DNS of Turbulent Combustion

Jacqueline H. Chen
Sandia National Laboratories
[email protected]
Cambridge Combustion
Institute Summer Shool
July 8-12, 2019
 Motivation, DNS Governing
Equations, Numerical Methods and
HPC Considerations
Jacqueline H. Chen
Sandia National Laboratories
[email protected]
Cambridge Combustion
Institute Summer Shool
July 8-12, 2019
Combustion and Technology Advancement

Industrial Revolution 1760-1840

time
Energy consumption – where does it all go?

IEA World Energy Balances 2018

Source of Transportation World Energy
Consumption

Quadrillion Btu

DOE IEA Oulook 2016

Source of World Net Electricity Generation

Trillion kWh

DOE IEA Oulook 2016

 Temperature’s rising

Global mean surface temperature change from 1880 to 2014, relative to the 1951–1980
mean. The black line is the annual mean and the red line is the 5-year
running mean. The green bars show uncertainty estimates. Source: NASA GISS.
Where are the greenhouses coming from?
Hard to beat the energy density of petroleum!

All of the energy concentrated in one gallon of gasoline is

enough to charge an iPhone once a day for almost 20
years.
Mitigation Strategies for Greenhouse Gas Emissions

• CO2-sequestration (Carbon Capture and Storage, CCS)

• Bio-fuels. (Ethanol)
• Hydrogen-rich fuels
• Increase fuel efficiency
• Other….

Sound combustion foundation for

technology development

24
Combustion is Multi-physics & Multi-scale

• Large range of length and time

scales
– In-cylinder geometry (cm)
– Turbulence-chemistry
(microns-mm)
–Soot inception (nanometer)
• Chemical complexity
– large number of species and
reactions (100’s of species,
thousands of reactions)
• Multi-Physics complexity
– multiphase (sprays, gas
phase, soot)
– thermal radiation
Diesel Engine Autoignition, Soot Incandescence
Chuck Mueller, Sandia National Laboratories
All these are tightly coupled
 Motivation: Changing World of Fuels
and Engines
• Fuel streams are rapidly evolving
– Heavy hydrocarbons
• Oil sands
• Oil shale
• Coal
– New renewable fuel sources
• Ethanol
• Bio-diesel
• Bio-gas

• New engine technologies

• Direct Injection (DI)
• Homogeneous Charge
Compression Ignition (HCCI)
• Low-temperature combustion

• Mixed modes of combustion

• Predictive, validated multi-scale

models!
Comparison of Engines

12
 Fuel- and Load-Flexible Power Generation
Towards Carbon Free Renewable Electricity

Courtesy Y. Ju
Challenges With Hydrogen-Rich Combustion

• Optimized hydrogen fuel transport and combustion system will

enable safe operation within a broad range of fuels and operating
conditions - enable fast fuel switches or blending (e.g. between
natural gas and hydrogen-rich syngas) during fast start-ups and
load-shifts.

• Compared to natural gas, hydrogen has higher stoichiometric

temperatures, reduced ignition delay, smaller calorific value,
requires higher volumetric flow rates to avoid preignition and to
ensure flashback into premixer section doesn’t occur

• Fundamental challenges to lean premixed hydrogen-enriched

combustion in gas turbines:
• Flame flashback
• Fuel injection: flame holding and stabilization
• Thermo-acoustics
 Fuel and Load Flexibility Through Staging

• Originally developed by ABB for high efficiency, load flexibility

and low emissions
• Recently improved and simplified (reduced cost) for the H-class
GT36
• First (premix) combustion stage based on flame propagation
• Second (sequential) combustion stage based on auto-ignition
 Fuel and Load Flexibility Through Staging

• Adjusting firing temperature of 1st stage allows control of tign in

2nd stage
 Fuel and Load Flexibility Through Staging

Hydrogen fuel
• Flashback in 1st stage
• Early auto-ignition in 2nd stage

• 2nd stage is mainly auto-ignition stabilized

• 2nd stage inlet temperature needs to be decreased and not 2nd
stage flame temperature
 Fuel and Load Flexibility Through Staging
Hydrogen fuel
• Flashback in 1st stage
• Early auto-ignition in 2nd stage

• 2nd stage is mainly auto-ignition stabilized

• 2nd stage inlet temperature needs to be decreased and not 2nd
stage flame temperature
• 1st stage de-rating is compensated by shifting fuel to 2nd stage
 Outline

• Motivation
• Research paradigms
• Governing equations, numerical methods, high
performance computing
 Focus: Turbulence-Chemistry Interactions

• Turbulence entrains, advects, strains and

wrinkles a flame creating more area for
burning

• Through turbulence cascade eventually

reactants are molecularly mixed

• Chemical reactions are enhanced with

mixing to a limit – extinction - and create
heat release at dissipation scales

• Heat release, dilatation reduce turbulence

intensity through density, and property
changes and may alter spectral energy
transfer
 Combustion Research Paradigms:
Experiment/Computation/Theory
• Experiments in laboratory scale flames/ignition at ambient
and elevated pressure with advanced laser diagnostics are
critical.
http://www.sandia.gov/ecn/
• High-fidelity simulations (DNS and LES) play a
complementary role:

– Precise control of parameters

– Desired level of detail
– Full specification of boundary and initial
conditions (scalars, velocity)
– Less expensive, faster design-cycle

• Computational power and algorithms are advancing

rapidly.

• Theory needed for interpretation, scaling, for

identifying reduced manifolds and closure models
Turbulent Combustion: Computational
Considerations

• Combustion in most practical devices occurs in turbulent

flows where the ratio of inertial to viscous forces is large
• It is both:
– Multi-scale ( ~ 4-6 decades in space and time).
– Multi-physics (fluid dynamics, molecular transport, chemical kinetics, multi-
phase).

• Governing equations represent a tightly coupled system with

large dimensionality (5 + Nspecies).
• Computational cost becomes prohibitive very quickly.
Spatial Scales (Turbulence, Mixing, and Flames)
Spatial Resolution – Nonreacting Turbulence

-5/3

log(E)

Δ ≈η 1/L log(k) 1/η

Ngrid ≈(L/η)3 Ngrid ≈ Re9/4

L/η ≈Re3/4
 Spatial Resolution Reacting Turbulence

• For turbulent combustion: L≈ 10-1,Re ≈ 104, η ≈ 10-4m

– δ<η
δ ≈ (diffusivity/flame speed) ≈ 10-5m
Ngrid ≈ Re9/4 (η/δ)3

– No. of equations is large

-5/3

log(E
)
Nspecies ≈ O(101-102) 1/δ

1/L log(k) 1/η

Temporal Resolution
Multi-scale Modeling of Combustion Processes

• Multi-scale modeling describes

combustion processes, from
quantum scales up to device-level,
continuum scales
• Multi-scale Strategy:
• Use petascale computing power to perform
direct simulation at the atomistic and fine-
continuum scales (~4 decades), and
develop new parameterizations that will
enable bootstrapping information upscale
 DNS –LES –RANS
• Reynolds Averaged Navier Stokes modeled
(RANS)
(RANS) resolved
– No attempt to resolve any scales (DNS)
-5/3

log(E)
– Inexpensive
resolved modeled
• Large Eddy Simulation (LES) LES

– Energy containing scales are resolved

– Model subgrid physics
1/L log(k) 1/η 1/δ
• Direct Numerical Simulation (DNS)
Filtered
– All continuum scales are resolved DNS DNS LES RANS
– Most expensive
 Combustion Simulation Approaches
(DNS/LES/RANS)

• RANS and LES require models for

Chemistry

• Active mostly at small scales

Turbulence
• Coupling is highly non-linear

• Driven by large scales of motion Transport

• L, u’, µ drive the cascade

• DNSis often used to develop/validate models.

• DNSresolves all turbulence scales, no turbulenceclosure.
• DNSis not free of modeling (needs chemical and molecular transport
models).
 Outline

• Motivation
• Research paradigms

• DNS mathematical formulation, numerical methods, and

high performance computing
 Multi-scale Modeling of Combustion Processes

•Multi-scale modeling describes

combustion processes, from quantum
scales up to device-level, continuum
scales
• Multi-scale Strategy:
Use petascale computing power to perform
direct simulation at the atomistic and fine-
continuum scales (~4 decades), and develop
new parameterizations that will enable
bootstrapping information upscale
 Governing Equations – Compressible

Chen et al., Comp. Sci. Disc., 2009

Governing Equations - Compressible
Constitutive Relations
Mass Diffusion Flux
Mass Diffusion Flux / Heat Flux
 Direct Numerical Simulation Code – S3D
Chen et al., Comp. Sci. Disc., 2009

• Used to perform first-principles-based

DNS of reacting flows
• Solves compressible reacting Navier-
Stokes equations
• High-fidelity numerical methods
• Detailed reaction kinetics and
molecular transport models
• Multi-physics: sprays, radiation and
soot
• Ported to all major platforms, scales
well
• Particle tracking cabability

Physical Engineering
DNS models CFD codes
(RANS, LES)
S3D Numerical Methods

– Finite difference solver (physical space).

