Testing of Reaction Mechanisms for the Combustion of NH3/H2 Mixtures Using a Large Amount of Experimental Data

Abstract

Introduction
Ammonia (NH3) is a promising carbon-free fuel because it can be used in a sustainable and recyclable loop for energy production. It is often blended with other fuels in practical applications. One of the most often used co-fuels is hydrogen (H2). Accurate chemical kinetic models are needed that can describe the combustion of fuel mixtures containing NH3 under typical conditions of industrial applications. In this work, we compare the performance of 18 reaction mechanisms against a large amount of experimental data on the combustion of neat NH3 and NH3/H2 fuel mixtures. This work is an extension of our previous work on the same combustion system (Szanthoffer et al., Appl. Energ. Combust. Sci. 14 (2023) 100127.).

Full text

András Gy. Szanthoffer*, István Gy. Zsély, László Kawka, Máté Papp, Tamás Turányi
Testing of Reaction Mechanisms
for the Combustion of NH3/H2Mixtures
Using a Large Amount of Experimental Data
Introduction
•Ammonia (NH3) is a promising carbon-free
fuel because it can be used in a sustainable and
recyclable loop for energy production.
•It is often blended with other fuels in practical
applications. One of the most often used co-fuels
is hydrogen (H2).
•Accurate chemical kinetic models are needed
that can describe the combustion of fuel mixtures
containing NH3under typical conditions of
industrial applications.
•In this work we compare the performance of 18
reaction mechanisms against a large amount
of experimental data on the combustion of
neat NH3and NH3/H2fuel mixtures.
•This work is an extension of our previous work on
the same combustion system [1].
Acknowledgments
This work was supported by
NKFIH grant OTKA K132109
of the Hungarian National
Research, Development and
Innovation Office.
References
Institute of Chemistry, ELTE Eötvös Loránd University, Budapest, Hungary
Experimental data and simulations
•Experimental data were utilized from
46 publications from the literature:
Reaction mechanisms investigated Visual mechanism comparison
Exp. typea
XML/Ds./Dp.b
T/ K p/ atm φ
ST-IDTc41/41/315
1023
–
2489
1.01
–
41.65
0.50
–
2.07
JSRd44/153/2348 500–1452
1.00
–
1.40
0.01
–
1.50
FRe23/87/1125 451–1973
1.00
–
98.69
0.00
–
2.44
LBVf
188/188/3329
295–584g
0.50
–
36.58
0.40
–
2.00
Overall:
296/469/7117
a: Type of experiment
b: Number of XML data files/data series/data points
c: Shock tube ignition delay time measurements
d: Concentration measurements in jet stirred reactors
e: Concentration measurements in flow reactors
f: Laminar burning velocity measurements
g: Cold side temperatures of the flames Quantitative mechanism comparison –Results, discussion, conclusions
Experimental data were encoded in
ReSpecTh Kinetics Data (RKD)
version 2.4 format XML files [2].
1
Program Optima++ [3] read the XML data
file, set up the simulation environment, and
called OpenSMOKE++ [4] simulation code
to carry out the simulations.
2
All simulations were carried out with
OpenSMOKE++ [4] using a
0D (ST-IDT, JSR, FR) or 1D (LBV) models.
3
Error function
•Quantitative assessment of
mechanism performance was based on the
evaluation of an error function:
N
: number of data series
Nf
: number of XML data files
Nfs
: number of data series in the f-th XML data file
Nfsd
: number of data points in the s-th data series of the f-th XML data file

