composition vector (φ) is composed of all intensive variables that define the
thermochemical state of a system. It includes enthalpy (H) and all species mass fractions
(Y
i
, i=1,2,…,n
s
), and is written as φ = (H, Y
1
, Y
2
,…,Y
ns
). The equations for the
composition φ can be written as (Tomlin, et. al. 1997)
) , ( )] , ( [
) , (
t x t x S
dt
t x d
Γ + · φ
φ
(1)
where S is the rate of change due to reactions (for enthalpy, S = 0) and Γ is the rate of
change due to transport effects (convection, diffusion, etc).
2.2.2 Chemical and Physical Process Time Scales
Chemical mechanisms typically have a much larger range of time scales than
necessary for the accurate calculation of a reacting flow system. Typically, reaction time
scales are on the order of 10
9
seconds (e.g. some radical reactions) to 10
2
seconds (e.g.
NO formation in coal combustion) in reactions important to chemical combustion. On
the other hand, the time scales necessary to model physical processes are on the order of
10
4
to 10
2
seconds. Figure 1 illustrates these ranges (Maas and Pope 1992b).
10
Figure 1  Time Scales Important to Reacting Flow
Species involved in reactions that happen much faster than the physical processes
can be decoupled and eliminated (Blasenbrey, et. al. 1998).
2.2.3 Decoupling Fast Chemical Time Scales
Mass and Pope used an eigenvalue analysis to determine if a system is on a lower
dimensional manifold and is governed by slow chemistry. The eigenvalues are those of
the Jacobian (dS
i
/dφ
j
). A large negative eigenvalue associated with an elementary
reaction means that the reaction is governed by fast chemistry and is in local equilibrium.
Conserved variables (e.g. pressure in an isobaric system) create eigenvectors that have
eigenvalues of zero (see Table 1).
Table 1  Eigenvalue Analysis
Eigenvalue (real part) Response to Physical Perturbations
Positive Perturbation will increase (instability)
Zero Perturbation will not change with time (change
of a conserved variable)
Negative Perturbation will relax to zero
For every reacting flow calculation, there is a time scale for the flow (τ
flow
) based
on the resolution of the transport calculations. If the real part of an eigenvalue is greater
11
than –1/τ
flow
, then the corresponding eigenvector is in the slow subspace (Yang and Pope
1998a). However, if the eigenvalue is less than –1/τ
flow
, then the corresponding
eigenvector is in the fast subspace and can be decoupled from the reaction system. The
number of progress variables that cannot be decoupled form a reduced set. The progress
variables in the reduced set are the only ones that must calculated for a laminar or
turbulent combustion calculation.
2.2.4 Manifold Dimension
The number of variables in the reduced set determines the dimension of the
manifold. Since it is difficult to implement an adaptivedimensional manifold into a CFD
(Computational Fluid Dynamics) code, the dimension of the manifold should be fixed
over the entire state space of the manifold (Blasenbrey, et. al. 1998). This makes the
implementation much less time consuming but also has some drawbacks. These
drawbacks include:
• Calculation of an ILDM is attempted over the whole state space even though
some domains of the state space are never accessed in practical applications.
• Fixed parameterization does not guarantee the uniqueness or existence of
solutions and can yield illconditioned equation systems for the manifold.
• Higherdimensional manifolds may be required in some regions of the state space.
This means that too few progress variables are tracked and could lead to errors
typical of an invalid steadystate approximation.
Despite these drawbacks, a manifold of constant dimension is the only practical way to
implement the ILDM method for combustion calculations at this time.
12
2.2.5 OneDimensional Manifolds
Up to this point, the manifold approach has been described mathematically.
Manifolds can also be observed through graphical techniques (Maas and Pope 1994).
The simplest case is a manifold plotted against one progress variable. Figure 2 shows the
specific mole number of H
2
O (mol of H
2
O/gm of mixture) plotted against the specific
mole number of CO
2
for COH
2
air as reactants (Maas and Pope 1992b). The beginning
of each trajectory represents an initial composition for the reaction. All initial
compositions have the same elemental fractions but different species fractions. The
trajectories represent the reaction progression and the square is the final equilibrium
condition.
Figure 2  OneDimensional Manifold
In this case, the specific mole number of CO
2
is the manifold coordinate. The
trajectories have the following properties (Maas and Pope 1992b):
13
• All approach one common point (equilibrium)
• All approach one common trajectory (manifold)
• On the manifold, only slow time scales govern the chemical reaction
• Once a thermochemical system is on the manifold, it will always remain on the
manifold if there are no perturbations (Γ = 0).
A thermochemical system is drawn to a unique manifold and progresses along it
until the system is perturbed through physical processes (e.g. molecular diffusion,
mixing, heat convection). If it is perturbed at a time scale slower than the fastest time
scale of the manifold, the manifold is an accurate representation of the chemical
dynamics. However, if the perturbations happen at a time scale that is faster than the
existing manifold, a higherdimensional manifold should be employed (Blasenbrey, et. al.
1998). A higherdimensional manifold simply means that more progress variables must
be added to the manifold to capture a larger number of time scales. This can dramatically
increase the storage requirements for the manifold.
2.2.6 TwoDimensional Manifolds
A twodimensional manifold has the same general characteristics as the one
dimensional manifold. The only difference is that two progress variables must be used to
describe the state of the system. Figure 3 is an example of a twodimensional manifold
(Blasenbrey, et. al. 1998).
14
Figure 3  TwoDimensional Manifold
The manifold is the grid surface. Each trajectory represents a starting point in the
state space. Each trajectory relaxes to the grid surface quickly and stays on that surface
until the thermochemical system reaches equilibrium (square point on the plot).
2.2.7 HigherDimensional Manifolds
At equilibrium, a 0D manifold (no progress variables) represents the
thermochemical system because there is no change. Near equilibrium (i.e. a final rate
determining step before equilibrium), a 1D (one progress variable) manifold adequately
characterizes the reaction system. The dimension of a manifold increases further from
equilibrium to account for faster reaction time scales. Figure 4 shows a multi
dimensional manifold (Bongers 2002).
15
Figure 4  MultiDimensional Manifold
Higherdimensional manifolds can be visualized through lowerdimensional plots
by taking slices of the full manifold. For example, a threedimensional manifold of the
CH
4
air system can be characterized by enthalpy, pressure, and mass fraction of CO
2
. If
the enthalpy is held constant, the manifold can be plotted as a twodimensional manifold.
Then the enthalpy can be held constant at another value and another twodimensional
manifold plot can be generated. In this way, a higherdimensional manifold can be
visualized and selected graphically.
2.2.8 Examples of Manifold Implementation
The ILDM method has been implemented successfully for laminar premixed CO
H
2
O
2
N
2
flames (Maas and Pope 1994). The COH
2
O
2
N
2
ILDM tracks all species,
including radicals, within 3% error. In addition, there is a speedup in the calculation of
the chemistry by a factor of 10.
The ILDM method has also been implemented successfully in turbulent flame
calculations. Three examples include:
16
• A numerical simulation of a turbulent nonpremixed CH
4
H
2
air flame shows that
the ILDM method models flames near extinction and near equilibrium (Xiao,
Schmidt, and Maas 1998).
• A piloted COH
2
N
2
air diffusion flame simulation shows that extinction can be
predicted within 5% of the experimental value (Norris and Pope 1995).
• A CH
4
air combustion system shows that the ILDM method is 1,500 times faster
in computing chemistry than a skeletal mechanism (Yang and Pope 1998b).
2.2.9 Recent Improvements to the Manifold Method
New strategies have been developed to overcome some of the shortcomings of the
original ILDM method. The drawbacks to the original ILDM method are (Yang and
Pope 1998b):
• The entire state space must be calculated for a fixed dimensional manifold.
• The storage space required to record the manifold can be very large and increases
dramatically for higherdimensional manifolds.
• For higherdimensional manifolds, the work to retrieve information is not trivial.
To overcome these shortcomings, the method of In Situ Adaptive Tabulation in Principal
Directions (ISAT) was created (Yang and Pope 1998b). The ISAT method operates
under the same principles as the ILDM method, but it calculates and stores a full
dimensional manifold during the reacting flow calculation. Thus, only areas of the state
space that are accessed are included in the manifold. Unlike the ILDM approach, the
dimension of the manifold is not reduced. All dimensions are stored (up to 20 species).
Even with the increased dimensions, a reduction by a factor of 1000 in computational
effort can still be observed (Pope 1997).
17
Another technique to overcome the drawbacks of the standard tabulation method
is to store the manifold as a piecewise polynomial. As a polynomial, only the
coefficients to the polynomial need to be stored. For a test case, this method reduces the
storage requirements of the manifold by a factor of 100 compared with the storage
requirements for individual points (Niemann, Schmidt, and Maas 1997).
18
19
Chapter 3 Manifold Generation for Gaseous HMX
One of the major challenges for the ILDM method has proved to be the
application of the theory to actual systems. The developers of the method have shown
success of the method for laminar and turbulent flame calculations (Maas and Pope
1992a, Maas and Pope 1994, Yang and Pope 1998a, Yang and Pope 1998b, Norris and
Pope 1995, Maas and Pope 1992b). The University of Utah CSAFE developers have
also used the ILDM approach in a CFD simulation of a heptane pool fire (CSAFE
Program 2001).
As stated previously, the goal of CSAFE is the simulation of the heatup and
pressurization of a container engulfed in a pool fire. Currently, the pool fire simulation
and the container simulation are performed independently. Within the container, a global
mechanism is used to model the gasphase reactions of HMX combustion.
A major development towards tighter coupling between the container dynamics
and the pool fire codes would be to include a more detailed model of the gasphase HMX
chemistry inside the container. This would provide a more fundamental representation of
the gasphase chemistry with respect to pressurization.
The ILDM method is an attractive alternative to detailed kinetics in the container
calculations. A threedimensional (based on mixture fraction, enthalpy, and the mass
fraction of CO2 +CO) manifold is already used to characterize the gasphase reactions
and soot chemistry in a turbulent heptane pool fire (CSAFE Program 2001). Based on
the success of the ILDM approach for hydrocarbon chemistry, it was proposed to
20
simplify the high explosive chemistry with the same approach. The following section
describes the generation of the HMX manifold that will be used in the simulation of the
gas phase of the container.
3.1 Manifold Generation
There are six steps to the generation of a fixeddimensional manifold: selecting
the dimension, generating starting points in the composition space, tracking each starting
point to equilibrium, identifying the manifold, choosing the parameterization, and storing
the manifold.
3.1.1 Dimension of the Manifold
The dimension of the manifold has been investigated through an eigenvalue
analysis of the Jacobian (a matrix formed during the iterative Newton’s method). The
number of eigenvalues that have a real portion greater than –1/τ
flow
determines the
dimension of the manifold. For gaseous HMX, an eigenvalue analysis could not be
performed because of numerical errors in calculating the eigenvalues. An analysis of the
Jacobian revealed that many of the array values approach zero while others are very
large. These numerical errors lead to extraneous eigenvalues. Because the eigenvalue
analysis did not reveal adequate information about the system, the dimension of the
manifold was initially fixed to three (enthalpy, pressure, and mass fraction of N
2
). Later,
a fourdimensional manifold was developed (mass fraction of NO added) to increase the
accuracy.
21
3.1.2 Starting Points in the Composition Space
To generate starting points in the composition space, all but one of the parameters
must be held constant. For the adiabatic batch reactor combustion code, enthalpy and
pressure were held constant. The other parameters (Y
N2
and Y
NO
) were recorded as the
simulated reactions proceeded. In the simulation, radicals were introduced to initiate the
reaction. The generation of radicals was simulated in an adiabatic Perfectly Stirred
Reactor (PSR).
A PSR code was developed by Glarborg et. al. (Glarborg, Kee, Grcar, and Miller
1992). Differing residence times for the PSR were attempted. However, because the
HMX reaction time scales are extremely fast, only fully reacted HMX or totally
unreacted HMX resulted. To generate sufficient radicals, fully reacted HMX were used.
These simulated reacted gases from the PSR were then mathematically mixed (in a
subroutine separate from the PSR code) in proportions of 5%, 20%, 35%, 50%, and 65%
with unreacted HMX on a mass basis according to the following equation:
Reactants Reactants Reactants
) 1 ( f Y f Y Y
j j
PSR j
+ − · (2)
where j is an index number for each species, Y is the mass fraction, Y
PSR
is the mass
fraction of PSR products, Y
Reactants
is the species mass fraction in the unreacted feed, and
f
Reactants
is the fraction of unreacted feed that is mixed with the PSR products.
After the mass fractions were evaluated, the temperature of the new mixture was
computed in a subroutine separate from the PSR code. Since the PSR is specified as an
adiabatic reactor, the products of the PSR had the same enthalpy as unreacted HMX.
Therefore, the mixture of PSR products and unreacted HMX had a known enthalpy. By
this means, the temperature of the mixture was solved iteratively. The secant method
22
(Burden and Faires 1997) was used to find the temperature with the initial guess midway
between the high temperature of the PSR products and the low temperature of the
reactants. Generally, the temperature met a 10
11
K convergence criteria in fewer than 10
iterations.
3.1.3 Reaction Progression
The calculated mixture of PSR products and unreacted HMX is the simulated
input stream to an adiabatic batch reactor program. Figure 5 illustrates the flow of a
trajectory calculation from the simulated initial reactants, through the simulated PSR, and
finally to the simulated batch reactor.
Figure 5  Reaction Path for Manifold Generation
The simulations use the ChemkinII subroutines developed at Sandia National
Laboratories (Kee, et. al. 1989) to calculate thermodynamic and kinetic properties. In
order to use the Chemkin Subroutine Library, a Chemkin link file must be created (see
Figure 6).
23
Figure 6 – Generating a Gas Phase Chemkin Link File
After the Chemkin link file is created, the manifold can be automatically
generated (see Figure 7).
Figure 7 – Manifold Generation
In the simulated batch reactor, mass fractions of the species, temperature, reaction
rate of the species, entropy, and enthalpy were computed starting at the initial time and
stepping forward until equilibrium is reached. The progression towards equilibrium
forms a trajectory when plotted against time. Figure 8 shows an example of the mass
24
fraction of CO
2
plotted against time for three different initial compositions at the same
enthalpy, pressure, and elemental composition.
Figure 8  Batch Reactor Results
Each point on the graph represents a time instance where all of the state space is
computed and recorded. Initially, the time step must be set by the user to be small to
account for fast reactions and to avoid errors in the numerical integrator. The original
batch reactor program only allowed for a constant time step equal to the smallest time
step during the entire calculation. Therefore, the batch reactor program was modified to
allow for changing time steps. The original program stopped when the numerical
integrator (DVODE) returned an error. Instead of stopping the execution of the program,
the modified program retries the numerical integration with a lower time step. If a certain
number of numerical integrations have passed since an error, then the time step is
increased. This modification speeds up the calculation by many orders of magnitude
without affecting the accuracy of the calculation.
25
3.1.4 Manifold Identification
During the course of a batch reactor simulation, the calculated species,
temperature, and reaction rates are stored in data files. These data files are analyzed to
identify the ILDM.
3.1.4.1 Graphical Method
Figure 9 shows an example of the graphical method for identifying the lower
dimensional manifold. Each trajectory represents a separate simulation performed in an
adiabatic batch reactor. Each starting point represents a different percentage of unreacted
HMX mixed with products of HMX combustion from a simulated PSR. The values in
the legend refer to the percent of unreacted HMX (see Equation 2). In this case, entropy
is plotted against the calculated mass fraction of N
2
. The initial temperature of the
unreacted HMX gas is 127 ºC and the pressure is 20 atm. All the trajectories are attracted
towards the manifold and end at the point that represents equilibrium.
Figure 9  Portion of HMX Manifold
26
To generate a complete manifold, the pressure and enthalpy must be varied over a
range necessitated by the state space of the combustion system that will employ the
manifold. With each change in pressure or enthalpy, a separate graph is generated,
analogous to Figure 9. Additional graphs can be made for all the species and temperature
as a function of N
2
mass fraction. These graphs can aid in identifying the ILDM.
3.1.4.2 Eigenvector Analysis
An eigenvector analysis requires more development time but results in a more
automated system for selecting a manifold. The eigenvector analysis allows for the
calculation of only one trajectory and identifies when that trajectory reaches the
manifold. For this project, the graphical method for selecting the manifold was
employed, so an eigenvector analysis was not developed.
3.1.5 Choice of Parameterization
Three (enthalpy, pressure, and mass fraction of N
2
) or four (enthalpy, pressure,
and mass fraction of N
2
and NO) parameters were chosen as dependent variables of the
manifold. Therefore, if these three or four parameters are known, all other species
concentrations are also known.
A special requirement of one of the species progress variables is that it must be
monotonically increasing or decreasing in the area of the manifold. Additional species
progress variables must also meet this specification if aliasing occurs. Major products
(i.e. N
2
, CO
2
, H
2
O) are often considered as the first species progress variable instead of
intermediates (i.e. NO, CO, HCN).
27
3.2 Manifold Storage and Retrieval
Once the manifold is identified, it is stored in a data file for future use in a
combustion code. Because the project uses computers with varying operating systems,
the files are stored in ASCII format instead of binary.
At runtime, the data files are loaded into arrays within the combustion application.
A multivariable linear interpolator looks up values in the tables based on enthalpy,
pressure, and mass fractions of N
2
and NO. One of the challenges to the retrieval process
is accounting for values that are outside of the manifold state space. For example, at a
pressure of 20 atm and an enthalpy of 10
10
erg/gm, the lowest accessible value for mass
fraction of N
2
may be 0.10. What values are returned when the code attempts to access
0.05 for a mass fraction of N
2
? Blasenbrey, et. al. suggest a linear extrapolation between
the lowest available value and the initial concentration (Blasenbrey, et. al. 1998).
Another option is to include reaction trajectories that have not reached the manifold. The
latter option was implemented in this project because it represents more realistic
conditions for the simulated reactions.
28
29
Chapter 4 3D Manifold Validation with 1D Simulation
In order to validate the effectiveness of the HMX manifold, results from a full
kinetic mechanism and a manifold approach are compared in a onedimensional (in
space) combustion code. The code that is used to test the manifold includes a condensed
(solid and liquid) phase and a gas phase (Davidson 1996). The onedimensional
combustion model is simplified further with a steadystate assumption by specifying that
the liquidgas interface as stationary. In such a Lagrangian reference, the simulated
combustion products move through advection. With high advection, diffusion is assumed
to be negligible. Figure 10 shows how the ILDM replaces the full kinetics calculation in
the 1D, steadystate laminar simulation.
Figure 10  ILDM Implementation into 1D, SteadyState Laminar Simulation
In addition to the gas phase Chemkin link file (see Figure 6), a condensed phase
Chemkin link file (see Figure 11), and a transport properties Chemkin link file (see
Figure 12) must be created.
30
Figure 11 – Generating a Condensed Phase Chemkin Link File
Figure 12 – Generating a Transport Properties Chemkin Link File
Once the required link files are created, the 1D, steadystate laminar simulation
can be run (see Figure 13).
31
Figure 13 – 1D, SteadyState Laminar Simulation
Since the calculation of the gas phase kinetics typically requires more than 99%
of the CPU time, the manifold is only used to simplify the calculation of the gasphase
reactions. The three result files are the solutions to direct substitution of the manifold
(using the full kinetics solution of enthalpy and mass fraction of N
2
), full kinetics, and
iterative solution of the manifold.
