Presentation at the 9th US National Combustion Meeting in May 2015.
"Identification, Correction, and Comparison of Detailed Kinetic Models"
Extended abstract available from http://www.northeastern.edu/comocheng/2015/05/uscombustionmeeting/
This material is based upon work supported by the National Science Foundation under Grant Number 1403171. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
Chemical Tests; flame test, positive and negative ions test Edexcel Internati...
2015 US Combustion Meeting - West - Identification, Correction, and Comparison of Detailed Kinetic Models
1. .edu/comocheng
Victor R. Lambert & Richard H. West
9th US National Combustion Meeting
20 June 2015
1
r.west@neu.edu richardhwest rwest
Identification, Correction, and Comparison
of Detailed Kinetic Models
Grant No. 1403171
2. Identification, Correction, and Comparison
of Detailed Kinetic Models
!
!"
!
!" #
!
!"
!
!
!"
!
!"#
$
!"
!
!" !
" #$
!"#
!" !"
!"#
"#! !"#
!
!"
!
!"
!
" #$
#$!
"
!
" #$
!
!" !""
#
!""
#
!"#
!"#$"
!
" #$
#$!
!
!"
!"
!
!
!
!"
!
!"#
!
!
!
!
!"""#
!""#
$
!""#
$
!"
!
" #$
!
!
" #$
%!
!
" #$
%
!
" #$
%
!
" #$
%!
Elizabeth
Becky
Eliza
Liz
Beth
with many names
for the same thing...
...it is difficult to
compare models...
...researchers
give molecules
nicknames.
...and easy to
make mistakes!
...and create
a unified
database...
with this we can:..
To publish in
"CHEMKIN"
format...
Our tool to
identify
species...
...allows us to
analyze models.
...find common
parameters,..
...detect
mistakes,..
...identify
controversial
rates,..
...of all
kinetic
models!
4. Transforming data into knowledge—ProcessInformatics for combustion chemistry
Michael Frenklach *Department of Mechanical Engineering, University of California, and Environmental Energy Technologies
Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
Abstract
The present frontier of combustion chemistry is the development of predictive reaction models,
namely, chemical kinetics models capable of accurate numerical predictions with quantifiable uncertain-
ties. While the usual factors like deficient knowledge of reaction pathways and insufficient accuracy of
individual measurements and/or theoretical calculations impede progress, the key obstacle is the incon-
sistency of accumulating data and proliferating reaction mechanisms. Process Informatics introduces a
new paradigm. It relies on three major components: proper organization of scientific data, availability
of scientific tools for analysis and processing of these data, and engagement of the entire scientific com-
munity in the data collection and analysis. The proper infrastructure will enable a new form of scien-
tific method by considering the entire content of information available, assessing and assuring mutual
scientific consistency of the data, rigorously assessing data uncertainty, identifying problems with the
available data, evaluating model predictability, suggesting new experimental and theoretical work with
the highest possible impact, reaching community consensus, and merging the assembled data into new
knowledge.
Ó 2006 The Combustion Institute. Published by Elsevier Inc. All rigKeywo
Proceedings of the Combustion Institute 31 (2007) 125–140
www.elsevier.com/locate/proci
Proceedings
of the
Combustion
Institute
5. “the key obstacle is the inconsistency of accumulating
data and proliferating reaction mechanisms.
Process Informatics introduces a new paradigm. It relies
on three major components:
• proper organization of scientific data,
• availability of scientific tools for analysis and
processing of these data,
• and engagement of the entire scientific community
in the data collection and analysis.”