:fsd-th experimental result

:fsd-th simulation result
 
 : standard deviation of 

E
= 9 → 3

error on average
Mechanism ID
Species Reactions Ref.
Mathieu
-2015 33 160 [7]
Otomo
-2018 32 213 [8]
Li
-2019 34 252 [9]
POLIMI
-2020 31 203 [10]
Han
-2020 32 163 [11]
KAUST
-2021 34 262 [12]
Konnov
-2021 36 298 [13]
Dai
-2021 33 211 [14]
Bertolino
-2021 31 203 [15]
CEU
-2022 32 140 [16]
Sun
-2022 36 229 [17]
Tang
-2022 33 211 [18]
Shrestha
-2022 33 270 [19]
Zhou
-2023 33 233 [20]
Glarborg
-2023 34 228 [21]
POLIMI
-2023 31 205 [22]
KAUST
-2023 32 243 [23]
Shrestha
-2023 34 283 [24]
[1] A. Gy. Szanthoffer et al., Appl. Energ. Combust. Sci. 14 (2023) 100127.
[2] T. Varga et al., ReSpecTh Kinetics Data Format Specification v2.4, 2022.
http://ReSpecTh.hu/
[3] M. Papp et al., Optima++. http://ReSpecTh.hu
[4] The CRECK Modeling Group (POLIMI), OpenSMOKE++.
https://www.opensmokepp.polimi.it/
[5] T. Nagy, T. Turányi, ECM 2021,14–15 April, Naples, Italy. Paper 336.
[6] K. N. Osipova et al., Fuel 310 (2022) 122202.
[7] O. Mathieu and E. L. Petersen, Combust. Flame 162 (2015) 554–570.
[8] J. Otomo et al., Int. J. Hydrog. Energy 43 (2018) 3004–3014.
[9] R. Li et al., Fuel 257 (2019) 116059.
[10]A. Stagni et al., React. Chem. Eng. 5 (2020) 696–711.
[11] X. Han et al., Combust. Flame 213 (2020) 1–13.
[12] X. Zhang et al., Combust. Flame 234 (2021) 111653.
[13] X. Han et al., Combust. Flame 228 (2021) 13–28.
[14]L. Dai et al., Combust. Flame 227 (2021) 120–134.
[15] A. Bertolino et al., Combust. Flame 229 (2021) 111366.
[16] S. Wang et al., Combust. Flame 236 (2022) 111788.
[17] Z. Sun et al., Combust. Flame 243 (2022) 112015.
[18] R. Tang et al., Combust. Flame 240 (2022) 112007.
[19]K. P. Shrestha et al., Fuel Comm. 10 (2022) 100051.
[20] S. Zhou et al., Combust. Flame 248 (2023) 112536.
[21]P. Marshall and P. Glarborg, J. Phys. Chem. A 127 (2023) 2601–2607.
[22]A. Stagni et al., Proc. Combust. Inst. (2022)
[23] X. Zhang et al., Fuel 341 (2023) 127676.
[24] M. V. Manna et al., Proc. Combust. Inst. (2022)
[25] X. Han et al., Combust. Flame 206 (2019) 214–226.
[26] V. J. Wargadalam et al., Combust. Flame 120 (2000) 465–478.
A quantitative indicator is more useful
to assess the performance of mechanisms
on a large collection of experimental data.
ST-IDT measurement by Mathieu et al. [7] LBV measurement by Han et al. [25]
JSR measurement by Zhang et al. [12] FR measurement by Wargadalam et al. [26]
Mechanism ID
EST-IDT EJSR EFR ELBV
Eaverage
KAUST
-2023 17.7 48.4 19.4 14.1 24.9
POLIMI
-2020 9.8 53.8 23.4 14.3 25.3
Bertolino
-
2021
7.8 56.2 22.7 15.0 25.4
KAUST
-2021 21.3 48.6 18.2 14.9 25.7
POLIMI
-2023 10.4 55.4 23.7 14.2 25.9
Tang
-2022 6.6 44.8 23.5 32.0 26.8
Han
-2020 4.7 64.0 27.6 14.8 27.8
Otomo
-2018 6.5 54.9 25.9 24.0 27.8
Konnov
-2021 6.7 67.6 27.6 15.9 29.5
CEU
-2022 6.7 65.2 31.6 15.0 29.6
Mathieu
-2015 7.7 58.5 22.3 31.2 29.9
Sun
-2022 17.3 51.3 21.6 29.4 29.9
Li
-2019 11.9 63.0 31.4 19.3 31.4
Shrestha
-
2022
18.3 62.1 24.4 25.2 32.5
Shrestha
-
2023
16.1 48.0 34.0 33.6 32.9
Glarborg
-
2023
11.6 58.1 26.4 41.6 34.4
Dai
-2021 14.1 48.2 21.7 65.8 37.5
Zhou
-2023 67.8 59.0 28.3 13.5 42.1
E:
< 9 ≤
<
16
≤
<
25
≤
<
36
≤
<
49
≤
< 64 ≤
Estimation of the standard deviation of experimental data
•To estimate  
 , both the uncertainty reported by
the experimenters (

exp,
fsd
)and the estimated statistical
scatter of the fs-th data series (

fit,
fs
) were considered:
•

fit,
fs
was obtained by fitting a trendline (spline or
polynomial function) to the data points of the fs-th dataset
using the code Minimal Spline Fit [5]. Experimental data: Osipova et al. [6]
* Email address:
andras.gyorgy.szanthoffer@ttk.elte.hu





 
 

 


 
 
  
 
  
 
 
  


Averaged Evalues
•Comparison of the mechanisms by
computing their averaged Evalues for each
type of experiment and overall.
•Smaller Evalue →
better mechanism performance.
•The performance of a mechanism changes
significantly with the type of experiment.
Distribution of Efsd values
•Comparison of the mechanisms based on the
distribution of the 



 values →
stacked bar plot.
•More data points within the 3σlimits (Efsd < 9) →
better mechanism performance.
Conclusions
•The best-performing mechanisms are
almost the same in both cases.
•The performances of the investigated models are
very different.
•None of the models has Eaverage value smaller than 9
→ further mechanism development is necessary!