4.1 3D Manifold Direct Substitution
A quick check to see how well the manifold follows the detailed kinetics is to
substitute manifold values for detailed kinetics values. The final solution of enthalpy and
mass fraction of N
2
from the detailed kinetics calculation is used to look up manifold
values. Figure 14 shows the concentration of CO
2
versus distance from the surface.
32
Figure 14 – Direct Substitution of 3D Manifold
Near the surface (0 to 60 µm) the manifold does not agree with the detailed
kinetics while away from the surface (60 µm to infinity) the two are exactly equal. This
is due to the varying time scales along the flame. The time scale of the onedimensional
flow is computed by the following equation:
velocity
distance ∆
·
flow
τ (3)
Figure 15 shows the time scale versus distance.
33
Figure 15  Flow Time Scale
Because the time scale is so small near the surface, a higherdimensional manifold
should be used to get accurate results. As long as a sufficiently high dimension of the
manifold is used, the manifold should accurately represent the detailed kinetics
everywhere.
4.2 3D Manifold Implementation in 1D Laminar Simulation
Using detailed kinetics requires the solution to the energy equation and to the
continuity equations. For HMX, this means that 45 ordinary differential equations must
be solved. Using a lowdimensional manifold reduces the number of dependent
variables. The number of differential equations that must be solved is equal to the
dimension of the manifold. For example, using a 4D manifold means that only 4
ordinary differential equations must be solved.
34
4.2.1 The Equations
In an isobaric simulation, the HMX manifold is only a function of enthalpy and
mass fraction of N
2
. Therefore, the equations that best describe the chemistry are the
energy equation in the enthalpy form and the continuity equation for N
2
. The energy
equation in the enthalpy form is:
T
dx
d
k
dx
d
dx
dH
n
mass
· (4)
(where H is enthalpy, n
mass
is the mass flux, k is the thermal conductivity, and T is the
temperature). The N
2
continuity equation is:
2 2 2
N N N mass
W Y
dx
d
n ω · (5)
(where Y
N2
is the mass fraction of N
2
, ω
Ν2
is the reaction rate of N
2
, and W
N2
is the
molecular weight of N
2
).
4.2.2 Discretization of the Equations
The enthalpy equation and N
2
continuity equation are discretized using the control
volume layout as shown in Figure 16.
Figure 16  Control Volume Layout
35
The control volume approach integrates the equations over a distance from e
(east) to w (west). Using the control volume approach and integrating between the
control surfaces for P, the energy equation becomes
w
w
e
e w e mass
dx
dT
k
dx
dT
k H H n
,
`
.

−
,
`
.

· − ) ( (6)
and the N
2
conservation equation becomes
P P P N N mass
x W Y Y n
w e
∆ ⋅ ⋅ · − ϖ ) (
2 2
(7)
where ∆x
P
is the distance across the control volume for P from e to w.
Assuming a linear temperature profile between nodes, using an underrelaxation
parameter (α < 1) to ensure the stability of the solver, and using an upwind scheme (valid
for high Peclet numbers) the enthalpy equation becomes
]
]
]
]
]
]
]
−
,
`
.

,
`
.
 −
⋅ −
−
⋅ + ⋅
+ ·
− − 1 ,
,
1 , , i P
mass
w
W P
w
e
P E
e i W mass
i P i P
H
n
x
T T
k
x
T T
k H n
H H
α
δ δ
α (8)
and the N
2
conservation equation becomes
( )
]
]
]
]
]
]
−
∆ ⋅ ⋅ + ⋅
+ ·
− −
−
1 ,
,
1 , ,
2
1 2
2 2
) (
i P
i W
i P i P
N
mass
i P P P N mass
N N
Y
n
x W Y n
Y Y
α
ω
α (9)
where the subscript (i) denotes the current iteration value and the subscript (i1) denotes
the value from the previous iteration.
4.2.3 Iterative Solving Routine
Both of the discretized equations are explicit in time and distance yet the overall
solving routine is implicit. The overall solving routine is iterative because the manifold
36
is a function of both enthalpy and the mass fraction of N
2
. Once the enthalpy profile is
calculated, the mass fraction profile of N
2
is calculated. This iterative process repeats
until the solution converges (see Figure 17).
Figure 17  Iterative Solving Procedure
An interesting characteristic of the manifold solution is that the mass fraction of
N
2
and enthalpy continue to oscillate slightly around a solution but never stabilize to a
fixed solution. This is due to the discontinuities in the reaction trajectories when they
have not reached the manifold. Figure 18 shows the last iteration of 1000 total iterations
versus the temperature averaged over the last 100 iterations for grid spacing at 6 µm.
37
Figure 18  Oscillations in the Final Solution
The averaged solution over the last 100 iterations will be used to ensure
physically realistic results and to smooth out the oscillations.
4.2.4 Grid Independent Study
An important aspect of numerical analysis is a grid independent study. Generally,
a numerical solution becomes more accurate as more nodes are used. However, using
additional nodes also increases the required computer memory and computational time.
The appropriate number of nodes can be determined by increasing the number of nodes
until the numerical solution changes by less than a specified tolerance. Figure 19 shows
the calculated temperature versus distance for the grid independent study. The numbers
in the legend refer to the grid spacing.
38
Figure 19  Grid Independent Study
Grid spacing less than 6 µm gives a grid independent solution. However,
decreasing the grid spacing gives increased oscillations during the implicit solving
routine. The oscillations are likely the result of using reaction trajectories that have not
yet reached the lowerdimensional manifold. Small jumps in the temperature lead to
large values of the second derivative for temperature in the enthalpy equation. Large
values for the second derivative of temperature leads to large jumps in enthalpy. These
big jumps in enthalpy are handled by decreasing the underrelaxation parameter (α) and
attempting to solve the equation again. With large jumps the underrelaxation parameter
decreases to the point that reaching a converged solution is very computationally
expensive. Figure 20 shows the enthalpy profiles for a grid spacing of 3 µm and 0.6 µm.
39
Figure 20  Grid Independent Study
After 1000 iterations, the final value of the underrelaxation parameter for the
enthalpy equation is 0.125 for a grid spacing of 3.0 µm. For a grid spacing of 0.6 µm, the
underrelaxation parameter is 6.94 x 10
18
for the energy equation in the enthalpy form.
With an underrelaxation so low, a converged solution is not reached.
For the purposes of the grid independent study, any grid spacing less than 6 µm
will yield a grid independent solution. Any grid spacing less than 0.6 µm requires
excessive computational time to get a converged solution. A grid spacing of 3 µm is
recommended for the HMX manifold onedimensional calculations.
4.2.5 Results
One of the advantages of using the ILDM approach over reduced mechanism
approaches is that all of the species profiles, even the minor radicals, are calculated.
Graphs of all the species profiles are included in Appendix A. To illustrate the
capabilities of the ILDM approach, the species profiles of NO, CO
2
, H
2
O, and the
40
temperature profile for the manifold and full kinetic mechanism are shown in Figures 21 
24. Reasonable agreement between the 3D manifold and full kinetics is found beyond
60 µm. Even though it is not shown in the plots, excellent agreement is found beyond
100 µm, as equilibrium is approached.
Figure 21  Results of OneDimensional, SteadyState Simulation
Figure 22  Results of OneDimensional, SteadyState Simulation
41
Figure 23  Results of OneDimensional, SteadyState Simulation
Figure 24  Results of OneDimensional, SteadyState Simulation
The manifold lags the detailed mechanism in the formation of the major product
species. This is due to a dark zone where NO is the limiting reactant and dominant
species near the surface. One method to overcome this lag is to include another species
in the manifold, such as the mass fraction of NO. Increasing the dimension of the
42
manifold poses new challenges in the creation and implementation of the manifold. A 4
D manifold is discussed in the next chapter.
43
Chapter 5 4D Manifold Validation with 1D Simulation
An additional reaction parameter can be added to increase the accuracy of the
ILDM approximation to the onedimensional laminar simulation. The addition of another
reaction parameter raises the complexity of the ILDM generation, and implementation
into a CFD code. Rather than review the complete procedure for the generation and
implementation of the 4D ILDM, only the major additions to the 3D ILDM approach
are discussed.
5.1 4D Manifold Direct Substitution
An indication of how well the higherdimensional manifold follows the detailed
kinetics is to substitute manifold values at every node using the solution of enthalpy,
mass fraction of N
2
, and mass fraction of NO from the detailed kinetics calculation.
Since the correct solution is already known, there is no reason to approximate the
solution. Direct substitution is not used to predict chemical kinetics, only to predict the
accuracy of the manifold. The power of direct substitution is that it can give a quick
indication of the accuracy of the manifold. Figure 25 shows the concentration of NO
versus distance from the surface using the detailed mechanism, 3D manifold, and 4D
manifold.
44
Figure 25 – Full Kinetic Mechanism Compared to Manifold Results
The 3D manifold agrees with the detailed kinetics above ~60 µm away from the
surface (see Figure 2119). The advantage of the 4D manifold over the 3D manifold is
that the 4D manifold is closer to the exact solution from 12 µm and larger. Overall, as
the dimension of the manifold is increased, the manifold more accurately represents the
detailed kinetics. The 4D manifold is expected to be more accurate than the 3D
manifold.
5.2 4D Manifold Implementation in 1D Laminar Simulation
5.2.1 The Equations
An additional equation (continuity equation of NO) must be solved to implement
the 4D manifold. The 4D HMX manifold is a function of enthalpy, pressure, mass
fraction of N
2
, and mass fraction of NO. The equations for enthalpy and mass fraction of
45
N
2
are the same as those in the 3D manifold implementation. The additional NO species
conservation equation is
NO NO NO mass
W Y
dx
d
n ω · (10)
where Y
NO
is the mass fraction of NO, ω
ΝΟ
is the reaction rate of NO, and W
NO
is the
molecular weight of NO.
5.2.2 Discretization of the Equations
The control volume approach is also used to discretize the NO species equation.
The resulting equation is equivalent to the discretized equation for N
2
(see Equation 7)
when NO is substituted for N
2
.
5.2.3 Iterative Solving Routine
As with the 3D manifold, the 4D manifold solving routine is implicit. The 4D
solving routine adds the solution to the NO profile directly after the solution to the N
2
profile. The iterative routine repeats until the solution converges (see Figure 17 and
Figure 26).
46
Figure 26  Iterative Solving Procedure
For the 3D manifold, an average value was taken to average out an oscillating
solution. However, for the 4D manifold the solution is more stable and therefore, an
average value is not necessary.
5.2.4 Grid Refinement for Enthalpy Profile
One of the causes of instability using a 3D manifold is the enthalpy solution. In
the region near the surface where the 3D manifold does not track the fast kinetics,
temperature oscillations result in unrealistic enthalpy values. When unrealistic enthalpy
values are computed, the enthalpy equation underrelaxation factor is automatically
decreased, and the variables are reinitialized. Without grid refinement, the under
relaxation factor can become so small that a grid independent solution is computationally
prohibitive (see Section 4.2.4). To overcome this problem, the iterative solving routine
includes a gridrefinement for the enthalpy profile. When an enthalpy profile is outside
of the manifold state space, the iterative solving routine is reinitialized with a coarser
47
enthalpy grid. A coarser enthalpy grid reduces the temperature oscillations and decreases
the computational cost. The solution procedure with enthalpy refinement is shown
Figure 27.
Figure 27  Data Flow with Enthalpy Grid Refinement
To prevent grid dependent results, a restriction is imposed on the maximum grid
spacing of the enthalpy profile. When the enthalpy grid spacing becomes greater than a
specified value, the enthalpy grid layout is restored to the original configuration and the
underrelaxation factor is decreased. Through a grid independent study, the maximum
allowable enthalpy grid spacing at 20 atm was found to be 30 µm.
48
5.2.5 Grid Independent Study
Figure 28 shows the calculated mass fraction of NO plotted versus distance (up to
20 µm) for the grid independent study. The values in the legend refer to the distance
between nodes.
Figure 28  Grid Independence Study (050 µm)
Even with significant grid refinement, the calculated NO profiles of the 4D
manifold solution and full kinetics solution do not produce gridindependent results in the
20 µm nearest the burning surface. This is most likely due to the issue that the 4D
manifold does not describe detailed kinetics in that region. As shown with direct
substitution (see Figure 25), the 4D manifold does not represent the fast reaction kinetics
in the nearsurface region.
A similar grid refinement study was performed on the region greater than 25 µm
(see Figure 29). The idea is to completely skip the region where the NO profile diverges.
This is accomplished by using the full kinetics solution at 25 µm as the initial point for
the 4D manifold solution.
49
Figure 29  Grid Independence Study (>25 µm)
When starting the solution at 25 µm, a grid spacing smaller than 0.7 µm appears
to be adequate.
5.2.6 A Model for NearSurface Kinetics
Because the 4D manifold does not accurately represent fast kinetics near the
burning surface, a model based on the full kinetics solution is used, and the 4D manifold
is employed where it accurately represents the full kinetic solution. Figure 30 shows the
calculated mass fraction of NO plotted as a function of distance for various starting points
using the combined approach. The values in the legend are the distance from the burning
surface where the 4D manifold is used. Based on the grid independent study, the grid
spacing is fixed at 0.25 µm.
50
Figure 30  NO Profile for Various Starting Points at 20 atm
The 4D manifold accurately represents detailed kinetics beyond 19 µm. Previous
results (see Figure 25) suggest good agreement beyond 12 µm. The cause of this 7 µm
discrepancy is unknown. Previous results, such as direct substitution, can only suggest an
approximate range where the manifold will produce good results. The real test of the 4D
manifold’s performance is through an iterative solution as shown in Figure 30.
A challenge for implementing the nearsurface model is recognizing the correct
point to begin solving with the 4D manifold. While at 20 atm, a starting point of 19 µm
appears to be adequate. However, as pressure changes the starting distance from the
surface should change. To evaluate this trend, 4D manifolds were generated at 5 and 60
atm. Figures 31 and 32 show the results of using these 4D manifolds in the 1D, steady
state laminar combustion code. The different curves represent different starting points
for the 4D manifold. Grid spacing of 0.6 µm was used at 5 atm, and 0.1 µm was used at
60 atm for the 4D manifold solutions.
51
Figure 31  NO Profile for Various Starting Points at 5 atm
Figure 32  NO Profile for Various Starting Points at 60 atm
At 5 atm, the 4D manifold accurately starts at ~92 µm. At 60 atm, the 4D
manifold accuracy is sufficient at ~6 µm. In the region before the 4D manifold is
sufficiently accurate, the NO solution diverges. With an upwind approximation, the
“downwind” solution is strongly influenced by deviations “upwind”. “Downwind” errors
52
caused “upwind” deviations are apparent in the curve with a starting point at 21 µm
(Figure 31) and the curve with a starting point at 5 µm (Figure 32).
One method to approximate the beginning point of the 4D manifold solution is
obtained by observing the enthalpy profile. Figure 33 shows the enthalpy profile at 60
atm. A grid spacing of 0.1 µm was used.
Figure 33  Enthalpy Profile at 60 atm
At 60 atm the 4D manifold represents the detailed kinetics after the maximum
enthalpy is reached. The curve starting at 5 µm shows major deviations while the curves
starting at 6 and 7 µm represent the enthalpy profile reasonably well. The difference is
that the latter curves begin after the maximum enthalpy. This same trend holds for 20
atm and 5 atm profiles. Therefore, the nearsurface model should extend to the maximum
enthalpy and the 4D manifold should be used thereafter.
53
5.2.7 Results
To illustrate the capabilities of the 4D manifold approach with a nearsurface
model, the species profiles of N
2
, CO
2
, H
2
O, and the temperature profile for the manifold
and full kinetic mechanism are shown in Figures 34  37. For comparison, the 3D
manifold results are also shown. For the 4D manifold, the calculations were performed
with a grid spacing of 0.25 µm and a starting location of 19 µm. For the 3D manifold
the grid spacing is 3 µm (see Section 4.2.4) and the starting location is 0 µm. All
calculations were performed at 20 atm.
Figure 34  Results of OneDimensional, SteadyState Simulation
54
Figure 35  Results of OneDimensional, SteadyState Simulation
Figure 36  Results of OneDimensional, SteadyState Simulation
55
Figure 37  Results of OneDimensional, SteadyState Simulation
Graphs of all the species profiles for steadystate combustion at 20 atm are
included in Appendix A. Generally, there is good agreement between the 4D manifold
and the full kinetics, even with the minor species. A few of the curves (i.e. CO
2
) have
discontinuities at ~25 µm (see Figure 35). This is due to discontinuities within the 4D
manifold at that point. The problem can be minimized by generating and using a
manifold with greater resolution.
56
57
Chapter 6 Container Simulation
The overall goal of the HMX manifold generation and testing is to incorporate the
ILDM approach in the gasphase of the CSAFE container simulation. Simulations have
been performed by CSAFE to investigate deformation of the container, cracking of the
solid high explosive, and heating of the solid high explosive. In addition, cookoff tests
(heating a container of high explosive until explosion) have been performed to validate
the models. A few of these studies are reported and referenced in this report.
It was suggested that one way to increase the accuracy of the container
simulations would be to include a simulation of the gas phase chemistry inside the
container through the ILDM approach. Further work is necessary to fulfill this objective.
This section discusses results from three simplified simulations of the gas phase with the
gas at equilibrium. Although the simulations do not implement an ILDM, they are
insightful with respect to heat transfer at the solidgas interfaces. Also, these simplified
simulations reveal information about gas flow and flame propagation within the
container. The intention of this work is to provide a foundation for future work in
implementing the HMX manifold.
6.1 CSAFE Container Simulation
As noted previously, the CSAFE project includes simulation of a container filled
with PBX9501. PBX9501 is an explosive with 95% energetic material (HMX) and 5%
binder (2.5% Estane and 2.5% BDNPF/A). Experimental tests have been performed to
determine the time to explosion of a heated container filled with PBX9501. A
58
cylindrical container (4inch diameter x 12inch length) filled with the explosive was
subjected to a high heat flux (~10
4
W/m
2
). Ignition times ranged from 1.62.4 minutes
(Eddings, Sarofim, Ciro, and Beckstead 2000).
A simulation by the University of Utah models the pressurization of the container
after ignition (CSAFE Program 2001). Figure 38 shows a stress simulation performed
on the container and high explosive.
Figure 38  Pressurization of Container
To reduce computational cost, the size of the container in the simulation is smaller
than the size of the experimental validation container. The outer shell is a steel container.
The interior is high explosive with temperature contours that show the thermal wave
penetration. A twodimensional cross section of the container shows the pressurization in
three snapshots as it progresses in time.
6.2 A PostIgnition Model with Gas Products at Equilibrium
The CSAFE container simulation includes heat transfer, a global mechanism for
HMX reaction rate, deformation of the container, and deformation of the high explosive.
In addition, a onedimensional ignition model is available to predict time to ignition
59
(Beckstead 1994). Work has also been performed on multidimensional effects of
ignition (Baer 1994). No known work has been performed to model multidimensional
effects after ignition for fast heating rates. This postignition study explores the multi
dimensional effect of flame spreading.
In this model, HMX properties are used to approximate those of PBX9501.
Figure 39 shows a simplified model of a gap that could exist between the container and
the HMX. This model is used to qualitatively investigate the point at which the HMX
will ignite.
Figure 39  Model of Container
In the proposed situation for the CSAFE simulation, the container is one meter
above a pool fire and receives an assumed heat flux (~10
4
W/m
2
) at the outside wall
(Beckstead 1999). This heat conducts through the steel container and into the HMX.
