PHD Parente Full

Università di Pisa
Scuola di Dottorato in Ingegneria "Leonardo da Vinci"
Corso di Dottorato di Ricerca in
Ingegneria Chimica e dei Materiali
Tesi di Dottorato di Ricerca
Experimental and Numerical Investigation

of Advanced Systems for
Hydrogen-based Fuel Combustion
Autore:
Alessandro Parente
Relatori:
Prof. Ing. Leonardo Tognotti
Dott. Ing. Chiara Galletti
Anno 2008
Università di Pisa
Scuola di Dottorato in Ingegneria "Leonardo da Vinci"
Corso di Dottorato di Ricerca in
Ingegneria Chimica e dei Materiali
Tesi di Dottorato di Ricerca
Sperimentazione e Modellazione
di Sistemi Avanzati per la
Combustione di Idrogeno e sue Miscele
Autore:
Alessandro Parente
Relatori:
Prof. Ing. Leonardo Tognotti
Dott. Ing. Chiara Galletti
Anno 2008
Copyright © Alessandro Parente 2008
All Rights Reserved
Ad Aurora
Il est très simple: on ne voit bien qu’avec le cœur.

L’essentiel est invisible pour les yeux.
(Le Petit Prince - Antoine de Saint-Exupéry)
Summary
The hydrogen economy has received considerable attention during the last years,
in the academic, industrial and political fields, for its potential to deal with
the urgent issues related to the world energy scenario. Hydrogen can be pro-
duced directly from all primary energy sources, allowing fuel diversification
and energy independence. In particular, hydrogen containing fuels can be ob-
tained from the thermochemical conversion of coal and renewable sources such
as biomasses. The combination of the gasification technologies with carbon cap-
ture and storage (CCS) is particularly attracting, to increase H2 purity while
reducing greenhouse gas emissions, i.e. CO2 .
However, numerous challenges related to production, distribution and end
use still need to be faced for H2 to become an energy carrier. For example,
hydrogen purity is a major issue in fuel cells, as impurities can adversely affect
performances and durability. Fuel composition does not represent a priori a
concern for combustion systems; however, H2 properties may negatively affect
conventional combustion systems, leading to stability problems and large NOx
emissions. Therefore, efforts must be spent to develop technologies able to deal
with the complexities of hydrogen containing mixtures. To this purpose, hydro-
gen/natural gas fuels represent a more realistic alternative to pure hydrogen in
a short term perspective, as they provide a ready alternative to pure fossil fuels
and retain the H2 potentials from the point of view of greenhouse gas reduction
and efficiency increase.
The present Thesis reports numerical and experimental investigations of
advanced systems for hydrogen-based fuel combustion. In particular, atten-
tion is devoted to a novel combustion technology, named flameless combustion,
which allows to control NOx emissions while ensuring very high combustion
and thermal efficiencies. Flameless combustion is based on the modification
of the traditional flame structure: the system is driven towards homogeneous
(temperature and species) conditions, thus allowing to smooth the effect of
the oxidation process on the temperature distribution. Such effect is certainly
beneficial for controlling NOx formation and shows potentials for limiting the
reactivity of hydrogen-based fuels.
Three different case studies have been taken into account. Two of them are
semi-industrial systems, both installed at ENEL Ricerca facilities of Livorno,
Italy: a self recuperative burner used in the steel industry for heating appli-
7
Summary
cations and a micro-CHP unit for the distributed cogeneration of heat and
power. The third system is a lab-scale burner, developed at the Politecnico
di Milano. An integrated approach, based on both numerical modeling and
experiments, has been employed to assess the feasibility of flameless combus-
tion with hydrogen-enriched fuels, being the investigated devices designed for
burning natural gas.
Recognizing the complexity of the aforementioned systems, a hierarchical
approach is proposed to address the different chemical and physical processes
which are involved in the overall operations. In particular, from a modeling
prospective, the proper representation of turbulence-chemistry interactions is
fundamental to correctly capture the principal features of the combustion pro-
cess. Therefore, a fundamental study on turbulence-chemistry interactions in
turbulent reacting flows has been carried out, to determine the modeling ingre-
dients required for an accurate description of the flameless combustion regime.
A novel methodology based on Principal Component Analysis is presented for
the identification of the parameters controlling the evolution of a reacting sys-
tem and for the development of optimal combustion models. To this purpose,
high fidelity experimental and numerical data, available in the literature for
reference systems, have been used.
The results obtained from such fundamental activity have supported the
numerical modeling of the burners for hydrogen-enriched fuel combustion. The
major focus of the numerical simulation is represented by the choice of the com-
bustion model and kinetic mechanism and the sensitivity of the final predictions
on such choices. It is known that flameless combustion relies on a competition
between chemistry and turbulent mixing, ensured by either exhaust recircu-
lation or fuel dilution. Therefore, turbulence-chemistry interactions require
special treatment and simple combustion models are unsuited, as they cannot
capture the volumetric and diffuse features of the regime. Recent works on
the topic have suggested that only the use of detailed kinetic mechanisms can
lead to reliable results; however further investigation is needed, especially when
hydrogen is added to the fuel.
The numerical simulations of the flameless systems have been carried out
with commercial numerical tools; however, state-of-the-art physical models
from the literature have been coupled to the main code solvers, to enhance
their modeling capabilities. In particular, heat transfer models have been im-
plemented to simulate the system interactions with the surroundings and non-
conventional NO formation routes, have been introduced to account for NO
formation at low temperatures and with H2 in the fuel.
The quantitative validation of the computational approaches has repre-
sented a fundamental moment for the present Thesis, to critically identify
potentials and limits in the mathematical models and to plan further improve-
ments and developments. Therefore, the availability of the experimental data
has been crucial for judging the actual predictive capabilities of the numerical
simulations, over a wide range of operating conditions. In particular, the as-
8
Summary
sessment of the level of agreement between experimental data and numerical

simulations has been based on the Verification and Validation (V&V) method-
ology, which allows to determine the uncertainty in the modeling results by
estimating both the experimental and numerical errors.
9
Summary
10
Contents
List of Figures v
List of Tables xv
Nomenclature xix
1 Advanced systems for hydrogen-based fuel combustion 1

1.1 NOx formation and control in combustion processes . . . . . . . 2
1.1.1 Homogeneous NOx formation mechanisms . . . . . . . . 3
1.1.1.1 Thermal NO mechanism . . . . . . . . . . . . . 5
1.1.1.2 N2 O intermediate NO mechanism . . . . . . . 7
1.1.1.3 NNH intermediate NO mechanism . . . . . . . 7
1.1.1.4 Prompt NO mechanism . . . . . . . . . . . . . 8
1.1.1.5 Fuel NO mechanism . . . . . . . . . . . . . . . 10
1.1.2 NOx control in combustion . . . . . . . . . . . . . . . . 10
1.2 Flameless combustion . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Operating principles of flameless combustion . . . . . . 12
1.2.2 Review of the relevant literature in flameless combustion 15
1.2.2.1 Flameless combustion of hydrogen-enriched fuels 20
1.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 Turbulent combustion modeling 23

2.1 Governing equations for turbulent reacting flows . . . . . . . . 25
2.1.1 Constitutive equations . . . . . . . . . . . . . . . . . . . 26
2.2 Turbulent combustion . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Reynolds and Favre averaging . . . . . . . . . . . . . . . 29
2.2.2 Closure of Favre Averaged Navier-Stokes Equations (FANS) 31
2.3 Combustion models for mean reaction rates . . . . . . . . . . . 32
2.3.1 Eddy Break-up model and Eddy Dissipation Model . . . 34
2.3.2 Eddy Dissipation Concept . . . . . . . . . . . . . . . . . 35
2.3.2.1 Energy cascade model . . . . . . . . . . . . . . 36
2.3.2.2 Fine structures . . . . . . . . . . . . . . . . . . 38
2.3.2.3 Reactor model . . . . . . . . . . . . . . . . . . 39
2.3.3 The primitive variable approach . . . . . . . . . . . . . . 41
i
Contents
2.3.3.1 The mixture fraction variable . . . . . . . . . . 42

2.3.4 The Burke-Schumann solution . . . . . . . . . . . . . . . 44
2.3.5 Flamelet model . . . . . . . . . . . . . . . . . . . . . . . 45
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3 Methodology and Case Studies 49

3.1 Experimental equipment . . . . . . . . . . . . . . . . . . . . . . 50
3.2 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Verification and Validation . . . . . . . . . . . . . . . . . . . . . 53
3.3.1 Validation hierarchies . . . . . . . . . . . . . . . . . . . 55
3.3.2 Solution Verification . . . . . . . . . . . . . . . . . . . . 56
3.3.3 Assessing the level of agreement: validation metrics . . . 58
3.3.3.1 Definition of a confidence interval for the exper-
imental data . . . . . . . . . . . . . . . . . . . 58
3.3.3.2 Validation metrics based on confidence intervals 59
3.3.3.3 Global metrics . . . . . . . . . . . . . . . . . . 61
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4 Principal Components Analysis for turbulence-chemistry inter-

action modeling 63
4.1 Definition and derivation of Principal Components . . . . . . . 65
4.2 Sample PCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.1 Optimal properties of the PCA reduction . . . . . . . . 68
4.2.2 Data preprocessing: centering and scaling . . . . . . . . 69
4.2.2.1 Outlier detection and removal with PCA . . . 71
4.2.3 Choosing a subset of Principal Components or Variables 74
4.2.3.1 Cumulative percentage of total variance . . . . 74
4.2.3.2 Variance of Principal Components . . . . . . . 75
4.2.3.3 Broken Stick Model . . . . . . . . . . . . . . . 75
4.2.3.4 Scree plot . . . . . . . . . . . . . . . . . . . . . 76
4.2.3.5 Choosing a subset of Variables . . . . . . . . . 76
4.2.4 Interpretation of principal components . . . . . . . . . . 79
4.3 Local Principal Components Analysis . . . . . . . . . . . . . . . 81
4.4 Data sets for model validation . . . . . . . . . . . . . . . . . . . 84
4.4.1 Experimental data . . . . . . . . . . . . . . . . . . . . . 84
4.4.2 Numerical data . . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.1 PCA for the identification of low-dimensional manifolds 86
4.5.1.1 GPCA of experimental data sets . . . . . . . . 87
4.5.1.2 LPCA of experimental and numerical data sets 106
4.6 Development of a PCA based combustion model . . . . . . . . 122
4.6.1 Transport equations for the PCs . . . . . . . . . . . . . 124
4.6.2 PCA Modeling Approach . . . . . . . . . . . . . . . . . 125
4.6.3 Parametrizing the State Variables . . . . . . . . . . . . . 126
ii
Contents
4.6.4 Parametrizing Source Terms . . . . . . . . . . . . . . . . 129

4.6.5 Global versus Semi-Local PCA . . . . . . . . . . . . . . 131
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5 Experimental and numerical investigation of a self-recuperative

flameless burner 135
5.1 Description of the burner . . . . . . . . . . . . . . . . . . . . . 135
5.1.1 Experimental campaign n.1 . . . . . . . . . . . . . . . . 140
5.1.1.1 Extrapolation of NO emissions to steady state 142
5.1.2 Experimental campaign n. 2 . . . . . . . . . . . . . . . . 146
5.2 Numerical modeling of the self-recuperative burner . . . . . . . 146
5.2.1 Modeling approach for experimental campaign n.1 . . . 149
5.2.1.1 Flameless combustion of methane . . . . . . . 149
5.2.1.2 Flameless combustion of methane-hydrogen mix-
tures . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2.2 Modeling approach for experimental campaign n.2 . . . 158
5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.3.1.1 Flameless combustion of methane . . . . . . . 162
5.3.1.2 Flameless oxidation of methane-hydrogen mix-
tures . . . . . . . . . . . . . . . . . . . . . . . . 169
5.3.2.1 CFD analysis . . . . . . . . . . . . . . . . . . . 186
5.3.2.2 Kinetic post-processing . . . . . . . . . . . . . 194
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
6 Experimental and numerical investigation of the flameless com-

bustion of hydrogen-enriched fuels in a lab-scale burner 197
6.1 Description of the burner and experimental campaigns . . . . . 197
6.2 Numerical modeling of the lab-scale burner . . . . . . . . . . . 200
6.2.1 Computational domain and grid . . . . . . . . . . . . . 200
6.2.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . 203
6.2.3 Physical models . . . . . . . . . . . . . . . . . . . . . . . 203
6.2.3.1 Turbulence/chemistry interactions and kinetic
mechanisms . . . . . . . . . . . . . . . . . . . . 203
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
6.3.1 Flow-field characterization . . . . . . . . . . . . . . . . . 204
6.3.2 Effect of turbulence/chemistry interaction models . . . . 205
6.3.3 Effect of dilution on the combustion regime . . . . . . . 209
6.3.4 Model validation: comparison of simulations and experi-
ments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
iii
Contents
7 Experimental and numerical investigation of a micro-CHP flame-

less unit 219
7.1 Distributed combined heat and power generation . . . . . . . . 220
7.2 Description of the system . . . . . . . . . . . . . . . . . . . . . 220
7.2.1 Stirling engine . . . . . . . . . . . . . . . . . . . . . . . 224
7.3 Experimental campaign . . . . . . . . . . . . . . . . . . . . . . 226
7.4 Numerical modeling of the micro-CHP flameless unit . . . . . . 229
7.4.1 Computational domain and grid . . . . . . . . . . . . . 229
7.4.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . 231
7.4.3 Physical models . . . . . . . . . . . . . . . . . . . . . . . 234
7.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
7.5.1 Flow field and main operating features . . . . . . . . . . 236
7.5.2 Effect of the kinetic mechanism on the flame structure . 238
7.5.3 Effect of H2 addition to the fuel . . . . . . . . . . . . . . 241
7.5.4 Effect of burner load . . . . . . . . . . . . . . . . . . . . 242
7.5.5 Model validation . . . . . . . . . . . . . . . . . . . . . . 246
7.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Concluding remarks 251
Bibliography 253
List of Publications 265
iv
List of Figures
1.1 Combustion of hydrogen containing fuels: issues and solutions. 2

1.2 Flammable range for methane-air mixture as a function of the
recirculation degree, kv [33]. . . . . . . . . . . . . . . . . . . . . 13
1.3 Stability diagram for methane-air combustion, as a function of
the recirculation degree and operating temperature [2]. . . . . . 14
1.4 Idealized flameless oxidation process [2]. . . . . . . . . . . . . . 14
1.5 High velocity burner for operation in flame and flameless regime. 15
1.6 Experimental apparatus used by Wünning and Wünning [2]. . . 16
1.7 FLOX® recuperative burner [2]. . . . . . . . . . . . . . . . . . 16
1.8 Time resolved temperature measurement for flame, unstable and
flameless regime [2]. . . . . . . . . . . . . . . . . . . . . . . . . 17
1.9 Temperature and OH radical concentration in flame and flame-
less regimes [35]. . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1 Verification and Validation methodology [1]. . . . . . . . . . . . 24

2.2 Energy cascade model [65, 66, 71]. . . . . . . . . . . . . . . . . 37
2.3 Mass flow through a cylinder. . . . . . . . . . . . . . . . . . . . 39
2.4 PSR model for the fine structures. . . . . . . . . . . . . . . . . 40
2.5 Burke-Schumann solution as a function of the mixture fraction. 45
3.1 Lab-scale burner (a), FLOX® burner with air preheater (b) and
SOLO Stirling 161 Cogeneration Unit (c). . . . . . . . . . . . . 51
3.2 Integrated CFD–experimental approach. . . . . . . . . . . . . . 52
3.3 Verification and Validation: objectives to quantify and tools. . . 53
3.4 Two phases of validation activities . . . . . . . . . . . . . . . . 54
3.5 V&V hierarchy for a flameless system. . . . . . . . . . . . . . . 56
4.1 Principal components scores with (a) and without (b) outliers. . 73
4.2 Eigenvalues size with (a) and without (b) outliers. . . . . . . . 73
4.3 Size reduction process with PCA. . . . . . . . . . . . . . . . . . 74
4.4 Schematic illustration of the VARIMAX rotation [110]. . . . . . 81
4.5 Schematic illustration of the VQPCA algorithm [92] . . . . . . . 83
4.6 Schematic illustration of the FPCA algorithm [92] for a CO/H2
flame [112].. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
v
List of Figures
4.7 Scree-graph and histograms of the q largest eigenvalues for the

jet flame data set, preprocessed with auto scaling. . . . . . . . . 88
4.8 Parity plots of temperature (a), H2 O (b), H2 (c), CO (d), OH
(e) and NO (f) mass fractions illustrating the GPCA (q = 2)
reduction of the jet flame data set. Scaling criterion adopted:
auto scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.9 Parity plots of temperature (a), H2 O (b), H2 (c), CO (d), OH
reduction of the jet flame data set. Scaling criterion adopted:
auto scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.10 Scree-graph and histograms of the q largest eigenvalues for Flame
D (a), Flame F (b) and JHC (c). Scaling criterion adopted: auto
scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.11 Parity plots of temperature (a), H2 O (b), CO (c), H2 (d), OH
reduction of Flame F. Scaling criterion adopted: auto scaling. 98
4.12 Parity plots of temperature (a), H2 O (b), CO (c), H2 (d), OH
reduction of Flame F. Scaling criterion adopted: auto scaling. 99
4.13 Parity plots of temperature (a), H2 O (b), CO (c), H2 (d), OH (e)
and NO (f) mass fractions illustrating the GPCA (q = 4) reduc-
tion of JHC data set. Scaling criterion adopted: auto scaling.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.14 Parity plots of temperature (a), H2 O (b), H2 (c), CO (d), OH (e)
and NO (f) mass fractions illustrating the GPCA (q = 5) reduc-
tion of JHC data set. Scaling criterion adopted: auto scaling.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.15 Scree-graph and histograms of the q largest eigenvalues for the
jet flame data set, preprocessed with range (a), vast (b), level
(c) and max (d) scaling. . . . . . . . . . . . . . . . . . . . . . . 107
and NO (f) mass fractions illustrating the VQPCA (q = 2, k =
8) reduction of the jet flame data set. GSRE,n = 0.08. . . . . . 109
8) reduction of Flame F data set. GSRE,n = 0.08. . . . . . . . . 110
6) reduction of JHC data set. GSRE,n = 0.08. . . . . . . . . . . 111
4.19 Original (a) and conditioned (b) temperature field for DNS2 data
set at time step t = 1.5e − 03 s. . . . . . . . . . . . . . . . . . . 113
4.20 Contour plots of original and recovered temperature (a, a’) and
OH mass fraction (b, b’) distribution for DNS1. VQPCA reduc-
tion with q = 3 and k = 8. GSRE,n = 0.01. . . . . . . . . . . . 115
vi
List of Figures
4.21 Parity plots of original and recovered temperature (a, a’) and OH
mass fraction (b, b’) distribution for DNS1. VQPCA reduction
with q = 3 and k = 8. GSRE,n = 0.04. . . . . . . . . . . . . . . 116
4.22 Contour plots of original (a, b) and recovered (a’, b’) temper-
ature distribution for DNS2, at two different time steps, i.e.
t = 1.5e − 03 s (a, a’) and t = 2.0e − 03 s (b, b’). VQPCA
reduction with q = 4 and k = 8. GSRE,n = 0.04. . . . . . . . . 117
4.23 Parity plots of temperature (a) and OH (b) mass fraction illus-
trating the VQPCA (q = 4, k = 8) reduction of DNS2 data set.
GSRE,n = 0.04. . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.24 Values of GSRE,n as a function of the number of clusters, k, and
retained PCs, q, for the jet flame, Flame F and JHC data sets. 118
4.25 Values of GSRE,n as a function of the number of clusters, k, and
retained PCs, q, for the DNS1 and DNS2 data sets. . . . . . . . 119
4.26 Temperature as a function of mixture fraction in the two clusters
selected by VQPCA for the jet flame. q = 2 and GSRE,n = 0.21. 119
selected by VQPCA for Flame F. q = 3 and GSRE,n = 0.21. . . 120
selected by VQPCA for the JHC data set. q = 3 and GSRE,n =
0.21. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6) reduction of JHC data set. GSRE,n = 0.08. . . . . . . . . . . 123
4.30 CPU time associated with the FPCA and VQPCA reductions.as
a function of the number of clusters, k, and retained PCs, q, for
the experimental (a) and numerical (b) data sets. . . . . . . . . 123
4.31 Parametrization of temperature at fst by χ (a) and z1 (b) for
case B. Solid lines are the doubly-conditional mean temperature.
R2 is calculated from Eq. (4.54). . . . . . . . . . . . . . . . . . 127
4.32 Parametrization of OH mass fraction at fs t by χ (a) and z1 (b)
for case B. Solid lines are the doubly-conditional mean temper-
ature. R2 is calculated from Eq. (4.54). . . . . . . . . . . . . . 129
4.33 Parametrization of ω̇z1 at fst by z1 for case B. Solid line: doubly-
conditional mean value of ω̇z1 . R2 is calculated from Eq. (4.54). 130
5.1 Longitudinal section of the FLOX® burner. . . . . . . . . . . . 136

5.2 3D view of the combustion chamber. . . . . . . . . . . . . . . . 137
5.3 Outer surface of the air preheater. . . . . . . . . . . . . . . . . 137
5.4 Inconel® shield (a) and outer insulation layer (b). . . . . . . . 138
5.5 Water heat exchanger during the first (a) and second (b) exper-
imental campaigns. . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.6 Outer surface of the air preheater. . . . . . . . . . . . . . . . . 139
vii
List of Figures
5.7 NO concentration as a function of time during a typical run of

the first experimental campaign. . . . . . . . . . . . . . . . . . 140
5.8 O2 concentration in the flue gases with NG (a) and NG-H2 mix-
tures (b), for the experimental campaign n. 1 on the FLOX®
burner. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.9 NO emissions for NG and NG-H2 mixtures for the experimental
campaign n. 1 on FLOX® burner. . . . . . . . . . . . . . . . . 143
5.10 Raw data, experimental mean and standard deviation for the
NO emissions vs. time. Experimental run n. 5, Table 5.1. . . . 144
5.11 Experimental mean plus 98% confidence intervals and compari-
son with the mathematical models for the NO emissions vs. time.
Experimental run n. 5, Table 5.11. . . . . . . . . . . . . . . . . 145
5.12 Estimated error plus 98% confidence interval showing the level
of agreement between experimental measurements and mathe-
matical models. Experimental run n. 5, Table 5.12. . . . . . . . 145
5.13 NO emissions as a function of the H2 load (a) and O2 concentra-
tion in the flue gases (b). Experimental campaign n. 2 on the
FLOX® burner (Table 5.3). . . . . . . . . . . . . . . . . . . . . 148
5.14 Temperature measurements on the radiant tube (a), on the in-
ternal (b) and external (c) surface of the Inconel® shield and on
the internal surface of the water heat exchanger (d). H2 ther-
mal input = 30%. Experimental campaign n. 2 on the FLOX®
burner (Table 5.3). . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.15 3D (a) and axisymmetric (b) grids . . . . . . . . . . . . . . . . 151
5.16 Heat exchange between the coaxial cylindrical shields represent-
ing the radiant tube, the internal and external insulation layers
of the Inconel® shield and the water heat exchanger. . . . . . . 152
5.17 Grid Convergence Index (GCI) as a function of the coarsening
factor, r, for the set of 2D grids investigated in the numerical
simulation of the FLOX® burner. Assumed order of convergence
p = 1 [81]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.18 Computational mesh for FLUENT simulations. . . . . . . . . . 156
5.19 Estimated error, δRE , for the profiles of axial velocity obtained
at x = 0.25 m (a), x = 0.35 m (b) and x = 0.45 m (c) with the
selected grid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.20 Conceptual scheme of the methodology adopted for the numerical sim-
ulation of the experimental campaign n. 2 on the FLOX burner ® . . 161
5.21 Measured and computed temperature profiles along the radiant
tube for two different burner loads, i.e. Q̇in = 8.5 kW and
Q̇in = 9.5 kW. . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.22 Flow pattern inside the combustion chamber, determined by the
high-momentum air jet issuing into the flame tube. Run 1, Table
5.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
viii
List of Figures
5.23 Temperature distribution in the combustion chamber for Aair,in =

339 mm2 (a), Aair,in = 201 mm2 (b), Aair,in = 88 mm2 (c),
Aair,in = 64 mm2 (d), Aair,in = 46 mm2 (e) and Aair,in =
41 mm2 (f). Run 1, Table 5.4. . . . . . . . . . . . . . . . . . . . 164
5.24 Axial temperature profiles at r = 7 mm for different air inlet
cross-sectional areas. Run 1, Table 5.4. . . . . . . . . . . . . . 165
5.25 Recirculation degree, kR , and NO emissions as a function of the
air inlet cross-sectional area. Run 1, Table 5.4. . . . . . . . . . 166
5.26 Temperature (colored map) and Damkohler number (contours)
distributions for Aair,in = 339 mm2 (a), Aair,in = 88 mm2 (b)
and Aair,in = 41 mm2 (c). Run 1 Table 5.4. . . . . . . . . . . . 168
5.27 Radiant tube temperature for different air inlet cross-sectional
area, Aair,in . Run 1 Table 5.4. . . . . . . . . . . . . . . . . . . . 169
5.28 Preheater and radiant rube efficiencies as a function of the recir-
culation degree, kR . Run 1 Table 5.4. . . . . . . . . . . . . . . . 170
5.29 Temperature distribution in the burner fed with CH4 /H2 (Run
20, Table 5.4) predicted by ED/FR with global chemistry (a),
EDC with global chemistry (b) , EDC with DRM-19 (c) and
EDC with GRI-3.0 (d). . . . . . . . . . . . . . . . . . . . . . . . 171
5.30 Radial profiles of temperature at different axial locations along
the axis, i.e. x = 0.075 m (a), x = 0.15 m (b) and x = 0.25 m,
predicted by different combustion models and kinetic mecha-
nisms. Run 20, Table 5.4. . . . . . . . . . . . . . . . . . . . . . 173
5.31 Temperature distribution predicted by EDC with GRI-3.0 in the
burner fed with increasing H2 mass fraction (Runs 5, 20, 1m and
2m, Table 5.4): 0% (a), 5.5% (b), 10% (c) and 20% (d). . . . . 174
the axis, i.e. x = 0.075 m (a), x = 0.15 m (b) and x = 0.25 m,
predicted by different combustion models and kinetic mecha-
nisms. Runs 5, 20, 1m and 2m, Table 5.4. . . . . . . . . . . . . 176
5.33 OH radical distribution predicted by EDC with GRI-3.0 in the
burner fed with increasing H2 mass fraction (Runs 5, 20, 1m and
2m, Table 5.4): 0% (a), 5.5% (b), 10% (c) and 20% (d). . . . . 177
5.34 Radial profiles of OH radical mass fraction at different axial lo-
cations along the axis, i.e. x = 0.075 m (a), x = 0.15 m (b)
and x = 0.25 m, predicted by different combustion models and
kinetic mechanisms. Runs 5, 20, 1m and 2m, Table 5.4. . . . . 178
5.35 Axial profile at r = 0 of OH radical and CH2 O species mass
fraction. Runs 5, 20, 1m and 2m, Table 5.4. . . . . . . . . . . . 179
5.36 CH2 O species mass fraction distribution predicted by EDC with
GRI-3.0 in the burner fed with increasing H2 mass fraction (Runs
5, 20, 1m and 2m, Table 5.4): 0% (a), 5.5% (b), 10% (c) and
20% (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
ix
List of Figures
5.37 Comparison of predicted and measured NO emissions in the flue

gases. Runs 5, 20, 22 and 23 in Table 5.4): . . . . . . . . . . . . 183
5.38 Effect of molecular diffusion on the radial profiles of H2 mass
fraction (a, b), CO2 mass fraction (c, d) and temperature (e, f)
at different axial locations along the burner axis. Combustion
model and kinetic mechanism: EDC with DRM- 19. Runs 5, 20,
1m and 2m, Table 5.4. . . . . . . . . . . . . . . . . . . . . . . . 184
5.39 Laminar to turbulent H2 diffusion coefficient with increasing H2
mass fraction in the fuel: 0% (a), 5.5% (b), 10% (c) and 20%
(d). Runs 5, 20, 1m and 2m, Table 5.4. . . . . . . . . . . . . . . 185
5.40 Temperature field in the burner fed with increasing H2 in the
fuel. H2 content corresponding to 10% (a), 20% (b), 30% (c),
50% (d), 60% (e) and 70% (f) of the total burner thermal input,
10 kW. Kinetic mechanism: KEE-58. Runs 4-9, Table 5.3. . . . 186
5.41 OH radical mass fraction distribution in the burner fed with in-
creasing H2 in the fuel. H2 content corresponding to 10% (a),
20% (b), 30% (c), 50% (d), 60% (e) and 70% (f) of the to-
tal burner thermal input, 10 kW. Kinetic mechanism: KEE-58.
Runs 4-9, Table 5.3. . . . . . . . . . . . . . . . . . . . . . . . . 187
5.42 CH2 O radical mass fraction distribution in the burner fed with
increasing H2 in the fuel. H2 content corresponding to 10% (a),
20% (b), 30% (c), 50% (d), 60% (e) and 70% (f) of the to-
tal burner thermal input, 10 kW. Kinetic mechanism: KEE-58.
Runs 4-9, Table 5.3. . . . . . . . . . . . . . . . . . . . . . . . . 187
5.43 Sample mean and standard deviation for the average radiant
tube temperature (a) and NO concentration in the flue gases
(b), when increasing the hydrogen thermal input from 10% to
70%. Runs 4-9, Table 5.3. The sample mean and the standard
deviations are plotted on different scales. . . . . . . . . . . . . . 188
5.44 95% confidence intervals for the average temperature of the ra-
diant tube and comparison with the numerical results (a). 95%
confidence interval for the true error and estimated errors for the
average temperature of the radiant tube (b). Runs 4-9, Table 5.3. 189
5.45 95% confidence intervals for the average NO emission from the
burner and comparison with the numerical results (a). 95% con-
fidence intervals for the true error and estimated errors for the
average NO emissions from the burner (b). Runs 4-9, Table 5.3 190
5.46 Radial profiles of temperature, H, O and OH radical mass frac-
tions at different axial locations along the burner, as predicted
by the KEE-58 and DRM-19 kinetic mechanisms, for a hydrogen
thermal input of 20%. Run 4, Table 5.3 . . . . . . . . . . . . . 192
x
List of Figures
5.47 Radial profiles of temperature, H, O and OH radical mass frac-

tions at different axial locations along the burner, as predicted
by the KEE-58 and DRM-19 kinetic mechanisms, for a hydrogen
thermal input of 70%. Run 9, Table 5.3 . . . . . . . . . . . . . 193
5.48 Relative importance of NO formation routes as a function of
hydrogen thermal input for the complete NO mechanism. . . . . 193
6.1 Laboratory scale experimental equipment. Sketch and main di-

mensions [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6.2 Sketch of the exhaust gas outlet sections. . . . . . . . . . . . . . 199
6.3 3D computational domain and grid. . . . . . . . . . . . . . . . . 202
6.4 2D computational domain and grid. . . . . . . . . . . . . . . . . 202
6.5 Velocity streamlines. Run 6, Table 6.1. . . . . . . . . . . . . . . 205
6.6 Temperature distribution in the burner fed with CH4 (Run 1,
Table 6.1), predicted by ED/FR with global chemistry (a), EDC
with global chemistry (b), EDC with KEE-58 (c), DRM-19 (d)
and GRI-3.0 (e). . . . . . . . . . . . . . . . . . . . . . . . . . . 206
the axis, i.e. x = 0.06 m (a) and x = 0.10 m (b), predicted
by different combustion models and kinetic mechanisms. Run 1,
Table 6.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
6.8 Temperature distribution in the burner fed with a mixture con-
taining 60% by vol. of H2 (Run 6, Table 6.1), predicted by
ED/FR with global chemistry (a), EDC with global chemistry
(b), EDC with KEE-58 (c), DRM-19 (d) and GRI-3.0 (e). . . . 208
the axis, i.e. x = 0.06 m (a) and x = 0.10 m (b), predicted
by different combustion models and kinetic mechanisms. Run 6,
Table 6.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.10 Temperature distribution in the combustion chamber fed with
CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for dif-
ferent dilution ratios, kv : 7.3 (a), 10.6 (b), 11.2 (c), 13.2 (d) and
19.1 (e). Runs 2-6, Table 6.1. . . . . . . . . . . . . . . . . . . . 210
6.11 Radial profiles of temperature (a, b) and OH radical mass frac-
tion (a’, b’), at different axial locations along the axis, i.e. x =
0.06 m (a, a’) and x = 0.10 m (b, b’), predicted with EDC/GRI-
3.0 for different dilution ratios, kv : 7.3 (FLAME), 10.6 (TR1),
11.2 (TR2), 13.2 (TR3) and 19.1 (MILD). Runs 2-6, Table 6.1. 211
6.12 OH radical distribution in the combustion chamber fed with
CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for dif-
ferent dilution ratios, kv : 10.6 (a), 11.2 (b), 13.2 (c) and 19.1
(d). Runs 3-6, Table 6.1. . . . . . . . . . . . . . . . . . . . . . . 212
xi
List of Figures
6.13 CH2 O mass fraction distribution in the combustion chamber fed

with CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0
for different dilution ratios, kv : 10.6 (a), 11.2 (b), 13.2 (c) and
19.1 (d). Runs 3-6, Table 6.1. . . . . . . . . . . . . . . . . . . . 213
6.14 Comparison of measured and predicted temperatures inside the
combustion chamber, Tmiddle , x = 0.18 m and r = 0.014 m, Runs
2-6, Table 6.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
6.15 NO emissions in the flue gases obtained with CH4 /H2 fuel (60%
H2 by vol.) for different dilution ratios, kv . Temperature and
species fields predicted by different combustion models and ki-
netic mechanisms. Runs 2-6, Table 6.1. . . . . . . . . . . . . . . 216
6.16 Contribution of different formation routes to total NO emissions
with CH4 /H2 fuel (60% H2 by vol.), for different dilution ratios,
kv . Temperature and species fields predicted by EDC with GRI-
3.0. Runs 2-6, Table 6.1. . . . . . . . . . . . . . . . . . . . . . . 217
7.1 SOLO Stirling 161 Cogeneration Unit. . . . . . . . . . . . . . . 221

7.2 Flame tube (a) and finned heat exchanger (b). . . . . . . . . . 222
7.3 Air (a) and fuel (b) feeding system. . . . . . . . . . . . . . . . . 223
7.4 Schematic diagram of flow streams and heat exchanges. . . . . 224
7.5 Thermocouples on the heat exchanger (a) and on the top of the
expansion cylinder (b). . . . . . . . . . . . . . . . . . . . . . . . 225
7.6 Functional diagram of Stirling engine. . . . . . . . . . . . . . . 225
7.7 Sketch of the combustion chamber and details about the feeding
system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
7.8 Computational domain and grid with details of the injection noz-
zles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.9 Grid Convergence Index (GCI) as a function of the coarsening
factor, r, for the set of 3D grids investigated in the numerical
simulation of the FLOX® burner. Assumed order of convergence
p = 1 [81]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
7.10 Estimated error, δRE , for the profiles of axial velocity obtained
at x = 0.25 m (a), x = 0.35 m (b) and x = 0.45 m (c) with the
selected grid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
7.11 Control region for the evaluation of the heat transferred to the
operating fluid of the Stirling cycle. . . . . . . . . . . . . . . . . 234
7.12 Angular Section (180 degrees) of the fluid volume corresponding
to the flue gases/helium heat exchanger. Run 12, Table 7.2. . . 235
7.13 Flow pattern inside the combustion chamber, determined by the
high-momentum air jet issuing into the flame tube. Run 8, Table
7.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
CH4 , predicted by the 4-step (a) and KEE-58 (b) kinetic mech-
anisms. PHe = 90 bar. Runs 5, Table 7.2. . . . . . . . . . . . . 239
xii
List of Figures

CH4 /H2 (H2 9% by wt.), predicted by the 4-step (a) and KEE-
58 (b) kinetic mechanisms. PHe = 90 bar. Runs 8, Table 7.2. . 240
7.16 Radial temperature profiles at different axial locations, i.e. x =
0.075 m (a), x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m
(d), predicted by the 4-step and KEE-58 mechanisms for a hydro-
gen content in the fuel equal to 7 and 9% (by wt.), respectively..
PHe = 90 bar. Runs 7 and 8, Table 7.2. . . . . . . . . . . . . . 241
CH4 /H2 mixtures containing 0 (a), 7 (b) and 9 (c) % (by wt.) of
H2 , predicted by the KEE-58 kinetic mechanism. PHe = 90 bar.
Runs 5, 7 and 8, Table 7.2. . . . . . . . . . . . . . . . . . . . . 243
7.18 OH radical mass fraction distribution in the combustion chamber
fed with CH4 /H2 mixtures containing 0 (a), 7 (b) and 9 (c) %
(by wt.) of H2 , predicted by the KEE-58 kinetic mechanism.
PHe = 90 bar. Runs 5, 7 and 8, Table 7.2. . . . . . . . . . . . . 244
7.19 Radial temperature profiles at different axial locations, i.e. x =
0.075 m (a), x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m
(d), predicted by the KEE-58 kinetic mechanism for a hydrogen
content in the fuel equal to 0, 7 and 9% (by wt.), respectively..
PHe = 90, 120 bar. Runs 5, 7-9, 11 and 12, Table 7.2. . . . . . 245
CH4 , predicted by the KEE-58 kinetic mechanisms. PHe =
120 bar. Runs 9, Table 7.2. . . . . . . . . . . . . . . . . . . . . 246
7.21 Comparison between measured and predicted CO (a) and NO
(b) emissions. Experimental data with 95% confidence intervals.
Runs 5, 7-9, 11 and 12, Table 7.2. . . . . . . . . . . . . . . . . . 247
xiii
List of Figures
xiv
List of Tables
1.1 Reaction parameters for the extended Zeldovich mechanism and

for the NO2 intermediate NO mechanism. Units: mol, cm, s, K. 6
4.1 Covariance matrix for the jet flame data set. Scaling criterion
adopted: auto scaling. . . . . . . . . . . . . . . . . . . . . . . . 89
4.2 Total, tq , and individual variance, tq,j , accounted for the jet flame
data set, as a function of the number of retained PCs, q, and the
preprocessing criterion. . . . . . . . . . . . . . . . . . . . . . . . 90
4.3 Covariance matrix for Flame D data set. Scaling criterion adopted:
auto scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4 Covariance matrix for Flame F data set. Scaling criterion adopted:
auto scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5 Covariance matrix for JHC data set. Scaling criterion adopted:
auto scaling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.6 Total, tq , and individual variance, tq,j , accounted for Flame D,
F and JHC data sets by the GPCA reduction, as a function of
the number of retained PCs, q. . . . . . . . . . . . . . . . . . . 102
4.7 Retained (a) and rotated (b) eigenvectors for the jet flame data
set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.8 Retained (a) and rotated (b) eigenvectors for Flame D data set. 103
4.9 Retained (a) and rotated (b) eigenvectors for Flame F data set. 103
4.10 Retained (a) and rotated (b) eigenvectors for JHC data set. . . 104
4.11 Principal variables for the jet flame data set, as provided by the
different methods described in Section 4.2.3.5. . . . . . . . . . 105
4.12 Principal variables for Flame D, F and the JHC data set. PV
method: MC2 (Section 4.2.3.5). . . . . . . . . . . . . . . . . . 105
4.13 Values of GSRE,n associated with the GPCA, VQPCA and FPCA
reconstructions of the jet flame, flame F and JHC data set, as a
function of the number of clusters, k, and retained PCs, q. . . . 107
4.14 Total, tq , and individual variance, tq,j , accounted for by the re-
tained PCs for the DNS1 and DNS2 data sets as a function of
the number of components, q. . . . . . . . . . . . . . . . . . . . 112
xv
List of Tables
4.15 Rotated eigenvector in the first (a) and second (b) cluster iden-
tified by VQPCA for Flame F. q = 3 and GSRE,n = 0.21 . . . . 121
4.16 Rotated eigenvector in the first (a) and second (b) cluster iden-
tified by VQPCA for the JHC data set. q = 3 and GSRE,n = 0.21122
4.17 R2 values defined by Eq. (4.54). Also shown are results for the χ
parametrization. All results are at f = fst = 0.4375. . . . . . . . . . 128
4.18 R2 values defined by Eq. (4.54) for PC source terms, sη . . . . . . . . 130
4.19 R2 values at f = 0.2 using the PCA obtained at fst . Also shown are
results for the χ parametrization. . . . . . . . . . . . . . . . . . . 132
4.20 R2 values at f = 0.6 using the PCA obtained at fst . Also shown are
results for the χ parametrization. . . . . . . . . . . . . . . . . . . 133
5.1 Summary of the experimental campaign n.1 on the FLOX® burner.141

5.2 Extrapolated NO emissions for a selected subset of experimental
runs listed in Table 5.1. . . . . . . . . . . . . . . . . . . . . . . 146
5.5 Global 4-step mechanism [23]. Reaction rates units: kmol · m−3 · s−1
. Activation energy units: J · kmol−1 . . . . . . . . . . . . . . . . . 160
5.6 Computed maximum temperatures and, predicted and measured,
NO emissions with different combustion models and kinetic mech-
anisms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.7 Average and maximum global validation metrics for the average
temperature of the radiant tube. . . . . . . . . . . . . . . . . . 189
5.8 Average and maximum global validation metrics for the average
NO emissions from the burner. All average metrics are ±2.1%,
with 95% confidence. . . . . . . . . . . . . . . . . . . . . . . . . 191
6.1 Summary of the experimental campaign on the lab-scale burner. 201

6.2 Location of the recirculation loop, aerodynamic recirculation ra-
tio, RA , and dilution factor, kv . . . . . . . . . . . . . . . . . . . 206
6.3 Average global validation metrics for temperature measurements
at two locations (x = 0.10 m r = 0.014 m and x = 0.18 m r =
0.014 m) and NO emissions from the burner. . . . . . . . . . . 214
7.1 Typical performances for the SOLO Stirling 161 Cogeneration

Unit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.2 Summary of the experimental campaign on the SOLO Stirling
CHP unit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
7.3 Flue gas composition and main operation outputs for the runs
investigated in Table 7.2. . . . . . . . . . . . . . . . . . . . . . . 228
7.4 Global 4-step mechanism Jones and Lindstedt [60]. Reaction rates
units: kmol · m−3 · s−1 . Activation energy units: J · kmol−1 . . . . . 236
xvi
List of Tables
7.5 Macro indicators of the main operating features of the SOLO

Stirling CHP unit. Data extracted from the numerical simula-
tions carried out with the KEE-58 kinetic mechanism. . . . . . 238
7.6 Relative and global validation metrics for NO and CO measure-
ments at different burner loads, obtained with the 4-step and
KEE-58 kinetic mechanisms. Runs 5, 7-9, 11 and 12, Table 7.2. 248
xvii
Nomenclature
a Oxygen reaction order in the one-step prompt

mechanism
ak Eigenvector corresponding to the kth largest
eigenvalue of S (or Σ)
A Pre-exponential factor, reaction dependent
A Constant of the Eddy Dissipation Model
A Matrix containing the eigenvectors of S
A∗ Exchange surface of the fine structures, m2
bke Number of atoms of element e in a molecule of
species k
B Constant of the Eddy Dissipation Model
B Rotation matrix
BC Boundary Conditions
C Confidence level adopted for the definition of the CIs
CD1 , CD2 Constants in the energy cascade model
CEBU Constant of the Eddy Break-up Model
Cµ Constant in the k − model
cp Specific heat at constant pressure, J kg-1 K-1
CAE Computer Aided Engineering
CF D Computational Fluid Dynamics
CHP Combined Heat and Power
CI Confidence interval
CM C Conditional Moment Closure
d Scaling parameter for variable x, variable dependent
det Determinant of a matrix
D Diffusion coefficient, m2 s-1
D Matrix of scaling parameters, variable dependent
DM Mahalanobis distance
Da Damkohler number
DN S Direct Numerical Simulation
DO Discrete Ordinate Radiation Model
E True error
E Expectation operator
Ea Activation energy, J kmol-1
xix
Nomenclature
E
e Estimated error
EBU Eddy Break-up Model
EDM Eddy Dissipation Model
ED/F R Eddy Dissipation Finite Rate Model
f Volume force, m s-2
f Mixture fraction
F Model solution of the SRQ
Fij View factor between shields i and j
Fs Safety factor for the estimation of the solution
uncertainty, Usver
F0 Exact value of the SRQ
F AN S Favre Averaged Navier Stokes
F P CA Mixture Fraction Principal Components Analysis
FR Finite Rate Model
GCI Grid Convergence Index
GP CA Global Principal Components Analysis
GRE Global reconstruction error
GSRE Global scaled reconstruction error
h Specific enthalpy, m2 s-2
h Grid spacing
ht Total specific enthalpy, m2 s-2
h∗ Height of the cylindrically shaped fine structure, m
HIT AC High Temperature Air Combustion
I Identity matrix
IC Initial Conditions
IF CM Infinitely Fast Chemistry Model
JHC Jet issuing in a Hot Co-flow
k Kinetic energy, m2 s-2
Keq Equilibrium constant
kf Forward kinetic constant, reaction dependent
kr Backward kinetic constant, reaction dependent
kR Recirculation degree
kv Dilution factor
KinP P Kinetic post-processing
lk kth largest eigenvalue of the sample covariance
matrix S
l Mean value of the eigenvalues of the sample
covariance matrix S
L Length, m
L Diagonal matrix containing the eigenvalues of S in
descending order
Le Lewis number
LEM Linear Eddy Model
LES Large Eddy Simulation
xx
Nomenclature
LP CA Local Principal Components Analysis

ṁ Mass flow rate, kg s-1
ṁ∗ Specific mass flow rate across the fine structures, s-1
M Symbol for chemical species
Ṁ Mass flow across the fine structures, kg s-1
M ILD Moderate and Intense Low Oxygen Dilution
MMS Method of Manufactured Solutions
n Number of experimental replicates for the definition
of CIs
ne Total number of chemical elements
N Total number of species
NG Natural gas
ODT One Dimensional Turbulence
p Pressure, bar
p Convergence order
p Number of random variables of X
P AH Polycyclic Aromatic Hydrocarbons
PC Principal Component
P CA Principal Components Analysis
P CC Principal Component Classifier
P DE Partial Differential Equation
P DF Probability Density Function
PF Principal features
Pr Prandtl number
P rt Turbulent Prandtl number
PV Principal variables
q Energy flux, J s-1 s-2
q Energy dissipation in the energy cascade model, m2
s-3
Q Flow field property
Qj Rate of progress of reaction j, kg m-3 s-1
Q̇ Heat source, W m-3
0
Q Property fluctuation
00
Q Favre property fluctuation
Q Property mean value
Qe Favre property mean value
r Radial coordinate, m
rh Grid refinement ratio
R2 Correlation coefficient
RA Aerodynamic recirculation ratio
RAN S Reynolds Averaged Navier Stokes
Re Reynolds number
RHS Right hand side
xxi
Nomenclature
RSM Reynolds Stress Model

RT E Radiative Transfer Equation
s Sample standard deviation
S Sample covariance matrix
S 22,1 Partial covariance matrix for X (2) given X (1)
Sc Schmidt number
Sct Turbulent Schmidt number
SLF M Steady Laminar Flamelet Model
SRQ System Response Quantity
SV ER Solution Verification
t Time, s
t 2,α ,ν 1 − α2 quantile of a t-distribution with ν degrees of
freedom
tq Total variance explained by a q-dimensional subset of
PCs
tq,j Variance accounted for each variable by a
q-dimensional subset of PCs
tr Trace of a matrix
T Temperature, K
T Standardized random variable
TNF Turbulent Non-premixed Flames
u Velocity, m s-1
Usver Solution verification uncertainty
U DF User Defined Function
U HC Unburned Hydrocarbon
V Diffusion velocity, m s-1
V Volume, m3
V Transposed eigenvector matrix A
VM AX Quantity maximized by the VARIMAX criterion
V &V Verification and Validation
V AST Variable Stability Scaling
V QP CA Vector Quantization Principal Components Analysis
w Production of turbulent kinetic energy, m2 s-2
W Molecular weight, kg kmol-1
W SGG Weighted Sum of Gray Gases Model
x Axial coordinate, m
x Observed mean for random variable x
x Random variable
x
e Scaled random variable
X Mole fraction
X Sample matrix
X (1) , X (2) Subsets of X
X Vector of sample means
xxii
Nomenclature
X
f Scaled sample matrix
ye Sample mean based on n experiments
ym Value of the SRQ given by the computational model
Y Species mass fraction
zk kth PC of S (or Σ)
Z Matrix of PC scores
Ẑ Approximation of Z based on a subset of m variables
of X
Greek symbols
α Constant in the Richardson extrapolation
β Coupling function for the definition of the mixture
fraction
β Temperature exponent
γ∗ Fine structure mass fraction
δ Eigenvalues of S 22,1
δij Kronecker symbol
δRE Numerical error estimate
Turbulent energy dissipation rate, m2 s-3
Error from the PCA reconstruction of X
air Air excess
λ Thermal conductivity, W m-1 K-1
λ Lagrange multiplier
λk kth largest eigenvalue of S (or Σ)
µ Dynamic viscosity, kg m-1 s-1
µ True mean
µt Turbulent viscosity, kg m-1 s-1
µ Mean of the infinite population X
ν Degrees of freedom of the t-distribution
ρ Density, kg m-3
ρ Canonical correlations between X (1) and X (2)
σ Boltzmann constant, J K-1
σj2 Residual variance in predicting xj from Z
Σ Covariance matrix of the infinite population X
τ Characteristic time, s
τ∗ Residence time within the fine structures, s
τij Viscous force tensor, kg m-1 s-2
υ Kinematic viscosity, m2 s-1
υ Net mass stoichiometric coefficient
0
υ Molar stoichiometric coefficient for forward reaction
xxiii
Nomenclature
00
υ Molar stoichiometric coefficient for backward reaction
φ Lagrange multiplier
φ Optimal rotation angle determined with the
VARIMAX criterion
χ Scalar dissipation rate
χ1 Probability of coexistence of reactants for the
definition of χf s
χ2 Degree of heating for the definition of χf s
χ3 Lack of reactants for the definition of χf s
χf s Reacting fraction of the fine structures, χf s = χ1 χ2 χ3
ω Strain rate, s-1
ω̇k Reaction rate for species k, kg m-3 s-1
Subscripts
1 Fuel stream
2 Oxidizer stream
A Air
c Chemical
e Chemical element
em Emitter (referred to the radiant tube surface)
E Exhaust gases
F Fuel
i Direction index
i Observation index
ins Insulation (referred to the insulation shields)
in Incoming
j Direction index
j Reaction index
j Variable index
k Species index
m Number of retained variables
O Oxidizer
P Products
q Dimensionality of the truncated set of PCs
st Stoichiometric conditions
t Turbulent
u Unburnt status
xxiv
Nomenclature
Superscripts
0
Transpose operator
0 00
, , . . . ,n . . . ,∗ Levels in the energy cascade model
∗ Fine structure properties
1, 2, . . . , n Experiment replications for the definition of CIs
k Cluster index
xxv
Nomenclature
xxvi
Chapter 1
Advanced systems for

hydrogen-based fuel combustion
Hydrogen has recently received special attention in the scientific community due
to its promising role as future energy carrier. Hydrogen can be relatively easily
stored and its utilization does not cause any local emission of CO2 . Moreover,
hydrogen combustion is relevant in many applications, including fuel cells after-
burners, power production with CO2 sequestration and thermal conversion of
CO2 neutral fuels such as biomasses and wastes.
However, the use of hydrogen as a fuel is not straightforward due to some
specific properties which strongly deviate from those characterizing traditional
fuels such as methane and which make the existing combustion equipment un-
suited (Figure 1.1). Hydrogen has a high molecular diffusivity, a wide flamma-
bility range, a high laminar flame speed and a high energy density per unit
mass. Therefore, hydrogen combustion could lead to very large NOx emissions,
when operating with conventional diffusion burners. On the other hand, NOx
emissions could be controlled by means of lean and partially premixed combus-
tion. However, this may lead to stability problems and flashback phenomena.
A major technological challenge related to hydrogen utilization is, then,
to work with high hydrogen contents in the fuel ensuring, at the same time,
acceptable levels of NOx emissions. This requires the re-conceptualization of
the burner design with two main objectives: i) to upgrade conventional burners
and combustion chambers for H2 -enriched fuels and ii) to design and develop
new devices for pure H2 combustion.
Computational Fluid Dynamics (CFD) represents an established tool for
the design and development of combustion technologies. Efforts need to be
spent to make such tool as reliable as possible, through a conscious choice of
validated models and sub-models, e.g. turbulence and combustion models, ki-
netic mechanisms [1]. The present Chapter will firstly focus on a brief review of
the main mechanisms responsible for the formation of NOx in flames. Partic-
ular emphases will be devoted to homogeneous hydrocarbon-hydrogen flames;
1
Chapter 1. Advanced systems for hydrogen-based fuel combustion
Figure 1.1: Combustion of hydrogen containing fuels: issues and solutions.
however, some references regarding solid fuel combustion, i.e. coal combustion,
will be provided. Then, comprehensive modeling approaches for NOx forma-
tion will be discussed. Finally, advanced technologies adopted to mitigate NOx
formation during the combustion process and to control emissions in the post
combustion phase will be presented. In particular, attention will be devoted
to a novel combustion regime, known as flameless combustion [2] (or MILD1
[3] or HITAC2 [4]. Such combustion regime appears particularly promising for
hydrogen-based fuels, also because it overcomes the paradigm of the conflict
between energy saving, usually achieved with an increase in energy efficiency
via air preheating, and NO reduction.
1.1 NOx formation and control in combustion pro-

cesses
Modeling NOx formation in turbulent reacting systems requires the description
of several, inherently coupled, physical processes, including the general fluid
dynamic, the local mixing process, heat transfer and chemical kinetics. The
literature on nitrogen pollutant is extraordinary rich of kinetic mechanisms for
the description of the net rate of NOx formation in different types of flames.
However, it is still unfeasible to directly couple very large kinetic mechanisms
with the turbulent mixing process into a CFD code. Moreover, it is recognized
that nitrogen pollutant chemistry barely affects the flame structure, which is
dominated by the fast fuel-oxidizer reactions. Therefore, NOx formation models
are often decoupled from the generalized combustion models and they are gen-
1
Moderate and Intense Low Oxygen Dilution
2
High Temperature Air Combustion
2
1.1. NOx formation and control in combustion processes
erally executed after the flame structure has developed. This approach is very
computationally efficient and less complicated than solving the transport equa-
tions for the pollutants and the reacting species jointly. It should be also noted
that the pollutant formation sub-models typically converge in a small fraction
(typically around 10%) of the CPU time required to solve the reacting species
equation, thus allowing to investigate several NOx formation mechanisms on
pre-calculated flame structures.
The arguments provided above regarding the poor effect of NOx chemistry
on the overall temperature and flow fields have prompted the development
of post-processing tools of CFD results which incorporate very large kinetic
mechanism for the prediction of pollutant emissions, i.e. NOx , CO, UHC3 and
PAH4 [5, 6, 7, 8, 9]. According to such approach, the flow and temperature
field are taken from the CFD simulation and a detailed kinetic calculation is
carried out on a reactor network built from the CFD grid using criteria based
on the similarities of local temperature and species concentration.
1.1.1 Homogeneous NOx formation mechanisms

Traditionally, the description of nitrogen oxides formation in gas-phase takes
into account the following three main routes:
• Thermal NO (Zeldovich mechanism)
• Prompt NO
• Fuel NO
However, it has been widely recognized in the literature that homogeneous NO
formation may occur with alternative routes:
• N2 O intermediate NO mechanism
• NNH route
These formation mechanisms can become relevant and contribute equally or
even more than the main routes mentioned above, when operating in non con-
ventional conditions. In particular, technologies promoting low-temperature
combustion and employing hydrogen enriched fuels may be characterized by
very large contributions of the N2 O and NNH routes. For example, combustion
equipment implementing flameless combustion typically operate at low tem-
peratures, thus resulting in NOx emissions which cannot be explained only by
the thermal and prompt kinetic mechanisms, because of the significant role
played by the N2 O and NNH intermediates. The latter may become dominant,
when hydrogen is added to the fuel, to stabilize lean hydrocarbon flames and
to reduce greenhouse gas emissions.
3
Unburned Hydrocarbon
4
Polycyclic Aromatic Hydrocarbons
3
NO formation via NNH formation was first proposed by Bozzelli and Dean
[10], who deduced a large kinetic constant for the reaction employing the di-
rect oxidation of NNH to NO. Such result was confirmed by Hayhurst and
Hutchinson [11], who performed measurements of NOx in a laminar, premixed,
flat CH4 /H2 . Results showed NOx concentration too high to be explained en-
tirely by the Zeldovich mechanism. The inclusion of the NNH route proposed
by Bozzelli and Dean [10] gave reasonable good agreement with experimental
observations.
Konnov et al. [12] proposed an updated detailed H/N/O mechanism for the
analysis of NO formation during hydrogen combustion in well stirred reactors.
Numerical results were compared to experiments carried out in the temperature
range 1500-2200 K. The authors observed that the NNH route represents the
dominant source of NO at 1500 K, not only in rich conditions [13], but also
in lean mixtures and at stoichiometric conditions. They observed that, at
stoichiometric and rich conditions, thermal NO formation became predominant
to the NNH route later than in lean conditions, when increasing the operating
temperature. This is easily explained considering that NO formation via NNH
is proportional to H radicals concentration, which is higher in stoichiometric
and rich mixtures. Finally, a possible new route for NO formation in rich
mixture and at low temperature via N2 H3 radicals was identified.
Rørtveit et al. [14] investigated the effect of diluent addition on NOx forma-
tion in hydrogen laminar counter-flow flames. They observed that the addition
of CO2 and He to the fuel stream resulted in significantly lower temperature
peaks and NOx with respect to the diluent N2 . The authors also found sig-
nificant contributions of the N2 O and NNH mechanisms on the overall NOx
production at low temperatures (∼1800 K). When raising the temperature up
to 2100 K, these mechanisms showed a very mild temperature dependency, dif-
ferently from the thermal mechanism, which increased 16 times in the 300 K
temperature range investigated, becoming the dominant contribution.
Guo et al. [15] studied the effect of hydrogen addition to ultra-lean counter-
flow CH4 /air premixed flames on the flammability limits and NOx emissions.
Results showed that the addition of hydrogen led to a substantial enlargement
of the flammable range and to an increase of the NO emissions, when operating
at constant equivalence ratios. With regards to the source of NO, the authors
found that the N2 O and NNH routes played a major role on NO formation,
while the thermal and prompt mechanisms resulted in minor contributions,
due to the lean conditions and the low temperatures.
Löffler et al. [13] recently developed a new simplified reaction scheme to
model NOx formation in natural gas combustion. The model was tested in
a wide range of conditions and compared to more detailed kinetic schemes
[16] under premixed plug-flow reactor conditions. Results confirmed that NO
formation is dominated by the thermal mechanism at temperatures higher than
1800 K. At lower temperatures and with increasing oxygen availability, the
N2 O mechanism gave significant contribution, due to the increased O radical
4
concentration. On the contrary, NNH dominated below 1600 K and in fuel rich
conditions, when the concentration of H radicals is higher.
Skottene and Rian [17] compared different detailed mechanisms for the pre-
diction of NOx formation in laminar hydrogen flames and turbulent hydrogen
jet flames. They considered the scheme by Glarborg et al. [16], the San Diego
mechanism [18], and two other mechanisms obtained by replacing the H2 /O2
chemistry of the first two schemes with the kinetic model proposed by Li et al.
[19]. Results showed the important role of the NNH pathway in NO forma-
tion for all investigated flames. The performances of the different models were
found to be strongly dependent on their ability to properly capture the NNH
route. Therefore, the main drawback of the San Diego mechanism was related
to the omission of an explicit pathway converting NNH to NO. On the other
hand, the mechanism by Glarborg et al. [16] was found to perform satisfactory
for the turbulent jet flame, while its modified version with Li et al. [19] H2 /O2
chemistry gave the best prediction for the diffusion flames, although resulting
in a slight overprediction of the NO levels.
1.1.1.1 Thermal NO mechanism