M. Frenklach, Proc. Combust. Inst. 31 (2006)
6. Model Complexity is still increasing
2-methylalkanes/Sarathy
n-Heptane/Mehl
n-Octane/Ji
Heptamethylnonane/Westbrook
Isobutene/Dias
Ethanol/Leplat
1-butanol/Grana
Acetaldehyde/Leplat
Ethylbenzene/Therrien
Methylcrotonate/Bennadji
Methyloleate/Naik
Furan/Yasunaga
Methylhexanoate/Westbrook
Methydecanoate/Herbinet
H2
/KlippensteinH2
/ConnaireH2
/Marinov
GRI-Mech
Methane/Leeds
Ethane
C2
/Karbach
Ethylene/Egolfopoulos
C3
/Appel
USC-MechMarinov
Propane/Qin
n-Butane
Neopentane/Wang
n-Heptane/Curran
n-Heptane/Babushok
10
100
1000
10000
1995 2000 2005 2010
Numberofspeciesinmodel
Year of publication
7. Model Complexity is still increasing
2-methylalkanes/Sarathy
n-Heptane/Mehl
n-Octane/Ji
Heptamethylnonane/Westbrook
Isobutene/Dias
Ethanol/Leplat
1-butanol/Grana
Acetaldehyde/Leplat
Ethylbenzene/Therrien
Methylcrotonate/Bennadji
Methyloleate/Naik
Furan/Yasunaga
Methylhexanoate/Westbrook
Methydecanoate/Herbinet
H2
/KlippensteinH2
/ConnaireH2
/Marinov
GRI-Mech
Methane/Leeds
Ethane
C2
/Karbach
Ethylene/Egolfopoulos
C3
/Appel
USC-MechMarinov
Propane/Qin
n-Butane
Neopentane/Wang
n-Heptane/Curran
n-Heptane/Babushok
10
100
1000
10000
1995 2000 2005 2010
Numberofspeciesinmodel
Year of publication
11. APPENDIX D
THERMO DATA (FORMAT AND LISTING)
The order and format of the input data cards in this appendix are given in the follow-
ing table"
Card
order
(a)
(Final
card)
Contents
Format Card
column
THERMO
Temperature ranges for 2 sets of coefficients:
lowest T. common T, and highest T
Species name
Date
Atomic symbols and formula
Phase of species (S, L, or G for solid, liquid,
or gas, respectively)
Temperature range
Integer t
Coefficients ai(i = 1 to 5) in equations (90) to (92)
(for upper temperature interval)
Integer 2
Coefficients in equations (90) to (92) (a6, a T for
upper temperature interval, and a 1, a2, and a 3
for lower)
Integer 3
Coefficients in equations (90) to (92) (a4, a5, a6,
a, i for lower temperature interval)
Integer 4
Repeat cards numbered 1 to 4 in cc 80 for each
species
END (Indicates end of thermodynamic data)
3A4
3F10.3
3A4
2A3
4(A2, F3. O)
A1
2F10.3
I15
5(E15.8)
15
4(E15.8)
I20
3A4
1 to6
1 to 30
1 to 12
19 to 24
25 to 44
45
46 to 65
80
1 to75
80
1 to75
80
1 to 60
8O
ito3
aGaseous species and condensed species with only one condensed phase can be in
any order. However. the sets for two or more condensed phases of the same
species must be adjacent. If there are more than two condensed phases of a
species, their sets must be either in increasing or decreasing order according
to their temperature intervals.
168
ORIGINAL PAGE IS
OF. POOR QUALITY
nq
12. NASA (Chemkin) format is very dense –
not much room for species identifiers
Chemical Formula Parameters for H(T), S(T)Species Name
13. Space constraints led to “creative”
naming schemes
MVOX
IIC4H7Q2-T
C3H5-A C3H5-SC6H101OOH5-4
TC4H8O2H-I
C4H8O1-3
C3KET21
CH3COCH2O2H
14. Sometimes the same species appears twice
in a model, with different names
• C3KET21 is generated from alkyl peroxy radical
isomerization pathway
• CH3COCH2O2H is generated from low-temperature
oxidation of acetone
MVOX
IIC4H7Q2-T
C3H5-A C3H5-SC6H101OOH5-4
TC4H8O2H-I
C4H8O1-3
C3KET21
CH3COCH2O2H
16. Reaction Mechanism Generator (RMG-Py)
has many of the required features
• Represent molecules and recognize duplicates
• Propose reactions from templates and rules
• Estimate parameters quickly
• Expand model based on known species
• Modular and Extensible!
⇌RMG
17. Algorithm is like solving a Sudoku puzzle.
• Identify species with only one possible structure
• CO2
• H2O
• C3H8
• Then species with "borrowed" thermochemistry
• Then “boot-strap” based on how these react…
18. Identifying ‘sc3h5co’ from its reactions:
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
19. Identifying ‘sc3h5co’ from its reactions:
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
20. Identifying ‘sc3h5co’ from its reactions:
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
21. Identifying ‘sc3h5co’ from its reactions:
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
22. Identifying ‘sc3h5co’ from its reactions:
R1-R6
R7
The first thing the tool will do is to check previously imported models for matching thermochemistry
blocks. For example, it could tell the user that the LLNL model for n-Heptane [29] has a species with
the same parameters that has already been identified as but-2-enoyl ( ), and that LLNL
called it “SC3H5CHO”. This would probably be enough evidence for the user to confirm the match.