The thermal contact resistance between the steel casing and solid is equivalent to a 13
mm air gap (Eddings, et. al. 2000). The proposed CSAFE simulation predicts the heat
60
up, reaction of the HMX, pressurization, and deformation of the container. The proposed
simulation continues until the time of rupture of the container.
6.2.1 Heat Transfer Analysis
A simplified simulation of the container was performed to reveal the most likely
initiation point of the HMX reaction. A crosssection of the container was modeled in
Fluent™ to determine a realistic temperature profile. Figure 40 shows the overall cross
section of a 30 cm diameter cylinder after 2,000 seconds. An arbitrary gap, filled with
simulated combustion gases is ~2 cm at the widest part. In this heat transfer simulation,
the outside of the container is subjected to a constant temperature boundary condition of
550 K. The melting temperature of HMX is slightly greater than 550 K. The boundary
condition temperature of 550 K was selected to ensure the validity of the solid HMX
assumption.
61
Figure 40  Heat Transfer Simulation
The container is assumed to be made out of stainless steel (ANSI 316) with a
thermal conductivity of 13.4 W/mK. The heated solid HMX has a thermal conductivity
of 0.4 W/mK (Beckstead 1999, Bedrov, Smith, and Sewell 2000). The gap between the
container and solid HMX has the same properties as air, and the heat transfer coefficient
at the airsolid interface is assumed to be 25 W/m
2
K. This figure shows that the solid
HMX is an excellent insulator. Because of this, reaction will likely occur where the solid
HMX comes into contact with the container, and where the thermal contact resistance is
the lowest. Figure 41 shows the same heat transfer simulation zoomed in at the solidgas
interface.
62
Figure 41  Heat Transfer at the SolidGas Interface
Heat transfer by conduction from the container to the solid HMX is larger than the
heat transfer from the gas to the solid HMX. Accordingly, the highest temperature at the
solidgas interface is at the contact points.
6.2.2 Open System Simulation
In an open system reaction scenario the reaction occurs at the separation point and
proceeds around the outside of the solid HMX. Since the gap thickness is much less than
the container diameter, the angular effects are assumed to be negligible (Eddings, et. al.
2000). The simulation is approximated as a straight sheet of HMX covered by stainless
steel.
63
6.2.2.1 The Model
A simplified model of the HMX combustion in the container is shown in Figure
42.
Figure 42  Open System Model
In this simulation, ignition is assumed to have occurred at the left, vertical plane.
Heat feedback is propagating further burning into the preheated HMX near the
steel/HMX interface. This creates a moving burning surface, with combustion gases that
leave the burning surface at 3200 K. The combustion gas flows in a gap (assumed to be 5
mm wide) between the steel and unignited HMX, and transfers heat to both solids.
6.2.2.2 Ignition Model
If enough heat is transferred to the horizontal solid HMX it will ignite and begin
to burn, releasing additional combustion gases. Figure 43 shows an ignition line
generated from experimental heating tests for solid HMX (Beckstead 1999).
64
Figure 43  Ignition Data from Heating Tests
Knowing the heat flux to a surface and using data from Figure 43, the time to
ignition can be estimated. If the hot gases do not transfer enough heat to the unignited
HMX surface for ignition, then a rigorous combustion simulation should only have one
burning front. However, if the hot gases from one surface cause other HMX surfaces to
ignite then twodimensional flame spreading must be modeled.
Using a Lagrangian approach, the regressing burning surface is taken to be
stationary. From a onedimensional, steadystate combustion simulation developed at
BYU (Davidson 1996), the burning rate of HMX preheated to 550K at 20 atm is
calculated to be 1.22 gm/cm
2
sec. This is equivalent to a surface regression rate of 0.71
cm/sec, based on the HMX density of 1.9 gm/cm
3
(Beckstead 1999). With the burning
surface in a stationary reference frame, the steel and solid HMX are assumed to move at
the surface regression rate.
The gap between the steel and HMX is taken to be 5 mm, while a steel thickness
of 150 mm (the thickness is arbitrarily large to allow adiabatic boundary conditions) and
65
an HMX thickness of 100 mm are assumed. The gap thickness is much smaller than the
solid thickness so that adiabatic boundary conditions can be employed around the outer
solid edges. The regressing surface has an inflow boundary condition of 12.2 kg/seccm
2
and the gas exit an outflow boundary condition. Since the Reynolds number for the
flowing gases is under 100, all flow is laminar. Consequently, a locally parabolic
assumption at the gas outflow is valid.
6.2.2.3 Grid Independent Study
A simulation was performed in Fluent™ to determine the heat flux to the solid
HMX surface. The grid layout is shown in Figure 44.
Figure 44  Grid Layout
In order to reduce the total number of nodes used, the grid is clustered in the areas
of high temperature gradients. Gambit™ was used to create the unstructured grid. An
unstructured grid allows for a very high density of nodes to be clustered around the gas
chamber and fewer nodes to be used at the extremes of the model. A grid independent
66
study dictates the use of 200 nodes with grid clustering around the gas cavity where there
are high temperature and velocity gradients.
Various discretization schemes were also studied. A discretization scheme of 2
nd
order for the pressure and QUICK (Quadratic Upstream Interpolation for Convection
Dominated Kinetics) for the energy and momentum equations was selected. QUICK and
2
nd
order were chosen because they are higher order and tend to be more accurate than
lower order schemes such as 1
st
order, upwind. Also, SIMPLE (SemiImplicit Method
for PressureLinked Equations) was the method used to couple the pressure and velocity
terms in the continuity and momentum equations.
6.2.2.4 Simulation Results
Figure 45 illustrates the temperature contours (in Kelvin), showing the area of
high temperature gradients around the flowing channel.
67
Figure 45  Temperature Contours
The heat transfer from the combustion gases to the solid surfaces is greatest near
the reacting surface. As the gases cool in the boundary layer, the heat flux to the solid
surfaces decreases.
Time for ignition data is based on a constant heat flux. In the simulation, the heat
flux has a stationary profile. However, with a Lagrangian approach, the surface is
moving and the heat flux that an individual point sees is changing. To account for the
movement of the surfaces, the heat flux is computed through the following expression:
t
t q
q
n
i
i i
∆
∆
·
∑
·1
' '
' ' (13)
where q″
i
is the instantaneous heat flux at a node, ∆t
i
is the time the differential surface is
subjected to the heat flux, and ∆t is the time to travel from the burning surface to the final
68
position. Figure 46 shows the average and instantaneous heat fluxes across the gas/HMX
interface. The distance on the xaxis refers to the distance from the regressing surface.
Figure 46  Comparison of Heat Fluxes
Using the average heat flux, the distance to ignition is computed by locating the
point (in space) at which the heat flux is sufficient to ignite the surface. Ignition data
relates heat flux to time for ignition (see Figure 43). The time of the incident heat flux is
related to the distance away from the regressing surface by the surface regression velocity
(∆x = v ∆t ). Figure 47 shows the calculated results of average heat flux versus distance
along the gas/HMX interface, compared to the empirical data correlation.
69
Figure 47  Distance to Ignition for a 5 mm Gap
At 0.23 mm the average heat flux reaches a value sufficient for ignition. This
indicates that with the open system geometry, when ignition occurs at one point, the
resulting combustion gases will ignite other surfaces nearby.
6.2.2.5 Gap Thickness Parametric Study
A parametric study was performed to study the effect of gap thickness on flame
spreading. A preheated band of HMX varies in thickness depending on the rate of
heating to the outside of the container. At high heating rates (~10
5
W/m
2
) the thermal
wave only penetrates ~23 mm into the solid HMX before ignition. At low heating rates
(~10
4
W/m
2
) the thermal wave penetrates up to ~40 mm into the solid HMX before
ignition (Beckstead 1999). After ignition the burning front preferentially burns through
the preheated HMX because of the 40% faster burning rate over nonpreheated HMX
(0.72 cm/sec compared to 0.52 cm/sec). By burning the preheated HMX, a gap forms
between the steel and unignited HMX. Simulations were performed with gap widths of 1
70
mm, 2 mm, 5 mm and 10 mm. Figure 48 shows the results of the simulations with the
ignition heat flux line.
Figure 48  Ignition Prediction for Various Gap Widths
The ignition distance for a gap of 10 mm is ~0.55 mm while a gap of 1 mm
requires only ~0.07 mm to ignite. This study shows that as the gap width increases,
ignition of the horizontal surface is delayed. The important point is that flame spreading
is significant for all gap widths tested.
6.2.2.6 Conclusions for the Open System Simulation
The results of this study indicate that the single burning front assumption is not
valid after ignition of the HMX. Once a single point ignites, the combustion gases ignite
other surfaces nearby, regardless of the width of the gap. This spreading effect leads to
accelerated combustion through flame propagation. In a rigorous simulation of the high
explosive combustion, twodimensional flame spreading must be modeled.
71
6.2.3 Closed System Simulation
For an open system flame spreading is significant. However, in a closed system
the velocity of the gases is lower and therefore, flame spreading may not be significant.
Another important factor is pressurization. When solid HMX begins to react in the
closed system, the container pressurizes. As the container pressurizes, the reaction rate of
HMX increases. The closed system exhibits positive feedback, and the reaction rate
accelerates.
There is a void volume in the center of the HMX annulus that will accumulate
combustion gases and slow down the pressurization of the container. In addition, the
volume of the container can expand. This expansion is due to a ballooning effect of the
container walls. Experimental results show that the volume of the container can increase
by as much as 25% before rupture (Eddings, et. al. 2000). The combined void volume of
the annulus and expansion of the container decreases the rate of pressurization and delays
the rupture of the container.
6.2.3.1 The Model
The closed system model of the HMX combustion in the container is shown in
Figure 49.
72
Figure 49  Closed System Model
This simulation is similar to the open system simulation in that the left surface is
assumed to regress at the steady state burning rate of 0.72 cm/sec with combustion gases
leaving the burning surface at the equilibrium temperature of 3200 K. The combustion
gases enter the pocket (assumed 5 mm x 10 mm) between the steel and unignited HMX,
and transfer heat to both solids. The difference between this simulation and the open
system simulation is that the right boundary is closed. With a closed system,
pressurization occurs and time must be added as another variable in the calculation. The
simulation was run for 0.01 seconds with time steps of 0.00005 seconds. Larger time
steps lead to solver instability in Fluent™.
In an unsteady calculation, initial conditions must be specified. The initial
temperature for the simulation is set at 550 K, the melting temperature of HMX. The
initial pressure is 1 atm, and the combustion gases are modeled as an ideal gas with the
same properties as air.
The surface regression enlarges the area of the pocket by less than 1% during the
total simulation time of 0.01 seconds. This enlarging of the pocket is assumed to be
73
negligible and, therefore, all vertical boundaries are taken to be stationary. The most
significant enlargement of the pocket will occur as the pressure increases. High pressure
in the gap will deform the steel or HMX solid and cause them to separate. This
deformation is beyond the scope of the closed system simulation.
6.2.3.2 Ignition Model
The ignition model for the closed system simulation is the same as the open
simulation ignition model (discussed in Section 6.2.2.2).
6.2.3.3 Simulation Results
As in the open system simulation, the closed system surface regression rate is
assumed to be 1.22 gm/cm
2
sec. With that surface regression rate and the given
geometry, the ideal gas pressure is calculated to increase linearly from 1 atm initially to
150 atm in 0.01 seconds. Because of unrestricted flow within the closed pocket, the flow
should be laminar and there should not be recirculation of the gases. Figure 50 shows the
velocity profile at 0.00005 seconds with the vectors corresponding to the velocity
direction and magnitude (in m/s).
74
Figure 50  Initial Velocity Profile
The combustion gases reach equilibrium §µm from the burning surface.
Because the gases reach equilibrium quickly, thermal effects are more significant than
chemical effects, such as caused by radicals, in igniting the HMX. Figure 51 shows the
temperature contours (in Kelvin) after 0.01 sec of simulation time, showing the area of
high thermal gradients within and around the pocket.
75
Figure 51  Temperature Profile in the Closed System Simulation
The temperature rises above the equilibrium temperature of 3200 K because of
pressurization. An adiabatic system with ideal gas undergoing pressurization increases in
temperature according to the following equation.
γ
γ 1
1
2
1
2
−
,
`
.

·
P
P
T
T
(15)
Even though this simulation is not an adiabatic system, the same trend applies. The
pressurization increases the temperature of the gases. The rise in temperature also
increases the heat flux to the solid HMX and steel. In the simulation, a pocket of gas
approaches 7500 K due to pressurization. As the temperature increases, many of the
equilibrium products will dissociate into the elements and moderate the rise in
temperature. A more rigorous simulation, such as an ILDM gasphase calculation
76
(provided the full kinetic mechanism accounts for dissociation), would account for
species dissociation, and would result in a much lower temperature.
Another shortcoming of the closed system simulation is that phase changes are
not modeled. In the simulation, the surface temperature of the solid HMX rises above
1500 K. The surface temperature of the liquidgas interface in the steadystate one
dimensional model is 775 K. This implies that it is near the HMX boiling point and that
vaporization of the HMX should occur in the closed system simulation. A more rigorous
simulation would account for the vaporization and reaction of HMX in the gas phase.
The heat flux to the unignited HMX is calculated to predict flame spreading in the
closed pocket. Figure 52 shows the results of this analysis.
Figure 52  Heat Flux Analysis for Flame Propagation
The heat flux along the horizontal surface of HMX does not reach a value
sufficient for ignition within 0.01 seconds.
77
6.2.3.4 Conclusions for the Closed System Simulation
The flame will not spread within the pocket for this model because the heat flux
to the solid HMX does not reach ignition values for the horizontal surface. Extrapolating
the results suggests that ignition may occur around 0.02 seconds. By this time the
pressure within the pocket is expected to increase to ~400 atm. The total time of the
closed system simulation would be dictated by the yield stress of the steel container.
After the pressure in the container is sufficient to overcome the yield stress, the fixed
boundary closed system simulation is no longer valid.
78
79
Chapter 7 Conclusions
One of the major obstacles to manifold implementation in CFD codes is the lack
of automation for manifold generation and implementation. This report describes a
method that automates the manifold process. The automated manifold generation is
accomplished through the Chemkin paradigm of interchangeable reaction mechanisms.
The manifold generation code reads a mechanism file, simulates a PSR reactor, calculates
the composition and temperature of a mixture of reactants and PSR products, and
computes the reaction trajectory of that mixture in a batch reactor. The batch reactor
results are then used to generate the manifold. As new mechanisms are developed for
HMX or as other solid propellants are modeled, a new manifold can be generated simply
by switching out mechanism files.
This project also provides an additional example of using the ILDM method. Up
to this point, the applications of the ILDM method have been with simple combustion
reactants such as shortchained hydrocarbons, H
2
, and O
2
. This project is the first known
attempt to use the manifold method for a complex reactant molecule such as HMX.
A 3D (in enthalpy, pressure, and mass fraction of N
2
) manifold was implemented
in a 1D (in space) steadystate simulation. There were large errors near the surface of
the burning HMX (< 60 µm) and good agreement with full kinetics beyond that.
A 4D (in enthalpy, pressure, mass fraction of N
2
, and mass fraction of NO)
manifold increases the accuracy of the approximation. At 20 atm the 4D manifold
provides an accurate representation of detailed kinetics starting at the peak of the
80
enthalpy profile (~19 µm). Within 19 µm of the burning surface, a nearsurface model is
required. The nearsurface model is generated by extracting the first 19 µm of the
steadystate, 1D full kinetics solution.
If coarse gridding (>60 µm) were used in the CSAFE simulation, the 3D
manifold would be sufficient. If coarse gridding is not sufficient, the 4D manifold with
a nearsurface model should be used.
With the ILDM method, only a limited number of differential equations (equal to
the dimension of the manifold) must be solved. This speeds up the convergence by
eliminating much of the stiffness in the differential equations. In this project,
computational cost was reduced by ~10 times for the ILDM method.
The multidimensional effects of HMX combustion were explored with a 2D (in
space) simulation using gases that approximate the properties of HMX combustion gases
at equilibrium. In the opensystem simulation, flame spreading is shown to be
significant. In the closedsystem simulation, flame spreading is delayed by the restriction
to flowing gases.
81
Chapter 8 Future Work
The project described in this report is the beginning of the process to generate,
implement, and evaluate an ILDM for HMX in the University of Utah’s 3D CFD code,
ARCHES. The same techniques that were used in this research project to implement the
manifold method into the onedimensional (in space) combustion code will be used to
implement the manifold method into ARCHES. Future work will need to be performed
by another researcher to accomplish this goal.
Modeling the reaction of HMX in the gas space of a container will give insight to
heat feedback, local quenching, and flow of gases in order to predict time to detonation or
deflagration. The simplified open and closed system simulations performed in this
project offer insights into some of the physical processes, such as heat transfer.
However, a more accurate representation of the open and closed systems should include
species involved in chemical reactions as well as deformation of the HMX and steel due
to pressurization. The kinetics calculation can be accomplished by implementing the
ILDM for HMX. Deformation of the container and its contents is currently being
investigated by other CSAFE researchers.
82
83
Chapter 9 References
Baer, M. R., Gross R. J., Gartling D. K., and Hobbs M. L., "Multidimensional Thermal
Chemical Cookoff Modeling", 1994 JANNAF Propulsion Systems Hazards
Meeting, 1994.
Beckstead, M. W., "An Ignition Model for Modern Double Base Propellants", 31st
JANNAF Combustion Meeting, 1994, CPIA No. 620, Vol. II, pp. 163180.
Beckstead, M. W. and Hendershot R. J., "Calculated Conditions for Fast CookOff", 36th
JANNAF Combustion Meeting, 1999, CPIA No. 691, Vol I, pp. 123132.
Bedrov, D., Smith, G. D., and Sewell, T. D., "Thermal Conductivity of Liquid
Octahydro1,3,5,7tetranitro1,3,5,7tetrazocine (HMX) from Molecular Dynamics
Simulations", Chemical Physics Letters, 2000, 324, pp. 6468.
Blasenbrey, T., Schmidt, D., and Maas, U., "Automatically simplified Chemical Kinetics
and Molecular Transport and its Application in Premixed and NonPremixed
Laminar Flame calculations", TwentySeventh Symposium (International) on
Combustion, 1998, The Combustion Institute, Pittsburgh, pp. 505511.
Bongers, H. Reduced Reaction Mechanisms, an Introduction. URL:
http://www.combustion.tue.nl/~hbongers/, 2002.
Burden, R. L. and Faires, J. D. Numerical Analysis. Pacific Grove, CA: Brooks/Cole
Publishing Company, 1997.
CSAFE Program. CSAFE, Center for the Simulation of Accidental Fires and
Explosions. URL: http://www.csafe.utah.edu, 2001.
Davidson, J. E. and Beckstead M. W., "A ThreePhase Model of HMX Combustion",
26th Symposium (Int'l) on Combustion, 1996, The Combustion Institute, pp.
19891996.
Eddings, E. G., Sarofim, A., Ciro, W., and Beckstead, M. W., "Fast CookOff Tests of an
HMXbased Propellant in Pool Fires", 19th JANNAF Propulsion Systems
Hazards Meeting, 2000, CPIA No. 704, Vol I, pp. 6372.
Glarborg, P., Kee, R. J., Grcar, J. F., and Miller, J. A., "PSR: A Fortran Program for
Modeling WellStirred Reactors", Sandia Report, 1992, SAND868209, UC4.
84
Kee, R. J., Rupley, F. M.., and Miller, J. A., "A Fortran Chemical Kinetics Package for
the Analysis of Gas Phase Chemical Kinetics", Sandia Report, 1989, SAND89
8009B, UC706.