Thermal NO is formed from the oxidation of atmospheric nitrogen at relatively
high temperatures in fuel-lean conditions. The process is described by the
so-called Zeldovich two-step mechanism:
kf,1
−−
N2 + O )−*
− NO + N (1.1)
kr,1
kf,2
−−
N + O2 )−*
− N O + O. (1.2)
kr,2
An additional elementary reaction is added to the thermal mechanism when

operating in fuel-rich conditions to obtain the so called extended Zeldovich
mechanism:
kf,3
−−
N + OH )−*
− N O + H. (1.3)
kr,3
The extended Zeldovich mechanism includes the effects of O, H and OH

radicals. Typically, their concentration is obtained from equilibrium consid-
erations; however, this can result in a significant underestimation of the NO
formation rates. Better agreement can be usually achieved if the radicals con-
centration is estimated by means of detailed kinetic mechanisms for the gas
phase oxidation [20]. The rate parameters for reactions (1.1)-(1.3) are listed in
Table 1.1.
Reaction (1.1) is usually considered the rate limiting step of the thermal
formation process due to its high activation energy, which determines the high
temperature sensitivity of the mechanism. Assuming the N radical to be in
quasi steady state, the overall rate of NO formation is given by:
5
Table 1.1: Reaction parameters for the extended Zeldovich mechanism and for the
NO2 intermediate NO mechanism. Units: mol, cm, s, K.
Reaction expressiona Forward Backward

A β Ea /R A β Ea /R
kr,1
−−
N2 + O )−*
− NO + N 1.71 · 1014 0 38367 3.30 · 1012 0.3 0
kf,1
kr,2
N + O2 −
)−
−*
− NO + O 6.40 · 109 1 3160 4.92 · 1012 0 20399
kf,2
kr,3
N2 + O −
)−
−*
− NO + N 3.80 · 1013 0 0 1.10 · 1012 0 23911
kf,3
kr,4
N2 + O + M −
)−
−*
− N2 O + M 4.13 · 1013 0 7890 4.00 · 1014 0 28200
kf,4
kr,5
−−
N2 O + O )−*
− NO + NO 6.60 · 1013 0 13390 1.64 · 1012 0 32057
kf,5
kr,4
N2 O + O −
)−
−*
− N2 + O2 1.00 · 1014 0 14080 5.26 · 1013 0 54745
kf,4
a k = AT β exp (−E/RT ).
kr,1 kr,2 [N O]2

 
d [N O]  kf,1 [N2 ] − 
kf,2 [O2 ]
= 2 [O] kr,1 [N O]
. (1.4)
dt  1+ 
kf,2 [O2 ]+kf,3 [OH]
Assuming that the initial concentrations of NO and OH are small, only the
forward rates of the Zeldovich mechanism are significant and Equation (1.4)
becomes:
d [N O]
= 2kf,1 [O] [N2 ] . (1.5)
dt
Equations (1.4) and (1.5) are coupled to the fuel oxidation process through
competition of the O and OH radicals. Their local concentration can be esti-
mated if a comprehensive kinetic model is adopted during the CFD calculation.
Otherwise, estimation procedures based on equilibrium and partial-equilibrium
approaches are taken into account. Following Westenberg [21], oxygen atoms
can be assumed to be in equilibrium with O2 in fuel lean zones, where CO is
oxidized to CO. Therefore:
1
[O] = Keq [O2 ] 2 . (1.6)
An improvement to the equilibrium approach is obtained by accounting for

third-body reactions in the O2 dissociation-recombination process [22]:
6
−
O2 + M *
−O+O+M
)
which generally leads to higher O radical concentrations. In case the OH radical
is not negligible with respect to the O radical, i.e. fuel-rich conditions, the local
concentration of OH can be deduced from a comprehensive kinetic model or
from a partial equilibrium assumption [23].
At mean temperatures below 1600-1800 K, thermal NO formation via the
Zeldovich scheme is significantly reduced and it does not represent a major
source of NO. Other routes have been proposed to extend the thermal mecha-
nism described above, in order to describe the observed NO emissions from real
systems. In particular, the so called N2 O and NNH routes will be presented
briefly.
1.1.1.2 N2 O intermediate NO mechanism

Malte and Pratt [24] first proposed the formation of NO via the intermediate
species N2 O. The mechanism can contribute as much as 90% of the total NO
formed when operating at temperatures below 1800 K, elevated pressures and
fuel-lean conditions. Therefore, this mechanism can be particularly relevant
for gas turbine conditions. The N2 O intermediate route is described by the
reactions:
kf,4
−−
N2 + O + M )−*
− N2 O + M (1.7)
kr,4
kf,5
−−
N2 O + O )−*
− NO + NO (1.8)
kr,5
kf,6
−−
N2 O + O )−*
− N2 + O2 . (1.9)
kr,6
The rate constants for Reactions (1.7)-(1.9) are listed in Table 1.1.
1.1.1.3 NNH intermediate NO mechanism

Bozzelli and Dean [10] proposed an alternative route for NO formation via NNH
radicals. The mechanism can be particularly relevant at low temperatures, in
fuel rich conditions and for hydrogen enriched flames. The essential reactions
of the NNH route are:
kr,7
−−
N2 + H )−*
− NNH (1.10)
kf,7
kr,8
NNH + O −
)−
−*
− N O + N H. (1.11)
kf,8
The NNH route is also linked to the N2 O route via the reaction:
7
kr,9
−−
NNH + O )−*
− N2 O + H. (1.12)
kf,9
Assuming that the NH radicals formed in reaction (1.11) are completely

converted to NO and that reaction (1.10) is equilibrated, it is possible to derive
a rate expression for NO formation via NNH radicals as follows:
d [N O]N N H
= 2kf,8 Kp,7 [N2 ] [O] XH (1.13)
dt
where Kp,7 is the equilibrium constant of reaction (1.10) expressed in partial
pressures and XH is the mole fraction of H radicals. Hayhurst and Hutchinson
[11] estimated the product kf,8 Kp,7 in the range between 1800-2500 K, using
experimental measurements of temperature and OH radical along the axis of
laminar, premixed, flat CH4 /air and H2 /air flames. The NNH contribution
was estimated as the NO exceeding the Zeldovich mechanism, leading to the
following value for the product kf,8 Kp,7 :
kf,78 Kp,7 = 2.3 · 10−15 exp (−2760/T ) cm3 /molecules · s−1 . (1.14)
Konnov et al. [25] extended the temperature range for kf,8 Kp,7 from 1800-
2500 K [11] to 1400-2500 K, by comparing experiments on the lean combustion
of hydrogen and air in stirred reactors with kinetic modeling, carried out with
different mechanisms. The resulting value for kf,8 Kp,7 is:
kf,8 Kp,7 = 2.5 · 10−15 exp (−3600/T ) cm3 /molecules · s−1 . (1.15)
1.1.1.4 Prompt NO mechanism

Prompt NO is formed by the reaction of atmospheric nitrogen with hydrocar-
bon radicals in fuel rich regions of the flame with subsequent oxidation of the
intermediate species to NO. The mechanism was first reported by Fenimore [26]
and it can be described by the following sequence of reactions:
−
N2 + CHx *
− HCN + N + . . .
) (1.16)
−
*
− 2CN
N2 + C2 ) (1.17)
−
*
− N O + H.
N + OH ) (1.18)
The mechanism requires a hydrocarbon to initiate the reaction with nitro-
gen; therefore, it is particularly relevant in rich conditions or in rich regions of
the flame, at temperatures below 1800 K. The detailed prompt mechanism is
8
usually neglected in most NO models for CFD calculations due to the increased
complexity of nitrogen chemistry and its intimate coupling with the fuel oxida-
tion chemistry. Instead, simplified approaches are usually taken into account.
De Soete [27] suggested the following one-step kinetic mechanism:
d [N O]prompt
= kprompt [O2 ]a [N2 ] [F ] (1.19)
dt
where

TA,prompt
kprompt = Aprompt exp − . (1.20)
T
In Eq. (1.20) a is the oxygen reaction order and F denotes the fuel. The kinetic
parameters are fuel dependent. De Soete [27] suggested the following values for
acetylene fuel:
TA,prompt = 30215 K Aprompt = 1.2 · 106 s−1 . (1.21)
As for the oxygen reaction order, it is uniquely related to oxygen mole fraction
in the flame [27]:



1.0 XO2 ≤ 4.1 · 10−3
4.1 · 10−3 ≤ XO2 ≤ 1.11 · 10−2

−3.95 − 0.9 ln X
O2
a= . (1.22)
−0.35 − 0.1 ln XO2

 1.11 · 10−2 ≤ XO2 ≤ 0.03

0 XO2 ≥ 0.03
Eq. (1.21) was proposed for acetylene. Therefore, it may cause errors if
used for other hydrocarbons, especially in fuel-rich conditions. The commercial
CFD package FLUENT® employs a modified version of the model by De Soete
[27], fitted using experimental data available from Backmier et al. [28]:
d [N O]prompt 0
= f kprompt [O2 ]a [N2 ] [F ] (1.23)
dt
with f being a correction factor, equal to:
f = 4.75 + 0.0819n − 23.2φ + 32φ2 − 12.2φ3 . (1.24)
In Eq. (1.24), n is the number of carbon atoms per molecule for the hydrocarbon
fuel and φ is the equivalence ratio. The correction factor is valid for equivalence
0
ratios between 0.6 and 1.6. The values proposed for kprompt are:
0
TA,prompt = 36510 K Aprompt = 6.4 · 106 s−1 . (1.25)
9
1.1.1.5 Fuel NO mechanism

Nitrogen-containing organic compounds can contribute to the total NO formed
during the combustion process through the fuel mechanism. Fuel NO is formed
from nitrogen bound in the fuel, via intermediate species such as HCN and NH3 ,
which are oxidized to NO while being competitively reduced to N2 according
to the overall reactions:
−
HCN/N H3 + O2 *
− NO + . . .
) (1.26)
−
N O + HCN/N H3 *
− N2 + . . .
) (1.27)
Fuel NO is a relevant source of NO for residual fuel oil and coal, which
typically contain 0.3-2% nitrogen by weight. On the other hand, natural gas
and hydrogen combustion is not interested by such formation mechanism.
1.1.2 NOx control in combustion

Nitrogen oxides control is a main goal for Researchers and Engineers. NOx
represent a significant threat for the environment and combustion systems are
a major source of such pollutants. Many efforts have been spent during the
last decades to improve the understanding of NOx formation mechanisms and,
then, to develop effective NOx control technologies. In general, when refer-
ring to NOx control, it is possible to distinguish between post-combustion NOx
clean up and modification of the combustion process. Typical post-flame tech-
niques include selective catalytic reduction (SCR) and selective non-catalytic
reduction (SNCR). In SCR ammonia is injected near the combustor exit, al-
lowing NOx reductions by as much as 90%. SCR is very effective in a broad
range of applications but it can be relatively expensive, due to the high cost
of the catalyst. With SNCR, ammonia is injected in a higher temperature re-
gion (1173-1373 K), to ensure the effective action of NH3 . SNCR is a viable
option only in a narrow temperature range, which precludes its use in many
applications.
The suppression of NOx formation during combustion can be accomplished
with different approaches, often used in combination with each other. The most
common are burner design, fuel staging, flame cooling, ultra-lean combustion
and flue-gas recirculation. All these approaches share the same objective of
suppressing the temperature peaks and reducing the residence time and the
oxygen concentration in the high-temperature regions. Low-NOx burners for
coal combustion accomplish this with swirling and an appropriate combination
of secondary and tertiary air, to generate a fuel rich region where NOx can
be reduced to N2 . Staging is often used to create a fuel rich region near the
burner through the staged introduction of fuel, fuel staging, and oxidizer, air
staging. Staging can result in significant NOx suppression, especially for hydro-
gen combustion, due to the absence of prompt mechanism for such flames [29].
10
1.2. Flameless combustion
However, the implementation costs of staging are usually very high, resulting
in lower overall performances. Flame cooling adopts steam injection in the
combustion chamber to reduce the temperature levels, thus allowing to reduce
NOx formation but also determining significant CO emissions in the exhausts.
Ultra-lean combustion can be effectively exploited to reduce NOx emissions due
to the reduced temperature peaks. However, careful attention must be paid to
the flame stability, to avoid extinction and ensure safety of operation. As men-
tioned before, the addition of hydrogen to these flames may be of interest to
enlarge the flammability limits [15, 14]. Mastorakos et al. [30] observed that
the techniques involving flue-gas recirculation are the most effective in accom-
plishing NOx reduction. For sake of clarity, it is worth distinguishing between
external and internal recirculation. The first option is accomplished by recy-
cling part of the exhaust gases to the inlet section, to dilute the fresh reactants.
On the other hand, internal recirculation is achieved through a special aero-
dynamic design of the burner, based on the entrainment of the exhaust gases
driven by the high momentum of the fresh reactants.
Flameless combustion is based on the internal recirculation of exhaust gases.
The investigation of advanced combustion systems operating in flameless com-
bustion has represented the main objective for the present Thesis, as they show
great potentials, not only for the efficient and environmental friendly oxidation
of conventional fuels, but also for the exploitation of more complex ones, such
as hydrocarbon-hydrogen mixtures and low-calorific value fuels enriched with
H2 . The theory and applications of flameless combustion are discussed in the
following Sections.
1.2 Flameless combustion

Flameless combustion couples high combustion efficiencies with very low pol-
lutant emissions [2, 4, 3]. Such a combustion regime needs the reactants to
be preheated above the self-ignition temperature and enough inert combustion
products to be entrained in the reaction region. The former requirement en-
sures high thermal efficiency, whereas the latter allows diluting the flame and
reducing the final temperature well below the adiabatic flame temperature.
As a result, a flame front is no longer identifiable, thus the name flameless,
and the combustion process is no longer restricted to the flame front region
but extended to a larger portion of the combustion chamber. The system ap-
proaches perfectly stirred reactor (PSR) conditions and it is characterized by
a more uniform temperature field than traditional combustion systems. By
avoiding temperature peaks, NOx formation is largely reduced and the effect
on the materials is beneficial. Soot formation is also suppressed, due to the
lean conditions and the large CO2 concentration in the exhausts [3]. From
the operation point of view, it has been observed that ignition and extinction
phenomena do not occur in flameless combustion because of the small temper-
11
ature difference between burnt and unburnt gases. Flame stabilization occurs
naturally as the reactants temperature exceeds the self-ignition temperature.
Therefore, a large degree of freedom in the choice of the fluid dynamical con-
figuration of the combustion chamber is allowed. Several studies have been
devoted to understanding the operating principles of flameless combustion [31]
as well as its mechanisms and critical parameters [32]. An extensive review on
flameless combustion features considering physical, chemical, and thermody-
namic aspects has been provided by Cavaliere and de Joannon [3]. Gupta [4]
also provides a detailed review on HITAC, describing benefits (energy savings,
CO2 and NOx reduction, equipment size reduction) and main features of the
technology.
However, flameless combustion appears to be still worthy of further inves-
tigations and attention. In particular, from the modeling perspective, it is still
not clear how to treat turbulence-chemistry interactions in such regime and
to which extent the kinetic mechanisms adopted need to be detailed. More-
over, the uncertainly of these topics is increased by the wide range of solutions
adopted to achieve flameless operations, through the competition of turbulence
mixing and chemical kinetics.
In the following Section a brief description of the operating principles of
flameless combustion is presented, followed by an exhaustive review of the rele-
vant literature in the field. In particular, emphasis will be put on the literature
regarding the flameless combustion of hydrogen-based fuels.
1.2.1 Operating principles of flameless combustion

In contrast to stabilized flame combustion, flameless oxidation is mixture and
temperature controlled, and it is achieved with specific flow and temperature
conditions. A prerequisite for a stable flame in traditional combustion systems
is represented by the balance between flow velocity and flame velocity. Creating
flow conditions for flame stabilization is an essential burner design criterion,
for both premixed and diffusion flames. Usually, swirl or bluff body are used
to create stagnation points or areas of low velocity for stabilization. Internal
exhaust gas recirculation is sometimes used to reduce NOx formation. Following
Wünning and Wünning [2], it is possible to quantify the amount of exhausts
recirculated in the reaction zone as:
ṁE
kR = (1.28)
ṁA + m˙F
where ṁE is the net mass flow rate of recirculated flue gas, whereas ṁF and
ṁA are the fuel and air mass flow rates, respectively. It has been shown [33]
that it is not possible to achieve flammable mixtures of hydrocarbon and air
for values of kR ≥ 0.5 (Figure 1.2). Above this value, extinction occurs, due to
the lower oxygen concentration and higher inerts in the mixture.
However, it was found [34] that a stable form of combustion is also possible
12
Figure 1.2: Flammable range for methane-air mixture as a function of the

recirculation degree, kv [33].
for much higher recirculation rates. Under ideal conditions, such combustion
process takes place without any visible or audible flame. Figure 1.3 shows
a stability diagram for the combustion of methane in air, as a function of
the recirculation degree, kR , and the operating temperature. Three different
combustion modes or regimes are identified: stable flames, unstable flames
and flameless oxidation. Burner stabilized flames (A) are possible over the
entire range of operating temperatures, but only for recirculation degrees up
to 30%. For higher recirculation values of kR , the flame becomes unstable
(B), lifts off and finally blow out, for temperatures below self ignition. If the
furnace temperature is sufficiently high, the fuel can react in a steady, stable
form of flameless oxidation (C), without limitations of the recirculation degree.
As it can be deduced from Figure 1.3, it is not possible to operate a burner
in flameless mode in a cold combustion chamber. Therefore, the combustion
chamber must be heated up in conventional mode and switched to flameless
oxidation only when the operating temperature exceeds the fuel self-ignition
temperature.
An idealized operating scheme for flameless oxidation is shown in Figure
1.4. Combustion air is mixed with the exhaust gases (region I); then, fuel is
added in region II and combustion takes place. The maximum temperature rise
due to the oxidation process can be effectively reduced to few hundred Kelvin, if
enough product gases are recirculated in the reaction zone. In region III, energy
is withdrawn from the combustion products to control the temperature level in
the combustion chamber. However, the temperature has to be kept at a level
high enough to guarantee reaction in region II. To increase energy efficiency,
the gases leaving the system could be used for air-preheating, although this is
not compelling to ensure the flameless oxidation process.
The technical realization of the idealized process in Figure 1.4 can be ac-
complished by means of a high velocity burner (Figure 1.5). The burner is
provided with two different air supplies, which allow to operate with a burner
13
Figure 1.3: Stability diagram for methane-air combustion, as a function of the

recirculation degree and operating temperature [2].
Figure 1.4: Idealized flameless oxidation process [2].
14
Figure 1.5: High velocity burner for operation in flame and flameless regime.
stabilized flame (flame mode) or in flameless oxidation mode (flameless mode).

In flame mode, the fuel is fed to the primary combustion chamber (5) through
the fuel supply (1) and the gas nozzle (4). Air is also supplied to the same
chamber, through nozzle (2). The burner operates like a conventional high ve-
locity burner with a spark plug ignition and a stabilized flame. At the end of
the heat up cycle, when the operating temperature exceeds the auto-ignition
temperature of the fuel (∼1100 K), the valve connected to the air supply (2)
is closed and air is fed through inlet (3) and the nozzles (6), concentric to
the combustion chamber (5). The air jet (A) acts like an ejector, causing the
entrainment of the exhaust gases from the surrounding (D). This causes the
reduction of oxygen concentration and the inertization of the reacting mixture,
whose temperature becomes close to that of the combustion products. The
oxidation process takes place in (C), when the fuel jet (B) meets the diluted
air stream. As a result, the temperature rise due to combustion is limited and
combustion takes place without a visible flame, as the entire volume containing
the reacting mixture is at conditions which can sustain the oxidation process.
Thus, the system approaches the conditions of a perfectly stirred reactor.
The separation of the air and fuel stream shown in Figure 1.5 is adopted
to ensure that reaction occurs only after mixing with the combustion products
takes place. However, it can be shown [2, 31] that it is possible to achieve
flameless combustion also in premixed conditions. Moreover, when operating in
very lean conditions, the heat removal process described in Figure 1.4 could be
avoided, making the flameless oxidation process feasible for adiabatic systems.
1.2.2 Review of the relevant literature in flameless combustion

The main advantage of flameless combustion is the great potential of the
technology for the reduction of NO emissions. The absence of temperature
peaks reduces NO formation through the thermal mechanism even with large
fuel/oxidizer preheating. This was pointed out by Wünning and Wünning [2],
using an experimental apparatus for temperature and flow field measurements
in flameless and flame regimes (Figure 1.6). Recuperative burners (Figure 1.7)
were mounted at the bottom of the chamber, firing vertically into a cylindrical,
air cooled chamber. A non-cooled probe was inserted from the top and it was
equipped to measure temperature, pressure and species compositions.
15
Figure 1.6: Experimental apparatus used by Wünning and Wünning [2].
Figure 1.7: FLOX® recuperative burner [2].
16
Figure 1.8: Time resolved temperature measurement for flame, unstable and
flameless regime [2].
Figure 1.8 shows time resolved temperature measurements in the burner op-
erated in flame, lifted flame and flameless mode. It can be observed how both
flame and flameless oxidation modes show steady temperature conditions. NO
and noise are substantially higher in flame mode, compared to flameless mode.
NO and noise are in-between for the lifted flame but the temperature signal is
highly unstable. Moreover, Wünning and Wünning [2] supported experimen-
tal investigation with numerical simulation finding good agreement between
measured and calculated temperature and velocity profiles, thus showing the
suitability of CFD as predicting tool for flameless combustion.
The first experimental evidence regarding the structure of the combustion
zone in flameless combustion was provided by Plessing et al. [35], who mea-
sured temperature and species concentration with optical techniques in flame
and flameless conditions. Figure 1.9 shows a comparison of temperature and
OH radicals concentration between the two cases. It can be observed that the
maximum temperatures decrease, from 1900 to 1650 °C, when switching from
flame to flameless mode. In addition OH radicals are present in lower concentra-
tion (below 10 ppm) in flameless regime, thus reducing NO prompt formation.
Cavigiolo et al. [31] investigated the effect of the recirculation degree on the
transition from flame to flameless combustion regimes in a pilot burner fed with
methane and ethane. The transition was found to be a critical phenomenon,
showing a drastic reduction of NO emission from values of 500 ppm, typical of
conventional combustion equipment, to 10 ppm.
Flameless combustion appears very well suited for those industrial processes
which require high and homogeneous temperature distribution within the com-
bustion chamber (e.g. in glass and ceramic industry, steel thermal treatments).
For these processes, the energy recovery represents a primary issue to ensure
acceptable energy efficiency. Energy saving can be achieved with either recu-
perative or regenerative burners according to the exchange area and heating
17
Figure 1.9: Temperature and OH radical concentration in flame and flameless

regimes [35].
rate required. These burners are usually designed to operate both in flame and
flameless mode, by varying the reactants’ feeding mode. The combination of
recuperative burners with radiant tubes is a widely applied solution in many
thermal treatments of material surfaces, to avoid any contact or contamination
of the flue gases with the stock surface to be treated [36]. Choi and Kat-
suki [37] investigated the feasibility of flameless oxidation in industrial glass
furnaces. They found that combustion was sustained even with low calorific
value fuels and low oxygen concentrations, if the combustion air was preheated
above the fuel self-ignition temperature. Results also showed that NOx for-
mation was controlled by the mixing process between fuel and the preheated
air. Flamme [38] investigated the application of flameless combustion to glass
melting furnaces operating with process temperature around 1600°C and air
preheating temperatures up to 1350°C. Experimental results on a 300 kW fur-
nace showed that the replacement of conventional burners with recuperative
burners led to dramatic reduction of NOx emission, from 1500 ppm down to
values safely below 100 ppm. The application of flameless combustion in dif-
ferent fields than the thermal treatment processes is also very attractive. The
advantages of a clean and quite combustion process could be exploited by sev-
eral fields of application, including power generation, micro co-generation and
low-temperature applications. Flamme [39] showed the applicability of recu-
perative burners to gas turbine in order to overcome the oscillating problems
typical of lean premixed combustion occurring in gas turbines. Recently Wang
et al. [40] performed a technical, environmental and economic analysis of NOx
reduction technologies over a gas turbine power station. Results showed that
the use of recuperative burners could offer an effective method for reducing
18
NOx emissions, the technology being much cheaper than SCR and determining
only a slight increase in capital cost and electricity selling price with respect to
the simple cycle gas turbine power plants.
Given the industrial interest, an increasing attention has been paid not only
to the experimental characterization of flameless combustion burners, but also
to their modeling through Computational Fluid Dynamics (CFD). In particular
CFD may help optimizing burner performances by investigating geometrical
details, such as injection nozzle configurations and internal devices for flue gas
recirculation. Moreover, data on lab-scale burners are difficult to be transferred
to industrial burners operating in flameless combustion, as usual scaling criteria
are not adequate. Therefore, the support of CFD simulations becomes crucial
[41]. From the modeling side, an accurate description of turbulence-chemistry
interactions appears essential to capture the features of the oxidation process,
due the enhanced mixing levels and the diluted conditions which characterize
the combustion regime. To which extent such description needs to be detailed
still represents an open question, which the present Thesis will try to address.
Coelho and Peters [42] simulated a furnace operating in flameless combus-
tion mode by applying the flamelet approach to describe turbulence-chemistry
interactions. They compared predictions with measurements supplied by Pless-
ing et al. [35] and ? ]. Although a qualitative agreement between the experi-
mental and predicted flow field was observed, some considerable discrepancies
were present. In addition the residence time was underestimated by the nu-
merical model, and the flamelet approach was found to be unable of describing
correctly the NO formation. Better results could be achieved by means of
the unsteady flamelets [43]. Orsino et al. [44] investigated the performance of
different combustion models (Eddy Break-Up, Eddy Dissipation Concept with
chemical equilibrium and PDF/Mixture Fraction) for predicting the combustion
of natural gas with high-temperature air and large quantities of flue gas. They
found that all models failed in the vicinity of the fuel injection. However, the
authors claimed that they could obtain excellent NO predictions through the
steady flamelet library. Christo and Dally [45] pointed out difficulties arising
when modeling flameless combustion burners. They investigated numerically a
jet issuing in a hot diluted co-flow (JHC), replicating flameless conditions. The
burner was characterized experimentally by Dally et al. [46]. The authors used
different turbulence, combustion and kinetic models, comparing their perfor-
mances. In particular they showed that Mixture Fraction/Probability Density
Function and Flamelet approaches perform poorly for the flameless combustion
regime. The Eddy Dissipation Model also gave unsatisfactory results. Better
predictions were achieved through combustion models considering both chem-
istry and turbulence effects, such as the Eddy Dissipation Concept. In addition,
the aforementioned authors observed that molecular diffusion may play an im-
portant role in flameless combustion: for instance the predicted temperatures
were higher than the measured ones by approximately 10%, if differential diffu-
sion effects were neglected. It is worth noting that the general view in turbulent
19
flames modeling is to neglect the molecular diffusion term and to take into ac-
count only turbulent diffusion.
1.2.2.1 Flameless combustion of hydrogen-enriched fuels
Although it is recognized that stable flameless combustion is achievable with

pure hydrocarbons or methane-derived fuels, less information are available con-
cerning the sustainability of flameless combustion with hydrogen-enriched fuels
and pure hydrogen. Derudi et al. [47] experimentally investigated the sustain-
ability of flameless combustion of hydrogen containing fuels on a laboratory
scale combustor; in particular the influence of the hydrogen content in the fuel
was evaluated, using mixtures at 30% and 60% (by vol.) H2 and pure hydrogen.
The authors observed that the flameless combustion of hydrogen enriched mix-
tures required larger jet velocities (thus larger recirculation of exhaust gases)
with respect to methane; however, stable flameless conditions was obtained at
lower temperatures. Sabia et al. [48] investigated the effect of H2 addition (up
to 0.9% by vol.) on the flameless combustion of methane, by paying attention
to the dynamic features of the process. The small amounts of H2 regarded
its use as fuel enhancer, to improve the characteristics of poor mixtures. The
authors performed experiments in a jet stirred reactor, which were validated
further through numerical tools for the simulation of perfectly stirred reactor
conditions. The operating region of thermo-kinetics oscillations, i.e. instabili-
ties, was found to decrease with increasing H2 content; however, the importance
of irregular and double oscillations increased.
Medwell et al. [49] provided detailed measurements of temperature, OH
and formaldehyde for a turbulent jet of CH4 /H2 mixture (1/1 by vol.) in a
hot and diluted co-flow emulating flameless combustion conditions [46]. The
authors found that a decrease of O2 content in the co-flow led both to a reduc-
tion of OH levels, related to the smaller temperatures, and to a broadening of
the OH layer. More recently the authors extended a similar investigation to
ethylene/hydrogen mixtures [50].
As emerged from the above review, most of the works on flameless combus-
tion of H2 containing fuels are aimed at investigating particular features and
regard lab-scale devices. However, there is lack of data on practical industrial
devices, which need to be upgraded to allow burning H2 -enriched fuels. More-
over, being flameless combustion profoundly influenced by fluid dynamics, it
becomes crucial to analyze the specific burner conditions. The present Thesis
will specifically address the characterization of industrial-like devices for the ef-
ficient and environmentally affordable oxidation of hydrogen-enriched mixtures,
investigating the potentials of the flameless combustion regime.
20
1.3. Summary
1.3 Summary
The present Chapter has presented the main features of advanced combus-
tion systems which can be effectively exploited for the oxidation of hydrogen-
enriched mixtures.
The addition of hydrogen to traditional fuels in conventional combustion
equipment results in several technological issues, including high temperature
regions and very large NO emissions. To overcome such limitations, advanced
systems designed to operate in the so-called flameless regime could provide a
viable solution, as they couple very high combustion efficiencies with very low
pollutant emissions. Flameless combustion is achieved by means of a charac-
teristic burner aerodynamic, which allows the recirculation of large amounts
of exhaust gases in the reaction region. As a result, the combustion zone is
not limited to a diffusive flame front but spread over the entire volume in the
combustion chamber.
Recognizing the strong potentials of such combustion technology, the present
Thesis focuses on the experimental and numerical investigation of combustion
equipment operating in flameless regime and fed with hydrogen-enriched mix-
tures, to assess the feasibility of flameless combustion with H2 -enriched fuels.
21
Chapter 2
Turbulent combustion modeling
Our society massively relies on combustion systems for power production and
transportation. Therefore, pollution control and environmental protection rep-
resent critical issues, to be faced with scientific knowledge and responsibility, to
ensure citizens safety together with the necessary technological development.
Nowadays, computer simulations represent an essential tool to produce
multi-scale information on complex systems (such as combustion systems),
which could not be available by using any experimental technique. However,
good experimental practice is still strongly needed when a physical phenomenon
has to be fully clarified; moreover, experiments represent an unavoidable refer-
ence to improve the accuracy of the numerical models.
Computational Fluid Dynamics (CFD) is effectively used to improve the
design of systems involving fluid flow, heat transfer and related phenomena
such as chemical reactions. In particular, the use of CFD for aircraft and ve-
hicle aerodynamics, gas turbine engines and turbo-pumps appears mandatory
to achieve real progresses in design. However, users and developers of compu-
tational tools and CFD face today the critical issue of determining how the
confidence in modeling and simulations should be critically assessed, to effec-
tively help decision making in new design and regulation. It becomes, then,
crucial to estimate the inaccuracy in the numerical simulations arising from
each step of the process involving the schematization of the physical phenom-
ena by means of partial differential equations, their discretization and numerical
solution. In particular, it is possible to identify four main sources of inaccu-
racy, which ultimately lead to discrepancies between the numerical results and
reality:
• The partial differential equations describing the system may contain ap-
proximations or idealizations (modeling errors).
• The translation of the partial differential equations into the computerized

model may contain errors (programming errors).
• The discretization of the partial differential equations in space and time
23
Chapter 2. Turbulent combustion modeling
Figure 2.1: Verification and Validation methodology [1].
introduce an approximation (discretization error).
• The iterative methods used to solve the discretized equations may not
converge to their exact solution (iteration error).
Verification and Validation (V&V) [1] of computational simulations provides a
primary method for building and identifying the confidence in the numerical
simulations, by quantifying the aforementioned sources of errors. The method-
ology is cyclical (Figure 2.1) and involves the processes of model verification
and model validation. Briefly [51], verification is the process of determining
that a model implementation accurately represents the developer’s conceptual
description of the model and the solution to the model. On the other hand, val-
idation represents the process of determining the degree to which a model is an
accurate representation of the real world from the perspective of the intended
uses of the model. As it was pointed out by Roache [52], verification is purely
a mathematical activity, whereas validation concerns the actual relationship
between the numerical simulation and real world. V&V methodology can be
effectively exploited for the systematic investigation of complex systems, as the
ones involved in combustion applications, allowing the decomposition of the
highly coupled physical phenomena which ultimately characterize the complex
engineering system of interest.
This appears particularly interesting for the study of turbulent combustion,
which generally involves a wide range of coupled problems. In particular, mod-
eling industrial reacting flows involve the following inherently coupled physics:
• Fluid mechanics properties of the combustion system. This is crucial
to describe reactants mixing and all the other transport phenomena oc-
curring in turbulent flames, including heat transfer, molecular diffusion,
convection, turbulent transport etc . . . .
24
2.1. Governing equations for turbulent reacting flows
• Detailed chemical mechanisms. This is crucial to describe the formation

of pollutant species and to predict ignition and extinction phenomena.
• Multiphase systems. Many industrial applications involve the injection

of liquid and/or solid fuels. The numerical modeling of a multiphase
system is complicated by the complex interactions which exist between
different phenomena such as liquid sheets breakdown, liquid vaporization,
turbulent mixing, droplet combustion, particle devolatilization and char
burnout.
• Radiative heat transfer. This may represent a very important phenomenon

in industrial systems, due to the influence of the walls on the combustion
process through radiative transfer.
The aim of the present Chapter is to provide an overview of the main model-
ing approaches adopted to calculate mean burning rates in turbulent flames,
following renowned sources in the field [53, 54, 55]. The reference to the word
mean is essential, as the whole discussion will regard combustion modeling in
the context of Reynolds Averaged Navier-Stokes (RANS) equations. Moreover,
emphasis will be devoted to the modeling of turbulent non-premixed flames,
although modeling approaches valid in premixed conditions will be also pre-
sented.
As far as the V&V methodology is concerned, Chapter 3 will show the ap-
plication of the V&V methodology to the subject of the present Thesis. In
particular, a problem decomposition approach will be presented for the sys-
tematic investigation of combustion devices employing flameless combustion
(Chapter 1) and for their characterization in terms of energetic performances
and pollutant emissions.
2.1 Governing equations for turbulent reacting flows

The mathematical modeling of fluid flow relies on the so called continuum hy-
pothesis, according to which the fluid is considered as a continuous body despite
of being physically constituted of discrete molecules. This means that it is pos-
sible to define the value of any intensive variable (e.g., density, temperature,
pressure) at every point occupied by the fluid as the average value of the same
property in an elementary volume surrounding the considered point. Then,
properties are assumed to vary continuously from one point to another. The
hypothesis is valid only if the volume of fluid is small enough to be considered
as infinitesimal compared to the macroscopic quantities variations. At the same
time, the volume must be large enough to contain a number of molecules such
that the value of the variable is statistically meaningful and stationary.
Physical phenomena involving fluid flows, heat transfer and chemical re-
actions are governed by conservation principles such as mass, momentum and
energy conservation equations. These principles are mathematically expressed
25
through a set of partial differential equations which describes the behavior of

an infinitesimal volume of fluid. The basic set of balance equations, known as
Navier-Stokes equations, can be expressed in instantaneous local form as:
• Continuity:
∂ρ ∂ρuj
+ =0 (2.1)
∂t ∂xj
where ρ denotes the density and uj is velocity component along direction xj .
• Species (k = 1, . . . , N for N species)
∂ρYk ∂
+ [ρ (uj + Vk,j ) Yk ] = ω̇k (2.2)
∂t ∂xj
where Yk is the mass fraction of species k, Vk,j is the jth component of the
diffusion velocity Vk of species k and ω̇k represent the net mass reaction rate
per unit volume of species k. By definition:
N
X N
X
Yk Vk,j = 0 and ω̇k = 0 (2.3)
k=1 k=1
• Momentum (i = 1, 2, 3):
N
∂ρui ∂ρuj ui ∂p ∂τij X
+ =− + +ρ Yk fk,i (2.4)
∂t ∂xj ∂xi ∂xj
k=1
where p is the pressure, τij denotes the viscous force tensor and fk,i is the
volume force acting on species k in direction i.
• Total enthalpy ht = h + ui ui
2
N
∂ρht ∂ρuj ht ∂p ∂qj ∂ (ui τij ) X
+ = − + + Q̇ + ρ Yk fk,j (uj + Vk,j ) (2.5)
∂t ∂xj ∂t ∂xj ∂xj
k=1
where Q̇ is the heat source

P term, not to be confused with the heat released by
combustion. ui τij and ρ N k=1 Yk fk,j (uj + Vk,j ) denote the power generated by
viscous forces and by volume forces fk acting on species k, respectively.
2.1.1 Constitutive equations

Eqs. (2.1)-(2.5) are closed by expressions for the diffusive heat flux, the species
diffusive flux and the viscous stress tensor. Under the assumption of Newtonian
fluid, the viscous tensor is given by the Newton law:

∂ui ∂uj 2 ∂uk
τij = µ + − µδij (2.6)
∂xj ∂xi 3 ∂xk
26
2.1. Governing equations for turbulent reacting flows
where µ represents the molecular viscosity and δij is the Kronecker symbol.
The most popular approach used to express the diffusion velocity, Vk,j , is
based on a simplified approach, known as Fick’s law:
∂Yk µ ∂Yk
Vk,j Yk = −Dk =− (2.7)
∂xj Sck ∂xj
where Dk is the diffusion coefficient of species k into the mixture and Sck is
the Schmidt number for species k,
µ
Sck = . (2.8)
ρDk
Following Eq. (2.7), Eq. (2.2) becomes:

∂ρYk ∂ρuj Yk ∂ ∂Yk
+ = ρDk + ω̇k . (2.9)
∂t ∂xj ∂xj ∂xj
Fick’s law is a reasonable approximation for diffusion velocities because Lewis
numbers of individual species usually vary very slightly in flame fronts. How-
ever, Fick’s law does not necessarily satisfy the conservation of mass. In other
words, it is not guaranteed that:
N
X
Yk = 1. (2.10)
k=1
It can easily shown that summing all the species Eqs. (2.2) and using relations
(2.3), the mass conservation equation is recovered:
" N # N
∂ρ ∂ρuj ∂ X X
+ = ρ Yk Vk,j + ω̇k = 0. (2.11)
∂t ∂xj ∂xj
k=1 k=1
Eq. (2.11) shows that there are only N independent equations and any of the
N species equations or the mass conservation equation could be eliminated.
However, when Fick’s law is used, the RHS term of Eq. (2.11) becomes:
" N #
∂ X ∂Yk
ρ Dk (2.12)
∂xj ∂xj
k=1
which is zero only if all diffusion coefficients are equal (Dk = D). Otherwise,
global mass is not conserved. Despite this inconsistency, Fick’s law is commonly
used in CFD codes, due to the simplicity of the approach. The method generally
adopted to ensure global mass conservation with Fick’s law is to solve the global
mass conservation equation and only N − 1 species equations. The last species
mass fraction (usually a bulk species such as N2 ), is obtained by difference as:
N
X −1
YN = 1 − Yk
k=1
27
which carries all the inconsistencies deriving from Fick’s law. The approxima-
tion introduced by Fick’s law is negligible when the last species is a diluent
as it happens for N2 in air flames. Moreover, the assumption holds when all
the diffusion coefficients are assumed to be equal and this hypothesis appears
well suited in the context of turbulent combustion, being turbulent diffusion
generally predominant over molecular processes.
The energy flux qj in Eq. (2.5) can be expressed as:
N
∂T X
qj = −λ +ρ hk Yk Vk,j (2.13)
∂xj
k=1
where λ is the thermal conductivity. The flux given by Eq. (2.13) includes a
heat diffusion term, expressed by the Fourier’s law and a second term associated
with the diffusion of species with different enthalpies, hk . Assuming Fick’s law
for mass diffusion, Eq. (2.13) becomes:
N N
∂T X ∂Yk µcp ∂T X ∂Yk
qj = −λ +ρ hk Dk = +ρ hk Dk (2.14)
∂xj ∂xj P r ∂xj ∂xj
k=1 k=1
where P r is the Prandtl number, which compares the diffusive transport of

momentum and temperature. It can be expressed as:
µcp
Pr = (2.15)
λ
where cp is the specific heat at constant pressure. Then, the Lewis number of
species k, Lek , can be introduced to compare thermal and mass diffusivities:
Sck λ
Lek = = (2.16)
Pr ρcp Dk
Under the assumption of unity Lewis number, the enthalpy diffusive flux
in Eq. (2.14) is simplified and the species and enthalpy balance equations
are formally identical, if the power due to viscous and body forces and the
pressure gradient are negligible. This assumption is generally verified at low
Mach numbers [55].
2.2 Turbulent combustion

Combustion usually takes place under turbulent flow conditions in most prac-
tical systems, i.e. rockets, internal combustion or aircraft engines, industrial
burners and furnaces. Turbulent flows are needed to increase mixing and en-
hance combustion. At the same time combustion releases heat, which gener-
ates flow instability through gas expansion and buoyancy, thus enhancing the
transition to turbulence. The numerical simulation of turbulent reacting flows
28
2.2. Turbulent combustion
represents a fast growing area; however, it remains a very challenging task, due
to the following reasons:
• Combustion, even without turbulence, is an intrinsically complex pro-

cess involving a wide range of time and length scales. A comprehensive
description of the chemistry of laminar flames may require hundreds of
species and thousand of reactions, leading to considerable numerical dif-
ficulties.
• Turbulent is probably the most complex, and still unresolved, phenomenon

in fluid mechanics, involving various temporal and spatial scales.
• Turbulent combustion arises from the interaction between turbulence and

combustion. The heat released by combustion within the flame front de-
termines a flow acceleration, thus leading to the so-called flame-generated
turbulence. On the other hand, the large changes in kinematic viscosity
associated with temperature (kinematic viscosity of air increases roughly
as T 1.7 ) changes may determine the laminarization of the flow after igni-
tion (re-laminarization).
2.2.1 Reynolds and Favre averaging

The direct numerical simulation of the instantaneous balance equations is not
possible for practical combustion systems and it is limited to very simplified
cases, where the range of time and length scales is not too broad. To overcome
such difficulty, the balance equations are averaged to describe only the mean
flow field. Each property Q can be split into a mean, Q, and a fluctuation,
0
denoted as Q :
0
Q=Q+Q with Q0 = 0 . (2.17)
Relation (2.17) can be substituted into the instantaneous balance equations to
derive transport equations for the mean. The procedure, known as Reynold
averaging, leads to the so called (RANS) equations. The Reynolds averaged
continuity equation is of the form:
∂ ρ̄ ∂ 0

+ ρuj + ρ0 uj = 0 (2.18)
∂t ∂xj
0
where the velocity/density fluctuations correlations ρ0 uj appears. To avoid the
explicit modeling of such correlations, a Favre average Qe is introduced and any
quantity is decomposed into:
e + Q00
Q=Q (2.19)
where
29
Q
e= ρQ f00 = ρ(Q−Qe) = 0
and Q (2.20)
ρ ρ
Using Eq. (2.20), the Favre averaged continuity equations becomes:
∂ ρ̄ ∂ρuej
+ = 0. (2.21)
∂t ∂xj
The Favre Averaged Navier Stokes (FANS) equations are formally identical
to RANS equations for constant density flows, as the correlation terms ρ0 Q0
vanish for constant density. It is noteworthy that the relation between the
Favre and Reynolds averages is not simple and requires the knowledge of the
density fluctuation correlations, hidden in Favre average definition:
e = ρQ = (ρ + ρ0 ) Q + Q0 = ρQ + ρ0 Q0 . (2.22)

ρQ
Summarizing, the FANS equations are:
• Continuity
∂ρ ∂ρuej
+ = 0. (2.23)
∂t ∂xj
• Species (k = 1, . . . , N for N species)
^00 00
∂ρY
fk ∂ρuej Y
fk ∂ρuj Yk ∂Vk,j Yk
+ =− − + ω̇ k . (2.24)
• Momentum (i = 1, 2, 3):
00 00 N
∂ρuei ∂ρuej uei ∂ρu]
j ui ∂p ∂τij X
+ =− − + +ρ Yk fk,i . (2.25)
∂t ∂xj ∂xj ∂xi ∂xj
k=1
• Total enthalpy ht = h + ui ui
2
00 00
∂ρhet ∂ρuej het ∂ρuj ht ∂p ∂qj
+ =− + − +
∂t ∂xj ∂xj ∂t ∂xj
N
∂ui τij X
+ + Q̇ + ρ Yk fk,j (uj + Vk,j ). (2.26)
∂xj
k=1
30
2.2. Turbulent combustion
2.2.2 Closure of Favre Averaged Navier-Stokes Equations (FANS)

The objective of turbulent combustion modeling is to provide closures for the
unknown quantities in the averaged Navier-Stokes equations. In particular,
closure must be provided for the following terms:
00 00
• Reynolds stresses u]j ui : turbulence models are used to close these terms.
The closure can be done directly or by deriving balance equations for the
Reynolds stresses. Most of the studies on turbulent combustion make
use of turbulence models derived under the Boussinesq hypothesis [56],
which assumes for the Reynolds stresses the same dependency used for
the viscous tensor of Newtonian fluids:

00 00 ∂ uei ∂ uej 2 ∂f uk 2
ρu]
j ui = −µt + − δij + ρk (2.27)
∂xj ∂xi 3 ∂xk 3
where µt is the turbulent viscosity, k is the kinetic energy, k = 1/2 3k=1 u]

00 00
k uk ,
P
and δij is the Kronecker symbol. The role of the turbulence model is, then,
to provide an estimation of the turbulent viscosity. Three possible ap-
proaches are available: algebraic expressions [57], one and two equations
models. The latter are usually adopted in turbulent combustion simula-
tion. In particular, the k − model [58] is very popular because of its
simplicity and cost effectiveness. However, it should be stressed that such
modeling approaches have been developed for non-reacting flows and they
are simply rewritten in terms of Favre averaging when used in combustion
applications. Therefore, heat release effects on the Reynolds stresses are
generally not explicitly taken into account.
Popular turbulence models such as k − assume isotropic turbulence;
however, practical flows often show strong anisotropic features. Such
phenomena can be incorporated by relaxing the Boussinesq hypothesis
(Eq. (2.27)) and solving transport equations for the Reynolds stresses
00 00
u] u . This leads to the so called Reynolds Stress Models (RSM), which
j i
requires six additional transport equations and closure hypothesis for su-
perior terms (e.g. triple velocity products).
00 00 00 00
• Species, ρu
^
j Yk , and enthalpy, ρuj ht , turbulent fluxes
These fluxes are usually closed using a gradient assumption:
00 00 µt ∂ Y
fk
^
− ρuj Yk = − (2.28)
Sckt ∂xj
00 00 µt ∂ het
− ρu]
j ht = − . (2.29)
P rt ∂xj
31
In the expressions above, Sckt and P rt are the turbulent Schmidt and
Prandtl numbers, respectively. In premixed systems, the gradient as-
sumption can be wrong and counter-diffusion phenomena may be ob-
served.
• Laminar diffusion fluxes for species and enthalpy

These terms are generally neglected as they vanish when compared to
turbulent transport, at sufficient turbulence levels. However, they may be
retained adding a laminar diffusivity to Eqs. (2.28)-(2.29). For example,
species laminar diffusion fluxes are generally modeled as:
∂Yk ∂Y
fk
Vk,j Yk = −ρDk ≈ −ρDk (2.30)
∂xj ∂xj
where Dk is a mean species molecular diffusion coefficients.
• Species mean reaction rates ω̇ k

Turbulent combustion modeling focuses on developing closure models for
such terms. It should be noted, however, that combustion models in
the FANS context can only provide mean quantities, which may result
quite different from the instantaneous values of the same quantities. In
particular, the knowledge of steady statistical means may be not sufficient
when strong unsteady mixing effects are observed in turbulent reacting
systems. When, necessary, Large Eddy Simulation (LES) may provide a
valid alternative to FANS. The objective of LES is to explicitly compute
the largest structures of the flow and to model the effect of the smaller
structures. As it happens for FANS, models are needed to deal with
the complex interactions occurring at the unresolved scales level. The
application of LES to reacting systems is still at an early stage and retains
many open questions. However, LES potentially shows some attractive
properties. First, it is expected that the turbulence model can perform
more effectively when dealing only with the smaller scales of the system,
which show more universal properties with respect to larger scales, highly
dependent on the geometry of the system. Moreover, the possibility of
computing the large scale mixing is very interesting, to properly describe
the interactions between flame and turbulence. Finally, LES is a very
powerful tool to identify and help avoiding thermo-acoustic (humming)
instabilities in turbines.
2.3 Combustion models for mean reaction rates

A combustion process can be described in terms of M elementary reactions
involving N species:
32
2.3. Combustion models for mean reaction rates
N N
0 00
X X
υk,j Mk = υk,j Mk j = 1, 2, . . . , M (2.31)
k=1 k=1
0 00
where υk,j and υk,j are the molar stoichiometric coefficients of species k in
reaction j and Mk is the symbol for species k.
The net mass reaction rate for species k is given by:
M M
00 0
X X
ω̇k = ω̇kj = Wk υk,j − υk,j Qj (2.32)
j=1 j=1
where Wk is the molecular weight of species k. The rate of progress for reaction
j, Qj , is expressed as:
0 00
N N
ρYk υkj ρYk υkj
Y Y
Qj = kf j − krj (2.33)
Wk Wk
k=1 k=1
where kf j and krj are the forward and backward kinetic constants of reaction
j, while ρYk/Wk is the molar fraction of species k. The forward kinetic constant
is expressed using the so called Arrhenius law,
E
aj
− RT
kf j = Af j T e βj
, (2.34)
k
while reverse constants are derived from equilibrium constants krj = Keq,j fj
.
In order to express the rate of progress Qj for each reaction, data for the
pre-exponential factor, Aj , the temperature exponent, βj , and the activation
energy, Eaj , must be provided. This aspect is generally covered by chemical
schemes, which are provided for different classes of fuels and at different levels
of detail (i.e. global, quasi-global, reduced and detailed mechanisms). The con-
struction of chemical schemes and their validation is beyond the scope of the
present Chapter. In most CFD application, the kinetic mechanism represents
only one of the data elements which should be available for the computation.
The present Thesis has been focused on the combustion of methane-hydrogen
mixtures. Therefore, different approaches have been adopted, including global
mechanisms for hydrogen and methane oxidation [59, 23, 60], quasi-global re-
action schemes [23, 5] , reduced [61, 62] and detailed kinetic mechanisms [63].
The interactions between chemical kinetics and turbulent mixing represent
the main interest of turbulent combustion modeling. With regard to non-
premixed combustion, it is possible to identify two conditions which allow to
completely decouple the problem:
• Infinitely fast chemistry: the combustion process is dominated by tur-

bulent mixing; therefore no kinetic modeling is required and the mean
reaction rate is simply given by the mixing rate.
33
• Finite rate chemistry: The combustion process is limited by chemical

reactions. The mean reaction rate is simply given by the Arrhenius based
kinetic rates.
All the intermediate conditions require a certain degree of coupling between

turbulence and chemistry. Both mixing and chemistry are needed to explain
flame ignition and stabilization, to capture finite-rate chemistry effects and
to account for more complex phenomena such as localized extinction and re-
ignition.
The present Section will discuss some of the common approaches for the
estimation of the mean species reaction rates, ω̇ k in turbulent flames. First, the
models based on the reaction rate approach will be presented. Such models are
based on the solution of balance equations for the species and temperature and
provide models for the mean reaction rates. Then, a brief discussion regarding
the primitive variable method will follow.
2.3.1 Eddy Break-up model and Eddy Dissipation Model

The Eddy Break-up and Eddy Dissipation models are both developed under the
hypothesis of infinitely fast chemistry. Under this assumption, the combustion
process can be expressed by a one-step irreversible reaction:
0 0 00
υF F + υF O → υP P, (2.35)
for example CH4 + 2O2 → CO2 + 2H2 O for methane fuel. The mass fraction
of fuel and oxidizer correspond to stoichiometric conditions when
0
YO υO WO
= 0 = υ, (2.36)
YF st υF WF
where υ is the mass stoichiometric ratio. Therefore, Eq. (2.35) can be expressed
as:
F + υO → (1 + υ)P. (2.37)
Spalding [64] was the first to propose a closure for the chemical source
term, ω̇ k . He argued that the energy cascade, describing the energy transfer
from large to small scales, also controls the chemical reaction, as long as the
overall reaction rate is mixing rather than chemistry controlled. The model,
formulated primarily for premixed combustion, is known as Eddy Break-up
(EBU) and it is based on the following expression for the mean reaction rate
of products:
00 2 12
ω̇ P = ρCEBU Y (2.38)
k P
00
where YP is the variance of the product mass fraction and CEBU is the Eddy
Break-up constant.
34
The Eddy Dissipation Model (EDM), derived by [65], directly extends EBU
to non-premixed combustion. EDM is simply obtained from EBU by replacing
00
YP 2 with the mean mass fraction of the deficient species, i.e. fuel for lean and
oxygen for rich mixtures. The model, known as the Eddy Dissipation Model
(EDM) takes the minimum of three rates, the one defined by the mean fuel
mass fraction

ω̇ F = −ρAY F , (2.39)
k
the one defined by the mean oxidizer mass fraction
YO
ω̇ O = −ρA , (2.40)
υ k
and the one defined by the mean product mass fraction
YP
ω̇ P = ρAB , (2.41)
1+υk
to calculate the mean chemical source term. In Eqs. (2.39)-(2.41), A and B are
model constants and υ is the stoichiometric oxygen to fuel mass ratio. When
the EBU and EDM models are used in CFD codes, the coefficients CEBU , A
and B must be tuned within a certain range to obtain reasonable results for a
particular problem.
The Eddy Dissipation/Finite Rate Model (ED/FR) represents an extension
of EDM to the case of finite-rate chemistry. The mean reaction rate is estimated
as the minimum rate between the EDM equation and a rate based on the
Arrhenius equation:
(2.42)

ω̇ F = min ω̇ F EDM
, ω̇ F Arrhenius
.
In practice, the Arrhenius rate, given by Eq. (2.32), acts as a kinetic “switch”,
preventing reaction before the flame is actually stabilized. Once the flame is
ignited, the eddy-dissipation rate is generally smaller than the Arrhenius rate,
and reactions are mixing-limited
2.3.2 Eddy Dissipation Concept

The Eddy Dissipation Concept represents an extension of the Eddy Dissipation
Model [65] to the case of detailed chemistry. The model has been developed by
Magnussen [66, 67] and later extended by Gran and Magnussen [68], Magnussen
[69]. EDC provides an empirical expression for the mean reaction rate based
on the assumption that the chemical reactions occur in the region of the flow
where the dissipation of turbulent kinetic energy takes place. According to EDC
the viscous dissipation of energy into heat is concentrated in isolated regions,
whose entire volume is only a fraction of the fluid volume. These regions are
denoted as fine structures and they are believed to be vortex tubes, sheets or
35
slabs, whose characteristic dimensions are of the same order of magnitude of

the Kolmogorov length scale, in one or two dimensions, but not in the third
[56]. EDC is applicable to both premixed and diffusive flames and allows a
detailed kinetic treatment for non-premixed systems with significant finite-rate
effects. According to the model, the fine structures are treated as homogeneous
reactors, which exchange mass and energy with the surrounding fluid. The
connection between such structures and the main flow is based on a turbulent
energy cascade model, which provides the parameters required to model the
fine structures (i.e. reacting volume, V ∗ , and mass flow rate, ṁ∗ ) from the
macroscopic features of the flow (i.e. turbulent kinetic energy, k, and energy
dissipation rate, ). According to such model, mechanical energy is transferred
from the mean flow to large eddies and then further to smaller and smaller
eddies [70]. The large eddies carry the major part of the kinetic energy, while
smaller eddies whirl faster (higher frequency) but contain less energy. Viscous
friction, which is responsible for the transformation of mechanical energy into
heat, occurs at all scales but is more significant for the smaller eddies.
2.3.2.1 Energy cascade model