If the species cannot be found in a previously imported model, the reactions give additional clues.
The seven reactions containing sc3h5co in the methyl butanoate model are:
R1 sc3h5cho + o2 ⇌ sc3h5co + ho2 + ⇌ + sc3h5co
R2 sc3h5cho + oh ⇌ sc3h5co + h2o + ⇌ + sc3h5co
R3 sc3h5cho + o ⇌ sc3h5co + oh + ⇌ + sc3h5co
R4 sc3h5cho + ch3 ⇌ sc3h5co + ch4 + ⇌ + sc3h5co
R5 sc3h5cho + h ⇌ sc3h5co + h2 + ⇌ + sc3h5co
R6 sc3h5cho + ho2 ⇌ sc3h5co + h2o2 + ⇌ + sc3h5co
R7 sc3h5co ⇌ c3h5-s + co
sc3h5co ⇌ +
The first six are all hydrogen abstractions
from but-2-enal (assume for now that this
species has already been identified), which
has four types of hydrogen atom, implying
sc3h5co could be one of four possible radi-
cals. The seventh reaction is the decomposi-
tion into propenyl and carbon monoxide,
also limiting sc3h5co to four possible spe-
cies. The Venn diagram in Fig. 4 shows that
only one species satisfies all seven reactions:
but-2-enoyl. The tool will present the user
with this data, as well as a comparison of
Fig. 4. Venn diagram showing identification of
“sc3h5co” from reactions R1 through R7.
24. Human-Computer team can
identify species more quickly
• Our new tool uses RMG to generate reactions to
compare with the target model
• A human reviews the evidence and confirms matches.
Generate reactions,
check for equivalents,
propose matches
Review evidence,
confirm matches.
proposed
matches
proposed
matches
Human Computer
confirmed
matches
confirmed
matches
25.
26. • Model compiled over many years by many authors
• Recently replaced core chemistry, merged methyl
cyclohexane, replaced toluene, and still updating
other submechanisms…
• We found 51 unintended duplicates, now fixed.
• Identifying species allowed reactions to be
classified and correlated uncertainties estimated, for
Uncertainty Quantification (talk 114RK-0445)
Identified all 1.7k species in latest LLNL model
27. 60 papers in Reaction Kinetics division
33 mention or use Chemkin software
28 have supplementary material
17 are kinetic models
13 are usable
We analyzed all mechanisms in proceedings
of 34th Combustion Symposium
28. Of 13 models in the 34th Symposium,
77% are at least 77% identified
113
113
852
137
1686
380
1064
202
488
296
94
125
335
113
121
877
137
1924
1924
1350
202
662
355
277
125
392
0 500 1000 1500 2000
Veloo p.599
Sheen p.527
Darcy p.411
Liu p.401
Malewicki p.361
Malewicki p.353
Wang p.335
Husson p.325
Herbinet p.297
Dagaut p.289
Matsugi p.269
Labbe p.259
Somers p.225 Identified
Unidentified
C/H/O
Species
(as of last week).