– Rectangular Cartesian regular mesh.
– Explicit 8th order central difference (Kennedy & Carpenter, Appl. Num.
Math., 1994, vol. 14).
– Explicit 4th order 6-stage Runge-Kutta (Kennedy & Carpenter, Appl. Num.
Math., 2000, vol. 35).
– 10th order filter to remove spurious numerical noise.
– Δx, Δt resolve all relevant spatio-temporal scales (fluid-dynamic, thermo-
chemical, acoustic).
– Boundary Conditions:
• NSCBC for Non-reflecting inflow/outflow (Poinsot and Lele, J. Comp.
Phys. 1992, Sutherland and Kennedy, J. Comp. Phys. 2003, Yoo and
Im, Comb. Theory Mod. 2007)
• Periodic
• Isothermal or adiabatic wall
 Chemical Complexity vs Scale

2-methyl alkanes (LLNL)

methyl palmitate (CNRS)

Ranzi mechanism
comlete, ver 1201 Biodiesel (LLNL)
4
10 C16 (LLNL)
MD (LLNL)
Number of reactions, I

C12 (LLNL)
C10 (LLNL) C14 (LLNL)
iso-octane (LLNL) PRF (LLNL)
iso-octane (ENSIC-CNRS)
n-heptane (LLNL)
Gasoline (Raj et al) JetSURF 2.0
CH4 (Konnov)
103 USC C1-C4
skeletal iso-octane (Lu & Law)
skeletal n-heptane (Lu & Law)
USC C2H4 n-butane (LLNL)
1,3-Butadiene
C1-C3 (Qin et al) DME (Curran)
GRI3.0 neo-pentane (LLNL)
before 2000
C2H4 (San Diego)
GRI1.2
CH4 (Leeds)
2000-2004
2 2005-2009
10
I = 5K since 2010

101 102 103 104

Number of species, K
T. Lu and C. K. Law, PECS 35, 2009
 A Systematic Procedure for Dimension
Reduction & Stiffness Removal
Detailed mechanisms Skeletal mechanisms
C2H4: 70 species C2H4: 30 species
nC7H16: 500 species nC7H16: 100 species
Dimension Reduc5on Time Scale Reduc5on
With DRG With QSSA

Reduced mechanisms
C2H4: 20 species
Minimal diffusive species nC7H16: 60 species
C2H4: 9 groups
nC7H16: 20 groups
On-the-fly Diffusive SpeciesBundling
S5ffness Removal

Non-stiff reduced mechanisms

C2H4: 20 species
nC7H16: 60 species

Lu and Law 2005, Lu and Law 2006,2008, Lu and Law 2007, Lu et al. 2009
 S3D Parallelization – Data Parallelism

• Algorithm:
– fixed mesh – trivial 3D domaindecomposition
(each MPI process computes on a piece of the 3D
domain)
– MPI only parallelism (hybrid MPI+OpenMP,
Open ACC,OpenMP +CUDA). On JAGUAR, at NCCS (Oak Ridge National Lab)

– ideal load balancing (all processes havesame

amount of work).
– communication is almost entirely between
nearest neighbors – ghost cells, large
message sizes, nonblocking sends/receives
– parallel collective I/O
– per node performance limited by memory
bandwidth.
– extremely scalable and portable (CRAY,IBM,
INTEL).
 Domain Decomposition and Bulk Synchrony

Bulk Sychronous

1 2

3 4

.Each MPI process is in charge of a piece of the 3D domain.

Inter-processor communication is only between nearest neighbors
 Compressible DNS Solver – Main Loop
• Computes rate of change of N conserved quantities at everygrid
point
– d/dt (Qk) = (Advection) + (Diffusion) +(Source)
– Sum of all the terms that contribute to the time derivative iscalled
the RHS
• d/dt (Qk) is integrated explicitly in time throughRunge-Kutta (4th
order – 6 stage, Kennedy and Carpenter)
• RHS contains multiple terms that are functions ofQk, variables
derived from Qk
• Advection and diffusion require spatial finite differencingand
nearest neighbor ghost zone communication, MPI
• Source terms are point-wise functions
• Thermodynamic, chemical and molecular transport properties are
point-wise functions of Qk
 Computational Intensity of Chemistry
• Called as ckwyp or getrates (Chemkin)
• Chemical reaction rate computed using Arrhenius model
– A + B = C+ D
– Forward reaction rate =C*[A]*[B]*Taexp(-Ta/T)
– Equilibrium constant gives reverse reaction rates
– More terms for thirdbody efficiency, collision efficiency,
pressure corrections …
• The source term for a species is the sum of the rates of all
reactions in which it participates
• The kernel uses exp/log heavily
 Computational intensity of DNS scales with Moore’s Law
Lean CH4Piloted
3D turbulent 2-stage laboratory-scale
3Dturbulent autoignition at 25 bar jet flame in the
Jet flames with O(35-116)species distributed
With O(20) reaction zones
species regime withNOx

2D unsteadyflows [6]
With global step 1.9e11
chemistry [5]
[4] 1.8x1011
9.7x1010

[3]
3.5x109

[2]
2.3x108
problemsize=# grid pts x# variables
[1] (e.g. chemicalspecies)
3.0x106

1 T.Echekki, J.H. Chen, Comb. Flame, 1996, vol.106.

2 T.Echekki, J.H, Chen, Proc. Comb. Inst., 2002, vol. 29.
3 R. Sankaran, E.R. Hawkes, J.H. Chen, Proc. Comb. Inst., 2007, vol. 31.
4 E.R. Hawkes, O. Chatakonda, H. Kolla, A.R. Kerstein, J.H. Chen, Comb. Flame, 2012, vol. 159.
5 2015 submission for Gordon Bell prize
6 H. Wang, E. Hawkes, J. H. Chen, Comb. Flame 2017
 DNS Benchmarks
(Physical Insights & Model Development)
CO/H2 and C2H4 jet
Lifted H2 and C2H4 jet
flames extinction/
flames in hot coflow
reignition
Di-Methyl Ether HCCI

Lean
premixed
and
stratified
CH4/air
Bunsen
flames

H2 Flame-Boundary Layer H2 and CO/H2 Jet in

Interaction - Flashback Crossflow
 Constraints imposed by exascale architecture
• Power: primary design constraint
for future HPC system design On-Package
Memory Core ... Core
• Cost: Data movement dominates:
optimize to minimize data
NUMA Domain

External Interconnect

External Network
movement Network-on-Chip
DRAM

Interface
• Concurrency: Exponential growth of NUMA Domain

parallelism within chips Core ... Core

NVRAM

• Locality: must reason about data

locality and possibly topology Accelerator

• Memory Scaling: Compute growing On-Package

2x faster than capacity or Acc. Memory

bandwidth, Node

• Heterogeneity:Architectural and Conceptual model of future HPC node

Figure 2.1: Conceptual Model of a Future High Performance Computing
performance nonuniformity

Express data locality and independence, express massive parallelism,

minimize data movement and reduce synchronization, detect and address faults

On-Package
DOE Exascale Computing Project (ECP) will
achieve capable exascale machines in 2021-2023

Application Software Hardware Exascale

Development Technology Technology Systems
Science and Scalable and Hardware Integrated
mission productive technology exascale
applications software stack elements supercomputers
applica
AppliCo-Design
Correctness Visualization DataAnalysis

ti Programmingmodels,
development environmen t, Math libraries and
Tools
Frameworks

Workflows
Resilience

andruntimes

System Software, resource

management threading, Data
scheduling, monitoring, and Memory management
andBurst I/O and file
control buffer system
Node OS,runtimes

Hardwareinterface

ECP’s work encompasses applications, system software, hardware

technologies and architectures, and workforce development

From Paul Messina’s ASCAC talk April 19, 2017

 ECP application: transforming combustion science and
technology through exascale simulation (Pele)
PeleC and PeleLM: Block-structured
adaptive mesh refinement, multi-physics:
spray, soot, and radiation, real gas,
complex geometry

Effects of reactivity
stratification at:
• high pressure
• high turbulence
• fuel blends
on:
• ignition delay
• combustion
Automated Mechanism Generation
rates
• emissions Fuel Conditions

Rate Rules
Mechanism Generator (RMG)
Specific Rates
Uncertainty
Mechanism

1st Iteration
No Yes

S3D: multi-block compressible reacting Chemical Sim.

Theoretical Rate
Constants (EStokTP)
Method

Reaction(s)

DNS multi-physics validation: spray, Observables

Dakota
Sensitivities

soot, radiation Chem. Sim.

UQ range
Acceptable
yes no

Mechanism Reduction Engine Sim.