The turbulent energy cascade model for EDC is shown in Figure 2.2. In the
0
Figure w represents the production of turbulent kinetic energy, while the sum
0 00
of all the dissipation terms, q + q + . . . + q ∗ is the turbulent dissipation rate, .
Each level of the cascade is characterized by a velocity scale, u, a length scale,
L, and a strain rate, ω, expressed as:
u
ω= . (2.43)
L
The first level in the cascade contains the large, energy-rich eddies and it
0 0 0
is characterized by quantities (u , L , ω ) determined by a specific turbulence
model (k − or Reynolds stresses). This level represents the whole turbulence
spectrum as it contains the effect of the smaller scales. According to the cascade
model, the strain rate doubles at each energy level so that the second level has
00 0
a characteristic frequency given by ω = 2ω ; similarly, the strain rate at the
nth level is ωn = 2ωn−1 . The last level of the cascade can be described by
scales, u∗ , L∗ and ω ∗ , of the same order of magnitude of the Kolmogorov scales.
The production of mechanical energy, wi , and the viscous dissipation, qi , are
modeled in analogy to the production of turbulent kinetic energy, k, and to the
dissipation term in the energy equation. In particular:
• The transfer of mechanical energy between levels n − 1 and n is modeled

to be proportional to the square of a characteristic fluctuation at the nth
level and to the strain rate at level n − 1:
3 3 3 u3
wn = CD1 2u2n ωn−1 = CD1 u2n ωn = CD1 n . (2.44)
2 2 2 Ln
36
Figure 2.2: Energy cascade model [65, 66, 71].
• The dissipation of mechanical energy at the nth level is assumed to be

proportional to the viscosity and the square of a strain rate.
qn = CD1 υωn2 . (2.45)

In Eqs. (2.44)-(2.45), CD1 and CD2 represent model constants, set equal
to 0.135 and 0.5, respectively. The reason for the choice of such values will be
clear in the following discussion.
The energy cascade model in EDC is used to provide a relation between the
large and small eddies, where dissipation and combustion take place. At the
highest energy levels, the viscous dissipation is negligible; therefore, the energy
dissipation can be expressed as:
0
0 0 00 3 u3 00
= w = q + w ≈ w = CD1 0 . (2.46)
2 L
Moreover, it can be shown that is also given by:
0
= q ∗ + . . . + qn + . . . + q =
0
= CD2 υω ∗2 + . . . + CD2 υωn2 + . . . + CD2 υω 2 =
1 1
= CD2 υω ∗2 + CD2 υ ω ∗2 + CD2 υ ω ∗2 + . . . =
4 16
∞
X 1 4 ∗ 4 u∗2
= q∗ = q = C D2 υ . (2.47)
22k 3 3 L∗2
k=1
From a balance on the last level of the energy cascade q ∗ = w∗ , then:
4 4 u∗3
= q ∗ = w∗ = 2CD1 ∗ . (2.48)
3 3 L
37
Eqs. (2.47)-(2.48) give the characteristic scales for the fine structures:
3 14 41
υ3

∗ 2 3CD2
L = 2 (2.49)
3 CD1
1
∗ CD2 4 1
u = 2 (υ) 4 (2.50)
3CD1
2 CD2
Re∗ = . (2.51)
3 CD1
With regard to the model constants, CD1 and CD2 , they were chosen to obtain
the best fit with respect to several types of flow. However, it is still possible
to provide a physical interpretation for them. For example, the value of 0.135
for CD1 was chosen by using the approximation given by Eq. (2.46). Following
0 0 0 0
Eq. (2.46), the turbulent viscosity, υT = u L , can be expressed as υT = u L =
0
3/2CD1 u 4/ = 2/3CD1 k2/. Then, 2/3CD1 corresponds to the constant Cµ (= 0.09)
used in the k − model [58].
2.3.2.2 Fine structures

According to EDC combustion takes place in the fine structures. These are
concentrated into fine structure regions, whose linear dimensions are consider-
able larger than the fine structures therein. The fraction of the flow occupied
by such regions is, on a mass basis:
3 3 3
u∗

∗ 3CD2 4 υ 4
γ = = . (2.52)
u0 2
4CD1 k2
Eq. (2.52) implies, together with Eqs. (2.46) and (2.47), that:
L∗
γ∗ =
. (2.53)
L0
This is very similar to the model by Corrsin [72], which assumes the dissipative
regions to be thin sheets, with a characteristic length L∗ in one dimension.
Recently [69], the expression for γ ∗ has been modified as follows:
2 1 1
u∗

∗ 3CD2 2 υ 2
γ2005 = = . (2.54)
u0 2
4CD1 k2
In the following, the original definition of γ ∗ will be retained as the modifi-
cation introduced by Eq. (2.54) has not been fully clarified. Moreover, the
definition of γ ∗ according to Eq. (2.52) appears in the large majority of EDC
implementations.
If we assume that the fine structures are cylindrically shaped (Figure 2.3),
the mass flow across the cylinder boundary, M ∗ , is given by:
38
Figure 2.3: Mass flow through a cylinder.
πL∗ h
Ṁ ∗ = ρ∗ u∗ A∗ = ρ∗ u∗ (2.55)
2
where the cylinder diameter is given by the characteristic length scale L∗ and
h is the height. The mass flow rate per unit mass, ṁ∗ , is then given by:
∗ 1 1
Ṁ ∗ ρ∗ u∗ πL2 h u∗

∗ 3 2 2
ṁ = ∗ ∗ = ∗2 = 2 = . (2.56)
ρ V ρ∗ πh L4 L∗ CD2 υ
The residence time in the fine structures, that is the fluid-dynamic time
scale for the chemical reactions, is defined as:
1
τ= . (2.57)
ṁ∗
An expression for the mass transfer between the fine structures and the
surrounding is needed to model the mean reaction rate. The mass exchange
takes place between the fine structures and the fine structures regions; therefore,
Magnussen [67] proposed to model the exchange term as:
1 1
γ∗

∗ 3 12CD2 4 υ 4
ṁ = ṁ 1 = 2 . (2.58)
γ∗ 3 4CD1 CD1 k k
2.3.2.3 Reactor model

In the previous Sections, relations have been derived to model the smallest
dissipative scales using the quantities provided by the turbulence models, i.e.
k and , and to express the mass exchange between the smallest scales and the
surroundings. Now an expression for the reaction rate for a species k is needed.
The fine structures are assumed to behave as Perfectly Stirred Reactors
(PSR) reactors (Figure 2.4). Therefore, the reaction rate is given by:
ρ∗ Yk0 − Yk∗

∗ ∗ ∗ ∗
0
(2.59)

ω̇k = −ṁ ρ Yk − Yk = − .
τ∗
39
Figure 2.4: PSR model for the fine structures.
If the volume of interest is extended to the surrounding fluid, an expression

for the mean reaction rate is obtained:
ρṁχf s e
ω̇ k = −ρṁχf s Yk0 − Yk∗ = − ∗
(2.60)

Yk − Y k .
1 − γ ∗ χf s
In Eq. (2.59), χf s represents the reacting fraction of the fine structures, while
Yk0 and Yk∗ are the mass fractions of species k in the surrounding fluid and
within the fine structures, respectively. Eq. (2.60) introduces a mean mass
fraction between fine structures and surrounding fluid, Yek , necessary for CFD
calculations and defined as:
Yek = γ ∗ χf s Yk∗ + (1 − γ ∗ χf s ) Yk0 . (2.61)

The knowledge of Yk∗ is then required to calculate ω̇ k . Two solutions are possible
and correspond to the fast chemistry limit and finite-rate chemistry approach.
Fast chemistry limit The fast chemistry limit is obtained by assuming that
there is enough time to achieve equilibrium in the fine structures. The combus-
tion process can be expressed by a one-step irreversible reaction (Eq. (2.37))
and the mean reaction rate for the fuel is then expressed as:
ρṁχf s e
ω̇ k = − Ymin
1 − γ ∗ χf s

where Yemin = min Yf Yf
F, υ
O
.
Gran and Magnussen [68] suggest the following model for the determination
of the reacting fraction of the fine structures:
χf s = χ1 χ2 χ3 (2.62)
where χ1 is the probability of coexistence of reactants,
40

Yf
Y] min + 1+υ
P
χ1 = , (2.63)
Yf Yf
Yf
F + 1+υ
P
Yf O + 1+υ
P
χ2 expresses the degree of heating,

 
Yf
P
1
χ2 = min  1 f 1+υ , 1 , (2.64)
γ ∗ 3 1+υ
YP
+Y ]
min
and χ3 expresses the limits due to the lack of reactants

 1

γ∗ 3 Yf
P
1+υ +Y
]min
χ3 = min  , 1 . (2.65)
Y
]min
Finite-rate chemistry approach The effect of chemical kinetic can be

taken into account by treating the fine structures as constant pressure ho-
mogeneous reactors. Assuming steady state, the following set of equations is
obtained for the mass fractions involved in the selected kinetic mechanism:
ρ∗
Yk0 − Yk∗ (2.66)

ω̇k = − ∗
τ
The set of algebraic Eqs. (2.66) is very non-linear and requires a robust time-
stepping algorithm for its solution. The solution of the system provides Yk∗ ,
which can be substituted in the expression for the mean reaction rate to get:
ρṁχf s e ∗
ρṁ e ∗

ω̇ k = − Yk − Y k = − Y k − Yk
1 − γ ∗ χf s 1 − γ∗
where the reacting fraction of the fine structure, χf s , has been set to 1, following
Gran and Magnussen [68], thus leaving the determination of finite-rate effects
to the kinetic mechanism.
2.3.3 The primitive variable approach

In the context of the primitive variable method, assumptions are made regard-
ing the flame structure to provide the averaged quantities which define the
thermochemical state of the system. These information can be extracted from
flamelet libraries or solving balance equations, as in Conditional Moment Clo-
sure (CMC) [73]. Species mass fractions and temperature balance equations
are no longer required and mean reaction rates are not modeled. The RANS
codes only solve for the flow and mixture fraction variables (see below).
41
2.3.3.1 The mixture fraction variable

Mixture fraction, f , represents a very important quantity in non-premixed com-
bustion. In a two-feed system where a fuel stream, ṁ1 , is mixed with an oxidizer
stream, ṁ2 , the mixture fraction f is defined at any location in the system as
the local ratio:
ṁ1
f= . (2.67)
ṁ1 + ṁ2
If the system is homogeneous, or if the inhomogeneous system is character-
ized by equal diffusivities of fuel, oxygen and inert substances, the local fuel
mass fraction can be related to the mass fraction of the fuel in the stream ṁ1
as:
YF,u = YF,1 f (2.68)

where the subscript u denotes the unburned status. Similarly, the oxygen local
mass fraction can be expressed as a function of the oxygen mass fraction in the
oxidizer stream, YO2 ,2 :
YO2 ,u = YO2 ,2 (1 − f ) (2.69)

being (1 − f ) the local mass fraction of the oxidizer stream.
If we describe the combustion process with Eq. (2.35), it is possible to relate
the mass variation of fuel and oxidizer as:
υYF − YO2 = υYF,u − YO2 ,u . (2.70)

Substituting Eqs. (2.68) and (2.69) in Eq. (2.70), it is possible to relate the
mass fraction of fuel and oxygen to the mixture fraction:
υYF − YO2 + YO2 ,2

f= . (2.71)
υYF,1 + YO2 ,2
For stoichiometric mixtures, the RHS of Eq. (2.70) vanishes, since both oxygen
and fuel mass fractions are zero; therefore, the stoichiometric mixture fraction
is given by:
YO2 ,2
fst = (2.72)
υYF,1 + YO2 ,2
It is possible to define the mixture fraction in a more general way, as a
quantity related to chemical elements. While the mass of species changes during
combustion, the mass of elements is conserved. Denoting with bke the number
of atoms of element e in a molecule of species k and with We the molecular
weight of that atom, it is possible to write the mass fraction of element e in the
system as:
42
N
X bke We
fe = Yk e = 1, 2, . . . , ne (2.73)
Wk
k=1
where ne is the total number of elements (i.e. C, H, O). Summing Eqs. (2.9)
for the mass fractions Yk as in Eq. 2.73, one obtains:

∂ρfe ∂ρuj fe ∂ ∂fe
+ = ρDk (2.74)
where no chemical source term, since
N
X
bke Wk υkj = 0, (2.75)
k=1
for each element e in any reaction j. Eq. (2.74) shows that element mass
fraction is conserved during combustion.
If we denote with fC , fH and fO the element mass fractions of C, H and
O, we have, for a hydrocarbon Cn Hm :
WC
fC = n YF,u fH = m W
WF YF,u fO = 2YO2 ,u .
H
(2.76)
WF
From Eqs. (2.76) and (2.36) it can be shown that the coupling function1
fC fH fO
β= + −2 0 (2.77)
nWC mWH υO2 WO2
vanishes at stoichiometric conditions. Eq. (2.77) corresponds to the original
definition of a conserved scalar of Burke and Schumann [74]. It can be normal-
ized between 0 and 1 as
β − β2
f= , (2.78)
β1 − β2
to obtain Bilger [75] definition of the mixture fraction:
fC fH (YO2 ,2 −fO )
nWC + mWH +2 0
υO WO2
f= fC,1 fH,1
2
YO2 ,2
(2.79)
nWC + mWH +2 0
υO WO2
2
If the elements C, H and O are in stoichiometric proportions, Eq. (2.79) reduces

to (2.72).
If all the species diffusivities, Dk , are equal to D, it follows from Eqs. (2.74),
(2.77) and (2.78) that f satisfies the balance equation:
0
1
Here it is assumed that υF = 1.
43

∂ρf ∂ρuj f ∂ ∂f
+ = ρDk . (2.80)
The diffusion coefficient in Eq. (2.80) controls the location of the reaction zone.
It has been shown (Section 2.1.1) that under the hypothesis of unity Lewis
number, the transport equations for species mass fractions and temperature
are formally identical. Therefore, the thermal diffusivity is usually adopted as
diffusion coefficient in the mixture fraction equation, since diffusion of enthalpy
is the most important transport process in the mixture fraction space.
2.3.4 The Burke-Schumann solution

In the limit of an infinitely fast one-step irreversible reaction (Eq. (2.37)), the
mass fractions of fuel and oxidizer are either zero (Eq. (2.71)) or they are
piecewise linear fraction of f :
−fst
YF = YF,1 f1−f Y = 0 f or f ≥ fst
st O2 . (2.81)
f
YO2 = YO2 ,2 1 − fst YF = 0 f or f ≤ fst
If all the Lewis number are equal to unity, also temperature is a piecewise linear
function of f :

Q̇ Y
Tu (f ) + c0 F,1 f
 f ≤ fst
cp υF WF
T (f ) = Q̇c YO2 ,2 (2.82)
Tu (f ) +
 0 (1 − f ) f ≥ fst
cp υO WO2
2
where Q̇c is the heat released during combustion and Tu (f ) is given by
Tu (f ) = T2 + f (T1 − T2 ) . (2.83)
The Burke-Schumann solution is shown in Figure 2.5. The expressions
obtained for fuel, oxygen and temperature may be integrated over probability
density functions (PDF) for f , p (f ), to obtain mean quantities. The PDF can
be thought of as the fraction of time that the fluid spends in the vicinity of the
state identified by the instantaneous value of the mixture fraction. Generally, a
beta function [55] is chosen to model the shape of the PDF, which is ultimately
determined by the mean mixture fraction, fe, and its variance, fg 00 2
. Thus,
density-weighted mean species mass fractions and temperature can be expressed
as:
R1
YeF = 0 YFIF CM (f ) p (f ) df
R1
YeO2 = 0 YOIF CM (f ) p (f ) df (2.84)
R 1 IF2CM
Te = 0 T (f ) p (f ) df
where the superscript IFCM has been added to indicate that the mean flame
structure is known from an infinitely fast-chemistry model. The closure requires
44
Figure 2.5: Burke-Schumann solution as a function of the mixture fraction.
the solution of two transport equation for fe and fg

00 2
, for the determination of
the β-PDF. Therefore, the IFCM is a two-parameter model for turbulent non-
premixed combustion.
2.3.5 Flamelet model

Experiments in jet flames and direct numerical simulations suggest that there
are situations where the chemistry is fast, but not infinitely fast. The applica-
tion of the IFCM for such systems is unsatisfactory and different approaches
have been proposed. Among them, the flamelet models [76] are derived assum-
ing that the local balance between diffusion and reaction in a turbulent system
is similar to the one found in a prototype laminar flame. The model can be
viewed as a direct extension of IFCM, since it uses the same formalism, but
with an additional parameter, the scalar dissipation rate, χ, defined as:

k ∂ 2 f
χ= . (2.85)
ρcp ∂x2j

This parameter measures the inverse of a diffusive time through the stoichio-
metric surface. As this time decreases, mass and heat transfer are enhanced,
until a critical dissipation is reached and the flame extinguishes. As a non-
equilibrium parameter, χ can help capturing finite-rate chemistry effects. If
the joint PDF of f and χ is known, or presumed, the mean properties of the
flame can be calculated as:
SLF M
(2.86)
R R
Yei = f χ Yi (f, χ) p (f, χ) df dχ
where YiSLF M (f, χ) is the local flame structure in mixture fraction space and
p (f, χ) is the joint probability density function of f and χ. SLFM stands for
45
Steady Laminar Flamelet Model.

Similarly to IFCM, SLFM requires the knowledge of fe , fg
00 2
and, addition-
ally, χ
e. Two issues emerge:
• The local flame structure YiSLF M (f, χ) must be determined and tabu-
lated under particular hypothesis and choosing a determined flame pro-
totype. A widely adopted flame configuration is that of the counter-flow
diffusion flames. In steady state conditions and at a given value of χ the
flame is described by:
χ ∂ 2 Yk
ω̇ = − . (2.87)
Lek ∂f 2
The solution of Eq. (2.87) for given concentrations and temperature
boundary conditions, and for different χ provides a flamelet library. Var-
ious techniques are available to build these tables.
• The joint probability density function p (f, χ) must be presumed. Again,

the choice of the β-PDF is very common.
2.4 Summary
The present Chapter has focused on the description of common approaches for
the closure of the reaction term in the Favre Averaged Navier Stokes Equations
(FANS). The purpose is not to provide an exhaustive review of all the existing
models, but to focus on the approaches which have been adopted within the
context of the present PhD Thesis.
Therefore, advanced approaches such as the Conditional Moment Closure
(CMC) [73] and Probability Density Function methods for turbulent combus-
tion [77] have not been discussed, also considering the significant computational
cost associated to such models when coupled to a detailed description of chem-
istry. The attention has been concentrated on empirical models (i.e. ED/FR,
EDC) for turbulence-chemistry interactions, suitable for the affordable descrip-
tion of industrial equipment.
Moreover, the mixture fraction based approaches, i.e. flamelet, have been
taken into account because they belong to that class of combustion models
based on the re-parametrization of the thermo-chemical state with a small
number of parameters (f and χ for the flamelet model). This offers the op-
portunity to strongly reduce the dimensionality of the problem and to take
into account, at the same time, finite-rate chemistry. When considerable non-
equilibrium effects are present, such models fail as they prescribe the size of the
subset that the thermo-chemistry may access, without any quantitative error
analysis. Such consideration has driven our interest towards the development
of a novel methodology for the selection of optimal progress variables for the
46
2.4. Summary
re-parametrization of the thermo-chemical state. In particular, Principal Com-

ponents Analysis has been taken into account, as it will be discussed in Chapter
4.
47
Chapter 3
Methodology and Case Studies
The aim of the present Thesis is the characterization of advanced systems for
the combustion of hydrogen based fuels. As introduced in Chapter 1, burning
H2 -enriched fuels in conventional combustion equipment may lead to several
difficulties (Figure 1.1). In particular, diffusive burners can operate with very
stable flames; however, temperature levels are generally very high, thus deter-
mining adverse effects on materials and pollutant emissions, i.e. NOx . On the
other hand, the utilization of premixed burners could allow, in principle, to
control NOx with very lean mixtures, thus leading, however, to possible flash-
back phenomena and stability problems. A possible solutions to this problem is
provided by the flameless combustion systems, which exploit the flameless com-
bustion concept introduced in Chapter 1, based on the development, within the
combustion chamber, of a volumetric combustion region approaching perfectly
stirred reactor conditions. Such operating conditions may be achieved only if
the reactants are heated above the self ignition temperature of the fuel and the
reacting mixture is sufficiently diluted by the exhausts, so that the chemical
and mixing times become comparable and a significant volume of fluid is at
temperatures which allow to sustain the oxidation process.
For industrial systems, these conditions are generally achieved with a spe-
cific aerodynamic of the burners: high velocity burners (Chapter 1) are used to
generate a strong recirculation within the combustion chamber, which allows
both the preheating and the dilution of the reacting mixture in the exhausts
stream. For lab-scale burners, other approaches are used. Jet issuing in hot vi-
tiated co-flow [46] can be exploited to emulate flameless combustion conditions
for diffusive systems. On the other hand, lab-scale premixed burners [31, 47]
usually operate in flameless regime using diluted mixtures and preheating the
feed by means of electric ovens.
In the following, the systems investigated in the present Thesis will be in-
troduced, together with the methodology adopted for the characterization of
their operating features. Recognizing the complexity of the considered devices,
a hierarchical approach based on Verification and Validation (V&V) has been
49
Chapter 3. Methodology and Case Studies
exploited, to couple the numerical and experimental investigation of the com-

bustion equipments.
3.1 Experimental equipment
Three systems have been taken into account for the present Thesis: a lab-scale
and two semi-industrial flameless systems. The lab-scale burner (Figure 3.1
(a)) has been developed at the Politecnico di Milano, by the group headed by
Prof. Rota. Experimental data regarding the flameless combustion of vari-
ous gaseous fuels (CH4 , CH4 -H2 and H2 ) have been collected and provided as
experimental reference for the development of a suitable modeling approach.
Interestingly, the experimental data on the lab-scale burner also include infor-
mation regarding the transition between the flame and flameless combustion
regimes. From a modeling point of view, this has been extremely important to
determine the different requirements, in terms of the computational approaches,
of the two combustion regimes. The mathematical modeling of this system will
be presented in Chapter 6.
The first semi-industrial burner is a self-recuperative burner manufactured
by WS and installed at the ENEL Ricerca facilities of Livorno, Italy. The
burner, known as FLOX® (Figure 3.1 (b)), achieves the flameless conditions
through a specific internal burner aerodynamic, characterized by a massive
recirculation of the flue gases in the reaction region, enclosed by a flame tube.
Two experimental campaigns have been carried out in collaboration with ENEL
Ricerca, to investigated the effect of hydrogen addition to the fuel (natural
gas) on the combustion regime and NO emissions from the burner, at different
operating conditions, i.e. moderately lean and ultra-lean. The experimental
and numerical investigation of the FLOX® burner is discussed in Chapter 5.
The third system investigated is the SOLO Stirling 161 Cogeneration Unit
(Figure 3.1 (c)), developed by SOLO® and installed at the ENEL Ricerca fa-
cilities of Livorno. Similarly to the FLOX® burner, the flameless combustion
regime is achieved through a particular internal aerodynamic which allows the
entrainment of large amounts of flue gases in the reaction region before reac-
tions take place. The peculiarity of the system is that the flue gases cross a
finned heat exchanger to heat a helium stream, involved in a Stirling cycle.
Therefore, the unit is used for the micro-generation of thermal and electrical
power (CHP). An experimental campaign has been carried out on this system,
in collaboration with ENEL Ricerca, to analyze the burner behavior at differ-
ent operating conditions and with hydrogen in the fuel. The experimental and
numerical investigation of the SOLO Stirling burner is discussed in Chapter 7.
50
3.1. Experimental equipment
(a)
(b)
(b)
Figure 3.1: Lab-scale burner (a), FLOX® burner with air preheater (b) and
SOLO Stirling 161 Cogeneration Unit (c).
51
Figure 3.2: Integrated CFD–experimental approach.
3.2 Approach
The characterization of the systems described in Section 3.1 is quite complex, as

they show non trivial interactions between different chemical and physical pro-
cesses. Therefore, an integrated approach based on a constructive combination
of Computational Fluid Dynamics (CFD) and experiments has been adopted
(Figure 3.2).
CFD is an essential tool for the investigation of complex systems, as it al-
lows to carry out “virtual prototyping” and “virtual testing” activities, without
the extremely high cost and time associated with laboratory testing of com-
ponents and complete systems. Therefore, CFD has represented an effective
resource for characterizing the system operations at a resolution which is not
provided by any experimental campaign, and for proposing possible design im-
provements. Moreover, CFD has been successfully exploited to help planning
and understanding the outcomes of the experimental campaigns.
On the other hand, it is not possible to state that the so-called Computer
Aided Engineering (CAE) is, to date, a self-sufficient science. In particular,
users and developers of computational tools face today the critical issue of
determining how the confidence in modeling and simulations should be criti-
cally assessed, to effectively help decision making in new design and regulation.
Therefore, the availability of experimental data has been crucial for the present
Thesis, to provide an essential reference for model validation.
The validation of the computational models have been carried out using
the Verification and Validation (V&V) methodology, introduced in Chapter
2. V&V provides a systematic methodology for building and identifying the
confidence in the numerical simulations. Several ingredients contribute to a
complete V&V activity, as pointed out in the next Section.
52
3.3. Verification and Validation
Figure 3.3: Verification and Validation: objectives to quantify and tools.
3.3 Verification and Validation

The definition of Verification and Validation has been already proposed in
Chapter 2, according to the recommendations of AIAA [51]. In practice, verifi-
cation addresses the accuracy of the numerical solution produced by the com-
puter code as compared to the exact solution of the conceptual model. On
the other hand, validation addresses the accuracy of the conceptual model as
compared to the real world, i.e. experimental measurements (Figure 3.3).
Verification is composed of two types of activities: code verification and so-
lution verification. Code verification basically deals with assessing the adequacy
of the numerical algorithms and the fidelity of the computer programming,
to detect shortcomings in the numerical algorithms or errors in the computer
code, such as coding mistakes. Such activity is generally carried out comparing
the numerical solution with classical analytical and exact solutions, using the
Method of Manufactured Solutions (MMS) [52]. This analysis was not carried
out during the present Thesis as it was assumed that a proper code verification
had been conducted on the computational tools used, which are commercial
CFD softwares.
Attention has been focused on the second activity in verification, solution
verification, which basically addresses the estimation of the numerical error,
using a posteriori methods based on Richardson extrapolation. Richardson ex-
trapolation estimates the numerical error based on the analysis of two or more
numerical solutions, obtained on grids of different spatial or temporal resolu-
53
Figure 3.4: Two phases of validation activities
tion. The method can be applied to any discretization procedure for differential
or integral equations. Logically, code verification and numerical error estima-
tion should be completed before model validation activities are conducted, to
ensure that the computational results are not distorted or polluted by coding
errors or by large numerical solution errors.
Model validation can be regarded as a two phases activity, as indicated
in Figure 3.4. First, the quantitative comparison of computational and experi-
mental results is carried out for the output quantity of interest; then, a criterion
is established to determine whether or not the observed level of agreement is
acceptable. The accuracy of the solutions is quantified by means of validation
metrics (Section 3.3.3), which measure the agreement (or difference) between
model and reality, taking into account also the experimental uncertainty.
It is not always possible to carry out an exhaustive validation activity for
the object of the modeling activity. For example, the systems under investi-
gation in the present Thesis (Figure 3.1) are characterized by a very limited
availability of experimental observations, due to their industrial characteris-
tics. Macro indicators, such as pollutant emissions and walls temperature, are
usually provided and no information regarding the internal flame structure is
known. Moreover, an estimate of the experimental uncertainty is not always
provided and hypothesis must be made before performing model validation.
From the above considerations it follows that hierarchical approaches must
be pursued for the validation of the numerical modeling of very complex sys-
tems, as the one investigated in the present Thesis (Section 3.1). Therefore, a
validation process can be built, recognizing that the system of interest may be
decomposed into several blocks, each of them describing physical processes rele-
vant to the complete system, but characterized by different levels of complexity
and experimental data availability.
54
3.3.1 Validation hierarchies

Validation hierarchies [1] provide a methodology for the systematic validation
of complex systems. The purpose of a validation hierarchy is to help identi-
fying possible separation of coupled physics and levels of complexity for the
engineering system of interest, so that numerical simulations can be performed
at different levels to ultimately lead to the validation of a comprehensive math-
ematical model. The approach recognizes that the quantity and accuracy of
experimental data are dramatically affected by the complexity of the investi-
gated system; therefore, V&V is carried out on single computational modules,
progressively coupled to ultimately characterize the intended use of the simu-
lation tool, in other words, the full system to be simulated.
In the present context, the complex system under study is represented by a
flameless system, which needs to be characterized in terms of pollutants emis-
sions. In principle, it is possible to perform a decomposition of all the relevant
physics for such system to derive a hierarchy as the one depicted in Figure
3.5. As we move down from the apex of the hierarchy, the technical complexity
at each level decreases; moreover, the amount of experimental data available
for comparison increases and the related uncertainties decrease. Data at the
highest level of the hierarchy can be directly applied to the intended use, al-
though they are limited in scope and accuracy. Therefore, validation at this
level is often qualitative, due to the lack of quantitative data. The follow-
ing level in the hierarchy consists of benchmark cases, which provide accurate
experimental data for validation from simplified fully coupled problems. The
third level in the hierarchy is represented by unit problems, characterized by
a large availability of accurate and reliable experimental data. The last level
in the hierarchy is represented by the molecular processes, which split the unit
physical models in their fundamental components. The high fidelity data avail-
able at the unit problem level is also available at this level. It is noteworthy
to mention that all the bricks of the hierarchy must contain features which are
relevant to the intended, full-scale, application. This explain, for example, the
choice of a benchmark case such as the jet issuing in a hot co-flow (JHC) [46],
because such system shows features which are relevant for the characterization
of turbulence-chemistry interactions in flameless combustion regime.
The objective of the present Thesis is represented by the subsystem cases
level in Figure 3.5, corresponding to the combustion systems described in Sec-
tion 3.1. The development of a validated computational approach at this level
can be directly extended to the complete system level, for the characteriza-
tion of a general industrial application employing flameless combustion, e.g. a
furnace equipped with flameless burners.
The main challenge related to the numerical modeling of flameless com-
bustion is represented by the complex turbulence-chemical interactions which
characterize such combustion regime [45, 78, 79]. Thus, following the con-
ceptual scheme in Figure 3.5, a fundamental study on turbulence-chemistry
55
Figure 3.5: V&V hierarchy for a flameless system.
interactions in turbulent reacting flows has been carried out during the present
Thesis, in the framework of a collaboration with the research group headed
by Prof. Philip J. Smith at the University of Utah, Salt Lake City, USA.
Such activity has exploited the availability of high fidelity experimental data
on reference systems for the combustion community, to develop an approach
for the numerical modeling of turbulent reacting flows (Chapter 2). In par-
ticular, Figure 3.5 shows that, beside the well known Sandia flames [80], the
JHC system [46] has been taken into account, to better understand the nature
of turbulence-chemistry interactions in flameless combustion. The conclusions
drawn from this activity have been conveyed and exploited to develop a suit-
able computational approach for the systems at the higher levels. The proposed
methodology, its applications and results are presented in Chapter 4.
In the following Sections, a brief description of the verification and valida-
tion tools available at each level of the validation hierarchy is provided. In par-
ticular, methods for solution verification, based on Richardson extrapolation,
and validation metrics for the comparison between simulations and experiments
are discussed.
3.3.2 Solution Verification

The solution verification methods discussed in this Section are a posteriori
methods based on the so-called Richardson extrapolation, which estimates the
numerical error comparing two or more numerical solutions obtained with grids
of different resolution. Richardson extrapolation techniques are based on the
56
following expansion for the Partial Differential Equation (PDE) of interest:
F0 = Fi + αhpi + Θ hp+1 (3.1)

where F0 is the exact value of the system response quantity (SRQ), Fi is the
SRQ model solution with grid spacing hi , α is a constant and p is the conver-
gence order. If Fi and p are known, then two numerical solutions with different
grid sizes (different choices of h) are required to compute the two unknowns F0
and α in Eq. 3.1. Then, the discretization error can be estimated from the two
numerical solutions. If p is not known a priori, three grids are needed to obtain
an estimate of the convergence order, p, F0 and α.
For any grid triplet or series of three grids, with a constant grid refinement
ratio, rh = h2/h1 = h3/h2 , we can calculate p, neglecting higher order terms, as:

log FF01 −F
−F2
3
p= . (3.2)
log (rh )
More general methods for evaluating p for non-constant rh are discussed dis-
cussed by Roache [52]. Based on the empirical order of convergence p, an
estimate of the exactly converged solution, F0∗ , is [52]:
F1 − F2
F0∗ = F1 + . (3.3)
rhp − 1
A numerical error estimate for grid i can be, then, defined as:
δRE = E1 = Fi − F0∗ = αhpi . (3.4)

A percent error estimate based on the extrapolated exact solution is:
E1
E= · 100%. (3.5)
F0∗
Based on the notation above, Logan and Nitta [81] proposed several meth-
ods for the estimation of the numerical error. Among them, two methods
were selected to carry out solution verification for the cases investigated in the
present Thesis, methods 3 and 4:
• Method 3. According to this method, p is found to maximize the cor-

relation coefficient R2 of the linear fit of Fi versus hpi . Then, a safety
factor Fs = 3 is assumed for the estimation of a Grid Convergence Index
(GCI), given by Eq. 3.4 multiplied by Fs . The resulting GCI represents
a 1-sigma estimate of the numerical solution uncertainty, Usver .
• Method 4. This method is one step simpler than Method 3. Fi is plot

versus hpi , choosing first p = 1 and then p = 2. The maximum of the two
uncorrected δRE is taken as the 1-sigma estimate of Usver .
57
Once a verification activity is completed, the level of agreement between com-

putational results and experimental data is carried out by means of quantitative
metrics.
3.3.3 Assessing the level of agreement: validation metrics

The comparison of experimental data and computational results is usually per-
formed graphically. The quantity of interest (SRQ) is plotted together with
the experimentally measured response over a range of one or more input pa-
rameters. Then, the computational model is considered validated if it generally
agrees with the experimental data. However, graphical comparisons cannot pro-
vide any quantification of the numerical solution error or quantification of the
computational uncertainty, due to uncertain boundary conditions and variabil-
ity in modeling parameters. Moreover, graphical comparisons cannot indicate
how the agreement between measured and predicted data varies over the range
of the independent variables of interest. Therefore, efforts have been spent to
develop tools for comparing computational and experimental results. In the
present Thesis, the validation metrics proposed by Oberkampf and Barone [82]
are taken into account.
3.3.3.1 Definition of a confidence interval for the experimental data

The first step in the creation of a validation metric is represented by the deter-
mination of a confidence interval for the experimental data, which contains the
observed mean value of the measured population within a defined confidence.
For an arbitrary number of experiments, it is possible to define a standardized
random variable T as:
x−µ
T = √ (3.6)
s/ n
where x is the sample mean, µ is the true mean, s is the sample standard
deviation and n is the number of experiments. It can be shown that T has
a probability distribution called student t-distribution only if x comes from a
normal distribution. The t-distribution is governed by the parameter ν = n−1,
which represent the degrees of freedom of the population. When the number of
experiments is large (n ≥ 16), ν is large and the distribution tend to approach
the normal distribution. For lower values of n, the t-distribution is characterized
by a lower peak and fatter tails with respect to the normal distribution, denoting
higher probabilities far from the sample mean, due to the reduced number of
observations.
It is possible to define a confidence interval for µ , the true mean of the
population, as follows:

s s
µ ∼ x − t 2 ,ν √ , x + t 2 ,ν √
α α (3.7)
n n
58
where t 2,α ,ν is the 1 − α2 quantile of t-distribution with ν degrees of freedom.

Therefore, it is possible to state that the true mean is contained in the interval
given by Eq. (3.18) with a confidence, C, equal to 100 · (1 − α) %. Common
values for C are 0.9, 0.95, 0.98 and 0.99 which correspond to values of α equal
to 0.1, 0.05, 0.02 and 0.01, respectively.
3.3.3.2 Validation metrics based on confidence intervals

The construction of an error validation metric requires the evaluation of two
quantities: the estimated error and the interval containing the true error. The
estimated error of the computational model can be defined as:
e = ym − y e
E (3.8)
where ym is the value of the SRQ provided by the computational model and y e
is the sample mean based on n experiments,
n
1X i
ye = ye . (3.9)
n
i=1
In Eq. (3.9) ye1 , ye2 , . . . , yen are the measured SRQ from each experiment. Simi-
larly to the estimated error, we can define a true error as:
E = ym − µ (3.10)
where µ is the true mean of the population. Following Eq. (3.7) and changing
the notation from x to y e , it is possible to define a confidence interval for µ in
inequality form as:
s s
y e − t α2 ,ν √ < µ < y e + t α2 ,ν √ (3.11)
n n
where s is the sample standard deviation,
" n
#1
2
1 X i 2
s= ye − y e . (3.12)
n−1
i=1
Multiplying Eq. (3.11) by −1 and adding ym to each term, we get:

s s
ym − y e + t 2,α ,ν √ > ym − µ > ym − y e − t 2,α ,ν √ . (3.13)
n n
Employing the expressions for the true error (3.10), and the estimated error
(3.8), the following expression for the confidence interval of the true error is
obtained:

s e s
E ∼ E − t 2 ,ν √ , E + t 2 ,ν √
e α α (3.14)
n n
59
With respect to the metric defined in Eq. (3.14), it is important to mention

the following characteristics:
• The confidence intervals are defined solely on the basis of the experimental
data, since the true mean is the real unknown of the problem.
• The estimated confidence intervals are symmetric around the estimated

error. The amplitude of the interval decreases with the number of exper-
√
iments by a factor equal to 1/ n. However, for large values of n, the
confidence interval is valid regardless of the type of probability distribu-
tion representing measurement uncertainty
• For small numbers of experiments, the measurement uncertainty is as-

sumed to be normally distributed. Although this is a very common
assumption in experimental uncertainty estimation, it is rarely demon-
strated to be true.
Eqs. (3.8)-(3.14) can be easily extended to the case of a system quantity re-
sponse monitored over a range of the input variable. The expression of the
estimated error as a function of the input variable x is:
e (x) = ym (x) − y e (x)

E (3.15)
where ym (x) is the computed SRQ and ȳe (x) is the estimated mean from
multiple experimental realizations,
n
1X i
y e (x) = ye (x) . (3.16)
n
i=1
The 100 (1 − α) % confidence interval for the true mean, µ, as a function of x

is given by:

s (x) s (x)
µ (x) ∼ y e (x) − t 2,α ,ν √ , y e (x) + t 2,α ,ν √ (3.17)
n n
with
" n
#1
2
1 X i 2
s (x) = ye (x) − y e (x) . (3.18)
n−1
i=1
Finally, the 100 (1 − α) % confidence interval for the true error is:

s (x) e s (x)
E (x) ∼ E (x) − t 2,α ,ν √ , E (x) + t 2,α ,ν √
e . (3.19)
n n
The above equations hold also when the SRQ is evaluated at discrete values of
the input variable x, xi , i = 1, 2, . . . , n.
60
3.3.3.3 Global metrics

Eqs. (3.15), (3.17) and (3.19) allow the construction of validation metrics as
a function of x; however, it is sometimes desirable to provide global metrics,
able to summarize the level of agreement for a large number of computational
models and experimental data.
It is possible to define an average absolute error, normalizing the absolute
error by the estimated sample mean and integrating, or summing, over the
range of the input variable:
Z xu e
xu

ym (x) − ȳe (x)
E Z
e
1 1 E (x)
= dx = dx

ȳe (xu − xl ) ȳe (x) (xu − xl ) xi ȳe (x)

xi
avg
(3.20)
or, for discrete functions of x,
p p

E
e
1 X ym (xi ) − ȳe (xi ) 1 X Ee (xi )
= = p . (3.21)
ȳe p ȳe (xi ) ȳe (xi )

avg i=1 i=1
The confidence interval to be associated with this average relative error is the
average confidence interval normalized by the value of the estimated experi-
mental mean over the range of the data:

t α2 ,ν Z xu
CI s (x)
= √ dx (3.22)
ȳe
avg (xu − xl ) n xi ȳe (x)
p
1 t α2 ,ν X s (xi )

CI

ȳe = √ ȳe (xi ) . (3.23)
avg p n
i=1
It is possible to define also a maximum value of the relative error over the range
of the data, to capture localized large errors. We can define a maximum relative
error metric as:

E
e
ym (x) − ȳe (x)
= max (3.24)
ȳe ȳe (x)

xl ≤x≤xu
max

E
e
ym (xi ) − ȳe (xi )
= max . (3.25)
ȳe ȳe (xi )

xi ,i=1,2,...,p
max
The confidence interval to be associated with the maximum error metric is the
confidence interval normalized by the experimental mean, at the point where
the maximum deviation occur, x̂ :
61
t α2 ,ν

CI s (x̂)

ȳe = √ ȳe (x̂) .
(3.26)
max n
3.4 Summary
The present Section has defined the methodology adopted in the present Thesis
for the investigation of hydrogen based fuel combustion in lab-scale and semi-
industrial systems. In particular, recognizing the complex interactions which
characterize such systems, a hierarchical approach has been proposed for the
development of a comprehensive computational approach. This has allowed to
better identify the objectives of the present Thesis:
• Fundamental study on turbulence-chemistry interactions in turbulent re-

acting flows. This step has exploited the availability of high fidelity exper-
imental and numerical data on reference systems, to identify the param-
eters controlling the evolution of a reacting system and develop optimal
combustion models. Moreover, the investigation of operating conditions
characteristic of the flameless combustion regime, i.e. JHC burner (Sec-
tion 3.3.1), has allowed to support the simulation activity carried out
on the advanced burners for hydrogen-enriched fuel combustion (Section
3.1), in particular from the point of view of turbulence-chemistry inter-
action modeling.
• Numerical simulation of lab-scale and semi-industrial burners, for the

characterization of the flameless combustion regime and the development
of a simplified NO formation approach.
Finally, tools for the verification and validation of numerical results have been
discussed, to provide a quantitative basis for the comparison between simula-
tions and experiments.
62
Chapter 4
Principal Components Analysis

for turbulence-chemistry
interaction modeling
As discussed in Chapter 2, turbulent combustion modeling is a very broad

subject and includes a wide range of coupled physics. Our primary interest is
focused on combustion models, adopted in the framework of RANS and LES ap-
proaches to close the reacting scalars source terms. Such models must provide
an adequate coupling of turbulent mixing and chemical reactions for the unre-
solved scales. Moreover, an accurate kinetic description is needed in situations
where finite-rate chemistry effects may be relevant, as in flameless combustion
regime [78, 79, 45]. However, detailed combustion mechanisms for fuels as sim-
ple as methane involve 53 species and 325 reactions [63] and the number of
species and reactions dramatically increases with the molecular weight of the
hydrocarbon fuel [83]. Therefore, solution of the species transport equations
for turbulent reacting system can become very computationally intensive, if a
reaction rate approach is adopted and no simplification is made.
The reduction of the number of species equations to be solved can be ac-
complished in two ways:
• Reduction of the kinetic mechanism [84, 85, 86]. This approach is based
on the analysis of the dominant reaction rates at the conditions of inter-
est and proceeds through the elimination of species and reactions in the
original kinetic mechanism, ultimately leading to a reduced set of species
equations to be solved.
• State space parametrization. This approach relies on the assumption

that the thermodynamic state of a reacting system relaxes onto a low-
dimensional, strongly attracting, manifold in chemical state space [87,
88]. The thermochemical state of a single-phase reacting fluid having Ns
species is uniquely determined by Ns + 1 parameters (T , p, and Ns − 1
63
Chapter 4. Principal Components Analysis for turbulence-chemistry
species mass fractions). Yet, if a set of “optimal” variables is identified, the

whole thermochemical state can be re-parametrized with a lower number
of variables, which nevertheless must provide a satisfactory approxima-
tion of the system in a lower dimensional space [87]. In the context of
turbulent non-premixed flames, a widely used approach is represented by
the flamelet model .(Chapter 2). The model adopts two parameters for
the parametrization of the thermochemical state, the mixture fraction f
and the scalar dissipation rate, χ. The choice of f is particularly appeal-
ing, as this parameter is a conserved scalar if all species diffusivities are
equal [88], and it does not require any source term closure.
The present Chapter focuses on the second of these approaches, the parametriza-
tion of the thermochemical state by a small number of parameters based on the
existence of a low-dimensional manifold. Most of the existing models exploiting
the existence of such manifolds are based on the a priori prescription of the
manifold dimensionality. For example, the flamelet model [76, 89] assumes that
the system can be satisfactory described by means of two parameters. How-
ever, such an approach restricts the subspace that thermochemistry may access,
without providing any quantitative error analysis. Indeed, as mixing and reac-
tion timescales increasingly overlap, the dimensionality of a manifold increases,
as does the error associated with a parametrization of fixed dimensionality [88].
Such consideration has prompted our interest towards the development of a
methodology to automate the selection of the optimal basis for the representa-
tion of the manifolds in thermochemical space. Principal Components Analysis
(PCA) [90, 91] offers this potential, as it provides a rigorous mathematical
formalism for reducing the dimensionality of a data set consisting of a large
number of correlated variables, while retaining most of the variation present in
the original data. The reduction is achieved by transforming to a new set of
variables, called the principal components (PCs), which are uncorrelated and
ordered so that the first few account for most of the variation present in all the
original variables. PCA provides an optimal representation of the system based
on q optimal variables, the PCs, which are linear combination of the Ns + 1
primitive variables T , p, and Yi . The linearity of the method is a particularly
appealing aspect since, once the reaction variables are identified, a few linear
combinations of the original variables could be transported in a numerical simu-
lation, if a proper closure for the source terms is employed. Nevertheless, since
the reaction variables provided by PCA are not conserved scalars, an a pri-
ori analysis of the ability of principal components to parametrize their source
terms must be assessed, as a mandatory requirement for the generated manifold
method. One of the main advantages of PCA lies in the potential to obtain
the principal components from a target system and to apply them to a similar
system. This potential could remove one of the main drawback which, to date,
affects PCA. In fact, the derivation of the manifold model via PCA requires
the availability of a data set for the extraction of principal components.
64
4.1. Definition and derivation of Principal Components
In the following, a background on PCA is provided. In particular, the

definition and derivation of the PCs, the criteria used for selecting optimal
parameters, or variables, from large data sets and the approaches to help PCs
interpretability and deal with highly non linear systems will be discussed. Then,
various examples of the application of PCA to turbulent reacting systems will
be presented, to investigate the feasibility of PCA for the identification of the
low-dimensional manifolds in thermochemical state. Finally, the possibility of
exploiting PCA as a predictive model approach will be investigated.
All the analysis carried out in this Chapter have been performed with a
MATLAB® code written on purpose, available under request. The results
have been published in [92, 93].
4.1 Definition and derivation of Principal Compo-

nents
A primary goal when dealing with multivariate data is to reduce their dimen-
sionality to the smallest number of meaningful dimensions, in order to help
data exploration and any further processing. Principal Components Analysis
(PCA) can be successfully exploited for this purpose. PCA was first introduced
by Pearson in the early 1900’s [94]. A formal treatment of the method is due
to Hotelling [95] and Rao [96].
Suppose that X is a vector of p random variables, i.e. X = (x1 , x2 , . . . , xp ),
with mean µ and covariance matrix Σ. The (i, j)th element of Σ represents the
covariance between the ith and jth variables of X, if i 6= j, or the variance of the
jth element of X, if i = j. PCA is concerned with finding a few (<< p) derived
variables, called Principal Components (PCs), which nevertheless preserve most
of the information present in the original data. The PCs are linear combinations
of the original variables; moreover, they are uncorrelated (i.e. orthogonal) and
derived so that the variance on the jth component is maximal.
The first PC of X is defined as the linear combination:
z1 = Xa1 . (4.1)
0
To determine z1 , a vector a1 is sought so that var (z1 ) = a1 Σa1 is maximized,
0
subject to the constraint a1 a1 = 1. If we adopt the standard approach of
Lagrange multipliers to solve this constrained problem, we need to maximize:
0
0
a1 Σa1 − λ a1 a1 − 1 (4.2)
where λ is a Lagrange multiplier. Differentiating with respect to a1 gives:
Σa1 − λa1 = (Σ − λIp ) a1 = 0 (4.3)

where Ip is the (p x p) identity matrix. Thus, λ is an eigenvalue of Σ and a1 is
the corresponding eigenvector. It is easy to proof that the eigenvector a1 which
65
maximizes the variance of z1 is the one corresponding to the largest eigenvalue

of Σ, being:
0 0 0
a1 Σa1 = a1 λa1 = λa1 a1 = λ = λ1 . (4.4)
0
The second PC, z2 = Xa2 , maximizes the variance var (z2 ) = a2 Σa2
subject to the constraints cov (Xa1 , Xa2 ) = 0 (z1 and z2 uncorrelated) and
0
a2 a2 = 1. Being
0 0 0 0
cov (Xa1 , Xa2 ) = a1 Σa2 = a2 Σa1 = λ1 a1 a2 = λ1 a2 a1 , (4.5)
any of the equations
0 0 0 0
a1 Σa2 = 0 a2 Σa1 = 0 a1 a2 = 0 a2 a1 = 0 (4.6)
could be used to specify no correlation between z1 and z2 . Choosing arbitrarily

the last expression in Eq. (4.6), the quantity to be maximized becomes:
0
0 0
a2 Σa2 − λ a2 a2 − 1 − φa2 a1 (4.7)
where and φ are Lagrange multipliers. Differentiating with respect to a2 and

pre-multiplying by a1 , we get:
0 0 0
a1 Σa2 − λa1 a2 − φa1 a1 = 0 (4.8)
which reduces to
φ = 0, (4.9)
being
0 0
a1 Σa2 = λa1 a2 (4.10)
due to the constraint of z1 and z2 being uncorrelated. Then, Eq. (4.8) reduces
to:
Σa2 − λa2 = 0. (4.11)
Then, λ is once more an eigenvalue of Σ and a2 is the corresponding eigenvec-

0
tor. Again, a2 Σa2 = λ. Thus, assuming that Σ has all different eigenvalues,
λ is the second largest eigenvalue of Σ, λ2 , and a2 is the corresponding eigen-
vector.
0
In general, the kth PC of is zk = Xak and var (zk ) = ak Σak = λk , where
λk is the kth largest eigenvalue of Σ and ak is the corresponding eigenvector.
66
4.2. Sample PCA
4.2 Sample PCA

In Section 4.1, the definition and derivation of PCs have been discussed for an
infinite population of measures. In practice, a random sample of n observations
of the p variables is available, so that Xi = (xi1 , xi2 , . . . , xip ) represents the ith
observation from the data set. Thus, the data available for PCA is a (n x p)
data matrix and an unbiased estimator of Σ, S 1 , is employed.
For a single observation of X, Xi , zi1 is given by:
zi1 = Xi a1 , i = 1, 2 . . . , n (4.12)
where the vector of coefficients a1 is chosen to maximize the variance
n
1 X
(zi1 − z 1 )2 , (4.13)
(n − 1)
i=1
0
subject to the constraint a1 a1 = 1. Then, for the second PC:
zi2 = Xi a2 , i = 1, 2 . . . , n (4.14)
where a2 is chosen to maximize the sample variance of z12 , subject to the
0
constraints a2 a2 = 1 and cov (Xa1 , Xa2 ) = 0.
Continuing the process in an obvious manner, zk = Xak is the kth sam-
ple PC (k = 1, 2, . . . , p) and zik is the score for the ith observation on the
kth sample PC. If the derivation of Section 4.1 is followed, but replacing popu-
lation quantities with sample variances and covariances, then it turns out that
the sample variance of the kth sample PC is lk , the kth largest eigenvalue of
the sample covariance matrix S, and that ak is the corresponding eigenvector
of S.
Let Z be the (n x p) matrix of PCs scores, with (i, k)th element equal to
zik ; then, Z and X are related by
Z = XA (4.15)
where A is the (p x p) orthogonal matrix whose columns are the eigenvectors of

S. Here we assume, for simplicity, that the variables have zero mean, otherwise
the mean of each variable is subtracted from the columns of X before PCA is
applied. Then, the sample covariance matrix, S, of X can be defined as:
1 0
S= X X. (4.16)
n−1
Recalling the eigenvector decomposition of a symmetric, non-singular matrix,
S can be decomposed as:
1
The matrix S represents the approximation of Σ for a finite population, i.e. the random
sample consisting of n observations for p variables.
67
0
S = A LA (4.17)
where L is a (p x p) diagonal matrix containing the eigenvalues of S in de-
scending order, l1 > l2 > . . . > lp .
The linear transformation given by Eq. (4.15) simply recast the original
variables into a set of new uncorrelated variables, whose coordinate axes are
described by A. Then, the original variables can be stated as a function of the
PCs as:
0
X = ZA (4.18)
0
being A orthonormal and, hence, A−1 = A . This means that, given Z, the
values of the original variables can be uniquely recovered. However, the main
objective of PCA is to replace the p elements of X with a much smaller number,
q, of PCs, which nevertheless discard a small fraction of the variance originally
contained in the data. If a subset of size q << p is used, the truncated PCs are
defined as:
Zq = XAq . (4.19)
Eq. (4.19) can be inverted to obtain:
0
Xq = Zq Aq . (4.20)
The linear transformation provided by Eq. (4.20) is particularly appealing
for size reduction in multivariate data analysis due to some optimal properties
of the PCA transformation, described in the following text.
4.2.1 Optimal properties of the PCA reduction

It can be shown [90] that PCA satisfies the following optimal properties:
• Property 1 : For any integer q, 1 ≤ q ≤ p, consider the orthonormal

0
transformation Z = XB, where B is a (p x q) matrix. Let Sz = B LB
be the variance-covariance matrix for Z. The, the trace of Sz , tr (Sz ),
is maximized by taking B = Aq , where Aq contains the first q columns
of A. Property 1 emphasizes that the PCs explain, successively, as much
as possible of the total univariate variance in the original data, being
tr (Sz ) = tr (S)
• Property 2. Consider the orthogonal transformation Z = XB. Then,

tr (Sz ) is minimized by taking B = A∗q , where A∗q consist of the last q
columns of A. The statistical implication of property 2 is that the last few
PCs are not simply unstructured leftovers after removing the important
PCs. Being the variances of the last PCs small, they can help to detect
unsuspected near-constant linear dependencies among the element of X,
68
4.2. Sample PCA
to select a subset of variables from X in regression analysis or to detect

outliers from a data set.
• Property 3. The sample covariance matrix, S, can be expressed as (spec-

tral decomposition):
0 0 0
S = l1 a1 a1 + l2 a2 a2 + . . . + lp ap ap . (4.21)
This result shows that we can decompose the whole covariance matrix
into decreasing contribution due to each PC.
• Property 4. Consider the orthogonal transformation Z = XB. Then,

the determinant of Sz , det (Sz ), is maximized by taking B = Aq . The
statistical importance of this property follows because the determinant of
a covariance matrix, called generalized variance, can be used as a simple
measure of spread for a multivariate random variable.
• Property 5. Each element of X can be predicted by a linear function of Z,

Z = XB. If σj2 is the residual variance in predicting xj from Z , then
Pp
j=1 σj is minimized by taking B = Aq . The statistical implication
2
of property 5 is that Eq. (4.20) is the best linear predictor of X in a

q-dimensional subspace, in terms of squared prediction error.
4.2.2 Data preprocessing: centering and scaling

As it was anticipated in Section 4.2, data are usually centered before PCA is
carried out. When the variable means are subtracted from the data sample, all
the observations are converted to fluctuations, thus leaving only the relevant
variation for analysis. Moreover, when working with centered variables, cen-
tered PCs are obtained. Centering is usually used with all the scaling criteria
described below.
Scaling is an essential operation when the elements of X are in different
units or when they have very different variances. These aspects have both
to be faced when analyzing the thermochemical state of a reacting system
since temperature and species concentrations have different units. Moreover,
temperature may range from ambient conditions to thousands of degrees while
species mass fractions vary between zero and one. Besides, even among species
mass fractions, there may be need for scaling. For example, radicals appear in
small concentrations and their mass fractions may range from zero to something
far less than one (i.e. 10−3 − 10−6 ), while major species mass fractions range
from 0 to 1. Taking into account centering, it is possible to define a scaled
variable, xej as:
xj − xj
x
ej = (4.22)
dj
69
where xj = n1 ni=1 xij , j = 1, 2, . . . , p, while dj is the scaling parameter relative

P
to variable xj . In matrix form, Eq. (4.22) becomes:
f = X − X D −1 (4.23)

X
where D is the diagonal matrix containing the scaling parameters, dj . When
scaling is applied, Eqs. (4.15)-(4.19) are modified as:
−1
0
1
X − X X − X D −1

S= n−1 D
Z = X − X D −1 A . (4.24)
−1
Zq = X − X D Aq
The choice of the scaling parameters is very important, and has a potentially
strong impact on the resulting eigenvectors. The following choices are available:
1. Auto scaling, also called unit variance scaling. It is commonly applied and
uses the standard deviation, sj , as the scaling factor. After auto scaling,
all the elements of X have a standard deviation equal to 1 and therefore
the data is analyzed on the basis of correlations instead of covariances
2. Vast scaling [97]. Vast is an acronym of variable stability scaling and
it is an extension of auto scaling. It focuses on stable variables, the
variables that do not show strong variation, using the standard deviation
and the so-called coefficient of variation as scaling factors. The use of the
coefficient of variation, defined as the ratio of the standard deviation and
the mean: sj/xj , results in a higher importance for variables with a small
relative standard deviation
3. Range scaling. Range scaling adopts the difference between the minimal
and the maximal value, (xj,max − xj,min ), as scaling factor. A disadvan-
tage of range scaling with respect to other scaling methods is that only
two values are used to estimate the range, while for the standard devia-
tion all measurements are taken into account. This makes range scaling
more sensitive to outliers. To increase the robustness of range scaling,
the range could also be determined by using robust range estimators or
after the outliers have been removed.
4. Level scaling. The mean values of the variables, xj , are used as scaling
factors. Level scaling converts deviations from the mean (the mean is
always subtracted) in percentages compared to the mean values. As for
the range scaling, also level scaling can be affected by outliers. Then,
a more robust estimator of the mean, the median, could be used or the
mean could be determined after outlier removal. Level scaling can be
used when large relative changes are of specific interest. However, in
the case of the thermochemical state of a system, this could lead to an
overestimation of the role of chemical species which appear in very small
concentrations, i.e. radicals.
70
4.2. Sample PCA
5. Max scaling. The variables are normalized by their maximum values,

xj,max , so that they are all bounded between zero and one. As for the
range and level scaling, a robust estimator of maximum values or a pro-
cedure for outliers removal should be employed.
As it was pointed out in the above discussion, range, level and max scaling
can be affected by the presence of outliers in the sample X. This problem can
be overcome by means of robust estimators of the quantities of interest (i.e.
median in place of sample average); however, a procedure for the detection and
removal of outlier observations could provide a viable solution to the problem.
PCA can be effectively employed for outlier detection in large data sets.
4.2.2.1 Outlier detection and removal with PCA