29. Some species have many names
• Note that A1 means either benzene or phenyl
Species Names
propen-2-yl radical C3H5-T, ch3cch2, TC3H5
benzene A1, A, C6H6#, c6h6
phenyl radical A1J, A1, C6H5#, c6h5
1-butyl radical PC4H9, R20C4H9, NC4H9, NC4H9P
2-hexyl radical R72C6H13, hex2yl, C6H13-2
isoprene b13de2m, IC5H8
30. Thermochemistry:
Of 1039 Species found in 2 or more models…
😀 408 (39%) have identical thermo
😊 731 (70%) span < 5 kJ/mol
😟 28 (3%) span > 50 kJ/mol
Spread in ∆Hf(298K) in kJ/mol
#ofSpecies
31. Of 60 species also in Argonne/Ruscic’s
Active Thermochemical Tables…
😀 3 are always within the ATcT
uncertainty bounds
😊 23 (38%) are always within 1 kJ/mol
of ATcT uncertainty
😟 13 (22%) are sometimes more than
10 kJ/mol outside ATcT uncertainty
Worst error in ∆Hf(298K) beyond uncertainty, in kJ/mol
#ofSpecies
32. Of 60 species also in Argonne/Ruscic’s
Active Thermochemical Tables…
😊 52 (87%) are sometimes within 1 kJ/
mol of ATcT uncertainty
😧 4 are always more than 10 kJ/mol
outside ATcT uncertainty
Least error in ∆Hf(298K) beyond uncertainty, in kJ/mol
#ofSpecies
33. Kinetics:
Of 1303 reactions in 3 or more models…
😉 432 (33%) have identical rates
😊 721 (55%) agree within a factor of 2
😟 233 (18%) disagree by > 10x
😬 45 (3%) disagree by > 1000x
span in log10(k @ 1000K)
#ofReactions
(16 outliers > 1010)
34. Often, unannounced changes are made when
merging from one model to another.
⇄
⟶ + +
+
Examples of reactions slowed by 50 orders of magnitude without comment:
The present X
model was coupled
to the C5–C7 LLNL
n-Heptane sub-
mechanisms
...and some
rates were
divided by
10
50
37. log10(k)
@1000K
A
(cm3/mol/s)
n
(T=1K)
Ea
(cal/mol)
Reference
–19.4 4.5E-19 0 5,020
Curran (1998)
Discusses but gives no numbers
11.1 1.26E+11 0 0
Tsang (1988)
Literature Review
11.4 1.2E+12 0 3,000
Ranzi / Milan
(in models since pre-2008)
11.4 1.4E+12 0 5,000
Battin-Leclerc / Nancy
(Generated by EXGAS in 2001)
11.2 6.7E+20 -3.02 2,504
DeSain, Klippenstein, et al. 2003.
Ab initio/ master equation
The slow rates all come from one source
(through a variety of citations)
• Used by 5 of the 13 models
• “C3 sub-models” attributed to Ji (2012), Sarathy (2012),
Sarathy (2011), Sivaramakrishnan (2007), Peterson (2007),
Curran (2002), Pope (2000), Curran (1998)…
• Trail leads to Curran, Gaffuri, Pitz, Westbrook, Combust.
Flame, 114 (1998). Page 154 talks in depth about this family of
reactions, which are chemically activated via excited adduct,
but no rates are given.
⇄+ +
38. log10(k)
@1000K
A
(cm3/mol/s)
n
(T=1K)
Ea
(cal/mol)
Reference
–19.4 4.5E-19 0 5,020
Curran (1998)
Discusses but gives no numbers
11.1 1.26E+11 0 0
Tsang (1988)
Literature Review
11.4 1.2E+12 0 3,000
Ranzi / Milan
(in models since pre-2008)
11.4 1.4E+12 0 5,000
Battin-Leclerc / Nancy
(Generated by EXGAS in 2001)
11.2 6.7E+20 -3.02 2,504
DeSain, Klippenstein, et al. 2003.
Ab initio/ master equation*
The fast rates come from a variety of estimates
⇄+ +
*for comparison. Not used in these models
39. .edu/comocheng
!
!
!" #
!
!"
!
!
!"
!
!"#
$
!"
!
!" !
" #$
!"#
!" !"
!"#
"#! !"#
!
!"
!
!"
!
" #$
#$!
"
!
" #$
!
!" !""
#
!""
#
!"#
!"#$"
!
" #$
#$!
!
!"
!"
!
!
!
!"
!
!"#
!
!
!
!
!"""#
!""#
$
!""#
$
!"
!
" #$
!
!
" #$
%!
!
" #$
%
!
" #$
%
!
" #$
%!
Elizabeth
Becky
Eliza
Liz
Beth
with many names
for the same thing...
...it is difficult to
compare models...
...researchers
give molecules
nicknames.
...and easy to
make mistakes!
...and create
a unified
database...
with this we can:..
To publish in
"CHEMKIN"
format...
Our tool to
identify
species...
...allows us to
analyze models.
...find common
parameters,..
...detect
mistakes,..
...identify
controversial
rates,..
...of all
kinetic
models!
Richard H. West r.west@neu.edu Grant No. 1403171