DRGASA Acceptable
yes no

CFD Simulations
of RCCI
DONE
Programming Environment Critical to
Performance
Effective use of exascale hardware will require programming
environment that effectively maps algorithms to hardware

• Driven by programmability of combustion applications and

characterization of algorithms on different designs of
architectures
– Simplify programming to express locality and independence
– Automate discovery of parallelism and hide latencies
– Simplify programming of extensible workflows, block-structured
PDE’s, analytics, UQ for performance, scalability, portability, and
productivity on heterogeneous architectures
 Parallel Programming 101 - Productivity

Functionally correct Mapping to target

application code machine

Management of
Extraction of parallelism
data transfers

Task scheduling Data-Dependent

Latency hiding Behavior

52 Exascale Computing Project

 Parallel Programming 101

Functionally correct Mapping to target

application code machine

Compiler/Runtime
Management of
Extraction of parallelism understanding of
data transfers
data

Task scheduling Data-Dependent

Latency hiding Behavior

53 Exascale Computing Project

 Legion Programming System applied to S3D
• A data-centric parallel programming system
• A programming model for heterogeneous, distributed machines
– Automates many aspects of achieving high performance, such as extracting task- and
data-level parallelism
– Automates details of scheduling tasks and data movement (performance optimization)
– Separates the specification of tasks and data from the mapping onto a machine
(performance portability)
• Legion application example: S3D
S3D
– Production combustion simulation
Production combustion simulation
– Written in ~200K lines of Fortran
– Direct numerical simulation using explicit
Written in ~200K lines of Fortran

methods
Direct numerical simulation using explicit methods

36 S3D performance Legion vs. MPI

S. Treichler et al., “S3D-Legion: An Exascale Software for Direct Numerical

Simulation (DNS) of Turbulent Combustion with Complex Multicomponent
Chemistry,” CRC Book on Exascale Scientific Applications: Programming
Approaches for Scalability Performance and Portability, 2017.
54 Exascale Computing Project, www.exascaleproject.org
 In-situ Data Analytics in Legion
Chemical Explosive Mode Analytics (CEMA)
CEMA: eigenvalue solve on the reaction rate Jacobian to determine the mode of
combustion

Run CEMA at each time step as a diagnostic to steer mesh refinement

CEMA computation takes longer than a single explicit RK stage (6 stages/timestep)
Dividing CEMA across RK stages and interleaving with other computation so as not to
impact other critical operations would be hard to schedule manually
Asynchronous task execution, schedule CEMA on CPU resources
Interoperate Fortran CEMA with Legion code – took a day to implement
 Execution Overhead of In-situ Analytics
(CEMA) in S3D-Legion (Titan & Piz Daint)

84%
98% reduction
reduction

56 Exascale Computing Project

 Legion S3D Lessons Learned

• Legion
– S3D shows potential of data-centric, task-based models
– Enables new simulation capabilities (physics, and in situ analytics)
– Code is easier to modify and maintain
• Ports are just new mappings, easy to tune for performance
• New functionality usually just means new tasks
• Legion will figure out the dependences and scheduling
• Productivity requires higher level abstraction layer for scientists to write in

• Co-Design and ECP

– The Legion/S3D experience is a tribute to co-design
– Computer and computational scientists worked closely
– Major progress on important problems resulted

57 Exascale Computing Project

DNS Relevant to Compression Ignition
Engines and Stationary Reheat Gas
Turbines
Jacqueline H. Chen
Sandia National Laboratories
[email protected]
Cambridge Combustion
Institute Summer Shool
July 8-12, 2019
 Outline

• Compression Ignition Engines:

– 2D lifted DME laminar flames spanning NTC regime
– 2D DME mixing layer ignition
– 3D turbulent autoignition in n-dodecane temporal jet
– 3D multi-injection mixing and combustion at diesel
conditions
• Reheat Combustion
 Background
Dec SAE 1997-97-0873
• Diesel engines will used for trucks
and shipping for the foreseeable
future
• Regulations for pollutant emissions
are stringent
• Engine design spans a large
parameter space, wide operating
Figure: Dec’s model for conventional diesel
regimes, complex multi-physics. combustion.
Requires predictive engineering
Chapter 1. Introduction 3

• Stanton et al. SAE 2013-23-0094

models to aid design.

• Model development requires more
and higher-fidelity data.

Figure: Historical pollutant reduction with improving

Figure 1.1: Historical reductions in specific NOx and soot emissions for heavy

technology
duty diesel engines [10].

This requires detailed observations at real device conditions in order to develop and
validate the required theory and models.
 Experiments and conceptual models
Chapter 2. Literature review
Observations 31
Downloaded from SAE International by University of New South W

• Great progress from community LII

10% O2, 1000 K
LII

efforts (e.g. see Engine Combustion Downloaded from SAE International by University of New South Wales, Wednesday, May 25, 2016

Skeen et al / SAE Int. J. Engines / Volume 9, Issue 2 (June 2016)

Network at:
888

HCHO/PAH P
HCHO
ignition delay is around twice as long as the experimental value at the the modeling parameters influencing ignition delay and lift-off length

http://www.sandia.gov/ecn/
Spray A condition, which complicates the comparison of time- among the various models is warranted P and should be the subject of
resolved soot results later. future ECN work.
10 10
5
OH* 5
OH*

• Mean flame structure, global

Lift-off Length
0 Soot Optical Properties and TEM Imaging for Soot0
Five institutions participating in the ECN have measured the -5 Morphology -5
quasi-steady lift-off length at the Spray A baseline condition, with -10 Knowledge of the in-situ soot optical properties and morphology -10 is

observables (LOL, !) well- two institutions contributing data at lower ambient temperature and
only SNL providing measurements at 1000 K and above. Figure 3
Pickett et al, SAE 2005-01-3843
displaysIsolated
the experimental and simulated quasi-steady lift-off lengths
Figure 2.12: Chemiluminescence imaging of diesel combustion from Ref. [73].
0necessary
10 to20
of soot volume
30
fractionfrom
Idicheria, Pickett, SAE 2006-01-3434
40 50
(fv) ininjector
60
[mm]
flames. For
70
reduce uncertainty when making optical measurements
Distance
80
LII measurements
calibrated by laser extinction or extinction imaging measurements as
0 10

documented.
pockets are observed upstream of the flame base due to autoignition.
as a function of ambient temperature under the same conditions as Figure 5. LII,will formaldehyde/PAH PLIF and chemiluminescence images for low-soot
be presented below, the dimensionless extinction coefficient (ke)
and 21% O2, 900 K (right).
Fig.as2.a two-stage
For the ambient temperature
process, since substantial conditions
levels of CH2 O atand
900CHOK and werebelow,
detected. is a critical parameter in relating the measured optical attenuation
theObservations
standard errorof LTC among
prior tothethedifferent contributing
main autoginition institutions
event have is ferred
been observed in to as(I/Ithe) tolow-soot
fv throughcondition),
Bouguer’s Law. Becauseinkethe
as shown is notLIIonly dependent
would form
0

• Diagnostics: Schlieren, LII, PLIF, CL,

provided.
subsequent For 1000atKsimilar
studies and above,
conditionsthe[20,
standard error isofderived
74–76]. Imaging fromimages in Fig.
the quasi-steady 5.
on particle refractive index, but also morphology, care must be creasing taken a
twolifted
independent SNL experimental
flame also detected significant LTC campaigns.
upstream of the The
meanliftoff lengthloca-
stabilisation to account for the effect of soot particle aggregation on the scattering- the continu
results
tion. for
The the Spray
presence A baseline
of the case are asalso
LTC was significant tabulated in
it demonstrated figure The PLIF images
theautoignition
that to-absorptionin Fig. 5 (α
ratio show
sa
). two
Provided distinct
the regions
appropriate for
particle and off length

etc – much can be observed.

forchemistry
reference. canThe
existANL resultof at
in advance the900 K is thelocation,
stabilisation average andofis15 LES
likely both 10% and
to a↵ect aggregate O2 conditions.
21% parameters can beFirst, similarthe
determined, to influence tion zone—
the no- of aggregation

Validation data Conceptual models fortions. The

realizations,
the LOL either while for the other
by transitioning ambient
to the temperatures
main autoignition event, oronly a singlethe
by modifying soot condition,
can beacaptured
signal by begins upstream of the approximation
the Rayleigh-Debye-Gans lift-off
realization was performed. The tabulated ANL result for Spray A also location and then goes away near the hypothesized fuel- close to th
upstream propagation of a premixed flame feature. fractal aggregates (RDG-FA). The information on particle size and
includes the standard error. rich premixed reaction zone. However, unlike the no- and PLIF i
morphology necessary to derive ke via RDG-FA calculations includes

• Some things can not be measured,

The role of LTC on diesel flame stabilisation was further investigated by Idicheria soot condition, a signal reappears downstream of this upstream
location andthe extends
complex refractive
over the index
length (m), of
primary
the particle re-
image diameter (dp),stream, wh
and Pickett [20]. In that study, the oxygen concentration and sooting propensity of
fractal prefactor
gion. For example, in the(k10%
f
), fractal
O2, dimension (Df), and aggregate
1000 K condition, the sizea(Nmixture
p
). o
This section provides
mm aand briefreappears
explanation at of the
60 soot
mm.refractive index

most details can not be resolved

signal disappears at 50
Likewise, in the 21% O2, 900 K condition, the signal dis-
selected for use by the ECN along with a review of TEM The result
imaging
appears atresults
40 mm fromandECNeventually
participants reappears
and others at 50 mm.
providing no-soot co
the necessary
From the arguments presented
determine kin the previous section ing conditi

(small scales at high pressure).

information to e
in high-pressure spray flames.
and the similarity of the upstream signal structure, any was unkno
LIF signal Numerous
upstreamstudies of thehavelift-off length isthelikely
investigated formal-
refractive served in
index of flame-
dehyde, formed
formed soot as well as other types of carbonaceous particles as caused by
in a cool-flame that precedes high-
temperature reaction. Also similar to the no-soot condi- at the con