Lam, S. H. and Goussis, D. A., "Understanding Complex Chemical Kinetics with
Computational Singular Perturbation", TwentySecond Symposium
(International) on Combustion, 1988, The Combustion Institute, Pittsburgh, pp.
931941.
Maas, U. and Pope, S. B., "Implementation of Simplified Chemical Kinetics Based on
Intrinsic LowDimensional Manifolds", TwentyFourth Symposium
(International) on Combustion, 1992a, The Combustion Institute, Pittsburgh, pp.
103112.
Maas, U. and Pope, S. B., "Simplifying Chemical Kinetics: Intrinsic LowDimensional
Manifolds in Composition Space", Combustion and Flame, 1992b, 88, pp. 239
264.
Maas, U. and Pope, S. B., "Laminar Flame Calculations Using Simplified Chemical
Kinetics Based on Intrinsic LowDimensional Manifolds", TwentyFifth
Symposium (International) on Combustion, 1994, The Combustion Institute,
Pittsburgh, pp. 13491356.
Niemann, H., Schmidt, D., and Maas, U., "An Efficient Storage Scheme for Reduced
Chemical Kinetics Based on Orthogonal Polynomials", Journal of Engineering
Mathematics, 1997, 31, pp. 131142.
Norris, A. T. and Pope, S. B., "Modeling of Extinction in Turbulent Diffusion Flames by
the VelocityDissipationComposition PDF Method", Combustion and Flame,
1995, 100, pp. 211220.
Pope, S. B., "Computationally Efficient Implementation of Combustion Chemistry Using
In Situ Adaptive Tabulation", Combustion Theory Modeling, 1997, 1, pp. 4163.
Tomlin, A. S., Turanyi, T., and Pilling, M. J., "Mathematical Tools for the Construction,
Investigation and Reduction of Combustion Mechanisms", Chap. 4 in
Comprehensive Chemical Kinetics edited by Compton, R. G. and Hancock, G.,
LowTemperature Combustion and Autoignition, 1997, 35, pp. 293437.
Xiao, K., Schmidt, D., and Maas, U., "PDF Simulation of Turbulent NonPremixed
CH
4
/H
2
Air Flames Using Automatically Reduced Chemical Kinetics", Twenty
Seventh Symposium (International) on Combustion, 1998, The Combustion
Institute, Pittsburgh, pp. 10731080.
Yang, B. and Pope, S. B., "An Investigation of the Accuracy of Manifold Methods and
Splitting Schemes in the Computational Implementation of Combustion
Chemistry", Combustion and Flame, 1998a, 112, pp. 1632.
85
Yang, B. and Pope, S. B., "Treating Chemistry in Combustion with Detailed
Mechanisms—In Situ Adaptive Tabulation in Principal Directions—Premixed
Combustion", Combustion and Flame, 1998b, 112, pp. 85112.
86
87
Appendix A
Results from the 1D, steadystate simulation using a 3D manifold, a 4D manifold, and
full kinetics.
List of Graphs in Appendix A
Properties Intermediate Species (cont.)
TEMPERATURE CNO
ENTHALPY NCO
RATE N2 NCN
Major Products C2N2
H2O CN
CO2 N2O
CO HCN
N2 HONO
Intermediate Species HNO
NO NNH
H2COHNNO2 NH3
HNC NH2
HMXRO NH
HMXR NO2
HMX N
H2CNNO2 HCO
H2CNNO CH2O
H2CNO H2O2
H2CNH HO2
H2CN OH
HNO3 H
NO3 O
HCNO O2
HOCN H2
HNCO
88
0
500
1000
1500
2000
2500
3000
3500
4000
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
T
E
M
P
E
R
A
T
U
R
E
(
K
)
3D Manifold
4D Manifold
Full Kinetics
0.00E+00
5.00E+09
1.00E+10
1.50E+10
2.00E+10
2.50E+10
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
E
N
T
H
A
L
P
Y
(
e
r
g
/
g
m
)
3D Manifold
4D Manifold
Full Kinetics
89
0
2
4
6
8
10
12
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
R
A
T
E
N
2
(
m
o
l
/
c
m
3
/
s
e
c
)
3D Manifold
4D Manifold
Full Kinetics
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
O
3D Manifold
4D Manifold
Full Kinetics
90
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
O
2
3D Manifold
4D Manifold
Full Kinetics
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
O
3D Manifold
4D Manifold
Full Kinetics
91
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
2
3D Manifold
4D Manifold
Full Kinetics
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
O
3D Manifold
4D Manifold
Full Kinetics
92
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
O
H
N
N
O
2
3D Manifold
4D Manifold
Full Kinetics
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
N
C
3D Manifold
4D Manifold
Full Kinetics
93
0.00000
0.00005
0.00010
0.00015
0.00020
0.00025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
M
X
R
O
3D Manifold
4D Manifold
Full Kinetics
0.00000
0.00005
0.00010
0.00015
0.00020
0.00025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
M
X
R
3D Manifold
4D Manifold
Full Kinetics
94
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
M
X
3D Manifold
4D Manifold
Full Kinetics
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
N
N
O
2
3D Manifold
4D Manifold
Full Kinetics
95
0.00000
0.00002
0.00004
0.00006
0.00008
0.00010
0.00012
0.00014
0.00016
0.00018
0.00020
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
N
N
O
3D Manifold
4D Manifold
Full Kinetics
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
N
O
3D Manifold
4D Manifold
Full Kinetics
96
0.00000
0.00005
0.00010
0.00015
0.00020
0.00025
0.00030
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
N
H
3D Manifold
4D Manifold
Full Kinetics
0.000
0.005
0.010
0.015
0.020
0.025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
C
N
3D Manifold
4D Manifold
Full Kinetics
97
0.00
0.00
0.00
0.01
0.01
0.01
0.01
0.01
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
N
O
3
3D Manifold
4D Manifold
Full Kinetics
0.00E+00
5.00E+03
1.00E+04
1.50E+04
2.00E+04
2.50E+04
3.00E+04
3.50E+04
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
O
3
3D Manifold
4D Manifold
Full Kinetics
98
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
C
N
O
3D Manifold
4D Manifold
Full Kinetics
0.0000
0.0005
0.0010
0.0015
0.0020
0.0025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
O
C
N
3D Manifold
4D Manifold
Full Kinetics
99
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
N
C
O
3D Manifold
4D Manifold
Full Kinetics
0.00000
0.00002
0.00004
0.00006
0.00008
0.00010
0.00012
0.00014
0.00016
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
N
O
3D Manifold
4D Manifold
Full Kinetics
100
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
C
O
3D Manifold
4D Manifold
Full Kinetics
0.00000
0.00002
0.00004
0.00006
0.00008
0.00010
0.00012
0.00014
0.00016
0.00018
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
C
N
3D Manifold
4D Manifold
Full Kinetics
101
0.000
0.005
0.010
0.015
0.020
0.025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
2
N
2
3D Manifold
4D Manifold
Full Kinetics
0.0000
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
N
3D Manifold
4D Manifold
Full Kinetics
102
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
2
O
3D Manifold
4D Manifold
Full Kinetics
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
C
N
3D Manifold
4D Manifold
Full Kinetics
103
0.000
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
O
N
O
3D Manifold
4D Manifold
Full Kinetics
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
N
O
3D Manifold
4D Manifold
Full Kinetics
104
0.00E+00
5.00E07
1.00E06
1.50E06
2.00E06
2.50E06
3.00E06
3.50E06
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
N
H
3D Manifold
4D Manifold
Full Kinetics
0.000
0.001
0.002
0.003
0.004
0.005
0.006
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
H
3
3D Manifold
4D Manifold
Full Kinetics
105
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
0.0014
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
H
2
3D Manifold
4D Manifold
Full Kinetics
0.00000
0.00005
0.00010
0.00015
0.00020
0.00025
0.00030
0.00035
0.00040
0.00045
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
H
3D Manifold
4D Manifold
Full Kinetics
106
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
O
2
3D Manifold
4D Manifold
Full Kinetics
0.00000
0.00001
0.00002
0.00003
0.00004
0.00005
0.00006
0.00007
0.00008
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
N
3D Manifold
4D Manifold
Full Kinetics
107
0.00000
0.00002
0.00004
0.00006
0.00008
0.00010
0.00012
0.00014
0.00016
0.00018
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
C
O
3D Manifold
4D Manifold
Full Kinetics
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
C
H
2
O
3D Manifold
4D Manifold
Full Kinetics
108
0.00E+00
5.00E06
1.00E05
1.50E05
2.00E05
2.50E05
3.00E05
3.50E05
4.00E05
4.50E05
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
O
2
3D Manifold
4D Manifold
Full Kinetics
0.00E+00
5.00E06
1.00E05
1.50E05
2.00E05
2.50E05
3.00E05
3.50E05
4.00E05
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
O
2
3D Manifold
4D Manifold
Full Kinetics
109
0.000
0.005
0.010
0.015
0.020
0.025
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
O
H
3D Manifold
4D Manifold
Full Kinetics
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
0.0014
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
3D Manifold
4D Manifold
Full Kinetics
110
0.0000
0.0005
0.0010
0.0015
0.0020
0.0025
0.0030
0.0035
0.0040
0.0045
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
O
3D Manifold
4D Manifold
Full Kinetics
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
O
2
3D Manifold
4D Manifold
Full Kinetics
111
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0 10 20 30 40 50 60 70 80 90 100
DISTANCE (µm)
H
2
3D Manifold
4D Manifold
Full Kinetics
BRIGHAM YOUNG UNIVERSITY
GRADUATE COMMITTEE APPROVAL
of a thesis submitted by John D. Hedengren
This thesis has been read by each member of the following graduate committee and by a majority vote has been found satisfactory.
__________________________ Date __________________________ Date __________________________ Date
_________________________________________ Merrill W. Beckstead, Chair _________________________________________ Jennifer P. Spinti _________________________________________ Larry L. Baxter
ii
BRIGHAM YOUNG UNIVERSITY
As chair of the candidate’s graduate committee, I have read the dissertation of John D. Hedengren in its final form and have found that (1) its format, citations, and bibliographical style are consistent and acceptable and fulfill university requirements; (2) its illustrative materials including figures, tables, and charts are in place; and (3) the final manuscript is satisfactory to the graduate committee and is ready for submission to the university library.
______________________________ Date
____________________________________ Merrill W. Beckstead Chair, Graduate Committee
Accepted for the Department ____________________________________ Thomas H. Fletcher Graduate Coordinator
Accepted for the College ____________________________________ Douglas M. Chabries Dean, College of Engineering and Technology
iii
7Tetrazocine) mechanism that contains 44 species and 232 reactions. A onedimensional (in space) BYU combustion program has been used to compare the three. A threedimensional manifold tracked detailed chemistry based on enthalpy.3. and implement an Intrinsic LowDimensional Manifold (ILDM) has been developed.ABSTRACT IMPLEMENTATION OF AUTOMATICALLY SIMPLIFIED CHEMICAL KINETICS THROUGH INTRINSIC LOWDIMENSIONAL MANIFOLDS FOR GASEOUS HMX John D.7Tetranitro1.5.and fourdimensional manifolds with a full kinetic mechanism. Simulations show that the threedimensional manifold is an adequate representation of the full kinetic mechanism away from the reacting surface (> 60 µm at 20 atm). A fourdimensional manifold adds the mass fraction of NO. Hedengren Department of Chemical Engineering Master of Science An automated method to generate. validate. pressure. and mass fraction of N2. This method has been applied to a detailed gaseous HMX (Octahydro1.3. The fourdimensional manifold is an adequate representation of the iv .5.
One application of interest is the University of Utah’s threedimensional (in space) CFD code. Further work is necessary to implement the ILDM approach in similar container simulations. this advantage can be offset by the development time required to create and implement the ILDM method. An advantage of the ILDM method is that computational time is reduced by an order of magnitude. However. The methodology used to implement the fourdimensional ILDM into the onedimensional (in space) combustion code is available to be used in other applications.full kinetic mechanism closer to the reacting surface (>19 µm at 20 atm). Another portion of this project is twodimensional (in space) simulations that model high explosives in a container. it gives insight to meaningful simulations that can be performed with an ILDM. Heat feedback is calculated to investigate flame propagation and burning characteristics of the solid HMX. The simulations explore heat feedback of equilibrium gases in a heated container filled with solid HMX. v . Even though this portion of the project does not implement an ILDM. ARCHES.
In addition.ACKNOWLEDGEMENTS I would like to thank all the people who have helped me with this project. Also. I appreciate opportunities he has given me to present my work at a conference and poster session. Beckstead provided helpful comments and guidance from the start. Ephraim Washburn and Karl Meredith have both contributed by helping me understand the computer codes available to our research group. Ben Hardy. especially my thesis advisor. The occasional evening or Saturday spent without me is a tribute to her support. Jennifer P. I would like to thank Dr. Ryan Stevens. Phillip J. Dr. exploring tangent ideas. I would especially like to thank the people at the University of Utah’s CSAFE program for funding through this project. Merrill W. Dr. I have thoroughly enjoyed the autonomy that he has given me while working on this project. In addition. Dr. Beckstead. Sarah. and Matt Gross have each helped at various points in the development of this project by generating plots. and by supplying stress relief for the research group. Ephraim’s suggestions were invaluable during the project’s development. vi . Spinti for the practical suggestions in developing the manifold code. My wife. has been supportive when deadlines approached. Smith contributed seminal ideas and helpful comments throughout this project.
...................................3 Decoupling Fast Chemical Time Scales...3 Iterative Solving Routine ........................2......6 2.............................2.................14 2........20 3..................25 3...........1.......................5 2...................1 Reduced Mechanism Methods .........5 OneDimensional Manifolds..............34 4........................................................................................................2 Intrinsic LowDimensional Manifold (ILDM)............................33 The Equations ...............1............2..................................37 vii .....................4 Grid Independent Study ......................................11 2..................................................................9 2....................................................................................................................................29 4....10 2..........................8 Examples of Manifold Implementation....................1.........4 Manifold Dimension ..21 3.......2 Starting Points in the Composition Space .............................7 2.................................25 3.............1.....2 Species Lumping Procedures ..................................................................................2.................................................15 2......1....................................................................................2 Eigenvector Analysis.............................................................................................................19 3.................................................7 HigherDimensional Manifolds ...................................1.......................9 2...........4 Manifold Identification .....................................27 CHAPTER 4 3D MANIFOLD VALIDATION WITH 1D SIMULATION .2......................................2 Discretization of the Equations .2 Manifold Storage and Retrieval .........................................................................5 Choice of Parameterization ................1.......................................................................................1 4....................................................................................................................4..............................................................1........31 4...............................2.................1 Graphical Method .................................6 2....1.......................................12 2....................................................................2......................2................................13 2.......................................................................7 2......2....................................1 Dimension of the Manifold ...................................................16 CHAPTER 3 MANIFOLD GENERATION FOR GASEOUS HMX.............26 3.................................................2 3D Manifold Implementation in 1D Laminar Simulation .......1 The Equations ...........................................3 Reaction Progression....................3 Computational Singular Perturbation (CSP) Technique ........1 3D Manifold Direct Substitution .......................6 TwoDimensional Manifolds ......................................................................................2.........................................................9 Recent Improvements to the Manifold Method ...........................................................................................................................2...1 Manifold Generation .......................................................................2.............34 4.......................1....................................................35 4.........................................................................22 3...................26 3.....................5 2...............Table of Contents CHAPTER 1 INTRODUCTION ..............4................................................................2 Chemical and Physical Process Time Scales ..........................................1 Sensitivity Analysis..................................................................1 CHAPTER 2 LITERATURE REVIEW .2.................20 3..................................................
.........2..................................................................................58 6.............60 6...................................2........................................2.............................................3................6 Conclusions for the Open System Simulation ..................................................2 A PostIgnition Model with Gas Products at Equilibrium ....................................46 5....2..............................2............................................................................................................53 CHAPTER 6 CONTAINER SIMULATION .................................................6 A Model for NearSurface Kinetics ..................81 CHAPTER 9 REFERENCES ............................2................3 Closed System Simulation...............................79 CHAPTER 8 FUTURE WORK.....57 6..........................................................................................65 6.............2.............2......3.....................1 The Model..............................................3 Iterative Solving Routine.............45 5.....71 6....................2 Ignition Model ............71 6..........3 Simulation Results ...................2 Open System Simulation .....5 Results ..................................................................................2.............2...............39 CHAPTER 5 4D MANIFOLD VALIDATION WITH 1D SIMULATION ............................2.........................................43 5....................2 Ignition Model .............................................................................2 4D Manifold Implementation in 1D Laminar Simulation.2............1 CSAFE Container Simulation ..................................................................70 6................................................1 The Model........................................................................................................................................................2..............................3.....................................1 Heat Transfer Analysis................................................................................................................................................................83 APPENDIX A............................43 5......2.........................................................................4 Grid Refinement for Enthalpy Profile .......48 5.................7 Results ...2.......................77 CHAPTER 7 CONCLUSIONS ............1 The Equations.3 Grid Independent Study ......2.....................................2......................................69 6.................................4 Conclusions for the Closed System Simulation .......2 Discretization of the Equations ......................2.............................57 6....2...2............49 5..........5 Grid Independent Study...63 6.................................................................................2.............................................................73 6....................................................4 Simulation Results .......4............................................................................44 5.....................44 5...............................................2...................1 4D Manifold Direct Substitution ...2..................................................................................................................73 6..........................63 6...................2..................................................45 5..5 Gap Thickness Parametric Study ..3.............................66 6.....................................................2......87 viii .........................................................2..........................................2.............................................................................................................62 6.........................................................
..... 51 Figure 33 ............ 46 Figure 27 ...... 34 Figure 17 ... 50 Figure 31 ........................... 12 Figure 3 ................ 47 Figure 28 .........................................Grid Independence Study (>25 µm)................................................ 15 Figure 5 ...............Reaction Path for Manifold Generation ............. SteadyState Simulation ....... 32 Figure 15 ............NO Profile for Various Starting Points at 60 atm ....... 22 Figure 6 – Generating a Gas Phase Chemkin Link File........NO Profile for Various Starting Points at 5 atm ....................................... 37 Figure 19 .......................................................... 41 Figure 24 .............................................. SteadyState Simulation ....Enthalpy Profile at 60 atm........................ 48 Figure 29 ..... 55 Figure 38 ...............................................MultiDimensional Manifold ....................................................... 58 Figure 39 .............................................. SteadyState Laminar Simulation ......................................................................Results of OneDimensional.......... 36 Figure 18 ......................................... SteadyState Simulation .......................................................................................... 44 Figure 26 .... 54 Figure 36 ............................................. 41 Figure 25 – Full Kinetic Mechanism Compared to Manifold Results......................Oscillations in the Final Solution..................................................................Results of OneDimensional...Results of OneDimensional......... 31 Figure 14 – Direct Substitution of 3D Manifold .....................Flow Time Scale ............ 30 Figure 12 – Generating a Transport Properties Chemkin Link File ............................ 24 Figure 9 .................................... 39 Figure 21 ................................... 14 Figure 4 .................................Results of OneDimensional.........................................Model of Container .................................... 33 Figure 16 .......Results of OneDimensional.... SteadyState Simulation .................. 23 Figure 8 ...............Pressurization of Container .....................TwoDimensional Manifold.Batch Reactor Results .................................................................................................. 53 Figure 35 ..........OneDimensional Manifold ....................................................................................... 40 Figure 23 ......Grid Independent Study ......................... 29 Figure 11 – Generating a Condensed Phase Chemkin Link File ..............Time Scales Important to Reacting Flow............. SteadyState Simulation ....... 40 Figure 22 .............................................................................................NO Profile for Various Starting Points at 20 atm . SteadyState Simulation ............. SteadyState Simulation ..................Iterative Solving Procedure.............................................ILDM Implementation into 1D.... 59 ix ............................................................................ 23 Figure 7 – Manifold Generation ...........Iterative Solving Procedure................................................................... 38 Figure 20 ........................................Results of OneDimensional.................Data Flow with Enthalpy Grid Refinement ...................................List of Figures Figure 1 . 30 Figure 13 – 1D... 25 Figure 10 ................................................................... 52 Figure 34 ....................Results of OneDimensional............................................................Grid Independent Study ..................Portion of HMX Manifold ................ 49 Figure 30 ....... SteadyState Laminar Simulation..Control Volume Layout ...................... 51 Figure 32 ....Results of OneDimensional..........................................................................................................Grid Independence Study (050 µm) ................. 10 Figure 2 .............................. 54 Figure 37 ........... SteadyState Simulation ............