Experimental data sets usually contain a few unusual observations. If we refer
to a one-dimensional problem, the outliers can be classified as those observa-
tions which are either very large or very small with respect to the others. In
high dimensions, there can be outliers that do not appear as outlying observa-
tions when considering each dimension separately and, therefore, they will not
be detected using univariate criteria. Thus, a multivariate approach must be
pursued.
The usual procedure for outlier detection in multivariate data analysis is
to measure the distance of each observation, Xi = (xi1 , xi2 , . . . , xip ), from the
data center, using the so called Mahalanobis distance:
0
DM = Xi − X S −1 Xi − X . (4.25)

The observations associated to large values of DM are considered outliers and,

then, they are discarded. However, a procedure must be employed to ensure a
robust estimation of the sample mean. To this purpose, PCA can be effectively
exploited [98].
PCA has long been used for multivariate outlier detection. The sum of
squares of the PCs, standardized by the eigenvalue size, equals the Mahalanobis
distance for an observation i:
p 2
X z2 2
zi1 z2 zip
ik
= + i2 + . . . + = DM . (4.26)
lk l1 l2 lp
k=1
The first few principal components have large variances and explain most of
the variation in X. Therefore, these major components are strongly affected
by variables with relatively large variances and covariances. Consequently, the
observations that are outliers with respect to the first few components usually
correspond to outliers on one or more of the original variables. On the other
hand, the last few principal components represent linear functions of the orig-
inal variables with minimal variance. These components are sensitive to the
71
observations that are inconsistent with the correlation structure of the data but
are not outliers with respect to the original individual variables.
Based on the above considerations, the following detection scheme can be
proposed [98]:
1. Multivariate trimming. A fraction γ (0.005-0.01) of the data points char-
acterized by the largest value of DM is classified as outliers and removed.
New trimmed estimators for X and S are then computed from the re-
maining observations. The trimming process can be iterated to ensure
that X and S are resistant to outliers.
2. Principal components classifier (PCC). The PCC consists of two func-
z2
tions, one from the major PCs, qk=1 lik , and one from the minor PCs,
P
k
Pp 2
zik
k=p−r+1 lk . The first function can easily detect observations with large
values on some of the original variables; in addition, the second function
helps detect the observations that do not conform to the correlation struc-
ture of the sample. The number of major components, q, is determined
to explain about 50% of the original data variance, while r is chosen so
that the minor components used for the definition of the PCC are those
whose variance is less than 0.20, thus indicating the existence of almost
linear relations among the variables. Based on the PCC definition, an
observation Xi is classified as outlier if:
Pq zik 2 2
zik
(4.27)
Pp
k=1 lk > c1 k=p−r+1 lk > c2
where c1 and c2 are chosen as the 0.99 quantile of the empirical distribu-
z2 2
zik
tions of qk=1 lik
Pp
and .
P
k k=p−r+1 lk
An example of application of the outlined detection scheme in shown in Figure

4.1 and Figure 4.2, with reference to a data set consisting of 62766 observations
of 10 variables [99]. Outliers have been artificially introduced in the experi-
mental data: specifically, 1000 random numbers between 0 and 1 have been
generated and successively scaled using the standard deviation, sj , of the vari-
ables xj . The effect of the outliers on the PCs is very clear from Figure 4.1.
The introduced unusual observation are outliers with respect to the original
variables and they are reflected in the first few PCs; a small cluster of points,
separated from the majority of observations, appears in the plot of the first and
second PCs scores (Figure 4.1 (a)). If the outlier detection scheme is applied,
the introduced outliers are completely removed (Figure 4.1 (b)); in addition,
outliers present in the original experimental data set, affecting the last PCs
scores (multivariate outliers), are also detected with the procedure described.
Outliers must be treated with care as they can strongly affect the determination
of the covariance matrix, thus leading to the identification of false PCs (Figure
4.2).
All the data sets analyzed in the present Chapter have been preprocessed
to remove outlying observations.
72
4.2. Sample PCA
(a) (b)
Figure 4.1: Principal components scores with (a) and without (b) outliers.
(a) (b)
Figure 4.2: Eigenvalues size with (a) and without (b) outliers.
73
Figure 4.3: Size reduction process with PCA.
4.2.3 Choosing a subset of Principal Components or Variables

The major objective in many applications of PCA is to replace the p elements
of X by a much smaller number q of PCs, which nevertheless discard very little
information. Then, it is crucial to determine how small q can be taken without
serious information loss. Several criteria have been proposed and they will be
discussed in the following Sections. Once a criteria for selecting q is adopted,
size reduction can be accomplished, as described by Figure 4.3 for a simple two
dimensional case. The procedure outlined in Figure 4.3 is extremely attractive
for the analysis of turbulent reacting systems, as it provides a mathematical
formalism for the automate selection of the parameters which maximize the
variance in state space.
4.2.3.1 Cumulative percentage of total variance

The most obvious criterion for choosing q is to select a cumulative fraction
of the total variance that the PCs have to account for, i.e. 0.8 or 0.9. The
required number of PCs, q, is then the smallest value of q for which this chosen
percentage is exceeded. As discussed in Section 4.2, the PCs are successively
selected to maximize
Pp their
Pp variance, expressed by the associated eigenvalue lk .
Thus, being k=1 lk = j=1 var (xj ), the desired fraction of the total variance
can be defined as:
Pq
lk
tq = Ppk=1 . (4.28)
k=1 lk
Following the derivation of tq , an appropriate measure of lack-of-fit of the rank

q approximation of X is:
p
n X
X p
X
(xq,ij − xij )2 = lk . (4.29)
i=1 j=1 k=q+1
74
4.2. Sample PCA
For a given number of retained PCs, it is possible to define the variance

accounted for each variable by the retained eigenvectors as:
q √ 2
X ajk lk
tq,j = (4.30)
sj
k=1
where ajk is the weight of the jth variable on the kth eigenvector and sj is the
standard deviation of variable xj .
4.2.3.2 Variance of Principal Components
The rule described in this Section originally applies to correlation matrices,

although it can be easily adapted to covariance matrices.
The idea behind this rule is that if all elements of X are independent, then
the PCs are the same as the original variables and they are characterized by
unit variances in the case of a correlation matrix (e xj = xj/sj ). Thus, any PC
with variance less than 1 explains less variation than the original variables and
can be discarded. According to this rule, also known as Kaiser’s rule [100], only
those PCs whose variances lq exceed 1 are retained. However, a cut-off at lq =
1 could lead to discard important information in some circumstances. In fact,
if a variable from the sample is more or less independent of all others, it will be
characterized by small coefficients in (p − 1) PCs but it will dominate one PC,
whose variance will be close to 1 when using the correlation matrix. However,
deletion will occur if Kaiser’s rule is used, and if, due to sampling variation,
lq < 1. It is therefore advisable to choose a cut-off lower than 1, to allow for
sampling variation. Jolliffe [101] suggests that 0.7 is roughly the correct level.
The rule just described can be easily extended to covariance matrices by taking
as cut-off l∗ equal to the average value, l, of the eigenvalues, or a somewhat
lower cut-off such as l∗ = 0.7l .
4.2.3.3 Broken Stick Model
Frontier [102] proposed a broken-stick method to select the number of PCs. If

we have a stick of unit length, broken into p segments, then it can be shown
that the expected length of the qth longest segment is:
q
1X1
lq∗ = . (4.31)
p k
k=1
This method actually compares the eigenvalues from the observed sample
with the eigenvalues from random data. Based on Eq. (4.31), the observed
eigenvalues are considered interpretable if they exceed lq∗ .
75
4.2.3.4 Scree plot

Another common method to determine the number of PCs is the Scree Plot
(Figure 4.2). This is a simple plot of the eigenvalues sorted in descending order
against their indexes. The number of eigenvalues to retain is based on the
observation of the index q at which the slopes of lines joining the plotted points
are steep to the left of q, and not steep to the right of it. Cattell [103] originally
proposed that the points to the left of the straight line, defined by the smaller
eigenvalues (three components for the data in Figure 4.2), should be considered
important. Afterwards, Cattell and Vogelmann [104] concluded that also the
first eigenvalue to the right of this point should be included (four components
for the data in Figure 4.2). Often the Scree Plot approach is complicated by
either the leak of any obvious break or the possibility of multiple break points.
4.2.3.5 Choosing a subset of Variables

Principal components (PCs) are linear combinations of all variables available.
However, these variables are not necessarily equally important to the formation
of PCs: some of the variables might be critical, but some might be redundant.
Motivated by this fact, it was attempted to link the PCs back to a subset of
the original variables, selecting critical variables or eliminating irrelevant ones.
Such an approach is very appealing as it overcomes one of the major issues
related to PCA. Being linear combination of all the original variables, the PCs
often lack of physical meaning. Therefore, working in terms of the original
variables could be helpful and straightforward. However, it should be noticed
that for a given number, m, of retained variables, an equal number of PCs will
always explain more of the original data variance.
A number of methods exist for selecting a subset of m original variables
which preserve most of the variation in X. Some of them are directly related
to the PCs.
• B4 Forward method [101]. PCA is performed on the original matrix of

p variables and n observation, i.e. size (X) = (n, p). The eigenvalues of
the covariance/correlation matrix are then computed and a criterion is
chosen to retain q of them (l∗ = 0.7l [101]). If p1 components have eigen-
values less than l∗ , the eigenvectors associated with the remaining p − p1
eigenvalues are evaluated starting with the first component. The variable
associated with the highest eigenvector coefficient is then retained from
each of the p − p1 variables, as it is highly correlated with an important
PC. A second PCA is performed such that p − p1 − p2 variables remain.
PCA is then repeated until all the components have eigenvalues larger
than l∗ .
• B2 Backward method [101]. PCA is performed on the original matrix of

p variables and n observation, i.e. size (X) = (n, p). The eigenvalues
76
4.2. Sample PCA
of the covariance/correlation matrix are then computed and a criterion

is chosen to retain q of them (l∗ = 0.7l [101]). If p1 components have
eigenvalues larger than l∗ , the eigenvectors associated with the remain-
ing p − p1 eigenvalues are evaluated starting with the last component.
The variable associated with the highest eigenvector coefficient is then
discarded from each of the p − p1 variables, as it is highly correlated with
an unimportant PC. A second PCA is performed such that p − p1 − p2
variables are discarded. PCA is then repeated until all the components
have eigenvalues larger than l∗ .
• M2 backward method [105]. PCA is performed on the original matrix of

p variables and n observation, i.e. size (X) = (n, p). The eigenvalues
of the covariance/correlation matrix are then computed and a criterion
is chosen to retain q of them. The (n x q) matrix of PCs scores, Zq , is
then evaluated. The goal is to select m (m < p and m ≥ q) variables
from X which preserve most of the data variation. The PCs scores from
the reduced data are denoted by Ẑ. Being q the true dimensionality of
the data, as determined with the PCA analysis, Zq is the true matrix of
scores, also indicated as true configuration, while Ẑ is the corresponding
approximation based on m variables. The discrepancy between the two
configuration is evaluated with a Procrustes Analysis2 . The idea is to
compare the shape of Zq and Ẑ, to establish which set of m original
variables better reproduces the true configuration Zq . In practice, this
consists in the following steps:
– Find the sum of squared differences between corresponding points of

Zq and Ẑ, after they have been matched as well as possible under
translation, rotation and reflection.
∗ Matching under translation is ensured by centering both Zq and
Ẑ
∗ Matching under rotation and reflection is ensured by considering
Zq as the fixed configuration and transforming Ẑ.
– The quantity which is to be minimized in the selection of variables is
the following sum of squared differences between the configurations:
0 0 0 0

M 2 = trace Zq Zq + Ẑ Ẑ − 2ẐQ Zq =
0 0
(4.32)
= trace Zq Zq + Ẑ Ẑ − 2Σ
0 0 0
Q=VU Ẑ Zq = U ΣV (4.33)
0
where Σ is the matrix of singular values from the SVD of Ẑ Zq .
2
In statistics, Procrustes analysis is a form of statistical shape analysis used to analyze
the distribution of a set of shapes.
77
The M2 algorithm employs the following backward elimination procedure

for the retention of m variables from the original data:
1. Initially, set m = p and, for a fixed q, compute the matrix of PCs

scores, Zq .
2. Obtain and store the matrix of PCs scores obtained by deleting in
turn each variable from X.
3. Compute M 2 for each matrix of scores and identify the variable xu
which yields the smallest M 2. Let Ẑu denote the corresponding
matrix of scores.
4. Delete variable xj . Set Zq = Ẑu and return to stage 2 with p − 1
variables. Continue the cycle until only m variables are left.
• McCabe criteria [106]. This approach originates from the observation that
the PCs satisfy a certain number of optimality criteria (Section 4.2.1). A
subset of the original variables that optimizes one of these criteria is
called a set of principal variables by McCabe [106]. Suppose that the
set of variables of X is partitioned into subsets X (1) and X (2) . The
covariance matrix of X can be partitioned as:

S 11 S 12
S= . (4.34)
S 21 S 22
Then, the partial covariance matrix for X (2) given X (1) is:
S 22,1 = S 22 − S 21 S −1
11 S 12 . (4.35)
The criteria proposed by McCabe [106] for the definition of the principal
variables are:
min m
Q
MC1 max |S 11 | = min |S 22,1 | = P k=1 δk
MC2 min tr (S 22,1 ) = min m δ
2 Pmk=1 2k (4.36)
MC3 min
Pr kS 22,1 k = min k=1 δk
MC4 max k=1 ρ2k , with r = min (m, p − m)
where δk are the eigenvalues of S 22,1 and ρk are the canonical correla-
tions between the selected and not selected variables. As McCabe [106]
points out, after the selection of the PVs, S 22,1 represents the information
left in the remaining unselected variables and, then, it is quite plausible
that three of the optimality criteria should be functions of this matrix.
McCabe [106] criteria are very appealing as they satisfy well defined prop-
erties. For instance, criterion MC1 maximizes the variance of the data
explained by the subset of variables, while MC2 and MC3 both minimize
the reconstruction error. However, the criteria rapidly becomes compu-
tationally unfeasible for very large data sets.
78
4.2. Sample PCA
• Principal Features [107]. Using this method the dimension reduction is

accomplished by choosing a subset of the original features that contains
most of the essential information, both in the sense of maximum vari-
ability of the variables in the lower dimensional space and in the sense
of minimizing the reconstruction error. The rows of the eigenvector ma-
trix, Aq , denoted as Vj , represent the projection of the jth variable onto
the lower dimensional space, that is, the q elements of Vj correspond
to the weights of xj on each axis of the subspace. The key observation
of the method is that variables that are highly correlated or have high
mutual information will have similar weight vectors Vj . On the two ex-
treme sides, two independent variables have maximally separated weight
vectors; while two fully correlated variables have identical weight vectors
(up to a change of sign). Therefore, the structure of the rows Vj is first
analyzed, to find the subsets of variables that are highly correlated; then,
a variable is extracted from each subset. The chosen variables represent
each group optimally in terms of spread in the lower dimension, recon-
struction and insensitivity to noise. The principal features algorithm can
be summarized in the following five steps [107]:
1. Compute the sample covariance/correlation matrix, S.

2. Compute the PCs and eigenvalues of S, lk .
3. Choose the subspace dimension q and construct the matrix Aq .
4. Cluster the q vectors, Vj , to m ≥ q clusters using the k-means algo-
rithm [108]. The distance measure used for the k-means algorithm
is the Euclidean distance. Choosing m greater than q is usually
necessary if the same variability of the PCs subset is desired.
5. For each cluster, find the corresponding vector Vj which is closest to
the mean of the cluster. Choose the corresponding feature, xj , as
a principal variable. This step yields the choice of m features. The
reason for choosing the vector nearest to the mean is twofold. This
feature can be thought of as the central feature of that cluster, the
one dominant in it, and which holds the least redundant information
of features in other clusters.
4.2.4 Interpretation of principal components

The principal components are, by construction, linear combinations of all the
measured variables. Therefore, their physical interpretation is usually not
straightforward. Various attempts have been made to overcome this difficulty;
among them, PCs rotation represents a very common solution. Through rota-
tion, the weights can be redefined to meet alternative criteria. In particular,
rotation is aimed at attaining a simple structure for Aq , so that weights on a
principal component are either close to unity or close to zero and, thus, variables
have large weights on only few or (ideally) one principal component.
79
More formally, rotation is concerned with finding an orthogonal matrix, T ,

so that orthogonally rotated weights (or loadings) for the PCs can be defined
as:
Bq = Aq T . (4.37)
The matrix T is chosen to optimize one of many simplicity criteria available

[109]. The most common orthogonal rotation is based on the maximization of
the VARIMAX criterion [110]:
 !2 
q p p
1 1X
X X
VM AX (Aq ) = a4ik − a2ik . (4.38)
p p
k=1 i=1 i=1
Kaiser [110] refers to this as raw VARIMAX, but it is the version that has
become most popular. Verbally, this is simply the sum of the column-wise
variances of the squared elements of Aq . In other words, a criterion is defined
to maximize the amount of variance explained for any of the original variables
on single PCs. After VARIMAX rotation, Aq will generally have fewer large
loadings in its columns, thereby making the columns more easily interpretable.
A simple analytical solution for the maximization of the criterion in Eq. (4.38)
exist for the two-dimensional case [110]. Indicating the columns of Aq with k
and l, the two-dimensional solutions is:
Pp 2 − a2 (2a a ) +

t = 2p
P 2 i=1 aik il
Pp ik il
ahik − a2il

−2 i=1 (2aik ail ) i
2 , (4.39)
b = p pi=1 a2ik − a2il − (2aik ail )2 +
P
2 2
− [ pi=1 (2aik ail )]
P 2
aik − a2il
P
−
thus leading to the following definition of the optimal rotation angle, φ
1
φ = arctan (t, b) . (4.40)
4
The optimal rotation matrix is then given by:

cos (φ) −sin (φ)
B2 = . (4.41)
sin (φ) cos (φ)
The two-dimensional solution just presented can be extended to the more

general case of a sample X with dimensionality p. In this case, a possible
algorithm consists in applying the planar solution over all the (2 x 2) orthogonal
B2 matrices, through a sequence of iterates converging to a final solution. Little
is known about the convergence behavior of the algorithm, but in practice it
usually converges to a local maximizer of the VARIMAX criterion.
80
4.3. Local Principal Components Analysis
Figure 4.4: Schematic illustration of the VARIMAX rotation [110].
4.3 Local Principal Components Analysis

The PCA transformation described in Section 4.2 can suffer from its reliance on
second order statistics. In fact, the PCs are uncorrelated, i.e. their second-order
product moment is zero, but they can still be highly statistically dependent.
This is particularly important when the relationships among the correlated
variables are non-linear, as it usually happens for a reacting system. In this
case, PCA fails to find the most compact description of the data and it usually
requires a larger number of components to model the low-dimensional hyper
plane embedded in the original space, with respect to a non-linear technique.
This simple realization has prompted the development of non-linear alternatives
to PCA. A considerable amount of work has been done in the context of neural
networks. Nevertheless, here we are more interested in a different approach,
introduced by Kambhatla and Leen [111] in the field of images processing and
known as Local Principal Components Analysis (LPCA).
LPCA employs a local linear approach to reduce the statistical dependency
between the variables of a sample and to achieve the desired optimal dimension
reduction. According to LPCA, a Vector Quantization (VQ) algorithm first par-
titions the data space into disjoint regions and then PCA is performed in each
cluster, relying on the observation that, if the local regions are small enough,
the data manifold will not curve much over the extent of the region and the
linear model will be a good fit. For the LPCA to be effective, the VQ algorithm
should not be independent of the PCA analysis. For example, a partitioning
based on the Euclidean distance is very intuitive and easy to implement but
the sample clustering is carried out without any connection with the following
projection onto the lower-dimensional subspace. For this reason, Kambhatla
and Leen [111] introduce a VQ algorithm based on a reconstruction error met-
ric. Given an observation from the sample X, Xi , a global reconstruction error
for each observation can be defined as:

(k)
GRE Xi , X = kXi − Xi,q k =

(k)
= Xi − X + Zi,q A(k) (4.42)

q
81
(k)
where X is kth cluster centroid, Xi,q is the rank q approximation of Xi ,
Zi,q is the ith value of the truncated set of PCs, Zq , and A(k)q is the matrix ob-
tained by retaining only the first q eigenvectors of the covariance matrix, S (k) ,
associated to the kth cluster. In the context of reacting systems Eq. (4.42)
needs to be modified to take into account the differences in size and units of
the state variables. In fact, a clustering based on GRE would lead to an op-
timization with respect to temperature only. Therefore, the original LPCA
algorithm from Kambhatla and Leen [111] was modified [92] to include data
preprocessing (Section 4.2.2) in the quantization scheme. A very stable algo-
rithm is obtained by using a global scaled reconstruction error metric, GSRE ,
defined as:

(k)
GSRE Xi , X , D = X − X (4.43)
f
i i,q
f
where X ei is the ith observation of the sample scaled by D, the diagonal matrix
whose jth diagonal element is the scaling factor dj associated to xj . The
proposed LPCA algorithm, briefly referred as VQPCA, can be summarized as
follows:
(k)
1. Initialization: the cluster centroids, X , are randomly chosen from the
data set and S (k) is initialized to the identity matrix for each cluster.
2. Partition: each observation from the sample is assigned to a cluster using

the squared reconstruction distance given by GSRE .
3. Update: the cluster centroids are updated on the basis of partitioning

carried out at step 2.
4. Local PCA: PCA is performed in each disjoint region of the sample.
5. Steps 2-4 are iterated until convergence is reached.
The VQPCA algorithm is illustrated in Figure 4.5. The vector quantization step
partitions the data into cluster, trying to follow the curvature of the manifold
in the low-dimensional space. Then, the points are assigned to the clusters
depending on their low-dimensional projection on each of the identified clusters.
The goodness of reconstruction given by VQPCA is measured with respect
to the mean variance in the data as:
E (GSRE )
GSRE,n = (4.44)
E [var (e
xj )]
where E denotes the expectation operator and x ej is the scaled jth variable
from X. If auto scaling is employed in data preprocessing, Eq. (4.44) reduces
to:
GSRE,n = E (GSRE ) . (4.45)
82
4.3. Local Principal Components Analysis
Figure 4.5: Schematic illustration of the VQPCA algorithm [92] .
Convergence can be judged using the following criteria:

1. The normalized global scaled reconstruction error, GSRE,n , is below a
specific threshold, ∗GSRE,n .
2. The relative change in cluster centroids between two successive iterations

is below a fixed threshold, i.e. 10−8 .
3. The relative change in GSRE,n between two successive iterations is below

a fixed threshold, i.e. 10−8 .
Requirements 2 and 3 are particularly useful if an explanatory analysis on
the performances of VQPCA in terms of GSRE,n is of interest. In this case,
requirement 1 can be relaxed and the variation of GSRE,n as a function of the
number of eigenvalues and clusters can be analyzed by enforcing requirements
2 and 3. Otherwise, all the three conditions can be used and an iterative
procedure for the determination of the number of eigenvalues required to achieve
a fixed GSRE,n can be employed. Staring with q = 1, the number of eigenvalues
can be increased progressively until the desired error level is reached.
The VQPCA approach is based on the unsupervised partitioning of the data
into clusters, based on the minimization of the reconstruction error. Therefore,
the approach is optimal from the point of view of error minimization; how-
ever, being the partitioning iterative, the approach can result computationally
intensive for very large data sets (i.e. data from DNS). Therefore, a viable
alternative to VQPCA could be represented by a supervised partition of data
into clusters, based on the a priori knowledge on the conditioning variable.
In the context of turbulent non-premixed combustion, the obvious choice is
represented by mixture fraction. Therefore, a Mixture fraction PCA (FPCA)
algorithm can be proposed as follows [92]:
1. Partition. The data is partitioned into bins of mixture fractions.
2. Local PCA: PCA performed in each of mixture fraction bin.

The FPCA approach is schematized in Figure 4.6. The data are partitioned
into two mixture fraction bins, i.e. rich and lean regions, and a one-dimensional
83
Figure 4.6: Schematic illustration of the FPCA algorithm [92] for a CO/H2
flame [112]..
coordinate system is identified in each cluster. With respect to the VQPCA

approach, FPCA allows a very fast clustering. However, it is not possible to
state a priori that the choice of the mixture fraction as conditioning variable
is the best available.
In the following, the local approaches will be compared to the classic ap-
proach consisting in the application of PCA to compete sets of data, i.e. taking
k = 1, and denoted with Global PCA (GPCA).
4.4 Data sets for model validation

A prerequisite for the application of the PCA methodology described in the
previous Sections is the availability of data sets for the extraction of the PCs.
Both experimental and numerical data were used in the present study. The
data to be analyzed with PCA have been organized in (n x p) matrices, X,
whose rows represent instantaneous spatial snapshots of the reacting species
concentrations and temperature.
4.4.1 Experimental data

High fidelity experimental data provided under the framework of the Workshop
on Measurement and Computation of Turbulent Non-premixed Flames (TNF
workshop) [80] have been used to assess the PCA methodology.
The first flame investigated in the present study is a turbulent non-premixed
CO/H2 /N2 (0.4/0.3/0.3 by vol.) jet flame [112], hereafter called simply jet
flame, selected as base case for the analysis due to its favorable properties. In
84
4.4. Data sets for model validation
fact, the flame does not experience any liftoff or localized extinction and retains
the simple flow geometry of the hydrogen jet flames [80], while adding a modest
level of chemical kinetic complexity. Moreover, the flame is fully characterized
in terms of scalar data. Simultaneous Raman/Rayleigh/LIF measurements
of temperature and species concentrations were conducted at Sandia National
Laboratories, California. About 800 to 1000 measurements were taken at dif-
ferent spatial locations, for a total of 66.275 data points of nine different state
variables (T, N2 , O2 , H2 O, H2 , CO, CO2 , OH, NO). The mixture fraction
calculated following Bilger [75] is also available in the experimental results.
The stoichiometric value of the mixture fraction for this flame is 0.295. The
experimental uncertainties can be obtained from [112].
The second flame is a CH4 flame, part of a series of four piloted jet flames,
C, D, E and F, investigated by Barlow and Frank [99]. Starting from Flame
D, the velocity, and then the Reynolds number, associated to the main jet
is raised, thus increasing the probability of local extinction phenomena. The
flames of interest for the present study are Flame D and, in particular, Flame F.
Flame F shows severe non-equilibrium effects and it is close to global extinction
in the downstream part of the flame; therefore, it represents a challenging
system to judge PCA capabilities in terms of chemical manifolds identification
and parametrization. Likewise the jet flame, Raman/LIF measurements of
temperature and species’ concentrations (T, N2 , O2 , H2 O, H2 , CH4 , CO, CO2 ,
OH, NO), are provided at different spatial locations for a total of 62.766 data
points. The mixture fraction is defined according to Bilger [75], except that
only the elemental mass fractions of hydrogen and carbon are included. The
mixture fraction for this flame is 0.351. The experimental uncertainties can be
obtained from [99, 113].
The third system investigated is the jet in hot co-flow (JHC) burner [46,
49, 50], hereafter denoted with JHC, designed to emulate the flameless com-
bustion regime (Section 1). It consists of a central fuel jet (80% CH4 and 20%
H2 ) within an annular co-flow of hot exhaust products from a secondary burner
mounted upstream of the jet exit plane. The O2 level of the co-flow is controlled
at three different levels, i.e. 3, 6 and 9% (by vol.), while the temperature and
exit velocity are kept constant. Similarly to the other flames, around 56.000
observations are provided for temperature and species concentrations (T, N2 ,
O2 , H2 O, H2 , CH4 , CO, CO2 , OH, NO). The mixture fraction is defined accord-
ing to Bilger [75], The experimental uncertainties can be obtained from [46].
The availability of experimental data for the JHC system has been particularly
important for the present Thesis, as it has allowed to give insights for the CFD
analysis of the combustion systems investigated in Chapters 5-7. In particular,
information regarding turbulence/chemistry interactions in flameless combus-
tion regime have been extracted from the PCA analysis.
85
4.4.2 Numerical data

In conjunction with high fidelity experimental data, numerical results from the
Direct Numerical Simulation (DNS) of CO/H2 oxidation with detailed chem-
istry [114] have also been considered. Details about the DNS simulations and
code can be found in Sutherland et al. [88]. Two DNS data sets have been con-
sidered: a spatially evolving and a temporally evolving jet, characterized by a
significant degree of extinction. For the first data set, indicated as DNS1, three
temporal slices, each consisting of approximately 1.500.000 scalar observations
(T, H2 , O2 , O, OH, H2 O, H, HO2 , H2 O2 , CO, CO2 , HCO), are available; the
second data set, DNS2, consists of twelve temporal slices, each one comprising
around 700.000 observations of the same variables.
The advantage of DNS data with respect to experimental data lies in the
large amount of data accessible. Moreover, DNS simulations give access to
many additional variables, beside scalar values, which are not provided by any
experimental campaign. In particular, the scalar source terms can be extracted
from DNS simulations, thus allowing to judge the capabilities of the extracted
PCs to parametrize not only the original variables, but also their source terms.
Of course, in the perspective of adopting PCA as a predictive model, the gener-
ation of data with DNS for PCs extraction does not represent a viable solution
and other approaches, such as One Dimensional Turbulence (ODT) [115], could
be pursued.
4.5 Results
This section describes the results of the PCA methodology applied to the ex-
perimental and numerical data sets described in Section 4.4 are here presented.
First, the capabilities of PCA for the identification of low-dimensional mani-
folds in turbulent reacting systems is investigated. In particular, the effect of
the preprocessing strategies and modeling approaches (i.e. GPCA vs. LPCA)
on the manifold dimensionality is thoroughly discussed, trying to provide also
a physical interpretation for the extracted PCs.
Then, the feasibility of a PCA based combustion model is discussed. The
PCA model is validated a priori using the DNS data sets and its performances
are compared to those of an ideal flamelet parametrization (Chapter 2).
4.5.1 PCA for the identification of low-dimensional manifolds

The objective of the present Section is to provide a methodology i) to in-
vestigate the existence of low-dimensional manifolds in turbulent flames, ii)
to find the most compact representation for them and iii) to guide the se-
lection of “optimal” reaction variables able to accurately reproduce the state
space of a reacting system. PCA has been previously applied to combustion.
Frouzakis et al. [116] applied PCA for data reduction of two-dimensional DNS
86
4.5. Results
data of opposed jet flames. The analysis was aimed at identifying the num-
ber of components required to accurately approximate the original data. To
this purpose, the correlations among velocities, pressure and species concentra-
tions at different times were taken into account, thus leading to eigenvectors
which are linear combination of the temporal snapshots considered. Similarly,
Danby and Echekki [117] implemented PCA for the analysis of an unsteady
two-dimensional direct numerical simulation of auto ignition in homogeneous
hydrogen air mixtures, with the main purpose of determining the requirements
to reproduce passive and reactive scalars during the process of auto ignition.
The approach presented here is quite different from the ones described above.
The main purpose of the developed PCA methodology is to find correlations
among the state variables (temperature and species concentration) to allow
an optimal approximation of the system in a low-dimensional space. Such an
approach leads to the determination of eigenvectors which are linear combina-
tions of the original variables in a way that allows reducing the dimension of the
system. A similar method was proposed by Maas and Thévenin [118] for the
analysis of DNS data. However, they only considered a very small sampling in
state space. The current study provides significantly more depth in its analysis,
and applies PCA to both experimental and numerical data sets.
4.5.1.1 GPCA of experimental data sets
Figure 4.7 shows the magnitude of the eigenvalues associated with the PCA
reduction of the jet flame data set, together with the contribution of the q
largest eigenvalues to the amount of variance explained by the new basis vectors.
The eigenvalue distribution reflects the covariance structure of the data set,
shown in Table 4.1, and obtained by applying the auto scaling criterion. It is
clear that the first two eigenvalues alone account for more than the 92% of the
total variance in the data. On the other hand, the last four smallest eigenvalues
are very close to zero; therefore, they contain no useful information and only
explain linear dependencies among the original variables. Therefore, a strong
size reduction, from 9 to 2 or 3, can be accomplished by using PCA, through
the identification of the most active directions in the original data. The total,
tq , and individual variance, tq,j , accounted for the jet flame by the first two
or three eigenvalues are listed in the first two columns of Table 4.2. It can be
observed that, by choosing q = 2, it is possible to capture more than 90% of the
individual variances of all the main species and temperature, while the minor
species, OH and NO, require an additional component, q = 3, to reach levels
of approximation comparable to the other state variables.
This is confirmed by the analysis of the parity plots of temperature and
species mass fractions given by the PCA reconstruction for the cases q = 2
(Figure 4.8) and q = 3 (Figures 4.9). It can be observed that the addition of a
component has a small effect on temperature and main species, whose variation
is mainly explained by the first two components (Table 4.2). Moreover, the par-
87
Figure 4.7: Scree-graph and histograms of the q largest eigenvalues for the jet
flame data set, preprocessed with auto scaling.
ity plots of temperature (Figures 4.8 and 4.9 (a)), H2 O mass fraction (Figures
4.8 and 4.9 (d)) and minor species such as OH and NO (Figures 4.8 and 4.9
(e, f)) point out the existence of non linear deviations in the recovered data,
which can be probably ascribed to non linear dependencies among the original
variables. This result suggests that the low-dimensional projection of the ther-
mochemical state shows significant non linearities which cannot be taken into
account with a global linear approach. Therefore, specific algorithms perform-
ing PCA in locally linear regions of the data (Section 4.3) could be taken into
account, to improve the accuracy of the parametrization.
Figure 4.10 shows the eigenvalue size distribution and the contribution of
the q largest eigenvalues to the total explained variance, tq , for Flame D, F and
JHC data sets. The covariance matrices for the data sets are shown in Table
4.3-4.5. Similarly to the jet flame, a significant size reduction can be achieved
for D and F flames, although an additional component is required, q = 3 or
q = 4, due to the higher complexity of the piloted flames (Section 4.4.1). On
the other hand, the JHC data set shows a higher dimensionality and at least 4
components are needed to explain as much as 90% of the total variance in the
original data. The number of required PCs, q, increases to 5 if an individual
variance, tq,i , above 90% is desired for all the variables, as indicated in Table
4.6. Such result is particularly interesting for the present Thesis, as it confirms
the complexity in the numerical modeling of the flameless combustion regime
[45, 78, 79], caused by the overlap between chemical and mixing scale and,
thus, by the need of optimal progress variables for the description of complex
interactions which take place in such regime.
Table 4.6 lists the values of tq and tq,j accounted for Flame D, F and for
JHC. It is interesting to observe the very strong similarities between Flame
D and F, confirmed by the analysis of their covariance structure (Tables 4.3
88
4.5. Results
Table 4.1: Covariance matrix for the jet flame data set. Scaling criterion adopted: auto scaling.
 
T Y O2 YN 2 YH2 YH2 O YCO YCO2 YOH YN O

 T 1.000 −0.825 −0.512 0.005 0.938 0.117 0.984 0.771 0.815 


 YO2 1.000 0.887 −0.541 −0.909 −0.646 −0.767 −0.562 −0.558 


 YN2 1.000 −0.835 −0.667 −0.902 −0.438 −0.266 −0.256 

89

 YH2 1.000 0.196 0.973 −0.082 −0.168 −0.170 


 YH2 O 1.000 0.329 0.892 0.725 0.678 


 YCO 1.000 0.024 −0.081 −0.113 


 YCO2 1.000 0.793 0.855 

 YOH 1.000 0.639 
YN O 1.000
Table 4.2: Total, tq , and individual variance, tq,j , accounted for the jet flame data set, as a function of the number of retained
PCs, q, and the preprocessing criterion.
tq,i (%)
auto range max vast level
q=2 q=3 q=2 q=3 q=2 q=3 q=2 q=3 q=2 q=3
T 0.971 0.973 0.983 0.991 0.979 0.990 0.992 0.992 0.896 0.943
YO2 0.986 0.986 0.994 0.994 0.997 0.997 0.975 0.978 0.942 0.961
0.986 0.986 0.981 0.981 0.971 0.971 1.000 1.000 0.965 0.970
90
YN2
YH2 0.968 0.969 0.962 0.963 0.957 0.960 0.945 0.947 0.991 0.991
YH2 O 0.930 0.936 0.945 0.945 0.944 0.944 0.940 0.978 0.870 0.884
YCO 0.994 0.994 0.995 0.997 0.990 0.994 0.979 0.980 0.987 0.987
YCO2 0.973 0.977 0.979 0.987 0.977 0.988 0.981 0.985 0.908 0.959
YOH 0.738 0.940 0.731 0.991 0.745 0.992 0.660 0.687 0.870 0.993
YN O 0.772 0.930 0.728 0.795 0.729 0.802 0.744 0.970 0.701 0.926
tq (%) 0.924 0.966 0.946 0.975 0.942 0.975 0.992 0.996 0.949 0.980
4.5. Results
(a) (b)
(c) (d)
(e) (f)
Figure 4.8: Parity plots of temperature (a), H2 O (b), H2 (c), CO (d), OH (e)
and NO (f) mass fractions illustrating the GPCA (q = 2) reduction of the jet
flame data set. Scaling criterion adopted: auto scaling.
91
(a) (b)
(c) (d)
(e) (f)
and NO (f) mass fractions illustrating the GPCA (q = 3) reduction of the jet
flame data set. Scaling criterion adopted: auto scaling.
92
4.5. Results
and 4.4), thus indicating that the relations between the state variables are not
strongly affected by the increase in Reynolds number from one flame to the
other.
A closer look at the covariance matrices structure indicate that, with the
exception of the JHC data set, there is always a strong correlation between
temperature, oxidation products (CO2 , H2 O), OH and NO (Table 4.1, Tables
4.3-4.4), as it is expected for a turbulent non premixed flame. The covariance
matrix for the JHC data set still shows a strong correlation between temper-
ature and product’s mass fractions; however, the covariance between tempera-
ture and the minor species, i.e. OH and NO, is lower. Once again, this indicate
the existence of a more complex flame structure, arising from a balance between
turbulent mixing and chemical kinetics.
Figure 4.11 and Figure 4.12 show the GPCA reconstruction of Flame F,
with q = 3 and q = 4, respectively. Similarly to the jet flame, the addition
of a PC barely affects the accuracy in the prediction of the major species, as
it mainly acts on the prediction of the minor species, i.e. OH and NO (Table
4.6). Very similar results are observed for Flame D.
With regard to the JHC system, very large (non linear) deviations are ob-
served for temperature (Figure 4.13 (a)), CO (Figure 4.13 (c)) and OH (Figure
4.13 (e)), for the case q = 4. The increase of the number of PCs to q = 5
strongly improves the prediction of CO (Figure 4.14 (c)) and OH (Figure 4.14
(e)), but not temperature (Figure 4.13 (a)) and other species, i.e. CO2 . It is
noteworthy that NO is very well captured, even with q = 4. This results sug-
gests that one of the retained PCs is highly correlated with NO, thus leading
to the observed result.
PCs interpretation and rotation It is interesting to provide an interpre-

tation of the results described above by looking at the structure of the eigenvec-
tors matrices for the different experimental data sets. Tables 4.7-4.10 report the
weights of the original variables on the retained principal components, before
(a) and after applying VARIMAX rotation, for the jet flame, Flame D, Flame F
and JHC, respectively. As it was pointed out in Section 4.2.4, the PCs weights
are determined to maximize variance and not physical interpretability. How-
ever, PCs rotation can help overcome such difficulty, through the determination
of a simpler structure for the eigenvectors.
The analysis of the rotated eigenvectors matrices shows a common pattern
for the different systems, again with the exception of the JHC data set. It
can be observed how the first (rotated) PC is always an ensemble component,
consisting of temperature, oxidizer, product species and NO. This component
has the effect of capturing as much as possible of the original data variance in
the data, trying to explain the (non linear) relations among the state variables
with a single parameter. The other PCs differs from one data set to the other.
For the jet flame, the second PC consists of reactants (CO, H2 , Air)., while
the third is basically OH, which is determinant for capturing the reaction zone
93
(a)
(b)
(c)
Figure 4.10: Scree-graph and histograms of the q largest eigenvalues for Flame
D (a), Flame F (b) and JHC (c). Scaling criterion adopted: auto scaling.
94
4.5. Results
Table 4.3: Covariance matrix for Flame D data set. Scaling criterion adopted: auto scaling.
 
T Y O2 YN2 YH2 YH2 O YCH4 YCO YCO2 YOH YN O

 T 1.000 −0.960 −0.134 0.418 0.979 −0.295 0.535 0.984 0.681 0.912 

 YO2 1.000 0.323 −0.589 −0.977 0.093 −0.688 −0.932 −0.645 −0.859 
 
 YN2 1.000 −0.473 −0.194 −0.867 −0.451 −0.109 −0.061 −0.052 
 
 
YH2 1.000 0.548 0.194 0.919 0.320 0.056
95
 0.240 
 
 YH2 O 1.000 −0.257 0.658 0.949 0.666 0.883 
 
 YCH4 1.000 0.102 −0.312 −0.221 −0.329 
 
 YCO 1.000 0.442 0.213 0.372 
 
 YCO2 1.000 0.708 0.933 
 
 YOH 1.000 0.688 
YN O 1.000
Table 4.4: Covariance matrix for Flame F data set. Scaling criterion adopted: auto scaling.
 
T YO2 YN 2 YH2 YH2 O YCH4 YCO YCO2 YOH YN O

 T 1.000 −0.968 −0.073 0.418 0.984 −0.312 0.543 0.981 0.745 0.824 


 YO2 1.000 0.241 −0.545 −0.976 0.128 −0.660 −0.940 −0.748 −0.790  

 YN2 1.000 −0.378 −0.109 −0.882 −0.349 −0.053 −0.057 −0.026  
YH2 1.000 0.512 0.124 0.926 0.305 0.189 0.229 
 
96

YH2 O 1.000 −0.296 0.636 0.956 0.754 0.816 
 

YCH4 1.000 0.041 −0.327 −0.232 −0.297 
 

YCO 1.000 0.432 0.262 0.331 
 

 
 YCO2 1.000 0.767 0.851 
 
 YOH 1.000 0.633 
YN O 1.000
4.5. Results
Table 4.5: Covariance matrix for JHC data set. Scaling criterion adopted: auto scaling.
 
T YO2 YN 2 YH2 YH2 O YCH4 YCO YCO2 YOH YN O

 T 1.000 −0.476 0.616 −0.534 0.892 −0.534 0.292 0.913 0.427 0.388 
 YO2 1.000 0.306 −0.420 −0.619 −0.418 −0.378 −0.614 −0.150 −0.126 
 
 YN2 1.000 −0.990 0.483 −0.991 0.139 0.465 0.216 0.253 
 
 
YH2
97
 1.000 −0.392 0.998 −0.085 −0.376 −0.195 −0.240 
 
 YH2 O 1.000 −0.398 0.516 0.927 0.340 0.389 
 
 YCH4 1.000 −0.089 −0.381 −0.196 −0.241 
 
 YCO 1.000 0.266 −0.072 0.123 
 
 YCO2 1.000 0.362 0.376 
 
 YOH 1.000 0.214 
YN O 1.000
(a) (b)
(c) (d)
(e) (f)
Figure 4.11: Parity plots of temperature (a), H2 O (b), CO (c), H2 (d), OH (e)
and NO (f) mass fractions illustrating the GPCA (q = 3) reduction of Flame
F. Scaling criterion adopted: auto scaling.
98
4.5. Results
(a) (b)
(c) (d)
(e) (f)
and NO (f) mass fractions illustrating the GPCA (q = 4) reduction of Flame
F. Scaling criterion adopted: auto scaling.
99
(a) (b)
(c) (d)
(e) (f)
and NO (f) mass fractions illustrating the GPCA (q = 4) reduction of JHC
data set. Scaling criterion adopted: auto scaling.
100
4.5. Results
(a) (b)
(c) (d)
(e) (f)
and NO (f) mass fractions illustrating the GPCA (q = 5) reduction of JHC
data set. Scaling criterion adopted: auto scaling.
101
Table 4.6: Total, tq , and individual variance, tq,j , accounted for Flame D, F
and JHC data sets by the GPCA reduction, as a function of the number of
retained PCs, q.
tq,i (%)
Flame D Flame F JHC
q=3 q=4 q=3 q=4 q=4 q=5
T 0.971 0.985 0.967 0.971 0.932 0.948
YO2 0.982 0.986 0.978 0.979 0.961 0.974
YN2 0.979 0.981 0.979 0.980 0.991 0.991
YH2 0.959 0.966 0.969 0.970 0.998 0.998
YH2 O 0.987 0.988 0.983 0.984 0.966 0.966
YCH4 0.984 0.986 0.984 0.984 0.999 0.999
YCO 0.940 0.965 0.961 0.969 0.757 0.998
YCO2 0.965 0.985 0.969 0.974 0.911 0.970
YOH 0.743 1.000 0.711 0.978 0.735 1.000
YN O 0.902 0.932 0.792 0.892 0.999 1.000
tq (%) 0.941 0.977 0.946 0.968 0.925 0.984
correctly.
Moving on to Flame D and F, the second PC observed for the jet flame is
somehow split into two components, one representative of a mixture fraction
(both N2 and CH4 are very highly correlated to mixture fraction) and one rep-
resentative of the intermediate product species (CO, H2 ). The last component
is again OH, the flame marker. The eigenvectors structures of Flame D and
F are very similar. There is only a significant difference which could be high-
lighted, namely the NO weights on the fourth component. For Flame D, NO
does not appear as a relevant weight on the last PC, whereas it is no negligible
for the fourth component of Flame F, thus reflecting the lower correlations be-
Table 4.7: Retained (a) and rotated (b) eigenvectors for the jet flame data set.
a1 a2 a3 a1,r a2,r a3,r

T 0.40 0.18 -0.07 T 0.44 0.00 0.03
YO2 -0.41 0.15 0.00 YO2 -0.30 0.31 -0.07
YN2 -0.33 0.38 0.01 YN2 -0.13 0.48 -0.02
YH2 0.14 -0.55 -0.05 YH2 -0.10 -0.56 -0.07
(a) (b)
YH2 O 0.41 0.05 0.12 YH2 O 0.36 -0.13 0.20
YCO 0.18 -0.53 0.02 YCO -0.06 -0.56 0.01
YCO2 0.39 0.24 -0.11 YCO2 0.46 0.05 0.00
YOH 0.31 0.27 0.74 YOH 0.22 0.11 0.81
YN O 0.31 0.29 -0.65 YN O 0.54 0.13 -0.54
102
4.5. Results
Table 4.8: Retained (a) and rotated (b) eigenvectors for Flame D data set.
a1 a2 a3 a3 a1,r a2,r a3,r a3

T 0.40 -0.08 0.07 0.11 T 0.46 0.02 0.00 0.02
YO2 -0.40 -0.06 -0.07 -0.04 YO2 -0.39 0.08 0.13 0.03
YN2 -0.06 -0.59 -0.38 -0.05 YN2 -0.07 0.69 0.05 0.02
YH2 0.23 0.39 -0.55 0.04 YH2 0.00 -0.01 -0.69 0.08
(a) YH2 O 0.41 -0.03 0.00 0.05 (b) YH2 O 0.38 0.04 -0.15 -0.05
YCH4 -0.10 0.57 0.41 0.01 YCH4 -0.07 -0.72 0.06 0.02
YCO 0.28 0.33 -0.49 -0.14 YCO 0.01 0.02 -0.67 -0.07
YCO2 0.39 -0.13 0.18 0.11 YCO2 0.48 0.00 0.09 0.01
YOH 0.32 -0.11 0.25 -0.83 YOH 0.00 0.00 0.01 -0.99
YN O 0.34 -0.15 0.22 0.51 YN O 0.50 0.01 0.16 0.04
Table 4.9: Retained (a) and rotated (b) eigenvectors for Flame F data set.
a1 a2 a3 a3 a1,r a2,r a3,r a3

T 0.40 0.10 0.04 0.20 T 0.43 -0.03 -0.02 -0.01
YO2 -0.40 0.06 -0.03 -0.11 YO2 -0.39 -0.08 0.11 0.07
YN2 -0.10 0.56 -0.39 -0.08 YN2 -0.07 -0.70 0.04 0.00
YH2 0.23 -0.40 -0.49 -0.14 YH2 0.03 0.01 -0.70 0.12
(a) YH2 O 0.41 0.03 -0.05 0.07 (b) YH2 O 0.39 -0.04 -0.11 -0.05
YCH4 -0.08 -0.55 0.45 0.08 YCH4 -0.08 0.70 0.04 0.00
YCO 0.28 -0.34 -0.44 -0.26 YCO 0.04 -0.01 -0.66 -0.09
YCO2 0.39 0.14 0.13 0.24 YCO2 0.45 -0.01 0.09 -0.03
YOH 0.29 0.19 0.40 -0.84 YOH 0.13 0.00 0.06 -0.92
YN O 0.37 0.18 0.16 0.29 YN O 0.53 0.02 0.19 0.35
tween the two variables (Table 4.4), probably determined by the higher physical
complexity of the system.
Finally, regarding the eigenvectors of the JHC system, it can be observed
how the first rotated component does not show a large influence of NO, differ-
ently from all the other systems. This can be explained by taking into account
that the first PC tries to explain as much as possible of the data variability.
It is well known [3, 4, 2, 45] that NO formation in flameless combustion is
more homogeneous than in traditional non premixed combustion, due to the
smoother temperature gradients; therefore, NO is characterized by less vari-
ability and disappears from the first PC. The second and third PCs are, again,
representative of reactant and intermediate combustion products (Table 4.5),
reflecting a similar pattern to that observed for Flame F (and D). Differently
from the piloted flames, the fourth component is exclusively NO, thus meaning
that none of the previous components can take into account NO formation and
a specific PC is needed. Then, the OH component, present in all the other
103
Table 4.10: Retained (a) and rotated (b) eigenvectors for JHC data set.
a1 a2 a3 a4 a5
T 0.42 -0.16 0.08 0.12 0.16
YO2 -0.11 0.59 0.02 -0.11 -0.15
YN2 0.38 0.34 -0.10 0.07 0.03
YH2 -0.35 -0.39 0.08 -0.04 0.00
(a) YH2 O 0.40 -0.27 -0.10 0.06 0.03
YCH4 -0.35 -0.39 0.08 -0.04 0.00
YCO 0.16 -0.23 -0.67 -0.10 -0.64
YCO2 0.39 -0.26 0.08 0.11 0.31
YOH 0.19 -0.07 0.68 0.20 -0.67
YN O 0.22 -0.07 0.21 -0.95 0.00
a1,r a2,r a3,r a4,r a5,r

T 0.47 0.14 0.06 0.00 -0.06
YO2 -0.49 0.37 0.08 -0.06 -0.04
YN2 0.09 0.51 -0.02 0.02 0.01
YH2 -0.02 -0.53 0.00 0.00 0.00
(b) YH2 O 0.46 0.06 -0.18 -0.03 -0.01
YCH4 -0.02 -0.53 0.01 0.00 0.00
YCO -0.02 0.02 -0.97 0.00 0.00
YCO2 0.56 0.04 0.14 -0.01 0.05
YOH 0.01 -0.02 0.00 0.00 -1.00
YN O 0.01 -0.01 0.00 -1.00 0.00
system, is also found here as last PC.
Principal variables As it was pointed out in the previous Section, the orig-
inal variables do not contribute equally in the determination of the PCs and
rotation can improve eigenvectors interpretability, transforming them to meet
a simpler structure. Another way to help interpretation is to extract the so
called Principal Variables (PVs), described in Section 4.2.3.5.
Table 4.11 lists the PVs determined using the methods outlined in Section
4.2.3.5 for the jet flame. At first glance, the results may appear to vary widely,
going from one method to the other. However, a more careful analysis shows
the existence of many similarities. In particular, methods B4, B2, M2, MC2
and MC3 lead to very similar results, identifying a major variable (T, CO2
or H2 O), a fuel species (CO or H2 ) and OH as PVs. In fact, T, CO2 of H2 O
are highly correlated (Table 4.1), cov (T, CO2 ) = 0.984, cov (T, H2 O) = 0.938
and cov (H2 O, CO2 ) = 0.892. Similarly, H2 and CO are also exchangeable
variables, being cov (H2 O, CO2 ) = 0.973. Method MC1 replaces T (or CO2
or H2 O) with NO, which shows a strong correlation with T, although weaker
104
4.5. Results
Table 4.11: Principal variables for the jet flame data set, as provided by the
different methods described in Section 4.2.3.5.
Method Principal Variables

B4 H2 O, H2 , OH
B2 CO2 , H2 , OH
M3 T, CO, OH
MC1 NO, CO, OH
MC2 CO2 , CO, OH
MC3 CO2 , CO, OH
PF T, O2 , H2
Table 4.12: Principal variables for Flame D, F and the JHC data set. PV
method: MC2 (Section 4.2.3.5).
Data set Principal Variables

Flame D CH4 , CO, CO2 , OH
Flame F CH4 , CO, CO2 , OH
JHC CH4 , CO, CO2 , NO, OH
than that CO2 and H2 O. Finally, the PF method provides a different solution,
neglecting OH as PV and replacing it with O2 . However, this solution was
considered unreliable, being very far from the pattern identified by all the other
methods.
On the basis of the results obtained for the jet flame case, it was chosen
to adopt the MC2 method for the extraction of the PVs, as it provides results
comparable to most of the other models and satisfies a very appealing prop-
erty of PCA, the minimization of the reconstruction error. Applying the MC2
methods to the other data sets, we get the results in Table 4.12. It is very
interesting to observe that the same considerations derived from the analysis of
the rotated PCs can be done here, with a clearer physical interpretation. The
PVs selected for Flame D and F reflect the patters of the PCs, as they include
a mixture fraction variable, an intermediate and a product species and OH.
Finally, for the JHC system, the same set of PVs obtained for Flame D (and F)
is recovered, although augmented with NO, thus confirming the need to take
explicitly into account the formation of such pollutant species.
Effect of preprocessing strategies on the manifold dimensionality In

this Paragraph, the effect of preprocessing strategies on the PCA reduction is
presented, focusing on the jet flame data set. The performances of auto scaling
have been compared to those of other scaling criteria, presented in Section 4.2.2.
Figure 4.15 shows the eigenvalue size distribution and the contribution of the
q largest eigenvalues to the total variance explained, when applying scaling
105
criterion different than auto scaling, namely range (a), vast (b), level (c) and
max (d) scaling. If we compare Figure 4.15 to Figure 4.7, it is clear that all
the methods identify the manifold dimensionality to be equal to three, with
the exception of vast scaling (Figure 4.15 (b)), which identifies only two PCs,
with a very dominant first PC. Columns 3-10 in Table 4.2 show the values of
tq and tq,j obtained by applying range, max, vast and level scaling to the jet
flame data set. Results confirm that auto scaling is the only criterion able to
provide a uniform reconstruction of the state variables, leading to comparable
values of tq,j for all of them. Range and max scaling (columns 3-6, Table
4.2), very similar as expected, perform slightly better than auto scaling for
most of the main species and temperature. However, they proves to be unable
to properly capture NO variation, even with q = 3. Similarly, vast scaling
(columns 7-8, Table 4.2) concentrates on extremely stable variables, i.e. N2 ,
but completely fails in recovering OH properly. Then, the higher values of tq
given by range, max and vast scaling, compared to auto scaling, are due to the
higher variance explained for the major variables; however, these approaches
miss very important features, such as the parametrization of NO and OH. The
variance explained by auto scaling for OH and NO is up to 16% and 25% higher,
respectively, than that explained by the other scaling methods.
On the opposite, level scaling (columns 9-10, Table 4.2) focuses on variables
characterized by large relative changes and leads to an overestimation of the
role of minor species in the PCA reduction. Therefore, the prediction of minor
species such as OH and NO is very accurate, but major species such as H2 O
are badly recovered.
On the basis of the described sensitivity, it was chosen to adopt the auto
scaling as default preprocessing criterion for the analysis. Obviously, in ap-
plications which do not require the accurate parametrization of minor species,
other options could provide better results than to auto scaling.
4.5.1.2 LPCA of experimental and numerical data sets

The GPCA analysis presented in Section 4.5.1.1 has shown the existence of se-
vere non linearities in the parity plots of observed and predicted state variables.
Therefore, the determination of the manifold dimensionality with GPCA can
be somehow biased, as a globally linear approach is adopted to model complex
non linear interactions. In this context, LPCA (Section 4.3) can provide locally
linear models, able to follow the non linear development of the thermochemical
manifold in low-dimensional space.
Table 4.13 lists the values of the error metric, GSRE,n (Section 4.3), given
by GPCA, VQPCA and FPCA for the jet flame, Flame F and the JHC data
set, as a function of the number of clusters, k, and retained PCs, q.
It is interesting to observe (Table 4.13) that, when the reconstruction er-
ror is evaluated on a scaled basis, all the state variables become relevant and
the goodness of reconstruction can be properly judged. It should be recalled
106
4.5. Results
(a) (b)
(c) (d)
Figure 4.15: Scree-graph and histograms of the q largest eigenvalues for the
jet flame data set, preprocessed with range (a), vast (b), level (c) and max (d)
scaling.
Table 4.13: Values of GSRE,n associated with the GPCA, VQPCA and FPCA
reconstructions of the jet flame, flame F and JHC data set, as a function of the
number of clusters, k, and retained PCs, q.
k Jet flame Flame F JHC

q=2 q=3 q=3 q=4 q=3 q=4
GPCA 1 0.681 0.309 0.707 0.320 1.552 0.752
VQPCA 2 0.208 0.106 0.205 0.119 0.214 0.093
4 0.112 0.056 0.131 0.076 0.099 0.050
6 0.091 0.046 0.095 0.052 0.078 0.037
8 0.079 0.034 0.090 0.040 0.058 0.028
FPCA 2 0.214 0.084 0.263 0.147 0.410 0.105
4 0.121 0.066 0.158 0.087 0.123 0.059
6 0.103 0.051 0.134 0.067 0.112 0.052
8 0.092 0.045 0.122 0.063 0.093 0.044
107
here that (Section 4.3) GSRE,n is a mean global scaled reconstruction error,
normalized by the mean variance present in the original data. So, for exam-
ple, the value of GSRE,n associated with the GPCA reduction of the jet flame
with q = 2 is fairly large, GSRE,n = 0.68, thus reflecting the large deviations
observed in Figure 4.8 for some state variables. Even when q in increased to
three, a significant error, GSRE,n = 0.309 is obtained, confirming the persis-
tence of mainly non-linear departures from the original data. If lower values
of GSRE,n are desired, i.e. < 0.1, the value of q should be increased to 4 or 5
(GSRE,n = 0.098 and GSRE,n = 0.046, respectively). However, the manifold
dimensionality obtained could not be regarded as a true manifold dimension-
ality. It can be argued that fake components need to be added to account for
non linear interactions because of the global linear nature of the adopted model.
However, if a locally linear model is employed, i.e. VQPCA, much higher per-
formances in terms of GSRE,n are obtained, even for smaller values of q, i.e.
q = 2 or q = 3. Figure 4.16 shows the VQPCA reconstruction of temperature
(a), H2 O (b), CO (c), H2 (d), OH (e) and NO (f), with k = 8 and q = 2. A
much better agreement between original and reconstructed data is observed, as
it is confirmed by the value of 0.08 obtained for GSRE,n . A similar value of
GSRE,n would require q = 5 if GPCA is applied. Moreover, the parity plots for
the state variables in Figure 4.16 show how, after partitioning, the relationships
between the original and reconstructed data are mainly linear.
The values of GSRE,n for Flame F, as provided by the different approaches,
are also shown Table 4.13. Similarly to the jet flame, the reconstruction error,
GSRE,n , associated with the GPCA reductions (Figure 4.11 and Figure 4.12)
obtained by choosing q = 3 and q = 4 are very high, GSRE,n = 0.71 and
GSRE,n = 0.32, respectively, thus confirming the inability of GPCA to deter-
mine the most compact description of the data in a lower dimensional manifold.
Figure 4.17 shows the VQPCA reconstruction of temperature (a), H2 O (b), CO
(c), H2 (d), OH (e) and NO (f), with k = 8 and q = 3. The value of GSRE,n
obtained is 0.09, almost eight and four times smaller than the values given by
GPCA with q = 3 and q = 4, respectively. Also, GPCA would require 6 PCs
to give a value of GSRE,n as small as 0.09.
Finally, the same considerations hold for the JHC data set. In fact, the
effect of data partitioning is even more evident for this case with respect to the
jet flame and Flame F. Table 4.13 clearly points out that GSRE,n dramatically
decreases when VQPCA with k = 2 is employed, going from 1.552 and 0.752
to 0.241 and 0.093, for q = 3 and q = 4, respectively. This corresponds to a
reduction of around 85% for both cases, which is not observed for any of the
other investigated data sets. Such result is extremely interesting and suggests
the existence of different flame structures within the JHC system. Figure 4.18
shows the VQPCA reconstruction of temperature (a), H2 O (b), CO (c), H2
(d), OH (e) and NO (f), with k = 6 and q = 3. With respect to the jet flame
and Flame F, it is possible to reach values of GSRE,n below 0.1 with a smaller
number, i.e. k = 6, of clusters.
108
4.5. Results
(a) (b)
(c) (d)
(e) (f)
and NO (f) mass fractions illustrating the VQPCA (q = 2, k = 8) reduction of
the jet flame data set. GSRE,n = 0.08.
109
(a) (b)
(c) (d)
(e) (f)
Flame F data set. GSRE,n = 0.08.
110
4.5. Results
(a) (b)
(c) (d)
(e) (f)
JHC data set. GSRE,n = 0.08.
111
Table 4.14: Total, tq , and individual variance, tq,j , accounted for by the re-
tained PCs for the DNS1 and DNS2 data sets as a function of the number of
components, q.
k DNS1 DNS2
q=2 q=3 q=3 q=4
GPCA 1 3.130 1.830 1.800 1.130
VQPCA 2 0.816 0.176 0.734 0.369
4 0.307 0.065 0.235 0.076
6 0.116 0.025 0.141 0.046
8 0.043 0.010 0.114 0.038
10 0.036 0.009 0.096 0.033
FPCA 2 0.625 0.216 0.773 0.417
4 0.243 0.052 0.263 0.081
6 0.122 0.030 0.204 0.066
8 0.062 0.020 0.167 0.054
10 0.046 0.015 0.140 0.043
VQPCA has been exploited also for the analysis of the DNS data sets, DNS1
and DNS2, described in Section 4.4.2. Regarding the DNS2 data set, multiple
time steps have been merged before analyzing the data, namely t = 1.5e − 03 s,
t = 2.0e − 03 s, t = 2.5e − 03 s and t = 3.0e − 03 s. However, the resulting data
set (∼3.800.000 data points) have been conditioned in mixture fraction space,
between f = 0.1 and f = 0.8, to overcome memory issues (Figure 4.19).
Table 4.14 lists the values of GSRE,n given by GPCA, VQPCA and FPCA
for the DNS data sets. Similarly to the JHC case, the first partition is charac-
terized by a dramatic reduction of GSRE,n also for the DNS data sets. This
indicates, once again, that a global approach would lead to misleading estima-
tion of the manifold dimensionality. Table 4.14 also shows that DNS2 requires
an additional PC with respect to DNS1 to reach acceptable levels of accuracy.
This is determined by the higher complexity of the DNS2 data set, characterized
by a significant degree of extinction.
Figure 4.20 shows the contour plots of original and recovered temperature
and OH mass fraction distribution for the DNS1 data set. It can be observed
how a VQPCA approach with q = 3 and k = 8 allows to capture with great
accuracy the flame features, resulting in a very small reconstruction error,
GSRE,n = 0.01. This is a very appealing result, indicating that VQPCA could
be effectively exploited for the compression of DNS data sets, characterized by
very large storage requirements, for visualization and post-processing purposes.
Very strong compressions could be achieved, as shown here, prescribing the de-
sired accuracy of the recovered data. For a given manifold dimensionality, the
dimensions of the reduced data sets are independent of the number of clusters;
therefore, the parameters q and k can be varied to optimize the accuracy and
112
4.5. Results
(a)
(b)
Figure 4.19: Original (a) and conditioned (b) temperature field for DNS2 data
set at time step t = 1.5e − 03 s.
113
the disk space required for any data set.