• Open questions remain (flame

reviewed by Smyth and Shaddix [46] and Bond and Bergstrom [47].
tion results, the fact that the LIF signal disappears com- sors (PAH
pletely whenWhile
movingthe refractive
downstreamindex applied
suggests for extinction measurementsinternal
that formal- can re
be dependent on a variety of
dehyde is consumed in a high-temperature, premixed parameters including the wavelength
sion of flame

stabilization mechanism/ignition the incident light and soot composition (i.e., carbon-to-hydrogen
reaction zone.
ratio), the ECN community has adopted the refractive index proposed
The reappearance
by Williams of etan
al. LIF
[48] signal downstream
of 1.75-1.03i. Althoughof Williams
By changin
condition t
the et al. PAH forma

dynamics, regime diagram) Figure 3. Experimental and simulated quasi-steady lift-off lengths (H) for
SpraySkeen et al. variants
A and its parametric SAE in2016-01-0734
ambient temperature.
high-temperature
LIF of PAH,
indicatedreaction
635anm,
further downstream. The fact
the visible spectrum
zone is most
that this refractive
precursor
we apply to thissoot
index likely
valueformation
that an the
to simplify
caused
was specific
thatacross
as the standard
LIFcomparison
occurs
by
to measurements
measurements
signal
at
sumption
in
of ECN soot the no-soo
signal disap-
dow

In comparing the simulated lift-off lengths with the experimental pears and then forms again
measurements is useful
performed for interpretation
by different institutions andof with different
results, we observe good agreement from the POL and UNSW tPDF
results in several ways. First, it is unlikely that the reap- Summariz
incident wavelengths. The consequences of choosing a constant
pearing signal is from formaldehyde because tempera- condition:
models, while the UNSW well-mixed model diverges from the refractive index for the analysis of soot in spray flames is discussed
tures near the jet center increase with increasing axial stream of
experiment at 800 K and 1000 K. The ETHZ lift-off length also distance due latertointhe
the high-temperature,
context of the ke derived from RDG-FA
premixed reaction calcualtions.temperatu
More
diverges at these two temperatures; however, in contrast to the information
zone and further on this
mixing subject
with hot can be found in products
combustion Manin et al. [11]. of the hig
UNSW well-mixed model, the lift-off length at 800 K is instead from the diffusion flame. Formaldehyde that is con- formaldehy
under-predicted by ETHZ. The ANL lift-off length trend is nearly sumed during Withthe high-temperature
regard to primary particlereactions
size, TEMwould
analysisnot
published inthe fuel je
linear, with a shorter lift-off length than the experiment at 800 K and be expected to form
Cenker et al.again if the atemperature
[14] showed count median continues
diameter of 9.1 nm for formaldehy
soot
a longer value at 900 K and above; however, this might be due to the to rise. Second,
sampledit60-mm is also fromunlikely that orifice
the injector PAH under
molecules eling sectio
the Spray A condition.
insufficient number of LES realizations for temperatures other than While this was the only spatial location sampled for Spray A, TEM
7
the Spray A case. Although the standard error of the lift-off length for analysis was also performed on samples extracted at three axial
the 15 LES realizations is quite small, fluctuations in lift-off length locations, namely, 36 mm, 45 mm, and 60 mm from the 21% O2
extend down to 17 mm and up to 24 mm during the transient parametric variant of Spray A. Although the higher ambient oxygen
 LES and RANS
• LES and RANS needed for industry
• Performance is good for global observables, less good for fine details
• Difficult to test and discriminate between models due to sparse (and unresolved) data
• Motivation for DNS – augment existing knowledge with fully-resolved information. Care
Downloaded from SAE International by University of New South Wales, Wednesday, May 25, 2016

needed! Skeen et al / SAE Int. J. Engines / Volume 9, Issue 2 (June 2016) 891

Skeen et al. SAE 2016-01-0734 (2016) [5]

Pei et al. CNF 168 (2016) [6]

Figure: Comparison of models against experiment.

Figure 5. Top two rows: Time-series of fv (IFPEN, LII) and KL (SNL, DBIEI) for Spray A (1.5-ms injection duration). The dashed lines designate the location selected
for the radial cross section profiles shown in Fig. 6. Bottom five rows: Time-series of fv as simulated by ANL, ETHZ, POL, UNSW, and UW. Note that the UNSW tPDF
and WM results are presented in a single image.
Figure: Ignition transient. TPDF modelling of
Review of ECN progress on soot modelling (Sibendu
With the exception of the earliest timing shown in Fig. 6, the Fig. 7. For reference, we have included SNL data acquired in 2012 Spray A.
Som et al. 2016)
experimental radial cross sections of fv from IFPEN and SNL are with the original DBIEI setup using 406-nm and 519-nm incident
remarkably similar both in shape and in amplitude. This is in spite light. Recall that the IFPEN data were acquired as a single shot at a
of the requirement, discussed above, that the projected KL data be specific instant in time for several repeated injection events, while
.
interpolated for multiple viewing angles before tomographic
reconstruction. The larger fv measured by DBIEI at 0.9 ms may be
the DBIEI data were acquired at 30-45 kHz using high-speed
imaging. With the exception of the soot mass determined from the
 Importance of Premixed Flames and Ignition
in Diesel
Importance of Spray Flames
Premixed Flames and Ignition in Diesel
Spray Flames

• Ignition delay is controlled by the combined effect of

autoignition kinetics of a given fuel together with the mixing
history

• Lift-off length is affected by the local premixed combustion mode

due to competition between autoignition and premixed flame
propagation

• Need to be accounted for in diesel spray flame modeling

Regime Classification for Exothermic
Reaction Fronts in Stratified Mixtures
 Deflagration (Premixed Flames) Versus
Spontaneous Ignition (Zeldovich 1980)
• Mixtures can contain a spatial gradient of ignition
delay time:
– Spatial gradient of temperature
– Spatial gradient of concentration
– Spatial gradient of reactivity (mixing of two fuels
with different ignition properties)

• Large spatial gradients in ignition delay lead to

deflagration
– Balance between reaction and transport

• Small gradients in ignition delay lead to spontaneous

ignition
– Reaction dominates over transport

• Turbulent mixing can modify scalar gradients

 Background on high-temperature ignition
68 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97

8 945
7 Li et al.
Del Alamo et al.
6
O’Conaire et al.
• Homogeneous PSR calculations 5 Yetter et al. 940

show the existence of x value, xm,

ignition delay tign [ms]

4
935
where ignition delay has a

T [K]
3

minimum; 930
2

• Flamelet simulations show the 925

ignition delay increases with scalar

dissipation N until a critical value is
Fig. 7. Conditionally-averaged reaction rate against mixture fraction from one-step
66
1
0 0.02 0.04 0.06 0.08 0.1
920
E. Mastorakos / Progress in Energy and Combustion Science 35

reached;
DNS. The times at which the conditional averages were made, normalized by the mixture fraction Z [-]
resulting autoignition time for this simulation (with autoignition defined as the first
appearance of a burning spot), are 0.67 (open circles), 0.86 (filled circles) and 0.99 3.0 with one-step
Fig. 8. Calculations of autoignition time in homogeneous hydrogen-air mixtures at
(triangles). Reprinted from Ref. [11], with permission from Elsevier. 1 bar with various detailed
detailedmechanism
mechanisms: mixture fractio
Li et al. [104], Yetter et al. [80], Del Alamo
− VODPK
• In practical systems, ignition occurs et al. [105], and O’Conaire
reducedet
Y1,0 ¼ 0.13 (the restreduced
al. [106]. −The
mechanism
in N2), mechanism
T2,0 ¼ 945 K,
hydrogen mass fraction in the fuel stream is
CHEMEQ2
T1,0 ¼ 720 K. The initial species mass fractions
− VODPK
than a straigh
detailed chemis

at locations
are doneclose tocomputational
xm where cost) to N is
us to project with some confidence homogeneous calculations (that
with modest inhomogeneous
and temperature
2.5 (dashed line; right axis) are functions of the mixture fraction cor-
responding to frozen mixing. sref is the minimum value (that depends on the scheme)
mechanism in 2
decaying. Probl

ignition delay [ms]

and xMR the corresponding mixture fraction. Calculations performed by Frouzakis and
low. The ignition
situations.
delay
If the fuel is distributed in theis longer
oxidizer in a manner that global
Kerkemeier from ETH Zurich. Reproduced from [102]. (in 2-D) [100],
3-D) [101], and

than in PSR;
mixing may occur quickly and the extrema of x shrink so that the
nominal x (from one of the methods discussed above) ceases to
MR
2.0
temperature) kernels are located at mixture fractions leaner than
air (in 2-D) [102
results compare
exist in the flow, the evolution of temperature in x space can be xst, although a quantification of xMR was not attempted. Later datapeak of the con
• Question: which features of high-T
somewhat different than the above picture [95]. Nevertheless, the
fundamental finding that a competition between the high
from a methane jet [111], with the co-flow temperature now at
1350 1.5K, also showed that the highest temperature across mixture
a large amount
shift depends o

ignition carry over to low-T ignition?

temperature at low x and the high concentrations at higher x leads
to a preferred mixture fraction and that this value does not change
fraction space evolves from a very lean x to the stoichiometric with
downstream distance (Fig. 9), in qualitative agreement with the
may not be si
mechanism use
much with flow configuration, remains valid. shapes from the examples of laminar flamelet simulations shown in pressure on xMR
1.0 observe in Fig. 7
A situation more relevant to HCCI combustion, with autoignition Fig. 4 and1 the DNS (Fig.10 6). In the methane 100 case, it looks as if the
1000
achieved by compression, has also been studied. DNS with one-step initial temperature increment occurs at very
peak scalar dissipation rate N0 [1/s] lean mixture fractions,mixture fraction
chemistry has shown that if the fuel and air have the same consistent with homogeneous autoignition calculations of sign in low activation e
temperature, autoignition happens around the stoichiometric that
Fig. paper that
5. Calculations show a minimum
of autoignition at inananequivalence
time of heptane unsteady flameletratio for that simulat
of 0.05,
between
mixture fraction [107], although not many details were included for Figure: ignition delay time in PSR and
corresponding
hot to xunity
air and cold fuel with MR ¼ 0.01
Lewis for the
numbers jetconstant
under dilutions
scalarused. 4
The
dissipation relatively wide i
withdevel-