... 69 Figure 48 .................................................................. 67 Figure 46 .......Temperature Contours..........Heat Transfer at the SolidGas Interface................. 72 Figure 50 ........................................... 68 Figure 47 .....................Heat Flux Analysis for Flame Propagation.......Figure 40 ................... 76 x ........................... 64 Figure 44 ..Closed System Model ..............Open System Model.........Grid Layout ........ 63 Figure 43 .......................... 61 Figure 41 ........................ 65 Figure 45 .Initial Velocity Profile................ 70 Figure 49 .....................................................................................Heat Transfer Simulation ...................................................................Ignition Prediction for Various Gap Widths ............................ 75 Figure 52 .............Comparison of Heat Fluxes........................................................Ignition Data from Heating Tests............................................ 74 Figure 51 .............Temperature Profile in the Closed System Simulation .............................................................................. 62 Figure 42 .........................................................................................................................................................................................................Distance to Ignition for a 5 mm Gap........................
...........................List of Tables Table 1 ...... 10 xi ...................Eigenvalue Analysis ........................................
xii .
and finally bursts. Part of the program is delegated to the University of Utah through CSAFE (Center for the Simulation of Accidental Fires and Explosions). swells. The challenge of such a simulation lies in calculating detailed transport and complex chemical kinetics simultaneously. and Pilling 1997). As the high explosive combusts.7Tetrazocine) as the explosive and 5% binder. Turanyi. Currently. 1 . This initiative seeks to simulate the rapid heating of a container filled with a plastic bonded explosive. the container pressurizes.e.3. This challenge is being explored by DOE’s ASCI (Accelerated Strategic Computing Initiative) program. it is very computationally expensive to calculate both. reduce the chemical kinetics. To carry out this program.7Tetranitro1. CSAFE’s goal is to simulate this scenario from just after the ignition of the pool fire up to the rupture of the container. PBX9501. in a large jetfuel pool fire. Therefore.5. Reynolds Averaged NavierStokes methods).5. In the simulation. PBX9501 is composed of 95% HMX (Octahydro1.3. the University of Utah received funding from the DOE ASCI (Accelerated Strategic Computational Initiative) program to develop new tools for the simulation of accidental fires and explosions. In 1996. To simplify the chemical kinetics. the options are to make simplified transport calculations (i. or both. heat from the jetfuel fire ignites the high explosive. the University of Utah created CSAFE. the high explosive is modeled as pure HMX.Chapter 1 Introduction Modeling of reacting flows is a computational challenge even for the world’s largest supercomputers (Tomlin.
All of these methods fall under one of two general categories: (1) reduced mechanism methods or (2) ILDM methods. The objectives of this project were to: • • • Automate a system for generating an ILDM for gaseous HMX. gaseous HMX mechanism. and Maas 1998). The appeal of the ILDM method over reduced mechanism methods is that much of the detailed chemistry is maintained while still showing a dramatic computational speedup for chemistry calculations (Yang and Pope 1998b). • Lay a foundation for higherdimensional (in space) simulations that apply ILDMsimplified chemistry by investigating 2D (in space) simulations with equilibrium gas properties. the constrained equilibrium approach. steadystate laminar simulation. repromodeling. Implement the ILDM in a 1D (in space). Instead of simplifying a reaction mechanism by eliminating reactions or species. The manifold method ignores reactions with fast time scales yet is still based on detailed kinetics.Efforts to reduce chemical kinetics include systematically reduced mechanisms. computational singular perturbation. and dynamic dimension reduction (Blasenbrey. the manifold method explores time scales for chemical reactions. steadystate laminar simulation. the progression of reactions with fast time scales (compared to flow time scales) can be ignored. the method of Intrinsic LowDimensional Manifolds (ILDM). the ILDM method for reducing chemical kinetics has been applied to a detailed. To assist in the CSAFE effort of providing a fullphysics simulation at a reasonable computational cost. Evaluate the performance of the HMX ILDM by comparison with full kinetics in the 1D (in space). For reacting flow calculations. 2 . Schmidt.
Because of shortcomings of the 3D ILDM. a model for nearsurface kinetics is proposed. Chapter 6 describes 2D (in space) simulations with equilibrium HMX products. The ILDM method is described mathematically. Because of the 4D ILDM inadequacy in the fast reaction zone. Chapter 3 describes original work of this project in generating a 3D (in pressure. steadystate laminar simulation is described in Chapter 4. Chapter 2 reviews two methods for simplifying chemical kinetics: reduced mechanism methods and ILDM methods. steadystate laminar simulation. enthalpy. Chapter 8 describes future work that can be done by other researchers to meet CSAFE’s objective of implementing more accurate HMX chemistry into container simulations. These simulations lay a foundation for higherdimensional (in space) simulations that apply ILDMsimplified chemistry.This thesis describes the achievement of these objectives. The application of the 3D HMX ILDM in a 1D (in space). and mass fraction of N2) ILDM for HMX. enthalpy. and successful applications of this method are reviewed. Specifically. a 4D (in pressure. Conclusions and general observations about the ILDM method are given in Chapter 7. Chapter 5 describes the application of the 4D ILDM in the 1D (in space). 3 . and mass fractions of N2 and NO) ILDM was created.
4 .
al. it became possible. though still very expensive. 2. the size of the kinetic mechanisms was limited.1 Reduced Mechanism Methods Traditionally. the resultant reduced mechanism only works for a 5 . In the precomputer era. The ILDM method. With the development of faster computers. in particular. The final two sections of the chapter show applications of ILDM techniques to practical combustion simulations and recent improvements to the method. This person selected chemical species and reactions that would be important under given conditions. the chemical expert would systematically reduce the complex mechanisms into skeletal mechanisms. The drawbacks to this method are that it requires substantial knowledge of a detailed mechanism to generate a reduced mechanism. The advantage of using skeletal mechanisms over analytical solutions is that less development time is required. In addition. Because analytical solutions were necessary. To reduce the cost of computing. 1997). it required the technical expertise of someone familiar with the reaction mechanism (Tomlin. when chemical kinetic mechanisms were constructed manually.Chapter 2 Literature Review This chapter discusses efforts to reduce chemical kinetics that fall into two general categories: reduced mechanism methods and ILDM methods (Yang and Pope 1998b). the chemical expert would apply either the steadystate or partialequilibrium approximation to various elementary reactions in the mechanism to come up with an analytical solution. et. to model complex chemical reactions with detailed kinetics. is described in theoretical detail.
In this way. The range of the reduced mechanism is limited because reactions can depend greatly on physical conditions. there is a change in reaction pathway as the combustion moves from fuelrich to fuellean conditions.). Three of these techniques are described below.limited range of temperature. This technique is founded on the assumption that chemically similar species have similar rate parameters and can therefore be treated as a single compound. Lumping procedures can be used to reduce the reaction system to a lowerdimensional system of equations (Tomlin. pressure. temperature.1. The sensitivity analysis identifies reactions that have little effect on the overall reaction by measuring a normalized local sensitivity. et. al. etc.1. 2. Accordingly. 2. For example. 1997). reactions that involve principal components remain in the new. et.2 Species Lumping Procedures Even after a reduced mechanism is achieved. the number of reactions can still be too great to implement the mechanism into a practical combustion code. and composition. pressure. 1997). Other techniques have also been developed to assist in reduced mechanism development. reduced mechanism.1 Sensitivity Analysis A sensitivity analysis or principal component analysis is one of the tools to reduce complex chemical kinetics (Tomlin. The normalized local sensitivity represents the amount that a species concentration changes after a small perturbation in one of the system parameters (e.g. al. the dimension of the lumped mechanism is equal to the 6 . Reactions that contain species with low sensitivities can be eliminated from the detailed mechanism.
The chemistry calculation is then allowed to proceed at the time scale of the next fastest reaction. However. 2. The weakness of this method is that it is a compromise between a detailed mechanism and a global mechanism. Mass and Pope proposed a new method for reducing chemical kinetics based on Intrinsic LowDimensional Manifolds (ILDM) (Maas and 7 . al. To overcome this deficiency.e. and/or species’ concentrations (i.number of unique classifications of chemical species. it is labeled as a dead mode and is discarded from the calculation. the reduced mechanisms are generally limited to a range of temperature. Outside of this defined state space. 2. the system of equations becomes less stiff and therefore less computationally expensive to solve. As the reaction proceeds. Once the fastest reaction is exhausted.1. et. The procedure begins by evaluating time scales of reactions and ordering them from fastest to slowest. much of the detailed reaction information is lost. 1997). As the fastest reactions are eliminated.3 Computational Singular Perturbation (CSP) Technique The CSP technique is a formal way to apply partialequilibrium approximations on an a priori basis (Lam and Goussis 1988). pressure.2 Intrinsic LowDimensional Manifold (ILDM) Many of the previously mentioned methods for creating reduced mechanisms rely on steadystate or partialequilibrium approximations. large errors can occur. state space). the fastest reaction becomes exhausted and its reaction rate approaches zero. By lumping species together. The entire chemistry calculation proceeds with the time scale of the fastest reaction. CSP has been used to investigate mechanisms and to reduce them rather than to provide a computationally efficient technique for combustion modeling (Tomlin.
g. The ILDM method is especially suited for nonpremixed reacting flow calculations where mixing controls much of the chemical reaction. temperature. A thermochemical system can be characterized by fast and slow reactions. Maas and Pope’s original work proposed a method to automatically apply the steadystate and partialequilibrium approximations over all possible thermochemical states of the system (state space).Pope 1992a. Once these variables are calculated. are known. three variables (e. and reaction rates. only these variables must be calculated. a limited number of progress variables can be used to characterize the entire thermochemical system. for a particular mechanism. including mass fractions of chemical species. Since the progress variables completely describe the system. all statespace variables. the decoupled fast reactions (those that are faster than the flow time scales) can be ignored while those with slow time scales are tracked using progress variables. It is assumed that the fastest reactions can be decoupled from the slow ones. 8 . This leads to a dramatic reduction of CPU time for solving the chemistry in a reacting flow calculation. Mathematically. fewer of the fast time scales can be decoupled. An ILDM is a topological space or surface defined by independent thermochemical parameters. a manifold is a topological space or surface. enthalpy. In premixed reacting flow calculations. Then. entropy. For example. This can lead to higherdimensional manifolds that are more difficult to implement in realistic combustion simulations. and mass fraction of N2) could be selected as the progress variables. pressure. Maas and Pope 1992b). By this means.
NO formation in coal combustion) in reactions important to chemical combustion. Figure 1 illustrates these ranges (Maas and Pope 1992b).2. some radical reactions) to 102 seconds (e. 2.ns). diffusion. al.2. et. and is written as φ = (H.Yns). etc). It includes enthalpy (H) and all species mass fractions (Yi.1 The Equations The composition vector (φ) is composed of all intensive variables that define the thermochemical state of a system. 1997) dφ ( x .…. 9 . t ) = S [φ ( x. Typically. t ) dt (1) where S is the rate of change due to reactions (for enthalpy. the time scales necessary to model physical processes are on the order of 104 to 102 seconds. Y2.…. Y1. The equations for the composition φ can be written as (Tomlin. reaction time scales are on the order of 109 seconds (e. On the other hand. i=1.2. t )] + Γ( x.2. S = 0) and Γ is the rate of change due to transport effects (convection.2 Chemical and Physical Process Time Scales Chemical mechanisms typically have a much larger range of time scales than necessary for the accurate calculation of a reacting flow system.g.g.
3 Decoupling Fast Chemical Time Scales Mass and Pope used an eigenvalue analysis to determine if a system is on a lowerdimensional manifold and is governed by slow chemistry.2. 2.Time Scales Important to Reacting Flow Species involved in reactions that happen much faster than the physical processes can be decoupled and eliminated (Blasenbrey. Table 1 . Conserved variables (e.g.Figure 1 . there is a time scale for the flow (τflow) based on the resolution of the transport calculations. The eigenvalues are those of the Jacobian (dSi/dφj). 1998). al. If the real part of an eigenvalue is greater 10 . et. pressure in an isobaric system) create eigenvectors that have eigenvalues of zero (see Table 1). A large negative eigenvalue associated with an elementary reaction means that the reaction is governed by fast chemistry and is in local equilibrium.Eigenvalue Analysis Eigenvalue (real part) Response to Physical Perturbations Positive Perturbation will increase (instability) Zero Perturbation will not change with time (change of a conserved variable) Negative Perturbation will relax to zero For every reacting flow calculation.
The progress variables in the reduced set are the only ones that must calculated for a laminar or turbulent combustion calculation. This means that too few progress variables are tracked and could lead to errors typical of an invalid steadystate approximation. if the eigenvalue is less than –1/τflow. • Fixed parameterization does not guarantee the uniqueness or existence of solutions and can yield illconditioned equation systems for the manifold. al. a manifold of constant dimension is the only practical way to implement the ILDM method for combustion calculations at this time. • Higherdimensional manifolds may be required in some regions of the state space.2. 1998). The number of progress variables that cannot be decoupled form a reduced set. the dimension of the manifold should be fixed over the entire state space of the manifold (Blasenbrey. However. 2. Despite these drawbacks. This makes the implementation much less time consuming but also has some drawbacks. Since it is difficult to implement an adaptivedimensional manifold into a CFD (Computational Fluid Dynamics) code. then the corresponding eigenvector is in the fast subspace and can be decoupled from the reaction system. These drawbacks include: • Calculation of an ILDM is attempted over the whole state space even though some domains of the state space are never accessed in practical applications. 11 . then the corresponding eigenvector is in the slow subspace (Yang and Pope 1998a). et.than –1/τflow.4 Manifold Dimension The number of variables in the reduced set determines the dimension of the manifold.
The trajectories have the following properties (Maas and Pope 1992b): 12 .5 OneDimensional Manifolds Up to this point. The trajectories represent the reaction progression and the square is the final equilibrium condition. Figure 2 . The simplest case is a manifold plotted against one progress variable. Manifolds can also be observed through graphical techniques (Maas and Pope 1994).2.2.OneDimensional Manifold In this case. All initial compositions have the same elemental fractions but different species fractions. Figure 2 shows the specific mole number of H2O (mol of H2O/gm of mixture) plotted against the specific mole number of CO2 for COH2air as reactants (Maas and Pope 1992b). the manifold approach has been described mathematically. The beginning of each trajectory represents an initial composition for the reaction. the specific mole number of CO2 is the manifold coordinate.
If it is perturbed at a time scale slower than the fastest time scale of the manifold. 1998). 1998). heat convection). mixing. molecular diffusion. a higherdimensional manifold should be employed (Blasenbrey. it will always remain on the manifold if there are no perturbations (Γ = 0). A thermochemical system is drawn to a unique manifold and progresses along it until the system is perturbed through physical processes (e. et.• • • • All approach one common point (equilibrium) All approach one common trajectory (manifold) On the manifold.2. al. However. 2. if the perturbations happen at a time scale that is faster than the existing manifold.g.6 TwoDimensional Manifolds A twodimensional manifold has the same general characteristics as the onedimensional manifold. al. et. the manifold is an accurate representation of the chemical dynamics. A higherdimensional manifold simply means that more progress variables must be added to the manifold to capture a larger number of time scales. Figure 3 is an example of a twodimensional manifold (Blasenbrey. The only difference is that two progress variables must be used to describe the state of the system. 13 . This can dramatically increase the storage requirements for the manifold. only slow time scales govern the chemical reaction Once a thermochemical system is on the manifold.
TwoDimensional Manifold The manifold is the grid surface. Each trajectory relaxes to the grid surface quickly and stays on that surface until the thermochemical system reaches equilibrium (square point on the plot). 14 . Near equilibrium (i.Figure 3 . The dimension of a manifold increases further from equilibrium to account for faster reaction time scales.e. Figure 4 shows a multidimensional manifold (Bongers 2002). a 0D manifold (no progress variables) represents the thermochemical system because there is no change. a 1D (one progress variable) manifold adequately characterizes the reaction system.7 HigherDimensional Manifolds At equilibrium. 2. Each trajectory represents a starting point in the state space. a final rate determining step before equilibrium).2.
Then the enthalpy can be held constant at another value and another twodimensional manifold plot can be generated. and mass fraction of CO2. a threedimensional manifold of the CH4air system can be characterized by enthalpy.MultiDimensional Manifold Higherdimensional manifolds can be visualized through lowerdimensional plots by taking slices of the full manifold. pressure. For example.Figure 4 . including radicals. In addition. a higherdimensional manifold can be visualized and selected graphically. If the enthalpy is held constant. within 3% error.2. the manifold can be plotted as a twodimensional manifold. 2. there is a speedup in the calculation of the chemistry by a factor of 10. Three examples include: 15 . The ILDM method has also been implemented successfully in turbulent flame calculations.8 Examples of Manifold Implementation The ILDM method has been implemented successfully for laminar premixed COH2O2N2 flames (Maas and Pope 1994). The COH2O2N2 ILDM tracks all species. In this way.
• A numerical simulation of a turbulent nonpremixed CH4H2air flame shows that the ILDM method models flames near extinction and near equilibrium (Xiao. The drawbacks to the original ILDM method are (Yang and Pope 1998b): • • The entire state space must be calculated for a fixed dimensional manifold. Unlike the ILDM approach. and Maas 1998). All dimensions are stored (up to 20 species). Thus. but it calculates and stores a fulldimensional manifold during the reacting flow calculation. 2.9 Recent Improvements to the Manifold Method New strategies have been developed to overcome some of the shortcomings of the original ILDM method. • A CH4air combustion system shows that the ILDM method is 1. • For higherdimensional manifolds. only areas of the state space that are accessed are included in the manifold. The ISAT method operates under the same principles as the ILDM method. the work to retrieve information is not trivial.500 times faster in computing chemistry than a skeletal mechanism (Yang and Pope 1998b). 16 . the method of In Situ Adaptive Tabulation in Principal Directions (ISAT) was created (Yang and Pope 1998b). • A piloted COH2N2air diffusion flame simulation shows that extinction can be predicted within 5% of the experimental value (Norris and Pope 1995). Schmidt. Even with the increased dimensions.2. The storage space required to record the manifold can be very large and increases dramatically for higherdimensional manifolds. a reduction by a factor of 1000 in computational effort can still be observed (Pope 1997). To overcome these shortcomings. the dimension of the manifold is not reduced.
and Maas 1997).Another technique to overcome the drawbacks of the standard tabulation method is to store the manifold as a piecewise polynomial. Schmidt. 17 . As a polynomial. For a test case. this method reduces the storage requirements of the manifold by a factor of 100 compared with the storage requirements for individual points (Niemann. only the coefficients to the polynomial need to be stored.