The quality of the VQPCA reconstruction can be also qualitatively assessed
by means of contour plots of recovered and observed state variables (Figure
4.21). As it was pointed out for the experimental data, after the VQPCA
reduction, the parity plots show a linear behavior throughout the domain of the
state variable, confirming the capability of the VQPCA to follow the curvature
of the low-dimensional manifold in the reduced state space.
Figure 4.22 shows the contour plots of (conditioned) original and recovered
temperature distribution at time steps t = 1.5e−03 s (a, a’) and t = 2.0e−03 s
(b’). As for Flame F (Figure 4.17), VQPCA is able to capture flame extinc-
tion and re-ignition remarkably well, resulting in the quantitative agreement
between observed and recovered variables shown in Figure 4.23.
Comparison of VQPCA and FPCA Rows 6-9 in Table 4.13 and rows 7-11
in Table 4.14 list the values of GSRE,n given by FPCA for the experimental
and numerical data sets, respectively, as a function of the number of clusters, k,
and retained PCs, q. These values can be weighed against those obtained with
the VQPCA algorithm (rows 2-5 in Table 4.13 and rows 2-6 in Table 4.14) for
the same values of q and k. The comparison is illustrated graphically in Figure
4.24 for the experimental data, and in Figure 4.25 for the DNS data sets.
For the jet flame (Figure 4.24 (a)), VQPCA performs generally better with
respect to FPCA. The values of GSRE,n given by VQPCA are 6 − 25% lower
than those provided by FPCA with the exception of the cases corresponding to
k = 2 and q = 3. However, the performances of FPCA can be considered very
satisfying and promising, especially from the point of view of model implemen-
tation. In fact, FPCA partitioning is much simpler and straightforward than
the one underlying the VQPCA algorithm. Moreover, it is interesting to inves-
tigate how the VQPCA partitioning is reflected in the mixture fraction space.
To this purpose, the indexes of the observations with respect to the original
data matrix, X,f were stored and used to reconstruct the mixture fraction vec-
tors in each cluster identified by VQPCA. Figure 4.26 shows the temperature
as a function of mixture fraction for the two clusters selected by VQPCA, with
q = 2. It can be observed that the data are almost clustered into two regions
corresponding to the rich and lean zones of the flame, being the stoichiometric
mixture fraction of the jet flame flame equal to 0.295. This result suggests that,
for the jet flame, the mixture fraction can be considered an optimal variable for
the parametrization of the thermochemical state of the system, as it is gener-
ally assumed in many models for non-premixed combustion. The information
added here is that mixture fraction is an optimal variable from the point of
view of reconstruction error minimization
As far as Flame F flame is concerned, Table 4.13 and Figure 4.24 (b) point
out that VQPCA outperforms FPCA in all cases, providing values of GSRE,n
13-46% lower than those given by FPCA. These results confirm the notorious
complexity of this flame, characterized by significant local extinction and re-
114
4.5. Results
(a)
(a’)
(b)
(b’)
Figure 4.20: Contour plots of original and recovered temperature (a, a’) and
OH mass fraction (b, b’) distribution for DNS1. VQPCA reduction with q = 3
and k = 8. GSRE,n = 0.01.
115
(a) (b)
Figure 4.21: Parity plots of original and recovered temperature (a, a’) and OH
mass fraction (b, b’) distribution for DNS1. VQPCA reduction with q = 3 and
k = 8. GSRE,n = 0.04.
ignition. In the context of the Conditional Moment Closure [73], for example,
it has been recognized [119] that conditioning on mixture fraction is not suffi-
cient for Flame F and a second conditioning variable should be used. Figure
4.27 shows the temperature as a function of mixture fraction for the two clus-
ters selected by VQPCA with q = 3. It can be observed that, differently from
the jet flame, the VQPCA algorithm extracts features from the whole mixture
fraction space in order to achieve the best q-dimensional representation of the
thermochemical state of the system. Then, it can be concluded that, for Flame
F, mixture fraction does not represent an optimal reaction variables. There-
fore, VQPCA could provide an appealing alternative to guide the selection of
the most compact subset of reaction variables needed to properly describe the
thermochemical state of such reacting system.
Figure 4.24 (c) shows the comparison between the reconstruction provided
by VQPCA and FPCA for the JHC data set. Similarly to Flame F, the VQPCA
algorithm provides GSRE,n , 10-50% lower than those obtained with FPCA.
The mixture fraction partitioning does not optimally follow the curvature of
the manifold in state space, indicating the complexity of turbulence/chemistry
interactions for the system. This is further confirmed by Figure 4.28, which
shows the partition of temperature in the two clusters selected by VQPCA
with q = 3, in mixture fraction space. The algorithm selects a first cluster
characterized by a lean branch and part of a rich region, with an aspect charac-
teristic of a non premixed flame. On the other hand, the second cluster shows
important non equilibrium phenomena, such as extinction, similarly to cluster
2 for Flame F (Figure 4.27 (b)).
To better understand the underlying mechanism of the VQ partitioning
algorithm, it is possible to analyze the structure of the rotated eigenvectors in
116
4.5. Results
(a) (a’)
(b) (b’)
Figure 4.22: Contour plots of original (a, b) and recovered (a’, b’) temperature
distribution for DNS2, at two different time steps, i.e. t = 1.5e − 03 s (a,
a’) and t = 2.0e − 03 s (b, b’). VQPCA reduction with q = 4 and k = 8.
GSRE,n = 0.04.
(a) (b)
Figure 4.23: Parity plots of temperature (a) and OH (b) mass fraction illustrat-
ing the VQPCA (q = 4, k = 8) reduction of DNS2 data set. GSRE,n = 0.04.
117
118
Figure 4.24: Values of GSRE,n as a function of the number of clusters, k, and retained PCs, q, for the jet flame, Flame F and
JHC data sets.
4.5. Results
Figure 4.25: Values of GSRE,n as a function of the number of clusters, k, and

retained PCs, q, for the DNS1 and DNS2 data sets.
(a) (b)
Figure 4.26: Temperature as a function of mixture fraction in the two clusters

selected by VQPCA for the jet flame. q = 2 and GSRE,n = 0.21.
119
(a) (b)

selected by VQPCA for Flame F. q = 3 and GSRE,n = 0.21.
(a) (b)

selected by VQPCA for the JHC data set. q = 3 and GSRE,n = 0.21.
120
4.5. Results
Table 4.15: Rotated eigenvector in the first (a) and second (b) cluster identified
by VQPCA for Flame F. q = 3 and GSRE,n = 0.21
a1,r a2,r a1 a2
T -0.01 0.55 T 0.30 -0.01
YO2 -0.09 -0.44 YO2 -0.34 -0.09
YN2 -0.70 -0.07 YN2 -0.12 -0.12
YH2 -0.01 -0.04 YH2 0.05 0.64
(a) YH2 O -0.03 0.44 (b) YH2 O 0.33 -0.09
YCH4 0.71 -0.10 YCH4 -0.01 0.02
YCO 0.00 0.07 YCO 0.12 0.70
YCO2 0.00 0.51 YCO2 0.35 -0.11
YOH 0.00 0.05 YOH 0.60 -0.17
YN O -0.02 0.18 YN O 0.42 -0.13
the two clusters identified by VQPCA for Flame F (Table 4.15) and for the
JHC data set (Table 4.16). For Flame F, the eigenvectors associated to the
first cluster (Table 4.15 (a)) are a mixture fraction 3 and a linear combination
of major species and temperature, respectively. This supports the graphical
observation provided by Figure 4.27 (a), which shows the first cluster to be
characterized by the lean and rich branches of the flame. On the other hand,
the reaction region identified by Figure 4.27 (b), needs to be described by means
of parameters with a strong contribution of intermediate and minor species, as
it is shown in Table 4.15 (b).
With regard to the JHC data set, the structure of the rotated eigenvec-
tors prompts very interesting considerations. In particular, the second cluster
Table (4.16 (b)) is parametrized by a first component with significant weights
on the fuel species, intermediate species and temperature, whereas the second
PC reduces to OH. Thus, VQPCA is able to extract the subset of the data set
dominated by finite rate chemistry effects by means of progress variables able to
capture the ignition process. In the context of the numerical modeling of flame-
less combustion, such result confirms the need of combustion models suited for
the description of turbulence-chemistry interactions in such combustion regime.
As far as the numerical data are concerned, the VQPCA and FPCA reduc-
tions appear comparable for the DNS1 data set (Table 4.14), while VQPCA
outperforms FPCA for DNS2 (Table 4.14).This confirms that mixture fraction
is not optimal from the point of view of error minimization when the physics
under investigation become too complex. This is somehow expected, as mixture
fraction is only a measure of the local system stoichiometry and, then, it can
only cover relatively fast scales.
The small discrepancies between FPCA and VQPCA for DNS1 (and for
3
The denomination mixture fraction is used here because the variables which define the
first PC are highly correlated with the f , cov (f, N2 ) = 0.97 and cov (f, CH4 ) = 0.90.
121
Table 4.16: Rotated eigenvector in the first (a) and second (b) cluster identified
by VQPCA for the JHC data set. q = 3 and GSRE,n = 0.21
a1 a2 a1,r a2,r
T 0.48 0.01 T -0.25 0.21
YO2 -0.51 0.03 YO2 -0.13 -0.03
YN2 0.07 -0.02 YN2 -0.50 0.00
YH2 0.00 0.00 YH2 0.48 -0.06
(a) YH2 O 0.46 0.04 (b) YH2 O -0.23 0.06
YCH4 0.00 0.00 YCH4 0.49 0.00
YCO 0.05 -0.03 YCO 0.15 -0.07
YCO2 0.53 0.04 YCO2 -0.32 0.05
YOH 0.09 0.01 YOH 0.10 0.96
YN O -0.03 1.00 YN O -0.05 0.14
the jet flame) suggest that VQPCA actually tends to FPCA when dealing with
relatively simple systems, characterized by fast chemistry and a small degree
of extinction. This is confirmed by Figure 4.29, showing the VQPCA (a-d) and
FPCA (a’-d’) partition of the DNS1 data. Both the approaches identify a rich
and lean region, together with a rich and lean reacting layer.
Computational cost of the analysis In the above discussion, VQPCA

has been showed to be generally superior to FPCA from the point of view
of reconstruction error minimization. However, it should be reminded that
VQPCA is an iterative algorithm, whereas FPCA is based on the supervised
partitioning of data into bins of mixture fraction (Section 4.3). Therefore,
the CPU time associated with VQPCA is certainly higher that that of FPCA;
moreover, it increases with k, as shown in Figure 4.30, for an experimental
and numerical data set. It is clear how the CPU associated with VQPCA can
reach values of the order of minutes, for the experimental data sets (Figure
4.30 (a)), and hours, for the numerical data sets (Figure 4.30 (b)), whereas
the corresponding CPU time of FPCA is of the order of seconds and minutes.
Therefore, FPCA represents certainly a valid solution for applications similar
to the jet or the DNS1 flame, as it optimizes both CPU time and accuracy of
predictions.
4.6 Development of a PCA based combustion model

In the previous Section, a methodology based on Principal Components Anal-
ysis (PCA) has proposed for the identification of low-dimensional manifolds
in turbulent flames, the estimation of their dimensionality and the selection
of optimal reaction variables. The reduced representation given by PCA has
great potential, especially in its local formulations, i.e. VQPCA and FPCA.
122
4.6. Development of a PCA based combustion model
(a) (b)
(c) (d)
(a’) (b’)
(c’) (d’)
JHC data set. GSRE,n = 0.08.
Figure 4.30: CPU time associated with the FPCA and VQPCA reductions.as a
function of the number of clusters, k, and retained PCs, q, for the experimental
(a) and numerical (b) data sets.
123
In fact, the selection of optimal variables for turbulent reacting systems could
be exploited for the development of turbulence-chemistry interaction models.
In this context, the linearity of the PCA method is extremely appealing. In
fact, if a set of reaction variables is selected, only few linear combinations of
the original variables need to be transported in a numerical simulation.
4.6.1 Transport equations for the PCs

Recalling the conservation equation for a reacting species (Chapter 2), it is
possible to show that the equation

∂ρYk ∂ρuj Yk ∂ ∂Yk
+ = ρDk + ω̇k
can be transformed into a transport equation for the reacting species. Intro-
ducing the material derivative and the Lewis number, the species equation
becomes:
λ ∂2

DYk ∂Yk
ρ = + ω̇k (4.46)
Dt cp Lek ∂x2j ∂xj
where we have assumed the density and species diffusivity to be constant.
If the species mean, Y k , is subtracted and a scaling factor, dk , is applied to
the centered variable, we get:

(Yk −Y k )
D " #
dk λ ∂2 Yk − Y k ω̇k
= + . (4.47)
Dt ρcp Lek ∂x2j dk ρdk
Indicating with aki the weight of the kth variable on the ith PC, the following
equation is obtained:

(Yk −Y k )
D aki " #
dk λ ∂2 Yk − Y k ω̇k aki
= aki + . (4.48)
Dt ρcp Lek ∂x2j dk ρdk
Summing over all the variables we have:

Pp (Yk −Y k )
D aki " p #
k=1 dk λ ∂ 2 X Yk − Y k
= aki +
Dt ρcp Lek ∂x2j dk
k=1
p
1 X ω̇k aki
+ . (4.49)
ρ dk
k=1
124
(Y −Y )
But pk=1 k dk k aki is simply the definition of the ith PC, zi . Therefore Eq.
P
(4.49) can be rewritten as:
Dzi λ ∂ 2 zi
= + ω̇zi (4.50)
Dt cp Lei ∂x2j
where ω̇zi is the source term for zi
p
1 X ω̇k aki
ω̇zi = . (4.51)
ρ dk
k=1
If temperature is also included in the PCs definition, Eq. (4.51) becomes:

p
Qr 1 X ω̇k aki
ω̇zi = + (4.52)
ρcp dT ρ dk
k=1
where Qr is the heat released by reaction, akT is the weight of temperature on

the ith PC and dT is the scaling factor for temperature.
A prerequisite for the validity of the above equation is that the matrix of
PCs coefficients A is constant. Being the proposed PCA analysis based on the
processing of a multitude of observations in both space and time, A is constant
by construction,
More compactly, we can express Eq (4.50) as:
D
(Z) = −∇ · (jZ ) + (ω̇Z ) , (4.53)
Dt
where jZ is the mass diffusive flux of Z In Eq. (4.53), the source terms of
temperature and all species contribute to the source term for each PC.
4.6.2 PCA Modeling Approach

A complete PCA modeling approach requires several ingredients. First, the PCs
must be identified using the procedure outlined in Section 4.1. This identifica-
tion requires high-fidelity, fully-resolved data including source terms. Once the
PCs are selected, transport equations may be derived for each PC as described
in Section 4.6.1.
Second, the initial conditions (ICs) and boundary conditions (BCs) on the
PCs must be defined using the transformation matrix A. For Dirichlet BCs
on all the original variables, we obtain Dirichlet conditions on the PCs (ICs
are analogously defined). Likewise, Neumann conditions on X yield Neumann
conditions on Z. Mixed conditions on X yield Robin boundary conditions on
Z.
Diffusion terms in the transport equations for Z require evaluation of the
diffusive fluxes for each component of X. In turbulent flow calculations, the
molecular diffusion term is typically augmented by a “turbulent diffusion” term
125
arising from closure of the convective term. In many cases, and particularly
at high Reynolds number, the molecular diffusion term is small relative to the
turbulent diffusion term and it is neglected. However, even when one wishes
to retain the full description of molecular diffusion, the treatment with PCA is
straightforward. First, X is approximated from Z. Next, the diffusive terms for
X, jX , are constructed. Finally, the diffusive fluxes for the PCs are calculated
as jZ = jX A.
Source terms for the PCs, ω̇Z , can be parametrized by Z and tabulated
a priori to avoid run-time calculation. The accurate parametrization of the
source terms is crucial for the successful application of PCA as a modeling
approach. Therefore, the data adopted for PCs extraction must have source
terms for all X, which is currently impossible to obtain from experimental
data. Then, the decisive step for moving on from data analysis (Section 4.5.1)
to predictive modeling is the availability of computational data generated from
reliable chemical mechanisms, using methods such as DNS or ODT. Further-
more, the reliability of PCA as a modeling approach also hinges on the relative
invariance of PCs from one data set to another which is nearby in parameter
space.
4.6.3 Parametrizing the State Variables

This section presents the results of PCA applied to two DNS data sets of
non-premixed CO/H2 combustion. The DNS data-sets (Case A and B) have
been obtained using a code with 8th order spatial and 4th order temporal dis-
cretization. Detailed kinetics of CO/H2 oxidation have been used [114], along
with mixture-averaged transport approximations. The fuel stream is 0.45% CO,
0.05% H2 , and 0.5% N2 , giving a stoichiometric mixture fraction of fst = 0.4375,
and both fuel and air streams are at 300 K.
Case A is a spatially-evolving jet with an initial χmax = 25 s−1 , while case
B is a temporally-evolving jet with an initial χmax = 125 s−1 . The primary
difference between the two data-sets is the initial scalar dissipation rate (χ)
and turbulence intensity, which affects the degree of extinction observed; case
A exhibits virtually no extinction, while case B exhibits moderate extinction.
The existence of moderate extinction in case B is shown qualitatively in Figure
4.31 (a), which shows T versus χ at fst 4 . Additional details of the DNS code
and simulation configuration may be found elsewhere [120, 88].
To quantify the error in representing the data in low-dimensional space,
parametrized by Z, we calculate the R2 value,
n
" #" n #−1
X X
2
R =1− (xij − x∗ij )2 (xij − x¯j )2
(4.54)
i=1 i=1
4
The results shown in this Section refer to data conditioned on mixture fraction, f , since
this is a convenient variable to “force” as the first component.
126
(a) (b)
Figure 4.31: Parametrization of temperature at fst by χ (a) and z1 (b) for case
B. Solid lines are the doubly-conditional mean temperature. R2 is calculated
from Eq. (4.54).
where xi is the ith observation of the jth variable, x∗ij is its parametrized ap-
proximation, and x̄j is the mean of xj . For the state variables, R2 is equivalent
to the parameter tq,j , introduced in Section 4.2.3.1. However, for the source
term, such parameter is not available and the R2 can be directly calculated.
Figures 4.31 (a) and 4.31 (b) show the parametrization (at fst ) of T by χ
and the first PC, z1 , respectively, for Case B. Examining Figure 4.31 (b), we
see that z1 acts as a progress variable, capturing the extinction process remark-
ably well. This has also been observed for other choices of progress variables
such as CO2 [88]. Comparing the two-parameter PCA approach with the (f, χ)
parametrization is reasonable since both are two-parameter models, although
the second parameter (χ versus z1 ) represents different physical phenomena
(gradient versus chemical state). Figure 4.32 shows the parametrization of the
OH mass fraction by the common (f, χ) and the proposed (f, z1 ) parametriza-
tions. This demonstrates that the PCA approach can be used to represent a
wide range of the state variables, not temperature alone.
Also shown on Figures 4.31 (a) and 4.31 (b) is the R2 value as calculated by
Eq.(4.54). Table 4.17 lists R2 values for reconstruction of the temperature and
all species mass fractions as a function of the number of parameters adopted,
q. These values are a concise, quantitative representation of the information
presented graphically in Figs. 4.31 and 4.32. For example, for Case B with
q = 1, we obtain RT2 = 0.967, corresponding to Figure 4.31 (b). For comparison,
Table 4.17 also lists the R2 values given by the (f, χ) parametrization, RT2 =
0.801 (Figure 4.31 (a)). Clearly, the two-parameter (f, z1 ) parametrization
reconstructs the temperature and most other state variables with much more
accuracy than the (f, χ) parametrization. It should be noted that the results for
the (f, χ) parametrization represent the best possible performance of a model
based on (f, χ); the steady laminar flamelet model typically does not perform
ideally [88].
127
Table 4.17: R2 values defined by Eq. (4.54). Also shown are results for the χ parametrization. All results are at f = fst = 0.4375.
q T H2 O2 O OH H2 O H HO2 H2 O 2 CO CO2 HCO
A χ 0.789 0.344 0.811 0.718 0.165 0.085 0.695 0.839 0.816 0.803 0.827 0.828
1 0.983 0.259 0.976 0.930 0.240 0.178 0.823 0.986 0.916 0.978 0.956 0.980
128
2 0.983 0.936 0.968 0.958 0.963 0.924 0.964 0.980 0.985 0.969 0.976 0.980
B χ 0.801 0.509 0.807 0.697 0.426 0.186 0.648 0.665 0.729 0.810 0.058 0.817
1 0.967 0.370 0.910 0.614 0.736 0.531 0.524 0.940 0.849 0.907 0.094 0.901
2 0.996 0.845 0.982 0.882 0.931 0.990 0.858 0.974 0.941 0.981 0.378 0.984
3 0.990 0.904 0.982 0.984 0.979 0.991 0.985 0.977 0.933 0.981 0.854 0.980
(a) (b)
Figure 4.32: Parametrization of OH mass fraction at fs t by χ (a) and z1 (b)

for case B. Solid lines are the doubly-conditional mean temperature. R2 is
calculated from Eq. (4.54).
Table 4.17 also demonstrates that increasing the number of retained PCs
increases the accuracy with which the state variables are represented. This
indicates that one may select a desired error threshold and then determine the
minimum number of PCs required to achieve that accuracy. Conversely, one
may choose the number of PCs and estimate a priori the associated error.
4.6.4 Parametrizing Source Terms
The PCs are not conserved variables and their source terms must be parametrized
by the PCs. In this section we explore the ability of PCA to parametrize
source terms. Any function of X may be approximated by F(X) ≈ F (XAq ) .
However, it is more accurate to calculate F(X) directly from the data in p-
dimensional space and then project it onto Z by calculating the conditional
mean hF (X) |Z i. Thus, source terms are calculated directly from the original
observables, X, and their conditional means are projected onto Z. Figure 4.33
illustrates this for the two-dimensional (f, z1 ) parametrization of ω̇z1 .
Table 4.18 summarizes the ability of an q-dimensional PCA to parametrize
the source terms of the PCs. We first consider the columns describing the
results at fst . For case A, a two-dimensional parametrization (f, z1 ) captures
ω̇z1 with Rω̇2 z = 0.978. For case B, 3 PCs are required to parametrize ω̇z1 to
1
a similar degree of accuracy. Comparing the dimensionality requirements for
parametrizing ω̇Z with those for parametrizing the state variables (Table 4.17),
we see that parametrizing the source terms does not require more PCs than
parametrization of the state variables themselves, an encouraging result.
129
(a)
Figure 4.33: Parametrization of ω̇z1 at fst by z1 for case B. Solid line: doubly-
conditional mean value of ω̇z1 . R2 is calculated from Eq. (4.54).
Table 4.18: R2 values defined by Eq. (4.54) for PC source terms, sη .
f = 0.2 f = fst = 0.4375 f = 0.6

q 1 2 3 1 2 3 1 2 3
A ω̇z1 0.993 0.985 - 0.978 0.985 - 0.923 0.934 -
ω̇z2 - 0.996 - - 0.922 - - 0.876 -
B ω̇z1 0.270 0.844 0.967 0.815 0.932 0.958 0.809 0.852 0.902
ω̇z2 - 0.835 0.955 - 0.951 0.961 - 0.883 0.909
ω̇z3 - - 0.976 - - 0.731 - - 0.831
130
4.7. Summary
4.6.5 Global versus Semi-Local PCA

The results presented thus far have been obtained “locally” at fst . One may
consider whether a PCA performed at fst is applicable at other f . We term
this a “semi-local” PCA. If the PCA is highly dependent on mixture fraction,
then one of two options must be considered
• Eliminate the mixture fraction as a parameter and seek a global PCA on
the entire data set. This approach typically requires more PCs than a
PCA obtained at fst (Section 4.5.1).
• Perform a local PCA (Section 4.5.1) in f -space and derive transport equa-
tions for Z|f . These equations would have exchange terms representing
transport in mixture fraction space. This approach is further complicated
by the fact that the definition of the PCs would vary with f .
If the PCA obtained at fst reasonably represents the data at other f , then
the transport equations derived in Section 4.6.2 may be used directly at all f ,
eliminating the need for conditional equations in f -space.
Tables 4.19 and 4.20 provide parametrization errors for the state variables
at f = 0.2 and f = 0.6, respectively. Table 4.18 shows the parametrization
errors for ω̇Z at f = 0.2 and f = 0.6. Interestingly, the parametrizations do
not perform well at lean conditions (especially for CASE B); the same is true for
the (f, χ) parametrization. A posteriori testing is necessary to fully determine
the parametrization accuracy required. However, these results show promise
for the ability to use a PCA obtained at fst globally.
4.7 Summary
In the first part of the present Chapter, a novel methodology based on Principal
Components Analysis (PCA) has been proposed for the identification of low-
dimensional manifolds in turbulent flames, the estimation of their dimensional-
ity and the selection of optimal reaction variables. To this purpose, high fidelity
experimental and numerical data sets have been investigated. Three different
PCA approaches are proposed. A global PCA analysis, GPCA, has been com-
pared to two local PCA models, VQPCA and FPCA, based on the partitioning
of the experimental data into separate clusters where PCA is performed locally.
However, the partitioning algorithm used by VQPCA is unsupervised and based
on reconstruction error minimization while FPCA conditions the data a priori
on the mixture fraction. Results show that the local PCA approaches (VQPCA
and FPCA) outperform the global approach in all cases. Indeed, GPCA is un-
able to provide a compact representation of the data in a low-dimensional space
due to the highly non-linear relationships existing among the state variables.
Regarding the local approaches, the performances of VQPCA and FPCA are
comparable for a simple jet flames, while FPCA proves unable to capture im-
portant features for systems characterized by complex equilibrium phenomena
131
Table 4.19: R2 values at f = 0.2 using the PCA obtained at fst . Also shown are results for the χ parametrization.
q T H2 O2 O OH H2 O H HO2 H 2 O2 CO CO2 HCO
A χ 0.097 0.798 0.169 0.774 0.736 0.245 0.827 0.812 0.811 0.580 0.432 0.881
1 0.500 0.413 0.816 0.212 0.188 0.134 0.319 0.433 0.398 0.666 0.555 0.619
132
2 0.968 0.910 0.881 0.868 0.859 0.940 0.888 0.838 0.855 0.867 0.940 0.934
B χ 0.497 0.542 0.390 0.303 0.329 0.269 0.558 0.537 0.390 0.417 0.206 0.689
1 0.979 0.741 0.866 0.337 0.219 0.749 0.127 0.805 0.858 0.859 0.513 0.403
2 0.996 0.877 0.945 0.819 0.822 0.994 0.806 0.970 0.960 0.958 0.737 0.860
3 0.990 0.963 0.958 0.989 0.977 0.994 0.978 0.984 0.982 0.968 0.808 0.955
4.7. Summary
Table 4.20: R2 values at f = 0.6 using the PCA obtained at fst . Also shown are results for the χ parametrization.
q T H2 O2 O OH H2 O H HO2 H2 O 2 CO CO2 HCO

A χ 0.676 0.190 0.740 0.642 0.548 0.073 0.434 0.741 0.727 0.467 0.572 0.555
1 0.956 0.287 0.958 0.887 0.587 0.076 0.542 0.966 0.867 0.836 0.868 0.751
2 0.959 0.962 0.949 0.923 0.775 0.826 0.768 0.955 0.898 0.911 0.919 0.889
133
B χ 0.628 0.081 0.593 0.662 0.112 0.246 0.365 0.508 0.616 0.521 0.268 0.570
1 0.964 0.134 0.904 0.804 0.197 0.755 0.721 0.938 0.650 0.844 0.442 0.896
2 0.984 0.612 0.928 0.836 0.373 0.986 0.822 0.960 0.791 0.873 0.542 0.930
3 0.986 0.769 0.948 0.888 0.543 0.991 0.913 0.967 0.839 0.909 0.841 0.941
(e. g. Flame F, JHC, DNS2), resulting in higher reconstruction error with

respect to VQPCA.
In the second part of the Chapter, a modeling approach based on PCA has
been proposed and tested a priori using DNS data. This modeling approach
is complete, with the exception of a turbulent closure model which would be
required if this model were used in a LES or RANS context. The model is based
on a rotation of the thermochemical state basis from one based on temperature,
pressure, and Ns − 1 species mass fractions to one which best represents the
variance in the data. Implementation of the model requires transport equa-
tions for the principal components, which are reacting scalars. Results from
a quantitative a priori analysis of this approach using DNS data show great
promise. State variables and source terms both are parametrized well by a
two-parameter (f, z1 ) model, and adding additional parameters provides a
significant increase in accuracy for all state variables and their reaction rates.
Results also indicate a uniformly better representation of the DNS data using
an (f, z1 ) parametrization over the commonly used (f, χ). There are many
potential applications of this modeling approach. For example, laminar flame
calculations could benefit from PCA modeling approaches to provide rapid
solutions using a reduced set of equations. Once a full calculation has been
performed, subsequent calculations may be performed using PCs rather than
the full set of species and energy equations. The number of PCs retained can be
chosen by the desired accuracy. This could be particularly useful for stochastic
models such as the Linear Eddy Model (LEM) [121] and the one-dimensional
turbulence (ODT) model [115, 122], which require many realizations of a flow
field. The first realization could employ full chemistry while subsequent realiza-
tions utilize a reduced set of equations defined by PCA. Another application is
in modeling turbulent flows, where a compact parametrization of the thermo-
chemical state is key to achieving affordable simulations. Additionally, while
the analysis presented herein has been applied to non-premixed combustion, the
PCA approach applies in principle to all combustion regimes from premixed to
non-premixed. For application to turbulent flows, additional closure models
are required for the unresolved convective and source terms. In the context of
transported PDF methods, a PCA modeling approach could drastically reduce
the computational cost by significantly reducing the thermochemical dimen-
sionality while maintaining a quantified error bound on the thermochemical
reduction.
Future work will focus on examining the feasibility of PCA with various
fuels and exploring the universality of the PCA, i.e. the applicability of a PCA
obtained under one set of conditions to be applied to a simulation at another
set of conditions. Also, a posteriori tests will be conducted to determine the
effect of nonlinear propagation of errors in source term parametrization.
134
Chapter 5
Experimental and numerical

investigation of a
self-recuperative flameless
burner
The present Chapter describes the experimental and numerical activity carried
out on a self-recuperative burner designed to operate in flameless combustion
regime. The characterization has proceeded throughout the duration of the
PhD program and has allowed to investigate the system under widely different
operating conditions. In particular the burner is designed to operate with
natural gas; however, the utilization of hydrogen-methane mixtures has been
also explored, to assess the feasibility of flameless combustion with hydrogen
enriched fuels. Moreover, the burner has been studied in ultra-lean conditions,
to evaluate the possibility of exploiting flameless combustion for the reduction
of NO emissions in the oxidation of low calorific value fuels, enriched with
hydrogen.
This Chapter starts with the description of the burner, followed by a brief
summary of the experimental campaigns carried out at the experimental fa-
cilities of ENEL Ricerca in Livorno, Italy. Then, the numerical modeling of
the burner is presented, with particular emphasis on the modeling choices
adopted to achieve predictivity in the characterization of the environmental
performances of the system, i.e. NO emissions.
The results showed in the present Chapter have been published in [78, 79].
5.1 Description of the burner

The burner is a 13 kW nominal power FLOX® burner, designed and manu-
factured by WS, and belongs to ENEL Ricerca, Livorno, Italy. A longitudi-
nal section of the burner is shown in Figure 5.1. The combustion chamber is
135
Chapter 5. Experimental and numerical investigation of a self-recuperative
flameless burner
Figure 5.1: Longitudinal section of the FLOX® burner.
delimited by a radiant tube closed at the upper end, to ensure that the com-
bustion products are kept separated from the process atmosphere. Such design
is commonly adopted in furnaces devoted to the steel thermal treatment, i.e.
annealing process [2]. Usually, the furnaces are equipped with several radiant
tubes which provide indirect heating to the steel stocks placed in a hydrogen
rich atmosphere. The use of flameless burners in conjunction with radiant tubes
is particularly appealing for the steel industry, as it provides a more uniform
temperature environment within the combustion chamber, thus improving the
effectiveness of the thermal treatment and extending the time to replacement
of the heating units.
The combustion chamber is cylindrical with a radius of 0.045 m and a length
of 0.58 m. It operates with an internal recirculation of exhaust gases which is
promoted by a 0.41 m long flame tube with a radius of 0.02 m, positioned inside
the combustion chamber. The flame tube is equipped with three windows in
the lower part which allow the recirculation of exhaust gases in the reaction
region. A 3D view of the combustion chamber with the recirculation windows
is depicted in Figure 5.2. The burner is self-recuperative, which means that the
inlet air is preheated with exhaust gases by means of a corrugated-surface heat
exchanger (Figure 5.3).
The burner can work in either flame or flameless combustion modes. Ini-
tially, the combustion chamber is heated up with a conventional burner attached
flame, as flameless combustion requires process temperatures higher than the
self-ignition temperature of the fuel. Once such temperature levels are reached,
the burner switches to flameless mode. Fuel and air are fed coaxially into the
136
5.1. Description of the burner
Figure 5.2: 3D view of the combustion chamber.
Figure 5.3: Outer surface of the air preheater.
137
flameless burner
(a) (b)
Figure 5.4: Inconel® shield (a) and outer insulation layer (b).
combustion chamber through separated jets, so that fresh air can entrain large
amounts of burnt gases before reaction takes place. Mixing of products and
reactants allows both to preheat the fuel above its self ignition temperature
and to dilute the reacting mixture, thus promoting the development of a volu-
metric combustion process. For safety reasons, a ON/OFF valve is present in
the fuel feeding line. Such device interrupts the feeding of the fuel when the
temperature in the burner exceeds a certain limit.
With the exception of the fuel feeding system, which is steel made, the
burner is made of silicon carbide (SiC). Therefore, the burner can operate at
higher temperatures, around 1300°C, thus reaching very high thermal efficien-
cies with low maintenance costs, due to the excellent chemical behavior of SiC.
The use of silicon carbide is very appealing for combustion systems; however,
its application is limited by the high costs and fragility of the material.
Outside the combustion chamber, and coaxially to the radiant tube, an
Inconel® shield (Figure 5.4 (a)) and a water heat exchanger (Figure 5.5) are
placed in the experimental apparatus, to replicate heat losses towards the sur-
rounding of real burner operations. Moreover, two insulation layers cover the
inner and outer surfaces of the Inconel® shield (Figure 5.4 (b)).
With regard to the experimental measurements, the flow rates of inlet air,
fuel and water, together with the chemical composition of exhaust gases (CO,
O2 , CO2 , NO), are available. Moreover, the temperature profiles along the
radiant tube and Inconel® shield are characterized with multiple thermocouples
(Figure 5.6 (a)). Unfortunately, no measurements are taken inside the burner,
due to its industrial characteristics and the absence of any access window. A
picture of the burner assembly is shown in Figure 5.6 (b).
The experimental campaigns on the FLOX® burner have been carried out
in two different periods, November-December 2004 and December 2006. Dur-
ing the first campaign, some operating issues were encountered, namely the
138
(a) (b)
Figure 5.5: Water heat exchanger during the first (a) and second (b) experi-
mental campaigns.
(a) (b)
Figure 5.6: Outer surface of the air preheater.
139
flameless burner
Figure 5.7: NO concentration as a function of time during a typical run of the

first experimental campaign.
maximum heat removed by the water heat exchanger (Figure 5.5 (a)) resulted
not sufficient to ensure stable operations of the systems. In other words, the
ON/OFF valve continuously interrupted the fuel flow inside the burner to pre-
vent system damage, thus leading to an unsteady behavior of all the output
variables. Figure 5.7 shows a typical plot of NO concentration versus time dur-
ing an experimental run. A considerable amount of work has been done in order
to extract valuable information from this experiment campaign. In particular, a
procedure based on data extrapolation, validated using the V&V methodology
described in Chapter (3) has been used, as described in the following text.
Finally, during the planning of the second experimental campaign (Decem-
ber 2006), the water heat exchanger has been re-designed in order to satisfy
the heat exchange requirements, thus avoiding the control associated with the
ON/OFF valve. The new water jacket consists of a spiral cooling coil, realized
through fins welded on a cylindrical surface (Figure 5.5 (b)), providing signif-
icantly more exchange surface with respect to the original system (Figure 5.5
(a)). The dimension of the spiral coil has been designed in order to optimize
the heat transfer coefficient, because the water flow rate is fixed.
5.1.1 Experimental campaign n.1

Table 5.1 summarizes the experimental campaign carried out on the FLOX®
burner during November-December 2004. The burner was investigated both
with natural gas (NG) and hydrogen enriched mixtures, as indicated in Table
5.1.
As mentioned in Section 5.1, during the first experimental campaign the
140
Table 5.1: Summary of the experimental campaign n.1 on the FLOX® burner.
Run Q̇in [kW] ṁF [kg/s] YH2 ṁair [kg/s] air [%] Q̇H2/Q̇in
1 13.4 2.67E-04 0.0 5.24E-03 14 0

2 12.5 2.50E-04 0.0 4.80E-03 11 0
3 12.2 2.44E-04 0.0 4.53E-03 8 0
4 13.5 2.70E-04 0.0 5.23E-03 13 0
5 13.0 2.61E-04 0.0 4.80E-03 7 0
6 12.5 2.49E-04 0.0 4.53E-03 6 0
7 13.7 2.73E-04 0.0 5.12E-03 9 0
8 12.8 2.55E-04 0.0 4.80E-03 9 0
9 11.9 2.39E-04 0.0 4.49E-03 9 0
10 9.9 1.98E-04 0.0 3.69E-03 9 0
11 9.9 1.99E-04 0.0 3.69E-03 8 0
12 9.1 1.82E-04 0.0 3.31E-03 6 0
13 9.8 1.96E-04 0.0 3.69E-03 10 0
14 9.0 1.81E-04 0.0 3.31E-03 6 0
15 13.4 2.62E-04 1.6 5.01E-03 10 3.7
16 13.5 2.52E-04 5.3 5.01E-03 10 11.8
17 13.6 2.46E-04 7.5 5.01E-03 10 16.3
18 13.0 2.40E-04 5.4 4.65E-03 7 12.1
19 13.0 2.50E-04 2.4 4.65E-03 6 5.6
20 13.0 2.40E-04 5.5 4.65E-03 7 12.2
21 13.0 2.43E-04 4.5 4.39E-03 1 10.2
22 13.0 2.50E-04 4.1 4.39E-03 0 9.3
23 13.0 2.50E-04 2.0 4.39E-03 0 4.6
24 13.7 2.55E-04 5.6 5.23E-03 13 12.5
141
flameless burner
Figure 5.8: O2 concentration in the flue gases with NG (a) and NG-H2 mixtures
(b), for the experimental campaign n. 1 on the FLOX® burner.
water heat exchanger was insufficient to completely remove the radiative heat
flux coming from the radiant tube, thus leading to a continuous action of the
temperature based control, leading to unsteady conditions (Figure 5.7). The
regulation was found to occur more frequently with NG/H2 mixture than with
NG, reflecting a more rapid increase of temperature with hydrogen. Figure 5.8
shows the effect of the regulation cycles on the O2 content in the flue gases,
for NG and NG/H2 mixtures. It can be observed that the regulation occurs
approximately every 150 s for NG/H2 mixtures, whereas a 180 s interval is
observed with NG.
Figure 5.9 shows NO emissions for NG and NG/H2 mixtures, apparently
indicating no correlation between NO emissions and H2 content in the fuel.
Actually, a more careful analysis of NO emissions in the flue gases (Figure
5.7) shows that the interruption of the fuel feeding system occurs before NO
has reached steady values. Consequently, efforts need to be spent to extract
useful information from the available measurements, to make them comparable
to the successive numerical modeling activities. To this purpose, the V&V
methodology described in Chapter 3 has been exploited to determine reliable
asymptotic values for the NO measurements. The procedure adopted is briefly
discussed in the following Section.
5.1.1.1 Extrapolation of NO emissions to steady state

The extrapolation of the NO emissions to their steady values has been carried
out by means of exponential functions, due to the observation that the plot
of NO emissions as a function of time follows an exponential law and tend to
142
Figure 5.9: NO emissions for NG and NG-H2 mixtures for the experimental
campaign n. 1 on FLOX® burner.
an asymptote representing the steady value. The proposed models have been
compared to the experiments in the range of measurements available, to choose
the most reliable model for carrying out the extrapolation. To this purpose,
the V&V validation metrics have been adopted (Chapter 3), to quantitatively
assess the level of agreement between experiments and models.
The first step towards the construction of the validation metrics is the ex-
traction of statistics from the experimental data, to determine the parameters,
i.e. mean and standard deviation, required to estimate the confidence intervals
for the experiments. Figure 5.10 show the experimental measurements, the
estimated mean and standard deviation of NO concentration versus time, for
experimental run 5 in Table 5.1. It can be observed how the standard deviation
is initially very large and progressively reduces as we move to the asymptotic
region.
To model such observed behavior, two mathematical models have been
adopted. The first is the classic Box-Lucas exponential function:
N Osteady = A 1 − e−Bt (5.1)

where A and B are parameters to be determined using the experimental mea-

surements. The second model, Exp3P, assumes a different form for the asymp-
totic function:
B
N Osteady = Ae− C+t (5.2)
where A, B and C are, again, parameters to be determined. Actually the num-
143
flameless burner
Figure 5.10: Raw data, experimental mean and standard deviation for the NO
emissions vs. time. Experimental run n. 5, Table 5.1.
bers of parameters in Eq. (5.2) reduce to 2, forcing the curve to equal the
mean observed value at t = 0. The determination of the unknown parameters
for the two models is carried out using a quadratic error minimization algo-
rithm available in the software package MATLAB® . Figure 5.11 shows the
98% confidence intervals for the estimated NO mean, together with the model
predictions. It can be observed that both models are within the experimental
confidence intervals for small values of t. However, the Box-Lucas model (Eq.
(5.1)) tends to overpredict the initial NO increase and ultimately underpredicts
NO values in the asymptotic region, stepping out from the estimated confidence
intervals. On the other hand, the Exp3P model indicates a better agreement
with the experimental mean, especially in the asymptotic region.
This is further confirmed by Figure 5.12, which shows the estimated error
for the Box-Lucas and Exp3P models, with the 98% confidence intervals for
the experimental mean. The prediction of the Exp3P model are almost always
within the experimental confidence interval, especially close to the steady state
region, whereas the Box-Lucas results in an underprediction of NO values in this
zone. Therefore, on the basis of the validation activity described, it was chosen
to adopt the Exp3P model for the extrapolation of the steady value of NO
emissions. It is true that such approach carries a certain degree of uncertainty;
however, the V&V methodology provides also a way to estimate the confidence
in the prediction, by means of the confidence intervals1 .
1
Here, it is assumed that the experimental behavior observed at t = 9 s, which corresponds
to the last experimental observation, can be extrapolated at steady state.
144
Figure 5.11: Experimental mean plus 98% confidence intervals and comparison
with the mathematical models for the NO emissions vs. time. Experimental
run n. 5, Table 5.11.
Figure 5.12: Estimated error plus 98% confidence interval showing the level
of agreement between experimental measurements and mathematical models.
Experimental run n. 5, Table 5.12.
145
flameless burner
Table 5.2: Extrapolated NO emissions for a selected subset of experimental

runs listed in Table 5.1.
Run Q̇in [kW] YH2 [-] XN Om [ppmv ] XN Om [ppmex ]

1 13.4 0 53 62
5 13.0 0 67 84
16 13.5 5.3 51 89
20 13.0 5.5 61 132
22 13.0 4.1 56 100
23 13.0 2.0 52 73
The original and extrapolated NO values for some of the runs listed in Table
5.1 are shown in Table 5.2.
5.1.2 Experimental campaign n. 2

The experimental campaign n. 2 was carried out in December 2006 at ENEL
Ricerca. The FLOX® burner was fed with NG-H2 mixtures in ultra-lean con-
ditions, to allow the hydrogen content to be increased up to 50% by weight. A
summary of the runs performed is shown in Table 5.3.
It can be noticed that experiments were carried out at 3 different burner
thermal inputs (i.e. 8, 10 and 12 kW) and for different H2 loads. Specific
runs (10, 11, 12 and 13) were planned to investigate the effect of air excess
on flameless operation mode. During the experimental campaign n. 2, the
burner was equipped with the new cooling system (Figure 5.5 (b)), described
in Section 5.1.1; therefore, the ON/OFF safety valve was not found to act on
the fuel feeding system.
Figure 5.13 (a) shows the NO emissions for different burner thermal inputs
and H2 loads, whereas Figure 5.13 (b) shows NO emissions obtained with dif-
ferent air excesses at a fixed burner load, i.e. 10 kW, and H2 thermal input, i.e.
0.3. Figure 5.14 shows temperature measurements along the radiant tube (a),
the inner (b) and outer (c) surfaces of the Inconel® shield and the inner sur-
face of the water heat exchanger (d). Thermocouples denoted with DOWN and
UP are positioned at about 1/4 and 3/4 of the burner length from the injection,
respectively.
5.2 Numerical modeling of the self-recuperative burner

The present Section reports the modeling choices adopted in the numerical
modeling of the self-recuperative burner. The approach varied during the de-
velopment of the PhD program, depending on the fuel characteristics (methane
146
5.2. Numerical modeling of the self-recuperative burner
Run Q̇in [kW] ṁF [kg/s] YH2 ṁair [kg/s] air [%] Q̇H2/Q̇in
1 8 1.38E-04 0.09 7.87E-03 217 0.2

2 8 1.34E-04 0.15 7.88E-03 208 0.3
3 10 2.07E-04 0.00 7.46E-03 118 0.0
4 10 1.96E-04 0.04 7.35E-03 117 0.1
5 10 1.90E-04 0.09 7.74E-03 126 0.2
6 10 1.65E-04 0.15 7.71E-03 144 0.3
7 10 1.45E-04 0.29 7.87E-03 152 0.5
8 10 1.29E-04 0.39 7.73E-03 158 0.6
9 10 1.18E-04 0.50 8.01E-03 170 0.7
10 10 1.67E-04 0.15 8.10E-03 153 0.3
11 10 1.69E-04 0.15 6.39E-03 99 0.3
12 10 1.75E-04 0.14 5.64E-03 70 0.3
13 10 1.75E-04 0.14 4.74E-03 43 0.3
14 12 2.45E-04 0.00 7.29E-03 81 0.0
15 12 2.36E-04 0.04 7.25E-03 79 0.1
16 12 2.19E-04 0.11 7.19E-03 78 0.2
17 12 2.04E-04 0.16 8.16E-03 108 0.3
18 12 2.01E-04 0.15 7.48E-03 94 0.3
147
flameless burner
Figure 5.13: NO emissions as a function of the H2 load (a) and O2 concentration

in the flue gases (b). Experimental campaign n. 2 on the FLOX® burner (Table
5.3).
Figure 5.14: Temperature measurements on the radiant tube (a), on the internal
(b) and external (c) surface of the Inconel® shield and on the internal surface
of the water heat exchanger (d). H2 thermal input = 30%. Experimental
campaign n. 2 on the FLOX® burner (Table 5.3).
148
or hydrogen enriched fuels), the operating conditions (lean and ultra-lean condi-
tions) and, importantly, the features of the commercial codes used. Therefore, a
separated discussion will be provided for the numerical modeling of the two ex-
perimental campaigns and, within experimental campaign n.1, for the flameless
combustion of methane and methane-hydrogen mixtures.
5.2.1 Modeling approach for experimental campaign n.1

A subset of the experimental runs shown in Table 5.1 was selected to carry
out numerical simulations, as indicated in Table 5.4. From Table 5.4 it can be
observed that runs 1m and 2m were simulated only numerically, to study the
effects of higher concentrations of H2 in the fuel (up to 20% by wt.) on the
results. The thermal input, Q̇in , and air excess air (Column 5, Table 5.4), of
runs 1m and 2m have been set to make them directly comparable to runs 5 and
20.
The mathematical modeling has been carried out with the commercial codes
CFX 5.7.1 and FLUENT 6.3, both owned by Ansys Inc. In particular, CFX has
been adopted in a preliminary stage, to simulate the flameless combustion of
methane and to investigate the effect of the geometrical model, i.e. 2D vs. 3D,
and of the air inlet velocity2 on the results. Then, FLUENT has been used to
model the flameless combustion of methane-hydrogen mixtures. In this phase,
detailed kinetic mechanisms have been incorporated into the code to study the
effect of turbulence-chemistry interaction models on the NO predictions. The
mathematical models employed in the two stages are discussed in the following
Sections.
5.2.1.1 Flameless combustion of methane

Runs 1, 4, 7 and 8 in Table 5.4 have been simulated with the commercial code
CFX 5.7.1 by Ansys Inc. and refer to the flameless combustion of methane.
For Run 1 (Table 5.4), the air inlet cross-sectional area has been reduced from
339 down to 41 mm2 , to change the recirculation degree of exhaust gases, such
a recirculation being driven by the jet momentum of inlet air. Actually, the
real burner operates with just one air inlet cross-sectional area, i.e. 88 mm2 .
Values of Aairin for each run are listed in the second column of Table 5.4.
Computational domain and grid The computational domain consists of

three sub-domains: a fluid domain representing the real combustion chamber,
and two solid domains representing the flame and the radiant tubes. Two
different approaches have been adopted for the generation of the computational
grid.
2
The air jet momentum is responsible for the entrainment of the exhaust gases, which
determines the development of the diluted combustion regime. The air inlet velocity has
been varied, by reducing/increasing the air inlet cross-sectional area.
149
Run Aairin kR [%] YO2 ,m air [%] Exp CFD Grid Combustion Kinetic Code
[mm2 ] Model Mechanism
1 41, 46, 64, 136 0.120 14 X X 2D, 3D ED/FR 1-step CFX
88, 201, 339
4 88 135 0.118 13 X X 3D ED/FR 1-step CFX
5 88 129 0.118 7 X X 2D EDC 1-step, DRM, FLUENT
GRI
7 88 134 0.115 9 X X 3D ED/FR 1-step CFX
8 88 134 0.115 9 X X 3D ED/FR 1-step CFX
150
GRI
GRI
GRI
GRI
flameless burner
1m 88 132 0.118 7 - X 2D EDC 2-step, DRM, FLUENT

GRI
2m 88 136 0.120 7 - X 2D EDC 2-step, DRM, FLUENT
GRI
(a)
(b)
Figure 5.15: 3D (a) and axisymmetric (b) grids
Due to the presence of three recirculation windows (Figure 5.2), a 120 de-
grees angular sector has been modeled. Three different grids have been gener-
ated and a grid convergence analysis has been carried out (Chapter 3) to choose
the optimal mesh resolution. The selected grid consists of around 350.000 cells
and contains parts meshed with tetras (near burner zone) and parts meshed
with hexahedrons (reaction and reverse flow zone, flame tube and radiant tube).
Specifically, there are 143.000 tetras and 207.000 hexahedrons. The choice of
a hybrid grid is justified to reduce the number of cells required to define the
computational domain accurately, as a grid constituted only of tetras would
require a huge amount of cells. The computational grid is shown in Figure 5.15
(a).
Beside the detailed geometrical modeling, a simplified approach has been
also tested, assuming the axial symmetry hypothesis and neglecting the pres-
ence of the recirculation windows. This was made on purpose to evaluate the
effect of recirculation windows; nevertheless simulations available in literature
on the FLOX® burners are usually 2D simulations [123]. The axisymmetric
model is shown in Figure 5.15 (b). The structured grid consists of 120.000
hexahedrons, arranged on four layers describing an angular sector of 5°.
Boundary conditions As far as boundary conditions are concerned, fuels

are fed at room temperature (298 K). Conversely, the air inlet temperature is
151
flameless burner
Figure 5.16: Heat exchange between the coaxial cylindrical shields representing
the radiant tube, the internal and external insulation layers of the Inconel®
shield and the water heat exchanger.
evaluated as a function of exhaust gas temperature, on the basis of experimen-

tal data on the air preheater available from the burner supplier. A function
adjusting the air inlet temperature on the basis of the exhaust gas temperature
has been implemented into the code by means of a user-defined function.
Particular attention has been devoted to the boundary conditions of the ra-
diant tube. As it was pointed out in Section 5.1, the main mode of operation of
the burner is by radiation towards the surroundings. It becomes, then, crucial
to properly capture the radiative heat flux emitted in order to close the energy
balance in the burner. To accomplish this, the code has been coupled with a
subroutine for the evaluation of the heat exchange between the coaxial cylindri-
cal shields representing the radiant tube, the internal and external insulation
layers of the Inconel® shield and the water heat exchanger (Figure 5.16).
The radiative heat flux between radiant tube, 1, and inner insulation layer,
2’, is:

Q̇120 = 120 σA1 T14 − T240 (5.3)
where 120 = 1−1

1
R
1−
0
.
1
+F1 +R 1 0
2
0 0
12 2 2
The conductive heat transfer within the insulation layers 2’ and 2” is given
by:

2πLem kins T20 − T200
Q̇20 200 = (5.4)
ln R200/R20
neglecting the thermal resistance of the Inconel® shield. The radiative heat
flux between the outer insulation layer, 2”, and the water heat exchanger, 3, is:

Q̇200 3 = 200 3 σA200 T2400 − T34 (5.5)
152
where 200 3 = 1− 00

1
R 00
1−3
.
2 1
00
+F + R2 3
00 3
2 2 3
In the above equations T1 is the radiant tube temperature, T3 is the water
heat exchanger temperature, whereas T20 and T200 are the temperatures of the
inner and outer Inconel® insulation layers, respectively. 1 , 20 , 200 and 3 are
the emissivities of the radiant tube, the inner and outer insulation layers of the
Inconel® shield and the water heat exchanger, respectively. Lem is the emitter
length, i.e. the length of the coaxial shield radiating towards the surroundings
whereas kins is the thermal conductivity of the insulation layers. F120 and F100 2
are the view factors between coaxial cylinders.
At steady state, expressing T20 and T200 as a function of T1 and T3 , the
following expression for Q̇ is obtained:
  4
!1
4

1  Q̇ Q̇
T14 − T34 +  T14 −

Q̇ = −

1 1 σ120 A1 2πL
em kins

σ 0 A1
+ σ00 A 00


ln R200 /R 0
12 2 3 2 2
!)
Q̇
− T14 − (5.6)
σ120 A1
Equation (5.6) is implicit in Q̇; therefore, an iterative procedure is adopted for

the calculation of the heat flux.
Physical models Favre-averaged Navier-Stokes equations are solved using

the standard k- turbulence model.
As indicated in Table 5.4, a combined Eddy Dissipation Model/Finite Rate
Chemistry (ED/FR) (Chapter 2) has been chosen for the numerical simulation
of the flameless combustion of methane. Such model allows a very simplified
treatment of the finite-rate chemistry effects and it can only handle global ki-
netic mechanisms. If detailed models are taken into account, the ED/FR model
will likely produce incorrect results since the rate of production/destruction of
the species has to be controlled by the Arrhenius parameters through an ac-
curate description of turbulence-chemistry interactions. Moreover, the use of
an ED/FR approach for detailed kinetic schemes could lead to numerical in-
stabilities as it could result in sudden variations of the controlling rates within
the reaction zone. Such consideration may become particularly important in
flameless combustion, due to the enhanced turbulent mixing levels.
The limitations of ED/FR can become relevant for methane-hydrogen mix-
tures, due to the higher complexity of the resulting reaction region [45, 78, 79].
However, with regard to the oxidation of pure methane, the approach has been
considered suited to characterize the combustion regime, allowing, at the same
time, to investigate the effects of the geometric model and air inlet velocity on
153
flameless burner
the numerical results. A simple one-step global mechanism has been used to
model methane oxidation [23].
The radiative heat transfer equation (RTE) is solved using the Discrete
Transfer radiation model with a number of rays equal to 16. A coarsening
rate equal to 24 is applied to the actual grid to facilitate the radiation calcula-
tion. The radiation properties of the reacting mixture are calculated with the
Weighted-Sum-of-Grey-Gases (WSGG) model, by using the coefficients pro-
posed by Smith et al. [124], implemented into the code with a user-defined
subroutine written on purpose.
As far as NO formation is concerned, two possible sources have been taken
into account, i.e. thermal and prompt. Thermal NO formation is modeled with
a Finite Rate Chemistry combustion (FR) model and a simplified one-step
kinetic mechanism, available in the code [125] and obtained from the Zeldovich
scheme by assuming a steady state for the N radicals and relating the O radical
concentration to that of oxygen by means of the dissociation reaction [21].
The prompt NO formation from methane is modeled using a similar approach,
according to the one-step mechanism proposed by De Soete [27]. For both
thermal and prompt NO formations, the Arrhenius equation is integrated over
a probability density function (PDF) for temperature to take turbulence effects
into account.
The effect of molecular diffusion on the results has been also considered,
following Christo and Dally [45], with the Slattery and Bird [126] equation for
binary diffusion coefficients. The molecular species are assumed to diffuse in a
mixture manly constituted by nitrogen, due to the large amount of inert gases
recirculating in the reaction zone.
5.2.1.2 Flameless combustion of methane-hydrogen mixtures

The simulation of runs 5, 16 , 20, 22, 23, 1m and 2m in Table 5.4 have been
performed with the commercial code FLUENT 6.3 by Ansys Inc.
Computational domain and grid In order to emphasize the role of the

combustion model and kinetic mechanism on the results, a 2D computational
model has been adopted, to reduce the CPU time associated with the numerical
simulations. From the sensitivity analysis on the geometrical model (Section
5.2.1.1), we are aware that such simplification leads to an overestimation of the
recirculation degree by 10–15% with respect to a 3D model [78]. However, this
approximation has been considered acceptable to allow the implementation of
detailed kinetic mechanisms.
A set of three 2D grids with a number of cells ranging from 12.000 to 30.000
has been investigated. Starting from the finer grid, the other meshes have been
generated by applying a coarsening factor equal to 1.25 in each direction. The
grids are not uniform. A higher resolution has been adopted in the near burner
region and along the axis, to properly describe the reaction region; moreover,
154
the air and fuel ducts have been extended 50 mm upstream to ensure the flow
to be fully developed at the burner exit.
The solution verification methodology described in Section 3 has been ap-
plied for the selection of the computational mesh. The estimation of the solution
uncertainty, Usver , has been carried out using Logan and Nitta [81] criterion
n. 4 and assuming a convergence order p = 1, as this was the one providing
the higher values for the extrapolated error, δRE . Then, the Grid Convergence
Index (GCI) has been estimated applying a safety factor, Fs = 3, to δRE . The
estimation of the numerical error has been carried out for a cold flow simulation
with boundary conditions taken from Run 5 in Table 5.4.
Figure 5.17 shows the values of δRE and GCI for three different grids with
constant coarsening factor (h2/h1 = h3/h2 = 1.25). The plotted points have been
obtained from the radial profiles of axial velocity at different distances along
the axis, i.e. x = 0.25 m, x = 0.35 m and x = 0.45 m, averaging over all the
punctual error estimates. Following Logan and Nitta [81], the δRE and GCI
values can be regarded as the 1-sigma (68.4%) and 3-sigma (99.7%) estimates
of the solution verification uncertainty, Usver , respectively. Interestingly, very
similar values of δRE (and GCI) are obtained for different axial positions inside
the combustion chamber, thus indicating an almost uniform behavior of solution
convergence within the computational domain.
Based on Figure 5.17, it was chosen to adopt the intermediate grid as base
grid for the numerical simulations (Figure 5.18). The selected grid consists
of about 18.000 cells, with a number of finite volumes in the axial and radial
directions equal to 316 and 60, respectively. The choice of such grid have
been motivated by the need of reducing as much as possible the weight of the
computational mesh on the CPU time, to allow the implementation of detailed
kinetic mechanisms, while ensuring an acceptable numerical error.
Figure 5.19 shows the numerical error estimate for the radial profiles of
axial velocity obtained at x = 0.25 m (a), x = 0.35 m (b) and x = 0.45 m (c)
with the selected grid. The error estimate is based on the local δRE values and
provides a local measure of the intrinsic uncertainty related to the numerical
simulation.
Boundary conditions As far as the boundary conditions are concerned, the

two user-defined functions implemented in CFX (Section 5.2.1.1) have been
coupled also to FLUENT, to evaluate the air inlet temperature as a function of
the exhaust gas temperature, and to estimate the radiative heat flux emitted
by the radiant tube.
Physical models Regarding the turbulence-chemistry interaction approaches,

the ED/FR model adopted for methane combustion has been compared to EDC
(Chapter 3), which provides a more accurate coupling between the mixing pro-
cess and chemical reactions. According to EDC, combustion occurs in the re-
gions of the flow where the dissipation of turbulent kinetic energy takes place.
155
flameless burner
Figure 5.17: Grid Convergence Index (GCI) as a function of the coarsening

factor, r, for the set of 2D grids investigated in the numerical simulation of the
FLOX® burner. Assumed order of convergence p = 1 [81].
Figure 5.18: Computational mesh for FLUENT simulations.
156
Figure 5.19: Estimated error, δRE , for the profiles of axial velocity obtained at
x = 0.25 m (a), x = 0.35 m (b) and x = 0.45 m (c) with the selected grid.
157
flameless burner
Such regions are denoted as fine structures and they can be described as per-
fectly stirred reactors (PSR). The mass fraction of the fine structures and the
mean residence time of the fluid within the fine structures are provided by an
energy cascade model, which describes the energy dissipation process as a func-
tion of the characteristic scales. Within the context of flameless combustion,
the EDC hypothesis is expected to provide satisfactory results, the system ap-
proaching perfectly stirred conditions due to the intense recirculation promoted
by the internal aerodynamics. Moreover, the effect of combustion process on
turbulence is expected to be less important than in a traditional flame, due to
the flattening of the local temperature gradients.
Three different kinetic mechanisms have been used to describe the oxida-
tion of methane/hydrogen mixtures. Global kinetic rates are applied with the
ED/FR model for both CH4 [23] and H2 [59, 127]. When using the EDC model,
the global reaction approach has been compared to two detailed mechanisms,
the DRM-19 [62] and the GRI-3.0 [63]. The DRM-19 mechanism is a subset of
the GRI-1.2 [128] full mechanism, developed to obtain the smallest set of reac-
tions needed to closely reproduce the main combustion characteristics predicted
by the full mechanism, i.e. ignition delays and laminar flame speeds. It consists
of 19 species, plus Ar and N2 , for a total of 84 reversible reactions. The GRI-3.0
mechanism has been implemented without the NOx reactions, resulting in 219
reversible reactions involving 36 chemical species.
NO formation has been modeled with the same approach used in CFX
(Section 5.2.1.1), taking into account the thermal and prompt formation routes
[129, 27] and integrating the Arrhenius equation over a probability density
function (PDF) for temperature, in order to take into account the effect of
turbulent fluctuations on the mean reaction rates
Finally, the RTE equation is solved with the Discrete Ordinate (DO) model
with a number of directions equal to 64, adopting, as in CFX, the WSGG
model for predicting the spectral properties [124]. The influence of molecular
diffusion has been also taken into account with a more accurate approach than
that adopted in CFX. The binary diffusion coefficients are calculated following
the kinetic theory and a modification of the Chapman–Enskog formula [130].
Then, an effective diffusion coefficient of the species in the mixture is obtained
by applying the Wilke’s mixing rule [131]. The influence of molecular diffusion
has been investigated in the numerical simulations carried out with the EDC
model and the DRM-19 kinetic mechanism.
5.2.2 Modeling approach for experimental campaign n.2

Runs 3-9 in Table 5.3 have been selected for the numerical modeling of the ex-
perimental campaign n. 2 on the FLOX® burner. The runs are characterized
by a constant burner load, i.e. 10 kW, and by an increasing mass fraction of
hydrogen in the fuel stream. These runs were specifically planned to investi-
gate the effect of hydrogen on the combustion regime and NO formation. As it
158
was anticipated in Section 5.1.2, the experimental campaign n. 2 was carried

out using very large air excess values, to simulate the effect of H2 addition
to low-calorific value fuels. From the modeling point of view, this drastically
complicated the modeling approaches, due to the non-conventional conditions
to be analyzed. Ultra-lean combustion is characterized by moderately low tem-
peratures and reduced NO emissions. Moreover, when operating with flame-
less combustion burners, the temperature peaks are further smoothed due to
the internal aerodynamic of the system. Therefore, NO formation cannot be
modeled using conventional formation routes, i.e. thermal and prompt, and
low-temperature pathways must be taken into account, i.e. N2 O [24] (espe-
cially in lean conditions). Finally, H2 can significantly influence NO formation,
providing an intermediate species, NNH [10], which is directly converted to NO.
As far as the CFD modeling is concerned, a very similar approach to that
described in Section 5.2.1.2 has been adopted. Only the EDC model and de-
tailed kinetic schemes have been taken into account. In particular, the DRM-19
mechanism has been compared to a smaller kinetic scheme, known as KEE-58
[61], consisting of 17 species and 58 reversible reactions. As far as NO forma-
tion is concerned, two additional formation routes have been considered, beside
the conventional thermal and prompt ones. In particular, the N2 O intermedi-
ate and NNH [10] routes have been also included in the modeling approach,
to account for low-temperature formation mechanisms and for the presence of
H2 in the fuel. The NNH route is not directly available in the code; therefore,
it has been implemented by means of a user-defined function (UDF) following
Konnov et al. [25]. For the thermal and NNH intermediate mechanisms, the
radical mass concentrations, i.e. O, H, OH, are supplied by the detailed kinetic
mechanism without any equilibrium or partial equilibrium hypothesis (Chapter
1).
The drawback of the described approach lies in the simplified treatment
of NO formation, which is actually determined by hundreds of elementary re-
actions [132]. To investigate the effect of such approximation, our modeling
strategy has been compared to a different methodology, developed by the re-
search group headed by Prof. Ranzi at the Politecnico di Milano and based on
the kinetic-processing (KinPP) of CFD simulations with very detailed kinetic
schemes.
The KinPP approach is based on the general concept of reactor network
analysis [5, 6] and it has been applied to several industrial burners [133, 134]
for the prediction of different pollutants (in particular NOx , CO and unburned
hydrocarbons). The first step for the KinPP application is the CFD calculation
of the flame structure using a simplified kinetic scheme, based on a small num-
ber of species and global reactions. The knowledge of the thermo-fluid dynamic
field, as evaluated by the CFD code, allows to identify critical and non criti-
cal zones in the overall reacting system and to lump or group several adjacent
and similar cells into single equivalent reactors. The grouping or clustering of
several cells, similar from a kinetic point of view, into a single lumped reac-
159
flameless burner
tor reduces the dimensions of the overall system. A fixed average temperature
is assumed in each reactor and the rates of all the reactions involved in the
kinetic scheme are evaluated accordingly. However, in turbulent combustion
conditions, these reaction rates cannot simply be calculated as a function of
the mean temperature and composition, mainly due to the highly non-linear
dependence of reaction rates on temperature. A proper correction must be ap-
plied to take into account the effects of temperature fluctuations [133]. These
corrections are very relevant for thermal NOx reactions, due to the high activa-
tion energies. The fixed temperature inside these reactors reduces the high non
linearity of the system, mainly related to the reaction rates and to the coupling
between mass and energy balances. CFD results are then used to define the
steady-state mass balance equations for all the chemical species involved in the
detailed kinetic scheme.
For all the equivalent reactors, the mass balance of each species accounts
for convection, diffusion and chemical reaction terms. The global Newton or
modified Newton methods are not robust enough to solve the resulting system of
non-algebraic equations using CFD results as a first-guess solution. Therefore,
a better estimate of the solution is obtained by iteratively solving the sequence
of individual reactors with successive substitutions. Each reactor is solved by
using a local Newton method with the possible use of a false transient method
(time stepping), to improve the initial guess or to approach the solution. Only
when the residuals of all the equations reach sufficiently low values, a modified
global Newton method is applied to the whole system. Additional details on
the Kinetic Post-Processor and its numerical method are reported in [133, 134].
Regarding the numerical simulations to be post-processed with kinPP, it
has considered appropriate to refer to a simplified (4-step) kinetic scheme for
the CFD simulations, as indicated in Table 5.5.
Table 5.5: Global 4-step mechanism [23]. Reaction rates units: kmol · m−3 · s−1 .
Activation energy units: J · kmol−1 .
Reaction Reaction rates

2·108
CH4 + 32 O2 → CO + 2H2 O 5.01 · 1011 · e− RT 0.7
CCH 4
0.8
CO 2
1.7·108
CO + 12 O2 → CO2 2.24 · 1012 · e− RT 0.25 0.5
CCO CO 2
CH2 O
1.7·108
CO2 → CO + 12 O2 5 · 108 · e− RT CCO2
7
− 3.1·10
H2 + 12 O2 → H2 O 9.87 · 108 · e RT C H2 C O 2
The detailed kinetic scheme used in the post-processing step of the CFD
results is able to describe the oxidation of heavy hydrocarbon fuels, up to diesel
and jet fuels, and its main features were already discussed in the literature [83].
The chemistry of nitrogen compounds is also discussed elsewhere [135, 136].
The thermodynamic properties are taken from CHEMKIN Thermodynamic
160
5.3. Results
Figure 5.20: Conceptual scheme of the methodology adopted for the numerical sim-
ulation of the experimental campaign n. 2 on the FLOX burner ® .
Database [137]. The overall mechanism is freely available at the following web
address: www.chem.polimi.it/CRECKModeling.
A scheme of the conceptual path followed in the present study in shown in
Figure 5.20.
5.3 Results
The present Section describes the results of the numerical modeling of the
self-recuperative burner and the comparison between the computational results
and experimental data. The presentation of the results will follow the scheme
adopted for the description of the experimental campaigns.

The results of the numerical simulations are presented separately for the methane
and methane-hydrogen cases, respectively. As pointed out in Section 5.2.1.1,
161
flameless burner
Figure 5.21: Measured and computed temperature profiles along the radiant
tube for two different burner loads, i.e. Q̇in = 8.5 kW and Q̇in = 9.5 kW.
the numerical simulation of the flameless combustion of methane has rep-

resented a fundamental step for determining the modeling requirements, in
terms of computational domain and turbulence-chemistry interaction models,
for the successive simulation of methane-hydrogen flameless combustion (Sec-
tion 5.2.1.2).
5.3.1.1 Flameless combustion of methane

As a first step in the analysis of the numerical results, the performances of the
subroutine for the evaluation of the heat flux emitted by the radiant tube (Sec-
tion 5.2.1.1) have been evaluated. Figure 5.21 shows the comparison between
predicted and measured temperatures along the radiant tube for two different
burner loads. The agreement is satisfactory, as the predicted temperature pro-
files fit measured data well. This confirms that the proposed model for radiative
heat losses is somewhat appropriate.
Effect of increasing air inlet velocity The flow pattern in the burner (Fig-
ure 5.22) confirms the mechanism that promotes the exhaust gas recirculation.
The burnt gases are entrained by the fresh air jet and deviated from their path,
so that they recirculate toward the combustion zone. The degree of recircula-
tion can be evaluated with the recirculation degree, kR , defined in Chapter 1,
as:
ṁE
kR =
ṁA + m˙F
where ṁE is the mass flow rate of the exhaust gases recirculating into the
reaction zone, whereas ṁF and ṁA represent the fuel and air mass flow rates
162
5.3. Results
Figure 5.22: Flow pattern inside the combustion chamber, determined by the
high-momentum air jet issuing into the flame tube. Run 1, Table 5.4.
fed to the burner, respectively. kR is found to increase when the air inlet cross-
sectional area is reduced, varying between kR = 25% with Aair,in = 339 mm2
and kR = 235% with Aair,in = 41 mm2 , for run 1 in Table 5.4. The real burner
operates with Aair,in = 88 mm2 , thus with a recirculation degree of kR = 136%.
Values of kR for each run are listed in the third column of Table 5.4.
The mixing of fresh air with the flue gases determines a reduction of the
oxygen mass fraction available for the oxidation process, YO2 , from the atmo-
spheric value of 0.232 to about 0.12 (column 4, Table 5.4). thus allowing the
achievement of the dilution conditions which characterize the flameless com-
bustion regime.
Figure 5.23 shows the temperature distributions in the burner for different
recirculation degrees. A more homogeneous temperature distribution in the
combustion chamber can be observed for large values of kR . The maximum
temperatures are reduced by more than 100 K when halving the actual air inlet
cross-sectional area, i.e. Aair,in = 88 mm2 , from Tmax = 2080 K for kR = 136%
down to Tmax = 1972 K for kR = 235%. Simulations performed with air inlet
cross-sectional areas larger than the real one exhibit high temperature peaks
of 2572 K and 2351 K, for kR = 23% and kR = 59%. In addition, these
predictions indicate the existence of a well defined flame front, characteristic of
a lifted non-premixed flame.
Figure 5.24 illustrates the axial temperature profiles at a radial distance
r = 7 mm and for different air inlet cross-sectional areas. The radial distance
r = 7 mm corresponds to the centerline of the inlet air jet. It can be noticed
that the temperature increase is by about 850 K for the real burner (Aair,in =
88 mm2 ), whereas it is reduced down to 850 K for the largest recirculation
degree (Aair,in = 41 mm2 ). The methane self-ignition temperature is 853 K;
therefore, according to the criterion by Cavaliere and de Joannon [3], the real
burner operation is in the limit of flameless operation mode3 . For smaller
3
According to Cavaliere and de Joannon [3], a combustion process is named MILD (or
163
flameless burner
(a)
(b)
(c)
(d)
(e)
(f)
Figure 5.23: Temperature distribution in the combustion chamber for Aair,in =

339 mm2 (a), Aair,in = 201 mm2 (b), Aair,in = 88 mm2 (c), Aair,in = 64 mm2
(d), Aair,in = 46 mm2 (e) and Aair,in = 41 mm2 (f). Run 1, Table 5.4.
164
5.3. Results
Figure 5.24: Axial temperature profiles at r = 7 mm for different air inlet

cross-sectional areas. Run 1, Table 5.4.
recirculation degree, the temperature increase due to combustion heat release

is dramatic, for instance being of 1300 K for Aair,in = 339 mm2 , thus indicating
that the burner would not operate in flameless conditions. In addition, the
shape of temperature profiles obtained for kR = 23% and kR = 59% differs
considerably from that obtained with larger recirculation degrees. For large kR ,
the high temperature of recirculating exhaust gases increases the mean reactant
temperature, leading to temperature profiles characterized by a double peak
(Figure 5.24). Indeed, the first temperature peak, located at x = 100−150 mm,
is due to the mixing of fresh reactants with exhaust gases. Conversely, for
smaller recirculation degrees (Aair,in = 339 mm2 and Aair,in = 201 mm2 ), the
contribution of exhaust gases to the reactants’ heat capacity is negligible, so
that the first temperature peak disappears.
Effect of the geometrical model Figure 5.25 shows the recirculation de-
gree (a) and the NO emissions (b) as a function of the air inlet cross-sectional
area, obtained using a 2D and 3D geometrical model, respectively. It can be
observed that the kR values predicted through the axisymmetric model are
larger than those predicted with the 3D model by at most 10%, for the small-
est air inlet cross-sectional areas. This is reasonable, as a 2D model does not
take into account the presence of recirculation windows, thus considering no
restriction for the entrainment of the flue gases. For the largest air inlet cross-
sectional areas, there are smaller differences between 2D and 3D models, being
flameless) when the inlet temperature of the reactant mixture is higher than mixture self-
ignition temperature whereas the maximum allowable temperature increase with respect to
inlet temperature during combustion is lower than the mixture self-ignition temperature (in
Kelvin)
165
flameless burner
Figure 5.25: Recirculation degree, kR , and NO emissions as a function of the

air inlet cross-sectional area. Run 1, Table 5.4.
the exhausts recirculation anyway limited in those cases. As a consequence of

the overpredicted recirculation degree, the 2D model approach underestimates
the NO emissions, by a factor which is roughly around 15% for all the simu-
lated cases. On the basis of these results, it can be argued that the choice of
a 2D model does not result in dramatic differences with respect to a detailed
geometrical model. Therefore, the error arising from the axysimmetric hy-
pothesis can be considered acceptable, to allow the implementation of detailed
kinetic mechanisms. Indeed, the coupling of detailed geometrical models and
detailed chemical mechanisms would result in unaffordable CPU times. An-
other possibility is to adopt a 2D model and consider the error deriving from
such approximation as an additional source of numerical uncertainty, beside
the discretization error (Section 5.2.1.2),.
166
5.3. Results
Turbulence-chemistry interactions Flameless combustion is character-

ized by a particular kind of turbulence–chemistry interaction. Specifically,
higher turbulence levels with respect to conventional flames occur in the re-
action region because of the strong recirculation, whereas slower chemical rates
are observed because of the large dilution of reacting species. Therefore, the re-
sulting flame structure is characterized by a stronger competition between mix-
ing and chemistry. In order to characterize the present system from the point of
view of turbulence-chemistry interactions, the analysis of the Damkohler num-
ber, Da, distribution in the combustion chamber has been carried out. Da is
defined as:
τt
Da = (5.7)
τc
where τt and τc are the turbulent and chemical time scales, respectively. The
estimation of τt and τc is straightforward: τt is determined from the turbu-
lence quantities, k and , whereas τc is calculated from kinetic constant of the
one-step methane oxidation reaction. Being the chemical timescale based on
one-step kinetics, results cannot provide a quantitative answer but give qualita-
tive evidence of the effect of exhaust gas recirculation on turbulence-chemistry
interactions.
Figure 5.26 shows the Damkohler number distribution in the burner through
contours, for different recirculation degrees. The contours have been overlapped
to the temperature distribution to facilitate the interpretation. Figure 5.26
(a) refers to the largest air inlet cross-sectional area, for which the burner
is operating in flame mode (kR = 23%). It can be noticed that the flame
front is characterized by large Da values. Conversely, in 5.26 (b) and (c), the
Damkohler number values are low and near unity even in the reaction region,
thus proving that both turbulence and chemistry control the reaction rates.
Such result is particularly meaningful and it confirms the need for finite-rate
chemistry models for the description of the flameless combustion regime, as
also indicated by the analysis of the JHC system in Chapter 4. The ED/FR
approach can provide reasonable results in this case, being the fuel methane
(Run 1, Table 5.4). However, when hydrogen is added to the fuel, detailed
kinetic mechanisms are needed to correctly capture the fuel behavior, especially
in a combustion regime where the chemical and mixing rates are comparable.
Therefore, on the basis of such observation, it has been chosen to adopt the EDC
turbulence-chemistry interaction model with detailed kinetics for modeling the
flameless oxidation of methane-hydrogen mixtures.
Burner efficiency As mentioned in Chapter 1, recuperative burners enclosed

by radiant tubes are widely used in continuous annealing furnaces for electric
steel strip. In such applications, it would be desirable to achieve smooth tem-
perature profiles along the radiant tube in order to ensure good steel quality
and safe conditions for the material constituting the radiant tube. Therefore,
167
flameless burner
(a)
(b)
(c)
Figure 5.26: Temperature (colored map) and Damkohler number (contours)

distributions for Aair,in = 339 mm2 (a), Aair,in = 88 mm2 (b) and Aair,in =
41 mm2 (c). Run 1 Table 5.4.
168
5.3. Results
Figure 5.27: Radiant tube temperature for different air inlet cross-sectional
area, Aair,in . Run 1 Table 5.4.
local temperature peaks and gradients should be minimized to avoid the col-
lapse of the radiant tube itself due to localized phenomena, such as deformation
and mechanical stresses arising from nonuniform thermal expansion [36].
Figure 5.27 shows temperature profiles along the radiant tube for run 1
(Table 5.4), obtained by varying the air inlet cross-sectional area and, then,
the recirculation degree. A large temperature interval, greater than 230 K,
is observed in flame combustion mode, i.e. Aair,in = 339 mm2 , whereas in
flameless regime, i.e. Aair,in = 88 − 41 mm2 , the temperature profiles appear
more uniform, covering temperature intervals of about 150 K. Therefore this
condition is more suited to ensure homogeneous conditions.
Another criterion to evaluate the burner performance is based on the ther-
mal efficiency. In particular two different efficiencies can be calculated: a ra-
diation efficiency based on heat transfer from the radiant tube and a recovery
efficiency characterizing the air preheater performance. The former efficiency
may be calculated as ηrad = Q̇/Q̇in , whereas the recovery efficiency is given by
ηrad = Q̇air/Q̇in . Figure 5.28 shows the two efficiencies as a function of the
recirculation degree. It is worth noting that the sum of the two efficiencies
approaches unity as in the type of burner works through radiation towards the
surroundings. It can be also observed that the radiation efficiency increases with
increasing recirculation degree; thus the burner ensures better performances in
the flameless combustion regime than in the flame mode.
5.3.1.2 Flameless oxidation of methane-hydrogen mixtures

This section describes the results of the CFD analysis of the flameless com-
bustion of methane-hydrogen mixtures. The main objectives of the discussion
are: i) to investigate the effect of turbulence-chemistry interaction models and
169
flameless burner
Figure 5.28: Preheater and radiant rube efficiencies as a function of the recir-
culation degree, kR . Run 1 Table 5.4.
kinetic mechanisms on the results, ii) to examine the effect of H2 addition to

the fuel on the temperature distributions and NO formation and iii) to assess
the influence of molecular diffusion on the numerical results.
Effect of turbulence-chemistry interaction models and kinetic mech-

anisms Figure 5.29 (a–d) shows the contour plots of temperature in the
burner fed with a mixture containing a H2 mass fraction equal to 5.5% (Run
10, Table 5.4). In such conditions, the use of the ED/FR model (Figure 5.29
(a)) shows a strong influence of H2 reactivity on the temperature distribution.
The reaction zone is attached to the burner and the flame shape resembles that
of a typical diffusion flame. If the EDC model is employed (Figure 5.29 (b)),
the reaction zone is no longer attached to the burner but shifted downstream
along the axis. When the detailed kinetic mechanisms are used in conjunction
with EDC (Figure 5.29 (c, d)), the resulting temperature distribution is more
homogeneous, as expected when operating in flameless combustion regime. As
for the comparison between DRM-19 and GRI-3.0, a deeper penetration of the
air jet is observed with GRI-3.0, indicating that a higher reactivity is predicted
by the DRM-19 mechanism.
Such considerations can be quantitatively confirmed by the analysis of the
radial profiles of temperature at different locations, x, along the burner axis
(Figure 5.30 (a–c)). It can be observed how the use of the ED/FR model with
global kinetic results in a significant temperature increase in the near burner
region, at an axial distance x equal to 0.075 m (Figure 5.30 (a)). This behavior
can be ascribed to the high reactivity of H2 resulting from the decoupled treat-
ment of CH4 and H2 oxidation processes, which leads to the early oxidation of
H2 . On the other hand, the temperature profile predicted by the EDC model
with global chemistry does not significantly deviate from the profiles obtained
170
5.3. Results
Figure 5.29: Temperature distribution in the burner fed with CH4 /H2 (Run
20, Table 5.4) predicted by ED/FR with global chemistry (a), EDC with global
chemistry (b) , EDC with DRM-19 (c) and EDC with GRI-3.0 (d).
171
flameless burner
with the detailed mechanisms. This result suggests that the near burner be-
havior is largely affected by the ability of the turbulence-chemistry interaction
model to capture the ignition process. The discrepancies between the results
provided by the global and detailed kinetic mechanisms are, again, due to the
effect of H2 reactivity, enhanced by the global chemistry approach. As the dis-
tance from the burner exit increases (Figure 5.30 (b-c)), the influence of the
kinetic mechanisms on the heat release and, hence, on the temperature distri-
bution becomes significant, leading to the overlap of the profiles provided by
the ED/FR and EDC models with global chemistry at x = 0.25 m (Figure 5.30
(c)). Regarding the detailed kinetic mechanisms, the DRM-19 and GRI-3.0
mechanisms are almost collapsed onto a single line and provide a homogeneous
temperature distribution in the burner, without local peaks. Small differences,
below 10% with respect to the GRI-3.0 case, are observed at the axial location
x = 0.15 m, indicating the earlier ignition predicted by the DRM-19 mechanism
(Figure 5.30 (b)).
Effect of H2 addition to the fuel Figure 5.31 (a-d) compares the temper-
ature distributions in the burner fed with mixtures containing mass fractions
of H2 up to 20% (Runs 5, 20, 1m and 2m, Table 5.4), obtained with the EDC
combustion model and the GRI-3.0 kinetic mechanism. The contour plots show
that the addition of H2 determines the increase of the temperature levels in the
burner as well as the reduction of the lift-off length and the shift of the reac-
tion zone towards the burner exit. The temperature increase is not only due
to the higher specific energy content of the H2 blended fuel with respect to
methane, but also to the reduced radiation losses from the flame, determined
by the decrease of CO2 formation [138, 139].
The influence of the kinetic mechanisms on the results can be assessed by
comparing the radial profiles of temperature at different heights along the axis,
as provided by the global and detailed chemistry approaches (Figure 5.32 (a-c)).
It can be observed how, in the near burner region (Figure 5.32 (a)), the use of
a global kinetic scheme determines the formation of temperature peaks which
become more and more pronounced as the H2 content in the fuel increases.
Instead, when the GRI-3.0 mechanism is associated to the EDC model (Figure
5.32 (a)), no influence of H2 addition on the temperature profiles is observed at
the axial location x = 0.075 m, the temperature simply increasing from the air
inlet temperature to the recirculating flue gas temperature. The effect of H2 be-
comes relevant for both the kinetic mechanisms at x = 0.15 m (Figure 5.32 (b)).
The radial temperature profiles show, for both cases, a generalized increase of
the temperature levels in the burner when increasing the mass fraction of H2
in the fuel stream. Moreover, H2 addition causes the maximum temperature to
shift towards the burner axis, denoting a minor penetration of the exhaust gases
in the fresh mixture, due to the higher reactivity of the fuel. Such behavior can
be explained by the direct effect of hydrogen on the chain carrying radicals,
i.e. H, O and OH, which globally results in the enhancement of the combustion
172
5.3. Results
Figure 5.30: Radial profiles of temperature at different axial locations along

the axis, i.e. x = 0.075 m (a), x = 0.15 m (b) and x = 0.25 m, predicted by
different combustion models and kinetic mechanisms. Run 20, Table 5.4.
173
flameless burner
Figure 5.31: Temperature distribution predicted by EDC with GRI-3.0 in the

burner fed with increasing H2 mass fraction (Runs 5, 20, 1m and 2m, Table
5.4): 0% (a), 5.5% (b), 10% (c) and 20% (d).
174
5.3. Results
rate of methane and in the shortening of the visible flame length, as observed
experimentally by Choudhuri and Gollahalli [140] for turbulent jet diffusion
flames. However, it should be underlined how the detailed mechanisms predict
temperature levels in the burner far below the (unrealistic) ones given by the
global approach, and provide a significantly more uniform temperature distri-
bution, as one would expect in a flameless combustion regime. With respect
to the CH4 case, the EDC with the GRI-3.0 mechanism predicts an increase of
the maximum temperature in the burner by 150 K with a mass fraction of H2
in the fuel equal to 20%, whereas the global chemistry approach would lead,
in the same conditions, to an increase higher than 400 K. Such temperature
increase is expected to affect dramatically the estimated NO emissions from
the burner, as described in the following text.
The effect of H2 on the burner operation is also visible from the contour
plots of OH radical in the burner with increasing H2 mass fraction (Runs 5,
20, 1m and 2m, Table 5.4), obtained with the EDC model and the GRI-3.0
mechanism (Figure 5.33 (a–d)). It is clear how the addition of H2 causes an
increase of the OH concentration in the burner and the formation of a high
reactivity core, as shown by the temperature distributions (Figure 5.31 (a-d)).
Moreover, the profiles of OH radical mass fraction (Figure 5.34 (a-c)) show a
radial spread of OH when increasing H2 in the fuel. Such behavior is expected,
since hydrogen increases the production of OH through the chain branching
reactions [140, 141]. Finally, regarding the comparison between DRM-19 and
GRI-3.0, the OH radial profiles of Figure 5.34 (a-c) show that the two kinetic
mechanisms are in good agreement, being the maximum observed differences
below to 20% (with respect to the GRI-3.0 predictions). This can be considered
acceptable, especially because such differences are reflected in discrepancies on
the predicted temperature fields safely below 10%.
Figure 5.35 shows the axial profiles of the OH radical and CH2 O interme-
diate species mass fractions, with a H2 mass fraction in the fuel ranging from
0 to 20%. The figure confirms the main effect of H2 on the flame structure
inside the burner. When the mass fraction of H2 is increased, the OH peak
rises and shifts backward along the axis. The existence of a diffuse and ex-
tended reaction zone is confirmed by the axial profiles of OH reaching almost a
plateau value. Moreover, Figure 5.35 clearly indicates the role of the interme-
diate species CH2 O as ignition marker, in contrast to the flame marker OH,
as experimentally observed by [49, 50]. The CH2 O peak anticipates the OH
maximum and indicates the attachment height of the reaction zone. Again, it
can be observed how ignition occurs faster when H2 is added to the fuel. This
is further confirmed by the contour plots of CH2 O species mass fraction in the
burner, when increasing the H2 mass fraction in the fuel (Figure 5.36).
Table 5.6 lists the computed maximum temperatures in the burner and
the NO emissions, predicted and measured, as a function of the H2 content in
the fuel stream. As it was pointed out in Section 5.1, there is no access to
the burner interior. Therefore, the numerical models adopted can be assessed
175
flameless burner
Figure 5.32: Radial profiles of temperature at different axial locations along

the axis, i.e. x = 0.075 m (a), x = 0.15 m (b) and x = 0.25 m, predicted by
different combustion models and kinetic mechanisms. Runs 5, 20, 1m and 2m,
Table 5.4.
176
5.3. Results
Figure 5.33: OH radical distribution predicted by EDC with GRI-3.0 in the

burner fed with increasing H2 mass fraction (Runs 5, 20, 1m and 2m, Table
5.4): 0% (a), 5.5% (b), 10% (c) and 20% (d).
177
flameless burner
Figure 5.34: Radial profiles of OH radical mass fraction at different axial lo-
cations along the axis, i.e. x = 0.075 m (a), x = 0.15 m (b) and x = 0.25 m,
predicted by different combustion models and kinetic mechanisms. Runs 5, 20,
1m and 2m, Table 5.4.
178
5.3. Results
Figure 5.35: Axial profile at r = 0 of OH radical and CH2 O species mass

fraction. Runs 5, 20, 1m and 2m, Table 5.4.
only through macro-indicators such as the NO emissions available in the ex-

perimental measurements. Moreover, a proper prediction of the NO trends is
mandatory, being flameless combustion an appealing technology for reducing
NOx pollutants. It can be observed (Table 5.6) how the maximum temperature
in the burner increases with the hydrogen fraction with all the turbulence-
chemistry interaction models and kinetic mechanisms used, with the exception
of the case corresponding to a H2 mass fraction equal to 2% (run 23, Table 5.4).
However, this experimental run was performed nearly at stoichiometric condi-
tions, thus at conditions not directly comparable to runs 5, 20, 1m and 2m.
Table 2 also indicates that, following the maximum temperatures, also the NO
emissions show a generalized increasing trend with the H2 fraction. Considering
only the experimental data, it can be observed how the measured NO emissions
are almost doubled, from about 50 to 100 ppmv when increasing the H2 mass
fraction from 0 to 5.5%. Such an increase could be regarded as unacceptable
with respect to existing environmental regulation. Then, a burner modification
could be proposed, by reducing the air inlet cross sectional area in order to
enhance the recirculation of exhaust gases in the reaction zone and the dilution
effects on the reacting mixture (Section 5.3.1.1). Finally, Table 5.6 underlines
that the EDC model with global chemistry results in higher temperatures and,
consequently NO levels with respect to ED/FR, when the H2 mass fraction in
the fuel is higher than 5.5%.
The comparison between the predicted and measured NO emissions (columns
6-10, Table 5.6) is shown graphically in Figure 5.37. It can be clearly observed
179
flameless burner
Figure 5.36: CH2 O species mass fraction distribution predicted by EDC with
GRI-3.0 in the burner fed with increasing H2 mass fraction (Runs 5, 20, 1m
and 2m, Table 5.4): 0% (a), 5.5% (b), 10% (c) and 20% (d).
180
5.3. Results
Table 5.6: Computed maximum temperatures and, predicted and measured, NO emissions with different combustion models
and kinetic mechanisms.
Tmax [K] NO
[ppmv ]
Run ED/FR EDC EDC EDC ED/FR EDC EDC EDC Exp
1,2-step DRM-19 GRI-3.0 1,2-step DRM-19 GRI-3.0
181
5 2510 2345 2082 2100 348 276 67 79 69
20 2488 2324 2092 2107 280 105 53 58 59
22 2558 2377 2115 2132 337 128 74 87 80
23 2635 2492 2114 2130 786 291 98 109 105
1m 2690 2710 2170 2160 1266 1244 136 158 -
2m 2685 2765 2243 2250 1655 1810 165 192 -
flameless burner
how it is possible to get close to the experimental measurements only by us-

ing an accurate description of turbulence-chemistry interactions and chemical
kinetics. In all cases, the ED/FR and EDC models with global chemistry lead
to a significant overprediction of the NO emissions. For instance, when us-
ing the EDC model, the NO emissions estimated for the CH4 case using the
temperature field given by the global kinetic mechanism are about 350 ppmv ,
thus much larger than those obtained from the temperature field provided by
the DRM-19 and GRI-3.0, i.e. 70 and 80 ppmv (the NO measured value is
79 ppmv ). This effect becomes more pronounced when H2 is added to the
fuel. For a H2 mass fraction of 20%, NO emissions of about 1810 ppmv are
obtained using global chemistry, whereas NO emissions of about 165 and 190
ppmv are calculated from the temperature distributions provided by the DRM-
19 and GRI- 3.0 mechanisms, respectively. Figure 5.37 seems to suggest that
better predictions can be achieved with the DRM-19 mechanism. However,
it is not possible to derive any conclusion from the observed behavior as this
could be imputed to the overestimation of the NO rates determined by the
simplified NO formation mechanism adopted. In conclusion, the analysis of the
computational results and their comparison to the experimental data suggest
that the conditions of uniform temperature distribution, characteristic of the
flameless combustion regime, allow capturing NO trends even with a simplified
NO formation model, as the one used in the present work, given an adequate
characterization of the temperature distribution in the burner. Obviously, the
prerequisite for the validity of such a result is that the NO formation is domi-
nated by the thermal mechanism and that the temperature is almost uniform
inside the burner. However, flameless combustion typically occurs at lower
temperatures so that other mechanisms, such as the N2 O route, may become
dominant [142]. Moreover, when hydrogen is added to the fuel, the NNH path
may play a significant role [17]. Such considerations have prompted our inter-
est towards the numerical modeling of the ultra-lean flameless combustion of
methane-hydrogen mixtures (Section 5.2.2), to investigate the potential of two
distinct approaches. The first is a natural extension of the one just proposed,
with the inclusion of simplified one-step rates for N2 O and NNH routes. The
second approach, known as kinetic post-processing [5, 6, 133, 134], has been
already applied to flameless combustion [143] and is based on the calculation
of NO with very detailed kinetic schemes, in a post-processing step of CFD
simulations, carried out with simplified mechanisms for gas-phase combustion.
Such activity is described in Section 5.3.2.
Effect of molecular diffusion Figure 5.38 shows the effect of molecular

diffusion on the H2 (a, b), CO2 (c, d) and temperature (e, f) radial profiles at
two different axial locations along the axis. It can be observed (Figure 5.38 (a,
b)) that the laminar diffusion has some influences on the H2 distribution in the
burner, when the H2 content in the fuel is higher than 10%. The maximum
difference between the radial profiles obtained with and without differential
182
5.3. Results
Figure 5.37: Comparison of predicted and measured NO emissions in the flue

gases. Runs 5, 20, 22 and 23 in Table 5.4):
diffusion effects are observed at the axial location x = 0.15 m and are of the
order of 15%, for a H2 mass fraction equal to 0.2. However, this affects very
slightly the distribution of the major species, such as CO2 , and temperature.
In fact, Figure 5.38 (c–f) points out how the variations in the radial profiles
of CO2 and temperature, when including the effects of molecular diffusivity,
are negligible, being in all cases lower than 5%. This is in contrast to what
observed by Christo and Dally [45], who found the differential diffusion to play
a major role in the numerical simulation of a jet issuing in a hot diluted co-flow.
However, the different findings of the present study could be ascribed to the
intense recirculation and to the enhanced turbulence levels in the combustion
chamber, which have a direct effect on the turbulent viscosity and, therefore,
on the turbulent contribution to the diffusive flux. Such consideration has been
quantitatively confirmed by the analysis of the distribution of the H2 molecular
to effective diffusion ratio inside the combustion chamber (Figure 5.39). The
turbulent contribution is dominant in the near burner region, where the ignition
process takes place, representing around 85% of the effective diffusion coefficient
for the 20% hydrogen case.

This Section described the results of the numerical modeling of the second
experimental campaign on the self-recuperative burner. The discussion will
be focused on the comparison between the different approaches adopted for the
prediction of NO emissions form the burner (Section 5.2.2). First, the results of
the direct coupling of simplified NO mechanisms with CFD simulations carried
out with detailed kinetic schemes for gas-phase combustion is presented. Then,
183
flameless burner
Figure 5.38: Effect of molecular diffusion on the radial profiles of H2 mass

fraction (a, b), CO2 mass fraction (c, d) and temperature (e, f) at different axial
locations along the burner axis. Combustion model and kinetic mechanism:
EDC with DRM- 19. Runs 5, 20, 1m and 2m, Table 5.4.
184
5.3. Results
Figure 5.39: Laminar to turbulent H2 diffusion coefficient with increasing H2

mass fraction in the fuel: 0% (a), 5.5% (b), 10% (c) and 20% (d). Runs 5, 20,
1m and 2m, Table 5.4.
185
flameless burner
Figure 5.40: Temperature field in the burner fed with increasing H2 in the fuel.
H2 content corresponding to 10% (a), 20% (b), 30% (c), 50% (d), 60% (e) and
70% (f) of the total burner thermal input, 10 kW. Kinetic mechanism: KEE-58.
Runs 4-9, Table 5.3.
the comparison with the kinetic post-processing approach is shown.
5.3.2.1 CFD analysis

This section describes the results of the CFD analysis of the self-recuperative
burner. The contour plots of temperature, OH and CH2 O radical mass fractions
in the burner fed with increasing H2 fractions (from 10 to 70% of the total
thermal input) are shown in Figure 5.40-5.42, respectively.
Similarly to the results illustrated in Section 5.3.1.2, it can be observed how
the addition of hydrogen results in a combustion region which is progressively
shifted toward the burner exit as confirmed, in particular, by the OH and CH2 O
radical mass fraction in the burner. The maximum temperature is not observed
for the case corresponding to the highest hydrogen content in the fuel; however,
this is due to the fact that the air-excess increases almost linearly (from 120
to 170 %) when increasing the hydrogen thermal input (from 10 to 70%), as
indicated in Table 5.3.
Differently from the experimental campaign n. 1, raw data were available
from the experimental campaign. Therefore, the statistics available from the
temporal measurements have been used to apply the validation methodology
introduced in Chapter 3, for the assessment of the level of agreement between
the experiments and numerical simulations. Figure 5.43 shows the average
186
5.3. Results
Figure 5.41: OH radical mass fraction distribution in the burner fed with in-
creasing H2 in the fuel. H2 content corresponding to 10% (a), 20% (b), 30%
(c), 50% (d), 60% (e) and 70% (f) of the total burner thermal input, 10 kW.
Kinetic mechanism: KEE-58. Runs 4-9, Table 5.3.
Figure 5.42: CH2 O radical mass fraction distribution in the burner fed with
increasing H2 in the fuel. H2 content corresponding to 10% (a), 20% (b), 30%
(c), 50% (d), 60% (e) and 70% (f) of the total burner thermal input, 10 kW.
Kinetic mechanism: KEE-58. Runs 4-9, Table 5.3.
187
flameless burner
(a) (b)
Figure 5.43: Sample mean and standard deviation for the average radiant tube
temperature (a) and NO concentration in the flue gases (b), when increasing
the hydrogen thermal input from 10% to 70%. Runs 4-9, Table 5.3. The sample
mean and the standard deviations are plotted on different scales.
radiant tube temperature and the NO emissions from the burner, together with
the associated standard deviation, when increasing the hydrogen thermal input
from 10% to 70%. It can be observed that the fluctuations associated to the
experimental measurements are quite limited, thus resulting in small values of
the standard deviations, especially for the NO measurements. This is somehow
expected, being the flue gases sampled outside the combustion chamber, when
the reaction process is already completed.
The statistical information illustrated in Figure 5.43 can be used to build
confidence intervals for the experimental data and to derive validation metrics
for the quantitative comparison of measurements and model predictions. In
particular, ten minutes of data acquisition were available for each experimental
run. Therefore, it was decided to build five replicates of each run, averaging
the data over a time period of two minutes, to evaluate the reproducibility
of the measurements, according to the approach described by Oberkampf and
Trucano [1].
Figure 5.44 (a) offers a comparison between the calculated and measured
average temperatures on the radiant tube, over the range of fuel composition fed
to the burner. The computed average temperatures have been obtained using
the model described in Section 5.2.1, for the estimation of the heat flux leaving
the radiant tube towards the Inconel® shield and the water heat exchanger.
The numerical results are compared to the experimental values with estimated
95% confidence intervals (Chapter 3), for the simulations carried out with the
KEE-58 mechanism. To help visualizing the entity of the error associated to
the numerical simulations, the estimated error is shown in Figure 5.44 (b), with
95% confidence intervals for the true error (Chapter 3). It can be observed how
the numerical results are very close to the experimental mean for the 10%
and 20% cases, while the proposed model underpredict the average radiant
tube temperature for all the other cases, although the calculated values are
188
5.3. Results
(a) (b)
Figure 5.44: 95% confidence intervals for the average temperature of the radiant
tube and comparison with the numerical results (a). 95% confidence interval for
the true error and estimated errors for the average temperature of the radiant
tube (b). Runs 4-9, Table 5.3.
Table 5.7: Average and maximum global validation metrics for the average
temperature of the radiant tube.

Model
Ee CI Ee CI
ȳe ȳe ȳe ȳe
avg avg max max
KEE-58 0.013 ±0.051 0.024 ±0.047

DRM 0.014 ±0.051 0.025 ±0.047
always within the experimental confidence interval. The maximum departure

is observed for the 70% H2 thermal input and is around 23 K. This slight under
prediction is actually expected, due to the geometric simplification and the
choice of a 2D model for the burner, instead of a 3D grid, which determines an
over prediction of the recirculation degree, thus leading to a reduction of the
temperature levels (Section 5.3.1.1).
Table 5.7 lists the global validation metrics (Chapter 3) for the prediction of
the average temperature of the radiant tube. As expected, the results obtained
using the KEE-58 and DRM-19 kinetic mechanisms are very similar: the aver-
age error metric is approximately 1.3% (±5.1%), whereas the maximum error
metric is around 2.5%(±4.7%). Such result can be considered acceptable, given
all the experimental and numerical uncertainties of the problem.
Figure 5.45 (a) shows the confidence intervals for the average NO emissions
from the burner, together with the comparison between experiments and model
predictions. Different variants of the model including NNH have been tested,
due to an uncertainty in the activation energy for the NO formation reaction
via NNH [11]. It can be observed that NO formation is very sensible to such
parameter, especially at high H2 concentrations. Figure 5.45 shows that an im-
189
flameless burner
(a) (b)
Figure 5.45: 95% confidence intervals for the average NO emission from the
burner and comparison with the numerical results (a). 95% confidence intervals
for the true error and estimated errors for the average NO emissions from the
burner (b). Runs 4-9, Table 5.3
pressive range of solutions are obtained by simply acting on the possible source
of NO emissions. This represents a strong input towards the development of
methodologies for the constructive comparison between numerical results and
experimental data, to effectively assess the reliability of the model predictions.
With regard to Figure 5.45, it is also noteworthy to remind that the confi-
dence intervals for NO are very narrow, limited to fractions of ppmv , being the
measurements carried out after the combustion process, thus, in a very stable
environment.
Figure 5.45 (b) reports the estimated error for the different NO models over
the range of hydrogen thermal inputs. It is clear how the role of the NNH route
is critical to explain the NO emissions. Without considering such route the re-
sulting error would be very large. A considerable improvement in the prediction
is also obtained by varying the activation energy for the NNH route within the
uncertainty range (30 ± 5 kJ/mol) indicated by Hayhurst and Hutchinson [11].
The value of 2.5 kJ/kmol seems the one providing, in all cases. better agree-
ment with the experimental data. Moreover, the KEE-58 mechanism appears
to provide always better results than the DRM-19 mechanism.
Such analysis can be quantitatively supported by the global validation met-
rics provided in Table 5.8. The values confirm that KEE-58 mechanism is able
to better capture the trend of NO emissions over the range of hydrogen thermal
inputs, when NNH is coupled to the NO model. In particular the combination
of KEE-58 and of a NO model with an activation energy for the NNH route
increased by 2.5 kJ/mol, results in an average error metric of 8.9%±2.1% (with
95% confidence). This is a very interesting result, considering the extremely
challenging conditions for the prediction, i.e. ultra-lean and hydrogen-rich.
The maximum relative error metric associated to the same model is 0.19±0.04,
190
5.3. Results
Table 5.8: Average and maximum global validation metrics for the average NO
emissions from the burner. All average metrics are ±2.1%, with 95% confidence.