[1] E. Mastorakos, PECS (2009), pp. 57-97 flame simulatio

reduced and detailed mechanisms. Reprinted from Ref. [84], with permission from
the temperature in x space. A 2-D simulation using a detailed opment between these lean mixture fractions and the stoichio-
Elsevier. The potential effect on the results of the numerical solver is also evident. 1
scheme for hydrogen and including temperature non-uniformities in nonpremixed flamelet simulation
metric, when examined in terms of conditional averages, seems Is there a w
but uniform composition, showed the difference between local gradual (see also Fig. 5 of Ref. [82] for transient flamelet methane methods may b
The same observations have been found in turbulent situations, First, a series o
autoignition and deflagration following a nearby autoignition site calculations at 40 bar demonstrating the same trend). These
 How to make a DNS diesel-relevant?
• The actual diesel condition may never be feasible with DNS
– ReJET ~ 5e5, variable and extremely complex fuel, multi-phase behaviors,
radiation, sparse validation data (…and many more limits!)
• Need to target facets of the problem, play to the strength of DNS
– Identify useful but achievable questions of interest that can not be answered
by other means
– Contrive simplest possible DNS to conduct investigation

Question
What role does NTC and two-stage ignition have on the ignition and flame-
stabilization processes?

Configurations
Targeting high pressure and NTC behaviour
1. Lifted laminar flames 2. Mixing layer ignition 3. Turbulent jet ignition
 Partially premixed flames – edge flames
• Partially premixed combustion at
sufficiently low temperature conditions
•Leading edge with three flame
branches (triple point)
•Lean premixed
•Diffusion
•Rich premixed
Right Schlieren images from H. Phillips,
PROCI (1965)

• 3D DNS of turbulent lifted jet flames

•Y. Mizobuchi et al., 2002: H2, 1atm

Fuel

Fuel
Air

Air
•C. S. Yoo et al., 2009: H2, 1 atm
•C. S. Yoo et al., 2011: Ethylene, 1 atm
•S. Karami et al., 2014: 1atm
 Lifted laminar flames
Overview
,-(./01
• Conventional diesel combustion resembles
quasi-steady lifted flames.
• Stabilization mechanism unclear, but has
important consequences.
• Goal: Investigate flame structure and

$%&&'()*+
,-(./01
stabilization mechanism across range of
diesel-relevant temperatures (span NTC). 3*) 2-'/
• Fuel is DME, features NTC, two-stage
ignition. Similar cetane number to diesel. 4
• 30 species reduced mech, validated.
Conditions
Ø P = 40 atm
Ø TOX = 700à1500 K
Ø Fuel: DME 70% N2 30%. #
1
Ø Oxidiser: O2 21% N2 79%. " 0

Figure: Initial and boundary conditions.

Krisman, 2016
 Chapter 4. Lifted laminar flames

Lifted laminar flames Table 4.1: Values of variable parameters for eac

Stabilization transition Case

1
2
Tox [K]
700
900
U [ms 1 ]
1.25
3.50
Krisman et al. PROCI 35 (2015)
3 1100 5.75
4 1300 10.00
5 1500 45.00

Table: Parametric variation across cases.

near-constant lift-o↵ length. Table 1 lists the oxidiser tempera
for each case.

The chemical mechanism is a 30 species reduced mechanism d

and introduced in section 3.1.5.1. As earlier discussed in secti
nism was obtained by a reduction of the mechanism develope

Edge flame using reduction methods outlined by Lu and Law [134]. Th

Hybrid Autoignition
propagation duced mechanism was validated against the detailed mechan

Increasing T and inlet velocity mixture-averaged molecular transport model was employed [1

• Complex transition. One or two upstream branches

observed (tetrabrachial, pentabrachial). 4.3 Grid convergence
• Transition smooth, wide range of hybrid structures.
• Flame speeds (non-turbulent) O(10) m/s, comparable
To assess the chemical length scale, a stabilised, lifted, one-d
Figure: OH budget analysis reveals
with stabilization location in diesel flame simulations
of a stoichiometric and premixed
gradual DME flame
flame to ignition was conducte
transition
atmospheres and (Krisman et al. 2015)
a temperature of 900 K. These conditions w
 Lifted laminar flames
Upstream branches
• Upstream branches due
to both first and second
stage of auto ignition.
• Location of upstream Figure: Zero-D ignition delay times
branches correlated to 0D [Krisman et al. 2015]
ignition delay times.
• LTC radical species clearly
mark ‘cool-flame’ branch.
• Formaldehyde upstream
of stabilization point does
not necessarily indicate
LTC.

Figure: Chemical structure of polybrachial flames

(Krisman et al. 2015)
 Mixing layer ignition - overview
• Aim – observe ignition transient at Krisman et al. PROCI 36 (2016)

same conditions where polybrachial

flames observed.
• Configuration – 2D, no-shear,
isotropic pseudo-turbulent mixing
layer. Da=0.4.
• Thermochemical state – identical to
Figure: Initial and boundary conditions
900 K case from lifted laminar study.

Figure: Snapshots of HRR multi-stage, multi-mode ignition process

Tetrabrachial edge flames
 Mixing layer ignition - low temperature
chemistry
• LTC initiates similar
to 0D case.
• Rapidly develops
diffusive ’cool
flame’ nature. Figure: Cool-flame propagates in
composition space [Krisman et al. 2016]
• Propagates into
rich mixtures and
advances ignition.
• HTC ignition very
different to 0D
case.
Figure: Cool-flame advances
ignition, shifts to richer mixtures
[Krisman et al. 2016]
Figure: Autoignition to flame transition
for LTC [Krisman et al. 2016]
DNS of a Turbulent Autoigniting n-Dodecane
temporal jet at 25 Bar
Giulio Borghesi1, Alex Krisman1, Tianfeng Lu2 and Jackie Chen1
1Combustion Research Facility, Sandia National Laboratories

2University of Connecticut

Ketohydroperoxide mass fraction

Borghesi et al. Combustion and Flame, 2018

 Background and Objective

• Low-temperature combustion
(LTC) aims at increasing fuel
efficiency and reducing emissions
• Under LTC conditions, combustion
occurs in a mixed mode and in time
multiple ignition stages
• Ignition is now very sensitive to
the fuel chemistry, especially to
the low temperature reactions
branch

Question: How does transport and low-temperature chemistry affect

ignition in low-temperature diesel combustion?
 Low Temperature Diesel Combustion
Experiments – Engine Combustion Network

Skeen et al. PROCI 35 (2015) Dahms et al. PROCI 36 (2017)

 DNS Configuration and Physical Parameters
• Pressure: 25 bar
• Air stream: 15% XO2+85% XN2, T=960 K
• Fuel stream: n-dodecane at 5=0.3, T=450 K
• Kinetics: 35-species non-stiff reduced (Lu)
• Fuel jet velocity: 21 m/s, Rej = 7000, Ret ~ 950
• Code and cost: S3D Legion, 60M CPUh
• Setup:
– 3 billion grids
– 3 microns spatial grid resolution
– Dimensions: 3.6 mm x 14.0 mm x 3.0 mm
– 1 ms of physical time with 4 ns timesteps
to observe ignition and propagation of
burning flames throughout the domain Figure: H2O2 mass fraction at
t=0.17 ms after start of
– BCs: X and Z periodic, Y NSCBC outflows reactions
Homogeneous Multi-Stage Autoignition

5st

2nd Stage, High-T

1st Stage, Low-T 3-4X

Homogeneous ignition delay

time

Temporal evolution of
selected reactive scalars
 Dynamics of 2-Stage Ndodecane Ignition in a
Jet at Diesel Conditions

Rendering by Chris Ye, Min Shih, Franz Sauer, and Kwan-Liu Ma

Ketohydroperoxide and T,K (>1150K) H2O2 mass fraction

 Conditional statistics reveal ignition dynamics

Temperature

H2O2

Ketohydro-
peroxide

time
 Conditional mean temperature, KET, and H2O2 reveal
cool flame propagation and spontaneous ignition
assisted by turbulent diffusion

Low scalar
Low N
Dissipation time

High scalar
Dissipation
High N

Temperature KET H2O2

For low scalar dissipation, cool flames propagate as a wave towards rich conditions
while for high dissipation, turbulent diffusion of heat and mass leads to
spontaneous ignition where low-T ignition is impeded whereas high-T ignition is
accelerated
 Propagation mechanism for low-T reaction front
Flames +
Flames spontaneous Flames
ignition

i mee
t

Figure: Fraction of reactive low-T fronts propagating as a flame

Sketch of reaction / diffusion balance along
normal for KET for flame and ignition kernel
<7= (<7? @89: )
6789: = < 5 is a cool flame
<B= (<7? CDEE89: )

Low-T fronts propagate through diffusively supported flame

 Effect of Scalar Dissipation Rate on Low
Temperature Ignition
temp KET H2O2 CDF of F

t
i
m
Inc F e

time
 Turbulent versus homogeneous ignition

0.5
0.01

High T

tio n&
o p aga
e p r io n
l f la diffus
m
Coo ulent
b
Low T Tur

Low-T and high-T ignition in jet can be faster and than in a PSR !