18 .
Chapter 3 Manifold Generation for Gaseous HMX One of the major challenges for the ILDM method has proved to be the application of the theory to actual systems. The University of Utah CSAFE developers have also used the ILDM approach in a CFD simulation of a heptane pool fire (CSAFE Program 2001). Maas and Pope 1992b). Norris and Pope 1995. the goal of CSAFE is the simulation of the heatup and pressurization of a container engulfed in a pool fire. a global mechanism is used to model the gasphase reactions of HMX combustion. As stated previously. the pool fire simulation and the container simulation are performed independently. A major development towards tighter coupling between the container dynamics and the pool fire codes would be to include a more detailed model of the gasphase HMX chemistry inside the container. enthalpy. The ILDM method is an attractive alternative to detailed kinetics in the container calculations. and the mass fraction of CO2 +CO) manifold is already used to characterize the gasphase reactions and soot chemistry in a turbulent heptane pool fire (CSAFE Program 2001). This would provide a more fundamental representation of the gasphase chemistry with respect to pressurization. Within the container. Based on the success of the ILDM approach for hydrocarbon chemistry. it was proposed to 19 . Maas and Pope 1994. Yang and Pope 1998b. Yang and Pope 1998a. The developers of the method have shown success of the method for laminar and turbulent flame calculations (Maas and Pope 1992a. Currently. A threedimensional (based on mixture fraction.
1. pressure. the dimension of the manifold was initially fixed to three (enthalpy. identifying the manifold. and mass fraction of N2). 3. These numerical errors lead to extraneous eigenvalues.1 Manifold Generation There are six steps to the generation of a fixeddimensional manifold: selecting the dimension. The following section describes the generation of the HMX manifold that will be used in the simulation of the gas phase of the container. 3. Because the eigenvalue analysis did not reveal adequate information about the system. Later.simplify the high explosive chemistry with the same approach. An analysis of the Jacobian revealed that many of the array values approach zero while others are very large. a fourdimensional manifold was developed (mass fraction of NO added) to increase the accuracy. an eigenvalue analysis could not be performed because of numerical errors in calculating the eigenvalues.1 Dimension of the Manifold The dimension of the manifold has been investigated through an eigenvalue analysis of the Jacobian (a matrix formed during the iterative Newton’s method). The number of eigenvalues that have a real portion greater than –1/τflow determines the dimension of the manifold. generating starting points in the composition space. choosing the parameterization. For gaseous HMX. 20 . tracking each starting point to equilibrium. and storing the manifold.
20%. fully reacted HMX were used. The secant method 21 . YReactants is the species mass fraction in the unreacted feed. Kee. and 65% with unreacted HMX on a mass basis according to the following equation: Y j = YPSR j (1 − f Reactants ) + YReactants j f Reactants where j is an index number for each species. Therefore. Differing residence times for the PSR were attempted. After the mass fractions were evaluated. YPSR is the mass (2) fraction of PSR products. Since the PSR is specified as an adiabatic reactor. These simulated reacted gases from the PSR were then mathematically mixed (in a subroutine separate from the PSR code) in proportions of 5%. radicals were introduced to initiate the reaction. The other parameters (YN2 and YNO) were recorded as the simulated reactions proceeded. the temperature of the new mixture was computed in a subroutine separate from the PSR code. the mixture of PSR products and unreacted HMX had a known enthalpy.3. all but one of the parameters must be held constant. 50%. The generation of radicals was simulated in an adiabatic Perfectly Stirred Reactor (PSR). al. Grcar. 35%. For the adiabatic batch reactor combustion code. the temperature of the mixture was solved iteratively. because the HMX reaction time scales are extremely fast. and fReactants is the fraction of unreacted feed that is mixed with the PSR products. enthalpy and pressure were held constant. A PSR code was developed by Glarborg et. the products of the PSR had the same enthalpy as unreacted HMX. only fully reacted HMX or totally unreacted HMX resulted. To generate sufficient radicals.1. In the simulation. However. By this means. Y is the mass fraction. (Glarborg.2 Starting Points in the Composition Space To generate starting points in the composition space. and Miller 1992).
al.Reaction Path for Manifold Generation The simulations use the ChemkinII subroutines developed at Sandia National Laboratories (Kee.3 Reaction Progression The calculated mixture of PSR products and unreacted HMX is the simulated input stream to an adiabatic batch reactor program. et. the temperature met a 1011K convergence criteria in fewer than 10 iterations. Figure 5 . Generally. a Chemkin link file must be created (see Figure 6). through the simulated PSR. 22 . Figure 5 illustrates the flow of a trajectory calculation from the simulated initial reactants. 1989) to calculate thermodynamic and kinetic properties. and finally to the simulated batch reactor.1. In order to use the Chemkin Subroutine Library.(Burden and Faires 1997) was used to find the temperature with the initial guess midway between the high temperature of the PSR products and the low temperature of the reactants. 3.
reaction rate of the species. temperature. the manifold can be automatically generated (see Figure 7). Figure 8 shows an example of the mass 23 .Figure 6 – Generating a Gas Phase Chemkin Link File After the Chemkin link file is created. Figure 7 – Manifold Generation In the simulated batch reactor. entropy. and enthalpy were computed starting at the initial time and stepping forward until equilibrium is reached. mass fractions of the species. The progression towards equilibrium forms a trajectory when plotted against time.
Instead of stopping the execution of the program. This modification speeds up the calculation by many orders of magnitude without affecting the accuracy of the calculation. then the time step is increased. the time step must be set by the user to be small to account for fast reactions and to avoid errors in the numerical integrator. Initially. 24 . pressure. the batch reactor program was modified to allow for changing time steps.fraction of CO2 plotted against time for three different initial compositions at the same enthalpy. and elemental composition. Therefore. the modified program retries the numerical integration with a lower time step. If a certain number of numerical integrations have passed since an error. The original batch reactor program only allowed for a constant time step equal to the smallest time step during the entire calculation. Figure 8 .Batch Reactor Results Each point on the graph represents a time instance where all of the state space is computed and recorded. The original program stopped when the numerical integrator (DVODE) returned an error.
entropy is plotted against the calculated mass fraction of N2. and reaction rates are stored in data files. the calculated species.1. In this case.1. Each starting point represents a different percentage of unreacted HMX mixed with products of HMX combustion from a simulated PSR. Each trajectory represents a separate simulation performed in an adiabatic batch reactor. Figure 9 . All the trajectories are attracted towards the manifold and end at the point that represents equilibrium. The initial temperature of the unreacted HMX gas is 127 ºC and the pressure is 20 atm.3. temperature.Portion of HMX Manifold 25 . These data files are analyzed to identify the ILDM.4. 3.1 Graphical Method Figure 9 shows an example of the graphical method for identifying the lowerdimensional manifold. The values in the legend refer to the percent of unreacted HMX (see Equation 2).4 Manifold Identification During the course of a batch reactor simulation.
CO2. if these three or four parameters are known.2 Eigenvector Analysis An eigenvector analysis requires more development time but results in a more automated system for selecting a manifold. the graphical method for selecting the manifold was employed. 3. pressure. so an eigenvector analysis was not developed. a separate graph is generated.1. Additional graphs can be made for all the species and temperature as a function of N2 mass fraction. For this project.1. and mass fraction of N2) or four (enthalpy. 26 . H2O) are often considered as the first species progress variable instead of intermediates (i. CO. NO.5 Choice of Parameterization Three (enthalpy. all other species concentrations are also known. HCN). the pressure and enthalpy must be varied over a range necessitated by the state space of the combustion system that will employ the manifold. 3. and mass fraction of N2 and NO) parameters were chosen as dependent variables of the manifold. The eigenvector analysis allows for the calculation of only one trajectory and identifies when that trajectory reaches the manifold. N2. With each change in pressure or enthalpy. Major products (i. A special requirement of one of the species progress variables is that it must be monotonically increasing or decreasing in the area of the manifold. pressure. analogous to Figure 9. Additional species progress variables must also meet this specification if aliasing occurs. These graphs can aid in identifying the ILDM.e.4.e. Therefore.To generate a complete manifold.
05 for a mass fraction of N2? Blasenbrey. 1998). al. For example. A multivariable linear interpolator looks up values in the tables based on enthalpy. it is stored in a data file for future use in a combustion code. suggest a linear extrapolation between the lowest available value and the initial concentration (Blasenbrey. pressure.2 Manifold Storage and Retrieval Once the manifold is identified.10. At runtime.3. 27 . at a pressure of 20 atm and an enthalpy of 1010 erg/gm. The latter option was implemented in this project because it represents more realistic conditions for the simulated reactions. al. and mass fractions of N2 and NO. the data files are loaded into arrays within the combustion application. the lowest accessible value for mass fraction of N2 may be 0. One of the challenges to the retrieval process is accounting for values that are outside of the manifold state space. et. Because the project uses computers with varying operating systems. et. the files are stored in ASCII format instead of binary. Another option is to include reaction trajectories that have not reached the manifold. What values are returned when the code attempts to access 0.
28 .
Figure 10 shows how the ILDM replaces the full kinetics calculation in the 1D. steadystate laminar simulation. and a transport properties Chemkin link file (see Figure 12) must be created. The onedimensional combustion model is simplified further with a steadystate assumption by specifying that the liquidgas interface as stationary. SteadyState Laminar Simulation In addition to the gas phase Chemkin link file (see Figure 6).ILDM Implementation into 1D. With high advection. Figure 10 .Chapter 4 3D Manifold Validation with 1D Simulation In order to validate the effectiveness of the HMX manifold. diffusion is assumed to be negligible. The code that is used to test the manifold includes a condensed (solid and liquid) phase and a gas phase (Davidson 1996). a condensed phase Chemkin link file (see Figure 11). the simulated combustion products move through advection. 29 . In such a Lagrangian reference. results from a full kinetic mechanism and a manifold approach are compared in a onedimensional (in space) combustion code.
Figure 11 – Generating a Condensed Phase Chemkin Link File Figure 12 – Generating a Transport Properties Chemkin Link File Once the required link files are created. steadystate laminar simulation can be run (see Figure 13). the 1D. 30 .
SteadyState Laminar Simulation Since the calculation of the gas phase kinetics typically requires more than 99% of the CPU time.Figure 13 – 1D. the manifold is only used to simplify the calculation of the gasphase reactions.1 3D Manifold Direct Substitution A quick check to see how well the manifold follows the detailed kinetics is to substitute manifold values for detailed kinetics values. 31 . and iterative solution of the manifold. full kinetics. 4. The three result files are the solutions to direct substitution of the manifold (using the full kinetics solution of enthalpy and mass fraction of N2). Figure 14 shows the concentration of CO2 versus distance from the surface. The final solution of enthalpy and mass fraction of N2 from the detailed kinetics calculation is used to look up manifold values.
32 . This is due to the varying time scales along the flame. The time scale of the onedimensional flow is computed by the following equation: τ flow = ∆ distance velocity (3) Figure 15 shows the time scale versus distance.Figure 14 – Direct Substitution of 3D Manifold Near the surface (0 to 60 µm) the manifold does not agree with the detailed kinetics while away from the surface (60 µm to infinity) the two are exactly equal.
using a 4D manifold means that only 4 ordinary differential equations must be solved. For example. a higherdimensional manifold should be used to get accurate results. The number of differential equations that must be solved is equal to the dimension of the manifold.2 3D Manifold Implementation in 1D Laminar Simulation Using detailed kinetics requires the solution to the energy equation and to the continuity equations. 33 .Flow Time Scale Because the time scale is so small near the surface.Figure 15 . 4. For HMX. the manifold should accurately represent the detailed kinetics everywhere. As long as a sufficiently high dimension of the manifold is used. Using a lowdimensional manifold reduces the number of dependent variables. this means that 45 ordinary differential equations must be solved.
4. nmass is the mass flux. Figure 16 . 4. the HMX manifold is only a function of enthalpy and mass fraction of N2.2 Discretization of the Equations The enthalpy equation and N2 continuity equation are discretized using the control volume layout as shown in Figure 16. Therefore. ωΝ2 is the reaction rate of N2. and WN2 is the molecular weight of N2). The energy equation in the enthalpy form is: n mass dH d d k T = dx dx dx (4) (where H is enthalpy.2.2. the equations that best describe the chemistry are the energy equation in the enthalpy form and the continuity equation for N2. The N2 continuity equation is: n mass d YN = ω N 2 W N 2 dx 2 (5) (where YN2 is the mass fraction of N2.Control Volume Layout 34 . k is the thermal conductivity.1 The Equations In an isobaric simulation. and T is the temperature).
i −1 n mass α YN 2 P . using an underrelaxation parameter (α < 1) to ensure the stability of the solver.i = YN 2 P .i −1 (8) and the N2 conservation equation becomes n mass ⋅ YN 2W . the energy equation becomes dT dT nmass ( H e − H w ) = k e − kw dx e dx w and the N2 conservation equation becomes n mass (YN 2e − YN 2w ) = ϖ P ⋅ WP ⋅ ∆x P (6) (7) where ∆xP is the distance across the control volume for P from e to w.The control volume approach integrates the equations over a distance from e (east) to w (west).i + k e ⋅ E δx e δx w = H P . The overall solving routine is iterative because the manifold 35 .i −1 + α nmass α −H P .i + (ω P ⋅ WP ⋅ ∆x P ) i −1 +α − Y N 2 P . and using an upwind scheme (valid for high Peclet numbers) the enthalpy equation becomes H P .3 Iterative Solving Routine Both of the discretized equations are explicit in time and distance yet the overall solving routine is implicit. Using the control volume approach and integrating between the control surfaces for P. 4.2.i −1 ( ) (9) where the subscript (i) denotes the current iteration value and the subscript (i1) denotes the value from the previous iteration. Assuming a linear temperature profile between nodes.i T − TW T − TP − kw ⋅ P nmass ⋅ H W .
Figure 17 .is a function of both enthalpy and the mass fraction of N2. Figure 18 shows the last iteration of 1000 total iterations versus the temperature averaged over the last 100 iterations for grid spacing at 6 µm. 36 . Once the enthalpy profile is calculated. This iterative process repeats until the solution converges (see Figure 17).Iterative Solving Procedure An interesting characteristic of the manifold solution is that the mass fraction of N2 and enthalpy continue to oscillate slightly around a solution but never stabilize to a fixed solution. the mass fraction profile of N2 is calculated. This is due to the discontinuities in the reaction trajectories when they have not reached the manifold.
2.4 Grid Independent Study An important aspect of numerical analysis is a grid independent study.Figure 18 . using additional nodes also increases the required computer memory and computational time. Generally. The appropriate number of nodes can be determined by increasing the number of nodes until the numerical solution changes by less than a specified tolerance. However. 4.Oscillations in the Final Solution The averaged solution over the last 100 iterations will be used to ensure physically realistic results and to smooth out the oscillations. a numerical solution becomes more accurate as more nodes are used. The numbers in the legend refer to the grid spacing. 37 . Figure 19 shows the calculated temperature versus distance for the grid independent study.
6 µm. With large jumps the underrelaxation parameter decreases to the point that reaching a converged solution is very computationally expensive. Small jumps in the temperature lead to large values of the second derivative for temperature in the enthalpy equation. 38 .Figure 19 . However. Figure 20 shows the enthalpy profiles for a grid spacing of 3 µm and 0. These big jumps in enthalpy are handled by decreasing the underrelaxation parameter (α) and attempting to solve the equation again.Grid Independent Study Grid spacing less than 6 µm gives a grid independent solution. Large values for the second derivative of temperature leads to large jumps in enthalpy. The oscillations are likely the result of using reaction trajectories that have not yet reached the lowerdimensional manifold. decreasing the grid spacing gives increased oscillations during the implicit solving routine.
even the minor radicals.2.5 Results One of the advantages of using the ILDM approach over reduced mechanism approaches is that all of the species profiles. H2O. the underrelaxation parameter is 6. With an underrelaxation so low. any grid spacing less than 6 µm will yield a grid independent solution. To illustrate the capabilities of the ILDM approach. For a grid spacing of 0. Any grid spacing less than 0.0 µm.6 µm. For the purposes of the grid independent study.Grid Independent Study After 1000 iterations. Graphs of all the species profiles are included in Appendix A.Figure 20 . and the 39 .125 for a grid spacing of 3. the final value of the underrelaxation parameter for the enthalpy equation is 0. the species profiles of NO. CO2. are calculated.6 µm requires excessive computational time to get a converged solution. 4. a converged solution is not reached.94 x 1018 for the energy equation in the enthalpy form. A grid spacing of 3 µm is recommended for the HMX manifold onedimensional calculations.
Reasonable agreement between the 3D manifold and full kinetics is found beyond 60 µm. SteadyState Simulation Figure 22 . Even though it is not shown in the plots. SteadyState Simulation 40 . Figure 21 .Results of OneDimensional.Results of OneDimensional. excellent agreement is found beyond 100 µm.temperature profile for the manifold and full kinetic mechanism are shown in Figures 21 24. as equilibrium is approached.
One method to overcome this lag is to include another species in the manifold.Results of OneDimensional. SteadyState Simulation The manifold lags the detailed mechanism in the formation of the major product species.Figure 23 . Increasing the dimension of the 41 . SteadyState Simulation Figure 24 .Results of OneDimensional. This is due to a dark zone where NO is the limiting reactant and dominant species near the surface. such as the mass fraction of NO.
manifold poses new challenges in the creation and implementation of the manifold. 42 . A 4D manifold is discussed in the next chapter.
and 4D manifold. only the major additions to the 3D ILDM approach are discussed. 3D manifold. and mass fraction of NO from the detailed kinetics calculation. Rather than review the complete procedure for the generation and implementation of the 4D ILDM. only to predict the accuracy of the manifold. there is no reason to approximate the solution. The power of direct substitution is that it can give a quick indication of the accuracy of the manifold. mass fraction of N2. Since the correct solution is already known. 43 .1 4D Manifold Direct Substitution An indication of how well the higherdimensional manifold follows the detailed kinetics is to substitute manifold values at every node using the solution of enthalpy. and implementation into a CFD code. 5. Direct substitution is not used to predict chemical kinetics.Chapter 5 4D Manifold Validation with 1D Simulation An additional reaction parameter can be added to increase the accuracy of the ILDM approximation to the onedimensional laminar simulation. Figure 25 shows the concentration of NO versus distance from the surface using the detailed mechanism. The addition of another reaction parameter raises the complexity of the ILDM generation.
Figure 25 – Full Kinetic Mechanism Compared to Manifold Results The 3D manifold agrees with the detailed kinetics above ~60 µm away from the surface (see Figure 2119). The 4D manifold is expected to be more accurate than the 3D manifold.2 4D Manifold Implementation in 1D Laminar Simulation 5. as the dimension of the manifold is increased. 5. The 4D HMX manifold is a function of enthalpy.1 The Equations An additional equation (continuity equation of NO) must be solved to implement the 4D manifold. The advantage of the 4D manifold over the 3D manifold is that the 4D manifold is closer to the exact solution from 12 µm and larger. pressure. and mass fraction of NO. the manifold more accurately represents the detailed kinetics. The equations for enthalpy and mass fraction of 44 . mass fraction of N2. Overall.2.