Model
Ee Ee CI
ȳe ȳe ȳe
avg max max
KEE - Th+Pr 0.992 0.996 ±0.022

KEE - Th+Pr+N2 O 0.635 0.740 ±0.022
KEE - Th+Pr+N2 O+NNH 0.181 0.291 ±0.022
KEE - Th+Pr+N2 O+NNH
0.089 0.191 ±0.040
(Ea +2.5 kJ/mol)
KEE - Th+Pr+N2 O+NNH
0.138 0.276 ±0.040
(Ea +5.0 kJ/mol)
DRM - Th+Pr 0.992 0.995 ±0.022
DRM - Th+Pr+N2 O 0.648 0.726 ±0.022
DRM - Th+Pr+N2 O+NNH 0.244 0.349 ±0.022
DRM - Th+Pr+N2 O+NNH
0.159 0.309 ±0.040
(Ea +2.5 kJ/mol)
DRM - Th+Pr+N2 O+NNH
0.178 0.371 ±0.040
(Ea +5.0 kJ/mol)
kinPP - No T fluctuations 0.216 0.250 ±0.008
kinPP - T fluctuations 0.070 0.098 ±0.022
which means that the maximum error can range from ∼23% to ∼15%, with 95%
confidence. All the other values for Ee/ȳe avg and Ee/ȳe max are well above the

values indicated for the selected model, which will be referenced in the follow-
ing simply as complete NO mechanism. For example, a conventional approach
based on the thermal and prompt mechanisms would determine an error close
to 100%, which is clearly unacceptable.
The discrepancy in the performances of the KEE-58 and DRM-19 mecha-
nisms can be explained by analyzing the radial profiles of temperature, H, O
and OH radical mass fractions at different locations along the axis, for the cases
corresponding to 20% and 70% hydrogen thermal input. Beside temperature,
the formation of NO via NNH is affected directly by H and O radicals. Figure
5.46 makes clear the reasons for the higher NO levels predicted when using the
KEE-58 mechanisms, for the 20% hydrogen thermal input (Run 4, Table 5.3).
All the radical species are predicted in higher concentrations by KEE-58 with
respect to DRM-19 and the same happens for temperature, thus justifying the
higher NO production via NNH. This trend inverts above 50% hydrogen ther-
191
flameless burner
Figure 5.46: Radial profiles of temperature, H, O and OH radical mass fractions

at different axial locations along the burner, as predicted by the KEE-58 and
DRM-19 kinetic mechanisms, for a hydrogen thermal input of 20%. Run 4,
Table 5.3
mal input case, as shown in Figure 5.47 for the 70% hydrogen thermal input
case (Run 9, Table 5.3). Higher temperatures and radical concentrations are
predicted by the DRM-19 model, thus explaining the higher predicted NO val-
ues. It should be recalled [62], however, that the DRM-19 mechanism has been
developed for high temperatures applications. At lower temperatures, as the
ones investigated in the present study, the accuracy of the reduced mechanism
is expected to drop, thus providing a possible explanation for the differences
between the results provided by KEE-58 and DRM-19.
The relative importance of the different NO formation routes is shown in

Figure 5.48, for the KEE-58 model in combination with the complete NO mech-
anism. It can be observed that the thermal and prompt routes are almost neg-
ligible in all cases, due to the low temperatures and very lean conditions. N2 O
and NNH routes play a major role in the overall NO formation. In particular,
the N2 O contribution appears to be stable and around the tens of ppmv , almost
independently on the hydrogen fraction in the fuel. On the contrary, the NNH
route is highly sensible to H2 , leading to contribution to NO formation higher
as the hydrogen thermal input increases. This is expected since the availability
of H radicals has a direct impact on the formation of the intermediate species
NNH, which is ultimately oxidized to NO.
192
5.3. Results
Figure 5.47: Radial profiles of temperature, H, O and OH radical mass fractions

at different axial locations along the burner, as predicted by the KEE-58 and
DRM-19 kinetic mechanisms, for a hydrogen thermal input of 70%. Run 9,
Table 5.3
Figure 5.48: Relative importance of NO formation routes as a function of hy-

drogen thermal input for the complete NO mechanism.
193
flameless burner
5.3.2.2 Kinetic post-processing

The results shown in the present Section have been kindly provided by the
research group headed by Prof. Ranzi at the Politecnico di Milano, to allow
a comparison between the NO predictions obtained using a fully coupled CFD
approach and a CFD+kinetic post-processing methodology (Section 5.2.2).
The results of the kinetic post-processing of the CFD calculation, carried
out with a simplified kinetic mechanism (Section 5.2.2), are also summarized
in Figure 5.45 and Table 5.8. The comparison with the experimental data can
be considered satisfactory: the effect of H2 addition is well represented by the
model and the quantitative agreement is reasonably good. In particular, the
average validation metrics in Table 5.8 show an improvement in the predictions
with respect to the approach based on the direct coupling of simplified NO
mechanisms in the the CFD simulation, decreasing from the value of 15.5%,
given by the complete NO mechanism, to 7.7%.
Figure 5.45 also shows that the predicted NOx emissions are always larger
than the experimental values and this seems to indicate that the CFD simula-
tion tends to over-predict the temperature. This result is somehow expected,
being the numerical simulation carried out with a global kinetic scheme. Of
course, the more accurate is the prediction of the thermal field by the CFD
calculation, the more reliable the KinPP predictions are expected to be. The
effects of temperature fluctuations on the formation of NO have been also in-
vestigated. Even if in ultra-lean conditions the temperature fluctuations are
relatively small, their effect on the NO emissions is not negligible. Figure 5.45
shows a decrease of 25-30% of the NO emissions, when temperature fluctuations
are not taken into account. This leads to an increase in the average validation
metrics, as pointed out in Table 5.8, thus confirming the strong influence of
temperature on NO formation.
From the comparison between the two modeling approaches, it is possible to
conclude that a simplified treatment of NO formation can provide a reasonable
agreement between simulations and experiments, only by taking into account
detailed kinetic schemes for gas-phase combustion and including all the relevant
sources of NO at the investigated operating conditions (e.g. NNH route). On
the other hand, the kinetic post-processing allows to adopt simpler kinetic
mechanisms in the CFD simulations, as the pollutant formation is computed
using very detailed schemes.
5.4 Summary
In the present Chapter, the experimental and modeling activity carried out on
a self-recuperative burner operating in flameless combustion regime has been
presented.
With regard to the experimental activity, different operating conditions
have been studied. In particular, the effect of hydrogen addition to the fuel
194
5.4. Summary
(natural gas) on the combustion regime has been investigated, to assess the
feasibility of mild combustion with hydrogen enriched fuels. Moreover, ultra-
lean combustion conditions have also been explored, to evaluate the opportunity
of exploiting flameless combustion for the oxidation of low-calorific value fuels,
enriched with hydrogen.
As far as the mathematical modeling is concerned, different modeling ap-
proaches have been tested to develop the most realistic representation of the
burner. Model validation has been carried out using the data available from
the experimental campaigns, i.e. furnace wall temperatures and NO emissions
in the flue gases. The comparison between the experimental measurements has
demonstrated that the developed model is able to provide a satisfactory de-
scription of the system over the wide range of operating conditions analyzed.
In particular, regarding NO emissions, reasonable predictions of NO forma-
tion is achieved only with an accurate characterization of the temperature field
and by taking into account all the relevant formation routes at the investi-
gated operating conditions. For instance, the inclusion of non-conventional NO
formation routes, N2 O and NNH, is crucial for characterizing the pollutant
emissions from the burner at low temperatures and when hydrogen is added to
the fuel.
195
Chapter 6

investigation of the flameless
combustion of
hydrogen-enriched fuels in a
lab-scale burner
In the present Chapter, the experimental and numerical investigation of a lab-

scale flameless burner fed with hydrogen-enriched mixtures is presented. This
activity has been carried out in collaboration with the Politecnico di Milano,
where the research group headed by Prof. Rota performed the experimental
campaigns. Then, a comprehensive numerical approach has been developed
for the characterization of the system. In particular, the availability of exper-
imental data regarding the transition between the flame and flameless com-
bustion regimes allowed investigating the modeling requirements, in terms of
turbulence/chemistry interaction models, to reproduce the observed modes of
operation of the burner. Specifically, a simplified NO formation approach for
capturing NO emissions in both flame and flameless regimes have been pro-
posed.
The results shown in the present Section have been submitted for publica-
tion.
6.1 Description of the burner and experimental cam-

paigns
Figure 6.1 shows a sketch of the lab-scale experimental apparatus investigated
in the present study. The system consists of a vertical quartz reactor divided
in two sections, the combustion chamber and the air preheater. Both sections
197
Chapter 6. Experimental and numerical investigation of the flameless
combustion of hydrogen-enriched fuels in a lab-scale burner
Figure 6.1: Laboratory scale experimental equipment. Sketch and main dimen-
sions [31].
of the burner are placed in refractory insulated electrical ovens. The lower
one provides air preheating up to 1300 °C, simulating a regenerative heat ex-
changer, while the upper part is used to reduce heat losses from the combustion
chamber towards the surrounding, emulating adiabatic conditions. The vertical
combustion chamber is cylindrical with a radius of 0.025 m and a length of 0.34
m. The burner core is constituted by a single, high velocity, nozzle located at
the bottom of the combustion chamber, feeding a mixture of air and fuel. The
fuel is introduced perpendicularly to the injection nozzle into the air stream,
by means of a t-shaped pipe (Figure 6.1).
Fuel and combustion air are partially premixed into the nozzle body; how-
ever, no combustion takes place before the mixture enters the combustion cham-
ber, as indicated by the experimental evidence, due to the small pipe diameter
and the short residence time. This allows to replicate the operation of a high
velocity burner (Chapter 1), leading to the entrainment of a large amount of
exhaust gases and, then, to the dilution of the feeding stream and the preheat-
ing of the fuel mixture above its self-ignition temperature. Moreover, beside
the internal, aerodynamic driven, recirculation, an external recirculation of flue
gases can be simulated by vitiating the combustion air with an inert gas, such
as nitrogen. The burner is also provided with a secondary inlet for the com-
bustion air, located on the bottom of the preheating section, used to start-up
the system in conventional stabilized flame conditions. When the temperature
levels are high enough to sustain the combustion process, the secondary air
stream is interrupted and the system is operated in flameless mode. The ex-
198
6.1. Description of the burner and experimental campaigns
Figure 6.2: Sketch of the exhaust gas outlet sections.
haust gases leave the combustion chamber from one central and three eccentric
holes placed at the top of the combustion chamber (Figure 6.2).
Regarding the experimental data, the composition of the flue gases is mea-
sured, as well as the temperatures at three positions inside the combustion
chamber. The experimental apparatus is designed to allow changing indepen-
dently the values of those parameters which directly affect the achievement of
the flameless combustion regime, namely the jet velocity, process temperature
and dilution ratio.
The dilution ratio, kv , is nothing but the recirculation degree, kR , intro-
duced in Chapter 1:
ṁE − ṁAII
kv = kR = (6.1)
ṁA + m˙F
where ṁA is the sum of the primary air mass flow rate, ṁAI , fed with the
fuel through the capillary pipe, and the secondary air mass flow rate, ṁAII .
In the present context, the notation kv is preferred, due to the mechanism
which drives the burner to operate in flameless combustion regime, i.e. dilution
with nitrogen. An aerodynamic recirculation ratio, RA , can be introduced to
compare the recirculated and inlet (air and fuel) mass flow rates:
ṁE
RA = . (6.2)
ṁAI + ṁF + ṁI
Eq. (6.2) has been derived assuming that the secondary air is completely en-
trained by the jet issuing into the combustion chamber. Substituting Eq. (6.2)
in Eq. (6.1), the following expression for kv is recovered:
RA (ṁAI + ṁF + ṁI ) − ṁAII

kv = . (6.3)
ṁA + m˙F
199
where ṁI represents the mass flow rate of the dilution inert, i.e. N2 , introduced
to simulate external flue gas recirculation.
A large experimental activity has been carried out at the Politecnico di Mi-
lano. Some experimental runs have been selected to carry out the numerical
simulations described in the present Thesis, as shown in Table 6.1. Two cases,
corresponding to a hydrogen volume fraction in the fuel of 0 and 60 % respec-
tively, have been considered. For the 0% case, only the flameless combustion
mode has been modeled, whereas the transition between flame and flameless
modes has been simulated for the 60% case (Table 6.1).
6.2 Numerical modeling of the lab-scale burner

The numerical simulation of the lab-scale burner has been carried out with the
commercial code FLUENT 6.3 by Ansys Inc, although some simulations were
also performed with CFX 5.7.1 by Ansys Inc, to investigate the feasibility of a
2D simplified geometrical approach with respect to the full 3D simulation. The
simulation activity was particularly focused on the development of a computa-
tional approach suitable to capture the transition of the system from flame to
flameless (runs 2-6, Table 6.1), from the point of view of burner operation and
pollutant emissions.
6.2.1 Computational domain and grid

A preliminary study has been carried out to investigate the suitability of a 2D
model for representing the system. Actually, the presence of three eccentric
outlets (Figure 6.2) indicates that the computational domain should consist of
at least a 120° angular sector of the combustion chamber. The sketches of the
3D and 2D grids are shown in Figure 6.3 and Figure 6.4, respectively.
A 2D geometrical model was built with an anular outlet section equivalent to
the sum of the 3 eccentric holes. The comparison between the results provided
by the 2D and 3D geometrical models shows very small discrepancies, maximum
differences of predicted temperature across the burner being smaller than 10
K.
Based on such results, it has been chosen to adopt a 2D model, to al-
low concentrating the computational resources on the accurate description of
turbulence/chemistry interactions and gas-phase combustion. Similarly to the
self-recuperative burner (Chapter 5), a grid independence analysis has been
carried out on the 2D domain, by generating grids of different resolution, with
a number of elements ranging from 6k to 26k. Based on the results of the sen-
sitivity study, a grid consisting of around 11k elements has been selected. The
mesh (Figure 6.4) is structured and non uniform; a special grid refinement is
adopted in the near burner region, to allow a more accurate resolution of the
reaction zone structure.
200
Table 6.1: Summary of the experimental campaign on the lab-scale burner.
Run Mode ṁCH4 [kg/s] ṁH2 [kg/s] ṁAI [kg/s] ṁN2 [kg/s] ṁAII [kg/s] XH2 [% vol]
1 flameless 5.36E-06 0 9.85E-05 0 0 0
2 flame 4.40E-06 8.26E-07 9.50E-05 0 1.25E-05 60
201
3 transition 2.63E-06 4.93E-07 6.42E-05 4.98E-05 0 60
4 transition 2.63E-06 4.93E-07 6.42E-05 6.81E-05 0 60
6.2. Numerical modeling of the lab-scale burner
5 transition 2.63E-06 4.93E-07 6.42E-05 9.05E-05 0 60

6 flameless 2.63E-06 4.93E-07 6.42E-05 1.58E-04 0 60
Figure 6.3: 3D computational domain and grid.
Figure 6.4: 2D computational domain and grid.
202
6.2. Numerical modeling of the lab-scale burner
6.2.2 Boundary conditions

As far as the boundary conditions are concerned, the inlet gas temperature
is assumed to be linearly dependent on the preheater nominal temperature.
Special attention has been devoted to the boundary condition at the combustion
chamber walls. The real wall temperature is not known: the electric oven
nominal temperature is fixed; however, quartz wool is inserted between the
oven and the walls, but uncertainties exist regarding its thickness. Therefore,
the wall temperature is determined by identifying the temperature value which
provides a flue gas temperature at the outlet equal to the measured one. In
other words, the wall temperature is set to satisfy the energy balance in the
burner. As a result of such procedure, the wall temperature is found to be 100-
150 K higher than the electric oven nominal temperature, thus, in agreement
with a conduction model based on a presumed thickness of the quartz wool.
6.2.3 Physical models

Favre-averaged Navier-Stokes equations are solved using the standard k- tur-
bulence model. A modified k- model [144] for round jets has also been tested.
However, being such model developed for non reacting jets, it is still not fully
clarified whether its performance for reacting round jets is superior with re-
spect to standard k-; therefore, the standard k- turbulence model has been
considered a reasonable choice.
The radiative transfer equation is solved with the Discrete Ordinate tech-
nique. The weighted sum of gray gases model (WSGG) with coefficients from
Smith et al. [145] is used to evaluate the spectral properties of the participating
medium.
6.2.3.1 Turbulence/chemistry interactions and kinetic mechanisms
Special attention has been paid to turbulence/chemistry interactions through

the choice of the combustion model. In particular, the Eddy Dissipation/Finite
Rate (ED/FR) model and the Eddy Dissipation Concept (EDC) have been
considered, as they account for finite-rate chemistry effects, although only
EDC allows to implement detailed kinetic mechanisms for gas-phase combus-
tion (Chapters 1, 5). More details regarding the aforementioned combustion
models and their applicability in the context of flameless combustion are given
in Chapter 2, 5.
Four different kinetic mechanisms have been used to describe the oxidation
of methane-hydrogen mixtures. Global kinetic rates are used in conjunction
with ED/FR and EDC [23, 59], whereas detailed reaction mechanisms are cou-
pled to EDC only. Two reduced kinetic schemes, the KEE-58 [61] and DRM-19
[62], are compared to a reference detailed mechanism for methane oxidation,
the GRI-3.0 [63]. The latter has been implemented into FLUENT without the
203
NOx subset, not to exceed the maximum number of species transport equations
which can be solved by the code.
Regarding NOx formation, the same approach adopted for modeling the
second experimental campaign on the self-recuperative burner (Chapter 5) has
been used. In particular, simplified one-step rates are taken into account for
thermal and prompt mechanisms. With regard to the thermal route, equilib-
rium and partial equilibrium hypothesis have been adopted to obtain O and
OH radical concentrations, when global kinetic rates are applied; otherwise, the
radical distributions are known from the detailed mechanisms.
Moreover, the low temperature levels in the combustion chamber and the
significant hydrogen content in the fuel stream suggest that NO formation is
dominated by other mechanisms, such as the N2 O [24] and NNH [10] interme-
diates routes. A simplified treatment of NO formation via N2 O is available in
FLUENT; for the NNH route, a bespoke subroutine has been coupled to the
CFD solver, following the one-step model by Konnov et al. [25]. It is worth
noting that the NNH route requires the knowledge of O and H radical distri-
butions; therefore it can be applied only with one of the three detailed kinetic
mechanisms (KEE-58, DRM-19 and GRI-3.0).
6.3 Results
The present Section shows the main results provided by the CFD analysis
of the lab-scale burner. First, the main operating features of the system are
discussed; then, a sensitivity analysis regarding the effect of the physical models
on the results is carried out. The influence of the reactant dilution is also taken
into account, investigating the transition of the system from flame to flameless
regimes. Finally, the validation of the computational approach is proposed,
comparing experiments and numerical simulations.
6.3.1 Flow-field characterization

Figure 6.5 shows the flow field streamlines in the combustion chamber obtained
for Run 6 in Table 6.1. A well recognizable recirculation loop can be observed
in Figure 6.5, thus confirming that the flameless combustion regime is ensured
by the internal aerodynamics which ensures both the preheating and dilution
of the fresh stream through mixing with the exhaust gases. Moreover, a low
velocity loop is observed in the lower part of the chamber, thus indicating the
existence of a dead zone when the secondary air supply is interrupted.
The numerical simulations can be exploited to characterize the different
runs investigated in terms of aerodynamic recirculation ratio, RA , and dilution
factor, kv . The determination of RA is not straightforward as for the self-
recuperative burner (Chapter 5) because the flue gases are not conveyed in
a well defined region such as a flame tube. Therefore, the calculation of the
recirculated mass flow rate, ṁE , must be performed through the identification
204
6.3. Results
Figure 6.5: Velocity streamlines. Run 6, Table 6.1.
of the recirculation loop center, r = R1 , and the integration of the radial

velocity profile from R1 to the walls, r = R. Thus:
RR
2π R1 ρE vE (r) rdr
RA = . (6.4)
ṁAI + ṁF + ṁI
The values of RA and kv for the runs in Table 6.1 are listed in Table 6.2,
together with the location of the recirculation loop. It can be observed that
the values of RA are comprised between 5.3 and 6.1 for all the investigated
runs, except for case 2, which has a larger RA , i.e. 8.3. However, the dilution
factor, kv , progressively increases when going from flame (Run 2, Table 6.1) to
flameless (Run 6, Table 6.1), due to the effect of N2 addition to the fuel stream
(Eq. (6.3)).
6.3.2 Effect of turbulence/chemistry interaction models

The influence of turbulence/chemistry interaction models on the temperature
distribution in the combustion chamber is first investigated. Figure 6.6 shows
the temperature distribution inside the combustion chamber obtained with dif-
ferent combustion models and kinetic mechanisms, with the burner operating
in flameless mode and fed with pure CH4 (Run 1, Table 6.1). It can be noticed
that the ED/FR model with one-step chemistry (Figure 6.6 (a)) predicts a high
temperature reaction zone near the burner nozzle, with a peak value of 1720
K.
205
Table 6.2: Location of the recirculation loop, aerodynamic recirculation ratio,

RA , and dilution factor, kv .
Run Mode x[m] r [m] RA [-] kv

1 flameless 0.087 0.014 5.3 5.3
2 flame 0.100 0.014 8.3 7.3
3 transition 0.090 0.014 6.1 10.6
4 transition 0.089 0.014 5.6 11.2
5 transition 0.090 0.014 5.6 13.2
6 flameless 0.091 0.014 5.7 19.1
Figure 6.6: Temperature distribution in the burner fed with CH4 (Run 1, Table
6.1), predicted by ED/FR with global chemistry (a), EDC with global chemistry
(b), EDC with KEE-58 (c), DRM-19 (d) and GRI-3.0 (e).
206
6.3. Results
Figure 6.7: Radial profiles of temperature at different axial locations along the
axis, i.e. x = 0.06 m (a) and x = 0.10 m (b), predicted by different combustion
models and kinetic mechanisms. Run 1, Table 6.1.
If the EDC model is employed with a single step chemistry (Figure 6.6
(b)), the high temperature region disappears and a more uniform distribu-
tion is observed, with maximum temperatures of 1500 K. Therefore turbu-
lence/chemistry interactions play a major role in the determination of the com-
bustion features. As far as the detailed mechanisms are concerned, the KEE-58
(Figure 6.6 (c)), DRM-19 (Figure 6.6 (d)) and GRI-3.0 (Figure 6.6 (e)) behave
in a similar manner, maximum temperature being 1505 K, 1530 K and 1520 K,
respectively.
Figure 6.7 shows the radial profiles of temperature at two axial locations
along the axis, i.e. x = 0.06 m (a) and x = 0.10 m (b). Figure 6.7 confirms
that the ED/FR model and the simplified one-step chemistry approach result in
higher temperatures in the region close to the axis, i.e. r < 0.1 m, whereas all
the profiles collapse onto a single line moving towards the walls, i.e. r > 0.14 m.
This is expected since the recirculation loop generated by the incoming jet
is located at a radius of approximately 0.014 m (Table 6.2) and, then, the
temperature of the gases above r = 0.14 m is that of the flue gases. Regarding
the effect of the kinetic mechanism, Figure 6.7 (b) shows that the temperature
profile provided by EDC with a one-step approach is flatter than those obtained
with the detailed mechanisms, thus indicating that the oxidation process is
already completed and the temperature distribution is uniform.
Figure 6.8 shows the temperature distributions in flameless regime for the
60% H2 (by vol.) case (Run 6, Table 6.1), obtained with different combustion
models and kinetic mechanisms. It can be noticed that simple (2-step) kinetic
mechanisms are strongly affected by hydrogen reactivity, resulting in an ignition
region which is attached to the nozzle for the ED/FR case (Figure 6.8 (a)), and
inside the injection pipe for the EDC case1 (Figure 6.8 (b)). Conversely, the
1
As a result of the ignition in the feeding pipe, the simulation carried out with EDC/2-step
207
Figure 6.8: Temperature distribution in the burner fed with a mixture contain-
ing 60% by vol. of H2 (Run 6, Table 6.1), predicted by ED/FR with global
chemistry (a), EDC with global chemistry (b), EDC with KEE-58 (c), DRM-19
(d) and GRI-3.0 (e).
temperature distributions obtained with the reduced (Figures 6.8 (c, d)) and
detailed mechanisms (Figure 6.8 (e)) denote the existence of a lifted and uniform
reaction region, as one would expect when operating in flameless combustion
regime.
To better investigate the sensitivity to the turbulence/chemistry interaction
approach, radial profiles of temperature taken at different axial distances, i.e.
x = 0.06 m and x = 0.10 m, are reported and compared in Figure 6.9. The
temperature values near the nozzle, x = 0.06 m (Figure 6.9 (a)) confirm that
the global kinetic approach results in an early ignition of the fuel, for both
ED/FR and EDC combustion models, temperatures along the axis are approx-
imately 100 K higher than those predicted with the detailed kinetic schemes.
chemistry is extremely unstable. A fully converged solution was not obtained.
208
6.3. Results
Figure 6.9: Radial profiles of temperature at different axial locations along the
axis, i.e. x = 0.06 m (a) and x = 0.10 m (b), predicted by different combustion
models and kinetic mechanisms. Run 6, Table 6.1.
As far as the detailed kinetic schemes are concerned, the temperature profiles
obtained with KEE-58, DRM-19 and GRI-3.0 are in very good agreement at
x = 0.06 m; small differences between DRM-19 and the other mechanisms are
observed at x = 0.10 m. It should be recalled [62], however, that the DRM-
19 mechanism has been developed for high temperature applications; at lower
temperatures, as the ones investigated in the present study, the accuracy of
the reduced mechanism is expected to drop. Similarly to the methane case
(Figure 6.7), all profiles tend to collapse onto a single line when increasing the
radial distance from the axis, with the exception of the EDC/2-step case, which
provides lower temperature levels. Such behavior is confirmed at higher axial
distances, x = 0.10 m (Figure 6.9 (b)), and can be ascribed to the partial ig-
nition which takes place in the injection pipe and results in lower temperature
levels in the combustion chamber.
6.3.3 Effect of dilution on the combustion regime

Figure 6.10 shows the temperature distribution in the combustion chamber
fed with CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for different
dilution ratios, kv . The increase of kv from 7.1 to 19.1 has been obtained, both
experimentally and numerically, by increasing the amount of nitrogen in the
primary air.
In flame conditions (Figure 6.10 (a)), secondary air at ambient temperature
is also fed coaxially to the fuel jet, to stabilize the flame. It can be observed
that the CFD simulations can qualitatively capture the transition from flame
to flameless regimes. The maximum temperatures in the combustion chamber
decrease dramatically when increasing kv , from 2500 K, for kv = 7.3 (Figure
6.10 (a)), to approximately 1400 K, for kv = 19.1 (Figure 6.10 (e)). Moreover,
the structure of the reaction zone is strongly affected by dilution: an attached
209
Figure 6.10: Temperature distribution in the combustion chamber fed with

CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for different dilution
ratios, kv : 7.3 (a), 10.6 (b), 11.2 (c), 13.2 (d) and 19.1 (e). Runs 2-6, Table 6.1.
flame is no longer observed above kv = 11.2 (Figure 6.10 (c)) and the high
temperature region is spread over a large portion of the available volume in the
combustion chamber.
Such considerations are also supported by the analysis of the temperature

and OH mass fraction radial profiles at x = 0.06 m and x = 0.10 m, shown in
Figure 6.11. In particular, with regard to the OH radical mass fraction profiles
(Figure 6.11 (a’, b’)), it can be observed that OH levels are much higher in
flame regime, thus explaining the earlier ignition.
Conversely, OH distribution is much more uniformly distributed in flameless

regime, over a region whose extension is significantly larger than a tradition
diffusive flame front, as pointed out by Figure 6.12.
The CH2 O distribution in the combustion chamber can be effectively used

to describe the ignition process, as indicated by Figure 6.13. This supports
the experimental findings of Medwell et al. [49], who stressed the role of the
intermediate species CH2 O at low temperatures, as the ones found in flameless
combustion regime. It would be desirable, then, to include such species in the
experimental measurements usually carried out on laboratory apparatus [80],
especially in the perspective of a modeling approach based on the determina-
tion of optimal progress variables for the low-dimensional representation of the
thermo-chemical state of a reacting system (Chapter 4).
210
6.3. Results
Figure 6.11: Radial profiles of temperature (a, b) and OH radical mass fraction
(a’, b’), at different axial locations along the axis, i.e. x = 0.06 m (a, a’) and
x = 0.10 m (b, b’), predicted with EDC/GRI-3.0 for different dilution ratios,
kv : 7.3 (FLAME), 10.6 (TR1), 11.2 (TR2), 13.2 (TR3) and 19.1 (MILD). Runs
2-6, Table 6.1.
211
Figure 6.12: OH radical distribution in the combustion chamber fed with

CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for different dilution
ratios, kv : 10.6 (a), 11.2 (b), 13.2 (c) and 19.1 (d). Runs 3-6, Table 6.1.
6.3.4 Model validation: comparison of simulations and exper-

iments
The validation of the numerical simulations is carried out by comparing the

computational results with the available experimental data. In particular, the
comparison between the predicted and measured temperatures inside the com-
bustion chamber is presented. Moreover, the reliability of the simplified NO
mechanisms adopted is assessed, using the measured NO values at burner exit.
No comparison is provided for the flue gas outlet temperature, as this value
was used to tune the boundary condition at the walls.
Figure 6.14 shows the comparison between the measured and calculated
temperatures inside the quartz tube at x = 0.18 m and r = 0.014 m, for runs
2-6 in Table 6.1. The experimental data are provided with an uncertainty of
±50 K, due to the uncertainty in the actual position of the thermocouples.
This can have an impact on the measurements, especially in the regions of
larger gradients. The laboratory at the Politecnico di Milano has recently
implemented a new system for the automatic positioning of the thermocouples,
to reduce the uncertainty in the measurements. It can be observed that, with
the exception of the ED/FR model, the calculated temperatures well represent
the measured values, although they show an overprediction of the temperature
levels.
212
6.3. Results
Figure 6.13: CH2 O mass fraction distribution in the combustion chamber fed
with CH4 /H2 (60% H2 by vol.), predicted with EDC/GRI-3.0 for different
dilution ratios, kv : 10.6 (a), 11.2 (b), 13.2 (c) and 19.1 (d). Runs 3-6, Table
6.1.
Figure 6.14: Comparison of measured and predicted temperatures inside the

combustion chamber, Tmiddle , x = 0.18 m and r = 0.014 m, Runs 2-6, Table
6.1.
213
Table 6.3: Average global validation metrics for temperature measurements at

two locations (x = 0.10 m r = 0.014 m and x = 0.18 m r = 0.014 m) and NO
emissions from the burner.

Ee
ȳe
avg
Model Tbottom (±4%) Tmiddle (±4%) NO

ED/FR - 2-step 0.09 0.04 2.65
EDC - 2-step 0.05 0.03 4.26
EDC - KEE-58 0.06 0.03 0.41
EDC - DRM-19 0.06 0.03 0.41
EDC - GRI-3.0 0.07 0.03 0.28
The second columns in Table 6.3 lists the average error metrics (Chapter
3) for the predicted temperatures at the same location of Figure 6.14. With
the exception of the ED/FR case, all the other models are characterized by
an average error around 3%. It should be observed, however, that the mea-
surements are taken at a radial position of 0.014 m, which corresponds to the
radial coordinate of the center of the recirculation loop for all the simulated
runs (Table 6.2). Therefore, the monitored temperature is that of the flue
gases, thus explaining the observed level of agreement. Measurements taken
inside the reacting fuel jet would certainly provide more useful information
from the point of view of model validation, as they could capture the tendency
of the global chemistry approaches to overpredict the temperature levels in the
axis region. To this purpose, CFD simulations could provide an effective tool
to optimize the experimental campaign, based on the knowledge of the flow
and temperature field in the combustion chamber. The comparison between
computed and measured temperatures is carried out for an additional position,
x = 0.10 m r = 0.014 m, as shown in Table 6.3. Similar results to those
obtained for position x = 0.18 m r = 0.014 m are observed; however, larger
discrepancies are found, due to the strong gradients which characterize the
investigated zone.
From the analysis of Figure 6.14 and the first two columns of Table 6.3 it
could be concluded that the approach EDC/2-step is the best for the description
of the system. However, Figure 6.15 and the third column of Table 6.3 show
that this would determine unacceptable errors from the point of view of NO
emissions.
Figure 6.15 compares on a log-scale measured (black solid squares) and
predicted NO emissions at different values of the dilution ratio, kv , for runs
2-6 in Table 6.1. The NO measurements are provided without an estimate
of the experimental associated error. Predictions have been obtained from
214
6.3. Results
the temperature fields provided by different combustion models and kinetic

mechanisms, by considering simple one-step equation for thermal, prompt, N2 O
and NNH routes. The latter is only used in conjunction with the detailed gas-
phase mechanisms, as it needs the knowledge of H and O radical distributions.
In flame conditions (kv = 7.3) a global kinetic approach (with both EDC
and ED/FR) provides a temperature distribution which leads to a significant
overprediction of the NO emissions, by more than one order of magnitude.
This is due to the large temperature peaks predicted by EDFR/2-step and
EDC/2-step, which strongly affect the thermal NO formation2 . Results ob-
tained with reduced and detailed mechanisms (KEE-58, DRM-19 and GRI-3.0)
are in good agreement with each other and in reasonable accordance with the
experimental evidences, maximum errors between measured and computed val-
ues being of the order of 10%. An increase of kv leads to a decrease of the
measured NO, as expected in the transition from flame to flameless regimes.
NO predictions obtained from the temperature fields provided by the reduced
and detailed schemes are very close to the experimental values, whereas simple
schemes predict NO levels lower than the measured ones by one or more orders
of magnitude. This is mainly due to the fact that the NNH intermediate mech-
anism cannot be taken into account with simplified chemistry approaches. In
flameless conditions (kv = 19.1) the measured NO are about 10 ppmv . The NO
predictions provided by GRI-3.0 are excellent, i.e. 13.6 ppmv , and satisfactory
values are also obtained with KEE-58 and the DRM-19 (23.3 and 22.8 ppmv ,
respectively). On the contrary the EDC/2-step and EDFR/2-step approaches
provide NO emissions of 0.01 and 7E-05 ppmv , respectively.
The average error metrics in Table 6.3 quantitatively confirms the aforemen-
tioned considerations. The use of global kinetic approaches, i.e. EDFR/2-step
and EDC-2step, result in very large error metrics, 265 and 426 %, respectively.
The predictions are strongly improved by employing a reduced kinetic mech-
anism (both KEE-58 and DRM-19 provide an average error metric of 41%);
however, only with a detailed mechanism error reduction is achieved. The
value of the average error metric provided by GRI-3.0, i..e. 28%, is extremely
satisfactory for the non-conventional conditions investigated. Reaching predic-
tivity for such small absolute values of NO emissions is extremely challenging
for any modeling approach, since differences of even a few ppmv lead to large
relative errors, when normalized to the measured values.
In order to better understand the role of the different formation routes on
NO emissions, their contribution to the total NO is analyzed in Figure 6.16, for
different values of the dilution ratio, kv . In flame regime, the thermal route is
sufficient to explain the observed NO in the flue gases; however, when increasing
the value of kv , the thermal route becomes negligible and NO formation is
2
As already mentioned, the high temperature peaks predicted by EDFR/2-step and
EDC/2-step are not reflected in the error metrics for temperature measurements (first and
second column of Table 6.3), because the latters are taken at a radial distance from the axis,
r = 0.014 m, characterized by the back-flow of the flue gases.
215
Figure 6.15: NO emissions in the flue gases obtained with CH4 /H2 fuel (60%
H2 by vol.) for different dilution ratios, kv . Temperature and species fields
predicted by different combustion models and kinetic mechanisms. Runs 2-6,
Table 6.1.
dominated by NNH and N2 O pathways. Therefore, it appears clear that only

the inclusion of these two mechanisms can allow a satisfactory calculation of
NO emissions in flameless conditions with hydrogen-enriched fuels. However,
these models are not always available in commercial CFD codes and they need
to be implemented by means of user-defined subroutines.
6.4 Summary
In the present Chapter, the experimental and numerical investigation of a lab-
scale burner operating in flameless conditions and fed with hydrogen-enriched
mixtures has been presented.
Modeling results indicate that both the combustion model and the kinetic
mechanism play a strong role in the determination of the main combustion
features. This is particularly evident when hydrogen is added to the fuel, as
hydrogen reactivity is usually overstimated by global kinetic approaches. The
choice of the combustion model and kinetic mechanism appears decisive for the
determination of accurate temperature and species fields, to be post-processed
for NO calculation. With such prerequisite, a reasonable agreement between
predicted and measured NO can be obtained by means of simplified one-step
approaches for all the relevant NO sources at the investigated conditions. In
particular, the moderately low and homogeneous temperatures and the presence
of hydrogen in the fuel prevent capturing NO emissions with the conventional
216
6.4. Summary
Figure 6.16: Contribution of different formation routes to total NO emissions

with CH4 /H2 fuel (60% H2 by vol.), for different dilution ratios, kv . Tem-
perature and species fields predicted by EDC with GRI-3.0. Runs 2-6, Table
6.1.
thermal and prompt routes, available in commercial codes. Therefore, the

N2 O intermediate route has been included in the model and a user-defined
subroutine has bee written on purpose to account for NO formation via the
so called NNH intermediate mechanism. Results indicate that detailed and
reduced kinetic mechanism for gas-phase combustion are required to provide
acceptable estimates of the NO emissions. The observed agreement is certainly
satisfactory and encourages further investigation.
217
Chapter 7

investigation of a micro-CHP
flameless unit
The present Chapter reports a numerical and experimental activity carried out
on a flameless system for the micro-cogeneration of heat and power (micro-
CHP). Similarly to the self-recuperative burner (Chapter 5), an experimental
campaign have been performed in collaboration with ENEL Ricerca, Livorno
- Italy, with the purpose of characterizing the system operation in off-design
conditions, through the addition of hydrogen to natural gas.
As for the FLOX® burner, the flameless combustion regime is achieved
through a particular internal aerodynamic which allows the entrainment of large
amounts of flue gases in the reaction region before combustion takes place. The
peculiarity of the system is that the flue gases cross a finned heat exchanger to
exchange heat with helium stream, involved in a Stirling cycle. Therefore, the
unit is used for the micro co-generation of thermal and electrical power (CHP).
The experiments have been integrated with a modeling activity, to help un-
derstanding the main features of the system, which shows complex geometry,
due to its industrial characteristics, and operation, due to the integration be-
tween combustion process and Stirling cycle. The available experimental data
are, then, crucial for a constructive validation of the computational approach,
in particular of sub-models adopted for describing turbulence/chemistry inter-
actions.
Given the CHP nature of the investigated system, a brief introduction re-
garding the potentials of the micro-cogeneration technologies is provided, fol-
lowed by the description of the system and the main findings of the experimental
campaign. Then, the CFD modeling of the CHP unit is discussed, showing the
comparison between numerical simulations and available experimental data.
219
Chapter 7. Experimental and numerical investigation of a micro-CHP
flameless unit
7.1 Distributed combined heat and power generation

In the perspective of exploiting hydrogen as energy carrier, attention has been
devoted to systems for the distributed co-generation of heat and power from
a single energy stream, i.e. coal, oil, natural gas, . . . Cogeneration technology
ensures higher energy conversion efficiency compared to conventional methods
of generating electricity and heat separately. This can result in lower costs and
greenhouse gas (GHG) emissions. Cogeneration is particular interesting for
micro-utilities, such as residential areas, hospitals, sport centers, commercial
centers, etc . . . A recent Life Cycle Assessment analysis on micro-cogeneration
[146] has indicated micro-CHP systems are superior in terms of GHG emissions
to both average electricity and heat supply as well as to separate production of
electricity in gas power plants and heat in condensing boilers.
Among the existing possibilities (i.e. fuel cells, micro-turbines), cogener-
ation systems based on Stirling cycles appear particularly appealing, as they
ensure high efficiency, fuel flexibility, low emissions, low noise/vibration levels
and good performance at partial load [147]. Since the combustion process takes
places outside the engine, different fuels may be used; moreover the combustion
process can be controlled independently from the engine operation. Besides,
the maintenance of Stirling engines is less critical compared to other recipro-
cating internal combustion engines, because of fewer moving parts leading to
quieter and smoother operations. Stirling engines represents a promising tech-
nology and are now beginning to appear on the market as an effective solution
for distributed cogeneration [147].
The versatility regarding the fuel topology is extremely important as it
implies that non conventional fuels such as biomasses and low-calorific gases
enriched with hydrogen could be effectively exploited, without damaging the en-
gine. Indeed, the performances of the combustion system in terms of efficiency
and pollutants emissions need to be assessed when working with non conven-
tional fuels. To this purpose the implementation of flameless burners represent
an attractive solution, due to the great potentials of the flameless technology
to deal with complex fuels. Then, such considerations have prompted the ex-
perimental and modeling activity on the CHP flameless unit.
7.2 Description of the system

The investigated system is the SOLO Stirling 161 Cogeneration Unit, developed
by SOLO® and installed at the ENEL Ricerca facilities of Livorno, Italy. The
burner is equipped with a flameless burner for the oxidation of the gaseous fuel
and a finned heat exchanger is placed inside the combustion chamber to supply
a Stirling cycle. A picture of the burner is shown in Figure 7.1. The typical
performances of the CHP unit are listed in Table 7.1. The burner is designed
to be fed with natural gas, however the effect of H2 mass fractions in the fuel
up to 10% has been investigated, to study the burner behavior in off design
220
7.2. Description of the system
Figure 7.1: SOLO Stirling 161 Cogeneration Unit.
conditions.
The SOLO Stirling CHP module consists of several sub-assemblies (Figure
7.1).
The overall housing of the unit provides a sound and heat barrier. It also
prevents from contact with hot parts and assists with the collection of possible
leaking fluids. The Stirling engine is mounted in the front part with the gen-
erator coupled directly to it. The helium storage bottle is connected to such
section.
The cylindrical and horizontally arranged combustion chamber (0.31 m
long) is closed at the end and delimited by an external (steel) case with a
diameter varying from 0.021 to 0.0235 mm. The fresh gases are fed into a
flame tube (Figure 7.2 (a)) with a diameter of 0.087 m, where the oxidation
process takes place. The flue gases flow along the axis until they reach the
chamber closed end and reverse direction, where the external enclosure diam-
eter is larger. Then, the exhausts cross the finned heat exchanger (Figure 7.2
(b)) which supplies the heat required by the Stirling cycle.
221
flameless unit
Table 7.1: Typical performances for the SOLO Stirling 161 Cogeneration Unit.
Performances
Electrical power 2 − 9 kW (±5%)
Thermal power 8 − 26 kW
Electric efficiency 22 − 24.5 % (±1)
Total efficiency 92 − 90 %
Figure 7.2: Flame tube (a) and finned heat exchanger (b).
222
Figure 7.3: Air (a) and fuel (b) feeding system.
Finally, exhaust gases are entrained into the flame tube by the incoming
air, fed to the combustion chamber through six nozzles inclined by -25 degrees
with respect to the horizontal (Figure 7.3 (a)). The high momentum air jet also
entrains the fuel, which is injected at low velocity through six nozzles inclined by
65 degrees with respect to the horizontal (Figure 7.3 (b)). The recirculation of
the flue gases in the reaction region is essential to achieve a flameless combustion
regime in the combustion chamber, as it ensures the dilution of the fresh gases
as well as their preheating above the fuel self ignition temperature. The fraction
of exhausts which do not recirculate in the reaction zone leaves the combustion
chamber, crossing a honeycomb heat exchanger for air preheating. This allows
to increase the overall efficiency, providing air at about 823 K.
The working temperatures in the combustion chamber are between 1500
and 2300 K. Before the chimney, exhaust gases are cooled down in a water-gas
heat exchanger, for the generation of thermal power, i.e. hot water, which can
be effectively exploited to satisfy, for example, the requirements of a domestic
network. An additional water-water heat exchanger is also included in the CHP
system, to recover the heat released by helium. The overall operation of the
CHP system with the main fluid streams and the heat exchanges is shown in
Figure 7.4.
The combustion chamber, the flame tube and the exhausts/helium heat
exchanger are steel made; ceramic fibers are used to insulate the external walls,
to emulate adiabatic conditions.
Regarding the experimental measurements available, five thermocouples are
placed on the external surface of the exhausts/helium heat exchanger to mea-
sure the flue gas temperature (Figure 7.5 (a)); moreover the temperatures on
top of the expansion cylinder (Figure 7.5 (b)) and before the chimney are also
monitored. As for the species composition, the composition of the exhaust
stream leaving the combustion chamber is characterized (O2 , CO2 , CO and
223
flameless unit
Figure 7.4: Schematic diagram of flow streams and heat exchanges.
NO).
The system performances can be adjusted by varying the Helium working
pressure in the range 40-150 bar, corresponding to an electrical output from 2
to 9 kW. The pressure level in the storage tank next to the machine is always
higher than the engine pressure. Helium can flow into the engine or can return
to the bottle. A magnetically controlled supply valve regulates helium from
the storage tank to the engine, whereas a small piston pump located above the
compression piston (discharge valve) is used to send helium back to the tank.
Two pressure sensors read the medium process pressure (engine pressure) and
tank pressure.
7.2.1 Stirling engine

The finned heat exchanger (Figure 7.2), also denoted as heater, provides heat
to an operating gas, i.e. helium, which drives a Stirling engine (Figure 7.6).
This Stirling engine consists of a compression and an expansion cylinder placed
at 90°. Heater, regenerator and gas cooler are positioned between the two cylin-
ders. The working gas, i.e. helium, is moved between the cylinders, undergoing
a closed thermodynamic cycle.
Inside the compression cylinder the gas is isothermally compressed at low
temperature by water cooling; then, it is moved through the regenerator to-
wards the expansion cylinder. The regenerator, hot in this case, heats the
helium stream at temperature levels of about 923 K. During the isothermal ex-
pansion the gas is constantly heated by the heater; afterwards, the gas is moved
back through the regenerator (cold in this case), where it is cooled down, to
the compression cylinder.
The heater consists of ring shaped small tubes which are heated up to
approximately 1023 K from the flue gases in the combustion chamber. The
224
Figure 7.5: Thermocouples on the heat exchanger (a) and on the top of the
expansion cylinder (b).
Figure 7.6: Functional diagram of Stirling engine.
225
flameless unit
working gas cooler is a small heat exchanger cooled by water. The regenerator
is a compressed metal fabric screen which provides a thermal storage during
the Stirling cycle, providing heat to the cold helium stream and absorbing heat
after the gas expansion. The use of a regenerator considerably improves the
cycle efficiency.
Combustion is continuous. A heat sensor attached to a small tube contin-
uously checks the temperature of the burner and relays data to the computer.
The piston rods are connected to the crankshaft by connecting rods, the
dry-running pistons in the high pressure chambers are sealed against the oil-
lubricated crankcase by (especially developed material) piston seals.
7.3 Experimental campaign
The experimental campaign of the SOLO Stirling unit has been carried out in
December 2005. Different operating conditions have been investigated, varying
the hydrogen content in the fuel stream and the operating pressure of helium.
Table 7.2 reports the boundary conditions for the experimental runs.
Table 7.3 shows the main results of the experimental campaign. Runs 1 and
6 have been removed from the Table because some problems on the natural gas
feeling line were encountered during the experiments.
Columns 2-4 in Table 7.3 report the composition of the flue gases for the
different investigated runs. It can be observed that oxygen content in the flue
gases increases with the hydrogen mass fraction. This is expected as the air
excess is higher for hydrogen-enriched mixtures1 , as indicated in Table 7.2. In
addition CO and CO2 levels decrease with H2 , because of the lower relative
carbon fraction in the fuel. Finally, regarding NO emissions, a slight decrease
is observed with increasing H2 , for higher burner loads (i.e. helium pressures,
PHe , of 90 and 120 bar). This is probably due to the increase of air excess when
hydrogen is added to the fuel, which leads to a higher dilution of the flue gases.
Columns 6-10 summarize the energetic performances of the system at the
investigated conditions. The electrical, Ẇel , and thermal power, Q̇th , are listed,
together with the efficiencies (electric, thermal and overall). It can be observed
how all the efficiencies rise with the burner load, i.e. helium pressure, reaching
values around 0.97-0.98. At lower burner loads the system performances are
lower, as expected, as the system operates in off-design conditions and the
influence of the heat losses is more significant.
1
This is due to the feeding system, which is designed to determine the volumetric air flow
rate based on that of the fuel, assumed to be natural gas.
226
Table 7.2: Summary of the experimental campaign on the SOLO Stirling CHP unit.
Run ṁN G [kg/s] ṁH2 [kg/s] YH2 [-] ṁAir [kg/s] air [%] Q̇in [kW] Q̇H2/Q̇in [%] pHe [bar]
1 2.32E-04 0 0 6.55E-03 70 10.8 0 40
2 2.65E-04 7.60E-06 0.03 8.62E-03 86 13.3 6.9 40
7.3. Experimental campaign
3 2.38E-04 1.60E-05 0.07 8.85E-03 100 12.98 14.5 40

4 2.27E-04 2.30E-05 0.10 9.52E-03 113 13.3 20.4 40
5 5.23E-04 0 0 1.29E-02 48 24.4 0 90
227
6 2.44E-04 1.30E-05 0.06 7.19E-03 59 13 12.4 90
7 4.47E-04 2.80E-05 0.07 1.44E-02 73 24.2 14 90
8 4.42E-04 3.90E-05 0.09 1.56E-02 82 25.3 18.6 90
9 6.39E-04 0 0 1.54E-02 43 29.8 0 120
10 6.00E-04 1.70E-05 0.03 1.66E-02 57 30 7 120
11 5.62E-04 3.60E-05 0.07 1.77E-02 69 30.5 14.1 120
12 5.37E-04 5.10E-05 0.09 1.88E-02 79 31.1 19.5 120
Table 7.3: Flue gas composition and main operation outputs for the runs investigated in Table 7.2.
Run CO [ppmv ] NO [ppmv ] O2 [% vol.] CO2 [% vol.] Ẇel [kW] Q̇th [kW] ηe [-] ηth [-] ηtot [-]
2 2.3 55.4 9.2 6.2 2.9 9.0 0.63 0.20 0.83
3 1.6 58.6 10.0 5.4 2.9 9.0 0.64 0.21 0.85
4 1.3 62.0 10.6 4.8 2.9 9.0 0.62 0.20 0.83
5 12.0 50.8 6.4 8.1 7.0 17.0 0.65 0.27 0.92
228
7 5.8 45.8 8.5 6.1 7.2 17.6 0.67 0.28 0.95
8 7.0 46.4 7.8 6.7 7.3 17.8 0.65 0.27 0.92
9 17.6 48.9 5.9 8.3 9.1 22.0 0.69 0.28 0.97
10 10.8 44.3 7.1 7.3 9.1 22.4 0.69 0.28 0.97
11 7.8 40.8 8.0 6.4 9.2 23.0 0.70 0.28 0.98
12 6.8 38.7 8.6 5.8 9.1 23.5 0.70 0.27 0.97
flameless unit
7.4. Numerical modeling of the micro-CHP flameless unit
7.4 Numerical modeling of the micro-CHP flameless

unit
The numerical modeling of the SOLO Stirling CHP unit has been carried out
with the commercial software FLUENT 6.3 by Ansys Inc., although preliminary
studies have been carried out with CFX 5.7.1 by Ansys Inc., to support the
experimental campaign.
A subset of the experimental runs in Table 7.2 has been selected to carry
out numerical simulations. In particular, the runs corresponding to a hydrogen
mass fraction, YH2 , in the fuel equal to 0, 7 and 9 % have been chosen, for both
the pHe = 90 bar (runs 5,7 and 8, Table 7.2) and pHe = 120 bar (runs 9,11
and 12, Table 7.2) cases. This allows to investigate the effect of hydrogen on
the reaction structure, as well as the operation of the unit at partial loads, i.e.
pHe = 90 bar.
The numerical modeling has been focused on the combustion chamber. Hy-
potheses regarding the overall operation of the CHP unit have been made to
determine the heat exchange between the flue gases and the operating fluid,
i.e. helium, of the Stirling cycle. The following Sections describe the approach
adopted.
7.4.1 Computational domain and grid

The geometry of the combustion chamber is rather complex, due to the presence
of six injection nozzles for both air and fuel streams (Figure 7.7). Moreover, a
detailed description of the flue gases/helium heat exchanger (Figure 7.2 (b)) is
not feasible, because of the large number of grid cells which would be required to
model the geometrical details of the heater. Simplifications need to be made,
to minimize the burden of the mesh on the overall computational resources,
while ensuring an acceptable accuracy in the representation.
Following the above considerations, it has been chosen to model a 60 degree
angular section of the real combustion chamber. As for the heat exchanger,
it has been treated as a fluid sub-domain with a heat sink equal to the heat
transferred from the flue gases to helium. Details about the calculation of
the heat removed by the heater are given in Section 7.4.3. Finally, the flame
tube (Figure 7.2 (a)) has been modeled as a solid domain placed inside the
combustion chamber.
To reduce the number of grid cells, a hybrid approach has been adopted
for building the mesh (Figure 7.8). For the fluid domain, a hexahedral mesh
has been used everywhere with the exception for the injection region, where a
tetrahedral mesh is preferred to reproduce accurately injection nozzles (Figure
7.8). The air and injection nozzles have been extended upstream the nozzles
of about five characteristic diameters, to ensure the flow to be fully developed
at the burner exit plane. The cell size is not uniform and a higher density
is observable in the injection region, to provide a better characterization of
229
flameless unit
Figure 7.7: Sketch of the combustion chamber and details about the feeding
system.
the reaction structure. The flame tube is also meshed with hexahedra. The
resulting total number of cells for the base mesh is about 382.000.
A grid sensitivity analysis has been carried out (Chapter 3), to provide
an estimate of the solution error associated to the discretization adopted. To
this purpose, two additional grids have been taken into account, a finer and a
coarser one. The finer grid has been obtained by applying a uniform refining
ratio, i.e. rhf = hf/hb = 0.8, to the base mesh, in axial and radial directions2 .
Similarly, the coarser grid is obtained by applying a constant coarsening factor
rhc = hc/hb = 1.25 to the base mesh. The total number of grid cells for the finer
and coarser grid is 550.000 and 270.000, respectively. The meshes have been
compared in a cold flow simulation, with boundary conditions taken from Run
9 in Table 7.2.
The solution uncertainty, Usver , has been estimated following Logan and
Nitta [81] criterion n. 4 (Chapter 3), assuming a convergence order p = 1.
The Grid Convergence Index (GCI) has been estimated from δRE applying a
safety factor, Fs = 3. Figure 7.9 shows the values δRE , and GCI for the grids
investigated. The plotted points have been obtained from the radial profiles of
axial velocity at different distances along the axis, i.e. x = 0.2 m, x = 0.3 m and
x = 0.4 m, averaging over all the local error estimates. It can be observed that
lower values of δRE , and GCI are obtained at higher axial distances, when the
mixing of fresh gases and exhausts is completed. This is expected, being the
injection region characterized by larger gradients due to the mixing process.
2
The number of divisions in tangential direction has been kept constant.
230
Figure 7.8: Computational domain and grid with details of the injection nozzles.
However, the extrapolated error, δRE , for the base grid is always below 5-7
%, which can be considered a reasonable compromise between the CPU time
requirements and the numerical accuracy.
Figure 7.10 shows the numerical error estimate for the radial profiles of
axial velocity obtained at x = 0.2 m (a), x = 0.3 m (b) and x = 0.4 m (c) with
the selected grid. Following Logan and Nitta [81], the plotted values of δRE
represent a 1-sigma (68.4%) estimate of the solution verification uncertainty,
Usver .
7.4.2 Boundary conditions

As far as boundary conditions (BCs) are concerned, mass flow rate BCs are used
to specify the inlets (Table 7.2). Fuels, i.e. methane and hydrogen, are fed at
room temperature whereas air temperature downstream of the preheater is not
known from the experimental measurements. Therefore, it has been evaluated
from the exhaust gas temperature by assuming a preheating efficiency of 50%.
Such value ensures a preheating temperature of about 823 K for nominal oper-
ating conditions, with the burner fed with natural gas and a helium pressure of
120 bar, as indicated by the burner manufacturer (www.stirling-engine.de).
A user-defined subroutine has been developed to model the heat exchanger
behavior and to estimated the air inlet temperature. As anticipated in Section
231
flameless unit
Figure 7.9: Grid Convergence Index (GCI) as a function of the coarsening

factor, r, for the set of 3D grids investigated in the numerical simulation of the
FLOX® burner. Assumed order of convergence p = 1 [81].
7.4.1, the heater has been modeled as a fluid domain with a heat sink. There-
fore, the external subroutine performs an energy balance on the overall CHP
unit to determine the heat removed by the exhaust stream. In particular, the
following operations are performed:
1. The average temperature of the exhaust gases is calculated, by averaging
the temperatures of each cell adjacent to the outlet boundary condition.
2. Assuming a preheating efficiency equal to 0.5, the air inlet temperature

is estimated from the exhaust gas temperature, Tair = 0.6 · Tf g .
3. The heat recoverable from the exhaust gases, Q̇f g , and the air preheating
heat, Q̇air , are calculated, given the flue gas temperature at the chim-
ney, Tf g,out , and the ambient air temperature, Tair,0 , known from the
experimental campaign:
(7.1)

Q̇f g = ṁf g cp,f g T f g (Tf g − Tf g,out )
(7.2)

Q̇air = ṁair cp,air T air (Tair − Tair,0 )
where cp,f g T f g and cp,f g T air are the heat capacity at constant pres-

sure calculated at average flue gas and air temperatures, respectively, i.e.
T f g = (Tf g +Tf g,out )/2 and T f g = (Tair +Tair,0 )/2. The coefficients for the
evaluation of the specific heat are taken from the CHEMKIN Thermody-
namic Database [137].
232
Figure 7.10: Estimated error, δRE , for the profiles of axial velocity obtained at
x = 0.25 m (a), x = 0.35 m (b) and x = 0.45 m (c) with the selected grid.
233
flameless unit
Figure 7.11: Control region for the evaluation of the heat transferred to the
operating fluid of the Stirling cycle.
4. On the basis of the operating scheme of the system (Figure 7.4), an energy
balance is carried out on the control region indicated in Figure 7.11,
to determine the heat transferred from the flue gases to helium in the
combustion chamber. From Figure 7.11 it is clear that the heat removed
in the flue gases/helium heat exchanger, Q̇rem , is:
Q̇rem = Q̇in − Q̇f g + Q̇air (7.3)

where Q̇in is known from the experimental campaign (Table 7.2).
5. The heat, Q̇rem , calculated at step 4 is specified as a volumetric heat sink

for the fluid region corresponding to the flue gases/helium heat exchanger
(Figure 7.7). Therefore, the flue gases crossing such fluid region are cooled
down accordingly, leading to a characteristic temperature distribution
shown in Figure 7.12.
Regarding the other boundary conditions, standard wall functions are used at
the walls and symmetry conditions are applied to the planes delimiting the
modeled angular sector.
7.4.3 Physical models

Favre-averaged Navier-Stokes equations are solved across the computational
domain, using the standard k- turbulence model for Reynolds stresses.
Following the results obtained for the self-recuperative and lab-scale burners
(Chapters 5 and 6), only the Eddy Dissipation Concept (EDC) model (Chap-
ter 2) is taken into account for turbulence chemistry interactions. A sensitiv-
ity analysis on the effect of combustion models is not viable for such system,
234
Figure 7.12: Angular Section (180 degrees) of the fluid volume corresponding
to the flue gases/helium heat exchanger. Run 12, Table 7.2.
given the significant CPU requirements associated with the selected geometri-
cal model. Therefore, the EDC model has been chosen, as it appears suited to
describe the peculiarities of the flameless combustion regime, characterized by
comparable turbulent and mixing timescales, due to the diluted characteristic
of the reaction region,
The effect of the kinetic mechanisms is investigated, comparing the results
provided by a global kinetic approach and a detailed kinetic scheme. The 4-
step global kinetic scheme by Jones and Lindstedt [60] is used for both methane
and hydrogen-enriched fuels, with kinetic parameters shown in Table 7.4. The
detailed scheme is the so-called KEE-58 [61], which involves 17 species in 58
reversible reactions.
As far as NO formation is concerned, thermal and prompt NO formation
are considered, due to the relatively high temperatures of the oxidation process,
which make the contributions of other routes, i.e. N2 O and NNH (Chapters 5
and 6), negligible. Thermal NO formation is modeled with a simplified one-step
kinetic mechanism, available in the code [125] and obtained from the Zeldovich
scheme by assuming a steady state for the N radicals and relating the O radi-
cal concentration to that of oxygen by means of the dissociation reaction [21].
The prompt NO formation from methane is modeled using a similar approach,
according to the one-step mechanism proposed by Soete [27]. For both thermal
and prompt NO formations, the Arrhenius equation is integrated over a prob-
ability density function (PDF) for temperature in order to take into account
turbulence effects.
Finally, the RTE equation is solved with the P1 radiation model. Such
model belongs to the family of the so called PN based on the expansion of the
radiation intensity into an orthogonal series of spherical harmonics [148]. The
P1 can provide reasonable results when radiation is emitted isotropically, as it
235
flameless unit
Table 7.4: Global 4-step mechanism Jones and Lindstedt [60]. Reaction rates units:
kmol · m−3 · s−1 . Activation energy units: J · kmol−1 .
Reaction Reaction rates

1.25·108
CH4 + 12 O2 → CO + 2H2 4.4 · 1011 · e− RT 0.5
CCH 4
1.25
CO 2
8
− 1.25·10
CH4 + H2 O → CO + 3H2 3.0 · 108 · e RT CCH4 CH2 O
8
1 15 −1 − 1.67·10 0.25 1.5
H2 + 2 O2 → H2 O 6.8 · 10 ·T ·e RT CH 2
CO2
7
− 8.36·10
CO + H 2 O → CO2 + H2 2.75 · 109 · e RT CCO CH2 O
happens for combustion applications; more accurate radiation models as the

Discrete Ordinate (DO) have not been used, being much more computation-
ally intensive than P1 . Similarly to the simulations of the other systems, the
Weighted-Sum-of-Grey-Gases (WSGG) model is used for predicting the spec-
tral properties [124].
The influence of molecular diffusion has been also taken into account. The
binary diffusion coefficients are calculated following the kinetic theory and a
modification of the Chapman–Enskog formula [130]. Then, an effective diffusion
coefficient of the species in the mixture is obtained by applying the Wilke’s
mixing rule [131].
7.5 Results
This Section reports the main results of the numerical activity on the CHP
flameless unit. First, the main features of the flow field inside the combustion
chamber are presented. Then, the effect of the kinetic mechanism adopted for
gas-phase oxidation is analyzed, followed by a discussion regarding the influence
of hydrogen addition to the structure of the reaction zone in the combustion
chamber. Finally the validation of the computational approach is shown, com-
paring the numerical results with the available experimental data.
7.5.1 Flow field and main operating features