Borghesi et al. 2018, Borghesi and Mastorakos, 2015, and Krisman et al. 2017
 Conclusions
• Low-temperature reactions create the conditions for high-temperature
ignition to occur faster than under homogeneous conditions;
• Low-temperature front appears to propagate through a diffusively
supported cool flame;
• High scalar dissipation appears to delay low-temperature ignition;
however, it leads to faster ignition at very rich mixture conditions;
• High-T ignition starts at conditions richer-than-homogeneous conditions
(x=0.16 compared to x=0.12). Edge flames are seen to form around xst.
High-T flame ignites mainly by propagation of rich premixed flames
following hot ignition to xst.
DOE Exascale Computing Project (ECP) will
achieve capable exascale machines in 2021-2023

Application Software Hardware Exascale

Development Technology Technology Systems
Science and Scalable and Hardware Integrated
mission productive technology exascale
applications software stack elements supercomputers
applica
AppliCo-Design
Correctness Visualization DataAnalysis

ti Programmingmodels,
development environmen t, Math libraries and
Tools
Frameworks

Workflows
Resilience

andruntimes

System Software, resource

management threading, Data
scheduling, monitoring, and Memory management
andBurst I/O and file
control buffer system
Node OS,runtimes

Hardwareinterface

ECP’s work encompasses applications, system software, hardware

technologies and architectures, and workforce development

From Paul Messina’s ASCAC talk April 19, 2017

 ECP application: transforming combustion science and
technology through exascale simulation (Pele)
PeleC and PeleLM: Block-structured
adaptive mesh refinement, multi-physics:
spray, soot, and radiation, real gas,
complex geometry

Effects of reactivity
stratification at:
• high pressure
• high turbulence
• fuel blends
on:
• ignition delay
• combustion
Automated Mechanism Generation
rates
• emissions Fuel Conditions

Rate Rules
Mechanism Generator (RMG)
Specific Rates
Uncertainty
Mechanism

1st Iteration
No Yes

S3D: multi-block compressible reacting Chemical Sim.

Theoretical Rate
Constants (EStokTP)
Method

Reaction(s)

DNS multi-physics validation: spray, Observables

Dakota
Sensitivities

soot, radiation Chem. Sim.

UQ range
Acceptable
yes no

Mechanism Reduction Engine Sim.

DRGASA Acceptable
yes no

CFD Simulations
of RCCI
DONE
Direct Numerical Simulation of Multi-Injection
Mixing and Combustion at Compression Ignition
Engine Conditions
M. Rietha, M. Dayb, T. Luc, M Kweon d, J. Temme d,
and J.H. Chena
aSandia National Laboratories
bLawrence Berkeley National Laboratory
cUniversity of Connecticut
dUS Army

xxxxxxxx
 Objectives
• Multi-injection is a strategy used in DI-CI engines to improve emissions
and noise while max. fuel economy

Injection
1 s t inj. 2 n d inj.

rate
Pilot Dw ell M a i n

time

Sandia Spray A experim ent (www.crf.gov)

Skeen e t al., JSTOR, 2 0 1 5

• Main injection sees very different conditions compared to pilot injection

• How do mixing and different thermo-chemical conditions affect ignition
of the main injection? (e.g. what determines whether cool flame or
high-temperature products are produced by pilot)
xxxxxxxx
Two Cases

Ground transportation
High pressure (60atm), moderate temperature
(900K) Exhaust gas recirculation
Pilot 0.5 ms, dwell 0.5 ms, main 0.5 ms
Improve mixture formation, reduce
emissions

High-altitude operation (UAV)

Moderate pressure (10atm), low temperature
(750K) No exhaust gas recirculation
Pilot 0.208 ms, dwell 0.992 ms, main 1.138 ms
Reduce signature (IR, smoke), improve ignition
reliability, improve operation with various fuels

Conditions lead to significantly different ignition processes

Images: www.challenger.com, http://insideunmannedsystems.com

xxxxxxxx
Physical Parameters for Multi-Injection Pulsed Jet DNS

Ground conditions
Purely gaseous n-dodecane/air jet @T=470K, Z=0.45
Ambient conditions: 900K, 15% O2, 85% N2, 60atm
Conditions adapted from Dalakoti et al. (ICDERS, 2017.)
→ downscaled ECN Spray A, keeping Da constant (Re≈19,000, Dajet≈0.02)
Multi-injection: 0.5 ms pilot, 0.5 ms dwell, 0.5 ms main

UAV
Purely gaseous n-dodecane/air jet @T=450K, Z=0.37
Ambient conditions: 750K, 21% O2, 79% N2, 10atm
Multi-injection: Pilot 0.208 ms, dwell 0.992 ms, main 1.138 ms

xxxxxxxx
Ground and UAV operation –multi-stage
homogeneous ignition
Exemplary ignition sequence
Z = 0.0417 ⇥ 1011
2000 0.75 0.015
OC 12 H23 OOH
High-temperature H2 O2 ·3

HRR [J/m 3 /s]

1500
Low-temperature ignition 0.50 0.010
OH ·5
T [K]

Y [-]
CH 2 O
ignition
0.25 0.005
1000
0.00 0.000

time time
T [K] T [K]

0.14 . UAV
0.14 Ground 2200
2500

0.12 0.12 2250

2000
Mixture fraction Z [-]

Mixture fraction Z [-]

0.10 2000
0.10 1800
1750
1600 0.08
0.08
1400 Stoich. mix frac 1500
0.06 0.06
1250
Stoich. mix frac. 1200
0.04 0.04 1000

High T
1000

Low T
High T
Low T

0.02 800 0.02 750

0.0000 0.0002 0.0004 0.0006 0.0008 0.0010 0.000 0.005 0.010 0.015 0.020
time t [s] time t [s]

xxxxxxxx
 PeleLM Code & Numerical Setup
• PeleLM – low-Mach adaptive mesh refinement code based on AMReX
• Spectral deferred correction scheme for fluid dynamics-chemistry coupling
• Multi-physics: soot, radiation, sprays, embedded boundary for geometry
• Code is open-source at https://amrex-combustion.github.io/
• Resolution ~1.25 micron required for rich premixed cool flames,
currently 5 micron for full multi-injection run
• Size of simulation: ~1B cells (O(100)B cells full run without AMR)
• 35 species reduced n-dodecane mechanism (Yao et al., 2017;
Borghesi et al., 2018)

Emmett et al., arxiv, 2018.

E. Motheau AMReX gallery.

37
Multi-injection ignition sequence (ground)
Low High-
*Movie does not show full domain temperature temperature
species species

Mixture fractions track fluid from individual injections (two additional

transport equations)
38
 Homogeneous n-dodecane ignition (ground)

Temperature H 2O 2 OH

Oxidizer now consists of an equilibrated lean mixture with Z=0.01

Temperature H 2O 2 OH

Low-T High-T

Stoich. Mix frac

39

• N-dodecane exhibits two-stage ignition with minimum ignition delay at ‘preferred mixture fraction’ for
each stage1
• H2O2 is a marker for low temperature combustion; OH is a marker for high temperature combustion
• Low temperature ignition shifts toward richer mixtures (by almost a factor of 3, has a much shorter
ignition delay while hot ignition remains the same) xxxxxxxx
 Conditional statistics (ground)
vs. Zpilot vs. Zmain

0.55⌧inj 0.91⌧inj 2.09⌧inj 2.56⌧inj 2.93⌧inj

2000 2000 2000 2000 2000

End 1st Mid 2nd End 2nd

1500 1500 1500 1500 1500
injection injection
T [K]

T [K]

T [K]

T [K]

T [K]
Mid 1st injection
Start 2nd
Temperature

1000 injection 1000 1000 1000 1000

injection
500 500 500 500 500
0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4
Zpilot Zpilot Zpilot Zmain Zmain
0.004 0.004 0.004 0.004 0.004

0.003 0.003 0.003 0.003 0.003

Y H2 O 2

Y H2 O 2

Y H2 O 2

Y H2 O 2

Y H2 O 2
0.002 0.002 0.002 0.002 0.002
H2O2

0.001 0.001 0.001 0.001 0.001

0.000 0.000 0.000 0.000 0.000

0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4
Zpilot Zpilot Zpilot Zmain Zmain
0.0008 0.0008 0.0008 0.0008 0.0008

0.0006 0.0006 0.0006 0.0006 0.0006

YOH

YOH

YOH

YOH

YOH
0.0004 0.0004 0.0004 0.0004 0.0004
OH

0.0002 0.0002 0.0002 0.0002 0.0002

0.0000 0.0000 0.0000 0.0000 0.0000

0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4 0.0 0.2 0.4
Zpilot Zpilot Zpilot Zmain Zmain

40
 P( pilot )  0.25
0.000008 0.25 < P ( pilot )  0.50
0.0006 0.50 < P ( pilot )  0.75
0.000006
Effect of mixing on ignition 0.75 < P ( pilot )  1.00

0.55⌧inj 0.91⌧inj 0.55⌧inj 0.91⌧inj

YOH

YOH
1000
inc G
0.0004 0.0008
P( pilot )  0.25
0.000004 Mid 1st 2000 0.000008 2
900 injection0.55⌧inj pilot =0.91⌧
2D(@Zinj pilot /@x j)
0.0006
0.25 < P (
0.50 < P (
pilot )

pilot )
0.00080.000006

Bins based on pilot scalar

0.75 < P ( pilot )
800 1500
0.0002 P( pilot )  0.25
T [K]