2.2.3 Iterative Solving Routine As with the 3D manifold. ωΝΟ is the reaction rate of NO. The additional NO species conservation equation is n mass d YNO = ω NOW NO dx (10) where YNO is the mass fraction of NO. The 4D solving routine adds the solution to the NO profile directly after the solution to the N2 profile.2 Discretization of the Equations The control volume approach is also used to discretize the NO species equation. The iterative routine repeats until the solution converges (see Figure 17 and Figure 26). 5. The resulting equation is equivalent to the discretized equation for N2 (see Equation 7) when NO is substituted for N2. the 4D manifold solving routine is implicit.N2 are the same as those in the 3D manifold implementation. 5. 45 . and WNO is the molecular weight of NO.
4).Iterative Solving Procedure For the 3D manifold. for the 4D manifold the solution is more stable and therefore.2. temperature oscillations result in unrealistic enthalpy values.2. an average value was taken to average out an oscillating solution. the iterative solving routine is reinitialized with a coarser 46 . and the variables are reinitialized. When an enthalpy profile is outside of the manifold state space.4 Grid Refinement for Enthalpy Profile One of the causes of instability using a 3D manifold is the enthalpy solution. the iterative solving routine includes a gridrefinement for the enthalpy profile. In the region near the surface where the 3D manifold does not track the fast kinetics.Figure 26 . To overcome this problem. the underrelaxation factor can become so small that a grid independent solution is computationally prohibitive (see Section 4. However. the enthalpy equation underrelaxation factor is automatically decreased. an average value is not necessary. 5. When unrealistic enthalpy values are computed. Without grid refinement.
a restriction is imposed on the maximum grid spacing of the enthalpy profile. The solution procedure with enthalpy refinement is shown Figure 27.Data Flow with Enthalpy Grid Refinement To prevent grid dependent results. A coarser enthalpy grid reduces the temperature oscillations and decreases the computational cost. 47 . the enthalpy grid layout is restored to the original configuration and the underrelaxation factor is decreased. Figure 27 . the maximum allowable enthalpy grid spacing at 20 atm was found to be 30 µm. When the enthalpy grid spacing becomes greater than a specified value.enthalpy grid. Through a grid independent study.
This is accomplished by using the full kinetics solution at 25 µm as the initial point for the 4D manifold solution. As shown with direct substitution (see Figure 25). the 4D manifold does not represent the fast reaction kinetics in the nearsurface region. The idea is to completely skip the region where the NO profile diverges.2.5. A similar grid refinement study was performed on the region greater than 25 µm (see Figure 29). The values in the legend refer to the distance between nodes. the calculated NO profiles of the 4D manifold solution and full kinetics solution do not produce gridindependent results in the 20 µm nearest the burning surface. 48 .Grid Independence Study (050 µm) Even with significant grid refinement. Figure 28 . This is most likely due to the issue that the 4D manifold does not describe detailed kinetics in that region.5 Grid Independent Study Figure 28 shows the calculated mass fraction of NO plotted versus distance (up to 20 µm) for the grid independent study.
Figure 30 shows the calculated mass fraction of NO plotted as a function of distance for various starting points using the combined approach. 5. The values in the legend are the distance from the burning surface where the 4D manifold is used.25 µm. Based on the grid independent study. a grid spacing smaller than 0. the grid spacing is fixed at 0.6 A Model for NearSurface Kinetics Because the 4D manifold does not accurately represent fast kinetics near the burning surface. 49 . a model based on the full kinetics solution is used.7 µm appears to be adequate.2.Figure 29 .Grid Independence Study (>25 µm) When starting the solution at 25 µm. and the 4D manifold is employed where it accurately represents the full kinetic solution.
While at 20 atm. Grid spacing of 0. To evaluate this trend. The different curves represent different starting points for the 4D manifold.NO Profile for Various Starting Points at 20 atm The 4D manifold accurately represents detailed kinetics beyond 19 µm. can only suggest an approximate range where the manifold will produce good results. The real test of the 4D manifold’s performance is through an iterative solution as shown in Figure 30. such as direct substitution.Figure 30 .1 µm was used at 60 atm for the 4D manifold solutions.6 µm was used at 5 atm. 4D manifolds were generated at 5 and 60 atm. Figures 31 and 32 show the results of using these 4D manifolds in the 1D. 50 . The cause of this 7 µm discrepancy is unknown. A challenge for implementing the nearsurface model is recognizing the correct point to begin solving with the 4D manifold. and 0. a starting point of 19 µm appears to be adequate. Previous results (see Figure 25) suggest good agreement beyond 12 µm. steadystate laminar combustion code. Previous results. as pressure changes the starting distance from the surface should change. However.
the 4D manifold accuracy is sufficient at ~6 µm. With an upwind approximation. the “downwind” solution is strongly influenced by deviations “upwind”. At 60 atm.NO Profile for Various Starting Points at 60 atm At 5 atm. the NO solution diverges. “Downwind” errors 51 . the 4D manifold accurately starts at ~92 µm. In the region before the 4D manifold is sufficiently accurate.Figure 31 .NO Profile for Various Starting Points at 5 atm Figure 32 .
the nearsurface model should extend to the maximum enthalpy and the 4D manifold should be used thereafter.caused “upwind” deviations are apparent in the curve with a starting point at 21 µm (Figure 31) and the curve with a starting point at 5 µm (Figure 32). A grid spacing of 0. One method to approximate the beginning point of the 4D manifold solution is obtained by observing the enthalpy profile. Figure 33 .1 µm was used.Enthalpy Profile at 60 atm At 60 atm the 4D manifold represents the detailed kinetics after the maximum enthalpy is reached. The curve starting at 5 µm shows major deviations while the curves starting at 6 and 7 µm represent the enthalpy profile reasonably well. Figure 33 shows the enthalpy profile at 60 atm. Therefore. The difference is that the latter curves begin after the maximum enthalpy. This same trend holds for 20 atm and 5 atm profiles. 52 .
37. the calculations were performed with a grid spacing of 0.5.7 Results To illustrate the capabilities of the 4D manifold approach with a nearsurface model.2. All calculations were performed at 20 atm. H2O. the species profiles of N2.25 µm and a starting location of 19 µm. SteadyState Simulation 53 . For the 4D manifold.Results of OneDimensional. Figure 34 . and the temperature profile for the manifold and full kinetic mechanism are shown in Figures 34 . the 3D manifold results are also shown. CO2.2. For comparison. For the 3D manifold the grid spacing is 3 µm (see Section 4.4) and the starting location is 0 µm.
Results of OneDimensional.Figure 35 .Results of OneDimensional. SteadyState Simulation Figure 36 . SteadyState Simulation 54 .
Generally. CO2) have discontinuities at ~25 µm (see Figure 35).e.Results of OneDimensional.Figure 37 . even with the minor species. A few of the curves (i. 55 . The problem can be minimized by generating and using a manifold with greater resolution. SteadyState Simulation Graphs of all the species profiles for steadystate combustion at 20 atm are included in Appendix A. This is due to discontinuities within the 4D manifold at that point. there is good agreement between the 4D manifold and the full kinetics.
56 .
In addition.1 CSAFE Container Simulation As noted previously. Further work is necessary to fulfill this objective. these simplified simulations reveal information about gas flow and flame propagation within the container.Chapter 6 Container Simulation The overall goal of the HMX manifold generation and testing is to incorporate the ILDM approach in the gasphase of the CSAFE container simulation. Also. A 57 . A few of these studies are reported and referenced in this report.5% Estane and 2. It was suggested that one way to increase the accuracy of the container simulations would be to include a simulation of the gas phase chemistry inside the container through the ILDM approach. Simulations have been performed by CSAFE to investigate deformation of the container. they are insightful with respect to heat transfer at the solidgas interfaces. and heating of the solid high explosive. the CSAFE project includes simulation of a container filled with PBX9501. cracking of the solid high explosive. Experimental tests have been performed to determine the time to explosion of a heated container filled with PBX9501. This section discusses results from three simplified simulations of the gas phase with the gas at equilibrium. 6.5% BDNPF/A). The intention of this work is to provide a foundation for future work in implementing the HMX manifold. cookoff tests (heating a container of high explosive until explosion) have been performed to validate the models. Although the simulations do not implement an ILDM. PBX9501 is an explosive with 95% energetic material (HMX) and 5% binder (2.
cylindrical container (4inch diameter x 12inch length) filled with the explosive was subjected to a high heat flux (~104W/m2). Ignition times ranged from 1.62.4 minutes (Eddings, Sarofim, Ciro, and Beckstead 2000). A simulation by the University of Utah models the pressurization of the container after ignition (CSAFE Program 2001). Figure 38 shows a stress simulation performed on the container and high explosive.
Figure 38  Pressurization of Container
To reduce computational cost, the size of the container in the simulation is smaller than the size of the experimental validation container. The outer shell is a steel container. The interior is high explosive with temperature contours that show the thermal wave penetration. A twodimensional cross section of the container shows the pressurization in three snapshots as it progresses in time.
6.2 A PostIgnition Model with Gas Products at Equilibrium
The CSAFE container simulation includes heat transfer, a global mechanism for HMX reaction rate, deformation of the container, and deformation of the high explosive. In addition, a onedimensional ignition model is available to predict time to ignition 58
(Beckstead 1994). Work has also been performed on multidimensional effects of ignition (Baer 1994). No known work has been performed to model multidimensional effects after ignition for fast heating rates. This postignition study explores the multidimensional effect of flame spreading. In this model, HMX properties are used to approximate those of PBX9501. Figure 39 shows a simplified model of a gap that could exist between the container and the HMX. This model is used to qualitatively investigate the point at which the HMX will ignite.
Figure 39  Model of Container
In the proposed situation for the CSAFE simulation, the container is one meter above a pool fire and receives an assumed heat flux (~104 W/m2) at the outside wall (Beckstead 1999). This heat conducts through the steel container and into the HMX. The thermal contact resistance between the steel casing and solid is equivalent to a 13 mm air gap (Eddings, et. al. 2000). The proposed CSAFE simulation predicts the heat
59
up, reaction of the HMX, pressurization, and deformation of the container. The proposed simulation continues until the time of rupture of the container.
6.2.1 Heat Transfer Analysis
A simplified simulation of the container was performed to reveal the most likely initiation point of the HMX reaction. A crosssection of the container was modeled in Fluent™ to determine a realistic temperature profile. Figure 40 shows the overall crosssection of a 30 cm diameter cylinder after 2,000 seconds. An arbitrary gap, filled with simulated combustion gases is ~2 cm at the widest part. In this heat transfer simulation, the outside of the container is subjected to a constant temperature boundary condition of 550 K. The melting temperature of HMX is slightly greater than 550 K. The boundary condition temperature of 550 K was selected to ensure the validity of the solid HMX assumption.
60
This figure shows that the solid HMX is an excellent insulator. Figure 41 shows the same heat transfer simulation zoomed in at the solidgas interface. Bedrov.Heat Transfer Simulation The container is assumed to be made out of stainless steel (ANSI 316) with a thermal conductivity of 13. reaction will likely occur where the solid HMX comes into contact with the container. The gap between the container and solid HMX has the same properties as air. 61 .4 W/mK (Beckstead 1999. and where the thermal contact resistance is the lowest. and Sewell 2000). Smith.4 W/mK. The heated solid HMX has a thermal conductivity of 0.Figure 40 . and the heat transfer coefficient at the airsolid interface is assumed to be 25 W/m2K. Because of this.
Accordingly. The simulation is approximated as a straight sheet of HMX covered by stainless steel. the highest temperature at the solidgas interface is at the contact points.2. 2000). 62 .Figure 41 .Heat Transfer at the SolidGas Interface Heat transfer by conduction from the container to the solid HMX is larger than the heat transfer from the gas to the solid HMX. the angular effects are assumed to be negligible (Eddings. et. al. 6. Since the gap thickness is much less than the container diameter.2 Open System Simulation In an open system reaction scenario the reaction occurs at the separation point and proceeds around the outside of the solid HMX.
with combustion gases that leave the burning surface at 3200 K. 63 . ignition is assumed to have occurred at the left. vertical plane. releasing additional combustion gases. 6.2. Figure 43 shows an ignition line generated from experimental heating tests for solid HMX (Beckstead 1999). This creates a moving burning surface.Open System Model In this simulation.2 Ignition Model If enough heat is transferred to the horizontal solid HMX it will ignite and begin to burn. and transfers heat to both solids.1 The Model A simplified model of the HMX combustion in the container is shown in Figure 42.2. Figure 42 .2.2. Heat feedback is propagating further burning into the preheated HMX near the steel/HMX interface.6. The combustion gas flows in a gap (assumed to be 5 mm wide) between the steel and unignited HMX.
71 cm/sec. the steel and solid HMX are assumed to move at the surface regression rate. If the hot gases do not transfer enough heat to the unignited HMX surface for ignition. With the burning surface in a stationary reference frame. the burning rate of HMX preheated to 550K at 20 atm is calculated to be 1.22 gm/cm2sec.Ignition Data from Heating Tests Knowing the heat flux to a surface and using data from Figure 43. then a rigorous combustion simulation should only have one burning front. based on the HMX density of 1. while a steel thickness of 150 mm (the thickness is arbitrarily large to allow adiabatic boundary conditions) and 64 . the regressing burning surface is taken to be stationary. This is equivalent to a surface regression rate of 0. if the hot gases from one surface cause other HMX surfaces to ignite then twodimensional flame spreading must be modeled.9 gm/cm3 (Beckstead 1999). The gap between the steel and HMX is taken to be 5 mm. From a onedimensional. Using a Lagrangian approach. steadystate combustion simulation developed at BYU (Davidson 1996). However. the time to ignition can be estimated.Figure 43 .
An unstructured grid allows for a very high density of nodes to be clustered around the gas chamber and fewer nodes to be used at the extremes of the model.an HMX thickness of 100 mm are assumed.2. The grid layout is shown in Figure 44.2. A grid independent 65 . the grid is clustered in the areas of high temperature gradients. The regressing surface has an inflow boundary condition of 12.2 kg/seccm2 and the gas exit an outflow boundary condition. a locally parabolic assumption at the gas outflow is valid. Consequently. Figure 44 .3 Grid Independent Study A simulation was performed in Fluent™ to determine the heat flux to the solid HMX surface. all flow is laminar. 6.Grid Layout In order to reduce the total number of nodes used. The gap thickness is much smaller than the solid thickness so that adiabatic boundary conditions can be employed around the outer solid edges. Gambit™ was used to create the unstructured grid. Since the Reynolds number for the flowing gases is under 100.
SIMPLE (SemiImplicit Method for PressureLinked Equations) was the method used to couple the pressure and velocity terms in the continuity and momentum equations.2. showing the area of high temperature gradients around the flowing channel. 6. upwind. 66 . QUICK and 2nd order were chosen because they are higher order and tend to be more accurate than lower order schemes such as 1st order. A discretization scheme of 2nd order for the pressure and QUICK (Quadratic Upstream Interpolation for ConvectionDominated Kinetics) for the energy and momentum equations was selected.2. Various discretization schemes were also studied.4 Simulation Results Figure 45 illustrates the temperature contours (in Kelvin).study dictates the use of 200 nodes with grid clustering around the gas cavity where there are high temperature and velocity gradients. Also.
the surface is moving and the heat flux that an individual point sees is changing.Figure 45 .Temperature Contours The heat transfer from the combustion gases to the solid surfaces is greatest near the reacting surface. the heat flux has a stationary profile. ∆ti is the time the differential surface is subjected to the heat flux. the heat flux to the solid surfaces decreases. Time for ignition data is based on a constant heat flux. the heat flux is computed through the following expression: q' ' = ∑ q ∆t i =1 '' i n i ∆t (13) where q″i is the instantaneous heat flux at a node. However. As the gases cool in the boundary layer. In the simulation. and ∆t is the time to travel from the burning surface to the final 67 . To account for the movement of the surfaces. with a Lagrangian approach.
position. compared to the empirical data correlation. Figure 47 shows the calculated results of average heat flux versus distance along the gas/HMX interface. The distance on the xaxis refers to the distance from the regressing surface. The time of the incident heat flux is related to the distance away from the regressing surface by the surface regression velocity (∆x = v ∆t ). Figure 46 .Comparison of Heat Fluxes Using the average heat flux. Figure 46 shows the average and instantaneous heat fluxes across the gas/HMX interface. Ignition data relates heat flux to time for ignition (see Figure 43). the distance to ignition is computed by locating the point (in space) at which the heat flux is sufficient to ignite the surface. 68 .
52 cm/sec).5 Gap Thickness Parametric Study A parametric study was performed to study the effect of gap thickness on flame spreading.Distance to Ignition for a 5 mm Gap At 0. when ignition occurs at one point. This indicates that with the open system geometry. At low heating rates (~104 W/m2) the thermal wave penetrates up to ~40 mm into the solid HMX before ignition (Beckstead 1999).2. a gap forms between the steel and unignited HMX.Figure 47 . After ignition the burning front preferentially burns through the preheated HMX because of the 40% faster burning rate over nonpreheated HMX (0. A preheated band of HMX varies in thickness depending on the rate of heating to the outside of the container. 6.72 cm/sec compared to 0.2. Simulations were performed with gap widths of 1 69 . At high heating rates (~105 W/m2) the thermal wave only penetrates ~23 mm into the solid HMX before ignition. the resulting combustion gases will ignite other surfaces nearby.23 mm the average heat flux reaches a value sufficient for ignition. By burning the preheated HMX.
Figure 48 . the combustion gases ignite other surfaces nearby. ignition of the horizontal surface is delayed.2. 70 . Once a single point ignites.07 mm to ignite.mm. The important point is that flame spreading is significant for all gap widths tested. Figure 48 shows the results of the simulations with the ignition heat flux line.55 mm while a gap of 1 mm requires only ~0.Ignition Prediction for Various Gap Widths The ignition distance for a gap of 10 mm is ~0.6 Conclusions for the Open System Simulation The results of this study indicate that the single burning front assumption is not valid after ignition of the HMX. In a rigorous simulation of the high explosive combustion. 6. This spreading effect leads to accelerated combustion through flame propagation.2. This study shows that as the gap width increases. 5 mm and 10 mm. regardless of the width of the gap. twodimensional flame spreading must be modeled. 2 mm.
71 . 2000). the volume of the container can expand. However. Experimental results show that the volume of the container can increase by as much as 25% before rupture (Eddings. This expansion is due to a ballooning effect of the container walls. There is a void volume in the center of the HMX annulus that will accumulate combustion gases and slow down the pressurization of the container. As the container pressurizes. Another important factor is pressurization.2. When solid HMX begins to react in the closed system. 6.2. al. and the reaction rate accelerates.3 Closed System Simulation For an open system flame spreading is significant. the container pressurizes.3. The closed system exhibits positive feedback. The combined void volume of the annulus and expansion of the container decreases the rate of pressurization and delays the rupture of the container. the reaction rate of HMX increases. in a closed system the velocity of the gases is lower and therefore.1 The Model The closed system model of the HMX combustion in the container is shown in Figure 49. flame spreading may not be significant. et.6. In addition.
In an unsteady calculation. Larger time steps lead to solver instability in Fluent™.00005 seconds.01 seconds with time steps of 0.Closed System Model This simulation is similar to the open system simulation in that the left surface is assumed to regress at the steady state burning rate of 0. pressurization occurs and time must be added as another variable in the calculation. The simulation was run for 0. The initial pressure is 1 atm. and transfer heat to both solids. This enlarging of the pocket is assumed to be 72 . and the combustion gases are modeled as an ideal gas with the same properties as air. initial conditions must be specified. the melting temperature of HMX.72 cm/sec with combustion gases leaving the burning surface at the equilibrium temperature of 3200 K.01 seconds. The surface regression enlarges the area of the pocket by less than 1% during the total simulation time of 0.Figure 49 . The difference between this simulation and the open system simulation is that the right boundary is closed. The initial temperature for the simulation is set at 550 K. With a closed system. The combustion gases enter the pocket (assumed 5 mm x 10 mm) between the steel and unignited HMX.