Figure 7.13 shows, through vectors, the flow field in the combustion chamber.
It can be observed that, similarly to the self-recuperative burner (Chapter 5),
the flue gases are entrained into the flame tube by the high velocity air jet.
Fuel is also entrained by air and mixed with flue gas stream. The recirculation
degree in the combustion chamber has been determined following the definition
given in Chapter 5:
ṁE
kR =
ṁA + m˙F
\where ṁE is the mass flow rate of the exhaust gases recirculating into the
reaction zone, whereas ṁF and ṁA represent the fuel and air mass flow rates
236
7.5. Results
Figure 7.13: Flow pattern inside the combustion chamber, determined by the
high-momentum air jet issuing into the flame tube. Run 8, Table 7.2.
fed to the burner. The values of kR obtained for the different runs are listed
in the second column of Table 7.5: the burner generates a recirculation which
is about the same, i.e. 106%-112%, of the fresh gases fed to the system. The
average conditions in terms of oxygen availability, YO2 ,m (Column 2, Table
7.5), and temperature, Tm (Column 3, Table 7.5) after the flue and fresh gas
mixing process are approximately YO2 ,m = 0.15 and Tm = 1240 K, respectively.
Therefore, recirculation allows to achieve both dilution and preheating of the
reactants, required to operate in flameless regime.
The fifth column in Table 7.5 lists the maximum temperature increase ob-
served for the different investigated runs, as provided by the KEE-58 kinetic
mechanism. Following the criterion of Cavaliere and de Joannon [3] introduced
in Chapter 5, the burner operates in flameless mode only when fed with CH4 ,
being the observed ∆Tmax below the fuel self-ignition temperature (i.e. 853 K).
When hydrogen is added to the fuel, the observed ∆Tmax values are higher than
823 K, i.e. ∼1020-1050 K, thus indicating a strong effect of hydrogen reactiv-
ity on the flame structure. Therefore, the system operation cannot be strictly
regarded as flameless; however, further investigation of the temperature distri-
bution in the combustion chamber is needed to better clarify the characteristic
behavior of the burner when operating with hydrogen-enriched fuels.
The last column in Table 7.5 reports the ratio Ẇel/Q̇rem obtained from the nu-
merical simulations, using the user-defined function described in Section 7.4.2.
237
flameless unit
Table 7.5: Macro indicators of the main operating features of the SOLO Stirling
CHP unit. Data extracted from the numerical simulations carried out with the
KEE-58 kinetic mechanism.
Run kR [%] YO2 ,m Tm [%] ∆Tmax [K[ Ẇel/Q̇rem [-]

5 107 0.15 1231 817 0.45
7 110 0.15 1234 1057 0.48
8 112 0.16 1251 1059 0.47
9 106 0.14 1196 675 0.44
11 108 0.16 1203 1016 0.47
12 109 0.16 1205 1022 0.46
Such ratio is related to the thermal efficiency of the Stirling cycle by the regen-
erator efficiency:
Ẇel
ηth,st = (7.4)
Q̇rem + ηreg Q̇reg
where Qreg and ηreg are the heat supplied to the working fluid in the regenera-
tor and the regenerator efficiency, respectively. The non-ideal operation of the
regenerator results in values of ηreg smaller than the corresponding Ẇel/Q̇rem
ratios. Therefore, the results provided by the developed model are very sat-
isfying, considering that the expected performances of a Stirling engine with
regenerator are in the range 35-40% [149].
7.5.2 Effect of the kinetic mechanism on the flame structure

Figure 7.14 shows the contour plots of temperature inside the combustion cham-
ber fed with CH4 , at a helium pressure, PHe = 90 bar, predicted by the EDC
model with the 4-step (a) and KEE-58 (b) kinetic mechanisms. A major in-
fluence of the kinetic scheme on the structure of the reaction zone is observed,
more pronounced than for the self-recuperative and lab-scale burners (Chapters
5 and 6). The global chemistry approach results in a flame shape characteristic
of a non-premixed flame with a thin flame front and high temperature peaks
along the burner axis. Conversely, when the KEE-58 scheme is employed, the
reaction region is fully detached from the burner nozzles and shifted along the
axis, towards the closed end of the combustion chamber. Moreover the temper-
ature distribution in the flame tube is much more uniform and the maximum
predicted temperature, i.e. 2050 K, is significantly smaller than that provided
by the 4-step mechanism, i.e. 2585 K.
238
7.5. Results

CH4 , predicted by the 4-step (a) and KEE-58 (b) kinetic mechanisms. PHe =
90 bar. Runs 5, Table 7.2.
239
flameless unit

CH4 /H2 (H2 9% by wt.), predicted by the 4-step (a) and KEE-58 (b) kinetic
mechanisms. PHe = 90 bar. Runs 8, Table 7.2.
Figure 7.15 illustrates the temperature distribution in the combustion cham-

ber fed with a CH4 /H2 mixture containing 9% (by wt.) of H2 , at a helium
operating pressure PHe = 90 bar. When hydrogen is added to the fuel, the
qualitative differences between the results provided by the 4-step and KEE-58
kinetic mechanism are smaller. Both schemes indicate the ignition of the fuel
mixture in the region immediately downstream the fuel injection nozzles, with a
defined flame region developed along the burner axis. However, the results pro-
vided by KEE-58 indicate a more diffuse reaction zone, without the thin flame
sheets characterizing the temperature distribution predicted by the global ap-
proach. Finally, the temperature levels are significantly lower with KEE-58,
being the maximum predicted temperature, i.e. 2309 K, approximately 350 K
lower than that obtained with the 4-step mechanism.
Such considerations are quantitatively supported by the analysis of the ra-
dial profile of temperature provided by the 4-step and KEE-58 mechanisms,
shown in Figure 7.16 for different axial locations, i.e. x = 0.075 m (a),
x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m (d). Two cases, correspond-
ing to a H2 mass fraction in the fuel equal to 7 and 9% are taken into account, at
240
7.5. Results
Figure 7.16: Radial temperature profiles at different axial locations, i.e. x =

0.075 m (a), x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m (d), predicted
by the 4-step and KEE-58 mechanisms for a hydrogen content in the fuel equal
to 7 and 9% (by wt.), respectively.. PHe = 90 bar. Runs 7 and 8, Table 7.2.
a helium pressure PHe = 90 bar. The profiles provided by the global chemistry
approach not only indicate higher temperature levels with respect to KEE-58,
but they also show a different flame structure. With KEE-58, the temperature
distribution in the high temperature region (x = 0.075 m and x = 0.150 m)
is always flatter than that obtained with the 4-step scheme, which results in
large peaks indicating the existence of a thin diffusive flame front (Figure 7.16
(a, b)). When the distance from the burner is increased (x = 0.250 m and
x = 0.350 m), the profiles obtained with the two approaches are qualitatively
very similar, although global chemistry results in higher temperatures.
7.5.3 Effect of H2 addition to the fuel

Figure 7.17 shows the effect of hydrogen addition, i.e. 0 (a), 7 (b) and 9 (c)
% (by wt.), on the flame structure in the combustion chamber, at a helium
pressure PHe = 90 bar. Hydrogen has a significant impact on the radical pool
[117] and, then, it directly affects the ignition process. For the methane case
(7.17 (a)), the oxidation occurs close to the exit section of the flame tube, with
a homogeneous temperature distribution. When hydrogen is added to the fuel
241
flameless unit
(7.17 (b, c)), the ignition zone is shifted backward, progressively closer to the
burner, and a well defined high temperature region is observed. Moreover, the
flame becomes shorter and thinner, when increasing the H2 mass fractions.
The visual evidence provided by 7.17 supports the conclusions suggested by
the ∆Tmax values, discussed in Section 7.5.1. When H2 is added to the fuel, a
homogeneous temperature distribution is no longer observed in the combustion
chamber, being the effect of hydrogen reactivity dominant with respect to flue
gas dilution. This is in contrast to what observed for the self-recuperative
burner (Chapter 5), where a flameless like regime is observed even for higher
values of H2 mass fractions in the fuel. However, a possible explanation for
such discrepancy lies in the more enhanced recirculation which characterizes the
self-recuperative burner, i.e. kR = 130%, with respect to the present system,
kR = 110%. Such consideration also provides a possible solution to the problem:
the burner feeding system of the SOLO Stirling unit could be re-designed to
modify the internal aerodynamic of the combustion chamber, allowing to reach
higher values of kR .
The effect of hydrogen on the reaction zone structure and ignition process is
further confirmed by the analysis of the OH radical mass fraction distribution
in the conditions investigated (Figure 7.18).
7.5.4 Effect of burner load

Figure 7.19 shows the radial temperature profiles obtained at different axial lo-
cations, i.e. x = 0.075 m (a), x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m
(d), when increasing the H2 mass fraction in the fuel, for two different helium
pressures, PHe = 90 bar and PHe = 120 bar, corresponding to burner loads of
∼24 and ∼30 kW, respectively. Figure 7.19 quantitatively confirms the effect of
hydrogen on fuel ignition, as discussed in Section 7.5.3. Regarding the compar-
ison between the different operating pressures, it could appear unexpected that
the lower burner load results in higher temperature levels, especially closer to
the burner (Figure 7.19 (a, b)). However, such behavior can be easily explained
by taking into account that the burner is designed for the highest load, so that
the internal fluid dynamic is optimized for such conditions; when operating
with reduced flow rates, ignition occurs before the fresh gases have entrained a
sufficient amount of flue gases, thus resulting in higher temperature peaks.
Figure 7.19 also points out that the reaction zone is shifted towards the
end of the flame tube for the CH4 fuel case. For PHe = 90 bar, the reaction
region is collocated between x = 0.250 m ( 7.19 (c)) and x = 0.350 m (7.19
(d)), as indicated by the temperature rise at the axis, from about 1250 K to
approximately 2050 K. On the other hand, for the PHe = 120 bar case, no
ignition is observed up to an axial distance of x = 0.350 m (7.19 (d)), the
temperature profile simply increasing from the mixed temperature, Tm , value
(Column 4, Table 7.5) to the flame tube temperature. Then, the reaction zone is
shifted further towards the closed end of the combustion chamber with respect
242
7.5. Results

CH4 /H2 mixtures containing 0 (a), 7 (b) and 9 (c) % (by wt.) of H2 , predicted
by the KEE-58 kinetic mechanism. PHe = 90 bar. Runs 5, 7 and 8, Table 7.2.
243
flameless unit
Figure 7.18: OH radical mass fraction distribution in the combustion chamber

fed with CH4 /H2 mixtures containing 0 (a), 7 (b) and 9 (c) % (by wt.) of H2 ,
predicted by the KEE-58 kinetic mechanism. PHe = 90 bar. Runs 5, 7 and 8,
Table 7.2.
244
7.5. Results
Figure 7.19: Radial temperature profiles at different axial locations, i.e. x =

0.075 m (a), x = 0.150 m (b), x = 0.250 m (c) and x = 0.350 m (d), predicted
by the KEE-58 kinetic mechanism for a hydrogen content in the fuel equal to
0, 7 and 9% (by wt.), respectively.. PHe = 90, 120 bar. Runs 5, 7-9, 11 and
12, Table 7.2.
245
flameless unit

CH4 , predicted by the KEE-58 kinetic mechanisms. PHe = 120 bar. Runs 9,
Table 7.2.
to the PHe = 90 bar case, as confirmed by Figure 7.20.
7.5.5 Model validation

A first validation of the modeling approach has been already provided in Section
7.5.1, being the estimated heat flux to the working fluid of the Stirling engine
compatible with the values of thermal efficiencies found in the literature for
similar systems [149].
The present section provides a comparison between predicted and measured
CO and NO emissions from the unit. Although temperature measurements are
available on the external walls of the flue gases/helium heat exchanger, they
have not been used for validation due to the significant uncertainty regarding
their location and the large temperature variation of the fluid across the fluid
volume corresponding to the heater (Figure 7.12).
Figure 7.21 shows the comparison between measured and calculated CO (a)
and NO (b) emissions from the combustion chamber, with the different kinetic
mechanisms adopted. It is very clear that the global kinetic approach results in
major discrepancies for both CO and NO values, indicating its inadequacy to
properly describe the oxidation process; in particular, the large NO emissions
reflect a non negligible overprediction of the temperature levels by the 4-step
scheme, as suggested in Section 7.5.3.
With regard to the KEE-58 kinetic mechanism, a very good agreement with
the experimental data is observed for the PHe = 90 bar case. An excellent quan-
titative agreement is observed for both NO and CO emissions, as confirmed by
Table 7.6. The average error metrics (Chapter 3) are 15%±2% and 14%±21%
for NO and CO, respectively, with 95% confidence. The higher confidence in-
246
7.5. Results
Figure 7.21: Comparison between measured and predicted CO (a) and NO (b)
emissions. Experimental data with 95% confidence intervals. Runs 5, 7-9, 11
and 12, Table 7.2.
terval, i.e. 21%, for the CO case reflects the strong variations observed for such
chemical species during the experimental campaign and implies that the error
can range from 7% to 35%, with 95% confidence. Such level of predictivity
is extremely satisfying, especially if we take into account the absolute values
of the monitored output variables: CO and NO ranges are ∼6-12 ppmv and
∼46-51 ppmv , respectively.
The level of agreement provided by the KEE-58 mechanism is less satis-

factory for the PHe = 120 bar, especially for the NO emissions, which are
underpredicted over the entire range of the H2 mass fractions. The average er-
ror metrics (Table 7.6) for NO and CO are 58%±1% and 45%±2%, respectively,
thus indicating a significant departure from the experimental data. Indeed, re-
sults provided by KEE-58 represent a strong improvement with respect to the
global chemistry approach, which provides CO and NO values higher than the
measured ones by more than a order of magnitude. The poor performance of
KEE-58 for the PHe = 120 bar case, with respect to the PHe = 90 bar, is
probably related to the strong approximation made for estimating the air inlet
temperature (Section 7.4.2). The preheating efficiency can be significantly af-
fected by the operating conditions, i.e. flow rates, and considering an average
value of 0.6 can be inappropriate in some cases, especially at the higher loads,
when the preheater is expected to perform better. Therefore, it would be desir-
able to have such temperature measurement from the experimental campaign,
to avoid the consequences of such rough approximation. To this purpose, three
additional thermocouples have been placed near the air inlet section for the
future experimental campaigns.
247
Table 7.6: Relative and global validation metrics for NO and CO measurements at different burner loads, obtained with the
4-step and KEE-58 kinetic mechanisms. Runs 5, 7-9, 11 and 12, Table 7.2.
CO (ppmv ) Error NO (ppmv ) Error

Model Exp. ȳEe CI Model Exp ȳEe CI
e e
avg ȳe
avg avg ȳe
avg
H2 (% wt.) 4-step KEE 4-step KEE 4-step KEE 4-step KEE
248
0 284 7.9 12.0 2090 71.8 50.5
pHe = 90 bar 7 280 5.9 6.2 46.0 0.14 0.21 2480 44.7 45.6 47.0 0.15 0.01
9 423 6.3 5.9 2270 46.8 45.8
0 268 15.4 17.4 1260 25.2 49.0
pHe = 120 bar 7 291 11.2 7.9 31.6 0.45 0.02 1510 15.0 40.8 33.0 0.58 0.01
9 300 11.9 6.6 1520 14.7 39.0
flameless unit
7.6. Summary
7.6 Summary
The present Chapter has presented a numerical and experimental investigation
of a micro-CHP unit, equipped with a flameless burner which supplies the
heat required by a Stirling engine for the production of electrical power. The
numerical modeling has been focused on the combustion chamber, to investigate
the effect of hydrogen on the combustion regime features and, thus, to assess
the feasibility of flameless combustion with hydrogen enriched fuels.
The numerical modeling of the combustion chamber requires, however, the
knowledge of boundary conditions which couple the oxidation process to the
overall operation of the CHP unit. In particular, the heat transferred to the
working fluid of the Stirling cycle is required, to properly reproduce the energy
balance of the system. To this purpose, a user-defined subroutine has been
developed, to evaluate such energy flux, based on information provided by the
experimental campaign, i.e. burner load, and by the numerical simulation itself,
i.e. exhaust temperature and air inlet temperature. The external subroutine
has proved to perform satisfactory, providing values of the heat source for the
Stirling cycle which lead to expected thermal efficiencies in agreement with
those indicated in the literature for similar systems.
Regarding the main features of the combustion process, results show more
uniform temperature distribution in the combustion chamber for the higher
loads, being the system designed to optimize the mixing process for such op-
erating conditions. Finally, the addition of hydrogen to the fuel is found to
deeply affect the flame structure, resulting in a reaction region which is pro-
gressively shifted towards the burner as H2 in the fuel increases. Moreover, the
temperature distribution with hydrogen-enriched fuels denotes the existence of
a hot core, characterized by steep gradients, indicating that the system oper-
ates in an intermediate regime between flameless oxidation and a conventional
diffusion flame.
Finally, the comparison between the predicted and experimental NO mea-
surements indicate a reasonable agreement for the medium burner loads, whereas
the NO emissions at full load are underpredicted, although the experimental
trend is well captured. The reason for such inaccuracies are mainly to be found
in the large uncertainty related to the estimation of the air inlet temperature,
which directly affects the temperature increase due to the oxidation process and
ultimately influences the energy balance performed by the user-defined subrou-
tine. Following such considerations, a new experimental campaign has been
planned after a modification of the experimental apparatus: three additional
thermocouples have been placed in the combustion chamber to measure the air
temperature after preheating.
249
Concluding remarks
The present Section is aimed at providing a brief discussion regarding the main
accomplishments and original contributions of the present PhD Thesis, beside
the specific comments provided at the end of each Chapter.
A multi-scale, multi-physics approach for the investigation of hydrogen-
based fuel combustion in advanced systems operating in flameless regime has
been presented. Three different case studies have been taken into account, a
lab-scale burner, belonging to the Politecnico di Milano, and two semi-industrial
systems, installed at the experimental facilities of ENEL Ricerca in Livorno,
Italy. The numerical modeling of such systems have been integrated with an
experimental characterization, required to assess the actual predictivity of the
developed computational approaches.
A fundamental study on turbulence/chemistry interactions in turbulent
flames has been carried out in parallel with the modeling activity, recogniz-
ing the inherent coupling between turbulence and chemistry in flameless regime
and, therefore, the need of adequate approaches for describing the main features
of the oxidation process. A novel methodology based on Principal Component
Analysis has been developed and tested using high fidelity data on a variety of
turbulent reacting systems. The methodology shows excellent potentials for the
development of optimal combustion models, as it provides a rigorous mathe-
matical formalism for the extraction of those parameters in the thermo-chemical
state which control the evolution of a turbulent reacting system.
Among the experimental data analyzed with the PCA methodology, a well-
characterized flameless system, i.e. JHC burner, has been also taken into ac-
count, to investigate the main features of the flameless combustion regime. Re-
sults of the analysis have confirmed the complexity of such regime and the need
for modeling approaches able to take into account important non-equilibrium
effects. Such considerations have supported the simulation activity carried out
on the flameless burners, providing important insights regarding the modeling
choices.
The numerical modeling of the flameless systems has been carried out with
commercial Computational Fluid Dynamics (CFD) softwares. However, these
have been customized by means of state-of-the-art physical models from the
literature, to enhance their modeling capabilities. In particular, additional
models have been implemented to describe the interaction between the systems
251
Concluding remarks
and the surrounding, i.e. heat transfer, and to incorporate NO formation routes
not commonly available in the CFD codes but relevant at specific conditions.
In particular, the moderately low temperatures and fluctuations which charac-
terize flameless operation and the presence of hydrogen in the fuel may result
in dominant contributions from NO formation mechanisms different from the
conventional thermal and prompt routes. Therefore, additional pathways, i.e.
N2 O and NNH routes, have been coupled to the code.
The developed modeling approach has proved to be able to well characterize
the investigated systems in terms of thermal and environmental performances,
over a wide range of operating conditions. Specific requirements related to the
numerical modeling of the flameless combustion regime have been identified,
through the quantitative validation of the numerical results. In particular, a
satisfactory level of agreement between measured and predicted NO emissions
are achieved only with an accurate characterization of the temperature field
inside the combustion chamber and by taking into account all the relevant
formation routes at the investigated operating conditions, for instance N2 O
and NNH intermediate mechanisms.
Finally, the validation of the computational models with the V&V method-
ology has pointed out the extreme importance of code customization for the
development of actually predictive tools. This supports the approach under-
taken in the present Thesis and encourages further developments. In particular,
the implementation of a PCA based model into a CFD code for the numeri-
cal simulation of a flameless combustion process represents a very attracting
short-term perspective.
252
Bibliography
[1] W. L. Oberkampf, T. G. Trucano, Verification and Validation in Com-

putational Fluid Dynamics, Progress in Aerospace Sciences 38 (2002)
209–272.
[2] J. A. Wünning, J. G. Wünning, Flameless oxidation to reduce thermal

NO-formation, Progress in Energy and Combustion Science 23 (1997)
81–94.
[3] A. Cavaliere, M. de Joannon, Mild Combustion, Progress in Energy and

Combustion Science 30 (2004) 329–366.
[4] A. K. Gupta, Thermal characteristics of gaseous fuel flames using high

temperature air, Journal of Engineering for Gas Turbines and Power 126
(2004) 9–19.
[5] M. Falcitelli, S. Pasini, L. Tognotti, Modelling practical combustion sys-

tems and predicting NOx emissions with an integrated CFD based ap-
proach, Computers and Chemical Engineering 26 (2002) 1171–1183.
[6] M. Falcitelli, S. Pasini, N. Rossi, L. Tognotti, CFD+reactor network anal-

ysis: an integrated methodology for the modelling and optimisation of
industrial systems for energy saving and pollution reduction, Applied
Thermal Engineering 22 (2002) 971–981.
[7] T. Faravelli, L. Bua, A. Fassoldati, A. Antifora, L. Tognotti, E. Ranzi, A

new procedure for predicting NOx emissions from furnaces, Computers
and Chemical Engineering 25 (2001) 613–618.
[8] A. Frassoldati, S. Frigerio, E. Colombo, F. Inzoli, T. Faravelli, Determi-

nation of NOx emissions from strong swirling confined flames with an in-
tegrated CFD-based procedure, Chemical Engineering Science 60 (2005)
2851–2869.
[9] M. S. Skjøth-Rasmussen, O. Holm-Christensen, M. Østberg, T. S. Chris-

tensen, T. Johannessen, A. D. Jensen, P. Glarborg, H. Livbjerg, Post-
processing of detailed chemical kinetic mechanisms onto CFD simula-
tions, Computers and Chemical Engineering 28 (2004) 2351–2361.
253
Bibliography
[10] J. W. Bozzelli, A. M. Dean, O + NNH — A Possible New Route for

NOx Formation in Flames, International Journal of Chemical Kinetics 27
(1995) 1097–1109.
[11] A. N. Hayhurst, E. M. Hutchinson, Evidence for a New Way of Producing

NO via NNH in Fuel-Rich Flames at Atmospheric Pressure, Combustion
and Flame 114 (1998) 274–279.
[12] A. A. Konnov, G. Colson, J. D. Ruyck, NO formation rates for hydrogen

combustion in stirred reactors, Fuel 80 (2001) 49–65.
[13] G. Löffler, R. Sieber, M. Harasek, H. Hofbauer, R. Hauss, J. Landauf,

NOx formation in natural gas combustion: Evaluation of simplified reac-
tion schemes for CFD calculations, Fuel 85 (2006) 513–523.
[14] G. J. Rørtveit, J. E. Hustad, S. C. Li, F. A. Williams, Effects of dilu-

ents on NOx formation in hydrogen counterflow flames, Combustion and
Flame 130 (2002) 48–61.
[15] H. Guo, G. J. Smallwood, F. Liu, Y. Ju, . L. Gülder, The effect of hy-

drogen addition on flammability limit and NOx emission in ultra-lean
counterflow CH4/air premixed flames, Proceedings of the Combustion
Institute 30 (2005) 303–311.
[16] P. Glarborg, M. Alzueta, K. Dam-Johansen, J. Miller, Kinetic modelling

of hydrocarbon/nitric oxide interactions in a flow reactor, Combustion
and Flame 115 (1998) 1–27.
[17] M. Skottene, K. E. Rian, A study of NOx in hydrogen flames, Hydrogen

Energy 32 (2007) 3572–3585.
[18] The San Diego mechanism URL http://maeweb.ucsd.edu/

~combustion/cermech/.
[19] J. Li, Z. Zhao, A. Kazakov, F. L. Dryer, An updated comprehensive

kinetic model of hydrogen combustion, International Journal of Chemical
Kinetics 36 (2004) 566–575.
[20] S. C. Hill, L. D. Smoot, Modeling of nitrogen oxides formation and de-

struction in combustion systems, Progress in Energy and Combustion
Science 26 (2000) 417–458.
[21] A. A. Westenberg, Kinetics of NO and CO in lean, premixed hydrocarbon-

air flames, Combustion Science and Technology 4 (1971) 59–64.
[22] J. Warnatz, U. Maas, R. W. Dibble, Combustion. Physical and chemical

fundamentals, modeling and simulation, experiments, pollutant forma-
tion, Springer, Berlin, 2001.
254
Bibliography
[23] C. Westbrook, F. Dryer, Chemical Kinetic Modelling of Hydrocarbon

Combustion, Progress in Energy and Combustion Science 10 (1984) 1–
57.
[24] P. Malte, D. Pratt, Measurement of atomic oxygen and nitrogen oxides

in jet-stirred combustion, Proceedings of the Combustion Institute 15
(1975) 1061–1070.
[25] A. A. Konnov, G. Colson, J. D. Ruyck, The New Route Forming NO Via

NNH, Combustion and Flame 121 (2000) 548–550.
[26] C. P. Fenimore, Formation of nitric oxide in premixed hydrocarbon

flames, Proceedings of the Combustion Institute 13 (1971) 373–380.
[27] G. G. D. Soete, Overall Reaction Rates of NO and N2 Formation from

Fuel Nitrogen, Proceedings of the Combustion Institute 15 (1974) 1093–
1102.
[28] F. Backmier, K. H. Eberius, T. Just, The Formation of Nitric Oxide

and the Detection of HCN in Premixed Hydrocarbon-Air Flames at 1
Atmosphere, Combustion Science and Technology 7 (1973) 77–84.
[29] T. Shudo, K. Omori, O. Hiyama, NOx reduction and NO2 emission char-
acteristics in rich-lean combustion of hydrogen, International Journal of
Hydrogen Energy 33 (2008) 4689–4693.
[30] E. Mastorakos, A. M. Taylor, J. H. Whitelaw, Turbulent Counterflow

Flames with Reactants Diluted by Hot Products, in: T. C. Institute
(Ed.), Joint Meeting of the British and German Sections, 1993.
[31] A. Cavigiolo, M. A. Galbiati, A. Effuggi, D. Gelosa, R. Rota, Mild Com-

bustion in a Laboratory-Scale Apparatus, Combustion Science and Tech-
nology 175 (2003) 1347–1367.
[32] M. de Joannon, A. Cavaliere, T. Faravelli, E. Ranzi, P. Sabia, A. Tre-

grossi, Analysis of process parameters for steady operations in methane
mild combustion technology, Proceedings of the Combustion Institute 30
(2005) 2605–2612.
[33] M. G. Zabetakis, Flammability Characteristics of Combustible Gases and

Vapours, Tech. Rep. Bulletin 627, U.S.Department of the Interior, Bureau
of Mines, 1965.
[34] J. G. Wünning, Flammlose Oxidation yon Brennstoff mit

hochvorgewärmter Luft, Chemie Ingenieur Technik 63 (1991) 1243–
1245.
255
Bibliography
[35] T. Plessing, N. Peters, J. G. Wünning, Laseroptical investigation of

Highly preheated combustion with strong exhaust gas recirculation, Pro-
ceedings of the Combustion Institute 27 (1998) 3197–3204.
[36] A. Milani, A. Saponaro, Diluted combustion technologies, IFRF Combus-

tion Journal 200101.
[37] G. M. Choi, M. Katsuki, Advanced low NOx combustion using highly

preheated air, Energy Conversion and Management 42 (2001) 639–652.
[38] M. Flamme, Low NOx combustion technologies for high temperature ap-
plications, Energy Conversion and Management 42 (2001) 1919–1935.
[39] M. Flamme, New combustion systems for gas turbines (NGT), Applied
thermal Engineering 24 (2004) 1551–1559.
[40] Y. D. Wang, Y. Huang, D. Mcllveen-Wright, J. McMullan, N. Hewitt,

P. Eames, S. Rezvani, A techno-economic analysis of the application of
continuous staged-combustion and flameless oxidation to the combustor
design in gas turbines, Fuel Processing Technology 87 (2006) 727–736.
[41] S. Kumar, P. J. Paul, H. S. Mukunda, Investigations of the Scaling Cri-

teria for a Mild Combustion Burner, Proceedings of the Combustion In-
stitute 30 (2005) 2613–2621.
[42] P. J. Coelho, N. Peters, Numerical Simulation of a Mild Combustion

Burner, Combustion and Flame 124 (2001) 503–518.
[43] B. B. Dally, E. Riesmeier, N. Peters, Effect of Fuel Mixture on MILD

Combustion, Combustion and Flame 137 (2004) 418–431.
[44] S. Orsino, R. Weber, U. Bollettini, Numerical Simulation of Combus-

tion of Natural Gas with High-Temperature Air, Combustion Science
and Technology 170 (2001) 1–34.
[45] F. C. Christo, B. B. Dally, Modelling Turbulent Reacting Jets Issuing into

a Hot and Diluted Coflow, Combustion and Flame 142 (2005) 117–129.
[46] B. B. Dally, A. N. Karpetis, R. S. Barlow, Structure of turbulent non-

premixed jet flames in a diluted hot coflow, Proceedings of the Combus-
tion Institute 29 (2002) 1147–1154.
[47] M. Derudi, A. Villani, R. Rota, Mild combustion of industrial hydrogen-

containing byproducts, Industrial & Engineering Chemistry Research 46
(2007) 6806–6811.
[48] P. Sabia, M. de Joannon, S. Fierro, A. Tregrossi, A. Cavaliere, Hydrogen-

enriched methane Mild combustion in a well stirred reactor, Experimental
Thermal and Fluid Science 31 (2007) 469–475.
256
Bibliography
[49] P. R. Medwell, P. A. M. Kalt, B. B. Dally, Simultaneous imaging of OH,

formaldehyde, and temperature of turbulent nonpremixed jet flames in a
heated and diluted coflow, Combustion and Flame 148 (2007) 48–61.
[50] P. R. Medwell, P. A. M. Kalt, B. B. Dally, Imaging of diluted turbulent
ethylene flames stabilized on a Jet in Hot Coflow (JHC) burner, Com-
bustion and Flame 152 (2008) 100–113.
[51] AIAA, Guide for the verification and validation of computational fluid
dynamics simulations. AIAA-G-077-1998, Tech. Rep., American Institute
of Aeronautics and Astronautics, Reston, VA, 1998.
[52] P. J. Roache, Verification and validation in computational science and
engineering, Hermosa Publishers, Albuquerque, NM, 1998.
[53] N. Peters, N. Peters, Cambridge University Press, 2001.
[54] T. Poinsot, D. Veynante, Theoretical and Numerical Combustion, R.T.
Edwards, Inc., 2001.
[55] D. Veynante, L. Vervisch, Turbulent Combustion Modeling, Progress in
Energy and Combustion Science 28 (2002) 193–266.
[56] H. Tennekes, J. Lumley, A First Course In Turbulence, MIT Press, 1972.
[57] L. Prandtl, Investigation on turbulent flow, Zeitcchrift fur angewandte
Mathematik und Mechanik 5 (1925) 136–139.
[58] W. P. Jones, B. E. Launder, The Prediction of Laminarization with a
Two-Equation Model of Turbulence, Journal of Heat and Mass Transfer
15 (1972) 301–314.
[59] A. Marathe, H. Mukunda, V. Jain, Some Studies on Hydrogen-Oxygen
Diffusion Flame, Combustion Science and Technology 15 (1977) 49–64.
[60] W. P. Jones, R. P. Lindstedt, Global reaction schemes for hydrocarbon
combustion, Combustion and Flame 73 (1988) 233–249.
[61] R. Bilger, S. Starner, R. Kee, On Reduced Mechanisms for Methane-Air
Combustion in Nonpremixed Flames, Combustion and Flame 80 (1990)
135–149.
[62] A. Kazakov, M. Frenklach URL http://www.me.berkeley.edu/drm/.
[63] G. P. Smith, D. M. Golden, M. Frenklach, N. W. Moriarty, B. Eiteneer,
M. Goldenberg, C. T. Bowman, R. K. Hanson, S. Song, W. C. Gardiner,
V. V. Lissianski, Z. Qin URL http://www.me.berkeley.edu/gri-mech/.
[64] D. B. Spalding, Mixing and Chemical Reaction in Steady Confined Turbu-
lent Flames, Proceedings of the Combustion Institute 13 (1971) 649–657.
257
Bibliography
[65] B. Magnussen, B. Hjertager, On Mathematcal Modeling of Turbulent

Combustion with Special Emphasis on Soot Formation and Combustion,
Proceedings of the Combustion Institute 16 (1977) 719–729.
[66] B. F. Magnussen, On the Structure of Turbulence and a Generalized Eddy

Dissipation Concept for Chemical Reaction in Turbulent Flow, in: 19th
AIAA Aerospace Science Meeting, 1981.
[67] B. F. Magnussen, Modeling of NOx and soot formation by the Eddy Dis-
sipation Concept, in: Int. Flame Research Foundation, 1st topic Oriented
Technical Meeting, 1989.
[68] I. R. Gran, B. F. Magnussen, A Numerical Study of a Bluff-Body Sta-

bilized Diffusion Flame. Part 2. Influence of Combustion Modeling and
Finite-Rate Chemistry, Combustion Science and Technology 119 (1996)
191–217.
[69] B. F. Magnussen, The Eddy Dissipation Concept, A Bridge between Sci-

ence and Technology, in: Eccomas Thematic Conference on Computa-
tional Computation, 2005.
[70] L. F. Richardson, Weather Prediction by Numerical Process, Cambridge,

1922.
[71] I. S. Ertesvåg, B. F. Magnussen, The Eddy Dissipation Turbulent Energy

Cascade Model, Combustion Science and Technology 159 (2001) 213–236.
[72] S. Corrsin, Turbulent dissipation fluctuation, Physics of Fluids 5 (1962)

1301–1302.
[73] A. Y. Klimenko, R. W. Bilger, Conditional Moment Closure for Turbulent

Combustion, Progress in Energy and Combustion Science 25 (1999) 595–
687.
[74] S. P. Burke, T. E. W. Schumann, Diffusion. Flames, Industrial & Engi-

neering Chemistry Research 20 (1928) 998–1004.
[75] R. Bilger, The structure of turbulent non-premixed flames, Proceedings

of the Combustion Institute 22 (1988) 475–488.
[76] N. Peters, Laminar flamelet concept in turbulent combustion, Proceed-

ings of the Combustion Institute 21 (1986) 1231–1250.
[77] S. B. Pope, PDF methods for turbulent reactive flows, Progress in Energy
and Combustion Science 11 (1985) 119–192.
[78] C. Galletti, A. Parente, L. Tognotti, Numerical and experimental inves-

tigation of a mild combustion burner, Combustion and Flame 151 (2007)
649–664.
258
Bibliography
[79] A. Parente, C. Galletti, L. Tognotti, Effect of the combustion model and

kinetic mechanism on the MILD combustion in an industrial burner fed
with hydrogen enriched fuels, International Journal of Hydrogen Energy
33 (2008) 7553–7564.
[80] TNF Workshop URL http://www.ca.sandia.gov/TNF/abstract.html.
[81] R. W. Logan, C. K. Nitta, Comparing 10 methods for solution verifica-

tion, and linking to model validation, Journal of Aerospace Computing,
Information and Communication (2006) 354–373.
[82] W. L. Oberkampf, M. F. Barone, Measures of agreement between com-

putation and experiment: validation metrics, Journal of Computational
Physics 217 (2006) 5–38.
[83] E. Ranzi, M. Dente, A. Goldaniga, G. Bozzano, T. Faravelli, Lumping

procedures in detailed kinetic modeling of gasification, pyrolysis, partial
oxidation and combustion of hydrocarbon mixtures, Progress in Energy
and Combustion Science 27 (2001) 99–139.
[84] T. Lu, Y. Ju, C. K. Law, Theory of Complex CSP for Chemistry Reduc-
tion and Analysis, Combustion and Flame 126 (2001) 1445–1455.
[85] B. Bhattacharjee, D. A. Schwer, P. I. Barton, W. H. G. Jr., Optimally-

reduced kinetic models: Reaction elimination in large-scale kinetic mech-
anisms, Combustion and Flame 135 (2003) 191–208.
[86] T. Lu, C. Law, A Directed Relation Graph Method for Mechanism Re-
duction, Proceedings of the Combustion Institute 30 (2004) 1333–1341.
[87] U. Maas, S. Pope, Implementation of simplified chemical kinetics based

on intrinsic low-dimensional manifolds, Proceedings of the Combustion
Institute 24 (1993) 103–112.
[88] J. C. Sutherland, P. J. Smith, J. H. Chen, A Quantitative Method for A

Priori Evaluation of Combustion Reaction Models, Combustion Theory
& Modelling 11 (2007) 287–303.
[89] J. van Oijen, L. de Goey, Modelling of premixed counterflow flames using

the flamelet-generated manifold method, Combustion Theory and Mod-
elling 6 (2002) 463–478.
[90] I. T. Jolliffe, Principal Component Analysis, Springer, New York, NY,

1986.
[91] E. J. Jackson, A User’s Guide to Principal Components, Wiley, New York,

NY, 1991.
259
Bibliography
[92] A. Parente, J. C. Sutherland, L. Tognotti, P. Smith, Identification of

low-dimensional manifolds in turbulent flames, Proceedings of the Com-
bustion Institute 32 (2009) 579–1586.
[93] J. C. Sutherland, A. Parente, Combustion modeling using Principal Com-

ponent Analysis, Proceedings of the Combustion Institute 32 (2009)
1563–1570.
[94] K. Pearson, On lines and planes of closest fit to systems of points in space,
The Philosophical Magazine 3 (1901) 559–572.
[95] H. Hotelling, Analysis of a complex of statistical variables into principal

components, Journal of Educational Psychology 24 (1933) 417–441.
[96] C. R. Rao, The use and interpretation of principal component analysis in

applied research, Sankhya A. 26 (1964) 329–358.
[97] H. C. Keun, T. M. D. Ebbels, H. Antti, M. B. Bollard, O. Beckonert,

B. Holmes, J. C. Lindon, J. K. Nicholson, Improved analysis of mul-
tivariate data by variable stability scaling: application to NMR-based
metabolic profiling, Analytica Chimica Acta 490 (2003) 265–276.
[98] M. Shyu, S. Chen, K. Sarinnapakorn, L. Chang, A novel anomaly detec-

tion scheme based on principal component classifier, in: Proceedings of
the IEEE Foundations and New Directions of Data Mining Workshop,
in conjunction with the Third IEEE International Conference on Data
Mining, 172–179, 2003.
[99] R. S. Barlow, J. H. Frank, Effects of turbulence on species mass fractions

in methane-air jet flames, Proceedings of the Combustion Institute 27
(1998) 1087–1095.
[100] H. F. Kaiser, The application of electronic computers to factor analysis,

Educational and Psychological Measurement 20 (1960) 141–151.
[101] I. T. Jolliffe, Discarding Variables in a Principal Component Analysis I:

Artificial Data, Applied Statistics 21 (1972) 160–173.
[102] S. Frontier, Étude de la decroissance des valeurs propres dans une analyze
en composantes principales: comparison avec le modèle de baton brisé,
Journal of Experimental Marine Biology and Ecology 25 (1976) 67–75.
[103] R. B. Cattell, The scree test for the number of factors, Multivariate Be-
havioral Research 2 (1966) 245–276.
[104] R. B. Cattell, S. Vogelmann, A comprehensive trial of the scree and K.G.

criteria for determining the number of factors, Multivariate Behavioral
Research 12 (1977) 289–325.
260
Bibliography
[105] W. Krzanowski, Selection of variables to preserve multivariate structure,

using principal components, Applied Statistics 36 (1987) 22–33.
[106] G. P. McCabe, Principal Variables, Technometrics 26 (1984) 137–144.
[107] I. Cohen, Q. Tian, X. Sean, Z. Thomas, S. Huang, Feature selection using

principal feature analysis, 2002.
[108] H. Steinhaus, Sur la division des corp materiels en parties, Bulletin

L’Acadmie Polonaise des Science C1. III (IV) (1956) 801–804.
[109] M. B. Richman, Rotation of principal components, International Journal

of Climatology 6 (1986) 293–335.
[110] H. F. Kaiser, The Varimax Criterion for Analytic Rotation in Factor

Analysis, Psychometrika 23 (1958) 187–200.
[111] N. Kambhatla, T. Leen, Dimension Reduction by Local Principal Com-

ponent Analysis, Neural Computation 9 (1997) 1493–1516.
[112] R. S. Barlow, G. J. Fiechtner, C. D. Carter, J.-Y. Chen, Experiments on

the Scalar Structure of Turbulent CO/H2 /Nx Jet Flames, Combustion
and Flame 120 (2000) 549–569.
[113] R. S. Barlow, J. H. Frank, A. N. Karpetis, J.-Y. Chen, Piloted

methane/air jet flames: Transport effects and aspects of scalar structure,
Combustion and Flame 143 (2005) 433–449.
[114] R. A. Yetter, F. L. Dryer, H. Rabitz, A Comprehensive Reaction Mecha-

nism for Carbon Monoxide/Hydrogen/Oxygen Kinetics, Combustion Sci-
ence and Technology 79 (1991) 97–129.
[115] A. R. Kerstein, One-dimensional turbulence: model formulation and ap-

plication to homogeneous turbulence, shear flows, and buoyant stratified
flows, Journal of Fluid Mechanics 392 (1999) 277–334.
[116] C. E. Frouzakis, Y. G. Kevrekidis, J. Lee, K. Boulouchos, A. A. Alonso,

Proper orthogonal decomposition of direct numerical simulation data:
Data reduction and observer construction, Proceedings of the Combustion
Institute 28 (2000) 75–81.
[117] S. J. Danby, T. Echekki, Proper orthogonal decomposition analysis of

autoignition simulation data of nonhomogeneous hydrogen–air mixtures,
Combustion and Flame 144 (2006) 126–138.
[118] U. Maas, D. Thévenin, Correlation Analysis of Direct Numerical Simula-

tion Data of Turbulent Non-Premixed Flames, Proceedings of the Com-
261
Bibliography
[119] M. R. Roomina, R. W. Bilger, Conditional Moment Closure (CMC) Pre-

dictions of a Turbulent Methane-Air Jet Flame, Combustion and Flame
125 (2001) 1176–1195.
[120] J. C. Sutherland, Evaluation of Mixing and Reaction Models for Large-

Eddy Simulation of Nonpremixed Combustion Using Direct Numerical
Simulation, Ph.D. thesis, University of Utah, Salt Lake City, UT, May
2004.
[121] P. A. McMurtry, S. Menon, A. R. Kerstein, A Linear Eddy Subgrid Model

for Turbulent Reacting Flows: Application to Hydrogen-Air Combustion,
Proceedings of the Combustion Institute 24 (1992) 271–278.
[122] T. Echekki, A. Kerstein, J.-Y. Chen, T. D. Dreeben, One-dimensional

turbulence simulation of turbulent jet diffusion flames: model formulation
and illustrative applications, Combustion and Flame 125 (2001) 1083–
1105.
[123] A. Al-Halbouni, A. Giese, M. Flamme, M. Brune, Flameless Oxidation

and Continued Staged Air Combustion Systems for Gas Turbines, Clean
Air 5 (2004) 391–405.
[124] T. F. Smith, Z. F. Shen, J.N.Friedman, Evaluation of coefficients for the

weighted sum of gray gases model, Journal of Heat Transfer-transactions
of the ASME 104 (1982) 602–608.
[125] CFX 5.7.1 User Guide, 2004.
[126] J. C. Slattery, R. B. Bird, AIChE Journal 42 (1958) 137–142.
[127] B. Chakraborty, P. Paul, H. S. Mukunda, Evaluation of combustion mod-

els for high speed H2 /air confined mixing layer using DNS data, Combus-
tion and Flame (2000) 195–209.
[128] M. Frenklach, H. Wang, C.-L. Yu, M. Goldenberg, C. Bowman, R. Han-

son, D. Davidson, E. Chang, G. Smith, D. Golden, W. Gardiner, V. Lis-
sianski URL http://www.me.berkeley.edu/gri_mech/.
[129] Fluent 6.3 User Guide, 2006.
[130] H. A. McGee, Molecular engineering, McGraw-Hill, New York, NY, 1991.
[131] S. P. Wilke, A viscosity equation for gas mixtures, Journal of Chemical

Physics 18 (1950) 517–519.
[132] J. A. Miller, T. C. Bowman, Mechanism and modeling of nitrogen chem-

istry in combustion, Progress Energy Combustion Science 15 (1989) 287–
338.
262
Bibliography
[133] A. Cuoci, A. Frassoldati, G. Buzzi-Ferraris, T. Faravelli, E. Ranzi, The

ignition, combustion and flame structure of carbon monoxide/hydrogen
mixtures. Note 2: Fluid dynamics and kinetic aspects of syngas combus-
tion, International Journal of Hydrogen Energy 32 (2007) 3486–3500.
[134] A. Frassoldati, A. Cuoci, T. Faravelli, E. Ranzi, S. Colantuoni, P. D.

Martino, G. Cinque, Experimental and modeling study of a low NOx
combustor for aero-engine turbofan, Combustion Science and Technology
181 (2009) in press.
[135] T. Faravelli, A. Frassoldati, E. Ranzi, Kinetic Modeling of Mutual Inter-

actions in NO-Hydrocarbon High Temperature Oxidation, Combustion
and Flame 135 (2003) 97–112.
[136] T. Faravelli, A. Frassoldati, E. Ranzi, Kinetic Modeling of Mutual Inter-

actions in NO-Hydrocarbon Low Temperature Oxidation, Combustion
and Flame 133 (2003) 188–207.
[137] R. J. Kee, F. Rupley, J. A. Miller, The CHEMKIN Thermodynamic Data

Base, Report SAND89-8008, Sandia National Laboratories, Livermore,
CA, 1989.
[138] M. Ilbas, The effect of thermal radiation and radiation models on hydro-
gen–hydrocarbon combustion modeling, International Journal of Hydro-
gen Energy 30 (2005) 1113–1126.
[139] A. R. Choudhuri, S. R. Gollahalli, Global characteristics of hydro-

gen–hydrocarbon composite fuel turbulent jet flames, International Jour-
nal of Hydrogen Energy 28 (2003) 445–454.
[140] R. Choudhuri, S. R. Gollahalli, Intermediate radical concentrations in

hydrogen–natural gas blended fuel jet, International Journal of Hydrogen
Energy 29 (2004) 1293–1302.
[141] T. Echekki, J. H. Chen, Direct numerical simulation of autoignition in

non-homogeneous hydrogen–air mixtures, Combustion and Flame 134
(2003) 169–191.
[142] G. G. Szego, B. B. Dally, G. J. Nathan, Scaling of NOx emissions from

a laboratory-scale mild combustion furnace, Combustion and Flame 154
(2008) 281–295.
[143] M. Mancini, P. Schwöppe, R. Weber, S. Orsino, On mathematical mod-

elling of flameless combustion, Combustion and Flame 150 (2007) 54–59.
[144] A. P. Morse, Axisymmetric Turbulent Shear Flows with and without

Swirl, Ph.D. thesis, London University, 1977.
263
Bibliography
[145] F. Smith, Z. F. Shen, J. N. Friedman, Evaluation of coefficients for the

weighted sum of gray gases model, Journal of Heat Transfer 104 (1982)
602–608.
[146] M. Pehnt, Environmental impacts of distributed energy systemm - The

case of micro cogeneration, Environmental Science & Policy 11 (2008)
25–37.
[147] H. I. Onovwiona, V. I. Ugursal, Residential cogeneration systems: review

of the current technology, Renewable and Sustainable Energy Reviews 10
(2006) 389–431.
[148] G. D. Raithby, Equations of motion for reacting, particle-laden flows,

Tech. Rep., Thermal Science Ltd., 1991.
[149] Y. Timoumi, I. Tlili, S. B. Nasrallah, Performance optimization of Stirling

engines, Renewable Energy 33 (2008) 2134–2144.
264
List of Publications
International Journals
1. A. Parente, J. C. Sutherland, L. Tognotti, P. Smith, Identification of
low-dimensional manifolds in turbulent flames, Proceedings of the Com-
2. J. C. Sutherland, A. Parente, Combustion modeling using Principal Com-
ponent Analysis, Proceedings of the Combustion Institute 32 (2009) 1563–
1570.
3. A. Parente, C. Galletti, L. Tognotti, Effect of the combustion model and
kinetic mechanism on the MILD combustion in an industrial burner fed
with hydrogen enriched fuels, International Journal of Hydrogen Energy
33 (2008) 7553–7564.
4. C. Galletti, A. Parente, L. Tognotti, Numerical and experimental inves-
tigation of a mild combustion burner, Combustion and Flame 151 (2007)
649–664.
Submitted to International Journals

1. C. Galletti, A. Parente, M. Derudi, R. Rota, L. Tognotti, Numerical and
experimental analysis of NO emissions from a lab-scale burner fed with
hydrogen-enriched fuels and operating in MILD combustion, International
Journal of Hydrogen Energy.
In preparation for International Journals

1. A. Parente, C. Galletti, L. Tognotti, Experimental and numerical inves-
tigation of a burner for the micro-cogeneration of heat and power.
2. A. Parente, C. Galletti, L. Tognotti, NO formation in flameless combus-
tion: a simplified modeling approach.
3. A. Parente, J. C. Sutherland, L. Tognotti, P. Smith, Principal Compo-
nents Analysis for turbulence-chemistry interaction modeling.
265
International Conferences
1. A. Parente, E. Cresci, C. Galletti, M. Schiavetti, J. Riccardi, L. Tognotti,
Effect of H2 Content in the Fuel on the Flameless Combustion in Indus-
trial Burners, in: 31st Meeting of the Italian Section of The Combustion
Institute, 17-20 June 2006, Torino, Italy (oral presentation).
2. C. Galletti, A. Parente, M. Derudi, L. Tognotti, R. Rota, Numerical and
Experimental Analysis of the Mild Combustion of Hydrogen-Containing
Fuels, in: 31st Meeting of the Italian Section of The Combustion Institute,
17-20 June 2006, Torino, Italy.
3. M. Falcitelli, A. Parente, L. Tognotti, N. Rossi, Methods of Reactor Net-
work Analysis for Combustion Chambers Simulated with Commercial
CFD Codes, in: 31st Meeting of the Italian Section of The Combustion
Institute, 17-20 June 2006, Torino, Italy.
4. A. Parente, E. Cresci, C. Galletti, L. Tognotti, Mild Combustion of H2 -
Enriched Fuels: A Numerical and Experimental Study, in: SIAM Inter-
national Conference on Numerical Combustion (NC08), March 31 - April
2, 2008. Monterey, California, USA (oral presentation).
5. A. Parente, J. C. Sutherland, L. Tognotti, P. J. Smith, Capturing Ex-
tinction and Reignition in Turbulent Flames: a Principal Component
Analysis Approach, in: SIAM International Conference on Numerical
Combustion (NC08), March 31 - April 2, 2008. Monterey, California,
USA (oral presentation).
6. J. C. Sutherland, A. Parente, Combustion Reaction Model Generation
using Principal Component Analysis, in: 2007 Fall Meeting of the Western
States Section of the Combustion Institute, 16 – 17 October 2007, Sandia
National Laboratories, Livermore, CA, USA.
7. A. Parente, J. C. Sutherland, L. Tognotti, P. J. Smith, Identification of
Low-Dimensional Manifolds in Turbulent Flames, in: 2007 Fall Meet-
ing of the Western States Section of the Combustion Institute, 16 – 17
October 2007, Sandia National Laboratories, Livermore, CA, USA (oral
presentation).
8. C. Galletti, A. Parente, L. Tognotti, M. Derudi, A. Villani, R. Rota,
Experimental and Numerical Investigation of a Burner Operated in Mild
Combustion Conditions, in: Third European Combustion Meeting, 11-13
April 2007, Crete, Greece.
9. E. Biagini, A. Parente, C. Galletti, L. Tognotti, Integrated Systems of
Hydrogen Production and its Use in Distributed Generation, in: First
Mediterranean Congress Chemical Engineering for Environment, 4–6 Oc-
tober 2006, Venice, Italy.
266
10. C. Galletti, A. Parente, L. Tognotti, Mild Combustion with Hydrogen

Enriched Fuels, in: 31st International Symposium on Combustion (Work
in Progress), 6-11 August 2006, Heidelberg, Germany.
11. C. Galletti, P. Gheri, G. Gigliucci, A. Parente, M. Schiavetti, S. Soricetti,

L.Tognotti, Flameless Combustion of H2 -Enriched Fuels: a CFD Aided
Experimental Investigation, in: 29th Meeting of the Italian Section of
The Combustion Institute, 14-17 June 2006, Pisa, Italy.
12. C. Galletti, M. Lenzi, A. Parente, L. Tognotti, Numerical Modeling of

Hydrogen Diffusion Jet Flames: the Role of the Combustion Model, in:
29th Meeting of the Italian Section of The Combustion Institute, 14-17
June 2006, Pisa, Italy (oral presentation).
13. C. Galletti, A. Parente, L. Tognotti, CFD Simulations of Mild Combus-

tion, in: ECCOMAS Thematic Conference on Computational Combus-
tion, 21-24 June 2005, Lisbon, Portugal.
14. C. Galletti, A. Parente, L. Tognotti, Flameless Combustion with H2 En-

riched Fuels, in: ICheaP-7 Conference, 15 - 18 May 2005, Giardini Naxos
- Taormina, Italy.
267
Acknowledgments
I feel extremely grateful to many people who made my PhD a terrific human
and professional experience.
My appreciation and gratitude go to my Thesis advisor, Prof. Leonardo
Tognotti, for the enthusiasm, the inspiration and the continuous support. I
hope I awarded his appreciation and trust with my work. I would also like to
thank Dr. Chiara Galletti for her constant scientific aid, the fruitful discussion
and for her friendship.
I have spent more than a year at the University of Utah, and I would have
stayed there even more! I express my sincere gratitude to Prof. Philip J. Smith,
for giving me the opportunity of joining his group, for his expertise, kindness
and unique ability to be an intellectual leader and to inspire passion in research.
I would like to thank Prof. Adel Sarofim for allowing me to join the Utah group
and Prof. James C. Sutherland for the key support provided throughout the
development of my research activity. Thanks to Dr. Jennifer Spinti and Dr.
Jeremy Thornock, for being always available to answer my questions and for
making me feel less alone, so far from home.
Thanks to my friends, in Pisa and Salt Lake City, in particular Alberto,
Lucrezia, Niveditha and Enrico. I have always been sure to have someone close
to me, to share the good and bad of my everyday life!
Vorrei concludere ringraziando la mia famiglia. Anche se viviamo lontani da
molti anni ormai, mi sono sempre sentito sospinto dall’amore dei miei genitori,
Titti e Augusto, e di mia sorella Daniela. Infine, grazie Aurora, soprattutto
per la pazienza. Ma non ti preoccupare, questo “terzo round”, vissuto ancora
insieme, conclude definitivamente la nostra vita da studenti!
. . . And to really conclude, a picture I recently discovered (thanks Marco!),
which finally clarified to me the real meaning of my research. . .
269
from: The Stunning Science of Everything, Arnold Nick, Scholastic Reference,
2006
270

PHD Parente Full

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

PHD Parente Full

Hochgeladen von

Copyright:

Verfügbare Formate

Università di Pisa

Scuola di Dottorato in Ingegneria "Leonardo da Vinci"

Corso di Dottorato di Ricerca in

Ingegneria Chimica e dei Materiali

Tesi di Dottorato di Ricerca

Experimental and Numerical Investigation

Prof. Ing. Leonardo Tognotti

Dott. Ing. Chiara Galletti

Scuola di Dottorato in Ingegneria "Leonardo da Vinci"

Corso di Dottorato di Ricerca in

Ingegneria Chimica e dei Materiali

Tesi di Dottorato di Ricerca

Prof. Ing. Leonardo Tognotti

Dott. Ing. Chiara Galletti

Il est très simple: on ne voit bien qu’avec le cœur.

sessment of the level of agreement between experimental data and numerical

1 Advanced systems for hydrogen-based fuel combustion 1

2 Turbulent combustion modeling 23

2.3.3.1 The mixture fraction variable . . . . . . . . . . 42

3 Methodology and Case Studies 49

4 Principal Components Analysis for turbulence-chemistry inter-

4.6.4 Parametrizing Source Terms . . . . . . . . . . . . . . . . 129

5 Experimental and numerical investigation of a self-recuperative

6 Experimental and numerical investigation of the flameless com-

7 Experimental and numerical investigation of a micro-CHP flame-

Concluding remarks 251

List of Publications 265

1.1 Combustion of hydrogen containing fuels: issues and solutions. 2

2.1 Verification and Validation methodology [1]. . . . . . . . . . . . 24

4.7 Scree-graph and histograms of the q largest eigenvalues for the

5.1 Longitudinal section of the FLOX® burner. . . . . . . . . . . . 136

5.7 NO concentration as a function of time during a typical run of

5.23 Temperature distribution in the combustion chamber for Aair,in =

5.37 Comparison of predicted and measured NO emissions in the flue

5.47 Radial profiles of temperature, H, O and OH radical mass frac-

6.1 Laboratory scale experimental equipment. Sketch and main di-

6.13 CH2 O mass fraction distribution in the combustion chamber fed

7.1 SOLO Stirling 161 Cogeneration Unit. . . . . . . . . . . . . . . 221

7.15 Temperature distribution in the combustion chamber fed with

1.1 Reaction parameters for the extended Zeldovich mechanism and

5.1 Summary of the experimental campaign n.1 on the FLOX® burner.141

6.1 Summary of the experimental campaign on the lab-scale burner. 201

7.1 Typical performances for the SOLO Stirling 161 Cogeneration

7.5 Macro indicators of the main operating features of the SOLO

a Oxygen reaction order in the one-step prompt

LP CA Local Principal Components Analysis

RSM Reynolds Stress Model

Advanced systems for

Figure 1.1: Combustion of hydrogen containing fuels: issues and solutions.

1.1 NOx formation and control in combustion pro-

1.1.1 Homogeneous NOx formation mechanisms

1.1.1.1 Thermal NO mechanism

An additional elementary reaction is added to the thermal mechanism when

The extended Zeldovich mechanism includes the effects of O, H and OH

Reaction expressiona Forward Backward

kr,1 kr,2 [N O]2

An improvement to the equilibrium approach is obtained by accounting for

1.1.1.2 N2 O intermediate NO mechanism

1.1.1.3 NNH intermediate NO mechanism

Assuming that the NH radicals formed in reaction (1.11) are completely

1.1.1.4 Prompt NO mechanism

TA,prompt = 30215 K Aprompt = 1.2 · 106 s−1 . (1.21)