T [K]
0.000002

YOH

YOH
0.000008
700 End 1st 0.000004 0.25 < P ( pilot )  0.50
0.0004

dissipation rate
1000 injection
600 0.00060.000002 0.50 < P ( pilot )  0.75
0.0002
0.000000
0.000006
500 0.0000 0.75 < P ( pilot )  1.00
500 0.000000 0.0000
0.0
0.0 0.2 0.2
0.4 0.4
0.0 0.2 0.4 0.0 0.0 0.2
0.2 0.4 0.4 0.0 0.2
YOH

YOH
Zpilot
Zpilot Zpilot 0.0004 Zpilot
Zpilot Zpilot
0.000004 2.56⌧inj 2.93⌧inj 2.56⌧inj 2.93⌧inj
cross = 2D(@Zpilot /@xj · @Z0.0008 main /@xj )
2000 2.56⌧inj 2.93⌧inj P( )  0.50 cross
G-
0.000002 1750
2000 0.0008
0.0002 0.0006 0.0006
0.50 < P (
P(
)
)  0.50 +
cross

Bins based on cross scalar

G+ P( cross )  0.50 cross
0.50 < P ( +
cross )
1750
1500 0.50 < P (  1.00
0.0004 cross )
T [K]

T [K]

0.0006
YOH

YOH
Mid 2nd End 2nd 0.0004

dissipation rate
1500 +
0.000000
1250 injection 0.0006
0.0000
injection P( cross )  0.50
0.0002 +
1000 0.0 0.2 1250
0.4 0.0 0.50
0.2< P ( cross ) 0.4
1.00
0.0002

0.0004
750 Zpilot 1000 0.0000
Zpilot Zmain0.0000
YOH

YOH

0.0 0.2 0.4 0.0 0.2 0.0004

0.4 0.0
Zpilot
0.2 0.4 0.0 0.2
Zmain 2.56⌧ inj Zmain 2.93⌧
Zmain inj Zmain
0.0008 Z
P( cross )  0.50
0.0002 0.0002 0.50x < P (  1.00 41
cross )
 Effect of mixing on ignition

1st injection 2nd injection

Correlation between Correlation between Only negative
progress variable progress variable values of cross
and pilot scalar and cross scalar scalar dissipation
dissipation rate dissipation rate rate taken into
account
0.0
0.1

0.0 0.2 Zmain

cross
Zpilot
pilot

0.1 0.4
⇢ Yp ,

⇢ Yp ,

Z
0.2
0.6
0.3 0.55⌧inj 2.56⌧inj x
0.91⌧inj 0.8 2.93⌧inj
0.4
0.0 0.2 0.4 0.0 0.2 0.4
Zpilot Zmain

1st injection consistent with results from Borghesi et al., Comb. Flame, 2018.

42
Conclusions (Ground)

• First- and second stage pilot ignition consistent with previous numerical
studies

• Accelerated ignition for main injection observed consistent with

experiments

• Strong mixing inhibits ignition of first injection, promotes ignition of

second injection

43
Ignition sequence UAV

t = 0.212 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

Pilot Main Temp (K) KET CH2O H2O2 OH

mixture mixture
fraction fraction 8/17
Ignition sequence UAV

t = 0.484 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

Pilot Main Temp (K) KET CH2O H2O2 OH

mixture mixture
fraction fraction 8/17
Ignition sequence UAV

t = 0.824 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 1.256 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 1.505 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 1.786 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 2.066 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 2.348 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 2.773 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 3.575 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 3.778 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 3.994 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 4.228 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 4.471 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 4.735 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 4.986 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 5.277 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 5.623 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 5.981 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
Ignition sequence UAV

t = 6.386 ms

Z pilot [-] Z m a i n [-] T [K] YO C 1 2 H 2 3 O O H [-] YC H 2 O YH 2 O 2 [-] Y OH [-]

0.030

0.025

0.020
z [m]

0.015

0.010

0.005

0.000
− 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004 − 0 . 0 0 4 0. 000 0. 004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.0 0.3 0.0 0.3 1000 2000 0.00 0. 05 0. 00 0.01 0.000 0.003 0.000 0.005

8/17
 For UAV conditions pilot accelerates ignition

t = 3.591 ms
Z pilot [-] Z m a i n [-] T [K] Y O C 1 2 H 2 3 O O H [-] YC H 2 O Y H 2 O 2 [-] Y O H [-]
0 .0 3 0

0 .0 2 5

0 .0 2 0

No pilot (t=3.6ms after

start of main injection)
z [m]

0 .0 1 5

0 .0 1 0

0 .0 0 5

0 .0 0 0
− 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

t = 4.735 ms
0 .0 0 0 .0 2 0 .0 0 .1 1000 2000 0 .0 0 0 .0 5 0 .0 0 0 .0 1 0 .0 0 0 0 .0 0 3 0 .0 0 0 0 .0 0 5
Z pilot [-] Z m a i n [-] T [K] Y O C 1 2 H 2 3 O O H [-] YC H 2 O Y H 2 O 2 [-] Y O H [-]
0 .0 3 0

0 .0 2 5
Pilot (t=3.6ms after
0 .0 2 0
start of main injection)
z [m]

0 .0 1 5

0 .0 1 0
→ clear acceleration of
0 .0 0 5
the ignition process
0 .0 0 0
− 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0 .0 0 .3 0 .0 0 .3 1000 2000 0 .0 0 0.05 0.00 0 .0 1 0 .0 0 0 0.003 0.000 0 .0 0 5

Zmain Zpilot T (K) KET CH2O H2O2 OH

9/17
For UAV conditions (10 atm, 750K ndodecane)
pilot accelerates ignition
t = 4.601 ms
Z pilot [-] Z m a i n [-] T [K] Y O C 1 2 H 2 3 O O H [-] YC H 2 O Y H 2 O 2 [-] Y O H [-]
0 .0 3 0

0 .0 2 5

0 .0 2 0

No pilot (t=4.6ms after

start of main injection)
z [m]

0 .0 1 5

0 .0 1 0

0 .0 0 5

0 .0 0 0
− 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

t = 5.802 ms
0 .0 0 0 .0 2 0 .0 0 .1 1000 2000 0 .0 0 0 .0 5 0 .0 0 0 .0 1 0 .0 0 0 0 .0 0 3 0 .0 0 0 0 .0 0 5
Z pilot [-] Z m a i n [-] T [K] Y O C 1 2 H 2 3 O O H [-] YC H 2 O Y H 2 O 2 [-] Y O H [-]
0 .0 3 0

0 .0 2 5
Pilot (t=4.6ms after
0 .0 2 0 start of main injection)
z [m]

0 .0 1 5

0 .0 1 0
→ clear acceleration of
0 .0 0 5
the ignition process
0 .0 0 0
− 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004 − 0 . 0 0 4 0.000 0.004
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0 .0 0 .3 0 .0 0 .3 1000 2000 0 .0 0 0.05 0.00 0 .0 1 0 .0 0 0 0.003 0.000 0 .0 0 5

Zmain Zpilot T (K) KET CH2O H2O2 OH

Joint PDFs of mixture fractions show pilot and main
fluid is well mixed
0.030 0.030 105
t=3.58 ms t=4.17 ms
0.025 ρ corr =0.72 104 0.025 ρ corr =0.78 104

0.020
cov=6.94e-06 0.020
cov=6.17e-06
103 103
Zpilot

Zpilot
jpdf

jpdf
0.015 0.015
102 102
0.010 0.010

0.005 101 101

0.005

0.000 0.000
0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
Z main Z main
0.030 105 0.030 105
t=4.74 ms t=5.80 ms
0.025 ρ corr =0.83 104
0.025 ρ corr =0.86 104

0.020
cov=6.17e-06 0.020
cov=5.44e-06
103 103
Zpilot

Zpilot
jpdf

jpdf
0.015 0.015
102 102
0.010 0.010

101 101
0.005 0.005

0.000 0.000
0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
Z main Z main

Pilot and main fluid well-mixed at time of low-/high-

temperature ignition (bottom two plots) xxxxxxxx
Mixing behavior – scalar dissipation rates
Scalar dissipation rates at t=2.3ms and t=3.6ms (before low-T ignition)
χ Zpilot χ Zmain. si gn(χ cross ) × log(χ cross ) . χ cross

0.020
• Early time: Mostly negative
0.015
cross scalar dissipation rates
z [m]

0.010

• Later time: Mostly positive

0.005

cross scalar dissipation rates

0.000
−0. 005 0.000 0.005 −0. 005 0.000 0.005 −0. 005 0.000 0.005 −0. 005 0.000 0.005
x [m] x [m] x [m] x [m]

1 0 − 4 1 0 − 2 10 0 10 2 1 0 − 4 1 0 − 2 10 0 10 2 2 1 0 1 200 100 0 Z p ilot Z m ain

χ Zp i l o t χ Zm a i n sign(χcross) × lo g 1 0 [ m a x ( 1 , | χ c r o s s |)] χ cross

0.020 Hcross < 0 z

0.015
x
z [m]

0.010

z
0.005
Hcross > 0
x
0.000
−0. 005 0.000 0.005 −0. 005 0.000 0.005 −0. 005 0.000 0.005 −0. 005 0.000 0.005
x [m] x [m] x [m] x [m]