This deformation is beyond the scope of the closed system simulation.2.2. 6.2).3 Simulation Results As in the open system simulation.01 seconds.negligible and. the ideal gas pressure is calculated to increase linearly from 1 atm initially to 150 atm in 0. all vertical boundaries are taken to be stationary. therefore.2 Ignition Model The ignition model for the closed system simulation is the same as the open simulation ignition model (discussed in Section 6.2. The most significant enlargement of the pocket will occur as the pressure increases.2. Because of unrestricted flow within the closed pocket.22 gm/cm2sec. the flow should be laminar and there should not be recirculation of the gases. High pressure in the gap will deform the steel or HMX solid and cause them to separate. 6. Figure 50 shows the velocity profile at 0. the closed system surface regression rate is assumed to be 1.3.00005 seconds with the vectors corresponding to the velocity direction and magnitude (in m/s).3. 73 . With that surface regression rate and the given geometry.
in igniting the HMX.Initial Velocity Profile The combustion gases reach equilibrium § µm from the burning surface. Figure 51 shows the temperature contours (in Kelvin) after 0. thermal effects are more significant than chemical effects. Because the gases reach equilibrium quickly. 74 .01 sec of simulation time.Figure 50 . such as caused by radicals. showing the area of high thermal gradients within and around the pocket.
such as an ILDM gasphase calculation 75 . T2 P2 = T1 P1 γ −1 γ (15) Even though this simulation is not an adiabatic system.Temperature Profile in the Closed System Simulation The temperature rises above the equilibrium temperature of 3200 K because of pressurization. A more rigorous simulation. many of the equilibrium products will dissociate into the elements and moderate the rise in temperature. An adiabatic system with ideal gas undergoing pressurization increases in temperature according to the following equation. The pressurization increases the temperature of the gases. In the simulation. As the temperature increases. a pocket of gas approaches 7500 K due to pressurization. The rise in temperature also increases the heat flux to the solid HMX and steel.Figure 51 . the same trend applies.
A more rigorous simulation would account for the vaporization and reaction of HMX in the gas phase. and would result in a much lower temperature. Figure 52 shows the results of this analysis. In the simulation. Another shortcoming of the closed system simulation is that phase changes are not modeled. The heat flux to the unignited HMX is calculated to predict flame spreading in the closed pocket. This implies that it is near the HMX boiling point and that vaporization of the HMX should occur in the closed system simulation. 76 . the surface temperature of the solid HMX rises above 1500 K. The surface temperature of the liquidgas interface in the steadystate onedimensional model is 775 K. would account for species dissociation.01 seconds. Figure 52 .Heat Flux Analysis for Flame Propagation The heat flux along the horizontal surface of HMX does not reach a value sufficient for ignition within 0.(provided the full kinetic mechanism accounts for dissociation).
Extrapolating the results suggests that ignition may occur around 0.3. By this time the pressure within the pocket is expected to increase to ~400 atm. After the pressure in the container is sufficient to overcome the yield stress. The total time of the closed system simulation would be dictated by the yield stress of the steel container.2. the fixedboundary closed system simulation is no longer valid. 77 .6.02 seconds.4 Conclusions for the Closed System Simulation The flame will not spread within the pocket for this model because the heat flux to the solid HMX does not reach ignition values for the horizontal surface.
78 .
This project also provides an additional example of using the ILDM method. pressure. A 3D (in enthalpy. pressure. This project is the first known attempt to use the manifold method for a complex reactant molecule such as HMX. and mass fraction of NO) manifold increases the accuracy of the approximation. and O2. the applications of the ILDM method have been with simple combustion reactants such as shortchained hydrocarbons. The automated manifold generation is accomplished through the Chemkin paradigm of interchangeable reaction mechanisms. The manifold generation code reads a mechanism file. Up to this point. This report describes a method that automates the manifold process. and computes the reaction trajectory of that mixture in a batch reactor. H2. At 20 atm the 4D manifold provides an accurate representation of detailed kinetics starting at the peak of the 79 . simulates a PSR reactor. mass fraction of N2.Chapter 7 Conclusions One of the major obstacles to manifold implementation in CFD codes is the lack of automation for manifold generation and implementation. The batch reactor results are then used to generate the manifold. A 4D (in enthalpy. and mass fraction of N2) manifold was implemented in a 1D (in space) steadystate simulation. a new manifold can be generated simply by switching out mechanism files. calculates the composition and temperature of a mixture of reactants and PSR products. As new mechanisms are developed for HMX or as other solid propellants are modeled. There were large errors near the surface of the burning HMX (< 60 µm) and good agreement with full kinetics beyond that.
Within 19 µm of the burning surface. only a limited number of differential equations (equal to the dimension of the manifold) must be solved. computational cost was reduced by ~10 times for the ILDM method. With the ILDM method. If coarse gridding is not sufficient. In this project. This speeds up the convergence by eliminating much of the stiffness in the differential equations. flame spreading is shown to be significant. If coarse gridding (>60 µm) were used in the CSAFE simulation. The nearsurface model is generated by extracting the first 19 µm of the steadystate. 1D full kinetics solution.enthalpy profile (~19 µm). In the closedsystem simulation. the 3D manifold would be sufficient. flame spreading is delayed by the restriction to flowing gases. the 4D manifold with a nearsurface model should be used. a nearsurface model is required. 80 . In the opensystem simulation. The multidimensional effects of HMX combustion were explored with a 2D (in space) simulation using gases that approximate the properties of HMX combustion gases at equilibrium.
Deformation of the container and its contents is currently being investigated by other CSAFE researchers. such as heat transfer. The simplified open and closed system simulations performed in this project offer insights into some of the physical processes. local quenching. and flow of gases in order to predict time to detonation or deflagration. The kinetics calculation can be accomplished by implementing the ILDM for HMX.Chapter 8 Future Work The project described in this report is the beginning of the process to generate. 81 . ARCHES. a more accurate representation of the open and closed systems should include species involved in chemical reactions as well as deformation of the HMX and steel due to pressurization. However. and evaluate an ILDM for HMX in the University of Utah’s 3D CFD code. The same techniques that were used in this research project to implement the manifold method into the onedimensional (in space) combustion code will be used to implement the manifold method into ARCHES. implement. Modeling the reaction of HMX in the gas space of a container will give insight to heat feedback. Future work will need to be performed by another researcher to accomplish this goal.
82 .
"Thermal Conductivity of Liquid Octahydro1. Eddings. 31st JANNAF Combustion Meeting. 1998... D. J. 26th Symposium (Int'l) on Combustion.nl/~hbongers/. W. CA: Brooks/Cole Publishing Company. UC4. Blasenbrey. U. 19th JANNAF Propulsion Systems Hazards Meeting. Gross R. E. 2000. R. pp.utah. "Calculated Conditions for Fast CookOff". Gartling D. Pacific Grove. M. Vol. Chemical Physics Letters. CPIA No.. Vol I. and Faires. Glarborg.. "Automatically simplified Chemical Kinetics and Molecular Transport and its Application in Premixed and NonPremixed Laminar Flame calculations". The Combustion Institute. and Beckstead M.. 6468.. CPIA No. Bedrov. W. Kee. Center for the Simulation of Accidental Fires and Explosions. A. Grcar. L.. 1994. Sarofim. Sandia Report. and Hendershot R.. E.. D.. 36th JANNAF Combustion Meeting. P.. R. II. CPIA No. 1994 JANNAF Propulsion Systems Hazards Meeting. 123132.. T. 505511. J.5. W. 691. pp. L. "PSR: A Fortran Program for Modeling WellStirred Reactors". SAND868209. CSAFE Program... 1992. and Sewell.. URL: http://www. J. J.. 1996. pp. and Hobbs M. "A ThreePhase Model of HMX Combustion". 324. J. pp. 704.5.tue. Schmidt.. Vol I. an Introduction. W.edu.combustion. M.7tetranitro1. "An Ignition Model for Modern Double Base Propellants". F. D.. Burden. "Fast CookOff Tests of an HMXbased Propellant in Pool Fires". K. Pittsburgh. and Beckstead. "Multidimensional ThermalChemical Cookoff Modeling". Beckstead. Davidson. pp. J. D. 1997. 19891996. Smith. CSAFE. and Miller. and Maas. 620. Ciro. 1994.csafe. 2001. M. 6372. A. G. D.7tetrazocine (HMX) from Molecular Dynamics Simulations".3.. G. H. pp.. Beckstead.3. Bongers. J. W. 2002. URL: http://www. T. M. 163180. Numerical Analysis. 2000. 1999. The Combustion Institute. TwentySeventh Symposium (International) on Combustion.. 83 . Reduced Reaction Mechanisms.Chapter 9 References Baer. R.
Kee, R. J., Rupley, F. M.., and Miller, J. A., "A Fortran Chemical Kinetics Package for the Analysis of Gas Phase Chemical Kinetics", Sandia Report, 1989, SAND898009B, UC706. Lam, S. H. and Goussis, D. A., "Understanding Complex Chemical Kinetics with Computational Singular Perturbation", TwentySecond Symposium (International) on Combustion, 1988, The Combustion Institute, Pittsburgh, pp. 931941. Maas, U. and Pope, S. B., "Implementation of Simplified Chemical Kinetics Based on Intrinsic LowDimensional Manifolds", TwentyFourth Symposium (International) on Combustion, 1992a, The Combustion Institute, Pittsburgh, pp. 103112. Maas, U. and Pope, S. B., "Simplifying Chemical Kinetics: Intrinsic LowDimensional Manifolds in Composition Space", Combustion and Flame, 1992b, 88, pp. 239264. Maas, U. and Pope, S. B., "Laminar Flame Calculations Using Simplified Chemical Kinetics Based on Intrinsic LowDimensional Manifolds", TwentyFifth Symposium (International) on Combustion, 1994, The Combustion Institute, Pittsburgh, pp. 13491356. Niemann, H., Schmidt, D., and Maas, U., "An Efficient Storage Scheme for Reduced Chemical Kinetics Based on Orthogonal Polynomials", Journal of Engineering Mathematics, 1997, 31, pp. 131142. Norris, A. T. and Pope, S. B., "Modeling of Extinction in Turbulent Diffusion Flames by the VelocityDissipationComposition PDF Method", Combustion and Flame, 1995, 100, pp. 211220. Pope, S. B., "Computationally Efficient Implementation of Combustion Chemistry Using In Situ Adaptive Tabulation", Combustion Theory Modeling, 1997, 1, pp. 4163. Tomlin, A. S., Turanyi, T., and Pilling, M. J., "Mathematical Tools for the Construction, Investigation and Reduction of Combustion Mechanisms", Chap. 4 in Comprehensive Chemical Kinetics edited by Compton, R. G. and Hancock, G., LowTemperature Combustion and Autoignition, 1997, 35, pp. 293437. Xiao, K., Schmidt, D., and Maas, U., "PDF Simulation of Turbulent NonPremixed CH4/H2Air Flames Using Automatically Reduced Chemical Kinetics", TwentySeventh Symposium (International) on Combustion, 1998, The Combustion Institute, Pittsburgh, pp. 10731080. Yang, B. and Pope, S. B., "An Investigation of the Accuracy of Manifold Methods and Splitting Schemes in the Computational Implementation of Combustion Chemistry", Combustion and Flame, 1998a, 112, pp. 1632.
84
Yang, B. and Pope, S. B., "Treating Chemistry in Combustion with Detailed Mechanisms—In Situ Adaptive Tabulation in Principal Directions—Premixed Combustion", Combustion and Flame, 1998b, 112, pp. 85112.
85
86
and full kinetics.) TEMPERATURE CNO ENTHALPY NCO RATE N2 NCN Major Products C2N2 H2O CN CO2 N2O CO HCN N2 HONO Intermediate Species HNO NO NNH H2COHNNO2 NH3 HNC NH2 HMXRO NH HMXR NO2 HMX N H2CNNO2 HCO H2CNNO CH2O H2CNO H2O2 H2CNH HO2 H2CN OH HNO3 H NO3 O HCNO O2 HOCN H2 HNCO 87 . a 4D manifold. steadystate simulation using a 3D manifold.Appendix A Results from the 1D. List of Graphs in Appendix A Properties Intermediate Species (cont.
00E+09 0.00E+00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 88 .00E+10 3D Manifold 4D Manifold Full Kinetics ENTHALPY (erg/gm) 1.50E+10 2.00E+10 5.50E+10 1.4000 3500 3000 TEMPERATURE (K) 2500 2000 1500 1000 3D Manifold 500 4D Manifold Full Kinetics 0 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 2.
02 0.12 10 3D Manifold 4D Manifold Full Kinetics RATE N2 (mol/cm /sec) 8 3 6 4 2 0 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.14 0.08 0.10 H2O 0.16 0.04 3D Manifold 0.18 0.12 0.06 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 4D Manifold Full Kinetics 89 .
04 0.0.00 0.10 0.15 0.02 3D Manifold 4D Manifold Full Kinetics 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.25 4D Manifold Full Kinetics 0.35 0.05 0.12 0.10 CO2 0.08 0.20 CO 0.14 0.06 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 90 .16 0.30 3D Manifold 0.
00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.15 0.05 0.35 0.05 4D Manifold Full Kinetics 0.35 3D Manifold 0.10 0.40 0.0.25 0.20 NO 0.15 0.20 0.30 0.30 4D Manifold Full Kinetics 0.25 N2 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 91 .10 3D Manifold 0.
00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics H2COHNNO2 0.014 3D Manifold 4D Manifold Full Kinetics 0.010 HNC 0.70 0.016 0.00 0.004 0.006 0.008 0.50 0.90 0.012 0.10 0.80 0.1.40 0.30 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 92 .20 0.60 0.002 0.
00025 3D Manifold 4D Manifold Full Kinetics 0.0.00015 HMXR 0.00020 0.00010 0.00015 HMXRO 0.00005 0.00005 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 93 .00025 3D Manifold 4D Manifold Full Kinetics 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.00020 0.00010 0.
020 H2CNNO2 0.005 0.10 0.010 0.40 HMX 0.30 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 94 .60 3D Manifold 4D Manifold Full Kinetics 0.0.20 0.030 3D Manifold 4D Manifold Full Kinetics 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.015 0.025 0.50 0.
00018 0.00008 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 0.00012 H2CNNO 0.0004 0.00020 0.0006 0.00016 0.0.0008 H2CNO 0.00010 0.00014 0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 95 .0012 0.0002 0.0010 3D Manifold 4D Manifold Full Kinetics 0.00006 0.00004 0.00002 0.
00015 0.00010 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 96 .020 3D Manifold 4D Manifold Full Kinetics 0.015 H2CN 0.00030 3D Manifold 4D Manifold Full Kinetics 0.00020 H2CNH 0.005 0.010 0.0.00025 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.00005 0.025 0.
01 0.00E+04 5.00 0.50E+04 3D Manifold 4D Manifold Full Kinetics 3.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3.00E+04 NO3 1.50E+04 1.01 3D Manifold 4D Manifold Full Kinetics 0.00 0.00E+00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 97 .50E+04 2.01 HNO3 0.0.01 0.00E+04 2.00E+03 0.01 0.
014 0.0005 0.0015 HOCN 0.0.002 0.012 3D Manifold 4D Manifold Full Kinetics 0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 98 .004 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.0020 0.006 0.0025 3D Manifold 4D Manifold Full Kinetics 0.008 HCNO 0.0010 0.010 0.
00016 3D Manifold 4D Manifold Full Kinetics 0.0.00002 0.00012 0.05 3D Manifold 4D Manifold Full Kinetics 0.02 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 99 .06 0.00006 0.00004 0.00010 CNO 0.01 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.04 HNCO 0.03 0.00014 0.00008 0.
00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 100 .0002 0.0008 NCO 0.00010 NCN 0.0004 0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.0.00006 0.00008 0.00016 0.00012 0.00002 0.0012 0.00014 0.00018 0.0010 3D Manifold 4D Manifold Full Kinetics 0.0006 0.00004 0.
0.025
3D Manifold 0.020 4D Manifold Full Kinetics
0.015 C2N2 0.010 0.005 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.0006
0.0005
3D Manifold 4D Manifold Full Kinetics
0.0004
CN
0.0003
0.0002
0.0001
0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100
101
0.12
0.10
3D Manifold 4D Manifold Full Kinetics
0.08
N2O
0.06
0.04
0.02
0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100
0.30
0.25
3D Manifold 4D Manifold Full Kinetics
0.20
HCN
0.15
0.10
0.05
0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100
102
0.070 3D Manifold 4D Manifold Full Kinetics
0.060
0.050
0.040 HONO 0.030 0.020 0.010 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.014
0.012
3D Manifold 4D Manifold Full Kinetics
0.010
0.008 HNO 0.006 0.004 0.002 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100
103
006 0.50E06 3D Manifold 4D Manifold Full Kinetics 3.003 0.004 NH3 0.50E06 1.002 0.00E+00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.3.005 3D Manifold 4D Manifold Full Kinetics 0.00E07 0.00E06 5.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 104 .001 0.00E06 NNH 1.50E06 2.00E06 2.
00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 105 .0014 0.0.0002 0.0010 0.0006 0.00005 0.00015 0.00020 0.0004 0.00030 0.00010 0.0012 3D Manifold 4D Manifold Full Kinetics 0.00040 0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.0008 NH2 0.00025 NH 0.00035 0.00045 0.
12 3D Manifold 4D Manifold Full Kinetics 0.00004 N 0.00005 0.04 0.08 0.00001 0.16 0.02 0.00002 0.14 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.00008 0.10 NO2 0.0.06 0.00007 0.00003 0.00006 3D Manifold 4D Manifold Full Kinetics 0.00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 106 .
00010 HCO 0.00012 0.00004 0.01 0.02 0.00014 0.00002 0.08 0.07 0.00006 0.05 CH2O 0.00018 0.06 3D Manifold 4D Manifold Full Kinetics 0.03 0.00016 0.00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 107 .00000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 0.0.00008 0.04 0.
00E05 1.4.00E06 0.00E05 5.50E05 1.00E06 0.50E05 3.00E05 3.50E05 1.00E05 3.00E+00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 108 .00E+00 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 4.50E05 HO2 2.00E05 2.50E05 4.50E05 H2O2 2.00E05 5.50E05 3.00E05 1.00E05 3D Manifold 4D Manifold Full Kinetics 2.
005 0.010 0.0002 0.0004 0.025 0.0010 0.0012 3D Manifold 4D Manifold Full Kinetics 0.0014 0.020 3D Manifold 4D Manifold Full Kinetics 0.0.0006 0.0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 109 .000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 0.015 OH 0.0008 H 0.
004 0.0025 O 0.008 0.010 O2 0.0010 0.0020 0.0005 0.0035 0.016 3D Manifold 4D Manifold Full Kinetics 0.0040 0.012 0.006 0.0015 0.0045 0.014 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 110 .0000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 3D Manifold 4D Manifold Full Kinetics 0.0.0030 0.002 0.
008 H2 0.002 0.004 0.000 0 10 20 30 40 50 DISTANCE (µm) 60 70 80 90 100 111 .0.012 3D Manifold 4D Manifold Full Kinetics 0.010 0.006 0.