1 0 − 4 1 0 − 3 1 0 − 2 1 0 − 1 10 0 10 − 1 4 0 − 1 3 0 − 1 2 0 − 1 10 0 10 1 10 2 −1 0 1 − 1 0 0 10

xxxxxxxx
Mixing behavior – scalar dissipation rate (SDR) PDFs

100 100 100 2.35 ms

3.58 ms
10− 1 10− 1 10− 1
--- 5.80 ms
log-normal
10− 2 10− 2 10− 2
pdf [-]

pdf [-]

pdf [-]
10− 3 10− 3 10− 3

10− 4 10− 4 10− 4

10− 5 10− 5 10− 5

10− 6 10− 6 10− 6

−5 0 5 −5 0 5 − 100 0 100
log(χ pilot ) log(χ main ) χ cross

From left to right: Pilot SDR, main SDR, cross SDR

Pilot and main SDR nearly log-normal (Hawkes et al., PCI,2007)
Cross SDR has stretched exponential tails (similar: Chaudhuri et al., C&F, 2017)

xxxxxxxx
 Conditional averages of ketohydroperoxide, OC12H23OOH

0.030 0.010 0.030 0.010 0.030 0.010

t=2.35 ms t=2.77 ms t=3.58 ms

0.025 0.025 0.025
0.008 0.008 0.008

YOC12H23OOH

YOC12H23OOH

YOC12H23OOH
0.020 0.020 0.020
0.006 0.006 0.006
Zpilot

Zpilot

Zpilot
0.015 0.015 0.015
0.004 0.004 0.004
0.010 0.010 0.010

0.002 0.002 0.002

0.005 0.005 0.005

0.000 0.000 0.000 0.000 0.000 0.000

0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10

0.030 Zmain 0.030 0.030 Zmain 0.030 0.030 Zmain 0.030

t=4.17 ms t=4.74 ms t=5.80 ms

0.025 0.025 0.025 0.025 0.025 0.025
YOC12H23OOH

YOC12H23OOH

YOC12H23OOH
0.020 0.020 0.020 0.020 0.020 0.020
Zpilot

Zpilot

Zpilot
0.015 0.015 0.015 0.015 0.015 0.015

0.010 0.010 0.010 0.010 0.010 0.010

0.005 0.005 0.005 0.005 0.005 0.005

0.000 0.000 0.000 0.000 0.000 0.000

0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10 0.00 0.02 0.04 0.06 0.08 0.10
Z main Z main Z main

Formation starts in regions with pilot fluid, moves into richer conditions
Change of maximum range between top and bottom row (bottomxxxxxxxx
factor 3 higher) Bottom
row: before main jet low-T ign., start of low-T ign., start of high-T ign.
Conditional averages of OC12H23OOH

0.006
multi
0.005 single

0.004
YOC12 H23 OOH

0.003

0.002

0.001

0.000
0.00 0.02 0.04 0.06 0.08 0.10
Zsum

OC12H23OOH compared at the same time, 3.59 s with respect to start of

simulations. Pilot is able to provide OC12H23OOH to rich regions.

xxxxxxxx
 Chemical Explosive Mode Analysis (code by T. Lu)

• Eigen-analysis of the Jacobian of the local chemical source term

diffusion
D!(y) Dy @!
= J! = J! (! + s), J! =
Dt Dt y
reaction
= be · J! · ae
e
• CEM is associated with λe that has a positive real part
• Projection onto CEM to distinguish reaction or diffusion dominated regions
D!(y) Dy D ! Dbe
be = b e · J! = e be · (! + s) = e e+ e s+ · !(y)
Dt Dt Dt Dt
• Combustion mode indicator 1  ↵  1 : auto-ignition (AI)
↵= s/ w ↵< 1 : di↵usion dominates (extinction)
↵ > 1 : flame propagation
71
Lu et al., JFM, 2010; Xu et al., PROCI, 2019.
 Chemical Explosive Mode Analysis – 4ms (2.8ms
main) t = 3.994 ms

Zsum T [K] sign( exp ) ⇥ log10 [max(1, exp )] alpha YH2 O2 [-] YOC12 H23 OOH [-] Fuel consumpt.
0.030

Fuel
Zsum Temp. Expl. Combust H2O2 ket
consumpt
0.025
mode .
..
mode
0.020
z [m]

0.015

0.010

0.005

0.000
0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.00 0.05 1000 2000 5 0 5 1 1 0.00 0.01 0.00 0.02

8
10

Integrated fuel 10 9
Preliminary
consumption rate 10 requires further
10
conditioned on Zsum D
research for two-stage
and combustion mode 10 11 AI
FP 72 ignition
0.00 0.02 0.04 0.06 0.08
 Chemical Explosive Mode Analysis – 4.6ms
(3.4ms main) t = 4.602 ms

Zsum T [K] sign( exp ) ⇥ log10 [max(1, exp )] alpha YH2 O2 [-] YOC12 H23 OOH [-] Fuel consumpt.
Fuel
0.030

Zsum Temp. Expl. Combust H2O2 ket

consumpt
mode .
..
0.025

mode
0.020
z [m]

0.015

0.010

0.005

0.000
0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005 0.005 0.000 0.005
x [m] x [m] x [m] x [m] x [m] x [m] x [m]

0.00 0.05 1000 2000 5 0 5 1 1 0.00 0.01 0.00 0.02 100 104

8
10
Integrated fuel Preliminary
consumption rate 10 10

requires further
conditioned on Zsum
research for two-stage
D
12
10
and combustion mode AI
FP 73 ignition
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07
 Conclusions (UAV case)

• Different conditions lead to very different ignition behavior

• Pilot injection does not ignite prior to mixing with main injection for UAV
• Clear differences between cases with and without pilot

• Pilot mixes well with fluid from main injection

• Pilot supplies low-temperature species that enhance ignition of the main injection
• CEMA analysis confirms predominant role of autoignition over deflagration

zzzzzzzzz
 Direct Numerical Simulation of flame stabilization
assisted by auto-ignition at reheat conditions
Konduri Adityaa, Andrea Gruberb, Mirko Bothienc and Jacqueline H. Chena

aCombustion Research Facility, Sandia National Laboratories, Livermore, CA, USA

bSINTEF Energy Research, Trondheim, Norway
cAnsaldo Energia, Baden, Switzerland

Acknowledgements:
Basic Energy Sciences, Office of Science, DOE
OLCF, NERSC, Norwegian CCS Research Centre (NCCS)
Aditya et al. PCI 2018
 Staged gas turbine combustion

• Originally developed by ABB for high efficiency, load flexibility

and low emissions
• Recently improved and simplified (reduced cost) for the H-class
GT36
• First (premix) combustion stage based on flame propagation
• Second (sequential) combustion stage based on auto-ignition
 Reheat burner
DNS of idealized reheat burner configuration from Ansaldo Energia
Operating conditions:
• Inlet temperature: ∼ 1100 K
• Pressure: ∼ 20 atm

Scaled conditions:
• Mean inlet temperature: 1100 K
• Pressure: 1 atm
• Fuel: hydrogen

Objective:
• Understand the flame stabilization
• Identify the modes of combustion
• Quantify the role of autoignition
 Simulation details

• Chemical mechanism: 9 species hydrogen-air (Li et al., 2004)

• Inflow composition: premixed H2 + O2 + N2 + H2O (φ = 0.35)
• Ubulk = 200m/s, uʹ = 20m/s, Tinlet = 1100K, Twall = 750K
• Inflow profile: feed from DNS of a fully developed channel flow,
Reb ∼13, 000, Reτ= 378 (viscous length scale)

Feed data
sampling plane
 Enstrophy conditioned on temperature, and heat release rate (red)

Iso-surfaces of
vorticity magnitude
colored by
temperature

Two combustion configurations are observed:

• Design state: mainly auto-ignition in the combustion chamber
• Intermittent auto-ignition state: ignition in mixing section
Design combustion state

Combustion modes:
• Autoignition along center-
line
• Flame propagation near
corners
• HO2: indicative of chain
branching
 Combustion mode: OH budget analysis

• Auto-ignition: balance between

advection and reaction
• Flame propagation: balance between
diffusion and reaction
Intermittent auto-ignition state

• Early auto-ignition in the

mixing section
• Ignition kernel advects
downstream
• Occurs intermittently
 Chemical Explosive Mode Analysis
• α = φs /φω : ratio of the projected non-chemical source term and
the projected chemical source term (C. Xu et al., PROCI 2018 )

Three mode are identified:

• Assisted-ignition (α > 1): diffusion significantly promotes reaction
• Auto-ignition (−1 < α < 1): chemistry plays a dominant role
• Extinction zone (α < −1): diffusion dominates chemistry and
suppresses ignition
 Intermittent auto-ignition state

Auto-ignition Contours of
heat release

Compression wave

• Local rise in pressure due

to intermittent
constructive interference
pattern
• Increases local
temperature by 20-30 K
• High reactivity of hydrogen
Contours of instantaneous dilatation
• Decrease in ignition delay
time (30%)
 Conclusions
• Performed DNS of a reheat burner at scaled conditions
• Two states of hydrogen/air combustion have been observed:
• design state: flame propagation and auto-ignition in the combustor
• intermittent auto-ignition in mixing section
• Premature auto-ignition arises due to pressure (and following
temperature) rise in mixing section
• Quantified the contribution of different modes towards heat release
using chemical explosive mode analysis (CEMA)
• Future work:
• characterize the unstable flame behavior and the conditions
leading to it
• find the inlet conditions for statistically stationary reheat flame
• perform 2D and 3D simulations with varying fuel composition and
its stratification